platform/adaptation/renesas_rcar/renesas_kernel.git
13 years agothp: mm: define MADV_HUGEPAGE
Andrea Arcangeli [Thu, 13 Jan 2011 23:46:31 +0000 (15:46 -0800)]
thp: mm: define MADV_HUGEPAGE

Define MADV_HUGEPAGE.

Signed-off-by: Andrea Arcangeli <aarcange@redhat.com>
Acked-by: Rik van Riel <riel@redhat.com>
Acked-by: Arnd Bergmann <arnd@arndb.de>
Acked-by: Mel Gorman <mel@csn.ul.ie>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
13 years agothp: transparent hugepage support documentation
Andrea Arcangeli [Thu, 13 Jan 2011 23:46:30 +0000 (15:46 -0800)]
thp: transparent hugepage support documentation

Documentation/vm/transhuge.txt

Signed-off-by: Andrea Arcangeli <aarcange@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
13 years agothp: fix bad_page to show the real reason the page is bad
Andrea Arcangeli [Thu, 13 Jan 2011 23:46:29 +0000 (15:46 -0800)]
thp: fix bad_page to show the real reason the page is bad

page_count shows the count of the head page, but the actual check is done
on the tail page, so show what is really being checked.

Signed-off-by: Andrea Arcangeli <aarcange@redhat.com>
Acked-by: Rik van Riel <riel@redhat.com>
Acked-by: Mel Gorman <mel@csn.ul.ie>
Reviewed-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
13 years agothp: ksm: free swap when swapcache page is replaced
Hugh Dickins [Thu, 13 Jan 2011 23:46:28 +0000 (15:46 -0800)]
thp: ksm: free swap when swapcache page is replaced

When a swapcache page is replaced by a ksm page, it's best to free that
swap immediately.

Reported-by: Andrea Arcangeli <aarcange@redhat.com>
Signed-off-by: Hugh Dickins <hughd@google.com>
Signed-off-by: Andrea Arcangeli <aarcange@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
13 years agowriteback: avoid unnecessary determine_dirtyable_memory call
Minchan Kim [Thu, 13 Jan 2011 23:46:27 +0000 (15:46 -0800)]
writeback: avoid unnecessary determine_dirtyable_memory call

I think determine_dirtyable_memory() is a rather costly function since it
need many atomic reads for gathering zone/global page state.  But when we
use vm_dirty_bytes && dirty_background_bytes, we don't need that costly
calculation.

This patch eliminates such unnecessary overhead.

NOTE : newly added if condition might add overhead in normal path.
       But it should be _really_ small because anyway we need the
       access both vm_dirty_bytes and dirty_background_bytes so it is
       likely to hit the cache.

[akpm@linux-foundation.org: fix used-uninitialised warning]
Signed-off-by: Minchan Kim <minchan.kim@gmail.com>
Cc: Wu Fengguang <fengguang.wu@intel.com>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
13 years agomm: set correct numa_zonelist_order string when configured on the kernel command...
Volodymyr G. Lukiianyk [Thu, 13 Jan 2011 23:46:26 +0000 (15:46 -0800)]
mm: set correct numa_zonelist_order string when configured on the kernel command line

When numa_zonelist_order parameter is set to "node" or "zone" on the
command line it's still showing as "default" in sysctl.  That's because
early_param parsing function changes only user_zonelist_order variable.
Fix this by copying user-provided string to numa_zonelist_order if it was
successfully parsed.

Signed-off-by: Volodymyr G Lukiianyk <volodymyrgl@gmail.com>
Acked-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>
13 years agomm: kswapd: use the classzone idx that kswapd was using for sleeping_prematurely()
Mel Gorman [Thu, 13 Jan 2011 23:46:26 +0000 (15:46 -0800)]
mm: kswapd: use the classzone idx that kswapd was using for sleeping_prematurely()

When kswapd is woken up for a high-order allocation, it takes account of
the highest usable zone by the caller (the classzone idx).  During
allocation, this index is used to select the lowmem_reserve[] that should
be applied to the watermark calculation in zone_watermark_ok().

When balancing a node, kswapd considers the highest unbalanced zone to be
the classzone index.  This will always be at least be the callers
classzone_idx and can be higher.  However, sleeping_prematurely() always
considers the lowest zone (e.g.  ZONE_DMA) to be the classzone index.
This means that sleeping_prematurely() can consider a zone to be balanced
that is unusable by the allocation request that originally woke kswapd.
This patch changes sleeping_prematurely() to use a classzone_idx matching
the value it used in balance_pgdat().

Signed-off-by: Mel Gorman <mel@csn.ul.ie>
Reviewed-by: Minchan Kim <minchan.kim@gmail.com>
Reviewed-by: Eric B Munson <emunson@mgebm.net>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Simon Kirby <sim@hostway.ca>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Shaohua Li <shaohua.li@intel.com>
Cc: Dave Hansen <dave@linux.vnet.ibm.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Rik van Riel <riel@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
13 years agomm: kswapd: treat zone->all_unreclaimable in sleeping_prematurely similar to balance_...
Mel Gorman [Thu, 13 Jan 2011 23:46:24 +0000 (15:46 -0800)]
mm: kswapd: treat zone->all_unreclaimable in sleeping_prematurely similar to balance_pgdat()

After DEF_PRIORITY, balance_pgdat() considers all_unreclaimable zones to
be balanced but sleeping_prematurely does not.  This can force kswapd to
stay awake longer than it should.  This patch fixes it.

Signed-off-by: Mel Gorman <mel@csn.ul.ie>
Reviewed-by: Eric B Munson <emunson@mgebm.net>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Simon Kirby <sim@hostway.ca>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Shaohua Li <shaohua.li@intel.com>
Cc: Dave Hansen <dave@linux.vnet.ibm.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Rik van Riel <riel@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
13 years agomm: kswapd: reset kswapd_max_order and classzone_idx after reading
Mel Gorman [Thu, 13 Jan 2011 23:46:23 +0000 (15:46 -0800)]
mm: kswapd: reset kswapd_max_order and classzone_idx after reading

When kswapd wakes up, it reads its order and classzone from pgdat and
calls balance_pgdat.  While its awake, it potentially reclaimes at a high
order and a low classzone index.  This might have been a once-off that was
not required by subsequent callers.  However, because the pgdat values
were not reset, they remain artifically high while balance_pgdat() is
running and potentially kswapd enters a second unnecessary reclaim cycle.
Reset the pgdat order and classzone index after reading.

Signed-off-by: Mel Gorman <mel@csn.ul.ie>
Reviewed-by: Minchan Kim <minchan.kim@gmail.com>
Reviewed-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Reviewed-by: Eric B Munson <emunson@mgebm.net>
Cc: Simon Kirby <sim@hostway.ca>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Shaohua Li <shaohua.li@intel.com>
Cc: Dave Hansen <dave@linux.vnet.ibm.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Rik van Riel <riel@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
13 years agomm: kswapd: use the order that kswapd was reclaiming at for sleeping_prematurely()
Mel Gorman [Thu, 13 Jan 2011 23:46:22 +0000 (15:46 -0800)]
mm: kswapd: use the order that kswapd was reclaiming at for sleeping_prematurely()

Before kswapd goes to sleep, it uses sleeping_prematurely() to check if
there was a race pushing a zone below its watermark.  If the race
happened, it stays awake.  However, balance_pgdat() can decide to reclaim
at order-0 if it decides that high-order reclaim is not working as
expected.  This information is not passed back to sleeping_prematurely().
The impact is that kswapd remains awake reclaiming pages long after it
should have gone to sleep.  This patch passes the adjusted order to
sleeping_prematurely and uses the same logic as balance_pgdat to decide if
it's ok to go to sleep.

Signed-off-by: Mel Gorman <mel@csn.ul.ie>
Reviewed-by: Minchan Kim <minchan.kim@gmail.com>
Reviewed-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Reviewed-by: Eric B Munson <emunson@mgebm.net>
Cc: Simon Kirby <sim@hostway.ca>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Shaohua Li <shaohua.li@intel.com>
Cc: Dave Hansen <dave@linux.vnet.ibm.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Rik van Riel <riel@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
13 years agomm: kswapd: keep kswapd awake for high-order allocations until a percentage of the...
Mel Gorman [Thu, 13 Jan 2011 23:46:21 +0000 (15:46 -0800)]
mm: kswapd: keep kswapd awake for high-order allocations until a percentage of the node is balanced

When reclaiming for high-orders, kswapd is responsible for balancing a
node but it should not reclaim excessively.  It avoids excessive reclaim
by considering if any zone in a node is balanced then the node is
balanced.  In the cases where there are imbalanced zone sizes (e.g.
ZONE_DMA with both ZONE_DMA32 and ZONE_NORMAL), kswapd can go to sleep
prematurely as just one small zone was balanced.

This alters the sleep logic of kswapd slightly.  It counts the number of
pages that make up the balanced zones.  If the total number of balanced
pages is more than a quarter of the zone, kswapd will go back to sleep.
This should keep a node balanced without reclaiming an excessive number of
pages.

Signed-off-by: Mel Gorman <mel@csn.ul.ie>
Reviewed-by: Minchan Kim <minchan.kim@gmail.com>
Reviewed-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Reviewed-by: Eric B Munson <emunson@mgebm.net>
Cc: Simon Kirby <sim@hostway.ca>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Shaohua Li <shaohua.li@intel.com>
Cc: Dave Hansen <dave@linux.vnet.ibm.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Rik van Riel <riel@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
13 years agomm: kswapd: stop high-order balancing when any suitable zone is balanced
Mel Gorman [Thu, 13 Jan 2011 23:46:20 +0000 (15:46 -0800)]
mm: kswapd: stop high-order balancing when any suitable zone is balanced

Simon Kirby reported the following problem

   We're seeing cases on a number of servers where cache never fully
   grows to use all available memory.  Sometimes we see servers with 4 GB
   of memory that never seem to have less than 1.5 GB free, even with a
   constantly-active VM.  In some cases, these servers also swap out while
   this happens, even though they are constantly reading the working set
   into memory.  We have been seeing this happening for a long time; I
   don't think it's anything recent, and it still happens on 2.6.36.

After some debugging work by Simon, Dave Hansen and others, the prevaling
theory became that kswapd is reclaiming order-3 pages requested by SLUB
too aggressive about it.

There are two apparent problems here.  On the target machine, there is a
small Normal zone in comparison to DMA32.  As kswapd tries to balance all
zones, it would continually try reclaiming for Normal even though DMA32
was balanced enough for callers.  The second problem is that
sleeping_prematurely() does not use the same logic as balance_pgdat() when
deciding whether to sleep or not.  This keeps kswapd artifically awake.

A number of tests were run and the figures from previous postings will
look very different for a few reasons.  One, the old figures were forcing
my network card to use GFP_ATOMIC in attempt to replicate Simon's problem.
 Second, I previous specified slub_min_order=3 again in an attempt to
reproduce Simon's problem.  In this posting, I'm depending on Simon to say
whether his problem is fixed or not and these figures are to show the
impact to the ordinary cases.  Finally, the "vmscan" figures are taken
from /proc/vmstat instead of the tracepoints.  There is less information
but recording is less disruptive.

The first test of relevance was postmark with a process running in the
background reading a large amount of anonymous memory in blocks.  The
objective was to vaguely simulate what was happening on Simon's machine
and it's memory intensive enough to have kswapd awake.

POSTMARK
                                            traceonly          kanyzone
Transactions per second:              156.00 ( 0.00%)   153.00 (-1.96%)
Data megabytes read per second:        21.51 ( 0.00%)    21.52 ( 0.05%)
Data megabytes written per second:     29.28 ( 0.00%)    29.11 (-0.58%)
Files created alone per second:       250.00 ( 0.00%)   416.00 (39.90%)
Files create/transact per second:      79.00 ( 0.00%)    76.00 (-3.95%)
Files deleted alone per second:       520.00 ( 0.00%)   420.00 (-23.81%)
Files delete/transact per second:      79.00 ( 0.00%)    76.00 (-3.95%)

MMTests Statistics: duration
User/Sys Time Running Test (seconds)         16.58      17.4
Total Elapsed Time (seconds)                218.48    222.47

VMstat Reclaim Statistics: vmscan
Direct reclaims                                  0          4
Direct reclaim pages scanned                     0        203
Direct reclaim pages reclaimed                   0        184
Kswapd pages scanned                        326631     322018
Kswapd pages reclaimed                      312632     309784
Kswapd low wmark quickly                         1          4
Kswapd high wmark quickly                      122        475
Kswapd skip congestion_wait                      1          0
Pages activated                             700040     705317
Pages deactivated                           212113     203922
Pages written                                 9875       6363

Total pages scanned                         326631    322221
Total pages reclaimed                       312632    309968
%age total pages scanned/reclaimed          95.71%    96.20%
%age total pages scanned/written             3.02%     1.97%

proc vmstat: Faults
Major Faults                                   300       254
Minor Faults                                645183    660284
Page ins                                    493588    486704
Page outs                                  4960088   4986704
Swap ins                                      1230       661
Swap outs                                     9869      6355

Performance is mildly affected because kswapd is no longer doing as much
work and the background memory consumer process is getting in the way.
Note that kswapd scanned and reclaimed fewer pages as it's less aggressive
and overall fewer pages were scanned and reclaimed.  Swap in/out is
particularly reduced again reflecting kswapd throwing out fewer pages.

The slight performance impact is unfortunate here but it looks like a
direct result of kswapd being less aggressive.  As the bug report is about
too many pages being freed by kswapd, it may have to be accepted for now.

The second test is a streaming IO benchmark that was previously used by
Johannes to show regressions in page reclaim.

MICRO
 traceonly  kanyzone
User/Sys Time Running Test (seconds)         29.29     28.87
Total Elapsed Time (seconds)                492.18    488.79

VMstat Reclaim Statistics: vmscan
Direct reclaims                               2128       1460
Direct reclaim pages scanned               2284822    1496067
Direct reclaim pages reclaimed              148919     110937
Kswapd pages scanned                      15450014   16202876
Kswapd pages reclaimed                     8503697    8537897
Kswapd low wmark quickly                      3100       3397
Kswapd high wmark quickly                     1860       7243
Kswapd skip congestion_wait                    708        801
Pages activated                               9635       9573
Pages deactivated                             1432       1271
Pages written                                  223       1130

Total pages scanned                       17734836  17698943
Total pages reclaimed                      8652616   8648834
%age total pages scanned/reclaimed          48.79%    48.87%
%age total pages scanned/written             0.00%     0.01%

proc vmstat: Faults
Major Faults                                   165       221
Minor Faults                               9655785   9656506
Page ins                                      3880      7228
Page outs                                 37692940  37480076
Swap ins                                         0        69
Swap outs                                       19        15

Again fewer pages are scanned and reclaimed as expected and this time the
test completed faster.  Note that kswapd is hitting its watermarks faster
(low and high wmark quickly) which I expect is due to kswapd reclaiming
fewer pages.

I also ran fs-mark, iozone and sysbench but there is nothing interesting
to report in the figures.  Performance is not significantly changed and
the reclaim statistics look reasonable.

Tgis patch:

When the allocator enters its slow path, kswapd is woken up to balance the
node.  It continues working until all zones within the node are balanced.
For order-0 allocations, this makes perfect sense but for higher orders it
can have unintended side-effects.  If the zone sizes are imbalanced,
kswapd may reclaim heavily within a smaller zone discarding an excessive
number of pages.  The user-visible behaviour is that kswapd is awake and
reclaiming even though plenty of pages are free from a suitable zone.

This patch alters the "balance" logic for high-order reclaim allowing
kswapd to stop if any suitable zone becomes balanced to reduce the number
of pages it reclaims from other zones.  kswapd still tries to ensure that
order-0 watermarks for all zones are met before sleeping.

Signed-off-by: Mel Gorman <mel@csn.ul.ie>
Reviewed-by: Minchan Kim <minchan.kim@gmail.com>
Reviewed-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Reviewed-by: Eric B Munson <emunson@mgebm.net>
Cc: Simon Kirby <sim@hostway.ca>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Shaohua Li <shaohua.li@intel.com>
Cc: Dave Hansen <dave@linux.vnet.ibm.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Rik van Riel <riel@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
13 years agomm: remove likely() from grab_cache_page_write_begin()
Steven Rostedt [Thu, 13 Jan 2011 23:46:18 +0000 (15:46 -0800)]
mm: remove likely() from grab_cache_page_write_begin()

Running the annotated branch profiler on a box doing average work
(firefox, evolution, xchat, distcc farm), the likely() used in
grab_cache_page_write_begin() was incorrect most of the time:

 correct incorrect  %        Function                  File              Line
 ------- ---------  -        --------                  ----              ----
 1924262 71332401  97 grab_cache_page_write_begin    filemap.c           2206

Adding a trace_printk() and running the function tracer limited to
just this function I can see:

        gconfd-2-2696  [000]  4467.268935: grab_cache_page_write_begin: page=          (null) mapping=ffff8800676a9460 index=7
        gconfd-2-2696  [000]  4467.268946: grab_cache_page_write_begin <-ext3_write_begin
        gconfd-2-2696  [000]  4467.268947: grab_cache_page_write_begin: page=          (null) mapping=ffff8800676a9460 index=8
        gconfd-2-2696  [000]  4467.268959: grab_cache_page_write_begin <-ext3_write_begin
        gconfd-2-2696  [000]  4467.268960: grab_cache_page_write_begin: page=          (null) mapping=ffff8800676a9460 index=9
        gconfd-2-2696  [000]  4467.268972: grab_cache_page_write_begin <-ext3_write_begin
        gconfd-2-2696  [000]  4467.268973: grab_cache_page_write_begin: page=          (null) mapping=ffff8800676a9460 index=10
        gconfd-2-2696  [000]  4467.268991: grab_cache_page_write_begin <-ext3_write_begin
        gconfd-2-2696  [000]  4467.268992: grab_cache_page_write_begin: page=          (null) mapping=ffff8800676a9460 index=11
        gconfd-2-2696  [000]  4467.269005: grab_cache_page_write_begin <-ext3_write_begin

Which shows that a lot of calls from ext3_write_begin will result in the
page returned by "find_lock_page" will be NULL.

Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Acked-by: Nick Piggin <npiggin@kernel.dk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
13 years agomm: remove unlikely() from page_mapping()
Steven Rostedt [Thu, 13 Jan 2011 23:46:17 +0000 (15:46 -0800)]
mm: remove unlikely() from page_mapping()

page_mapping() has a unlikely that the mapping has PAGE_MAPPING_ANON set.
But running the annotated branch profiler on a normal desktop system doing
vairous tasks (xchat, evolution, firefox, distcc), it is not really that
unlikely that the mapping here will have the PAGE_MAPPING_ANON flag set:

 correct incorrect  %        Function                  File              Line
 ------- ---------  -        --------                  ----              ----
35935762 1270265395  97 page_mapping                   mm.h                 659
1306198001   143659   0 page_mapping                   mm.h                 657
203131478   121586   0 page_mapping                   mm.h                 657
 5415491     1116   0 page_mapping                   mm.h                 657
74899487     1116   0 page_mapping                   mm.h                 657
203132845      224   0 page_mapping                   mm.h                 659
 5415464       27   0 page_mapping                   mm.h                 659
   13552        0   0 page_mapping                   mm.h                 657
   13552        0   0 page_mapping                   mm.h                 659
  242630        0   0 page_mapping                   mm.h                 657
  242630        0   0 page_mapping                   mm.h                 659
74899487        0   0 page_mapping                   mm.h                 659

The page_mapping() is a static inline, which is why it shows up multiple
times.

The unlikely in page_mapping() was correct a total of 1909540379 times and
incorrect 1270533123 times, with a 39% being incorrect.  With this much of
an error, it's best to simply remove the unlikely and have the compiler
and branch prediction figure this out.

Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Nick Piggin <npiggin@kernel.dk>
Cc: Rik van Riel <riel@redhat.com>
Cc: Lee Schermerhorn <Lee.Schermerhorn@hp.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
13 years agomm: remove likely() from mapping_unevictable()
Steven Rostedt [Thu, 13 Jan 2011 23:46:16 +0000 (15:46 -0800)]
mm: remove likely() from mapping_unevictable()

The mapping_unevictable() has a likely() around the mapping parameter.
This mapping parameter comes from page_mapping() which has an unlikely()
that the page will be set as PAGE_MAPPING_ANON, and if so, it will return
NULL.  One would think that this unlikely() means that the mapping
returned by page_mapping() would not be NULL, but where page_mapping() is
used just above mapping_unevictable(), that unlikely() is incorrect most
of the time.  This means that the "likely(mapping)" in
mapping_unevictable() is incorrect most of the time.

Running the annotated branch profiler on my main box which runs firefox,
evolution, xchat and is part of my distcc farm, I had this:

 correct incorrect  %        Function                  File              Line
 ------- ---------  -        --------                  ----              ----
12872836 1269443893  98 mapping_unevictable            pagemap.h            51
35935762 1270265395  97 page_mapping                   mm.h                 659
1306198001   143659   0 page_mapping                   mm.h                 657
203131478   121586   0 page_mapping                   mm.h                 657
 5415491     1116   0 page_mapping                   mm.h                 657
74899487     1116   0 page_mapping                   mm.h                 657
203132845      224   0 page_mapping                   mm.h                 659
 5415464       27   0 page_mapping                   mm.h                 659
   13552        0   0 page_mapping                   mm.h                 657
   13552        0   0 page_mapping                   mm.h                 659
  242630        0   0 page_mapping                   mm.h                 657
  242630        0   0 page_mapping                   mm.h                 659
74899487        0   0 page_mapping                   mm.h                 659

The page_mapping() is a static inline, which is why it shows up multiple
times.  The mapping_unevictable() is also a static inline but seems to be
used only once in my setup.

The unlikely in page_mapping() was correct a total of 1909540379 times and
incorrect 1270533123 times, with a 39% being incorrect.  Perhaps this is
enough to remove the unlikely from page_mapping() as well.

Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Reviewed-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Acked-by: Nick Piggin <npiggin@kernel.dk>
Acked-by: Rik van Riel <riel@redhat.com>
Cc: Lee Schermerhorn <Lee.Schermerhorn@hp.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
13 years agovmalloc: remove redundant unlikely()
Tobias Klauser [Thu, 13 Jan 2011 23:46:15 +0000 (15:46 -0800)]
vmalloc: remove redundant unlikely()

IS_ERR() already implies unlikely(), so it can be omitted here.

Signed-off-by: Tobias Klauser <tklauser@distanz.ch>
Reviewed-by: Minchan Kim <minchan.kim@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
13 years agomempolicy: remove tasklist_lock from migrate_pages
KOSAKI Motohiro [Thu, 13 Jan 2011 23:46:14 +0000 (15:46 -0800)]
mempolicy: remove tasklist_lock from migrate_pages

Today, tasklist_lock in migrate_pages doesn't protect anything.
rcu_read_lock() provide enough protection from pid hash walk.

Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Reported-by: Peter Zijlstra <peterz@infradead.org>
Cc: Christoph Lameter <cl@linux-foundation.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
13 years agomlock: do not hold mmap_sem for extended periods of time
Michel Lespinasse [Thu, 13 Jan 2011 23:46:14 +0000 (15:46 -0800)]
mlock: do not hold mmap_sem for extended periods of time

__get_user_pages gets a new 'nonblocking' parameter to signal that the
caller is prepared to re-acquire mmap_sem and retry the operation if
needed.  This is used to split off long operations if they are going to
block on a disk transfer, or when we detect contention on the mmap_sem.

[akpm@linux-foundation.org: remove ref to rwsem_is_contended()]
Signed-off-by: Michel Lespinasse <walken@google.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Nick Piggin <npiggin@kernel.dk>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: David Howells <dhowells@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
13 years agomm: move VM_LOCKED check to __mlock_vma_pages_range()
Michel Lespinasse [Thu, 13 Jan 2011 23:46:12 +0000 (15:46 -0800)]
mm: move VM_LOCKED check to __mlock_vma_pages_range()

Use a single code path for faulting in pages during mlock.

The reason to have it in this patch series is that I did not want to
update both code paths in a later change that releases mmap_sem when
blocking on disk.

Signed-off-by: Michel Lespinasse <walken@google.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Nick Piggin <npiggin@kernel.dk>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: David Howells <dhowells@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
13 years agomm: add FOLL_MLOCK follow_page flag.
Michel Lespinasse [Thu, 13 Jan 2011 23:46:11 +0000 (15:46 -0800)]
mm: add FOLL_MLOCK follow_page flag.

Move the code to mlock pages from __mlock_vma_pages_range() to
follow_page().

This allows __mlock_vma_pages_range() to not have to break down work into
16-page batches.

An additional motivation for doing this within the present patch series is
that it'll make it easier for a later chagne to drop mmap_sem when
blocking on disk (we'd like to be able to resume at the page that was read
from disk instead of at the start of a 16-page batch).

Signed-off-by: Michel Lespinasse <walken@google.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Nick Piggin <npiggin@kernel.dk>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: David Howells <dhowells@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
13 years agomlock: only hold mmap_sem in shared mode when faulting in pages
Michel Lespinasse [Thu, 13 Jan 2011 23:46:10 +0000 (15:46 -0800)]
mlock: only hold mmap_sem in shared mode when faulting in pages

Currently mlock() holds mmap_sem in exclusive mode while the pages get
faulted in.  In the case of a large mlock, this can potentially take a
very long time, during which various commands such as 'ps auxw' will
block.  This makes sysadmins unhappy:

real    14m36.232s
user    0m0.003s
sys     0m0.015s
(output from 'time ps auxw' while a 20GB file was being mlocked without
being previously preloaded into page cache)

I propose that mlock() could release mmap_sem after the VM_LOCKED bits
have been set in all appropriate VMAs.  Then a second pass could be done
to actually mlock the pages, in small batches, releasing mmap_sem when we
block on disk access or when we detect some contention.

This patch:

Before this change, mlock() holds mmap_sem in exclusive mode while the
pages get faulted in.  In the case of a large mlock, this can potentially
take a very long time.  Various things will block while mmap_sem is held,
including 'ps auxw'.  This can make sysadmins angry.

I propose that mlock() could release mmap_sem after the VM_LOCKED bits
have been set in all appropriate VMAs.  Then a second pass could be done
to actually mlock the pages with mmap_sem held for reads only.  We need to
recheck the vma flags after we re-acquire mmap_sem, but this is easy.

In the case where a vma has been munlocked before mlock completes, pages
that were already marked as PageMlocked() are handled by the munlock()
call, and mlock() is careful to not mark new page batches as PageMlocked()
after the munlock() call has cleared the VM_LOCKED vma flags.  So, the end
result will be identical to what'd happen if munlock() had executed after
the mlock() call.

In a later change, I will allow the second pass to release mmap_sem when
blocking on disk accesses or when it is otherwise contended, so that it
won't be held for long periods of time even in shared mode.

Signed-off-by: Michel Lespinasse <walken@google.com>
Tested-by: Valdis Kletnieks <Valdis.Kletnieks@vt.edu>
Cc: Hugh Dickins <hughd@google.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Nick Piggin <npiggin@kernel.dk>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: David Howells <dhowells@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
13 years agomlock: avoid dirtying pages and triggering writeback
Michel Lespinasse [Thu, 13 Jan 2011 23:46:09 +0000 (15:46 -0800)]
mlock: avoid dirtying pages and triggering writeback

When faulting in pages for mlock(), we want to break COW for anonymous or
file pages within VM_WRITABLE, non-VM_SHARED vmas.  However, there is no
need to write-fault into VM_SHARED vmas since shared file pages can be
mlocked first and dirtied later, when/if they actually get written to.
Skipping the write fault is desirable, as we don't want to unnecessarily
cause these pages to be dirtied and queued for writeback.

Signed-off-by: Michel Lespinasse <walken@google.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Kosaki Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Nick Piggin <npiggin@kernel.dk>
Cc: Theodore Tso <tytso@google.com>
Cc: Michael Rubin <mrubin@google.com>
Cc: Suleiman Souhlal <suleiman@google.com>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Christoph Hellwig <hch@infradead.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
13 years agodo_wp_page: clarify dirty_page handling
Michel Lespinasse [Thu, 13 Jan 2011 23:46:08 +0000 (15:46 -0800)]
do_wp_page: clarify dirty_page handling

Reorganize the code so that dirty pages are handled closer to the place
that makes them dirty (handling write fault into shared, writable VMAs).
No behavior changes.

Signed-off-by: Michel Lespinasse <walken@google.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Kosaki Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Nick Piggin <npiggin@kernel.dk>
Cc: Theodore Tso <tytso@google.com>
Cc: Michael Rubin <mrubin@google.com>
Cc: Suleiman Souhlal <suleiman@google.com>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Christoph Hellwig <hch@infradead.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
13 years agodo_wp_page: remove the 'reuse' flag
Michel Lespinasse [Thu, 13 Jan 2011 23:46:07 +0000 (15:46 -0800)]
do_wp_page: remove the 'reuse' flag

mlocking a shared, writable vma currently causes the corresponding pages
to be marked as dirty and queued for writeback.  This seems rather
unnecessary given that the pages are not being actually modified during
mlock.  It is understood that for non-shared mappings (file or anon) we
want to use a write fault in order to break COW, but there is just no such
need for shared mappings.

The first two patches in this series do not introduce any behavior change.
 The intent there is to make it obvious that dirtying file pages is only
done in the (writable, shared) case.  I think this clarifies the code, but
I wouldn't mind dropping these two patches if there is no consensus about
them.

The last patch is where we actually avoid dirtying shared mappings during
mlock.  Note that as a side effect of this, we won't call page_mkwrite()
for the mappings that define it, and won't be pre-allocating data blocks
at the FS level if the mapped file was sparsely allocated.  My
understanding is that mlock does not need to provide such guarantee, as
evidenced by the fact that it never did for the filesystems that don't
define page_mkwrite() - including some common ones like ext3.  However, I
would like to gather feedback on this from filesystem people as a
precaution.  If this turns out to be a showstopper, maybe block
preallocation can be added back on using a different interface.

Large shared mlocks are getting significantly (>2x) faster in my tests, as
the disk can be fully used for reading the file instead of having to share
between this and writeback.

This patch:

Reorganize the code to remove the 'reuse' flag.  No behavior changes.

Signed-off-by: Michel Lespinasse <walken@google.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Kosaki Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Nick Piggin <npiggin@kernel.dk>
Cc: Theodore Tso <tytso@google.com>
Cc: Michael Rubin <mrubin@google.com>
Cc: Suleiman Souhlal <suleiman@google.com>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Christoph Hellwig <hch@infradead.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
13 years agomm: clear PageError bit in msync & fsync
Rik van Riel [Thu, 13 Jan 2011 23:46:06 +0000 (15:46 -0800)]
mm: clear PageError bit in msync & fsync

Temporary IO failures, eg.  due to loss of both multipath paths, can
permanently leave the PageError bit set on a page, resulting in msync or
fsync returning -EIO over and over again, even if IO is now getting to the
disk correctly.

We already clear the AS_ENOSPC and AS_IO bits in mapping->flags in the
filemap_fdatawait_range function.  Also clearing the PageError bit on the
page allows subsequent msync or fsync calls on this file to return without
an error, if the subsequent IO succeeds.

Unfortunately data written out in the msync or fsync call that returned
-EIO can still get lost, because the page dirty bit appears to not get
restored on IO error.  However, the alternative could be potentially all
of memory filling up with uncleanable dirty pages, hanging the system, so
there is no nice choice here...

Signed-off-by: Rik van Riel <riel@redhat.com>
Acked-by: Valerie Aurora <vaurora@redhat.com>
Acked-by: Jeff Layton <jlayton@redhat.com>
Cc: Theodore Ts'o <tytso@mit.edu>
Acked-by: Jan Kara <jack@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
13 years agooom: allow a non-CAP_SYS_RESOURCE proces to oom_score_adj down
Mandeep Singh Baines [Thu, 13 Jan 2011 23:46:05 +0000 (15:46 -0800)]
oom: allow a non-CAP_SYS_RESOURCE proces to oom_score_adj down

We'd like to be able to oom_score_adj a process up/down as it
enters/leaves the foreground.  Currently, it is not possible to oom_adj
down without CAP_SYS_RESOURCE.  This patch allows a task to decrease its
oom_score_adj back to the value that a CAP_SYS_RESOURCE thread set it to
or its inherited value at fork.  Assuming the thread that has forked it
has oom_score_adj of 0, each process could decrease it back from 0 upon
activation unless a CAP_SYS_RESOURCE thread elevated it to something
higher.

Alternative considered:

* a setuid binary
* a daemon with CAP_SYS_RESOURCE

Since you don't wan't all processes to be able to reduce their oom_adj, a
setuid or daemon implementation would be complex.  The alternatives also
have much higher overhead.

This patch updated from original patch based on feedback from David
Rientjes.

Signed-off-by: Mandeep Singh Baines <msb@chromium.org>
Acked-by: David Rientjes <rientjes@google.com>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Ying Han <yinghan@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
13 years agomm: unify module_alloc code for vmalloc
David Rientjes [Thu, 13 Jan 2011 23:46:02 +0000 (15:46 -0800)]
mm: unify module_alloc code for vmalloc

Four architectures (arm, mips, sparc, x86) use __vmalloc_area() for
module_init().  Much of the code is duplicated and can be generalized in a
globally accessible function, __vmalloc_node_range().

__vmalloc_node() now calls into __vmalloc_node_range() with a range of
[VMALLOC_START, VMALLOC_END) for functionally equivalent behavior.

Each architecture may then use __vmalloc_node_range() directly to remove
the duplication of code.

Signed-off-by: David Rientjes <rientjes@google.com>
Cc: Christoph Lameter <cl@linux.com>
Cc: Russell King <linux@arm.linux.org.uk>
Cc: Ralf Baechle <ralf@linux-mips.org>
Cc: "David S. Miller" <davem@davemloft.net>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: "H. Peter Anvin" <hpa@zytor.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>
13 years agomm: remove gfp mask from pcpu_get_vm_areas
David Rientjes [Thu, 13 Jan 2011 23:46:01 +0000 (15:46 -0800)]
mm: remove gfp mask from pcpu_get_vm_areas

pcpu_get_vm_areas() only uses GFP_KERNEL allocations, so remove the gfp_t
formal and use the mask internally.

Signed-off-by: David Rientjes <rientjes@google.com>
Cc: Christoph Lameter <cl@linux.com>
Cc: Tejun Heo <tj@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
13 years agomm: remove unused get_vm_area_node
David Rientjes [Thu, 13 Jan 2011 23:46:00 +0000 (15:46 -0800)]
mm: remove unused get_vm_area_node

get_vm_area_node() is unused in the kernel and can thus be removed.

Signed-off-by: David Rientjes <rientjes@google.com>
Cc: Christoph Lameter <cl@linux.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
13 years agomm: vmscan: rename lumpy_mode to reclaim_mode
Mel Gorman [Thu, 13 Jan 2011 23:46:00 +0000 (15:46 -0800)]
mm: vmscan: rename lumpy_mode to reclaim_mode

With compaction being used instead of lumpy reclaim, the name lumpy_mode
and associated variables is a bit misleading.  Rename lumpy_mode to
reclaim_mode which is a better fit.  There is no functional change.

Signed-off-by: Mel Gorman <mel@csn.ul.ie>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Rik van Riel <riel@redhat.com>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Cc: Andy Whitcroft <apw@shadowen.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
13 years agomm: compaction: perform a faster migration scan when migrating asynchronously
Mel Gorman [Thu, 13 Jan 2011 23:45:59 +0000 (15:45 -0800)]
mm: compaction: perform a faster migration scan when migrating asynchronously

try_to_compact_pages() is initially called to only migrate pages
asychronously and kswapd always compacts asynchronously.  Both are being
optimistic so it is important to complete the work as quickly as possible
to minimise stalls.

This patch alters the scanner when asynchronous to only consider
MIGRATE_MOVABLE pageblocks as migration candidates.  This reduces stalls
when allocating huge pages while not impairing allocation success rates as
a full scan will be performed if necessary after direct reclaim.

Signed-off-by: Mel Gorman <mel@csn.ul.ie>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Rik van Riel <riel@redhat.com>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Cc: Andy Whitcroft <apw@shadowen.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
13 years agomm: migration: cleanup migrate_pages API by matching types for offlining and sync
Mel Gorman [Thu, 13 Jan 2011 23:45:58 +0000 (15:45 -0800)]
mm: migration: cleanup migrate_pages API by matching types for offlining and sync

With the introduction of the boolean sync parameter, the API looks a
little inconsistent as offlining is still an int.  Convert offlining to a
bool for the sake of being tidy.

Signed-off-by: Mel Gorman <mel@csn.ul.ie>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Rik van Riel <riel@redhat.com>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Cc: Andy Whitcroft <apw@shadowen.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
13 years agomm: migration: allow migration to operate asynchronously and avoid synchronous compac...
Mel Gorman [Thu, 13 Jan 2011 23:45:57 +0000 (15:45 -0800)]
mm: migration: allow migration to operate asynchronously and avoid synchronous compaction in the faster path

Migration synchronously waits for writeback if the initial passes fails.
Callers of memory compaction do not necessarily want this behaviour if the
caller is latency sensitive or expects that synchronous migration is not
going to have a significantly better success rate.

This patch adds a sync parameter to migrate_pages() allowing the caller to
indicate if wait_on_page_writeback() is allowed within migration or not.
For reclaim/compaction, try_to_compact_pages() is first called
asynchronously, direct reclaim runs and then try_to_compact_pages() is
called synchronously as there is a greater expectation that it'll succeed.

[akpm@linux-foundation.org: build/merge fix]
Signed-off-by: Mel Gorman <mel@csn.ul.ie>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Rik van Riel <riel@redhat.com>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Cc: Andy Whitcroft <apw@shadowen.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
13 years agomm: vmscan: reclaim order-0 and use compaction instead of lumpy reclaim
Mel Gorman [Thu, 13 Jan 2011 23:45:56 +0000 (15:45 -0800)]
mm: vmscan: reclaim order-0 and use compaction instead of lumpy reclaim

Lumpy reclaim is disruptive.  It reclaims a large number of pages and
ignores the age of the pages it reclaims.  This can incur significant
stalls and potentially increase the number of major faults.

Compaction has reached the point where it is considered reasonably stable
(meaning it has passed a lot of testing) and is a potential candidate for
displacing lumpy reclaim.  This patch introduces an alternative to lumpy
reclaim whe compaction is available called reclaim/compaction.  The basic
operation is very simple - instead of selecting a contiguous range of
pages to reclaim, a number of order-0 pages are reclaimed and then
compaction is later by either kswapd (compact_zone_order()) or direct
compaction (__alloc_pages_direct_compact()).

[akpm@linux-foundation.org: fix build]
[akpm@linux-foundation.org: use conventional task_struct naming]
Signed-off-by: Mel Gorman <mel@csn.ul.ie>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Rik van Riel <riel@redhat.com>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Cc: Andy Whitcroft <apw@shadowen.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
13 years agomm: vmscan: convert lumpy_mode into a bitmask
Mel Gorman [Thu, 13 Jan 2011 23:45:55 +0000 (15:45 -0800)]
mm: vmscan: convert lumpy_mode into a bitmask

Currently lumpy_mode is an enum and determines if lumpy reclaim is off,
syncronous or asyncronous.  In preparation for using compaction instead of
lumpy reclaim, this patch converts the flags into a bitmap.

Signed-off-by: Mel Gorman <mel@csn.ul.ie>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Andy Whitcroft <apw@shadowen.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
13 years agomm: compaction: add trace events for memory compaction activity
Mel Gorman [Thu, 13 Jan 2011 23:45:54 +0000 (15:45 -0800)]
mm: compaction: add trace events for memory compaction activity

In preparation for a patches promoting the use of memory compaction over
lumpy reclaim, this patch adds trace points for memory compaction
activity.  Using them, we can monitor the scanning activity of the
migration and free page scanners as well as the number and success rates
of pages passed to page migration.

Signed-off-by: Mel Gorman <mel@csn.ul.ie>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Andy Whitcroft <apw@shadowen.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
13 years agomm: smaps: export mlock information
Nikanth Karthikesan [Thu, 13 Jan 2011 23:45:53 +0000 (15:45 -0800)]
mm: smaps: export mlock information

Currently there is no way to find whether a process has locked its pages
in memory or not.  And which of the memory regions are locked in memory.

Add a new field "Locked" to export this information via the smaps file.

Signed-off-by: Nikanth Karthikesan <knikanth@suse.de>
Acked-by: Balbir Singh <balbir@linux.vnet.ibm.com>
Acked-by: Wu Fengguang <fengguang.wu@intel.com>
Cc: Matt Mackall <mpm@selenic.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
13 years agomm: convert sprintf_symbol to %pS
Joe Perches [Thu, 13 Jan 2011 23:45:52 +0000 (15:45 -0800)]
mm: convert sprintf_symbol to %pS

Signed-off-by: Joe Perches <joe@perches.com>
Acked-by: Pekka Enberg <penberg@kernel.org>
Cc: Jiri Kosina <trivial@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
13 years agofs/mpage.c: consolidate code
Hai Shan [Thu, 13 Jan 2011 23:45:51 +0000 (15:45 -0800)]
fs/mpage.c: consolidate code

Merge mpage_end_io_read() and mpage_end_io_write() into mpage_end_io() to
eliminate code duplication.

[akpm@linux-foundation.org: coding-style fixes]
Signed-off-by: Hai Shan <shan.hai@windriver.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
13 years agomm: find_get_pages_contig fixlet
Nick Piggin [Thu, 13 Jan 2011 23:45:51 +0000 (15:45 -0800)]
mm: find_get_pages_contig fixlet

Testing ->mapping and ->index without a ref is not stable as the page
may have been reused at this point.

Signed-off-by: Nick Piggin <npiggin@kernel.dk>
Reviewed-by: Wu Fengguang <fengguang.wu@intel.com>
Reviewed-by: Minchan Kim <minchan.kim@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
13 years agovmscan: factor out kswapd sleeping logic from kswapd()
KOSAKI Motohiro [Thu, 13 Jan 2011 23:45:50 +0000 (15:45 -0800)]
vmscan: factor out kswapd sleeping logic from kswapd()

Currently, kswapd() has deep nesting and is slightly hard to read.  Clean
this up.

Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Mel Gorman <mel@csn.ul.ie>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
13 years agomm/page-writeback.c: fix __set_page_dirty_no_writeback() return value
Bob Liu [Thu, 13 Jan 2011 23:45:49 +0000 (15:45 -0800)]
mm/page-writeback.c: fix __set_page_dirty_no_writeback() return value

__set_page_dirty_no_writeback() should return true if it actually
transitioned the page from a clean to dirty state although it seems nobody
uses its return value at present.

Signed-off-by: Bob Liu <lliubbo@gmail.com>
Acked-by: Wu Fengguang <fengguang.wu@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
13 years agosync_inode_metadata: fix comment
Andrew Morton [Thu, 13 Jan 2011 23:45:48 +0000 (15:45 -0800)]
sync_inode_metadata: fix comment

Use correct function name, remove incorrect apostrophe

Cc: Christoph Hellwig <hch@lst.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
13 years agowriteback: avoid livelocking WB_SYNC_ALL writeback
Jan Kara [Thu, 13 Jan 2011 23:45:48 +0000 (15:45 -0800)]
writeback: avoid livelocking WB_SYNC_ALL writeback

When wb_writeback() is called in WB_SYNC_ALL mode, work->nr_to_write is
usually set to LONG_MAX.  The logic in wb_writeback() then calls
__writeback_inodes_sb() with nr_to_write == MAX_WRITEBACK_PAGES and we
easily end up with non-positive nr_to_write after the function returns, if
the inode has more than MAX_WRITEBACK_PAGES dirty pages at the moment.

When nr_to_write is <= 0 wb_writeback() decides we need another round of
writeback but this is wrong in some cases!  For example when a single
large file is continuously dirtied, we would never finish syncing it
because each pass would be able to write MAX_WRITEBACK_PAGES and inode
dirty timestamp never gets updated (as inode is never completely clean).
Thus __writeback_inodes_sb() would write the redirtied inode again and
again.

Fix the issue by setting nr_to_write to LONG_MAX in WB_SYNC_ALL mode.  We
do not need nr_to_write in WB_SYNC_ALL mode anyway since
write_cache_pages() does livelock avoidance using page tagging in
WB_SYNC_ALL mode.

This makes wb_writeback() call __writeback_inodes_sb() only once on
WB_SYNC_ALL.  The latter function won't livelock because it works on

- a finite set of files by doing queue_io() once at the beginning
- a finite set of pages by PAGECACHE_TAG_TOWRITE page tagging

After this patch, program from http://lkml.org/lkml/2010/10/24/154 is no
longer able to stall sync forever.

[fengguang.wu@intel.com: fix locking comment]
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Jan Engelhardt <jengelh@medozas.de>
Cc: Jens Axboe <axboe@kernel.dk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
13 years agowriteback: stop background/kupdate works from livelocking other works
Jan Kara [Thu, 13 Jan 2011 23:45:47 +0000 (15:45 -0800)]
writeback: stop background/kupdate works from livelocking other works

Background writeback is easily livelockable in a loop in wb_writeback() by
a process continuously re-dirtying pages (or continuously appending to a
file).  This is in fact intended as the target of background writeback is
to write dirty pages it can find as long as we are over
dirty_background_threshold.

But the above behavior gets inconvenient at times because no other work
queued in the flusher thread's queue gets processed.  In particular, since
e.g.  sync(1) relies on flusher thread to do all the IO for it, sync(1)
can hang forever waiting for flusher thread to do the work.

Generally, when a flusher thread has some work queued, someone submitted
the work to achieve a goal more specific than what background writeback
does.  Moreover by working on the specific work, we also reduce amount of
dirty pages which is exactly the target of background writeout.  So it
makes sense to give specific work a priority over a generic page cleaning.

Thus we interrupt background writeback if there is some other work to do.
We return to the background writeback after completing all the queued
work.

This may delay the writeback of expired inodes for a while, however the
expired inodes will eventually be flushed to disk as long as the other
works won't livelock.

[fengguang.wu@intel.com: update comment]
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Jan Engelhardt <jengelh@medozas.de>
Cc: Jens Axboe <axboe@kernel.dk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
13 years agowriteback: trace wakeup event for background writeback
Wu Fengguang [Thu, 13 Jan 2011 23:45:46 +0000 (15:45 -0800)]
writeback: trace wakeup event for background writeback

This tracks when balance_dirty_pages() tries to wakeup the flusher thread
for background writeback (if it was not started already).

Suggested-by: Christoph Hellwig <hch@infradead.org>
Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
Cc: Jan Kara <jack@suse.cz>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Jan Engelhardt <jengelh@medozas.de>
Cc: Jens Axboe <axboe@kernel.dk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
13 years agowriteback: integrated background writeback work
Jan Kara [Thu, 13 Jan 2011 23:45:44 +0000 (15:45 -0800)]
writeback: integrated background writeback work

Check whether background writeback is needed after finishing each work.

When bdi flusher thread finishes doing some work check whether any kind of
background writeback needs to be done (either because
dirty_background_ratio is exceeded or because we need to start flushing
old inodes).  If so, just do background write back.

This way, bdi_start_background_writeback() just needs to wake up the
flusher thread.  It will do background writeback as soon as there is no
other work.

This is a preparatory patch for the next patch which stops background
writeback as soon as there is other work to do.

Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Jan Engelhardt <jengelh@medozas.de>
Cc: Jens Axboe <axboe@kernel.dk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
13 years agomm: vmstat: use a single setter function and callback for adjusting percpu thresholds
Mel Gorman [Thu, 13 Jan 2011 23:45:43 +0000 (15:45 -0800)]
mm: vmstat: use a single setter function and callback for adjusting percpu thresholds

reduce_pgdat_percpu_threshold() and restore_pgdat_percpu_threshold() exist
to adjust the per-cpu vmstat thresholds while kswapd is awake to avoid
errors due to counter drift.  The functions duplicate some code so this
patch replaces them with a single set_pgdat_percpu_threshold() that takes
a callback function to calculate the desired threshold as a parameter.

[akpm@linux-foundation.org: readability tweak]
[kosaki.motohiro@jp.fujitsu.com: set_pgdat_percpu_threshold(): don't use for_each_online_cpu]
Signed-off-by: Mel Gorman <mel@csn.ul.ie>
Reviewed-by: Christoph Lameter <cl@linux.com>
Reviewed-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
13 years agomm: page allocator: adjust the per-cpu counter threshold when memory is low
Mel Gorman [Thu, 13 Jan 2011 23:45:41 +0000 (15:45 -0800)]
mm: page allocator: adjust the per-cpu counter threshold when memory is low

Commit aa45484 ("calculate a better estimate of NR_FREE_PAGES when memory
is low") noted that watermarks were based on the vmstat NR_FREE_PAGES.  To
avoid synchronization overhead, these counters are maintained on a per-cpu
basis and drained both periodically and when a threshold is above a
threshold.  On large CPU systems, the difference between the estimate and
real value of NR_FREE_PAGES can be very high.  The system can get into a
case where pages are allocated far below the min watermark potentially
causing livelock issues.  The commit solved the problem by taking a better
reading of NR_FREE_PAGES when memory was low.

Unfortately, as reported by Shaohua Li this accurate reading can consume a
large amount of CPU time on systems with many sockets due to cache line
bouncing.  This patch takes a different approach.  For large machines
where counter drift might be unsafe and while kswapd is awake, the per-cpu
thresholds for the target pgdat are reduced to limit the level of drift to
what should be a safe level.  This incurs a performance penalty in heavy
memory pressure by a factor that depends on the workload and the machine
but the machine should function correctly without accidentally exhausting
all memory on a node.  There is an additional cost when kswapd wakes and
sleeps but the event is not expected to be frequent - in Shaohua's test
case, there was one recorded sleep and wake event at least.

To ensure that kswapd wakes up, a safe version of zone_watermark_ok() is
introduced that takes a more accurate reading of NR_FREE_PAGES when called
from wakeup_kswapd, when deciding whether it is really safe to go back to
sleep in sleeping_prematurely() and when deciding if a zone is really
balanced or not in balance_pgdat().  We are still using an expensive
function but limiting how often it is called.

When the test case is reproduced, the time spent in the watermark
functions is reduced.  The following report is on the percentage of time
spent cumulatively spent in the functions zone_nr_free_pages(),
zone_watermark_ok(), __zone_watermark_ok(), zone_watermark_ok_safe(),
zone_page_state_snapshot(), zone_page_state().

vanilla                      11.6615%
disable-threshold            0.2584%

David said:

: We had to pull aa454840 "mm: page allocator: calculate a better estimate
: of NR_FREE_PAGES when memory is low and kswapd is awake" from 2.6.36
: internally because tests showed that it would cause the machine to stall
: as the result of heavy kswapd activity.  I merged it back with this fix as
: it is pending in the -mm tree and it solves the issue we were seeing, so I
: definitely think this should be pushed to -stable (and I would seriously
: consider it for 2.6.37 inclusion even at this late date).

Signed-off-by: Mel Gorman <mel@csn.ul.ie>
Reported-by: Shaohua Li <shaohua.li@intel.com>
Reviewed-by: Christoph Lameter <cl@linux.com>
Tested-by: Nicolas Bareil <nico@chdir.org>
Cc: David Rientjes <rientjes@google.com>
Cc: Kyle McMartin <kyle@mcmartin.ca>
Cc: <stable@kernel.org> [2.6.37.1, 2.6.36.x]
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
13 years agosched: remove long deprecated CLONE_STOPPED flag
Dave Jones [Thu, 13 Jan 2011 23:45:40 +0000 (15:45 -0800)]
sched: remove long deprecated CLONE_STOPPED flag

This warning was added in commit bdff746a3915 ("clone: prepare to recycle
CLONE_STOPPED") three years ago.  2.6.26 came and went.  As far as I know,
no-one is actually using CLONE_STOPPED.

Signed-off-by: Dave Jones <davej@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Tejun Heo <tj@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
13 years agoatmel_serial: fix RTS high after initialization in RS485 mode
Claudio Scordino [Thu, 13 Jan 2011 23:45:39 +0000 (15:45 -0800)]
atmel_serial: fix RTS high after initialization in RS485 mode

When working in RS485 mode, the atmel_serial driver keeps RTS high after
the initialization of the serial port.  It goes low only after the first
character has been sent.

[akpm@linux-foundation.org: simplify code]
Signed-off-by: Claudio Scordino <claudio@evidence.eu.com>
Signed-off-by: Arkadiusz Bubala <arkadiusz.bubala@gmail.com>
Tested-by: Arkadiusz Bubala <arkadiusz.bubala@gmail.com>
Cc: Nicolas Ferre <nicolas.ferre@atmel.com>
Cc: Greg KH <greg@kroah.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
13 years agoirq: use per_cpu kstat_irqs
Eric Dumazet [Thu, 13 Jan 2011 23:45:38 +0000 (15:45 -0800)]
irq: use per_cpu kstat_irqs

Use modern per_cpu API to increment {soft|hard}irq counters, and use
per_cpu allocation for (struct irq_desc)->kstats_irq instead of an array.

This gives better SMP/NUMA locality and saves few instructions per irq.

With small nr_cpuids values (8 for example), kstats_irq was a small array
(less than L1_CACHE_BYTES), potentially source of false sharing.

In the !CONFIG_SPARSE_IRQ case, remove the huge, NUMA/cache unfriendly
kstat_irqs_all[NR_IRQS][NR_CPUS] array.

Note: we still populate kstats_irq for all possible irqs in
early_irq_init().  We probably could use on-demand allocations.  (Code
included in alloc_descs()).  Problem is not all IRQS are used with a prior
alloc_descs() call.

kstat_irqs_this_cpu() is not used anymore, remove it.

Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Reviewed-by: Christoph Lameter <cl@linux.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Andi Kleen <andi@firstfloor.org>
Cc: Tejun Heo <tj@kernel.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
13 years agoMAINTAINERS: update entries affecting VIA Technologies
Bruce Chang [Thu, 13 Jan 2011 23:45:37 +0000 (15:45 -0800)]
MAINTAINERS: update entries affecting VIA Technologies

Since the original maintainer-Joseph Chan (josephchan@via.com.tw) doesn't
handle the Linux driver for VIA now, I would like to request to update the
maintainer for the SD/MMC CARD CONTROLLER DRIVER and VIA
UNICHROME(PRO)/CHROME9 FRAMEBUFFER DRIVER before we find a better one.

Signed-off-by: Bruce Chang <brucechang@via.com.tw>
Signed-off-by: Florian Tobias Schandinat <FlorianSchandinat@gmx.de>
Cc: Joseph Chan <JosephChan@via.com.tw>
Cc: Geert Uytterhoeven <geert@linux-m68k.org>
Cc: Harald Welte <HaraldWelte@viatech.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
13 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/agk/linux-2.6-dm
Linus Torvalds [Fri, 14 Jan 2011 01:30:47 +0000 (17:30 -0800)]
Merge git://git./linux/kernel/git/agk/linux-2.6-dm

* git://git.kernel.org/pub/scm/linux/kernel/git/agk/linux-2.6-dm: (32 commits)
  dm: raid456 basic support
  dm: per target unplug callback support
  dm: introduce target callbacks and congestion callback
  dm mpath: delay activate_path retry on SCSI_DH_RETRY
  dm: remove superfluous irq disablement in dm_request_fn
  dm log: use PTR_ERR value instead of ENOMEM
  dm snapshot: avoid storing private suspended state
  dm snapshot: persistent make metadata_wq multithreaded
  dm: use non reentrant workqueues if equivalent
  dm: convert workqueues to alloc_ordered
  dm stripe: switch from local workqueue to system_wq
  dm: dont use flush_scheduled_work
  dm snapshot: remove unused dm_snapshot queued_bios_work
  dm ioctl: suppress needless warning messages
  dm crypt: add loop aes iv generator
  dm crypt: add multi key capability
  dm crypt: add post iv call to iv generator
  dm crypt: use io thread for reads only if mempool exhausted
  dm crypt: scale to multiple cpus
  dm crypt: simplify compatible table output
  ...

13 years agoMerge branch 'for-linus' of git://neil.brown.name/md
Linus Torvalds [Fri, 14 Jan 2011 01:30:20 +0000 (17:30 -0800)]
Merge branch 'for-linus' of git://neil.brown.name/md

* 'for-linus' of git://neil.brown.name/md:
  md: Fix removal of extra drives when converting RAID6 to RAID5
  md: range check slot number when manually adding a spare.
  md/raid5: handle manually-added spares in start_reshape.
  md: fix sync_completed reporting for very large drives (>2TB)
  md: allow suspend_lo and suspend_hi to decrease as well as increase.
  md: Don't let implementation detail of curr_resync leak out through sysfs.
  md: separate meta and data devs
  md-new-param-to_sync_page_io
  md-new-param-to-calc_dev_sboffset
  md: Be more careful about clearing flags bit in ->recovery
  md: md_stop_writes requires mddev_lock.
  md/raid5: use sysfs_notify_dirent_safe to avoid NULL pointer
  md: Ensure no IO request to get md device before it is properly initialised.
  md: Fix single printks with multiple KERN_<level>s
  md: fix regression resulting in delays in clearing bits in a bitmap
  md: fix regression with re-adding devices to arrays with no metadata

13 years agoMerge branch 'release' of git://git.kernel.org/pub/scm/linux/kernel/git/aegl/linux-2.6
Linus Torvalds [Fri, 14 Jan 2011 01:29:53 +0000 (17:29 -0800)]
Merge branch 'release' of git://git./linux/kernel/git/aegl/linux-2.6

* 'release' of git://git.kernel.org/pub/scm/linux/kernel/git/aegl/linux-2.6:
  [IA64] fix build error - arch/ia64/kernel/perfmon.c

13 years agoRevert "gpiolib: annotate gpio-intialization with __must_check"
Linus Torvalds [Fri, 14 Jan 2011 01:26:46 +0000 (17:26 -0800)]
Revert "gpiolib: annotate gpio-intialization with __must_check"

This reverts commit 0fdae42d361bbb431ca0ab0efed5126a94821177, which
wasn't really supposed to go in, and causes lots of annoying warnings.

Quoth Andrew:
  "Complete brainfart - I meant to drop that patch ages ago."

Quoth Greg:
  "Ick, yeah, that patch isn't ok to go in as-is, all of the callers
   need to be fixed up first, which is what I thought we had agreed on..."

Reported-by: Stephen Rothwell <sfr@canb.auug.org.au>
Acked-by: Andrew Morton <akpm@linux-foundation.org>
Acked-by: Greg KH <greg@kroah.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
13 years agoecryptfs: fix broken build
Linus Torvalds [Fri, 14 Jan 2011 01:19:38 +0000 (17:19 -0800)]
ecryptfs: fix broken build

Stephen Rothwell reports that the vfs merge broke the build of ecryptfs.
The breakage comes from commit 66cb76666d69 ("sanitize ecryptfs
->mount()") which was obviously not even build tested. Tssk, tssk, Al.

This is the minimal build fixup for the situation, although I don't have
a filesystem to actually test it with.

Reported-by: Stephen Rothwell <sfr@canb.auug.org.au>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
13 years ago[IA64] fix build error - arch/ia64/kernel/perfmon.c
Tony Luck [Thu, 13 Jan 2011 22:49:56 +0000 (14:49 -0800)]
[IA64] fix build error - arch/ia64/kernel/perfmon.c

arch/ia64/kernel/perfmon.c:621: error: duplicate 'static'

Introduced by commit c74a1cbb3cac348f276fabc381758f5b0b4713b2

    pass default dentry_operations to mount_pseudo()

Signed-off-by: Tony Luck <tony.luck@intel.com>
13 years agomd: Fix removal of extra drives when converting RAID6 to RAID5
NeilBrown [Thu, 13 Jan 2011 22:14:34 +0000 (09:14 +1100)]
md: Fix removal of extra drives when converting RAID6 to RAID5

When a RAID6 is converted to a RAID5, the extra drive should
be discarded.  However it isn't due to a typo in a comparison.

This bug was introduced in commit e93f68a1fc6 in 2.6.35-rc4
and is suitable for any -stable since than.

As the extra drive is not removed, the 'degraded' counter is wrong and
so the RAID5 will not respond correctly to a subsequent failure.

Cc: stable@kernel.org
Signed-off-by: NeilBrown <neilb@suse.de>
13 years agomd: range check slot number when manually adding a spare.
NeilBrown [Thu, 13 Jan 2011 22:14:34 +0000 (09:14 +1100)]
md: range check slot number when manually adding a spare.

When adding a spare to an active array, we should check the slot
number, but allow it to be larger than raid_disks if a reshape
is being prepared.

Apply the same test when adding a device to an
array-under-construction.  It already had most of the test in place,
but not quite all.

Signed-off-by: NeilBrown <neilb@suse.de>
13 years agomd/raid5: handle manually-added spares in start_reshape.
NeilBrown [Thu, 13 Jan 2011 22:14:34 +0000 (09:14 +1100)]
md/raid5: handle manually-added spares in start_reshape.

It is possible to manually add spares to specific slots before
starting a reshape.
raid5_start_reshape should recognised this possibility and include
it in the accounting.

Signed-off-by: NeilBrown <neilb@suse.de>
13 years agomd: fix sync_completed reporting for very large drives (>2TB)
Rémi Rérolle [Thu, 13 Jan 2011 22:14:34 +0000 (09:14 +1100)]
md: fix sync_completed reporting for very large drives (>2TB)

The values exported in the sync_completed file are unsigned long, which
overflows with very large drives, resulting in wrong values reported.

Since sync_completed uses sectors as unit, we'll start getting wrong
values with components larger than 2TB.

This patch simply replaces the use of unsigned long by unsigned long long.

Signed-off-by: Rémi Rérolle <rrerolle@lacie.com>
Signed-off-by: NeilBrown <neilb@suse.de>
13 years agomd: allow suspend_lo and suspend_hi to decrease as well as increase.
NeilBrown [Thu, 13 Jan 2011 22:14:34 +0000 (09:14 +1100)]
md: allow suspend_lo and suspend_hi to decrease as well as increase.

The sysfs attributes 'suspend_lo' and 'suspend_hi' describe a region
to which read/writes are suspended so that the under lying data can be
manipulated without user-space noticing.
Currently the window they describe can only move forwards along the
device.  However this is an unnecessary restriction which will cause
problems with planned developments.
So relax this restriction and allow these endpoints to move
arbitrarily.

Signed-off-by: NeilBrown <neilb@suse.de>
13 years agomd: Don't let implementation detail of curr_resync leak out through sysfs.
NeilBrown [Thu, 13 Jan 2011 22:14:34 +0000 (09:14 +1100)]
md: Don't let implementation detail of curr_resync leak out through sysfs.

mddev->curr_resync has artificial values of '1' and '2' which are used
by the code which ensures only one resync is happening at a time on
any given device.

These values are internal and should never be exposed to user-space
(except when translated appropriately as in the 'pending' status in
/proc/mdstat).

Unfortunately they are as ->curr_resync is assigned to
->curr_resync_completed and that value is directly visible through
sysfs.

So change the assignments to ->curr_resync_completed to get the same
valued from elsewhere in a form that doesn't have the magic '1' or '2'
values.

Signed-off-by: NeilBrown <neilb@suse.de>
13 years agomd: separate meta and data devs
Jonathan Brassow [Thu, 13 Jan 2011 22:14:34 +0000 (09:14 +1100)]
md: separate meta and data devs

Allow the metadata to be on a separate device from the
data.

This doesn't mean the data and metadata will by on separate
physical devices - it simply gives device-mapper and userspace
tools more flexibility.

Signed-off-by: NeilBrown <neilb@suse.de>
13 years agomd-new-param-to_sync_page_io
Jonathan Brassow [Thu, 13 Jan 2011 22:14:33 +0000 (09:14 +1100)]
md-new-param-to_sync_page_io

Add new parameter to 'sync_page_io'.

The new parameter allows us to distinguish between metadata and data
operations.  This becomes important later when we add the ability to
use separate devices for data and metadata.

Signed-off-by: Jonathan Brassow <jbrassow@redhat.com>
13 years agomd-new-param-to-calc_dev_sboffset
Jonathan Brassow [Thu, 13 Jan 2011 22:14:33 +0000 (09:14 +1100)]
md-new-param-to-calc_dev_sboffset

When we allow for separate devices for data and metadata
in a later patch, we will need to be able to calculate
the superblock offset based on more than the bdev.

Signed-off-by: Jonathan Brassow <jbrassow@redhat.com>
13 years agomd: Be more careful about clearing flags bit in ->recovery
NeilBrown [Thu, 13 Jan 2011 22:14:33 +0000 (09:14 +1100)]
md: Be more careful about clearing flags bit in ->recovery

Setting ->recovery to 0 is generally not a good idea as it could clear
bits that shouldn't be cleared.  In particular, MD_RECOVERY_FROZEN
should only be cleared on explicit request from user-space.

So when we need to clear things, just clear the bits that need
clearing.

As there are a few different places which reap a resync process - and
some do an incomplte job - factor out the code for doing the from
md_check_recovery and call that function instead of open coding part
of it.

Signed-off-by: NeilBrown <neilb@suse.de>
Reported-by: Jonathan Brassow <jbrassow@redhat.com>
13 years agomd: md_stop_writes requires mddev_lock.
NeilBrown [Thu, 13 Jan 2011 22:14:33 +0000 (09:14 +1100)]
md: md_stop_writes requires mddev_lock.

As md_stop_writes manipulates the sync_thread and calls md_update_sb,
it need to be called with mddev_lock held.

In all internal cases it is, but the symbol is exported for dm-raid to
call and in that case the lock won't be help.
Do make an exported version which takes the lock, and an internal
version which does not.

Signed-off-by: NeilBrown <neilb@suse.de>
13 years agomd/raid5: use sysfs_notify_dirent_safe to avoid NULL pointer
Jonathan Brassow [Thu, 13 Jan 2011 22:14:33 +0000 (09:14 +1100)]
md/raid5: use sysfs_notify_dirent_safe to avoid NULL pointer

With the module parameter 'start_dirty_degraded' set,
raid5_spare_active() previously called sysfs_notify_dirent() with a NULL
argument (rdev->sysfs_state) when a rebuild finished.

Signed-off-by: Jonathan Brassow <jbrassow@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
13 years agomd: Ensure no IO request to get md device before it is properly initialised.
NeilBrown [Thu, 13 Jan 2011 22:14:33 +0000 (09:14 +1100)]
md: Ensure no IO request to get md device before it is properly initialised.

When an md device is in the process of coming on line it is possible
for an IO request (typically a partition table probe) to get through
before the array is fully initialised, which can cause unexpected
behaviour (e.g. a crash).

So explicitly record when the array is ready for IO and don't allow IO
through until then.

There is no possibility for a similar problem when the array is going
off-line as there must only be one 'open' at that time, and it is busy
off-lining the array and so cannot send IO requests.  So no memory
barrier is needed in md_stop()

This has been a bug since commit 409c57f3801 in 2.6.30 which
introduced md_make_request.  Before then, each personality would
register its own make_request_fn when it was ready.
This is suitable for any stable kernel from 2.6.30.y onwards.

Cc: <stable@kernel.org>
Signed-off-by: NeilBrown <neilb@suse.de>
Reported-by: "Hawrylewicz Czarnowski, Przemyslaw" <przemyslaw.hawrylewicz.czarnowski@intel.com>
13 years agomd: Fix single printks with multiple KERN_<level>s
Joe Perches [Thu, 13 Jan 2011 22:14:33 +0000 (09:14 +1100)]
md: Fix single printks with multiple KERN_<level>s

Noticed-by: Russell King <linux@arm.linux.org.uk>
Signed-off-by: Joe Perches <joe@perches.com>
Signed-off-by: NeilBrown <neilb@suse.de>
13 years agomd: fix regression resulting in delays in clearing bits in a bitmap
NeilBrown [Thu, 13 Jan 2011 22:13:53 +0000 (09:13 +1100)]
md: fix regression resulting in delays in clearing bits in a bitmap

commit 589a594be1fb (2.6.37-rc4) fixed a problem were md_thread would
sometimes call the ->run function at a bad time.

If an error is detected during array start up after the md_thread has
been started, the md_thread is killed.  This resulted in the ->run
function being called once.  However the array may not be in a state
that it is safe to call ->run.

However the fix imposed meant that  ->run was not called on a timeout.
This means that when an array goes idle, bitmap bits do not get
cleared promptly.  While the array is busy the bits will still be
cleared when appropriate so this is not very serious.  There is no
risk to data.

Change the test so that we only avoid calling ->run when the thread
is being stopped.  This more explicitly addresses the problem situation.

This is suitable for 2.6.37-stable and any -stable kernel to which
589a594be1fb was applied.

Cc: stable@kernel.org
Signed-off-by: NeilBrown <neilb@suse.de>
13 years agoMerge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/egtvedt...
Linus Torvalds [Thu, 13 Jan 2011 20:06:58 +0000 (12:06 -0800)]
Merge branch 'for-linus' of git://git./linux/kernel/git/egtvedt/avr32-2.6

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/egtvedt/avr32-2.6:
  avr32: update default configuration files for Atmel boards
  avr32: Convert to clocksource_register_hz
  avr32: make architecture sys_clone prototype match asm-generic prototype
  avr32: use syscall prototypes from asm-generic instead of arch
  avr32: disable kprobes for all default configurations
  avr32: boards: setup: use IS_ERR() instead of NULL check

13 years agoNFS: Fix NFSv3 exclusive open semantics
Trond Myklebust [Thu, 13 Jan 2011 19:15:50 +0000 (14:15 -0500)]
NFS: Fix NFSv3 exclusive open semantics

Commit c0204fd2b8fe047b18b67e07e1bf2a03691240cd (NFS: Clean up
nfs4_proc_create()) broke NFSv3 exclusive open by removing the code
that passes the O_EXCL flag down to nfs3_proc_create(). This patch
reverts that offending hunk from the original commit.

Reported-by: Nick Bowler <nbowler@elliptictech.com>
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
Cc: stable@kernel.org [2.6.37]
Tested-by: Nick Bowler <nbowler@elliptictech.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
13 years agodm: raid456 basic support
NeilBrown [Thu, 13 Jan 2011 20:00:02 +0000 (20:00 +0000)]
dm: raid456 basic support

This patch is the skeleton for the DM target that will be
the bridge from DM to MD (initially RAID456 and later RAID1).  It
provides a way to use device-mapper interfaces to the MD RAID456
drivers.

As with all device-mapper targets, the nominal public interfaces are the
constructor (CTR) tables and the status outputs (both STATUSTYPE_INFO
and STATUSTYPE_TABLE).  The CTR table looks like the following:

1: <s> <l> raid \
2: <raid_type> <#raid_params> <raid_params> \
3: <#raid_devs> <meta_dev1> <dev1> .. <meta_devN> <devN>

Line 1 contains the standard first three arguments to any device-mapper
target - the start, length, and target type fields.  The target type in
this case is "raid".

Line 2 contains the arguments that define the particular raid
type/personality/level, the required arguments for that raid type, and
any optional arguments.  Possible raid types include: raid4, raid5_la,
raid5_ls, raid5_rs, raid6_zr, raid6_nr, and raid6_nc.  (again, raid1 is
planned for the future.)  The list of required and optional parameters
is the same for all the current raid types.  The required parameters are
positional, while the optional parameters are given as key/value pairs.
The possible parameters are as follows:
 <chunk_size> Chunk size in sectors.
 [[no]sync] Force/Prevent RAID initialization
 [rebuild <idx>] Rebuild the drive indicated by the index
 [daemon_sleep <ms>] Time between bitmap daemon work to clear bits
 [min_recovery_rate <kB/sec/disk>] Throttle RAID initialization
 [max_recovery_rate <kB/sec/disk>] Throttle RAID initialization
 [max_write_behind <value>] See '-write-behind=' (man mdadm)
 [stripe_cache <sectors>] Stripe cache size for higher RAIDs

Line 3 contains the list of devices that compose the array in
metadata/data device pairs.  If the metadata is stored separately, a '-'
is given for the metadata device position.  If a drive has failed or is
missing at creation time, a '-' can be given for both the metadata and
data drives for a given position.

Examples:
# RAID4 - 4 data drives, 1 parity
# No metadata devices specified to hold superblock/bitmap info
# Chunk size of 1MiB
# (Lines separated for easy reading)
1960893648 raid \
raid4 1 2048 \
5 - 8:17 - 8:33 - 8:49 - 8:65 - 8:81

# RAID4 - 4 data drives, 1 parity (no metadata devices)
# Chunk size of 1MiB, force RAID initialization,
# min recovery rate at 20 kiB/sec/disk
1960893648 raid \
        raid4 4 2048 min_recovery_rate 20 sync\
        5 - 8:17 - 8:33 - 8:49 - 8:65 - 8:81

Performing a 'dmsetup table' should display the CTR table used to
construct the mapping (with possible reordering of optional
parameters).

Performing a 'dmsetup status' will yield information on the state and
health of the array.  The output is as follows:
1: <s> <l> raid \
2: <raid_type> <#devices> <1 health char for each dev> <resync_ratio>

Line 1 is standard DM output.  Line 2 is best shown by example:
1960893648 raid raid4 5 AAAAA 2/490221568
Here we can see the RAID type is raid4, there are 5 devices - all of
which are 'A'live, and the array is 2/490221568 complete with recovery.

Cc: linux-raid@vger.kernel.org
Signed-off-by: NeilBrown <neilb@suse.de>
Signed-off-by: Jonathan Brassow <jbrassow@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
13 years agodm: per target unplug callback support
NeilBrown [Thu, 13 Jan 2011 20:00:02 +0000 (20:00 +0000)]
dm: per target unplug callback support

Add per-target unplug callback support.

Cc: linux-raid@vger.kernel.org
Signed-off-by: NeilBrown <neilb@suse.de>
Signed-off-by: Jonathan Brassow <jbrassow@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
13 years agodm: introduce target callbacks and congestion callback
NeilBrown [Thu, 13 Jan 2011 20:00:01 +0000 (20:00 +0000)]
dm: introduce target callbacks and congestion callback

DM currently implements congestion checking by checking on congestion
in each component device.  For raid456 we need to also check if the
stripe cache is congested.

Add per-target congestion checker callback support.

Extending the target_callbacks structure with additional callback
functions allows for establishing multiple callbacks per-target (a
callback is also needed for unplug).

Cc: linux-raid@vger.kernel.org
Signed-off-by: NeilBrown <neilb@suse.de>
Signed-off-by: Jonathan Brassow <jbrassow@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
13 years agodm mpath: delay activate_path retry on SCSI_DH_RETRY
Chandra Seetharaman [Thu, 13 Jan 2011 20:00:01 +0000 (20:00 +0000)]
dm mpath: delay activate_path retry on SCSI_DH_RETRY

This patch adds a user-configurable 'pg_init_delay_msecs' feature.  Use
this feature to specify the number of milliseconds to delay before
retrying scsi_dh_activate, when SCSI_DH_RETRY is returned.

SCSI Device Handlers return SCSI_DH_IMM_RETRY if we could retry
activation immediately and SCSI_DH_RETRY in cases where it is better to
retry after some delay.

Currently we immediately retry scsi_dh_activate irrespective of
SCSI_DH_IMM_RETRY and SCSI_DH_RETRY.

The 'pg_init_delay_msecs' feature may be provided during table create or
load, e.g.:
    dmsetup create --table "0 20971520 multipath 3 queue_if_no_path \
pg_init_delay_msecs 2500 ..." mpatha

The default for 'pg_init_delay_msecs' is 2000 milliseconds.
Maximum configurable delay is 60000 milliseconds.  Specifying a
'pg_init_delay_msecs' of 0 will cause immediate retry.

Signed-off-by: Nikanth Karthikesan <knikanth@suse.de>
Signed-off-by: Chandra Seetharaman <sekharan@us.ibm.com>
Acked-by: Mike Christie <michaelc@cs.wisc.edu>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
13 years agodm: remove superfluous irq disablement in dm_request_fn
Kiyoshi Ueda [Thu, 13 Jan 2011 20:00:00 +0000 (20:00 +0000)]
dm: remove superfluous irq disablement in dm_request_fn

This patch changes spin_lock_irq() to spin_lock() in dm_request_fn().
This patch is just a clean-up and no functional change.

The spin_lock_irq() was leftover from the early request-based dm code,
where map_request() used to enable interrupts.
Since current map_request() never enables interrupts, we can change it
to spin_lock() to match the prior spin_unlock().

Auditing through the dm and block-layer code called from
map_request(), I confirmed all functions save/restore interrupt
status, so no function returning with interrupts enabled.
Also I haven't observed any problem on my test environment which
uses scsi and lpfc driver after heavy I/O testing with occasional
path down/up.

Added BUG_ON() to detect breakage in future.

Signed-off-by: Kiyoshi Ueda <k-ueda@ct.jp.nec.com>
Signed-off-by: Jun'ichi Nomura <j-nomura@ce.jp.nec.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
13 years agodm log: use PTR_ERR value instead of ENOMEM
Dan Carpenter [Thu, 13 Jan 2011 20:00:00 +0000 (20:00 +0000)]
dm log: use PTR_ERR value instead of ENOMEM

It's nicer to return the PTR_ERR() value instead of just returning
-ENOMEM.  In the current code the PTR_ERR() value is always equal to
-ENOMEM so this doesn't actually affect anything, but still...

In addition, dm_dirty_log_create() doesn't check for a specific -ENOMEM
return.  So this change is safe relative to potential for a non -ENOMEM
return in the future.

Signed-off-by: Dan Carpenter <error27@gmail.com>
Acked-by: Jonathan Brassow <jbrassow@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
13 years agodm snapshot: avoid storing private suspended state
Mike Snitzer [Thu, 13 Jan 2011 19:59:59 +0000 (19:59 +0000)]
dm snapshot: avoid storing private suspended state

Use dm_suspended() rather than having each snapshot target maintain a
private 'suspended' flag in struct dm_snapshot.

Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
13 years agodm snapshot: persistent make metadata_wq multithreaded
Tejun Heo [Thu, 13 Jan 2011 19:59:59 +0000 (19:59 +0000)]
dm snapshot: persistent make metadata_wq multithreaded

metadata_wq serves on-stack work items from chunk_io().  Even if
multiple chunk_io() are simultaneously in progress, each is
independent and queued only once, so multithreaded workqueue can be
safely used.

Switch metadata_wq to multithread and flush the work item instead of
the workqueue in chunk_io().

Signed-off-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
13 years agodm: use non reentrant workqueues if equivalent
Tejun Heo [Thu, 13 Jan 2011 19:59:58 +0000 (19:59 +0000)]
dm: use non reentrant workqueues if equivalent

kmirrord_wq, kcopyd_work and md->wq are created per dm instance and
serve only a single work item from the dm instance, so non-reentrant
workqueues would provide the same ordering guarantees as ordered ones
while allowing CPU affinity and use of the workqueues for other
purposes.  Switch them to non-reentrant workqueues.

Signed-off-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
13 years agodm: convert workqueues to alloc_ordered
Tejun Heo [Thu, 13 Jan 2011 19:59:57 +0000 (19:59 +0000)]
dm: convert workqueues to alloc_ordered

Convert all create[_singlethread]_work() users to the new
alloc[_ordered]_workqueue().  This conversion is mechanical and
doesn't introduce any behavior change.

Signed-off-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
13 years agodm stripe: switch from local workqueue to system_wq
Tejun Heo [Thu, 13 Jan 2011 19:59:57 +0000 (19:59 +0000)]
dm stripe: switch from local workqueue to system_wq

kstriped only serves sc->kstriped_ws which runs dm_table_event().
This doesn't need to be executed from an ordered workqueue w/ rescuer.
Drop kstriped and use the system_wq instead.  While at it, rename
kstriped_ws to trigger_event so that it's consistent with other dm
modules.

Signed-off-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
13 years agodm: dont use flush_scheduled_work
Tejun Heo [Thu, 13 Jan 2011 19:59:56 +0000 (19:59 +0000)]
dm: dont use flush_scheduled_work

flush_scheduled_work() is being deprecated.  Flush the used work
directly instead.  In all dm targets, the only work which uses
system_wq is ->trigger_event.

Signed-off-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
13 years agodm snapshot: remove unused dm_snapshot queued_bios_work
Tejun Heo [Thu, 13 Jan 2011 19:59:56 +0000 (19:59 +0000)]
dm snapshot: remove unused dm_snapshot queued_bios_work

dm_snapshot->queued_bios_work isn't used.  Remove ->queued_bios[_work]
from dm_snapshot structure, the flush_queued_bios work function and
ksnapd workqueue.

The DM snapshot changes that were going to use the ksnapd workqueue were
either superseded (fix for origin write races) or never completed
(deallocation of invalid snapshot's memory via workqueue).

Signed-off-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
13 years agodm ioctl: suppress needless warning messages
Milan Broz [Thu, 13 Jan 2011 19:59:55 +0000 (19:59 +0000)]
dm ioctl: suppress needless warning messages

The device-mapper should not send warning messages to syslog
if a device is not found. This can be done by userspace
according to the returned dm-ioctl error code.

So move these messages to debug level and use rate limiting
to not flood syslog.

Signed-off-by: Milan Broz <mbroz@redhat.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
13 years agodm crypt: add loop aes iv generator
Milan Broz [Thu, 13 Jan 2011 19:59:55 +0000 (19:59 +0000)]
dm crypt: add loop aes iv generator

This patch adds a compatible implementation of the block
chaining mode used by the Loop-AES block device encryption
system (http://loop-aes.sourceforge.net/) designed
by Jari Ruusu.

It operates on full 512 byte sectors and uses CBC
with an IV derived from the sector number, the data and
optionally extra IV seed.

This means that after CBC decryption the first block of sector
must be tweaked according to decrypted data.

Loop-AES can use three encryption schemes:
 version 1: is plain aes-cbc mode (already compatible)
 version 2: uses 64 multikey scheme with own IV generator
 version 3: the same as version 2 with additional IV seed
            (it uses 65 keys, last key is used as IV seed)

The IV generator is here named lmk (Loop-AES multikey)
and for the cipher specification looks like: aes:64-cbc-lmk

Version 2 and 3 is recognised according to length
of provided multi-key string (which is just hexa encoded
"raw key" used in original Loop-AES ioctl).

Configuration of the device and decoding key string will
be done in userspace (cryptsetup).
(Loop-AES stores keys in gpg encrypted file, raw keys are
output of simple hashing of lines in this file).

Based on an implementation by Max Vozeler:
  http://article.gmane.org/gmane.linux.kernel.cryptoapi/3752/

Signed-off-by: Milan Broz <mbroz@redhat.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
CC: Max Vozeler <max@hinterhof.net>
13 years agodm crypt: add multi key capability
Milan Broz [Thu, 13 Jan 2011 19:59:54 +0000 (19:59 +0000)]
dm crypt: add multi key capability

This patch adds generic multikey handling to be used
in following patch for Loop-AES mode compatibility.

This patch extends mapping table to optional keycount and
implements generic multi-key capability.

With more keys defined the <key> string is divided into
several <keycount> sections and these are used for tfms.

The tfm is used according to sector offset
(sector 0->tfm[0], sector 1->tfm[1], sector N->tfm[N modulo keycount])
(only power of two values supported for keycount here).

Because of tfms per-cpu allocation, this mode can be take
a lot of memory on large smp systems.

Signed-off-by: Milan Broz <mbroz@redhat.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
Cc: Max Vozeler <max@hinterhof.net>
13 years agodm crypt: add post iv call to iv generator
Milan Broz [Thu, 13 Jan 2011 19:59:54 +0000 (19:59 +0000)]
dm crypt: add post iv call to iv generator

IV (initialisation vector) can in principle depend not only
on sector but also on plaintext data (or other attributes).

Change IV generator interface to work directly with dmreq
structure to allow such dependence in generator.

Also add post() function which is called after the crypto
operation.

This allows tricky modification of decrypted data or IV
internals.

In asynchronous mode the post() can be called after
ctx->sector count was increased so it is needed
to add iv_sector copy directly to dmreq structure.
(N.B. dmreq always include only one sector in scatterlists)

Signed-off-by: Milan Broz <mbroz@redhat.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
13 years agodm crypt: use io thread for reads only if mempool exhausted
Milan Broz [Thu, 13 Jan 2011 19:59:53 +0000 (19:59 +0000)]
dm crypt: use io thread for reads only if mempool exhausted

If there is enough memory, code can directly submit bio
instead queing this operation in separate thread.

Try to alloc bio clone with GFP_NOWAIT and only if it
fails use separate queue (map function cannot block here).

Signed-off-by: Milan Broz <mbroz@redhat.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
13 years agodm crypt: scale to multiple cpus
Andi Kleen [Thu, 13 Jan 2011 19:59:53 +0000 (19:59 +0000)]
dm crypt: scale to multiple cpus

Currently dm-crypt does all the encryption work for a single dm-crypt
mapping in a single workqueue. This does not scale well when multiple
CPUs are submitting IO at a high rate. The single CPU running the single
thread cannot keep up with the encryption and encrypted IO performance
tanks.

This patch changes the crypto workqueue to be per CPU. This means
that as long as the IO submitter (or the interrupt target CPUs
for reads) runs on different CPUs the encryption work will be also
parallel.

To avoid a bottleneck on the IO worker I also changed those to be
per-CPU threads.

There is still some shared data, so I suspect some bouncing
cache lines. But I haven't done a detailed study on that yet.

Signed-off-by: Andi Kleen <ak@linux.intel.com>
Signed-off-by: Milan Broz <mbroz@redhat.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
13 years agodm crypt: simplify compatible table output
Milan Broz [Thu, 13 Jan 2011 19:59:52 +0000 (19:59 +0000)]
dm crypt: simplify compatible table output

Rename cc->cipher_mode to cc->cipher_string and store the whole of the cipher
information so it can easily be printed when processing the DM_DEV_STATUS ioctl.

Signed-off-by: Milan Broz <mbroz@redhat.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
13 years agodm log userspace: add version number to comms
Jonathan Brassow [Thu, 13 Jan 2011 19:59:52 +0000 (19:59 +0000)]
dm log userspace: add version number to comms

This patch adds a 'version' field to the 'dm_ulog_request'
structure.

The 'version' field is taken from a portion of the unused
'padding' field in the 'dm_ulog_request' structure.  This
was done to avoid changing the size of the structure and
possibly disrupting backwards compatibility.

The version number will help notify user-space daemons
when a change has been made to the kernel/userspace
log API.

Signed-off-by: Jonathan Brassow <jbrassow@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
13 years agodm log userspace: group clear and mark requests
Jonathan Brassow [Thu, 13 Jan 2011 19:59:51 +0000 (19:59 +0000)]
dm log userspace: group clear and mark requests

Allow the device-mapper log's 'mark' and 'clear' requests to be
grouped and processed in a batch.  This can significantly reduce the
amount of traffic going between the kernel and userspace (where the
processing daemon resides).

Signed-off-by: Jonathan Brassow <jbrassow@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
13 years agodm log userspace: split flush queue
Jonathan Brassow [Thu, 13 Jan 2011 19:59:50 +0000 (19:59 +0000)]
dm log userspace: split flush queue

Split the 'flush_list', which contained a mix of both 'mark' and 'clear'
requests, into two distinct lists ('mark_list' and 'clear_list').

The device mapper log implementations (used by various DM targets) are
allowed to cache 'mark' and 'clear' requests until a 'flush' is
received.  Until now, these cached requests were kept in the same list.
They will now be put into distinct lists to facilitate group processing
of these requests (in the next patch).

Signed-off-by: Jonathan Brassow <jbrassow@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
13 years agodm kcopyd: delay unplugging
Mikulas Patocka [Thu, 13 Jan 2011 19:59:50 +0000 (19:59 +0000)]
dm kcopyd: delay unplugging

Make kcopyd merge more I/O requests by using device unplugging.

Without this patch, each I/O request is dispatched separately to the device.
If the device supports tagged queuing, there are many small requests sent
to the device. To improve performance, this patch will batch as many requests
as possible, allowing the queue to merge consecutive requests, and send them
to the device at once.

In my tests (15k SCSI disk), this patch improves sequential write throughput:

  Sequential write throughput (chunksize of 4k, 32k, 512k)
  unpatched: 15.2, 18.5, 17.5 MB/s
  patched:   14.4, 22.6, 23.0 MB/s

In most common uses (snapshot or two-way mirror), kcopyd is only used for
two devices, one for reading and the other for writing, thus this optimization
is implemented only for two devices. The optimization may be extended to n-way
mirrors with some code complexity increase.

We keep track of two block devices to unplug (one for read and the
other for write) and unplug them when exiting "do_work" thread.  If
there are more devices used (in theory it could happen, in practice it
is rare), we unplug immediately.

Signed-off-by: Mikulas Patocka <mpatocka@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>