Skip to content

L
linux-davinci
Switch branch/tag
  1. 10 Oct, 2009 1 commit
    • KAMEZAWA Hiroyuki's avatar
      This is a patch for coalescing access to res_counter at charging by percpu · 7c97f8db
      KAMEZAWA Hiroyuki authored 
      caching.  At charge, memcg charges 64pages and remember it in percpu
cache.  Because it's cache, drain/flush if necessary.

This version uses public percpu area.
 2 benefits for using public percpu area.
 1. Sum of stocked charge in the system is limited to # of cpus
    not to the number of memcg. This shows better synchonization.
 2. drain code for flush/cpuhotplug is very easy (and quick)

The most important point of this patch is that we never touch res_counter
in fast path. The res_counter is system-wide shared counter which is modified
very frequently. We shouldn't touch it as far as we can for avoiding
false sharing.

On x86-64 8cpu server, I tested overheads of memcg at page fault by
running a program which does map/fault/unmap in a loop. Running
a task per a cpu by taskset and see sum of the number of page faults
in 60secs.

[without memcg config]
  40156968  page-faults              #      0.085 M/sec   ( +-   0.046% )
  27.67 cache-miss/faults

[root cgroup]
  36659599  page-faults              #      0.077 M/sec   ( +-   0.247% )
  31.58 cache miss/faults

[in a child cgroup]
  18444157  page-faults              #      0.039 M/sec   ( +-   0.133% )
  69.96 cache miss/faults

[ + coalescing uncharge patch]
  27133719  page-faults              #      0.057 M/sec   ( +-   0.155% )
  47.16 cache miss/faults

[ + coalescing uncharge patch + this patch ]
  34224709  page-faults              #      0.072 M/sec   ( +-   0.173% )
  34.69 cache miss/faults

Changelog (since Oct/2):
  - updated comments
  - replaced get_cpu_var() with __get_cpu_var() if possible.
  - removed mutex for system-wide drain. adds a counter instead of it.
  - removed CONFIG_HOTPLUG_CPU

Changelog (old):
  - rebased onto the latest mmotm
  - moved charge size check before __GFP_WAIT check for avoiding unnecesary
  - added asynchronous flush routine.
  - fixed bugs pointed out by Nishimura-san.
Signed-off-by: default avatarKAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: default avatarAndrew Morton <akpm@linux-foundation.org>
      7c97f8db
  3. 09 Nov, 2009 1 commit
    • KAMEZAWA Hiroyuki's avatar
      In massive parallel enviroment, res_counter can be a performance · a6404ad7
      KAMEZAWA Hiroyuki authored 
      bottleneck.  One strong techinque to reduce lock contention is reducing
calls by coalescing some amount of calls into one.

Considering charge/uncharge chatacteristic,
	- charge is done one by one via demand-paging.
	- uncharge is done by
		- in chunk at munmap, truncate, exit, execve...
		- one by one via vmscan/paging.

It seems we have a chance to coalesce uncharges for improving scalability
at unmap/truncation.

This patch is a for coalescing uncharge.  For avoiding scattering memcg's
structure to functions under /mm, this patch adds memcg batch uncharge
information to the task.  A reason for per-task batching is for making use
of caller's context information.  We do batched uncharge (deleyed
uncharge) when truncation/unmap occurs but do direct uncharge when
uncharge is called by memory reclaim (vmscan.c).

The degree of coalescing depends on callers
  - at invalidate/trucate... pagevec size
  - at unmap ....ZAP_BLOCK_SIZE
(memory itself will be freed in this degree.)
Then, we'll not coalescing too much.

On x86-64 8cpu server, I tested overheads of memcg at page fault by
running a program which does map/fault/unmap in a loop. Running
a task per a cpu by taskset and see sum of the number of page faults
in 60secs.

[without memcg config]
  40156968  page-faults              #      0.085 M/sec   ( +-   0.046% )
  27.67 cache-miss/faults
[root cgroup]
  36659599  page-faults              #      0.077 M/sec   ( +-   0.247% )
  31.58 miss/faults
[in a child cgroup]
  18444157  page-faults              #      0.039 M/sec   ( +-   0.133% )
  69.96 miss/faults
[child with this patch]
  27133719  page-faults              #      0.057 M/sec   ( +-   0.155% )
  47.16 miss/faults

We can see some amounts of improvement.
(root cgroup doesn't affected by this patch)
Another patch for "charge" will follow this and above will be improved more.

Changelog(since 2009/10/02):
 - renamed filed of memcg_batch (as pages to bytes, memsw to memsw_bytes)
 - some clean up and commentary/description updates.
 - added initialize code to copy_process(). (possible bug fix)

Changelog(old):
 - fixed !CONFIG_MEM_CGROUP case.
 - rebased onto the latest mmotm + softlimit fix patches.
 - unified patch for callers
 - added commetns.
 - make ->do_batch as bool.
 - removed css_get() at el. We don't need it.
Signed-off-by: default avatarKAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: default avatarAndrew Morton <akpm@linux-foundation.org>
      a6404ad7
  5. 10 Nov, 2009 1 commit
  7. 03 Nov, 2009 1 commit
  9. 25 Sep, 2009 1 commit
  11. 13 Nov, 2009 1 commit
  13. 09 Nov, 2009 1 commit
  15. 09 Oct, 2009 1 commit
  17. 09 Nov, 2009 1 commit
  19. 10 Nov, 2009 2 commits
  21. 25 Sep, 2009 1 commit
  23. 21 Nov, 2008 1 commit
    • Warren Turkal's avatar
      This is a patchset to change the way that the HFS+ filesystem detects · 0c0a4135
      Warren Turkal authored 
      whether a volume has a journal or not.

The code currently mounts an HFS+ volume read-only by default when a
journal is detected.  One can force a read/write mount by giving the
"force" mount option.  The current code has this behavior since there is
no support for the HFS+ journal.

My problem is that the detection of the journal could be better.  The
current code tests the attribute bit in the volume header that indicates
there is a journal.  If that bit is set, the code assumes that there is a
journal.

Unfortunately, this is not enough to really determine if there is a
journal or not.  When that bit is set, one must also examine the journal
info block field of the volume header.  If this field is 0, there is no
journal, and the volume can be mounted read/write.


This patch:

The HFS+ support in the kernel currently will mount an HFS+ volume
read-only if the volume header has the attribute bit set that indicates
there is a journal.  The kernel does this because there is no support for
a journalled HFS+ volume.

The problem is that this is only half of what needs to be checked to see
if there really is a journal.  There is also an entry in the volume header
that tells you where to find the journal info block.  In the kernel
version of the kernel, this 4 byte block is labeled reserved.  This patch
identifies the journal info block in the header.
Signed-off-by: default avatarWarren Turkal <wt@penguintechs.org>
Cc: Roman Zippel <zippel@linux-m68k.org>
Signed-off-by: default avatarAndrew Morton <akpm@linux-foundation.org>
      0c0a4135
  25. 03 Nov, 2009 2 commits
  27. 16 Oct, 2009 1 commit
  29. 13 Jul, 2009 1 commit
  31. 13 Nov, 2009 1 commit
  33. 10 Nov, 2009 1 commit
  35. 14 Nov, 2009 1 commit
  37. 11 Nov, 2009 1 commit
  39. 03 Nov, 2009 5 commits
  41. 14 Oct, 2009 1 commit
  43. 24 Sep, 2009 1 commit
  45. 14 Oct, 2009 2 commits
  47. 24 Sep, 2009 1 commit
  49. 25 Sep, 2009 1 commit
  51. 10 Nov, 2009 1 commit
  53. 14 Oct, 2009 1 commit
  55. 16 Oct, 2009 1 commit
  57. 09 Nov, 2009 1 commit
    • Anton Vorontsov's avatar
      RTC core won't allow wakeup alarms to be set if RTC devices' parent (i.e. · 0d0b9e35
      Anton Vorontsov authored 
      i2c_client or spi_device) isn't wakeup capable.

For I2C devices there is I2C_CLIENT_WAKE flag exists that we can pass via
board info, and if set, I2C core will initialize wakeup capability.  For
SPI devices there is no such flag at all.

I believe that it's not platform code responsibility to allow or disallow
wakeups, instead, drivers themselves should set the capability if a device
can trigger wakeups.

That's what drivers/base/power/sysfs.c says:

 * It is the responsibility of device drivers to enable (or disable)
 * wakeup signaling as part of changing device power states, respecting
 * the policy choices provided through the driver model.

I2C and SPI RTC devices send wakeup events via interrupt lines, so we
should set the wakeup capability if IRQ is routed.

Ideally we should also check irq for wakeup capability before setting
device's capability, i.e.

	if (can_irq_wake(irq))
		device_set_wakeup_capable(&client->dev, 1);

But there is no can_irq_wake() call exist, and it is not that trivial to
implement it for all interrupts controllers and complex/cascaded setups.

drivers/base/power/sysfs.c also covers these cases:

 * Devices may not be able to generate wakeup events from all power
 * states.  Also, the events may be ignored in some configurations;
 * for example, they might need help from other devices that aren't
 * active

So there is no guarantee that wakeup will actually work, and so I think
there is no point in being pedantic wrt checking IRQ wakeup capability.
Signed-off-by: default avatarAnton Vorontsov <avorontsov@ru.mvista.com>
Cc: David Brownell <dbrownell@users.sourceforge.net>
Cc: Ben Dooks <ben-linux@fluff.org>
Cc: Jean Delvare <khali@linux-fr.org>
Cc: Alessandro Zummo <a.zummo@towertech.it>
Signed-off-by: default avatarAndrew Morton <akpm@linux-foundation.org>
      0d0b9e35
  59. 10 Nov, 2009 1 commit
  61. 04 Nov, 2009 1 commit
  63. 16 Oct, 2009 2 commits
    • Atsushi Nemoto's avatar
      The to_platform_device macro itself uses container_of macro. Nested use · 94e2f167
      Atsushi Nemoto authored 
      of container_of macro causes following sparse warnings:

rtc-ds1553.c:259:3: warning: symbol '__mptr' shadows an earlier one
rtc-ds1553.c:259:3: originally declared here
Signed-off-by: default avatarAtsushi Nemoto <anemo@mba.ocn.ne.jp>
Cc: Alessandro Zummo <alessandro.zummo@towertech.it>
Cc: David Brownell <david-b@pacbell.net>
Cc: Andrew Sharp <andy.sharp@lsi.com>
Cc: Thomas Hommel <thomas.hommel@gefanuc.com>
Signed-off-by: default avatarAndrew Morton <akpm@linux-foundation.org>
      94e2f167
    • Atsushi Nemoto's avatar
      - Call dev_set_drvdata before rtc device creation. · 41684cba
      Atsushi Nemoto authored 
      - Use its own spinlock instead of rtc->irq_lock.  Because pdata->rtc
  must be initialized to use the irq_lock (pdata->rtc->irq_lock).  There
  is a small window which rtc methods can be called before pdata->rtc is
  initialized.

  And there is no need use the irq_lock to protect hardware registers.
  The driver's own spinlock shoule be enough.

- Check pdata->rtc before calling rtc_update_irq.

- Use alarm_irq_enable and remove ioctl routine.

- Use devres APIs and simplify error/remove path.

These fixes are ported from ds1553 driver and just compile-tested only.
Signed-off-by: default avatarAtsushi Nemoto <anemo@mba.ocn.ne.jp>
Cc: Alessandro Zummo <alessandro.zummo@towertech.it>
Cc: Thomas Hommel <thomas.hommel@gefanuc.com>
Cc: David Brownell <david-b@pacbell.net>
Signed-off-by: default avatarAndrew Morton <akpm@linux-foundation.org>
      41684cba