firefly-linux-kernel-4.4.55.git
16 years ago[S390] qdio: speed up multicast traffic on full HiperSocket queue
Ursula Braun [Fri, 10 Oct 2008 19:33:04 +0000 (21:33 +0200)]
[S390] qdio: speed up multicast traffic on full HiperSocket queue

If an asynchronous HiperSockets queue runs full, no further packet
can be sent. In this case the next initiative to give transmitted
skbs back to the stack is triggered only by a 10-seconds qdio timer.
This timer has been introduced for low multicast traffic scenarios
to guarantee freeing of skbs in a limited amount of time. For high
HiperSocket multicast traffic scenarios progress checking on the
outbound queue should be enforced by tasklet rescheduling.

Signed-off-by: Ursula Braun <ursula.braun@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
16 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/herbert/crypto-2.6
Linus Torvalds [Fri, 10 Oct 2008 18:20:42 +0000 (11:20 -0700)]
Merge git://git./linux/kernel/git/herbert/crypto-2.6

* git://git.kernel.org/pub/scm/linux/kernel/git/herbert/crypto-2.6:
  crypto: skcipher - Use RNG interface instead of get_random_bytes
  crypto: rng - RNG interface and implementation
  crypto: api - Add fips_enable flag
  crypto: skcipher - Move IV generators into their own modules
  crypto: cryptomgr - Test ciphers using ECB
  crypto: api - Use test infrastructure
  crypto: cryptomgr - Add test infrastructure
  crypto: tcrypt - Add alg_test interface
  crypto: tcrypt - Abort and only log if there is an error
  crypto: crc32c - Use Intel CRC32 instruction
  crypto: tcrypt - Avoid using contiguous pages
  crypto: api - Display larval objects properly
  crypto: api - Export crypto_alg_lookup instead of __crypto_alg_lookup
  crypto: Kconfig - Replace leading spaces with tabs

16 years agoMerge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/roland...
Linus Torvalds [Fri, 10 Oct 2008 18:16:33 +0000 (11:16 -0700)]
Merge branch 'for-linus' of git://git./linux/kernel/git/roland/infiniband

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/roland/infiniband: (29 commits)
  RDMA/nes: Fix slab corruption
  IB/mlx4: Set RLKEY bit for kernel QPs
  RDMA/nes: Correct error_module bit mask
  RDMA/nes: Fix routed RDMA connections
  RDMA/nes: Enhanced PFT management scheme
  RDMA/nes: Handle AE bounds violation
  RDMA/nes: Limit critical error interrupts
  RDMA/nes: Stop spurious MAC interrupts
  RDMA/nes: Correct tso_wqe_length
  RDMA/nes: Fill in firmware version for ethtool
  RDMA/nes: Use ethtool timer value
  RDMA/nes: Correct MAX TSO frags value
  RDMA/nes: Enable MC/UC after changing MTU
  RDMA/nes: Free NIC TX buffers when destroying NIC QP
  RDMA/nes: Fix MDC setting
  RDMA/nes: Add wqm_quanta module option
  RDMA/nes: Module parameter permissions
  RDMA/cxgb3: Set active_mtu in ib_port_attr
  RDMA/nes: Add support for 4-port 1G HP blade card
  RDMA/nes: Make mini_cm_connect() static
  ...

16 years agoMerge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/teigland/dlm
Linus Torvalds [Fri, 10 Oct 2008 18:13:55 +0000 (11:13 -0700)]
Merge branch 'for-linus' of git://git./linux/kernel/git/teigland/dlm

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/teigland/dlm:
  dlm: choose better identifiers
  dlm: remove bkl
  dlm: fix address compare
  dlm: fix locking of lockspace list in dlm_scand
  dlm: detect available userspace daemon
  dlm: allow multiple lockspace creates

16 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/agk/linux-2.6-dm
Linus Torvalds [Fri, 10 Oct 2008 18:11:47 +0000 (11:11 -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: detect lost queue
  dm: publish dm_vcalloc
  dm: publish dm_table_unplug_all
  dm: publish dm_get_mapinfo
  dm: export struct dm_dev
  dm crypt: avoid unnecessary wait when splitting bio
  dm crypt: tidy ctx pending
  dm crypt: fix async inc_pending
  dm crypt: move dec_pending on error into write_io_submit
  dm crypt: remove inc_pending from write_io_submit
  dm crypt: tidy write loop pending
  dm crypt: tidy crypt alloc
  dm crypt: tidy inc pending
  dm exception store: use chunk_t for_areas
  dm exception store: introduce area_location function
  dm raid1: kcopyd should stop on error if errors handled
  dm mpath: remove is_active from struct dm_path
  dm mpath: use more error codes

Fixed up trivial conflict in drivers/md/dm-mpath.c manually.

16 years agoFix barrier fail detection in XFS
Christoph Hellwig [Fri, 10 Oct 2008 06:28:29 +0000 (17:28 +1100)]
Fix barrier fail detection in XFS

Currently we disable barriers as soon as we get a buffer in xlog_iodone
that has the XBF_ORDERED flag cleared.  But this can be the case not only
for buffers where the barrier failed, but also the first buffer of a
split log write in case of a log wraparound.  Due to the disabled
barriers we can easily get directory corruption on unclean shutdowns.
So instead of using this check add a new buffer flag for failed barrier
writes.

This is a regression vs 2.6.26 caused by patch to use the right macro
to check for the ORDERED flag, as we previously got true returned for
every buffer.

Thanks to Toei Rei for reporting the bug.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Eric Sandeen <sandeen@sandeen.net>
Reviewed-by: David Chinner <david@fromorbit.com>
Signed-off-by: Tim Shimmin <tes@sgi.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/steve/gfs2-2.6-nmw
Linus Torvalds [Fri, 10 Oct 2008 18:02:22 +0000 (11:02 -0700)]
Merge git://git./linux/kernel/git/steve/gfs2-2.6-nmw

* git://git.kernel.org/pub/scm/linux/kernel/git/steve/gfs2-2.6-nmw:
  GFS2: Support for I/O barriers
  GFS2: Add UUID to GFS2 sb
  GFS2: high time to take some time over atime
  GFS2: The war on bloat
  GFS2: GFS2 will panic if you misspell any mount options
  GFS2: Direct IO write at end of file error
  GFS2: Use an IS_ERR test rather than a NULL test
  GFS2: Fix race relating to glock min-hold time
  GFS2: Fix & clean up GFS2 rename
  GFS2: rm on multiple nodes causes panic
  GFS2: Fix metafs mounts
  GFS2: Fix debugfs glock file iterator

16 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/jejb/scsi-misc-2.6
Linus Torvalds [Fri, 10 Oct 2008 17:53:26 +0000 (10:53 -0700)]
Merge git://git./linux/kernel/git/jejb/scsi-misc-2.6

* git://git.kernel.org/pub/scm/linux/kernel/git/jejb/scsi-misc-2.6: (37 commits)
  [SCSI] zfcp: fix double dbf id usage
  [SCSI] zfcp: wait on SCSI work to be finished before proceeding with init dev
  [SCSI] zfcp: fix erp list usage without using locks
  [SCSI] zfcp: prevent fc_remote_port_delete calls for unregistered rport
  [SCSI] zfcp: fix deadlock caused by shared work queue tasks
  [SCSI] zfcp: put threshold data in hba trace
  [SCSI] zfcp: Simplify zfcp data structures
  [SCSI] zfcp: Simplify get_adapter_by_busid
  [SCSI] zfcp: remove all typedefs and replace them with standards
  [SCSI] zfcp: attach and release SAN nameserver port on demand
  [SCSI] zfcp: remove unused references, declarations and flags
  [SCSI] zfcp: Update message with input from review
  [SCSI] zfcp: add queue_full sysfs attribute
  [SCSI] scsi_dh: suppress comparison warning
  [SCSI] scsi_dh: add Dell product information into rdac device handler
  [SCSI] qla2xxx: remove the unused SCSI_QLOGIC_FC_FIRMWARE option
  [SCSI] qla2xxx: fix printk format warnings
  [SCSI] qla2xxx: Update version number to 8.02.01-k8.
  [SCSI] qla2xxx: Ignore payload reserved-bits during RSCN processing.
  [SCSI] qla2xxx: Additional residual-count corrections during UNDERRUN handling.
  ...

16 years agoMerge branch 'for-2.6.28' of git://git.kernel.dk/linux-2.6-block
Linus Torvalds [Fri, 10 Oct 2008 17:52:45 +0000 (10:52 -0700)]
Merge branch 'for-2.6.28' of git://git.kernel.dk/linux-2.6-block

* 'for-2.6.28' of git://git.kernel.dk/linux-2.6-block: (132 commits)
  doc/cdrom: Trvial documentation error, file not present
  block_dev: fix kernel-doc in new functions
  block: add some comments around the bio read-write flags
  block: mark bio_split_pool static
  block: Find bio sector offset given idx and offset
  block: gendisk integrity wrapper
  block: Switch blk_integrity_compare from bdev to gendisk
  block: Fix double put in blk_integrity_unregister
  block: Introduce integrity data ownership flag
  block: revert part of d7533ad0e132f92e75c1b2eb7c26387b25a583c1
  bio.h: Remove unused conditional code
  block: remove end_{queued|dequeued}_request()
  block: change elevator to use __blk_end_request()
  gdrom: change to use __blk_end_request()
  memstick: change to use __blk_end_request()
  virtio_blk: change to use __blk_end_request()
  blktrace: use BLKTRACE_BDEV_SIZE as the name size for setup structure
  block: add lld busy state exporting interface
  block: Fix blk_start_queueing() to not kick a stopped queue
  include blktrace_api.h in headers_install
  ...

16 years agoMerge phase #1 of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip
Linus Torvalds [Fri, 10 Oct 2008 15:07:53 +0000 (08:07 -0700)]
Merge phase #1 of git://git./linux/kernel/git/tip/linux-2.6-tip

This merges phase 1 of the x86 tree, which is a collection of branches:

  x86/alternatives, x86/cleanups, x86/commandline, x86/crashdump,
  x86/debug, x86/defconfig, x86/doc, x86/exports, x86/fpu, x86/gart,
  x86/idle, x86/mm, x86/mtrr, x86/nmi-watchdog, x86/oprofile,
  x86/paravirt, x86/reboot, x86/sparse-fixes, x86/tsc, x86/urgent and
  x86/vmalloc

and as Ingo says: "these are the easiest, purely independent x86 topics
with no conflicts, in one nice Octopus merge".

* 'x86-v28-for-linus-phase1' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip: (147 commits)
  x86: mtrr_cleanup: treat WRPROT as UNCACHEABLE
  x86: mtrr_cleanup: first 1M may be covered in var mtrrs
  x86: mtrr_cleanup: print out correct type v2
  x86: trivial printk fix in efi.c
  x86, debug: mtrr_cleanup print out var mtrr before change it
  x86: mtrr_cleanup try gran_size to less than 1M, v3
  x86: mtrr_cleanup try gran_size to less than 1M, cleanup
  x86: change MTRR_SANITIZER to def_bool y
  x86, debug printouts: IOMMU setup failures should not be KERN_ERR
  x86: export set_memory_ro and set_memory_rw
  x86: mtrr_cleanup try gran_size to less than 1M
  x86: mtrr_cleanup prepare to make gran_size to less 1M
  x86: mtrr_cleanup safe to get more spare regs now
  x86_64: be less annoying on boot, v2
  x86: mtrr_cleanup hole size should be less than half of chunk_size, v2
  x86: add mtrr_cleanup_debug command line
  x86: mtrr_cleanup optimization, v2
  x86: don't need to go to chunksize to 4G
  x86_64: be less annoying on boot
  x86, olpc: fix endian bug in openfirmware workaround
  ...

16 years agoPnP: move pnpacpi/pnpbios_init to after PCI init
Linus Torvalds [Fri, 10 Oct 2008 15:00:17 +0000 (08:00 -0700)]
PnP: move pnpacpi/pnpbios_init to after PCI init

We already did that a long time ago for pnp_system_init, but
pnpacpi_init and pnpbios_init remained as subsys_initcalls, and get
linked into the kernel before the arch-specific routines that finalize
the PCI resources (pci_subsys_init).

This means that the PnP routines would either register their resources
before the PCI layer could, or would be unable to check whether a PCI
resource had already been registered.  Both are problematic.

I wanted to do this before 2.6.27, but every time we change something
like this, something breaks.  That said, _every_ single time we trust
some firmware (like PnP tables) more than we trust the hardware itself
(like PCI probing), the problems have been worse.

Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agoMerge branch 'upstream-2.6.28' of git://git.kernel.org/pub/scm/linux/kernel/git/jgarz...
Linus Torvalds [Fri, 10 Oct 2008 14:46:45 +0000 (07:46 -0700)]
Merge branch 'upstream-2.6.28' of git://git./linux/kernel/git/jgarzik/libata-dev

* 'upstream-2.6.28' of git://git.kernel.org/pub/scm/linux/kernel/git/jgarzik/libata-dev:
  ata_piix: IDE Mode SATA patch for Intel Ibex Peak DeviceIDs
  libata-eh: clear UNIT ATTENTION after reset
  ata_piix: add Hercules EC-900 mini-notebook to ich_laptop short cable list
  libata: reorder ata_device to remove 8 bytes of padding on 64 bits
  [libata] pata_bf54x: Add proper PM operation
  pata_sil680: convert CONFIG_PPC_MERGE to CONFIG_PPC
  libata: Implement disk shock protection support
  [libata] Introduce ata_id_has_unload()
  PATA: RPC now selects HAVE_PATA_PLATFORM for pata platform driver
  ata_piix: drop merged SCR access and use slave_link instead
  libata: implement slave_link
  libata: misc updates to prepare for slave link
  libata: reimplement link iterator
  libata: make SCR access ops per-link

16 years agodm: detect lost queue
Alasdair G Kergon [Fri, 10 Oct 2008 12:37:13 +0000 (13:37 +0100)]
dm: detect lost queue

Detect and report buggy drivers that destroy their request_queue.

Signed-off-by: Alasdair G Kergon <agk@redhat.com>
Cc: Stefan Raspl <raspl@linux.vnet.ibm.com>
Cc: Jens Axboe <jens.axboe@oracle.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
16 years agodm: publish dm_vcalloc
Mikulas Patocka [Fri, 10 Oct 2008 12:37:12 +0000 (13:37 +0100)]
dm: publish dm_vcalloc

Publish dm_vcalloc in include/linux/device-mapper.h because this function is
used by targets.

Signed-off-by: Mikulas Patocka <mpatocka@redhat.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
16 years agodm: publish dm_table_unplug_all
Mikulas Patocka [Fri, 10 Oct 2008 12:37:11 +0000 (13:37 +0100)]
dm: publish dm_table_unplug_all

Publish dm_table_unplug_all in include/linux/device-mapper.h because this
function is used by targets.

Signed-off-by: Mikulas Patocka <mpatocka@redhat.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
16 years agodm: publish dm_get_mapinfo
Mikulas Patocka [Fri, 10 Oct 2008 12:37:10 +0000 (13:37 +0100)]
dm: publish dm_get_mapinfo

Publish dm_get_mapinfo in include/linux/device-mapper.h because this function
is used by targets.

Signed-off-by: Mikulas Patocka <mpatocka@redhat.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
16 years agodm: export struct dm_dev
Mikulas Patocka [Fri, 10 Oct 2008 12:37:09 +0000 (13:37 +0100)]
dm: export struct dm_dev

Split struct dm_dev in two and publish the part that other targets need in
include/linux/device-mapper.h.

Signed-off-by: Mikulas Patocka <mpatocka@redhat.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
16 years agodm crypt: avoid unnecessary wait when splitting bio
Milan Broz [Fri, 10 Oct 2008 12:37:08 +0000 (13:37 +0100)]
dm crypt: avoid unnecessary wait when splitting bio

Don't wait between submitting crypt requests for a bio unless
we are short of memory.

There are two situations when we must split an encrypted bio:
  1) there are no free pages;
  2) the new bio would violate underlying device restrictions
(e.g. max hw segments).

In case (2) we do not need to wait.

Add output variable to crypt_alloc_buffer() to distinguish between
these cases.

Signed-off-by: Milan Broz <mbroz@redhat.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
16 years agodm crypt: tidy ctx pending
Milan Broz [Fri, 10 Oct 2008 12:37:08 +0000 (13:37 +0100)]
dm crypt: tidy ctx pending

Move the initialisation of ctx->pending into one place, at the
start of crypt_convert().

Introduce crypt_finished to indicate whether or not the encryption
is finished, for use in a later patch.

No functional change.

Signed-off-by: Milan Broz <mbroz@redhat.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
16 years agodm crypt: fix async inc_pending
Milan Broz [Fri, 10 Oct 2008 12:37:07 +0000 (13:37 +0100)]
dm crypt: fix async inc_pending

The pending reference count must be incremented *before* the async work is
queued to another thread, not after.  Otherwise there's a race if the
work completes and decrements the reference count before it gets incremented.

Signed-off-by: Milan Broz <mbroz@redhat.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
16 years agodm crypt: move dec_pending on error into write_io_submit
Milan Broz [Fri, 10 Oct 2008 12:37:06 +0000 (13:37 +0100)]
dm crypt: move dec_pending on error into write_io_submit

Make kcryptd_crypt_write_io_submit() responsible for decrementing
the pending count after an error.

Also fixes a bug in the async path that forgot to decrement it.

Signed-off-by: Milan Broz <mbroz@redhat.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
16 years agodm crypt: remove inc_pending from write_io_submit
Alasdair G Kergon [Fri, 10 Oct 2008 12:37:05 +0000 (13:37 +0100)]
dm crypt: remove inc_pending from write_io_submit

Make the caller reponsible for incrementing the pending count before calling
kcryptd_crypt_write_io_submit() in the non-async case to bring it into line
with the async case.

Signed-off-by: Alasdair G Kergon <agk@redhat.com>
16 years agodm crypt: tidy write loop pending
Milan Broz [Fri, 10 Oct 2008 12:37:04 +0000 (13:37 +0100)]
dm crypt: tidy write loop pending

Move kcryptd_crypt_write_convert_loop inside kcryptd_crypt_write_convert.
This change is needed for a later patch.

No functional change.

Signed-off-by: Milan Broz <mbroz@redhat.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
16 years agodm crypt: tidy crypt alloc
Milan Broz [Fri, 10 Oct 2008 12:37:03 +0000 (13:37 +0100)]
dm crypt: tidy crypt alloc

Factor out crypt io allocation code.
Later patches will call it from another place.

No functional change.

Signed-off-by: Milan Broz <mbroz@redhat.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
16 years agodm crypt: tidy inc pending
Milan Broz [Fri, 10 Oct 2008 12:37:02 +0000 (13:37 +0100)]
dm crypt: tidy inc pending

Move io pending to one place.

No functional change, usefull to simplify debugging.

Signed-off-by: Milan Broz <mbroz@redhat.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
16 years agodm exception store: use chunk_t for_areas
Mikulas Patocka [Fri, 10 Oct 2008 12:37:01 +0000 (13:37 +0100)]
dm exception store: use chunk_t for_areas

Change uint32_t into chunk_t to remove 32-bit limitation on the
number of chunks on systems with 64-bit sector numbers.

Signed-off-by: Mikulas Patocka <mpatocka@redhat.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
16 years agodm exception store: introduce area_location function
Mikulas Patocka [Fri, 10 Oct 2008 12:37:00 +0000 (13:37 +0100)]
dm exception store: introduce area_location function

Move this logic to a function, because it will be reused later.

Signed-off-by: Mikulas Patocka <mpatocka@redhat.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
16 years agodm raid1: kcopyd should stop on error if errors handled
Jonathan Brassow [Fri, 10 Oct 2008 12:36:59 +0000 (13:36 +0100)]
dm raid1: kcopyd should stop on error if errors handled

dm-raid1 is setting the 'DM_KCOPYD_IGNORE_ERROR' flag unconditionally
when assigning kcopyd work.  kcopyd is responsible for copying an
assigned section of disk to one or more other disks.  The
'DM_KCOPYD_IGNORE_ERROR' flag affects kcopyd in the following way:

When not set:
kcopyd will immediately stop the copy operation when an error is
encountered.

When set:
kcopyd will try to proceed regardless of errors and try to continue
copying any remaining amount.

Since dm-raid1 tracks regions of the address space that are (or
are not) in sync and it now has the ability to handle these
errors, we can safely enable this optimization.  This optimization
is conditional on whether mirror error handling has been enabled.

Signed-off-by: Jonathan Brassow <jbrassow@redhat.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
16 years agodm mpath: remove is_active from struct dm_path
Kiyoshi Ueda [Fri, 10 Oct 2008 12:36:58 +0000 (13:36 +0100)]
dm mpath: remove is_active from struct dm_path

This patch moves 'is_active' from struct dm_path to struct pgpath
as it does not need exporting.

Signed-off-by: Kiyoshi Ueda <k-ueda@ct.jp.nec.com>
Signed-off-by: Jun'ichi Nomura <j-nomura@ce.jp.nec.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
16 years agodm mpath: use more error codes
Benjamin Marzinski [Fri, 10 Oct 2008 12:36:57 +0000 (13:36 +0100)]
dm mpath: use more error codes

This patch allows path errors from the multipath ctr function to
propagate up to userspace as errno values from the ioctl() call.

This is in response to
  https://www.redhat.com/archives/dm-devel/2008-May/msg00000.html
and
  https://bugzilla.redhat.com/show_bug.cgi?id=444421

The patch only lets through the errors that it needs to in order to
get the path errors from parse_path().

Signed-off-by: Benjamin Marzinski <bmarzins@redhat.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
16 years agodoc/cdrom: Trvial documentation error, file not present
Randy Dunlap [Fri, 10 Oct 2008 06:22:44 +0000 (08:22 +0200)]
doc/cdrom: Trvial documentation error, file not present

The sbpcd tester program is not included in the kernel source tree,
so remove the reference to it.

Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com>
Reported-by: Nick Warne <nick@ukfsn.org>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
16 years agoMerge branches 'cma', 'cxgb3', 'ehca', 'ipath', 'ipoib', 'mad', 'misc', 'mlx4', ...
Roland Dreier [Fri, 10 Oct 2008 00:41:15 +0000 (17:41 -0700)]
Merge branches 'cma', 'cxgb3', 'ehca', 'ipath', 'ipoib', 'mad', 'misc', 'mlx4', 'mthca' and 'nes' into for-next

16 years agoRDMA/nes: Fix slab corruption
Chien Tung [Fri, 10 Oct 2008 00:41:05 +0000 (17:41 -0700)]
RDMA/nes: Fix slab corruption

Referencing cm_node after it is freed via rem_ref_cm_node() causes a
slab corruption.  There is no need to set cm_node->cm_id to NULL in
mini_cm_close().

Signed-off-by: Chien Tung <ctung@neteffect.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
16 years agoLinux 2.6.27
Linus Torvalds [Thu, 9 Oct 2008 22:13:53 +0000 (15:13 -0700)]
Linux 2.6.27

16 years agoDon't allow splice() to files opened with O_APPEND
Linus Torvalds [Thu, 9 Oct 2008 21:04:54 +0000 (14:04 -0700)]
Don't allow splice() to files opened with O_APPEND

This is debatable, but while we're debating it, let's disallow the
combination of splice and an O_APPEND destination.

It's not entirely clear what the semantics of O_APPEND should be, and
POSIX apparently expects pwrite() to ignore O_APPEND, for example.  So
we could make up any semantics we want, including the old ones.

But Miklos convinced me that we should at least give it some thought,
and that accepting writes at arbitrary offsets is wrong at least for
IS_APPEND() files (which always have O_APPEND set, even if the reverse
isn't true: you can obviously have O_APPEND set on a regular file).

So disallow O_APPEND entirely for now.  I doubt anybody cares, and this
way we have one less gray area to worry about.

Reported-and-argued-for-by: Miklos Szeredi <miklos@szeredi.hu>
Acked-by: Jens Axboe <ens.axboe@oracle.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agoMerge branch 'hwmon-for-linus' of git://jdelvare.pck.nerim.net/jdelvare-2.6
Linus Torvalds [Thu, 9 Oct 2008 19:23:54 +0000 (12:23 -0700)]
Merge branch 'hwmon-for-linus' of git://jdelvare.pck.nerim.net/jdelvare-2.6

* 'hwmon-for-linus' of git://jdelvare.pck.nerim.net/jdelvare-2.6:
  hwmon: (abituguru3) Enable DMI probing feature on Abit AT8 32X
  hwmon: (abituguru3) Enable reading from AUX3 fan on Abit AT8 32X
  hwmon: (adt7473) Fix some bogosity in documentation file
  hwmon: Define sysfs interface for energy consumption register
  hwmon: (it87) Prevent power-off on Shuttle SN68PT
  eeepc-laptop: Fix hwmon interface

16 years agoMerge branch 'fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/davej/cpufreq
Linus Torvalds [Thu, 9 Oct 2008 19:23:35 +0000 (12:23 -0700)]
Merge branch 'fixes' of git://git./linux/kernel/git/davej/cpufreq

* 'fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/davej/cpufreq:
  [CPUFREQ] correct broken links and email addresses

16 years agoSLOB: fix bogus ksize calculation fix
Matt Mackall [Wed, 8 Oct 2008 19:51:57 +0000 (14:51 -0500)]
SLOB: fix bogus ksize calculation fix

This fixes the previous fix, which was completely wrong on closer
inspection. This version has been manually tested with a user-space
test harness and generates sane values. A nearly identical patch has
been boot-tested.

The problem arose from changing how kmalloc/kfree handled alignment
padding without updating ksize to match. This brings it in sync.

Signed-off-by: Matt Mackall <mpm@selenic.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years ago[CPUFREQ] correct broken links and email addresses
Németh Márton [Thu, 9 Oct 2008 12:59:17 +0000 (14:59 +0200)]
[CPUFREQ] correct broken links and email addresses

Replace the no longer working links and email address in the
documentation and in source code.

Signed-off-by: Márton Németh <nm127@freemail.hu>
Signed-off-by: Dave Jones <davej@redhat.com>
16 years agohwmon: (abituguru3) Enable DMI probing feature on Abit AT8 32X
Alistair John Strachan [Thu, 9 Oct 2008 13:33:59 +0000 (15:33 +0200)]
hwmon: (abituguru3) Enable DMI probing feature on Abit AT8 32X

Enable driver checking of the DMI product name (when enabled) on
an Abit AT8 32X, instead of falling back to a manual probe. This
eliminates false negatives and eventually will help avoid
unnecessary bus probes on unsupported mainboards.

Signed-off-by: Alistair John Strachan <alistair@devzero.co.uk>
Tested-by: Daniel Exner <dex@dragonslave.de>
Acked-by: Hans de Goede <hdegoede@redhat.com>
Signed-off-by: Jean Delvare <khali@linux-fr.org>
16 years agohwmon: (abituguru3) Enable reading from AUX3 fan on Abit AT8 32X
Alistair John Strachan [Thu, 9 Oct 2008 13:33:59 +0000 (15:33 +0200)]
hwmon: (abituguru3) Enable reading from AUX3 fan on Abit AT8 32X

The table for the Abit AT8 32X was incorrectly missing an entry
for the sixth ("AUX3") fan. Add this entry, exporting the fan
reading to userspace.

Closes lm-sensors.org ticket #2339.

Signed-off-by: Alistair John Strachan <alistair@devzero.co.uk>
Tested-by: Daniel Exner <dex@dragonslave.de>
Acked-by: Hans de Goede <hdegoede@redhat.com>
Signed-off-by: Jean Delvare <khali@linux-fr.org>
16 years agohwmon: (adt7473) Fix some bogosity in documentation file
Darrick J. Wong [Thu, 9 Oct 2008 13:33:58 +0000 (15:33 +0200)]
hwmon: (adt7473) Fix some bogosity in documentation file

Signed-off-by: Darrick J. Wong <djwong@us.ibm.com>
Signed-off-by: Jean Delvare <khali@linux-fr.org>
16 years agohwmon: Define sysfs interface for energy consumption register
Darrick J. Wong [Thu, 9 Oct 2008 13:33:58 +0000 (15:33 +0200)]
hwmon: Define sysfs interface for energy consumption register

Describe the sysfs files that were introduced in the ibmaem driver.

Signed-off-by: Darrick J. Wong <djwong@us.ibm.com>
Signed-off-by: Jean Delvare <khali@linux-fr.org>
16 years agohwmon: (it87) Prevent power-off on Shuttle SN68PT
Jean Delvare [Thu, 9 Oct 2008 13:33:58 +0000 (15:33 +0200)]
hwmon: (it87) Prevent power-off on Shuttle SN68PT

On the Shuttle SN68PT, FAN_CTL2 is apparently not connected to a fan,
but to something else. One user has reported instant system power-off
when changing the PWM2 duty cycle, so we disable it.

I use the board name string as the trigger in case the same board is
ever used in other systems.

This closes lm-sensors ticket #2349:
pwmconfig causes a hard poweroff
http://www.lm-sensors.org/ticket/2349

Signed-off-by: Jean Delvare <khali@linux-fr.org>
16 years agoeeepc-laptop: Fix hwmon interface
Corentin Chary [Thu, 9 Oct 2008 13:33:57 +0000 (15:33 +0200)]
eeepc-laptop: Fix hwmon interface

Creates a name file in the sysfs directory, that
is needed for the libsensors library to work.
Also rename fan1_pwm to pwm1 and scale its value as needed.

This fixes bug #11520:
http://bugzilla.kernel.org/show_bug.cgi?id=11520

Signed-off-by: Corentin Chary <corentincj@iksaif.net>
Signed-off-by: Jean Delvare <khali@linux-fr.org>
16 years agoblock_dev: fix kernel-doc in new functions
Randy Dunlap [Thu, 9 Oct 2008 08:42:38 +0000 (10:42 +0200)]
block_dev: fix kernel-doc in new functions

Fix kernel-doc in new functions:

Error(mmotm-2008-1002-1617//fs/block_dev.c:895): duplicate section name 'Description'
Error(mmotm-2008-1002-1617//fs/block_dev.c:924): duplicate section name 'Description'
Warning(mmotm-2008-1002-1617//fs/block_dev.c:1282): No description found for parameter 'pathname'

Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com>
cc: Andrew Patterson <andrew.patterson@hp.com>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
16 years agoblock: add some comments around the bio read-write flags
Jens Axboe [Thu, 9 Oct 2008 07:01:10 +0000 (09:01 +0200)]
block: add some comments around the bio read-write flags

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
16 years agoblock: mark bio_split_pool static
Denis ChengRq [Thu, 9 Oct 2008 06:57:05 +0000 (08:57 +0200)]
block: mark bio_split_pool static

Since all bio_split calls refer the same single bio_split_pool, the bio_split
function can use bio_split_pool directly instead of the mempool_t parameter;

then the mempool_t parameter can be removed from bio_split param list, and
bio_split_pool is only referred in fs/bio.c file, can be marked static.

Signed-off-by: Denis ChengRq <crquan@gmail.com>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
16 years agoblock: Find bio sector offset given idx and offset
Martin K. Petersen [Thu, 2 Oct 2008 02:42:53 +0000 (22:42 -0400)]
block: Find bio sector offset given idx and offset

Helper function to find the sector offset in a bio given bvec index
and page offset.

Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
16 years agoblock: gendisk integrity wrapper
Martin K. Petersen [Thu, 2 Oct 2008 16:47:49 +0000 (18:47 +0200)]
block: gendisk integrity wrapper

This is a wrapper for accessing a gendisk's integrity bits.  It allows
the integrity support in MD to be compiled with BLK_DEV_INTEGRITY off.

Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
16 years agoblock: Switch blk_integrity_compare from bdev to gendisk
Martin K. Petersen [Wed, 1 Oct 2008 07:38:39 +0000 (03:38 -0400)]
block: Switch blk_integrity_compare from bdev to gendisk

The DM and MD integrity support now depends on being able to use
gendisks instead of block_devices when comparing integrity profiles.
Change function parameters accordingly.

Also update comparison logic so that two NULL profiles are a valid
configuration.

Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
16 years agoblock: Fix double put in blk_integrity_unregister
Martin K. Petersen [Wed, 1 Oct 2008 07:38:38 +0000 (03:38 -0400)]
block: Fix double put in blk_integrity_unregister

- kobject_del already puts the parent.

 - Set integrity profile to NULL to prevent stale data.

Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
16 years agoblock: Introduce integrity data ownership flag
Martin K. Petersen [Wed, 1 Oct 2008 07:38:37 +0000 (03:38 -0400)]
block: Introduce integrity data ownership flag

A filesystem might supply its own integrity metadata.  Introduce a
flag that indicates whether the filesystem or the block layer owns the
integrity buffer.

Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
16 years agoblock: revert part of d7533ad0e132f92e75c1b2eb7c26387b25a583c1
Jens Axboe [Thu, 2 Oct 2008 10:53:22 +0000 (12:53 +0200)]
block: revert part of d7533ad0e132f92e75c1b2eb7c26387b25a583c1

We need bdev_get_integrity() to support the pending md/dm patches.

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
16 years agobio.h: Remove unused conditional code
Alberto Bertogli [Thu, 2 Oct 2008 10:46:53 +0000 (12:46 +0200)]
bio.h: Remove unused conditional code

The whole bio_integrity() definition is inside an #ifdef
CONFIG_BLK_DEV_INTEGRITY, there's no need for the conditional code.

Signed-off-by: Alberto Bertogli <albertito@blitiri.com.ar>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
16 years agoblock: remove end_{queued|dequeued}_request()
Kiyoshi Ueda [Wed, 1 Oct 2008 14:14:46 +0000 (10:14 -0400)]
block: remove end_{queued|dequeued}_request()

This patch removes end_queued_request() and end_dequeued_request(),
which are no longer used.

As a results, users of __end_request() became only end_request().
So the actual code in __end_request() is moved to end_request()
and __end_request() is removed.

Signed-off-by: Kiyoshi Ueda <k-ueda@ct.jp.nec.com>
Signed-off-by: Jun'ichi Nomura <j-nomura@ce.jp.nec.com>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
16 years agoblock: change elevator to use __blk_end_request()
Kiyoshi Ueda [Wed, 1 Oct 2008 14:13:44 +0000 (10:13 -0400)]
block: change elevator to use __blk_end_request()

This patch converts elevator to use __blk_end_request() directly
so that end_{queued|dequeued}_request() can be removed.
Related 'uptodate' arguments is converted to 'error'.

Signed-off-by: Kiyoshi Ueda <k-ueda@ct.jp.nec.com>
Signed-off-by: Jun'ichi Nomura <j-nomura@ce.jp.nec.com>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
16 years agogdrom: change to use __blk_end_request()
Kiyoshi Ueda [Wed, 1 Oct 2008 14:13:02 +0000 (10:13 -0400)]
gdrom: change to use __blk_end_request()

This patch converts gdrom to use __blk_end_request() directly
so that end_{queued|dequeued}_request() can be removed.

gd.transfer is '1' in error cases and '0' in non-error cases,
so gdrom hasn't been propagating any error code to the block layer.
We can just convert error cases to '-EIO'.

Signed-off-by: Kiyoshi Ueda <k-ueda@ct.jp.nec.com>
Signed-off-by: Jun'ichi Nomura <j-nomura@ce.jp.nec.com>
Cc: Adrian McMenamin <adrian@mcmen.demon.co.uk>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
16 years agomemstick: change to use __blk_end_request()
Kiyoshi Ueda [Wed, 1 Oct 2008 14:12:15 +0000 (10:12 -0400)]
memstick: change to use __blk_end_request()

This patch converts memstick to use __blk_end_request() directly
so that end_{queued|dequeued}_request() can be removed.

Signed-off-by: Kiyoshi Ueda <k-ueda@ct.jp.nec.com>
Signed-off-by: Jun'ichi Nomura <j-nomura@ce.jp.nec.com>
Cc: Alex Dubov <oakad@yahoo.com>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
16 years agovirtio_blk: change to use __blk_end_request()
Kiyoshi Ueda [Wed, 1 Oct 2008 14:11:20 +0000 (10:11 -0400)]
virtio_blk: change to use __blk_end_request()

This patch converts virtio_blk to use __blk_end_request() directly
so that end_{queued|dequeued}_request() can be removed.
Related 'uptodate' argument is converted to 'error'.

Signed-off-by: Kiyoshi Ueda <k-ueda@ct.jp.nec.com>
Signed-off-by: Jun'ichi Nomura <j-nomura@ce.jp.nec.com>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
16 years agoblktrace: use BLKTRACE_BDEV_SIZE as the name size for setup structure
Jens Axboe [Wed, 1 Oct 2008 14:16:25 +0000 (16:16 +0200)]
blktrace: use BLKTRACE_BDEV_SIZE as the name size for setup structure

Define as 32, which is is what BDEVNAME_SIZE is/was as well. This keeps
the user interface the same and gets rid of the difference between
kernel and user api here.

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
16 years agoblock: add lld busy state exporting interface
Kiyoshi Ueda [Wed, 1 Oct 2008 14:12:15 +0000 (16:12 +0200)]
block: add lld busy state exporting interface

This patch adds an new interface, blk_lld_busy(), to check lld's
busy state from the block layer.
blk_lld_busy() calls down into low-level drivers for the checking
if the drivers set q->lld_busy_fn() using blk_queue_lld_busy().

This resolves a performance problem on request stacking devices below.

Some drivers like scsi mid layer stop dispatching request when
they detect busy state on its low-level device like host/target/device.
It allows other requests to stay in the I/O scheduler's queue
for a chance of merging.

Request stacking drivers like request-based dm should follow
the same logic.
However, there is no generic interface for the stacked device
to check if the underlying device(s) are busy.
If the request stacking driver dispatches and submits requests to
the busy underlying device, the requests will stay in
the underlying device's queue without a chance of merging.
This causes performance problem on burst I/O load.

With this patch, busy state of the underlying device is exported
via q->lld_busy_fn().  So the request stacking driver can check it
and stop dispatching requests if busy.

The underlying device driver must return the busy state appropriately:
    1: when the device driver can't process requests immediately.
    0: when the device driver can process requests immediately,
       including abnormal situations where the device driver needs
       to kill all requests.

Signed-off-by: Kiyoshi Ueda <k-ueda@ct.jp.nec.com>
Signed-off-by: Jun'ichi Nomura <j-nomura@ce.jp.nec.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
16 years agoblock: Fix blk_start_queueing() to not kick a stopped queue
Elias Oltmanns [Wed, 1 Oct 2008 14:02:33 +0000 (16:02 +0200)]
block: Fix blk_start_queueing() to not kick a stopped queue

blk_start_queueing() should act like the generic queue unplugging
and kicking and ignore a stopped queue. Such a queue may not be
run until after a call to blk_start_queue().

Signed-off-by: Elias Oltmanns <eo@nebensachen.de>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
16 years agoinclude blktrace_api.h in headers_install
Sven Schuetz [Fri, 26 Sep 2008 08:58:02 +0000 (10:58 +0200)]
include blktrace_api.h in headers_install

This header file is of interest for user space programming, i.e.
for tools that process blktrace data.

We would like to use it for a tool on-top of blktrace which processes
data provided by blktrace. For this purpose, it would be helpful
if the blktrace API would make it to /usr/include/linux.

The git tree for the blktrace tools comes with its own copy of this header
file. I didn't manage to replace that copy with the file generated
by the patch below yet. A few more cleanups would be needed.
For example, the blktrace ioctl numbers, which are currently defined in
usr/include/fs.h, might need to be moved. Should be feasible, though.

Signed-off-by: Sven Schuetz <sven@linux.vnet.ibm.com>
Signed-off-by: Martin Peschke <mp3@de.ibm.com>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
16 years agoblock: reserve some tags just for sync IO
Jens Axboe [Thu, 25 Sep 2008 09:42:41 +0000 (11:42 +0200)]
block: reserve some tags just for sync IO

By only allowing async IO to consume 3/4 ths of the tag depth, we
always have slots free to serve sync IO. This is important to avoid
having writes fill the entire tag queue, thus starving reads.

Original patch and idea from Linus Torvalds <torvalds@linux-foundation.org>

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
16 years agoblock: as/cfq ssd idle check update
Jens Axboe [Thu, 25 Sep 2008 09:37:50 +0000 (11:37 +0200)]
block: as/cfq ssd idle check update

We really need to know about the hardware tagging support as well,
since if the SSD does not do tagging then we still want to idle.
Otherwise have the same dependent sync IO vs flooding async IO
problem as on rotational media.

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
16 years agolibata: set queue SSD flag for SSD devices
Jens Axboe [Wed, 24 Sep 2008 11:05:10 +0000 (13:05 +0200)]
libata: set queue SSD flag for SSD devices

SSD devices should give an RPM setting of 1 in word 217 of the ID
page. If we see such a device, tell the block layer about it.

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
16 years agoblock: add queue flag for SSD/non-rotational devices
Jens Axboe [Wed, 24 Sep 2008 11:03:33 +0000 (13:03 +0200)]
block: add queue flag for SSD/non-rotational devices

We don't want to idle in AS/CFQ if the device doesn't have a seek
penalty. So add a QUEUE_FLAG_NONROT to indicate a non-rotational
device, low level drivers should set this flag upon discovery of
an SSD or similar device type.

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
16 years agofloppy: support arbitrary first-sector numbers
Keith Wansbrough [Mon, 22 Sep 2008 21:57:17 +0000 (14:57 -0700)]
floppy: support arbitrary first-sector numbers

The current floppy_struct allows floppies to number sectors starting
from 0 or 1.  This patch allows arbitrary first-sector numbers - for
example, 0xC1 for Amstrad CPC disks.

This extends the existing 1-bit field (FD_ZEROBASED, bit 2 of stretch)
to 8 bits (FD_SECTMASK, bits 2 to 9).

Currently 0x00 denotes a first sector number of 1, and 0x01 denotes a
first sector number of 0.  We extend this by interpreting FD_SECTMASK
as the first sector number with the LSB flipped.

Signed-off-by: Keith Wansbrough <keith@lochan.org>
Cc: Alain Knaff <alain@linux.lu>
Cc: Michael Kerrisk <mtk.manpages@googlemail.com>
Cc: Karel Zak <kzak@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
16 years agodrivers/block: Use DIV_ROUND_UP
Julia Lawall [Mon, 22 Sep 2008 21:57:16 +0000 (14:57 -0700)]
drivers/block: Use DIV_ROUND_UP

The kernel.h macro DIV_ROUND_UP performs the computation (((n) + (d) - 1) /
(d)) but is perhaps more readable.

An extract of the semantic patch that makes this change is as follows:
(http://www.emn.fr/x-info/coccinelle/)

// <smpl>
@haskernel@
@@

#include <linux/kernel.h>

@depends on haskernel@
expression n,d;
@@

(
- (n + d - 1) / d
+ DIV_ROUND_UP(n,d)
|
- (n + (d - 1)) / d
+ DIV_ROUND_UP(n,d)
)

@depends on haskernel@
expression n,d;
@@

- DIV_ROUND_UP((n),d)
+ DIV_ROUND_UP(n,d)

@depends on haskernel@
expression n,d;
@@

- DIV_ROUND_UP(n,(d))
+ DIV_ROUND_UP(n,d)
// </smpl>

Signed-off-by: Julia Lawall <julia@diku.dk>
Cc: <mike.miller@hp.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
16 years agocciss: Fix cciss SCSI rescan code to better notice device changes
scameron@beardog.cca.cpqcorp.net [Sat, 20 Sep 2008 01:27:47 +0000 (18:27 -0700)]
cciss: Fix cciss SCSI rescan code to better notice device changes

Fix cciss SCSI rescan code to better notice device changes.
If you hot-unplug a tape drive, then hot-plug a different
tape drive into the same slot in a storage enclosure,
the cciss driver wouldn't notice anything had changed, as
it was only looking at the LUN address and device type.
Now it looks at the inquiry page 0x83 device identifier,
and vendor and model strings as well.

Signed-off-by: Stephen M. Cameron <scameron@beardog.cca.cpqcorp.net>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
16 years agofix an example of scatterlists handling in DMA-API.txt
FUJITA Tomonori [Thu, 18 Sep 2008 16:35:28 +0000 (09:35 -0700)]
fix an example of scatterlists handling in DMA-API.txt

This example isn't the proper way to handle scatterlists (can't handle
sg chaining).

Signed-off-by: FUJITA Tomonori <fujita.tomonori@lab.ntt.co.jp>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
16 years agoblock: add a queue flag for request stacking support
Kiyoshi Ueda [Thu, 18 Sep 2008 14:46:13 +0000 (10:46 -0400)]
block: add a queue flag for request stacking support

This patch adds a queue flag to indicate the block device can be
used for request stacking.

Request stacking drivers need to stack their devices on top of
only devices of which q->request_fn is functional.
Since bio stacking drivers (e.g. md, loop) basically initialize
their queue using blk_alloc_queue() and don't set q->request_fn,
the check of (q->request_fn == NULL) looks enough for that purpose.

However, dm will become both types of stacking driver (bio-based and
request-based).  And dm will always set q->request_fn even if the dm
device is bio-based of which q->request_fn is not functional actually.
So we need something else to distinguish the type of the device.
Adding a queue flag is a solution for that.

The reason why dm always sets q->request_fn is to keep
the compatibility of dm user-space tools.
Currently, all dm user-space tools are using bio-based dm without
specifying the type of the dm device they use.
To use request-based dm without changing such tools, the kernel
must decide the type of the dm device automatically.
The automatic type decision can't be done at the device creation time
and needs to be deferred until such tools load a mapping table,
since the actual type is decided by dm target type included in
the mapping table.

So a dm device has to be initialized using blk_init_queue()
so that we can load either type of table.
Then, all queue stuffs are set (e.g. q->request_fn) and we have
no element to distinguish that it is bio-based or request-based,
even after a table is loaded and the type of the device is decided.

By the way, some stuffs of the queue (e.g. request_list, elevator)
are needless when the dm device is used as bio-based.
But the memory size is not so large (about 20[KB] per queue on ia64),
so I hope the memory loss can be acceptable for bio-based dm users.

Signed-off-by: Kiyoshi Ueda <k-ueda@ct.jp.nec.com>
Signed-off-by: Jun'ichi Nomura <j-nomura@ce.jp.nec.com>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
16 years agoblock: add request submission interface
Kiyoshi Ueda [Thu, 18 Sep 2008 14:45:38 +0000 (10:45 -0400)]
block: add request submission interface

This patch adds blk_insert_cloned_request(), a generic request
submission interface for request stacking drivers.
Request-based dm will use it to submit their clones to underlying
devices.

blk_rq_check_limits() is also added because it is possible that
the lower queue has stronger limitations than the upper queue
if multiple drivers are stacking at request-level.
Not only for blk_insert_cloned_request()'s internal use, the function
will be used by request-based dm when the queue limitation is
modified (e.g. by replacing dm's table).

Signed-off-by: Kiyoshi Ueda <k-ueda@ct.jp.nec.com>
Signed-off-by: Jun'ichi Nomura <j-nomura@ce.jp.nec.com>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
16 years agoblock: add request update interface
Kiyoshi Ueda [Thu, 18 Sep 2008 14:45:09 +0000 (10:45 -0400)]
block: add request update interface

This patch adds blk_update_request(), which updates struct request
with completing its data part, but doesn't complete the struct
request itself.
Though it looks like end_that_request_first() of older kernels,
blk_update_request() should be used only by request stacking drivers.

Request-based dm will use it in bio->bi_end_io callback to update
the original request when a data part of a cloned request completes.
Followings are additional background information of why request-based
dm needs this interface.

  - Request stacking drivers can't use blk_end_request() directly from
    the lower driver's completion context (bio->bi_end_io or rq->end_io),
    because some device drivers (e.g. ide) may try to complete
    their request with queue lock held, and it may cause deadlock.
    See below for detailed description of possible deadlock:
    <http://marc.info/?l=linux-kernel&m=120311479108569&w=2>

  - To solve that, request-based dm offloads the completion of
    cloned struct request to softirq context (i.e. using
    blk_complete_request() from rq->end_io).

  - Though it is possible to use the same solution from bio->bi_end_io,
    it will delay the notification of bio completion to the original
    submitter.  Also, it will cause inefficient partial completion,
    because the lower driver can't perform the cloned request anymore
    and request-based dm needs to requeue and redispatch it to
    the lower driver again later.  That's not good.

  - So request-based dm needs blk_update_request() to perform the bio
    completion in the lower driver's completion context, which is more
    efficient.

Signed-off-by: Kiyoshi Ueda <k-ueda@ct.jp.nec.com>
Signed-off-by: Jun'ichi Nomura <j-nomura@ce.jp.nec.com>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
16 years agoblock: blk_cleanup_queue() should call blk_sync_queue()
Jens Axboe [Thu, 18 Sep 2008 16:22:54 +0000 (09:22 -0700)]
block: blk_cleanup_queue() should call blk_sync_queue()

When a driver calls blk_cleanup_queue(), the device should be fully idle.
However, the block layer may have pending plugging timers and the IO
schedulers may have pending work in the work queues. So quisce the device
by waiting for the timer and flushing the work queues.

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
16 years agoblock: Expand Xen blkfront for > 16 xvd
Chris Lalancette [Wed, 17 Sep 2008 21:30:32 +0000 (14:30 -0700)]
block: Expand Xen blkfront for > 16 xvd

Until recently, the maximum number of xvd block devices you could attach
to a Xen domU was 16. This limitation turned out to be problematic for
some users, so it was expanded to handle a much larger number of disks.
However, this requires a couple of changes in the way that blkfront
scans for disks. This functionality is already present in the Xen
linux-2.6.18-xen.hg tree; the attached patch adds this functionality to
the mainline xen-blkfront implementation. I successfully tested it on a
2.6.25 tree, and build tested it on 2.6.27-rc3.

Signed-off-by: Chris Lalancette <clalance@redhat.com>
Acked-by: Jeremy Fitzhardinge <jeremy@goop.org>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
16 years agoblock: cleanup some of the integrity stuff in blkdev.h
Jens Axboe [Thu, 18 Sep 2008 16:31:53 +0000 (09:31 -0700)]
block: cleanup some of the integrity stuff in blkdev.h

Don't put functions that are only used in fs/bio-integrity.c in
blkdev.h, it's much cleaner to just keep it in there. Also kill
completely unused bdev_get_tag_size()

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
16 years agoblock: use rq complete marking in blk_abort_request()
Jens Axboe [Tue, 16 Sep 2008 16:54:11 +0000 (09:54 -0700)]
block: use rq complete marking in blk_abort_request()

We cannot abort a request if we raced with the timeout handler already,
or with the IO completion. So make blk_abort_request() mark the request
as complete, and only continue if we succeeded.

Found and suggested by Mike Anderson <andmike@linux.vnet.ibm.com>

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
16 years agoblock: add fault injection mechanism for faking request timeouts
Jens Axboe [Sun, 14 Sep 2008 12:56:33 +0000 (05:56 -0700)]
block: add fault injection mechanism for faking request timeouts

Only works for the generic request timer handling. Allows one to
sporadically ignore request completions, thus exercising the timeout
handling.

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
16 years agoblock: add bio_kmalloc()
Jens Axboe [Thu, 11 Sep 2008 11:17:37 +0000 (13:17 +0200)]
block: add bio_kmalloc()

Not all callers need (or want!) the mempool backing guarentee, it
essentially means that you can only use bio_alloc() for short allocations
and not for preallocating some bio's at setup or init time.

So add bio_kmalloc() which does the same thing as bio_alloc(), except
it just uses kmalloc() as the backing instead of the bio mempools.

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
16 years agoblock: adjust blkdev_issue_discard for swap
Hugh Dickins [Thu, 11 Sep 2008 08:57:55 +0000 (10:57 +0200)]
block: adjust blkdev_issue_discard for swap

Two mods to blkdev_issue_discard(), thinking ahead to its use on swap:

1. Add gfp_mask argument, so swap allocation can use it where GFP_KERNEL
   might deadlock but GFP_NOIO is safe.

2. Enlarge nr_sects argument from unsigned to sector_t: unsigned long is
   enough to cover a whole swap area, but sector_t suits any partition.

Change sb_issue_discard()'s nr_blocks to sector_t too; but no need seen
for a gfp_mask there, just pass GFP_KERNEL down to blkdev_issue_discard().

Signed-off-by: Hugh Dickins <hugh@veritas.com>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
16 years agosg: remove unnecessary blk_rq_unmap_user
FUJITA Tomonori [Tue, 2 Sep 2008 13:50:08 +0000 (22:50 +0900)]
sg: remove unnecessary blk_rq_unmap_user

blk_rq_unmap_user in sg_finish_rem_req can take care of all the cases.

Signed-off-by: FUJITA Tomonori <fujita.tomonori@lab.ntt.co.jp>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
16 years agosg: remove sg_read_xfer
FUJITA Tomonori [Tue, 2 Sep 2008 13:50:07 +0000 (22:50 +0900)]
sg: remove sg_read_xfer

sg_read_xfer was used to copy data to user space for READ
commands. blk_rq_unmap_user does the job so sg_read_xfer does nothing
useful.

Signed-off-by: FUJITA Tomonori <fujita.tomonori@lab.ntt.co.jp>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
16 years agosg: remove sg_write_xfer
FUJITA Tomonori [Tue, 2 Sep 2008 13:50:06 +0000 (22:50 +0900)]
sg: remove sg_write_xfer

sg_write_xfer was used to copy data from user space for WRITE
commands. blk_rq_map_user_iov and blk_rq_map_user do the job so
sg_write_xfer does nothing useful.

Signed-off-by: FUJITA Tomonori <fujita.tomonori@lab.ntt.co.jp>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
16 years agosg: incorporate sg_build_direct into sg_start_req
FUJITA Tomonori [Tue, 2 Sep 2008 13:50:05 +0000 (22:50 +0900)]
sg: incorporate sg_build_direct into sg_start_req

Calling blk_rq_map_user() at a single place is better than at
different two places. It makes the code more understandable.

Signed-off-by: FUJITA Tomonori <fujita.tomonori@lab.ntt.co.jp>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
16 years agosg: remove __sg_start_req
FUJITA Tomonori [Tue, 2 Sep 2008 13:50:04 +0000 (22:50 +0900)]
sg: remove __sg_start_req

__sg_start_req() was used temporarily to call blk_get_request() during
converting sg to use the block layer.

Now sg always calls blk_get_request() so we can move blk_get_request()
to sg_start_req(). We don't need __sg_start_req anymore.

Signed-off-by: FUJITA Tomonori <fujita.tomonori@lab.ntt.co.jp>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
16 years agosg: remove b_malloc_len in sg_scatter_hold struct
FUJITA Tomonori [Tue, 2 Sep 2008 13:50:03 +0000 (22:50 +0900)]
sg: remove b_malloc_len in sg_scatter_hold struct

It's not used for anything useful after the block layer conversion.

Signed-off-by: FUJITA Tomonori <fujita.tomonori@lab.ntt.co.jp>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
16 years agosg: remove SG_ALLOW_DIO_CODE define
FUJITA Tomonori [Tue, 2 Sep 2008 13:50:02 +0000 (22:50 +0900)]
sg: remove SG_ALLOW_DIO_CODE define

sg had lots of the own functions for the direct IO but now sg uses the
block layer functions for it. There are only five lines for the direct
IO. SG_ALLOW_DIO_CODE define was used to compile out the direct IO
code but we don't need the define. If someone wants to remove the
direct IO code, he can do easily without the define.

Signed-off-by: FUJITA Tomonori <fujita.tomonori@lab.ntt.co.jp>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
16 years agosg: rename sg_cmd_done sg_rq_end_io
FUJITA Tomonori [Tue, 2 Sep 2008 13:50:01 +0000 (22:50 +0900)]
sg: rename sg_cmd_done sg_rq_end_io

old sg_rq_end_io() was used to wrap sg_cmd_done during converting sg
to use the block layer (in order to cover the difference
scsi_execute_async and blk_execute_rq_nowait). Now we don't need it so
let's remove it.

Signed-off-by: FUJITA Tomonori <fujita.tomonori@lab.ntt.co.jp>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
16 years agodm: Call blk_abort_queue on failed paths
Mike Anderson [Fri, 29 Aug 2008 07:36:09 +0000 (09:36 +0200)]
dm: Call blk_abort_queue on failed paths

Signed-off-by: Mike Anderson <andmike@linux.vnet.ibm.com>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
16 years agoblock: Add interface to abort queued requests
Mike Anderson [Sat, 13 Sep 2008 18:31:27 +0000 (20:31 +0200)]
block: Add interface to abort queued requests

Signed-off-by: Mike Anderson <andmike@linux.vnet.ibm.com>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
16 years agoblock: unify request timeout handling
Jens Axboe [Sun, 14 Sep 2008 12:55:09 +0000 (05:55 -0700)]
block: unify request timeout handling

Right now SCSI and others do their own command timeout handling.
Move those bits to the block layer.

Instead of having a timer per command, we try to be a bit more clever
and simply have one per-queue. This avoids the overhead of having to
tear down and setup a timer for each command, so it will result in a lot
less timer fiddling.

Signed-off-by: Mike Anderson <andmike@linux.vnet.ibm.com>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
16 years agoCall flush_disk() after detecting an online resize.
Andrew Patterson [Thu, 4 Sep 2008 20:27:45 +0000 (14:27 -0600)]
Call flush_disk() after detecting an online resize.

We call flush_disk() to make sure the buffer cache for the disk is
flushed after a disk resize. There are two resize cases, growing and
shrinking. Given that users can shrink/then grow a disk before
revalidate_disk() is called, we treat the grow case identically to
shrinking. We need to flush the buffer cache after an online shrink
because, as James Bottomley puts it,

     The two use cases for shrinking I can see are

     1. planned: the fs is already shrunk to within the new boundaries
        and all data is relocated, so invalidate is fine (any dirty
        buffers that might exist in the shrunk region are there only
        because they were relocated but not yet written to their
        original location).
     2. unplanned:  In this case, the fs is probably toast, so whether
        we invalidate or not isn't going to make a whole lot of
        difference; it's still going to try to read or write from
        sectors beyond the new size and get I/O errors.

Immediately invalidating shrunk disks will cause errors for outstanding
I/Os for reads/write beyond the new end of the disk to be generated
earlier then if we waited for the normal buffer cache operation. It also
removes a potential security hole where we might keep old data around
from beyond the end of the shrunk disk if the disk was not invalidated.

Signed-off-by: Andrew Patterson <andrew.patterson@hp.com>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
16 years agoAdded flush_disk to factor out common buffer cache flushing code.
Andrew Patterson [Thu, 4 Sep 2008 20:27:40 +0000 (14:27 -0600)]
Added flush_disk to factor out common buffer cache flushing code.

We need to be able to flush the buffer cache for for more than
just when a disk is changed, so we factor out common cache flush code
in check_disk_change() to an internal flush_disk() routine.  This
routine will then be used for both disk changes and disk resizes (in a
later patch).

Include the disk name in the text indicating that there are busy
inodes on the device and increase the KERN severity of the message.

Signed-off-by: Andrew Patterson <andrew.patterson@hp.com>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
16 years agoSCSI sd driver calls revalidate_disk wrapper.
Andrew Patterson [Thu, 4 Sep 2008 20:27:35 +0000 (14:27 -0600)]
SCSI sd driver calls revalidate_disk wrapper.

Modify the SCSI disk driver to call the revalidate_disk()
wrapper. This allows us to do some housekeeping such as accounting for
a disk being resized online. The wrapper will call
sd_revalidate_disk() at the appropriate time.

Signed-off-by: Andrew Patterson <andrew.patterson@hp.com>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
16 years agoCheck for device resize when rescanning partitions
Andrew Patterson [Thu, 4 Sep 2008 20:27:30 +0000 (14:27 -0600)]
Check for device resize when rescanning partitions

Check for device resize in the rescan_partitions() routine. If the device
has been resized, the bdev size is set to match. The rescan_partitions()
routine is called when opening the device and when calling the
BLKRRPART ioctl.

Signed-off-by: Andrew Patterson <andrew.patterson@hp.com>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
16 years agoAdjust block device size after an online resize of a disk.
Andrew Patterson [Thu, 4 Sep 2008 20:27:25 +0000 (14:27 -0600)]
Adjust block device size after an online resize of a disk.

The revalidate_disk routine now checks if a disk has been resized by
comparing the gendisk capacity to the bdev inode size.  If they are
different (usually because the disk has been resized underneath the kernel)
the bdev inode size is adjusted to match the capacity.

Signed-off-by: Andrew Patterson <andrew.patterson@hp.com>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
16 years agoWrapper for lower-level revalidate_disk routines.
Andrew Patterson [Thu, 4 Sep 2008 20:27:20 +0000 (14:27 -0600)]
Wrapper for lower-level revalidate_disk routines.

This is a wrapper for the lower-level revalidate_disk call-backs such
as sd_revalidate_disk(). It allows us to perform pre and post
operations when calling them.

We will use this wrapper in a later patch to adjust block device sizes
after an online resize (a _post_ operation).

Signed-off-by: Andrew Patterson <andrew.patterson@hp.com>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
16 years agoblock: fix duplicate headers for /proc/partitions
Tejun Heo [Thu, 4 Sep 2008 07:17:31 +0000 (09:17 +0200)]
block: fix duplicate headers for /proc/partitions

seqf can be started multiple times for a read and the header should be
printed only for the initial one.  Fix it.

Signed-off-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>