This patchset introduces eventfd-based API for notifications in cgroups
and implements memory notifications on top of it.
It uses statistics in memory controler to track memory usage.
Output of time(1) on building kernel on tmpfs:
Root cgroup before changes:
make -j2 506.37 user 60.93s system 193% cpu 4:52.77 total
Non-root cgroup before changes:
make -j2 507.14 user 62.66s system 193% cpu 4:54.74 total
Root cgroup after changes (0 thresholds):
make -j2 507.13 user 62.20s system 193% cpu 4:53.55 total
Non-root cgroup after changes (0 thresholds):
make -j2 507.70 user 64.20s system 193% cpu 4:55.70 total
Root cgroup after changes (1 thresholds, never crossed):
make -j2 506.97 user 62.20s system 193% cpu 4:53.90 total
Non-root cgroup after changes (1 thresholds, never crossed):
make -j2 507.55 user 64.08s system 193% cpu 4:55.63 total
This patch:
Introduce the write-only file "cgroup.event_control" in every cgroup.
To register new notification handler you need:
- create an eventfd;
- open a control file to be monitored. Callbacks register_event() and
unregister_event() must be defined for the control file;
- write "<event_fd> <control_fd> <args>" to cgroup.event_control.
Interpretation of args is defined by control file implementation;
eventfd will be woken up by control file implementation or when the
cgroup is removed.
To unregister notification handler just close eventfd.
If you need notification functionality for a control file you have to
implement callbacks register_event() and unregister_event() in the
struct cftype.
[kamezawa.hiroyu@jp.fujitsu.com: Kconfig fix]
Signed-off-by: Kirill A. Shutemov <kirill@shutemov.name>
Reviewed-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Paul Menage <menage@google.com>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: Balbir Singh <balbir@linux.vnet.ibm.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Cc: Dan Malek <dan@embeddedalley.com>
Cc: Vladislav Buzov <vbuzov@embeddedalley.com>
Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Cc: Alexander Shishkin <virtuoso@slind.org>
Cc: Davide Libenzi <davidel@xmailserver.org>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Provides support for unloading modular subsystems.
This patch adds a new function cgroup_unload_subsys which is to be used
for removing a loaded subsystem during module deletion. Reference
counting of the subsystems' modules is moved from once (at load time) to
once per attached hierarchy (in parse_cgroupfs_options and
rebind_subsystems) (i.e., 0 or 1).
Signed-off-by: Ben Blum <bblum@andrew.cmu.edu>
Acked-by: Li Zefan <lizf@cn.fujitsu.com>
Cc: Paul Menage <menage@google.com>
Cc: "David S. Miller" <davem@davemloft.net>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Lai Jiangshan <laijs@cn.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Add interface between cgroups subsystem management and module loading
This patch implements rudimentary module-loading support for cgroups -
namely, a cgroup_load_subsys (similar to cgroup_init_subsys) for use as a
module initcall, and a struct module pointer in struct cgroup_subsys.
Several functions that might be wanted by modules have had EXPORT_SYMBOL
added to them, but it's unclear exactly which functions want it and which
won't.
Signed-off-by: Ben Blum <bblum@andrew.cmu.edu>
Acked-by: Li Zefan <lizf@cn.fujitsu.com>
Cc: Paul Menage <menage@google.com>
Cc: "David S. Miller" <davem@davemloft.net>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Lai Jiangshan <laijs@cn.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
This patch series provides the ability for cgroup subsystems to be
compiled as modules both within and outside the kernel tree. This is
mainly useful for classifiers and subsystems that hook into components
that are already modules. cls_cgroup and blkio-cgroup serve as the
example use cases for this feature.
It provides an interface cgroup_load_subsys() and cgroup_unload_subsys()
which modular subsystems can use to register and depart during runtime.
The net_cls classifier subsystem serves as the example for a subsystem
which can be converted into a module using these changes.
Patch #1 sets up the subsys[] array so its contents can be dynamic as
modules appear and (eventually) disappear. Iterations over the array are
modified to handle when subsystems are absent, and the dynamic section of
the array is protected by cgroup_mutex.
Patch #2 implements an interface for modules to load subsystems, called
cgroup_load_subsys, similar to cgroup_init_subsys, and adds a module
pointer in struct cgroup_subsys.
Patch #3 adds a mechanism for unloading modular subsystems, which includes
a more advanced rework of the rudimentary reference counting introduced in
patch 2.
Patch #4 modifies the net_cls subsystem, which already had some module
declarations, to be configurable as a module, which also serves as a
simple proof-of-concept.
Part of implementing patches 2 and 4 involved updating css pointers in
each css_set when the module appears or leaves. In doing this, it was
discovered that css_sets always remain linked to the dummy cgroup,
regardless of whether or not any subsystems are actually bound to it
(i.e., not mounted on an actual hierarchy). The subsystem loading and
unloading code therefore should keep in mind the special cases where the
added subsystem is the only one in the dummy cgroup (and therefore all
css_sets need to be linked back into it) and where the removed subsys was
the only one in the dummy cgroup (and therefore all css_sets should be
unlinked from it) - however, as all css_sets always stay attached to the
dummy cgroup anyway, these cases are ignored. Any fix that addresses this
issue should also make sure these cases are addressed in the subsystem
loading and unloading code.
This patch:
Make subsys[] able to be dynamically populated to support modular
subsystems
This patch reworks the way the subsys[] array is used so that subsystems
can register themselves after boot time, and enables the internals of
cgroups to be able to handle when subsystems are not present or may
appear/disappear.
Signed-off-by: Ben Blum <bblum@andrew.cmu.edu>
Acked-by: Li Zefan <lizf@cn.fujitsu.com>
Cc: Paul Menage <menage@google.com>
Cc: "David S. Miller" <davem@davemloft.net>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Lai Jiangshan <laijs@cn.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Current css_get() and css_put() increment/decrement css->refcnt one by
one.
This patch add a new function __css_get(), which takes "count" as a arg
and increment the css->refcnt by "count". And this patch also add a new
arg("count") to __css_put() and change the function to decrement the
css->refcnt by "count".
These coalesce version of __css_get()/__css_put() will be used to improve
performance of memcg's moving charge feature later, where instead of
calling css_get()/css_put() repeatedly, these new functions will be used.
No change is needed for current users of css_get()/css_put().
Signed-off-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Acked-by: Paul Menage <menage@google.com>
Cc: Balbir Singh <balbir@linux.vnet.ibm.com>
Acked-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Add a generic implementation of the ipc demultiplexer syscall. Except for
s390 and sparc64 all implementations of the sys_ipc are nearly identical.
There are slight differences in the types of the parameters, where mips
and powerpc as the only 64-bit architectures with sys_ipc use unsigned
long for the "third" argument as it gets casted to a pointer later, while
it traditionally is an "int" like most other paramters. frv goes even
further and uses unsigned long for all parameters execept for "ptr" which
is a pointer type everywhere. The change from int to unsigned long for
"third" and back to "int" for the others on frv should be fine due to the
in-register calling conventions for syscalls (we already had a similar
issue with the generic sys_ptrace), but I'd prefer to have the arch
maintainers looks over this in details.
Except for that h8300, m68k and m68knommu lack an impplementation of the
semtimedop sub call which this patch adds, and various architectures have
gets used - at least on i386 it seems superflous as the compat code on
x86-64 and ia64 doesn't even bother to implement it.
[akpm@linux-foundation.org: add sys_ipc to sys_ni.c]
Signed-off-by: Christoph Hellwig <hch@lst.de>
Cc: Ralf Baechle <ralf@linux-mips.org>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Paul Mundt <lethal@linux-sh.org>
Cc: Jeff Dike <jdike@addtoit.com>
Cc: Hirokazu Takata <takata@linux-m32r.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Ingo Molnar <mingo@elte.hu>
Reviewed-by: H. Peter Anvin <hpa@zytor.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Arnd Bergmann <arnd@arndb.de>
Cc: Heiko Carstens <heiko.carstens@de.ibm.com>
Cc: Martin Schwidefsky <schwidefsky@de.ibm.com>
Cc: "Luck, Tony" <tony.luck@intel.com>
Cc: James Morris <jmorris@namei.org>
Cc: Andreas Schwab <schwab@linux-m68k.org>
Acked-by: Jesper Nilsson <jesper.nilsson@axis.com>
Acked-by: Russell King <rmk+kernel@arm.linux.org.uk>
Acked-by: David Howells <dhowells@redhat.com>
Acked-by: Kyle McMartin <kyle@mcmartin.ca>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
If a timer callback leaks preempt_count we currently assert a
BUG(). That makes it unnecessarily hard to retrieve information about
the problem especially on laptops and headless stations.
There is a decent chance to survive the preempt_count leak by
restoring the preempt_count to the value before the callback. That
allows in many cases to get valuable information about the root cause
of the problem.
We carried that fixup in preempt-rt for years and were able to decode
such wreckage quite a few times.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Cc: Linux Torvalds <torvalds@linux-foundation.org>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Arjan van de Veen <arjan@infradead.org>
The ident level is starting to be annoying. More white space than
actual code. Split out the timer function call into its own function.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
A function scheduled with a timer must not exit with a different
preempt count than it was entered. To make helping users running into
the corresponding BUG() easier also print the name of the bad function
not only its address.
[ tglx: Sanitized printk ]
Signed-off-by: Uwe Kleine-König <u.kleine-koenig@pengutronix.de>
Cc: johnstul@us.ibm.com
Cc: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
warp_clock() currently accesses timekeeping internal state directly, which
is unnecessary. Convert it to use the proper timekeeping interfaces.
Signed-off-by: John Stultz <johnstul@us.ibm.com>
Cc: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Spread p->sighand->siglock locking scope to make sure that
fastpath_timer_check() never iterates over all threads. Without
locking there is small possibility that signal->cputimer will stop
running while we write values to signal->cputime_expires.
Calling thread_group_cputime() from fastpath_timer_check() is not only
bad because it is slow, also it is racy with __exit_signal() which can
lead to invalid signal->{s,u}time values.
Signed-off-by: Stanislaw Gruszka <sgruszka@redhat.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Hidetoshi Seto <seto.hidetoshi@jp.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
When user sets up a timer without associated signal and process does
not use any other cpu timers and does not exit, tsk->signal->cputimer
is enabled and running forever.
Avoid running the timer for no reason.
I used below program to check patch does not break current user space
visible behavior.
#include <sys/time.h>
#include <signal.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <time.h>
#include <unistd.h>
#include <assert.h>
void consume_cpu(void)
{
int i = 0;
int count = 0;
for(i=0; i<100000000; i++)
count++;
}
int main(void)
{
int i;
struct sigaction act;
struct sigevent evt = { };
timer_t tid;
struct itimerspec spec = { };
evt.sigev_notify = SIGEV_NONE;
assert(timer_create(CLOCK_PROCESS_CPUTIME_ID, &evt, &tid) == 0);
spec.it_value.tv_sec = 10;
assert(timer_settime(tid, 0, &spec, NULL) == 0);
for (i = 0; i < 30; i++) {
consume_cpu();
memset(&spec, 0, sizeof(spec));
assert(timer_gettime(tid, &spec) == 0);
printf("%lu.%09lu\n",
(unsigned long) spec.it_value.tv_sec,
(unsigned long) spec.it_value.tv_nsec);
}
assert(timer_delete(tid) == 0);
return 0;
}
Signed-off-by: Stanislaw Gruszka <sgruszka@redhat.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Hidetoshi Seto <seto.hidetoshi@jp.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
When a process deletes cpu timer or a timer expires we do not clear
the expiration cache sig->cputimer_expires.
As a result the fastpath_timer_check() which prevents us to loop over
all threads in case no timer is active is not working and we run the
slow path needlessly on every tick.
Zero sig->cputimer_expires in stop_process_timers().
Signed-off-by: Stanislaw Gruszka <sgruszka@redhat.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Hidetoshi Seto <seto.hidetoshi@jp.fujitsu.com>
Cc: Spencer Candland <spencer@bluehost.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
These functions forgot to run timer_stats_timer_clear_start_info(). It's
unobvious what effect this has and whether it matters much - we won't be
printing it out anyway if the timer's detached.
Untested, just an Ingo trollpatch.
[ Nevertheless correct - tglx ]
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: johnstul@us.ibm.com
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
The current logic which handles clock events programming failures can
increase min_delta_ns unlimited and even can cause overflows.
Sanitize it by:
- prevent zero increase when min_delta_ns == 1
- limiting min_delta_ns to a jiffie
- bail out if the jiffie limit is hit
- add retries stats for /proc/timer_list so we can gather data
Reported-by: Uwe Kleine-Koenig <u.kleine-koenig@pengutronix.de>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
This features has been enabled for quite a while, after testing showed that
easing preemption for light tasks was harmful to high priority threads.
Remove the feature flag.
Signed-off-by: Mike Galbraith <efault@gmx.de>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <1268301675.6785.44.camel@marge.simson.net>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Don't bother with selection when the current cpu is idle. Recent load
balancing changes also make it no longer necessary to check wake_affine()
success before returning the selected sibling, so we now always use it.
Signed-off-by: Mike Galbraith <efault@gmx.de>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <1268301369.6785.36.camel@marge.simson.net>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Allow LAST_BUDDY to kick in sooner, improving cache utilization as soon as
a second buddy pair arrives on scene. The cost is latency starting to climb
sooner, the tbenefit for tbench 8 on my Q6600 box is ~2%. No detrimental
effects noted in normal idesktop usage.
Signed-off-by: Mike Galbraith <efault@gmx.de>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <1268301285.6785.34.camel@marge.simson.net>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Now that we no longer depend on the clock being updated prior to enqueueing
on migratory wakeup, we can clean up a bit, placing calls to update_rq_clock()
exactly where they are needed, ie on enqueue, dequeue and schedule events.
In the case of a freshly enqueued task immediately preempting, we can skip the
update during preemption, as the clock was just updated by the enqueue event.
We also save an unneeded call during a migratory wakeup by not updating the
previous runqueue, where update_curr() won't be invoked.
Signed-off-by: Mike Galbraith <efault@gmx.de>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <1268301199.6785.32.camel@marge.simson.net>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Both avg_overlap and avg_wakeup had an inherent problem in that their accuracy
was detrimentally affected by cross-cpu wakeups, this because we are missing
the necessary call to update_curr(). This can't be fixed without increasing
overhead in our already too fat fastpath.
Additionally, with recent load balancing changes making us prefer to place tasks
in an idle cache domain (which is good for compute bound loads), communicating
tasks suffer when a sync wakeup, which would enable affine placement, is turned
into a non-sync wakeup by SYNC_LESS. With one task on the runqueue, wake_affine()
rejects the affine wakeup request, leaving the unfortunate where placed, taking
frequent cache misses.
Remove it, and recover some fastpath cycles.
Signed-off-by: Mike Galbraith <efault@gmx.de>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <1268301121.6785.30.camel@marge.simson.net>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Testing the load which led to this heuristic (nfs4 kbuild) shows that it has
outlived it's usefullness. With intervening load balancing changes, I cannot
see any difference with/without, so recover there fastpath cycles.
Signed-off-by: Mike Galbraith <efault@gmx.de>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <1268301062.6785.29.camel@marge.simson.net>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Entering nohz code on every micro-idle is costing ~10% throughput for netperf
TCP_RR when scheduling cross-cpu. Rate limiting entry fixes this, but raises
ticks a bit. On my Q6600, an idle box goes from ~85 interrupts/sec to 128.
The higher the context switch rate, the more nohz entry costs. With this patch
and some cycle recovery patches in my tree, max cross cpu context switch rate is
improved by ~16%, a large portion of which of which is this ratelimiting.
Signed-off-by: Mike Galbraith <efault@gmx.de>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <1268301003.6785.28.camel@marge.simson.net>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
This patch is an optimization in perf_event_task_sched_in() to avoid
scheduling the events twice in a row.
Without it, the perf_disable()/perf_enable() pair is invoked twice,
thereby pinned events counts while scheduling flexible events and we go
throuh hw_perf_enable() twice.
By encapsulating, the whole sequence into perf_disable()/perf_enable() we
ensure, hw_perf_enable() is going to be invoked only once because of the
refcount protection.
Signed-off-by: Stephane Eranian <eranian@google.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <1268288765-5326-1-git-send-email-eranian@google.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Put all statistic fields of sched_entity in one struct, sched_statistics,
and embed it into sched_entity.
This change allows to memset the sched_statistics to 0 when needed (for
instance when forking), avoiding bugs of non initialized fields.
Signed-off-by: Lucas De Marchi <lucas.de.marchi@gmail.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <1268275065-18542-1-git-send-email-lucas.de.marchi@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Since pick_next_highest_task_rt() already iterates all the cgroups and
is really only interested in tasks, skip over the !task entries.
Reported-by: Dhaval Giani <dhaval.giani@gmail.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Tested-by: Dhaval Giani <dhaval.giani@gmail.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
From : Ananth N Mavinakayanahalli <ananth@in.ibm.com>
When freeing the instruction slot, the arithmetic to calculate
the index of the slot in the page needs to account for the total
size of the instruction on the various architectures.
Calculate the index correctly when freeing the out-of-line
execution slot.
Reported-by: Sachin Sant <sachinp@in.ibm.com>
Reported-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Signed-off-by: Ananth N Mavinakayanahalli <ananth@in.ibm.com>
Signed-off-by: Masami Hiramatsu <mhiramat@redhat.com>
LKML-Reference: <4B9667AB.9050507@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Anton Blanchard found that he could reliably make the kernel hit a
BUG_ON in the slab allocator by taking a cpu offline and then online
while a system-wide perf record session was running.
The reason is that when the cpu comes up, we completely reinitialize
the ctx field of the struct perf_cpu_context for the cpu. If there is
a system-wide perf record session running, then there will be a struct
perf_event that has a reference to the context, so its refcount will
be 2. (The perf_event has been removed from the context's group_entry
and event_entry lists by perf_event_exit_cpu(), but that doesn't
remove the perf_event's reference to the context and doesn't decrement
the context's refcount.)
When the cpu comes up, perf_event_init_cpu() gets called, and it calls
__perf_event_init_context() on the cpu's context. That resets the
refcount to 1. Then when the perf record session finishes and the
perf_event is closed, the refcount gets decremented to 0 and the
context gets kfreed after an RCU grace period. Since the context
wasn't kmalloced -- it's part of a per-cpu variable -- bad things
happen.
In fact we don't need to completely reinitialize the context when the
cpu comes up. It's sufficient to initialize the context once at boot,
but we need to do it for all possible cpus.
This moves the context initialization to happen at boot time. With
this, we don't trash the refcount and the context never gets kfreed,
and we don't hit the BUG_ON.
Reported-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Paul Mackerras <paulus@samba.org>
Tested-by: Anton Blanchard <anton@samba.org>
Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: <stable@kernel.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Lars-Peter pointed out that the oneshot threaded interrupt handler
code has the following race:
CPU0 CPU1
hande_level_irq(irq X)
mask_ack_irq(irq X)
handle_IRQ_event(irq X)
wake_up(thread_handler)
thread handler(irq X) runs
finalize_oneshot(irq X)
does not unmask due to
!(desc->status & IRQ_MASKED)
return from irq
does not unmask due to
(desc->status & IRQ_ONESHOT)
This leaves the interrupt line masked forever.
The reason for this is the inconsistent handling of the IRQ_MASKED
flag. Instead of setting it in the mask function the oneshot support
sets the flag after waking up the irq thread.
The solution for this is to set/clear the IRQ_MASKED status whenever
we mask/unmask an interrupt line. That's the easy part, but that
cleanup opens another race:
CPU0 CPU1
hande_level_irq(irq)
mask_ack_irq(irq)
handle_IRQ_event(irq)
wake_up(thread_handler)
thread handler(irq) runs
finalize_oneshot_irq(irq)
unmask(irq)
irq triggers again
handle_level_irq(irq)
mask_ack_irq(irq)
return from irq due to IRQ_INPROGRESS
return from irq
does not unmask due to
(desc->status & IRQ_ONESHOT)
This requires that we synchronize finalize_oneshot_irq() with the
primary handler. If IRQ_INPROGESS is set we wait until the primary
handler on the other CPU has returned before unmasking the interrupt
line again.
We probably have never seen that problem because it does not happen on
UP and on SMP the irqbalancer protects us by pinning the primary
handler and the thread to the same CPU.
Reported-by: Lars-Peter Clausen <lars@metafoo.de>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Cc: stable@kernel.org
We are taking a wrong regs snapshot when a trace event triggers.
Either we use get_irq_regs(), which gives us the interrupted
registers if we are in an interrupt, or we use task_pt_regs()
which gives us the state before we entered the kernel, assuming
we are lucky enough to be no kernel thread, in which case
task_pt_regs() returns the initial set of regs when the kernel
thread was started.
What we want is different. We need a hot snapshot of the regs,
so that we can get the instruction pointer to record in the
sample, the frame pointer for the callchain, and some other
things.
Let's use the new perf_fetch_caller_regs() for that.
Comparison with perf record -e lock: -R -a -f -g
Before:
perf [kernel] [k] __do_softirq
|
--- __do_softirq
|
|--55.16%-- __open
|
--44.84%-- __write_nocancel
After:
perf [kernel] [k] perf_tp_event
|
--- perf_tp_event
|
|--41.07%-- lock_acquire
| |
| |--39.36%-- _raw_spin_lock
| | |
| | |--7.81%-- hrtimer_interrupt
| | | smp_apic_timer_interrupt
| | | apic_timer_interrupt
The old case was producing unreliable callchains. Now having
right frame and instruction pointers, we have the trace we
want.
Also syscalls and kprobe events already have the right regs,
let's use them instead of wasting a retrieval.
v2: Follow the rename perf_save_regs() -> perf_fetch_caller_regs()
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Masami Hiramatsu <mhiramat@redhat.com>
Cc: Jason Baron <jbaron@redhat.com>
Cc: Archs <linux-arch@vger.kernel.org>
Events that trigger overflows by interrupting a context can
use get_irq_regs() or task_pt_regs() to retrieve the state
when the event triggered. But this is not the case for some
other class of events like trace events as tracepoints are
executed in the same context than the code that triggered
the event.
It means we need a different api to capture the regs there,
namely we need a hot snapshot to get the most important
informations for perf: the instruction pointer to get the
event origin, the frame pointer for the callchain, the code
segment for user_mode() tests (we always use __KERNEL_CS as
trace events always occur from the kernel) and the eflags
for further purposes.
v2: rename perf_save_regs to perf_fetch_caller_regs as per
Masami's suggestion.
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Masami Hiramatsu <mhiramat@redhat.com>
Cc: Jason Baron <jbaron@redhat.com>
Cc: Archs <linux-arch@vger.kernel.org>
