Skip to content

L
linux
Switch branch/tag
  1. 14 Sep, 2008 1 commit
    • Frank Mayhar's avatar
      timers: fix itimer/many thread hang · f06febc9
      Frank Mayhar authored 
      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: default avatarFrank 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: default avatarIngo Molnar <mingo@elte.hu>
      f06febc9
  3. 13 Sep, 2008 32 commits
  5. 12 Sep, 2008 7 commits
    • Yinghai Lu's avatar
      PCI: re-add debug prints for unmodified BARs · 395a125c
      Yinghai Lu authored 
      Print out for device BAR values before the kernel tries to update them.
Also make related output use KERN_DEBUG.
Signed-off-by: default avatarYinghai Lu <yhlu.kernel@gmail.com>
Signed-off-by: default avatarJesse Barnes <jbarnes@virtuousgeek.org>
      395a125c
    • Santwona Behera's avatar
      niu: panic on reset · cff502a3
      Santwona Behera authored 
      The reset_task function in the niu driver does not reset the tx and rx
buffers properly. This leads to panic on reset. This patch is a
modified implementation of the previously posted fix.
Signed-off-by: default avatarSantwona Behera <santwona.behera@sun.com>
Signed-off-by: default avatarDavid S. Miller <davem@davemloft.net>
      cff502a3
    • David S. Miller's avatar
      sparc: Fix user_regset 'n' field values. · 7d4ee289
      David S. Miller authored 
      As noticed by Russell King, we were not setting this properly
to the number of entries, but rather the total size.

This results in the core dumping code allocating waayyyy too
much memory.
Signed-off-by: default avatarDavid S. Miller <davem@davemloft.net>
      7d4ee289
    • David S. Miller's avatar
      sparc64: Fix PCI error interrupt registry on PSYCHO. · 80a56ab6
      David S. Miller authored 
      We need to pass IRQF_SHARED, otherwise we get things like:

IRQ handler type mismatch for IRQ 33
current handler: PSYCHO_UE
Call Trace:
 [000000000048394c] request_irq+0xac/0x120
 [00000000007c5f6c] psycho_scan_bus+0x98/0x158
 [00000000007c2bc0] pcibios_init+0xdc/0x12c
 [0000000000426a5c] do_one_initcall+0x1c/0x160
 [00000000007c0180] kernel_init+0x9c/0xfc
 [0000000000427050] kernel_thread+0x30/0x60
 [00000000006ae1d0] rest_init+0x10/0x60

on e3500 and similar systems.

On a single board, the UE interrupts of two Psycho nodes
are funneled through the same interrupt, from of_debug=3
dump:

/pci@b,4000: direct translate 2ee --> 21
 ...
/pci@b,2000: direct translate 2ee --> 21

Decimal "33" mentioned above is the hex "21" mentioned here.

Thanks to Meelis Roos for dumps and testing.
Signed-off-by: default avatarDavid S. Miller <davem@davemloft.net>
      80a56ab6
    • David S. Miller's avatar
      Merge branch 'master' of git://git.kernel.org/pub/scm/linux/kernel/git/holtmann/bluetooth-2.6 · d58b622b
      David S. Miller authored
      d58b622b
    • Vegard Nossum's avatar
      netlink: fix overrun in attribute iteration · 1045b03e
      Vegard Nossum authored 
      kmemcheck reported this:

  kmemcheck: Caught 16-bit read from uninitialized memory (f6c1ba30)
  0500110001508abf050010000500000002017300140000006f72672e66726565
   i i i i i i i i i i i i i u u u u u u u u u u u u u u u u u u u
                                   ^

  Pid: 3462, comm: wpa_supplicant Not tainted (2.6.27-rc3-00054-g6397ab9-dirty #13)
  EIP: 0060:[<c05de64a>] EFLAGS: 00010296 CPU: 0
  EIP is at nla_parse+0x5a/0xf0
  EAX: 00000008 EBX: fffffffd ECX: c06f16c0 EDX: 00000005
  ESI: 00000010 EDI: f6c1ba30 EBP: f6367c6c ESP: c0a11e88
   DS: 007b ES: 007b FS: 00d8 GS: 0033 SS: 0068
  CR0: 8005003b CR2: f781cc84 CR3: 3632f000 CR4: 000006d0
  DR0: c0ead9bc DR1: 00000000 DR2: 00000000 DR3: 00000000
  DR6: ffff4ff0 DR7: 00000400
   [<c05d4b23>] rtnl_setlink+0x63/0x130
   [<c05d5f75>] rtnetlink_rcv_msg+0x165/0x200
   [<c05ddf66>] netlink_rcv_skb+0x76/0xa0
   [<c05d5dfe>] rtnetlink_rcv+0x1e/0x30
   [<c05dda21>] netlink_unicast+0x281/0x290
   [<c05ddbe9>] netlink_sendmsg+0x1b9/0x2b0
   [<c05beef2>] sock_sendmsg+0xd2/0x100
   [<c05bf945>] sys_sendto+0xa5/0xd0
   [<c05bf9a6>] sys_send+0x36/0x40
   [<c05c03d6>] sys_socketcall+0x1e6/0x2c0
   [<c020353b>] sysenter_do_call+0x12/0x3f
   [<ffffffff>] 0xffffffff

This is the line in nla_ok():

  /**
   * nla_ok - check if the netlink attribute fits into the remaining bytes
   * @nla: netlink attribute
   * @remaining: number of bytes remaining in attribute stream
   */
  static inline int nla_ok(const struct nlattr *nla, int remaining)
  {
          return remaining >= sizeof(*nla) &&
                 nla->nla_len >= sizeof(*nla) &&
                 nla->nla_len <= remaining;
  }

It turns out that remaining can become negative due to alignment in
nla_next(). But GCC promotes "remaining" to unsigned in the test
against sizeof(*nla) above. Therefore the test succeeds, and the
nla_for_each_attr() may access memory outside the received buffer.

A short example illustrating this point is here:

  #include <stdio.h>

  main(void)
  {
          printf("%d\n", -1 >= sizeof(int));
  }

...which prints "1".

This patch adds a cast in front of the sizeof so that GCC will make
a signed comparison and fix the illegal memory dereference. With the
patch applied, there is no kmemcheck report.
Signed-off-by: default avatarVegard Nossum <vegard.nossum@gmail.com>
Acked-by: default avatarThomas Graf <tgraf@suug.ch>
Signed-off-by: default avatarDavid S. Miller <davem@davemloft.net>
      1045b03e
    • Marcel Holtmann's avatar
      [Bluetooth] Fix regression from using default link policy · 7c6a329e
      Marcel Holtmann authored 
      To speed up the Simple Pairing connection setup, the support for the
default link policy has been enabled. This is in contrast to settings
the link policy on every connection setup. Using the default link policy
is the preferred way since there is no need to dynamically change it for
every connection.

For backward compatibility reason and to support old userspace the
HCISETLINKPOL ioctl has been switched over to using hci_request() to
issue the HCI command for setting the default link policy instead of
just storing it in the HCI device structure.

However the hci_request() can only be issued when the device is
brought up. If used on a device that is registered, but still down
it will timeout and fail. This is problematic since the command is
put on the TX queue and the Bluetooth core tries to submit it to
hardware that is not ready yet. The timeout for these requests is
10 seconds and this causes a significant regression when setting up
a new device.

The userspace can perfectly handle a failure of the HCISETLINKPOL
ioctl and will re-submit it later, but the 10 seconds delay causes
a problem. So in case hci_request() is called on a device that is
still down, just fail it with ENETDOWN to indicate what happens.
Signed-off-by: default avatarMarcel Holtmann <marcel@holtmann.org>
      7c6a329e
GitLab Nexedi Edition | About GitLab | About Nexedi | 沪ICP备2021021310号-2 | 沪ICP备2021021310号-7