[PATCH v1 OLK-6.6 0/3] kworker: Fix the problem of ipsan performance degradation on ARM architecture
From: shaodenghui <shaodenghui@huawei.com> When the current downstream FS tests IPSAN, it is found that the performance on ARM is much worse than that on X86, and the test data of IPSAN fluctuates greatly. After analysis, the reason is that when iscsi issues IO, the task is sent to kworker for processing by iscsi_xmitworker. The workqueue created by iscsi can automatically identify the CPU of the soft interrupt currently processed by iscsi, and automatically schedule the workqueue to the corresponding NUMA node. shaodenghui (3): iscsi: add member for NUMA aware order workqueue workqueue: implement NUMA affinity for single thread workqueue iscsi: use dynamic single thread workqueue to improve performance drivers/scsi/iscsi_tcp.c | 9 +++++++++ drivers/scsi/libiscsi.c | 17 ++++++++++++----- include/linux/workqueue.h | 1 + include/scsi/libiscsi.h | 1 + kernel/workqueue.c | 15 ++++++++++----- 5 files changed, 33 insertions(+), 10 deletions(-) -- 2.33.0
From: shaodenghui <shaodenghui@huawei.com> euleros inclusion category: feature bugzilla: https://gitee.com/openeuler/kernel/issues/I4IZNO CVE: NA ------------------------------------------------- Add member to struct iscsi_conn. Signed-off-by: Yang Yingliang <yangyingliang@huawei.com> Reviewed-By: Xie XiuQi <xiexiuqi@huawei.com> Signed-off-by: Yang Yingliang <yangyingliang@huawei.com> Signed-off-by: BiaoXiang Ye <yebiaoxiang@huawei.com> Reviewed-by: fang yi <eric.fangyi@huawei.com> Signed-off-by: Zheng Zengkai <zhengzengkai@huawei.com> Signed-off-by: shaodenghui <shaodenghui@huawei.com> --- include/linux/workqueue.h | 1 + include/scsi/libiscsi.h | 1 + 2 files changed, 2 insertions(+) diff --git a/include/linux/workqueue.h b/include/linux/workqueue.h index 961909875..485c0f5b2 100644 --- a/include/linux/workqueue.h +++ b/include/linux/workqueue.h @@ -415,6 +415,7 @@ enum { __WQ_ORDERED = 1 << 17, /* internal: workqueue is ordered */ __WQ_LEGACY = 1 << 18, /* internal: create*_workqueue() */ __WQ_ORDERED_EXPLICIT = 1 << 19, /* internal: alloc_ordered_workqueue() */ + __WQ_DYNAMIC = 1 << 25, /* internal: only support single work order WQ */ WQ_MAX_ACTIVE = 512, /* I like 512, better ideas? */ WQ_UNBOUND_MAX_ACTIVE = WQ_MAX_ACTIVE, diff --git a/include/scsi/libiscsi.h b/include/scsi/libiscsi.h index d253b82e9..ead17d222 100644 --- a/include/scsi/libiscsi.h +++ b/include/scsi/libiscsi.h @@ -266,6 +266,7 @@ struct iscsi_conn { /* custom statistics */ uint32_t eh_abort_cnt; uint32_t fmr_unalign_cnt; + int intimate_cpu; KABI_RESERVE(1) KABI_RESERVE(2) -- 2.33.0
From: shaodenghui <shaodenghui@huawei.com> euleros inclusion category: feature bugzilla: https://gitee.com/openeuler/kernel/issues/I4IZNO CVE: NA ------------------------------------------------- Currently, single thread workqueue only have single pwq, all of works are queued the same workerpool. This is not optimal on NUMA machines, will cause workers jump around across node. This patch add a new wq flags __WQ_DYNAMIC, this new kind of single thread workqueue creates a separate pwq covering the intersecting CPUS for each NUMA node which has online CPUS in @attrs->cpumask instead of mapping all entries of numa_pwq_tbl[] to the same pwq. After this, we can specify the @cpu of queue_work_on, so the work can be executed on the same NUMA node of the specified @cpu. This kind of wq only support single work, multi works can't guarantee the work's order. Signed-off-by: Biaoxiang Ye <yebiaoxiang@huawei.com> Acked-by: Hanjun Guo <guohanjun@huawei.com> Signed-off-by: Yang Yingliang <yangyingliang@huawei.com> Reviewed-by: fang yi <eric.fangyi@huawei.com> Signed-off-by: Zheng Zengkai <zhengzengkai@huawei.com> Signed-off-by: shaodenghui <shaodenghui@huawei.com> --- kernel/workqueue.c | 15 ++++++++++----- 1 file changed, 10 insertions(+), 5 deletions(-) diff --git a/kernel/workqueue.c b/kernel/workqueue.c index 1888741f5edd..fc733366d5f4 100644 --- a/kernel/workqueue.c +++ b/kernel/workqueue.c @@ -4355,6 +4355,9 @@ apply_wqattrs_prepare(struct workqueue_struct *wq, * it even if we don't use it immediately. */ copy_workqueue_attrs(new_attrs, attrs); + if (wq->flags & __WQ_DYNAMIC) + new_attrs->ordered = false; + wqattrs_actualize_cpumask(new_attrs, unbound_cpumask); cpumask_copy(new_attrs->__pod_cpumask, new_attrs->cpumask); ctx->dfl_pwq = alloc_unbound_pwq(wq, new_attrs); @@ -4591,10 +4594,12 @@ static int alloc_and_link_pwqs(struct workqueue_struct *wq) cpus_read_lock(); if (wq->flags & __WQ_ORDERED) { ret = apply_workqueue_attrs(wq, ordered_wq_attrs[highpri]); - /* there should only be single pwq for ordering guarantee */ - WARN(!ret && (wq->pwqs.next != &wq->dfl_pwq->pwqs_node || - wq->pwqs.prev != &wq->dfl_pwq->pwqs_node), - "ordering guarantee broken for workqueue %s\n", wq->name); + if (!(wq->flags & __WQ_DYNAMIC)) { + /* there should only be single pwq for ordering guarantee */ + WARN(!ret && (wq->pwqs.next != &wq->dfl_pwq->pwqs_node || + wq->pwqs.prev != &wq->dfl_pwq->pwqs_node), + "ordering guarantee broken for workqueue %s\n", wq->name); + } } else { ret = apply_workqueue_attrs(wq, unbound_std_wq_attrs[highpri]); } @@ -5798,7 +5803,7 @@ static int workqueue_apply_unbound_cpumask(const cpumask_var_t unbound_cpumask) continue; /* creating multiple pwqs breaks ordering guarantee */ - if (!list_empty(&wq->pwqs)) { + if (!list_empty(&wq->pwqs) && !(wq->flags & __WQ_DYNAMIC)) { if (wq->flags & __WQ_ORDERED_EXPLICIT) continue; wq->flags &= ~__WQ_ORDERED; -- 2.33.0
From: shaodenghui <shaodenghui@huawei.com> euleros inclusion category: feature bugzilla: https://gitee.com/openeuler/kernel/issues/I4IZNO CVE: NA ------------------------------------------------- On aarch64 NUMA machines, the kworker of iscsi created always jump around across node boundaries. If it work on the different node even different cpu package with the softirq of network interface, memcpy with in iscsi_tcp_segment_recv will be slow down, and iscsi got an terrible performance. In this patch, we trace the cpu of softirq, and tell queue_work_on to execute iscsi_xmitworker on the same NUMA node. The performance data as below: fio cmd: fio -filename=/dev/disk/by-id/wwn-0x6883fd3100a2ad260036281700000000 -direct=1 -iodepth=32 -rw=read -bs=64k -size=30G -ioengine=libaio -numjobs=1 -group_reporting -name=mytest -time_based -ramp_time=60 -runtime=60 before patch: Jobs: 1 (f=1): [R] [52.5% done] [852.3MB/0KB/0KB /s] [13.7K/0/0 iops] [eta 00m:57s] Jobs: 1 (f=1): [R] [53.3% done] [861.4MB/0KB/0KB /s] [13.8K/0/0 iops] [eta 00m:56s] Jobs: 1 (f=1): [R] [54.2% done] [868.2MB/0KB/0KB /s] [13.9K/0/0 iops] [eta 00m:55s] after pactch: Jobs: 1 (f=1): [R] [53.3% done] [1070MB/0KB/0KB /s] [17.2K/0/0 iops] [eta 00m:56s] Jobs: 1 (f=1): [R] [55.0% done] [1064MB/0KB/0KB /s] [17.3K/0/0 iops] [eta 00m:54s] Jobs: 1 (f=1): [R] [56.7% done] [1069MB/0KB/0KB /s] [17.1K/0/0 iops] [eta 00m:52s] cpu info: Architecture: aarch64 Byte Order: Little Endian CPU(s): 128 On-line CPU(s) list: 0-127 Thread(s) per core: 1 Core(s) per socket: 64 Socket(s): 2 NUMA node(s): 4 Model: 0 CPU max MHz: 2600.0000 CPU min MHz: 200.0000 BogoMIPS: 200.00 L1d cache: 64K L1i cache: 64K L2 cache: 512K L3 cache: 32768K NUMA node0 CPU(s): 0-31 NUMA node1 CPU(s): 32-63 NUMA node2 CPU(s): 64-95 NUMA node3 CPU(s): 96-127 Signed-off-by: Biaoxiang Ye <yebiaoxiang@huawei.com> Acked-by: Hanjun Guo <guohanjun@huawei.com> Signed-off-by: Yang Yingliang <yangyingliang@huawei.com> Reviewed-by: fang yi <eric.fangyi@huawei.com> Signed-off-by: Zheng Zengkai <zhengzengkai@huawei.com> Signed-off-by: shaodenghui <shaodenghui@huawei.com> --- drivers/scsi/iscsi_tcp.c | 9 +++++++++ drivers/scsi/libiscsi.c | 17 ++++++++++++----- 2 files changed, 21 insertions(+), 5 deletions(-) diff --git a/drivers/scsi/iscsi_tcp.c b/drivers/scsi/iscsi_tcp.c index 8e14cea15..54d01b9d3 100644 --- a/drivers/scsi/iscsi_tcp.c +++ b/drivers/scsi/iscsi_tcp.c @@ -170,6 +170,7 @@ static void iscsi_sw_tcp_data_ready(struct sock *sk) struct iscsi_sw_tcp_conn *tcp_sw_conn; struct iscsi_tcp_conn *tcp_conn; struct iscsi_conn *conn; + int current_cpu; trace_sk_data_ready(sk); @@ -180,6 +181,14 @@ static void iscsi_sw_tcp_data_ready(struct sock *sk) return; } tcp_conn = conn->dd_data; + + /* save intimate cpu when in softirq */ + if (!sock_owned_by_user_nocheck(sk)) { + current_cpu = smp_processor_id(); + if (conn->intimate_cpu != current_cpu) + conn->intimate_cpu = current_cpu; + } + tcp_sw_conn = tcp_conn->dd_data; if (tcp_sw_conn->queue_recv) diff --git a/drivers/scsi/libiscsi.c b/drivers/scsi/libiscsi.c index 0fda8905e..a6cbcbaff 100644 --- a/drivers/scsi/libiscsi.c +++ b/drivers/scsi/libiscsi.c @@ -89,9 +89,15 @@ inline void iscsi_conn_queue_xmit(struct iscsi_conn *conn) { struct Scsi_Host *shost = conn->session->host; struct iscsi_host *ihost = shost_priv(shost); + int intimate_cpu = conn->intimate_cpu; - if (ihost->workq) - queue_work(ihost->workq, &conn->xmitwork); + if (ihost->workq) { + /* we expect it to be excuted on the same numa of the intimate cpu */ + if ((intimate_cpu >= 0) && cpu_possible(intimate_cpu)) + queue_work_on(intimate_cpu, ihost->workq, &conn->xmitwork); + else + queue_work(ihost->workq, &conn->xmitwork); + } } EXPORT_SYMBOL_GPL(iscsi_conn_queue_xmit); @@ -2907,9 +2913,9 @@ struct Scsi_Host *iscsi_host_alloc(const struct scsi_host_template *sht, ihost = shost_priv(shost); if (xmit_can_sleep) { - ihost->workq = alloc_workqueue("iscsi_q_%d", - WQ_SYSFS | __WQ_LEGACY | WQ_MEM_RECLAIM | WQ_UNBOUND, - 1, shost->host_no); + /* this kind of workqueue only support single work */ + ihost->workq = alloc_ordered_workqueue("iscsi_q_%d", __WQ_LEGACY | WQ_MEM_RECLAIM | + __WQ_DYNAMIC, shost->host_no); if (!ihost->workq) goto free_host; } @@ -3190,6 +3196,7 @@ iscsi_conn_setup(struct iscsi_cls_session *cls_session, int dd_size, conn->c_stage = ISCSI_CONN_INITIAL_STAGE; conn->id = conn_idx; conn->exp_statsn = 0; + conn->intimate_cpu = -1; timer_setup(&conn->transport_timer, iscsi_check_transport_timeouts, 0); -- 2.33.0
-
jiangdongxu