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>
This adds support for setting the TAI value (International Atomic Time). The
value is reported back to userspace via timex (as we don't have a
ntp_gettime() syscall).
Signed-off-by: Roman Zippel <zippel@linux-m68k.org>
Cc: john stultz <johnstul@us.ibm.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
This adds the set_restore_sigmask() inline in <linux/thread_info.h> and
replaces every set_thread_flag(TIF_RESTORE_SIGMASK) with a call to it. No
change, but abstracts the details of the flag protocol from all the calls.
Signed-off-by: Roland McGrath <roland@redhat.com>
Cc: Oleg Nesterov <oleg@tv-sign.ru>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Martin Schwidefsky <schwidefsky@de.ibm.com>
Cc: Heiko Carstens <heiko.carstens@de.ibm.com>
Cc: "Luck, Tony" <tony.luck@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
* Modify sched_affinity functions to pass cpumask_t variables by reference
instead of by value.
* Use new set_cpus_allowed_ptr function.
Depends on:
[sched-devel]: sched: add new set_cpus_allowed_ptr function
Cc: Paul Jackson <pj@sgi.com>
Cc: Cliff Wickman <cpw@sgi.com>
Signed-off-by: Mike Travis <travis@sgi.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Convert all the nanosleep related users of restart_block to the
new nanosleep specific restart_block fields.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
hrtimer_nanosleep_restart() clears/restores restart_block->fn. This is
pointless and complicates its usage. Note that if sys_restart_syscall()
doesn't actually happen, we have a bogus "pending" restart->fn anyway,
this is harmless.
Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Cc: Alexey Dobriyan <adobriyan@sw.ru>
Cc: Pavel Emelyanov <xemul@sw.ru>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Toyo Abe <toyoa@mvista.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Spotted by Pavel Emelyanov and Alexey Dobriyan.
compat_sys_nanosleep() implicitly uses hrtimer_nanosleep_restart(), this can't
work. Make a suitable compat_nanosleep_restart() helper.
Introduced by commit c70878b4e0
hrtimer: hook compat_sys_nanosleep up to high res timer code
Also, set ->addr_limit = KERNEL_DS before doing hrtimer_nanosleep(), this func
was changed by the previous patch and now takes the "__user *" parameter.
Thanks to Ingo Molnar for fixing the bug in this patch.
Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Alexey Dobriyan <adobriyan@sw.ru>
Cc: Pavel Emelyanov <xemul@sw.ru>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Toyo Abe <toyoa@mvista.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
* ssh://master.kernel.org/pub/scm/linux/kernel/git/tglx/linux-2.6-hrt:
hrtimer: hook compat_sys_nanosleep up to high res timer code
hrtimer: Rework hrtimer_nanosleep to make sys_compat_nanosleep easier
Now we have high res timers on ppc64 I thought Id test them. It turns
out compat_sys_nanosleep hasnt been converted to the hrtimer code and so
is limited to HZ resolution.
The follow patch converts compat_sys_nanosleep to use high res timers.
Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
This patch implements the necessary compat code for the timerfd system call.
Signed-off-by: Davide Libenzi <davidel@xmailserver.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
I noticed that almost all architectures implemented exactly the same
sys32_sysinfo... except parisc, where a bug was to be found in handling of
the uptime. So let's remove a whole whack of code for fun and profit.
Cribbed compat_sys_sysinfo from x86_64's implementation, since I figured it
would be the best tested.
This patch incorporates Arnd's suggestion of not using set_fs/get_fs, but
instead extracting out the common code from sys_sysinfo.
Cc: Christoph Hellwig <hch@infradead.org>
Cc: <linux-arch@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
This is needed on bigendian 64bit architectures.
Signed-off-by: Stephen Rothwell <sfr@canb.auug.org.au>
Acked-by: Christoph Lameter <clameter@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
This means we can call it when the bitmap we want to fetch is declared
const.
Signed-off-by: Stephen Rothwell <sfr@canb.auug.org.au>
Cc: Christoph Lameter <clameter@engr.sgi.com>
Cc: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Revert Andrew Morton's patch to temporarily hack around the lack of a
declaration of sigset_t in linux/compat.h to make the block-disablement
patches build on IA64. This got accidentally pushed to Linus and should
be fixed in a different manner.
Also make linux/compat.h #include asm/signal.h to gain a definition of
sigset_t so that it can externally declare sigset_from_compat().
This has been compile-tested for i386, x86_64, ia64, mips, mips64, frv, ppc and
ppc64 and run-tested on frv.
Signed-off-by: David Howells <dhowells@redhat.com>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Create a new header file, fs/internal.h, for common definitions local to the
sources in the fs/ directory.
Move extern definitions that should be in header files from fs/*.c to
fs/internal.h or other main header files where they span directories.
Signed-Off-By: David Howells <dhowells@redhat.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
The clock_nanosleep() function does not return the time remaining when the
sleep is interrupted by a signal.
This patch creates a new call out, compat_clock_nanosleep_restart(), which
handles returning the remaining time after a sleep is interrupted. This
patch revives clock_nanosleep_restart(). It is now accessed via the new
call out. The compat_clock_nanosleep_restart() is used for compatibility
access.
Since this is implemented in compatibility mode the normal path is
virtually unaffected - no real performance impact.
Signed-off-by: Toyo Abe <toyoa@mvista.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Roland McGrath <roland@redhat.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
I'm testing glibc on MIPS64, little-endian, N32, O32 and N64 multilibs.
Among the NPTL test failures seen are some arising from sigsuspend problems
for N32: it blocks the wrong signals, so SIGCANCEL (SIGRTMIN) is blocked
despite glibc's carefully excluding it from sets of signals to block.
Specifically, testing suggests it blocks signal N^32 instead of signal N,
so (in the example tested) blocking SIGUSR1 (17) blocks signal 49 instead.
glibc's sigset_t uses an array of unsigned long, as does the kernel.
In both cases, signal N+1 is represented as
(1UL << (N % (8 * sizeof (unsigned long)))) in word number
(N / (8 * sizeof (unsigned long))).
Thus the N32 glibc uses an array of 32-bit words and the N64 kernel uses an
array of 64-bit words. For little-endian, the layout is the same, with
signals 1-32 in the first 4 bytes, signals 33-64 in the second, etc.; for
big-endian, userspace has that layout while in the kernel each 8 bytes have
the two halves swapped from the userspace layout.
The N32 sigsuspend syscall uses sigset_from_compat to convert the userspace
sigset to kernel format. If __COMPAT_ENDIAN_SWAP__ is *not* set, this uses
logic of the form
set->sig[0] = compat->sig[0] | (((long)compat->sig[1]) << 32 )
to convert the userspace sigset to a kernel one. This looks correct to me
for both big and little endian, given that in userspace compat->sig[1] will
represent signals 33-64, and so will the high 32 bits of set->sig[0] in the
kernel. If however __COMPAT_ENDIAN_SWAP__ *is* set, as it is for
__MIPSEL__, it uses
set->sig[0] = compat->sig[1] | (((long)compat->sig[0]) << 32 );
which seems incorrect for both big and little endian, and would
explain the observed symptoms.
This code is the only use of __COMPAT_ENDIAN_SWAP__, so if incorrect
then that macro serves no purpose, in which case something like the
following patch would seem appropriate to remove it.
Signed-off-by: Joseph Myers <joseph@codesourcery.com>
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
Cc: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
The definition of the third parameter is a pointer to an array of virtual
addresses which give us some trouble. The existing code calculated the
wrong address in the array since I used void to avoid having to specify a
type.
I now use the correct type "compat_uptr_t __user *" in the definition of
the function in kernel/compat.c.
However, I used __u32 in syscalls.h. Would have to include compat.h there
in order to provide the same definition which would generate an ugly
include situation.
On both ia64 and x86_64 compat_uptr_t is u32. So this works although
parameter declarations differ.
Signed-off-by: Christoph Lameter <clameter@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
sys_move_pages() support for 32bit (i386 plus x86_64 compat layer)
Add support for move_pages() on i386 and also add the compat functions
necessary to run 32 bit binaries on x86_64.
Add compat_sys_move_pages to the x86_64 32bit binary layer. Note that it is
not up to date so I added the missing pieces. Not sure if this is done the
right way.
[akpm@osdl.org: compile fix]
Signed-off-by: Christoph Lameter <clameter@sgi.com>
Cc: Andi Kleen <ak@muc.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
32-bit syscall compatibility support. (This patch also moves all futex
related compat functionality into kernel/futex_compat.c.)
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Arjan van de Ven <arjan@infradead.org>
Acked-by: Ulrich Drepper <drepper@redhat.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Create compat_sys_adjtimex and use it an all appropriate places.
Signed-off-by: Stephen Rothwell <sfr@canb.auug.org.au>
Cc: Arnd Bergmann <arnd@arndb.de>
Acked-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
A bunch of asm/bug.h includes are both not needed (since it will get
pulled anyway) and bogus (since they are done too early). Removed.
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
The TIF_RESTORE_SIGMASK flag allows us to have a generic implementation of
sys_rt_sigsuspend() instead of duplicating it for each architecture. This
provides such an implementation and makes arch/powerpc use it.
It also tidies up the ppc32 sys_sigsuspend() to use TIF_RESTORE_SIGMASK.
Signed-off-by: David Woodhouse <dwmw2@infradead.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
The comment in compat.c is wrong, every architecture provides a
get_compat_sigevent() for the IPC compat code already.
This basically moves the x86_64 version to common code and removes all the
others.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Acked-by: Paul Mackerras <paulus@samba.org>
Cc: Arnd Bergmann <arnd@arndb.de>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Martin Schwidefsky <schwidefsky@de.ibm.com>
Cc: "David S. Miller" <davem@davemloft.net>
Acked-by: Andi Kleen <ak@muc.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
I have no idea how a bug like this lasted so long. Anyways, obvious
memset()'ing of incorrect pointer.
Signed-off-by: David S. Miller <davem@davemloft.net>
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!
Frank Mayhar