- Kernel - mailweb.openeuler.org

[PATCH openEuler-1.0-LTS] mmc: via-sdmmc: fix return value check of mmc_add_host()
by Yi Yang 16 Jan '26

16 Jan '26

From: Yang Yingliang <yangyingliang(a)huawei.com> stable inclusion from stable-v4.19.270 commit 95025a8dd0ec015872f6c16473fe04d6264e68ca category: bugfix bugzilla: https://atomgit.com/src-openeuler/kernel/issues/13164 CVE: CVE-2022-50846 Reference: https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git/commit/?id… -------------------------------- [ Upstream commit e4e46fb61e3bb4628170810d3f2b996b709b90d9 ] mmc_add_host() may return error, if we ignore its return value, it will lead two issues: 1. The memory that allocated in mmc_alloc_host() is leaked. 2. In the remove() path, mmc_remove_host() will be called to delete device, but it's not added yet, it will lead a kernel crash because of null-ptr-deref in device_del(). Fix this by checking the return value and goto error path which will call mmc_free_host(). Fixes: f0bf7f61b840 ("mmc: Add new via-sdmmc host controller driver") Signed-off-by: Yang Yingliang <yangyingliang(a)huawei.com> Link: https://lore.kernel.org/r/20221108130949.1067699-1-yangyingliang@huawei.com Signed-off-by: Ulf Hansson <ulf.hansson(a)linaro.org> Signed-off-by: Sasha Levin <sashal(a)kernel.org> Signed-off-by: Yi Yang <yiyang13(a)huawei.com> --- drivers/mmc/host/via-sdmmc.c | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-) diff --git a/drivers/mmc/host/via-sdmmc.c b/drivers/mmc/host/via-sdmmc.c index 32c4211506fc..a86e4b3b4060 100644 --- a/drivers/mmc/host/via-sdmmc.c +++ b/drivers/mmc/host/via-sdmmc.c @@ -1158,7 +1158,9 @@ static int via_sd_probe(struct pci_dev *pcidev, pcidev->subsystem_device == 0x3891) sdhost->quirks = VIA_CRDR_QUIRK_300MS_PWRDELAY; - mmc_add_host(mmc); + ret = mmc_add_host(mmc); + if (ret) + goto unmap; return 0; -- 2.25.1

2 1

[PATCH openEuler-1.0-LTS] fbcon: Fix the issue of uninitialized charcount in the remaining consoles
by Luo Gengkun 15 Jan '26

15 Jan '26

HULK inclusion category: bugfix bugzilla: https://atomgit.com/src-openeuler/kernel/issues/12587 CVE: NA ---------------------------------------------------------------------- After commit 054a54161b88 ("fbdev: bitblit: bound-check glyph index in bit_putcs*") was merged, using alt+ctrl+f1 to switch the tty from tty0 to tty1 results in garbled display. The reason is the vc->vc_font.charcount is 0, it is clearly an uninitialized value. The mainline is fine because commit a1ac250a82a5 ("fbcon: Avoid using FNTCHARCNT() and hard-coded built-in font charcount") assigns the fvc->vc_font.charcount to vc->vc_font.charcount. Fixes: f9a6134c4b54 ("fbdev: bitblit: bound-check glyph index in bit_putcs*") Signed-off-by: Luo Gengkun <luogengkun2(a)huawei.com> --- drivers/video/fbdev/core/fbcon.c | 1 + 1 file changed, 1 insertion(+) diff --git a/drivers/video/fbdev/core/fbcon.c b/drivers/video/fbdev/core/fbcon.c index 83a4949e2497..4043b0b19140 100644 --- a/drivers/video/fbdev/core/fbcon.c +++ b/drivers/video/fbdev/core/fbcon.c @@ -1032,6 +1032,7 @@ static void fbcon_init(struct vc_data *vc, int init) fvc->vc_font.data); vc->vc_font.width = fvc->vc_font.width; vc->vc_font.height = fvc->vc_font.height; + vc->vc_font.charcount = fvc->vc_font.charcount; p->userfont = t->userfont; if (p->userfont) -- 2.34.1

2 1

[PATCH openEuler-1.0-LTS] fbcon: Fix the issue of uninitialized charcount in the remaining consoles
by Luo Gengkun 15 Jan '26

15 Jan '26

Offering: HULK hulk inclusion category: bugfix bugzilla: https://atomgit.com/src-openeuler/kernel/issues/12587 ---------------------------------------------------------------------- After commit 054a54161b88 ("fbdev: bitblit: bound-check glyph index in bit_putcs*") was merged, using alt+ctrl+f1 to switch the tty from tty0 to tty1 results in garbled display. The reason is the vc->vc_font.charcount is 0, it is clearly an uninitialized value. The mainline is fine because commit a1ac250a82a5 ("fbcon: Avoid using FNTCHARCNT() and hard-coded built-in font charcount") assigns the fvc->vc_font.charcount to vc->vc_font.charcount. Fixes: f9a6134c4b54 ("fbdev: bitblit: bound-check glyph index in bit_putcs*") Signed-off-by: Luo Gengkun <luogengkun2(a)huawei.com> --- drivers/video/fbdev/core/fbcon.c | 1 + 1 file changed, 1 insertion(+) diff --git a/drivers/video/fbdev/core/fbcon.c b/drivers/video/fbdev/core/fbcon.c index 83a4949e2497..4043b0b19140 100644 --- a/drivers/video/fbdev/core/fbcon.c +++ b/drivers/video/fbdev/core/fbcon.c @@ -1032,6 +1032,7 @@ static void fbcon_init(struct vc_data *vc, int init) fvc->vc_font.data); vc->vc_font.width = fvc->vc_font.width; vc->vc_font.height = fvc->vc_font.height; + vc->vc_font.charcount = fvc->vc_font.charcount; p->userfont = t->userfont; if (p->userfont) -- 2.34.1

2 1

[PATCH] mfs: fix the compile error of use demo tool mdsd_prefetch
by Kuang Kai 15 Jan '26

15 Jan '26

From: kuangkai <kuangkai(a)kylinos.cn> kylin inclusion category: bugfix bugzilla: https://atomgit.com/openeuler/kernel/issues/8354 CVE: NA -------------------------- Commit daa2c7d2bbd0 ("mfs: Add prefetch demo") use the pthread_create but not declared and linked, this cause compile mfs/tool with make fail: # cd tools/mfs && make mfsd_prefetch.cpp:118:23: error: 'pthread_create' was not declared in this scope Signed-off-by: kuangkai <kuangkai(a)kylinos.cn> --- tools/mfs/Makefile | 2 +- tools/mfs/mfsd_prefetch.cpp | 1 + 2 files changed, 2 insertions(+), 1 deletion(-) diff --git a/tools/mfs/Makefile b/tools/mfs/Makefile index f99c9cde828f..3fc331c7ebde 100644 --- a/tools/mfs/Makefile +++ b/tools/mfs/Makefile @@ -1,7 +1,7 @@ # SPDX-License-Identifier: GPL-2.0 # Makefile for mfs demo -CFLAGS = -Wall -Wextra +CFLAGS = -Wall -Wextra -lpthread PROGS := mfsd mfsd_prefetch diff --git a/tools/mfs/mfsd_prefetch.cpp b/tools/mfs/mfsd_prefetch.cpp index d35aeb31941a..a26b271b3a00 100644 --- a/tools/mfs/mfsd_prefetch.cpp +++ b/tools/mfs/mfsd_prefetch.cpp @@ -23,6 +23,7 @@ #include <sys/statfs.h> #include <sys/stat.h> #include <sys/mman.h> +#include <pthread.h> #include <map> #include <string> -- 2.39.2 (Apple Git-143)

1 0

[PATCH OLK-6.6 0/1] cpufreq/sched: Explicitly synchronize limits_changed flag handling
by Lifeng Zheng 15 Jan '26

15 Jan '26

From: Hongye Lin <linhongye(a)h-partners.com> driver inclusion category: bugfix bugzilla: https://atomgit.com/openeuler/kernel/issues/8343 ---------------------------------------------------------------------- Rafael J. Wysocki (1): cpufreq/sched: Explicitly synchronize limits_changed flag handling kernel/sched/cpufreq_schedutil.c | 28 ++++++++++++++++++++++++---- 1 file changed, 24 insertions(+), 4 deletions(-) -- 2.33.0

2 2

[PATCH OLK-5.10 0/2] kmemleak: Fix kmemleak false warnings
by Qi Xi 15 Jan '26

15 Jan '26

This series fixes kmemleak false warnings when reserving crash kernel memory. Patch 1 moves kmemleak_ignore_phys() to reserve_crashkernel_high() where memory is actually tracked by kmemleak. Patch 2 ensures kmemleak_ignore_phys() is only called when new memory is allocated, not when crash_base is pre-set via cmdline. ZhangPeng (2): arm64: kdump: fix kmemleak unknown object warning arm64: kdump: fix kmemleak unknown object warning when crash base is set kernel/crash_core.c | 14 ++++++-------- 1 file changed, 6 insertions(+), 8 deletions(-) -- 2.33.0

2 3

[PATCH OLK-6.6] armv8_pmuv3: Keep event consistency of hw_metric events
by Yushan Wang 15 Jan '26

15 Jan '26

driver inclusion category: bugfix bugzilla: https://atomgit.com/openeuler/kernel/issues/8310 CVE: NA ----------------------------------------------- hw_metric events should have same pmus. Check the event types and event pmus for event consistency. Signed-off-by: Yushan Wang <wangyushan12(a)huawei.com> Signed-off-by: Ying Jiang <jiangying44(a)h-partners.com> Signed-off-by: jiangying <jiangying44(a)h-partners.com> --- drivers/perf/arm_pmuv3.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) diff --git a/drivers/perf/arm_pmuv3.c b/drivers/perf/arm_pmuv3.c index 5d750678a70e..4f27c75cfdfa 100644 --- a/drivers/perf/arm_pmuv3.c +++ b/drivers/perf/arm_pmuv3.c @@ -1159,7 +1159,8 @@ static int armv8pmu_check_hw_metric_event(struct pmu_hw_events *cpuc, if (event == leader) return 0; - if (!armv8pmu_event_is_hw_metric(leader)) + if (leader->pmu != event->pmu || + !armv8pmu_event_is_hw_metric(leader)) return -EINVAL; for_each_sibling_event(sibling, leader) { -- 2.33.0

2 1

[PATCH OLK-5.10 0/4] fix kmemleak false warning
by Qi Xi 15 Jan '26

15 Jan '26

Patch #1 introduces the OBJECT_PHYS flag and create_object_phys() for physical address-allocated kmemleak objects. Patch #2 adds object_phys_tree_root to store physical address objects and uses raw physical addresses instead of virtual addresses. Patch #3 fixes false positives in KFENCE by using kmemleak_ignore_phys() for early memblock-allocated pools. Patch #4 fixes similar false positives in percpu by replacing kmemleak_free() with kmemleak_ignore_phys(). Patrick Wang (3): mm: kmemleak: add OBJECT_PHYS flag for objects allocated with physical address mm: kmemleak: add rbtree and store physical address for objects allocated with PA mm: percpu: use kmemleak_ignore_phys() instead of kmemleak_free() Yee Lee (1): mm: kfence: apply kmemleak_ignore_phys on early allocated pool mm/kfence/core.c | 18 ++--- mm/kmemleak.c | 166 +++++++++++++++++++++++++++++++++-------------- mm/percpu.c | 6 +- 3 files changed, 131 insertions(+), 59 deletions(-) -- 2.33.0

2 5

[PATCH v2 OLK-6.6] xsched: Add XSched manual
by Liu Kai 14 Jan '26

14 Jan '26

hulk inclusion category: feature bugzilla: https://atomgit.com/openeuler/kernel/issues/8232 -------------------------------- Add user manual for XSched. Signed-off-by: Liu Kai <liukai284(a)huawei.com> --- Documentation/scheduler/xsched.md | 355 ++++++++++++++++++++++++++++++ 1 file changed, 355 insertions(+) create mode 100644 Documentation/scheduler/xsched.md diff --git a/Documentation/scheduler/xsched.md b/Documentation/scheduler/xsched.md new file mode 100644 index 000000000000..89ef01d42fc0 --- /dev/null +++ b/Documentation/scheduler/xsched.md @@ -0,0 +1,355 @@ +# XSched 用户指南 + +## 1 环境部署 + +### 1.1 安装 NPU 原生软件栈 + +本特性依赖 NPU 原生软件栈，需要安装 NPU driver、firmware、CANN + +### 1.2 安装XSched内核 + +#### 1.2.1 获取内核源码 + +```bash +# 下载源码 +git clone https://atomgit.com/openeuler/kernel.git -b OLK-6.6 OLK-6.6 + +# 进入内核目录 +cd OLK-6.6 +``` + +#### 1.2.2 安装内核 + +##### 1.2.2.1 修改内核配置 + +```shell +# 生成默认配置 +make openeuler_defconfig + +# 修改并保存配置 +vim .config + +CONFIG_XCU_SCHEDULER=y +CONFIG_XCU_VSTREAM=y +CONFIG_XSCHED_NR_CUS=8 # 根据 NPU 卡的数量配置 +CONFIG_XCU_SCHED_RT=y +CONFIG_XCU_SCHED_CFS=y +CONFIG_CGROUP_XCU=y +``` + +##### 1.2.2.2 编译安装 + +以下方式**二选一**： + +* 源码编译安装 + + ```bash + make clean && make -j$(nproc) + make modules_install -j$(nproc) INSTALL_MOD_STRIP=1 && INSTALL_MOD_STRIP=1 make install + grub2-set-default 0 + ``` + +* RPM 包编译安装 + + ```bash + INSTALL_MOD_STRIP=1 make rpm-pkg -j `nproc` + # kernel 和 devel 包都要安装，否则会导致驱动无法重编 + rpm -ivh <path_to_rpm> --force + grub2-set-default 0 + ``` + +##### 1.2.2.3 修改 cmdline + +```shell +# 修改内核引导文件，根据实际情况编辑 +vim /etc/grub2-efi.cfg + +# 在XSched内核新增 cmdline 配置，关闭驱动签名校验、开启cgroup-v2 +module.sig_enforce=0 systemd.unified_cgroup_hierarchy=1 cgroup_no_v1=all +``` + +保存引导文件后，重启切换内核 + +### 1.3 重编驱动 + +**！！需先安装好原生驱动！！**，`npu-smi info` 检查驱动是否安装成功 + +#### 1.3.1 修改驱动 + +获取驱动源码 + +```shell +mkdir <new_driver_dir> + +./Ascend-hdk-910b-npu-driver_xx.x.x_linux-aarch64.run --tar -xvf -C <new_driver_dir> +``` + +下载驱动补丁 [XSched driver patch](https://gitee.com/openeuler/kernel/commit/8d50448f11b697a177b63d3ccb… 适配 NPU 驱动 + +```shell +cp 0001-Adapt-910b-npu-driver-for-xsched.patch <new_driver_dir>/driver/kernel/ +cd <new_driver_dir>/driver/kernel/ + +# 初始化 git 仓库并应用补丁 +git init +git add . +git commit -m "npu init" +# 如果有冲突则根据 .rej 文件适配冲突代码 +git am 0001-Adapt-910b-npu-driver-for-xsched.patch --reject +``` + +#### 1.3.2 替换驱动 + +```shell +# 返回驱动根目录 +cd ../../ + +# 备份原生驱动 ko +cp -r <new_driver_dir>/driver/host <new_driver_dir>/driver/host-bak +rm -f <new_driver_dir>/driver/host/* + +cd <new_driver_dir>/driver/script +# 如果强制重编则增加 --force 参数，驱动 ko 会生成到 <new_driver_dir>/driver/host/ +sh run_driver_ko_rebuild.sh [--force] + +# 替换驱动 +for line in `ls /lib/modules/$(uname -r)/updates`; do \cp ../host/$line /lib/modules/$(uname -r)/updates; done +``` + +#### 1.3.3 检查驱动 + +重启后检查驱动是否替换成功 + +```shell +reboot +npu-smi info +``` + +### 1.4 libXSched + +XSched 用户态拦截层，拦截 CANN API 并转发到 XSched 内核 + +#### 1.4.1 编译 + +获取源码 + +```bash +git clone https://atomgit.com/openeuler/libXSched.git +``` + +编译源码 + +```bash +# 准备头文件 +cp OLK-6.6/include/uapi/linux/xsched/xcu_vstream.h /usr/include/linux + +# 编译生成 libucc_engine.so +make clean && make +``` + +#### 1.4.2 使用 + +使用 `LD_PRELOAD` 加载 `libucc_engine.so` + +```bash +LD_PRELOAD=<path_to_libucc> <run_model_script> +``` + +### 1.5 验证 + +```shell +# 开启 XSched 日志 +echo "file kernel/xsched/* +p" > /sys/kernel/debug/dynamic_debug/control + +# pytorch 框架 +LD_PRELOAD=<path_to_libucc> python3 -c "import torch;import torch_npu; a = torch.randn(3, 4).npu(); print(a + a);" + +# mindspore 框架 +LD_PRELOAD=<path_to_libucc> python3 -c "import mindspore;mindspore.set_device('Ascend');mindspore.run_check()" +``` + +`dmesg` 查看内核日志，检查是否有 XSCHED 日志，有日志则说明 XSched 环境部署成功，验证完后关闭 XSched 日志 `echo "file kernel/xsched/* -p" > /sys/kernel/debug/dynamic_debug/control` + +## 2 使用指南 + +### 2.1 Cgroup 接口 + +#### 2.1.1 配置 XSched 策略 + +使能 XSched 控制器 + +```bash +echo "+xcu" > /sys/fs/cgroup/cgroup.subtree_control +``` + +##### 2.1.1.1 可配置接口 + +* `xcu.sched_class`：配置调度类，cfs 或 rt + +* `xcu.period_ms`：配置获取算力资源时间片周期，默认 100ms（需先设置为 cfs 调度类） + +* `xcu.quota_ms`：配置周期内可分配的时间片，比如配置 50ms，则每 100ms 内可分配 50ms，-1 则是不管控算力资源（需先设置为 cfs 调度类） + +* `xcu.shares`：配置 cfs 任务的权重，权重越大，相比于其他任务的优先级越高，默认为 1024（需先设置为 cfs 调度类） + +* `xcu.stat`：查看统计信息，仅支持 cfs 调度类 + +##### 2.1.1.2 Host 使用方法 + +在主机侧使用 XSched 管控 ai 任务，需手动创建 cgroup 组 + +```bash +mkdir -p /sys/fs/cgroup/xsched_group + +# 按需配置 XSched 策略，例如 echo cfs > /sys/fs/cgroup/xsched_group/xcu.sched_class +echo <config_value> > /sys/fs/cgroup/xsched_group/xcu.<config_item> +``` + +将 ai 任务加入到 cgroup 中管控算力资源 + +```bash +# 运行 ai 任务 +LD_PRELOAD=<path_to_libucc> <run_model_script> + +# 加入 ai 任务到 cgroup +echo <pid> > /sys/fs/cgroup/xsched_group/cgroup.procs +``` + +##### 2.1.1.3 Docker 使用方法 + +启动 docker 容器时会自动挂载到 cgroup 目录（`/sys/fs/cgroup/user.slice`）下，docker 内的任务的 pid 也会自动添加到 cgroup.procs 下，无需手动创建 cgroup 组和添加任务 + +1. 修改 docker 配置，修改挂载的 cgroup 目录和 docker 的 cgroup 版本为 v2（**docker 需升级到支持 cgroup-v2 的版本**） + + ```bash + vim /etc/docker/daemon.json + + # 添加以下配置 + # 挂载的 cgroup 目录，目录名任意，以 docker.slice 为例 + "cgroup-parent": "docker.slice" + # 使用 cgroup-v2 管理 docker + "exec-opts": ["native.cgroupdriver=systemd"] + + # 修改保存后重启 docker 服务 + systemctl restart docker + ``` + + 因为 docker 默认的挂载目录是 `/sys/fs/cgroup/user.slice` ，`user.slice` 下还会有其他的用户任务在运行，可能会无法修改 `user.slice` 的 XSched 配置，所以需要修改 docker 的挂载目录为 `/sys/fs/cgroup/docker.slice` 方便配置和管理 + +2. 配置 docker 的 xsched 调度策略 + + ```bash + # 修改调度类为 cfs + echo cfs > /sys/fs/cgroup/docker.slice/xcu.sched_class + + # 启动容器 + docker start <container_name> + + # 配置对应容器的 xsched 策略 + echo <config_value> > /sys/fs/cgroup/docker.slice/docker-<docker_id>/xcu.<config_item> + + # 进入容器 + docker exec -it <container_name> bash + + # 运行 ai 任务（容器内也有 libucc_engine.so），pid 会自动添加到 /sys/fs/cgroup/docker.slice/docker-<docker_id>/cgroup.procs + LD_PRELOAD=<path_to_libucc> <run_model_script> + ``` + +### 2.2 Syscall 接口 + +提供两个 syscall 接口用于设置/获取 rt 调度类的任务优先级（`xsched_setattr/xsched_getattr`） + +* `pid`：ai 任务 pid，pid=0 表示当前进程 + +* `struct xsched_attr`：XSched 可配置属性 + + ```c + struct xsched_attr { + __u32 xsched_class; // 调度类，当前仅支持 RT（xsched_class=0） + __u32 xsched_priority; // rt 优先级，可配置范围 [1,5]，数字越大，优先级越高 + }; + ``` + +调用 syscall 接口示例程序： + +```c +#include <stdio.h> +#include <stdlib.h> +#include <unistd.h> +#include <string.h> +#include <errno.h> +#include <sys/syscall.h> +#include <stdint.h> +#include <sys/types.h> + +// Define xsched_attr structure +struct xsched_attr { + uint32_t xsched_class; // Scheduling class: XSCHED_TYPE_RT, XSCHED_TYPE_CFS + uint32_t xsched_priority; // Priority level: 1~5 +}; + +// Scheduling class constants (should match kernel header definitions) +#define XSCHED_TYPE_RT 0 // Real-time scheduling class +#define XSCHED_TYPE_CFS 1 // Completely Fair Scheduler class + +// System call numbers (update these with actual numbers) +#define __NR_xsched_setattr 467 +#define __NR_xsched_getattr 468 + +// Wrapper functions for system calls +static inline long xsched_setattr(pid_t pid, const struct xsched_attr *attr) +{ + return syscall(__NR_xsched_setattr, pid, attr); +} + +static inline long xsched_getattr(pid_t pid, struct xsched_attr *attr) +{ + return syscall(__NR_xsched_getattr, pid, attr); +} + +void main(void) +{ + pid_t pid = getpid(); + struct xsched_attr attr; + long ret; + + // Get current configuration + ret = xsched_getattr(pid, &attr); + if (ret < 0) { + printf("Failed to get configuration!\n"); + } + + attr.xsched_class = XSCHED_TYPE_RT; + attr.xsched_priority = 3; + + ret = xsched_setattr(pid, &attr); + if (ret < 0) { + printf("Failed to set configuration\n"); + } + + // Verify the configuration + memset(&attr, 0, sizeof(attr)); + ret = xsched_getattr(pid, &attr); + if (ret < 0) { + printf("Failed to verify configuration!\n"); + } +} +``` + +### 2.3 调试日志 + +开启/关闭 XSched 调试日志 + +```shell +echo "file kernel/xsched/* +p" > /sys/kernel/debug/dynamic_debug/control +echo "file kernel/xsched/* -p" > /sys/kernel/debug/dynamic_debug/control +``` + +开启/关闭 NPU 驱动调试日志 + +```shell +echo "module ascend_trs_core +p" > /sys/kernel/debug/dynamic_debug/control +echo "module ascend_trs_pm_adapt +p" > /sys/kernel/debug/dynamic_debug/control +echo "module ascend_trs_core -p" > /sys/kernel/debug/dynamic_debug/control +echo "module ascend_trs_pm_adapt -p" > /sys/kernel/debug/dynamic_debug/control +``` -- 2.34.1

2 1

[PATCH 001/415] bpf: Add ability to pin bpf timer to calling CPU
by Luo Gengkun 14 Jan '26

14 Jan '26

From: David Vernet <void(a)manifault.com> mainline inclusion from mainline-v6.7-rc1 commit d6247ecb6c1e17d7a33317090627f5bfe563cbb2 category: feature bugzilla: https://atomgit.com/openeuler/kernel/issues/8335 Reference: https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?… ---------------------------------------------------------------------- BPF supports creating high resolution timers using bpf_timer_* helper functions. Currently, only the BPF_F_TIMER_ABS flag is supported, which specifies that the timeout should be interpreted as absolute time. It would also be useful to be able to pin that timer to a core. For example, if you wanted to make a subset of cores run without timer interrupts, and only have the timer be invoked on a single core. This patch adds support for this with a new BPF_F_TIMER_CPU_PIN flag. When specified, the HRTIMER_MODE_PINNED flag is passed to hrtimer_start(). A subsequent patch will update selftests to validate. Signed-off-by: David Vernet <void(a)manifault.com> Signed-off-by: Daniel Borkmann <daniel(a)iogearbox.net> Acked-by: Song Liu <song(a)kernel.org> Acked-by: Hou Tao <houtao1(a)huawei.com> Link: https://lore.kernel.org/bpf/20231004162339.200702-2-void@manifault.com Signed-off-by: Luo Gengkun <luogengkun2(a)huawei.com> --- include/uapi/linux/bpf.h | 4 ++++ kernel/bpf/helpers.c | 5 ++++- tools/include/uapi/linux/bpf.h | 4 ++++ 3 files changed, 12 insertions(+), 1 deletion(-) diff --git a/include/uapi/linux/bpf.h b/include/uapi/linux/bpf.h index 782d56c4ebca..3fa3f9b1a052 100644 --- a/include/uapi/linux/bpf.h +++ b/include/uapi/linux/bpf.h @@ -5122,6 +5122,8 @@ union bpf_attr { * **BPF_F_TIMER_ABS** * Start the timer in absolute expire value instead of the * default relative one. + * **BPF_F_TIMER_CPU_PIN** + * Timer will be pinned to the CPU of the caller. * * Return * 0 on success. @@ -7346,9 +7348,11 @@ struct bpf_core_relo { * Flags to control bpf_timer_start() behaviour. * - BPF_F_TIMER_ABS: Timeout passed is absolute time, by default it is * relative to current time. + * - BPF_F_TIMER_CPU_PIN: Timer will be pinned to the CPU of the caller. */ enum { BPF_F_TIMER_ABS = (1ULL << 0), + BPF_F_TIMER_CPU_PIN = (1ULL << 1), }; /* BPF numbers iterator state */ diff --git a/kernel/bpf/helpers.c b/kernel/bpf/helpers.c index f40085f4de31..c889a9dee6c6 100644 --- a/kernel/bpf/helpers.c +++ b/kernel/bpf/helpers.c @@ -1363,7 +1363,7 @@ BPF_CALL_3(bpf_timer_start, struct bpf_async_kern *, timer, u64, nsecs, u64, fla if (in_nmi()) return -EOPNOTSUPP; - if (flags > BPF_F_TIMER_ABS) + if (flags & ~(BPF_F_TIMER_ABS | BPF_F_TIMER_CPU_PIN)) return -EINVAL; __bpf_spin_lock_irqsave(&timer->lock); t = timer->timer; @@ -1377,6 +1377,9 @@ BPF_CALL_3(bpf_timer_start, struct bpf_async_kern *, timer, u64, nsecs, u64, fla else mode = HRTIMER_MODE_REL_SOFT; + if (flags & BPF_F_TIMER_CPU_PIN) + mode |= HRTIMER_MODE_PINNED; + hrtimer_start(&t->timer, ns_to_ktime(nsecs), mode); out: __bpf_spin_unlock_irqrestore(&timer->lock); diff --git a/tools/include/uapi/linux/bpf.h b/tools/include/uapi/linux/bpf.h index a457d4f82da2..fc0ed3b6abf1 100644 --- a/tools/include/uapi/linux/bpf.h +++ b/tools/include/uapi/linux/bpf.h @@ -5122,6 +5122,8 @@ union bpf_attr { * **BPF_F_TIMER_ABS** * Start the timer in absolute expire value instead of the * default relative one. + * **BPF_F_TIMER_CPU_PIN** + * Timer will be pinned to the CPU of the caller. * * Return * 0 on success. @@ -7349,9 +7351,11 @@ struct bpf_core_relo { * Flags to control bpf_timer_start() behaviour. * - BPF_F_TIMER_ABS: Timeout passed is absolute time, by default it is * relative to current time. + * - BPF_F_TIMER_CPU_PIN: Timer will be pinned to the CPU of the caller. */ enum { BPF_F_TIMER_ABS = (1ULL << 0), + BPF_F_TIMER_CPU_PIN = (1ULL << 1), }; /* BPF numbers iterator state */ -- 2.34.1

1 414