platform/kernel/linux-exynos.git
10 years agof2fs: avoid to calculate incorrect max orphan number
Chao Yu [Fri, 29 Nov 2013 08:37:00 +0000 (16:37 +0800)]
f2fs: avoid to calculate incorrect max orphan number

Because we will write node summaries when do_checkpoint with umount flag,
our number of max orphan blocks should minus NR_CURSEG_NODE_TYPE additional.

Signed-off-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Shu Tan <shu.tan@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
10 years agof2fs: remove unneeded code in punch_hole
Chao Yu [Fri, 22 Nov 2013 08:52:50 +0000 (16:52 +0800)]
f2fs: remove unneeded code in punch_hole

Because FALLOC_FL_PUNCH_HOLE flag must be ORed with FALLOC_FL_KEEP_SIZE
in fallocate, so we could remove the useless 'keep size' branch code which
will never be excuted in punch_hole.

Signed-off-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Fan Li <fanofcode.li@samsung.com>
[Jaegeuk Kim: remove an unnecessary parameter togather]
Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
10 years agof2fs: remove unnecessary condition checks
Jaegeuk Kim [Thu, 28 Nov 2013 03:55:13 +0000 (12:55 +0900)]
f2fs: remove unnecessary condition checks

This patch removes the unnecessary condition checks on:

fs/f2fs/gc.c:667 do_garbage_collect() warn: 'sum_page' isn't an ERR_PTR
fs/f2fs/f2fs.h:795 f2fs_put_page() warn: 'page' isn't an ERR_PTR

Reported-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
10 years agof2fs: bug fix on bit overflow from 32bits to 64bits
Jaegeuk Kim [Thu, 28 Nov 2013 03:44:05 +0000 (12:44 +0900)]
f2fs: bug fix on bit overflow from 32bits to 64bits

This patch fixes some bit overflows by the shift operations.

Dan Carpenter reported potential bugs on bit overflows as follows.

fs/f2fs/segment.c:910 submit_write_page()
warn: should 'blk_addr << ((sbi)->log_blocksize - 9)' be a 64 bit type?
fs/f2fs/checkpoint.c:429 get_valid_checkpoint()
warn: should '1 << ()' be a 64 bit type?
fs/f2fs/data.c:408 f2fs_readpage()
warn: should 'blk_addr << ((sbi)->log_blocksize - 9)' be a 64 bit type?
fs/f2fs/data.c:457 submit_read_page()
warn: should 'blk_addr << ((sbi)->log_blocksize - 9)' be a 64 bit type?
fs/f2fs/data.c:525 get_data_block_ro()
warn: should 'i << blkbits' be a 64 bit type?

Bug-Reported-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
10 years agof2fs: fix a potential out of range issue
Gu Zheng [Tue, 26 Nov 2013 08:44:16 +0000 (16:44 +0800)]
f2fs: fix a potential out of range issue

Fix a potential out of range issue introduced by commit:
22fb72225a
f2fs: simplify write_orphan_inodes for better readable

Signed-off-by: Gu Zheng <guz.fnst@cn.fujitsu.com>
Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
10 years agof2fs: remove unnecessary return value
Jaegeuk Kim [Tue, 26 Nov 2013 07:36:20 +0000 (16:36 +0900)]
f2fs: remove unnecessary return value

Let's remove the unnecessary return value.

Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
10 years agof2fs: add a new mount option: inline_data
Huajun Li [Sun, 10 Nov 2013 15:13:17 +0000 (23:13 +0800)]
f2fs: add a new mount option: inline_data

Add a mount option: inline_data. If the mount option is set,
data of New created small files can be stored in their inode.

Signed-off-by: Huajun Li <huajun.li@intel.com>
Signed-off-by: Haicheng Li <haicheng.li@linux.intel.com>
Signed-off-by: Weihong Xu <weihong.xu@intel.com>
Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
10 years agof2fs: add flags and helpers to support inline data
Huajun Li [Sun, 10 Nov 2013 15:13:16 +0000 (23:13 +0800)]
f2fs: add flags and helpers to support inline data

Add new inode flags F2FS_INLINE_DATA and FI_INLINE_DATA to indicate
whether the inode has inline data.

Inline data makes use of inode block's data indices region to save small
file. Currently there are 923 data indices in an inode block. Since
inline xattr has made use of the last 50 indices to save its data, there
are 873 indices left which can be used for inline data. When
FI_INLINE_DATA is set, the layout of inode block's indices region is
like below:

+-----------------+
|                 | Reserved. reserve_new_block() will make use of
| i_addr[0]       | i_addr[0] when we need to reserve a new data block
|                 | to convert inline data into regular one's.
|-----------------|
|                 | Used by inline data. A file whose size is less than
| i_addr[1~872]   | 3488 bytes(~3.4k) and doesn't reserve extra
|                 | blocks by fallocate() can be saved here.
|-----------------|
|                 |
| i_addr[873~922] | Reserved for inline xattr
|                 |
+-----------------+

Signed-off-by: Haicheng Li <haicheng.li@linux.intel.com>
Signed-off-by: Huajun Li <huajun.li@intel.com>
Signed-off-by: Weihong Xu <weihong.xu@intel.com>
Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
10 years agof2fs: send REQ_META or REQ_PRIO when reading meta area
Changman Lee [Sun, 24 Nov 2013 06:13:08 +0000 (15:13 +0900)]
f2fs: send REQ_META or REQ_PRIO when reading meta area

Let's send REQ_META or REQ_PRIO when reading meta area such as NAT/SIT
etc.

Signed-off-by: Changman Lee <cm224.lee@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
10 years agof2fs: add detailed information of bio types in the tracepoints
Jaegeuk Kim [Sun, 24 Nov 2013 05:42:23 +0000 (14:42 +0900)]
f2fs: add detailed information of bio types in the tracepoints

This patch inserts information of bio types in more detail.
So, we can now see REQ_META and REQ_PRIO too.

Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
10 years agof2fs: add a new function: f2fs_reserve_block()
Huajun Li [Sun, 10 Nov 2013 15:13:18 +0000 (23:13 +0800)]
f2fs: add a new function: f2fs_reserve_block()

Add the function f2fs_reserve_block() to easily reserve new blocks, and
use it to clean up more codes.

Signed-off-by: Huajun Li <huajun.li@intel.com>
Signed-off-by: Haicheng Li <haicheng.li@linux.intel.com>
Signed-off-by: Weihong Xu <weihong.xu@intel.com>
Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
10 years agof2fs: avoid lock debugging overhead
Jaegeuk Kim [Sun, 24 Nov 2013 04:50:35 +0000 (13:50 +0900)]
f2fs: avoid lock debugging overhead

If CONFIG_F2FS_CHECK_FS is unset, we don't need to add any debugging overhead.

Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
10 years agof2fs: read contiguous sit entry pages by merging for mount performance
Chao Yu [Fri, 22 Nov 2013 01:09:59 +0000 (09:09 +0800)]
f2fs: read contiguous sit entry pages by merging for mount performance

Previously we read sit entries page one by one, this method lost the chance
of reading contiguous page together. So we read pages as contiguous as
possible for better mount performance.

change log:
 o merge judgements/use 'Continue' or 'Break' instead of 'Goto' as Gu Zheng
   suggested.
 o add mark_page_accessed() before release page to delay VM reclaiming.
 o remove '*order' for simplification of function as Jaegeuk Kim suggested.

Signed-off-by: Chao Yu <chao2.yu@samsung.com>
[Jaegeuk Kim: fix a bug on the block address calculation]
Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
10 years agof2fs: adds a tracepoint for f2fs_submit_read_bio
Chao Yu [Sun, 24 Nov 2013 03:36:42 +0000 (12:36 +0900)]
f2fs: adds a tracepoint for f2fs_submit_read_bio

This patch adds a tracepoint for f2fs_submit_read_bio.

Signed-off-by: Chao Yu <chao2.yu@samsung.com>
[Jaegeuk Kim: integrate tracepoints of f2fs_submit_read(_write)_bio]
Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
10 years agof2fs: adds a tracepoint for submit_read_page
Chao Yu [Wed, 20 Nov 2013 08:40:10 +0000 (16:40 +0800)]
f2fs: adds a tracepoint for submit_read_page

This patch adds a tracepoint for submit_read_page.

Signed-off-by: Chao Yu <chao2.yu@samsung.com>
[Jaegeuk Kim: integrate tracepoints of f2fs_submit_read(_write)_page]
Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
10 years agof2fs: simplify IS_DATASEG and IS_NODESEG macro
Changman Lee [Thu, 21 Nov 2013 11:04:21 +0000 (20:04 +0900)]
f2fs: simplify IS_DATASEG and IS_NODESEG macro

It is not efficient comparing each segment type to find node or data.

Signed-off-by: Changman Lee <cm224.lee@samsung.com>
[Jaegeuk Kim: remove unnecessary white spaces]
Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
10 years agof2fs: merge read IOs at ra_nat_pages()
Jaegeuk Kim [Thu, 21 Nov 2013 04:54:23 +0000 (13:54 +0900)]
f2fs: merge read IOs at ra_nat_pages()

Change log from v1:
  o add mark_page_accessed() not to reclaim the nat pages.

This patch changes the policy of submitting read bios at ra_nat_pages.

Previously, f2fs submits small read bios with block plugging.
But, with this patch, f2fs itself merges read bios first and then submits a
large bio, which can reduce the bio handling overheads.

Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
10 years agof2fs: add a new function to support for merging contiguous read
Chao Yu [Wed, 20 Nov 2013 06:46:39 +0000 (14:46 +0800)]
f2fs: add a new function to support for merging contiguous read

For better read performance, we add a new function to support for merging
contiguous read as the one for write.

v1-->v2:
 o add declarations here as Gu Zheng suggested.
 o use new structure f2fs_bio_info introduced by Jaegeuk Kim.

Signed-off-by: Chao Yu <chao2.yu@samsung.com>
Acked-by: Gu Zheng <guz.fnst@cn.fujitsu.com>
10 years agof2fs: move the list_head initialization into the lock protection region
Gu Zheng [Tue, 19 Nov 2013 10:03:47 +0000 (18:03 +0800)]
f2fs: move the list_head initialization into the lock protection region

Signed-off-by: Gu Zheng <guz.fnst@cn.fujitsu.com>
Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
10 years agof2fs: simplify write_orphan_inodes for better readable
Gu Zheng [Tue, 19 Nov 2013 10:03:58 +0000 (18:03 +0800)]
f2fs: simplify write_orphan_inodes for better readable

Simplify write_orphan_inodes for better readable. Because we hold the
orphan_inode_mutex, so it's safe to use list_for_each_entry instead of
list_for_each_safe.

Signed-off-by: Gu Zheng <guz.fnst@cn.fujitsu.com>
Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
10 years agof2fs: convert inc/dec_valid_node_count to inc/dec one count
Gu Zheng [Tue, 19 Nov 2013 10:03:38 +0000 (18:03 +0800)]
f2fs: convert inc/dec_valid_node_count to inc/dec one count

Signed-off-by: Gu Zheng <guz.fnst@cn.fujitsu.com>
Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
10 years agof2fs: convert dev_valid_block_count to void
Gu Zheng [Tue, 19 Nov 2013 10:03:27 +0000 (18:03 +0800)]
f2fs: convert dev_valid_block_count to void

Signed-off-by: Gu Zheng <guz.fnst@cn.fujitsu.com>
Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
10 years agof2fs: convert remove_inode_page to void
Gu Zheng [Tue, 19 Nov 2013 10:03:18 +0000 (18:03 +0800)]
f2fs: convert remove_inode_page to void

Signed-off-by: Gu Zheng <guz.fnst@cn.fujitsu.com>
Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
10 years agof2fs: introduce a bio array for per-page write bios
Jaegeuk Kim [Tue, 19 Nov 2013 03:47:22 +0000 (12:47 +0900)]
f2fs: introduce a bio array for per-page write bios

The f2fs has three bio types, NODE, DATA, and META, and manages some data
structures per each bio types.

The codes are a little bit messy, thus, this patch introduces a bio array
which groups individual data structures as follows.

struct f2fs_bio_info {
struct bio *bio; /* bios to merge */
sector_t last_block_in_bio; /* last block number */
struct mutex io_mutex; /* mutex for bio */
};

struct f2fs_sb_info {
...
struct f2fs_bio_info write_io[NR_PAGE_TYPE]; /* for write bios */
...
};

The code changes from this new data structure are trivial.

Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
10 years agof2fs: disable the extent cache ops on high fragmented files
Jaegeuk Kim [Tue, 19 Nov 2013 01:41:54 +0000 (10:41 +0900)]
f2fs: disable the extent cache ops on high fragmented files

The f2fs manages an extent cache to search a number of consecutive data blocks
very quickly.

However it conducts unnecessary cache operations if the file is highly
fragmented with no valid extent cache.

In such the case, we don't need to handle the extent cache, but just can disable
the cache facility.

Nevertheless, this patch gives one more chance to enable the extent cache.

For example,
1. create a file
2. write data sequentially which produces a large valid extent cache
3. update some data, resulting in a fragmented extent
4. if the fragmented extent is too small, then drop extent cache
5. close the file

6. open the file again
7. give another chance to make a new extent cache
8. write data sequentially again which creates another big extent cache.
...

Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
10 years agof2fs: use sbi->write_mutex for write bios
Jaegeuk Kim [Mon, 18 Nov 2013 08:16:17 +0000 (17:16 +0900)]
f2fs: use sbi->write_mutex for write bios

This patch removes an unnecessary semaphore (i.e., sbi->bio_sem).
There is no reason to use the semaphore when f2fs submits read and write IOs.
Instead, let's use a write mutex and cover the sbi->bio[] by the lock.

Change log from v1:
 o split write_mutex suggested by Chao Yu

Chao described,
"All DATA/NODE/META bio buffers in superblock is protected by
'sbi->write_mutex', but each bio buffer area is independent, So we
should split write_mutex to three for DATA/NODE/META."

Signed-off-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
10 years agof2fs: clean up the do_submit_bio flow
Jaegeuk Kim [Mon, 18 Nov 2013 08:13:35 +0000 (17:13 +0900)]
f2fs: clean up the do_submit_bio flow

This patch introduces PAGE_TYPE_OF_BIO() and cleans up do_submit_bio() with it.

Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
10 years agof2fs: use f2fs_put_page to release page for uniform style
Chao Yu [Sat, 16 Nov 2013 06:15:59 +0000 (14:15 +0800)]
f2fs: use f2fs_put_page to release page for uniform style

We should use f2fs_put_page to release page for uniform style of f2fs code.

Signed-off-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
10 years agof2fs: add a tracepoint for f2fs_issue_discard
Jaegeuk Kim [Tue, 12 Nov 2013 08:01:00 +0000 (17:01 +0900)]
f2fs: add a tracepoint for f2fs_issue_discard

This patch adds a tracepoint for f2fs_issue_discard.

Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
10 years agof2fs: introduce f2fs_issue_discard() to clean up
Jaegeuk Kim [Tue, 12 Nov 2013 07:55:17 +0000 (16:55 +0900)]
f2fs: introduce f2fs_issue_discard() to clean up

Change log from v1:
 o fix 32bit drops reported by Dan Carpenter

This patch adds f2fs_issue_discard() to clean up blkdev_issue_discard() flows.

Dan carpenter reported:
"block_t is a 32 bit type and sector_t is a 64 bit type.  The upper 32
bits of the sector_t are not used because the shift will wrap."

Bug-Reported-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
10 years agof2fs: add a sysfs entry to control max_discards
Jaegeuk Kim [Tue, 12 Nov 2013 06:00:38 +0000 (15:00 +0900)]
f2fs: add a sysfs entry to control max_discards

If frequent small discards are issued to the device, the performance would
be degraded significantly.
So, this patch adds a sysfs entry to control the number of discards to be
issued during a checkpoint procedure.

By default, f2fs does not issue any small discards, which means max_discards
is zero.

Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
10 years agof2fs: add key functions for small discards
Jaegeuk Kim [Tue, 12 Nov 2013 05:49:56 +0000 (14:49 +0900)]
f2fs: add key functions for small discards

This patch adds key functions to activate the small discard feature.

Note that this procedure is conducted during the checkpoint only.

In flush_sit_entries(), when a new dirty sit entry is flushed, f2fs calls
add_discard_addrs() which searches candidates to be discarded.
The candidates should be marked *invalidated* and also previous checkpoint
recognizes it as *valid*.

At the end of a checkpoint procedure, f2fs throws discards based on the
discard entry list.

Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
10 years agof2fs: add a slab cache entry for small discards
Jaegeuk Kim [Fri, 15 Nov 2013 04:55:58 +0000 (13:55 +0900)]
f2fs: add a slab cache entry for small discards

This patch adds a slab cache entry for small discards.

Each entry consists of:

struct discard_entry {
struct list_head list; /* list head */
block_t blkaddr; /* block address to be discarded */
int len; /* # of consecutive blocks of the discard */
};

Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
10 years agof2fs: improve searching speed of __next_free_blkoff
Changman Lee [Fri, 15 Nov 2013 04:21:16 +0000 (13:21 +0900)]
f2fs: improve searching speed of __next_free_blkoff

To find a zero bit using the result of OR operation between ckpt_valid_map
and cur_valid_map is more fast than find a zero bit in each bitmap.

Signed-off-by: Changman Lee <cm224.lee@samsung.com>
[Jaegeuk Kim: adjust changed function name]
Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
10 years agof2fs: introduce __find_rev_next(_zero)_bit
Changman Lee [Fri, 15 Nov 2013 01:42:51 +0000 (10:42 +0900)]
f2fs: introduce __find_rev_next(_zero)_bit

When f2fs_set_bit is used, in a byte MSB and LSB is reversed,
in that case we can use __find_rev_next_bit or __find_rev_next_zero_bit.

Signed-off-by: Changman Lee <cm224.lee@samsung.com>
[Jaegeuk Kim: change the function names]
Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
10 years agoLinux 3.13-rc5 v3.13-rc5
Linus Torvalds [Sun, 22 Dec 2013 21:08:32 +0000 (13:08 -0800)]
Linux 3.13-rc5

10 years agoMerge tag 'fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/arm...
Linus Torvalds [Sun, 22 Dec 2013 19:13:02 +0000 (11:13 -0800)]
Merge tag 'fixes-for-linus' of git://git./linux/kernel/git/arm/arm-soc

Pull ARM SoC fixes from Olof Johansson:
 "Much smaller batch of fixes this week.

  Biggest one is a revert of an OMAP display change that removed some
  non-DT pinmux code that was still needed for 3.13 to get DSI displays
  to work.

  There's also a fix that resolves some misdescribed GPIO controller
  resources on shmobile.  The rest are mostly smaller fixes, a couple of
  MAINTAINERS updates, etc"

* tag 'fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/arm/arm-soc:
  Revert "ARM: OMAP2+: Remove legacy mux code for display.c"
  MAINTAINERS: Add keystone clock drivers
  MAINTAINERS: Add keystone git tree information
  ARM: s3c64xx: dt: Fix boot failure due to double clock initialization
  ARM: shmobile: r8a7790: Fix GPIO resources in DTS
  irqchip: renesas-intc-irqpin: Fix register bitfield shift calculation
  ARM: shmobile: lager: phy fixup needs CONFIG_PHYLIB

10 years agoMerge tag 'firewire-fix' of git://git.kernel.org/pub/scm/linux/kernel/git/ieee1394...
Linus Torvalds [Sun, 22 Dec 2013 19:11:57 +0000 (11:11 -0800)]
Merge tag 'firewire-fix' of git://git./linux/kernel/git/ieee1394/linux1394

Pull firewire fixlet from Stefan Richter:
 "A one-liner to reenable WRITE SAME over SBP-2 like in v3.8...v3.12.
  Buggy targets which could malfunction when being subjected to this
  command are already sufficiently protected by a scsi_level check in sd
  + SCSI core"

* tag 'firewire-fix' of git://git.kernel.org/pub/scm/linux/kernel/git/ieee1394/linux1394:
  firewire: sbp2: bring back WRITE SAME support

10 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/nab/target-pending
Linus Torvalds [Sun, 22 Dec 2013 19:11:20 +0000 (11:11 -0800)]
Merge git://git./linux/kernel/git/nab/target-pending

Pull SCSI target fixes from Nicholas Bellinger:
 "Mostly minor items this time around, the most notable being a FILEIO
  backend change to enforce hw_max_sectors based upon the current
  block_size to address a bug where large sized I/Os (> 1M) where being
  rejected"

* git://git.kernel.org/pub/scm/linux/kernel/git/nab/target-pending:
  qla2xxx: Fix scsi_host leak on qlt_lport_register callback failure
  target: Remove extra percpu_ref_init
  target/file: Update hw_max_sectors based on current block_size
  iser-target: Move INIT_WORK setup into isert_create_device_ib_res
  iscsi-target: Fix incorrect np->np_thread NULL assignment
  qla2xxx: Fix schedule_delayed_work() for target timeout calculations
  iser-target: fix error return code in isert_create_device_ib_res()
  iscsi-target: Fix-up all zero data-length CDBs with R/W_BIT set
  target: Remove write-only stats fields and lock from struct se_node_acl
  iscsi-target: return -EINVAL on oversized configfs parameter

10 years agoMerge git://git.kvack.org/~bcrl/aio-next
Linus Torvalds [Sun, 22 Dec 2013 19:03:49 +0000 (11:03 -0800)]
Merge git://git.kvack.org/~bcrl/aio-next

Pull AIO leak fixes from Ben LaHaise:
 "I've put these two patches plus Linus's change through a round of
  tests, and it passes millions of iterations of the aio numa
  migratepage test, as well as a number of repetitions of a few simple
  read and write tests.

  The first patch fixes the memory leak Kent introduced, while the
  second patch makes aio_migratepage() much more paranoid and robust"

* git://git.kvack.org/~bcrl/aio-next:
  aio/migratepages: make aio migrate pages sane
  aio: fix kioctx leak introduced by "aio: Fix a trinity splat"

10 years agoaio: clean up and fix aio_setup_ring page mapping
Linus Torvalds [Thu, 19 Dec 2013 20:11:12 +0000 (05:11 +0900)]
aio: clean up and fix aio_setup_ring page mapping

Since commit 36bc08cc01709 ("fs/aio: Add support to aio ring pages
migration") the aio ring setup code has used a special per-ring backing
inode for the page allocations, rather than just using random anonymous
pages.

However, rather than remembering the pages as it allocated them, it
would allocate the pages, insert them into the file mapping (dirty, so
that they couldn't be free'd), and then forget about them.  And then to
look them up again, it would mmap the mapping, and then use
"get_user_pages()" to get back an array of the pages we just created.

Now, not only is that incredibly inefficient, it also leaked all the
pages if the mmap failed (which could happen due to excessive number of
mappings, for example).

So clean it all up, making it much more straightforward.  Also remove
some left-overs of the previous (broken) mm_populate() usage that was
removed in commit d6c355c7dabc ("aio: fix race in ring buffer page
lookup introduced by page migration support") but left the pointless and
now misleading MAP_POPULATE flag around.

Tested-and-acked-by: Benjamin LaHaise <bcrl@kvack.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
10 years agoaio/migratepages: make aio migrate pages sane
Benjamin LaHaise [Sat, 21 Dec 2013 22:56:08 +0000 (17:56 -0500)]
aio/migratepages: make aio migrate pages sane

The arbitrary restriction on page counts offered by the core
migrate_page_move_mapping() code results in rather suspicious looking
fiddling with page reference counts in the aio_migratepage() operation.
To fix this, make migrate_page_move_mapping() take an extra_count parameter
that allows aio to tell the code about its own reference count on the page
being migrated.

While cleaning up aio_migratepage(), make it validate that the old page
being passed in is actually what aio_migratepage() expects to prevent
misbehaviour in the case of races.

Signed-off-by: Benjamin LaHaise <bcrl@kvack.org>
10 years agoaio: fix kioctx leak introduced by "aio: Fix a trinity splat"
Benjamin LaHaise [Sat, 21 Dec 2013 20:49:28 +0000 (15:49 -0500)]
aio: fix kioctx leak introduced by "aio: Fix a trinity splat"

e34ecee2ae791df674dfb466ce40692ca6218e43 reworked the percpu reference
counting to correct a bug trinity found.  Unfortunately, the change lead
to kioctxes being leaked because there was no final reference count to
put.  Add that reference count back in to fix things.

Signed-off-by: Benjamin LaHaise <bcrl@kvack.org>
Cc: stable@vger.kernel.org
10 years agoDon't set the INITRD_COMPRESS environment variable automatically
Linus Torvalds [Sat, 21 Dec 2013 00:52:45 +0000 (16:52 -0800)]
Don't set the INITRD_COMPRESS environment variable automatically

Commit 1bf49dd4be0b ("./Makefile: export initial ramdisk compression
config option") started setting the INITRD_COMPRESS environment variable
depending on which decompression models the kernel had available.

That is completely broken.

For example, we by default have CONFIG_RD_LZ4 enabled, and are able to
decompress such an initrd, but the user tools to *create* such an initrd
may not be availble.  So trying to tell dracut to generate an
lz4-compressed image just because we can decode such an image is
completely inappropriate.

Cc: J P <ppandit@redhat.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Jan Beulich <JBeulich@suse.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
10 years agoMerge tag 'xfs-for-linus-v3.13-rc5' of git://oss.sgi.com/xfs/xfs
Linus Torvalds [Fri, 20 Dec 2013 23:48:45 +0000 (15:48 -0800)]
Merge tag 'xfs-for-linus-v3.13-rc5' of git://oss.sgi.com/xfs/xfs

Pull xfs bugfixes from Ben Myers:
 "This contains fixes for some asserts
   related to project quotas, a memory leak, a hang when disabling group or
   project quotas before disabling user quotas, Dave's email address, several
   fixes for the alignment of file allocation to stripe unit/width geometry, a
   fix for an assertion with xfs_zero_remaining_bytes, and the behavior of
   metadata writeback in the face of IO errors.

   Details:
   - fix memory leak in xfs_dir2_node_removename
   - fix quota assertion in xfs_setattr_size
   - fix quota assertions in xfs_qm_vop_create_dqattach
   - fix for hang when disabling group and project quotas before
     disabling user quotas
   - fix Dave Chinner's email address in MAINTAINERS
   - fix for file allocation alignment
   - fix for assertion in xfs_buf_stale by removing xfsbdstrat
   - fix for alignment with swalloc mount option
   - fix for "retry forever" semantics on IO errors"

* tag 'xfs-for-linus-v3.13-rc5' of git://oss.sgi.com/xfs/xfs:
  xfs: abort metadata writeback on permanent errors
  xfs: swalloc doesn't align allocations properly
  xfs: remove xfsbdstrat error
  xfs: align initial file allocations correctly
  MAINTAINERS: fix incorrect mail address of XFS maintainer
  xfs: fix infinite loop by detaching the group/project hints from user dquot
  xfs: fix assertion failure at xfs_setattr_nonsize
  xfs: fix false assertion at xfs_qm_vop_create_dqattach
  xfs: fix memory leak in xfs_dir2_node_removename

10 years agomm: fix build of split ptlock code
Olof Johansson [Fri, 20 Dec 2013 22:28:05 +0000 (14:28 -0800)]
mm: fix build of split ptlock code

Commit 597d795a2a78 ('mm: do not allocate page->ptl dynamically, if
spinlock_t fits to long') restructures some allocators that are compiled
even if USE_SPLIT_PTLOCKS arn't used.  It results in compilation
failure:

  mm/memory.c:4282:6: error: 'struct page' has no member named 'ptl'
  mm/memory.c:4288:12: error: 'struct page' has no member named 'ptl'

Add in the missing ifdef.

Fixes: 597d795a2a78 ('mm: do not allocate page->ptl dynamically, if spinlock_t fits to long')
Signed-off-by: Olof Johansson <olof@lixom.net>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Hugh Dickins <hughd@google.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
10 years agoMerge tag 'arc-fixes-for-3.13-rc5' of git://git.kernel.org/pub/scm/linux/kernel/git...
Linus Torvalds [Fri, 20 Dec 2013 21:50:42 +0000 (13:50 -0800)]
Merge tag 'arc-fixes-for-3.13-rc5' of git://git./linux/kernel/git/vgupta/arc

Pull ARC fix from Vineet Gupta:
 "Fix busted syscall table due to unistd header inclusion issue"

* tag 'arc-fixes-for-3.13-rc5' of git://git.kernel.org/pub/scm/linux/kernel/git/vgupta/arc:
  ARC: Allow conditional multiple inclusion of uapi/asm/unistd.h

10 years agoMerge tag 'arm64-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/arm64/linux
Linus Torvalds [Fri, 20 Dec 2013 21:50:08 +0000 (13:50 -0800)]
Merge tag 'arm64-fixes' of git://git./linux/kernel/git/arm64/linux

Pull arm64 ptrace fix from Catalin Marinas.

* tag 'arm64-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/arm64/linux:
  arm64: ptrace: avoid using HW_BREAKPOINT_EMPTY for disabled events

10 years agopstore: Don't allow high traffic options on fragile devices
Luck, Tony [Wed, 18 Dec 2013 23:17:10 +0000 (15:17 -0800)]
pstore: Don't allow high traffic options on fragile devices

Some pstore backing devices use on board flash as persistent
storage. These have limited numbers of write cycles so it
is a poor idea to use them from high frequency operations.

Signed-off-by: Tony Luck <tony.luck@intel.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
10 years agoMerge tag 'dmaengine-fixes-3.13-rc4' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Fri, 20 Dec 2013 20:27:41 +0000 (12:27 -0800)]
Merge tag 'dmaengine-fixes-3.13-rc4' of git://git./linux/kernel/git/djbw/dmaengine

Pull dmaengine fixes from Dan Williams:

 - deprecation of net_dma to be removed in 3.14

 - crash regression fix in pl330 from the dmaengine_unmap rework

 - crash regression fix for any channel running raid ops without
   CONFIG_ASYNC_TX_DMA from dmaengine_unmap

 - memory leak regression in mv_xor from dmaengine_unmap

 - build warning regressions in mv_xor, fsldma, ppc4xx, txx9, and
   at_hdmac from dmaengine_unmap

 - sleep in atomic regression in dma_async_memcpy_pg_to_pg

 - new fix in mv_xor for handling channel initialization failures

* tag 'dmaengine-fixes-3.13-rc4' of git://git.kernel.org/pub/scm/linux/kernel/git/djbw/dmaengine:
  net_dma: mark broken
  dma: pl330: ensure DMA descriptors are zero-initialised
  dmaengine: fix sleep in atomic
  dmaengine: mv_xor: fix oops when channels fail to initialise
  dma: mv_xor: Use dmaengine_unmap_data for the self-tests
  dmaengine: fix enable for high order unmap pools
  dma: fix build warnings in txx9
  dmatest: fix build warning on mips
  dma: fix fsldma build warnings
  dma: fix build warnings in ppc4xx
  dmaengine: at_hdmac: remove unused function
  dma: mv_xor: remove mv_desc_get_dest_addr()

10 years agoMerge tag 'for-linus' of git://git.kernel.org/pub/scm/virt/kvm/kvm
Linus Torvalds [Fri, 20 Dec 2013 20:26:54 +0000 (12:26 -0800)]
Merge tag 'for-linus' of git://git./virt/kvm/kvm

Pull KVM fixes from Paolo Bonzini:
 "The PPC folks had a large amount of changes queued for 3.13, and now
  they are fixing the bugs"

* tag 'for-linus' of git://git.kernel.org/pub/scm/virt/kvm/kvm:
  KVM: PPC: Book3S HV: Don't drop low-order page address bits
  powerpc: book3s: kvm: Don't abuse host r2 in exit path
  powerpc/kvm/booke: Fix build break due to stack frame size warning
  KVM: PPC: Book3S: PR: Enable interrupts earlier
  KVM: PPC: Book3S: PR: Make svcpu -> vcpu store preempt savvy
  KVM: PPC: Book3S: PR: Export kvmppc_copy_to|from_svcpu
  KVM: PPC: Book3S: PR: Don't clobber our exit handler id
  powerpc: kvm: fix rare but potential deadlock scene
  KVM: PPC: Book3S HV: Take SRCU read lock around kvm_read_guest() call
  KVM: PPC: Book3S HV: Make tbacct_lock irq-safe
  KVM: PPC: Book3S HV: Refine barriers in guest entry/exit
  KVM: PPC: Book3S HV: Fix physical address calculations

10 years agomm: do not allocate page->ptl dynamically, if spinlock_t fits to long
Kirill A. Shutemov [Fri, 20 Dec 2013 11:35:58 +0000 (13:35 +0200)]
mm: do not allocate page->ptl dynamically, if spinlock_t fits to long

In struct page we have enough space to fit long-size page->ptl there,
but we use dynamically-allocated page->ptl if size(spinlock_t) is larger
than sizeof(int).

It hurts 64-bit architectures with CONFIG_GENERIC_LOCKBREAK, where
sizeof(spinlock_t) == 8, but it easily fits into struct page.

Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Acked-by: Hugh Dickins <hughd@google.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
10 years agomm: page_alloc: revert NUMA aspect of fair allocation policy
Johannes Weiner [Fri, 20 Dec 2013 14:54:12 +0000 (14:54 +0000)]
mm: page_alloc: revert NUMA aspect of fair allocation policy

Commit 81c0a2bb515f ("mm: page_alloc: fair zone allocator policy") meant
to bring aging fairness among zones in system, but it was overzealous
and badly regressed basic workloads on NUMA systems.

Due to the way kswapd and page allocator interacts, we still want to
make sure that all zones in any given node are used equally for all
allocations to maximize memory utilization and prevent thrashing on the
highest zone in the node.

While the same principle applies to NUMA nodes - memory utilization is
obviously improved by spreading allocations throughout all nodes -
remote references can be costly and so many workloads prefer locality
over memory utilization.  The original change assumed that
zone_reclaim_mode would be a good enough predictor for that, but it
turned out to be as indicative as a coin flip.

Revert the NUMA aspect of the fairness until we can find a proper way to
make it configurable and agree on a sane default.

Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Reviewed-by: Michal Hocko <mhocko@suse.cz>
Signed-off-by: Mel Gorman <mgorman@suse.de>
Cc: <stable@kernel.org> # 3.12
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
10 years agoRevert "mm: page_alloc: exclude unreclaimable allocations from zone fairness policy"
Mel Gorman [Fri, 20 Dec 2013 14:54:11 +0000 (14:54 +0000)]
Revert "mm: page_alloc: exclude unreclaimable allocations from zone fairness policy"

This reverts commit 73f038b863df.  The NUMA behaviour of this patch is
less than ideal.  An alternative approch is to interleave allocations
only within local zones which is implemented in the next patch.

Cc: stable@vger.kernel.org
Signed-off-by: Mel Gorman <mgorman@suse.de>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
10 years agomm: Fix NULL pointer dereference in madvise(MADV_WILLNEED) support
Kirill A. Shutemov [Fri, 20 Dec 2013 13:10:03 +0000 (15:10 +0200)]
mm: Fix NULL pointer dereference in madvise(MADV_WILLNEED) support

Sasha Levin found a NULL pointer dereference that is due to a missing
page table lock, which in turn is due to the pmd entry in question being
a transparent huge-table entry.

The code - introduced in commit 1998cc048901 ("mm: make
madvise(MADV_WILLNEED) support swap file prefetch") - correctly checks
for this situation using pmd_none_or_trans_huge_or_clear_bad(), but it
turns out that that function doesn't work correctly.

pmd_none_or_trans_huge_or_clear_bad() expected that pmd_bad() would
trigger if the transparent hugepage bit was set, but it doesn't do that
if pmd_numa() is also set. Note that the NUMA bit only gets set on real
NUMA machines, so people trying to reproduce this on most normal
development systems would never actually trigger this.

Fix it by removing the very subtle (and subtly incorrect) expectation,
and instead just checking pmd_trans_huge() explicitly.

Reported-by: Sasha Levin <sasha.levin@oracle.com>
Acked-by: Andrea Arcangeli <aarcange@redhat.com>
[ Additionally remove the now stale test for pmd_trans_huge() inside the
  pmd_bad() case - Linus ]
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
10 years agoMerge tag 'renesas-fixes-for-v3.13' of git://git.kernel.org/pub/scm/linux/kernel...
Kevin Hilman [Fri, 20 Dec 2013 19:27:12 +0000 (11:27 -0800)]
Merge tag 'renesas-fixes-for-v3.13' of git://git./linux/kernel/git/horms/renesas into fixes

From Simon Horman:
Renesas ARM based SoC fixes for v3.13

* r8a7790 (R-Car H1) SoC
  - Correct GPIO resources in DT.

    This problem has been present since GPIOs were added to the r8a7790 SoC
    by f98e10c88aa95bf7 ("ARM: shmobile: r8a7790: Add GPIO controller
    devices to device tree") in v3.12-rc1.

* irqchip renesas-intc-irqpin
  - Correct register bitfield shift calculation

    This bug has been present since the renesas-intc-irqpin driver was
    introduced by 443580486e3b9657 ("irqchip: Renesas INTC External IRQ pin
    driver") in v3.10-rc1

* Lager board
  - Do not build the phy fixup unless CONFIG_PHYLIB is enabled

    This problem was introduced by 48c8b96f21817aad

* tag 'renesas-fixes-for-v3.13' of git://git.kernel.org/pub/scm/linux/kernel/git/horms/renesas:
  ARM: shmobile: r8a7790: Fix GPIO resources in DTS
  irqchip: renesas-intc-irqpin: Fix register bitfield shift calculation
  ARM: shmobile: lager: phy fixup needs CONFIG_PHYLIB

Signed-off-by: Kevin Hilman <khilman@linaro.org>
10 years agoMerge tag 'signed-for-3.13' of git://github.com/agraf/linux-2.6 into kvm-master
Paolo Bonzini [Fri, 20 Dec 2013 18:13:58 +0000 (19:13 +0100)]
Merge tag 'signed-for-3.13' of git://github.com/agraf/linux-2.6 into kvm-master

Patch queue for 3.13 - 2013-12-18

This fixes some grave issues we've only found after 3.13-rc1:

  - Make the modularized HV/PR book3s kvm work well as modules
  - Fix some race conditions
  - Fix compilation with certain compilers (booke)
  - Fix THP for book3s_hv
  - Fix preemption for book3s_pr

Alexander Graf (4):
      KVM: PPC: Book3S: PR: Don't clobber our exit handler id
      KVM: PPC: Book3S: PR: Export kvmppc_copy_to|from_svcpu
      KVM: PPC: Book3S: PR: Make svcpu -> vcpu store preempt savvy
      KVM: PPC: Book3S: PR: Enable interrupts earlier

Aneesh Kumar K.V (1):
      powerpc: book3s: kvm: Don't abuse host r2 in exit path

Paul Mackerras (5):
      KVM: PPC: Book3S HV: Fix physical address calculations
      KVM: PPC: Book3S HV: Refine barriers in guest entry/exit
      KVM: PPC: Book3S HV: Make tbacct_lock irq-safe
      KVM: PPC: Book3S HV: Take SRCU read lock around kvm_read_guest() call
      KVM: PPC: Book3S HV: Don't drop low-order page address bits

Scott Wood (1):
      powerpc/kvm/booke: Fix build break due to stack frame size warning

pingfan liu (1):
      powerpc: kvm: fix rare but potential deadlock scene

10 years agoMerge tag 'stable/for-linus-3.13-rc4-tag' of git://git.kernel.org/pub/scm/linux/kerne...
Linus Torvalds [Fri, 20 Dec 2013 17:34:54 +0000 (09:34 -0800)]
Merge tag 'stable/for-linus-3.13-rc4-tag' of git://git./linux/kernel/git/xen/tip

Pull Xen bugfixes from Konrad Rzeszutek Wilk:
 - Fix balloon driver for auto-translate guests (PVHVM, ARM) to not use
   scratch pages.
 - Fix block API header for ARM32 and ARM64 to have proper layout
 - On ARM when mapping guests, stick on PTE_SPECIAL
 - When using SWIOTLB under ARM, don't call swiotlb functions twice
 - When unmapping guests memory and if we fail, don't return pages which
   failed to be unmapped.
 - Grant driver was using the wrong address on ARM.

* tag 'stable/for-linus-3.13-rc4-tag' of git://git.kernel.org/pub/scm/linux/kernel/git/xen/tip:
  xen/balloon: Seperate the auto-translate logic properly (v2)
  xen/block: Correctly define structures in public headers on ARM32 and ARM64
  arm: xen: foreign mapping PTEs are special.
  xen/arm64: do not call the swiotlb functions twice
  xen: privcmd: do not return pages which we have failed to unmap
  XEN: Grant table address, xen_hvm_resume_frames, is a phys_addr not a pfn

10 years agoMerge tag 'trace-fixes-v3.13-rc2' of git://git.kernel.org/pub/scm/linux/kernel/git...
Linus Torvalds [Fri, 20 Dec 2013 17:32:30 +0000 (09:32 -0800)]
Merge tag 'trace-fixes-v3.13-rc2' of git://git./linux/kernel/git/rostedt/linux-trace

Pull ftrace fix from Steven Rostedt:
 "This fixes a long standing bug in the ftrace profiler.  The problem is
  that the profiler only initializes the online CPUs, and not possible
  CPUs.  This causes issues if the user takes CPUs online or offline
  while the profiler is running.

  If we online a CPU after starting the profiler, we lose all the trace
  information on the CPU going online.

  If we offline a CPU after running a test and start a new test, it will
  not clear the old data from that CPU.

  This bug causes incorrect data to be reported to the user if they
  online or offline CPUs during the profiling"

* tag 'trace-fixes-v3.13-rc2' of git://git.kernel.org/pub/scm/linux/kernel/git/rostedt/linux-trace:
  ftrace: Initialize the ftrace profiler for each possible cpu

10 years agoMerge tag 'omap-for-v3.13/display-fix' of git://git.kernel.org/pub/scm/linux/kernel...
Kevin Hilman [Fri, 20 Dec 2013 16:30:50 +0000 (08:30 -0800)]
Merge tag 'omap-for-v3.13/display-fix' of git://git./linux/kernel/git/tmlind/linux-omap into fixes

I accidentally removed some mux code for omap4 that I thought was
dead code as omap4 has been booting with device tree only since
v3.10. Turns out I also removed some display related mux code,
so let's revert that except for the dead code parts.

* tag 'omap-for-v3.13/display-fix' of git://git.kernel.org/pub/scm/linux/kernel/git/tmlind/linux-omap: (439 commits)
  Revert "ARM: OMAP2+: Remove legacy mux code for display.c"
  +Linux 3.13-rc4

10 years agoMerge tag 'keystone/maintainer-file' of git://git.kernel.org/pub/scm/linux/kernel...
Kevin Hilman [Thu, 19 Dec 2013 23:12:23 +0000 (15:12 -0800)]
Merge tag 'keystone/maintainer-file' of git://git./linux/kernel/git/ssantosh/linux-keystone into fixes

From Santosh Shilimkar:
Couple of updates to MAINTAINERS file for Keystone

- Add git tree information
- Add clock drivers entry

* tag 'keystone/maintainer-file' of git://git.kernel.org/pub/scm/linux/kernel/git/ssantosh/linux-keystone:
  MAINTAINERS: Add keystone clock drivers
  MAINTAINERS: Add keystone git tree information

Signed-off-by: Kevin Hilman <khilman@linaro.org>
10 years agoqla2xxx: Fix scsi_host leak on qlt_lport_register callback failure
Nicholas Bellinger [Tue, 17 Dec 2013 09:51:22 +0000 (01:51 -0800)]
qla2xxx: Fix scsi_host leak on qlt_lport_register callback failure

This patch fixes a possible scsi_host reference leak in qlt_lport_register(),
when a non zero return from the passed (*callback) does not call drop the
local reference via scsi_host_put() before returning.

This currently does not effect existing tcm_qla2xxx code as the passed callback
will never fail, but fix this up regardless for future code.

Cc: Chad Dupuis <chad.dupuis@qlogic.com>
Signed-off-by: Nicholas Bellinger <nab@linux-iscsi.org>
10 years agotarget: Remove extra percpu_ref_init
Andy Grover [Tue, 26 Nov 2013 19:49:24 +0000 (11:49 -0800)]
target: Remove extra percpu_ref_init

lun->lun_ref is also initialized in core_tpg_post_addlun, so it doesn't
need to be done in core_tpg_setup_virtual_lun0.

(nab: Drop left-over percpu_ref_cancel_init in failure path)

Signed-off-by: Andy Grover <agrover@redhat.com>
Signed-off-by: Nicholas Bellinger <nab@linux-iscsi.org>
10 years agoarm64: ptrace: avoid using HW_BREAKPOINT_EMPTY for disabled events
Will Deacon [Tue, 17 Dec 2013 17:09:08 +0000 (17:09 +0000)]
arm64: ptrace: avoid using HW_BREAKPOINT_EMPTY for disabled events

Commit 8f34a1da35ae ("arm64: ptrace: use HW_BREAKPOINT_EMPTY type for
disabled breakpoints") fixed an issue with GDB trying to zero breakpoint
control registers. The problem there is that the arch hw_breakpoint code
will attempt to create a (disabled), execute breakpoint of length 0.

This will fail validation and report unexpected failure to GDB. To avoid
this, we treated disabled breakpoints as HW_BREAKPOINT_EMPTY, but that
seems to have broken with recent kernels, causing watchpoints to be
treated as TYPE_INST in the core code and returning ENOSPC for any
further breakpoints.

This patch fixes the problem by prioritising the `enable' field of the
breakpoint: if it is cleared, we simply update the perf_event_attr to
indicate that the thing is disabled and don't bother changing either the
type or the length. This reinforces the behaviour that the breakpoint
control register is essentially read-only apart from the enable bit
when disabling a breakpoint.

Cc: <stable@vger.kernel.org>
Reported-by: Aaron Liu <liucy214@gmail.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
Signed-off-by: Catalin Marinas <catalin.marinas@arm.com>
10 years agoMerge branch 'sched-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Thu, 19 Dec 2013 17:11:22 +0000 (09:11 -0800)]
Merge branch 'sched-urgent-for-linus' of git://git./linux/kernel/git/tip/tip

Pull scheduler fixes from Ingo Molnar:
 "An RT group-scheduling fix and the sched-domains topology setup fix
  from Mel"

* 'sched-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  sched/rt: Fix rq's cpupri leak while enqueue/dequeue child RT entities
  sched: Assign correct scheduling domain to 'sd_llc'

10 years agoMerge branch 'perf-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Thu, 19 Dec 2013 17:10:46 +0000 (09:10 -0800)]
Merge branch 'perf-urgent-for-linus' of git://git./linux/kernel/git/tip/tip

Pull perf fixes from Ingo Molnar:
 "An ABI documentation fix, and a mixed-PMU perf-info-corruption fix"

* 'perf-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  perf: Document the new transaction sample type
  perf: Disable all pmus on unthrottling and rescheduling

10 years agoMerge tag 'sound-3.13-rc5' of git://git.kernel.org/pub/scm/linux/kernel/git/tiwai...
Linus Torvalds [Thu, 19 Dec 2013 17:10:10 +0000 (09:10 -0800)]
Merge tag 'sound-3.13-rc5' of git://git./linux/kernel/git/tiwai/sound

Pull sound fixes from Takashi Iwai:
 "We have a bit more changes than usual in ASoC here, as it was slipped
  from the previous update.  There are one minr ASoC PCM code fix and
  ASoC dmaengine fix, in addition of a collection of small ASoC driver
  fixes.  The rest are a couple of HD-audio stable fixups, and a
  long-standing fix for the paused stream handling.

  So, all commits look not scary (and hopefully won't give you
  disastrous holiday season)"

* tag 'sound-3.13-rc5' of git://git.kernel.org/pub/scm/linux/kernel/git/tiwai/sound:
  ALSA: hda - Add Dell headset detection quirk for one more laptop model
  ASoC: wm8904: fix DSP mode B configuration
  ASoC: wm_adsp: Add small delay while polling DSP RAM start
  ALSA: Add SNDRV_PCM_STATE_PAUSED case in wait_for_avail function
  ASoC: kirkwood: Fix the CPU DAI rates
  ASoC: wm5110: Correct HPOUT3 DAPM route typo
  ALSA: hda - Add Dell headset detection quirk for three laptop models
  ALSA: hda - Add enable_msi=0 workaround for four HP machines
  ASoC: don't leak on error in snd_dmaengine_pcm_register
  ASoC: fsl: imx-wm8962: Don't update bias_level in machine driver
  ASoC: tegra: fix uninitialized variables in set_fmt
  ASoC: wm8962: Enable SYSCLK provisonally before fetching generated DSPCLK_DIV
  ASoC: sam9x5_wm8731: change to work in DSP A mode
  ASoC: atmel_ssc_dai: add dai trigger ops
  ASoC: soc-pcm: Use valid condition for snd_soc_dai_digital_mute() in hw_free()

10 years agoARC: Allow conditional multiple inclusion of uapi/asm/unistd.h
Vineet Gupta [Thu, 19 Dec 2013 13:25:58 +0000 (18:55 +0530)]
ARC: Allow conditional multiple inclusion of uapi/asm/unistd.h

Commit 97bc386fc12d "ARC: Add guard macro to uapi/asm/unistd.h"
inhibited multiple inclusion of ARCH unistd.h. This however hosed the system
since Generic syscall table generator relies on it being included twice,
and in lack-of an empty table was emitted by C preprocessor.

Fix that by allowing one exception to rule for the special case (just
like Xtensa)

Suggested-by: Chen Gang <gang.chen.5i5j@gmail.com>
Signed-off-by: Vineet Gupta <vgupta@synopsys.com>
10 years agoMerge tag 'asoc-v3.13-rc4' of git://git.kernel.org/pub/scm/linux/kernel/git/broonie...
Takashi Iwai [Thu, 19 Dec 2013 11:22:11 +0000 (12:22 +0100)]
Merge tag 'asoc-v3.13-rc4' of git://git./linux/kernel/git/broonie/sound into for-linus

ASoC: Fixes for v3.13

The fixes here are all driver specific ones, none of which particularly
stand out but all of which are useful to users of those drivers.

10 years agoMerge remote-tracking branches 'asoc/fix/adsp', 'asoc/fix/arizona', 'asoc/fix/atmel...
Mark Brown [Thu, 19 Dec 2013 10:25:27 +0000 (10:25 +0000)]
Merge remote-tracking branches 'asoc/fix/adsp', 'asoc/fix/arizona', 'asoc/fix/atmel', 'asoc/fix/fsl', 'asoc/fix/kirkwood', 'asoc/fix/tegra', 'asoc/fix/wm8904' and 'asoc/fix/wm8962' into asoc-linus

10 years agoMerge remote-tracking branch 'asoc/fix/dma' into asoc-linus
Mark Brown [Thu, 19 Dec 2013 10:25:27 +0000 (10:25 +0000)]
Merge remote-tracking branch 'asoc/fix/dma' into asoc-linus

10 years agoMerge remote-tracking branch 'asoc/fix/core' into asoc-linus
Mark Brown [Thu, 19 Dec 2013 10:25:26 +0000 (10:25 +0000)]
Merge remote-tracking branch 'asoc/fix/core' into asoc-linus

10 years agotarget/file: Update hw_max_sectors based on current block_size
Nicholas Bellinger [Thu, 12 Dec 2013 20:24:11 +0000 (12:24 -0800)]
target/file: Update hw_max_sectors based on current block_size

This patch allows FILEIO to update hw_max_sectors based on the current
max_bytes_per_io.  This is required because vfs_[writev,readv]() can accept
a maximum of 2048 iovecs per call, so the enforced hw_max_sectors really
needs to be calculated based on block_size.

This addresses a >= v3.5 bug where block_size=512 was rejecting > 1M
sized I/O requests, because FD_MAX_SECTORS was hardcoded to 2048 for
the block_size=4096 case.

(v2: Use max_bytes_per_io instead of ->update_hw_max_sectors)

Reported-by: Henrik Goldman <hg@x-formation.com>
Cc: <stable@vger.kernel.org> #3.5+
Signed-off-by: Nicholas Bellinger <nab@linux-iscsi.org>
10 years agoiser-target: Move INIT_WORK setup into isert_create_device_ib_res
Nicholas Bellinger [Thu, 12 Dec 2013 00:20:13 +0000 (16:20 -0800)]
iser-target: Move INIT_WORK setup into isert_create_device_ib_res

This patch moves INIT_WORK setup for cq_desc->cq_[rx,tx]_work into
isert_create_device_ib_res(), instead of being done each callback
invocation in isert_cq_[rx,tx]_callback().

This also fixes a 'INFO: trying to register non-static key' warning
when cancel_work_sync() is called before INIT_WORK has setup the
struct work_struct.

Reported-by: Or Gerlitz <ogerlitz@mellanox.com>
Cc: <stable@vger.kernel.org> #3.12+
Signed-off-by: Nicholas Bellinger <nab@linux-iscsi.org>
10 years agoiscsi-target: Fix incorrect np->np_thread NULL assignment
Nicholas Bellinger [Wed, 11 Dec 2013 23:45:32 +0000 (15:45 -0800)]
iscsi-target: Fix incorrect np->np_thread NULL assignment

When shutting down a target there is a race condition between
iscsit_del_np() and __iscsi_target_login_thread().
The latter sets the thread pointer to NULL, and the former
tries to issue kthread_stop() on that pointer without any
synchronization.

This patch moves the np->np_thread NULL assignment into
iscsit_del_np(), after kthread_stop() has completed. It also
removes the signal_pending() + np_state check, and only
exits when kthread_should_stop() is true.

Reported-by: Hannes Reinecke <hare@suse.de>
Cc: <stable@vger.kernel.org> #3.12+
Signed-off-by: Nicholas Bellinger <nab@linux-iscsi.org>
10 years agoMerge branch 'akpm' (incoming from Andrew)
Linus Torvalds [Thu, 19 Dec 2013 03:05:00 +0000 (19:05 -0800)]
Merge branch 'akpm' (incoming from Andrew)

Merge patches from Andrew Morton:
 "23 fixes and a MAINTAINERS update"

* emailed patches from Andrew Morton <akpm@linux-foundation.org>: (24 commits)
  mm/hugetlb: check for pte NULL pointer in __page_check_address()
  fix build with make 3.80
  mm/mempolicy: fix !vma in new_vma_page()
  MAINTAINERS: add Davidlohr as GPT maintainer
  mm/memory-failure.c: recheck PageHuge() after hugetlb page migrate successfully
  mm/compaction: respect ignore_skip_hint in update_pageblock_skip
  mm/mempolicy: correct putback method for isolate pages if failed
  mm: add missing dependency in Kconfig
  sh: always link in helper functions extracted from libgcc
  mm: page_alloc: exclude unreclaimable allocations from zone fairness policy
  mm: numa: defer TLB flush for THP migration as long as possible
  mm: numa: guarantee that tlb_flush_pending updates are visible before page table updates
  mm: fix TLB flush race between migration, and change_protection_range
  mm: numa: avoid unnecessary disruption of NUMA hinting during migration
  mm: numa: clear numa hinting information on mprotect
  sched: numa: skip inaccessible VMAs
  mm: numa: avoid unnecessary work on the failure path
  mm: numa: ensure anon_vma is locked to prevent parallel THP splits
  mm: numa: do not clear PTE for pte_numa update
  mm: numa: do not clear PMD during PTE update scan
  ...

10 years agomm/hugetlb: check for pte NULL pointer in __page_check_address()
Jianguo Wu [Thu, 19 Dec 2013 01:08:59 +0000 (17:08 -0800)]
mm/hugetlb: check for pte NULL pointer in __page_check_address()

In __page_check_address(), if address's pud is not present,
huge_pte_offset() will return NULL, we should check the return value.

Signed-off-by: Jianguo Wu <wujianguo@huawei.com>
Cc: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: qiuxishi <qiuxishi@huawei.com>
Cc: Hanjun Guo <guohanjun@huawei.com>
Acked-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
10 years agofix build with make 3.80
Jan Beulich [Thu, 19 Dec 2013 01:08:57 +0000 (17:08 -0800)]
fix build with make 3.80

According to Documentation/Changes, make 3.80 is still being supported
for building the kernel, hence make files must not make (unconditional)
use of features introduced only in newer versions.

Commit 1bf49dd4be0b ("./Makefile: export initial ramdisk compression
config option") however introduced "else ifeq" constructs which make
3.80 doesn't understand.  Replace the logic there with more conventional
(in the kernel build infrastructure) list constructs (except that the
list here is intentionally limited to exactly one element).

Signed-off-by: Jan Beulich <jbeulich@suse.com>
Cc: P J P <ppandit@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
10 years agomm/mempolicy: fix !vma in new_vma_page()
Wanpeng Li [Thu, 19 Dec 2013 01:08:56 +0000 (17:08 -0800)]
mm/mempolicy: fix !vma in new_vma_page()

BUG_ON(!vma) assumption is introduced by commit 0bf598d863e3 ("mbind:
add BUG_ON(!vma) in new_vma_page()"), however, even if

    address = __vma_address(page, vma);

and

    vma->start < address < vma->end

page_address_in_vma() may still return -EFAULT because of many other
conditions in it.  As a result the while loop in new_vma_page() may end
with vma=NULL.

This patch revert the commit and also fix the potential dereference NULL
pointer reported by Dan.

   http://marc.info/?l=linux-mm&m=137689530323257&w=2

  kernel BUG at mm/mempolicy.c:1204!
  invalid opcode: 0000 [#1] PREEMPT SMP DEBUG_PAGEALLOC
  CPU: 3 PID: 7056 Comm: trinity-child3 Not tainted 3.13.0-rc3+ #2
  task: ffff8801ca5295d0 ti: ffff88005ab20000 task.ti: ffff88005ab20000
  RIP: new_vma_page+0x70/0x90
  RSP: 0000:ffff88005ab21db0  EFLAGS: 00010246
  RAX: fffffffffffffff2 RBX: 0000000000000000 RCX: 0000000000000000
  RDX: 0000000008040075 RSI: ffff8801c3d74600 RDI: ffffea00079a8b80
  RBP: ffff88005ab21dc8 R08: 0000000000000004 R09: 0000000000000000
  R10: 0000000000000000 R11: 0000000000000000 R12: fffffffffffffff2
  R13: ffffea00079a8b80 R14: 0000000000400000 R15: 0000000000400000

  FS:  00007ff49c6f4740(0000) GS:ffff880244e00000(0000) knlGS:0000000000000000
  CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
  CR2: 00007ff49c68f994 CR3: 000000005a205000 CR4: 00000000001407e0
  DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
  DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
  Stack:
   ffffea00079a8b80 ffffea00079a8bc0 ffffea00079a8ba0 ffff88005ab21e50
   ffffffff811adc7a 0000000000000000 ffff8801ca5295d0 0000000464e224f8
   0000000000000000 0000000000000002 0000000000000000 ffff88020ce75c00
  Call Trace:
    migrate_pages+0x12a/0x850
    SYSC_mbind+0x513/0x6a0
    SyS_mbind+0xe/0x10
    ia32_do_call+0x13/0x13
  Code: 85 c0 75 2f 4c 89 e1 48 89 da 31 f6 bf da 00 02 00 65 44 8b 04 25 08 f7 1c 00 e8 ec fd ff ff 5b 41 5c 41 5d 5d c3 0f 1f 44 00 00 <0f> 0b 66 0f 1f 44 00 00 4c 89 e6 48 89 df ba 01 00 00 00 e8 48
  RIP  [<ffffffff8119f200>] new_vma_page+0x70/0x90
   RSP <ffff88005ab21db0>

Signed-off-by: Wanpeng Li <liwanp@linux.vnet.ibm.com>
Reported-by: Dave Jones <davej@redhat.com>
Reported-by: Sasha Levin <sasha.levin@oracle.com>
Reviewed-by: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com>
Reviewed-by: Bob Liu <bob.liu@oracle.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
10 years agoMAINTAINERS: add Davidlohr as GPT maintainer
Davidlohr Bueso [Thu, 19 Dec 2013 01:08:55 +0000 (17:08 -0800)]
MAINTAINERS: add Davidlohr as GPT maintainer

Add a new entry for the GPT standard.  Any future changes will now be
routed through linux-efi.

Signed-off-by: Davidlohr Bueso <davidlohr@hp.com>
Acked-by: Matt Fleming <matt.fleming@intel.com>
Cc: Jens Axboe <axboe@kernel.dk>
Acked-by: Matt Domsch <Matt_Domsch@dell.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
10 years agomm/memory-failure.c: recheck PageHuge() after hugetlb page migrate successfully
Jianguo Wu [Thu, 19 Dec 2013 01:08:54 +0000 (17:08 -0800)]
mm/memory-failure.c: recheck PageHuge() after hugetlb page migrate successfully

After a successful hugetlb page migration by soft offline, the source
page will either be freed into hugepage_freelists or buddy(over-commit
page).  If page is in buddy, page_hstate(page) will be NULL.  It will
hit a NULL pointer dereference in dequeue_hwpoisoned_huge_page().

  BUG: unable to handle kernel NULL pointer dereference at 0000000000000058
  IP: [<ffffffff81163761>] dequeue_hwpoisoned_huge_page+0x131/0x1d0
  PGD c23762067 PUD c24be2067 PMD 0
  Oops: 0000 [#1] SMP

So check PageHuge(page) after call migrate_pages() successfully.

Signed-off-by: Jianguo Wu <wujianguo@huawei.com>
Tested-by: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com>
Reviewed-by: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
10 years agomm/compaction: respect ignore_skip_hint in update_pageblock_skip
Joonsoo Kim [Thu, 19 Dec 2013 01:08:52 +0000 (17:08 -0800)]
mm/compaction: respect ignore_skip_hint in update_pageblock_skip

update_pageblock_skip() only fits to compaction which tries to isolate
by pageblock unit.  If isolate_migratepages_range() is called by CMA, it
try to isolate regardless of pageblock unit and it don't reference
get_pageblock_skip() by ignore_skip_hint.  We should also respect it on
update_pageblock_skip() to prevent from setting the wrong information.

Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Reviewed-by: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com>
Reviewed-by: Wanpeng Li <liwanp@linux.vnet.ibm.com>
Cc: Christoph Lameter <cl@linux.com>
Cc: Rafael Aquini <aquini@redhat.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Wanpeng Li <liwanp@linux.vnet.ibm.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Rik van Riel <riel@redhat.com>
Cc: Zhang Yanfei <zhangyanfei@cn.fujitsu.com>
Cc: <stable@vger.kernel.org> [3.7+]
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
10 years agomm/mempolicy: correct putback method for isolate pages if failed
Joonsoo Kim [Thu, 19 Dec 2013 01:08:51 +0000 (17:08 -0800)]
mm/mempolicy: correct putback method for isolate pages if failed

queue_pages_range() isolates hugetlbfs pages and putback_lru_pages()
can't handle these.  We should change it to putback_movable_pages().

Naoya said that it is worth going into stable, because it can break
in-use hugepage list.

Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Acked-by: Rafael Aquini <aquini@redhat.com>
Reviewed-by: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com>
Reviewed-by: Wanpeng Li <liwanp@linux.vnet.ibm.com>
Cc: Christoph Lameter <cl@linux.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Wanpeng Li <liwanp@linux.vnet.ibm.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Rik van Riel <riel@redhat.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Zhang Yanfei <zhangyanfei@cn.fujitsu.com>
Cc: <stable@vger.kernel.org> [3.12.x]
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
10 years agomm: add missing dependency in Kconfig
Sima Baymani [Thu, 19 Dec 2013 01:08:49 +0000 (17:08 -0800)]
mm: add missing dependency in Kconfig

Eliminate the following (rand)config warning by adding missing PROC_FS
dependency:

  warning: (HWPOISON_INJECT && MEM_SOFT_DIRTY) selects PROC_PAGE_MONITOR which has unmet direct dependencies (PROC_FS && MMU)

Signed-off-by: Sima Baymani <sima.baymani@gmail.com>
Suggested-by: David Rientjes <rientjes@google.com>
Acked-by: David Rientjes <rientjes@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
10 years agosh: always link in helper functions extracted from libgcc
Geert Uytterhoeven [Thu, 19 Dec 2013 01:08:48 +0000 (17:08 -0800)]
sh: always link in helper functions extracted from libgcc

E.g. landisk_defconfig, which has CONFIG_NTFS_FS=m:

  ERROR: "__ashrdi3" [fs/ntfs/ntfs.ko] undefined!

For "lib-y", if no symbols in a compilation unit are referenced by other
units, the compilation unit will not be included in vmlinux.  This
breaks modules that do reference those symbols.

Use "obj-y" instead to fix this.

http://kisskb.ellerman.id.au/kisskb/buildresult/8838077/

This doesn't fix all cases. There are others, e.g. udivsi3.
This is also not limited to sh, many architectures handle this in the
same way.

A simple solution is to unconditionally include all helper functions.
A more complex solution is to make the choice of "lib-y" or "obj-y" depend
on CONFIG_MODULES:

  obj-$(CONFIG_MODULES) += ...
  lib-y($CONFIG_MODULES) += ...

Signed-off-by: Geert Uytterhoeven <geert@linux-m68k.org>
Cc: Paul Mundt <lethal@linux-sh.org>
Tested-by: Nobuhiro Iwamatsu <nobuhiro.iwamatsu.yj@renesas.com>
Reviewed-by: Nobuhiro Iwamatsu <nobuhiro.iwamatsu.yj@renesas.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
10 years agomm: page_alloc: exclude unreclaimable allocations from zone fairness policy
Johannes Weiner [Thu, 19 Dec 2013 01:08:47 +0000 (17:08 -0800)]
mm: page_alloc: exclude unreclaimable allocations from zone fairness policy

Dave Hansen noted a regression in a microbenchmark that loops around
open() and close() on an 8-node NUMA machine and bisected it down to
commit 81c0a2bb515f ("mm: page_alloc: fair zone allocator policy").
That change forces the slab allocations of the file descriptor to spread
out to all 8 nodes, causing remote references in the page allocator and
slab.

The round-robin policy is only there to provide fairness among memory
allocations that are reclaimed involuntarily based on pressure in each
zone.  It does not make sense to apply it to unreclaimable kernel
allocations that are freed manually, in this case instantly after the
allocation, and incur the remote reference costs twice for no reason.

Only round-robin allocations that are usually freed through page reclaim
or slab shrinking.

Bisected by Dave Hansen.

Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Cc: Dave Hansen <dave.hansen@intel.com>
Cc: Mel Gorman <mgorman@suse.de>
Reviewed-by: Rik van Riel <riel@redhat.com>
Cc: <stable@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
10 years agomm: numa: defer TLB flush for THP migration as long as possible
Mel Gorman [Thu, 19 Dec 2013 01:08:46 +0000 (17:08 -0800)]
mm: numa: defer TLB flush for THP migration as long as possible

THP migration can fail for a variety of reasons.  Avoid flushing the TLB
to deal with THP migration races until the copy is ready to start.

Signed-off-by: Mel Gorman <mgorman@suse.de>
Reviewed-by: Rik van Riel <riel@redhat.com>
Cc: Alex Thorlton <athorlton@sgi.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
10 years agomm: numa: guarantee that tlb_flush_pending updates are visible before page table...
Mel Gorman [Thu, 19 Dec 2013 01:08:45 +0000 (17:08 -0800)]
mm: numa: guarantee that tlb_flush_pending updates are visible before page table updates

According to documentation on barriers, stores issued before a LOCK can
complete after the lock implying that it's possible tlb_flush_pending
can be visible after a page table update.  As per revised documentation,
this patch adds a smp_mb__before_spinlock to guarantee the correct
ordering.

Signed-off-by: Mel Gorman <mgorman@suse.de>
Acked-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Reviewed-by: Rik van Riel <riel@redhat.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
10 years agomm: fix TLB flush race between migration, and change_protection_range
Rik van Riel [Thu, 19 Dec 2013 01:08:44 +0000 (17:08 -0800)]
mm: fix TLB flush race between migration, and change_protection_range

There are a few subtle races, between change_protection_range (used by
mprotect and change_prot_numa) on one side, and NUMA page migration and
compaction on the other side.

The basic race is that there is a time window between when the PTE gets
made non-present (PROT_NONE or NUMA), and the TLB is flushed.

During that time, a CPU may continue writing to the page.

This is fine most of the time, however compaction or the NUMA migration
code may come in, and migrate the page away.

When that happens, the CPU may continue writing, through the cached
translation, to what is no longer the current memory location of the
process.

This only affects x86, which has a somewhat optimistic pte_accessible.
All other architectures appear to be safe, and will either always flush,
or flush whenever there is a valid mapping, even with no permissions
(SPARC).

The basic race looks like this:

CPU A CPU B CPU C

load TLB entry
make entry PTE/PMD_NUMA
fault on entry
read/write old page
start migrating page
change PTE/PMD to new page
read/write old page [*]
flush TLB
reload TLB from new entry
read/write new page
lose data

[*] the old page may belong to a new user at this point!

The obvious fix is to flush remote TLB entries, by making sure that
pte_accessible aware of the fact that PROT_NONE and PROT_NUMA memory may
still be accessible if there is a TLB flush pending for the mm.

This should fix both NUMA migration and compaction.

[mgorman@suse.de: fix build]
Signed-off-by: Rik van Riel <riel@redhat.com>
Signed-off-by: Mel Gorman <mgorman@suse.de>
Cc: Alex Thorlton <athorlton@sgi.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
10 years agomm: numa: avoid unnecessary disruption of NUMA hinting during migration
Mel Gorman [Thu, 19 Dec 2013 01:08:42 +0000 (17:08 -0800)]
mm: numa: avoid unnecessary disruption of NUMA hinting during migration

do_huge_pmd_numa_page() handles the case where there is parallel THP
migration.  However, by the time it is checked the NUMA hinting
information has already been disrupted.  This patch adds an earlier
check with some helpers.

Signed-off-by: Mel Gorman <mgorman@suse.de>
Reviewed-by: Rik van Riel <riel@redhat.com>
Cc: Alex Thorlton <athorlton@sgi.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
10 years agomm: numa: clear numa hinting information on mprotect
Mel Gorman [Thu, 19 Dec 2013 01:08:41 +0000 (17:08 -0800)]
mm: numa: clear numa hinting information on mprotect

On a protection change it is no longer clear if the page should be still
accessible.  This patch clears the NUMA hinting fault bits on a
protection change.

Signed-off-by: Mel Gorman <mgorman@suse.de>
Reviewed-by: Rik van Riel <riel@redhat.com>
Cc: Alex Thorlton <athorlton@sgi.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
10 years agosched: numa: skip inaccessible VMAs
Mel Gorman [Thu, 19 Dec 2013 01:08:40 +0000 (17:08 -0800)]
sched: numa: skip inaccessible VMAs

Inaccessible VMA should not be trapping NUMA hint faults. Skip them.

Signed-off-by: Mel Gorman <mgorman@suse.de>
Reviewed-by: Rik van Riel <riel@redhat.com>
Cc: Alex Thorlton <athorlton@sgi.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
10 years agomm: numa: avoid unnecessary work on the failure path
Mel Gorman [Thu, 19 Dec 2013 01:08:39 +0000 (17:08 -0800)]
mm: numa: avoid unnecessary work on the failure path

If a PMD changes during a THP migration then migration aborts but the
failure path is doing more work than is necessary.

Signed-off-by: Mel Gorman <mgorman@suse.de>
Reviewed-by: Rik van Riel <riel@redhat.com>
Cc: Alex Thorlton <athorlton@sgi.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
10 years agomm: numa: ensure anon_vma is locked to prevent parallel THP splits
Mel Gorman [Thu, 19 Dec 2013 01:08:38 +0000 (17:08 -0800)]
mm: numa: ensure anon_vma is locked to prevent parallel THP splits

The anon_vma lock prevents parallel THP splits and any associated
complexity that arises when handling splits during THP migration.  This
patch checks if the lock was successfully acquired and bails from THP
migration if it failed for any reason.

Signed-off-by: Mel Gorman <mgorman@suse.de>
Reviewed-by: Rik van Riel <riel@redhat.com>
Cc: Alex Thorlton <athorlton@sgi.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
10 years agomm: numa: do not clear PTE for pte_numa update
Mel Gorman [Thu, 19 Dec 2013 01:08:37 +0000 (17:08 -0800)]
mm: numa: do not clear PTE for pte_numa update

The TLB must be flushed if the PTE is updated but change_pte_range is
clearing the PTE while marking PTEs pte_numa without necessarily
flushing the TLB if it reinserts the same entry.  Without the flush,
it's conceivable that two processors have different TLBs for the same
virtual address and at the very least it would generate spurious faults.

This patch only unmaps the pages in change_pte_range for a full
protection change.

[riel@redhat.com: write pte_numa pte back to the page tables]
Signed-off-by: Mel Gorman <mgorman@suse.de>
Signed-off-by: Rik van Riel <riel@redhat.com>
Reviewed-by: Rik van Riel <riel@redhat.com>
Cc: Alex Thorlton <athorlton@sgi.com>
Cc: Chegu Vinod <chegu_vinod@hp.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
10 years agomm: numa: do not clear PMD during PTE update scan
Mel Gorman [Thu, 19 Dec 2013 01:08:36 +0000 (17:08 -0800)]
mm: numa: do not clear PMD during PTE update scan

If the PMD is flushed then a parallel fault in handle_mm_fault() will
enter the pmd_none and do_huge_pmd_anonymous_page() path where it'll
attempt to insert a huge zero page.  This is wasteful so the patch
avoids clearing the PMD when setting pmd_numa.

Signed-off-by: Mel Gorman <mgorman@suse.de>
Reviewed-by: Rik van Riel <riel@redhat.com>
Cc: Alex Thorlton <athorlton@sgi.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
10 years agomm: clear pmd_numa before invalidating
Mel Gorman [Thu, 19 Dec 2013 01:08:34 +0000 (17:08 -0800)]
mm: clear pmd_numa before invalidating

On x86, PMD entries are similar to _PAGE_PROTNONE protection and are
handled as NUMA hinting faults.  The following two page table protection
bits are what defines them

_PAGE_NUMA:set _PAGE_PRESENT:clear

A PMD is considered present if any of the _PAGE_PRESENT, _PAGE_PROTNONE,
_PAGE_PSE or _PAGE_NUMA bits are set.  If pmdp_invalidate encounters a
pmd_numa, it clears the present bit leaving _PAGE_NUMA which will be
considered not present by the CPU but present by pmd_present.  The
existing caller of pmdp_invalidate should handle it but it's an
inconsistent state for a PMD.  This patch keeps the state consistent
when calling pmdp_invalidate.

Signed-off-by: Mel Gorman <mgorman@suse.de>
Reviewed-by: Rik van Riel <riel@redhat.com>
Cc: Alex Thorlton <athorlton@sgi.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
10 years agomm: numa: call MMU notifiers on THP migration
Mel Gorman [Thu, 19 Dec 2013 01:08:33 +0000 (17:08 -0800)]
mm: numa: call MMU notifiers on THP migration

MMU notifiers must be called on THP page migration or secondary MMUs
will get very confused.

Signed-off-by: Mel Gorman <mgorman@suse.de>
Reviewed-by: Rik van Riel <riel@redhat.com>
Cc: Alex Thorlton <athorlton@sgi.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
10 years agomm: numa: serialise parallel get_user_page against THP migration
Mel Gorman [Thu, 19 Dec 2013 01:08:32 +0000 (17:08 -0800)]
mm: numa: serialise parallel get_user_page against THP migration

Base pages are unmapped and flushed from cache and TLB during normal
page migration and replaced with a migration entry that causes any
parallel NUMA hinting fault or gup to block until migration completes.

THP does not unmap pages due to a lack of support for migration entries
at a PMD level.  This allows races with get_user_pages and
get_user_pages_fast which commit 3f926ab945b6 ("mm: Close races between
THP migration and PMD numa clearing") made worse by introducing a
pmd_clear_flush().

This patch forces get_user_page (fast and normal) on a pmd_numa page to
go through the slow get_user_page path where it will serialise against
THP migration and properly account for the NUMA hinting fault.  On the
migration side the page table lock is taken for each PTE update.

Signed-off-by: Mel Gorman <mgorman@suse.de>
Reviewed-by: Rik van Riel <riel@redhat.com>
Cc: Alex Thorlton <athorlton@sgi.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
10 years agokexec: migrate to reboot cpu
Vivek Goyal [Thu, 19 Dec 2013 01:08:31 +0000 (17:08 -0800)]
kexec: migrate to reboot cpu

Commit 1b3a5d02ee07 ("reboot: move arch/x86 reboot= handling to generic
kernel") moved reboot= handling to generic code.  In the process it also
removed the code in native_machine_shutdown() which are moving reboot
process to reboot_cpu/cpu0.

I guess that thought must have been that all reboot paths are calling
migrate_to_reboot_cpu(), so we don't need this special handling.  But
kexec reboot path (kernel_kexec()) is not calling
migrate_to_reboot_cpu() so above change broke kexec.  Now reboot can
happen on non-boot cpu and when INIT is sent in second kerneo to bring
up BP, it brings down the machine.

So start calling migrate_to_reboot_cpu() in kexec reboot path to avoid
this problem.

Bisected by WANG Chao.

Reported-by: Matthew Whitehead <mwhitehe@redhat.com>
Reported-by: Dave Young <dyoung@redhat.com>
Signed-off-by: Vivek Goyal <vgoyal@redhat.com>
Tested-by: Baoquan He <bhe@redhat.com>
Tested-by: WANG Chao <chaowang@redhat.com>
Acked-by: H. Peter Anvin <hpa@linux.intel.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>