- 04 Dec, 2015 8 commits
-
-
Byungchul Park authored
The current code accounts for the time a task was absent from the fair class (per ATTACH_AGE_LOAD). However it does not work correctly when a task got migrated or moved to another cgroup while outside of the fair class. This patch tries to address that by aging on migration. We locklessly read the 'last_update_time' stamp from both the old and new cfs_rq, ages the load upto the old time, and sets it to the new time. These timestamps should in general not be more than 1 tick apart from one another, so there is a definite bound on things. Signed-off-by:
Byungchul Park <byungchul.park@lge.com> [ Changelog, a few edits and !SMP build fix ] Signed-off-by:
Peter Zijlstra (Intel) <peterz@infradead.org> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Mike Galbraith <efault@gmx.de> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Link: http://lkml.kernel.org/r/1445616981-29904-2-git-send-email-byungchul.park@lge.comSigned-off-by:
Ingo Molnar <mingo@kernel.org>
-
Ingo Molnar authored
Signed-off-by: -
Peter Zijlstra authored
Oleg noticed that its possible to falsely observe p->on_cpu == 0 such that we'll prematurely continue with the wakeup and effectively run p on two CPUs at the same time. Even though the overlap is very limited; the task is in the middle of being scheduled out; it could still result in corruption of the scheduler data structures. CPU0 CPU1 set_current_state(...) <preempt_schedule> context_switch(X, Y) prepare_lock_switch(Y) Y->on_cpu = 1; finish_lock_switch(X) store_release(X->on_cpu, 0); try_to_wake_up(X) LOCK(p->pi_lock); t = X->on_cpu; // 0 context_switch(Y, X) prepare_lock_switch(X) X->on_cpu = 1; finish_lock_switch(Y) store_release(Y->on_cpu, 0); </preempt_schedule> schedule(); deactivate_task(X); X->on_rq = 0; if (X->on_rq) // false if (t) while (X->on_cpu) cpu_relax(); context_switch(X, ..) finish_lock_switch(X) store_release(X->on_cpu, 0); Avoid the load of X->on_cpu being hoisted over the X->on_rq load. Reported-by:Oleg Nesterov <oleg@redhat.com> Signed-off-by:
-
Peter Zijlstra authored
Explain how the control dependency and smp_rmb() end up providing ACQUIRE semantics and pair with smp_store_release() in finish_lock_switch(). Signed-off-by:
-
Hiroshi Shimamoto authored
/proc/stats shows invalid gtime when the thread is running in guest. When vtime accounting is not enabled, we cannot get a valid delta. The delta is calculated with now - tsk->vtime_snap, but tsk->vtime_snap is only updated when vtime accounting is runtime enabled. This patch makes task_gtime() just return gtime without computing the buggy non-existing tickless delta when vtime accounting is not enabled. Use context_tracking_is_enabled() to check if vtime is accounting on some cpu, in which case only we need to check the tickless delta. This way we fix the gtime value regression on machines not running nohz full. The kernel config contains CONFIG_VIRT_CPU_ACCOUNTING_GEN=y and CONFIG_NO_HZ_FULL_ALL=n and boot without nohz_full. I ran and stop a busy loop in VM and see the gtime in host. Dump the 43rd field which shows the gtime in every second: # while :; do awk '{print $3" "$43}' /proc/3955/task/4014/stat; sleep 1; done S 4348 R 7064566 R 7064766 R 7064967 R 7065168 S 4759 S 4759 During running busy loop, it returns large value. After applying this patch, we can see right gtime. # while :; do awk '{print $3" "$43}' /proc/10913/task/10956/stat; sleep 1; done S 5338 R 5365 R 5465 R 5566 R 5666 S 5726 S 5726 Signed-off-by:Hiroshi Shimamoto <h-shimamoto@ct.jp.nec.com> Signed-off-by:
Frederic Weisbecker <fweisbec@gmail.com> Signed-off-by:
-
Xunlei Pang authored
root_domain::rto_mask allocated through alloc_cpumask_var() contains garbage data, this may cause problems. For instance, When doing pull_rt_task(), it may do useless iterations if rto_mask retains some extra garbage bits. Worse still, this violates the isolated domain rule for clustered scheduling using cpuset, because the tasks(with all the cpus allowed) belongs to one root domain can be pulled away into another root domain. The patch cleans the garbage by using zalloc_cpumask_var() instead of alloc_cpumask_var() for root_domain::rto_mask allocation, thereby addressing the issues. Do the same thing for root_domain's other cpumask memembers: dlo_mask, span, and online. Signed-off-by:
Xunlei Pang <xlpang@redhat.com> Signed-off-by:
-
Sasha Levin authored
Because wakeups can (fundamentally) be late, a task might not be in the expected state. Therefore testing against a task's state is racy, and can yield false positives. Signed-off-by:
Sasha Levin <sasha.levin@oracle.com> Signed-off-by:
Linus Torvalds <torvalds@linux-foundation.org> Cc: Mike Galbraith <efault@gmx.de> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: oleg@redhat.com Fixes: 9067ac85 ("wake_up_process() should be never used to wakeup a TASK_STOPPED/TRACED task") Link: http://lkml.kernel.org/r/1448933660-23082-1-git-send-email-sasha.levin@oracle.comSigned-off-by:
-
Peter Zijlstra authored
Vladimir reported getting RCU stall warnings and bisected it back to commit: 74316201 ("sched: Remove proliferation of wait_on_bit() action functions") That commit inadvertently reversed the calls to schedule() and signal_pending(), thereby not handling the case where the signal receives while we sleep. Reported-by:
Vladimir Murzin <vladimir.murzin@arm.com> Tested-by:
-
- 23 Nov, 2015 19 commits
-
-
Byungchul Park authored
The comment describing migrate_task_rq_fair() says that the caller should hold p->pi_lock. But in some cases the caller can hold task_rq(p)->lock instead of p->pi_lock. So the comment is broken and this patch fixes it. Signed-off-by:
-
Oleg Nesterov authored
1. Change this code to use preempt_count_inc/preempt_count_dec; this way it works even if CONFIG_PREEMPT_COUNT=n, and we avoid the unnecessary __preempt_schedule() check (stop_sched_class is not preemptible). And this makes clear that we only want to make preempt_count() != 0 for __might_sleep() / schedule_debug(). 2. Change WARN_ONCE() to use %pf to print the function name and remove kallsyms_lookup/ksym_buf. 3. Move "int ret" into the "if (work)" block, this looks more consistent. Signed-off-by:
Tejun Heo <tj@kernel.org> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Mike Galbraith <efault@gmx.de> Cc: Milos Vyletel <milos@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Prarit Bhargava <prarit@redhat.com> Cc: Thomas Gleixner <tglx@linutronix.de> Link: http://lkml.kernel.org/r/20151115193332.GA8281@redhat.comSigned-off-by:
-
Oleg Nesterov authored
Change cpu_stop_queue_work() and cpu_stopper_thread() to check done != NULL before cpu_stop_signal_done(done). This makes the code more clean imo, note that cpu_stopper_thread() has to do this check anyway. Signed-off-by:
-
Oleg Nesterov authored
Now that cpu_stop_done->executed becomes write-only (ignoring WARN_ON() checks) we can remove it. Signed-off-by:
-
Oleg Nesterov authored
Change queue_stop_cpus_work() to return true if it queues at least one work, this means that the caller should wait. __stop_cpus() can check the value returned by queue_stop_cpus_work() and avoid done.executed, just like stop_one_cpu() does. Signed-off-by:
-
Oleg Nesterov authored
Change stop_one_cpu() to return -ENOENT if cpu_stop_queue_work() fails. Otherwise we know that ->executed must be true after wait_for_completion() so we can just return done.ret. Signed-off-by:
-
Oleg Nesterov authored
Change cpu_stop_queue_work() to return true if the work was queued and change stop_one_cpu_nowait() to return the result of cpu_stop_queue_work(). This makes it more useful, for example now you can alloc cpu_stop_work for stop_one_cpu_nowait() and free it in the callback or if stop_one_cpu_nowait() fails, currently this is impossible because you can't know if @fn will be called or not. Also, this allows to kill cpu_stop_done->executed, see the next changes. Signed-off-by:
-
Oleg Nesterov authored
Now that stop_two_cpus() path does not check cpu_active() we can remove preempt_disable(), it was only needed to ensure that stop_machine() can not be called after we observe cpu_active() == T and before we queue the new work. Also, turn the pointless and confusing ->executed check into WARN_ON(). We know that both works must be executed, otherwise we have a bug. And in fact I think that done->executed should die, see the next changes. Signed-off-by:
-
Oleg Nesterov authored
stop_one_cpu_nowait(fn) will crash the kernel if the callback returns nonzero, work->done == NULL in this case. This needs more cleanups, cpu_stop_signal_done() is called right after we check done != NULL and it does the same check. Signed-off-by:
-
Geliang Tang authored
Use list_is_singular() to check if run_list has only one entry. Signed-off-by:
Geliang Tang <geliangtang@163.com> Signed-off-by:
-
Joonwoo Park authored
At present scheduler resets task's wait start timestamp when the task migrates to another rq. This misleads scheduler itself into reporting less wait time than actual by omitting time spent for waiting prior to migration and also more wait count than actual by counting migration as wait end event which can be seen by trace or /proc/<pid>/sched with CONFIG_SCHEDSTATS=y. Carry forward migrating task's wait time prior to migration and don't count migration as a wait end event to fix such statistics error. In order to determine whether task is migrating mark task->on_rq with TASK_ON_RQ_MIGRATING while dequeuing and enqueuing due to migration. Signed-off-by:
Joonwoo Park <joonwoop@codeaurora.org> Signed-off-by:
-
Rik van Riel authored
There is a fundamental mismatch between the runtime based NUMA scanning at the task level, and the wall clock time NUMA scanning at the mm level. On a severely overloaded system, with very large processes, this mismatch can cause the system to spend all of its time in change_prot_numa(). This can happen if the task spends at least two ticks in change_prot_numa(), and only gets two ticks of CPU time in the real time between two scan intervals of the mm. This patch ensures that a task never spends more than 3% of run time scanning PTEs. It does that by ensuring that in-between task_numa_work() runs, the task spends at least 32x as much time on other things than it did on task_numa_work(). This is done stochastically: if a timer tick happens, or the task gets rescheduled during task_numa_work(), we delay a future run of task_numa_work() until the task has spent at least 32x the amount of CPU time doing something else, as it spent inside task_numa_work(). The longer task_numa_work() takes, the more likely it is this happens. If task_numa_work() takes very little time, chances are low that that code will do anything, but we will not care. Reported-and-tested-by:
Jan Stancek <jstancek@redhat.com> Signed-off-by:
Rik van Riel <riel@redhat.com> Signed-off-by:
-
Byungchul Park authored
Usually the tick can be stopped for an idle CPU in NOHZ. However in NOHZ_FULL mode, a non-idle CPU's tick can also be stopped. However, update_cpu_load_nohz() does not consider the case a non-idle CPU's tick has been stopped at all. This patch makes the update_cpu_load_nohz() know if the calling path comes from NOHZ_FULL or idle NOHZ. Signed-off-by:
-
Byungchul Park authored
There are some cases where distance between ticks is more than one tick while the CPU is not idle, e.g. full NOHZ. However __update_cpu_load() assumes it is the idle tickless case if the distance between ticks is more than 1, even though it can be the active tickless case as well. Thus in the active tickless case, updating the CPU load will not be performed correctly. Where the current code assumes the load for each tick is zero, this is (obviously) not true in non-idle tickless case. We can approximately consider the load ~= this_rq->cpu_load[0] during tickless in non-idle tickless case. Signed-off-by:
-
Peter Zijlstra authored
Signed-off-by:
-
Peter Zijlstra authored
Kosuku reports that there were a fair number of buggy waitqueue_active() users and this function deserves a big comment in order to avoid growing more. Reported-by:
Kosuke Tatsukawa <tatsu@ab.jp.nec.com> Signed-off-by:
-
Dietmar Eggemann authored
Commit cd126afe ("sched/fair: Remove rq's runnable avg") got rid of rq->avg and so there is no need to update it any more when entering or exiting idle. Remove the now empty functions idle_{enter|exit}_fair(). Signed-off-by:
Dietmar Eggemann <dietmar.eggemann@arm.com> Signed-off-by:
-
Arnd Bergmann authored
The push_irq_work_func() function is conditionally defined only when both CONFIG_SMP and HAVE_RT_PUSH_IPI are defined, but the forward declaration remains visibile without HAVE_RT_PUSH_IPI, causing a gcc warning in ARM64 allnoconfig: kernel/sched/rt.c:68:13: warning: 'push_irq_work_func' declared 'static' but never defined [-Wunused-function] This changes the code to use the same condition for both the declaration and the function definition, which gets rid of the warning. As Peter Zijlstra, we can possibly get rid of the whole HAVE_RT_PUSH_IPI thing after: 8053871d ("smp: Fix smp_call_function_single_async() locking") Until that is done, this patch can be used to avoid the warning. Signed-off-by:
Arnd Bergmann <arnd@arndb.de> Signed-off-by:
Steven Rostedt <rostedt@goodmis.org> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Fixes: b6366f04 ("sched/rt: Use IPI to trigger RT task push migration instead of pulling") Link: http://lkml.kernel.org/r/3828565.oKfGk7yNIT@wuerfelSigned-off-by:
-
Linus Torvalds authored
-
- 22 Nov, 2015 13 commits
-
-
Linus Torvalds authored
Merge slub bulk allocator updates from Andrew Morton: "This missed the merge window because I was waiting for some repairs to come in. Nothing actually uses the bulk allocator yet and the changes to other code paths are pretty small. And the net guys are waiting for this so they can start merging the client code" More comments from Jesper Dangaard Brouer: "The kmem_cache_alloc_bulk() call, in mm/slub.c, were included in previous kernel. The present version contains a bug. Vladimir Davydov noticed it contained a bug, when kernel is compiled with CONFIG_MEMCG_KMEM (see commit 03ec0ed5: "slub: fix kmem cgroup bug in kmem_cache_alloc_bulk"). Plus the mem cgroup counterpart in kmem_cache_free_bulk() were missing (see commit 03374518 "slub: add missing kmem cgroup support to kmem_cache_free_bulk"). I don't consider the fix stable-material because there are no in-tree users of the API. But with known bugs (for memcg) I cannot start using the API in the net-tree" * emailed patches from Andrew Morton <akpm@linux-foundation.org>: slab/slub: adjust kmem_cache_alloc_bulk API slub: add missing kmem cgroup support to kmem_cache_free_bulk slub: fix kmem cgroup bug in kmem_cache_alloc_bulk slub: optimize bulk slowpath free by detached freelist slub: support for bulk free with SLUB freelists
-
git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/ttyLinus Torvalds authored
Pull tty/serial fixes from Greg KH: "Here are a few small tty/serial driver fixes for 4.4-rc2 that resolve some reported problems. All have been in linux-next, full details are in the shortlog below" * tag 'tty-4.4-rc2' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/tty: serial: export fsl8250_handle_irq serial: 8250_mid: Add missing dependency tty: audit: Fix audit source serial: etraxfs-uart: Fix crash serial: fsl_lpuart: Fix earlycon support bcm63xx_uart: Use the device name when registering an interrupt tty: Fix direct use of tty buffer work tty: Fix tty_send_xchar() lock order inversion
-
git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/stagingLinus Torvalds authored
Pull staging/IIO fixes from Greg KH: "Here are some staging and iio driver fixes for 4.4-rc2. All of these are in response to issues that have been reported and have been in linux-next for a while" * tag 'staging-4.4-rc2' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/staging: Revert "Staging: wilc1000: coreconfigurator: Drop unneeded wrapper functions" iio: adc: xilinx: Fix VREFN scale iio: si7020: Swap data byte order iio: adc: vf610_adc: Fix division by zero error iio:ad7793: Fix ad7785 product ID iio: ad5064: Fix ad5629/ad5669 shift iio:ad5064: Make sure ad5064_i2c_write() returns 0 on success iio: lpc32xx_adc: fix warnings caused by enabling unprepared clock staging: iio: select IRQ_WORK for IIO_DUMMY_EVGEN vf610_adc: Fix internal temperature calculation
-
git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/usbLinus Torvalds authored
Pull USB fixes from Greg KH: "Here are a number of USB fixes and new device ids for 4.4-rc2. All have been in linux-next and the details are in the shortlog" * tag 'usb-4.4-rc2' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/usb: (28 commits) usblp: do not set TASK_INTERRUPTIBLE before lock USB: MAINTAINERS: cxacru usb: kconfig: fix warning of select USB_OTG USB: option: add XS Stick W100-2 from 4G Systems xhci: Fix a race in usb2 LPM resume, blocking U3 for usb2 devices usb: xhci: fix checking ep busy for CFC xhci: Workaround to get Intel xHCI reset working more reliably usb: chipidea: imx: fix a possible NULL dereference usb: chipidea: usbmisc_imx: fix a possible NULL dereference usb: chipidea: otg: gadget module load and unload support usb: chipidea: debug: disable usb irq while role switch ARM: dts: imx27.dtsi: change the clock information for usb usb: chipidea: imx: refine clock operations to adapt for all platforms usb: gadget: atmel_usba_udc: Expose correct device speed usb: musb: enable usb_dma parameter usb: phy: phy-mxs-usb: fix a possible NULL dereference usb: dwc3: gadget: let us set lower max_speed usb: musb: fix tx fifo flush handling usb: gadget: f_loopback: fix the warning during the enumeration usb: dwc2: host: Fix remote wakeup when not in DWC2_L2 ...
-
git://git.linux-mips.org/pub/scm/ralf/upstream-linusLinus Torvalds authored
Pull MIPS fixes from Ralf Baechle: - Fix a flood of annoying build warnings - A number of fixes for Atheros 79xx platforms * 'upstream' of git://git.linux-mips.org/pub/scm/ralf/upstream-linus: MIPS: ath79: Add a machine entry for booting OF machines MIPS: ath79: Fix the size of the MISC INTC registers in ar9132.dtsi MIPS: ath79: Fix the DDR control initialization on ar71xx and ar934x MIPS: Fix flood of warnings about comparsion being always true.
-
git://git.kernel.org/pub/scm/linux/kernel/git/deller/parisc-linuxLinus Torvalds authored
Pull parisc update from Helge Deller: "This patchset adds Huge Page and HUGETLBFS support for parisc" Honestly, the hugepage support should have gone through in the merge window, and is not really an rc-time fix. But it only touches arch/parisc, and I cannot find it in myself to care. If one of the three parisc users notices a breakage, I will point at Helge and make rude farting noises. * 'parisc-4.4-2' of git://git.kernel.org/pub/scm/linux/kernel/git/deller/parisc-linux: parisc: Map kernel text and data on huge pages parisc: Add Huge Page and HUGETLBFS support parisc: Use long branch to do_syscall_trace_exit parisc: Increase initial kernel mapping to 32MB on 64bit kernel parisc: Initialize the fault vector earlier in the boot process. parisc: Add defines for Huge page support parisc: Drop unused MADV_xxxK_PAGES flags from asm/mman.h parisc: Drop definition of start_thread_som for HP-UX SOM binaries parisc: Fix wrong comment regarding first pmd entry flags
-
git://git.kernel.org/pub/scm/linux/kernel/git/tip/tipLinus Torvalds authored
Pull perf tool fixes from Thomas Gleixner: "A couple of fixes for perf tools: - Build system updates - Plug a memory leak in an error path of perf probe - Tear down probes correctly when adding fails - Fixes to the perf symbol handling - Fix ordering of event processing in buildid-list - Fix per DSO filtering in the histogram browser" * 'perf-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip: perf probe: Clear probe_trace_event when add_probe_trace_event() fails perf probe: Fix memory leaking on failure by clearing all probe_trace_events perf inject: Also re-pipe lost_samples event perf buildid-list: Requires ordered events perf symbols: Fix dso lookup by long name and missing buildids perf symbols: Allow forcing reading of non-root owned files by root perf hists browser: The dso can be obtained from popup_action->ms.map->dso perf hists browser: Fix 'd' hotkey action to filter by DSO perf symbols: Rebuild rbtree when adjusting symbols for kcore tools: Add a "make all" rule tools: Actually install tmon in the install rule
-
git://git.kernel.org/pub/scm/linux/kernel/git/tip/tipLinus Torvalds authored
Pull x86 fixes from Thomas Gleixner: "This update contains: - MPX updates for handling 32bit processes - A fix for a long standing bug in 32bit signal frame handling related to FPU/XSAVE state - Handle get_xsave_addr() correctly in KVM - Fix SMAP check under paravirtualization - Add a comment to the static function trace entry to avoid further confusion about the difference to dynamic tracing" * 'x86-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip: x86/cpu: Fix SMAP check in PVOPS environments x86/ftrace: Add comment on static function tracing x86/fpu: Fix get_xsave_addr() behavior under virtualization x86/fpu: Fix 32-bit signal frame handling x86/mpx: Fix 32-bit address space calculation x86/mpx: Do proper get_user() when running 32-bit binaries on 64-bit kernels -
Jesper Dangaard Brouer authored
Adjust kmem_cache_alloc_bulk API before we have any real users. Adjust API to return type 'int' instead of previously type 'bool'. This is done to allow future extension of the bulk alloc API. A future extension could be to allow SLUB to stop at a page boundary, when specified by a flag, and then return the number of objects. The advantage of this approach, would make it easier to make bulk alloc run without local IRQs disabled. With an approach of cmpxchg "stealing" the entire c->freelist or page->freelist. To avoid overshooting we would stop processing at a slab-page boundary. Else we always end up returning some objects at the cost of another cmpxchg. To keep compatible with future users of this API linking against an older kernel when using the new flag, we need to return the number of allocated objects with this API change. Signed-off-by:
Jesper Dangaard Brouer <brouer@redhat.com> Cc: Vladimir Davydov <vdavydov@virtuozzo.com> Acked-by:
Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Signed-off-by:
Andrew Morton <akpm@linux-foundation.org> Signed-off-by:
-
Jesper Dangaard Brouer authored
Initial implementation missed support for kmem cgroup support in kmem_cache_free_bulk() call, add this. If CONFIG_MEMCG_KMEM is not enabled, the compiler should be smart enough to not add any asm code. Incoming bulk free objects can belong to different kmem cgroups, and object free call can happen at a later point outside memcg context. Thus, we need to keep the orig kmem_cache, to correctly verify if a memcg object match against its "root_cache" (s->memcg_params.root_cache). Signed-off-by:
Vladimir Davydov <vdavydov@virtuozzo.com> Cc: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Signed-off-by:
-
Jesper Dangaard Brouer authored
The call slab_pre_alloc_hook() interacts with kmemgc and is not allowed to be called several times inside the bulk alloc for loop, due to the call to memcg_kmem_get_cache(). This would result in hitting the VM_BUG_ON in __memcg_kmem_get_cache. As suggested by Vladimir Davydov, change slab_post_alloc_hook() to be able to handle an array of objects. A subtle detail is, loop iterator "i" in slab_post_alloc_hook() must have same type (size_t) as size argument. This helps the compiler to easier realize that it can remove the loop, when all debug statements inside loop evaluates to nothing. Note, this is only an issue because the kernel is compiled with GCC option: -fno-strict-overflow In slab_alloc_node() the compiler inlines and optimizes the invocation of slab_post_alloc_hook(s, flags, 1, &object) by removing the loop and access object directly. Signed-off-by:
-
Jesper Dangaard Brouer authored
This change focus on improving the speed of object freeing in the "slowpath" of kmem_cache_free_bulk. The calls slab_free (fastpath) and __slab_free (slowpath) have been extended with support for bulk free, which amortize the overhead of the (locked) cmpxchg_double. To use the new bulking feature, we build what I call a detached freelist. The detached freelist takes advantage of three properties: 1) the free function call owns the object that is about to be freed, thus writing into this memory is synchronization-free. 2) many freelist's can co-exist side-by-side in the same slab-page each with a separate head pointer. 3) it is the visibility of the head pointer that needs synchronization. Given these properties, the brilliant part is that the detached freelist can be constructed without any need for synchronization. The freelist is constructed directly in the page objects, without any synchronization needed. The detached freelist is allocated on the stack of the function call kmem_cache_free_bulk. Thus, the freelist head pointer is not visible to other CPUs. All objects in a SLUB freelist must belong to the same slab-page. Thus, constructing the detached freelist is about matching objects that belong to the same slab-page. The bulk free array is scanned is a progressive manor with a limited look-ahead facility. Kmem debug support is handled in call of slab_free(). Notice kmem_cache_free_bulk no longer need to disable IRQs. This only slowed down single free bulk with approx 3 cycles. Performance data: Benchmarked[1] obj size 256 bytes on CPU i7-4790K @ 4.00GHz SLUB fastpath single object quick reuse: 47 cycles(tsc) 11.931 ns To get stable and comparable numbers, the kernel have been booted with "slab_merge" (this also improve performance for larger bulk sizes). Performance data, compared against fallback bulking: bulk - fallback bulk - improvement with this patch 1 - 62 cycles(tsc) 15.662 ns - 49 cycles(tsc) 12.407 ns- improved 21.0% 2 - 55 cycles(tsc) 13.935 ns - 30 cycles(tsc) 7.506 ns - improved 45.5% 3 - 53 cycles(tsc) 13.341 ns - 23 cycles(tsc) 5.865 ns - improved 56.6% 4 - 52 cycles(tsc) 13.081 ns - 20 cycles(tsc) 5.048 ns - improved 61.5% 8 - 50 cycles(tsc) 12.627 ns - 18 cycles(tsc) 4.659 ns - improved 64.0% 16 - 49 cycles(tsc) 12.412 ns - 17 cycles(tsc) 4.495 ns - improved 65.3% 30 - 49 cycles(tsc) 12.484 ns - 18 cycles(tsc) 4.533 ns - improved 63.3% 32 - 50 cycles(tsc) 12.627 ns - 18 cycles(tsc) 4.707 ns - improved 64.0% 34 - 96 cycles(tsc) 24.243 ns - 23 cycles(tsc) 5.976 ns - improved 76.0% 48 - 83 cycles(tsc) 20.818 ns - 21 cycles(tsc) 5.329 ns - improved 74.7% 64 - 74 cycles(tsc) 18.700 ns - 20 cycles(tsc) 5.127 ns - improved 73.0% 128 - 90 cycles(tsc) 22.734 ns - 27 cycles(tsc) 6.833 ns - improved 70.0% 158 - 99 cycles(tsc) 24.776 ns - 30 cycles(tsc) 7.583 ns - improved 69.7% 250 - 104 cycles(tsc) 26.089 ns - 37 cycles(tsc) 9.280 ns - improved 64.4% Performance data, compared current in-kernel bulking: bulk - curr in-kernel - improvement with this patch 1 - 46 cycles(tsc) - 49 cycles(tsc) - improved (cycles:-3) -6.5% 2 - 27 cycles(tsc) - 30 cycles(tsc) - improved (cycles:-3) -11.1% 3 - 21 cycles(tsc) - 23 cycles(tsc) - improved (cycles:-2) -9.5% 4 - 18 cycles(tsc) - 20 cycles(tsc) - improved (cycles:-2) -11.1% 8 - 17 cycles(tsc) - 18 cycles(tsc) - improved (cycles:-1) -5.9% 16 - 18 cycles(tsc) - 17 cycles(tsc) - improved (cycles: 1) 5.6% 30 - 18 cycles(tsc) - 18 cycles(tsc) - improved (cycles: 0) 0.0% 32 - 18 cycles(tsc) - 18 cycles(tsc) - improved (cycles: 0) 0.0% 34 - 78 cycles(tsc) - 23 cycles(tsc) - improved (cycles:55) 70.5% 48 - 60 cycles(tsc) - 21 cycles(tsc) - improved (cycles:39) 65.0% 64 - 49 cycles(tsc) - 20 cycles(tsc) - improved (cycles:29) 59.2% 128 - 69 cycles(tsc) - 27 cycles(tsc) - improved (cycles:42) 60.9% 158 - 79 cycles(tsc) - 30 cycles(tsc) - improved (cycles:49) 62.0% 250 - 86 cycles(tsc) - 37 cycles(tsc) - improved (cycles:49) 57.0% Performance with normal SLUB merging is significantly slower for larger bulking. This is believed to (primarily) be an effect of not having to share the per-CPU data-structures, as tuning per-CPU size can achieve similar performance. bulk - slab_nomerge - normal SLUB merge 1 - 49 cycles(tsc) - 49 cycles(tsc) - merge slower with cycles:0 2 - 30 cycles(tsc) - 30 cycles(tsc) - merge slower with cycles:0 3 - 23 cycles(tsc) - 23 cycles(tsc) - merge slower with cycles:0 4 - 20 cycles(tsc) - 20 cycles(tsc) - merge slower with cycles:0 8 - 18 cycles(tsc) - 18 cycles(tsc) - merge slower with cycles:0 16 - 17 cycles(tsc) - 17 cycles(tsc) - merge slower with cycles:0 30 - 18 cycles(tsc) - 23 cycles(tsc) - merge slower with cycles:5 32 - 18 cycles(tsc) - 22 cycles(tsc) - merge slower with cycles:4 34 - 23 cycles(tsc) - 22 cycles(tsc) - merge slower with cycles:-1 48 - 21 cycles(tsc) - 22 cycles(tsc) - merge slower with cycles:1 64 - 20 cycles(tsc) - 48 cycles(tsc) - merge slower with cycles:28 128 - 27 cycles(tsc) - 57 cycles(tsc) - merge slower with cycles:30 158 - 30 cycles(tsc) - 59 cycles(tsc) - merge slower with cycles:29 250 - 37 cycles(tsc) - 56 cycles(tsc) - merge slower with cycles:19 Joint work with Alexander Duyck. [1] https://github.com/netoptimizer/prototype-kernel/blob/master/kernel/mm/slab_bulk_test01.c [akpm@linux-foundation.org: BUG_ON -> WARN_ON;return] Signed-off-by:Alexander Duyck <alexander.h.duyck@redhat.com> Acked-by:
-
Jesper Dangaard Brouer authored
Make it possible to free a freelist with several objects by adjusting API of slab_free() and __slab_free() to have head, tail and an objects counter (cnt). Tail being NULL indicate single object free of head object. This allow compiler inline constant propagation in slab_free() and slab_free_freelist_hook() to avoid adding any overhead in case of single object free. This allows a freelist with several objects (all within the same slab-page) to be free'ed using a single locked cmpxchg_double in __slab_free() and with an unlocked cmpxchg_double in slab_free(). Object debugging on the free path is also extended to handle these freelists. When CONFIG_SLUB_DEBUG is enabled it will also detect if objects don't belong to the same slab-page. These changes are needed for the next patch to bulk free the detached freelists it introduces and constructs. Micro benchmarking showed no performance reduction due to this change, when debugging is turned off (compiled with CONFIG_SLUB_DEBUG). Signed-off-by:
-
