in scheduler-intense workloads native_read_tsc() overhead accounts for
20% of the system overhead:

 659567 system_call                              41222.9375
 686796 schedule                                 435.7843
 718382 __switch_to                              665.1685
 823875 switch_mm                                4526.7857
 1883122 native_read_tsc                          55385.9412
 9761990 total                                      2.8468

this is large part due to the rdtsc_barrier() that is done before
and after reading the TSC.

But sched_clock() is not a precise clock in the GTOD sense, using such
barriers is completely pointless. So remove the barriers and only use
them in vget_cycles().

This improves lat_ctx performance by about 5%.
Signed-off-by: NIngo Molnar <mingo@elte.hu>

fine-tune the HT sched-domains parameters as well.

On a HT capable box, this increases lat_ctx performance from 23.87
usecs to 1.49 usecs:

 # before

 $ ./lat_ctx -s 0 2

   "size=0k ovr=1.89
    2 23.87

 # after

 $ ./lat_ctx -s 0 2

   "size=0k ovr=1.84
     2 1.49
Signed-off-by: NIngo Molnar <mingo@elte.hu>

Tune SD_MC_INIT the same way as SD_CPU_INIT:
unset SD_BALANCE_NEWIDLE, and set SD_WAKE_BALANCE.

This improves vmark by 5%:

vmark         132102 125968 125497 messages/sec    avg 127855.66    .984
vmark         139404 131719 131272 messages/sec    avg 134131.66   1.033
Signed-off-by: NMike Galbraith <efault@gmx.de>
Acked-by: NPeter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: NIngo Molnar <mingo@elte.hu>

 # *DOCUMENTATION*

Impact: fix rare memory leak in the sched-domains manual reconfiguration code

In the failure path, rd is not attached to a sched domain,
so it causes a leak.
Signed-off-by: NLi Zefan <lizf@cn.fujitsu.com>
Acked-by: NPeter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: NIngo Molnar <mingo@elte.hu>

Impact: re-add incorrectly eliminated sched domain layers

(1) on i386 with SCHED_SMT and SCHED_MC enabled
	# mount -t cgroup -o cpuset xxx /mnt
	# echo 0 > /mnt/cpuset.sched_load_balance
	# mkdir /mnt/0
	# echo 0 > /mnt/0/cpuset.cpus
	# dmesg
	CPU0 attaching sched-domain:
	 domain 0: span 0 level CPU
	  groups: 0

(2) on i386 with SCHED_MC enabled but SCHED_SMT disabled
	# same with (1)
	# dmesg
	CPU0 attaching NULL sched-domain.

The bug is that some sched domains may be skipped unintentionally when
degenerating (optimizing) sched domains.
Signed-off-by: NLi Zefan <lizf@cn.fujitsu.com>
Acked-by: NPeter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: NIngo Molnar <mingo@elte.hu>

* git://git.kernel.org/pub/scm/linux/kernel/git/davem/net-2.6:
  net: Fix recursive descent in __scm_destroy().
  iwl3945: fix deadlock on suspend
  iwl3945: do not send scan command if channel count zero
  iwl3945: clear scanning bits upon failure
  ath5k: correct handling of rx status fields
  zd1211rw: Add 2 device IDs
  Fix logic error in rfkill_check_duplicity
  iwlagn: avoid sleep in softirq context
  iwlwifi: clear scanning bits upon failure
  Revert "ath5k: honor FIF_BCN_PRBRESP_PROMISC in STA mode"
  tcp: Fix recvmsg MSG_PEEK influence of blocking behavior.
  netfilter: netns ct: walk netns list under RTNL
  ipv6: fix run pending DAD when interface becomes ready
  net/9p: fix printk format warnings
  net: fix packet socket delivery in rx irq handler
  xfrm: Have af-specific init_tempsel() initialize family field of temporary selector

Merge branch 'x86-fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip

* 'x86-fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip:
  Revert "x86: default to reboot via ACPI"
  x86: align DirectMap in /proc/meminfo
  AMD IOMMU: fix lazy IO/TLB flushing in unmap path
  x86: add smp_mb() before sending INVALIDATE_TLB_VECTOR
  x86: remove VISWS and PARAVIRT around NR_IRQS puzzle
  x86: mention ACPI in top-level Kconfig menu
  x86: size NR_IRQS on 32-bit systems the same way as 64-bit
  x86: don't allow nr_irqs > NR_IRQS
  x86/docs: remove noirqbalance param docs
  x86: don't use tsc_khz to calculate lpj if notsc is passed
  x86, voyager: fix smp_intr_init() compile breakage
  AMD IOMMU: fix detection of NP capable IOMMUs

* master.kernel.org:/home/rmk/linux-2.6-arm:
  [ARM] xsc3: fix xsc3_l2_inv_range
  [ARM] mm: fix page table initialization
  [ARM] fix naming of MODULE_START / MODULE_END
  ARM: OMAP: Fix define for twl4030 irqs
  ARM: OMAP: Fix get_irqnr_and_base to clear spurious interrupt bits
  ARM: OMAP: Fix debugfs_create_*'s error checking method for arm/plat-omap
  ARM: OMAP: Fix compiler warnings in gpmc.c
  [ARM] fix VFP+softfloat binaries

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/ieee1394/linux1394-2.6:
  ieee1394: dv1394: fix possible deadlock in multithreaded clients
  ieee1394: raw1394: fix possible deadlock in multithreaded clients
  ieee1394: struct device - replace bus_id with dev_name(), dev_set_name()
  firewire: struct device - replace bus_id with dev_name(), dev_set_name()

* 'for-linus' of git://git.kernel.dk/linux-2.6-block:
  Block: use round_jiffies_up()
  Add round_jiffies_up and related routines
  block: fix __blkdev_get() for removable devices
  generic-ipi: fix the smp_mb() placement
  blk: move blk_delete_timer call in end_that_request_last
  block: add timer on blkdev_dequeue_request() not elv_next_request()
  bio: define __BIOVEC_PHYS_MERGEABLE
  block: remove unused ll_new_mergeable()

* git://git.kernel.org/pub/scm/linux/kernel/git/wim/linux-2.6-watchdog:
  [WATCHDOG] SAM9 watchdog - supported on all SAM9 and CAP9 processors
  [WATCHDOG] SAM9 watchdog - update for moved headers

* 'for-linus' of git://neil.brown.name/md:
  md: linear: Fix a division by zero bug for very small arrays.
  md: fix bug in raid10 recovery.
  md: revert the recent addition of a call to the BLKRRPART ioctl.

* 'merge' of git://git.kernel.org/pub/scm/linux/kernel/git/paulus/powerpc:
  powerpc: Fix "unused variable" warning in pci_dlpar.c
  powerpc/cell: Fix compile error in ras.c
  powerpc/ps3: Fix compile error in ps3-lpm.c

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/ericvh/v9fs:
  net/9p: fix printk format warnings
  unsigned fid->fid cannot be negative
  9p: rdma: remove duplicated #include
  p9: Fix leak of waitqueue in request allocation path
  9p: Remove unneeded free of fcall for Flush
  9p: Make all client spin locks IRQ safe
  9p: rdma: Set trans prior to requesting async connection ops

Merge branch 'sched-fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip

* 'sched-fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip:
  sched: re-tune balancing
  sched: fix buddies for group scheduling
  sched: backward looking buddy
  sched: fix fair preempt check
  sched: cleanup fair task selection

__scm_destroy() walks the list of file descriptors in the scm_fp_list
pointed to by the scm_cookie argument.

Those, in turn, can close sockets and invoke __scm_destroy() again.

There is nothing which limits how deeply this can occur.

The idea for how to fix this is from Linus.  Basically, we do all of
the fput()s at the top level by collecting all of the scm_fp_list
objects hit by an fput().  Inside of the initial __scm_destroy() we
keep running the list until it is empty.
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Fix the hrtimer_add_expires_ns() function.  It should take a 'u64 ns' argument,
but rather takes an 'unsigned long ns' argument - which might only be 32-bits.

On FRV, this results in the kernel locking up because hrtimer_forward() passes
the result of a 64-bit multiplication to this function, for which the compiler
discards the top 32-bits - something that didn't happen when ktime_add_ns() was
called directly.
Signed-off-by: NDavid Howells <dhowells@redhat.com>
Acked-by: NArjan van de Ven <arjan@linux.intel.com>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

* git://git.infradead.org/mtd-2.6:
  [JFFS2] fix race condition in jffs2_lzo_compress()
  [MTD] [NOR] Fix cfi_send_gen_cmd handling of x16 devices in x8 mode (v4)
  [JFFS2] Fix lack of locking in thread_should_wake()
  [JFFS2] Fix build failure with !CONFIG_JFFS2_FS_WRITEBUFFER
  [MTD] [NAND] OMAP2: remove duplicated #include

blkcnt_t type depends on CONFIG_LSF. Use unsigned long long always for
printk().  But lazy to type it, so add "llu" and use it.
Signed-off-by: NOGAWA Hirofumi <hirofumi@mail.parknet.co.jp>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

i_pos is 64bits value, hence it's not atomic to update.

Important place is fat_write_inode() only, other places without lock
are just for printk().

This adds lock for "BITS_PER_LONG == 32" kernel.
Signed-off-by: NOGAWA Hirofumi <hirofumi@mail.parknet.co.jp>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

mmu_private is 64bits value, hence it's not atomic to update.

So, the access rule for mmu_private is we must hold ->i_mutex.  But,
fat_get_block() path doesn't follow the rule on non-allocation path.

This fixes by using i_size instead if non-allocation path.
Signed-off-by: NOGAWA Hirofumi <hirofumi@mail.parknet.co.jp>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Signed-off-by: NOGAWA Hirofumi <hirofumi@mail.parknet.co.jp>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

fat_get_cluster() assumes the requested blocknr isn't truncated during
read. _fat_bmap() doesn't follow this rule.

This protects it by ->i_mutex.
Signed-off-by: NOGAWA Hirofumi <hirofumi@mail.parknet.co.jp>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

FAT has the ATTR_RO (read-only) attribute. But on Windows, the ATTR_RO
of the directory will be just ignored actually, and is used by only
applications as flag. E.g. it's setted for the customized folder by
Explorer.

http://msdn2.microsoft.com/en-us/library/aa969337.aspx

This adds "rodir" option. If user specified it, ATTR_RO is used as
read-only flag even if it's the directory. Otherwise, inode->i_mode
is not used to hold ATTR_RO (i.e. fat_mode_can_save_ro() returns 0).
Signed-off-by: NOGAWA Hirofumi <hirofumi@mail.parknet.co.jp>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

If inode->i_mode doesn't have S_WUGO, current code assumes it means
ATTR_RO.  However, if (~[ufd]mask & S_WUGO) == 0, inode->i_mode can't
hold S_WUGO. Therefore the updated directory entry will always have
ATTR_RO.

This adds fat_mode_can_hold_ro() to check it. And if inode->i_mode
can't hold, uses -i_attrs to hold ATTR_RO instead.

With this, we don't set ATTR_RO unless users change it via ioctl() if
(~[ufd]mask & S_WUGO) == 0.

And on FAT_IOCTL_GET_ATTRIBUTES path, this adds ->i_mutex to it for
not returning the partially updated attributes by FAT_IOCTL_SET_ATTRIBUTES
to userland.
Signed-off-by: NOGAWA Hirofumi <hirofumi@mail.parknet.co.jp>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

This adds three helpers:

fat_make_attrs() - makes FAT attributes from inode.
fat_make_mode()  - makes mode_t from FAT attributes.
fat_save_attrs() - saves FAT attributes to inode.

Then this replaces: MSDOS_MKMODE() by fat_make_mode(), fat_attr() by
fat_make_attrs(), ->i_attrs = attr & ATTR_UNUSED by fat_save_attrs().
And for root inode, those is used with ATTR_DIR instead of bogus
ATTR_NONE.
Signed-off-by: NOGAWA Hirofumi <hirofumi@mail.parknet.co.jp>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Use same style with vfat_lookup().
Signed-off-by: NOGAWA Hirofumi <hirofumi@mail.parknet.co.jp>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

d_invalidate() for positive dentry doesn't work in some cases
(vfsmount, nfsd, and maybe others). shrink_dcache_parent() by
d_invalidate() is pointless for vfat usage at all.

So, this kills it, and intead of it uses d_move().

To save old behavior, this returns alias simply for directory (don't
change pwd, etc..). the directory lookup shouldn't be important for
performance.
Signed-off-by: NOGAWA Hirofumi <hirofumi@mail.parknet.co.jp>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

- Add comments for handling dcache of vfat.

- Separate case-sensitive case and case-insensitive to
  vfat_revalidate() and vfat_ci_revalidate().

  vfat_revalidate() doesn't need to drop case-insensitive negative
  dentry on creation path.

- Current code is missing to set ->d_revalidate to the negative dentry
  created by unlink/etc..

  This sets ->d_revalidate always, and returns 1 for positive
  dentry. Now, we don't need to change ->d_op dynamically anymore,
  so this just uses sb->s_root->d_op to set ->d_op.

- d_find_alias() may return DCACHE_DISCONNECTED dentry. It's not
  the interesting dentry there. This checks it.

- Add missing LOOKUP_PARENT check. We don't need to drop the valid
  negative dentry for (LOOKUP_CREATE | LOOKUP_PARENT) lookup.

- For consistent filename on creation path, this drops negative dentry
  if we can't see intent.
Signed-off-by: NOGAWA Hirofumi <hirofumi@mail.parknet.co.jp>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Current vfat_lookup() creates negetive dentry blindly if vfat_find()
returned a error. It's wrong. If the error isn't -ENOENT, just return
error.
Signed-off-by: NOGAWA Hirofumi <hirofumi@mail.parknet.co.jp>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Use fat_detach() instead of opencoding it.
Signed-off-by: NOGAWA Hirofumi <hirofumi@mail.parknet.co.jp>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

This fixes the missing update for bhs/nr_bhs in case the caller
accessed from block boundary to first block of boundary.
Signed-off-by: NOGAWA Hirofumi <hirofumi@mail.parknet.co.jp>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

fat_hash() is using the algorithm known as bad. Instead of it, this
uses hash_32(). The following is the summary of test.

old hash:
	hash func (1000 times): 33489 cycles
	total inodes in hash table: 70926
	largest bucket contains: 696
	smallest bucket contains: 54

new hash:
	hash func (1000 times): 33129 cycles
	total inodes in hash table: 70926
	largest bucket contains: 315
	smallest bucket contains: 236
Signed-off-by: NOGAWA Hirofumi <hirofumi@mail.parknet.co.jp>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Coverity CID 2332 & 2333 RESOURCE_LEAK

In fat_search_long() if fat_parse_long() returns a -ve value we return
without first freeing unicode.  This patch free's them on this error path.

The above was false positive on current tree, but this change is more
clean, so apply as cleanup.

[hirofumi@mail.parknet.co.jp: fix coding style]
Signed-off-by: NDarren Jenkins <darrenrjenkins@gmail.com>
Signed-off-by: NOGAWA Hirofumi <hirofumi@mail.parknet.co.jp>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Since fat_dir_ioctl() was already fixed (i.e. called under ->i_mutex),
and __fat_readdir() doesn't take BKL anymore. So, BKL for ->llseek()
is pointless, and we have to use generic_file_llseek().
Signed-off-by: NOGAWA Hirofumi <hirofumi@mail.parknet.co.jp>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

This cleans date_dos2unix()/fat_date_unix2dos() up. New code should be
much more readable.

And this fixes those old functions. Those doesn't handle 2100
correctly. 2100 isn't leap year, but old one handles it as leap year.
Also, with this, centi sec is handled and is fixed.
Signed-off-by: NOGAWA Hirofumi <hirofumi@mail.parknet.co.jp>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

This splits __KERNEL__ stuff in include/msdos_fs.h into fs/fat/fat.h.
Signed-off-by: NOGAWA Hirofumi <hirofumi@mail.parknet.co.jp>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

This just moves those files, but change link order from MSDOS, VFAT to
VFAT, MSDOS.
Signed-off-by: NOGAWA Hirofumi <hirofumi@mail.parknet.co.jp>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

While debugging a sync mount regression on vfat I noticed that there were
mount options parsed by the driver that were not documented.

[hirofumi@mail.parknet.co.jp: fix some parts]
Signed-off-by: NBart Trojanowski <bart@jukie.net>
Signed-off-by: NOGAWA Hirofumi <hirofumi@mail.parknet.co.jp>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>