- 12 Jan, 2023 2 commits
-
-
Mathieu Desnoyers authored
Two new auxiliary vector entries are introduced for rseq without matching increment of the AT_VECTOR_SIZE_BASE, which causes failures with CONFIG_HARDENED_USERCOPY=y. Fixes: 317c8194 ("rseq: Introduce feature size and alignment ELF auxiliary vector entries") Reported-by:
Nathan Chancellor <nathan@kernel.org> Signed-off-by:
Mathieu Desnoyers <mathieu.desnoyers@efficios.com> Signed-off-by:
Peter Zijlstra (Intel) <peterz@infradead.org> Tested-by:
-
Mathieu Desnoyers authored
The mm_numa_cid related rseq patches from the series were not picked up into the tip tree, so enabling the mm_numa_cid test needs to be reverted. This reverts commit b344b8f2. Reported-by:
kernel test robot <oliver.sang@intel.com> Signed-off-by:
-
- 11 Jan, 2023 1 commit
-
-
Ingo Molnar authored
The following commit: c8997020 ("cputime: remove cputime_to_nsecs fallback") Removed an <asm/cputime.h> inclusion from <linux/sched/cputime.h>, but this broke the IA64 build: arch/ia64/kernel/time.c:110:6: warning: no previous prototype for 'arch_vtime_task_switch' [-Wmissing-prototypes] Add in the missing <asm/cputime.h> header to fix it. Fixes: c8997020 ("cputime: remove cputime_to_nsecs fallback") Reported-by:
kernel test robot <lkp@intel.com> Signed-off-by:
Ingo Molnar <mingo@kernel.org> Cc: linux-kernel@vger.kernel.org Cc: Nicholas Piggin <npiggin@gmail.com> Cc: Peter Zijlstra (Intel) <peterz@infradead.org> Cc: Alexander Gordeev <agordeev@linux.ibm.com> Signed-off-by:
-
- 07 Jan, 2023 5 commits
-
-
Michal Clapinski authored
Keep track of previously issued registrations and compare the result with MEMBARRIER_CMD_GET_REGISTRATIONS return value. Signed-off-by:
Michal Clapinski <mclapinski@google.com> Signed-off-by:
Paul E. McKenney <paulmck@kernel.org> Link: https://lore.kernel.org/r/20221207164338.1535591-3-mclapinski@google.com
-
Michal Clapinski authored
Provide a method to query previously issued registrations. Signed-off-by:
-
Lukasz Luba authored
The max CPU capacity is the same for all CPUs sharing frequency domain. There is a way to avoid heavy operations in a loop for each CPU by leveraging this knowledge. Thus, simplify the looping code in the sugov_next_freq_shared() and drop heavy multiplications. Instead, use simple max() to get the highest utilization from these CPUs. This is useful for platforms with many (4 or 6) little CPUs. We avoid heavy 2*PD_CPU_NUM multiplications in that loop, which is called billions of times, since it's not limited by the schedutil time delta filter in sugov_should_update_freq(). When there was no need to change frequency the code bailed out, not updating the sg_policy::last_freq_update_time. Then every visit after delta_ns time longer than the sg_policy::freq_update_delay_ns goes through and triggers the next frequency calculation code. Although, if the next frequency, as outcome of that, would be the same as current frequency, we won't update the sg_policy::last_freq_update_time and the story will be repeated (in a very short period, sometimes a few microseconds). The max CPU capacity must be fetched every time we are called, due to difficulties during the policy setup, where we are not able to get the normalized CPU capacity at the right time. The fetched CPU capacity value is than used in sugov_iowait_apply() to calculate the right boost. This required a few changes in the local functions and arguments. The capacity value should hopefully be fetched once when needed and then passed over CPU registers to those functions. Signed-off-by:
Lukasz Luba <lukasz.luba@arm.com> Signed-off-by:
-
Chengming Zhou authored
ttwu_do_activate() is used for a complete wakeup, in which we will activate_task() and use ttwu_do_wakeup() to mark the task runnable and perform wakeup-preemption, also call class->task_woken() callback and update the rq->idle_stamp. Since ttwu_runnable() is not a complete wakeup, don't need all those done in ttwu_do_wakeup(), so we can move those to ttwu_do_activate() to simplify ttwu_do_wakeup(), making it only mark the task runnable to be reused in ttwu_runnable() and try_to_wake_up(). This patch should not have any functional changes. Suggested-by:
Chengming Zhou <zhouchengming@bytedance.com> Signed-off-by:
-
Chengming Zhou authored
ttwu_runnable() is used as a fast wakeup path when the wakee task is running on CPU or runnable on RQ, in both cases we can just set its state to TASK_RUNNING to prevent a sleep. If the wakee task is on_cpu running, we don't need to update_rq_clock() or check_preempt_curr(). But if the wakee task is on_rq && !on_cpu (e.g. an IRQ hit before the task got to schedule() and the task been preempted), we should check_preempt_curr() to see if it can preempt the current running. This also removes the class->task_woken() callback from ttwu_runnable(), which wasn't required per the RT/DL implementations: any required push operation would have been queued during class->set_next_task() when p got preempted. ttwu_runnable() also loses the update to rq->idle_stamp, as by definition the rq cannot be idle in this scenario. Suggested-by:
Valentin Schneider <vschneid@redhat.com> Suggested-by:
Steven Rostedt (Google) <rostedt@goodmis.org> Link: https://lore.kernel.org/r/20221223103257.4962-1-zhouchengming@bytedance.com
-
- 05 Jan, 2023 2 commits
-
-
Qais Yousef authored
Add a document explaining the util clamp feature: what it is and how to use it. The new document hopefully covers everything one needs to know about uclamp. Signed-off-by:
Qais Yousef <qais.yousef@arm.com> Signed-off-by:
Qais Yousef (Google) <qyousef@layalina.io> Signed-off-by:
-
Bing Huang authored
sched_init_domains() is only used in initialization Signed-off-by:
Bing Huang <huangbing@kylinos.cn> Signed-off-by:
-
- 02 Jan, 2023 1 commit
-
-
Mathieu Desnoyers authored
sched_mm_cid_after_execve() does not expect NULL t->mm, but it may happen if a usermodehelper kthread fails when attempting to execute a binary. sched_mm_cid_fork() can be issued from a usermodehelper kthread, which has t->flags PF_KTHREAD set. Fixes: af7f588d ("sched: Introduce per-memory-map concurrency ID") Reported-by:
kernel test robot <yujie.liu@intel.com> Reported-by:
Borislav Petkov <bp@alien8.de> Signed-off-by:
-
- 27 Dec, 2022 28 commits
-
-
Nicholas Piggin authored
The archs that use cputime_to_nsecs() internally provide their own definition and don't need the fallback. cputime_to_usecs() unused except in this fallback, and is not defined anywhere. This removes the final remnant of the cputime_t code from the kernel. Signed-off-by:
Nicholas Piggin <npiggin@gmail.com> Signed-off-by:
Alexander Gordeev <agordeev@linux.ibm.com> Link: https://lore.kernel.org/r/20221220070705.2958959-1-npiggin@gmail.com
-
Hao Jia authored
When select_idle_capacity() starts scanning for an idle CPU, it starts with target CPU that has already been checked in select_idle_sibling(). So we start checking from the next CPU and try the target CPU at the end. Similarly for task_numa_assign(), we have just checked numa_migrate_on of dst_cpu, so start from the next CPU. This also works for steal_cookie_task(), the first scan must fail and start directly from the next one. Signed-off-by:
Hao Jia <jiahao.os@bytedance.com> Signed-off-by:
Vincent Guittot <vincent.guittot@linaro.org> Acked-by:
Mel Gorman <mgorman@techsingularity.net> Link: https://lore.kernel.org/r/20221216062406.7812-3-jiahao.os@bytedance.com
-
Hao Jia authored
In update_numa_stats() we try to find an idle cpu on the NUMA node, preferably an idle core. we can stop looking for the next idle core or idle cpu after finding an idle core. But we can't stop the whole loop of scanning the CPU, because we need to calculate approximate NUMA stats at a point in time. For example, the src and dst nr_running is needed by task_numa_find_cpu(). Signed-off-by:
-
Matthew Wilcox (Oracle) authored
With a modified container_of() that preserves constness, the compiler finds some pointers which should have been marked as const. task_of() also needs to become const-preserving for the !FAIR_GROUP_SCHED case so that cfs_rq_of() can take a const argument. No change to generated code. Signed-off-by:
Matthew Wilcox (Oracle) <willy@infradead.org> Signed-off-by:
-
Mathieu Desnoyers authored
Add mm_numa_cid tests to the run_param_test.sh test script. Signed-off-by:
-
Mathieu Desnoyers authored
Add the mm_cid field to the rseq_update event, allowing tracers to follow which mm_cid is observed by user-space, and whether negative mm_cid values are visible in case of internal scheduler implementation issues. Signed-off-by:
-
Mathieu Desnoyers authored
Report and abort when a negative concurrency ID value is observed by the spinlock test. Signed-off-by:
-
Mathieu Desnoyers authored
Adapt to the rseq.h API changes introduced by commits "selftests/rseq: <arch>: Template memory ordering and percpu access mode". Build a new param_test_mm_cid, param_test_mm_cid_benchmark, and param_test_mm_cid_compare_twice executables to test the new "mm_cid" rseq field. Signed-off-by:
-
Mathieu Desnoyers authored
Adapt to the rseq.h API changes introduced by commits "selftests/rseq: <arch>: Template memory ordering and percpu access mode". Build a new basic_percpu_ops_mm_cid_test to test the new "mm_cid" rseq field. Signed-off-by:
-
Mathieu Desnoyers authored
Introduce a rseq-riscv-bits.h template header which is internally included to generate the static inline functions covering: - relaxed and release memory ordering, - per-cpu-id and per-mm-cid per-cpu data access. Signed-off-by:
-
Mathieu Desnoyers authored
Introduce a rseq-s390-bits.h template header which is internally included to generate the static inline functions covering: - relaxed and release memory ordering, - per-cpu-id and per-mm-cid per-cpu data access. Signed-off-by:
-
Mathieu Desnoyers authored
Introduce a rseq-ppc-bits.h template header which is internally included to generate the static inline functions covering: - relaxed and release memory ordering, - per-cpu-id and per-mm-cid per-cpu data access. Signed-off-by:
-
Mathieu Desnoyers authored
Introduce a rseq-mips-bits.h template header which is internally included to generate the static inline functions covering: - relaxed and release memory ordering, - per-cpu-id and per-mm-cid per-cpu data access. Signed-off-by:
-
Mathieu Desnoyers authored
Introduce a rseq-arm64-bits.h template header which is internally included to generate the static inline functions covering: - relaxed and release memory ordering, - per-cpu-id and per-mm-cid per-cpu data access. Signed-off-by:
-
Mathieu Desnoyers authored
Introduce a rseq-arm-bits.h template header which is internally included to generate the static inline functions covering: - relaxed and release memory ordering, - per-cpu-id and per-mm-cid per-cpu data access. Signed-off-by:
-
Mathieu Desnoyers authored
Introduce a rseq-x86-bits.h template header which is internally included to generate the static inline functions covering: - relaxed and release memory ordering, - per-cpu-id and per-mm-cid per-cpu data access. This introduces changes to the rseq.h selftests API which require to update the rseq selftest programs. Similar API/templating changes need to be done for other architectures. Signed-off-by:
-
Mathieu Desnoyers authored
Add support for the mm_cid field (per-memory-map concurrency ID) of struct rseq to rseq selftests. Signed-off-by:
-
Mathieu Desnoyers authored
This code is not currently build by the test Makefile, adds complexity, and is not overall useful considering that the abort handling loops to retry the fast-path. Signed-off-by:
-
Mathieu Desnoyers authored
If a memory map has fewer threads than there are cores on the system, or is limited to run on few cores concurrently through sched affinity or cgroup cpusets, the concurrency IDs will be values close to 0, thus allowing efficient use of user-space memory for per-cpu data structures. Signed-off-by:
-
Mathieu Desnoyers authored
This feature allows the scheduler to expose a per-memory map concurrency ID to user-space. This concurrency ID is within the possible cpus range, and is temporarily (and uniquely) assigned while threads are actively running within a memory map. If a memory map has fewer threads than cores, or is limited to run on few cores concurrently through sched affinity or cgroup cpusets, the concurrency IDs will be values close to 0, thus allowing efficient use of user-space memory for per-cpu data structures. This feature is meant to be exposed by a new rseq thread area field. The primary purpose of this feature is to do the heavy-lifting needed by memory allocators to allow them to use per-cpu data structures efficiently in the following situations: - Single-threaded applications, - Multi-threaded applications on large systems (many cores) with limited cpu affinity mask, - Multi-threaded applications on large systems (many cores) with restricted cgroup cpuset per container. One of the key concern from scheduler maintainers is the overhead associated with additional spin locks or atomic operations in the scheduler fast-path. This is why the following optimization is implemented. On context switch between threads belonging to the same memory map, transfer the mm_cid from prev to next without any atomic ops. This takes care of use-cases involving frequent context switch between threads belonging to the same memory map. Additional optimizations can be done if the spin locks added when context switching between threads belonging to different memory maps end up being a performance bottleneck. Those are left out of this patch though. A performance impact would have to be clearly demonstrated to justify the added complexity. The credit goes to Paul Turner (Google) for the original virtual cpu id idea. This feature is implemented based on the discussions with Paul Turner and Peter Oskolkov (Google), but I took the liberty to implement scheduler fast-path optimizations and my own NUMA-awareness scheme. The rumor has it that Google have been running a rseq vcpu_id extension internally in production for a year. The tcmalloc source code indeed has comments hinting at a vcpu_id prototype extension to the rseq system call [1]. The following benchmarks do not show any significant overhead added to the scheduler context switch by this feature: * perf bench sched messaging (process) Baseline: 86.5±0.3 ms With mm_cid: 86.7±2.6 ms * perf bench sched messaging (threaded) Baseline: 84.3±3.0 ms With mm_cid: 84.7±2.6 ms * hackbench (process) Baseline: 82.9±2.7 ms With mm_cid: 82.9±2.9 ms * hackbench (threaded) Baseline: 85.2±2.6 ms With mm_cid: 84.4±2.9 ms [1] https://github.com/google/tcmalloc/blob/master/tcmalloc/internal/linux_syscall_support.h#L26Signed-off-by:
-
Mathieu Desnoyers authored
Test the NUMA node id extension rseq field. Compare it against the value returned by the getcpu(2) system call while pinned on a specific core. Signed-off-by:
-
Mathieu Desnoyers authored
Use the ELF auxiliary vector AT_RSEQ_FEATURE_SIZE to detect the RSEQ features supported by the kernel. Signed-off-by:
-
Mathieu Desnoyers authored
Adding the NUMA node id to struct rseq is a straightforward thing to do, and a good way to figure out if anything in the user-space ecosystem prevents extending struct rseq. This NUMA node id field allows memory allocators such as tcmalloc to take advantage of fast access to the current NUMA node id to perform NUMA-aware memory allocation. It can also be useful for implementing fast-paths for NUMA-aware user-space mutexes. It also allows implementing getcpu(2) purely in user-space. Signed-off-by:
-
Mathieu Desnoyers authored
Introduce the extensible rseq ABI, where the feature size supported by the kernel and the required alignment are communicated to user-space through ELF auxiliary vectors. This allows user-space to call rseq registration with a rseq_len of either 32 bytes for the original struct rseq size (which includes padding), or larger. If rseq_len is larger than 32 bytes, then it must be large enough to contain the feature size communicated to user-space through ELF auxiliary vectors. Signed-off-by:
-
Mathieu Desnoyers authored
Export the rseq feature size supported by the kernel as well as the required allocation alignment for the rseq per-thread area to user-space through ELF auxiliary vector entries. This is part of the extensible rseq ABI. Signed-off-by:
-
Mathieu Desnoyers authored
When linking the selftests against a libc which does not handle rseq registration (before 2.35), rseq thread registration silently succeed even with CONFIG_RSEQ=n because it erroneously thinks that libc is handling rseq registration. This is caused by setting the rseq ownership flag only after the rseq_available() check. It should rather be set before the rseq_available() check. Set the rseq_size to 0 (error value) immediately after the rseq_available() check fails rather than in the thread registration functions. Signed-off-by:
-
Josh Don authored
CFS bandwidth currently distributes new runtime and unthrottles cfs_rq's inline in an hrtimer callback. Runtime distribution is a per-cpu operation, and unthrottling is a per-cgroup operation, since a tg walk is required. On machines with a large number of cpus and large cgroup hierarchies, this cpus*cgroups work can be too much to do in a single hrtimer callback: since IRQ are disabled, hard lockups may easily occur. Specifically, we've found this scalability issue on configurations with 256 cpus, O(1000) cgroups in the hierarchy being throttled, and high memory bandwidth usage. To fix this, we can instead unthrottle cfs_rq's asynchronously via a CSD. Each cpu is responsible for unthrottling itself, thus sharding the total work more fairly across the system, and avoiding hard lockups. Signed-off-by:
Josh Don <joshdon@google.com> Signed-off-by:
-
Bing Huang authored
init_defrootdomain is only used in initialization Signed-off-by:
Randy Dunlap <rdunlap@infradead.org> Reviewed-by:
-
- 25 Dec, 2022 1 commit
-
-
Linus Torvalds authored
-
