Avoid atomic overhead in slab_alloc and slab_free

SLUB needs to use the slab_lock for the per cpu slabs to synchronize with
potential kfree operations.  This patch avoids that need by moving all free
objects onto a lockless_freelist.  The regular freelist continues to exist
and will be used to free objects.  So while we consume the
lockless_freelist the regular freelist may build up objects.

If we are out of objects on the lockless_freelist then we may check the
regular freelist.  If it has objects then we move those over to the
lockless_freelist and do this again.  There is a significant savings in
terms of atomic operations that have to be performed.

We can even free directly to the lockless_freelist if we know that we are
running on the same processor.  So this speeds up short lived objects.
They may be allocated and freed without taking the slab_lock.  This is
particular good for netperf.

In order to maximize the effect of the new faster hotpath we extract the
hottest performance pieces into inlined functions.  These are then inlined
into kmem_cache_alloc and kmem_cache_free.  So hotpath allocation and
freeing no longer requires a subroutine call within SLUB.

[I am not sure that it is worth doing this because it changes the easy to
read structure of slub just to reduce atomic ops.  However, there is
someone out there with a benchmark on 4 way and 8 way processor systems
that seems to show a 5% regression vs.  Slab.  Seems that the regression is
due to increased atomic operations use vs.  SLAB in SLUB).  I wonder if
this is applicable or discernable at all in a real workload?]
Signed-off-by: NChristoph Lameter <clameter@sgi.com>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Signed-off-by: NMathieu Desnoyers <mathieu.desnoyers@polymtl.ca>
Cc: Andi Kleen <ak@suse.de>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

* 'for-linus' of master.kernel.org:/pub/scm/linux/kernel/git/roland/infiniband:
  IB/mlx4: Add a driver Mellanox ConnectX InfiniBand adapters
  IB: Put rlimit accounting struct in struct ib_umem
  IB/uverbs: Export ib_umem_get()/ib_umem_release() to modules

* 'usb-move' of master.kernel.org:/pub/scm/linux/kernel/git/jgarzik/netdev-2.6:
  Move USB network drivers to drivers/net/usb.

* 'upstream-linus' of master.kernel.org:/pub/scm/linux/kernel/git/jgarzik/libata-dev:
  Doc Fix: remove mention of combined mode-related kernel parameters
  libata: fix kernel-doc parameters
  Fix pata_qdi.c probe code
  pata_scc: fix compilation
  sata_via: add missing PM hooks
  sata_nv: fix ADMA freeze/thaw/irq_clear issues
  pata_pcmcia.c: add card ident for jvc cdrom
  sata_promise: SATAII-150/300 TX4 port numbering fix
  sata_promise: fix another error decode regression
  libata-acpi: fix _GTF command protocol for ATAPI devices

This reverts commit 5b479c91.

Quoth Neil Brown:

  "It causes an oops when auto-detecting raid arrays, and it doesn't
   seem easy to fix.

   The array may not be 'open' when do_md_run is called, so
   bdev->bd_disk might be NULL, so bd_set_size can oops.

   This whole approach of opening an md device before it has been
   assembled just seems to get more and more painful.  I think I'm going
   to have to come up with something clever to provide both backward
   comparability with usage expectation, and sane integration into the
   rest of the kernel."
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

It is preferable to group drivers by usage (net, scsi, ATA, ...) than
by bus.  When reviewing drivers, the [PCI|USB|PCMCIA|...] maintainer
is probably less qualified on networking issues than a networking
maintainer.  Also, from a practical standpoint, chips often
appear on multiple buses, which is why we do not put drivers into
drivers/pci/net.
Signed-off-by: NGreg Kroah-Hartman <gregkh@suse.de>
Signed-off-by: NJeff Garzik <jeff@garzik.org>

Looks like you removed the combined_mode quirk (yay!) but didn't update
kernel-parameters.txt...  might confuse people.  Here's a patch to remove
mention of it from the documentation.
Signed-off-by: NJesse Barnes <jesse.barnes@intel.com>
Signed-off-by: NJeff Garzik <jeff@garzik.org>

Warning(linux-2.6.21-git4//drivers/ata/libata-core.c:904): No description found for parameter 'new_sectors'
Warning(linux-2.6.21-git4//drivers/ata/libata-core.c:941): No description found for parameter 'new_sectors'
Signed-off-by: NRandy Dunlap <randy.dunlap@oracle.com>
Signed-off-by: NJeff Garzik <jeff@garzik.org>

There is a small typo in the probe code of pata_qdi.c, here is a patch.
Signed-off-by: NJeff Garzik <jeff@garzik.org>

Signed-off-by: NAlexey Dobriyan <adobriyan@gmail.com>
Signed-off-by: NJeff Garzik <jeff@garzik.org>

For some reason, sata_via is missing PM hooks.  Add them.  Spotted by
Jeroen Janssen <jeroen.janssen@gmail.com>.
Signed-off-by: NTejun Heo <htejun@gmail.com>
Cc: Jeroen Janssen <jeroen.janssen@gmail.com>
Signed-off-by: NJeff Garzik <jeff@garzik.org>

This patch fixes some problems with ADMA-capable controllers with
regard to freeze, thaw and irq_clear libata callbacks. Freeze and
thaw didn't switch the ADMA-specific interrupts on or off, and more
critically the irq_clear function didn't respect the restriction that
the notifier clear registers for both ports have to be written at
the same time even when only one port is being cleared. This could
result in timeouts on one port when error handling (i.e. as a result
of hotplug) occurred on the other port.

As well, this fixes some issues in the interrupt handler: we shouldn't
check any ADMA status if the port has ADMA switched off because of
an ATAPI device, and it also checks to see if any ADMA interrupt has
been raised even when we are in port-register mode.
Signed-off-by: NRobert Hancock <hancockr@shaw.ca>
Signed-off-by: NJeff Garzik <jeff@garzik.org>

update pata_pcmcia to add card ident for JVC MP-CDX1 cdrom drive
card info:
PRODID_1="KME"
PRODID_2="KXLC005"
PRODID_3="00"
MANFID=0032,2904
Signed-off-by: NRichard Kennedy <richard@rsk.demon.co.uk>
Signed-off-by: NJeff Garzik <jeff@garzik.org>

There is a known problem with sata_promise on SATAII-150/300 TX4
controller cards: it enumerates drives in an order that differs
from the port numbers printed on the controller cards. However,
Promise's BIOS and Linux driver both get the order right.

I investigated Promise's Linux driver (v1.01.0.23), and found
that it explicitly changes the mapping from logical port number
to ATA engine MMIO address on the SATAII TX4 cards. It does this
on all SATAII TX4 cards, without inspecting revision etc. The
SATAII TX2plus cards continue to use the same mapping that was
used for the first-generation chips.

This patch updates sata_promise to use the new port number to
ATA engine mapping on SATAII TX4 cards, which fixes the drive
enumeration order problem on those cards. Tested on several
1st and 2nd generation TX2plus and TX4 chips.
Signed-off-by: NMikael Pettersson <mikpe@it.uu.se>
Signed-off-by: NJeff Garzik <jeff@garzik.org>

The sata_promise error decode update changed pdc_host_intr()
to return and not complete the qc after detecting an error.
Unfortunately not completing the qc:s causes them to always
time out on error, which is wrong and has nasty side-effects.

This patch updates pdc_error_intr() to call ata_port_abort(),
similar to ahci and sata_sil24. Doing this is important as it
makes EH see the original error and not a bogus timeout.
Signed-off-by: NMikael Pettersson <mikpe@it.uu.se>
Signed-off-by: NJeff Garzik <jeff@garzik.org>

_GTF command is never ATA_PROT_ATAPI_NODATA whether the device is
ATAPI or not.  It's always ATA_PROT_NODATA.
Signed-off-by: NTejun Heo <htejun@gmail.com>
Signed-off-by: NJeff Garzik <jeff@garzik.org>

Copied from b44 driver, but it works:

netconsole: device eth0 not up yet, forcing it
atl1: eth0 link is up 100 Mbps full duplex
netconsole: network logging started
Signed-off-by: NAlexey Dobriyan <adobriyan@gmail.com>
Signed-off-by: NJeff Garzik <jeff@garzik.org>

Ejecting a PCMCIA IBM Token Ring card that has not had its dev->open()
called will reliably trigger an uninitialized spinlock oops when
spinlock debugging is enabled. The system then hangs, occasionally
softlockup oopsing.  Apparently ibmtr.c:tok_interrupt() doesn't expect
to be called before tok_open(), but tok_interrupt() gets called anyway
when the card is ejected.  So, set an already-existing flag which
causes tok_interrupt() to bail out early upon card ejection. Tested by
inserting and removing the PCMCIA card several times.
Signed-off-by: NPaul Walmsley <paul@booyaka.com>
Signed-off-by: NJeff Garzik <jeff@garzik.org>

By default, the skge driver now enables wake on magic and wake on PHY.
This is a bad default (bug), wake on PHY means machine will never shutdown
if connected to a switch.

Signed-off-by: Stephen Hemminger <shemminger@linux-foundation.org>a
Signed-off-by: NJeff Garzik <jeff@garzik.org>

Merge branch 'upstream' of git://git.kernel.org/pub/scm/linux/kernel/git/linville/wireless-2.6 into upstream

* master.kernel.org:/pub/scm/linux/kernel/git/bart/ide-2.6:
  ide: fix PIO setup on resume for ATAPI devices
  ide: legacy PCI bus order probing fixes
  ide: add ide_proc_register_port()
  ide: add "initializing" argument to ide_register_hw()
  ide: cable detection fixes (take 2)
  ide: move IDE settings handling to ide-proc.c
  ide: split off ioctl handling from IDE settings (v2)
  ide: make /proc/ide/ optional
  ide: add ide_tune_dma() helper
  ide: rework the code for selecting the best DMA transfer mode (v3)
  ide: fix UDMA/MWDMA/SWDMA masks (v3)

* git://git.linux-nfs.org/pub/linux/nfs-2.6:
  NFS: Kill the obsolete NFS_PARANOIA
  NFS: use __set_current_state()
  sunrpc: fix crash in rpc_malloc()
  NFS: Clean up NFSv4 XDR error message
  NFS: NFS client underestimates how large an NFSv4 SETATTR reply can be
  SUNRPC: Fix pointer arithmetic bug recently introduced in rpc_malloc/free
  NFS: Remove redundant check in nfs_check_verifier()
  NFS: Fix a jiffie wraparound issue

PIO should be restored also for ATAPI devices during resume, fix it.
Signed-off-by: NBartlomiej Zolnierkiewicz <bzolnier@gmail.com>

IDE PCI host drivers should register themselves with IDE core only when
IDE driver is built-in, otherwise (IDE driver is modular and thus IDE PCI
host drivers are also modular) the code has no effect and just complicates
the probing.

Fix it by adding new config option CONFIG_IDEPCI_PCIBUS (defined only when
needed and invisible to the user) and covering by #ifdef/#endif the code
in question.  It turned out that "ide=reverse" was silently accepted but did
nothing in case when IDE driver was modular, this is fixed now.
Signed-off-by: NBartlomiej Zolnierkiewicz <bzolnier@gmail.com>

* create_proc_ide_interfaces() tries to add /proc entries for every probed
  and initialized IDE port, replace it by ide_proc_register_port() which does
  it only for the given port (also rename destroy_proc_ide_interface() to
  ide_proc_unregister_port() for consistency)
  
* convert {create,destroy}_proc_ide_interface[s]() users to use new functions

* pmac driver depended on proc_ide_create() to add /proc port entries, fix it
  
* au1xxx-ide, swarm and cs5520 drivers depended indirectly on ide-generic
  driver (CONFIG_IDE_GENERIC=y) to add port /proc entries, fix them

* there is now no need to add /proc entries for IDE ports in proc_ide_create()
  so don't do it

* proc_ide_create() needs now to be called before drivers are probed - fix it,
  while at it make proc_ide_create() create /proc "ide" directory
Signed-off-by: NBartlomiej Zolnierkiewicz <bzolnier@gmail.com>

Add "initializing" argument to ide_register_hw() and use it instead of ide.c
wide variable of the same name.  Update all users of ide_register_hw()
accordingly.
Signed-off-by: NBartlomiej Zolnierkiewicz <bzolnier@gmail.com>

Tejun's recent eighty_ninty_three() fix has inspired me to do more thorough
review of the cable detection code...

* print user-friendly warning about limiting the maximum transfer speed
  to UDMA33 (and the reason behind it) when 80-wire cable is not detected,
  also while at it cleanup eighty_ninty_three() a bit

* use eighty_ninty_three() in ide_ata66_check(), this actually fixes 3 bugs:
  - bit 14 (word 93 validity check) == 1 && bit 13 (80-wire cable test) == 1
    were used as 80-wire cable present test for CONFIG_IDEDMA_IVB=n case
    (please see FIXME comment in eighty_ninty_three() for more details)
  - CONFIG_IDEDMA_IVB=y/n cases were interchanged
  - check for SATA devices was missing

* remove private cable warnings from pdc_202xx{old,new} drivers now that core
  code provides this functionality (plus, in pdc202xx_new case the test could
  give false warnings for ATAPI devices because pdc202xx_new driver doesn't
  even support ATAPI DMA)

Cc: Tejun Heo <htejun@gmail.com>
Signed-off-by: NBartlomiej Zolnierkiewicz <bzolnier@gmail.com>

* move
	__ide_add_setting()
	ide_add_setting()
	__ide_remove_setting()
	auto_remove_settings()
	ide_find_setting_by_name()
	ide_read_setting()
	ide_write_setting()
	set_xfer_rate()
	ide_add_generic_settings()
	ide_register_subdriver()
	ide_unregister_subdriver()

  from ide.c to ide-proc.c

* set_{io_32bit,pio_mode,using_dma}() cannot be marked static now, fix it

* rename ide_[un]register_subdriver() to ide_proc_[un]register_driver(),
  update device drivers to use new names

* add CONFIG_IDE_PROC_FS=n versions of ide_proc_[un]register_driver()
  and ide_add_generic_settings()

* make ide_find_setting_by_name(), ide_{read,write}_setting()
  and ide_{add,remove}_proc_entries() static

* cover IDE settings code in device drivers with CONFIG_IDE_PROC_FS #ifdef,
  also while at it cover with CONFIG_IDE_PROC_FS #ifdef ide_driver_t.proc

* remove bogus comment from ide.h

* cover with CONFIG_IDE_PROC_FS #ifdef .proc and .settings in ide_drive_t

Besides saner code this patch results in the IDE core smaller by ~2 kB
(on x86-32) and IDE disk driver by ~1 kB (ditto) when CONFIG_IDE_PROC_FS=n.
Signed-off-by: NBartlomiej Zolnierkiewicz <bzolnier@gmail.com>

* do write permission and min/max checks in ide_procset_t functions

* ide-disk.c: drive->id is always available so cleanup "multcount" setting
  accordingly

* ide-disk.c: "address" setting was incorrectly defined as type TYPE_INTA,
  fix it by using type TYPE_BYTE and updating ide_drive_t->adressing field,
  the bug didn't trigger because this IDE setting uses custom ->set function

* ide.c: add set_ksettings() for handling HDIO_SET_KEEPSETTINGS ioctl

* ide.c: add set_unmaskirq() for handling HDIO_SET_UNMASKINTR ioctl

* handle ioctls directly in generic_ide_ioclt() and idedisk_ioctl()
  instead of using IDE settings to deal with them

* remove no longer needed ide_find_setting_by_ioctl() and {read,write}_ioctl
  fields from ide_settings_t, also remove now unused TYPE_INTA handling

v2:
* add missing EXPORT_SYMBOL_GPL(ide_setting_sem) needed now for ide-disk
Signed-off-by: NBartlomiej Zolnierkiewicz <bzolnier@gmail.com>

All important information/features should be already available through
sysfs and ioctl interfaces.

Add CONFIG_IDE_PROC_FS (CONFIG_SCSI_PROC_FS rip-off) config option,
disabling it makes IDE driver ~5 kB smaller (on x86-32).

While at it add CONFIG_PROC_FS=n versions of proc_ide_{create,destroy}()
and remove no longer needed #ifdefs.
Signed-off-by: NBartlomiej Zolnierkiewicz <bzolnier@gmail.com>

After reworking the code responsible for selecting the best DMA
transfer mode it is now possible to add generic ide_tune_dma() helper.

Convert some IDE PCI host drivers to use it (the ones left need more work).
Signed-off-by: NBartlomiej Zolnierkiewicz <bzolnier@gmail.com>

Depends on the "ide: fix UDMA/MWDMA/SWDMA masks" patch.

* add ide_hwif_t.udma_filter hook for filtering UDMA mask
  (use it in alim15x3, hpt366, siimage and serverworks drivers)
* add ide_max_dma_mode() for finding best DMA mode for the device
  (loosely based on some older libata-core.c code)
* convert ide_dma_speed() users to use ide_max_dma_mode()
* make ide_rate_filter() take "ide_drive_t *drive" as an argument instead
  of "u8 mode" and teach it to how to use UDMA mask to do filtering
* use ide_rate_filter() in hpt366 driver
* remove no longer needed ide_dma_speed() and *_ratemask()
* unexport eighty_ninty_three()

v2:
* rename ->filter_udma_mask to ->udma_filter
  [ Suggested by Sergei Shtylyov <sshtylyov@ru.mvista.com>. ]

v3:
* updated for scc_pata driver (fixes XFER_UDMA_6 filtering for user-space
  originated transfer mode change requests when 100MHz clock is used)
Signed-off-by: NBartlomiej Zolnierkiewicz <bzolnier@gmail.com>

* use 0x00 instead of 0x80 to disable ->{ultra,mwdma,swdma}_mask
* add udma_mask field to ide_pci_device_t and use it to initialize
  ->ultra_mask in aec62xx, cmd64x, pdc202xx_{new,old} and piix drivers
* fix UDMA masks to match with chipset specific *_ratemask()
  (alim15x3, hpt366, serverworks and siimage drivers need UDMA mask
   filtering method - done in the next patch)

v2:
* piix: fix cable detection for 82801AA_1 and 82372FB_1
  [ Noticed by Sergei Shtylyov <sshtylyov@ru.mvista.com>. ]
* cmd64x: use hwif->cds->udma_mask
  [ Suggested by Sergei Shtylyov <sshtylyov@ru.mvista.com>. ]
* aec62xx: fix newly introduced bug - check DMA status not command register
  [ Noticed by Sergei Shtylyov <sshtylyov@ru.mvista.com>. ]

v3:
* piix: use hwif->cds->udma_mask
  [ Suggested by Sergei Shtylyov <sshtylyov@ru.mvista.com>. ]
Signed-off-by: NBartlomiej Zolnierkiewicz <bzolnier@gmail.com>

Signed-off-by: NJesper Juhl <jesper.juhl@gmail.com>
Acked-by: NTrond Myklebust <Trond.Myklebust@netapp.com>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NTrond Myklebust <Trond.Myklebust@netapp.com>

use __set_current_state(TASK_*) instead of current->state = TASK_*, in fs/nfs
Signed-off-by: NMilind Arun Choudhary <milindchoudhary@gmail.com>
Cc: Trond Myklebust <trond.myklebust@fys.uio.no>
Cc: "J. Bruce Fields" <bfields@fieldses.org>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NTrond Myklebust <Trond.Myklebust@netapp.com>

While the comment says:
 * To prevent rpciod from hanging, this allocator never sleeps,
 * returning NULL if the request cannot be serviced immediately.

The function does not actually check for NULL pointers being returned.
Signed-off-by: NPeter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: NTrond Myklebust <Trond.Myklebust@netapp.com>

Make it more useful for debugging purposes.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NTrond Myklebust <Trond.Myklebust@netapp.com>

The maximum size of an NFSv4 SETATTR compound reply should include the
GETATTR operation that we send.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NTrond Myklebust <Trond.Myklebust@netapp.com>

Use a cleaner method to find the size of an rpc_buffer.  This actually
works on x86-64!
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NTrond Myklebust <Trond.Myklebust@netapp.com>