From: Greg KH <gregkh@suse.de>
To: linux-kernel@vger.kernel.org, stable@kernel.org
Cc: stable-review@kernel.org, torvalds@linux-foundation.org,
akpm@linux-foundation.org, alan@lxorguk.ukuu.org.uk,
Ingo Molnar <mingo@elte.hu>,
Peter Zijlstra <a.p.zijlstra@chello.nl>, Greg KH <greg@kroah.com>,
Suresh Siddha <suresh.b.siddha@intel.com>,
Mike Galbraith <efault@gmx.de>
Subject: [110/123] sched: Fix select_idle_sibling() logic in select_task_rq_fair()
Date: Sat, 18 Sep 2010 11:59:14 -0700 [thread overview]
Message-ID: <20100918190002.247368124@clark.site> (raw)
In-Reply-To: <20100918190024.GA14388@kroah.com>
[-- Attachment #1: sched-fix-select_idle_sibling-logic-in-select_task_rq_fair.patch --]
[-- Type: text/plain, Size: 5685 bytes --]
From: Suresh Siddha <suresh.b.siddha@intel.com>
commit 99bd5e2f245d8cd17d040c82d40becdb3efd9b69 upstream
Issues in the current select_idle_sibling() logic in select_task_rq_fair()
in the context of a task wake-up:
a) Once we select the idle sibling, we use that domain (spanning the cpu that
the task is currently woken-up and the idle sibling that we found) in our
wake_affine() decisions. This domain is completely different from the
domain(we are supposed to use) that spans the cpu that the task currently
woken-up and the cpu where the task previously ran.
b) We do select_idle_sibling() check only for the cpu that the task is
currently woken-up on. If select_task_rq_fair() selects the previously run
cpu for waking the task, doing a select_idle_sibling() check
for that cpu also helps and we don't do this currently.
c) In the scenarios where the cpu that the task is woken-up is busy but
with its HT siblings are idle, we are selecting the task be woken-up
on the idle HT sibling instead of a core that it previously ran
and currently completely idle. i.e., we are not taking decisions based on
wake_affine() but directly selecting an idle sibling that can cause
an imbalance at the SMT/MC level which will be later corrected by the
periodic load balancer.
Fix this by first going through the load imbalance calculations using
wake_affine() and once we make a decision of woken-up cpu vs previously-ran cpu,
then choose a possible idle sibling for waking up the task on.
Signed-off-by: Suresh Siddha <suresh.b.siddha@intel.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <1270079265.7835.8.camel@sbs-t61.sc.intel.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Mike Galbraith <efault@gmx.de>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
---
kernel/sched_fair.c | 82 +++++++++++++++++++++++++---------------------------
1 file changed, 40 insertions(+), 42 deletions(-)
--- a/kernel/sched_fair.c
+++ b/kernel/sched_fair.c
@@ -1386,29 +1386,48 @@ find_idlest_cpu(struct sched_group *grou
/*
* Try and locate an idle CPU in the sched_domain.
*/
-static int
-select_idle_sibling(struct task_struct *p, struct sched_domain *sd, int target)
+static int select_idle_sibling(struct task_struct *p, int target)
{
int cpu = smp_processor_id();
int prev_cpu = task_cpu(p);
+ struct sched_domain *sd;
int i;
/*
- * If this domain spans both cpu and prev_cpu (see the SD_WAKE_AFFINE
- * test in select_task_rq_fair) and the prev_cpu is idle then that's
- * always a better target than the current cpu.
+ * If the task is going to be woken-up on this cpu and if it is
+ * already idle, then it is the right target.
+ */
+ if (target == cpu && idle_cpu(cpu))
+ return cpu;
+
+ /*
+ * If the task is going to be woken-up on the cpu where it previously
+ * ran and if it is currently idle, then it the right target.
*/
- if (target == cpu && !cpu_rq(prev_cpu)->cfs.nr_running)
+ if (target == prev_cpu && idle_cpu(prev_cpu))
return prev_cpu;
/*
- * Otherwise, iterate the domain and find an elegible idle cpu.
+ * Otherwise, iterate the domains and find an elegible idle cpu.
*/
- for_each_cpu_and(i, sched_domain_span(sd), &p->cpus_allowed) {
- if (!cpu_rq(i)->cfs.nr_running) {
- target = i;
+ for_each_domain(target, sd) {
+ if (!(sd->flags & SD_SHARE_PKG_RESOURCES))
break;
+
+ for_each_cpu_and(i, sched_domain_span(sd), &p->cpus_allowed) {
+ if (idle_cpu(i)) {
+ target = i;
+ break;
+ }
}
+
+ /*
+ * Lets stop looking for an idle sibling when we reached
+ * the domain that spans the current cpu and prev_cpu.
+ */
+ if (cpumask_test_cpu(cpu, sched_domain_span(sd)) &&
+ cpumask_test_cpu(prev_cpu, sched_domain_span(sd)))
+ break;
}
return target;
@@ -1432,7 +1451,7 @@ select_task_rq_fair(struct rq *rq, struc
int cpu = smp_processor_id();
int prev_cpu = task_cpu(p);
int new_cpu = cpu;
- int want_affine = 0, cpu_idle = !current->pid;
+ int want_affine = 0;
int want_sd = 1;
int sync = wake_flags & WF_SYNC;
@@ -1472,36 +1491,13 @@ select_task_rq_fair(struct rq *rq, struc
}
/*
- * While iterating the domains looking for a spanning
- * WAKE_AFFINE domain, adjust the affine target to any idle cpu
- * in cache sharing domains along the way.
+ * If both cpu and prev_cpu are part of this domain,
+ * cpu is a valid SD_WAKE_AFFINE target.
*/
- if (want_affine) {
- int target = -1;
-
- /*
- * If both cpu and prev_cpu are part of this domain,
- * cpu is a valid SD_WAKE_AFFINE target.
- */
- if (cpumask_test_cpu(prev_cpu, sched_domain_span(tmp)))
- target = cpu;
-
- /*
- * If there's an idle sibling in this domain, make that
- * the wake_affine target instead of the current cpu.
- */
- if (!cpu_idle && tmp->flags & SD_SHARE_PKG_RESOURCES)
- target = select_idle_sibling(p, tmp, target);
-
- if (target >= 0) {
- if (tmp->flags & SD_WAKE_AFFINE) {
- affine_sd = tmp;
- want_affine = 0;
- if (target != cpu)
- cpu_idle = 1;
- }
- cpu = target;
- }
+ if (want_affine && (tmp->flags & SD_WAKE_AFFINE) &&
+ cpumask_test_cpu(prev_cpu, sched_domain_span(tmp))) {
+ affine_sd = tmp;
+ want_affine = 0;
}
if (!want_sd && !want_affine)
@@ -1532,8 +1528,10 @@ select_task_rq_fair(struct rq *rq, struc
#endif
if (affine_sd) {
- if (cpu_idle || cpu == prev_cpu || wake_affine(affine_sd, p, sync))
- return cpu;
+ if (cpu == prev_cpu || wake_affine(affine_sd, p, sync))
+ return select_idle_sibling(p, cpu);
+ else
+ return select_idle_sibling(p, prev_cpu);
}
while (sd) {
next prev parent reply other threads:[~2010-09-18 19:08 UTC|newest]
Thread overview: 132+ messages / expand[flat|nested] mbox.gz Atom feed top
[not found] <20100918185724.290702750@clark.site>
2010-09-18 19:00 ` [000/123] 2.6.32.22-stable review Greg KH
2010-09-18 18:57 ` [001/123] hwmon: (k8temp) Differentiate between AM2 and ASB1 Greg KH
2010-09-18 18:57 ` [002/123] xen: handle events as edge-triggered Greg KH
2010-09-18 18:57 ` [003/123] xen: use percpu interrupts for IPIs and VIRQs Greg KH
2010-09-18 18:57 ` [004/123] ALSA: hda - Rename iMic to Int Mic on Lenovo NB0763 Greg KH
2010-09-18 18:57 ` [005/123] sata_mv: fix broken DSM/TRIM support (v2) Greg KH
2010-09-18 18:57 ` [006/123] x86, tsc, sched: Recompute cyc2ns_offsets during resume from sleep states Greg KH
2010-09-18 18:57 ` [007/123] PCI: MSI: Remove unsafe and unnecessary hardware access Greg KH
2010-09-18 18:57 ` [008/123] PCI: MSI: Restore read_msi_msg_desc(); add get_cached_msi_msg_desc() Greg KH
2010-09-18 18:57 ` [009/123] sched: kill migration thread in CPU_POST_DEAD instead of CPU_DEAD Greg KH
2010-09-18 18:57 ` [010/123] sched: revert stable c6fc81a sched: Fix a race between ttwu() and migrate_task() Greg KH
2010-09-18 18:57 ` [011/123] staging: hv: Fix missing functions for net_device_ops Greg KH
2010-09-18 18:57 ` [012/123] staging: hv: Fixed bounce kmap problem by using correct index Greg KH
2010-09-18 18:57 ` [013/123] staging: hv: Fixed the value of the 64bit-hole inside ring buffer Greg KH
2010-09-18 18:57 ` [014/123] staging: hv: Increased storvsc ringbuffer and max_io_requests Greg KH
2010-09-18 18:57 ` [015/123] staging: hv: Fixed lockup problem with bounce_buffer scatter list Greg KH
2010-09-18 18:57 ` [016/123] fuse: flush background queue on connection close Greg KH
2010-09-18 18:57 ` [017/123] ath9k_hw: fix parsing of HT40 5 GHz CTLs Greg KH
2010-09-18 18:57 ` [018/123] ocfs2: Fix incorrect checksum validation error Greg KH
2010-09-18 18:57 ` [019/123] USB: ehci-ppc-of: problems in unwind Greg KH
2010-09-18 18:57 ` [020/123] USB: Fix kernel oops with g_ether and Windows Greg KH
2010-09-18 18:57 ` [021/123] USB: CP210x Add new device ID Greg KH
2010-09-18 18:57 ` [022/123] USB: cp210x: Add B&G H3000 link cable ID Greg KH
2010-09-18 18:57 ` [023/123] USB: ftdi_sio: Added custom PIDs for ChamSys products Greg KH
2010-09-18 18:57 ` [024/123] USB: serial: Extra device/vendor ID for mos7840 driver Greg KH
2010-09-18 18:57 ` [025/123] usb: serial: mos7840: Add USB ID to support the B&B Electronics USOPTL4-2P Greg KH
2010-09-18 18:57 ` [026/123] USB: mos7840: fix DMA buffers on stack and endianess bugs Greg KH
2010-09-18 18:57 ` [027/123] usb: serial: mos7840: Add USB IDs to support more B&B USB/RS485 converters Greg KH
2010-09-18 18:57 ` [028/123] USB: Exposing second ACM channel as tty for Nokia S60 phones Greg KH
2010-09-18 18:57 ` [029/123] USB: cdc-acm: add another device quirk Greg KH
2010-09-18 18:57 ` [030/123] USB: Expose vendor-specific ACM channel on Nokia 5230 Greg KH
2010-09-18 18:57 ` [031/123] USB: cdc-acm: Adding second ACM channel support for various Nokia and one Samsung phones Greg KH
2010-09-18 18:57 ` [032/123] USB: cdc-acm: Add pseudo modem without AT command capabilities Greg KH
2010-09-18 18:57 ` [033/123] USB: cdc-acm: Fixing crash when ACM probing interfaces with no endpoint descriptors Greg KH
2010-09-18 18:57 ` [034/123] ALSA: hda - Fix auto-parser of ALC269vb for HP pin NID 0x21 Greg KH
2010-09-18 18:57 ` [035/123] ALSA: seq/oss - Fix double-free at error path of snd_seq_oss_open() Greg KH
2010-09-18 18:58 ` [036/123] sysfs: checking for NULL instead of ERR_PTR Greg KH
2010-09-18 18:58 ` [037/123] tun: Dont add sysfs attributes to devices without sysfs directories Greg KH
2010-09-18 18:58 ` [038/123] oprofile: fix crash when accessing freed task structs Greg KH
2010-09-18 18:58 ` [039/123] oprofile, x86: fix init_sysfs error handling Greg KH
2010-09-18 18:58 ` [040/123] oprofile, x86: fix init_sysfs() function stub Greg KH
2010-09-18 18:58 ` [041/123] HID: usbhid: initialize interface pointers early enough Greg KH
2010-09-18 18:58 ` [042/123] HID: fix suspend crash by moving initializations earlier Greg KH
2010-09-18 18:58 ` [043/123] libata: skip EH autopsy and recovery during suspend Greg KH
2010-09-18 18:58 ` [044/123] tracing: Fix a race in function profile Greg KH
2010-09-18 18:58 ` [045/123] tracing: Do not allow llseek to set_ftrace_filter Greg KH
2010-09-18 18:58 ` [046/123] tracing: t_start: reset FTRACE_ITER_HASH in case of seek/pread Greg KH
2010-09-18 18:58 ` [047/123] irda: off by one Greg KH
2010-09-18 18:58 ` [048/123] gcov: fix null-pointer dereference for certain module types Greg KH
2010-09-18 18:58 ` [049/123] tmio_mmc: dont clear unhandled pending interrupts Greg KH
2010-09-18 18:58 ` [050/123] mmc: fix the use of kunmap_atomic() in tmio_mmc.h Greg KH
2010-09-18 18:58 ` [051/123] bounce: call flush_dcache_page() after bounce_copy_vec() Greg KH
2010-09-18 18:58 ` [052/123] kernel/groups.c: fix integer overflow in groups_search Greg KH
2010-09-18 18:58 ` [053/123] binfmt_misc: fix binfmt_misc priority Greg KH
2010-09-18 18:58 ` [054/123] Input: i8042 - fix device removal on unload Greg KH
2010-09-18 18:58 ` [055/123] memory hotplug: fix next block calculation in is_removable Greg KH
2010-09-18 18:58 ` [056/123] perf: Initialize callchains rootss childen hits Greg KH
2010-09-18 18:58 ` [057/123] p54: fix tx feedback status flag check Greg KH
2010-09-18 18:58 ` [058/123] ath5k: check return value of ieee80211_get_tx_rate Greg KH
2010-09-18 18:58 ` [059/123] wireless extensions: fix kernel heap content leak Greg KH
2010-09-18 18:58 ` [060/123] x86, tsc: Fix a preemption leak in restore_sched_clock_state() Greg KH
2010-09-18 18:58 ` [061/123] x86-64, compat: Test %rax for the syscall number, not %eax Greg KH
2010-09-18 18:58 ` [062/123] compat: Make compat_alloc_user_space() incorporate the access_ok() Greg KH
2010-09-18 18:58 ` [063/123] x86-64, compat: Retruncate rax after ia32 syscall entry tracing Greg KH
2010-09-18 18:58 ` [064/123] sched: Protect task->cpus_allowed access in sched_getaffinity() Greg KH
2010-09-18 20:32 ` [Stable-review] " Ben Hutchings
2010-09-18 22:19 ` Greg KH
2010-09-19 5:10 ` Mike Galbraith
2010-09-19 9:40 ` Mike Galbraith
2010-09-18 18:58 ` [065/123] sched: Protect sched_rr_get_param() access to task->sched_class Greg KH
2010-09-18 18:58 ` [066/123] sched: Consolidate select_task_rq() callers Greg KH
2010-09-18 18:58 ` [067/123] sched: Remove unused cpu_nr_migrations() Greg KH
2010-09-18 18:58 ` [068/123] sched: Remove rq->clock coupling from set_task_cpu() Greg KH
2010-09-18 18:58 ` [069/123] sched: Clean up ttwu() rq locking Greg KH
2010-09-18 18:58 ` [070/123] sched: Sanitize fork() handling Greg KH
2010-09-18 18:58 ` [071/123] sched: Remove forced2_migrations stats Greg KH
2010-09-18 18:58 ` [072/123] sched: Make wakeup side and atomic variants of completion API irq safe Greg KH
2010-09-18 18:58 ` [073/123] sched: Use rcu in sys_sched_getscheduler/sys_sched_getparam() Greg KH
2010-09-18 18:58 ` [074/123] sched: Use rcu in sched_get/set_affinity() Greg KH
2010-09-18 18:58 ` [075/123] sched: Use rcu in sched_get_rr_param() Greg KH
2010-09-18 18:58 ` [076/123] sched: Fix set_cpu_active() in cpu_down() Greg KH
2010-09-18 18:58 ` [077/123] sched: Use TASK_WAKING for fork wakups Greg KH
2010-09-18 18:58 ` [078/123] sched: Ensure set_task_cpu() is never called on blocked tasks Greg KH
2010-09-18 18:58 ` [079/123] sched: Make warning less noisy Greg KH
2010-09-18 18:58 ` [080/123] sched: Fix broken assertion Greg KH
2010-09-18 18:58 ` [081/123] sched: Fix sched_exec() balancing Greg KH
2010-09-18 18:58 ` [082/123] sched: Fix select_task_rq() vs hotplug issues Greg KH
2010-09-18 18:58 ` [083/123] sched: Add pre and post wakeup hooks Greg KH
2010-09-18 18:58 ` [084/123] sched: Remove the cfs_rq dependency from set_task_cpu() Greg KH
2010-09-18 18:58 ` [085/123] sched: Fix hotplug hang Greg KH
2010-09-18 18:58 ` [086/123] sched: Fix fork vs hotplug vs cpuset namespaces Greg KH
2010-09-18 18:58 ` [087/123] sched: Fix incorrect sanity check Greg KH
2010-09-18 18:58 ` [088/123] sched: Fix race between ttwu() and task_rq_lock() Greg KH
2010-09-18 18:58 ` [089/123] sched: Extend enqueue_task to allow head queueing Greg KH
2010-09-18 18:58 ` [090/123] sched: Implement head queueing for sched_rt Greg KH
2010-09-18 18:58 ` [091/123] sched: Queue a deboosted task to the head of the RT prio queue Greg KH
2010-09-18 18:58 ` [092/123] sched: set_cpus_allowed_ptr(): Dont use rq->migration_thread after unlock Greg KH
2010-09-18 18:58 ` [093/123] sched: Kill the broken and deadlockable cpuset_lock/cpuset_cpus_allowed_locked code Greg KH
2010-09-18 18:58 ` [094/123] sched: move_task_off_dead_cpu(): Take rq->lock around select_fallback_rq() Greg KH
2010-09-18 18:58 ` [095/123] sched: move_task_off_dead_cpu(): Remove retry logic Greg KH
2010-09-18 18:59 ` [096/123] sched: sched_exec(): Remove the select_fallback_rq() logic Greg KH
2010-09-18 18:59 ` [097/123] sched: _cpu_down(): Dont play with current->cpus_allowed Greg KH
2010-09-18 18:59 ` [098/123] sched: Make select_fallback_rq() cpuset friendly Greg KH
2010-09-18 18:59 ` [099/123] sched: Fix TASK_WAKING vs fork deadlock Greg KH
2010-09-18 18:59 ` [100/123] sched: Optimize task_rq_lock() Greg KH
2010-09-18 18:59 ` [101/123] sched: Fix nr_uninterruptible count Greg KH
2010-09-18 18:59 ` [102/123] sched: Fix rq->clock synchronization when migrating tasks Greg KH
2010-09-18 18:59 ` [103/123] sched: Remove unnecessary RCU exclusion Greg KH
2010-09-18 18:59 ` [104/123] sched: apply RCU protection to wake_affine() Greg KH
2010-09-18 18:59 ` [105/123] sched: Cleanup select_task_rq_fair() Greg KH
2010-09-18 18:59 ` [106/123] sched: More generic WAKE_AFFINE vs select_idle_sibling() Greg KH
2010-09-18 18:59 ` [107/123] sched: Fix vmark regression on big machines Greg KH
2010-09-18 18:59 ` [108/123] sched: Fix select_idle_sibling() Greg KH
2010-09-18 18:59 ` [109/123] sched: Pre-compute cpumask_weight(sched_domain_span(sd)) Greg KH
2010-09-18 18:59 ` Greg KH [this message]
2010-09-18 18:59 ` [111/123] sched: cpuacct: Use bigger percpu counter batch values for stats counters Greg KH
2010-09-18 18:59 ` [112/123] ALSA: hda - Handle missing NID 0x1b on ALC259 codec Greg KH
2010-09-18 18:59 ` [113/123] ALSA: hda - Handle pin NID 0x1a on ALC259/269 Greg KH
2010-09-18 18:59 ` [114/123] arm: fix really nasty sigreturn bug Greg KH
2010-09-18 18:59 ` [115/123] hwmon: (f75375s) Shift control mode to the correct bit position Greg KH
2010-09-18 18:59 ` [116/123] hwmon: (f75375s) Do not overwrite values read from registers Greg KH
2010-09-18 18:59 ` [117/123] apm_power: Add missing break statement Greg KH
2010-09-18 18:59 ` [118/123] NFS: Fix a typo in nfs_sockaddr_match_ipaddr6 Greg KH
2010-09-18 18:59 ` [119/123] SUNRPC: Fix race corrupting rpc upcall Greg KH
2010-09-18 18:59 ` [120/123] i915: return -EFAULT if copy_to_user fails Greg KH
2010-09-18 18:59 ` [121/123] i915_gem: " Greg KH
2010-09-18 18:59 ` [122/123] drm/i915: Prevent double dpms on Greg KH
2010-09-18 18:59 ` [123/123] drm: Only decouple the old_fb from the crtc is we call mode_set* Greg KH
2010-09-18 21:39 ` [Stable-review] [000/123] 2.6.32.22-stable review Willy Tarreau
2010-09-19 4:00 ` Greg KH
2010-09-19 5:45 ` Willy Tarreau
2010-09-19 4:34 ` David Miller
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20100918190002.247368124@clark.site \
--to=gregkh@suse.de \
--cc=a.p.zijlstra@chello.nl \
--cc=akpm@linux-foundation.org \
--cc=alan@lxorguk.ukuu.org.uk \
--cc=efault@gmx.de \
--cc=greg@kroah.com \
--cc=linux-kernel@vger.kernel.org \
--cc=mingo@elte.hu \
--cc=stable-review@kernel.org \
--cc=stable@kernel.org \
--cc=suresh.b.siddha@intel.com \
--cc=torvalds@linux-foundation.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox