- Kernel - mailweb.openeuler.org

openEuler Kernel SIG双周例会
by openEuler conference 18 Mar '26

18 Mar '26

您好！ Kernel 邀请您参加 2026-03-20 14:00 召开的WeLink会议(自动录制) 会议主题：openEuler Kernel SIG双周例会会议发起人： LiaoTao_Wave 会议内容： 1. 进展update 2. 议题征集中（新增议题可直接填报至会议纪要看板）会议链接：https://meeting.huaweicloud.com:36443/#/j/986250028 会议纪要：https://etherpad.openeuler.org/p/Kernel-meetings 更多资讯尽在：https://www.openeuler.org/zh/ Hello! Kernel invites you to attend the WeLink conference(auto recording) will be held at 2026-03-20 14:00 The subject of the conference is openEuler Kernel SIG双周例会 The sponsor of the conference is LiaoTao_Wave Summary: 1. 进展update 2. 议题征集中（新增议题可直接填报至会议纪要看板） You can join the meeting at https://meeting.huaweicloud.com:36443/#/j/986250028 Add topics at https://etherpad.openeuler.org/p/Kernel-meetings More information: https://www.openeuler.org/en/

1 0

[PATCH OLK-5.10] mm: make sure freeram is smaller than totalram
by Liu Shixin 17 Mar '26

17 Mar '26

From: Yang Yingliang <yangyingliang(a)huawei.com> hulk inclusion category: bugfix bugzilla: https://atomgit.com/openeuler/kernel/issues/8748 CVE: NA -------------------------------------------------- The memory stat is not done in real time, it have some gap with real value. In CPU-less NUMA node, the values of MemTotal and MemFree can be nearly equal, the gap make cause MemFree is bigger than MemTotal, it leads MemUsed is negative which print as a large positive number. cat /sys/devices/system/node/node17/meminfo Node 17 MemTotal: 4194304 kB Node 17 MemFree: 4195552 kB Node 17 MemUsed: 18446744073709550368 kB Node 17 Active: 52 kB Node 17 Inactive: 320 kB Node 17 Active(anon): 0 kB Node 17 Inactive(anon): 0 kB Node 17 Active(file): 52 kB Node 17 Inactive(file): 320 kB Node 17 Unevictable: 0 kB Node 17 Mlocked: 0 kB Node 17 Dirty: 0 kB Node 17 Writeback: 0 kB Node 17 FilePages: 372 kB Node 17 Mapped: 320 kB Node 17 AnonPages: 0 kB Node 17 Shmem: 0 kB Node 17 KernelStack: 0 kB Node 17 PageTables: 0 kB Node 17 NFS_Unstable: 0 kB Node 17 Bounce: 0 kB Node 17 WritebackTmp: 0 kB Node 17 KReclaimable: 0 kB Node 17 Slab: 0 kB Node 17 SReclaimable: 0 kB Node 17 SUnreclaim: 0 kB Node 17 AnonHugePages: 79872 kB Node 17 ShmemHugePages: 0 kB Node 17 ShmemPmdMapped: 0 kB Node 17 FileHugePages: 0 kB Node 17 FilePmdMapped: 0 kB Node 17 HugePages_Total: 0 Node 17 HugePages_Free: 0 Node 17 HugePages_Surp: 0 To avoid this exception, make MemFree equals MemTotal, when MemFree is bigger than MemTotal Signed-off-by: Yang Yingliang <yangyingliang(a)huawei.com> --- mm/page_alloc.c | 1 + 1 file changed, 1 insertion(+) diff --git a/mm/page_alloc.c b/mm/page_alloc.c index 72d5303f8000..02db21de93aa 100644 --- a/mm/page_alloc.c +++ b/mm/page_alloc.c @@ -5837,6 +5837,7 @@ void si_meminfo_node(struct sysinfo *val, int nid) val->totalram = managed_pages; val->sharedram = node_page_state(pgdat, NR_SHMEM); val->freeram = sum_zone_node_page_state(nid, NR_FREE_PAGES); + val->freeram = min(val->freeram, val->totalram); #ifdef CONFIG_HIGHMEM for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) { struct zone *zone = &pgdat->node_zones[zone_type]; -- 2.33.0

2 1

[PATCH OLK-5.10] mm: make sure freeram is smaller than totalram
by Liu Shixin 17 Mar '26

17 Mar '26

From: Yang Yingliang <yangyingliang(a)huawei.com> The memory stat is not done in real time, it have some gap with real value. In CPU-less NUMA node, the values of MemTotal and MemFree can be nearly equal, the gap make cause MemFree is bigger than MemTotal, it leads MemUsed is negative which print as a large positive number. cat /sys/devices/system/node/node17/meminfo Node 17 MemTotal: 4194304 kB Node 17 MemFree: 4195552 kB Node 17 MemUsed: 18446744073709550368 kB Node 17 Active: 52 kB Node 17 Inactive: 320 kB Node 17 Active(anon): 0 kB Node 17 Inactive(anon): 0 kB Node 17 Active(file): 52 kB Node 17 Inactive(file): 320 kB Node 17 Unevictable: 0 kB Node 17 Mlocked: 0 kB Node 17 Dirty: 0 kB Node 17 Writeback: 0 kB Node 17 FilePages: 372 kB Node 17 Mapped: 320 kB Node 17 AnonPages: 0 kB Node 17 Shmem: 0 kB Node 17 KernelStack: 0 kB Node 17 PageTables: 0 kB Node 17 NFS_Unstable: 0 kB Node 17 Bounce: 0 kB Node 17 WritebackTmp: 0 kB Node 17 KReclaimable: 0 kB Node 17 Slab: 0 kB Node 17 SReclaimable: 0 kB Node 17 SUnreclaim: 0 kB Node 17 AnonHugePages: 79872 kB Node 17 ShmemHugePages: 0 kB Node 17 ShmemPmdMapped: 0 kB Node 17 FileHugePages: 0 kB Node 17 FilePmdMapped: 0 kB Node 17 HugePages_Total: 0 Node 17 HugePages_Free: 0 Node 17 HugePages_Surp: 0 To avoid this exception, make MemFree equals MemTotal, when MemFree is bigger than MemTotal Signed-off-by: Yang Yingliang <yangyingliang(a)huawei.com> --- mm/page_alloc.c | 1 + 1 file changed, 1 insertion(+) diff --git a/mm/page_alloc.c b/mm/page_alloc.c index 72d5303f8000..02db21de93aa 100644 --- a/mm/page_alloc.c +++ b/mm/page_alloc.c @@ -5837,6 +5837,7 @@ void si_meminfo_node(struct sysinfo *val, int nid) val->totalram = managed_pages; val->sharedram = node_page_state(pgdat, NR_SHMEM); val->freeram = sum_zone_node_page_state(nid, NR_FREE_PAGES); + val->freeram = min(val->freeram, val->totalram); #ifdef CONFIG_HIGHMEM for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) { struct zone *zone = &pgdat->node_zones[zone_type]; -- 2.33.0

2 1

[PATCH OLK-5.10] l2tp: fix double dst_release() on sk_dst_cache race
by Li Xiasong 17 Mar '26

17 Mar '26

hulk inclusion category: bugfix bugzilla: https://atomgit.com/openeuler/kernel/issues/8696 Reference: https://lore.kernel.org/netdev/20251215145537.5085-1-m.lobanov@rosa.ru/ ---------------------------------------- A reproducible rcuref - imbalanced put() warning is observed under IPv6 L2TP (pppol2tp) traffic with blackhole routes, indicating an imbalance in dst reference counting for routes cached in sk->sk_dst_cache and pointing to a subtle lifetime/synchronization issue between the helpers that validate and drop cached dst entries. rcuref - imbalanced put() WARNING: CPU: 0 PID: 899 at lib/rcuref.c:266 rcuref_put_slowpath+0x1ce/0x240 lib/rcuref.c:266 Modules linked in: CPSocket connected tcp:127.0.0.1:48148,server=on <-> 127.0.0.1:33750 Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.16.3-debian-1.16.3-2 04/01/2014 RIP: 0010:rcuref_put_slowpath+0x1ce/0x240 lib/rcuref.c:266 Call Trace: <TASK> __rcuref_put include/linux/rcuref.h:97 [inline] rcuref_put include/linux/rcuref.h:153 [inline] dst_release+0x291/0x310 net/core/dst.c:167 __sk_dst_check+0x2d4/0x350 net/core/sock.c:604 __inet6_csk_dst_check net/ipv6/inet6_connection_sock.c:76 [inline] inet6_csk_route_socket+0x6ed/0x10c0 net/ipv6/inet6_connection_sock.c:104 inet6_csk_xmit+0x12f/0x740 net/ipv6/inet6_connection_sock.c:121 l2tp_xmit_queue net/l2tp/l2tp_core.c:1214 [inline] l2tp_xmit_core net/l2tp/l2tp_core.c:1309 [inline] l2tp_xmit_skb+0x1404/0x1910 net/l2tp/l2tp_core.c:1325 pppol2tp_sendmsg+0x3ca/0x550 net/l2tp/l2tp_ppp.c:302 sock_sendmsg_nosec net/socket.c:729 [inline] __sock_sendmsg net/socket.c:744 [inline] ____sys_sendmsg+0xab2/0xc70 net/socket.c:2609 ___sys_sendmsg+0x11d/0x1c0 net/socket.c:2663 __sys_sendmmsg+0x188/0x450 net/socket.c:2749 __do_sys_sendmmsg net/socket.c:2778 [inline] __se_sys_sendmmsg net/socket.c:2775 [inline] __x64_sys_sendmmsg+0x98/0x100 net/socket.c:2775 do_syscall_x64 arch/x86/entry/common.c:52 [inline] do_syscall_64+0x64/0x140 arch/x86/entry/common.c:83 entry_SYSCALL_64_after_hwframe+0x76/0x7e RIP: 0033:0x7fe6960ec719 </TASK> The race occurs between the lockless UDPv6 transmit path (udpv6_sendmsg() -> sk_dst_check()) and the locked L2TP/pppol2tp transmit path (pppol2tp_sendmsg() -> l2tp_xmit_skb() -> ... -> inet6_csk_xmit() → __sk_dst_check()), when both handle the same obsolete dst from sk->sk_dst_cache: the UDPv6 side takes an extra reference and atomically steals and releases the cached dst, while the L2TP side, using a stale cached pointer, still calls dst_release() on it, and together these updates produce an extra final dst_release() on that dst, triggering rcuref - imbalanced put(). The Race Condition: Initial: sk->sk_dst_cache = dst ref(dst) = 1 Thread 1: sk_dst_check() Thread 2: __sk_dst_check() ------------------------ ---------------------------- sk_dst_get(sk): rcu_read_lock() dst = rcu_dereference(sk->sk_dst_cache) rcuref_get(dst) succeeds rcu_read_unlock() // ref = 2 dst = __sk_dst_get(sk) // reads same dst from sk_dst_cache // ref still = 2 (no extra get) [both see dst obsolete & check() == NULL] sk_dst_reset(sk): old = xchg(&sk->sk_dst_cache, NULL) // old = dst dst_release(old) // drop cached ref // ref: 2 -> 1 RCU_INIT_POINTER(sk->sk_dst_cache, NULL) // cache already NULL after xchg dst_release(dst) // ref: 1 -> 0 dst_release(dst) // tries to drop its own ref after final put // rcuref_put_slowpath() -> "rcuref - imbalanced put()" Make L2TP's IPv6 transmit path stop using inet6_csk_xmit() (and thus __sk_dst_check()) and instead open-code the same routing and transmit sequence using the existing sk_dst_check() with dst reference counting and ip6_dst_lookup_flow(). The new code builds a flowi6 from the socket fields in the same way as inet6_csk_route_socket(), then performs dst lookup using sk_dst_check() with proper reference counting, attaches the resulting dst to the skb via skb_dst_set(), and finally invokes ip6_xmit() for transmission. This makes both the UDPv6 and L2TP IPv6 paths use the same dst-cache handling logic for a given socket and removes the possibility that the lockless sk_dst_check() and the locked __sk_dst_check() concurrently drop the same cached dst and trigger the rcuref - imbalanced put() warning under concurrent traffic. Found by Linux Verification Center (linuxtesting.org) with Syzkaller. Fixes: b0270e91014d ("ipv4: add a sock pointer to ip_queue_xmit()") Signed-off-by: Li Xiasong <lixiasong1(a)huawei.com> --- net/l2tp/l2tp_core.c | 60 +++++++++++++++++++++++++++++++++++++++++--- 1 file changed, 57 insertions(+), 3 deletions(-) diff --git a/net/l2tp/l2tp_core.c b/net/l2tp/l2tp_core.c index 97350350478d..80a20c079a9a 100644 --- a/net/l2tp/l2tp_core.c +++ b/net/l2tp/l2tp_core.c @@ -995,19 +995,73 @@ static int l2tp_build_l2tpv3_header(struct l2tp_session *session, void *buf) return bufp - optr; } +#if IS_ENABLED(CONFIG_IPV6) +static int l2tp_xmit_ipv6(struct sock *sk, struct sk_buff *skb) +{ + struct ipv6_pinfo *np = inet6_sk(sk); + struct inet_sock *inet = inet_sk(sk); + struct in6_addr *final_p, final; + struct dst_entry *dst; + struct flowi6 fl6; + int err; + + memset(&fl6, 0, sizeof(fl6)); + fl6.flowi6_proto = sk->sk_protocol; + fl6.daddr = sk->sk_v6_daddr; + fl6.saddr = np->saddr; + fl6.flowlabel = np->flow_label; + IP6_ECN_flow_xmit(sk, fl6.flowlabel); + + fl6.flowi6_oif = sk->sk_bound_dev_if; + fl6.flowi6_mark = sk->sk_mark; + fl6.fl6_sport = inet->inet_sport; + fl6.fl6_dport = inet->inet_dport; + fl6.flowi6_uid = sk->sk_uid; + + security_sk_classify_flow(sk, flowi6_to_flowi_common(&fl6)); + + rcu_read_lock(); + final_p = fl6_update_dst(&fl6, rcu_dereference(np->opt), &final); + rcu_read_unlock(); + + dst = sk_dst_check(sk, np->dst_cookie); + if (!dst) { + dst = ip6_dst_lookup_flow(sock_net(sk), sk, &fl6, final_p); + if (IS_ERR(dst)) { + sk->sk_err_soft = -PTR_ERR(dst); + sk->sk_route_caps = 0; + kfree_skb(skb); + return PTR_ERR(dst); + } + + ip6_dst_store(sk, dst_clone(dst), NULL, NULL); + } + + skb_dst_set(skb, dst); + fl6.daddr = sk->sk_v6_daddr; + + rcu_read_lock(); + err = ip6_xmit(sk, skb, &fl6, sk->sk_mark, rcu_dereference(np->opt), + np->tclass, sk->sk_priority); + rcu_read_unlock(); + return err; +} +#endif + /* Queue the packet to IP for output: tunnel socket lock must be held */ static int l2tp_xmit_queue(struct l2tp_tunnel *tunnel, struct sk_buff *skb, struct flowi *fl) { int err; + struct sock *sk = tunnel->sock; skb->ignore_df = 1; skb_dst_drop(skb); #if IS_ENABLED(CONFIG_IPV6) - if (l2tp_sk_is_v6(tunnel->sock)) - err = inet6_csk_xmit(tunnel->sock, skb, NULL); + if (l2tp_sk_is_v6(sk)) + err = l2tp_xmit_ipv6(sk, skb); else #endif - err = ip_queue_xmit(tunnel->sock, skb, fl); + err = ip_queue_xmit(sk, skb, fl); return err >= 0 ? NET_XMIT_SUCCESS : NET_XMIT_DROP; } -- 2.34.1

2 1

[PATCH OLK-6.6] l2tp: fix double dst_release() on sk_dst_cache race
by Li Xiasong 17 Mar '26

17 Mar '26

hulk inclusion category: bugfix bugzilla: https://atomgit.com/openeuler/kernel/issues/8696 Reference: https://lore.kernel.org/netdev/20251215145537.5085-1-m.lobanov@rosa.ru/ ---------------------------------------- A reproducible rcuref - imbalanced put() warning is observed under IPv6 L2TP (pppol2tp) traffic with blackhole routes, indicating an imbalance in dst reference counting for routes cached in sk->sk_dst_cache and pointing to a subtle lifetime/synchronization issue between the helpers that validate and drop cached dst entries. rcuref - imbalanced put() WARNING: CPU: 0 PID: 899 at lib/rcuref.c:266 rcuref_put_slowpath+0x1ce/0x240 lib/rcuref.c:266 Modules linked in: CPSocket connected tcp:127.0.0.1:48148,server=on <-> 127.0.0.1:33750 Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.16.3-debian-1.16.3-2 04/01/2014 RIP: 0010:rcuref_put_slowpath+0x1ce/0x240 lib/rcuref.c:266 Call Trace: <TASK> __rcuref_put include/linux/rcuref.h:97 [inline] rcuref_put include/linux/rcuref.h:153 [inline] dst_release+0x291/0x310 net/core/dst.c:167 __sk_dst_check+0x2d4/0x350 net/core/sock.c:604 __inet6_csk_dst_check net/ipv6/inet6_connection_sock.c:76 [inline] inet6_csk_route_socket+0x6ed/0x10c0 net/ipv6/inet6_connection_sock.c:104 inet6_csk_xmit+0x12f/0x740 net/ipv6/inet6_connection_sock.c:121 l2tp_xmit_queue net/l2tp/l2tp_core.c:1214 [inline] l2tp_xmit_core net/l2tp/l2tp_core.c:1309 [inline] l2tp_xmit_skb+0x1404/0x1910 net/l2tp/l2tp_core.c:1325 pppol2tp_sendmsg+0x3ca/0x550 net/l2tp/l2tp_ppp.c:302 sock_sendmsg_nosec net/socket.c:729 [inline] __sock_sendmsg net/socket.c:744 [inline] ____sys_sendmsg+0xab2/0xc70 net/socket.c:2609 ___sys_sendmsg+0x11d/0x1c0 net/socket.c:2663 __sys_sendmmsg+0x188/0x450 net/socket.c:2749 __do_sys_sendmmsg net/socket.c:2778 [inline] __se_sys_sendmmsg net/socket.c:2775 [inline] __x64_sys_sendmmsg+0x98/0x100 net/socket.c:2775 do_syscall_x64 arch/x86/entry/common.c:52 [inline] do_syscall_64+0x64/0x140 arch/x86/entry/common.c:83 entry_SYSCALL_64_after_hwframe+0x76/0x7e RIP: 0033:0x7fe6960ec719 </TASK> The race occurs between the lockless UDPv6 transmit path (udpv6_sendmsg() -> sk_dst_check()) and the locked L2TP/pppol2tp transmit path (pppol2tp_sendmsg() -> l2tp_xmit_skb() -> ... -> inet6_csk_xmit() → __sk_dst_check()), when both handle the same obsolete dst from sk->sk_dst_cache: the UDPv6 side takes an extra reference and atomically steals and releases the cached dst, while the L2TP side, using a stale cached pointer, still calls dst_release() on it, and together these updates produce an extra final dst_release() on that dst, triggering rcuref - imbalanced put(). The Race Condition: Initial: sk->sk_dst_cache = dst ref(dst) = 1 Thread 1: sk_dst_check() Thread 2: __sk_dst_check() ------------------------ ---------------------------- sk_dst_get(sk): rcu_read_lock() dst = rcu_dereference(sk->sk_dst_cache) rcuref_get(dst) succeeds rcu_read_unlock() // ref = 2 dst = __sk_dst_get(sk) // reads same dst from sk_dst_cache // ref still = 2 (no extra get) [both see dst obsolete & check() == NULL] sk_dst_reset(sk): old = xchg(&sk->sk_dst_cache, NULL) // old = dst dst_release(old) // drop cached ref // ref: 2 -> 1 RCU_INIT_POINTER(sk->sk_dst_cache, NULL) // cache already NULL after xchg dst_release(dst) // ref: 1 -> 0 dst_release(dst) // tries to drop its own ref after final put // rcuref_put_slowpath() -> "rcuref - imbalanced put()" Make L2TP's IPv6 transmit path stop using inet6_csk_xmit() (and thus __sk_dst_check()) and instead open-code the same routing and transmit sequence using the existing sk_dst_check() with dst reference counting and ip6_dst_lookup_flow(). The new code builds a flowi6 from the socket fields in the same way as inet6_csk_route_socket(), then performs dst lookup using sk_dst_check() with proper reference counting, attaches the resulting dst to the skb via skb_dst_set(), and finally invokes ip6_xmit() for transmission. This makes both the UDPv6 and L2TP IPv6 paths use the same dst-cache handling logic for a given socket and removes the possibility that the lockless sk_dst_check() and the locked __sk_dst_check() concurrently drop the same cached dst and trigger the rcuref - imbalanced put() warning under concurrent traffic. Found by Linux Verification Center (linuxtesting.org) with Syzkaller. Fixes: b0270e91014d ("ipv4: add a sock pointer to ip_queue_xmit()") Signed-off-by: Li Xiasong <lixiasong1(a)huawei.com> --- net/l2tp/l2tp_core.c | 60 +++++++++++++++++++++++++++++++++++++++++--- 1 file changed, 57 insertions(+), 3 deletions(-) diff --git a/net/l2tp/l2tp_core.c b/net/l2tp/l2tp_core.c index f85b383811d5..79a5f635b049 100644 --- a/net/l2tp/l2tp_core.c +++ b/net/l2tp/l2tp_core.c @@ -1000,19 +1000,73 @@ static int l2tp_build_l2tpv3_header(struct l2tp_session *session, void *buf) return bufp - optr; } +#if IS_ENABLED(CONFIG_IPV6) +static int l2tp_xmit_ipv6(struct sock *sk, struct sk_buff *skb) +{ + struct ipv6_pinfo *np = inet6_sk(sk); + struct inet_sock *inet = inet_sk(sk); + struct in6_addr *final_p, final; + struct dst_entry *dst; + struct flowi6 fl6; + int err; + + memset(&fl6, 0, sizeof(fl6)); + fl6.flowi6_proto = sk->sk_protocol; + fl6.daddr = sk->sk_v6_daddr; + fl6.saddr = np->saddr; + fl6.flowlabel = np->flow_label; + IP6_ECN_flow_xmit(sk, fl6.flowlabel); + + fl6.flowi6_oif = sk->sk_bound_dev_if; + fl6.flowi6_mark = sk->sk_mark; + fl6.fl6_sport = inet->inet_sport; + fl6.fl6_dport = inet->inet_dport; + fl6.flowi6_uid = sk->sk_uid; + + security_sk_classify_flow(sk, flowi6_to_flowi_common(&fl6)); + + rcu_read_lock(); + final_p = fl6_update_dst(&fl6, rcu_dereference(np->opt), &final); + rcu_read_unlock(); + + dst = sk_dst_check(sk, np->dst_cookie); + if (!dst) { + dst = ip6_dst_lookup_flow(sock_net(sk), sk, &fl6, final_p); + if (IS_ERR(dst)) { + WRITE_ONCE(sk->sk_err_soft, -PTR_ERR(dst)); + sk->sk_route_caps = 0; + kfree_skb(skb); + return PTR_ERR(dst); + } + + ip6_dst_store(sk, dst_clone(dst), NULL, NULL); + } + + skb_dst_set(skb, dst); + fl6.daddr = sk->sk_v6_daddr; + + rcu_read_lock(); + err = ip6_xmit(sk, skb, &fl6, sk->sk_mark, rcu_dereference(np->opt), + np->tclass, sk->sk_priority); + rcu_read_unlock(); + return err; +} +#endif + /* Queue the packet to IP for output: tunnel socket lock must be held */ static int l2tp_xmit_queue(struct l2tp_tunnel *tunnel, struct sk_buff *skb, struct flowi *fl) { int err; + struct sock *sk = tunnel->sock; skb->ignore_df = 1; skb_dst_drop(skb); #if IS_ENABLED(CONFIG_IPV6) - if (l2tp_sk_is_v6(tunnel->sock)) - err = inet6_csk_xmit(tunnel->sock, skb, NULL); + if (l2tp_sk_is_v6(sk)) + err = l2tp_xmit_ipv6(sk, skb); else #endif - err = ip_queue_xmit(tunnel->sock, skb, fl); + err = ip_queue_xmit(sk, skb, fl); return err >= 0 ? NET_XMIT_SUCCESS : NET_XMIT_DROP; } -- 2.34.1

2 1

[PATCH OLK-6.6 0/2] mptcp: fix soft lockup in mptcp_recvmsg()
by Li Xiasong 17 Mar '26

17 Mar '26

Li Xiasong (1): mptcp: fix soft lockup in mptcp_recvmsg() Paolo Abeni (1): mptcp: fix MSG_PEEK stream corruption net/mptcp/protocol.c | 44 +++++++++++++++++++++++++++++++++----------- 1 file changed, 33 insertions(+), 11 deletions(-) -- 2.34.1

2 3

[PATCH OLK-6.6] l2tp: fix double dst_release() on sk_dst_cache race
by Li Xiasong 17 Mar '26

17 Mar '26

hulk inclusion category: bugfix bugzilla: https://atomgit.com/openeuler/kernel/issues/8696 Reference: https://lore.kernel.org/netdev/20251215145537.5085-1-m.lobanov@rosa.ru/ ---------------------------------------- A reproducible rcuref - imbalanced put() warning is observed under IPv6 L2TP (pppol2tp) traffic with blackhole routes, indicating an imbalance in dst reference counting for routes cached in sk->sk_dst_cache and pointing to a subtle lifetime/synchronization issue between the helpers that validate and drop cached dst entries. rcuref - imbalanced put() WARNING: CPU: 0 PID: 899 at lib/rcuref.c:266 rcuref_put_slowpath+0x1ce/0x240 lib/rcuref.c:266 Modules linked in: CPSocket connected tcp:127.0.0.1:48148,server=on <-> 127.0.0.1:33750 Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.16.3-debian-1.16.3-2 04/01/2014 RIP: 0010:rcuref_put_slowpath+0x1ce/0x240 lib/rcuref.c:266 Call Trace: <TASK> __rcuref_put include/linux/rcuref.h:97 [inline] rcuref_put include/linux/rcuref.h:153 [inline] dst_release+0x291/0x310 net/core/dst.c:167 __sk_dst_check+0x2d4/0x350 net/core/sock.c:604 __inet6_csk_dst_check net/ipv6/inet6_connection_sock.c:76 [inline] inet6_csk_route_socket+0x6ed/0x10c0 net/ipv6/inet6_connection_sock.c:104 inet6_csk_xmit+0x12f/0x740 net/ipv6/inet6_connection_sock.c:121 l2tp_xmit_queue net/l2tp/l2tp_core.c:1214 [inline] l2tp_xmit_core net/l2tp/l2tp_core.c:1309 [inline] l2tp_xmit_skb+0x1404/0x1910 net/l2tp/l2tp_core.c:1325 pppol2tp_sendmsg+0x3ca/0x550 net/l2tp/l2tp_ppp.c:302 sock_sendmsg_nosec net/socket.c:729 [inline] __sock_sendmsg net/socket.c:744 [inline] ____sys_sendmsg+0xab2/0xc70 net/socket.c:2609 ___sys_sendmsg+0x11d/0x1c0 net/socket.c:2663 __sys_sendmmsg+0x188/0x450 net/socket.c:2749 __do_sys_sendmmsg net/socket.c:2778 [inline] __se_sys_sendmmsg net/socket.c:2775 [inline] __x64_sys_sendmmsg+0x98/0x100 net/socket.c:2775 do_syscall_x64 arch/x86/entry/common.c:52 [inline] do_syscall_64+0x64/0x140 arch/x86/entry/common.c:83 entry_SYSCALL_64_after_hwframe+0x76/0x7e RIP: 0033:0x7fe6960ec719 </TASK> The race occurs between the lockless UDPv6 transmit path (udpv6_sendmsg() -> sk_dst_check()) and the locked L2TP/pppol2tp transmit path (pppol2tp_sendmsg() -> l2tp_xmit_skb() -> ... -> inet6_csk_xmit() → __sk_dst_check()), when both handle the same obsolete dst from sk->sk_dst_cache: the UDPv6 side takes an extra reference and atomically steals and releases the cached dst, while the L2TP side, using a stale cached pointer, still calls dst_release() on it, and together these updates produce an extra final dst_release() on that dst, triggering rcuref - imbalanced put(). The Race Condition: Initial: sk->sk_dst_cache = dst ref(dst) = 1 Thread 1: sk_dst_check() Thread 2: __sk_dst_check() ------------------------ ---------------------------- sk_dst_get(sk): rcu_read_lock() dst = rcu_dereference(sk->sk_dst_cache) rcuref_get(dst) succeeds rcu_read_unlock() // ref = 2 dst = __sk_dst_get(sk) // reads same dst from sk_dst_cache // ref still = 2 (no extra get) [both see dst obsolete & check() == NULL] sk_dst_reset(sk): old = xchg(&sk->sk_dst_cache, NULL) // old = dst dst_release(old) // drop cached ref // ref: 2 -> 1 RCU_INIT_POINTER(sk->sk_dst_cache, NULL) // cache already NULL after xchg dst_release(dst) // ref: 1 -> 0 dst_release(dst) // tries to drop its own ref after final put // rcuref_put_slowpath() -> "rcuref - imbalanced put()" Make L2TP's IPv6 transmit path stop using inet6_csk_xmit() (and thus __sk_dst_check()) and instead open-code the same routing and transmit sequence using the existing sk_dst_check() with dst reference counting and ip6_dst_lookup_flow(). The new code builds a flowi6 from the socket fields in the same way as inet6_csk_route_socket(), then performs dst lookup using sk_dst_check() with proper reference counting, attaches the resulting dst to the skb via skb_dst_set(), and finally invokes ip6_xmit() for transmission. This makes both the UDPv6 and L2TP IPv6 paths use the same dst-cache handling logic for a given socket and removes the possibility that the lockless sk_dst_check() and the locked __sk_dst_check() concurrently drop the same cached dst and trigger the rcuref - imbalanced put() warning under concurrent traffic. Found by Linux Verification Center (linuxtesting.org) with Syzkaller. Fixes: b0270e91014d ("ipv4: add a sock pointer to ip_queue_xmit()") Signed-off-by: Li Xiasong <lixiasong1(a)huawei.com> --- net/l2tp/l2tp_core.c | 60 +++++++++++++++++++++++++++++++++++++++++--- 1 file changed, 57 insertions(+), 3 deletions(-) diff --git a/net/l2tp/l2tp_core.c b/net/l2tp/l2tp_core.c index f85b383811d5..79a5f635b049 100644 --- a/net/l2tp/l2tp_core.c +++ b/net/l2tp/l2tp_core.c @@ -1000,19 +1000,73 @@ static int l2tp_build_l2tpv3_header(struct l2tp_session *session, void *buf) return bufp - optr; } +#if IS_ENABLED(CONFIG_IPV6) +static int l2tp_xmit_ipv6(struct sock *sk, struct sk_buff *skb) +{ + struct ipv6_pinfo *np = inet6_sk(sk); + struct inet_sock *inet = inet_sk(sk); + struct in6_addr *final_p, final; + struct dst_entry *dst; + struct flowi6 fl6; + int err; + + memset(&fl6, 0, sizeof(fl6)); + fl6.flowi6_proto = sk->sk_protocol; + fl6.daddr = sk->sk_v6_daddr; + fl6.saddr = np->saddr; + fl6.flowlabel = np->flow_label; + IP6_ECN_flow_xmit(sk, fl6.flowlabel); + + fl6.flowi6_oif = sk->sk_bound_dev_if; + fl6.flowi6_mark = sk->sk_mark; + fl6.fl6_sport = inet->inet_sport; + fl6.fl6_dport = inet->inet_dport; + fl6.flowi6_uid = sk->sk_uid; + + security_sk_classify_flow(sk, flowi6_to_flowi_common(&fl6)); + + rcu_read_lock(); + final_p = fl6_update_dst(&fl6, rcu_dereference(np->opt), &final); + rcu_read_unlock(); + + dst = sk_dst_check(sk, np->dst_cookie); + if (!dst) { + dst = ip6_dst_lookup_flow(sock_net(sk), sk, &fl6, final_p); + if (IS_ERR(dst)) { + WRITE_ONCE(sk->sk_err_soft, -PTR_ERR(dst)); + sk->sk_route_caps = 0; + kfree_skb(skb); + return PTR_ERR(dst); + } + + ip6_dst_store(sk, dst_clone(dst), NULL, NULL); + } + + skb_dst_set(skb, dst); + fl6.daddr = sk->sk_v6_daddr; + + rcu_read_lock(); + err = ip6_xmit(sk, skb, &fl6, sk->sk_mark, rcu_dereference(np->opt), + np->tclass, sk->sk_priority); + rcu_read_unlock(); + return err; +} +#endif + /* Queue the packet to IP for output: tunnel socket lock must be held */ static int l2tp_xmit_queue(struct l2tp_tunnel *tunnel, struct sk_buff *skb, struct flowi *fl) { int err; + struct sock *sk = tunnel->sock; skb->ignore_df = 1; skb_dst_drop(skb); #if IS_ENABLED(CONFIG_IPV6) - if (l2tp_sk_is_v6(tunnel->sock)) - err = inet6_csk_xmit(tunnel->sock, skb, NULL); + if (l2tp_sk_is_v6(sk)) + err = l2tp_xmit_ipv6(sk, skb); else #endif - err = ip_queue_xmit(tunnel->sock, skb, fl); + err = ip_queue_xmit(sk, skb, fl); return err >= 0 ? NET_XMIT_SUCCESS : NET_XMIT_DROP; } -- 2.34.1

2 1

[PATCH OLK-5.10 0/3] arm64/mpam: Extending MPAM features and bugfix
by Quanmin Yan 17 Mar '26

17 Mar '26

Extending the MPAM features and bugfix in OLK-5.10. Quanmin Yan (3): arm64/mpam: Add quirk to shrink MATA PMG range arm64/mpam: Add quirk for MPAM MSMON_MBWU monitor NRDY bit arm64/mpam: Allow MBMIN to be set to 0 arch/arm64/kernel/mpam/mpam_device.c | 33 +++++++++++++++++++++++---- arch/arm64/kernel/mpam/mpam_resctrl.c | 8 ++++--- 2 files changed, 34 insertions(+), 7 deletions(-) -- 2.43.0

2 4

[PATCH OLK-5.10 0/3] arm64/mpam: Extending MPAM features and bugfix
by Quanmin Yan 17 Mar '26

17 Mar '26

Extending the MPAM features and bugfix in OLK-5.10. Quanmin Yan (3): arm64/mpam: Add quirk to shrink MATA PMG range arm64/mpam: Add quirk for MPAM MSMON_MBWU monitor NRDY bit arm64/mpam: Allow MBMIN to be set to 0 arch/arm64/kernel/mpam/mpam_device.c | 33 +++++++++++++++++++++++---- arch/arm64/kernel/mpam/mpam_resctrl.c | 8 ++++--- 2 files changed, 34 insertions(+), 7 deletions(-) -- 2.43.0

2 4

[PATCH OLK-6.6] bfq: Lock when clearing the q->elevator entry
by Zizhi Wo 16 Mar '26

16 Mar '26

hulk inclusion category: bugfix bugzilla: https://atomgit.com/openeuler/kernel/issues/8717 -------------------------------- In the error path out_free of bfq_init_queue(), the last reference count of eq->kobj is released, causing eq to be freed. Before the caller blk_mq_init_sched() executes "q->elevator = NULL", another process modifies the blkcgroup interface, such as disabling iocost, which triggers wbt_enable_default() to access the already freed elevator, leading to a UAF issue. Lock when clearing the q->elevator entry to solve this issue. And do cleanup on some repeated nulling processes. Fixes: 671fae5e5129 ("blk-wbt: don't enable throttling if default elevator is bfq") Signed-off-by: Zizhi Wo <wozizhi(a)huawei.com> --- block/bfq-iosched.c | 3 +++ block/blk-mq-sched.c | 4 ++-- block/blk-wbt.c | 2 ++ block/elevator.c | 1 - 4 files changed, 7 insertions(+), 3 deletions(-) diff --git a/block/bfq-iosched.c b/block/bfq-iosched.c index a8ebf3962f11..a12f0196dc51 100644 --- a/block/bfq-iosched.c +++ b/block/bfq-iosched.c @@ -7417,6 +7417,9 @@ static int bfq_init_queue(struct request_queue *q, struct elevator_type *e) return 0; out_free: + spin_lock_irq(&q->queue_lock); + q->elevator = NULL; + spin_unlock_irq(&q->queue_lock); kfree(bfqd); kobject_put(&eq->kobj); return -ENOMEM; diff --git a/block/blk-mq-sched.c b/block/blk-mq-sched.c index 7b48630b63a7..bb166403d311 100644 --- a/block/blk-mq-sched.c +++ b/block/blk-mq-sched.c @@ -499,8 +499,6 @@ int blk_mq_init_sched(struct request_queue *q, struct elevator_type *e) err_free_map_and_rqs: blk_mq_sched_free_rqs(q); blk_mq_sched_tags_teardown(q, flags); - - q->elevator = NULL; return ret; } @@ -550,5 +548,7 @@ void blk_mq_exit_sched(struct request_queue *q, struct elevator_queue *e) if (e->type->ops.exit_sched) e->type->ops.exit_sched(e); blk_mq_sched_tags_teardown(q, flags); + spin_lock_irq(&q->queue_lock); q->elevator = NULL; + spin_unlock_irq(&q->queue_lock); } diff --git a/block/blk-wbt.c b/block/blk-wbt.c index 6b81f2c47279..71ca08deb485 100644 --- a/block/blk-wbt.c +++ b/block/blk-wbt.c @@ -732,9 +732,11 @@ void wbt_enable_default(struct gendisk *disk) struct rq_qos *rqos; bool enable = IS_ENABLED(CONFIG_BLK_WBT_MQ); + spin_lock_irq(&q->queue_lock); if (q->elevator && test_bit(ELEVATOR_FLAG_DISABLE_WBT, &q->elevator->flags)) enable = false; + spin_unlock_irq(&q->queue_lock); /* Throttling already enabled? */ rqos = wbt_rq_qos(q); diff --git a/block/elevator.c b/block/elevator.c index ba072d8f660e..e80381765b3c 100644 --- a/block/elevator.c +++ b/block/elevator.c @@ -704,7 +704,6 @@ void elevator_disable(struct request_queue *q) elv_unregister_queue(q); elevator_exit(q); blk_queue_flag_clear(QUEUE_FLAG_SQ_SCHED, q); - q->elevator = NULL; q->nr_requests = q->tag_set->queue_depth; blk_add_trace_msg(q, "elv switch: none"); -- 2.39.2

2 1

[PATCH OLK-6.6] bfq: Lock when clearing the q->elevator entry
by Zizhi Wo 16 Mar '26

16 Mar '26

hulk inclusion category: bugfix bugzilla: https://atomgit.com/openeuler/kernel/issues/8717 -------------------------------- In the error path out_free of bfq_init_queue(), the last reference count of eq->kobj is released, causing eq to be freed. Before the caller blk_mq_init_sched() executes "q->elevator = NULL", another process modifies the blkcgroup interface, such as disabling iocost, which triggers wbt_enable_default() to access the already freed elevator, leading to a UAF issue. Lock when clearing the q->elevator entry to solve this issue. And do cleanup on some repeated nulling processes. Fixes: 671fae5e5129 ("blk-wbt: don't enable throttling if default elevator is bfq") Signed-off-by: Zizhi Wo <wozizhi(a)huawei.com> Signed-off-by: Zizhi Wo <wozizhi(a)huaweicloud.com> --- block/bfq-iosched.c | 3 +++ block/blk-mq-sched.c | 4 ++-- block/blk-wbt.c | 2 ++ block/elevator.c | 1 - 4 files changed, 7 insertions(+), 3 deletions(-) diff --git a/block/bfq-iosched.c b/block/bfq-iosched.c index a8ebf3962f11..a12f0196dc51 100644 --- a/block/bfq-iosched.c +++ b/block/bfq-iosched.c @@ -7417,6 +7417,9 @@ static int bfq_init_queue(struct request_queue *q, struct elevator_type *e) return 0; out_free: + spin_lock_irq(&q->queue_lock); + q->elevator = NULL; + spin_unlock_irq(&q->queue_lock); kfree(bfqd); kobject_put(&eq->kobj); return -ENOMEM; diff --git a/block/blk-mq-sched.c b/block/blk-mq-sched.c index 7b48630b63a7..bb166403d311 100644 --- a/block/blk-mq-sched.c +++ b/block/blk-mq-sched.c @@ -499,8 +499,6 @@ int blk_mq_init_sched(struct request_queue *q, struct elevator_type *e) err_free_map_and_rqs: blk_mq_sched_free_rqs(q); blk_mq_sched_tags_teardown(q, flags); - - q->elevator = NULL; return ret; } @@ -550,5 +548,7 @@ void blk_mq_exit_sched(struct request_queue *q, struct elevator_queue *e) if (e->type->ops.exit_sched) e->type->ops.exit_sched(e); blk_mq_sched_tags_teardown(q, flags); + spin_lock_irq(&q->queue_lock); q->elevator = NULL; + spin_unlock_irq(&q->queue_lock); } diff --git a/block/blk-wbt.c b/block/blk-wbt.c index 6b81f2c47279..71ca08deb485 100644 --- a/block/blk-wbt.c +++ b/block/blk-wbt.c @@ -732,9 +732,11 @@ void wbt_enable_default(struct gendisk *disk) struct rq_qos *rqos; bool enable = IS_ENABLED(CONFIG_BLK_WBT_MQ); + spin_lock_irq(&q->queue_lock); if (q->elevator && test_bit(ELEVATOR_FLAG_DISABLE_WBT, &q->elevator->flags)) enable = false; + spin_unlock_irq(&q->queue_lock); /* Throttling already enabled? */ rqos = wbt_rq_qos(q); diff --git a/block/elevator.c b/block/elevator.c index ba072d8f660e..e80381765b3c 100644 --- a/block/elevator.c +++ b/block/elevator.c @@ -704,7 +704,6 @@ void elevator_disable(struct request_queue *q) elv_unregister_queue(q); elevator_exit(q); blk_queue_flag_clear(QUEUE_FLAG_SQ_SCHED, q); - q->elevator = NULL; q->nr_requests = q->tag_set->queue_depth; blk_add_trace_msg(q, "elv switch: none"); -- 2.39.2

2 1

[PATCH OLK-6.6] xattr: switch to CLASS(fd)
by Yongjian Sun 16 Mar '26

16 Mar '26

From: Al Viro <viro(a)zeniv.linux.org.uk> mainline inclusion from mainline-v6.13-rc1 commit a71874379ec8c6e788a61d71b3ad014a8d9a5c08 category: bugfix bugzilla: https://atomgit.com/src-openeuler/kernel/issues/13855 CVE: CVE-2024-14027 Reference: https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?… -------------------------------- Reviewed-by: Christian Brauner <brauner(a)kernel.org> Signed-off-by: Al Viro <viro(a)zeniv.linux.org.uk> Conflicts: fs/xattr.c [mainline 88a2f6468d01 not applied] Signed-off-by: Yongjian Sun <sunyongjian1(a)huawei.com> --- fs/xattr.c | 25 +++++++++---------------- 1 file changed, 9 insertions(+), 16 deletions(-) diff --git a/fs/xattr.c b/fs/xattr.c index 7574d24b982e..44dffb1f90c5 100644 --- a/fs/xattr.c +++ b/fs/xattr.c @@ -810,16 +810,13 @@ SYSCALL_DEFINE4(lgetxattr, const char __user *, pathname, SYSCALL_DEFINE4(fgetxattr, int, fd, const char __user *, name, void __user *, value, size_t, size) { - struct fd f = fdget(fd); - ssize_t error = -EBADF; + CLASS(fd, f)(fd); if (!f.file) - return error; + return -EBADF; audit_file(f.file); - error = getxattr(file_mnt_idmap(f.file), f.file->f_path.dentry, + return getxattr(file_mnt_idmap(f.file), f.file->f_path.dentry, name, value, size); - fdput(f); - return error; } /* @@ -886,15 +883,12 @@ SYSCALL_DEFINE3(llistxattr, const char __user *, pathname, char __user *, list, SYSCALL_DEFINE3(flistxattr, int, fd, char __user *, list, size_t, size) { - struct fd f = fdget(fd); - ssize_t error = -EBADF; + CLASS(fd, f)(fd); if (!f.file) - return error; + return -EBADF; audit_file(f.file); - error = listxattr(f.file->f_path.dentry, list, size); - fdput(f); - return error; + return listxattr(f.file->f_path.dentry, list, size); } /* @@ -951,12 +945,12 @@ SYSCALL_DEFINE2(lremovexattr, const char __user *, pathname, SYSCALL_DEFINE2(fremovexattr, int, fd, const char __user *, name) { - struct fd f = fdget(fd); + CLASS(fd, f)(fd); char kname[XATTR_NAME_MAX + 1]; - int error = -EBADF; + int error; if (!f.file) - return error; + return -EBADF; audit_file(f.file); error = strncpy_from_user(kname, name, sizeof(kname)); @@ -971,7 +965,6 @@ SYSCALL_DEFINE2(fremovexattr, int, fd, const char __user *, name) f.file->f_path.dentry, kname); mnt_drop_write_file(f.file); } - fdput(f); return error; } -- 2.39.2

2 1

[PATCH OLK-6.6] arm64: Set __nocfi on swsusp_arch_resume()
by Lin Yujun 16 Mar '26

16 Mar '26

From: Zhaoyang Huang <zhaoyang.huang(a)unisoc.com> stable inclusion from stable-v6.6.122 commit 8557bdd9af8dd04911fba56ff92b17842b0b5c7f category: bugfix bugzilla: https://atomgit.com/src-openeuler/kernel/issues/13677 CVE: CVE-2026-23128 Reference: https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git/commit/?id… -------------------------------- commit e2f8216ca2d8e61a23cb6ec355616339667e0ba6 upstream. A DABT is reported[1] on an android based system when resume from hiberate. This happens because swsusp_arch_suspend_exit() is marked with SYM_CODE_*() and does not have a CFI hash, but swsusp_arch_resume() will attempt to verify the CFI hash when calling a copy of swsusp_arch_suspend_exit(). Given that there's an existing requirement that the entrypoint to swsusp_arch_suspend_exit() is the first byte of the .hibernate_exit.text section, we cannot fix this by marking swsusp_arch_suspend_exit() with SYM_FUNC_*(). The simplest fix for now is to disable the CFI check in swsusp_arch_resume(). Mark swsusp_arch_resume() as __nocfi to disable the CFI check. [1] [ 22.991934][ T1] Unable to handle kernel paging request at virtual address 0000000109170ffc [ 22.991934][ T1] Mem abort info: [ 22.991934][ T1] ESR = 0x0000000096000007 [ 22.991934][ T1] EC = 0x25: DABT (current EL), IL = 32 bits [ 22.991934][ T1] SET = 0, FnV = 0 [ 22.991934][ T1] EA = 0, S1PTW = 0 [ 22.991934][ T1] FSC = 0x07: level 3 translation fault [ 22.991934][ T1] Data abort info: [ 22.991934][ T1] ISV = 0, ISS = 0x00000007, ISS2 = 0x00000000 [ 22.991934][ T1] CM = 0, WnR = 0, TnD = 0, TagAccess = 0 [ 22.991934][ T1] GCS = 0, Overlay = 0, DirtyBit = 0, Xs = 0 [ 22.991934][ T1] [0000000109170ffc] user address but active_mm is swapper [ 22.991934][ T1] Internal error: Oops: 0000000096000007 [#1] PREEMPT SMP [ 22.991934][ T1] Dumping ftrace buffer: [ 22.991934][ T1] (ftrace buffer empty) [ 22.991934][ T1] Modules linked in: [ 22.991934][ T1] CPU: 0 PID: 1 Comm: swapper/0 Not tainted 6.6.98-android15-8-g0b1d2aee7fc3-dirty-4k #1 688c7060a825a3ac418fe53881730b355915a419 [ 22.991934][ T1] Hardware name: Unisoc UMS9360-base Board (DT) [ 22.991934][ T1] pstate: 804000c5 (Nzcv daIF +PAN -UAO -TCO -DIT -SSBS BTYPE=--) [ 22.991934][ T1] pc : swsusp_arch_resume+0x2ac/0x344 [ 22.991934][ T1] lr : swsusp_arch_resume+0x294/0x344 [ 22.991934][ T1] sp : ffffffc08006b960 [ 22.991934][ T1] x29: ffffffc08006b9c0 x28: 0000000000000000 x27: 0000000000000000 [ 22.991934][ T1] x26: 0000000000000000 x25: 0000000000000000 x24: 0000000000000820 [ 22.991934][ T1] x23: ffffffd0817e3000 x22: ffffffd0817e3000 x21: 0000000000000000 [ 22.991934][ T1] x20: ffffff8089171000 x19: ffffffd08252c8c8 x18: ffffffc080061058 [ 22.991934][ T1] x17: 00000000529c6ef0 x16: 00000000529c6ef0 x15: 0000000000000004 [ 22.991934][ T1] x14: ffffff8178c88000 x13: 0000000000000006 x12: 0000000000000000 [ 22.991934][ T1] x11: 0000000000000015 x10: 0000000000000001 x9 : ffffffd082533000 [ 22.991934][ T1] x8 : 0000000109171000 x7 : 205b5d3433393139 x6 : 392e32322020205b [ 22.991934][ T1] x5 : 000000010916f000 x4 : 000000008164b000 x3 : ffffff808a4e0530 [ 22.991934][ T1] x2 : ffffffd08058e784 x1 : 0000000082326000 x0 : 000000010a283000 [ 22.991934][ T1] Call trace: [ 22.991934][ T1] swsusp_arch_resume+0x2ac/0x344 [ 22.991934][ T1] hibernation_restore+0x158/0x18c [ 22.991934][ T1] load_image_and_restore+0xb0/0xec [ 22.991934][ T1] software_resume+0xf4/0x19c [ 22.991934][ T1] software_resume_initcall+0x34/0x78 [ 22.991934][ T1] do_one_initcall+0xe8/0x370 [ 22.991934][ T1] do_initcall_level+0xc8/0x19c [ 22.991934][ T1] do_initcalls+0x70/0xc0 [ 22.991934][ T1] do_basic_setup+0x1c/0x28 [ 22.991934][ T1] kernel_init_freeable+0xe0/0x148 [ 22.991934][ T1] kernel_init+0x20/0x1a8 [ 22.991934][ T1] ret_from_fork+0x10/0x20 [ 22.991934][ T1] Code: a9400a61 f94013e0 f9438923 f9400a64 (b85fc110) Co-developed-by: Jeson Gao <jeson.gao(a)unisoc.com> Signed-off-by: Jeson Gao <jeson.gao(a)unisoc.com> Signed-off-by: Zhaoyang Huang <zhaoyang.huang(a)unisoc.com> Acked-by: Will Deacon <will(a)kernel.org> Acked-by: Mark Rutland <mark.rutland(a)arm.com> Cc: <stable(a)vger.kernel.org> [catalin.marinas(a)arm.com: commit log updated by Mark Rutland] Signed-off-by: Catalin Marinas <catalin.marinas(a)arm.com> Signed-off-by: Greg Kroah-Hartman <gregkh(a)linuxfoundation.org> Signed-off-by: Yuan can <yuancan(a)huawei.com> Signed-off-by: Lin Yujun <linyujun809(a)h-partners.com> --- arch/arm64/kernel/hibernate.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/arch/arm64/kernel/hibernate.c b/arch/arm64/kernel/hibernate.c index 02870beb271e..17c436387f1b 100644 --- a/arch/arm64/kernel/hibernate.c +++ b/arch/arm64/kernel/hibernate.c @@ -394,11 +394,11 @@ int swsusp_arch_suspend(void) * Setup then Resume from the hibernate image using swsusp_arch_suspend_exit(). * * Memory allocated by get_safe_page() will be dealt with by the hibernate code, * we don't need to free it here. */ -int swsusp_arch_resume(void) +int __nocfi swsusp_arch_resume(void) { int rc; void *zero_page; size_t exit_size; pgd_t *tmp_pg_dir; -- 2.34.1

2 1

[PATCH OLK-5.10 0/3] arm64/mpam: Extending MPAM features and bugfix
by Quanmin Yan 16 Mar '26

16 Mar '26

Extending the MPAM features and bugfix in OLK-5.10. Quanmin Yan (3): arm64/mpam: Add quirk to shrink MATA PMG range arm64/mpam: Add quirk for MPAM MSMON_MBWU monitor NRDY bit arm64/mpam: Allow MBMIN to be set to 0 arch/arm64/kernel/mpam/mpam_device.c | 33 +++++++++++++++++++++++---- arch/arm64/kernel/mpam/mpam_resctrl.c | 7 +++--- 2 files changed, 33 insertions(+), 7 deletions(-) -- 2.43.0

2 4

[PATCH OLK-6.6] arm64/xcall: add research papers reference in file headers
by Yipeng Zou 16 Mar '26

16 Mar '26

hulk inclusion category: feature bugzilla: https://atomgit.com/openeuler/kernel/issues/8729 -------------------------------------------------------------------- Add research paper references to xcall feature source files to document the academic foundation of the design. Reference papers: 1. XPC (ISCA 2019) - XPC: Architectural Support for Secure and Efficient Cross Process Call Authors: Dong Du, Zhichao Hua, Yubin Xia, Binyu Zang, Haibo Chen 2. Boosting IPC (TOCS 2022) - Boosting Inter-Process Communication with Architectural Support Authors: Yubin Xia, Dong Du, Zhichao Hua, Binyu Zang, Haibo Chen, Haibing Guan Signed-off-by: Yipeng Zou <zouyipeng(a)huawei.com> --- arch/arm64/include/asm/xcall.h | 9 +++++++++ arch/arm64/kernel/xcall/core.c | 7 +++++++ arch/arm64/kernel/xcall/entry.S | 7 ++++++- arch/arm64/kernel/xcall/proc.c | 7 +++++++ arch/arm64/kernel/xcall/xcall.c | 7 ++++++- include/linux/xcall.h | 7 +++++++ 6 files changed, 42 insertions(+), 2 deletions(-) diff --git a/arch/arm64/include/asm/xcall.h b/arch/arm64/include/asm/xcall.h index 0d0c4c794010..6ea03cc6cb20 100644 --- a/arch/arm64/include/asm/xcall.h +++ b/arch/arm64/include/asm/xcall.h @@ -1,4 +1,13 @@ /* SPDX-License-Identifier: GPL-2.0 */ +/* + * ARM64-specific xcall definitions and inline functions. + * + * Inspired by: + * 1. XPC (ISCA 2019) - Dong Du, Zhichao Hua, Yubin Xia, Binyu Zang, Haibo Chen; + * 2. Boosting IPC (TOCS 2022) - Yubin Xia, Dong Du, Zhichao Hua, Binyu Zang, + * Haibo Chen, Haibing Guan. + */ + #ifndef __ASM_XCALL_H #define __ASM_XCALL_H diff --git a/arch/arm64/kernel/xcall/core.c b/arch/arm64/kernel/xcall/core.c index 5dc6e4aebf11..5dc8e3b505b4 100644 --- a/arch/arm64/kernel/xcall/core.c +++ b/arch/arm64/kernel/xcall/core.c @@ -1,6 +1,13 @@ // SPDX-License-Identifier: GPL-2.0-only /* + * xcall core implementation: syscall interception and xcall_area management. + * * Copyright (C) 2025 Huawei Limited. + * + * Inspired by: + * 1. XPC (ISCA 2019) - Dong Du, Zhichao Hua, Yubin Xia, Binyu Zang, Haibo Chen; + * 2. Boosting IPC (TOCS 2022) - Yubin Xia, Dong Du, Zhichao Hua, Binyu Zang, + * Haibo Chen, Haibing Guan. */ #define pr_fmt(fmt) "xcall: " fmt diff --git a/arch/arm64/kernel/xcall/entry.S b/arch/arm64/kernel/xcall/entry.S index f531f1e17f48..d5ed68db1547 100644 --- a/arch/arm64/kernel/xcall/entry.S +++ b/arch/arm64/kernel/xcall/entry.S @@ -1,10 +1,15 @@ /* SPDX-License-Identifier: GPL-2.0-only */ /* - * Low-level exception handling code + * xcall low-level exception handling and register saving. * * Copyright (C) 2012 ARM Ltd. * Authors: Catalin Marinas <catalin.marinas(a)arm.com> * Will Deacon <will.deacon(a)arm.com> + * + * Inspired by: + * 1. XPC (ISCA 2019) - Dong Du, Zhichao Hua, Yubin Xia, Binyu Zang, Haibo Chen; + * 2. Boosting IPC (TOCS 2022) - Yubin Xia, Dong Du, Zhichao Hua, Binyu Zang, + * Haibo Chen, Haibing Guan. */ .macro hw_xcall_save_base_regs diff --git a/arch/arm64/kernel/xcall/proc.c b/arch/arm64/kernel/xcall/proc.c index 9dea182340ef..f2fcc9c8eaa2 100644 --- a/arch/arm64/kernel/xcall/proc.c +++ b/arch/arm64/kernel/xcall/proc.c @@ -1,6 +1,13 @@ // SPDX-License-Identifier: GPL-2.0-only /* + * xcall proc interface for attach/detach operations. + * * Copyright (C) 2025 Huawei Limited. + * + * Inspired by: + * 1. XPC (ISCA 2019) - Dong Du, Zhichao Hua, Yubin Xia, Binyu Zang, Haibo Chen; + * 2. Boosting IPC (TOCS 2022) - Yubin Xia, Dong Du, Zhichao Hua, Binyu Zang, + * Haibo Chen, Haibing Guan. */ #include <linux/namei.h> diff --git a/arch/arm64/kernel/xcall/xcall.c b/arch/arm64/kernel/xcall/xcall.c index f3fc89ef8e2f..a239f8b1b88a 100644 --- a/arch/arm64/kernel/xcall/xcall.c +++ b/arch/arm64/kernel/xcall/xcall.c @@ -1,8 +1,13 @@ // SPDX-License-Identifier: GPL-2.0-only /* - * xcall related code + * xcall task lifecycle management and context switch implementation. * * Copyright (C) 2025 Huawei Ltd. + * + * Inspired by: + * 1. XPC (ISCA 2019) - Dong Du, Zhichao Hua, Yubin Xia, Binyu Zang, Haibo Chen; + * 2. Boosting IPC (TOCS 2022) - Yubin Xia, Dong Du, Zhichao Hua, Binyu Zang, + * Haibo Chen, Haibing Guan. */ #include <linux/bitmap.h> diff --git a/include/linux/xcall.h b/include/linux/xcall.h index 26c60cd3dd38..222bd2cfb843 100644 --- a/include/linux/xcall.h +++ b/include/linux/xcall.h @@ -1,6 +1,13 @@ /* SPDX-License-Identifier: GPL-2.0-only */ /* + * xcall data structures and function declarations. + * * Copyright (C) 2025 Huawei. + * + * Inspired by: + * 1. XPC (ISCA 2019) - Dong Du, Zhichao Hua, Yubin Xia, Binyu Zang, Haibo Chen; + * 2. Boosting IPC (TOCS 2022) - Yubin Xia, Dong Du, Zhichao Hua, Binyu Zang, + * Haibo Chen, Haibing Guan. */ #ifndef _LINUX_XCALL_H -- 2.34.1

2 1

[PATCH openEuler-26.09] tools: Add maio-utils tool
by Hongbo Li 16 Mar '26

16 Mar '26

hulk inclusion category: feature bugzilla: https://atomgit.com/openeuler/kernel/issues/8613 -------------------------------- This is a userspace tool for MFS to provide: programmable strategy framework. Some basic strategies have added in test, and more flexible strategies can be written based on this framework. Signed-off-by: Hongbo Li <lihongbo22(a)huawei.com> --- tools/mfs/maio-utils/CMakeLists.txt | 101 ++++ tools/mfs/maio-utils/README.md | 17 + tools/mfs/maio-utils/build/build.sh | 91 +++ .../mfs/maio-utils/include/client/fs_client.h | 17 + tools/mfs/maio-utils/include/inner/inner.h | 37 ++ tools/mfs/maio-utils/include/loader/loader.h | 10 + tools/mfs/maio-utils/include/maio.h | 53 ++ .../maio-utils/include/parser/req_parser.h | 13 + tools/mfs/maio-utils/include/policy/policy.h | 17 + tools/mfs/maio-utils/include/uapi/mfs.h | 62 ++ tools/mfs/maio-utils/src/CMakeLists.txt | 42 ++ .../mfs/maio-utils/src/client/CMakeLists.txt | 17 + tools/mfs/maio-utils/src/client/fs_client.c | 359 ++++++++++++ .../mfs/maio-utils/src/loader/CMakeLists.txt | 18 + tools/mfs/maio-utils/src/loader/loader.c | 144 +++++ tools/mfs/maio-utils/src/main.c | 156 +++++ .../mfs/maio-utils/src/parser/CMakeLists.txt | 16 + tools/mfs/maio-utils/src/parser/req_parser.c | 313 ++++++++++ .../mfs/maio-utils/src/policy/CMakeLists.txt | 16 + tools/mfs/maio-utils/src/policy/policy.c | 114 ++++ tools/mfs/maio-utils/test/Makefile | 26 + tools/mfs/maio-utils/test/README.md | 61 ++ tools/mfs/maio-utils/test/strategy_demo.c | 45 ++ .../mfs/maio-utils/test/strategy_generate.cpp | 296 ++++++++++ .../mfs/maio-utils/test/strategy_prefetch.cpp | 150 +++++ tools/mfs/maio-utils/test/strategy_prefetch.h | 21 + tools/mfs/maio-utils/test/strategy_replay.cpp | 540 ++++++++++++++++++ tools/mfs/maio-utils/test/strategy_template.h | 21 + tools/mfs/maio-utils/test/strategy_toend.c | 46 ++ tools/mfs/maio-utils/test/trace_analyse.py | 87 +++ tools/mfs/maio-utils/test/trace_gen.py | 116 ++++ 31 files changed, 3022 insertions(+) create mode 100644 tools/mfs/maio-utils/CMakeLists.txt create mode 100644 tools/mfs/maio-utils/README.md create mode 100644 tools/mfs/maio-utils/build/build.sh create mode 100644 tools/mfs/maio-utils/include/client/fs_client.h create mode 100644 tools/mfs/maio-utils/include/inner/inner.h create mode 100644 tools/mfs/maio-utils/include/loader/loader.h create mode 100644 tools/mfs/maio-utils/include/maio.h create mode 100644 tools/mfs/maio-utils/include/parser/req_parser.h create mode 100644 tools/mfs/maio-utils/include/policy/policy.h create mode 100644 tools/mfs/maio-utils/include/uapi/mfs.h create mode 100644 tools/mfs/maio-utils/src/CMakeLists.txt create mode 100644 tools/mfs/maio-utils/src/client/CMakeLists.txt create mode 100644 tools/mfs/maio-utils/src/client/fs_client.c create mode 100644 tools/mfs/maio-utils/src/loader/CMakeLists.txt create mode 100644 tools/mfs/maio-utils/src/loader/loader.c create mode 100644 tools/mfs/maio-utils/src/main.c create mode 100644 tools/mfs/maio-utils/src/parser/CMakeLists.txt create mode 100644 tools/mfs/maio-utils/src/parser/req_parser.c create mode 100644 tools/mfs/maio-utils/src/policy/CMakeLists.txt create mode 100644 tools/mfs/maio-utils/src/policy/policy.c create mode 100644 tools/mfs/maio-utils/test/Makefile create mode 100644 tools/mfs/maio-utils/test/README.md create mode 100644 tools/mfs/maio-utils/test/strategy_demo.c create mode 100644 tools/mfs/maio-utils/test/strategy_generate.cpp create mode 100644 tools/mfs/maio-utils/test/strategy_prefetch.cpp create mode 100644 tools/mfs/maio-utils/test/strategy_prefetch.h create mode 100644 tools/mfs/maio-utils/test/strategy_replay.cpp create mode 100644 tools/mfs/maio-utils/test/strategy_template.h create mode 100644 tools/mfs/maio-utils/test/strategy_toend.c create mode 100644 tools/mfs/maio-utils/test/trace_analyse.py create mode 100644 tools/mfs/maio-utils/test/trace_gen.py diff --git a/tools/mfs/maio-utils/CMakeLists.txt b/tools/mfs/maio-utils/CMakeLists.txt new file mode 100644 index 000000000000..a13fe2bf1d7f --- /dev/null +++ b/tools/mfs/maio-utils/CMakeLists.txt @@ -0,0 +1,101 @@ +#******************************************************************************* +#******************************************************************************* + +cmake_minimum_required(VERSION 3.0) +cmake_policy(SET CMP0057 NEW) + +project(maio) + +#******************************************************************************* +#******************************************************************************* + +add_definitions(-D__LINUX_USR__) +add_definitions(-D__USE_GNU) +add_definitions(-D__POSIX_SOURCE) +add_definitions(-D_GNU_SOURCE) +add_definitions(-D_FILE_OFFSET_BITS=64) + +#******************************************************************************* +#******************************************************************************* + +include(TestBigEndian) + +TEST_BIG_ENDIAN(BIGENDIAN) + if (${BIGENDIAN}) + message(">>> Big Endian") + add_definitions(-D_CMAKE_BIGENDIAN) + else() + message(">>> Little Endian") + endif(${BIGENDIAN}) + +#******************************************************************************* +#******************************************************************************* + +EXECUTE_PROCESS(COMMAND uname -m COMMAND tr -d '\n' OUTPUT_VARIABLE ARCH) +if(${ARCH} STREQUAL "x86_64" OR ${ARCH} STREQUAL "aarch64") + message(STATUS "Architecture: ${ARCH}") +else() + message(STATUS "Invalid Architecture: ${ARCH}") + return() +endif() + +#******************************************************************************* +#******************************************************************************* + +set(BASE_FLAGS "-g -Wall -Werror -fno-strict-overflow -rdynamic") +set(SP_C_FLAGS "-std=gnu99") +set(SP_C++_FLAGS "-std=c++11") +set(SP_RELEASE_FLAGS "-O2") +set(SP_DEBUG_FLAGS "-O0") +set(SEC_REQ_FLAGS "-fstack-protector-all -fPIE -pie -fPIC -Wl,-z,relro,-z,now,-z,noexecstack") +set(SEC_SUG_BASE_FLAGS "-Wformat=2") +set(SEC_SUG_SP_C_FLAGS "-Wshadow") + +set(C_FLAGS "${BASE_FLAGS} ${SP_C_FLAGS}") +set(C++_FLAGS "${BASE_FLAGS} ${SP_C++_FLAGS}") +set(SEC_C_FLAGS "${SEC_REQ_FLAGS} ${SEC_SUG_BASE_FLAGS} ${SEC_SUG_SP_C_FLAGS}") +set(SEC_C++_FLAGS "${SEC_REQ_FLAGS} ${SEC_SUG_BASE_FLAGS}") +set(DEPC_FLAGS "-Wno-error=deprecated-declarations -Wno-deprecated-declarations") + +if(CMAKE_BUILD_TYPE MATCHES "Release") + message(">>> Release Mode") + set(CMAKE_C_FLAGS "${C_FLAGS} ${SP_RELEASE_FLAGS} ${SEC_C_FLAGS}") + set(CMAKE_CXX_FLAGS "${C++_FLAGS} ${SP_RELEASE_FLAGS} ${SEC_C++_FLAGS} ${DEPC_FLAGS}") +else() + message(">>> Debug Mode") + add_definitions(-D_DEBUG) + set(CMAKE_C_FLAGS "${C_FLAGS} ${SP_DEBUG_FLAGS} ${SEC_C_FLAGS}") + set(CMAKE_CXX_FLAGS "${C++_FLAGS} ${SP_DEBUG_FLAGS} ${SEC_C++_FLAGS} ${DEPC_FLAGS}") + set(COVERAGE_FLAGS "-coverage") + set(COVERAGE_LINK "gcov") +endif() + +message(STATUS ">>> Source Dir: " ${PROJECT_SOURCE_DIR}) +message(STATUS ">>> Binary Dir: " ${PROJECT_BINARY_DIR}) +message(STATUS ">>> C_FLAGS : " ${CMAKE_C_FLAGS}) +message(STATUS ">>> CXX_FLAGS : " ${CMAKE_CXX_FLAGS}) + +#******************************************************************************* +#******************************************************************************* + +set(EXECUTABLE_OUTPUT_PATH ${PROJECT_BINARY_DIR}/bin) + +#******************************************************************************* +#******************************************************************************* + +if(DEFINED THIRD_PARTY) + message(">>> third_party: ${THIRD_PARTY}") +else() + set(THIRD_PARTY ${PROJECT_SOURCE_DIR}/lib) +endif() + +set(INFRA_DIR ${PROJECT_SOURCE_DIR}/infrastructure) +set(SECUREC_DIR ${THIRD_PARTY}/securec) +set(ZLOG_DIR ${THIRD_PARTY}/zlog) +set(LIBNUMA_DIR ${THIRD_PARTY}/libnuma) + +#******************************************************************************* +#******************************************************************************* + +add_subdirectory(infrastructure) +add_subdirectory(src) diff --git a/tools/mfs/maio-utils/README.md b/tools/mfs/maio-utils/README.md new file mode 100644 index 000000000000..40477bebf130 --- /dev/null +++ b/tools/mfs/maio-utils/README.md @@ -0,0 +1,17 @@ +# maio-utils + +#### How to build + +1. prepare thirdparty by using: +``` +git clone -b threadpool_bind https://github.com/hb-lee/infrastructure.git infrastructure +git clone -b 1.2.18 https://github.com/HardySimpson/zlog.git thirdparty/zlog +git clone https://github.com/chriszt/securec.git thirdparty/securec +git clone -b v2.0.19 https://github.com/numactl/numactl.git thirdparty/libnuma +``` +2. cd build +3. bash build.sh prepare # only once. +4. bash build.sh build # [debug|release] + +#### How to run +Run with `./maio -m ${MNTPOINT}`, here ${MNTPOINT} means the mfs mount point. If you want to use some strategies, you can run with `./maio -m ${MNTPOINT} -s ${STRATEGYLIB}` where ${STRATEGYLIB} is your strategy library. Some examples are provided in `test` directory. diff --git a/tools/mfs/maio-utils/build/build.sh b/tools/mfs/maio-utils/build/build.sh new file mode 100644 index 000000000000..91b49d3d6bf8 --- /dev/null +++ b/tools/mfs/maio-utils/build/build.sh @@ -0,0 +1,91 @@ +#!/bin/bash + +LOCAL_PATH="$(readlink -e "$(dirname "$0")")" +PRJ_PATH="$(readlink -e "${LOCAL_PATH}/..")" +THIRD_SUB="${PRJ_PATH}/thirdparty" +Lib_PATH="${PRJ_PATH}/lib" + +ARCH=$(uname -m) + +function build_zlog() { + local zlog_dir=${THIRD_SUB}/zlog + local build_path=${zlog_dir}/build + mkdir ${build_path} + mkdir -p ${Lib_PATH}/zlog + cd ${zlog_dir} + make PREFIX=`pwd`/build/zlog + make PREFIX=`pwd`/build/zlog install + cp -rdp ${build_path}/zlog/include ${Lib_PATH}/zlog + cp -rdp ${build_path}/zlog/lib ${Lib_PATH}/zlog + cd - +} + +function build_securec() { + local securec_dir=${THIRD_SUB}/securec + local securec_lib=${securec_dir}/build/securec/lib + mkdir -p ${securec_lib} + mkdir -p ${Lib_PATH}/securec + cd ${securec_dir} + if [ "${ARCH}" = "aarch64" ]; then + make -f aarch64-so.mk + else + make -f x86-so.mk + fi + cp -rdp include build/securec + cp libsecurec.so build/securec/lib + cp -rdp build/securec/include ${Lib_PATH}/securec + cp -rdp build/securec/lib ${Lib_PATH}/securec + cd - +} + +function build_libnuma() { + local libnuma_dir=${THIRD_SUB}/libnuma + local build_path=${libnuma_dir}/build + mkdir ${build_path} + mkdir -p ${Lib_PATH}/libnuma + cd ${libnuma_dir} + ./autogen.sh + ./configure --prefix=`pwd`/build + make -j4 + make install + cp -rdp build/include ${Lib_PATH}/libnuma + cp -rdp build/lib ${Lib_PATH}/libnuma + cd - +} + +function prepare_thirdparty() { + build_jsoncpp + build_zlog + build_securec + build_libnuma +} + +function build_program() { + local BUILD_TYPE=$1 + + set -x + if [ "${BUILD_TYPE}" == "release" ]; then + cmake .. -DCMAKE_BUILD_TYPE=Release + else + cmake .. -DCMAKE_BUILD_TYPE=Debug + fi + set +x + + make clean + make VERBOSE=1 -j +} + +function main() { + local ACTION="$1" + local PARAMETER="$2" + + if [ "${ACTION}" == "prepare" ]; then + prepare_thirdparty + fi + if [ "$ACTION" == "build" ]; then + build_program $PARAMETER + fi +} + +main $@ +exit $? diff --git a/tools/mfs/maio-utils/include/client/fs_client.h b/tools/mfs/maio-utils/include/client/fs_client.h new file mode 100644 index 000000000000..df3a086e317a --- /dev/null +++ b/tools/mfs/maio-utils/include/client/fs_client.h @@ -0,0 +1,17 @@ +// SPDX-License-Identifier: GPL-2.0 +/* Copyright(c) 2025 Huawei Technologies Co., Ltd + */ +#ifndef _FS_CLIENT_H_ +#define _FS_CLIENT_H_ + +#include <stdint.h> + +int fs_evict_by_path(const char *path, uint64_t off, uint64_t len); +int fs_load_by_path(const char *path, uint64_t off, uint64_t len); +int fs_load_by_fd(int fd, uint64_t off, uint64_t len); +int fs_sync_by_path(const char *path, void *buf, uint64_t off, uint64_t len); + +int fs_client_init(const char *source); +void fs_client_exit(void); + +#endif diff --git a/tools/mfs/maio-utils/include/inner/inner.h b/tools/mfs/maio-utils/include/inner/inner.h new file mode 100644 index 000000000000..fbb87e13b1bb --- /dev/null +++ b/tools/mfs/maio-utils/include/inner/inner.h @@ -0,0 +1,37 @@ +// SPDX-License-Identifier: GPL-2.0 +/* Copyright(c) 2025 Huawei Technologies Co., Ltd + */ +#ifndef _INNER_H_ +#define _INNER_H_ + +#include <time.h> +#include <stdint.h> + +static inline uint64_t _get_ts(void) +{ + struct timespec now = {0}; + (void)clock_gettime(CLOCK_MONOTONIC_COARSE, &now); + return (uint64_t)now.tv_sec; +} + +/* io context for loader and writeback */ +struct io_context { + uint32_t id; /* msg id */ + uint64_t off; /* io offset in fd */ + uint64_t len; /* io length */ + uint16_t numaid; /* numa node of this io, -1 means non-bind */ + int fd; /* io file handle */ + char *path; /* fullpath of this io */ + void *buf; /* flighting buffer for writeback */ + + uint64_t seq; + + void *private; + void (*end_io)(int ret, void *private); +}; + +void loader_update_curseq(uint64_t seq); +int loader_evict_submit(struct io_context *ioctx); +int loader_io_submit(struct io_context *ioctx); + +#endif diff --git a/tools/mfs/maio-utils/include/loader/loader.h b/tools/mfs/maio-utils/include/loader/loader.h new file mode 100644 index 000000000000..7bf0f163b292 --- /dev/null +++ b/tools/mfs/maio-utils/include/loader/loader.h @@ -0,0 +1,10 @@ +// SPDX-License-Identifier: GPL-2.0 +/* Copyright(c) 2025 Huawei Technologies Co., Ltd + */ +#ifndef _LOADER_H_ +#define _LOADER_H_ + +int loader_init(void); +void loader_exit(void); + +#endif diff --git a/tools/mfs/maio-utils/include/maio.h b/tools/mfs/maio-utils/include/maio.h new file mode 100644 index 000000000000..a508d5b1fa27 --- /dev/null +++ b/tools/mfs/maio-utils/include/maio.h @@ -0,0 +1,53 @@ +// SPDX-License-Identifier: GPL-2.0 +/* Copyright(c) 2025 Huawei Technologies Co., Ltd + */ +#ifndef _MAIO_H_ +#define _MAIO_H_ + +#ifdef __cplusplus +extern "C" { +#endif + +#include <stdint.h> + +struct maio_entry { + char *fpath; + uint64_t toff; + uint64_t tlen; + uint8_t tnuma; + uint64_t seq; +}; + +struct maio { + /* IN */ + int fd; + uint64_t off; + uint64_t len; + int pid; + uint8_t op; + uint64_t ts; + uint32_t cksz; + + /* OUT */ + uint16_t flags; +#define MAIO_WITH_SEQ 0x0001 + uint64_t curseq; + struct maio_entry entries[]; +}; + +struct maio_operation { + int max_io; + int (*init) (void); + void (*exit) (void); + int (*load) (struct maio **io); + int (*evict) (struct maio **io); +}; + +char *_get_fullpath(char *fpath, int fd); +void maio_preload(struct maio *maio, int nio); + +#ifdef __cplusplus +} +#endif + +#endif diff --git a/tools/mfs/maio-utils/include/parser/req_parser.h b/tools/mfs/maio-utils/include/parser/req_parser.h new file mode 100644 index 000000000000..bcb23474b58b --- /dev/null +++ b/tools/mfs/maio-utils/include/parser/req_parser.h @@ -0,0 +1,13 @@ +// SPDX-License-Identifier: GPL-2.0 +/* Copyright(c) 2025 Huawei Technologies Co., Ltd + */ +#ifndef __REQ_PARSER_H__ +#define __REQ_PARSER_H__ + +#include <stdint.h> + +int parser_init(uint32_t fs_mode, const char *mntpoint); +void parser_destory(void); +int maio_parse_req(void *buf, int size); + +#endif diff --git a/tools/mfs/maio-utils/include/policy/policy.h b/tools/mfs/maio-utils/include/policy/policy.h new file mode 100644 index 000000000000..954d25edc894 --- /dev/null +++ b/tools/mfs/maio-utils/include/policy/policy.h @@ -0,0 +1,17 @@ +// SPDX-License-Identifier: GPL-2.0 +/* Copyright(c) 2025 Huawei Technologies Co., Ltd + */ +#ifndef _POLICY_H_ +#define _POLICY_H_ + +#include "maio.h" + +int policy_max_io(void); +int policy_load(struct maio **io); +int policy_evict(struct maio **io); +int policy_register(const char *path); +void policy_unregister(void); +int policy_init(void); +void policy_exit(void); + +#endif diff --git a/tools/mfs/maio-utils/include/uapi/mfs.h b/tools/mfs/maio-utils/include/uapi/mfs.h new file mode 100644 index 000000000000..a7d5882b5e50 --- /dev/null +++ b/tools/mfs/maio-utils/include/uapi/mfs.h @@ -0,0 +1,62 @@ +/* SPDX-License-Identifier: GPL-2.0 WITH Linux-syscall-note */ + +#ifndef _UAPI_LINUX_MFS_H +#define _UAPI_LINUX_MFS_H + +#include <linux/types.h> +#include <linux/ioctl.h> + +enum mfs_opcode { + MFS_OP_READ = 0, + MFS_OP_FAULT, + MFS_OP_FAROUND, +}; + +enum { + MFS_MODE_NONE = 0, + MFS_MODE_LOCAL, + MFS_MODE_REMOTE, +}; + +struct mfs_ioc_ra { + __u64 off; + __u64 len; +}; + +struct mfs_ioc_done { + __u32 id; + __u32 ret; +}; + +struct mfs_ioc_rpath { + __u16 max; + __u16 len; + __u8 d[]; +}; + +#define MFS_IOC_RA _IOW(0xbc, 1, struct mfs_ioc_ra) +#define MFS_IOC_DONE _IOW(0xbc, 2, struct mfs_ioc_done) +#define MFS_IOC_RPATH _IOWR(0xbc, 3, struct mfs_ioc_rpath) + +struct mfs_ioc_fsinfo { + __u8 mode; /* 0: none, 1: local, 2: remote */ +}; + +#define MFS_IOC_FSINFO _IOR(0xbd, 1, struct mfs_ioc_fsinfo) + +struct mfs_msg { + __u8 version; + __u8 opcode; + __u16 len; + __u32 fd; + __u32 id; + __u8 data[]; +}; + +struct mfs_read { + __u64 off; + __u64 len; + __s32 pid; +}; + +#endif /* _UAPI_LINUX_MFS_H */ diff --git a/tools/mfs/maio-utils/src/CMakeLists.txt b/tools/mfs/maio-utils/src/CMakeLists.txt new file mode 100644 index 000000000000..fd053a8bd081 --- /dev/null +++ b/tools/mfs/maio-utils/src/CMakeLists.txt @@ -0,0 +1,42 @@ +#******************************************************************************* +#******************************************************************************* + +include_directories(${PROJECT_SOURCE_DIR}/include) +include_directories(${PROJECT_SOURCE_DIR}/include/inner) +include_directories(${PROJECT_SOURCE_DIR}/include/uapi) +include_directories(${PROJECT_SOURCE_DIR}/include/client) +include_directories(${PROJECT_SOURCE_DIR}/include/loader) +include_directories(${PROJECT_SOURCE_DIR}/include/policy) +include_directories(${PROJECT_SOURCE_DIR}/include/monitor) +include_directories(${PROJECT_SOURCE_DIR}/include/parser) + +include_directories(${INFRA_DIR}/include) +include_directories(${SECUREC_DIR}/include) +include_directories(${ZLOG_DIR}/include) +include_directories(${LIBNUMA_DIR}/include) + +#******************************************************************************* +#******************************************************************************* + +link_directories(${PROJECT_BINARY_DIR}/lib) +link_directories(${SECUREC_DIR}/lib) +link_directories(${ZLOG_DIR}/lib) +link_directories(${LIBNUMA_DIR}/lib) + +#******************************************************************************* +#******************************************************************************* + +add_subdirectory(client) +add_subdirectory(loader) +add_subdirectory(policy) +add_subdirectory(parser) + +add_executable(maio-utils main.c) +target_link_libraries(maio-utils "-Wl,--whole-archive" + client + loader + policy + parser + "-Wl,--no-whole-archive" + infra + dl) diff --git a/tools/mfs/maio-utils/src/client/CMakeLists.txt b/tools/mfs/maio-utils/src/client/CMakeLists.txt new file mode 100644 index 000000000000..ca1c91fd3057 --- /dev/null +++ b/tools/mfs/maio-utils/src/client/CMakeLists.txt @@ -0,0 +1,17 @@ +#******************************************************************************* +#******************************************************************************* + +include_directories(${CMAKE_CURRENT_SOURCE_DIR}) + +#******************************************************************************* +#******************************************************************************* + +#aux_source_directory(. CLIENT_SRC) +set(CLIENT_SRC fs_client.c) + +#******************************************************************************* +#******************************************************************************* + +add_library(client ${CLIENT_SRC}) +target_link_libraries(client infra + securec) diff --git a/tools/mfs/maio-utils/src/client/fs_client.c b/tools/mfs/maio-utils/src/client/fs_client.c new file mode 100644 index 000000000000..94ef1b70be97 --- /dev/null +++ b/tools/mfs/maio-utils/src/client/fs_client.c @@ -0,0 +1,359 @@ +// SPDX-License-Identifier: GPL-2.0 +/* Copyright(c) 2025 Huawei Technologies Co., Ltd + */ +#include "fs_client.h" +#include "mfs.h" +#include "inner.h" +#include "hashmap.h" +#include "hashfunc.h" +#include "stimer.h" +#include "atomic.h" +#include "log.h" +#include "list.h" +#include "spinlock.h" +#include "sysdef.h" +#include "securec.h" + +#include <fcntl.h> +#include <errno.h> +#include <string.h> +#include <unistd.h> +#include <sys/ioctl.h> + +#define TIMER_CYCLE 10000 /* 10s */ +#define FD_TMOUT 3600 /* 60s */ + +struct fs_client { + int rootfd; + hashmap_t *map; + stimer_t *timer; + list_head_t head; + spinlock_t lock; + uint64_t tmout; +}; + +struct fs_client g_client = {0}; + +typedef struct { + uint32_t plen; + char *path; +} fdkey_t; + +typedef struct { + fdkey_t key; + int fd; + int ref; + uint64_t ts; + list_head_t link; + hashlink_t hash; +} fd_t; + +static inline int fd_cmp(void *first, void *second) +{ + fdkey_t *key1 = (fdkey_t *)first; + fdkey_t *key2 = (fdkey_t *)second; + + if (key1->plen != key2->plen) + return -1; + + return strcmp(key1->path, key2->path); +} + +static inline uint32_t fd_hash(void *args) +{ + fdkey_t *key = (fdkey_t *)args; + uint64_t hash = hashstr(key->path, key->plen); + + return (uint32_t)hash; +} + +static inline void fd_inc(void *args, hashlink_t *link) +{ + fd_t *f = container_of(link, fd_t, hash); + int ref = atomic_s32_inc(&f->ref); + if (ref <= 0) { + log_error("fd(%d) for path(%s) with invalid ref(%d)", + f->fd, f->key.path, ref); + sys_assert(0); + } +} + +static inline int fd_dec(void *args, hashlink_t *link) +{ + fd_t *f = container_of(link, fd_t, hash); + int ref = atomic_s32_dec(&f->ref); + + if (ref < 0) { + log_error("fd(%d) for path(%s) with invalid ref(%d)", + f->fd, f->key.path, ref); + sys_assert(0); + } + /* update access timestamp for the last one */ + if (ref == 0) + f->ts = _get_ts(); + return -1; +} + +static inline int fd_expire(void *args, hashlink_t *link) +{ + fd_t *f = container_of(link, fd_t, hash); + int ref = atomic_s32_fetch(&f->ref); + uint64_t cur; + + if (ref > 0) + return -1; + if (ref < 0) { + log_error("fd(%d) for path(%s) with invalid ref(%d)", + f->fd, f->key.path, ref); + sys_assert(0); + } + cur = _get_ts(); + return ((cur - f->ts) >= g_client.tmout) ? 0 : -1; +} + +static fd_t *fd_alloc(const char *path) +{ + fd_t *f = (fd_t *)malloc(sizeof(fd_t)); + if (!f) { + log_error("alloc fd_t fail"); + return NULL; + } + + f->key.path = strdup(path); + if (!f->key.path) { + log_error("strdup(%s) fail", path); + free(f); + return NULL; + } + + f->key.plen = strlen(path) + 1; + f->fd = -1; + f->ref = 1; + f->hash.key = &f->key; + list_init(&f->link); + return f; +} + +static void fd_free(fd_t *f) +{ + if (f->fd > 0) { + int ret = close(f->fd); + if (ret != 0) { + log_error("close(%d) fail, err(%s)", f->fd, strerror(errno)); + } + f->fd = -1; + } + + if (f->key.path) { + free(f->key.path); + f->key.path = NULL; + } + list_del(&f->link); + free(f); +} + +static int fd_fetch(const char *path, fd_t **f_res, int dirfd) +{ + fdkey_t key; + key.plen = strlen(path) + 1; + key.path = (char *)path; + + /* fd in cache */ + hashlink_t *data = NULL; + if (EEXIST == hashmap_search(g_client.map, &key, &data, NULL, fd_inc)) { + *f_res = container_of(data, fd_t, hash); + (*f_res)->ts = _get_ts(); + return 0; + } + + /* new fd */ + fd_t *f = fd_alloc(path); + if (!f) + return -ENOMEM; + + /* open file */ + if (dirfd > 0) + f->fd = openat(dirfd, path, O_RDONLY, S_IRUSR | S_IRGRP); + else + f->fd = open(path, O_RDONLY, S_IRUSR | S_IRGRP); + if (f->fd < 0) { + if (errno != ENOENT) + log_error("open(%d,%s) fail, err(%s)", dirfd, path, strerror(errno)); + fd_free(f); + return errno; + } + + /* add to fd cache */ + hashlink_t *old = NULL; + if (EEXIST == hashmap_insert(g_client.map, &f->hash, &old, NULL, fd_inc)) { + fd_free(f); + f = container_of(old, fd_t, hash); + } + + spinlock_lock(&g_client.lock); + list_add_tail(&f->link, &g_client.head); + spinlock_unlock(&g_client.lock); + f->ts = _get_ts(); + *f_res = f; + return 0; +} + +static inline void fd_restore(fd_t *f) +{ + (void)hashmap_protect(g_client.map, &f->key, NULL, fd_dec); +} + +static void _fd_gc(void *args) +{ + list_head_t head, *curr, *next; + hashlink_t *data = NULL; + + list_init(&head); + spinlock_lock(&g_client.lock); + list_splice(&g_client.head, &head); + spinlock_unlock(&g_client.lock); + + list_foreach_safe(curr, next, &head) + { + fd_t *f = container_of(curr, fd_t, link); + if (atomic_s32_fetch(&f->ref) != 0) + continue; + /* try to destroy */ + if (0 != hashmap_delete(g_client.map, &f->key, &data, NULL, fd_expire)) + continue; + list_del(&f->link); + log_info("path:%s with fd:%d is expire.", f->key.path, f->fd); + fd_free(f); + } + + spinlock_lock(&g_client.lock); + list_splice(&head, &g_client.head); + spinlock_unlock(&g_client.lock); +} + +int fs_evict_by_path(const char *path, uint64_t off, uint64_t len) +{ + int ret; + fd_t *f; + uint64_t step = 4096 * 4; + + ret = fd_fetch(path, &f, 0); + if (ret) + return ret; + + while (len) { + if (step > len) + step = len; + ret = posix_fadvise(f->fd, off, len, POSIX_FADV_DONTNEED); + if (ret < 0) + break; + len -= step; + off += step; + } + + fd_restore(f); + return ret; +} + +int fs_load_by_path(const char *path, uint64_t off, uint64_t len) +{ + int ret; + fd_t *f; + uint64_t step = 16 * 4096; + char *buf = NULL; + + ret = fd_fetch(path, &f, 0); + if (ret) + return ret; + + buf = (char *)malloc(step); + + while (len) { + if (step > len) + step = len; + ret = pread(f->fd, buf, step, off); + if (ret <= 0) + break; + len -= step; + off += step; + } + free(buf); + fd_restore(f); + return ret; +} + +int fs_load_by_fd(int fd, uint64_t off, uint64_t len) +{ + struct mfs_ioc_ra ra; + + ra.off = off; + ra.len = len; + return ioctl(fd, MFS_IOC_RA, &ra); +} + +int fs_sync_by_path(const char *path, void *buf, uint64_t off, uint64_t len) +{ + ssize_t rsize = 0, ret; + uint64_t pos = 0; + fd_t *f; + + ret = fd_fetch(path + 1, &f, g_client.rootfd); + if (ret) + return ret; + + do { + ret = pread(f->fd, buf + pos, len - pos, off + pos); + if (ret == 0) + break; + if (ret < 0) { + log_error("failed to read:%s", strerror(errno)); + fd_restore(f); + return -1; + } + pos += ret; + rsize += ret; + } while (rsize < len); + fd_restore(f); + return 0; +} + +int fs_client_init(const char *source) +{ + int ret; + + g_client.rootfd = -1; + if (source) { + g_client.rootfd = open(source, O_RDONLY | O_DIRECTORY); + if (g_client.rootfd <= 0) { + log_error("open %s failed", source); + return -1; + } + } + ret = hashmap_create(1024, fd_cmp, fd_hash, &g_client.map); + if (ret) { + log_error("client fd map alloc failed"); + close(g_client.rootfd); + return ret; + } + + g_client.timer = stimer_create("LfsGc", TIMER_CYCLE, NULL, _fd_gc); + if (!g_client.timer) { + log_error("timer create failed"); + hashmap_destroy(g_client.map, NULL, NULL); + close(g_client.rootfd); + return -1; + } + list_init(&g_client.head); + spinlock_init(&g_client.lock); + g_client.tmout = FD_TMOUT; + return 0; +} + +void fs_client_exit(void) +{ + spinlock_destroy(&g_client.lock); + stimer_destroy(g_client.timer); + hashmap_destroy(g_client.map, NULL, NULL); + close(g_client.rootfd); +} + diff --git a/tools/mfs/maio-utils/src/loader/CMakeLists.txt b/tools/mfs/maio-utils/src/loader/CMakeLists.txt new file mode 100644 index 000000000000..08e1b637be26 --- /dev/null +++ b/tools/mfs/maio-utils/src/loader/CMakeLists.txt @@ -0,0 +1,18 @@ +#******************************************************************************* +#******************************************************************************* + +include_directories(${CMAKE_CURRENT_SOURCE_DIR}) + +#******************************************************************************* +#******************************************************************************* + +aux_source_directory(. LOADER_SRC) + +#******************************************************************************* +#******************************************************************************* + +add_library(loader ${LOADER_SRC}) +target_link_libraries(loader infra + numa + securec) + diff --git a/tools/mfs/maio-utils/src/loader/loader.c b/tools/mfs/maio-utils/src/loader/loader.c new file mode 100644 index 000000000000..f3460735c94e --- /dev/null +++ b/tools/mfs/maio-utils/src/loader/loader.c @@ -0,0 +1,144 @@ +// SPDX-License-Identifier: GPL-2.0 +/* Copyright(c) 2025 Huawei Technologies Co., Ltd + */ +#include "loader.h" + +#include "fs_client.h" +#include "threadpool.h" +#include "log.h" +#include "atomic.h" +#include "inner.h" +#include "sysdef.h" + +#include "securec.h" +#include "numa.h" +#include <errno.h> + +struct loader_mgr { + int numa_num; + threadpool_t **numa_thp; + uint64_t curseq; +}; + +struct loader_mgr g_local_mgr; + +static threadpool_t *choose_worker_pool(struct io_context *ioctx) +{ + static uint64_t cid; + int nid; + + if (ioctx->numaid == (uint16_t)-1) + nid = cid++ % g_local_mgr.numa_num; + else + nid = ioctx->numaid % g_local_mgr.numa_num; + return g_local_mgr.numa_thp[nid]; +} + +static void handle_io(void *args) +{ + struct io_context *ioctx = (struct io_context *)args; + int ret; + + if (g_local_mgr.curseq > ioctx->seq) { + if (ioctx->end_io) + ioctx->end_io(0, ioctx->private); + return; + } + ret = ioctx->path ? fs_load_by_path(ioctx->path, ioctx->off, ioctx->len) + : fs_load_by_fd(ioctx->fd, ioctx->off, ioctx->len); + if (ioctx->end_io) + ioctx->end_io(ret, ioctx->private); +} + +int loader_io_submit(struct io_context *ioctx) +{ + threadpool_t *worker_pools = choose_worker_pool(ioctx); + + threadpool_submit(worker_pools, ioctx, handle_io); + return 0; +} + +void loader_update_curseq(uint64_t seq) +{ + uint64_t curseq = atomic_u64_fetch(&g_local_mgr.curseq); + uint64_t old; + + while (seq > curseq) { + if (atomic_u64_cas(&g_local_mgr.curseq, curseq, seq, &old)) + break; + curseq = old; + } +} + +static void handle_evict(void *args) +{ + struct io_context *ioctx = (struct io_context *)args; + int ret; + + ret = ioctx->path ? fs_evict_by_path(ioctx->path, ioctx->off, ioctx->len) : 0; + if (ioctx->end_io) + ioctx->end_io(ret, ioctx->private); +} + +int loader_evict_submit(struct io_context *ioctx) +{ + threadpool_t *worker_pools = choose_worker_pool(ioctx); + + threadpool_submit(worker_pools, ioctx, handle_evict); + return 0; +} + +static void _free_loader(void) +{ + int i; + + if (!g_local_mgr.numa_thp) + return; + + for (i = 0; i < g_local_mgr.numa_num; i++) + threadpool_destroy(g_local_mgr.numa_thp[i]); + free(g_local_mgr.numa_thp); + g_local_mgr.numa_thp = NULL; +} + +int loader_init(void) +{ + int i, ret, wnum = 8; + char name[THD_NAME + 1] = {0}; + + if (numa_available() < 0) { + log_error("Current system does not support NUAM api"); + return -1; + } + + g_local_mgr.numa_num = numa_max_node() + 1; + g_local_mgr.curseq = 0; + ret = 0; + do { + g_local_mgr.numa_thp = + calloc(g_local_mgr.numa_num, sizeof(threadpool_t *)); + if (!g_local_mgr.numa_thp) { + log_error("alloc for numa threadpool failed"); + ret = -1; + break; + } + for (i = 0; i < g_local_mgr.numa_num; i++) { + sprintf_s(name, sizeof(name), "Loader-%d", i); + g_local_mgr.numa_thp[i] = threadpool_create(name, wnum, i); + if (!g_local_mgr.numa_thp[i]) { + log_error("alloc Loader-%d failed", i); + ret = -1; + break; + } + } + } while (0); + + if (ret != 0) + _free_loader(); + return ret; +} + +void loader_exit(void) +{ + _free_loader(); +} diff --git a/tools/mfs/maio-utils/src/main.c b/tools/mfs/maio-utils/src/main.c new file mode 100644 index 000000000000..ac5473d00011 --- /dev/null +++ b/tools/mfs/maio-utils/src/main.c @@ -0,0 +1,156 @@ +// SPDX-License-Identifier: GPL-2.0 +/* Copyright(c) 2025 Huawei Technologies Co., Ltd + */ +#include <stdlib.h> +#include <unistd.h> +#include <string.h> +#include <getopt.h> +#include <errno.h> +#include <fcntl.h> +#include <poll.h> + +#include "loader.h" +#include "log.h" +#include "maio.h" +#include "policy.h" +#include "req_parser.h" +#include "fs_client.h" +#include "mfs.h" +#include "securec.h" + +#include <sys/ioctl.h> +#include <sys/statfs.h> + +#define MAX_BUF_SIZE 1024 +#define MFS_SUPER_MAGIC 0x85428370 + +static int read_req(int fd) +{ + char *buf; + int ret; + + buf = calloc(MAX_BUF_SIZE, sizeof(char)); + if (!buf) { + log_error("failed to alloc buf\n"); + return -1; + } + ret = read(fd, buf, MAX_BUF_SIZE); + if (ret <= 0) { + if (ret < 0) + log_error("failed to read, ret:%d\n", ret); + return -1; + } + return maio_parse_req(buf, ret); +} + +static void dev_monitor(int dfd) +{ + struct pollfd pfd; + int ret; + + pfd.fd = dfd; + pfd.events = POLLIN; + while (1) { + ret = poll(&pfd, 1, -1); + if (ret < 0) { + log_error("poll failed\n"); + return; + } + if (ret == 0 || !(pfd.revents & POLLIN)) { + log_error("poll events error, ret:%d, revents:%x\n", ret, pfd.revents); + continue; + } + while (!read_req(pfd.fd)) {} + } +} + +int main(int argc, char *argv[]) +{ + const char *mntpoint = NULL, *strategylib = NULL; + struct mfs_ioc_fsinfo fsinfo = {0}; + int dfd, ret, opt; + struct statfs buf; + char devname[10]; + + while ((opt = getopt(argc, argv, "m:s:")) != -1) { + switch (opt) { + case 'm': + mntpoint = optarg; + break; + case 's': + strategylib = optarg; + break; + default: + fprintf(stderr, "Usage: %s -m ${mnt} [-s ${strategylib}]\n", argv[0]); + return -EINVAL; + } + } + if (!mntpoint) { + fprintf(stderr, "mount point should specify\n"); + fprintf(stderr, "Usage: %s -m ${mnt} [-s ${strategylib}]\n", argv[0]); + return -1; + } + + ret = statfs(mntpoint, &buf); + if (ret) { + log_error("statfs %s failed, errstr:%s\n", mntpoint, strerror(errno)); + return -1; + } + if (buf.f_type != MFS_SUPER_MAGIC) { + log_error("fstype(%x) is invalid, please check the mountpoint", buf.f_type); + return -1; + } + + sprintf_s(devname, sizeof(devname), "/dev/mfs%ld", buf.f_spare[0]); + /* Open with O_CLOEXEC to avoid potential fd leaks */ + dfd = open(devname, O_RDWR | O_CLOEXEC); + if (dfd < 0) { + log_error("open %s failed errstr:%s\n", devname, strerror(errno)); + return -1; + } + + ret = ioctl(dfd, MFS_IOC_FSINFO, (unsigned long)&fsinfo); + if (ret < 0) { + log_error("ioctl get fsinfo failed, ret:%d\n", ret); + goto close_fd; + } + if (fsinfo.mode != 1) { + log_error("fs runing mode(%d) is not supported", fsinfo.mode); + goto close_fd; + } + ret = loader_init(); + if (ret) { + log_error("loader init failed"); + goto close_fd; + } + ret = fs_client_init(NULL); + if (ret) { + log_error("clients init failed"); + goto free_loader; + } + ret = parser_init(fsinfo.mode, mntpoint); + if (ret) { + log_error("parser init failed"); + goto free_client; + } + ret = policy_init(); + if (ret) { + log_error("policy init failed"); + goto free_parser; + } + ret = policy_register(strategylib); + if (ret) { + log_info("register new policy failed, use default policy"); + } + + dev_monitor(dfd); +free_parser: + parser_destory(); +free_client: + fs_client_exit(); +free_loader: + loader_exit(); +close_fd: + close(dfd); + return ret; +} diff --git a/tools/mfs/maio-utils/src/parser/CMakeLists.txt b/tools/mfs/maio-utils/src/parser/CMakeLists.txt new file mode 100644 index 000000000000..72e5849da8e4 --- /dev/null +++ b/tools/mfs/maio-utils/src/parser/CMakeLists.txt @@ -0,0 +1,16 @@ +#******************************************************************************* +#******************************************************************************* + +include_directories(${CMAKE_CURRENT_SOURCE_DIR}) + +#******************************************************************************* +#******************************************************************************* + +aux_source_directory(. PARSER_SRC) + +#******************************************************************************* +#******************************************************************************* + +add_library(parser ${PARSER_SRC}) +target_link_libraries(parser infra + securec) diff --git a/tools/mfs/maio-utils/src/parser/req_parser.c b/tools/mfs/maio-utils/src/parser/req_parser.c new file mode 100644 index 000000000000..137eb8d8cb3f --- /dev/null +++ b/tools/mfs/maio-utils/src/parser/req_parser.c @@ -0,0 +1,313 @@ +// SPDX-License-Identifier: GPL-2.0 +/* Copyright(c) 2025 Huawei Technologies Co., Ltd + */ + +#include "inner.h" +#include "log.h" +#include "mfs.h" +#include "policy.h" +#include "securec.h" +#include "threadpool.h" + +#include <stdbool.h> +#include <errno.h> +#include <sys/ioctl.h> +#include <sys/stat.h> + +struct parser_mgr { + uint32_t fs_mode; + uint32_t chunk_size; + const char *mntpoint; + threadpool_t *parser; +}; + +struct parser_mgr g_parser_mgr; + +#define PARSER_NUM 8 +#define MAX_PATH_SIZE 1024 + +#define __round_mask(x, y) ((__typeof__(x))((y)-1)) +#define round_up(x, y) ((((x)-1) | __round_mask(x, y))+1) +#define round_down(x, y) ((x) & ~__round_mask(x, y)) + +#define DIV_ROUND_UP(n, d) (((n) + (d) - 1) / (d)) +#define MIN(x, y) ((x) < (y) ? (x) : (y)) + +bool is_power_of_2(unsigned long n) +{ + return (n != 0 && ((n & (n - 1)) == 0)); +} + +void ctx_cb(int ret, void *args) +{ + struct io_context *ctx = (struct io_context *)args; + + if (ctx->path) + free(ctx->path); + if (ctx->buf) + free(ctx->buf); + free(ctx); +} + +void _submit_local_io(int fd, struct maio_entry *entry) +{ + struct io_context *ioctx; + + ioctx = calloc(1, sizeof(struct io_context)); + if (!ioctx) { + log_error("ioctx alloc failed"); + return; + } + ioctx->off = entry->toff; + ioctx->len = entry->tlen; + ioctx->numaid = entry->tnuma; + ioctx->fd = fd; + ioctx->path = entry->fpath; + ioctx->seq = entry->seq; + ioctx->private = ioctx; + ioctx->end_io = ctx_cb; + loader_io_submit(ioctx); +} + +char *get_file_path(int fd) +{ + struct mfs_ioc_rpath *rp; + char *path; + int ret; + + rp = malloc(sizeof(struct mfs_ioc_rpath) + MAX_PATH_SIZE); + if (!rp) { + log_error("rpath alloc failed"); + return NULL; + } + rp->max = MAX_PATH_SIZE; + ret = ioctl(fd, MFS_IOC_RPATH, (unsigned long)rp); + if (ret) { + log_error("realpath failed for fd:%d, ret:%d", fd, ret); + free(rp); + return NULL; + } + rp->d[rp->len] = '\0'; + + path = malloc(strlen(g_parser_mgr.mntpoint) + rp->len); + if (!path) { + log_error("malloc path %s %s failed", g_parser_mgr.mntpoint, (const char *)rp->d); + return NULL; + } + sprintf(path, "%s%s", g_parser_mgr.mntpoint, (const char *)rp->d); + free(rp); + return path; +} + +char *_get_fullpath(char *fpath, int fd) +{ + struct mfs_ioc_rpath *rp; + char *path; + int ret; + + if (fpath) + return fpath; + rp = malloc(sizeof(struct mfs_ioc_rpath) + MAX_PATH_SIZE); + if (!rp) { + log_error("rpath alloc failed"); + return NULL; + } + rp->max = MAX_PATH_SIZE; + ret = ioctl(fd, MFS_IOC_RPATH, (unsigned long)rp); + if (ret) { + log_error("realpath failed for fd:%d, ret:%d", fd, ret); + free(rp); + return NULL; + } + rp->d[rp->len] = '\0'; + path = strdup((const char *)rp->d); + if (!path) + log_error("strdup path(%s) failed", (const char *)rp->d); + free(rp); + return path; +} + +void maio_preload(struct maio *maio, int nio) +{ + struct maio_entry *entry; + + for (int i = 0; i < nio; ++i) { + entry = &maio->entries[i]; + _submit_local_io(0, entry); + } +} + +uint8_t pid_to_numaid(int pid) +{ + return (uint8_t)-1; +} + +void _process_read_io(uint32_t id, struct maio *maio, int nio) +{ + struct maio_entry *entry; + int i; + + /* process the successor io */ + for (i = 0; i < nio; i++) { + entry = &maio->entries[i]; + _submit_local_io(maio->fd, entry); + } +} + +void _submit_evict(struct maio_entry *entry) +{ + struct io_context *ioctx; + + ioctx = calloc(1, sizeof(struct io_context)); + if (!ioctx) { + log_error("ioctx alloc failed"); + return; + } + ioctx->off = entry->toff; + ioctx->len = entry->tlen; + ioctx->numaid = -1; + ioctx->path = entry->fpath; + ioctx->private = ioctx; + ioctx->end_io = ctx_cb; + loader_evict_submit(ioctx); +} + +void _process_evict(struct mfs_msg *msg) +{ + struct maio_entry *entry; + struct mfs_read *read; + uint64_t soff, eoff; + struct maio *maio; + int ret; + + if (g_parser_mgr.fs_mode != MFS_MODE_LOCAL) + return; + + maio = calloc(1, sizeof(struct maio)); + if (!maio) { + log_error("failed to alloc maio"); + return; + } + + read = (void *)msg->data; + maio->cksz = g_parser_mgr.chunk_size; + soff = round_down(read->off, maio->cksz); + eoff = round_up(read->off + read->len, maio->cksz); + maio->fd = msg->fd; + maio->off = read->off; + maio->len = eoff - soff; + maio->pid = read->pid; + maio->op = msg->opcode; + + ret = policy_evict(&maio); + if (ret <= 0) + goto out; + + for (int i = 0; i < ret; ++i) { + entry = &maio->entries[i]; + _submit_evict(entry); + } +out: + free(maio); +} + +void _parse_read_req(struct mfs_msg *msg) +{ + struct mfs_read *read; + struct maio *maio; + uint64_t soff, eoff, ts; + int ret, io_num = policy_max_io(); + + _process_evict(msg); + + /* alloc maio slots */ + maio = calloc(1, sizeof(struct maio) + io_num * sizeof(struct maio_entry)); + if (!maio) { + log_error("failed to alloc maio, num:%d", io_num); + return; + } + read = (void *)msg->data; + maio->cksz = g_parser_mgr.chunk_size; + soff = round_down(read->off, maio->cksz); + eoff = round_up(read->off + read->len, maio->cksz); + ts = _get_ts(); + maio->fd = msg->fd; + maio->off = soff; + maio->len = eoff - soff; + maio->pid = read->pid; + maio->op = msg->opcode; + maio->ts = ts; + + /* fill slots from policy */ + ret = policy_load(&maio); + if (ret < 0) { + char *path = _get_fullpath(NULL, maio->fd); + if (path) { + log_warn("io on path:%s%s offset:%lu length:%lu dropped", + g_parser_mgr.mntpoint, path, read->off, read->len); + free(path); + } + free(maio); + return; + } + + if (maio->flags & MAIO_WITH_SEQ) + loader_update_curseq(maio->curseq); + + /* keep the input parameter */ + maio->fd = msg->fd; + maio->off = soff; + maio->len = eoff - soff; + maio->cksz = g_parser_mgr.chunk_size; + maio->pid = read->pid; + maio->op = msg->opcode; + maio->ts = ts; + + /* process each maio in slots */ + _process_read_io(msg->id, maio, ret); + free(maio); +} + +void _parse_one_req(void *buf) +{ + struct mfs_msg *msg = (struct mfs_msg *)buf; + + if (msg->opcode == MFS_OP_READ || msg->opcode == MFS_OP_FAULT) + _parse_read_req(msg); + + free(buf); +} + +int maio_parse_req(void *buf, int size) +{ + struct mfs_msg *msg = (struct mfs_msg *)buf; + + if (msg->len != size) + return -1; + + threadpool_submit(g_parser_mgr.parser, buf, _parse_one_req); + return 0; +} + +int parser_init(uint32_t fs_mode, const char *mntpoint) +{ + g_parser_mgr.fs_mode = fs_mode; + g_parser_mgr.mntpoint = strdup(mntpoint); + g_parser_mgr.chunk_size = 4096; + if (!is_power_of_2(g_parser_mgr.chunk_size)) { + log_error("chunk size must be power of 2"); + return -1; + } + g_parser_mgr.parser = threadpool_create("ReqParser", PARSER_NUM, -1); + if (!g_parser_mgr.parser) { + log_error("failed to alloc parser threadpool"); + return -1; + } + + return 0; +} + +void parser_destory(void) +{ + threadpool_destroy(g_parser_mgr.parser); +} diff --git a/tools/mfs/maio-utils/src/policy/CMakeLists.txt b/tools/mfs/maio-utils/src/policy/CMakeLists.txt new file mode 100644 index 000000000000..714023475e07 --- /dev/null +++ b/tools/mfs/maio-utils/src/policy/CMakeLists.txt @@ -0,0 +1,16 @@ +#******************************************************************************* +#******************************************************************************* + +include_directories(${CMAKE_CURRENT_SOURCE_DIR}) + +#******************************************************************************* +#******************************************************************************* + +aux_source_directory(. POLICY_SRC) + +#******************************************************************************* +#******************************************************************************* + +add_library(policy ${POLICY_SRC}) +target_link_libraries(policy infra + securec) diff --git a/tools/mfs/maio-utils/src/policy/policy.c b/tools/mfs/maio-utils/src/policy/policy.c new file mode 100644 index 000000000000..c77ab31752be --- /dev/null +++ b/tools/mfs/maio-utils/src/policy/policy.c @@ -0,0 +1,114 @@ +// SPDX-License-Identifier: GPL-2.0 +/* Copyright(c) 2025 Huawei Technologies Co., Ltd + */ +#include "policy.h" + +#include "maio.h" +#include "log.h" +#include "securec.h" +#include <dlfcn.h> + +typedef struct maio_operation* (*strategy_init)(); + +struct policy_mgr { + void *handle; + struct maio_operation *strategy; +}; + +struct policy_mgr g_policy; + +static int default_init(void) +{ + return 0; +} + +static void default_exit(void) +{ +} + +static int default_load(struct maio **io) +{ + return 0; +} + +static int default_evict(struct maio **io) +{ + return 0; +} + +struct maio_operation default_ops = { + .max_io = 1, + .init = default_init, + .exit = default_exit, + .load = default_load, + .evict = default_evict, +}; + +static void set_default_strategy(void) +{ + g_policy.strategy = &default_ops; +} + +int policy_max_io(void) +{ + return g_policy.strategy->max_io; +} + +/* return the filled num of io */ +int policy_load(struct maio **io) +{ + return g_policy.strategy->load(io); +} + +int policy_evict(struct maio **io) +{ + return g_policy.strategy->evict(io); +} + +int policy_register(const char *path) +{ + strategy_init sinit; + + if (!path || strlen(path) == 0) + return -1; + + g_policy.handle = dlopen(path, RTLD_LAZY); + if (!g_policy.handle) { + log_error("dlopen failed:%s", dlerror()); + return -1; + } + + sinit = dlsym(g_policy.handle, "register_strategy"); + if (!sinit) { + log_error("dlsym failed:%s", dlerror()); + dlclose(g_policy.handle); + g_policy.handle = NULL; + return -1; + } + g_policy.strategy = sinit(); + if (g_policy.strategy->init()) + return -1; + log_info("register strategy:%s done.", path); + return 0; +} + +void policy_unregister(void) +{ + if (g_policy.handle) { + g_policy.strategy->exit(); + dlclose(g_policy.handle); + g_policy.handle = NULL; + } + set_default_strategy(); +} + +int policy_init(void) +{ + g_policy.handle = NULL; + set_default_strategy(); + return 0; +} + +void policy_exit(void) +{ +} diff --git a/tools/mfs/maio-utils/test/Makefile b/tools/mfs/maio-utils/test/Makefile new file mode 100644 index 000000000000..6c87247aafa0 --- /dev/null +++ b/tools/mfs/maio-utils/test/Makefile @@ -0,0 +1,26 @@ +# SPDX-License-Identifier: GPL-2.0 + +CC := gcc +CFLAGS := -fPIC -Wall -Werror -O2 +CPPFLAGS := -fPIC -Wall -Werror -O2 -std=c++11 -g +LDFLAGS := -shared + +CFLAGS += $(addprefix -I,../include) +CPPFLAGS += $(addprefix -I,../include) + +C_SRCS := $(wildcard *.c) +CPP_SRCS := $(wildcard *.cpp) +TARGETS := $(C_SRCS:.c=.so) $(CPP_SRCS:.cpp=.so) + +all: $(TARGETS) + +%.so: %.c + $(CC) $(CFLAGS) $(LDFLAGS) -o $@ $< + +%.so: %.cpp + $(CC) $(CPPFLAGS) $(LDFLAGS) -o $@ $< + +clean: + rm -f $(TARGETS) + +.PHONY: all clean diff --git a/tools/mfs/maio-utils/test/README.md b/tools/mfs/maio-utils/test/README.md new file mode 100644 index 000000000000..a7b7980de933 --- /dev/null +++ b/tools/mfs/maio-utils/test/README.md @@ -0,0 +1,61 @@ +Here, we provide some strategies library for some cases. + +- strategy_demo +A simple strategy library shows how to write a demo. + +- strategy_prefetch +A prefetch strategy library to prefetch data. + +- strategy_toend +A readahead strategy until the end. + +- strategy_generate +A tools using maio-util to generate the io trace. + +- strategy_replay +A strategy to replay the io trace. + +### How to build + +Using `make -j` to build the libraries. + +### How to use + +Set the library path as the `-s` parameter's value to run with maio-utils. Here we show usage of the advanced replay strategy: + +- generate the io trace + +> assume you have mounted the mfs at /mnt/mfs + +This is the tracing period. + +``` +export MAIO_TRACE_FILE=trace.txt +export MAIO_MNTPOINT=/mnt/mfs # /mnt/mfs is the MFS mountpoint +./bin/maio-utils -m /mnt/mfs -s /home/lihb/workspace/hulk-6.6/maio-utils/test/strategy_generate.so +``` +Then, you can trigger the io. After finishing, you would get the trace file with the content likes: + +``` +xxx/model-00001-of-000008.safetensors 0 32768 0 +xxx/model-00001-of-000008.safetensors 32768 32768 0 +xxx/model-00001-of-000008.safetensors 65536 32768 0 +xxx/model-00001-of-000008.safetensors 98304 32768 0 +xxx/model-00001-of-000008.safetensors 131072 32768 0 +... + +``` + +The format is "<path> <offset> <length> <numaid>". + +- replay the io trace + +> assume you have mounted the mfs at /mnt/mfs and begin to launch the model. + +This is the running period. + +``` +export MAIO_TRACE_FILE=trace.txt +export MAIO_MNTPOINT=/mnt/mfs # /mnt/mfs is the MFS mountpoint +./bin/maio-utils -m /mnt/mfs -s /home/lihb/workspace/hulk-6.6/maio-utils/test/replay.so +``` diff --git a/tools/mfs/maio-utils/test/strategy_demo.c b/tools/mfs/maio-utils/test/strategy_demo.c new file mode 100644 index 000000000000..82a5cd178b03 --- /dev/null +++ b/tools/mfs/maio-utils/test/strategy_demo.c @@ -0,0 +1,45 @@ +// SPDX-License-Identifier: GPL-2.0 +/* Copyright(c) 2025 Huawei Technologies Co., Ltd + */ +#include "maio.h" +#include <stddef.h> + +int demo_init(void) +{ + return 0; +} + +void demo_exit(void) +{ +} + +int demo_load(struct maio **io) +{ + struct maio_entry *entry; + + entry = (*io)->entries; + entry[0].toff = (*io)->off + (*io)->len; + entry[0].tlen = (*io)->cksz; + entry[0].fpath = NULL; + entry[0].tnuma = -1; + return 1; +} + +int demo_evict(struct maio **io) +{ + ((void)io); + return 0; +} + +const struct maio_operation demo_strategy = { + .max_io = 1, + .init = demo_init, + .exit = demo_exit, + .load = demo_load, + .evict = demo_evict, +}; + +struct maio_operation *register_strategy(void) +{ + return (struct maio_operation *)&demo_strategy; +} diff --git a/tools/mfs/maio-utils/test/strategy_generate.cpp b/tools/mfs/maio-utils/test/strategy_generate.cpp new file mode 100644 index 000000000000..929be30e6a42 --- /dev/null +++ b/tools/mfs/maio-utils/test/strategy_generate.cpp @@ -0,0 +1,296 @@ +// SPDX-License-Identifier: GPL-2.0 +/* Copyright(c) 2025 Huawei Technologies Co., Ltd + */ +#include "maio.h" +#include "strategy_template.h" + +#include <array> +#include <atomic> +#include <cstring> +#include <condition_variable> +#include <fstream> +#include <sstream> +#include <iostream> +#include <mutex> +#include <queue> +#include <thread> +#include <unordered_map> +#include <vector> +#include <unistd.h> + +using namespace std; + +#define MAIO_MNTPOINT "MAIO_MNTPOINT" +#define MAIO_TRACE_FILE "MAIO_TRACE_FILE" +#define CACHEDIR "cachedir" + +#define MAXLINE 1024 + +/* Data struct definition */ +template<typename T> +class SafeQueue { +private: + queue<T> data_queue; + mutex mtx; + condition_variable cv; + +public: + // 添加数据到队列 + void push(T item) { + lock_guard<mutex> lock(mtx); + data_queue.push(move(item)); + cv.notify_one(); + } + + // 从队列取出数据（阻塞直到有数据） + T pop() { + unique_lock<mutex> lock(mtx); + cv.wait(lock, [this]{ return !data_queue.empty(); }); + T item = move(data_queue.front()); + data_queue.pop(); + return item; + } + + // 检查队列是否为空 + bool empty() { + lock_guard<mutex> lock(mtx); + return data_queue.empty(); + } +}; + +struct io { + string name; + uint64_t off; + uint64_t len; + uint8_t npu; + + io(const string& _name, uint64_t _off, uint64_t _len, uint8_t _npu) + : name(_name), off(_off), len(_len), npu(_npu) {} +}; + +/* Global variable definition */ +unordered_map<int, char*> fd_map; /* map fd to path */ +mutex fd_mutex; +string cachedir; + +vector<io> iovec; /* io traces */ +mutex iovec_mutex; +SafeQueue<vector<io>> iovec_queue; + +ofstream tfile; /* trace file */ + +thread writer; + +unordered_map<int, int> process_npu_map; +mutex pnpu_mutex; + +atomic<int> npu_smi_enable(-1); + +/* fd_map helper function */ +char* fd_path_find(int iofd) { + lock_guard<mutex> lock(fd_mutex); + auto it = fd_map.find(iofd); + if (it != fd_map.end()) { + return it->second; + } else { + char *path = _get_fullpath(NULL, iofd); + fd_map.insert({iofd, path}); + return path; + } +} + +/* iovec helper function */ +void io_insert_data(string name, uint64_t off, uint64_t len, uint8_t npu) +{ + lock_guard<mutex> lock(iovec_mutex); + iovec.emplace_back(name, off, len, npu); + + vector<io> copy_iovec = move(iovec); + iovec.clear(); + + iovec_queue.push(move(copy_iovec)); +} + +void get_mount_option(const char* mntpoint, const char* option) +{ + ifstream mounts("/proc/mounts"); + string line, target_mp; + + if (!mounts.is_open()) { + cerr << "Error: could not open /proc/mounts" << endl; + return; + } + + while (getline(mounts, line)) { + istringstream iss(line); + string device, mp, fs_type, options; + + if (!(iss >> device >> mp >> fs_type >> options)) + continue; + + if (mp.back() != '/') + mp += '/'; + target_mp = mntpoint; + if (target_mp.back() != '/') + target_mp += '/'; + + if (target_mp == mp) { + istringstream opt_stream(options); + string opt; + + while (getline(opt_stream, opt, ',')) { + if (opt.find(option) == 0) { + if (opt[strlen(option)] == '=') { + cachedir = opt.substr(strlen(option) + 1); + } + } + } + break; + } + } +} + +void writer_thread() { + while (true) { + vector<io> iovec_to_write = iovec_queue.pop(); + + for (const auto &item : iovec_to_write) + tfile << cachedir << item.name << " " << item.off << " " << item.len << " " << (unsigned int)item.npu << endl; + tfile.flush(); + } +} + +string exec_cmd(const char* cmd) { + array<char, 128> buffer; + string result; + unique_ptr<FILE, decltype(&pclose)> pipe(popen(cmd, "r"), pclose); + + if (!pipe) { + throw std::runtime_error("popen() failed!"); + } + while (fgets(buffer.data(), buffer.size(), pipe.get()) != nullptr) { + result += buffer.data(); + } + return result; +} + +bool isNumber(const std::string& str) { + return !str.empty() && str.find_first_not_of("0123456789") == std::string::npos; +} + +void get_npu_info(string &npuinfo) +{ + istringstream iss(npuinfo); + string line; + + while (getline(iss, line)) { + istringstream lineStream(line); + string token; + vector<string> tokens; + while (getline(lineStream, token, ' ')) { + if (!token.empty()) { + tokens.push_back(token); + } + } + if (tokens.size() >= 3) { + if (!isNumber(tokens[1]) || !isNumber(tokens[4])) + continue; + int npuId = stoi(tokens[1]); + int processId = stoi(tokens[4]); + process_npu_map[processId] = npuId; + } + } +} + +int process_npu_find(int processid) { + lock_guard<mutex> lock(pnpu_mutex); + auto it = process_npu_map.find(processid); + + if (npu_smi_enable.load() == -1) { + if (exec_cmd("command -v npu-smi").empty()) + npu_smi_enable.store(0); + else + npu_smi_enable.store(1); + } + + if (npu_smi_enable.load() == 0) + return 0; + + if (it != process_npu_map.end()) { + return it->second; + } else { + string result = exec_cmd("npu-smi info"); + get_npu_info(result); + auto it = process_npu_map.find(processid); + if (it != process_npu_map.end()) + return it->second; + else + return 0; + } +} + +int strategy_init(void) +{ + char *tfilename = getenv(MAIO_TRACE_FILE); + char *mntpoint = getenv(MAIO_MNTPOINT); + + if(!tfilename || !mntpoint) { + printf("no MAIO_TRACE_FILE or MAIO_MNTPOINT env variable\n"); + return -1; + } + get_mount_option(mntpoint, CACHEDIR); + string stfilename = tfilename; + tfile.open(stfilename, ios::trunc); + writer = thread(writer_thread); + if (!tfile.is_open()) { + cerr << "Error: could not open" << stfilename << endl; + return -1; + } + return 0; +} + +void strategy_exit(void) +{ + lock_guard<mutex> lock(iovec_mutex); + vector<io> copy_iovec = move(iovec); + iovec.clear(); + iovec_queue.push(move(copy_iovec)); + + tfile.close(); + writer.join(); +} + +int strategy_load(struct maio **io) +{ + char *path; + string pathStr; + int ret = 0; + + if (!tfile.is_open()) + return -1; + + path = fd_path_find((*io)->fd); + pathStr.assign(path); + io_insert_data(pathStr, (*io)->off, (*io)->len, process_npu_find((*io)->pid)); + + return ret; +} + +int strategy_evict(struct maio **io) +{ + ((void)io); + return 0; +} + +const struct maio_operation maio_strategy = { + .max_io = 1, + .init = strategy_init, + .exit = strategy_exit, + .load = strategy_load, + .evict = strategy_evict, +}; + +struct maio_operation *register_strategy() +{ + return (struct maio_operation *)&maio_strategy; +} diff --git a/tools/mfs/maio-utils/test/strategy_prefetch.cpp b/tools/mfs/maio-utils/test/strategy_prefetch.cpp new file mode 100644 index 000000000000..9ba205931cdc --- /dev/null +++ b/tools/mfs/maio-utils/test/strategy_prefetch.cpp @@ -0,0 +1,150 @@ +#include "strategy_prefetch.h" +#include "maio.h" +#include <iostream> +#include <map> +#include <cstring> +#include <string> +#include <stdint.h> +#include <sys/stat.h> +#include <sys/mman.h> +#include <dirent.h> +#include <fcntl.h> +#include <unistd.h> + +using namespace std; + +struct prefetch_mgr { + map<string, uint64_t> files; +}; + +struct prefetch_mgr g_prefetch; + +int get_files(const char *parent) +{ + DIR *dir; + struct dirent *entry; + string filepath; + struct stat buf; + + if ((dir = opendir(parent)) == NULL) { + cout<<"opening directory failed"<<endl; + return -1; + } + + while ((entry = readdir(dir)) != NULL) { + if (strcmp(entry->d_name, ".") == 0 || strcmp(entry->d_name, "..") == 0) + continue; + + filepath = string(parent) + "/" + string(entry->d_name); + if (stat(filepath.c_str(), &buf) == -1) { + cout<<"stat path:"<<filepath<<" failed"<<endl; + continue; + } + if (S_ISDIR(buf.st_mode)) + continue; + cout<<"Find file:"<<filepath<<" size:"<<buf.st_size<<endl; + g_prefetch.files.insert(pair<string, uint64_t>(filepath, buf.st_size)); + } + + map<string, uint64_t>::iterator it; + for (it = g_prefetch.files.begin(); it != g_prefetch.files.end(); ++it) { + cout<<"path:"<<it->first<<" size:"<<it->second<<endl; + } + return 0; +} + +int prefetch_init(void) +{ + char *parent = getenv("MODEL_WEIGHT_DIR"); + if (!parent) { + cout<<"please set the env MODEL_WEIGHT_DIR"<<endl; + return -1; + } + + if (get_files(parent) != 0) { + cout<<"get all files failed"<<endl; + return -1; + } + return 0; +} + +void prefetch_exit(void) +{ +} + +struct thread_ctx { + char *path; + uint64_t off; + uint64_t len; +}; + +void *fault(void *arg) +{ + struct thread_ctx *ctx = (struct thread_ctx *)arg; + int fd = open(ctx->path, O_RDONLY); + if (fd < 0) { + cout<<"open file:%s"<<ctx->path<<" failed"<<endl; + free(ctx); + return NULL; + } + + cout<<"FAULT path:"<<ctx->path<<" fd:"<<fd<<" off:"<<ctx->off<<" len:"<<ctx->len<<endl; + void *addr = mmap(NULL, ctx->len, PROT_READ, MAP_SHARED, fd, 0); + uint64_t idx; + char tmp, total, *buffer = (char *)addr; + total = 'B'; + for (idx = 0; idx < ctx->len; idx += 4096) { + tmp = buffer[idx]; + total += tmp; + } + cout<<"Fault calculate total:"<<total<<endl; + munmap(addr, ctx->len); + close(fd); + free(ctx); + return NULL; +} + +int prefetch_load(struct maio **io) +{ + map<string, uint64_t>::iterator it; + struct thread_ctx *ctx; + pthread_t t0; + int ret; + + for (it = g_prefetch.files.begin(); it != g_prefetch.files.end(); ++it) { + ctx = (struct thread_ctx *)malloc(sizeof(struct thread_ctx)); + if (!ctx) { + cout<<"malloc ctx failed"<<endl; + continue; + } + ctx->path = strdup(it->first.c_str()); + ctx->off = 0; + ctx->len = it->second; + ret = pthread_create(&t0, NULL, fault, ctx); + if (ret == 0) + pthread_detach(t0); + } + g_prefetch.files.clear(); + + return 0; +} + +int prefetch_evict(struct maio **io) +{ + ((void)io); + return 0; +} + +const struct maio_operation prefetch_strategy = { + .max_io = 1, + .init = prefetch_init, + .exit = prefetch_exit, + .load = prefetch_load, + .evict = prefetch_evict, +}; + +struct maio_operation *register_strategy() +{ + return (struct maio_operation *)&prefetch_strategy; +} + diff --git a/tools/mfs/maio-utils/test/strategy_prefetch.h b/tools/mfs/maio-utils/test/strategy_prefetch.h new file mode 100644 index 000000000000..e487e38b3d8b --- /dev/null +++ b/tools/mfs/maio-utils/test/strategy_prefetch.h @@ -0,0 +1,21 @@ +#ifndef STRATEGY_PREFETCH_H +#define STRATEGY_PREFETCH_H + +#ifdef __cplusplus +extern "C" { +#endif + +int prefetch_init(void); +void prefetch_exit(void); +int prefetch_load(struct maio *io); +int prefetch_evict(struct maio **io); + +extern const struct maio_operation prefetch_strategy; + +struct maio_operation *register_strategy(void); + +#ifdef __cplusplus +} +#endif + +#endif diff --git a/tools/mfs/maio-utils/test/strategy_replay.cpp b/tools/mfs/maio-utils/test/strategy_replay.cpp new file mode 100644 index 000000000000..85c1c7defedd --- /dev/null +++ b/tools/mfs/maio-utils/test/strategy_replay.cpp @@ -0,0 +1,540 @@ +// SPDX-License-Identifier: GPL-2.0 +/* Copyright(c) 2025 Huawei Technologies Co., Ltd + */ +#include "maio.h" +#include "strategy_template.h" + +#include <atomic> +#include <array> +#include <cstring> +#include <fstream> +#include <iostream> +#include <map> +#include <mutex> +#include <sstream> +#include <string> +#include <thread> +#include <unordered_map> +#include <vector> +#include <pthread.h> +#include <fcntl.h> +#include <unistd.h> + +using namespace std; + +#define MAIO_MNTPOINT "MAIO_MNTPOINT" +#define MAIO_TRACE_FILE "MAIO_TRACE_FILE" +#define CACHEDIR "cachedir" +#define MAXLINE 1024 + +/* Data structure definition */ +enum { + IO_NONE = 0, + IO_LOADED, + IO_EVICTED, +}; + +struct io { + string name; + uint64_t off; + uint64_t len; + uint8_t npu; + int flag; + + explicit io(const string& _name, uint64_t _off, uint64_t _len, uint8_t _npu) + : name(_name), off(_off), len(_len), npu(_npu), flag(IO_NONE) {} +}; + + +struct ioKey { + string name; + uint64_t off; + uint64_t len; + uint8_t npu; + + ioKey(const string& _name, uint64_t _off, uint64_t _len, uint8_t _npu) + : name(_name), off(_off), len(_len), npu(_npu) {} + + bool operator<(const ioKey &other) const { + return name < other.name || (name == other.name && off < other.off) || + (name == other.name && off == other.off && len < other.len); + } +}; + +vector<io> iovec; /* io traces */ +map<ioKey, int> io_to_index; +unordered_map<int, char*> fd_map; +mutex fd_mutex; +string cachedir; + +unordered_map<int, int> npu_numa_map; + +unordered_map<int, int> process_npu_map; +mutex pnpu_mutex; + +/* for evict */ +#define EVICT_DIST 10000 +#define EVICT_IO_MAX 5000 +thread_local size_t evict_idx = 0; + +atomic<int> npu_smi_enable(-1); +atomic<uint64_t> memory_usage(0); + +uint64_t memory_limit; +uint64_t memory_limit_bound; +string usage_in_bytes_path = "memory.usage_in_bytes"; +string limit_in_bytes_path = "memory.limit_in_bytes"; +string cgroup_mem=""; + +string get_cgroup_path() +{ + fstream cgroup_file("/proc/self/cgroup"); + if (!cgroup_file.is_open()) { + cerr << "Failed to open /proc/self/cgroup" << endl; + return ""; + } + + string line; + string cgroup_path; + while (getline(cgroup_file, line)) { + istringstream iss(line); + string hierarchy, controller, path; + if (!getline(iss, hierarchy, ':')) + continue; + if (!getline(iss, controller, ':')) + continue; + if (!getline(iss, path, ':')) + continue; + if (controller == "memory") { + cgroup_path = path; + break; + } + } + cgroup_file.close(); + return cgroup_path; +} + +/* read from /sys/fs/cgroup/memory/memory.[usage_in_bytes|limit_in_bytes] */ +uint64_t read_cgroup_memory(const string &file_path) +{ + if (cgroup_mem == "") + return -1; + string whole_path = "/sys/fs/cgroup/memory"+cgroup_mem+"/"+file_path; + ifstream cgroup_file(whole_path); + + if(!cgroup_file.is_open()) { + cerr << "Failed to open " << file_path << endl; + return -1; + } + + string line; + while(getline(cgroup_file, line)) { + try { + return stoll(line); + } catch (...) { + cerr << "Failed to parse line" << line << endl; + } + } + + cgroup_file.close(); + return -1; +} + +/* mount helper */ +void get_mount_option(const char* mntpoint, const char* option) +{ + ifstream mounts("/proc/mounts"); + string line, target_mp; + + if (!mounts.is_open()) { + cerr << "Error: could not open /proc/mounts" << endl; + return; + } + + while (getline(mounts, line)) { + istringstream iss(line); + string device, mp, fs_type, options; + + if (!(iss >> device >> mp >> fs_type >> options)) + continue; + + if (mp.back() != '/') + mp += '/'; + target_mp = mntpoint; + if (target_mp.back() != '/') + target_mp += '/'; + + if (target_mp == mp) { + istringstream opt_stream(options); + string opt; + + while (getline(opt_stream, opt, ',')) { + if (opt.find(option) == 0) { + if (opt[strlen(option)] == '=') { + cachedir = opt.substr(strlen(option) + 1); + } + } + } + break; + } + } +} + +/* fd_map helper function */ +char* fd_path_find(int iofd) { + lock_guard<mutex> lock(fd_mutex); + auto it = fd_map.find(iofd); + if (it != fd_map.end()) { + return it->second; + } else { + char *path = _get_fullpath(NULL, iofd); + size_t len = strlen(path) + cachedir.length() + 1; + char *realpath = (char*)malloc(len); + snprintf(realpath, len, "%s%s", cachedir.c_str(), path); + fd_map.insert({iofd, realpath}); + return realpath; + } +} + +string exec_cmd(const char* cmd) { + array<char, 128> buffer; + string result = ""; + unique_ptr<FILE, decltype(&pclose)> pipe(popen(cmd, "r"), pclose); + + if (!pipe) { + throw std::runtime_error("popen() failed!"); + } + while (fgets(buffer.data(), buffer.size(), pipe.get()) != nullptr) { + result += buffer.data(); + } + return result; +} + +bool isNumber(const std::string& str) { + return !str.empty() && str.find_first_not_of("0123456789") == std::string::npos; +} + +/* cpu_numa_map builder from lscpu */ +unordered_map<string, int> build_cpu_numa_map() { + unordered_map<string, int> cpuNumaMap; + string result = exec_cmd("lscpu"); + istringstream iss(result); + string line; + + while (getline(iss, line)) { + if (line.find("NUMA node") != string::npos) { + istringstream lineStream(line); + string key, value; + int numaNode; + getline(lineStream, value, ':'); + getline(lineStream, key); + /* what if numa is more than 10 */ + value = value.substr(value.find("NUMA node") + 9, 1); + if (!isNumber(value)) + continue; + numaNode = stoi(value); + key.erase(0, key.find_first_not_of(" \t")); + key.erase(key.find_last_not_of(" \t") + 1); + cpuNumaMap[key]=numaNode; + } + } + return cpuNumaMap; +} + +/* npu_numa_map builder, from npu-smi info -t topo */ +void build_npu_map_map() +{ + if (npu_smi_enable.load() == -1) { + if (exec_cmd("command -v npu-smi").empty()) + npu_smi_enable.store(0); + else + npu_smi_enable.store(1); + } + + if (npu_smi_enable.load() == 0) + return; + + unordered_map<string, int> cpuNumaMap = build_cpu_numa_map(); + string result = exec_cmd("npu-smi info -t topo"); + istringstream iss(result); + string line; + size_t target_token_size = 0; + + cout << "NPU\tCPU_AFFINITY\tNUMA_NODE" << endl; + while (getline(iss, line)) { + istringstream lineStream(line); + string npuStr, numaStr, token; + vector<string> tokens; + + while (lineStream >> token) { + tokens.push_back(token); + } + + if (tokens.size() >= 2 && tokens.back() == "Affinity") { + target_token_size = tokens.size() - 1 + 1; + continue; + } + + if (tokens.size() != target_token_size) + continue; + + int npuId = stoi(tokens[0].substr(3)); + + numaStr = tokens.back(); + cout << npuId << "\t" << numaStr << "\t\t" << cpuNumaMap[numaStr] << endl; + npu_numa_map[npuId] = cpuNumaMap[numaStr]; + } +} + +/* process_npu_map builder, from npu-smi info */ +void get_npu_info(string &npuinfo) +{ + istringstream iss(npuinfo); + string line; + + while (getline(iss, line)) { + istringstream lineStream(line); + string token; + vector<string> tokens; + while (getline(lineStream, token, ' ')) { + if (!token.empty()) { + tokens.push_back(token); + } + } + if (tokens.size() >= 3) { + if (!isNumber(tokens[1]) || !isNumber(tokens[4])) + continue; + int npuId = stoi(tokens[1]); + int processId = stoi(tokens[4]); + process_npu_map[processId] = npuId; + } + } +} + +/* process_npu_map helper */ +int process_npu_find(int processid) { + lock_guard<mutex> lock(pnpu_mutex); + auto it = process_npu_map.find(processid); + + if (npu_smi_enable.load() == 0) + return 0; + + if (it != process_npu_map.end()) { + return it->second; + } else { + string result = exec_cmd("npu-smi info"); + get_npu_info(result); + auto it = process_npu_map.find(processid); + if (it != process_npu_map.end()) + return it->second; + else + return 0; + } +} + +int strategy_init(void) +{ + const char *mntpoint = getenv(MAIO_MNTPOINT); + char *tfilename = getenv(MAIO_TRACE_FILE); + if(!tfilename || !mntpoint) { + fprintf(stderr, "no MAIO_TRACE_FILE or MAIO_MNTPOINT env variable\n"); + return -1; + } + ifstream tfile(tfilename); + string line; + + cgroup_mem=get_cgroup_path(); + memory_limit = read_cgroup_memory(limit_in_bytes_path); + memory_limit_bound = memory_limit * 4 / 5; + cout << "memory_limit: " << memory_limit << " memory_limit_bound: " << memory_limit_bound << endl; + get_mount_option(mntpoint, CACHEDIR); + build_npu_map_map(); + + if(!tfile.is_open()) { + cerr << "Failed to open file:" << tfilename << endl; + return -1; + } + + while(getline(tfile, line)) { + istringstream iss(line); + + string name; + uint64_t off; + uint64_t len; + unsigned int npu; + if (iss >> name >> off >> len >> npu) { + iovec.emplace_back(name, off, len, npu); + io_to_index[ioKey(name, off, len, npu)] = iovec.size() - 1; + } else { + cerr << "Failed to parse line:" << line << endl; + } + } + + tfile.close(); + return 0; +} + +void strategy_exit(void) +{ +} + +int strategy_load(struct maio **io) +{ + int npuid = process_npu_find((*io)->pid); + struct maio_entry *entry; + string pathStr; + char *path; + uint64_t idx = iovec.size(); + int ret = 0; + struct maio *new_io; + + path = fd_path_find((*io)->fd); + pathStr.assign(path); + + ioKey key(pathStr, (*io)->off, (*io)->len, npuid); + auto io_it = io_to_index.lower_bound(key); + if (io_it == io_to_index.end()) { + entry = (*io)->entries; + entry[0].toff = (*io)->off + (*io)->len; + entry[0].tlen = 0; + entry[0].fpath = NULL; + entry[0].tnuma = npu_numa_map[npuid]; + return 1; + } + if (key < io_it->first) { + if (io_it != io_to_index.begin()) { + --io_it; + if (io_it->first.name != pathStr) { + ++io_it; + } + } + } + + idx = io_it->second; + (*io)->flags |= MAIO_WITH_SEQ; + (*io)->curseq = idx; + entry = (*io)->entries; + int ios = 0; + uint64_t memory_usage = read_cgroup_memory(usage_in_bytes_path); + vector<int> io_extra; + + /* fill ios below max_io */ + while (idx < iovec.size()) { + if (iovec[idx].flag == IO_LOADED) { + idx++; + continue; + } + memory_usage += iovec[idx].len / 2; + if (memory_usage > memory_limit_bound) + break; + if (ios >= maio_strategy.max_io) { + io_extra.push_back(idx); + goto next_io; + } + entry[ios].seq = idx; + if (pathStr == iovec[idx].name) { + entry[ios].toff = iovec[idx].off; + entry[ios].tlen = iovec[idx].len; + entry[ios].tnuma = npu_numa_map[npuid]; + } else { + entry[ios].fpath = strdup(iovec[idx].name.c_str()); + entry[ios].toff = iovec[idx].off; + entry[ios].tlen = iovec[idx].len; + entry[ios].tnuma = npu_numa_map[iovec[idx].npu]; + } +next_io: + iovec[idx].flag = IO_LOADED; + idx++; + ios++; + } + + /* fill ios above max_io */ + if (ios > maio_strategy.max_io) { + new_io = (struct maio *)realloc(*io, sizeof(struct maio) + ios * sizeof(struct maio_entry)); + if (!new_io) { + printf("failed to alloc ios:%d maio\n", ios); + return -1; + } + *io = new_io; + entry = (*io)->entries; + for (int i = maio_strategy.max_io; i < ios; ++i) { + int io_idx = io_extra[i - maio_strategy.max_io]; + entry[i].seq = io_idx; + if (pathStr == iovec[io_idx].name) { + entry[i].toff = iovec[io_idx].off; + entry[i].tlen = iovec[io_idx].len; + entry[i].tnuma = npu_numa_map[npuid]; + } else { + entry[i].fpath = strdup(iovec[io_idx].name.c_str()); + entry[i].toff = iovec[io_idx].off; + entry[i].tlen = iovec[io_idx].len; + entry[i].tnuma = npu_numa_map[iovec[io_idx].npu]; + } + } + } + + // printf("file:%s off:%lu len:%lu generate_ios:%d from %d",path, (*io)->off, (*io)->len, ios, io_it->second); + ret = ios; + return ret; +} + +int strategy_evict(struct maio **io) +{ + string pathStr; + char *path; + path = fd_path_find((*io)->fd); + pathStr.assign(path); + ioKey key(pathStr, (*io)->off, (*io)->len, (*io)->pid); + auto io_it = io_to_index.lower_bound(key); + int dist = 0; + size_t i = 0, io_idx = io_it->second; + struct maio_entry *entry; + struct maio *new_io; + uint64_t evict_memory = 0; + uint64_t memory_usage = read_cgroup_memory(usage_in_bytes_path); + + if (memory_usage < (memory_limit_bound * 3 / 4)) + return 0; + dist = io_idx - evict_idx; + if (dist < 0) + return 0; + if (static_cast<size_t>(dist) < iovec.size() / 8 && io_it != io_to_index.end()) + return 0; + size_t evict_win = 0; + if (io_it != io_to_index.end()) { + evict_win = min(iovec.size() / 10, io_idx - 1); + } else { + evict_win = dist; + } + + new_io = (struct maio*)realloc(*io, sizeof(struct maio) + evict_win * sizeof(struct maio_entry)); + if (!new_io) + return -1; + *io = new_io; + entry = new_io->entries; + + for (i = 0; i < evict_win && evict_idx < iovec.size(); ++i) { + evict_memory += iovec[evict_idx].len; + if (evict_memory > memory_limit_bound / 8) + break; + entry[i].fpath = strdup(iovec[evict_idx].name.c_str()); + entry[i].toff = iovec[evict_idx].off; + entry[i].tlen = iovec[evict_idx].len; + evict_idx++; + iovec[evict_idx].flag = IO_EVICTED; + } + return i; +} + +const struct maio_operation maio_strategy = { + .max_io = 500, + .init = strategy_init, + .exit = strategy_exit, + .load = strategy_load, + .evict = strategy_evict, +}; + +struct maio_operation *register_strategy() +{ + return (struct maio_operation *)&maio_strategy; +} diff --git a/tools/mfs/maio-utils/test/strategy_template.h b/tools/mfs/maio-utils/test/strategy_template.h new file mode 100644 index 000000000000..c574d72e4a03 --- /dev/null +++ b/tools/mfs/maio-utils/test/strategy_template.h @@ -0,0 +1,21 @@ +#ifndef STRATEGY_TEMPLATE_H +#define STRATEGY_TEMPLATE_H + +#ifdef __cplusplus +extern "C" { +#endif + +int strategy_init(void); +void strategy_exit(void); +int strategy_load(struct maio **io); +int strategy_evict(struct maio *io); + +extern const struct maio_operation maio_strategy; + +struct maio_operation *register_strategy(void); + +#ifdef __cplusplus +} +#endif + +#endif diff --git a/tools/mfs/maio-utils/test/strategy_toend.c b/tools/mfs/maio-utils/test/strategy_toend.c new file mode 100644 index 000000000000..56a068d450e5 --- /dev/null +++ b/tools/mfs/maio-utils/test/strategy_toend.c @@ -0,0 +1,46 @@ +// SPDX-License-Identifier: GPL-2.0 +/* Copyright(c) 2025 Huawei Technologies Co., Ltd + */ +#include "maio.h" +#include <stddef.h> + +int demo_init(void) +{ + return 0; +} + +void demo_exit(void) +{ +} + +int demo_load(struct maio **io) +{ + struct maio_entry *entry; + + entry = (*io)->entries; + entry[0].toff = (*io)->off + (*io)->len; + /* Read Ahead until the end */ + entry[0].tlen = 0; + entry[0].fpath = NULL; + entry[0].tnuma = -1; + return 1; +} + +int demo_evict(struct maio **io) +{ + ((void)io); + return 0; +} + +const struct maio_operation demo_strategy = { + .max_io = 1, + .init = demo_init, + .exit = demo_exit, + .load = demo_load, + .evict = demo_evict, +}; + +struct maio_operation *register_strategy(void) +{ + return (struct maio_operation *)&demo_strategy; +} diff --git a/tools/mfs/maio-utils/test/trace_analyse.py b/tools/mfs/maio-utils/test/trace_analyse.py new file mode 100644 index 000000000000..c25d1ce364df --- /dev/null +++ b/tools/mfs/maio-utils/test/trace_analyse.py @@ -0,0 +1,87 @@ +def merge_intervals(intervals): + """ + 合并区间 + :param intervals: 区间列表，每个区间是一个元组 (start, end) + :return: 合并后的区间列表 + """ + if not intervals: + return [] + + # 按起始位置排序 + intervals.sort(key=lambda x: x[0]) + + merged = [intervals[0]] + for current in intervals[1:]: + last_merged = merged[-1] + if current[0] <= last_merged[1]: # 有重叠 + merged[-1] = (last_merged[0], max(last_merged[1], current[1])) + else: + merged.append(current) + + return merged + +def process_file(input_file): + """ + 处理输入文件 + :param input_file: 输入文件路径 + :return: 每个文件的合并区间和总大小 + """ + from collections import defaultdict + + # 使用字典按文件名分组 + file_intervals = defaultdict(list) + + with open(input_file, 'r') as f: + for line in f: + parts = line.strip().split() + if len(parts) < 3: + print(f"跳过无效行: {line.strip()}") + continue + + # 解析文件名、起始位置、大小 + filename, start, size = parts[:3] + start = int(start) + size = int(size) + + # 计算区间 + end = start + size + file_intervals[filename].append((start, end)) + + # 合并每个文件的区间 + merged_results = {} + for filename, intervals in file_intervals.items(): + merged_intervals = merge_intervals(intervals) + # 计算合并后的区间总大小 + total_size = sum(end - start for start, end in merged_intervals) + merged_results[filename] = { + "merged_intervals": merged_intervals, + "total_size": total_size + } + + return merged_results + +def main(): + import sys + + if len(sys.argv) != 2: + print("用法: python script.py <输入文件>") + sys.exit(1) + + input_file = sys.argv[1] + + merged_results = process_file(input_file) + + whole_result=0 + + for filename, result in merged_results.items(): + print(f"文件: {filename}") + # print("合并后的区间:") + # for interval in result["merged_intervals"]: + # print(f"起始位置: {interval[0]}, 结束位置: {interval[1]}") + print(f"所有区间内数据的总大小: {result['total_size']} 字节") + print("-" * 40) + whole_result+=result['total_size'] + print(f"所有文件内数据的总大小: {whole_result} 字节") + +if __name__ == "__main__": + main() diff --git a/tools/mfs/maio-utils/test/trace_gen.py b/tools/mfs/maio-utils/test/trace_gen.py new file mode 100644 index 000000000000..d050fa48801c --- /dev/null +++ b/tools/mfs/maio-utils/test/trace_gen.py @@ -0,0 +1,116 @@ +import argparse +import os +import sys + +FILENAME_IDX = 0 +OFF_IDX = 1 +LEN_IDX = 2 +NUMA_IDX = 3 +OFF_BOUND = 1024 * 1024 * 4096 + +class ReplayHintGenerator(object): + def __init__(self, trace_file, output_file): + self.output_file = output_file + self.trace_file = trace_file + self.data_list = [] + + def generate_hint(self): + with open(self.output_file, 'w') as f: + for data in self.data_list: + data_off = data[OFF_IDX] + data_len = data[LEN_IDX] + print(data_off) + step = (512 * 1024 * 1024) + while data_len: + if data_len < step: + step = data_len + print(f"{data[FILENAME_IDX]} {data_off} {step} {data[NUMA_IDX]}", file=f) + data_len-=step + data_off+=step + + def parse_trace(self): + file_map_index={} + with open(self.trace_file, 'r') as f: + for line in f: + parts = line.strip().split(' ') + + filename, off, length, numaid = parts + off = int(off) + length = int(length) + numaid = int(numaid) + if self.data_list: + tail_data = self.data_list[-1] + ''' + 首先尝试跟末尾的数据合并: + 如若跟末尾的数据是同一个文件跟numaid, + 两者之间的距离小于设定值, + 就进行合并 + ''' + if (tail_data[FILENAME_IDX] == filename and tail_data[NUMA_IDX] == numaid): + tail_off = tail_data[OFF_IDX] + tail_end = tail_data[OFF_IDX] + tail_data[LEN_IDX] + end = off + length + abs_off = max(abs(off - tail_end), abs(tail_off - end)) + if abs_off < OFF_BOUND: + tail_data[OFF_IDX] = min(tail_off, off) + tail_data[LEN_IDX] = max(tail_end, end) - tail_data[OFF_IDX] + continue + else: + ''' + 否则跟之前同一文件和numaid的数据进行合并 + 如若两者之间距离小于设定值， + 就进行合并 + ''' + key = (filename, numaid) + if key in file_map_index: + index = file_map_index[key] + data = self.data_list[index] + data_off = data[OFF_IDX] + data_end = data[OFF_IDX] + data[LEN_IDX] + end = off + length + abs_off = max(abs(off - data_end), abs(data_off - end)) + if abs_off < OFF_BOUND: + data[OFF_IDX] = min(data_off, off) + data[LEN_IDX] = max(data_end, end) - data[OFF_IDX] + continue + self.data_list.append([filename, off, length, numaid]) + file_map_index[(filename, numaid)] = len(self.data_list) - 1 + + +def main(agrv): + """ + Parse the argument + """ + parser = argparse.ArgumentParser('model trace parser') + parser.add_argument('--trace_file', + required=True, + type=str, + help='trace source') + parser.add_argument('--output_file', + required=True, + type=str, + help='output of hint file') + try: + args = parser.parse_args() + trace_file = args.trace_file + output_file = args.output_file + if not os.path.exists(trace_file): + print("{trace_file} not exist") + return -1 + + hint_generator=ReplayHintGenerator(trace_file, output_file) + hint_generator.parse_trace() + hint_generator.generate_hint() + return 0 + except Exception as e: + print(f"Parse model trace exception:{str(e)}") + return -1 + +if __name__ == '__main__': + try: + ret = main(sys.argv[1:]) + except Exception as main_e: + print(str(main_e)) + ret = -1 + sys.exit(ret) + -- 2.34.1

2 1

[PATCH OLK-5.10] fbdev: Add bounds checking in bit_putcs to fix vmalloc-out-of-bounds
by Long Li 16 Mar '26

16 Mar '26

From: Albin Babu Varghese <albinbabuvarghese20(a)gmail.com> stable inclusion from stable-v5.10.246 commit f0982400648a3e00580253e0c48e991f34d2684c category: bugfix bugzilla: https://atomgit.com/src-openeuler/kernel/issues/12588 CVE: CVE-2025-40304 Reference: https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git/commit/?h=… -------------------------------- [ Upstream commit 3637d34b35b287ab830e66048841ace404382b67 ] Add bounds checking to prevent writes past framebuffer boundaries when rendering text near screen edges. Return early if the Y position is off-screen and clip image height to screen boundary. Break from the rendering loop if the X position is off-screen. When clipping image width to fit the screen, update the character count to match the clipped width to prevent buffer size mismatches. Without the character count update, bit_putcs_aligned and bit_putcs_unaligned receive mismatched parameters where the buffer is allocated for the clipped width but cnt reflects the original larger count, causing out-of-bounds writes. Reported-by: syzbot+48b0652a95834717f190(a)syzkaller.appspotmail.com Closes: https://syzkaller.appspot.com/bug?extid=48b0652a95834717f190 Suggested-by: Helge Deller <deller(a)gmx.de> Tested-by: syzbot+48b0652a95834717f190(a)syzkaller.appspotmail.com Signed-off-by: Albin Babu Varghese <albinbabuvarghese20(a)gmail.com> Signed-off-by: Helge Deller <deller(a)gmx.de> Signed-off-by: Sasha Levin <sashal(a)kernel.org> Signed-off-by: Long Li <leo.lilong(a)huawei.com> --- drivers/video/fbdev/core/bitblit.c | 17 +++++++++++++++++ 1 file changed, 17 insertions(+) diff --git a/drivers/video/fbdev/core/bitblit.c b/drivers/video/fbdev/core/bitblit.c index f22a68603a61..786af65bbf7b 100644 --- a/drivers/video/fbdev/core/bitblit.c +++ b/drivers/video/fbdev/core/bitblit.c @@ -169,6 +169,11 @@ static void bit_putcs(struct vc_data *vc, struct fb_info *info, image.height = vc->vc_font.height; image.depth = 1; + if (image.dy >= info->var.yres) + return; + + image.height = min(image.height, info->var.yres - image.dy); + if (attribute) { buf = kmalloc(cellsize, GFP_ATOMIC); if (!buf) @@ -182,6 +187,18 @@ static void bit_putcs(struct vc_data *vc, struct fb_info *info, cnt = count; image.width = vc->vc_font.width * cnt; + + if (image.dx >= info->var.xres) + break; + + if (image.dx + image.width > info->var.xres) { + image.width = info->var.xres - image.dx; + cnt = image.width / vc->vc_font.width; + if (cnt == 0) + break; + image.width = cnt * vc->vc_font.width; + } + pitch = DIV_ROUND_UP(image.width, 8) + scan_align; pitch &= ~scan_align; size = pitch * image.height + buf_align; -- 2.39.2

2 1

[PATCH openEuler-26.09] tools: Add maio-utils tool
by Hongbo Li 16 Mar '26

16 Mar '26

hulk inclusion category: feature bugzilla: https://atomgit.com/openeuler/kernel/issues/8613 -------------------------------- This is a userspace tool for MFS to provide: programmable strategy framework. Some basic strategies have added in test, and more flexible strategies can be written based on this framework. Signed-off-by: Hongbo Li <lihongbo22(a)huawei.com> --- tools/mfs/maio-utils/CMakeLists.txt | 101 ++++ tools/mfs/maio-utils/README.md | 17 + tools/mfs/maio-utils/build/build.sh | 91 +++ .../mfs/maio-utils/include/client/fs_client.h | 17 + tools/mfs/maio-utils/include/inner/inner.h | 37 ++ tools/mfs/maio-utils/include/loader/loader.h | 10 + tools/mfs/maio-utils/include/maio.h | 53 ++ .../maio-utils/include/parser/req_parser.h | 13 + tools/mfs/maio-utils/include/policy/policy.h | 17 + tools/mfs/maio-utils/include/uapi/mfs.h | 62 ++ tools/mfs/maio-utils/src/CMakeLists.txt | 42 ++ .../mfs/maio-utils/src/client/CMakeLists.txt | 17 + tools/mfs/maio-utils/src/client/fs_client.c | 359 ++++++++++++ .../mfs/maio-utils/src/loader/CMakeLists.txt | 18 + tools/mfs/maio-utils/src/loader/loader.c | 144 +++++ tools/mfs/maio-utils/src/main.c | 156 +++++ .../mfs/maio-utils/src/parser/CMakeLists.txt | 16 + tools/mfs/maio-utils/src/parser/req_parser.c | 313 ++++++++++ .../mfs/maio-utils/src/policy/CMakeLists.txt | 16 + tools/mfs/maio-utils/src/policy/policy.c | 114 ++++ tools/mfs/maio-utils/test/Makefile | 26 + tools/mfs/maio-utils/test/README.md | 61 ++ tools/mfs/maio-utils/test/strategy_demo.c | 45 ++ .../mfs/maio-utils/test/strategy_generate.cpp | 296 ++++++++++ .../mfs/maio-utils/test/strategy_prefetch.cpp | 150 +++++ tools/mfs/maio-utils/test/strategy_prefetch.h | 21 + tools/mfs/maio-utils/test/strategy_replay.cpp | 540 ++++++++++++++++++ tools/mfs/maio-utils/test/strategy_template.h | 21 + tools/mfs/maio-utils/test/strategy_toend.c | 46 ++ tools/mfs/maio-utils/test/trace_analyse.py | 87 +++ tools/mfs/maio-utils/test/trace_gen.py | 116 ++++ 31 files changed, 3022 insertions(+) create mode 100644 tools/mfs/maio-utils/CMakeLists.txt create mode 100644 tools/mfs/maio-utils/README.md create mode 100644 tools/mfs/maio-utils/build/build.sh create mode 100644 tools/mfs/maio-utils/include/client/fs_client.h create mode 100644 tools/mfs/maio-utils/include/inner/inner.h create mode 100644 tools/mfs/maio-utils/include/loader/loader.h create mode 100644 tools/mfs/maio-utils/include/maio.h create mode 100644 tools/mfs/maio-utils/include/parser/req_parser.h create mode 100644 tools/mfs/maio-utils/include/policy/policy.h create mode 100644 tools/mfs/maio-utils/include/uapi/mfs.h create mode 100644 tools/mfs/maio-utils/src/CMakeLists.txt create mode 100644 tools/mfs/maio-utils/src/client/CMakeLists.txt create mode 100644 tools/mfs/maio-utils/src/client/fs_client.c create mode 100644 tools/mfs/maio-utils/src/loader/CMakeLists.txt create mode 100644 tools/mfs/maio-utils/src/loader/loader.c create mode 100644 tools/mfs/maio-utils/src/main.c create mode 100644 tools/mfs/maio-utils/src/parser/CMakeLists.txt create mode 100644 tools/mfs/maio-utils/src/parser/req_parser.c create mode 100644 tools/mfs/maio-utils/src/policy/CMakeLists.txt create mode 100644 tools/mfs/maio-utils/src/policy/policy.c create mode 100644 tools/mfs/maio-utils/test/Makefile create mode 100644 tools/mfs/maio-utils/test/README.md create mode 100644 tools/mfs/maio-utils/test/strategy_demo.c create mode 100644 tools/mfs/maio-utils/test/strategy_generate.cpp create mode 100644 tools/mfs/maio-utils/test/strategy_prefetch.cpp create mode 100644 tools/mfs/maio-utils/test/strategy_prefetch.h create mode 100644 tools/mfs/maio-utils/test/strategy_replay.cpp create mode 100644 tools/mfs/maio-utils/test/strategy_template.h create mode 100644 tools/mfs/maio-utils/test/strategy_toend.c create mode 100644 tools/mfs/maio-utils/test/trace_analyse.py create mode 100644 tools/mfs/maio-utils/test/trace_gen.py diff --git a/tools/mfs/maio-utils/CMakeLists.txt b/tools/mfs/maio-utils/CMakeLists.txt new file mode 100644 index 000000000000..f710e6c15f79 --- /dev/null +++ b/tools/mfs/maio-utils/CMakeLists.txt @@ -0,0 +1,101 @@ +#******************************************************************************* +#******************************************************************************* + +cmake_minimum_required(VERSION 3.0) +cmake_policy(SET CMP0057 NEW) + +project(maio) + +#******************************************************************************* +#******************************************************************************* + +add_definitions(-D__LINUX_USR__) +add_definitions(-D__USE_GNU) +add_definitions(-D__POSIX_SOURCE) +add_definitions(-D_GNU_SOURCE) +add_definitions(-D_FILE_OFFSET_BITS=64) + +#******************************************************************************* +#******************************************************************************* + +include(TestBigEndian) + +TEST_BIG_ENDIAN(BIGENDIAN) + if (${BIGENDIAN}) + message(">>> Big Endian") + add_definitions(-D_CMAKE_BIGENDIAN) + else() + message(">>> Little Endian") + endif(${BIGENDIAN}) + +#******************************************************************************* +#******************************************************************************* + +EXECUTE_PROCESS(COMMAND uname -m COMMAND tr -d '\n' OUTPUT_VARIABLE ARCH) +if(${ARCH} STREQUAL "x86_64" OR ${ARCH} STREQUAL "aarch64") + message(STATUS "Architecture: ${ARCH}") +else() + message(STATUS "Invaild Architecture: ${ARCH}") + return() +endif() + +#******************************************************************************* +#******************************************************************************* + +set(BASE_FLAGS "-g -Wall -Werror -fno-strict-overflow -rdynamic") +set(SP_C_FLAGS "-std=gnu99") +set(SP_C++_FLAGS "-std=c++11") +set(SP_RELEASE_FLAGS "-O2") +set(SP_DEBUG_FLAGS "-O0") +set(SEC_REQ_FLAGS "-fstack-protector-all -fPIE -pie -fPIC -Wl,-z,relro,-z,now,-z,noexecstack") +set(SEC_SUG_BASE_FLAGS "-Wformat=2") +set(SEC_SUG_SP_C_FLAGS "-Wshadow") + +set(C_FLAGS "${BASE_FLAGS} ${SP_C_FLAGS}") +set(C++_FLAGS "${BASE_FLAGS} ${SP_C++_FLAGS}") +set(SEC_C_FLAGS "${SEC_REQ_FLAGS} ${SEC_SUG_BASE_FLAGS} ${SEC_SUG_SP_C_FLAGS}") +set(SEC_C++_FLAGS "${SEC_REQ_FLAGS} ${SEC_SUG_BASE_FLAGS}") +set(DEPC_FLAGS "-Wno-error=deprecated-declarations -Wno-deprecated-declarations") + +if(CMAKE_BUILD_TYPE MATCHES "Release") + message(">>> Release Mode") + set(CMAKE_C_FLAGS "${C_FLAGS} ${SP_RELEASE_FLAGS} ${SEC_C_FLAGS}") + set(CMAKE_CXX_FLAGS "${C++_FLAGS} ${SP_RELEASE_FLAGS} ${SEC_C++_FLAGS} ${DEPC_FLAGS}") +else() + message(">>> Debug Mode") + add_definitions(-D_DEBUG) + set(CMAKE_C_FLAGS "${C_FLAGS} ${SP_DEBUG_FLAGS} ${SEC_C_FLAGS}") + set(CMAKE_CXX_FLAGS "${C++_FLAGS} ${SP_DEBUG_FLAGS} ${SEC_C++_FLAGS} ${DEPC_FLAGS}") + set(COVERAGE_FLAGS "-coverage") + set(COVERAGE_LINK "gcov") +endif() + +message(STATUS ">>> Source Dir: " ${PROJECT_SOURCE_DIR}) +message(STATUS ">>> Binary Dir: " ${PROJECT_BINARY_DIR}) +message(STATUS ">>> C_FLAGS : " ${CMAKE_C_FLAGS}) +message(STATUS ">>> CXX_FLAGS : " ${CMAKE_CXX_FLAGS}) + +#******************************************************************************* +#******************************************************************************* + +set(EXECUTABLE_OUTPUT_PATH ${PROJECT_BINARY_DIR}/bin) + +#******************************************************************************* +#******************************************************************************* + +if(DEFINED THIRD_PARTY) + message(">>> third_party: ${THIRD_PARTY}") +else() + set(THIRD_PARTY ${PROJECT_SOURCE_DIR}/lib) +endif() + +set(INFRA_DIR ${PROJECT_SOURCE_DIR}/infrastructure) +set(SECUREC_DIR ${THIRD_PARTY}/securec) +set(ZLOG_DIR ${THIRD_PARTY}/zlog) +set(LIBNUMA_DIR ${THIRD_PARTY}/libnuma) + +#******************************************************************************* +#******************************************************************************* + +add_subdirectory(infrastructure) +add_subdirectory(src) diff --git a/tools/mfs/maio-utils/README.md b/tools/mfs/maio-utils/README.md new file mode 100644 index 000000000000..4c5a66a83c60 --- /dev/null +++ b/tools/mfs/maio-utils/README.md @@ -0,0 +1,17 @@ +# maio-utils + +#### How to build + +1. prepare thirdparty by using: +``` +git clone -b threadpool_bind https://github.com/hb-lee/infrastructure.git infrastructure +git clone -b 1.2.18 https://github.com/HardySimpson/zlog.git thirdparty/zlog +git clone https://github.com/chriszt/securec.git thirdparty/securec +git clone -b v2.0.19 https://github.com/numactl/numactl.git thirdparty/libnuma +``` +2. cd build +3. bash build.sh prepare # only once. +4. bash build.sh build + +#### How to run +Run with `./maio -m ${MNTPOINT}`, here ${MNTPOINT} means the mfs mount point. If you want to use some strategies, you can run with `./maio -m ${MNTPOINT} -s ${STRATEGYLIB}` where ${STRATEGYLIB} is your strategy library. Some examples are provided in `test` directory. diff --git a/tools/mfs/maio-utils/build/build.sh b/tools/mfs/maio-utils/build/build.sh new file mode 100644 index 000000000000..91b49d3d6bf8 --- /dev/null +++ b/tools/mfs/maio-utils/build/build.sh @@ -0,0 +1,91 @@ +#!/bin/bash + +LOCAL_PATH="$(readlink -e "$(dirname "$0")")" +PRJ_PATH="$(readlink -e "${LOCAL_PATH}/..")" +THIRD_SUB="${PRJ_PATH}/thirdparty" +Lib_PATH="${PRJ_PATH}/lib" + +ARCH=$(uname -m) + +function build_zlog() { + local zlog_dir=${THIRD_SUB}/zlog + local build_path=${zlog_dir}/build + mkdir ${build_path} + mkdir -p ${Lib_PATH}/zlog + cd ${zlog_dir} + make PREFIX=`pwd`/build/zlog + make PREFIX=`pwd`/build/zlog install + cp -rdp ${build_path}/zlog/include ${Lib_PATH}/zlog + cp -rdp ${build_path}/zlog/lib ${Lib_PATH}/zlog + cd - +} + +function build_securec() { + local securec_dir=${THIRD_SUB}/securec + local securec_lib=${securec_dir}/build/securec/lib + mkdir -p ${securec_lib} + mkdir -p ${Lib_PATH}/securec + cd ${securec_dir} + if [ "${ARCH}" = "aarch64" ]; then + make -f aarch64-so.mk + else + make -f x86-so.mk + fi + cp -rdp include build/securec + cp libsecurec.so build/securec/lib + cp -rdp build/securec/include ${Lib_PATH}/securec + cp -rdp build/securec/lib ${Lib_PATH}/securec + cd - +} + +function build_libnuma() { + local libnuma_dir=${THIRD_SUB}/libnuma + local build_path=${libnuma_dir}/build + mkdir ${build_path} + mkdir -p ${Lib_PATH}/libnuma + cd ${libnuma_dir} + ./autogen.sh + ./configure --prefix=`pwd`/build + make -j4 + make install + cp -rdp build/include ${Lib_PATH}/libnuma + cp -rdp build/lib ${Lib_PATH}/libnuma + cd - +} + +function prepare_thirdparty() { + build_jsoncpp + build_zlog + build_securec + build_libnuma +} + +function build_program() { + local BUILD_TYPE=$1 + + set -x + if [ "${BUILD_TYPE}" == "release" ]; then + cmake .. -DCMAKE_BUILD_TYPE=Release + else + cmake .. -DCMAKE_BUILD_TYPE=Debug + fi + set +x + + make clean + make VERBOSE=1 -j +} + +function main() { + local ACTION="$1" + local PARAMETER="$2" + + if [ "${ACTION}" == "prepare" ]; then + prepare_thirdparty + fi + if [ "$ACTION" == "build" ]; then + build_program $PARAMETER + fi +} + +main $@ +exit $? diff --git a/tools/mfs/maio-utils/include/client/fs_client.h b/tools/mfs/maio-utils/include/client/fs_client.h new file mode 100644 index 000000000000..803179ade14c --- /dev/null +++ b/tools/mfs/maio-utils/include/client/fs_client.h @@ -0,0 +1,17 @@ +// SPDX-License-Identifier: GPL-2.0 +/* Copyright(c) 2025 Huawei Technologies Co., Ltd + */ +#ifndef _FS_CLIENT_H_ +#define _FS_CLIENT_H_ + +#include <stdint.h> + +int fs_evict_by_path(const char*path, uint64_t off, uint64_t len); +int fs_load_by_path(const char *path, uint64_t off, uint64_t len); +int fs_load_by_fd(int fd, uint64_t off, uint64_t len); +int fs_sync_by_path(const char *path, void *buf, uint64_t off, uint64_t len); + +int fs_client_init(const char *source); +void fs_client_exit(); + +#endif diff --git a/tools/mfs/maio-utils/include/inner/inner.h b/tools/mfs/maio-utils/include/inner/inner.h new file mode 100644 index 000000000000..c10f40a3712c --- /dev/null +++ b/tools/mfs/maio-utils/include/inner/inner.h @@ -0,0 +1,37 @@ +// SPDX-License-Identifier: GPL-2.0 +/* Copyright(c) 2025 Huawei Technologies Co., Ltd + */ +#ifndef _INNER_H_ +#define _INNER_H_ + +#include <time.h> +#include <stdint.h> + +static inline uint64_t _get_ts() +{ + struct timespec now = {0}; + (void)clock_gettime(CLOCK_MONOTONIC_COARSE, &now); + return (uint64_t)now.tv_sec; +} + +/* io context for loader and writeback */ +struct io_context { + uint32_t id; /* msg id */ + uint64_t off; /* io offset in fd */ + uint64_t len; /* io length */ + uint16_t numaid; /* numa node of this io, -1 means non-bind */ + int fd; /* io file handle */ + char *path; /* fullpath of this io */ + void *buf; /* flighting buffer for writeback */ + + uint64_t seq; + + void *private; + void (*end_io)(int ret, void *private); +}; + +void loader_update_curseq(uint64_t seq); +int loader_evict_submit(struct io_context *ioctx); +int loader_io_submit(struct io_context *ioctx); + +#endif diff --git a/tools/mfs/maio-utils/include/loader/loader.h b/tools/mfs/maio-utils/include/loader/loader.h new file mode 100644 index 000000000000..ff280989f9f9 --- /dev/null +++ b/tools/mfs/maio-utils/include/loader/loader.h @@ -0,0 +1,10 @@ +// SPDX-License-Identifier: GPL-2.0 +/* Copyright(c) 2025 Huawei Technologies Co., Ltd + */ +#ifndef _LOADER_H_ +#define _LOADER_H_ + +int loader_init(); +void loader_exit(); + +#endif diff --git a/tools/mfs/maio-utils/include/maio.h b/tools/mfs/maio-utils/include/maio.h new file mode 100644 index 000000000000..a508d5b1fa27 --- /dev/null +++ b/tools/mfs/maio-utils/include/maio.h @@ -0,0 +1,53 @@ +// SPDX-License-Identifier: GPL-2.0 +/* Copyright(c) 2025 Huawei Technologies Co., Ltd + */ +#ifndef _MAIO_H_ +#define _MAIO_H_ + +#ifdef __cplusplus +extern "C" { +#endif + +#include <stdint.h> + +struct maio_entry { + char *fpath; + uint64_t toff; + uint64_t tlen; + uint8_t tnuma; + uint64_t seq; +}; + +struct maio { + /* IN */ + int fd; + uint64_t off; + uint64_t len; + int pid; + uint8_t op; + uint64_t ts; + uint32_t cksz; + + /* OUT */ + uint16_t flags; +#define MAIO_WITH_SEQ 0x0001 + uint64_t curseq; + struct maio_entry entries[]; +}; + +struct maio_operation { + int max_io; + int (*init) (void); + void (*exit) (void); + int (*load) (struct maio **io); + int (*evict) (struct maio **io); +}; + +char *_get_fullpath(char *fpath, int fd); +void maio_preload(struct maio *maio, int nio); + +#ifdef __cplusplus +} +#endif + +#endif diff --git a/tools/mfs/maio-utils/include/parser/req_parser.h b/tools/mfs/maio-utils/include/parser/req_parser.h new file mode 100644 index 000000000000..908e7297558a --- /dev/null +++ b/tools/mfs/maio-utils/include/parser/req_parser.h @@ -0,0 +1,13 @@ +// SPDX-License-Identifier: GPL-2.0 +/* Copyright(c) 2025 Huawei Technologies Co., Ltd + */ +#ifndef __REQ_PARSER_H__ +#define __REQ_PARSER_H__ + +#include <stdint.h> + +int parser_init(uint32_t fs_mode, const char *mntpoint); +void parser_destory(); +int maio_parse_req(void *buf, int size); + +#endif diff --git a/tools/mfs/maio-utils/include/policy/policy.h b/tools/mfs/maio-utils/include/policy/policy.h new file mode 100644 index 000000000000..aef05c956b3b --- /dev/null +++ b/tools/mfs/maio-utils/include/policy/policy.h @@ -0,0 +1,17 @@ +// SPDX-License-Identifier: GPL-2.0 +/* Copyright(c) 2025 Huawei Technologies Co., Ltd + */ +#ifndef _POLICY_H_ +#define _POLICY_H_ + +#include "maio.h" + +int policy_max_io(); +int policy_load(struct maio **io); +int policy_evict(struct maio **io); +int policy_register(const char *path); +void policy_unregister(); +int policy_init(); +void policy_exit(); + +#endif diff --git a/tools/mfs/maio-utils/include/uapi/mfs.h b/tools/mfs/maio-utils/include/uapi/mfs.h new file mode 100644 index 000000000000..a7d5882b5e50 --- /dev/null +++ b/tools/mfs/maio-utils/include/uapi/mfs.h @@ -0,0 +1,62 @@ +/* SPDX-License-Identifier: GPL-2.0 WITH Linux-syscall-note */ + +#ifndef _UAPI_LINUX_MFS_H +#define _UAPI_LINUX_MFS_H + +#include <linux/types.h> +#include <linux/ioctl.h> + +enum mfs_opcode { + MFS_OP_READ = 0, + MFS_OP_FAULT, + MFS_OP_FAROUND, +}; + +enum { + MFS_MODE_NONE = 0, + MFS_MODE_LOCAL, + MFS_MODE_REMOTE, +}; + +struct mfs_ioc_ra { + __u64 off; + __u64 len; +}; + +struct mfs_ioc_done { + __u32 id; + __u32 ret; +}; + +struct mfs_ioc_rpath { + __u16 max; + __u16 len; + __u8 d[]; +}; + +#define MFS_IOC_RA _IOW(0xbc, 1, struct mfs_ioc_ra) +#define MFS_IOC_DONE _IOW(0xbc, 2, struct mfs_ioc_done) +#define MFS_IOC_RPATH _IOWR(0xbc, 3, struct mfs_ioc_rpath) + +struct mfs_ioc_fsinfo { + __u8 mode; /* 0: none, 1: local, 2: remote */ +}; + +#define MFS_IOC_FSINFO _IOR(0xbd, 1, struct mfs_ioc_fsinfo) + +struct mfs_msg { + __u8 version; + __u8 opcode; + __u16 len; + __u32 fd; + __u32 id; + __u8 data[]; +}; + +struct mfs_read { + __u64 off; + __u64 len; + __s32 pid; +}; + +#endif /* _UAPI_LINUX_MFS_H */ diff --git a/tools/mfs/maio-utils/src/CMakeLists.txt b/tools/mfs/maio-utils/src/CMakeLists.txt new file mode 100644 index 000000000000..fd053a8bd081 --- /dev/null +++ b/tools/mfs/maio-utils/src/CMakeLists.txt @@ -0,0 +1,42 @@ +#******************************************************************************* +#******************************************************************************* + +include_directories(${PROJECT_SOURCE_DIR}/include) +include_directories(${PROJECT_SOURCE_DIR}/include/inner) +include_directories(${PROJECT_SOURCE_DIR}/include/uapi) +include_directories(${PROJECT_SOURCE_DIR}/include/client) +include_directories(${PROJECT_SOURCE_DIR}/include/loader) +include_directories(${PROJECT_SOURCE_DIR}/include/policy) +include_directories(${PROJECT_SOURCE_DIR}/include/monitor) +include_directories(${PROJECT_SOURCE_DIR}/include/parser) + +include_directories(${INFRA_DIR}/include) +include_directories(${SECUREC_DIR}/include) +include_directories(${ZLOG_DIR}/include) +include_directories(${LIBNUMA_DIR}/include) + +#******************************************************************************* +#******************************************************************************* + +link_directories(${PROJECT_BINARY_DIR}/lib) +link_directories(${SECUREC_DIR}/lib) +link_directories(${ZLOG_DIR}/lib) +link_directories(${LIBNUMA_DIR}/lib) + +#******************************************************************************* +#******************************************************************************* + +add_subdirectory(client) +add_subdirectory(loader) +add_subdirectory(policy) +add_subdirectory(parser) + +add_executable(maio-utils main.c) +target_link_libraries(maio-utils "-Wl,--whole-archive" + client + loader + policy + parser + "-Wl,--no-whole-archive" + infra + dl) diff --git a/tools/mfs/maio-utils/src/client/CMakeLists.txt b/tools/mfs/maio-utils/src/client/CMakeLists.txt new file mode 100644 index 000000000000..0fb67d64eb03 --- /dev/null +++ b/tools/mfs/maio-utils/src/client/CMakeLists.txt @@ -0,0 +1,17 @@ +#******************************************************************************* +#******************************************************************************* + +include_directories(${CMAKE_CURRENT_SOURCE_DIR}) + +#******************************************************************************* +#******************************************************************************* + +#aux_source_directory(. CLIENT_SRC) +set(CLIENT_SRC fs_client.c) + +#******************************************************************************* +#******************************************************************************* + +add_library(client ${CLIENT_SRC}) +target_link_libraries(client infra + securec) diff --git a/tools/mfs/maio-utils/src/client/fs_client.c b/tools/mfs/maio-utils/src/client/fs_client.c new file mode 100644 index 000000000000..5d359b8f22e1 --- /dev/null +++ b/tools/mfs/maio-utils/src/client/fs_client.c @@ -0,0 +1,359 @@ +// SPDX-License-Identifier: GPL-2.0 +/* Copyright(c) 2025 Huawei Technologies Co., Ltd + */ +#include "fs_client.h" +#include "mfs.h" +#include "inner.h" +#include "hashmap.h" +#include "hashfunc.h" +#include "stimer.h" +#include "atomic.h" +#include "log.h" +#include "list.h" +#include "spinlock.h" +#include "sysdef.h" +#include "securec.h" + +#include <fcntl.h> +#include <errno.h> +#include <string.h> +#include <unistd.h> +#include <sys/ioctl.h> + +#define TIMER_CYCLE 10000 /* 10s */ +#define FD_TMOUT 3600 /* 60s */ + +struct fs_client { + int rootfd; + hashmap_t *map; + stimer_t *timer; + list_head_t head; + spinlock_t lock; + uint64_t tmout; +}; + +struct fs_client g_client = {0}; + +typedef struct { + uint32_t plen; + char *path; +} fdkey_t; + +typedef struct { + fdkey_t key; + int fd; + int ref; + uint64_t ts; + list_head_t link; + hashlink_t hash; +} fd_t; + +static inline int fd_cmp(void *first, void *second) +{ + fdkey_t *key1 = (fdkey_t *)first; + fdkey_t *key2 = (fdkey_t *)second; + + if (key1->plen != key2->plen) + return -1; + + return strcmp(key1->path, key2->path); +} + +static inline uint32_t fd_hash(void *args) +{ + fdkey_t *key = (fdkey_t *)args; + uint64_t hash = hashstr(key->path, key->plen); + + return (uint32_t)hash; +} + +static inline void fd_inc(void *args, hashlink_t *link) +{ + fd_t *f = container_of(link, fd_t, hash); + int ref = atomic_s32_inc(&f->ref); + if (ref <= 0) { + log_error("fd(%d) for path(%s) with invalid ref(%d)", + f->fd, f->key.path, ref); + sys_assert(0); + } +} + +static inline int fd_dec(void *args, hashlink_t *link) +{ + fd_t *f = container_of(link, fd_t, hash); + int ref = atomic_s32_dec(&f->ref); + + if (ref < 0) { + log_error("fd(%d) for path(%s) with invalid ref(%d)", + f->fd, f->key.path, ref); + sys_assert(0); + } + /* update access timestamp for the last one */ + if (ref == 0) + f->ts = _get_ts(); + return -1; +} + +static inline int fd_expire(void *args, hashlink_t *link) +{ + fd_t *f = container_of(link, fd_t, hash); + int ref = atomic_s32_fetch(&f->ref); + uint64_t cur; + + if (ref > 0) + return -1; + if (ref < 0) { + log_error("fd(%d) for path(%s) with invalid ref(%d)", + f->fd, f->key.path, ref); + sys_assert(0); + } + cur = _get_ts(); + return ((cur - f->ts) >= g_client.tmout) ? 0: -1; +} + +static fd_t *fd_alloc(const char *path) +{ + fd_t *f = (fd_t *)malloc(sizeof(fd_t)); + if (!f) { + log_error("alloc fd_t fail"); + return NULL; + } + + f->key.path = strdup(path); + if (!f->key.path) { + log_error("strdup(%s) fail", path); + free(f); + return NULL; + } + + f->key.plen = strlen(path) + 1; + f->fd = -1; + f->ref = 1; + f->hash.key = &f->key; + list_init(&f->link); + return f; +} + +static void fd_free(fd_t *f) +{ + if (f->fd > 0) { + int ret = close(f->fd); + if (ret != 0) { + log_error("close(%d) fail, err(%s)", f->fd, strerror(errno)); + } + f->fd = -1; + } + + if (f->key.path) { + free(f->key.path); + f->key.path = NULL; + } + list_del(&f->link); + free(f); +} + +static int fd_fetch(const char *path, fd_t **f_res, int dirfd) +{ + fdkey_t key; + key.plen = strlen(path) + 1; + key.path = (char *)path; + + /* fd in cache */ + hashlink_t *data = NULL; + if (EEXIST == hashmap_search(g_client.map, &key, &data, NULL, fd_inc)) { + *f_res = container_of(data, fd_t, hash); + (*f_res)->ts = _get_ts(); + return 0; + } + + /* new fd */ + fd_t *f = fd_alloc(path); + if (!f) + return -ENOMEM; + + /* open file */ + if (dirfd > 0) + f->fd = openat(dirfd, path, O_RDONLY, S_IRUSR | S_IRGRP); + else + f->fd = open(path, O_RDONLY, S_IRUSR | S_IRGRP); + if (f->fd < 0) { + if (errno != ENOENT) + log_error("open(%d,%s) fail, err(%s)", dirfd, path, strerror(errno)); + fd_free(f); + return errno; + } + + /* add to fd cache */ + hashlink_t *old = NULL; + if (EEXIST == hashmap_insert(g_client.map, &f->hash, &old, NULL, fd_inc)) { + fd_free(f); + f = container_of(old, fd_t, hash); + } + + spinlock_lock(&g_client.lock); + list_add_tail(&f->link, &g_client.head); + spinlock_unlock(&g_client.lock); + f->ts = _get_ts(); + *f_res = f; + return 0; +} + +static inline void fd_restore(fd_t *f) +{ + (void)hashmap_protect(g_client.map, &f->key, NULL, fd_dec); +} + +static void _fd_gc(void *args) +{ + list_head_t head, *curr, *next; + hashlink_t *data = NULL; + + list_init(&head); + spinlock_lock(&g_client.lock); + list_splice(&g_client.head, &head); + spinlock_unlock(&g_client.lock); + + list_foreach_safe(curr, next, &head) + { + fd_t *f = container_of(curr, fd_t, link); + if (atomic_s32_fetch(&f->ref) != 0) + continue; + /* try to destroy */ + if (0 != hashmap_delete(g_client.map, &f->key, &data, NULL, fd_expire)) + continue; + list_del(&f->link); + log_info("path:%s with fd:%d is expire.", f->key.path, f->fd); + fd_free(f); + } + + spinlock_lock(&g_client.lock); + list_splice(&head, &g_client.head); + spinlock_unlock(&g_client.lock); +} + +int fs_evict_by_path(const char*path, uint64_t off, uint64_t len) +{ + int ret; + fd_t *f; + uint64_t step = 4096 * 4; + + ret = fd_fetch(path, &f, 0); + if (ret) + return ret; + + while (len) { + if (step > len) + step = len; + ret = posix_fadvise(f->fd, off, len, POSIX_FADV_DONTNEED); + if (ret < 0) + break; + len -= step; + off += step; + } + + fd_restore(f); + return ret; +} + +int fs_load_by_path(const char *path, uint64_t off, uint64_t len) +{ + int ret; + fd_t *f; + uint64_t step = 16 * 4096; + char *buf = NULL; + + ret = fd_fetch(path, &f, 0); + if (ret) + return ret; + + buf = (char *)malloc(step); + + while (len) { + if (step > len) + step = len; + ret = pread(f->fd, buf, step, off); + if (ret <= 0) + break; + len -= step; + off += step; + } + free(buf); + fd_restore(f); + return ret; +} + +int fs_load_by_fd(int fd, uint64_t off, uint64_t len) +{ + struct mfs_ioc_ra ra; + + ra.off = off; + ra.len = len; + return ioctl(fd, MFS_IOC_RA, &ra); +} + +int fs_sync_by_path(const char *path, void *buf, uint64_t off, uint64_t len) +{ + ssize_t rsize = 0, ret; + uint64_t pos = 0; + fd_t *f; + + ret = fd_fetch(path + 1, &f, g_client.rootfd); + if (ret) + return ret; + + do { + ret = pread(f->fd, buf + pos, len - pos, off + pos); + if (ret == 0) + break; + if (ret < 0) { + log_error("failed to read:%s", strerror(errno)); + fd_restore(f); + return -1; + } + pos += ret; + rsize += ret; + } while (rsize < len); + fd_restore(f); + return 0; +} + +int fs_client_init(const char *source) +{ + int ret; + + g_client.rootfd = -1; + if (source) { + g_client.rootfd = open(source, O_RDONLY | O_DIRECTORY); + if (g_client.rootfd <= 0) { + log_error("open %s failed", source); + return -1; + } + } + ret = hashmap_create(1024, fd_cmp, fd_hash, &g_client.map); + if (ret) { + log_error("client fd map alloc failed"); + close(g_client.rootfd); + return ret; + } + + g_client.timer = stimer_create("LfsGc", TIMER_CYCLE, NULL, _fd_gc); + if (!g_client.timer) { + log_error("timer create failed"); + hashmap_destroy(g_client.map, NULL, NULL); + close(g_client.rootfd); + return -1; + } + list_init(&g_client.head); + spinlock_init(&g_client.lock); + g_client.tmout = FD_TMOUT; + return 0; +} + +void fs_client_exit() +{ + spinlock_destroy(&g_client.lock); + stimer_destroy(g_client.timer); + hashmap_destroy(g_client.map, NULL, NULL); + close(g_client.rootfd); +} + diff --git a/tools/mfs/maio-utils/src/loader/CMakeLists.txt b/tools/mfs/maio-utils/src/loader/CMakeLists.txt new file mode 100644 index 000000000000..08e1b637be26 --- /dev/null +++ b/tools/mfs/maio-utils/src/loader/CMakeLists.txt @@ -0,0 +1,18 @@ +#******************************************************************************* +#******************************************************************************* + +include_directories(${CMAKE_CURRENT_SOURCE_DIR}) + +#******************************************************************************* +#******************************************************************************* + +aux_source_directory(. LOADER_SRC) + +#******************************************************************************* +#******************************************************************************* + +add_library(loader ${LOADER_SRC}) +target_link_libraries(loader infra + numa + securec) + diff --git a/tools/mfs/maio-utils/src/loader/loader.c b/tools/mfs/maio-utils/src/loader/loader.c new file mode 100644 index 000000000000..61c5302204fc --- /dev/null +++ b/tools/mfs/maio-utils/src/loader/loader.c @@ -0,0 +1,144 @@ +// SPDX-License-Identifier: GPL-2.0 +/* Copyright(c) 2025 Huawei Technologies Co., Ltd + */ +#include "loader.h" + +#include "fs_client.h" +#include "threadpool.h" +#include "log.h" +#include "atomic.h" +#include "inner.h" +#include "sysdef.h" + +#include "securec.h" +#include "numa.h" +#include <errno.h> + +struct loader_mgr { + int numa_num; + threadpool_t **numa_thp; + uint64_t curseq; +}; + +struct loader_mgr g_local_mgr; + +static threadpool_t *choose_worker_pool(struct io_context *ioctx) +{ + static uint64_t cid = 0; + int nid; + + if (ioctx->numaid == (uint16_t)-1) + nid = cid++ % g_local_mgr.numa_num; + else + nid = ioctx->numaid % g_local_mgr.numa_num; + return g_local_mgr.numa_thp[nid]; +} + +static void handle_io(void *args) +{ + struct io_context *ioctx = (struct io_context *)args; + int ret; + + if (g_local_mgr.curseq > ioctx->seq) { + if (ioctx->end_io) + ioctx->end_io(0, ioctx->private); + return; + } + ret = ioctx->path ? fs_load_by_path(ioctx->path, ioctx->off, ioctx->len) + : fs_load_by_fd(ioctx->fd, ioctx->off, ioctx->len); + if (ioctx->end_io) + ioctx->end_io(ret, ioctx->private); +} + +int loader_io_submit(struct io_context *ioctx) +{ + threadpool_t *worker_pools = choose_worker_pool(ioctx); + + threadpool_submit(worker_pools, ioctx, handle_io); + return 0; +} + +void loader_update_curseq(uint64_t seq) +{ + uint64_t curseq = atomic_u64_fetch(&g_local_mgr.curseq); + uint64_t old; + + while (seq > curseq) { + if (atomic_u64_cas(&g_local_mgr.curseq, curseq, seq, &old)) + break; + curseq = old; + } +} + +static void handle_evict(void *args) +{ + struct io_context *ioctx = (struct io_context *)args; + int ret; + + ret = ioctx->path ? fs_evict_by_path(ioctx->path, ioctx->off, ioctx->len) : 0; + if (ioctx->end_io) + ioctx->end_io(ret, ioctx->private); +} + +int loader_evict_submit(struct io_context *ioctx) +{ + threadpool_t *worker_pools = choose_worker_pool(ioctx); + + threadpool_submit(worker_pools, ioctx, handle_evict); + return 0; +} + +static void _free_loader() +{ + int i; + + if (!g_local_mgr.numa_thp) + return; + + for (i = 0; i < g_local_mgr.numa_num; i++) + threadpool_destroy(g_local_mgr.numa_thp[i]); + free(g_local_mgr.numa_thp); + g_local_mgr.numa_thp = NULL; +} + +int loader_init() +{ + int i, ret, wnum = 8; + char name[THD_NAME + 1] = {0}; + + if (numa_available() < 0) { + log_error("Current system does not support NUAM api"); + return -1; + } + + g_local_mgr.numa_num = numa_max_node() + 1; + g_local_mgr.curseq = 0; + ret = 0; + do { + g_local_mgr.numa_thp = + calloc(g_local_mgr.numa_num, sizeof(threadpool_t *)); + if (!g_local_mgr.numa_thp) { + log_error("alloc for numa threadpool failed"); + ret = -1; + break; + } + for (i = 0; i < g_local_mgr.numa_num; i++) { + sprintf_s(name, sizeof(name), "Loader-%d", i); + g_local_mgr.numa_thp[i] = threadpool_create(name, wnum, i); + if (!g_local_mgr.numa_thp[i]) { + log_error("alloc Loader-%d failed", i); + ret = -1; + break; + } + } + } while (0); + + if (ret != 0) + _free_loader(); + return ret; +} + +void loader_exit() +{ + _free_loader(); +} diff --git a/tools/mfs/maio-utils/src/main.c b/tools/mfs/maio-utils/src/main.c new file mode 100644 index 000000000000..854c5b298521 --- /dev/null +++ b/tools/mfs/maio-utils/src/main.c @@ -0,0 +1,156 @@ +// SPDX-License-Identifier: GPL-2.0 +/* Copyright(c) 2025 Huawei Technologies Co., Ltd + */ +#include <stdlib.h> +#include <unistd.h> +#include <string.h> +#include <getopt.h> +#include <errno.h> +#include <fcntl.h> +#include <poll.h> + +#include "loader.h" +#include "log.h" +#include "maio.h" +#include "policy.h" +#include "req_parser.h" +#include "fs_client.h" +#include "mfs.h" +#include "securec.h" + +#include <sys/ioctl.h> +#include <sys/statfs.h> + +#define MAX_BUF_SIZE 1024 +#define MFS_SUPER_MAGIC 0x85428370 + +static int read_req(int fd) +{ + char *buf; + int ret; + + buf = calloc(MAX_BUF_SIZE, sizeof(char)); + if (!buf) { + log_error("failed to alloc buf\n"); + return -1; + } + ret = read(fd, buf, MAX_BUF_SIZE); + if (ret <= 0) { + if (ret < 0) + log_error("failed to read, ret:%d\n", ret); + return -1; + } + return maio_parse_req(buf, ret); +} + +static void dev_monitor(int dfd) +{ + struct pollfd pfd; + int ret; + + pfd.fd = dfd; + pfd.events = POLLIN; + while (1) { + ret = poll(&pfd, 1, -1); + if (ret < 0) { + log_error("poll failed\n"); + return; + } + if (ret == 0 || !(pfd.revents & POLLIN)) { + log_error("poll events error, ret:%d, revents:%x\n", ret, pfd.revents); + continue; + } + while(!read_req(pfd.fd)) {} + } +} + +int main(int argc, char *argv[]) +{ + const char *mntpoint = NULL, *strategylib = NULL; + struct mfs_ioc_fsinfo fsinfo = {0}; + int dfd, ret, opt; + struct statfs buf; + char devname[10]; + + while ((opt = getopt(argc, argv, "m:s:")) != -1) { + switch (opt) { + case 'm': + mntpoint = optarg; + break; + case 's': + strategylib = optarg; + break; + default: + fprintf(stderr, "Usage: %s -m ${mnt} [-s ${strategylib}]\n", argv[0]); + return -EINVAL; + } + } + if (!mntpoint) { + fprintf(stderr, "mount point should specify\n"); + fprintf(stderr, "Usage: %s -m ${mnt} [-s ${strategylib}]\n", argv[0]); + return -1; + } + + ret = statfs(mntpoint, &buf); + if (ret) { + log_error("statfs %s failed, errstr:%s\n", mntpoint, strerror(errno)); + return -1; + } + if (buf.f_type != MFS_SUPER_MAGIC) { + log_error("fstype(%x) is invalid, please check the mountpoint", buf.f_type); + return -1; + } + + sprintf_s(devname, sizeof(devname), "/dev/mfs%ld", buf.f_spare[0]); + /* Open with O_CLOEXEC to avoid potential fd leaks */ + dfd = open(devname, O_RDWR | O_CLOEXEC); + if (dfd < 0) { + log_error("open %s failed errstr:%s\n", devname, strerror(errno)); + return -1; + } + + ret = ioctl(dfd, MFS_IOC_FSINFO, (unsigned long)&fsinfo); + if (ret < 0) { + log_error("ioctl get fsinfo failed, ret:%d\n", ret); + goto close_fd; + } + if (fsinfo.mode != 1) { + log_error("fs runing mode(%d) is not supported", fsinfo.mode); + goto close_fd; + } + ret = loader_init(); + if (ret) { + log_error("loader init failed"); + goto close_fd; + } + ret = fs_client_init(NULL); + if (ret) { + log_error("clients init failed"); + goto free_loader; + } + ret = parser_init(fsinfo.mode, mntpoint); + if (ret) { + log_error("parser init failed"); + goto free_client; + } + ret = policy_init(); + if (ret) { + log_error("policy init failed"); + goto free_parser; + } + ret = policy_register(strategylib); + if (ret) { + log_info("register new policy failed, use default policy"); + } + + dev_monitor(dfd); +free_parser: + parser_destory(); +free_client: + fs_client_exit(); +free_loader: + loader_exit(); +close_fd: + close(dfd); + return ret; +} diff --git a/tools/mfs/maio-utils/src/parser/CMakeLists.txt b/tools/mfs/maio-utils/src/parser/CMakeLists.txt new file mode 100644 index 000000000000..72e5849da8e4 --- /dev/null +++ b/tools/mfs/maio-utils/src/parser/CMakeLists.txt @@ -0,0 +1,16 @@ +#******************************************************************************* +#******************************************************************************* + +include_directories(${CMAKE_CURRENT_SOURCE_DIR}) + +#******************************************************************************* +#******************************************************************************* + +aux_source_directory(. PARSER_SRC) + +#******************************************************************************* +#******************************************************************************* + +add_library(parser ${PARSER_SRC}) +target_link_libraries(parser infra + securec) diff --git a/tools/mfs/maio-utils/src/parser/req_parser.c b/tools/mfs/maio-utils/src/parser/req_parser.c new file mode 100644 index 000000000000..83a975276bc2 --- /dev/null +++ b/tools/mfs/maio-utils/src/parser/req_parser.c @@ -0,0 +1,313 @@ +// SPDX-License-Identifier: GPL-2.0 +/* Copyright(c) 2025 Huawei Technologies Co., Ltd + */ + +#include "inner.h" +#include "log.h" +#include "mfs.h" +#include "policy.h" +#include "securec.h" +#include "threadpool.h" + +#include <stdbool.h> +#include <errno.h> +#include <sys/ioctl.h> +#include <sys/stat.h> + +struct parser_mgr { + uint32_t fs_mode; + uint32_t chunk_size; + const char *mntpoint; + threadpool_t *parser; +}; + +struct parser_mgr g_parser_mgr; + +#define PARSER_NUM 8 +#define MAX_PATH_SIZE 1024 + +#define __round_mask(x, y) ((__typeof__(x))((y)-1)) +#define round_up(x, y) ((((x)-1) | __round_mask(x, y))+1) +#define round_down(x, y) ((x) & ~__round_mask(x, y)) + +#define DIV_ROUND_UP(n, d) (((n) + (d) - 1) / (d)) +#define MIN(x, y) ((x) < (y) ? (x) : (y)) + +bool is_power_of_2(unsigned long n) +{ + return (n != 0 && ((n & (n - 1)) == 0)); +} + +void ctx_cb(int ret, void *args) +{ + struct io_context *ctx = (struct io_context *)args; + + if (ctx->path) + free(ctx->path); + if (ctx->buf) + free(ctx->buf); + free(ctx); +} + +void _submit_local_io(int fd, struct maio_entry *entry) +{ + struct io_context *ioctx; + + ioctx = calloc(1, sizeof(struct io_context)); + if (!ioctx) { + log_error("ioctx alloc failed"); + return; + } + ioctx->off = entry->toff; + ioctx->len = entry->tlen; + ioctx->numaid = entry->tnuma; + ioctx->fd = fd; + ioctx->path = entry->fpath; + ioctx->seq = entry->seq; + ioctx->private = ioctx; + ioctx->end_io = ctx_cb; + loader_io_submit(ioctx); +} + +char *get_file_path(int fd) +{ + struct mfs_ioc_rpath *rp; + char *path; + int ret; + + rp = malloc(sizeof(struct mfs_ioc_rpath) + MAX_PATH_SIZE); + if (!rp) { + log_error("rpath alloc failed"); + return NULL; + } + rp->max = MAX_PATH_SIZE; + ret = ioctl(fd, MFS_IOC_RPATH, (unsigned long)rp); + if (ret) { + log_error("realpath failed for fd:%d, ret:%d", fd, ret); + free(rp); + return NULL; + } + rp->d[rp->len] = '\0'; + + path = malloc(strlen(g_parser_mgr.mntpoint) + rp->len); + if (!path) { + log_error("malloc path %s %s failed", g_parser_mgr.mntpoint, (const char *)rp->d); + return NULL; + } + sprintf(path, "%s%s", g_parser_mgr.mntpoint, (const char *)rp->d); + free(rp); + return path; +} + +char *_get_fullpath(char *fpath, int fd) +{ + struct mfs_ioc_rpath *rp; + char *path; + int ret; + + if (fpath) + return fpath; + rp = malloc(sizeof(struct mfs_ioc_rpath) + MAX_PATH_SIZE); + if (!rp) { + log_error("rpath alloc failed"); + return NULL; + } + rp->max = MAX_PATH_SIZE; + ret = ioctl(fd, MFS_IOC_RPATH, (unsigned long)rp); + if (ret) { + log_error("realpath failed for fd:%d, ret:%d", fd, ret); + free(rp); + return NULL; + } + rp->d[rp->len] = '\0'; + path = strdup((const char *)rp->d); + if (!path) + log_error("strdup path(%s) failed", (const char *)rp->d); + free(rp); + return path; +} + +void maio_preload(struct maio *maio, int nio) +{ + struct maio_entry *entry; + + for (int i = 0; i < nio; ++i) { + entry = &maio->entries[i]; + _submit_local_io(0, entry); + } +} + +uint8_t pid_to_numaid(int pid) +{ + return (uint8_t)-1; +} + +void _process_read_io(uint32_t id, struct maio *maio, int nio) +{ + struct maio_entry *entry; + int i; + + /* process the successor io */ + for (i = 0; i < nio; i++) { + entry = &maio->entries[i]; + _submit_local_io(maio->fd, entry); + } +} + +void _submit_evict(struct maio_entry *entry) +{ + struct io_context *ioctx; + + ioctx = calloc(1, sizeof(struct io_context)); + if (!ioctx) { + log_error("ioctx alloc failed"); + return; + } + ioctx->off = entry->toff; + ioctx->len = entry->tlen; + ioctx->numaid = -1; + ioctx->path = entry->fpath; + ioctx->private = ioctx; + ioctx->end_io = ctx_cb; + loader_evict_submit(ioctx); +} + +void _process_evict(struct mfs_msg *msg) +{ + struct maio_entry *entry; + struct mfs_read *read; + uint64_t soff, eoff; + struct maio *maio; + int ret; + + if (g_parser_mgr.fs_mode != MFS_MODE_LOCAL) + return; + + maio = calloc(1, sizeof(struct maio)); + if (!maio) { + log_error("failed to alloc maio"); + return; + } + + read = (void*)msg->data; + maio->cksz = g_parser_mgr.chunk_size; + soff = round_down(read->off, maio->cksz); + eoff = round_up(read->off + read->len, maio->cksz); + maio->fd = msg->fd; + maio->off = read->off; + maio->len = eoff - soff; + maio->pid = read->pid; + maio->op = msg->opcode; + + ret = policy_evict(&maio); + if (ret <= 0) + goto out; + + for (int i = 0; i < ret; ++i) { + entry = &maio->entries[i]; + _submit_evict(entry); + } +out: + free(maio); +} + +void _parse_read_req(struct mfs_msg *msg) +{ + struct mfs_read *read; + struct maio *maio; + uint64_t soff, eoff, ts; + int ret, io_num = policy_max_io(); + + _process_evict(msg); + + /* alloc maio slots */ + maio = calloc(1, sizeof(struct maio) + io_num * sizeof(struct maio_entry)); + if (!maio) { + log_error("failed to alloc maio, num:%d", io_num); + return; + } + read = (void *)msg->data; + maio->cksz = g_parser_mgr.chunk_size; + soff = round_down(read->off, maio->cksz); + eoff = round_up(read->off + read->len, maio->cksz); + ts = _get_ts(); + maio->fd = msg->fd; + maio->off = soff; + maio->len = eoff - soff; + maio->pid = read->pid; + maio->op = msg->opcode; + maio->ts = ts; + + /* fill slots from policy */ + ret = policy_load(&maio); + if (ret < 0) { + char *path = _get_fullpath(NULL, maio->fd); + if (path) { + log_warn("io on path:%s%s offset:%lu length:%lu dropped", + g_parser_mgr.mntpoint, path, read->off, read->len); + free(path); + } + free(maio); + return; + } + + if (maio->flags & MAIO_WITH_SEQ) + loader_update_curseq(maio->curseq); + + /* keep the input parameter */ + maio->fd = msg->fd; + maio->off = soff; + maio->len = eoff - soff; + maio->cksz = g_parser_mgr.chunk_size; + maio->pid = read->pid; + maio->op = msg->opcode; + maio->ts = ts; + + /* process each maio in slots */ + _process_read_io(msg->id, maio, ret); + free(maio); +} + +void _parse_one_req(void *buf) +{ + struct mfs_msg *msg = (struct mfs_msg *)buf; + + if (msg->opcode == MFS_OP_READ || msg->opcode == MFS_OP_FAULT) + _parse_read_req(msg); + + free(buf); +} + +int maio_parse_req(void *buf, int size) +{ + struct mfs_msg *msg = (struct mfs_msg *)buf; + + if (msg->len != size) + return -1; + + threadpool_submit(g_parser_mgr.parser, buf, _parse_one_req); + return 0; +} + +int parser_init(uint32_t fs_mode, const char *mntpoint) +{ + g_parser_mgr.fs_mode = fs_mode; + g_parser_mgr.mntpoint = strdup(mntpoint); + g_parser_mgr.chunk_size = 4096; + if (!is_power_of_2(g_parser_mgr.chunk_size)) { + log_error("chunk size must be power of 2"); + return -1; + } + g_parser_mgr.parser = threadpool_create("ReqParser", PARSER_NUM, -1); + if (!g_parser_mgr.parser) { + log_error("failed to alloc parser threadpool"); + return -1; + } + + return 0; +} + +void parser_destory() +{ + threadpool_destroy(g_parser_mgr.parser); +} diff --git a/tools/mfs/maio-utils/src/policy/CMakeLists.txt b/tools/mfs/maio-utils/src/policy/CMakeLists.txt new file mode 100644 index 000000000000..714023475e07 --- /dev/null +++ b/tools/mfs/maio-utils/src/policy/CMakeLists.txt @@ -0,0 +1,16 @@ +#******************************************************************************* +#******************************************************************************* + +include_directories(${CMAKE_CURRENT_SOURCE_DIR}) + +#******************************************************************************* +#******************************************************************************* + +aux_source_directory(. POLICY_SRC) + +#******************************************************************************* +#******************************************************************************* + +add_library(policy ${POLICY_SRC}) +target_link_libraries(policy infra + securec) diff --git a/tools/mfs/maio-utils/src/policy/policy.c b/tools/mfs/maio-utils/src/policy/policy.c new file mode 100644 index 000000000000..404358ff9b46 --- /dev/null +++ b/tools/mfs/maio-utils/src/policy/policy.c @@ -0,0 +1,114 @@ +// SPDX-License-Identifier: GPL-2.0 +/* Copyright(c) 2025 Huawei Technologies Co., Ltd + */ +#include "policy.h" + +#include "maio.h" +#include "log.h" +#include "securec.h" +#include <dlfcn.h> + +typedef struct maio_operation* (*strategy_init)(); + +struct policy_mgr { + void *handle; + struct maio_operation *strategy; +}; + +struct policy_mgr g_policy; + +static int default_init(void) +{ + return 0; +} + +static void default_exit(void) +{ +} + +static int default_load(struct maio **io) +{ + return 0; +} + +static int default_evict(struct maio **io) +{ + return 0; +} + +struct maio_operation default_ops = { + .max_io = 1, + .init = default_init, + .exit = default_exit, + .load = default_load, + .evict = default_evict, +}; + +static void set_default_strategy() +{ + g_policy.strategy = &default_ops; +} + +int policy_max_io(void) +{ + return g_policy.strategy->max_io; +} + +/* return the filled num of io */ +int policy_load(struct maio **io) +{ + return g_policy.strategy->load(io); +} + +int policy_evict(struct maio **io) +{ + return g_policy.strategy->evict(io); +} + +int policy_register(const char *path) +{ + strategy_init sinit; + + if (!path || strlen(path) == 0) + return -1; + + g_policy.handle = dlopen(path, RTLD_LAZY); + if (!g_policy.handle) { + log_error("dlopen failed:%s", dlerror()); + return -1; + } + + sinit = dlsym(g_policy.handle, "register_strategy"); + if (!sinit) { + log_error("dlsym failed:%s", dlerror()); + dlclose(g_policy.handle); + g_policy.handle = NULL; + return -1; + } + g_policy.strategy = sinit(); + if (g_policy.strategy->init()) + return -1; + log_info("register strategy:%s done.", path); + return 0; +} + +void policy_unregister() +{ + if (g_policy.handle) { + g_policy.strategy->exit(); + dlclose(g_policy.handle); + g_policy.handle = NULL; + } + set_default_strategy(); +} + +int policy_init() +{ + g_policy.handle = NULL; + set_default_strategy(); + return 0; +} + +void policy_exit() +{ +} diff --git a/tools/mfs/maio-utils/test/Makefile b/tools/mfs/maio-utils/test/Makefile new file mode 100644 index 000000000000..6c87247aafa0 --- /dev/null +++ b/tools/mfs/maio-utils/test/Makefile @@ -0,0 +1,26 @@ +# SPDX-License-Identifier: GPL-2.0 + +CC := gcc +CFLAGS := -fPIC -Wall -Werror -O2 +CPPFLAGS := -fPIC -Wall -Werror -O2 -std=c++11 -g +LDFLAGS := -shared + +CFLAGS += $(addprefix -I,../include) +CPPFLAGS += $(addprefix -I,../include) + +C_SRCS := $(wildcard *.c) +CPP_SRCS := $(wildcard *.cpp) +TARGETS := $(C_SRCS:.c=.so) $(CPP_SRCS:.cpp=.so) + +all: $(TARGETS) + +%.so: %.c + $(CC) $(CFLAGS) $(LDFLAGS) -o $@ $< + +%.so: %.cpp + $(CC) $(CPPFLAGS) $(LDFLAGS) -o $@ $< + +clean: + rm -f $(TARGETS) + +.PHONY: all clean diff --git a/tools/mfs/maio-utils/test/README.md b/tools/mfs/maio-utils/test/README.md new file mode 100644 index 000000000000..a7b7980de933 --- /dev/null +++ b/tools/mfs/maio-utils/test/README.md @@ -0,0 +1,61 @@ +Here, we provide some strategies library for some cases. + +- strategy_demo +A simple strategy library shows how to write a demo. + +- strategy_prefetch +A prefetch strategy library to prefetch data. + +- strategy_toend +A readahead strategy until the end. + +- strategy_generate +A tools using maio-util to generate the io trace. + +- strategy_replay +A strategy to replay the io trace. + +### How to build + +Using `make -j` to build the libraries. + +### How to use + +Set the library path as the `-s` parameter's value to run with maio-utils. Here we show usage of the advanced replay strategy: + +- generate the io trace + +> assume you have mounted the mfs at /mnt/mfs + +This is the tracing period. + +``` +export MAIO_TRACE_FILE=trace.txt +export MAIO_MNTPOINT=/mnt/mfs # /mnt/mfs is the MFS mountpoint +./bin/maio-utils -m /mnt/mfs -s /home/lihb/workspace/hulk-6.6/maio-utils/test/strategy_generate.so +``` +Then, you can trigger the io. After finishing, you would get the trace file with the content likes: + +``` +xxx/model-00001-of-000008.safetensors 0 32768 0 +xxx/model-00001-of-000008.safetensors 32768 32768 0 +xxx/model-00001-of-000008.safetensors 65536 32768 0 +xxx/model-00001-of-000008.safetensors 98304 32768 0 +xxx/model-00001-of-000008.safetensors 131072 32768 0 +... + +``` + +The format is "<path> <offset> <length> <numaid>". + +- replay the io trace + +> assume you have mounted the mfs at /mnt/mfs and begin to launch the model. + +This is the running period. + +``` +export MAIO_TRACE_FILE=trace.txt +export MAIO_MNTPOINT=/mnt/mfs # /mnt/mfs is the MFS mountpoint +./bin/maio-utils -m /mnt/mfs -s /home/lihb/workspace/hulk-6.6/maio-utils/test/replay.so +``` diff --git a/tools/mfs/maio-utils/test/strategy_demo.c b/tools/mfs/maio-utils/test/strategy_demo.c new file mode 100644 index 000000000000..54554b609f85 --- /dev/null +++ b/tools/mfs/maio-utils/test/strategy_demo.c @@ -0,0 +1,45 @@ +// SPDX-License-Identifier: GPL-2.0 +/* Copyright(c) 2025 Huawei Technologies Co., Ltd + */ +#include "maio.h" +#include <stddef.h> + +int demo_init(void) +{ + return 0; +} + +void demo_exit(void) +{ +} + +int demo_load(struct maio **io) +{ + struct maio_entry *entry; + + entry = (*io)->entries; + entry[0].toff = (*io)->off + (*io)->len; + entry[0].tlen = (*io)->cksz; + entry[0].fpath = NULL; + entry[0].tnuma = -1; + return 1; +} + +int demo_evict(struct maio **io) +{ + ((void)io); + return 0; +} + +const struct maio_operation demo_strategy = { + .max_io = 1, + .init = demo_init, + .exit = demo_exit, + .load = demo_load, + .evict = demo_evict, +}; + +struct maio_operation *register_strategy() +{ + return (struct maio_operation *)&demo_strategy; +} diff --git a/tools/mfs/maio-utils/test/strategy_generate.cpp b/tools/mfs/maio-utils/test/strategy_generate.cpp new file mode 100644 index 000000000000..929be30e6a42 --- /dev/null +++ b/tools/mfs/maio-utils/test/strategy_generate.cpp @@ -0,0 +1,296 @@ +// SPDX-License-Identifier: GPL-2.0 +/* Copyright(c) 2025 Huawei Technologies Co., Ltd + */ +#include "maio.h" +#include "strategy_template.h" + +#include <array> +#include <atomic> +#include <cstring> +#include <condition_variable> +#include <fstream> +#include <sstream> +#include <iostream> +#include <mutex> +#include <queue> +#include <thread> +#include <unordered_map> +#include <vector> +#include <unistd.h> + +using namespace std; + +#define MAIO_MNTPOINT "MAIO_MNTPOINT" +#define MAIO_TRACE_FILE "MAIO_TRACE_FILE" +#define CACHEDIR "cachedir" + +#define MAXLINE 1024 + +/* Data struct definition */ +template<typename T> +class SafeQueue { +private: + queue<T> data_queue; + mutex mtx; + condition_variable cv; + +public: + // 添加数据到队列 + void push(T item) { + lock_guard<mutex> lock(mtx); + data_queue.push(move(item)); + cv.notify_one(); + } + + // 从队列取出数据（阻塞直到有数据） + T pop() { + unique_lock<mutex> lock(mtx); + cv.wait(lock, [this]{ return !data_queue.empty(); }); + T item = move(data_queue.front()); + data_queue.pop(); + return item; + } + + // 检查队列是否为空 + bool empty() { + lock_guard<mutex> lock(mtx); + return data_queue.empty(); + } +}; + +struct io { + string name; + uint64_t off; + uint64_t len; + uint8_t npu; + + io(const string& _name, uint64_t _off, uint64_t _len, uint8_t _npu) + : name(_name), off(_off), len(_len), npu(_npu) {} +}; + +/* Global variable definition */ +unordered_map<int, char*> fd_map; /* map fd to path */ +mutex fd_mutex; +string cachedir; + +vector<io> iovec; /* io traces */ +mutex iovec_mutex; +SafeQueue<vector<io>> iovec_queue; + +ofstream tfile; /* trace file */ + +thread writer; + +unordered_map<int, int> process_npu_map; +mutex pnpu_mutex; + +atomic<int> npu_smi_enable(-1); + +/* fd_map helper function */ +char* fd_path_find(int iofd) { + lock_guard<mutex> lock(fd_mutex); + auto it = fd_map.find(iofd); + if (it != fd_map.end()) { + return it->second; + } else { + char *path = _get_fullpath(NULL, iofd); + fd_map.insert({iofd, path}); + return path; + } +} + +/* iovec helper function */ +void io_insert_data(string name, uint64_t off, uint64_t len, uint8_t npu) +{ + lock_guard<mutex> lock(iovec_mutex); + iovec.emplace_back(name, off, len, npu); + + vector<io> copy_iovec = move(iovec); + iovec.clear(); + + iovec_queue.push(move(copy_iovec)); +} + +void get_mount_option(const char* mntpoint, const char* option) +{ + ifstream mounts("/proc/mounts"); + string line, target_mp; + + if (!mounts.is_open()) { + cerr << "Error: could not open /proc/mounts" << endl; + return; + } + + while (getline(mounts, line)) { + istringstream iss(line); + string device, mp, fs_type, options; + + if (!(iss >> device >> mp >> fs_type >> options)) + continue; + + if (mp.back() != '/') + mp += '/'; + target_mp = mntpoint; + if (target_mp.back() != '/') + target_mp += '/'; + + if (target_mp == mp) { + istringstream opt_stream(options); + string opt; + + while (getline(opt_stream, opt, ',')) { + if (opt.find(option) == 0) { + if (opt[strlen(option)] == '=') { + cachedir = opt.substr(strlen(option) + 1); + } + } + } + break; + } + } +} + +void writer_thread() { + while (true) { + vector<io> iovec_to_write = iovec_queue.pop(); + + for (const auto &item : iovec_to_write) + tfile << cachedir << item.name << " " << item.off << " " << item.len << " " << (unsigned int)item.npu << endl; + tfile.flush(); + } +} + +string exec_cmd(const char* cmd) { + array<char, 128> buffer; + string result; + unique_ptr<FILE, decltype(&pclose)> pipe(popen(cmd, "r"), pclose); + + if (!pipe) { + throw std::runtime_error("popen() failed!"); + } + while (fgets(buffer.data(), buffer.size(), pipe.get()) != nullptr) { + result += buffer.data(); + } + return result; +} + +bool isNumber(const std::string& str) { + return !str.empty() && str.find_first_not_of("0123456789") == std::string::npos; +} + +void get_npu_info(string &npuinfo) +{ + istringstream iss(npuinfo); + string line; + + while (getline(iss, line)) { + istringstream lineStream(line); + string token; + vector<string> tokens; + while (getline(lineStream, token, ' ')) { + if (!token.empty()) { + tokens.push_back(token); + } + } + if (tokens.size() >= 3) { + if (!isNumber(tokens[1]) || !isNumber(tokens[4])) + continue; + int npuId = stoi(tokens[1]); + int processId = stoi(tokens[4]); + process_npu_map[processId] = npuId; + } + } +} + +int process_npu_find(int processid) { + lock_guard<mutex> lock(pnpu_mutex); + auto it = process_npu_map.find(processid); + + if (npu_smi_enable.load() == -1) { + if (exec_cmd("command -v npu-smi").empty()) + npu_smi_enable.store(0); + else + npu_smi_enable.store(1); + } + + if (npu_smi_enable.load() == 0) + return 0; + + if (it != process_npu_map.end()) { + return it->second; + } else { + string result = exec_cmd("npu-smi info"); + get_npu_info(result); + auto it = process_npu_map.find(processid); + if (it != process_npu_map.end()) + return it->second; + else + return 0; + } +} + +int strategy_init(void) +{ + char *tfilename = getenv(MAIO_TRACE_FILE); + char *mntpoint = getenv(MAIO_MNTPOINT); + + if(!tfilename || !mntpoint) { + printf("no MAIO_TRACE_FILE or MAIO_MNTPOINT env variable\n"); + return -1; + } + get_mount_option(mntpoint, CACHEDIR); + string stfilename = tfilename; + tfile.open(stfilename, ios::trunc); + writer = thread(writer_thread); + if (!tfile.is_open()) { + cerr << "Error: could not open" << stfilename << endl; + return -1; + } + return 0; +} + +void strategy_exit(void) +{ + lock_guard<mutex> lock(iovec_mutex); + vector<io> copy_iovec = move(iovec); + iovec.clear(); + iovec_queue.push(move(copy_iovec)); + + tfile.close(); + writer.join(); +} + +int strategy_load(struct maio **io) +{ + char *path; + string pathStr; + int ret = 0; + + if (!tfile.is_open()) + return -1; + + path = fd_path_find((*io)->fd); + pathStr.assign(path); + io_insert_data(pathStr, (*io)->off, (*io)->len, process_npu_find((*io)->pid)); + + return ret; +} + +int strategy_evict(struct maio **io) +{ + ((void)io); + return 0; +} + +const struct maio_operation maio_strategy = { + .max_io = 1, + .init = strategy_init, + .exit = strategy_exit, + .load = strategy_load, + .evict = strategy_evict, +}; + +struct maio_operation *register_strategy() +{ + return (struct maio_operation *)&maio_strategy; +} diff --git a/tools/mfs/maio-utils/test/strategy_prefetch.cpp b/tools/mfs/maio-utils/test/strategy_prefetch.cpp new file mode 100644 index 000000000000..9ba205931cdc --- /dev/null +++ b/tools/mfs/maio-utils/test/strategy_prefetch.cpp @@ -0,0 +1,150 @@ +#include "strategy_prefetch.h" +#include "maio.h" +#include <iostream> +#include <map> +#include <cstring> +#include <string> +#include <stdint.h> +#include <sys/stat.h> +#include <sys/mman.h> +#include <dirent.h> +#include <fcntl.h> +#include <unistd.h> + +using namespace std; + +struct prefetch_mgr { + map<string, uint64_t> files; +}; + +struct prefetch_mgr g_prefetch; + +int get_files(const char *parent) +{ + DIR *dir; + struct dirent *entry; + string filepath; + struct stat buf; + + if ((dir = opendir(parent)) == NULL) { + cout<<"opening directory failed"<<endl; + return -1; + } + + while ((entry = readdir(dir)) != NULL) { + if (strcmp(entry->d_name, ".") == 0 || strcmp(entry->d_name, "..") == 0) + continue; + + filepath = string(parent) + "/" + string(entry->d_name); + if (stat(filepath.c_str(), &buf) == -1) { + cout<<"stat path:"<<filepath<<" failed"<<endl; + continue; + } + if (S_ISDIR(buf.st_mode)) + continue; + cout<<"Find file:"<<filepath<<" size:"<<buf.st_size<<endl; + g_prefetch.files.insert(pair<string, uint64_t>(filepath, buf.st_size)); + } + + map<string, uint64_t>::iterator it; + for (it = g_prefetch.files.begin(); it != g_prefetch.files.end(); ++it) { + cout<<"path:"<<it->first<<" size:"<<it->second<<endl; + } + return 0; +} + +int prefetch_init(void) +{ + char *parent = getenv("MODEL_WEIGHT_DIR"); + if (!parent) { + cout<<"please set the env MODEL_WEIGHT_DIR"<<endl; + return -1; + } + + if (get_files(parent) != 0) { + cout<<"get all files failed"<<endl; + return -1; + } + return 0; +} + +void prefetch_exit(void) +{ +} + +struct thread_ctx { + char *path; + uint64_t off; + uint64_t len; +}; + +void *fault(void *arg) +{ + struct thread_ctx *ctx = (struct thread_ctx *)arg; + int fd = open(ctx->path, O_RDONLY); + if (fd < 0) { + cout<<"open file:%s"<<ctx->path<<" failed"<<endl; + free(ctx); + return NULL; + } + + cout<<"FAULT path:"<<ctx->path<<" fd:"<<fd<<" off:"<<ctx->off<<" len:"<<ctx->len<<endl; + void *addr = mmap(NULL, ctx->len, PROT_READ, MAP_SHARED, fd, 0); + uint64_t idx; + char tmp, total, *buffer = (char *)addr; + total = 'B'; + for (idx = 0; idx < ctx->len; idx += 4096) { + tmp = buffer[idx]; + total += tmp; + } + cout<<"Fault calculate total:"<<total<<endl; + munmap(addr, ctx->len); + close(fd); + free(ctx); + return NULL; +} + +int prefetch_load(struct maio **io) +{ + map<string, uint64_t>::iterator it; + struct thread_ctx *ctx; + pthread_t t0; + int ret; + + for (it = g_prefetch.files.begin(); it != g_prefetch.files.end(); ++it) { + ctx = (struct thread_ctx *)malloc(sizeof(struct thread_ctx)); + if (!ctx) { + cout<<"malloc ctx failed"<<endl; + continue; + } + ctx->path = strdup(it->first.c_str()); + ctx->off = 0; + ctx->len = it->second; + ret = pthread_create(&t0, NULL, fault, ctx); + if (ret == 0) + pthread_detach(t0); + } + g_prefetch.files.clear(); + + return 0; +} + +int prefetch_evict(struct maio **io) +{ + ((void)io); + return 0; +} + +const struct maio_operation prefetch_strategy = { + .max_io = 1, + .init = prefetch_init, + .exit = prefetch_exit, + .load = prefetch_load, + .evict = prefetch_evict, +}; + +struct maio_operation *register_strategy() +{ + return (struct maio_operation *)&prefetch_strategy; +} + diff --git a/tools/mfs/maio-utils/test/strategy_prefetch.h b/tools/mfs/maio-utils/test/strategy_prefetch.h new file mode 100644 index 000000000000..2c2475dc91f0 --- /dev/null +++ b/tools/mfs/maio-utils/test/strategy_prefetch.h @@ -0,0 +1,21 @@ +#ifndef STRATEGY_PREFETCH_H +#define STRATEGY_PREFETCH_H + +#ifdef __cplusplus +extern "C" { +#endif + +int prefetch_init(void); +void prefetch_exit(void); +int prefetch_load(struct maio *io); +int prefetch_evict(struct maio **io); + +extern const struct maio_operation prefetch_strategy; + +struct maio_operation *register_strategy(); + +#ifdef __cplusplus +} +#endif + +#endif diff --git a/tools/mfs/maio-utils/test/strategy_replay.cpp b/tools/mfs/maio-utils/test/strategy_replay.cpp new file mode 100644 index 000000000000..85c1c7defedd --- /dev/null +++ b/tools/mfs/maio-utils/test/strategy_replay.cpp @@ -0,0 +1,540 @@ +// SPDX-License-Identifier: GPL-2.0 +/* Copyright(c) 2025 Huawei Technologies Co., Ltd + */ +#include "maio.h" +#include "strategy_template.h" + +#include <atomic> +#include <array> +#include <cstring> +#include <fstream> +#include <iostream> +#include <map> +#include <mutex> +#include <sstream> +#include <string> +#include <thread> +#include <unordered_map> +#include <vector> +#include <pthread.h> +#include <fcntl.h> +#include <unistd.h> + +using namespace std; + +#define MAIO_MNTPOINT "MAIO_MNTPOINT" +#define MAIO_TRACE_FILE "MAIO_TRACE_FILE" +#define CACHEDIR "cachedir" +#define MAXLINE 1024 + +/* Data structure definition */ +enum { + IO_NONE = 0, + IO_LOADED, + IO_EVICTED, +}; + +struct io { + string name; + uint64_t off; + uint64_t len; + uint8_t npu; + int flag; + + explicit io(const string& _name, uint64_t _off, uint64_t _len, uint8_t _npu) + : name(_name), off(_off), len(_len), npu(_npu), flag(IO_NONE) {} +}; + + +struct ioKey { + string name; + uint64_t off; + uint64_t len; + uint8_t npu; + + ioKey(const string& _name, uint64_t _off, uint64_t _len, uint8_t _npu) + : name(_name), off(_off), len(_len), npu(_npu) {} + + bool operator<(const ioKey &other) const { + return name < other.name || (name == other.name && off < other.off) || + (name == other.name && off == other.off && len < other.len); + } +}; + +vector<io> iovec; /* io traces */ +map<ioKey, int> io_to_index; +unordered_map<int, char*> fd_map; +mutex fd_mutex; +string cachedir; + +unordered_map<int, int> npu_numa_map; + +unordered_map<int, int> process_npu_map; +mutex pnpu_mutex; + +/* for evict */ +#define EVICT_DIST 10000 +#define EVICT_IO_MAX 5000 +thread_local size_t evict_idx = 0; + +atomic<int> npu_smi_enable(-1); +atomic<uint64_t> memory_usage(0); + +uint64_t memory_limit; +uint64_t memory_limit_bound; +string usage_in_bytes_path = "memory.usage_in_bytes"; +string limit_in_bytes_path = "memory.limit_in_bytes"; +string cgroup_mem=""; + +string get_cgroup_path() +{ + fstream cgroup_file("/proc/self/cgroup"); + if (!cgroup_file.is_open()) { + cerr << "Failed to open /proc/self/cgroup" << endl; + return ""; + } + + string line; + string cgroup_path; + while (getline(cgroup_file, line)) { + istringstream iss(line); + string hierarchy, controller, path; + if (!getline(iss, hierarchy, ':')) + continue; + if (!getline(iss, controller, ':')) + continue; + if (!getline(iss, path, ':')) + continue; + if (controller == "memory") { + cgroup_path = path; + break; + } + } + cgroup_file.close(); + return cgroup_path; +} + +/* read from /sys/fs/cgroup/memory/memory.[usage_in_bytes|limit_in_bytes] */ +uint64_t read_cgroup_memory(const string &file_path) +{ + if (cgroup_mem == "") + return -1; + string whole_path = "/sys/fs/cgroup/memory"+cgroup_mem+"/"+file_path; + ifstream cgroup_file(whole_path); + + if(!cgroup_file.is_open()) { + cerr << "Failed to open " << file_path << endl; + return -1; + } + + string line; + while(getline(cgroup_file, line)) { + try { + return stoll(line); + } catch (...) { + cerr << "Failed to parse line" << line << endl; + } + } + + cgroup_file.close(); + return -1; +} + +/* mount helper */ +void get_mount_option(const char* mntpoint, const char* option) +{ + ifstream mounts("/proc/mounts"); + string line, target_mp; + + if (!mounts.is_open()) { + cerr << "Error: could not open /proc/mounts" << endl; + return; + } + + while (getline(mounts, line)) { + istringstream iss(line); + string device, mp, fs_type, options; + + if (!(iss >> device >> mp >> fs_type >> options)) + continue; + + if (mp.back() != '/') + mp += '/'; + target_mp = mntpoint; + if (target_mp.back() != '/') + target_mp += '/'; + + if (target_mp == mp) { + istringstream opt_stream(options); + string opt; + + while (getline(opt_stream, opt, ',')) { + if (opt.find(option) == 0) { + if (opt[strlen(option)] == '=') { + cachedir = opt.substr(strlen(option) + 1); + } + } + } + break; + } + } +} + +/* fd_map helper function */ +char* fd_path_find(int iofd) { + lock_guard<mutex> lock(fd_mutex); + auto it = fd_map.find(iofd); + if (it != fd_map.end()) { + return it->second; + } else { + char *path = _get_fullpath(NULL, iofd); + size_t len = strlen(path) + cachedir.length() + 1; + char *realpath = (char*)malloc(len); + snprintf(realpath, len, "%s%s", cachedir.c_str(), path); + fd_map.insert({iofd, realpath}); + return realpath; + } +} + +string exec_cmd(const char* cmd) { + array<char, 128> buffer; + string result = ""; + unique_ptr<FILE, decltype(&pclose)> pipe(popen(cmd, "r"), pclose); + + if (!pipe) { + throw std::runtime_error("popen() failed!"); + } + while (fgets(buffer.data(), buffer.size(), pipe.get()) != nullptr) { + result += buffer.data(); + } + return result; +} + +bool isNumber(const std::string& str) { + return !str.empty() && str.find_first_not_of("0123456789") == std::string::npos; +} + +/* cpu_numa_map builder from lscpu */ +unordered_map<string, int> build_cpu_numa_map() { + unordered_map<string, int> cpuNumaMap; + string result = exec_cmd("lscpu"); + istringstream iss(result); + string line; + + while (getline(iss, line)) { + if (line.find("NUMA node") != string::npos) { + istringstream lineStream(line); + string key, value; + int numaNode; + getline(lineStream, value, ':'); + getline(lineStream, key); + /* what if numa is more than 10 */ + value = value.substr(value.find("NUMA node") + 9, 1); + if (!isNumber(value)) + continue; + numaNode = stoi(value); + key.erase(0, key.find_first_not_of(" \t")); + key.erase(key.find_last_not_of(" \t") + 1); + cpuNumaMap[key]=numaNode; + } + } + return cpuNumaMap; +} + +/* npu_numa_map builder, from npu-smi info -t topo */ +void build_npu_map_map() +{ + if (npu_smi_enable.load() == -1) { + if (exec_cmd("command -v npu-smi").empty()) + npu_smi_enable.store(0); + else + npu_smi_enable.store(1); + } + + if (npu_smi_enable.load() == 0) + return; + + unordered_map<string, int> cpuNumaMap = build_cpu_numa_map(); + string result = exec_cmd("npu-smi info -t topo"); + istringstream iss(result); + string line; + size_t target_token_size = 0; + + cout << "NPU\tCPU_AFFINITY\tNUMA_NODE" << endl; + while (getline(iss, line)) { + istringstream lineStream(line); + string npuStr, numaStr, token; + vector<string> tokens; + + while (lineStream >> token) { + tokens.push_back(token); + } + + if (tokens.size() >= 2 && tokens.back() == "Affinity") { + target_token_size = tokens.size() - 1 + 1; + continue; + } + + if (tokens.size() != target_token_size) + continue; + + int npuId = stoi(tokens[0].substr(3)); + + numaStr = tokens.back(); + cout << npuId << "\t" << numaStr << "\t\t" << cpuNumaMap[numaStr] << endl; + npu_numa_map[npuId] = cpuNumaMap[numaStr]; + } +} + +/* process_npu_map builder, from npu-smi info */ +void get_npu_info(string &npuinfo) +{ + istringstream iss(npuinfo); + string line; + + while (getline(iss, line)) { + istringstream lineStream(line); + string token; + vector<string> tokens; + while (getline(lineStream, token, ' ')) { + if (!token.empty()) { + tokens.push_back(token); + } + } + if (tokens.size() >= 3) { + if (!isNumber(tokens[1]) || !isNumber(tokens[4])) + continue; + int npuId = stoi(tokens[1]); + int processId = stoi(tokens[4]); + process_npu_map[processId] = npuId; + } + } +} + +/* process_npu_map helper */ +int process_npu_find(int processid) { + lock_guard<mutex> lock(pnpu_mutex); + auto it = process_npu_map.find(processid); + + if (npu_smi_enable.load() == 0) + return 0; + + if (it != process_npu_map.end()) { + return it->second; + } else { + string result = exec_cmd("npu-smi info"); + get_npu_info(result); + auto it = process_npu_map.find(processid); + if (it != process_npu_map.end()) + return it->second; + else + return 0; + } +} + +int strategy_init(void) +{ + const char *mntpoint = getenv(MAIO_MNTPOINT); + char *tfilename = getenv(MAIO_TRACE_FILE); + if(!tfilename || !mntpoint) { + fprintf(stderr, "no MAIO_TRACE_FILE or MAIO_MNTPOINT env variable\n"); + return -1; + } + ifstream tfile(tfilename); + string line; + + cgroup_mem=get_cgroup_path(); + memory_limit = read_cgroup_memory(limit_in_bytes_path); + memory_limit_bound = memory_limit * 4 / 5; + cout << "memory_limit: " << memory_limit << " memory_limit_bound: " << memory_limit_bound << endl; + get_mount_option(mntpoint, CACHEDIR); + build_npu_map_map(); + + if(!tfile.is_open()) { + cerr << "Failed to open file:" << tfilename << endl; + return -1; + } + + while(getline(tfile, line)) { + istringstream iss(line); + + string name; + uint64_t off; + uint64_t len; + unsigned int npu; + if (iss >> name >> off >> len >> npu) { + iovec.emplace_back(name, off, len, npu); + io_to_index[ioKey(name, off, len, npu)] = iovec.size() - 1; + } else { + cerr << "Failed to parse line:" << line << endl; + } + } + + tfile.close(); + return 0; +} + +void strategy_exit(void) +{ +} + +int strategy_load(struct maio **io) +{ + int npuid = process_npu_find((*io)->pid); + struct maio_entry *entry; + string pathStr; + char *path; + uint64_t idx = iovec.size(); + int ret = 0; + struct maio *new_io; + + path = fd_path_find((*io)->fd); + pathStr.assign(path); + + ioKey key(pathStr, (*io)->off, (*io)->len, npuid); + auto io_it = io_to_index.lower_bound(key); + if (io_it == io_to_index.end()) { + entry = (*io)->entries; + entry[0].toff = (*io)->off + (*io)->len; + entry[0].tlen = 0; + entry[0].fpath = NULL; + entry[0].tnuma = npu_numa_map[npuid]; + return 1; + } + if (key < io_it->first) { + if (io_it != io_to_index.begin()) { + --io_it; + if (io_it->first.name != pathStr) { + ++io_it; + } + } + } + + idx = io_it->second; + (*io)->flags |= MAIO_WITH_SEQ; + (*io)->curseq = idx; + entry = (*io)->entries; + int ios = 0; + uint64_t memory_usage = read_cgroup_memory(usage_in_bytes_path); + vector<int> io_extra; + + /* fill ios below max_io */ + while (idx < iovec.size()) { + if (iovec[idx].flag == IO_LOADED) { + idx++; + continue; + } + memory_usage += iovec[idx].len / 2; + if (memory_usage > memory_limit_bound) + break; + if (ios >= maio_strategy.max_io) { + io_extra.push_back(idx); + goto next_io; + } + entry[ios].seq = idx; + if (pathStr == iovec[idx].name) { + entry[ios].toff = iovec[idx].off; + entry[ios].tlen = iovec[idx].len; + entry[ios].tnuma = npu_numa_map[npuid]; + } else { + entry[ios].fpath = strdup(iovec[idx].name.c_str()); + entry[ios].toff = iovec[idx].off; + entry[ios].tlen = iovec[idx].len; + entry[ios].tnuma = npu_numa_map[iovec[idx].npu]; + } +next_io: + iovec[idx].flag = IO_LOADED; + idx++; + ios++; + } + + /* fill ios above max_io */ + if (ios > maio_strategy.max_io) { + new_io = (struct maio *)realloc(*io, sizeof(struct maio) + ios * sizeof(struct maio_entry)); + if (!new_io) { + printf("failed to alloc ios:%d maio\n", ios); + return -1; + } + *io = new_io; + entry = (*io)->entries; + for (int i = maio_strategy.max_io; i < ios; ++i) { + int io_idx = io_extra[i - maio_strategy.max_io]; + entry[i].seq = io_idx; + if (pathStr == iovec[io_idx].name) { + entry[i].toff = iovec[io_idx].off; + entry[i].tlen = iovec[io_idx].len; + entry[i].tnuma = npu_numa_map[npuid]; + } else { + entry[i].fpath = strdup(iovec[io_idx].name.c_str()); + entry[i].toff = iovec[io_idx].off; + entry[i].tlen = iovec[io_idx].len; + entry[i].tnuma = npu_numa_map[iovec[io_idx].npu]; + } + } + } + + // printf("file:%s off:%lu len:%lu generate_ios:%d from %d",path, (*io)->off, (*io)->len, ios, io_it->second); + ret = ios; + return ret; +} + +int strategy_evict(struct maio **io) +{ + string pathStr; + char *path; + path = fd_path_find((*io)->fd); + pathStr.assign(path); + ioKey key(pathStr, (*io)->off, (*io)->len, (*io)->pid); + auto io_it = io_to_index.lower_bound(key); + int dist = 0; + size_t i = 0, io_idx = io_it->second; + struct maio_entry *entry; + struct maio *new_io; + uint64_t evict_memory = 0; + uint64_t memory_usage = read_cgroup_memory(usage_in_bytes_path); + + if (memory_usage < (memory_limit_bound * 3 / 4)) + return 0; + dist = io_idx - evict_idx; + if (dist < 0) + return 0; + if (static_cast<size_t>(dist) < iovec.size() / 8 && io_it != io_to_index.end()) + return 0; + size_t evict_win = 0; + if (io_it != io_to_index.end()) { + evict_win = min(iovec.size() / 10, io_idx - 1); + } else { + evict_win = dist; + } + + new_io = (struct maio*)realloc(*io, sizeof(struct maio) + evict_win * sizeof(struct maio_entry)); + if (!new_io) + return -1; + *io = new_io; + entry = new_io->entries; + + for (i = 0; i < evict_win && evict_idx < iovec.size(); ++i) { + evict_memory += iovec[evict_idx].len; + if (evict_memory > memory_limit_bound / 8) + break; + entry[i].fpath = strdup(iovec[evict_idx].name.c_str()); + entry[i].toff = iovec[evict_idx].off; + entry[i].tlen = iovec[evict_idx].len; + evict_idx++; + iovec[evict_idx].flag = IO_EVICTED; + } + return i; +} + +const struct maio_operation maio_strategy = { + .max_io = 500, + .init = strategy_init, + .exit = strategy_exit, + .load = strategy_load, + .evict = strategy_evict, +}; + +struct maio_operation *register_strategy() +{ + return (struct maio_operation *)&maio_strategy; +} diff --git a/tools/mfs/maio-utils/test/strategy_template.h b/tools/mfs/maio-utils/test/strategy_template.h new file mode 100644 index 000000000000..6599b68754c9 --- /dev/null +++ b/tools/mfs/maio-utils/test/strategy_template.h @@ -0,0 +1,21 @@ +#ifndef STRATEGY_TEMPLATE_H +#define STRATEGY_TEMPLATE_H + +#ifdef __cplusplus +extern "C" { +#endif + +int strategy_init(void); +void strategy_exit(void); +int strategy_load(struct maio **io); +int strategy_evict(struct maio *io); + +extern const struct maio_operation maio_strategy; + +struct maio_operation *register_strategy(); + +#ifdef __cplusplus +} +#endif + +#endif diff --git a/tools/mfs/maio-utils/test/strategy_toend.c b/tools/mfs/maio-utils/test/strategy_toend.c new file mode 100644 index 000000000000..f2208281f8ee --- /dev/null +++ b/tools/mfs/maio-utils/test/strategy_toend.c @@ -0,0 +1,46 @@ +// SPDX-License-Identifier: GPL-2.0 +/* Copyright(c) 2025 Huawei Technologies Co., Ltd + */ +#include "maio.h" +#include <stddef.h> + +int demo_init(void) +{ + return 0; +} + +void demo_exit(void) +{ +} + +int demo_load(struct maio **io) +{ + struct maio_entry *entry; + + entry = (*io)->entries; + entry[0].toff = (*io)->off + (*io)->len; + /* Read Ahead until the end */ + entry[0].tlen = 0; + entry[0].fpath = NULL; + entry[0].tnuma = -1; + return 1; +} + +int demo_evict(struct maio **io) +{ + ((void)io); + return 0; +} + +const struct maio_operation demo_strategy = { + .max_io = 1, + .init = demo_init, + .exit = demo_exit, + .load = demo_load, + .evict = demo_evict, +}; + +struct maio_operation *register_strategy() +{ + return (struct maio_operation *)&demo_strategy; +} diff --git a/tools/mfs/maio-utils/test/trace_analyse.py b/tools/mfs/maio-utils/test/trace_analyse.py new file mode 100644 index 000000000000..c25d1ce364df --- /dev/null +++ b/tools/mfs/maio-utils/test/trace_analyse.py @@ -0,0 +1,87 @@ +def merge_intervals(intervals): + """ + 合并区间 + :param intervals: 区间列表，每个区间是一个元组 (start, end) + :return: 合并后的区间列表 + """ + if not intervals: + return [] + + # 按起始位置排序 + intervals.sort(key=lambda x: x[0]) + + merged = [intervals[0]] + for current in intervals[1:]: + last_merged = merged[-1] + if current[0] <= last_merged[1]: # 有重叠 + merged[-1] = (last_merged[0], max(last_merged[1], current[1])) + else: + merged.append(current) + + return merged + +def process_file(input_file): + """ + 处理输入文件 + :param input_file: 输入文件路径 + :return: 每个文件的合并区间和总大小 + """ + from collections import defaultdict + + # 使用字典按文件名分组 + file_intervals = defaultdict(list) + + with open(input_file, 'r') as f: + for line in f: + parts = line.strip().split() + if len(parts) < 3: + print(f"跳过无效行: {line.strip()}") + continue + + # 解析文件名、起始位置、大小 + filename, start, size = parts[:3] + start = int(start) + size = int(size) + + # 计算区间 + end = start + size + file_intervals[filename].append((start, end)) + + # 合并每个文件的区间 + merged_results = {} + for filename, intervals in file_intervals.items(): + merged_intervals = merge_intervals(intervals) + # 计算合并后的区间总大小 + total_size = sum(end - start for start, end in merged_intervals) + merged_results[filename] = { + "merged_intervals": merged_intervals, + "total_size": total_size + } + + return merged_results + +def main(): + import sys + + if len(sys.argv) != 2: + print("用法: python script.py <输入文件>") + sys.exit(1) + + input_file = sys.argv[1] + + merged_results = process_file(input_file) + + whole_result=0 + + for filename, result in merged_results.items(): + print(f"文件: {filename}") + # print("合并后的区间:") + # for interval in result["merged_intervals"]: + # print(f"起始位置: {interval[0]}, 结束位置: {interval[1]}") + print(f"所有区间内数据的总大小: {result['total_size']} 字节") + print("-" * 40) + whole_result+=result['total_size'] + print(f"所有文件内数据的总大小: {whole_result} 字节") + +if __name__ == "__main__": + main() diff --git a/tools/mfs/maio-utils/test/trace_gen.py b/tools/mfs/maio-utils/test/trace_gen.py new file mode 100644 index 000000000000..d050fa48801c --- /dev/null +++ b/tools/mfs/maio-utils/test/trace_gen.py @@ -0,0 +1,116 @@ +import argparse +import os +import sys + +FILENAME_IDX = 0 +OFF_IDX = 1 +LEN_IDX = 2 +NUMA_IDX = 3 +OFF_BOUND = 1024 * 1024 * 4096 + +class ReplayHintGenerator(object): + def __init__(self, trace_file, output_file): + self.output_file = output_file + self.trace_file = trace_file + self.data_list = [] + + def generate_hint(self): + with open(self.output_file, 'w') as f: + for data in self.data_list: + data_off = data[OFF_IDX] + data_len = data[LEN_IDX] + print(data_off) + step = (512 * 1024 * 1024) + while data_len: + if data_len < step: + step = data_len + print(f"{data[FILENAME_IDX]} {data_off} {step} {data[NUMA_IDX]}", file=f) + data_len-=step + data_off+=step + + def parse_trace(self): + file_map_index={} + with open(self.trace_file, 'r') as f: + for line in f: + parts = line.strip().split(' ') + + filename, off, length, numaid = parts + off = int(off) + length = int(length) + numaid = int(numaid) + if self.data_list: + tail_data = self.data_list[-1] + ''' + 首先尝试跟末尾的数据合并: + 如若跟末尾的数据是同一个文件跟numaid, + 两者之间的距离小于设定值, + 就进行合并 + ''' + if (tail_data[FILENAME_IDX] == filename and tail_data[NUMA_IDX] == numaid): + tail_off = tail_data[OFF_IDX] + tail_end = tail_data[OFF_IDX] + tail_data[LEN_IDX] + end = off + length + abs_off = max(abs(off - tail_end), abs(tail_off - end)) + if abs_off < OFF_BOUND: + tail_data[OFF_IDX] = min(tail_off, off) + tail_data[LEN_IDX] = max(tail_end, end) - tail_data[OFF_IDX] + continue + else: + ''' + 否则跟之前同一文件和numaid的数据进行合并 + 如若两者之间距离小于设定值， + 就进行合并 + ''' + key = (filename, numaid) + if key in file_map_index: + index = file_map_index[key] + data = self.data_list[index] + data_off = data[OFF_IDX] + data_end = data[OFF_IDX] + data[LEN_IDX] + end = off + length + abs_off = max(abs(off - data_end), abs(data_off - end)) + if abs_off < OFF_BOUND: + data[OFF_IDX] = min(data_off, off) + data[LEN_IDX] = max(data_end, end) - data[OFF_IDX] + continue + self.data_list.append([filename, off, length, numaid]) + file_map_index[(filename, numaid)] = len(self.data_list) - 1 + + +def main(agrv): + """ + Parse the argument + """ + parser = argparse.ArgumentParser('model trace parser') + parser.add_argument('--trace_file', + required=True, + type=str, + help='trace source') + parser.add_argument('--output_file', + required=True, + type=str, + help='output of hint file') + try: + args = parser.parse_args() + trace_file = args.trace_file + output_file = args.output_file + if not os.path.exists(trace_file): + print("{trace_file} not exist") + return -1 + + hint_generator=ReplayHintGenerator(trace_file, output_file) + hint_generator.parse_trace() + hint_generator.generate_hint() + return 0 + except Exception as e: + print(f"Parse model trace exception:{str(e)}") + return -1 + +if __name__ == '__main__': + try: + ret = main(sys.argv[1:]) + except Exception as main_e: + print(str(main_e)) + ret = -1 + sys.exit(ret) + -- 2.34.1

2 1

[PATCH OLK-6.6 0/1] Fix TLBI broadcast optimization loss and incorrect dvmbm handling
by Tian Zheng 13 Mar '26

13 Mar '26

Fix 1 In the TLBI broadcast-optimization feature, the SYS_LSUDVM_CTRL_EL2 register controls whether the feature is enabled. When LPI low-power mode is enabled, this register is cleared as the pCPU powers down into LPI mode, causing the TLBI broadcast optimization to stop working. The fix saves and restores this control register in the callbacks for entering and exiting LPI mode. Fix 2 The current TLBI broadcast-optimization logic only supports normal VMs. If the feature is enabled globally, VMs that are not yet adapted for it, such as CCA will use an incorrect TLBI broadcast bitmap, which can cause a panic in CCA VMs. The fix moves the actual enable/disable of the dvmbm functionality into the vcpu load and vcpu put functions. Tian Zheng (1): KVM: arm64: Fix CCA guest panic when dvmbm is enabled arch/arm64/kvm/hisilicon/hisi_virt.c | 18 ++++++------------ 1 file changed, 6 insertions(+), 12 deletions(-) -- 2.33.0

1 1

[PATCH v13 00/12] arm64: entry: Convert to Generic Entry
by Jinjie Ruan 13 Mar '26

13 Mar '26

Currently, x86, Riscv, Loongarch use the Generic Entry which makes maintainers' work easier and codes more elegant. arm64 has already successfully switched to the Generic IRQ Entry in commit b3cf07851b6c ("arm64: entry: Switch to generic IRQ entry"), it is time to completely convert arm64 to Generic Entry. The goal is to bring arm64 in line with other architectures that already use the generic entry infrastructure, reducing duplicated code and making it easier to share future changes in entry/exit paths, such as "Syscall User Dispatch". This patch set is rebased on v7.0-rc1. And the performance was measured on Kunpeng 920 using "perf bench basic syscall" with "arm64.nopauth selinux=0 audit=1". On the basis of optimizing el0_svc_common(), the performance are below: | Metric | W/O Generic Entry | With Generic Entry opt| Change | | ---------- | ----------------- | ------------------ | ------ | | Total time | 2.487 [sec] | 2.264 [sec] | ↓9.0% | | usecs/op | 0.248780 | 0.226481 | ↓9.0% | | ops/sec | 4,019,620 | 4,415,383 | ↑9.8% | Therefore, after the optimization, ARM64 System Call performance improved by approximately 9%. It was tested ok with following test cases on kunpeng920 and QEMU virt platform: - Stress-ng CPU stress test. - Hackbench stress test. - "sud" selftest testcase. - get_set_sud, get_syscall_info, set_syscall_info, peeksiginfo in tools/testing/selftests/ptrace. - breakpoint_test_arm64 in selftests/breakpoints. - syscall-abi and ptrace in tools/testing/selftests/arm64/abi - fp-ptrace, sve-ptrace, za-ptrace in selftests/arm64/fp. - vdso_test_getrandom in tools/testing/selftests/vDSO - Strace tests. The test QEMU configuration is as follows: qemu-system-aarch64 \ -M virt,gic-version=3,virtualization=on,mte=on \ -cpu max,pauth-impdef=on \ -kernel Image \ -smp 8,sockets=1,cores=4,threads=2 \ -m 512m \ -nographic \ -no-reboot \ -device virtio-rng-pci \ -append "root=/dev/vda rw console=ttyAMA0 kgdboc=ttyAMA0,115200 \ earlycon preempt=voluntary irqchip.gicv3_pseudo_nmi=1" \ -drive if=none,file=images/rootfs.ext4,format=raw,id=hd0 \ -device virtio-blk-device,drive=hd0 \ Changes in v13: - Rebased on v7.0-rc3, so drop the firt applied arm64 patch. - Use generic TIF bits to enables RSEQ optimization. - Update most of the commit message to make it more clear. - Link to v12: https://lore.kernel.org/all/20260203133728.848283-5-ruanjinjie@huawei.com/ Changes in v12: - Rebased on "sched/core", so remove the four generic entry patches. - Move "Expand secure_computing() in place" and "Use syscall_get_arguments() helper" patch forward, which will group all non-functional cleanups at the front. - Adjust the explanation for moving rseq_syscall() before audit_syscall_exit(). - Link to v11: https://lore.kernel.org/all/20260128031934.3906955-1-ruanjinjie@huawei.com/ Changes in v11: - Remove unused syscall in syscall_trace_enter(). - Update and provide a detailed explanation of the differences after moving rseq_syscall() before audit_syscall_exit(). - Rebased on arm64 (for-next/entry), and remove the first applied 3 patchs. - syscall_exit_to_user_mode_work() for arch reuse instead of adding new syscall_exit_to_user_mode_work_prepare() helper. - Link to v10: https://lore.kernel.org/all/20251222114737.1334364-1-ruanjinjie@huawei.com/ Changes in v10: - Rebased on v6.19-rc1, rename syscall_exit_to_user_mode_prepare() to syscall_exit_to_user_mode_work_prepare() to avoid conflict. - Also inline syscall_trace_enter(). - Support aarch64 for sud_benchmark. - Update and correct the commit message. - Add Reviewed-by. - Link to v9: https://lore.kernel.org/all/20251204082123.2792067-1-ruanjinjie@huawei.com/ Changes in v9: - Move "Return early for ptrace_report_syscall_entry() error" patch ahead to make it not introduce a regression. - Not check _TIF_SECCOMP/SYSCALL_EMU for syscall_exit_work() in a separate patch. - Do not report_syscall_exit() for PTRACE_SYSEMU_SINGLESTEP in a separate patch. - Add two performance patch to improve the arm64 performance. - Add Reviewed-by. - Link to v8: https://lore.kernel.org/all/20251126071446.3234218-1-ruanjinjie@huawei.com/ Changes in v8: - Rename "report_syscall_enter()" to "report_syscall_entry()". - Add ptrace_save_reg() to avoid duplication. - Remove unused _TIF_WORK_MASK in a standalone patch. - Align syscall_trace_enter() return value with the generic version. - Use "scno" instead of regs->syscallno in el0_svc_common(). - Move rseq_syscall() ahead in a standalone patch to clarify it clearly. - Rename "syscall_trace_exit()" to "syscall_exit_work()". - Keep the goto in el0_svc_common(). - No argument was passed to __secure_computing() and check -1 not -1L. - Remove "Add has_syscall_work() helper" patch. - Move "Add syscall_exit_to_user_mode_prepare() helper" patch later. - Add miss header for asm/entry-common.h. - Update the implementation of arch_syscall_is_vdso_sigreturn(). - Add "ARCH_SYSCALL_WORK_EXIT" to be defined as "SECCOMP | SYSCALL_EMU" to keep the behaviour unchanged. - Add more testcases test. - Add Reviewed-by. - Update the commit message. - Link to v7: https://lore.kernel.org/all/20251117133048.53182-1-ruanjinjie@huawei.com/ Jinjie Ruan (11): arm64/ptrace: Refactor syscall_trace_enter/exit() to accept flags parameter arm64/ptrace: Expand secure_computing() in place arm64/ptrace: Use syscall_get_arguments() helper for audit arm64: ptrace: Move rseq_syscall() before audit_syscall_exit() arm64: syscall: Introduce syscall_exit_to_user_mode_work() arm64/ptrace: Define and use _TIF_SYSCALL_EXIT_WORK arm64/ptrace: Skip syscall exit reporting for PTRACE_SYSEMU_SINGLESTEP arm64: entry: Convert to generic entry arm64: Inline el0_svc_common() s390: Rename TIF_SINGLE_STEP to TIF_SINGLESTEP arm64: Use generic TIF bits for common thread flags kemal (1): selftests: sud_test: Support aarch64 arch/arm64/Kconfig | 3 +- arch/arm64/include/asm/entry-common.h | 76 ++++++++++++ arch/arm64/include/asm/syscall.h | 19 ++- arch/arm64/include/asm/thread_info.h | 76 ++++-------- arch/arm64/kernel/debug-monitors.c | 7 ++ arch/arm64/kernel/ptrace.c | 115 ------------------ arch/arm64/kernel/signal.c | 2 +- arch/arm64/kernel/syscall.c | 29 ++--- arch/s390/include/asm/thread_info.h | 7 +- arch/s390/kernel/process.c | 2 +- arch/s390/kernel/ptrace.c | 20 +-- arch/s390/kernel/signal.c | 6 +- arch/x86/include/asm/thread_info.h | 6 +- include/asm-generic/thread_info_tif.h | 5 + .../syscall_user_dispatch/sud_benchmark.c | 2 +- .../syscall_user_dispatch/sud_test.c | 4 + 16 files changed, 166 insertions(+), 213 deletions(-) -- 2.34.1

1 12

[PATCH v3 00/10] arm_mpam: Introduce the Narrow-PARTID feature for MPAM driver
by Zeng Heng 13 Mar '26

13 Mar '26

The patchset is based on mpam_resctrl_glue_v5_debugfs branch of https://gitlab.arm.com/linux-arm/linux-bh.git repository. The narrow-partid feature in MPAM allows for a more efficient use of PARTIDs by enabling a many-to-one mapping of reqpartids (requested PARTIDs) to intpartids (internal PARTIDs). This mapping reduces the number of unique PARTIDs needed, thus allowing more tasks or processes to be monitored and managed with the available resources. For a mixture of MSCs system, for MSCs that do not support narrow-partid, we use the PARTIDs exceeding the number of closids as reqPARTIDs for expanding the monitoring groups. Therefore, we will expand the information contained in the RMID, so that it includes not only PMG, but also reqPARTIDs information. The new RMID would be like: RMID = reqPARTID * NUM_PMG + PMG Each control group has m (req)PARTIDs, which are used to expand the number of monitoring groups under one control group. Therefore, the number of monitoring groups is no longer limited by the range of MPAM's PMG, which enhances the extensibility of the system's monitoring capabilities. --- Compared with v2: - Add dynamic reqPARTID allocation implementation. - Patch 4 has been merged into the repository, dropped from this series. Compared with v1: - Redefine the RMID information. - Refactor the resctrl_arch_rmid_idx_decode() and resctrl_arch_rmid_idx_encode(). - Simplify closid_rmid2reqpartid() to rmid2reqpartid() and replace it accordingly. Compared with RFC-v4: - Rebase the patch set on the v6.14-rc1 branch. Compared with RFC-v3: - Add limitation of the Narrow-PARTID feature (See Patch 2). - Remove redundant reqpartid2closid() and reqpartid_pmg2rmid(). - Refactor closid_rmid2reqpartid() partially. - Merge the PARTID conversion-related patches into a single patch for bisectability. - Skip adaptation of resctrl_arch_set_rmid() which is going to be removed. Compared with RFC-v2: - Refactor closid/rmid pair translation. - Simplify the logic of synchronize configuration. - Remove reqPARTID source bitmap. Compared with RFC-v1: - Rebase this patch set on latest MPAM driver of the v6.12-rc1 branch. v2: https://lore.kernel.org/all/20250222112448.2438586-1-zengheng4@huawei.com/ v1: https://lore.kernel.org/all/20250217031852.2014939-1-zengheng4@huawei.com/ RFC-v4: https://lore.kernel.org/all/20250104101224.873926-1-zengheng4@huawei.com/ RFC-v3: https://lore.kernel.org/all/20241207092136.2488426-1-zengheng4@huawei.com/ RFC-v2: https://lore.kernel.org/all/20241119135104.595630-1-zengheng4@huawei.com/ RFC-v1: https://lore.kernel.org/all/20241114135037.918470-1-zengheng4@huawei.com/ --- Ben Horgan (1): arm64/sysreg: Add MPAMSM_EL1 register Zeng Heng (9): arm_mpam: Add intPARTID and reqPARTID support for narrow PARTID feature arm_mpam: Disable Narrow-PARTID when MBA lacks support arm_mpam: Refactor rmid to reqPARTID/PMG mapping arm_mpam: Propagate control group config to sub-monitoring groups fs/resctrl: Add rmid_entry state helpers arm_mpam: Implement dynamic reqPARTID allocation for monitoring groups fs/resctrl: Wire up rmid expansion and reclaim functions arm64/mpam: Add mpam_sync_config() for dynamic rmid expansion fs/resctrl: Prevent rmid parsing errors by flushing limbo on umount Documentation/arch/arm64/index.rst | 1 + Documentation/arch/arm64/mpam.rst | 94 + Documentation/arch/arm64/silicon-errata.rst | 9 + arch/arm64/Kconfig | 6 +- arch/arm64/include/asm/el2_setup.h | 3 +- arch/arm64/include/asm/mpam.h | 96 + arch/arm64/include/asm/resctrl.h | 2 + arch/arm64/include/asm/thread_info.h | 3 + arch/arm64/kernel/Makefile | 1 + arch/arm64/kernel/cpufeature.c | 21 +- arch/arm64/kernel/mpam.c | 62 + arch/arm64/kernel/process.c | 7 + arch/arm64/kvm/hyp/include/hyp/switch.h | 12 +- arch/arm64/kvm/hyp/nvhe/hyp-main.c | 9 + arch/arm64/kvm/hyp/vhe/sysreg-sr.c | 16 + arch/arm64/kvm/sys_regs.c | 2 + arch/arm64/tools/sysreg | 8 + arch/x86/include/asm/resctrl.h | 7 + drivers/resctrl/Kconfig | 9 +- drivers/resctrl/Makefile | 1 + drivers/resctrl/mpam_devices.c | 340 ++- drivers/resctrl/mpam_internal.h | 112 +- drivers/resctrl/mpam_resctrl.c | 2094 +++++++++++++++++++ drivers/resctrl/test_mpam_resctrl.c | 364 ++++ fs/resctrl/monitor.c | 50 +- fs/resctrl/rdtgroup.c | 24 +- include/linux/arm_mpam.h | 49 + include/linux/resctrl.h | 21 + 28 files changed, 3368 insertions(+), 55 deletions(-) create mode 100644 Documentation/arch/arm64/mpam.rst create mode 100644 arch/arm64/include/asm/mpam.h create mode 100644 arch/arm64/include/asm/resctrl.h create mode 100644 arch/arm64/kernel/mpam.c create mode 100644 drivers/resctrl/mpam_resctrl.c create mode 100644 drivers/resctrl/test_mpam_resctrl.c -- 2.25.1

1 12

[PATCH OLK-6.6] arm64/mpam: Fix warning of schedule state in resctrl_arch_mon_ctx_alloc()
by Zeng Heng 13 Mar '26

13 Mar '26

hulk inclusion category: bugfix bugzilla: https://atomgit.com/openeuler/kernel/issues/8686 ------------------ resctrl_arch_mon_ctx_alloc() triggers a scheduling state warning when calling kmalloc() while the task is in a non-running state: do not call blocking ops when !TASK_RUNNING; state=1 set at prepare_to_wait WARNING: CPU: 78 PID: 12592 at kernel/sched/core.c:10584 _might_sleep Call trace: _might_sleep+0x1cc/0xf8 _kmalloc_cache_alloc_node+0x3a0/0x4e0 kmalloc_trace+0x44/0x120 resctrl_arch_mon_ctx_alloc+0xf0/0x178 mon_event_read+0xa0/0x308 mkdir_mondata_subdir+0x24c/0x330 mkdir_mondata_all+0x17c/0x248 rdt_get_tree+0x2e8/0x680 vfs_get_tree+0x5c/0x190 do_new_mount+0x27c/0x408 path_mount+0x2ac/0x448 arm64_sys_mount+0x1d4/0x228 Ensure the allocation is performed in TASK_RUNNING state to avoid schedule state corrupted. Fixes: 9119da143939 ("arm_mpam: resctrl: Allow resctrl to allocate monitors") Signed-off-by: Zeng Heng <zengheng4(a)huawei.com> --- drivers/platform/mpam/mpam_resctrl.c | 25 +++---------------------- 1 file changed, 3 insertions(+), 22 deletions(-) diff --git a/drivers/platform/mpam/mpam_resctrl.c b/drivers/platform/mpam/mpam_resctrl.c index 1a26a0ed2cff..b6681f328aaa 100644 --- a/drivers/platform/mpam/mpam_resctrl.c +++ b/drivers/platform/mpam/mpam_resctrl.c @@ -388,11 +388,11 @@ struct rdt_resource *resctrl_arch_get_resource(enum resctrl_res_level l) return &mpam_resctrl_exports[l].resctrl_res; } -static void *resctrl_arch_mon_ctx_alloc_no_wait(struct rdt_resource *r, - int evtid) +void *resctrl_arch_mon_ctx_alloc(struct rdt_resource *r, int evtid) { - u32 *ret = kmalloc(sizeof(*ret), GFP_KERNEL); + u32 *ret; + ret = kmalloc(sizeof(*ret), GFP_KERNEL); if (!ret) return ERR_PTR(-ENOMEM); @@ -400,25 +400,6 @@ static void *resctrl_arch_mon_ctx_alloc_no_wait(struct rdt_resource *r, return ret; } -void *resctrl_arch_mon_ctx_alloc(struct rdt_resource *r, int evtid) -{ - DEFINE_WAIT(wait); - void *ret; - - might_sleep(); - - do { - prepare_to_wait(&resctrl_mon_ctx_waiters, &wait, - TASK_INTERRUPTIBLE); - ret = resctrl_arch_mon_ctx_alloc_no_wait(r, evtid); - if (PTR_ERR(ret) == -ENOSPC) - schedule(); - } while (PTR_ERR(ret) == -ENOSPC && !signal_pending(current)); - finish_wait(&resctrl_mon_ctx_waiters, &wait); - - return ret; -} - void resctrl_arch_mon_ctx_free(struct rdt_resource *r, int evtid, void *arch_mon_ctx) { -- 2.25.1

2 1

[PATCH OLK-6.6] selftests/bpf: Add cross-build support for urandom_read et al
by Pu Lehui 12 Mar '26

12 Mar '26

From: Björn Töpel <bjorn(a)rivosinc.com> mainline inclusion from mainline-v6.7-rc1 commit 97a79e502e25e27a65b0506c9fb210cb2d89b52e category: bugfix bugzilla: https://atomgit.com/openeuler/kernel/issues/8638 Reference: https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?… -------------------------------- Some userland programs in the BPF test suite, e.g. urandom_read, is missing cross-build support. Add cross-build support for these programs Signed-off-by: Björn Töpel <bjorn(a)rivosinc.com> Signed-off-by: Andrii Nakryiko <andrii(a)kernel.org> Link: https://lore.kernel.org/bpf/20231004122721.54525-2-bjorn@kernel.org Conflicts: tools/testing/selftests/bpf/Makefile [ctx conflicts] Signed-off-by: Pu Lehui <pulehui(a)huawei.com> --- tools/testing/selftests/bpf/Makefile | 8 +++++--- 1 file changed, 5 insertions(+), 3 deletions(-) diff --git a/tools/testing/selftests/bpf/Makefile b/tools/testing/selftests/bpf/Makefile index 8ce50befca91..e97ec89fe83d 100644 --- a/tools/testing/selftests/bpf/Makefile +++ b/tools/testing/selftests/bpf/Makefile @@ -208,15 +208,17 @@ endif # do not fail. Static builds leave urandom_read relying on system-wide shared libraries. $(OUTPUT)/liburandom_read.so: urandom_read_lib1.c urandom_read_lib2.c $(call msg,LIB,,$@) - $(Q)$(CLANG) $(filter-out -static,$(CFLAGS) $(LDFLAGS)) \ + $(Q)$(CLANG) $(CLANG_TARGET_ARCH) \ + $(filter-out -static,$(CFLAGS) $(LDFLAGS)) \ $^ $(filter-out -static,$(LDLIBS)) \ -fuse-ld=$(LLD) -Wl,-znoseparate-code -Wl,--build-id=sha1 \ -fPIC -shared -o $@ $(OUTPUT)/urandom_read: urandom_read.c urandom_read_aux.c $(OUTPUT)/liburandom_read.so $(call msg,BINARY,,$@) - $(Q)$(CLANG) $(filter-out -static,$(CFLAGS) $(LDFLAGS)) $(filter %.c,$^) \ - -lurandom_read $(filter-out -static,$(LDLIBS)) -L$(OUTPUT) \ + $(Q)$(CLANG) $(CLANG_TARGET_ARCH) \ + $(filter-out -static,$(CFLAGS) $(LDFLAGS)) $(filter %.c,$^) \ + -lurandom_read $(filter-out -static,$(LDLIBS)) -L$(OUTPUT) \ -fuse-ld=$(LLD) -Wl,-znoseparate-code -Wl,--build-id=sha1 \ -Wl,-rpath=. -o $@ -- 2.34.1

2 1

[PATCH OLK-6.6] arm64/mpam: Fix warning of schedule state in resctrl_arch_mon_ctx_alloc()
by Zeng Heng 12 Mar '26

12 Mar '26

Offering: HULK hulk inclusion category: bugfix bugzilla: https://atomgit.com/openeuler/kernel/issues/8686 ------------------ resctrl_arch_mon_ctx_alloc() triggers a scheduling state warning when calling kmalloc() while the task is in a non-running state: do not call blocking ops when !TASK_RUNNING; state=1 set at prepare_to_wait WARNING: CPU: 78 PID: 12592 at kernel/sched/core.c:10584 _might_sleep Call trace: _might_sleep+0x1cc/0xf8 _kmalloc_cache_alloc_node+0x3a0/0x4e0 kmalloc_trace+0x44/0x120 resctrl_arch_mon_ctx_alloc+0xf0/0x178 mon_event_read+0xa0/0x308 mkdir_mondata_subdir+0x24c/0x330 mkdir_mondata_all+0x17c/0x248 rdt_get_tree+0x2e8/0x680 vfs_get_tree+0x5c/0x190 do_new_mount+0x27c/0x408 path_mount+0x2ac/0x448 arm64_sys_mount+0x1d4/0x228 Ensure the allocation is performed in TASK_RUNNING state to avoid schedule state corrupted. Fixes: 9119da143939 ("arm_mpam: resctrl: Allow resctrl to allocate monitors") Signed-off-by: Zeng Heng <zengheng4(a)huawei.com> --- drivers/platform/mpam/mpam_resctrl.c | 25 +++---------------------- 1 file changed, 3 insertions(+), 22 deletions(-) diff --git a/drivers/platform/mpam/mpam_resctrl.c b/drivers/platform/mpam/mpam_resctrl.c index 1a26a0ed2cff..b6681f328aaa 100644 --- a/drivers/platform/mpam/mpam_resctrl.c +++ b/drivers/platform/mpam/mpam_resctrl.c @@ -388,11 +388,11 @@ struct rdt_resource *resctrl_arch_get_resource(enum resctrl_res_level l) return &mpam_resctrl_exports[l].resctrl_res; } -static void *resctrl_arch_mon_ctx_alloc_no_wait(struct rdt_resource *r, - int evtid) +void *resctrl_arch_mon_ctx_alloc(struct rdt_resource *r, int evtid) { - u32 *ret = kmalloc(sizeof(*ret), GFP_KERNEL); + u32 *ret; + ret = kmalloc(sizeof(*ret), GFP_KERNEL); if (!ret) return ERR_PTR(-ENOMEM); @@ -400,25 +400,6 @@ static void *resctrl_arch_mon_ctx_alloc_no_wait(struct rdt_resource *r, return ret; } -void *resctrl_arch_mon_ctx_alloc(struct rdt_resource *r, int evtid) -{ - DEFINE_WAIT(wait); - void *ret; - - might_sleep(); - - do { - prepare_to_wait(&resctrl_mon_ctx_waiters, &wait, - TASK_INTERRUPTIBLE); - ret = resctrl_arch_mon_ctx_alloc_no_wait(r, evtid); - if (PTR_ERR(ret) == -ENOSPC) - schedule(); - } while (PTR_ERR(ret) == -ENOSPC && !signal_pending(current)); - finish_wait(&resctrl_mon_ctx_waiters, &wait); - - return ret; -} - void resctrl_arch_mon_ctx_free(struct rdt_resource *r, int evtid, void *arch_mon_ctx) { -- 2.25.1

2 1

[PATCH OLK-6.6] Revert "drivers: core: synchronize really_probe() and dev_uevent()"
by Lin Ruifeng 12 Mar '26

12 Mar '26

From: Dmitry Torokhov <dmitry.torokhov(a)gmail.com> stable inclusion from stable-v6.6.89 commit d29c2d53357fd35f24b7acba594725a5be6404d7 category: bugfix bugzilla: https://atomgit.com/openeuler/kernel/issues/8715 Reference: https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git/commit/?id… -------------------------------- commit dc1771f718548f7d4b93991b174c6e7b5e1ba410 upstream. This reverts commit c0a40097f0bc81deafc15f9195d1fb54595cd6d0. Probing a device can take arbitrary long time. In the field we observed that, for example, probing a bad micro-SD cards in an external USB card reader (or maybe cards were good but cables were flaky) sometimes takes longer than 2 minutes due to multiple retries at various levels of the stack. We can not block uevent_show() method for that long because udev is reading that attribute very often and that blocks udev and interferes with booting of the system. The change that introduced locking was concerned with dev_uevent() racing with unbinding the driver. However we can handle it without locking (which will be done in subsequent patch). There was also claim that synchronization with probe() is needed to properly load USB drivers, however this is a red herring: the change adding the lock was introduced in May of last year and USB loading and probing worked properly for many years before that. Revert the harmful locking. Cc: stable(a)vger.kernel.org Signed-off-by: Dmitry Torokhov <dmitry.torokhov(a)gmail.com> Reviewed-by: Masami Hiramatsu (Google) <mhiramat(a)kernel.org> Link: https://lore.kernel.org/r/20250311052417.1846985-1-dmitry.torokhov@gmail.com Signed-off-by: Greg Kroah-Hartman <gregkh(a)linuxfoundation.org> Signed-off-by: Lin Ruifeng <linruifeng4(a)huawei.com> --- drivers/base/core.c | 3 --- 1 file changed, 3 deletions(-) diff --git a/drivers/base/core.c b/drivers/base/core.c index 9872ce765eaa..8a4c3bc7bafa 100644 --- a/drivers/base/core.c +++ b/drivers/base/core.c @@ -2707,11 +2707,8 @@ static ssize_t uevent_show(struct device *dev, struct device_attribute *attr, if (!env) return -ENOMEM; - /* Synchronize with really_probe() */ - device_lock(dev); /* let the kset specific function add its keys */ retval = kset->uevent_ops->uevent(&dev->kobj, env); - device_unlock(dev); if (retval) goto out; -- 2.43.0

2 1

[PATCH OLK-6.6 0/3] Fix RT priority Setting
by Liu Kai 12 Mar '26

12 Mar '26

Fix RT priority Setting Liu Kai (3): xsched: refactor xse_this_grp_xcu to xse_this_cfs_rq for clarity and correctness xsched: fix typo in xcu cgroup subsys released callback name xsched: initialize tgid before configuring RT priority include/linux/xsched.h | 29 ++++++++++++++--------------- kernel/xsched/cfs.c | 2 +- kernel/xsched/cgroup.c | 4 ++-- kernel/xsched/core.c | 3 +-- 4 files changed, 18 insertions(+), 20 deletions(-) -- 2.34.1

2 4

[PATCH OLK-6.6 0/3] optimize lightweight load balancing
by Chen Jinghuang 12 Mar '26

12 Mar '26

*** BLURB HERE *** Chen Jinghuang (2): sched: Fix kabi breakage of struct sched_domain for newidle_stats sched: expand sysctl_sched_features to unsigned long Peter Zijlstra (Intel) (1): sched/fair: Proportional newidle balance include/linux/sched/topology.h | 5 ++++ kernel/sched/core.c | 5 +++- kernel/sched/fair.c | 50 +++++++++++++++++++++++++++++++--- kernel/sched/features.h | 5 ++++ kernel/sched/sched.h | 9 +++++- kernel/sched/topology.c | 6 ++++ 6 files changed, 74 insertions(+), 6 deletions(-) -- 2.34.1

2 4

[PATCH OLK-5.10] scsi: hisi_sas: Change the debugfs parameters dump count default value
by Yihang Li 12 Mar '26

12 Mar '26

driver inclusion category: feature bugzilla: https://atomgit.com/openeuler/kernel/issues/8713 ----------------------------------------------------------------------------- The memory allocation time of the DFX is changed from device initialization to dump occurs, so restore the default setting hisi_sas_debugfs_dump_count to 50. Signed-off-by: Yihang Li <liyihang9(a)h-partners.com> --- drivers/scsi/hisi_sas/hisi_sas_main.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/drivers/scsi/hisi_sas/hisi_sas_main.c b/drivers/scsi/hisi_sas/hisi_sas_main.c index a525d484d6e2..c58b77902782 100644 --- a/drivers/scsi/hisi_sas/hisi_sas_main.c +++ b/drivers/scsi/hisi_sas/hisi_sas_main.c @@ -2955,12 +2955,12 @@ EXPORT_SYMBOL_GPL(hisi_sas_remove); #if IS_ENABLED(CONFIG_SCSI_HISI_SAS_DEBUGFS_DEFAULT_ENABLE) #define DEBUGFS_ENABLE_DEFAULT "enabled" bool hisi_sas_debugfs_enable = true; +u32 hisi_sas_debugfs_dump_count = 50; #else #define DEBUGFS_ENABLE_DEFAULT "disabled" bool hisi_sas_debugfs_enable; -#endif - u32 hisi_sas_debugfs_dump_count = 1; +#endif EXPORT_SYMBOL_GPL(hisi_sas_debugfs_enable); module_param_named(debugfs_enable, hisi_sas_debugfs_enable, bool, 0444); -- 2.33.0

2 1

[PATCH OLK-5.10] nfsd: provide locking for v4_end_grace
by Li Lingfeng 12 Mar '26

12 Mar '26

From: NeilBrown <neil(a)brown.name> stable inclusion from stable-v5.10.248 commit ca97360860eb02e3ae4ba42c19b439a0fcecbf06 category: bugfix bugzilla: https://atomgit.com/src-openeuler/kernel/issues/13489 CVE: CVE-2026-22980 Reference: https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git/commit/?id… -------------------------------- [ Upstream commit 2857bd59feb63fcf40fe4baf55401baea6b4feb4 ] Writing to v4_end_grace can race with server shutdown and result in memory being accessed after it was freed - reclaim_str_hashtbl in particularly. We cannot hold nfsd_mutex across the nfsd4_end_grace() call as that is held while client_tracking_op->init() is called and that can wait for an upcall to nfsdcltrack which can write to v4_end_grace, resulting in a deadlock. nfsd4_end_grace() is also called by the landromat work queue and this doesn't require locking as server shutdown will stop the work and wait for it before freeing anything that nfsd4_end_grace() might access. However, we must be sure that writing to v4_end_grace doesn't restart the work item after shutdown has already waited for it. For this we add a new flag protected with nn->client_lock. It is set only while it is safe to make client tracking calls, and v4_end_grace only schedules work while the flag is set with the spinlock held. So this patch adds a nfsd_net field "client_tracking_active" which is set as described. Another field "grace_end_forced", is set when v4_end_grace is written. After this is set, and providing client_tracking_active is set, the laundromat is scheduled. This "grace_end_forced" field bypasses other checks for whether the grace period has finished. This resolves a race which can result in use-after-free. Reported-by: Li Lingfeng <lilingfeng3(a)huawei.com> Closes: https://lore.kernel.org/linux-nfs/20250623030015.2353515-1-neil@brown.name/… Fixes: 7f5ef2e900d9 ("nfsd: add a v4_end_grace file to /proc/fs/nfsd") Cc: stable(a)vger.kernel.org Signed-off-by: NeilBrown <neil(a)brown.name> Tested-by: Li Lingfeng <lilingfeng3(a)huawei.com> Reviewed-by: Jeff Layton <jlayton(a)kernel.org> Signed-off-by: Chuck Lever <chuck.lever(a)oracle.com> [ Adjust context ] Signed-off-by: Sasha Levin <sashal(a)kernel.org> Signed-off-by: Greg Kroah-Hartman <gregkh(a)linuxfoundation.org> Conflicts: fs/nfsd/nfs4state.c [Commit 5aa2c4a1fe28 ("NFSD: Add a nfsd4_file_hash_remove() helper") add declaration of nfsd4_file_hash_remove(); commit ea92c0768f98 ("nfsd: simplify nfsd_renew") change the return value; commit 438ef64bbfe4 ("NFSD: register/unregister of nfsd-client shrinker at nfsd startup/shutdown time") add unregister_shrinker() in nfs4_state_shutdown_net(); commit 67ef9e5fd737 ("NFSD: add courteous server support for thread with only delegation") move the position of declaration of laundry_wq.] Signed-off-by: Li Lingfeng <lilingfeng3(a)huawei.com> --- fs/nfsd/netns.h | 2 ++ fs/nfsd/nfs4state.c | 44 +++++++++++++++++++++++++++++++++++++++++--- fs/nfsd/nfsctl.c | 3 +-- fs/nfsd/state.h | 2 +- 4 files changed, 45 insertions(+), 6 deletions(-) diff --git a/fs/nfsd/netns.h b/fs/nfsd/netns.h index 8500acac09e0..ca5c386aafce 100644 --- a/fs/nfsd/netns.h +++ b/fs/nfsd/netns.h @@ -40,6 +40,8 @@ struct nfsd_net { struct lock_manager nfsd4_manager; bool grace_ended; + bool grace_end_forced; + bool client_tracking_active; time64_t boot_time; struct dentry *nfsd_client_dir; diff --git a/fs/nfsd/nfs4state.c b/fs/nfsd/nfs4state.c index 302bbad5cdc6..d17702a4a1ad 100644 --- a/fs/nfsd/nfs4state.c +++ b/fs/nfsd/nfs4state.c @@ -80,7 +80,7 @@ static u64 current_sessionid = 1; /* forward declarations */ static bool check_for_locks(struct nfs4_file *fp, struct nfs4_lockowner *lowner); static void nfs4_free_ol_stateid(struct nfs4_stid *stid); -void nfsd4_end_grace(struct nfsd_net *nn); +static void nfsd4_end_grace(struct nfsd_net *nn); static void _free_cpntf_state_locked(struct nfsd_net *nn, struct nfs4_cpntf_state *cps); /* Locking: */ @@ -5347,7 +5347,7 @@ nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, } void -nfsd4_end_grace(struct nfsd_net *nn) +static nfsd4_end_grace(struct nfsd_net *nn) { /* do nothing if grace period already ended */ if (nn->grace_ended) @@ -5379,6 +5379,34 @@ nfsd4_end_grace(struct nfsd_net *nn) */ } +static struct workqueue_struct *laundry_wq; +/** + * nfsd4_force_end_grace - forcibly end the NFSv4 grace period + * @nn: network namespace for the server instance to be updated + * + * Forces bypass of normal grace period completion, then schedules + * the laundromat to end the grace period immediately. Does not wait + * for the grace period to fully terminate before returning. + * + * Return values: + * %true: Grace termination schedule + * %false: No action was taken + */ +bool nfsd4_force_end_grace(struct nfsd_net *nn) +{ + if (!nn->client_tracking_ops) + return false; + spin_lock(&nn->client_lock); + if (nn->grace_ended || !nn->client_tracking_active) { + spin_unlock(&nn->client_lock); + return false; + } + WRITE_ONCE(nn->grace_end_forced, true); + mod_delayed_work(laundry_wq, &nn->laundromat_work, 0); + spin_unlock(&nn->client_lock); + return true; +} + /* * If we've waited a lease period but there are still clients trying to * reclaim, wait a little longer to give them a chance to finish. @@ -5388,6 +5416,8 @@ static bool clients_still_reclaiming(struct nfsd_net *nn) time64_t double_grace_period_end = nn->boot_time + 2 * nn->nfsd4_lease; + if (READ_ONCE(nn->grace_end_forced)) + return false; if (nn->track_reclaim_completes && atomic_read(&nn->nr_reclaim_complete) == nn->reclaim_str_hashtbl_size) @@ -5530,7 +5560,6 @@ nfs4_laundromat(struct nfsd_net *nn) return new_timeo; } -static struct workqueue_struct *laundry_wq; static void laundromat_main(struct work_struct *); static void @@ -7343,6 +7372,8 @@ static int nfs4_state_create_net(struct net *net) nn->unconf_name_tree = RB_ROOT; nn->boot_time = ktime_get_real_seconds(); nn->grace_ended = false; + nn->grace_end_forced = false; + nn->client_tracking_active = false; nn->nfsd4_manager.block_opens = true; INIT_LIST_HEAD(&nn->nfsd4_manager.list); INIT_LIST_HEAD(&nn->client_lru); @@ -7409,6 +7440,10 @@ nfs4_state_start_net(struct net *net) return ret; locks_start_grace(net, &nn->nfsd4_manager); nfsd4_client_tracking_init(net); + /* safe for laundromat to run now */ + spin_lock(&nn->client_lock); + nn->client_tracking_active = true; + spin_unlock(&nn->client_lock); if (nn->track_reclaim_completes && nn->reclaim_str_hashtbl_size == 0) goto skip_grace; printk(KERN_INFO "NFSD: starting %lld-second grace period (net %x)\n", @@ -7458,6 +7493,9 @@ nfs4_state_shutdown_net(struct net *net) struct nfsd_net *nn = net_generic(net, nfsd_net_id); cancel_delayed_work_sync(&nn->laundromat_work); + spin_lock(&nn->client_lock); + nn->client_tracking_active = false; + spin_unlock(&nn->client_lock); locks_end_grace(&nn->nfsd4_manager); INIT_LIST_HEAD(&reaplist); diff --git a/fs/nfsd/nfsctl.c b/fs/nfsd/nfsctl.c index 040340080b3d..e84c8cf8e2c0 100644 --- a/fs/nfsd/nfsctl.c +++ b/fs/nfsd/nfsctl.c @@ -1133,9 +1133,8 @@ static ssize_t write_v4_end_grace(struct file *file, char *buf, size_t size) case 'Y': case 'y': case '1': - if (!nn->nfsd_serv) + if (!nfsd4_force_end_grace(nn)) return -EBUSY; - nfsd4_end_grace(nn); break; default: return -EINVAL; diff --git a/fs/nfsd/state.h b/fs/nfsd/state.h index 9eae11a9d21c..0e4f98b41364 100644 --- a/fs/nfsd/state.h +++ b/fs/nfsd/state.h @@ -683,7 +683,7 @@ static inline void get_nfs4_file(struct nfs4_file *fi) struct nfsd_file *find_any_file(struct nfs4_file *f); /* grace period management */ -void nfsd4_end_grace(struct nfsd_net *nn); +bool nfsd4_force_end_grace(struct nfsd_net *nn); /* nfs4recover operations */ extern int nfsd4_client_tracking_init(struct net *net); -- 2.52.0

2 1

[PATCH OLK-6.6 0/2] arm64/mpam: Fix wrong closid config access when CDP is enabled
by Zeng Heng 12 Mar '26

12 Mar '26

Zeng Heng (2): Revert "fs/resctrl: Fix configuration to wrong control group when CDP is enabled" arm64/mpam: Fix wrong closid config access when CDP is enabled arch/x86/include/asm/resctrl.h | 4 ---- drivers/platform/mpam/mpam_resctrl.c | 35 +++++++++++++++++++++------- fs/resctrl/ctrlmondata.c | 28 +++------------------- include/linux/arm_mpam.h | 1 - 4 files changed, 29 insertions(+), 39 deletions(-) -- 2.25.1

2 3

[PATCH OLK-6.6 0/2] Fix TLBI broadcast optimization loss and incorrect dvmbm handling
by Tian Zheng 11 Mar '26

11 Mar '26

Fix 1 In the TLBI broadcast-optimization feature, the SYS_LSUDVM_CTRL_EL2 register controls whether the feature is enabled. When LPI low-power mode is enabled, this register is cleared as the pCPU powers down into LPI mode, causing the TLBI broadcast optimization to stop working. The fix saves and restores this control register in the callbacks for entering and exiting LPI mode. Fix 2 The current TLBI broadcast-optimization logic only supports normal VMs. If the feature is enabled globally, VMs that are not yet adapted for it, such as CCA will use an incorrect TLBI broadcast bitmap, which can cause a panic in CCA VMs. The fix moves the actual enable/disable of the dvmbm functionality into the vcpu load and vcpu put functions. Tian Zheng (2): KVM: arm64: Fix TLBI optimization broken in LPI mode KVM: arm64: Fix CCA guest panic when dvmbm is enabled arch/arm64/kvm/arm.c | 4 +++ arch/arm64/kvm/hisilicon/hisi_virt.c | 37 +++++++++++++++++++--------- arch/arm64/kvm/hisilicon/hisi_virt.h | 4 +++ 3 files changed, 33 insertions(+), 12 deletions(-) -- 2.33.0

1 2

[PATCH OLK-6.6 0/3] optimize lightweight load balancing
by Chen Jinghuang 11 Mar '26

11 Mar '26

*** BLURB HERE *** Chen Jinghuang (2): sched: Fix kabi breakage of struct sched_domain for newidle_stats sched: expand sysctl_sched_features to unsigned long Peter Zijlstra (Intel) (1): sched/fair: Proportional newidle balance include/linux/sched/topology.h | 7 ++++- kernel/sched/core.c | 5 +++- kernel/sched/fair.c | 48 +++++++++++++++++++++++++++++++--- kernel/sched/features.h | 5 ++++ kernel/sched/sched.h | 9 ++++++- kernel/sched/topology.c | 6 +++++ 6 files changed, 73 insertions(+), 7 deletions(-) -- 2.34.1

2 4

[PATCH OLK-6.6 0/3] optimize lightweight load balancing
by Chen Jinghuang 11 Mar '26

11 Mar '26

*** BLURB HERE *** Chen Jinghuang (2): sched: Fix kabi breakage of struct sched_domain for newidle_stats sched: expand sysctl_sched_features to unsigned long Peter Zijlstra (Intel) (1): sched/fair: Proportional newidle balance include/linux/sched/topology.h | 7 ++++- kernel/sched/core.c | 5 +++- kernel/sched/fair.c | 48 +++++++++++++++++++++++++++++++--- kernel/sched/features.h | 5 ++++ kernel/sched/sched.h | 7 +++++ kernel/sched/topology.c | 6 +++++ 6 files changed, 72 insertions(+), 6 deletions(-) -- 2.34.1

2 4

[PATCH OLK-6.6 0/2] Fix TLBI broadcast optimization loss and incorrect dvmbm handling
by Tian Zheng 11 Mar '26

11 Mar '26

Fix 1 In the TLBI broadcast-optimization feature, the SYS_LSUDVM_CTRL_EL2 register controls whether the feature is enabled. When LPI low-power mode is enabled, this register is cleared as the pCPU powers down into LPI mode, causing the TLBI broadcast optimization to stop working. The fix saves and restores this control register in the callbacks for entering and exiting LPI mode. Fix 2 The current TLBI broadcast-optimization logic only supports normal VMs. If the feature is enabled globally, VMs that are not yet adapted for it, such as CCA will use an incorrect TLBI broadcast bitmap, which can cause a panic in CCA VMs. The fix moves the actual enable/disable of the dvmbm functionality into the vcpu load and vcpu put functions. Tian Zheng (2): KVM: arm64: Fix TLBI optimization broken in LPI mode KVM: arm64: Fix CCA guest panic when dvmbm is enabled arch/arm64/kvm/arm.c | 4 ++++ arch/arm64/kvm/hisilicon/hisi_virt.c | 28 ++++++++++++++++++++++------ arch/arm64/kvm/hisilicon/hisi_virt.h | 4 ++++ 3 files changed, 30 insertions(+), 6 deletions(-) -- 2.33.0

1 2

[PATCH OLK-6.6] platform/x86: classmate-laptop: Add missing NULL pointer checks
by Zeng Heng 11 Mar '26

11 Mar '26

From: "Rafael J. Wysocki" <rafael.j.wysocki(a)intel.com> mainline inclusion from mainline-v6.19 commit fe747d7112283f47169e9c16e751179a9b38611e category: bugfix bugzilla: https://atomgit.com/src-openeuler/kernel/issues/13842 CVE: CVE-2026-23237 Reference: https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git/commit/?id… -------------------------------- In a few places in the Classmate laptop driver, code using the accel object may run before that object's address is stored in the driver data of the input device using it. For example, cmpc_accel_sensitivity_store_v4() is the "show" method of cmpc_accel_sensitivity_attr_v4 which is added in cmpc_accel_add_v4(), before calling dev_set_drvdata() for inputdev->dev. If the sysfs attribute is accessed prematurely, the dev_get_drvdata(&inputdev->dev) call in in cmpc_accel_sensitivity_store_v4() returns NULL which leads to a NULL pointer dereference going forward. Moreover, sysfs attributes using the input device are added before initializing that device by cmpc_add_acpi_notify_device() and if one of them is accessed before running that function, a NULL pointer dereference will occur. For example, cmpc_accel_sensitivity_attr_v4 is added before calling cmpc_add_acpi_notify_device() and if it is read prematurely, the dev_get_drvdata(&acpi->dev) call in cmpc_accel_sensitivity_show_v4() returns NULL which leads to a NULL pointer dereference going forward. Fix this by adding NULL pointer checks in all of the relevant places. Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki(a)intel.com> Link: https://patch.msgid.link/12825381.O9o76ZdvQC@rafael.j.wysocki Reviewed-by: Ilpo Järvinen <ilpo.jarvinen(a)linux.intel.com> Signed-off-by: Ilpo Järvinen <ilpo.jarvinen(a)linux.intel.com> Conflicts: drivers/platform/x86/classmate-laptop.c [Context conflicts.] Signed-off-by: Zeng Heng <zengheng4(a)huawei.com> --- drivers/platform/x86/classmate-laptop.c | 32 +++++++++++++++++++++++++ 1 file changed, 32 insertions(+) diff --git a/drivers/platform/x86/classmate-laptop.c b/drivers/platform/x86/classmate-laptop.c index 2edaea2492df..053c8a86d5ec 100644 --- a/drivers/platform/x86/classmate-laptop.c +++ b/drivers/platform/x86/classmate-laptop.c @@ -208,7 +208,12 @@ static ssize_t cmpc_accel_sensitivity_show_v4(struct device *dev, acpi = to_acpi_device(dev); inputdev = dev_get_drvdata(&acpi->dev); + if (!inputdev) + return -ENXIO; + accel = dev_get_drvdata(&inputdev->dev); + if (!accel) + return -ENXIO; return sprintf(buf, "%d\n", accel->sensitivity); } @@ -225,7 +230,12 @@ static ssize_t cmpc_accel_sensitivity_store_v4(struct device *dev, acpi = to_acpi_device(dev); inputdev = dev_get_drvdata(&acpi->dev); + if (!inputdev) + return -ENXIO; + accel = dev_get_drvdata(&inputdev->dev); + if (!accel) + return -ENXIO; r = kstrtoul(buf, 0, &sensitivity); if (r) @@ -257,7 +267,12 @@ static ssize_t cmpc_accel_g_select_show_v4(struct device *dev, acpi = to_acpi_device(dev); inputdev = dev_get_drvdata(&acpi->dev); + if (!inputdev) + return -ENXIO; + accel = dev_get_drvdata(&inputdev->dev); + if (!accel) + return -ENXIO; return sprintf(buf, "%d\n", accel->g_select); } @@ -274,7 +289,12 @@ static ssize_t cmpc_accel_g_select_store_v4(struct device *dev, acpi = to_acpi_device(dev); inputdev = dev_get_drvdata(&acpi->dev); + if (!inputdev) + return -ENXIO; + accel = dev_get_drvdata(&inputdev->dev); + if (!accel) + return -ENXIO; r = kstrtoul(buf, 0, &g_select); if (r) @@ -303,6 +323,8 @@ static int cmpc_accel_open_v4(struct input_dev *input) acpi = to_acpi_device(input->dev.parent); accel = dev_get_drvdata(&input->dev); + if (!accel) + return -ENXIO; cmpc_accel_set_sensitivity_v4(acpi->handle, accel->sensitivity); cmpc_accel_set_g_select_v4(acpi->handle, accel->g_select); @@ -551,7 +573,12 @@ static ssize_t cmpc_accel_sensitivity_show(struct device *dev, acpi = to_acpi_device(dev); inputdev = dev_get_drvdata(&acpi->dev); + if (!inputdev) + return -ENXIO; + accel = dev_get_drvdata(&inputdev->dev); + if (!accel) + return -ENXIO; return sprintf(buf, "%d\n", accel->sensitivity); } @@ -568,7 +595,12 @@ static ssize_t cmpc_accel_sensitivity_store(struct device *dev, acpi = to_acpi_device(dev); inputdev = dev_get_drvdata(&acpi->dev); + if (!inputdev) + return -ENXIO; + accel = dev_get_drvdata(&inputdev->dev); + if (!accel) + return -ENXIO; r = kstrtoul(buf, 0, &sensitivity); if (r) -- 2.25.1

2 1

[PATCH OLK-6.6 0/2] optimize lightweight load balancing
by Chen Jinghuang 11 Mar '26

11 Mar '26

*** BLURB HERE *** Chen Jinghuang (1): sched: Fix kabi breakage of struct sched_domain for newidle_stats Peter Zijlstra (Intel) (1): sched/fair: Proportional newidle balance include/linux/sched/topology.h | 7 ++++- kernel/sched/core.c | 3 +++ kernel/sched/fair.c | 48 +++++++++++++++++++++++++++++++--- kernel/sched/features.h | 5 ++++ kernel/sched/sched.h | 7 +++++ kernel/sched/topology.c | 6 +++++ 6 files changed, 71 insertions(+), 5 deletions(-) -- 2.34.1

2 3

[PATCH OLK-6.6 0/2] optimize lightweight load balancing
by Chen Jinghuang 11 Mar '26

11 Mar '26

*** BLURB HERE *** Chen Jinghuang (1): sched: Fix kabi breakage of struct sched_domain for newidle_stats Peter Zijlstra (Intel) (1): sched/fair: Proportional newidle balance include/linux/sched/topology.h | 8 +++++- kernel/sched/core.c | 3 +++ kernel/sched/fair.c | 48 +++++++++++++++++++++++++++++++--- kernel/sched/features.h | 5 ++++ kernel/sched/sched.h | 7 +++++ kernel/sched/topology.c | 6 +++++ 6 files changed, 72 insertions(+), 5 deletions(-) -- 2.34.1

2 3

[PATCH] arch/riscv: Add bitrev.h file to support rev8 and brev8
by Jinjie Ruan 11 Mar '26

11 Mar '26

The RISC-V Bit-manipulation Extension for Cryptography (Zbkb) provides the 'brev8' instruction, which reverses the bits within each byte. Combined with the 'rev8' instruction (from Zbb or Zbkb), which reverses the byte order of a register, we can efficiently implement 16-bit, 32-bit, and (on RV64) 64-bit bit reversal. This is significantly faster than the default software table-lookup implementation in lib/bitrev.c, as it replaces memory accesses and multiple arithmetic operations with just two or three hardware instructions. Select HAVE_ARCH_BITREVERSE and provide <asm/bitrev.h> to utilize these instructions when the Zbkb extension is available at runtime via the alternatives mechanism. Signed-off-by: Jinjie Ruan <ruanjinjie(a)huawei.com> --- arch/riscv/Kconfig | 1 + arch/riscv/include/asm/bitrev.h | 46 +++++++++++++++++++++++++++++++++ lib/bitrev.c | 3 --- 3 files changed, 47 insertions(+), 3 deletions(-) create mode 100644 arch/riscv/include/asm/bitrev.h diff --git a/arch/riscv/Kconfig b/arch/riscv/Kconfig index 90c531e6abf5..e7a04b96c0c8 100644 --- a/arch/riscv/Kconfig +++ b/arch/riscv/Kconfig @@ -128,6 +128,7 @@ config RISCV select HAS_IOPORT if MMU select HAVE_ALIGNED_STRUCT_PAGE select HAVE_ARCH_AUDITSYSCALL + select HAVE_ARCH_BITREVERSE select HAVE_ARCH_HUGE_VMALLOC if HAVE_ARCH_HUGE_VMAP select HAVE_ARCH_HUGE_VMAP if MMU && 64BIT select HAVE_ARCH_JUMP_LABEL if !XIP_KERNEL diff --git a/arch/riscv/include/asm/bitrev.h b/arch/riscv/include/asm/bitrev.h new file mode 100644 index 000000000000..b1977600638a --- /dev/null +++ b/arch/riscv/include/asm/bitrev.h @@ -0,0 +1,46 @@ +/* SPDX-License-Identifier: GPL-2.0 */ +#ifndef __ASM_BITREV_H +#define __ASM_BITREV_H + +#include <linux/types.h> +#include <asm/cpufeature-macros.h> +#include <asm/hwcap.h> + +extern u8 const byte_rev_table[256]; +static __always_inline __attribute_const__ u32 __arch_bitrev32(u32 x) +{ + if (IS_ENABLED(CONFIG_RISCV_ISA_ZBKB) && + riscv_has_extension_likely(RISCV_ISA_EXT_ZBKB)) { + unsigned long result = x; + + asm volatile( + ".option push\n" + ".option arch,+zbkb\n" + "rev8 %0, %0\n" + "brev8 %0, %0\n" + ".option pop" + : "+r" (result) + ); + + if (__riscv_xlen == 64) + return (u32)(result >> 32); + + return (u32)result; + } + + return (u32)byte_rev_table[x & 0xff] << 24 | + (u32)byte_rev_table[(x >> 8) & 0xff] << 16 | + (u32)byte_rev_table[(x >> 16) & 0xff] << 8 | + (u32)byte_rev_table[x >> 24]; +} + +static __always_inline __attribute_const__ u16 __arch_bitrev16(u16 x) +{ + return __arch_bitrev32((u32)x) >> 16; +} + +static __always_inline __attribute_const__ u8 __arch_bitrev8(u8 x) +{ + return __arch_bitrev32((u32)x) >> 24; +} +#endif diff --git a/lib/bitrev.c b/lib/bitrev.c index 81b56e0a7f32..05088231f31f 100644 --- a/lib/bitrev.c +++ b/lib/bitrev.c @@ -1,5 +1,4 @@ // SPDX-License-Identifier: GPL-2.0-only -#ifndef CONFIG_HAVE_ARCH_BITREVERSE #include <linux/types.h> #include <linux/module.h> #include <linux/bitrev.h> @@ -43,5 +42,3 @@ const u8 byte_rev_table[256] = { 0x1f, 0x9f, 0x5f, 0xdf, 0x3f, 0xbf, 0x7f, 0xff, }; EXPORT_SYMBOL_GPL(byte_rev_table); - -#endif /* CONFIG_HAVE_ARCH_BITREVERSE */ -- 2.34.1

1 0

[PATCH openEuler-26.09 0/3] Fix device memory leaks in XSched
by Liu Kai 11 Mar '26

11 Mar '26

Fix device memory leaks in XSched Liu Kai (3): xsched/dmem: ensure XSched Context creation before device memory allocation xsched/dmem: add size field to xsched_dmem_pool for dmem region tracking xsched/dmem: ensure dmem region release on AI task completion to prevent memory leaks include/linux/xsched.h | 2 ++ include/uapi/linux/xcu_vstream.h | 2 +- kernel/xsched/dmem.c | 38 +++++++++++++++++++++++++++----- kernel/xsched/vstream.c | 22 +++++++++++++++--- 4 files changed, 55 insertions(+), 9 deletions(-) -- 2.34.1

2 4

[PATCH OLK-6.6 0/2] optimize lightweight load balancing
by Chen Jinghuang 11 Mar '26

11 Mar '26

*** BLURB HERE *** Chen Jinghuang (1): sched: Fix kabi breakage of struct sched_domain for newidle_stats Peter Zijlstra (Intel) (1): sched/fair: Proportional newidle balance include/linux/sched/topology.h | 4 +++ kernel/sched/core.c | 3 +++ kernel/sched/fair.c | 48 +++++++++++++++++++++++++++++++--- kernel/sched/features.h | 5 ++++ kernel/sched/sched.h | 7 +++++ kernel/sched/topology.c | 6 +++++ 6 files changed, 69 insertions(+), 4 deletions(-) -- 2.34.1

2 3

[PATCH OLK-6.6 0/2] Fix TLBI broadcast optimization loss and incorrect dvmbm handling
by Tian Zheng 11 Mar '26

11 Mar '26

Fix 1 In the TLBI broadcast-optimization feature, the SYS_LSUDVM_CTRL_EL2 register controls whether the feature is enabled. When LPI low-power mode is enabled, this register is cleared as the pCPU powers down into LPI mode, causing the TLBI broadcast optimization to stop working. The fix saves and restores this control register in the callbacks for entering and exiting LPI mode. Fix 2 The current TLBI broadcast-optimization logic only supports normal VMs. If the feature is enabled globally, VMs that are not yet adapted for it, such as CCA will use an incorrect TLBI broadcast bitmap, which can cause a panic in CCA VMs. The fix moves the actual enable/disable of the dvmbm functionality into the vcpu load and vcpu put functions. Tian Zheng (2): KVM: arm64: Fix TLBI optimization broken in LPI mode KVM: arm64: Fix CCA guest panic when dvmbm is enabled arch/arm64/kvm/arm.c | 6 ++++++ arch/arm64/kvm/hisilicon/hisi_virt.c | 9 ++++----- arch/arm64/kvm/hisilicon/hisi_virt.h | 19 +++++++++++++++++++ 3 files changed, 29 insertions(+), 5 deletions(-) -- 2.33.0

1 2

[PATCH OLK-6.6 v4 0/4] Fix CVE-2025-21696.
by Ze Zuo 10 Mar '26

10 Mar '26

Fix CVE-2025-21696. Change since v3: -- internalize struct to reduce kABI exposure. David Hildenbrand (1): mm/mremap: fix WARN with uffd that has remap events disabled Lokesh Gidra (1): userfaultfd: move userfaultfd_ctx struct to header file Ryan Roberts (1): mm: clear uffd-wp PTE/PMD state on mremap() Ze Zuo (1): userfaultfd: fix kabi breakage due to struct userfaultfd_ctx fs/userfaultfd.c | 40 +---------------- include/linux/userfaultfd_k_ext.h | 62 ++++++++++++++++++++++++++ mm/huge_memory.c | 13 ++++++ mm/hugetlb.c | 15 ++++++- mm/mremap.c | 74 +++++++++++++++++++++++-------- 5 files changed, 145 insertions(+), 59 deletions(-) create mode 100644 include/linux/userfaultfd_k_ext.h -- 2.25.1

2 5

[PATCH OLK-6.6 0/2] optimize lightweight load balancing
by Chen Jinghuang 10 Mar '26

10 Mar '26

*** BLURB HERE *** Chen Jinghuang (1): sched: Fix kabi breakage of struct sched_domain for newidle_stats Peter Zijlstra (Intel) (1): sched/fair: Proportional newidle balance include/linux/sched/topology.h | 6 +++++ kernel/sched/core.c | 3 +++ kernel/sched/fair.c | 44 ++++++++++++++++++++++++++++++---- kernel/sched/features.h | 5 ++++ kernel/sched/sched.h | 7 ++++++ kernel/sched/topology.c | 6 +++++ 6 files changed, 67 insertions(+), 4 deletions(-) -- 2.34.1

2 3

[PATCH OLK-6.6 0/2] optimize lightweight load balancing
by Chen Jinghuang 10 Mar '26

10 Mar '26

*** BLURB HERE *** Chen Jinghuang (1): sched: Fix kabi breakage pf struct sched_domain for newidle_stats Peter Zijlstra (Intel) (1): sched/fair: Proportional newidle balance include/linux/sched/topology.h | 12 ++++++ kernel/sched/core.c | 3 ++ kernel/sched/fair.c | 78 ++++++++++++++++++++++++++++++++-- kernel/sched/features.h | 5 +++ kernel/sched/sched.h | 7 +++ kernel/sched/topology.c | 11 +++++ 6 files changed, 112 insertions(+), 4 deletions(-) -- 2.34.1

2 3

[PATCH OLK-6.6 1/2] sched/fair: Proportional newidle balance
by Chen Jinghuang 10 Mar '26

10 Mar '26

From: "Peter Zijlstra (Intel)" <peterz(a)infradead.org> stable inclusion from stable-v6.6.120 commit 51445190c10a36d292e70db085d0fb6cc3bec94f category: perf bugzilla: https://atomgit.com/openeuler/kernel/issues/8555 Reference: https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git/commit/?id… -------------------------------- commit 33cf66d88306663d16e4759e9d24766b0aaa2e17 upstream. Add a randomized algorithm that runs newidle balancing proportional to its success rate. This improves schbench significantly: 6.18-rc4: 2.22 Mrps/s 6.18-rc4+revert: 2.04 Mrps/s 6.18-rc4+revert+random: 2.18 Mrps/S Conversely, per Adam Li this affects SpecJBB slightly, reducing it by 1%: 6.17: -6% 6.17+revert: 0% 6.17+revert+random: -1% Signed-off-by: Peter Zijlstra (Intel) <peterz(a)infradead.org> Reviewed-by: Dietmar Eggemann <dietmar.eggemann(a)arm.com> Tested-by: Dietmar Eggemann <dietmar.eggemann(a)arm.com> Tested-by: Chris Mason <clm(a)meta.com> Link: https://lkml.kernel.org/r/6825c50d-7fa7-45d8-9b81-c6e7e25738e2@meta.com Link: https://patch.msgid.link/20251107161739.770122091@infradead.org [ Ajay: Modified to apply on v6.6 ] Signed-off-by: Ajay Kaher <ajay.kaher(a)broadcom.com> Signed-off-by: Greg Kroah-Hartman <gregkh(a)linuxfoundation.org> Conflicts: kernel/sched/core.c kernel/sched/fair.c [context conflicts] Signed-off-by: Chen Jinghuang <chenjinghuang2(a)huawei.com> --- include/linux/sched/topology.h | 3 +++ kernel/sched/core.c | 3 +++ kernel/sched/fair.c | 44 ++++++++++++++++++++++++++++++---- kernel/sched/features.h | 5 ++++ kernel/sched/sched.h | 7 ++++++ kernel/sched/topology.c | 6 +++++ 6 files changed, 64 insertions(+), 4 deletions(-) diff --git a/include/linux/sched/topology.h b/include/linux/sched/topology.h index 7eee852aa384..95e5d7772800 100644 --- a/include/linux/sched/topology.h +++ b/include/linux/sched/topology.h @@ -130,6 +130,9 @@ struct sched_domain { unsigned int nr_balance_failed; /* initialise to 0 */ /* idle_balance() stats */ + unsigned int newidle_call; + unsigned int newidle_success; + unsigned int newidle_ratio; u64 max_newidle_lb_cost; unsigned long last_decay_max_lb_cost; diff --git a/kernel/sched/core.c b/kernel/sched/core.c index 69d10fdb84d8..cf45ea94bb05 100644 --- a/kernel/sched/core.c +++ b/kernel/sched/core.c @@ -119,6 +119,7 @@ EXPORT_TRACEPOINT_SYMBOL_GPL(sched_util_est_se_tp); EXPORT_TRACEPOINT_SYMBOL_GPL(sched_update_nr_running_tp); DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); +DEFINE_PER_CPU(struct rnd_state, sched_rnd_state); #ifdef CONFIG_QOS_SCHED static void sched_change_qos_group(struct task_struct *tsk, struct task_group *tg); @@ -9976,6 +9977,8 @@ void __init sched_init_smp(void) sched_init_numa(NUMA_NO_NODE); set_sched_cluster(); + prandom_init_once(&sched_rnd_state); + /* * There's no userspace yet to cause hotplug operations; hence all the * CPU masks are stable and all blatant races in the below code cannot diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c index a4592d35da8b..d16c65d5fa34 100644 --- a/kernel/sched/fair.c +++ b/kernel/sched/fair.c @@ -13651,8 +13651,24 @@ void update_max_interval(void) max_load_balance_interval = HZ*num_online_cpus()/10; } -static inline bool update_newidle_cost(struct sched_domain *sd, u64 cost) +static inline void update_newidle_stats(struct sched_domain *sd, unsigned int success) { + sd->newidle_call++; + sd->newidle_success += success; + + if (sd->newidle_call >= 1024) { + sd->newidle_ratio = sd->newidle_success; + sd->newidle_call /= 2; + sd->newidle_success /= 2; + } +} + +static inline bool +update_newidle_cost(struct sched_domain *sd, u64 cost, unsigned int success) +{ + if (cost) + update_newidle_stats(sd, success); + if (cost > sd->max_newidle_lb_cost) { /* * Track max cost of a domain to make sure to not delay the @@ -13700,7 +13716,7 @@ static void rebalance_domains(struct rq *rq, enum cpu_idle_type idle) * Decay the newidle max times here because this is a regular * visit to all the domains. */ - need_decay = update_newidle_cost(sd, 0); + need_decay = update_newidle_cost(sd, 0, 0); max_cost += sd->max_newidle_lb_cost; /* @@ -14336,6 +14352,22 @@ static int newidle_balance(struct rq *this_rq, struct rq_flags *rf) break; if (sd->flags & SD_BALANCE_NEWIDLE) { + unsigned int weight = 1; + + if (sched_feat(NI_RANDOM)) { + /* + * Throw a 1k sided dice; and only run + * newidle_balance according to the success + * rate. + */ + u32 d1k = sched_rng() % 1024; + weight = 1 + sd->newidle_ratio; + if (d1k > weight) { + update_newidle_stats(sd, 0); + continue; + } + weight = (1024 + weight/2) / weight; + } pulled_task = load_balance(this_cpu, this_rq, sd, CPU_NEWLY_IDLE, @@ -14343,10 +14375,14 @@ static int newidle_balance(struct rq *this_rq, struct rq_flags *rf) t1 = sched_clock_cpu(this_cpu); domain_cost = t1 - t0; - update_newidle_cost(sd, domain_cost); - curr_cost += domain_cost; t0 = t1; + + /* + * Track max cost of a domain to make sure to not delay the + * next wakeup on the CPU. + */ + update_newidle_cost(sd, domain_cost, weight * !!pulled_task); } /* diff --git a/kernel/sched/features.h b/kernel/sched/features.h index 52ea0097c513..24a0c853a8a0 100644 --- a/kernel/sched/features.h +++ b/kernel/sched/features.h @@ -106,6 +106,11 @@ SCHED_FEAT(UTIL_EST_FASTUP, true) SCHED_FEAT(LATENCY_WARN, false) +/* + * Do newidle balancing proportional to its success rate using randomization. + */ +SCHED_FEAT(NI_RANDOM, false) + SCHED_FEAT(HZ_BW, true) SCHED_FEAT(IRQ_AVG, false) diff --git a/kernel/sched/sched.h b/kernel/sched/sched.h index bb581cbdae8d..0c3abb44f365 100644 --- a/kernel/sched/sched.h +++ b/kernel/sched/sched.h @@ -5,6 +5,7 @@ #ifndef _KERNEL_SCHED_SCHED_H #define _KERNEL_SCHED_SCHED_H +#include <linux/prandom.h> #include <linux/sched/affinity.h> #include <linux/sched/autogroup.h> #include <linux/sched/cpufreq.h> @@ -1377,6 +1378,12 @@ static inline bool is_migration_disabled(struct task_struct *p) } DECLARE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); +DECLARE_PER_CPU(struct rnd_state, sched_rnd_state); + +static inline u32 sched_rng(void) +{ + return prandom_u32_state(this_cpu_ptr(&sched_rnd_state)); +} #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu))) #define this_rq() this_cpu_ptr(&runqueues) diff --git a/kernel/sched/topology.c b/kernel/sched/topology.c index 3023e67da0fd..cf847fdf1063 100644 --- a/kernel/sched/topology.c +++ b/kernel/sched/topology.c @@ -1644,6 +1644,12 @@ sd_init(struct sched_domain_topology_level *tl, .last_balance = jiffies, .balance_interval = sd_weight, + + /* 50% success rate */ + .newidle_call = 512, + .newidle_success = 256, + .newidle_ratio = 512, + .max_newidle_lb_cost = 0, .last_decay_max_lb_cost = jiffies, .child = child, -- 2.34.1

2 2

[PATCH OLK-6.6 1/2] [Backport] sched/fair: Proportional newidle balance
by Chen Jinghuang 10 Mar '26

10 Mar '26

From: "Peter Zijlstra (Intel)" <peterz(a)infradead.org> stable inclusion from stable-v6.6.120 commit 51445190c10a36d292e70db085d0fb6cc3bec94f category: perf bugzilla: https://atomgit.com/openeuler/kernel/issues/8555 Reference: https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git/commit/?id… -------------------------------- commit 33cf66d88306663d16e4759e9d24766b0aaa2e17 upstream. Add a randomized algorithm that runs newidle balancing proportional to its success rate. This improves schbench significantly: 6.18-rc4: 2.22 Mrps/s 6.18-rc4+revert: 2.04 Mrps/s 6.18-rc4+revert+random: 2.18 Mrps/S Conversely, per Adam Li this affects SpecJBB slightly, reducing it by 1%: 6.17: -6% 6.17+revert: 0% 6.17+revert+random: -1% Signed-off-by: Peter Zijlstra (Intel) <peterz(a)infradead.org> Reviewed-by: Dietmar Eggemann <dietmar.eggemann(a)arm.com> Tested-by: Dietmar Eggemann <dietmar.eggemann(a)arm.com> Tested-by: Chris Mason <clm(a)meta.com> Link: https://lkml.kernel.org/r/6825c50d-7fa7-45d8-9b81-c6e7e25738e2@meta.com Link: https://patch.msgid.link/20251107161739.770122091@infradead.org [ Ajay: Modified to apply on v6.6 ] Signed-off-by: Ajay Kaher <ajay.kaher(a)broadcom.com> Signed-off-by: Greg Kroah-Hartman <gregkh(a)linuxfoundation.org> Conflicts: kernel/sched/core.c kernel/sched/fair.c [context conflicts] Signed-off-by: Chen Jinghuang <chenjinghuang2(a)huawei.com> --- include/linux/sched/topology.h | 3 +++ kernel/sched/core.c | 3 +++ kernel/sched/fair.c | 44 ++++++++++++++++++++++++++++++---- kernel/sched/features.h | 5 ++++ kernel/sched/sched.h | 7 ++++++ kernel/sched/topology.c | 6 +++++ 6 files changed, 64 insertions(+), 4 deletions(-) diff --git a/include/linux/sched/topology.h b/include/linux/sched/topology.h index 7eee852aa384..95e5d7772800 100644 --- a/include/linux/sched/topology.h +++ b/include/linux/sched/topology.h @@ -130,6 +130,9 @@ struct sched_domain { unsigned int nr_balance_failed; /* initialise to 0 */ /* idle_balance() stats */ + unsigned int newidle_call; + unsigned int newidle_success; + unsigned int newidle_ratio; u64 max_newidle_lb_cost; unsigned long last_decay_max_lb_cost; diff --git a/kernel/sched/core.c b/kernel/sched/core.c index 69d10fdb84d8..cf45ea94bb05 100644 --- a/kernel/sched/core.c +++ b/kernel/sched/core.c @@ -119,6 +119,7 @@ EXPORT_TRACEPOINT_SYMBOL_GPL(sched_util_est_se_tp); EXPORT_TRACEPOINT_SYMBOL_GPL(sched_update_nr_running_tp); DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); +DEFINE_PER_CPU(struct rnd_state, sched_rnd_state); #ifdef CONFIG_QOS_SCHED static void sched_change_qos_group(struct task_struct *tsk, struct task_group *tg); @@ -9976,6 +9977,8 @@ void __init sched_init_smp(void) sched_init_numa(NUMA_NO_NODE); set_sched_cluster(); + prandom_init_once(&sched_rnd_state); + /* * There's no userspace yet to cause hotplug operations; hence all the * CPU masks are stable and all blatant races in the below code cannot diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c index a4592d35da8b..d16c65d5fa34 100644 --- a/kernel/sched/fair.c +++ b/kernel/sched/fair.c @@ -13651,8 +13651,24 @@ void update_max_interval(void) max_load_balance_interval = HZ*num_online_cpus()/10; } -static inline bool update_newidle_cost(struct sched_domain *sd, u64 cost) +static inline void update_newidle_stats(struct sched_domain *sd, unsigned int success) { + sd->newidle_call++; + sd->newidle_success += success; + + if (sd->newidle_call >= 1024) { + sd->newidle_ratio = sd->newidle_success; + sd->newidle_call /= 2; + sd->newidle_success /= 2; + } +} + +static inline bool +update_newidle_cost(struct sched_domain *sd, u64 cost, unsigned int success) +{ + if (cost) + update_newidle_stats(sd, success); + if (cost > sd->max_newidle_lb_cost) { /* * Track max cost of a domain to make sure to not delay the @@ -13700,7 +13716,7 @@ static void rebalance_domains(struct rq *rq, enum cpu_idle_type idle) * Decay the newidle max times here because this is a regular * visit to all the domains. */ - need_decay = update_newidle_cost(sd, 0); + need_decay = update_newidle_cost(sd, 0, 0); max_cost += sd->max_newidle_lb_cost; /* @@ -14336,6 +14352,22 @@ static int newidle_balance(struct rq *this_rq, struct rq_flags *rf) break; if (sd->flags & SD_BALANCE_NEWIDLE) { + unsigned int weight = 1; + + if (sched_feat(NI_RANDOM)) { + /* + * Throw a 1k sided dice; and only run + * newidle_balance according to the success + * rate. + */ + u32 d1k = sched_rng() % 1024; + weight = 1 + sd->newidle_ratio; + if (d1k > weight) { + update_newidle_stats(sd, 0); + continue; + } + weight = (1024 + weight/2) / weight; + } pulled_task = load_balance(this_cpu, this_rq, sd, CPU_NEWLY_IDLE, @@ -14343,10 +14375,14 @@ static int newidle_balance(struct rq *this_rq, struct rq_flags *rf) t1 = sched_clock_cpu(this_cpu); domain_cost = t1 - t0; - update_newidle_cost(sd, domain_cost); - curr_cost += domain_cost; t0 = t1; + + /* + * Track max cost of a domain to make sure to not delay the + * next wakeup on the CPU. + */ + update_newidle_cost(sd, domain_cost, weight * !!pulled_task); } /* diff --git a/kernel/sched/features.h b/kernel/sched/features.h index 52ea0097c513..24a0c853a8a0 100644 --- a/kernel/sched/features.h +++ b/kernel/sched/features.h @@ -106,6 +106,11 @@ SCHED_FEAT(UTIL_EST_FASTUP, true) SCHED_FEAT(LATENCY_WARN, false) +/* + * Do newidle balancing proportional to its success rate using randomization. + */ +SCHED_FEAT(NI_RANDOM, false) + SCHED_FEAT(HZ_BW, true) SCHED_FEAT(IRQ_AVG, false) diff --git a/kernel/sched/sched.h b/kernel/sched/sched.h index bb581cbdae8d..0c3abb44f365 100644 --- a/kernel/sched/sched.h +++ b/kernel/sched/sched.h @@ -5,6 +5,7 @@ #ifndef _KERNEL_SCHED_SCHED_H #define _KERNEL_SCHED_SCHED_H +#include <linux/prandom.h> #include <linux/sched/affinity.h> #include <linux/sched/autogroup.h> #include <linux/sched/cpufreq.h> @@ -1377,6 +1378,12 @@ static inline bool is_migration_disabled(struct task_struct *p) } DECLARE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); +DECLARE_PER_CPU(struct rnd_state, sched_rnd_state); + +static inline u32 sched_rng(void) +{ + return prandom_u32_state(this_cpu_ptr(&sched_rnd_state)); +} #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu))) #define this_rq() this_cpu_ptr(&runqueues) diff --git a/kernel/sched/topology.c b/kernel/sched/topology.c index 3023e67da0fd..cf847fdf1063 100644 --- a/kernel/sched/topology.c +++ b/kernel/sched/topology.c @@ -1644,6 +1644,12 @@ sd_init(struct sched_domain_topology_level *tl, .last_balance = jiffies, .balance_interval = sd_weight, + + /* 50% success rate */ + .newidle_call = 512, + .newidle_success = 256, + .newidle_ratio = 512, + .max_newidle_lb_cost = 0, .last_decay_max_lb_cost = jiffies, .child = child, -- 2.34.1

2 3

[PATCH 1/2] [Backport] sched/fair: Proportional newidle balance
by Chen Jinghuang 10 Mar '26

10 Mar '26

From: "Peter Zijlstra (Intel)" <peterz(a)infradead.org> stable inclusion from stable-v6.6.120 commit 51445190c10a36d292e70db085d0fb6cc3bec94f category: perf bugzilla: https://atomgit.com/openeuler/kernel/issues/8555 Reference: https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git/commit/?id… -------------------------------- commit 33cf66d88306663d16e4759e9d24766b0aaa2e17 upstream. Add a randomized algorithm that runs newidle balancing proportional to its success rate. This improves schbench significantly: 6.18-rc4: 2.22 Mrps/s 6.18-rc4+revert: 2.04 Mrps/s 6.18-rc4+revert+random: 2.18 Mrps/S Conversely, per Adam Li this affects SpecJBB slightly, reducing it by 1%: 6.17: -6% 6.17+revert: 0% 6.17+revert+random: -1% Signed-off-by: Peter Zijlstra (Intel) <peterz(a)infradead.org> Reviewed-by: Dietmar Eggemann <dietmar.eggemann(a)arm.com> Tested-by: Dietmar Eggemann <dietmar.eggemann(a)arm.com> Tested-by: Chris Mason <clm(a)meta.com> Link: https://lkml.kernel.org/r/6825c50d-7fa7-45d8-9b81-c6e7e25738e2@meta.com Link: https://patch.msgid.link/20251107161739.770122091@infradead.org [ Ajay: Modified to apply on v6.6 ] Signed-off-by: Ajay Kaher <ajay.kaher(a)broadcom.com> Signed-off-by: Greg Kroah-Hartman <gregkh(a)linuxfoundation.org> Conflicts: kernel/sched/core.c kernel/sched/fair.c [context conflicts] Signed-off-by: Chen Jinghuang <chenjinghuang2(a)huawei.com> --- include/linux/sched/topology.h | 3 +++ kernel/sched/core.c | 3 +++ kernel/sched/fair.c | 44 ++++++++++++++++++++++++++++++---- kernel/sched/features.h | 5 ++++ kernel/sched/sched.h | 7 ++++++ kernel/sched/topology.c | 6 +++++ 6 files changed, 64 insertions(+), 4 deletions(-) diff --git a/include/linux/sched/topology.h b/include/linux/sched/topology.h index 7eee852aa384..95e5d7772800 100644 --- a/include/linux/sched/topology.h +++ b/include/linux/sched/topology.h @@ -130,6 +130,9 @@ struct sched_domain { unsigned int nr_balance_failed; /* initialise to 0 */ /* idle_balance() stats */ + unsigned int newidle_call; + unsigned int newidle_success; + unsigned int newidle_ratio; u64 max_newidle_lb_cost; unsigned long last_decay_max_lb_cost; diff --git a/kernel/sched/core.c b/kernel/sched/core.c index 69d10fdb84d8..cf45ea94bb05 100644 --- a/kernel/sched/core.c +++ b/kernel/sched/core.c @@ -119,6 +119,7 @@ EXPORT_TRACEPOINT_SYMBOL_GPL(sched_util_est_se_tp); EXPORT_TRACEPOINT_SYMBOL_GPL(sched_update_nr_running_tp); DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); +DEFINE_PER_CPU(struct rnd_state, sched_rnd_state); #ifdef CONFIG_QOS_SCHED static void sched_change_qos_group(struct task_struct *tsk, struct task_group *tg); @@ -9976,6 +9977,8 @@ void __init sched_init_smp(void) sched_init_numa(NUMA_NO_NODE); set_sched_cluster(); + prandom_init_once(&sched_rnd_state); + /* * There's no userspace yet to cause hotplug operations; hence all the * CPU masks are stable and all blatant races in the below code cannot diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c index a4592d35da8b..d16c65d5fa34 100644 --- a/kernel/sched/fair.c +++ b/kernel/sched/fair.c @@ -13651,8 +13651,24 @@ void update_max_interval(void) max_load_balance_interval = HZ*num_online_cpus()/10; } -static inline bool update_newidle_cost(struct sched_domain *sd, u64 cost) +static inline void update_newidle_stats(struct sched_domain *sd, unsigned int success) { + sd->newidle_call++; + sd->newidle_success += success; + + if (sd->newidle_call >= 1024) { + sd->newidle_ratio = sd->newidle_success; + sd->newidle_call /= 2; + sd->newidle_success /= 2; + } +} + +static inline bool +update_newidle_cost(struct sched_domain *sd, u64 cost, unsigned int success) +{ + if (cost) + update_newidle_stats(sd, success); + if (cost > sd->max_newidle_lb_cost) { /* * Track max cost of a domain to make sure to not delay the @@ -13700,7 +13716,7 @@ static void rebalance_domains(struct rq *rq, enum cpu_idle_type idle) * Decay the newidle max times here because this is a regular * visit to all the domains. */ - need_decay = update_newidle_cost(sd, 0); + need_decay = update_newidle_cost(sd, 0, 0); max_cost += sd->max_newidle_lb_cost; /* @@ -14336,6 +14352,22 @@ static int newidle_balance(struct rq *this_rq, struct rq_flags *rf) break; if (sd->flags & SD_BALANCE_NEWIDLE) { + unsigned int weight = 1; + + if (sched_feat(NI_RANDOM)) { + /* + * Throw a 1k sided dice; and only run + * newidle_balance according to the success + * rate. + */ + u32 d1k = sched_rng() % 1024; + weight = 1 + sd->newidle_ratio; + if (d1k > weight) { + update_newidle_stats(sd, 0); + continue; + } + weight = (1024 + weight/2) / weight; + } pulled_task = load_balance(this_cpu, this_rq, sd, CPU_NEWLY_IDLE, @@ -14343,10 +14375,14 @@ static int newidle_balance(struct rq *this_rq, struct rq_flags *rf) t1 = sched_clock_cpu(this_cpu); domain_cost = t1 - t0; - update_newidle_cost(sd, domain_cost); - curr_cost += domain_cost; t0 = t1; + + /* + * Track max cost of a domain to make sure to not delay the + * next wakeup on the CPU. + */ + update_newidle_cost(sd, domain_cost, weight * !!pulled_task); } /* diff --git a/kernel/sched/features.h b/kernel/sched/features.h index 52ea0097c513..24a0c853a8a0 100644 --- a/kernel/sched/features.h +++ b/kernel/sched/features.h @@ -106,6 +106,11 @@ SCHED_FEAT(UTIL_EST_FASTUP, true) SCHED_FEAT(LATENCY_WARN, false) +/* + * Do newidle balancing proportional to its success rate using randomization. + */ +SCHED_FEAT(NI_RANDOM, false) + SCHED_FEAT(HZ_BW, true) SCHED_FEAT(IRQ_AVG, false) diff --git a/kernel/sched/sched.h b/kernel/sched/sched.h index bb581cbdae8d..0c3abb44f365 100644 --- a/kernel/sched/sched.h +++ b/kernel/sched/sched.h @@ -5,6 +5,7 @@ #ifndef _KERNEL_SCHED_SCHED_H #define _KERNEL_SCHED_SCHED_H +#include <linux/prandom.h> #include <linux/sched/affinity.h> #include <linux/sched/autogroup.h> #include <linux/sched/cpufreq.h> @@ -1377,6 +1378,12 @@ static inline bool is_migration_disabled(struct task_struct *p) } DECLARE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); +DECLARE_PER_CPU(struct rnd_state, sched_rnd_state); + +static inline u32 sched_rng(void) +{ + return prandom_u32_state(this_cpu_ptr(&sched_rnd_state)); +} #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu))) #define this_rq() this_cpu_ptr(&runqueues) diff --git a/kernel/sched/topology.c b/kernel/sched/topology.c index 3023e67da0fd..cf847fdf1063 100644 --- a/kernel/sched/topology.c +++ b/kernel/sched/topology.c @@ -1644,6 +1644,12 @@ sd_init(struct sched_domain_topology_level *tl, .last_balance = jiffies, .balance_interval = sd_weight, + + /* 50% success rate */ + .newidle_call = 512, + .newidle_success = 256, + .newidle_ratio = 512, + .max_newidle_lb_cost = 0, .last_decay_max_lb_cost = jiffies, .child = child, -- 2.34.1

1 1

[PATCH 1/2] [Backport] sched/fair: Proportional newidle balance
by Chen Jinghuang 10 Mar '26

10 Mar '26

From: "Peter Zijlstra (Intel)" <peterz(a)infradead.org> stable inclusion from stable-v6.6.120 commit 51445190c10a36d292e70db085d0fb6cc3bec94f category: perf bugzilla: https://atomgit.com/openeuler/kernel/issues/8555 Reference: https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git/commit/?id… -------------------------------- commit 33cf66d88306663d16e4759e9d24766b0aaa2e17 upstream. Add a randomized algorithm that runs newidle balancing proportional to its success rate. This improves schbench significantly: 6.18-rc4: 2.22 Mrps/s 6.18-rc4+revert: 2.04 Mrps/s 6.18-rc4+revert+random: 2.18 Mrps/S Conversely, per Adam Li this affects SpecJBB slightly, reducing it by 1%: 6.17: -6% 6.17+revert: 0% 6.17+revert+random: -1% Signed-off-by: Peter Zijlstra (Intel) <peterz(a)infradead.org> Reviewed-by: Dietmar Eggemann <dietmar.eggemann(a)arm.com> Tested-by: Dietmar Eggemann <dietmar.eggemann(a)arm.com> Tested-by: Chris Mason <clm(a)meta.com> Link: https://lkml.kernel.org/r/6825c50d-7fa7-45d8-9b81-c6e7e25738e2@meta.com Link: https://patch.msgid.link/20251107161739.770122091@infradead.org [ Ajay: Modified to apply on v6.6 ] Signed-off-by: Ajay Kaher <ajay.kaher(a)broadcom.com> Signed-off-by: Greg Kroah-Hartman <gregkh(a)linuxfoundation.org> Conflicts: kernel/sched/core.c kernel/sched/fair.c [context conflicts] Signed-off-by: Chen Jinghuang <chenjinghuang2(a)huawei.com> --- include/linux/sched/topology.h | 3 +++ kernel/sched/core.c | 3 +++ kernel/sched/fair.c | 44 ++++++++++++++++++++++++++++++---- kernel/sched/features.h | 5 ++++ kernel/sched/sched.h | 7 ++++++ kernel/sched/topology.c | 6 +++++ 6 files changed, 64 insertions(+), 4 deletions(-) diff --git a/include/linux/sched/topology.h b/include/linux/sched/topology.h index 7eee852aa384..95e5d7772800 100644 --- a/include/linux/sched/topology.h +++ b/include/linux/sched/topology.h @@ -130,6 +130,9 @@ struct sched_domain { unsigned int nr_balance_failed; /* initialise to 0 */ /* idle_balance() stats */ + unsigned int newidle_call; + unsigned int newidle_success; + unsigned int newidle_ratio; u64 max_newidle_lb_cost; unsigned long last_decay_max_lb_cost; diff --git a/kernel/sched/core.c b/kernel/sched/core.c index 69d10fdb84d8..cf45ea94bb05 100644 --- a/kernel/sched/core.c +++ b/kernel/sched/core.c @@ -119,6 +119,7 @@ EXPORT_TRACEPOINT_SYMBOL_GPL(sched_util_est_se_tp); EXPORT_TRACEPOINT_SYMBOL_GPL(sched_update_nr_running_tp); DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); +DEFINE_PER_CPU(struct rnd_state, sched_rnd_state); #ifdef CONFIG_QOS_SCHED static void sched_change_qos_group(struct task_struct *tsk, struct task_group *tg); @@ -9976,6 +9977,8 @@ void __init sched_init_smp(void) sched_init_numa(NUMA_NO_NODE); set_sched_cluster(); + prandom_init_once(&sched_rnd_state); + /* * There's no userspace yet to cause hotplug operations; hence all the * CPU masks are stable and all blatant races in the below code cannot diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c index a4592d35da8b..d16c65d5fa34 100644 --- a/kernel/sched/fair.c +++ b/kernel/sched/fair.c @@ -13651,8 +13651,24 @@ void update_max_interval(void) max_load_balance_interval = HZ*num_online_cpus()/10; } -static inline bool update_newidle_cost(struct sched_domain *sd, u64 cost) +static inline void update_newidle_stats(struct sched_domain *sd, unsigned int success) { + sd->newidle_call++; + sd->newidle_success += success; + + if (sd->newidle_call >= 1024) { + sd->newidle_ratio = sd->newidle_success; + sd->newidle_call /= 2; + sd->newidle_success /= 2; + } +} + +static inline bool +update_newidle_cost(struct sched_domain *sd, u64 cost, unsigned int success) +{ + if (cost) + update_newidle_stats(sd, success); + if (cost > sd->max_newidle_lb_cost) { /* * Track max cost of a domain to make sure to not delay the @@ -13700,7 +13716,7 @@ static void rebalance_domains(struct rq *rq, enum cpu_idle_type idle) * Decay the newidle max times here because this is a regular * visit to all the domains. */ - need_decay = update_newidle_cost(sd, 0); + need_decay = update_newidle_cost(sd, 0, 0); max_cost += sd->max_newidle_lb_cost; /* @@ -14336,6 +14352,22 @@ static int newidle_balance(struct rq *this_rq, struct rq_flags *rf) break; if (sd->flags & SD_BALANCE_NEWIDLE) { + unsigned int weight = 1; + + if (sched_feat(NI_RANDOM)) { + /* + * Throw a 1k sided dice; and only run + * newidle_balance according to the success + * rate. + */ + u32 d1k = sched_rng() % 1024; + weight = 1 + sd->newidle_ratio; + if (d1k > weight) { + update_newidle_stats(sd, 0); + continue; + } + weight = (1024 + weight/2) / weight; + } pulled_task = load_balance(this_cpu, this_rq, sd, CPU_NEWLY_IDLE, @@ -14343,10 +14375,14 @@ static int newidle_balance(struct rq *this_rq, struct rq_flags *rf) t1 = sched_clock_cpu(this_cpu); domain_cost = t1 - t0; - update_newidle_cost(sd, domain_cost); - curr_cost += domain_cost; t0 = t1; + + /* + * Track max cost of a domain to make sure to not delay the + * next wakeup on the CPU. + */ + update_newidle_cost(sd, domain_cost, weight * !!pulled_task); } /* diff --git a/kernel/sched/features.h b/kernel/sched/features.h index 52ea0097c513..6cd6b5307430 100644 --- a/kernel/sched/features.h +++ b/kernel/sched/features.h @@ -106,6 +106,11 @@ SCHED_FEAT(UTIL_EST_FASTUP, true) SCHED_FEAT(LATENCY_WARN, false) +/* + * Do newidle balancing proportional to its success rate using randomization. + */ +SCHED_FEAT(NI_RANDOM, true) + SCHED_FEAT(HZ_BW, true) SCHED_FEAT(IRQ_AVG, false) diff --git a/kernel/sched/sched.h b/kernel/sched/sched.h index bb581cbdae8d..0c3abb44f365 100644 --- a/kernel/sched/sched.h +++ b/kernel/sched/sched.h @@ -5,6 +5,7 @@ #ifndef _KERNEL_SCHED_SCHED_H #define _KERNEL_SCHED_SCHED_H +#include <linux/prandom.h> #include <linux/sched/affinity.h> #include <linux/sched/autogroup.h> #include <linux/sched/cpufreq.h> @@ -1377,6 +1378,12 @@ static inline bool is_migration_disabled(struct task_struct *p) } DECLARE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); +DECLARE_PER_CPU(struct rnd_state, sched_rnd_state); + +static inline u32 sched_rng(void) +{ + return prandom_u32_state(this_cpu_ptr(&sched_rnd_state)); +} #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu))) #define this_rq() this_cpu_ptr(&runqueues) diff --git a/kernel/sched/topology.c b/kernel/sched/topology.c index 3023e67da0fd..cf847fdf1063 100644 --- a/kernel/sched/topology.c +++ b/kernel/sched/topology.c @@ -1644,6 +1644,12 @@ sd_init(struct sched_domain_topology_level *tl, .last_balance = jiffies, .balance_interval = sd_weight, + + /* 50% success rate */ + .newidle_call = 512, + .newidle_success = 256, + .newidle_ratio = 512, + .max_newidle_lb_cost = 0, .last_decay_max_lb_cost = jiffies, .child = child, -- 2.34.1

1 1

[PATCH 1/2] [Backport] sched/fair: Proportional newidle balance
by Chen Jinghuang 10 Mar '26

10 Mar '26

From: "Peter Zijlstra (Intel)" <peterz(a)infradead.org> stable inclusion from stable-v6.6.120 commit 51445190c10a36d292e70db085d0fb6cc3bec94f category: perf bugzilla: https://atomgit.com/openeuler/kernel/issues/8555 Reference: https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git/commit/?id… -------------------------------- commit 33cf66d88306663d16e4759e9d24766b0aaa2e17 upstream. Add a randomized algorithm that runs newidle balancing proportional to its success rate. This improves schbench significantly: 6.18-rc4: 2.22 Mrps/s 6.18-rc4+revert: 2.04 Mrps/s 6.18-rc4+revert+random: 2.18 Mrps/S Conversely, per Adam Li this affects SpecJBB slightly, reducing it by 1%: 6.17: -6% 6.17+revert: 0% 6.17+revert+random: -1% Signed-off-by: Peter Zijlstra (Intel) <peterz(a)infradead.org> Reviewed-by: Dietmar Eggemann <dietmar.eggemann(a)arm.com> Tested-by: Dietmar Eggemann <dietmar.eggemann(a)arm.com> Tested-by: Chris Mason <clm(a)meta.com> Link: https://lkml.kernel.org/r/6825c50d-7fa7-45d8-9b81-c6e7e25738e2@meta.com Link: https://patch.msgid.link/20251107161739.770122091@infradead.org [ Ajay: Modified to apply on v6.6 ] Signed-off-by: Ajay Kaher <ajay.kaher(a)broadcom.com> Signed-off-by: Greg Kroah-Hartman <gregkh(a)linuxfoundation.org> Conflicts: kernel/sched/core.c kernel/sched/fair.c [context conflicts] Signed-off-by: Chen Jinghuang <chenjinghuang2(a)huawei.com> --- include/linux/sched/topology.h | 3 +++ kernel/sched/core.c | 3 +++ kernel/sched/fair.c | 44 ++++++++++++++++++++++++++++++---- kernel/sched/features.h | 5 ++++ kernel/sched/sched.h | 7 ++++++ kernel/sched/topology.c | 6 +++++ 6 files changed, 64 insertions(+), 4 deletions(-) diff --git a/include/linux/sched/topology.h b/include/linux/sched/topology.h index 7eee852aa384..95e5d7772800 100644 --- a/include/linux/sched/topology.h +++ b/include/linux/sched/topology.h @@ -130,6 +130,9 @@ struct sched_domain { unsigned int nr_balance_failed; /* initialise to 0 */ /* idle_balance() stats */ + unsigned int newidle_call; + unsigned int newidle_success; + unsigned int newidle_ratio; u64 max_newidle_lb_cost; unsigned long last_decay_max_lb_cost; diff --git a/kernel/sched/core.c b/kernel/sched/core.c index 69d10fdb84d8..cf45ea94bb05 100644 --- a/kernel/sched/core.c +++ b/kernel/sched/core.c @@ -119,6 +119,7 @@ EXPORT_TRACEPOINT_SYMBOL_GPL(sched_util_est_se_tp); EXPORT_TRACEPOINT_SYMBOL_GPL(sched_update_nr_running_tp); DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); +DEFINE_PER_CPU(struct rnd_state, sched_rnd_state); #ifdef CONFIG_QOS_SCHED static void sched_change_qos_group(struct task_struct *tsk, struct task_group *tg); @@ -9976,6 +9977,8 @@ void __init sched_init_smp(void) sched_init_numa(NUMA_NO_NODE); set_sched_cluster(); + prandom_init_once(&sched_rnd_state); + /* * There's no userspace yet to cause hotplug operations; hence all the * CPU masks are stable and all blatant races in the below code cannot diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c index a4592d35da8b..d16c65d5fa34 100644 --- a/kernel/sched/fair.c +++ b/kernel/sched/fair.c @@ -13651,8 +13651,24 @@ void update_max_interval(void) max_load_balance_interval = HZ*num_online_cpus()/10; } -static inline bool update_newidle_cost(struct sched_domain *sd, u64 cost) +static inline void update_newidle_stats(struct sched_domain *sd, unsigned int success) { + sd->newidle_call++; + sd->newidle_success += success; + + if (sd->newidle_call >= 1024) { + sd->newidle_ratio = sd->newidle_success; + sd->newidle_call /= 2; + sd->newidle_success /= 2; + } +} + +static inline bool +update_newidle_cost(struct sched_domain *sd, u64 cost, unsigned int success) +{ + if (cost) + update_newidle_stats(sd, success); + if (cost > sd->max_newidle_lb_cost) { /* * Track max cost of a domain to make sure to not delay the @@ -13700,7 +13716,7 @@ static void rebalance_domains(struct rq *rq, enum cpu_idle_type idle) * Decay the newidle max times here because this is a regular * visit to all the domains. */ - need_decay = update_newidle_cost(sd, 0); + need_decay = update_newidle_cost(sd, 0, 0); max_cost += sd->max_newidle_lb_cost; /* @@ -14336,6 +14352,22 @@ static int newidle_balance(struct rq *this_rq, struct rq_flags *rf) break; if (sd->flags & SD_BALANCE_NEWIDLE) { + unsigned int weight = 1; + + if (sched_feat(NI_RANDOM)) { + /* + * Throw a 1k sided dice; and only run + * newidle_balance according to the success + * rate. + */ + u32 d1k = sched_rng() % 1024; + weight = 1 + sd->newidle_ratio; + if (d1k > weight) { + update_newidle_stats(sd, 0); + continue; + } + weight = (1024 + weight/2) / weight; + } pulled_task = load_balance(this_cpu, this_rq, sd, CPU_NEWLY_IDLE, @@ -14343,10 +14375,14 @@ static int newidle_balance(struct rq *this_rq, struct rq_flags *rf) t1 = sched_clock_cpu(this_cpu); domain_cost = t1 - t0; - update_newidle_cost(sd, domain_cost); - curr_cost += domain_cost; t0 = t1; + + /* + * Track max cost of a domain to make sure to not delay the + * next wakeup on the CPU. + */ + update_newidle_cost(sd, domain_cost, weight * !!pulled_task); } /* diff --git a/kernel/sched/features.h b/kernel/sched/features.h index 52ea0097c513..6cd6b5307430 100644 --- a/kernel/sched/features.h +++ b/kernel/sched/features.h @@ -106,6 +106,11 @@ SCHED_FEAT(UTIL_EST_FASTUP, true) SCHED_FEAT(LATENCY_WARN, false) +/* + * Do newidle balancing proportional to its success rate using randomization. + */ +SCHED_FEAT(NI_RANDOM, true) + SCHED_FEAT(HZ_BW, true) SCHED_FEAT(IRQ_AVG, false) diff --git a/kernel/sched/sched.h b/kernel/sched/sched.h index bb581cbdae8d..0c3abb44f365 100644 --- a/kernel/sched/sched.h +++ b/kernel/sched/sched.h @@ -5,6 +5,7 @@ #ifndef _KERNEL_SCHED_SCHED_H #define _KERNEL_SCHED_SCHED_H +#include <linux/prandom.h> #include <linux/sched/affinity.h> #include <linux/sched/autogroup.h> #include <linux/sched/cpufreq.h> @@ -1377,6 +1378,12 @@ static inline bool is_migration_disabled(struct task_struct *p) } DECLARE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); +DECLARE_PER_CPU(struct rnd_state, sched_rnd_state); + +static inline u32 sched_rng(void) +{ + return prandom_u32_state(this_cpu_ptr(&sched_rnd_state)); +} #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu))) #define this_rq() this_cpu_ptr(&runqueues) diff --git a/kernel/sched/topology.c b/kernel/sched/topology.c index 3023e67da0fd..cf847fdf1063 100644 --- a/kernel/sched/topology.c +++ b/kernel/sched/topology.c @@ -1644,6 +1644,12 @@ sd_init(struct sched_domain_topology_level *tl, .last_balance = jiffies, .balance_interval = sd_weight, + + /* 50% success rate */ + .newidle_call = 512, + .newidle_success = 256, + .newidle_ratio = 512, + .max_newidle_lb_cost = 0, .last_decay_max_lb_cost = jiffies, .child = child, -- 2.34.1

1 1

[PATCH OLK-6.6] tools/interference: add ifstool utility for kernel interference statistics
by Tengda Wu 10 Mar '26

10 Mar '26

hulk inclusion category: feature bugzilla: https://atomgit.com/openeuler/kernel/issues/7429 -------------------------------- As real-time and high-performance workloads become more sensitive to execution jitter, observability into kernel-induced "noise" is critical. The CONFIG_CGROUP_IFS infrastructure provides this telemetry, but raw data from cgroups is difficult to parse and visualize manually. Introduce ifstool, a userspace utility to monitor and analyze the interference.stat interface. Key capabilities include: 1. High-Resolution Monitoring: - Samples raw counters and latency distributions at sub-second intervals. - Outputs structured CSV data for long-term profiling and integration with external post-processing tools. 2. Interactive Offline Reporting: - Generates self-contained HTML dashboards using Plotly. - Visualizes "Total Time Delta" trends to identify temporal spikes. - Renders latency heatmaps to expose the magnitude and frequency of interference events. Signed-off-by: Tengda Wu <wutengda2(a)huawei.com> --- tools/Makefile | 11 +- tools/kspect/Makefile | 28 +++ tools/kspect/README | 65 ++++++ tools/kspect/ifstool | 409 ++++++++++++++++++++++++++++++++++ tools/kspect/ifstool.1 | 87 ++++++++ tools/kspect/requirements.txt | 2 + 6 files changed, 598 insertions(+), 4 deletions(-) create mode 100644 tools/kspect/Makefile create mode 100644 tools/kspect/README create mode 100644 tools/kspect/ifstool create mode 100644 tools/kspect/ifstool.1 create mode 100644 tools/kspect/requirements.txt diff --git a/tools/Makefile b/tools/Makefile index 37e9f6804832..153817f0cc17 100644 --- a/tools/Makefile +++ b/tools/Makefile @@ -22,6 +22,7 @@ help: @echo ' hv - tools used when in Hyper-V clients' @echo ' iio - IIO tools' @echo ' intel-speed-select - Intel Speed Select tool' + @echo ' kspect - KSPECT tools' @echo ' kvm_stat - top-like utility for displaying kvm statistics' @echo ' leds - LEDs tools' @echo ' nolibc - nolibc headers testing and installation' @@ -69,7 +70,7 @@ acpi: FORCE cpupower: FORCE $(call descend,power/$@) -cgroup counter firewire hv guest bootconfig spi usb virtio mm bpf iio gpio objtool leds wmi pci firmware debugging tracing: FORCE +cgroup counter firewire hv guest bootconfig spi usb virtio mm bpf iio gpio objtool leds wmi pci firmware debugging tracing kspect: FORCE $(call descend,$@) bpf/%: FORCE @@ -120,7 +121,8 @@ all: acpi cgroup counter cpupower gpio hv firewire \ perf selftests bootconfig spi turbostat usb \ virtio mm bpf x86_energy_perf_policy \ tmon freefall iio objtool kvm_stat wmi \ - pci debugging tracing thermal thermometer thermal-engine + pci debugging tracing thermal thermometer thermal-engine \ + kspect acpi_install: $(call descend,power/$(@:_install=),install) @@ -128,7 +130,7 @@ acpi_install: cpupower_install: $(call descend,power/$(@:_install=),install) -cgroup_install counter_install firewire_install gpio_install hv_install iio_install perf_install bootconfig_install spi_install usb_install virtio_install mm_install bpf_install objtool_install wmi_install pci_install debugging_install tracing_install: +cgroup_install counter_install firewire_install gpio_install hv_install iio_install perf_install bootconfig_install spi_install usb_install virtio_install mm_install bpf_install objtool_install wmi_install pci_install debugging_install tracing_install kspect_install: $(call descend,$(@:_install=),install) selftests_install: @@ -161,7 +163,8 @@ install: acpi_install cgroup_install counter_install cpupower_install gpio_insta virtio_install mm_install bpf_install x86_energy_perf_policy_install \ tmon_install freefall_install objtool_install kvm_stat_install \ wmi_install pci_install debugging_install intel-speed-select_install \ - tracing_install thermometer_install thermal-engine_install + tracing_install thermometer_install thermal-engine_install \ + kspect_install acpi_clean: $(call descend,power/acpi,clean) diff --git a/tools/kspect/Makefile b/tools/kspect/Makefile new file mode 100644 index 000000000000..ffb380ddc885 --- /dev/null +++ b/tools/kspect/Makefile @@ -0,0 +1,28 @@ +# SPDX-License-Identifier: GPL-2.0 + +PREFIX ?= /usr/local +BINDIR = $(PREFIX)/bin +MANDIR = $(PREFIX)/share/man/man1 + +MAN1 = ifstool.1 +TARGET = ifstool + +all: man + +man: $(MAN1) + +install-man: + install -d $(MANDIR) + install -m 0644 $(MAN1) $(MANDIR) + +install-tools: + install -d $(BINDIR) + install -m 0755 $(TARGET) $(BINDIR) + +install: install-tools install-man + +uninstall: + rm -f $(BINDIR)/$(TARGET) + rm -f $(MANDIR)/$(MAN1) + +.PHONY: all install-tools install-man install diff --git a/tools/kspect/README b/tools/kspect/README new file mode 100644 index 000000000000..618e02a202f7 --- /dev/null +++ b/tools/kspect/README @@ -0,0 +1,65 @@ +IFSTOOL - Interference Statistics Analytical Utility + +Overview +======== +IFSTOOL is a specialized userspace utility designed to facilitate the +monitoring and analysis of Interference Statistics (CONFIG_CGROUP_IFS). + +The IFS infrastructure is a kernel-level framework that provides critical +observability into execution jitter (noise) that disrupts task determinism. +It monitors and quantifies the CPU time stolen by kernel-level activities, +such as: interrupt handling, softirqs, and lock contention. The framework +exposes this telemetry via the interference.stat control file within the +cgroup hierarchy. + +IFSTOOL interfaces with the interference.stat file to export raw metrics +into structured CSV data and interactive HTML-based distribution reports. + +Setup +===== +The host environment must meet the following criteria: + +* Kernel: Compiled with `CONFIG_CGROUP_IFS=y`. +* Boot Parameters: `cgroup_ifs=1` added to the kernel command line (optional, + not needed if `CONFIG_CGROUP_IFS_DEFAULT_ENABLED=y`). +* Python Runtime: Python 3.x with `pandas` and `plotly` libraries. +* Cgroup Hierarchy: Either v2 unified or v1 with the cpu subsystem mounted. + +Build +===== +IFSTOOL provides a Makefile to streamline the installation of the +executable and its documentation. + + $ make install + +Run +=== +IFSTOOL operates via two primary functional modes: monitor and report. + +1. **Monitor:** Capture raw interference data from a target cgroup. + + $ ifstool monitor --cgroup docker/<cid> --interval 1 --output capture.csv + +2. **Report:** Transform captured CSV data into an interactive HTML dashboard. + + - Provide only --base for a single-session deep dive: + + $ ifstool report --base capture.csv + + - Provide both --base and --curr to render a differential report, + ideal for validating optimizations: + + $ ifstool report --base baseline.csv --curr current.csv + +HTML Description +================ +The generated HTML report provides a multi-dimensional view of kernel noise: + +- Total Time Delta Trend: A time-series line chart illustrating the incremental + nanoseconds of interference per category (e.g., irq, spinlock). +- Latency Heatmaps: A frequency-domain visualization of the kernel's internal + histogram. + - X-axis: Wall-clock time of the trace. + - Y-axis: Latency magnitude (logarithmic buckets from ns to s). + - Color Intensity: Represents the density (event count) of interference + within that specific latency window. diff --git a/tools/kspect/ifstool b/tools/kspect/ifstool new file mode 100644 index 000000000000..c17de045263a --- /dev/null +++ b/tools/kspect/ifstool @@ -0,0 +1,409 @@ +#!/usr/bin/env python3 +# SPDX-License-Identifier: GPL-2.0-only +# +# ifstool: A tool to monitor and report interference statistics (CONFIG_CGROUP_IFS). +# +# Copyright(c) 2026. Huawei Technologies Co., Ltd +# +# Authors: +# Tengda Wu <wutengda2(a)huawei.com> + +import os +import time +import re +import csv +import argparse +import pandas as pd +import plotly.graph_objects as go +from plotly.subplots import make_subplots +from plotly.offline import get_plotlyjs + + +class InterferenceTool: + """ + A utility class to monitor Linux cgroup interference statistics (CONFIG_CGROUP_IFS) + and generate visual comparison reports between baseline and current data. + """ + + def __init__(self): + # Default path for cgroup v2 interference stats + self.base_path = "/sys/fs/cgroup" + + def parse_stat(self, content): + """ + Parses the raw content of the interference.stat file. + + Args: + content (str): Raw string content from the stat file. + Returns: + tuple: (total_times dict, distributions dict) + """ + + total_times, distributions = {}, {} + + # Split content into sections: Top-level totals and various distributions + # Uses positive lookahead to split before a word followed by ' distribution' + sections = re.split(r"\n(?=[a-z]+ distribution)", content) + + # Parse global total times (first section) + for line in sections[0].strip().split("\n"): + match = re.match(r"^([a-z]+)\s+(\d+)$", line.strip()) + if match: + total_times[match.group(1)] = int(match.group(2)) + + # Parse histogram distributions (subsequent sections) + for section in sections[1:]: + lines = section.strip().split("\n") + # Extract header name (e.g., 'spinlock distribution') + header = lines[0].replace(" distribution", "").strip() + # Parse bucket key-value pairs (e.g., '[64 ns, 128 ns) : 143791') + dist_data = { + l.split(":")[0].strip(): int(l.split(":")[1].strip()) + for l in lines[1:] + if ":" in l + } + distributions[header] = dist_data + + return total_times, distributions + + def monitor(self, cgroup_id, interval, duration, output_csv): + """ + Periodically samples interference stats and saves results to a CSV file. + + Args: + cgroup_id (str): The specific cgroup folder name. For example, + /sys/kernel/cgroup/A/B/C, when want to monitor level B, the + cgroup_id is A/B. + interval (float): Seconds between samples. + duration (int): Total monitoring time in seconds. + output_csv (str): Filename to save sampled data. + """ + + candidates = [ + os.path.join(self.base_path, cgroup_id), + os.path.join(self.base_path, "cpu", cgroup_id), # cgroup v1 + ] + path = next((p for p in candidates if os.path.exists(p)), None) + + if not path: + print( + f"\n[!] No access to cgroup: {os.path.join(self.base_path, cgroup_id)}" + ) + print(" Hint: You can find the correct cid by executing:") + print(" cat /proc/<PID>/cgroup") + return + + path = os.path.join(path, "interference.stat") + if not os.path.exists(path): + print(f"\n[!] Interface file '{path}' not found.") + print(" This usually happens due to one of the following:") + print(" 1. Kernel is not compiled with CONFIG_CGROUP_IFS=y") + print( + " 2. CONFIG_CGROUP_IFS_DEFAULT_ENABLED is not set to y", + "and the boot parameter 'cgroup_ifs=1' is not added", + ) + print(" 3. The current cgroup is not managed by the IFS controller") + return + + print( + f"[*] Starting monitor: {cgroup_id}, interval: {interval}s, duration: {duration}s" + ) + + data_list = [] + start_time = time.time() + + try: + while (time.time() - start_time) < duration: + ts = time.strftime("%H:%M:%S") + # Using fractional seconds in timestamp for sub-second intervals + if interval < 1.0: + ts = time.strftime("%H:%M:%S") + f".{int((time.time()%1)*100):02d}" + + with open(path, "r") as f: + total_times, dists = self.parse_stat(f.read()) + + # Append total time metrics + for cat, val in total_times.items(): + data_list.append([ts, cgroup_id, cat, "total_time_ns", val]) + + # Append bucket distribution metrics + for cat, dist in dists.items(): + for b, c in dist.items(): + data_list.append([ts, cgroup_id, cat, f"bucket_{b}", c]) + + time.sleep(interval) + except KeyboardInterrupt: + print("\n[!] Monitoring interrupted by user.") + + # Write buffered data to CSV + with open(output_csv, "w", newline="") as f: + writer = csv.writer(f) + writer.writerow( + ["timestamp", "cgroup_id", "category", "metric_type", "value"] + ) + writer.writerows(data_list) + print(f"[+] Data exported successfully: {output_csv}") + + def bucket_key(self, bucket_str): + """ + Parsing logic for histogram bucket labels to enable correct numerical sorting. + Example input: "bucket_[67.10 ms, 134.21 ms)" + + Returns: + float: The value converted to nanoseconds (ns). + """ + try: + # Regex to capture the first number and its unit (ns|us|ms|s) + match = re.search(r"(\d+\.?\d*)\s*(ns|us|ms|s)", bucket_str) + if not match: + return 0 + + value = float(match.group(1)) + unit = match.group(2).lower() + + # Conversion factors to nanoseconds + factors = {"ns": 1, "us": 1000, "ms": 1000000, "s": 1000000000} + + return value * factors.get(unit, 1) + except Exception: + return 0 + + def report(self, base_csv, curr_csv, output_html): + """ + Loads two CSV files and generates an interactive HTML dashboard. + + Args: + base_csv (str): Path to baseline data. + curr_csv (str): Path to current data. + output_html (str): Path to generate the HTML report. + """ + + df_b = pd.read_csv(base_csv) + single_mode = curr_csv is None or base_csv == curr_csv + df_c = None if single_mode else pd.read_csv(curr_csv) + + # Get unique union of categories present in both datasets + all_dfs = [df_b] if single_mode else [df_b, df_c] + categories = sorted( + list(set().union(*(df.category.unique() for df in all_dfs))) + ) + + # Color mapping to ensure consistent colors for categories across plots + color_sequence = [ + "#636EFA", + "#EF553B", + "#00CC96", + "#AB63FA", + "#FFA15A", + "#19D3F3", + "#FF6692", + "#B6E880", + "#FF97FF", + "#FECB52", + ] + color_map = { + cat: color_sequence[i % len(color_sequence)] + for i, cat in enumerate(categories) + } + + # Build HTML content with CSS for layout + html_parts = [ + "<html><head><title>IFS Analysis Dashboard</title>", + f"<script type='text/javascript'>{get_plotlyjs()}</script>", + ] + col_width = "100%" if single_mode else "49.5%" + html_parts.append( + f"""<style> + body{{font-family:'Segoe UI',sans-serif; background:#f0f2f5; padding:20px; color:#333;}} + .card{{background:white; padding:15px; margin-bottom:20px; border-radius:10px; box-shadow:0 2px 5px rgba(0,0,0,0.05);}} + .row-container{{display:flex; gap:15px; justify-content:space-between;}} + .col-item{{width:{col_width}; min-width:0;}} + .stat-box{{display:flex; gap:10px; margin:5px 0; font-size:0.85em; color:#555;}} + .stat-item{{background:#f1f3f5; padding:4px 8px; border-radius:4px; border-left:3px solid #007bff;}} + .section-header{{margin:30px 0 15px 0; padding-bottom:10px; border-bottom:2px solid #ddd;}} + </style></head><body>""" + ) + + title = "Performance Analysis" if single_mode else "Performance Comparison" + html_parts.append(f"<h2>{title} (CONFIG_CGROUP_IFS)</h2>") + info = ( + f"File: {base_csv}" + if single_mode + else f"Baseline: {base_csv} | Current: {curr_csv}" + ) + html_parts.append(f"<p style='color:#666;'>{info}</p>") + + # --- Part 1: Total Time Trends (Line Charts) --- + html_parts.append("<div class='card'>") + html_parts.append("<h3 style='margin-top:0;'>Total Time Delta (ns) Trend</h3>") + + # Subplots ensure Y-axis can be matched for direct visual comparison + fig_line = make_subplots( + rows=1, + cols=1 if single_mode else 2, + horizontal_spacing=0.05, + subplot_titles=("Latency",) if single_mode else ("Baseline", "Current"), + ) + + plot_configs = ( + [("Data", df_b)] if single_mode else [("Baseline", df_b), ("Current", df_c)] + ) + for i, (name, df) in enumerate(plot_configs, 1): + for cat in categories: + sub = df[ + (df["category"] == cat) & (df["metric_type"] == "total_time_ns") + ].sort_values("timestamp") + if not sub.empty: + y_val = sub["value"].diff().fillna(0) + fig_line.add_trace( + go.Scatter( + x=sub["timestamp"], + y=y_val, + name=f"{cat} ({name})", + legendgroup=cat, + mode="lines+markers", + line=dict(color=color_map[cat], width=2), + marker=dict(color=color_map[cat], size=6), + ), + row=1, + col=i, + ) + + # Sync Y-axes scale for baseline and current plots + if not single_mode: + fig_line.update_yaxes(matches="y", row=1, col=2) + fig_line.update_layout( + height=500, template="plotly_white", margin=dict(t=50, b=20) + ) + html_parts.append(fig_line.to_html(full_html=False, include_plotlyjs=False)) + html_parts.append("</div>") + + html_parts.append( + "<h3 class='section-header'>Detailed Latency Distribution</h3>" + ) + + # --- Part 2: Distribution Heatmaps --- + for cat in categories: + + def get_m_and_stats(df): + """Processes raw metrics into a delta-count matrix for heatmap visualization.""" + sub = df[ + (df["category"] == cat) + & (df["metric_type"].str.startswith("bucket_")) + ] + if sub.empty: + return pd.DataFrame(), {} + + # Transform to matrix (Buckets vs Time) + p = sub.pivot( + index="metric_type", columns="timestamp", values="value" + ).sort_index(axis=1) + # Sort Y-axis buckets based on numerical time value + p = p.reindex(sorted(p.index, key=self.bucket_key)) + # Calculate incremental change (delta) + delta = p.diff(axis=1).fillna(0) + return delta, { + "Total": int(delta.values.sum()), + "Peak": int(delta.values.max()), + } + + m_b, stats_b = get_m_and_stats(df_b) + m_c, stats_c = (None, None) if single_mode else get_m_and_stats(df_c) + + # Determine global max for heatmap color scaling consistency + z_vals = [m_b.values.max() if not m_b.empty else 0] + if not single_mode: + z_vals.append(m_c.values.max() if not m_c.empty else 0) + global_max_z = max(z_vals + [1]) + + html_parts.append(f"<div class='card'><h4>Category: {cat.upper()}</h4>") + html_parts.append("<div class='row-container'>") + + configs = ( + [("Data", m_b, stats_b)] + if single_mode + else [("Baseline", m_b, stats_b), ("Current", m_c, stats_c)] + ) + for name, m, stats in configs: + html_parts.append("<div class='col-item'>") + if not m.empty: + # Display summary statistics + html_parts.append("<div class='stat-box'>") + for k, v in stats.items(): + html_parts.append( + f"<div class='stat-item'><b>{k}:</b> {v}</div>" + ) + html_parts.append("</div>") + + # Generate Heatmap + fig = go.Figure( + data=go.Heatmap( + z=m.values, + x=m.columns, + y=m.index, + colorscale="Viridis", + zmin=0, + zmax=global_max_z, + colorbar=dict(title="Count", thickness=10, len=0.7), + ) + ) + fig.update_layout( + height=400, + margin=dict(l=120, r=0, t=10, b=30), + template="plotly_white", + ) + html_parts.append( + fig.to_html(full_html=False, include_plotlyjs=False) + ) + else: + html_parts.append( + "<div style='height:100px; display:flex; align-items:center; justify-content:center; background:#fafafa; color:#ccc;'>No Data</div>" + ) + html_parts.append("</div>") + + html_parts.append("</div></div>") # end row-container & card + + # Save the final HTML report + with open(output_html, "w") as f: + f.writelines(html_parts + ["</body></html>"]) + print(f"[+] Comparison report generated: {output_html}") + + +if __name__ == "__main__": + parser = argparse.ArgumentParser() + sub = parser.add_subparsers(dest="mode") + + # Monitor Mode Arguments + p_mon = sub.add_parser("monitor") + p_mon.add_argument( + "-G", + "--cgroup", + required=True, + help="Cgroup name (e.g. 'docker/cid'). " + "Tip: Find this via 'cat /proc/<PID>/cgroup'", + ) + p_mon.add_argument( + "-i", "--interval", type=float, default=1.0, help="Sampling interval (sec)" + ) + p_mon.add_argument( + "-d", "--duration", type=int, default=30, help="Sampling duration (sec)" + ) + p_mon.add_argument("-o", "--output", default="capture.csv") + + # Report Mode Arguments + p_comp = sub.add_parser("report") + p_comp.add_argument("-b", "--base", required=True, help="Baseline CSV file") + p_comp.add_argument( + "-c", "--curr", default=None, help="Current CSV file (optional, for comparison)" + ) + p_comp.add_argument("-o", "--output", default="report.html") + + args = parser.parse_args() + tool = InterferenceTool() + + if args.mode == "monitor": + tool.monitor(args.cgroup, args.interval, args.duration, args.output) + elif args.mode == "report": + tool.report(args.base, args.curr, args.output) + else: + parser.print_help() diff --git a/tools/kspect/ifstool.1 b/tools/kspect/ifstool.1 new file mode 100644 index 000000000000..7ff9d297fc08 --- /dev/null +++ b/tools/kspect/ifstool.1 @@ -0,0 +1,87 @@ +.TH IFSTOOL 1 "March 2026" "Linux" "User Commands" +.SH NAME +ifstool \- Analyze kernel interference statistics (CONFIG_CGROUP_IFS) +.SH SYNOPSIS +.B ifstool monitor +[\fB\-G\fR|\fB\-\-cgroup\fR \fICGROUP_ID\fR] [\fB\-d\fR|\fB\-\-duration\fR \fISEC\fR] [\fB\-i\fR|\fB\-\-interval\fR \fISEC\fR] [\fB\-o\fR|\fB\-\-output\fR \fICSV\fR] +.br +.B ifstool report +[\fB\-b\fR|\fB\-\-base\fR \fICSV\fR] [\fB\-c\fR|\fB\-\-curr\fR \fICSV\fR] [\fB\-o\fR|\fB\-\-output\fR \fIHTML\fR] + +.SH DESCRIPTION +.B IFSTOOL +is a specialized userspace utility designed to facilitate the monitoring and analysis of Interference Statistics (\fBCONFIG_CGROUP_IFS\fR). + +The \fBIFS infrastructure\fR is a kernel-level framework providing critical observability into execution jitter (noise) that disrupts task determinism. It quantifies CPU time stolen by kernel activities such as interrupt handling, softirqs, and lock contention. This telemetry is exposed via the \fBinterference.stat\fR control file within the cgroup hierarchy. + +\fBIFSTOOL\fR interfaces with this file to export raw metrics into structured CSV data and interactive HTML-based distribution reports. + +.SH SETUP +The host environment must meet the following criteria: +.IP \[bu] 2 +\fBKernel:\fR Compiled with \fBCONFIG_CGROUP_IFS=y\fR. +.IP \[bu] 2 +\fBBoot Parameters:\fR \fBcgroup_ifs=1\fR added to the kernel command line (optional, not needed if \fBCONFIG_CGROUP_IFS_DEFAULT_ENABLED=y\fR). +.IP \[bu] 2 +\fBPython Runtime:\fR Python 3.x with \fBpandas\fR and \fBplotly\fR libraries. +.IP \[bu] 2 +\fBCgroup Hierarchy:\fR Either v2 unified or v1 (with the \fBcpu\fR subsystem mounted). + +.SH COMMANDS +.SS monitor +Capture raw interference data from a target cgroup. +.TP +\fB\-G, \-\-cgroup\fR \fICGROUP_ID\fR +Specify the cgroup identifier (e.g., \fIdocker/<cid>\fR). For example, /sys/kernel/cgroup/A/B/C, when you want to monitor level B, the cgroup_id is A/B. Tip: Find this via cat \fI/proc/PID/cgroup\fR. +.TP +\fB\-d, \-\-duration\fR \fISEC\fR +Total collection time in seconds (default: 30). +.TP +\fB\-i, \-\-interval\fR \fISEC\fR +Sampling interval in seconds; supports floating point (default: 1.0). +.TP +\fB\-o, \-\-output\fR \fICSV\fR +Path to the output CSV file where raw sampled data will be stored (default: \fBcapture.csv\fR). + +.SS report +Transform captured CSV data into an interactive HTML dashboard. +.TP +\fB\-b, \-\-base\fR \fICSV\fR +Primary capture file for analysis. +.TP +\fB\-c, \-\-curr\fR \fICSV\fR +Optional secondary file for differential (comparison) analysis. +.TP +\fB\-o, \-\-output\fR \fIHTML\fR +Path to the generated interactive HTML dashboard (default: \fBreport.html\fR). + +.SH HTML REPORT STRUCTURE +The generated HTML report provides a multi-dimensional view of kernel noise: +.IP "Total Time Delta Trend" 4 +A time-series line chart illustrating the incremental nanoseconds of interference per category (e.g., irq, spinlock). +.IP "Latency Heatmaps" 4 +A frequency-domain visualization of the kernel's internal histogram: +.RS 8 +.IP "X-axis:" 8 +Wall-clock time of the trace. +.IP "Y-axis:" 8 +Latency magnitude (logarithmic buckets from ns to s). +.IP "Color Intensity:" 8 +Represents the event count (density) of interference within that specific latency window. +.RE + +.SH EXAMPLES +Monitor a Docker container for 10 seconds: +.IP +.B $ ifstool monitor \-\-cgroup docker/<cid> \-\-duration 10 \-\-interval 1 +.PP +Generate a single-session deep dive: +.IP +.B $ ifstool report \-\-base capture.csv +.PP +Generate a differential report between two traces: +.IP +.B $ ifstool report \-\-base baseline.csv \-\-curr current.csv + +.SH AUTHOR +Tengda Wu <wutengda2(a)huawei.com> \ No newline at end of file diff --git a/tools/kspect/requirements.txt b/tools/kspect/requirements.txt new file mode 100644 index 000000000000..8e5e376a0e9a --- /dev/null +++ b/tools/kspect/requirements.txt @@ -0,0 +1,2 @@ +pandas==3.* +plotly==6.* -- 2.34.1

2 1