Overview
This patch reworks the handling of POSIX CPU timers, including the
ITIMER_PROF, ITIMER_VIRT timers and rlimit handling. It was put together
with the help of Roland McGrath, the owner and original writer of this code.
The problem we ran into, and the reason for this rework, has to do with using
a profiling timer in a process with a large number of threads. It appears
that the performance of the old implementation of run_posix_cpu_timers() was
at least O(n*3) (where "n" is the number of threads in a process) or worse.
Everything is fine with an increasing number of threads until the time taken
for that routine to run becomes the same as or greater than the tick time, at
which point things degrade rather quickly.
This patch fixes bug 9906, "Weird hang with NPTL and SIGPROF."
Code Changes
This rework corrects the implementation of run_posix_cpu_timers() to make it
run in constant time for a particular machine. (Performance may vary between
one machine and another depending upon whether the kernel is built as single-
or multiprocessor and, in the latter case, depending upon the number of
running processors.) To do this, at each tick we now update fields in
signal_struct as well as task_struct. The run_posix_cpu_timers() function
uses those fields to make its decisions.
We define a new structure, "task_cputime," to contain user, system and
scheduler times and use these in appropriate places:
struct task_cputime {
cputime_t utime;
cputime_t stime;
unsigned long long sum_exec_runtime;
};
This is included in the structure "thread_group_cputime," which is a new
substructure of signal_struct and which varies for uniprocessor versus
multiprocessor kernels. For uniprocessor kernels, it uses "task_cputime" as
a simple substructure, while for multiprocessor kernels it is a pointer:
struct thread_group_cputime {
struct task_cputime totals;
};
struct thread_group_cputime {
struct task_cputime *totals;
};
We also add a new task_cputime substructure directly to signal_struct, to
cache the earliest expiration of process-wide timers, and task_cputime also
replaces the it_*_expires fields of task_struct (used for earliest expiration
of thread timers). The "thread_group_cputime" structure contains process-wide
timers that are updated via account_user_time() and friends. In the non-SMP
case the structure is a simple aggregator; unfortunately in the SMP case that
simplicity was not achievable due to cache-line contention between CPUs (in
one measured case performance was actually _worse_ on a 16-cpu system than
the same test on a 4-cpu system, due to this contention). For SMP, the
thread_group_cputime counters are maintained as a per-cpu structure allocated
using alloc_percpu(). The timer functions update only the timer field in
the structure corresponding to the running CPU, obtained using per_cpu_ptr().
We define a set of inline functions in sched.h that we use to maintain the
thread_group_cputime structure and hide the differences between UP and SMP
implementations from the rest of the kernel. The thread_group_cputime_init()
function initializes the thread_group_cputime structure for the given task.
The thread_group_cputime_alloc() is a no-op for UP; for SMP it calls the
out-of-line function thread_group_cputime_alloc_smp() to allocate and fill
in the per-cpu structures and fields. The thread_group_cputime_free()
function, also a no-op for UP, in SMP frees the per-cpu structures. The
thread_group_cputime_clone_thread() function (also a UP no-op) for SMP calls
thread_group_cputime_alloc() if the per-cpu structures haven't yet been
allocated. The thread_group_cputime() function fills the task_cputime
structure it is passed with the contents of the thread_group_cputime fields;
in UP it's that simple but in SMP it must also safely check that tsk->signal
is non-NULL (if it is it just uses the appropriate fields of task_struct) and,
if so, sums the per-cpu values for each online CPU. Finally, the three
functions account_group_user_time(), account_group_system_time() and
account_group_exec_runtime() are used by timer functions to update the
respective fields of the thread_group_cputime structure.
Non-SMP operation is trivial and will not be mentioned further.
The per-cpu structure is always allocated when a task creates its first new
thread, via a call to thread_group_cputime_clone_thread() from copy_signal().
It is freed at process exit via a call to thread_group_cputime_free() from
cleanup_signal().
All functions that formerly summed utime/stime/sum_sched_runtime values from
from all threads in the thread group now use thread_group_cputime() to
snapshot the values in the thread_group_cputime structure or the values in
the task structure itself if the per-cpu structure hasn't been allocated.
Finally, the code in kernel/posix-cpu-timers.c has changed quite a bit.
The run_posix_cpu_timers() function has been split into a fast path and a
slow path; the former safely checks whether there are any expired thread
timers and, if not, just returns, while the slow path does the heavy lifting.
With the dedicated thread group fields, timers are no longer "rebalanced" and
the process_timer_rebalance() function and related code has gone away. All
summing loops are gone and all code that used them now uses the
thread_group_cputime() inline. When process-wide timers are set, the new
task_cputime structure in signal_struct is used to cache the earliest
expiration; this is checked in the fast path.
Performance
The fix appears not to add significant overhead to existing operations. It
generally performs the same as the current code except in two cases, one in
which it performs slightly worse (Case 5 below) and one in which it performs
very significantly better (Case 2 below). Overall it's a wash except in those
two cases.
I've since done somewhat more involved testing on a dual-core Opteron system.
Case 1: With no itimer running, for a test with 100,000 threads, the fixed
kernel took 1428.5 seconds, 513 seconds more than the unfixed system,
all of which was spent in the system. There were twice as many
voluntary context switches with the fix as without it.
Case 2: With an itimer running at .01 second ticks and 4000 threads (the most
an unmodified kernel can handle), the fixed kernel ran the test in
eight percent of the time (5.8 seconds as opposed to 70 seconds) and
had better tick accuracy (.012 seconds per tick as opposed to .023
seconds per tick).
Case 3: A 4000-thread test with an initial timer tick of .01 second and an
interval of 10,000 seconds (i.e. a timer that ticks only once) had
very nearly the same performance in both cases: 6.3 seconds elapsed
for the fixed kernel versus 5.5 seconds for the unfixed kernel.
With fewer threads (eight in these tests), the Case 1 test ran in essentially
the same time on both the modified and unmodified kernels (5.2 seconds versus
5.8 seconds). The Case 2 test ran in about the same time as well, 5.9 seconds
versus 5.4 seconds but again with much better tick accuracy, .013 seconds per
tick versus .025 seconds per tick for the unmodified kernel.
Since the fix affected the rlimit code, I also tested soft and hard CPU limits.
Case 4: With a hard CPU limit of 20 seconds and eight threads (and an itimer
running), the modified kernel was very slightly favored in that while
it killed the process in 19.997 seconds of CPU time (5.002 seconds of
wall time), only .003 seconds of that was system time, the rest was
user time. The unmodified kernel killed the process in 20.001 seconds
of CPU (5.014 seconds of wall time) of which .016 seconds was system
time. Really, though, the results were too close to call. The results
were essentially the same with no itimer running.
Case 5: With a soft limit of 20 seconds and a hard limit of 2000 seconds
(where the hard limit would never be reached) and an itimer running,
the modified kernel exhibited worse tick accuracy than the unmodified
kernel: .050 seconds/tick versus .028 seconds/tick. Otherwise,
performance was almost indistinguishable. With no itimer running this
test exhibited virtually identical behavior and times in both cases.
In times past I did some limited performance testing. those results are below.
On a four-cpu Opteron system without this fix, a sixteen-thread test executed
in 3569.991 seconds, of which user was 3568.435s and system was 1.556s. On
the same system with the fix, user and elapsed time were about the same, but
system time dropped to 0.007 seconds. Performance with eight, four and one
thread were comparable. Interestingly, the timer ticks with the fix seemed
more accurate: The sixteen-thread test with the fix received 149543 ticks
for 0.024 seconds per tick, while the same test without the fix received 58720
for 0.061 seconds per tick. Both cases were configured for an interval of
0.01 seconds. Again, the other tests were comparable. Each thread in this
test computed the primes up to 25,000,000.
I also did a test with a large number of threads, 100,000 threads, which is
impossible without the fix. In this case each thread computed the primes only
up to 10,000 (to make the runtime manageable). System time dominated, at
1546.968 seconds out of a total 2176.906 seconds (giving a user time of
629.938s). It received 147651 ticks for 0.015 seconds per tick, still quite
accurate. There is obviously no comparable test without the fix.
Signed-off-by: Frank Mayhar <fmayhar@google.com>
Cc: Roland McGrath <roland@redhat.com>
Cc: Alexey Dobriyan <adobriyan@gmail.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
signal_struct->tsk points to the ->group_leader and thus we have the nasty
code in de_thread() which has to change it and restart ->real_timer if the
leader is changed.
Use "struct pid *leader_pid" instead. This also allows us to kill now
unneeded send_group_sig_info().
Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Acked-by: "Eric W. Biederman" <ebiederm@xmission.com>
Cc: Davide Libenzi <davidel@xmailserver.org>
Cc: Pavel Emelyanov <xemul@openvz.org>
Acked-by: Roland McGrath <roland@redhat.com>
Acked-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
As scheduled, do_setitimer() now returns -EINVAL for invalid timeval.
Signed-off-by: Adrian Bunk <bunk@stusta.de>
Acked-by: Thomas Gleixner <tglx@linutronix.de>
Cc: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Remove includes of <linux/smp_lock.h> where it is not used/needed.
Suggested by Al Viro.
Builds cleanly on x86_64, i386, alpha, ia64, powerpc, sparc,
sparc64, and arm (all 59 defconfigs).
Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Fix potential setitimer DoS with high-res timers by pushing itimer rearm
processing to process context.
[Fixes from: Ingo Molnar <mingo@elte.hu>]
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Cc: john stultz <johnstul@us.ibm.com>
Cc: Roman Zippel <zippel@linux-m68k.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Implement high resolution timers on top of the hrtimers infrastructure and the
clockevents / tick-management framework. This provides accurate timers for
all hrtimer subsystem users.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Cc: john stultz <johnstul@us.ibm.com>
Cc: Roman Zippel <zippel@linux-m68k.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
- hrtimers did not use the hrtimer_restart enum and relied on the implict
int representation. Fix the prototypes and the functions using the enums.
- Use seperate name spaces for the enumerations
- Convert hrtimer_restart macro to inline function
- Add comments
No functional changes.
[akpm@osdl.org: fix input driver]
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Cc: john stultz <johnstul@us.ibm.com>
Cc: Roman Zippel <zippel@linux-m68k.org>
Cc: Dmitry Torokhov <dtor@mail.ru>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
The nanosleep cleanup allows to remove the data field of hrtimer. The
callback function can use container_of() to get it's own data. Since the
hrtimer structure is anyway embedded in other structures, this adds no
overhead.
Signed-off-by: Roman Zippel <zippel@linux-m68k.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Pass current time to hrtimer_forward(). This allows to use the softirq time
in the timer base when the forward function is called from the timer callback.
Other places pass current time with a call to timer->base->get_time().
Signed-off-by: Roman Zippel <zippel@linux-m68k.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
According to the specification the timevals must be validated and an
errorcode -EINVAL returned in case the timevals are not in canonical form.
This check was never done in Linux.
The pre 2.6.16 code converted invalid timevals silently. Negative timeouts
were converted by the timeval_to_jiffies conversion to the maximum timeout.
hrtimers and the ktime_t operations expect timevals in canonical form.
Otherwise random results might happen on 32 bits machines due to the
optimized ktime_add/sub operations. Negative timeouts are treated as
already expired. This might break applications which work on pre 2.6.16.
To prevent random behaviour and API breakage the timevals are checked and
invalid timevals sanitized in a simliar way as the pre 2.6.16 code did.
Invalid timevals are reported with a per boot limited number of kernel
messages so applications which use this misfeature can be corrected.
After a grace period of one year the sanitizing should be replaced by a
correct validation check. This is also documented in
Documentation/feature-removal-schedule.txt
The validation and sanitizing is done inside do_setitimer so all callers
(sys_setitimer, compat_sys_setitimer, osf_setitimer) are catched.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
alarm() calls the kernel with an unsigend int timeout in seconds. The
value is stored in the tv_sec field of a struct timeval to setup the
itimer. The tv_sec field of struct timeval is of type long, which causes
the tv_sec value to be negative on 32 bit machines if seconds > INT_MAX.
Before the hrtimer merge (pre 2.6.16) such a negative value was converted
to the maximum jiffies timeout by the timeval_to_jiffies conversion. It's
not clear whether this was intended or just happened to be done by the
timeval_to_jiffies code.
hrtimers expect a timeval in canonical form and treat a negative timeout as
already expired. This breaks the legitimate usage of alarm() with a
timeout value > INT_MAX seconds.
For 32 bit machines it is therefor necessary to limit the internal seconds
value to avoid API breakage. Instead of doing this in all implementations
of sys_alarm the duplicated sys_alarm code is moved into a common function
in itimer.c
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
This resolves bugzilla bug#5617. The oldvalue of the timer was read after the
timer was cancelled, so the remaining time was always zero.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
The itimer conversion removed the locking which protects the timer and
variables in the shared signal structure. Steven Rostedt found the problem in
the latest -rt patches.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Fix the recent off-by-one fix in the itimer code:
1. The repeating timer is figured using the requested time
(not +1 as we know where we are in the jiffie).
2. The tests for interval too large are left to the time_val to jiffie code.
Signed-off-by: George Anzinger <george@mvista.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
As Steven Rostedt pointed out, there are 2 problems with ITIMER_REAL
timers.
1. do_setitimer() does not call del_timer_sync() in case
when the timer is not pending (it_real_value() returns 0).
This is wrong, the timer may still be running, and it can
rearm itself.
2. It calls del_timer_sync() with tsk->sighand->siglock held.
This is deadlockable, because timer's handler needs this
lock too.
Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Acked-by: Steven Rostedt <rostedt@goodmis.org>
Cc: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
It seems that the code responsible for this is in kernel/itimer.c:126:
p->signal->real_timer.expires = jiffies + interval;
add_timer(&p->signal->real_timer);
If you request an interval of, lets say 900 usecs, the interval given by
timeval_to_jiffies will be 1.
If you request this when we are half-way between two timer ticks, the
interval will only give 400 usecs.
If we want to guarantee that we never ever give intervals less than
requested, the simple solution would be to change that to:
p->signal->real_timer.expires = jiffies + interval + 1;
This however will produce pathological cases, like having a idle system
being requested 1 ms timeouts will give systematically 2 ms timeouts,
whereas currently it simply gives a few usecs less than 1 ms.
The complex (and more computationally expensive) solution would be to
check the gettimeofday time, and compute the correct number of jiffies.
This way, if we request a 300 usecs timer 200 usecs inside the timer
tick, we can wait just one tick, but not if we are 800 usecs inside the
tick. This would also mean that we would have to lock preemption during
these computations to avoid races, etc.
I've searched the archives but couldn't find this particular issue being
discussed before.
Attached is a patch to do the simple solution, in case anybody thinks
that it should be used.
Signed-Off-By: Paulo Marques <pmarques@grupopie.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Initial git repository build. I'm not bothering with the full history,
even though we have it. We can create a separate "historical" git
archive of that later if we want to, and in the meantime it's about
3.2GB when imported into git - space that would just make the early
git days unnecessarily complicated, when we don't have a lot of good
infrastructure for it.
Let it rip!
Heiko Carstens
Frank Mayhar
Oleg Nesterov
Daniel Walker
Adrian Bunk
Randy Dunlap
Thomas Gleixner
Roman Zippel
George Anzinger
Paulo Marques
Linus Torvalds