platform/kernel/linux-exynos.git
13 years agoxen/pci: Move check for acpi_sci_override_gsi to xen_setup_acpi_sci.
Konrad Rzeszutek Wilk [Wed, 6 Jul 2011 13:43:16 +0000 (09:43 -0400)]
xen/pci: Move check for acpi_sci_override_gsi to xen_setup_acpi_sci.

Previously we would check for acpi_sci_override_gsi == gsi every time
a PCI device was enabled. That works during early bootup, but later
on it could lead to triggering unnecessarily the acpi_gsi_to_irq(..) lookup.
The reason is that acpi_sci_override_gsi was declared in __initdata and
after early bootup could contain bogus values.

This patch moves the check for acpi_sci_override_gsi to the
site where the ACPI SCI is preset.

CC: stable@kernel.org
Reported-by: Raghavendra D Prabhu <rprabhu@wnohang.net>
Tested-by: Raghavendra D Prabhu <rprabhu@wnohang.net>
[http://lists.xensource.com/archives/html/xen-devel/2011-07/msg00154.html]
Suggested-by: Ian Campbell <ijc@hellion.org.uk>
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
13 years agoxen/pci: Use the INT_SRC_OVR IRQ (instead of GSI) to preset the ACPI SCI IRQ.
Konrad Rzeszutek Wilk [Thu, 30 Jun 2011 13:18:27 +0000 (09:18 -0400)]
xen/pci: Use the INT_SRC_OVR IRQ (instead of GSI) to preset the ACPI SCI IRQ.

In the past we would use the GSI value to preset the ACPI SCI
IRQ which worked great as GSI == IRQ:

ACPI: INT_SRC_OVR (bus 0 bus_irq 9 global_irq 9 low level)

While that is most often seen, there are some oddities:

ACPI: INT_SRC_OVR (bus 0 bus_irq 9 global_irq 20 low level)

which means that GSI 20 (or pin 20) is to be overriden for IRQ 9.
Our code that presets the interrupt for ACPI SCI however would
use the GSI 20 instead of IRQ 9 ending up with:

xen: sci override: global_irq=20 trigger=0 polarity=1
xen: registering gsi 20 triggering 0 polarity 1
xen: --> pirq=20 -> irq=20
xen: acpi sci 20
.. snip..
calling  acpi_init+0x0/0xbc @ 1
ACPI: SCI (IRQ9) allocation failed
ACPI Exception: AE_NOT_ACQUIRED, Unable to install System Control Interrupt handler (20110413/evevent-119)
ACPI: Unable to start the ACPI Interpreter

as the ACPI interpreter made a call to 'acpi_gsi_to_irq' which got nine.
It used that value to request an IRQ (request_irq) and since that was not
present it failed.

The fix is to recognize that for interrupts that are overriden (in our
case we only care about the ACPI SCI) we should use the IRQ number
to present the IRQ instead of the using GSI. End result is that we get:

xen: sci override: global_irq=20 trigger=0 polarity=1
xen: registering gsi 20 triggering 0 polarity 1
xen: --> pirq=20 -> irq=9 (gsi=9)
xen: acpi sci 9

which fixes the ACPI interpreter failing on startup.

CC: stable@kernel.org
Reported-by: Liwei <xieliwei@gmail.com>
Tested-by: Liwei <xieliwei@gmail.com>
[http://lists.xensource.com/archives/html/xen-devel/2011-06/msg01727.html]
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
13 years agoxen/mmu: Fix for linker errors when CONFIG_SMP is not defined.
Konrad Rzeszutek Wilk [Thu, 30 Jun 2011 13:12:40 +0000 (09:12 -0400)]
xen/mmu: Fix for linker errors when CONFIG_SMP is not defined.

Simple enough - we use an extern defined symbol which is not
defined when CONFIG_SMP is not defined. This fixes the linker
dying.

CC: stable@kernel.org
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
13 years agoxen/setup: Fix for incorrect xen_extra_mem_start.
Konrad Rzeszutek Wilk [Thu, 16 Jun 2011 17:07:19 +0000 (13:07 -0400)]
xen/setup: Fix for incorrect xen_extra_mem_start.

The earlier attempts (24bdb0b62cc82120924762ae6bc85afc8c3f2b26)
at fixing this problem caused other problems to surface (PV guests
with no PCI passthrough would have SWIOTLB turned on - which meant
64MB of precious contingous DMA32 memory being eaten up per guest).
The problem was: "on xen we add an extra memory region at the end of
the e820, and on this particular machine this extra memory region
would start below 4g and cross over the 4g boundary:

[0xfee01000-0x192655000)

Unfortunately e820_end_of_low_ram_pfn does not expect an
e820 layout like that so it returns 4g, therefore initial_memory_mapping
will map [0 - 0x100000000), that is a memory range that includes some
reserved memory regions."

The memory range was the IOAPIC regions, and with the 1-1 mapping
turned on, it would map them as RAM, not as MMIO regions. This caused
the hypervisor to complain. Fortunately this is experienced only under
the initial domain so we guard for it.

Acked-by: Stefano Stabellini <stefano.stabellini@eu.citrix.com>
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
13 years agoxen: When calling power_off, don't call the halt function.
Tom Goetz [Mon, 16 May 2011 19:06:26 +0000 (15:06 -0400)]
xen: When calling power_off, don't call the halt function.

.. As it won't actually power off the machine.

Reported-by: Sven Köhler <sven.koehler@gmail.com>
Tested-by: Sven Köhler <sven.koehler@gmail.com>
Signed-off-by: Tom Goetz <tom.goetz@virtualcomputer.com>
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
13 years agoxen: Fix compile warning when CONFIG_SMP is not defined.
Konrad Rzeszutek Wilk [Wed, 15 Jun 2011 18:24:29 +0000 (14:24 -0400)]
xen: Fix compile warning when CONFIG_SMP is not defined.

.. which is quite benign.

drivers/xen/events.c:398: warning: unused variable ‘desc’

Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
13 years agoxen: support CONFIG_MAXSMP
Andrew Jones [Fri, 18 Dec 2009 09:31:31 +0000 (10:31 +0100)]
xen: support CONFIG_MAXSMP

The MAXSMP config option requires CPUMASK_OFFSTACK, which in turn
requires we init the memory for the maps while we bring up the cpus.
MAXSMP also increases NR_CPUS to 4096. This increase in size exposed an
issue in the argument construction for multicalls from
xen_flush_tlb_others. The args should only need space for the actual
number of cpus.

Also in 2.6.39 it exposes a bootup problem.

BUG: unable to handle kernel NULL pointer dereference at           (null)
IP: [<ffffffff8157a1d3>] set_cpu_sibling_map+0x123/0x30d
...
Call Trace:
[<ffffffff81039a3f>] ? xen_restore_fl_direct_reloc+0x4/0x4
[<ffffffff819dc4db>] xen_smp_prepare_cpus+0x36/0x135
..

CC: stable@kernel.org
Signed-off-by: Andrew Jones <drjones@redhat.com>
[v2: Updated to compile on 3.0]
[v3: Updated to compile when CONFIG_SMP is not defined]
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
13 years agoxen: partially revert "xen: set max_pfn_mapped to the last pfn mapped"
Stefano Stabellini [Fri, 3 Jun 2011 09:51:34 +0000 (09:51 +0000)]
xen: partially revert "xen: set max_pfn_mapped to the last pfn mapped"

We only need to set max_pfn_mapped to the last pfn mapped on x86_64 to
make sure that cleanup_highmap doesn't remove important mappings at
_end.

We don't need to do this on x86_32 because cleanup_highmap is not called
on x86_32. Besides lowering max_pfn_mapped on x86_32 has the unwanted
side effect of limiting the amount of memory available for the 1:1
kernel pagetable allocation.

This patch reverts the x86_32 part of the original patch.

CC: stable@kernel.org
Signed-off-by: Stefano Stabellini <stefano.stabellini@eu.citrix.com>
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
13 years agoxen: off by one errors in multicalls.c
Dan Carpenter [Fri, 3 Jun 2011 04:45:28 +0000 (07:45 +0300)]
xen: off by one errors in multicalls.c

b->args[] has MC_ARGS elements, so the comparison here should be
">=" instead of ">".  Otherwise we read past the end of the array
one space.

CC: stable@kernel.org
Signed-off-by: Dan Carpenter <error27@gmail.com>
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Signed-off-by: Jeremy Fitzhardinge <jeremy.fitzhardinge@citrix.com>
13 years agoxen: use the trigger info we already have to choose the irq handler
Stefano Stabellini [Wed, 25 May 2011 11:33:23 +0000 (12:33 +0100)]
xen: use the trigger info we already have to choose the irq handler

Do not use pirq_needs_eoi to decide which irq handler to use because Xen
always returns true if the guest does not support pirq_eoi_map.
Use the trigger information we already have from MP-tables and ACPI.

Signed-off-by: Stefano Stabellini <stefano.stabellini@eu.citrix.com>
Reported-by: Thomas Goetz <tom.goetz@virtualcomputer.com>
Tested-by: Thomas Goetz <tom.goetz@virtualcomputer.com>
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
13 years agoLinux 3.0-rc1 v3.0-rc1
Linus Torvalds [Mon, 30 May 2011 00:43:36 +0000 (17:43 -0700)]
Linux 3.0-rc1

.. except there are various scripts that really know that there are
three numbers, so it calls itself "3.0.0-rc1".

Hopefully by the time the final 3.0 is out, we'll have that extra zero
all figured out.

Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
13 years agoMerge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/ecryptfs...
Linus Torvalds [Sun, 29 May 2011 21:13:25 +0000 (14:13 -0700)]
Merge branch 'for-linus' of git://git./linux/kernel/git/ecryptfs/ecryptfs-2.6

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/ecryptfs/ecryptfs-2.6:
  eCryptfs: Remove ecryptfs_header_cache_2
  eCryptfs: Cleanup and optimize ecryptfs_lookup_interpose()
  eCryptfs: Return useful code from contains_ecryptfs_marker
  eCryptfs: Fix new inode race condition
  eCryptfs: Cleanup inode initialization code
  eCryptfs: Consolidate inode functions into inode.c

13 years agoMerge branch 'pnfs-submit' of git://git.open-osd.org/linux-open-osd
Linus Torvalds [Sun, 29 May 2011 21:10:13 +0000 (14:10 -0700)]
Merge branch 'pnfs-submit' of git://git.open-osd.org/linux-open-osd

* 'pnfs-submit' of git://git.open-osd.org/linux-open-osd: (32 commits)
  pnfs-obj: pg_test check for max_io_size
  NFSv4.1: define nfs_generic_pg_test
  NFSv4.1: use pnfs_generic_pg_test directly by layout driver
  NFSv4.1: change pg_test return type to bool
  NFSv4.1: unify pnfs_pageio_init functions
  pnfs-obj: objlayout_encode_layoutcommit implementation
  pnfs: encode_layoutcommit
  pnfs-obj: report errors and .encode_layoutreturn Implementation.
  pnfs: encode_layoutreturn
  pnfs: layoutret_on_setattr
  pnfs: layoutreturn
  pnfs-obj: osd raid engine read/write implementation
  pnfs: support for non-rpc layout drivers
  pnfs-obj: define per-inode private structure
  pnfs: alloc and free layout_hdr layoutdriver methods
  pnfs-obj: objio_osd device information retrieval and caching
  pnfs-obj: decode layout, alloc/free lseg
  pnfs-obj: pnfs_osd XDR client implementation
  pnfs-obj: pnfs_osd XDR definitions
  pnfs-obj: objlayoutdriver module skeleton
  ...

13 years agoarm gpio drivers: make them 'depends on ARM'
Linus Torvalds [Sun, 29 May 2011 21:06:42 +0000 (14:06 -0700)]
arm gpio drivers: make them 'depends on ARM'

We had a few drivers move from arch/arm into drivers/gpio, but they
don't actually compile without the ARM platform headers etc.  As a
result they were messing up allyesconfig on x86.

Make them depend on ARM.

Reported-by: Ingo Molnar <mingo@elte.hu>
Cc: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
13 years agoeCryptfs: Remove ecryptfs_header_cache_2
Tyler Hicks [Tue, 24 May 2011 10:11:12 +0000 (05:11 -0500)]
eCryptfs: Remove ecryptfs_header_cache_2

Now that ecryptfs_lookup_interpose() is no longer using
ecryptfs_header_cache_2 to read in metadata, the kmem_cache can be
removed and the ecryptfs_header_cache_1 kmem_cache can be renamed to
ecryptfs_header_cache.

Signed-off-by: Tyler Hicks <tyhicks@linux.vnet.ibm.com>
13 years agoeCryptfs: Cleanup and optimize ecryptfs_lookup_interpose()
Tyler Hicks [Tue, 24 May 2011 09:56:23 +0000 (04:56 -0500)]
eCryptfs: Cleanup and optimize ecryptfs_lookup_interpose()

ecryptfs_lookup_interpose() has turned into spaghetti code over the
years. This is an effort to clean it up.

 - Shorten overly descriptive variable names such as ecryptfs_dentry
 - Simplify gotos and error paths
 - Create helper function for reading plaintext i_size from metadata

It also includes an optimization when reading i_size from the metadata.
A complete page-sized kmem_cache_alloc() was being done to read in 16
bytes of metadata. The buffer for that is now statically declared.

Signed-off-by: Tyler Hicks <tyhicks@linux.vnet.ibm.com>
13 years agoeCryptfs: Return useful code from contains_ecryptfs_marker
Tyler Hicks [Mon, 2 May 2011 05:39:54 +0000 (00:39 -0500)]
eCryptfs: Return useful code from contains_ecryptfs_marker

Instead of having the calling functions translate the true/false return
code to either 0 or -EINVAL, have contains_ecryptfs_marker() return 0 or
-EINVAL so that the calling functions can just reuse the return code.

Also, rename the function to ecryptfs_validate_marker() to avoid callers
mistakenly thinking that it returns true/false codes.

Signed-off-by: Tyler Hicks <tyhicks@linux.vnet.ibm.com>
13 years agoeCryptfs: Fix new inode race condition
Tyler Hicks [Tue, 24 May 2011 08:49:02 +0000 (03:49 -0500)]
eCryptfs: Fix new inode race condition

Only unlock and d_add() new inodes after the plaintext inode size has
been read from the lower filesystem. This fixes a race condition that
was sometimes seen during a multi-job kernel build in an eCryptfs mount.

https://bugzilla.kernel.org/show_bug.cgi?id=36002

Signed-off-by: Tyler Hicks <tyhicks@linux.vnet.ibm.com>
Reported-by: David <david@unsolicited.net>
Tested-by: David <david@unsolicited.net>
13 years agoMerge branch 'for_linus' of git://git.kernel.org/pub/scm/linux/kernel/git/mjg59/platf...
Linus Torvalds [Sun, 29 May 2011 18:44:33 +0000 (11:44 -0700)]
Merge branch 'for_linus' of git://git./linux/kernel/git/mjg59/platform-drivers-x86

* 'for_linus' of git://git.kernel.org/pub/scm/linux/kernel/git/mjg59/platform-drivers-x86: (43 commits)
  acer-wmi: support integer return type from WMI methods
  msi-laptop: fix section mismatch in reference from the function load_scm_model_init
  acer-wmi: support to set communication device state by new wmid method
  acer-wmi: allow 64-bits return buffer from WMI methods
  acer-wmi: check the existence of internal 3G device when set capability
  platform/x86:delete two unused variables
  support wlan hotkey on Acer Travelmate 5735Z
  platform-x86: intel_mid_thermal: Fix memory leak
  platform/x86: Fix Makefile for intel_mid_powerbtn
  platform/x86: Simplify intel_mid_powerbtn
  acer-wmi: Delete out-of-date documentation
  acerhdf: Clean up includes
  acerhdf: Drop pointless dependency on THERMAL_HWMON
  acer-wmi: Update MAINTAINERS
  wmi: Orphan ACPI-WMI driver
  tc1100-wmi: Orphan driver
  acer-wmi: does not allow negative number set to initial device state
  platform/oaktrail: ACPI EC Extra driver for Oaktrail
  thinkpad_acpi: Convert printks to pr_<level>
  thinkpad_acpi: Correct !CONFIG_THINKPAD_ACPI_VIDEO warning
  ...

13 years agomm: Fix boot crash in mm_alloc()
Linus Torvalds [Sun, 29 May 2011 18:32:28 +0000 (11:32 -0700)]
mm: Fix boot crash in mm_alloc()

Thomas Gleixner reports that we now have a boot crash triggered by
CONFIG_CPUMASK_OFFSTACK=y:

    BUG: unable to handle kernel NULL pointer dereference at   (null)
    IP: [<c11ae035>] find_next_bit+0x55/0xb0
    Call Trace:
     [<c11addda>] cpumask_any_but+0x2a/0x70
     [<c102396b>] flush_tlb_mm+0x2b/0x80
     [<c1022705>] pud_populate+0x35/0x50
     [<c10227ba>] pgd_alloc+0x9a/0xf0
     [<c103a3fc>] mm_init+0xec/0x120
     [<c103a7a3>] mm_alloc+0x53/0xd0

which was introduced by commit de03c72cfce5 ("mm: convert
mm->cpu_vm_cpumask into cpumask_var_t"), and is due to wrong ordering of
mm_init() vs mm_init_cpumask

Thomas wrote a patch to just fix the ordering of initialization, but I
hate the new double allocation in the fork path, so I ended up instead
doing some more radical surgery to clean it all up.

Reported-by: Thomas Gleixner <tglx@linutronix.de>
Reported-by: Ingo Molnar <mingo@elte.hu>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
13 years agoMerge branch 'for-linus' of git://git390.marist.edu/pub/scm/linux-2.6
Linus Torvalds [Sun, 29 May 2011 18:30:20 +0000 (11:30 -0700)]
Merge branch 'for-linus' of git://git390.marist.edu/linux-2.6

* 'for-linus' of git://git390.marist.edu/pub/scm/linux-2.6:
  [S390] mm: fix mmu_gather rework
  [S390] mm: fix storage key handling

13 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/cmetcalf/linux-tile
Linus Torvalds [Sun, 29 May 2011 18:29:28 +0000 (11:29 -0700)]
Merge git://git./linux/kernel/git/cmetcalf/linux-tile

* git://git.kernel.org/pub/scm/linux/kernel/git/cmetcalf/linux-tile:
  arch/tile: more /proc and /sys file support

13 years agoMerge branch 'x86-platform-next' into x86-platform
Matthew Garrett [Sun, 29 May 2011 18:27:13 +0000 (14:27 -0400)]
Merge branch 'x86-platform-next' into x86-platform

13 years agoMerge branch 'for-2.6.40' of git://linux-nfs.org/~bfields/linux
Linus Torvalds [Sun, 29 May 2011 18:21:12 +0000 (11:21 -0700)]
Merge branch 'for-2.6.40' of git://linux-nfs.org/~bfields/linux

* 'for-2.6.40' of git://linux-nfs.org/~bfields/linux: (22 commits)
  nfsd: make local functions static
  NFSD: Remove unused variable from nfsd4_decode_bind_conn_to_session()
  NFSD: Check status from nfsd4_map_bcts_dir()
  NFSD: Remove setting unused variable in nfsd_vfs_read()
  nfsd41: error out on repeated RECLAIM_COMPLETE
  nfsd41: compare request's opcnt with session's maxops at nfsd4_sequence
  nfsd v4.1 lOCKT clientid field must be ignored
  nfsd41: add flag checking for create_session
  nfsd41: make sure nfs server process OPEN with EXCLUSIVE4_1 correctly
  nfsd4: fix wrongsec handling for PUTFH + op cases
  nfsd4: make fh_verify responsibility of nfsd_lookup_dentry caller
  nfsd4: introduce OPDESC helper
  nfsd4: allow fh_verify caller to skip pseudoflavor checks
  nfsd: distinguish functions of NFSD_MAY_* flags
  svcrpc: complete svsk processing on cb receive failure
  svcrpc: take advantage of tcp autotuning
  SUNRPC: Don't wait for full record to receive tcp data
  svcrpc: copy cb reply instead of pages
  svcrpc: close connection if client sends short packet
  svcrpc: note network-order types in svc_process_calldir
  ...

13 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/agk/linux-2.6-dm
Linus Torvalds [Sun, 29 May 2011 18:20:48 +0000 (11:20 -0700)]
Merge git://git./linux/kernel/git/agk/linux-2.6-dm

* git://git.kernel.org/pub/scm/linux/kernel/git/agk/linux-2.6-dm:
  dm kcopyd: return client directly and not through a pointer
  dm kcopyd: reserve fewer pages
  dm io: use fixed initial mempool size
  dm kcopyd: alloc pages from the main page allocator
  dm kcopyd: add gfp parm to alloc_pl
  dm kcopyd: remove superfluous page allocation spinlock
  dm kcopyd: preallocate sub jobs to avoid deadlock
  dm kcopyd: avoid pointless job splitting
  dm mpath: do not fail paths after integrity errors
  dm table: reject devices without request fns
  dm table: allow targets to support discards internally

13 years agoMerge branch 'nfs-for-2.6.40' of git://git.linux-nfs.org/projects/trondmy/nfs-2.6
Linus Torvalds [Sun, 29 May 2011 18:20:02 +0000 (11:20 -0700)]
Merge branch 'nfs-for-2.6.40' of git://git.linux-nfs.org/projects/trondmy/nfs-2.6

* 'nfs-for-2.6.40' of git://git.linux-nfs.org/projects/trondmy/nfs-2.6:
  SUNRPC: Support for RPC over AF_LOCAL transports
  SUNRPC: Remove obsolete comment
  SUNRPC: Use AF_LOCAL for rpcbind upcalls
  SUNRPC: Clean up use of curly braces in switch cases
  NFS: Revert NFSROOT default mount options
  SUNRPC: Rename xs_encode_tcp_fragment_header()
  nfs,rcu: convert call_rcu(nfs_free_delegation_callback) to kfree_rcu()
  nfs41: Correct offset for LAYOUTCOMMIT
  NFS: nfs_update_inode: print current and new inode size in debug output
  NFSv4.1: Fix the handling of NFS4ERR_SEQ_MISORDERED errors
  NFSv4: Handle expired stateids when the lease is still valid
  SUNRPC: Deal with the lack of a SYN_SENT sk->sk_state_change callback...

13 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/pkl/squashfs-linus
Linus Torvalds [Sun, 29 May 2011 18:19:45 +0000 (11:19 -0700)]
Merge git://git./linux/kernel/git/pkl/squashfs-linus

* git://git.kernel.org/pub/scm/linux/kernel/git/pkl/squashfs-linus:
  Squashfs: Fix sanity check patches on big-endian systems

13 years agoMerge branch 'release' of git://git.kernel.org/pub/scm/linux/kernel/git/lenb/linux...
Linus Torvalds [Sun, 29 May 2011 18:19:16 +0000 (11:19 -0700)]
Merge branch 'release' of git://git./linux/kernel/git/lenb/linux-acpi-2.6

* 'release' of git://git.kernel.org/pub/scm/linux/kernel/git/lenb/linux-acpi-2.6:
  ACPI EC: remove redundant code
  ACPI: Add D3 cold state
  ACPI: processor: fix processor_physically_present in UP kernel
  ACPI: Split out custom_method functionality into an own driver
  ACPI: Cleanup custom_method debug stuff
  ACPI EC: enable MSI workaround for Quanta laptops
  ACPICA: Update to version 20110413
  ACPICA: Execute an orphan _REG method under the EC device
  ACPICA: Move ACPI_NUM_PREDEFINED_REGIONS to a more appropriate place
  ACPICA: Update internal address SpaceID for DataTable regions
  ACPICA: Add more methods eligible for NULL package element removal
  ACPICA: Split all internal Global Lock functions to new file - evglock
  ACPI: EC: add another DMI check for ASUS hardware
  ACPI EC: remove dead code
  ACPICA: Fix code divergence of global lock handling
  ACPICA: Use acpi_os_create_lock interface
  ACPI: osl, add acpi_os_create_lock interface
  ACPI:Fix goto flows in thermal-sys

13 years agoMerge branch 'idle-release' of git://git.kernel.org/pub/scm/linux/kernel/git/lenb...
Linus Torvalds [Sun, 29 May 2011 18:18:09 +0000 (11:18 -0700)]
Merge branch 'idle-release' of git://git./linux/kernel/git/lenb/linux-idle-2.6

* 'idle-release' of git://git.kernel.org/pub/scm/linux/kernel/git/lenb/linux-idle-2.6:
  x86 idle: deprecate mwait_idle() and "idle=mwait" cmdline param
  x86 idle: deprecate "no-hlt" cmdline param
  x86 idle APM: deprecate CONFIG_APM_CPU_IDLE
  x86 idle floppy: deprecate disable_hlt()
  x86 idle: EXPORT_SYMBOL(default_idle, pm_idle) only when APM demands it
  x86 idle: clarify AMD erratum 400 workaround
  idle governor: Avoid lock acquisition to read pm_qos before entering idle
  cpuidle: menu: fixed wrapping timers at 4.294 seconds

13 years agocifs/ubifs: Fix shrinker API change fallout
Al Viro [Sun, 29 May 2011 12:46:08 +0000 (13:46 +0100)]
cifs/ubifs: Fix shrinker API change fallout

Commit 1495f230fa77 ("vmscan: change shrinker API by passing
shrink_control struct") changed the API of ->shrink(), but missed ubifs
and cifs instances.

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
13 years agopnfs-obj: pg_test check for max_io_size
Boaz Harrosh [Wed, 25 May 2011 18:25:29 +0000 (21:25 +0300)]
pnfs-obj: pg_test check for max_io_size

Implement pg_test vector to test for max IO sizes. We calculate
a max_io_size member only once, and cache it in lseg so to not
do so on every page insert.

Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
[simplify logic]
Signed-off-by: Benny Halevy <bhalevy@panasas.com>
13 years agoNFSv4.1: define nfs_generic_pg_test
Boaz Harrosh [Sun, 29 May 2011 08:45:39 +0000 (11:45 +0300)]
NFSv4.1: define nfs_generic_pg_test

By default, unless pnfs is used coalesce pages until pg_bsize
(rsize or wsize) is reached.

pnfs layout drivers define their own pg_test methods that use
pnfs_generic_pg_test and need to define their own I/O size
limits (e.g. based on the file stripe size).

[Move a check from nfs_pageio_do_add_request to nfs_generic_pg_test]
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
Signed-off-by: Benny Halevy <bhalevy@panasas.com>
13 years agoNFSv4.1: use pnfs_generic_pg_test directly by layout driver
Benny Halevy [Wed, 25 May 2011 17:54:40 +0000 (20:54 +0300)]
NFSv4.1: use pnfs_generic_pg_test directly by layout driver

Signed-off-by: Benny Halevy <bhalevy@panasas.com>
13 years agoNFSv4.1: change pg_test return type to bool
Benny Halevy [Wed, 25 May 2011 18:03:56 +0000 (21:03 +0300)]
NFSv4.1: change pg_test return type to bool

Signed-off-by: Benny Halevy <bhalevy@panasas.com>
13 years agoNFSv4.1: unify pnfs_pageio_init functions
Benny Halevy [Wed, 25 May 2011 17:25:22 +0000 (20:25 +0300)]
NFSv4.1: unify pnfs_pageio_init functions

Use common code for pnfs_pageio_init_{read,write} and use
a common generic pg_test function.

Note that this function always assumes the the layout driver's
pg_test method is implemented.

[Fix BUG]
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
Signed-off-by: Benny Halevy <bhalevy@panasas.com>
13 years agopnfs-obj: objlayout_encode_layoutcommit implementation
Boaz Harrosh [Sun, 22 May 2011 16:54:13 +0000 (19:54 +0300)]
pnfs-obj: objlayout_encode_layoutcommit implementation

* Define API for io-engines to report delta_space_used in IOs
* Encode the osd-layout specific information of the layoutcommit
  XDR buffer.

Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
Signed-off-by: Benny Halevy <bhalevy@panasas.com>
13 years agopnfs: encode_layoutcommit
Benny Halevy [Sun, 22 May 2011 16:53:48 +0000 (19:53 +0300)]
pnfs: encode_layoutcommit

Add a layout driver method to encode the layout type specific
opaque part of layout commit in-line in the xdr stream.

Currently, the pnfs-objects layout driver uses it to encode metadata hints
to the MDS and the blocks layout driver to commit provisionally allocated
extents to the file.

Signed-off-by: Benny Halevy <bhalevy@panasas.com>
13 years agopnfs-obj: report errors and .encode_layoutreturn Implementation.
Boaz Harrosh [Thu, 26 May 2011 18:49:46 +0000 (21:49 +0300)]
pnfs-obj: report errors and .encode_layoutreturn Implementation.

An io_state pre-allocates an error information structure for each
possible osd-device that might error during IO. When IO is done if all
was well the io_state is freed. (as today). If the I/O has ended with an
error, the io_state is queued on a per-layout err_list. When eventually
encode_layoutreturn() is called, each error is properly encoded on the
XDR buffer and only then the io_state is removed from err_list and
de-allocated.

It is up to the io_engine to fill in the segment that fault and the type
of osd_error that occurred. By calling objlayout_io_set_result() for
each failing device.

In objio_osd:
* Allocate io-error descriptors space as part of io_state
* Use generic objlayout error reporting at end of io.

Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
Signed-off-by: Benny Halevy <bhalevy@panasas.com>
13 years agopnfs: encode_layoutreturn
Andy Adamson [Sun, 22 May 2011 16:53:10 +0000 (19:53 +0300)]
pnfs: encode_layoutreturn

Add a layout driver method to encode the layout type specific
opaque part of layout return in-line in the xdr stream.

Currently the pnfs-objects layout driver uses it to encode i/o error
information on LAYOUTRETURN.

Signed-off-by: Andy Adamson <andros@netapp.com>
[fixup layout header pointer for encode_layoutreturn]
Signed-off-by: Benny Halevy <bhalevy@panasas.com>
13 years agopnfs: layoutret_on_setattr
Benny Halevy [Wed, 14 Jul 2010 19:43:57 +0000 (15:43 -0400)]
pnfs: layoutret_on_setattr

With the objects layout security model, we have object capabilities
that are associated with the layout and we anticipate that the server
will issue a cb_layoutrecall for any setattr that changes security
related attributes (user/group/mode/acl) or truncates the file.

Therefore, the layout is returned before issuing the setattr to avoid
the anticipated cb_layoutrecall.

Signed-off-by: Benny Halevy <bhalevy@panasas.com>
13 years agopnfs: layoutreturn
Benny Halevy [Sun, 22 May 2011 16:52:37 +0000 (19:52 +0300)]
pnfs: layoutreturn

NFSv4.1 LAYOUTRETURN implementation

Currently, does not support layout-type payload encoding.

Signed-off-by: Alexandros Batsakis <batsakis@netapp.com>
Signed-off-by: Andy Adamson <andros@citi.umich.edu>
Signed-off-by: Andy Adamson <andros@netapp.com>
Signed-off-by: Dean Hildebrand <dhildeb@us.ibm.com>
Signed-off-by: Fred Isaman <iisaman@citi.umich.edu>
Signed-off-by: Fred Isaman <iisaman@netapp.com>
Signed-off-by: Marc Eshel <eshel@almaden.ibm.com>
Signed-off-by: Zhang Jingwang <zhangjingwang@nrchpc.ac.cn>
[call pnfs_return_layout right before pnfs_destroy_layout]
[remove assert_spin_locked from pnfs_clear_lseg_list]
[remove wait parameter from the layoutreturn path.]
[remove return_type field from nfs4_layoutreturn_args]
[remove range from nfs4_layoutreturn_args]
[no need to send layoutcommit from _pnfs_return_layout]
[don't wait on sync layoutreturn]
[fix layout stateid in layoutreturn args]
[fixed NULL deref in _pnfs_return_layout]
[removed recaim member of nfs4_layoutreturn_args]
Signed-off-by: Benny Halevy <bhalevy@panasas.com>
13 years agopnfs-obj: osd raid engine read/write implementation
Boaz Harrosh [Sun, 22 May 2011 16:52:19 +0000 (19:52 +0300)]
pnfs-obj: osd raid engine read/write implementation

With the use of the in-kernel osd library. Implement read/write
of data from/to osd-objects according to information specified
in the objects-layout.

Support for stripping over mirrors with a received stripe_unit.
There are however a few constrains which are not supported:
 1. Stripe Unit must be a multiple of PAGE_SIZE
 2. stripe length (stripe_unit * number_of_stripes) can not be
    bigger then 32bit.

Also support raid-groups and partial-layout. Partial-layout is
when not all the groups are received on the line, addressing
only a partial range of the file.

TODO:
  Only raid0! raid 4/5/6 support will come at later stage

A none supported layout will send IO through the MDS

[Important fallout from the last rebase]
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
[gfp_flags]
Signed-off-by: Benny Halevy <bhalevy@panasas.com>
13 years agopnfs: support for non-rpc layout drivers
Benny Halevy [Sun, 22 May 2011 16:52:03 +0000 (19:52 +0300)]
pnfs: support for non-rpc layout drivers

Non-rpc layout driver such as for objects and blocks
implement their own I/O path and error handling logic.
Therefore bypass NFS-based error handling for these layout drivers.

[fix lseg ref-count bugs, and null de-refs]
[Fall out from: non-rpc layout drivers]
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
[get rid of PNFS_USE_RPC_CODE]
[get rid of __nfs4_write_done_cb]
[revert useless change in nfs4_write_done_cb]
Signed-off-by: Benny Halevy <bhalevy@panasas.com>
13 years agopnfs-obj: define per-inode private structure
Benny Halevy [Sun, 22 May 2011 16:51:48 +0000 (19:51 +0300)]
pnfs-obj: define per-inode private structure

allocate and deallocate per-inode private pnfs_layout_hdr
in preparation for I/O implementation.

Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
Signed-off-by: Benny Halevy <bhalevy@panasas.com>
13 years agopnfs: alloc and free layout_hdr layoutdriver methods
Benny Halevy [Sun, 22 May 2011 16:51:33 +0000 (19:51 +0300)]
pnfs: alloc and free layout_hdr layoutdriver methods

[gfp_flags]
Signed-off-by: Benny Halevy <bhalevy@panasas.com>
13 years agopnfs-obj: objio_osd device information retrieval and caching
Boaz Harrosh [Thu, 26 May 2011 18:45:34 +0000 (21:45 +0300)]
pnfs-obj: objio_osd device information retrieval and caching

When a new layout is received in objio_alloc_lseg all device_ids
referenced are retrieved. The device information is queried for from MDS
and then the osd_device is looked-up from the osd-initiator library. The
devices are cached in a per-mount-point list, for later use. At unmount
all devices are "put" back to the library.

objlayout_get_deviceinfo(), objlayout_put_deviceinfo() middleware
API for retrieving device information given a device_id.

TODO: The device cache can get big. Cap its size. Keep an LRU and start
      to return devices which were not used, when list gets to big, or
      when new entries allocation fail.

[pnfs-obj: Bugs in new global-device-cache code]
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
[gfp_flags]
[use global device cache]
[use layout driver in global device cache]
Signed-off-by: Benny Halevy <bhalevy@panasas.com>
13 years agopnfs-obj: decode layout, alloc/free lseg
Boaz Harrosh [Sun, 22 May 2011 16:50:20 +0000 (19:50 +0300)]
pnfs-obj: decode layout, alloc/free lseg

objlayout_alloc_lseg prepares an xdr_stream and calls the
raid engins objio_alloc_lseg() to allocate a private
pnfs_layout_segment.

objio_osd.c::objio_alloc_lseg() uses passed xdr_stream to
decode and store the layout_segment information in an
objio_segment struct, using the pnfs_osd_xdr.h API for
the actual parsing the layout xdr.

objlayout_free_lseg calls objio_free_lseg() to free the
allocated space.

Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
[gfp_flags]
[removed "extern" from function definitions]
Signed-off-by: Benny Halevy <bhalevy@panasas.com>
13 years agopnfs-obj: pnfs_osd XDR client implementation
Boaz Harrosh [Sun, 22 May 2011 16:49:57 +0000 (19:49 +0300)]
pnfs-obj: pnfs_osd XDR client implementation

* Add the fs/nfs/objlayout/pnfs_osd_xdr_cli.c file, which will
  include the XDR encode/decode implementations for the pNFS
  client objlayout driver.

[Wrong type in comments]
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
Signed-off-by: Benny Halevy <bhalevy@panasas.com>
13 years agopnfs-obj: pnfs_osd XDR definitions
Benny Halevy [Sun, 22 May 2011 16:49:32 +0000 (19:49 +0300)]
pnfs-obj: pnfs_osd XDR definitions

* Add the pnfs_osd_xdr.h header

* defintions the pnfs_osd_layout structure including all it's
  sub-types and constants.
* Declare the pnfs_osd_xdr_decode_layout API + all needed
  inline helpers.

* Define the pnfs_osd_deviceaddr structure and all its subtypes and
  constants.
* Declare API for decoding of a pnfs_osd_deviceaddr from XDR stream.

* Define the pnfs_osd_ioerr structure, its substructures and constants.
* Declare API for encoding of a pnfs_osd_ioerr into XDR stream.

* Define the pnfs_osd_layoutupdate structure and its substructures.
* Declare API for encoding of a pnfs_osd_layoutupdate into XDR stream.

[Remove server definitions]
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
Signed-off-by: Benny Halevy <bhalevy@panasas.com>
13 years agopnfs-obj: objlayoutdriver module skeleton
Benny Halevy [Sun, 22 May 2011 16:49:06 +0000 (19:49 +0300)]
pnfs-obj: objlayoutdriver module skeleton

* Define the PNFS_OBJLAYOUT Kconfig option in the nfs
  master Kconfig file.
* Add the objlayout driver to the Kernel's Kbuild system.
* Add the fs/nfs/objlayout/Kbuild file for building the
  objlayoutdriver.ko driver
* Define fs/nfs/objlayout/objio_osd.c, register the driver on module
  initialization and unregister on exit.

[pnfs-obj: remove of CONFIG_PNFS fallout]
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
[added "unsure" clause]
[depend on NFS_V4_1]
Signed-off-by: Benny Halevy <bhalevy@panasas.com>
13 years agopnfs: client stats
J. Bruce Fields [Sun, 22 May 2011 16:48:21 +0000 (19:48 +0300)]
pnfs: client stats

A pNFS client auto-negotiates a lot of features (minorversion level,
pNFS layout type, etc.).  This is convenient, but makes certain kinds of
failures hard for a user to detect.

For example, if the client falls back on 4.0, or falls back to MDS IO
because the user didn't connect to the right iscsi disks before
mounting, the only symptoms may be reduced performance, which may not be
noticed till long after the actual failure, and may be difficult for a
user to diagnose.

However, such "failures" may also be perfectly normal in some cases, so
we don't want to spam the system logs with them.

One approach would be to put some more information into
/proc/self/mountstats.

Signed-off-by: J. Bruce Fields <bfields@fieldses.org>
Signed-off-by: Benny Halevy <bhalevy@panasas.com>
[pnfs: add commit client stats]
[fixup data types for "ret" variables in pnfs_try_to* inline funcs.]
Signed-off-by: Benny Halevy <bhalevy@panasas.com>
[fix definition of show_pnfs for !CONFIG_PNFS]
Signed-off-by: Benny Halevy <bhalevy@panasas.com>
[nfs41: Fix show_sessions in the not CONFIG_NFS_V4_1 case]
    There is a build error when CONFIG_NFS_V4 is set but
    CONFIG_NFS_V4_1 is *not* set. show_sessions() prototype
    was unbalanced between the two cases.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
[pnfs: super.c remove CONFIG_PNFS]
Signed-off-by: Andy Adamson <andros@netapp.com>
Signed-off-by: Benny Halevy <bhalevy@panasas.com>
13 years agopnfs: Use byte-range for cb_layoutrecall
Benny Halevy [Sun, 22 May 2011 16:48:02 +0000 (19:48 +0300)]
pnfs: Use byte-range for cb_layoutrecall

Use recalled range to invalidate particular layout segments in the layout cache.

Signed-off-by: Benny Halevy <bhalevy@panasas.com>
13 years agopnfs: align layoutget requests on page boundaries
Benny Halevy [Sun, 22 May 2011 16:47:46 +0000 (19:47 +0300)]
pnfs: align layoutget requests on page boundaries

Signed-off-by: Benny Halevy <bhalevy@panasas.com>
13 years agopnfs: Use byte-range for layoutget
Benny Halevy [Sun, 22 May 2011 16:47:26 +0000 (19:47 +0300)]
pnfs: Use byte-range for layoutget

Add offset and count parameters to pnfs_update_layout and use them to get
the layout in the pageio path.

Order cache layout segments in the following order:
* offset (ascending)
* length (descending)
* iomode (RW before READ)

Test byte range against the layout segment in use in pnfs_{read,write}_pg_test
so not to coalesce pages not using the same layout segment.

[fix lseg ordering]
[clean up pnfs_find_lseg lseg arg]
[remove unnecessary FIXME]
[fix ordering in pnfs_insert_layout]
[clean up pnfs_insert_layout]
Signed-off-by: Benny Halevy <bhalevy@panasas.com>
13 years agoSUNRPC: introduce xdr_init_decode_pages
Benny Halevy [Thu, 19 May 2011 18:16:47 +0000 (14:16 -0400)]
SUNRPC: introduce xdr_init_decode_pages

Initialize xdr_stream and xdr_buf using an array of page pointers
and length of buffer.

Signed-off-by: Benny Halevy <bhalevy@panasas.com>
13 years agoNFSv4.1: use layout driver in global device cache
Benny Halevy [Tue, 24 May 2011 15:04:02 +0000 (18:04 +0300)]
NFSv4.1: use layout driver in global device cache

pnfs deviceids are unique per server, per layout type.
struct nfs_client is currently used to distinguish deviceids from
different nfs servers, yet these may clash between different layout
types on the same server.  Therefore, use the layout driver associated
with each deviceid at insertion time to look it up, unhash, or
delete it.

Signed-off-by: Benny Halevy <bhalevy@panasas.com>
13 years agopnfs: CB_NOTIFY_DEVICEID
Marc Eshel [Sun, 22 May 2011 16:47:09 +0000 (19:47 +0300)]
pnfs: CB_NOTIFY_DEVICEID

Note: This functionlaity is incomplete as all layout segments referring to
the 'to be removed device id' need to be reaped, and all in flight I/O drained.

[use be32 res in nfs4_callback_devicenotify]
[use nfs_client to qualify deviceid for cb_notify_deviceid]
[use global deviceid cache for CB_NOTIFY_DEVICEID]
[refactor device cache _lookup_deviceid]
[refactor device cache _find_get_deviceid]
Signed-off-by: Benny Halevy <bhalevy@panasas.com>
[Bug in new global-device-cache code]
[layout_driver MUST set free_deviceid_node if using dev-cache]
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
Signed-off-by: Benny Halevy <bhalevy@panasas.com>
13 years agoeCryptfs: Cleanup inode initialization code
Tyler Hicks [Tue, 24 May 2011 07:16:51 +0000 (02:16 -0500)]
eCryptfs: Cleanup inode initialization code

The eCryptfs inode get, initialization, and dentry interposition code
has two separate paths. One is for when dentry interposition is needed
after doing things like a mkdir in the lower filesystem and the other
is needed after a lookup. Unlocking new inodes and doing a d_add() needs
to happen at different times, depending on which type of dentry
interposing is being done.

This patch cleans up the inode get and initialization code paths and
splits them up so that the locking and d_add() differences mentioned
above can be handled appropriately in a later patch.

Signed-off-by: Tyler Hicks <tyhicks@linux.vnet.ibm.com>
Tested-by: David <david@unsolicited.net>
13 years agoNFSv4.1: purge deviceid cache on nfs_free_client
Benny Halevy [Fri, 20 May 2011 11:47:33 +0000 (13:47 +0200)]
NFSv4.1: purge deviceid cache on nfs_free_client

Use the pnfs_layoutdriver_type both as a qualifier for the deviceid,
distinguishing deviceid from different layout types on the server,
and for freeing the layout-driver allocated structure containing the
nfs4_deviceid_node.

[BUG in _deviceid_purge_client]
[layout_driver MUST set free_deviceid_node if using dev-cache]
[let ver < 4.1 compile]
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
[removed EXPORT_SYMBOL_GPL(nfs4_deviceid_purge_client)]
Signed-off-by: Benny Halevy <bhalevy@panasas.com>
13 years agoeCryptfs: Consolidate inode functions into inode.c
Tyler Hicks [Tue, 24 May 2011 02:18:20 +0000 (21:18 -0500)]
eCryptfs: Consolidate inode functions into inode.c

These functions should live in inode.c since their focus is on inodes
and they're primarily used by functions in inode.c.

Also does a simple cleanup of ecryptfs_inode_test() and rolls
ecryptfs_init_inode() into ecryptfs_inode_set().

Signed-off-by: Tyler Hicks <tyhicks@linux.vnet.ibm.com>
Tested-by: David <david@unsolicited.net>
13 years agomm, rmap: Add yet more comments to page_get_anon_vma/page_lock_anon_vma
Peter Zijlstra [Sun, 29 May 2011 08:33:44 +0000 (10:33 +0200)]
mm, rmap: Add yet more comments to page_get_anon_vma/page_lock_anon_vma

Inspired by an analysis from Hugh on why again all this doesn't explode
in our face.

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
13 years agodm kcopyd: return client directly and not through a pointer
Mikulas Patocka [Sun, 29 May 2011 12:03:13 +0000 (13:03 +0100)]
dm kcopyd: return client directly and not through a pointer

Return client directly from dm_kcopyd_client_create, not through a
parameter, making it consistent with dm_io_client_create.

Signed-off-by: Mikulas Patocka <mpatocka@redhat.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
13 years agodm kcopyd: reserve fewer pages
Mikulas Patocka [Sun, 29 May 2011 12:03:11 +0000 (13:03 +0100)]
dm kcopyd: reserve fewer pages

Reserve just the minimum of pages needed to process one job.

Because we allocate pages from page allocator, we don't need to reserve
a large number of pages.  The maximum job size is SUB_JOB_SIZE and we
calculate the number of reserved pages based on this.

Signed-off-by: Mikulas Patocka <mpatocka@redhat.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
13 years agodm io: use fixed initial mempool size
Mikulas Patocka [Sun, 29 May 2011 12:03:09 +0000 (13:03 +0100)]
dm io: use fixed initial mempool size

Replace the arbitrary calculation of an initial io struct mempool size
with a constant.

The code calculated the number of reserved structures based on the request
size and used a "magic" multiplication constant of 4.  This patch changes
it to reserve a fixed number - itself still chosen quite arbitrarily.
Further testing might show if there is a better number to choose.

Note that if there is no memory pressure, we can still allocate an
arbitrary number of "struct io" structures.  One structure is enough to
process the whole request.

Signed-off-by: Mikulas Patocka <mpatocka@redhat.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
13 years agodm kcopyd: alloc pages from the main page allocator
Mikulas Patocka [Sun, 29 May 2011 12:03:07 +0000 (13:03 +0100)]
dm kcopyd: alloc pages from the main page allocator

This patch changes dm-kcopyd so that it allocates pages from the main
page allocator with __GFP_NOWARN | __GFP_NORETRY flags (so that it can
fail in case of memory pressure). If the allocation fails, dm-kcopyd
allocates pages from its own reserve.

Signed-off-by: Mikulas Patocka <mpatocka@redhat.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
13 years agodm kcopyd: add gfp parm to alloc_pl
Mikulas Patocka [Sun, 29 May 2011 12:03:04 +0000 (13:03 +0100)]
dm kcopyd: add gfp parm to alloc_pl

Introduce a parameter for gfp flags to alloc_pl() for use in following
patches.

Signed-off-by: Mikulas Patocka <mpatocka@redhat.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
13 years agodm kcopyd: remove superfluous page allocation spinlock
Mikulas Patocka [Sun, 29 May 2011 12:03:02 +0000 (13:03 +0100)]
dm kcopyd: remove superfluous page allocation spinlock

Remove the spinlock protecting the pages allocation.  The spinlock is only
taken on initialization or from single-threaded workqueue.  Therefore, the
spinlock is useless.

The spinlock is taken in kcopyd_get_pages and kcopyd_put_pages.

kcopyd_get_pages is only called from run_pages_job, which is only
called from process_jobs called from do_work.

kcopyd_put_pages is called from client_alloc_pages (which is initialization
function) or from run_complete_job. run_complete_job is only called from
process_jobs called from do_work.

Another spinlock, kc->job_lock is taken each time someone pushes or pops
some work for the worker thread.  Once we take kc->job_lock, we
guarantee that any written memory is visible to the other CPUs.

Signed-off-by: Mikulas Patocka <mpatocka@redhat.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
13 years agodm kcopyd: preallocate sub jobs to avoid deadlock
Mikulas Patocka [Sun, 29 May 2011 12:03:00 +0000 (13:03 +0100)]
dm kcopyd: preallocate sub jobs to avoid deadlock

There's a possible theoretical deadlock in dm-kcopyd because multiple
allocations from the same mempool are required to finish a request.
Avoid this by preallocating sub jobs.

There is a mempool of 512 entries. Each request requires up to 9
entries from the mempool. If we have at least 57 concurrent requests
running, the mempool may overflow and mempool allocations may start
blocking until another entry is freed to the mempool. Because the same
thread is used to free entries to the mempool and allocate entries from
the mempool, this may result in a deadlock.

This patch changes it so that one mempool entry contains all 9 "struct
kcopyd_job" required to fulfill the whole request. The allocation is
done only once in dm_kcopyd_copy and no further mempool allocations are
done during request processing.

If dm_kcopyd_copy is not run in the completion thread, this
implementation is deadlock-free.

MIN_JOBS needs reducing accordingly and we've chosen to reduce it
further to 8.

Signed-off-by: Mikulas Patocka <mpatocka@redhat.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
13 years agodm kcopyd: avoid pointless job splitting
Mikulas Patocka [Sun, 29 May 2011 12:02:58 +0000 (13:02 +0100)]
dm kcopyd: avoid pointless job splitting

Don't split SUB_JOB_SIZE jobs

If the job size equals SUB_JOB_SIZE, there is no point in splitting it.
Splitting it just unnecessarily wastes time, because the split job size
is SUB_JOB_SIZE too.

Signed-off-by: Mikulas Patocka <mpatocka@redhat.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
13 years agodm mpath: do not fail paths after integrity errors
Martin K. Petersen [Sun, 29 May 2011 12:02:55 +0000 (13:02 +0100)]
dm mpath: do not fail paths after integrity errors

Integrity errors need to be passed to the owner of the integrity
metadata for processing. Consequently EILSEQ should be passed up the
stack.

Cc: stable@kernel.org
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Acked-by: Mike Snitzer <snitzer@redhat.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
13 years agodm table: reject devices without request fns
Milan Broz [Sun, 29 May 2011 12:02:52 +0000 (13:02 +0100)]
dm table: reject devices without request fns

This patch adds a check that a block device has a request function
defined before it is used.  Otherwise, misconfiguration can cause an oops.

Because we are allowing devices with zero size e.g. an offline multipath
device as in commit 2cd54d9bedb79a97f014e86c0da393416b264eb3
("dm: allow offline devices") there needs to be an additional check
to ensure devices are initialised.  Some block devices, like a loop
device without a backing file, exist but have no request function.

Reproducer is trivial: dm-mirror on unbound loop device
(no backing file on loop devices)

dmsetup create x --table "0 8 mirror core 2 8 sync 2 /dev/loop0 0 /dev/loop1 0"

and mirror resync will immediatelly cause OOps.

BUG: unable to handle kernel NULL pointer dereference at   (null)
 ? generic_make_request+0x2bd/0x590
 ? kmem_cache_alloc+0xad/0x190
 submit_bio+0x53/0xe0
 ? bio_add_page+0x3b/0x50
 dispatch_io+0x1ca/0x210 [dm_mod]
 ? read_callback+0x0/0xd0 [dm_mirror]
 dm_io+0xbb/0x290 [dm_mod]
 do_mirror+0x1e0/0x748 [dm_mirror]

Signed-off-by: Milan Broz <mbroz@redhat.com>
Reported-by: Zdenek Kabelac <zkabelac@redhat.com>
Acked-by: Mike Snitzer <snitzer@redhat.com>
Cc: stable@kernel.org
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
13 years agodm table: allow targets to support discards internally
Mike Snitzer [Sun, 29 May 2011 11:52:55 +0000 (12:52 +0100)]
dm table: allow targets to support discards internally

Permit a target to support discards regardless of whether or not all its
underlying devices do.

Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
13 years ago[S390] mm: fix mmu_gather rework
Heiko Carstens [Sun, 29 May 2011 10:40:51 +0000 (12:40 +0200)]
[S390] mm: fix mmu_gather rework

Quite a few functions that get called from the tlb gather code require that
preemption must be disabled. So disable preemption inside of the called
functions instead.
The only drawback is that rcu_table_freelist_finish() doesn't get necessarily
called on the cpu(s) that filled the free lists. So we may see a delay, until
we finally see an rcu callback. However over time this shouldn't matter.

So we get rid of lots of "BUG: using smp_processor_id() in preemptible"
messages.

Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>
13 years ago[S390] mm: fix storage key handling
Heiko Carstens [Sun, 29 May 2011 10:40:50 +0000 (12:40 +0200)]
[S390] mm: fix storage key handling

page_get_storage_key() and page_set_storage_key() expect a page address
and not its page frame number. This got inconsistent with 2d42552d
"[S390] merge page_test_dirty and page_clear_dirty".

Result is that we read/write storage keys from random pages and do not
have a working dirty bit tracking at all.
E.g. SetPageUpdate() doesn't clear the dirty bit of requested pages, which
for example ext4 doesn't like very much and panics after a while.

Unable to handle kernel paging request at virtual user address (null)
Oops: 0004 [#1] PREEMPT SMP DEBUG_PAGEALLOC
Modules linked in:
CPU: 1 Not tainted 2.6.39-07551-g139f37f-dirty #152
Process flush-94:0 (pid: 1576, task: 000000003eb34538, ksp: 000000003c287b70)
Krnl PSW : 0704c00180000000 0000000000316b12 (jbd2_journal_file_inode+0x10e/0x138)
           R:0 T:1 IO:1 EX:1 Key:0 M:1 W:0 P:0 AS:3 CC:0 PM:0 EA:3
Krnl GPRS: 0000000000000000 0000000000000000 0000000000000000 0700000000000000
           0000000000316a62 000000003eb34cd0 0000000000000025 000000003c287b88
           0000000000000001 000000003c287a70 000000003f1ec678 000000003f1ec000
           0000000000000000 000000003e66ec00 0000000000316a62 000000003c287988
Krnl Code: 0000000000316b04f0a0000407f4       srp     4(11,%r0),2036,0
           0000000000316b0ab9020022           ltgr    %r2,%r2
           0000000000316b0ea7740015           brc     7,316b38
          >0000000000316b12e3d0c0000024       stg     %r13,0(%r12)
           0000000000316b184120c010           la      %r2,16(%r12)
           0000000000316b1c4130d060           la      %r3,96(%r13)
           0000000000316b20e340d0600004       lg      %r4,96(%r13)
           0000000000316b26c0e50002b567       brasl   %r14,36d5f4
Call Trace:
([<0000000000316a62>] jbd2_journal_file_inode+0x5e/0x138)
 [<00000000002da13c>] mpage_da_map_and_submit+0x2e8/0x42c
 [<00000000002daac2>] ext4_da_writepages+0x2da/0x504
 [<00000000002597e8>] writeback_single_inode+0xf8/0x268
 [<0000000000259f06>] writeback_sb_inodes+0xd2/0x18c
 [<000000000025a700>] writeback_inodes_wb+0x80/0x168
 [<000000000025aa92>] wb_writeback+0x2aa/0x324
 [<000000000025abde>] wb_do_writeback+0xd2/0x274
 [<000000000025ae3a>] bdi_writeback_thread+0xba/0x1c4
 [<00000000001737be>] kthread+0xa6/0xb0
 [<000000000056c1da>] kernel_thread_starter+0x6/0xc
 [<000000000056c1d4>] kernel_thread_starter+0x0/0xc
INFO: lockdep is turned off.
Last Breaking-Event-Address:
 [<0000000000316a8a>] jbd2_journal_file_inode+0x86/0x138

Reported-by: Sebastian Ott <sebott@linux.vnet.ibm.com>
Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>
13 years agoNFSv4.1: make deviceid cache global
Benny Halevy [Fri, 20 May 2011 02:14:47 +0000 (22:14 -0400)]
NFSv4.1: make deviceid cache global

Move deviceid cache from the pnfs files layout driver to the
generic layer in preparation for the objects layout driver.

Signed-off-by: Benny Halevy <bhalevy@panasas.com>
13 years agopnfs: resolve header dependency in pnfs.h
Benny Halevy [Thu, 5 May 2011 05:28:46 +0000 (08:28 +0300)]
pnfs: resolve header dependency in pnfs.h

Some definitions in the header file depend on nfs_fs.h so pnfs.h can't
be included independently.

Signed-off-by: Benny Halevy <bhalevy@panasas.com>
13 years agoNFSv4.1: use struct nfs_client to qualify deviceid
Benny Halevy [Fri, 20 May 2011 08:45:05 +0000 (10:45 +0200)]
NFSv4.1: use struct nfs_client to qualify deviceid

deviceids are unique per server, per layout type.
Therefore, in the global cache in the files layout driver
deviceids from different servers may clash so we need
to qualify them with a struct nfs_client that represents
the nfs server that returned the deviceid.

Introduced in 2.6.39 commit ea8eecdd
"NFSv4.1 move deviceid cache to filelayout driver"

Signed-off-by: Benny Halevy <bhalevy@panasas.com>
13 years agoNFSv4.1: fix typo in filelayout_check_layout
Jim Rees [Wed, 23 Feb 2011 00:31:57 +0000 (19:31 -0500)]
NFSv4.1: fix typo in filelayout_check_layout

Signed-off-by: Jim Rees <rees@umich.edu>
Signed-off-by: Benny Halevy <bhalevy@panasas.com>
13 years agoSquashfs: Fix sanity check patches on big-endian systems
Phillip Lougher [Sat, 28 May 2011 23:38:46 +0000 (00:38 +0100)]
Squashfs: Fix sanity check patches on big-endian systems

le64 values should be swapped when accessing on
big-endian systems.

Signed-off-by: Phillip Lougher <phillip@lougher.demon.co.uk>
13 years agoMerge branch 'ec-cleanup' into release
Len Brown [Sun, 29 May 2011 08:40:39 +0000 (04:40 -0400)]
Merge branch 'ec-cleanup' into release

Conflicts:
drivers/platform/x86/compal-laptop.c

13 years agoMerge branches 'acpica', 'aml-custom', 'bugzilla-16548', 'bugzilla-20242', 'd3-cold...
Len Brown [Sun, 29 May 2011 08:38:48 +0000 (04:38 -0400)]
Merge branches 'acpica', 'aml-custom', 'bugzilla-16548', 'bugzilla-20242', 'd3-cold', 'ec-asus' and 'thermal-fix' into release

13 years agox86 idle: deprecate mwait_idle() and "idle=mwait" cmdline param
Len Brown [Fri, 1 Apr 2011 19:46:09 +0000 (15:46 -0400)]
x86 idle: deprecate mwait_idle() and "idle=mwait" cmdline param

mwait_idle() is a C1-only idle loop intended to be more efficient
than HLT on SMP hardware that supports it.

But mwait_idle() has been replaced by the more general
mwait_idle_with_hints(), which handles both C1 and deeper C-states.
ACPI uses only mwait_idle_with_hints(), and never uses mwait_idle().

Deprecate mwait_idle() and the "idle=mwait" cmdline param
to simplify the x86 idle code.

After this change, kernels configured with
(!CONFIG_ACPI=n && !CONFIG_INTEL_IDLE=n) when run on hardware
that support MWAIT will simply use HLT.  If MWAIT is desired
on those systems, cpuidle and the cpuidle drivers above
can be used.

cc: x86@kernel.org
cc: stable@kernel.org # .39.x
Signed-off-by: Len Brown <len.brown@intel.com>
13 years agox86 idle: deprecate "no-hlt" cmdline param
Len Brown [Fri, 1 Apr 2011 19:41:17 +0000 (15:41 -0400)]
x86 idle: deprecate "no-hlt" cmdline param

We'd rather that modern machines not check if HLT works on
every entry into idle, for the benefit of machines that had
marginal electricals 15-years ago.  If those machines are still running
the upstream kernel, they can use "idle=poll".  The only difference
will be that they'll now invoke HLT in machine_hlt().

cc: x86@kernel.org # .39.x
Signed-off-by: Len Brown <len.brown@intel.com>
13 years agox86 idle APM: deprecate CONFIG_APM_CPU_IDLE
Len Brown [Fri, 1 Apr 2011 19:19:23 +0000 (15:19 -0400)]
x86 idle APM: deprecate CONFIG_APM_CPU_IDLE

We don't want to export the pm_idle function pointer to modules.
Currently CONFIG_APM_CPU_IDLE w/ CONFIG_APM_MODULE forces us to.

CONFIG_APM_CPU_IDLE is of dubious value, it runs only on 32-bit
uniprocessor laptops that are over 10 years old.  It calls into
the BIOS during idle, and is known to cause a number of machines
to fail.

Removing CONFIG_APM_CPU_IDLE and will allow us to stop exporting
pm_idle.  Any systems that were calling into the APM BIOS
at run-time will simply use HLT instead.

cc: x86@kernel.org
cc: Jiri Kosina <jkosina@suse.cz>
cc: stable@kernel.org # .39.x
Signed-off-by: Len Brown <len.brown@intel.com>
13 years agox86 idle floppy: deprecate disable_hlt()
Len Brown [Fri, 1 Apr 2011 19:08:48 +0000 (15:08 -0400)]
x86 idle floppy: deprecate disable_hlt()

Plan to remove floppy_disable_hlt in 2012, an ancient
workaround with comments that it should be removed.

This allows us to remove clutter and a run-time branch
from the idle code.

WARN_ONCE() on invocation until it is removed.

cc: x86@kernel.org
cc: stable@kernel.org # .39.x
Signed-off-by: Len Brown <len.brown@intel.com>
13 years agox86 idle: EXPORT_SYMBOL(default_idle, pm_idle) only when APM demands it
Len Brown [Fri, 1 Apr 2011 19:28:09 +0000 (15:28 -0400)]
x86 idle: EXPORT_SYMBOL(default_idle, pm_idle) only when APM demands it

In the long run, we don't want default_idle() or (pm_idle)() to
be exported outside of process.c.  Start by not exporting them
to modules, unless the APM build demands it.

cc: x86@kernel.org
cc: Jiri Kosina <jkosina@suse.cz>
Signed-off-by: Len Brown <len.brown@intel.com>
13 years agox86 idle: clarify AMD erratum 400 workaround
Len Brown [Fri, 1 Apr 2011 20:59:53 +0000 (16:59 -0400)]
x86 idle: clarify AMD erratum 400 workaround

The workaround for AMD erratum 400 uses the term "c1e" falsely suggesting:
1. Intel C1E is somehow involved
2. All AMD processors with C1E are involved

Use the string "amd_c1e" instead of simply "c1e" to clarify that
this workaround is specific to AMD's version of C1E.
Use the string "e400" to clarify that the workaround is specific
to AMD processors with Erratum 400.

This patch is text-substitution only, with no functional change.

cc: x86@kernel.org
Acked-by: Borislav Petkov <borislav.petkov@amd.com>
Signed-off-by: Len Brown <len.brown@intel.com>
13 years agoACPI EC: remove redundant code
Zhang Rui [Tue, 26 Apr 2011 08:29:55 +0000 (16:29 +0800)]
ACPI EC: remove redundant code

ec->handle is set in ec_parse_device(), so don't bother to set it again.

Signed-off-by: Zhang Rui <rui.zhang@intel.com>
Signed-off-by: Len Brown <len.brown@intel.com>
13 years agoACPI: Add D3 cold state
Lin Ming [Wed, 4 May 2011 14:56:43 +0000 (22:56 +0800)]
ACPI: Add D3 cold state

_SxW returns an Integer containing the lowest D-state supported in state
Sx. If OSPM has not indicated that it supports _PR3, then the value “3”
corresponds to D3.  If it has indicated _PR3 support, the value “3”
represents D3hot and the value “4” represents D3cold.

Linux does set _OSC._PR3, so we should fix it to expect that _SxW can
return 4.

Signed-off-by: Lin Ming <ming.m.lin@intel.com>
Acked-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Len Brown <len.brown@intel.com>
13 years agoACPI: processor: fix processor_physically_present in UP kernel
Lin Ming [Mon, 16 May 2011 01:11:00 +0000 (09:11 +0800)]
ACPI: processor: fix processor_physically_present in UP kernel

Usually, there are multiple processors defined in ACPI table, for
example

    Scope (_PR)
    {
        Processor (CPU0, 0x00, 0x00000410, 0x06) {}
        Processor (CPU1, 0x01, 0x00000410, 0x06) {}
        Processor (CPU2, 0x02, 0x00000410, 0x06) {}
        Processor (CPU3, 0x03, 0x00000410, 0x06) {}
    }

processor_physically_present(...) will be called to check whether those
processors are physically present.

Currently we have below codes in processor_physically_present,

cpuid = acpi_get_cpuid(...);
if ((cpuid == -1) && (num_possible_cpus() > 1))
        return false;
return true;

In UP kernel, acpi_get_cpuid(...) always return -1 and
num_possible_cpus() always return 1, so
processor_physically_present(...) always returns true for all passed in
processor handles.

This is wrong for UP processor or SMP processor running UP kernel.

This patch removes the !SMP version of acpi_get_cpuid(), so both UP and
SMP kernel use the same acpi_get_cpuid function.

And for UP kernel, only processor 0 is valid.

https://bugzilla.kernel.org/show_bug.cgi?id=16548
https://bugzilla.kernel.org/show_bug.cgi?id=16357

Tested-by: Anton Kochkov <anton.kochkov@gmail.com>
Tested-by: Ambroz Bizjak <ambrop7@gmail.com>
Signed-off-by: Lin Ming <ming.m.lin@intel.com>
Signed-off-by: Len Brown <len.brown@intel.com>
13 years agoMerge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/vapier...
Linus Torvalds [Sun, 29 May 2011 06:12:28 +0000 (23:12 -0700)]
Merge branch 'for-linus' of git://git./linux/kernel/git/vapier/blackfin

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/vapier/blackfin:
  Blackfin: debug-mmrs: include RSI_PID[4567] MMRs
  Blackfin: bf51x: fix up RSI_PID# MMR defines
  Blackfin: bf52x/bf54x: fix up usb MMR defines
  Blackfin: debug-mmrs: fix typos with gptimers/mdma/ppi
  Blackfin: gptimers: add structure for hardware register layout
  Blackfin: wire up new sendmmsg syscall
  Blackfin: mach/bfin_serial_5xx.h: punt now-unused header
  Blackfin: bfin_serial.h: turn default port wrappers into stubs

13 years agoscsi: fix scsi_proc new kernel-doc warning
Randy Dunlap [Sun, 29 May 2011 00:26:40 +0000 (17:26 -0700)]
scsi: fix scsi_proc new kernel-doc warning

Fix kernel-doc warnings in scsi_proc.c:

  Warning(drivers/scsi/scsi_proc.c:390): No description found for parameter 'dev'
  Warning(drivers/scsi/scsi_proc.c:390): No description found for parameter 'data'
  Warning(drivers/scsi/scsi_proc.c:390): Excess function parameter 's' description in 'always_match'
  Warning(drivers/scsi/scsi_proc.c:390): Excess function parameter 'p' description in 'always_match'

Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
13 years agoACPI: Split out custom_method functionality into an own driver
Thomas Renninger [Thu, 26 May 2011 10:26:24 +0000 (12:26 +0200)]
ACPI: Split out custom_method functionality into an own driver

With /sys/kernel/debug/acpi/custom_method root can write
to arbitrary memory and increase his priveleges, even if
these are restricted.

-> Make this an own debug .config option and warn about the
security issue in the config description.

-> Still keep acpi/debugfs.c which now only creates an empty
   /sys/kernel/debug/acpi directory. There might be other
   users of it later.

Signed-off-by: Thomas Renninger <trenn@suse.de>
Acked-by: Rafael J. Wysocki <rjw@sisk.pl>
Acked-by: rui.zhang@intel.com
Signed-off-by: Len Brown <len.brown@intel.com>
13 years agoACPI: Cleanup custom_method debug stuff
Thomas Renninger [Thu, 26 May 2011 10:26:23 +0000 (12:26 +0200)]
ACPI: Cleanup custom_method debug stuff

- Move param aml_debug_output to other params into sysfs.c
- Split acpi_debugfs_init to prepare custom_method to be
  an own .config option and driver.

Signed-off-by: Thomas Renninger <trenn@suse.de>
Acked-by: Rafael J. Wysocki <rjw@sisk.pl>
Acked-by: rui.zhang@intel.com
Signed-off-by: Len Brown <len.brown@intel.com>
13 years agoACPI EC: enable MSI workaround for Quanta laptops
Zhang Rui [Tue, 26 Apr 2011 08:30:02 +0000 (16:30 +0800)]
ACPI EC: enable MSI workaround for Quanta laptops

Enable MSI workaround for Quanta laptops.
https://bugzilla.kernel.org/show_bug.cgi?id=20242

Tested-by: Jan-Matthias Braun <jan_braun@gmx.net>
Signed-off-by: Zhang Rui <rui.zhang@intel.com>
Signed-off-by: Len Brown <len.brown@intel.com>
13 years agoidle governor: Avoid lock acquisition to read pm_qos before entering idle
Tim Chen [Fri, 11 Feb 2011 20:49:04 +0000 (12:49 -0800)]
idle governor: Avoid lock acquisition to read pm_qos before entering idle

Thanks to the reviews and comments by Rafael, James, Mark and Andi.
Here's version 2 of the patch incorporating your comments and also some
update to my previous patch comments.

I noticed that before entering idle state, the menu idle governor will
look up the current pm_qos target value according to the list of qos
requests received.  This look up currently needs the acquisition of a
lock to access the list of qos requests to find the qos target value,
slowing down the entrance into idle state due to contention by multiple
cpus to access this list.  The contention is severe when there are a lot
of cpus waking and going into idle.  For example, for a simple workload
that has 32 pair of processes ping ponging messages to each other, where
64 cpu cores are active in test system, I see the following profile with
37.82% of cpu cycles spent in contention of pm_qos_lock:

-     37.82%          swapper  [kernel.kallsyms]          [k]
_raw_spin_lock_irqsave
   - _raw_spin_lock_irqsave
      - 95.65% pm_qos_request
           menu_select
           cpuidle_idle_call
         - cpu_idle
              99.98% start_secondary

A better approach will be to cache the updated pm_qos target value so
reading it does not require lock acquisition as in the patch below.
With this patch the contention for pm_qos_lock is removed and I saw a
2.2X increase in throughput for my message passing workload.

cc: stable@kernel.org
Signed-off-by: Tim Chen <tim.c.chen@linux.intel.com>
Acked-by: Andi Kleen <ak@linux.intel.com>
Acked-by: James Bottomley <James.Bottomley@suse.de>
Acked-by: mark gross <markgross@thegnar.org>
Signed-off-by: Len Brown <len.brown@intel.com>
13 years agocpuidle: menu: fixed wrapping timers at 4.294 seconds
Tero Kristo [Thu, 24 Feb 2011 15:19:23 +0000 (17:19 +0200)]
cpuidle: menu: fixed wrapping timers at 4.294 seconds

Cpuidle menu governor is using u32 as a temporary datatype for storing
nanosecond values which wrap around at 4.294 seconds. This causes errors
in predicted sleep times resulting in higher than should be C state
selection and increased power consumption. This also breaks cpuidle
state residency statistics.

cc: stable@kernel.org # .32.x through .39.x
Signed-off-by: Tero Kristo <tero.kristo@nokia.com>
Signed-off-by: Len Brown <len.brown@intel.com>
13 years agomm: fix page_lock_anon_vma leaving mutex locked
Hugh Dickins [Sat, 28 May 2011 20:20:21 +0000 (13:20 -0700)]
mm: fix page_lock_anon_vma leaving mutex locked

On one machine I've been getting hangs, a page fault's anon_vma_prepare()
waiting in anon_vma_lock(), other processes waiting for that page's lock.

This is a replay of last year's f18194275c39 "mm: fix hang on
anon_vma->root->lock".

The new page_lock_anon_vma() places too much faith in its refcount: when
it has acquired the mutex_trylock(), it's possible that a racing task in
anon_vma_alloc() has just reallocated the struct anon_vma, set refcount
to 1, and is about to reset its anon_vma->root.

Fix this by saving anon_vma->root, and relying on the usual page_mapped()
check instead of a refcount check: if page is still mapped, the anon_vma
is still ours; if page is not still mapped, we're no longer interested.

Signed-off-by: Hugh Dickins <hughd@google.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
13 years agomm: fix kernel BUG at mm/rmap.c:1017!
Hugh Dickins [Sat, 28 May 2011 20:17:04 +0000 (13:17 -0700)]
mm: fix kernel BUG at mm/rmap.c:1017!

I've hit the "address >= vma->vm_end" check in do_page_add_anon_rmap()
just once.  The stack showed khugepaged allocation trying to compact
pages: the call to page_add_anon_rmap() coming from remove_migration_pte().

That path holds anon_vma lock, but does not hold mmap_sem: it can
therefore race with a split_vma(), and in commit 5f70b962ccc2 "mmap:
avoid unnecessary anon_vma lock" we just took away the anon_vma lock
protection when adjusting vma->vm_end.

I don't think that particular BUG_ON ever caught anything interesting,
so better replace it by a comment, than reinstate the anon_vma locking.

Signed-off-by: Hugh Dickins <hughd@google.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
13 years agotmpfs: fix race between truncate and writepage
Hugh Dickins [Sat, 28 May 2011 20:14:09 +0000 (13:14 -0700)]
tmpfs: fix race between truncate and writepage

While running fsx on tmpfs with a memhog then swapoff, swapoff was hanging
(interruptibly), repeatedly failing to locate the owner of a 0xff entry in
the swap_map.

Although shmem_writepage() does abandon when it sees incoming page index
is beyond eof, there was still a window in which shmem_truncate_range()
could come in between writepage's dropping lock and updating swap_map,
find the half-completed swap_map entry, and in trying to free it,
leave it in a state that swap_shmem_alloc() could not correct.

Arguably a bug in __swap_duplicate()'s and swap_entry_free()'s handling
of the different cases, but easiest to fix by moving swap_shmem_alloc()
under cover of the lock.

More interesting than the bug: it's been there since 2.6.33, why could
I not see it with earlier kernels?  The mmotm of two weeks ago seems to
have some magic for generating races, this is just one of three I found.

With yesterday's git I first saw this in mainline, bisected in search of
that magic, but the easy reproducibility evaporated.  Oh well, fix the bug.

Signed-off-by: Hugh Dickins <hughd@google.com>
Cc: stable@kernel.org
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>