pandora-kernel.git
17 years agoMerge branch 'for-linus' of git://www.atmel.no/~hskinnemoen/linux/kernel/avr32
Linus Torvalds [Tue, 12 Dec 2006 02:28:59 +0000 (18:28 -0800)]
Merge branch 'for-linus' of git://atmel.no/~hskinnemoen/linux/kernel/avr32

* 'for-linus' of git://www.atmel.no/~hskinnemoen/linux/kernel/avr32:
  [AVR32] Add missing #include <linux/param.h> to delay.c
  [AVR32] Pass dev parameter to dma_cache_sync()
  [AVR32] Implement intc_get_pending()
  [AVR32] Don't include <asm/delay.h>
  [AVR32] Put the chip in "stop" mode when halting the system
  [AVR32] Set flow handler for external interrupts
  [AVR32] Remove unused file
  [AVR32] Remove mii_phy_addr and eth_addr from eth_platform_data
  [AVR32] Move ethernet tag parsing to board-specific code
  [AVR32] Add macb1 platform_device
  [AVR32] Portmux API update

17 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/paulus/powerpc
Linus Torvalds [Tue, 12 Dec 2006 02:24:58 +0000 (18:24 -0800)]
Merge git://git./linux/kernel/git/paulus/powerpc

* git://git.kernel.org/pub/scm/linux/kernel/git/paulus/powerpc: (36 commits)
  [POWERPC] Generic BUG for powerpc
  [PPC] Fix compile failure do to introduction of PHY_POLL
  [POWERPC] Only export __mtdcr/__mfdcr if CONFIG_PPC_DCR is set
  [POWERPC] Remove old dcr.S
  [POWERPC] Fix SPU coredump code for max_fdset removal
  [POWERPC] Fix irq routing on some 32-bit PowerMacs
  [POWERPC] ps3: Add vuart support
  [POWERPC] Support ibm,dynamic-reconfiguration-memory nodes
  [POWERPC] dont allow pSeries_probe to succeed without initialising MMU
  [POWERPC] micro optimise pSeries_probe
  [POWERPC] Add SPURR SPR to sysfs
  [POWERPC] Add DSCR SPR to sysfs
  [POWERPC] Fix 440SPe CPU table entry
  [POWERPC] Add support for FP emulation for the e300c2 core
  [POWERPC] of_device_register: propagate device_create_file return code
  [POWERPC] Fix mmap of PCI resource with hack for X
  [POWERPC] iSeries: head_64.o needs to depend on lparmap.s
  [POWERPC] cbe_thermal: Fix initialization of sysfs attribute_group
  [POWERPC] Remove QE header files from lite5200.c
  [POWERPC] of_platform_make_bus_id(): make `magic' int
  ...

17 years ago[MIPS] paging_init(): use highend_pfn/highstart_pfn
Franck Bui-Huu [Wed, 6 Dec 2006 15:48:28 +0000 (16:48 +0100)]
[MIPS] paging_init(): use highend_pfn/highstart_pfn

This patch makes paging_init() use highend_pfn/highstart_pfn globals.

It removes the need of 'high' local which was needed only by HIGHMEM config.

More important perhaps, it fixes a bug when HIGHMEM is set but there's
actually no physical highmem (highend_pfn = 0)

Signed-off-by: Franck Bui-Huu <fbuihuu@gmail.com>
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
17 years ago[MIPS] Malta: Resurrect MTD support for onboard flash.
Ralf Baechle [Mon, 11 Dec 2006 15:33:30 +0000 (15:33 +0000)]
[MIPS] Malta: Resurrect MTD support for onboard flash.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
17 years ago[MIPS] Discard .exit.text and .exit.data at runtime.
Ralf Baechle [Mon, 11 Dec 2006 11:54:52 +0000 (11:54 +0000)]
[MIPS] Discard .exit.text and .exit.data at runtime.

While the recent cset 86384d544157db23879064cde36061cdcafc6794 did improve
things it didn't resolve all the problems.  So bite the bullet and discard
.exit.text and .exit.data at runtime.  Which of course sucks because it
bloats binaries with code that will never ever be used but it's the only
thing that will work reliable as demonstrated by the function sd_major() in
drivers/scsi/sd.c.

Gcc may compile sd_major() using a jump table which it will put into
.rodata.  If it also inlines sd_major's function body into exit_sd() which
gcc > 3.4.x does.  If CONFIG_BLK_DEV_SD has been set to y we would like ld
to discard exit_sd's code at link time.  However sd_major happens to
contain a switch statement which gcc will compile using a jump table in
.rodata on the architectures I checked.  So, when ld later discards
.exit.text only the jump table in .rodata with its stale references to
the discard .exit.text will be left which any no antique ld will honor
with a link error.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
17 years ago[MIPS] IP27: Don't drag <asm/sn/arch.h> into topology.h.
Ralf Baechle [Mon, 11 Dec 2006 04:17:30 +0000 (04:17 +0000)]
[MIPS] IP27: Don't drag <asm/sn/arch.h> into topology.h.

Another way that old SGI types were getting dragged into generic code.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
17 years ago[MIPS] IP27: Move definition of nic_t to its sole user.
Ralf Baechle [Mon, 11 Dec 2006 04:07:40 +0000 (04:07 +0000)]
[MIPS] IP27: Move definition of nic_t to its sole user.

This also fixes the duplicate definition of nic_t in the s2io driver.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
17 years ago[MIPS] IP27: Don't include <asm/sn/arch.h>.
Ralf Baechle [Mon, 11 Dec 2006 04:00:37 +0000 (04:00 +0000)]
[MIPS] IP27: Don't include <asm/sn/arch.h>.

Nothing <asm/sn/arch.h> defines is used.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
17 years ago[MIPS] compat.h uses struct pt_regs so needs to include ptrace.h.
Ralf Baechle [Mon, 11 Dec 2006 00:43:58 +0000 (00:43 +0000)]
[MIPS] compat.h uses struct pt_regs so needs to include ptrace.h.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
17 years ago[PATCH] Fix typo in 'EXPERIMENTAL' in CC_STACKPROTECTOR on x86_64
Brice Goglin [Mon, 11 Dec 2006 19:14:15 +0000 (20:14 +0100)]
[PATCH] Fix typo in 'EXPERIMENTAL' in CC_STACKPROTECTOR on x86_64

Fix typo in 'EXPERIMENTAL' in config CC_STACKPROTECTOR in arch/x86_64/Kconfig.

Signed-off-by: Brice Goglin <brice@myri.com>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] LOG2: Make powerpc's __ilog2_u64() take a 64-bit argument
David Howells [Mon, 11 Dec 2006 13:16:05 +0000 (13:16 +0000)]
[PATCH] LOG2: Make powerpc's __ilog2_u64() take a 64-bit argument

Make powerpc's __ilog2_u64() take a 64-bit argument.

Signed-off-by: David Howells <dhowells@redhat.com>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years agoMerge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/drzeus/mmc
Linus Torvalds [Mon, 11 Dec 2006 20:26:03 +0000 (12:26 -0800)]
Merge branch 'for-linus' of git://git./linux/kernel/git/drzeus/mmc

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/drzeus/mmc:
  AT91 MMC update for 2.6.19
  mmc: Change SDHCI iomem error to a warning
  mmc: fix "prev->state: 2 != TASK_RUNNING??" problem on SD/MMC  card removal
  AT91 MMC 5 : Minor cleanups
  AT91 MMC 4 : Interrupt handler cleanup
  AT91 MMC 3 : Move global mci_clk variable
  AT91 MMC 2 : Use platform resources
  AT91 MMC 1: Pass host structure.

17 years agoMerge branch 'upstream-linus' of master.kernel.org:/pub/scm/linux/kernel/git/jgarzik...
Linus Torvalds [Mon, 11 Dec 2006 20:22:58 +0000 (12:22 -0800)]
Merge branch 'upstream-linus' of /linux/kernel/git/jgarzik/netdev-2.6

* 'upstream-linus' of master.kernel.org:/pub/scm/linux/kernel/git/jgarzik/netdev-2.6:
  [PATCH] smc91x: Kill off excessive versatile hooks.
  [PATCH] myri10ge: update driver version to 1.1.0
  [PATCH] myri10ge: fix big_bytes in case of vlan frames
  [PATCH] myri10ge: Full vlan frame in small_bytes
  [PATCH] myri10ge: drop contiguous skb routines
  [PATCH] myri10ge: switch to page-based skb
  [PATCH] myri10ge: add page-based skb routines
  [PATCH] myri10ge: indentation cleanups
  [PATCH] chelsio: working NAPI
  [PATCH] MACB: Use __raw register access
  [PATCH] MACB: Use struct delayed_work instead of struct work_struct
  [PATCH] ucc_geth: Initialize mdio_lock.
  [PATCH] ucc_geth: compilation error fixes

17 years agoMake sure we populate the initroot filesystem late enough
Linus Torvalds [Mon, 11 Dec 2006 20:12:04 +0000 (12:12 -0800)]
Make sure we populate the initroot filesystem late enough

We should not initialize rootfs before all the core initializers have
run.  So do it as a separate stage just before starting the regular
driver initializers.

Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years agoMake SLES9 "get_kernel_version" work on the kernel binary again
Linus Torvalds [Mon, 11 Dec 2006 17:28:46 +0000 (09:28 -0800)]
Make SLES9 "get_kernel_version" work on the kernel binary again

As reported by Andy Whitcroft, at least the SLES9 initrd build process
depends on getting the kernel version from the kernel binary.  It does
that by simply trawling the binary and looking for the signature of the
"linux_banner" string (the string "Linux version " to be exact. Which
is really broken in itself, but whatever..)

That got broken when the string was changed to allow /proc/version to
change the UTS release information dynamically, and "get_kernel_version"
thus returned "%s" (see commit a2ee8649ba6d71416712e798276bf7c40b64e6e5:
"[PATCH] Fix linux banner utsname information").

This just restores "linux_banner" as a static string, which should fix
the version finding.  And /proc/version simply uses a different string.

To avoid wasting even that miniscule amount of memory, the early boot
string should really be marked __initdata, but that just causes the same
bug in SLES9 to re-appear, since it will then find other occurrences of
"Linux version " first.

Cc: Andy Whitcroft <apw@shadowen.org>
Acked-by: Herbert Poetzl <herbert@13thfloor.at>
Cc: Andi Kleen <ak@suse.de>
Cc: Andrew Morton <akpm@osdl.org>
Cc: Steve Fox <drfickle@us.ibm.com>
Acked-by: Olaf Hering <olaf@aepfle.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] smc91x: Kill off excessive versatile hooks.
Paul Mundt [Mon, 11 Dec 2006 10:30:06 +0000 (19:30 +0900)]
[PATCH] smc91x: Kill off excessive versatile hooks.

This looks like a result of too many auto-merges. The
CONFIG_ARCH_VERSATILE case was handled a total of 6 times.
This kills 5 of them.

Signed-off-by: Paul Mundt <lethal@linux-sh.org>
--

 drivers/net/smc91x.h |   90 ---------------------------------------------------
 1 file changed, 90 deletions(-)
Signed-off-by: Jeff Garzik <jeff@garzik.org>
17 years ago[PATCH] myri10ge: update driver version to 1.1.0
Brice Goglin [Mon, 11 Dec 2006 10:27:55 +0000 (11:27 +0100)]
[PATCH] myri10ge: update driver version to 1.1.0

Update driver version to 1.1.0.

Signed-off-by: Brice Goglin <brice@myri.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
17 years ago[PATCH] myri10ge: fix big_bytes in case of vlan frames
Brice Goglin [Mon, 11 Dec 2006 10:27:19 +0000 (11:27 +0100)]
[PATCH] myri10ge: fix big_bytes in case of vlan frames

Fix sizing of big_bytes in the case of vlan frames. The 4
VLAN_HLEN bytes were omitted, leading to sizing the big buffer
4 bytes smaller than it should be.  Due to how rx buffers are
carved from pages, this was harmless for the common (9000, 1500)
byte MTUs, but could lead to data corruption for some MTUs.

Signed-off-by: Brice Goglin <brice@myri.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
17 years ago[PATCH] myri10ge: Full vlan frame in small_bytes
Brice Goglin [Mon, 11 Dec 2006 10:26:38 +0000 (11:26 +0100)]
[PATCH] myri10ge: Full vlan frame in small_bytes

Receive full vlan frames into smalls when running with a jumbo MTU.

Signed-off-by: Brice Goglin <brice@myri.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
17 years ago[PATCH] myri10ge: drop contiguous skb routines
Brice Goglin [Mon, 11 Dec 2006 10:26:12 +0000 (11:26 +0100)]
[PATCH] myri10ge: drop contiguous skb routines

Drop the old routines that used the physically contigous skb now
that we use the physical pages. And rename myri10ge_page_rx_done()
to myri10ge_rx_done() as it was previously.

Signed-off-by: Brice Goglin <brice@myri.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
17 years ago[PATCH] myri10ge: switch to page-based skb
Brice Goglin [Mon, 11 Dec 2006 10:25:42 +0000 (11:25 +0100)]
[PATCH] myri10ge: switch to page-based skb

Switch to physical page skb, by calling the new page-based
allocation routines and using myri10ge_page_rx_done().

Signed-off-by: Brice Goglin <brice@myri.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
17 years ago[PATCH] myri10ge: add page-based skb routines
Brice Goglin [Mon, 11 Dec 2006 10:25:09 +0000 (11:25 +0100)]
[PATCH] myri10ge: add page-based skb routines

Add physical page skb allocation routines and page based rx_done,
to be used by upcoming patches.

Signed-off-by: Brice Goglin <brice@myri.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
17 years ago[PATCH] myri10ge: indentation cleanups
Brice Goglin [Mon, 11 Dec 2006 10:24:37 +0000 (11:24 +0100)]
[PATCH] myri10ge: indentation cleanups

Indentation cleanups to synchronize to our tree which is automatically
indent'ed.

Signed-off-by: Brice Goglin <brice@myri.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
17 years ago[PATCH] chelsio: working NAPI
Stephen Hemminger [Fri, 8 Dec 2006 19:08:33 +0000 (11:08 -0800)]
[PATCH] chelsio: working NAPI

This driver tries to enable/disable NAPI at runtime, but
does so in an unsafe manner, and the NAPI interrupt handling is
a mess. Replace it with a compile time selected NAPI implementation.

Signed-off-by: Stephen Hemminger <shemminger@osdl.org>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
17 years ago[PATCH] MACB: Use __raw register access
Haavard Skinnemoen [Fri, 8 Dec 2006 13:38:30 +0000 (14:38 +0100)]
[PATCH] MACB: Use __raw register access

Since macb is a chip-internal device, use __raw_readl and
__raw_writel instead of readl/writel. This will perform native-endian
accesses, which is the right thing to do on both AVR32 and ARM devices.

Signed-off-by: Haavard Skinnemoen <hskinnemoen@atmel.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
17 years ago[PATCH] MACB: Use struct delayed_work instead of struct work_struct
Haavard Skinnemoen [Fri, 8 Dec 2006 13:37:35 +0000 (14:37 +0100)]
[PATCH] MACB: Use struct delayed_work instead of struct work_struct

The macb driver calls schedule_delayed_work() and friends, so we need
to use a struct delayed_work along with it. The conversion was
explained by David Howells on lkml Dec 5 2006:

http://lkml.org/lkml/2006/12/5/269

Signed-off-by: Haavard Skinnemoen <hskinnemoen@atmel.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
17 years ago[PATCH] ucc_geth: Initialize mdio_lock.
Scott Wood [Thu, 7 Dec 2006 19:31:26 +0000 (13:31 -0600)]
[PATCH] ucc_geth: Initialize mdio_lock.

Signed-off-by: Scott Wood <scottwood@freescale.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
17 years ago[PATCH] ucc_geth: compilation error fixes
Scott Wood [Thu, 7 Dec 2006 19:31:07 +0000 (13:31 -0600)]
[PATCH] ucc_geth: compilation error fixes

Fix compilation failures when building the ucc_geth driver with spinlock
debugging.

Signed-off-by: Scott Wood <scottwood@freescale.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
17 years agoAT91 MMC update for 2.6.19
Andrew Victor [Mon, 11 Dec 2006 11:40:23 +0000 (12:40 +0100)]
AT91 MMC update for 2.6.19

The driver is usable on the newer SAM9 processors so replace all text
references to AT91RM9200 with just AT91.

The controller bug where all the words are byte-swapped is fixed on the
AT91SAM9 processors.  The byte-swapping work-around therefore only needs
to be done if cpu_is_at91rm9200().
[Original patch from Wojtek Kaniewski]

The AT91RM9200 and AT91SAM9260 processors support two MMC/SD slots - the
slot which is connected is now passed via the platform_data and the
correct slot selected in the AT91_MCI_SDCR register.

The driver should not be calling at91_set_gpio_output() since the VCC
pin should have already been configured as an output in the
processor/board setup code.  The driver should call
at91_set_gpio_value().

Signed-off-by: Andrew Victor <andrew@sanpeople.com>
Signed-off-by: Pierre Ossman <drzeus@drzeus.cx>
17 years agommc: Change SDHCI iomem error to a warning
Pierre Ossman [Thu, 7 Dec 2006 18:17:20 +0000 (19:17 +0100)]
mmc: Change SDHCI iomem error to a warning

Some controllers report an invalid iomem size, but seem to work
correctly anyway. Change our current error to just a warning and
hope it doesn't cause too much problems.

Signed-off-by: Pierre Ossman <drzeus@drzeus.cx>
17 years agommc: fix "prev->state: 2 != TASK_RUNNING??" problem on SD/MMC card removal
Vitaly Wool [Thu, 7 Dec 2006 19:08:02 +0000 (20:08 +0100)]
mmc: fix "prev->state: 2 != TASK_RUNNING??" problem on SD/MMC  card removal

Currently on SD/MMC card removal the system exhibits the following message (the platform is ARM Versatile):

    prev->state: 2 != TASK_RUNNING??
    mmcqd/762[CPU#0]: BUG in __schedule at linux-2.6/kernel/sched.c:3826

(akpm: someone tried to fix this, but it's still wrong)

Signed-off-by: Vitaly Wool <vitalywool@gmail.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Pierre Ossman <drzeus@drzeus.cx>
17 years agoAT91 MMC 5 : Minor cleanups
Andrew Victor [Mon, 23 Oct 2006 12:53:20 +0000 (14:53 +0200)]
AT91 MMC 5 : Minor cleanups

A number of small cleanups to the AT91RM9200 MMC driver:
 - fix warnings generated by pr_debug().
 - prepend "AT91 MMC:" to printk() messages.

Signed-off-by: Andrew Victor <andrew@sanpeople.com>
Signed-off-by: Pierre Ossman <drzeus@drzeus.cx>
17 years agoAT91 MMC 4 : Interrupt handler cleanup
Andrew Victor [Mon, 23 Oct 2006 12:50:09 +0000 (14:50 +0200)]
AT91 MMC 4 : Interrupt handler cleanup

This patch simplifies the AT91RM9200 MMC interrupt handler code so that
it doesn't re-read the Interrupt Status and Interrupt Mask registers
multiple times.

Also defined AT91_MCI_ERRORS instead of using the hard-coded 0xffff0000.

Signed-off-by: Andrew Victor <andrew@sanpeople.com>
Signed-off-by: Pierre Ossman <drzeus@drzeus.cx>
17 years agoAT91 MMC 3 : Move global mci_clk variable
Andrew Victor [Mon, 23 Oct 2006 12:46:54 +0000 (14:46 +0200)]
AT91 MMC 3 : Move global mci_clk variable

Move the global 'mci_clk' variable into the local 'at91mci_host'
structure.

Signed-off-by: Andrew Victor <andrew@sanpeople.com>
Signed-off-by: Pierre Ossman <drzeus@drzeus.cx>
17 years agoAT91 MMC 2 : Use platform resources
Andrew Victor [Mon, 23 Oct 2006 12:44:40 +0000 (14:44 +0200)]
AT91 MMC 2 : Use platform resources

Use the I/O base-address and IRQ passed to the driver via the
platform_device resources instead of using hardcoded values.

Signed-off-by: Andrew Victor <andrew@sanpeople.com>
Signed-off-by: Pierre Ossman <drzeus@drzeus.cx>
17 years agoAT91 MMC 1: Pass host structure.
Andrew Victor [Wed, 25 Oct 2006 17:42:38 +0000 (19:42 +0200)]
AT91 MMC 1: Pass host structure.

The I/O base address is now stored in the 'at91mci_host' structure.  We
therefore have to pass this structure to at91_mci_read() and
at91_mci_write().

Signed-off-by: Andrew Victor <andrew@sanpeople.com>
Signed-off-by: Pierre Ossman <drzeus@drzeus.cx>
17 years ago[POWERPC] Generic BUG for powerpc
Jeremy Fitzhardinge [Fri, 8 Dec 2006 11:30:41 +0000 (03:30 -0800)]
[POWERPC] Generic BUG for powerpc

This makes powerpc use the generic BUG machinery.  The biggest reports the
function name, since it is redundant with kallsyms, and not needed in general.

There is an overall reduction of code, since module_32/64 duplicated several
functions.

Unfortunately there's no way to tell gcc that BUG won't return, so the BUG
macro includes a goto loop.  This will generate a real jmp instruction, which
is never used.

[akpm@osdl.org: build fix]
[paulus@samba.org: remove infinite loop in BUG_ON]
Signed-off-by: Jeremy Fitzhardinge <jeremy@goop.org>
Cc: Andi Kleen <ak@muc.de>
Cc: Hugh Dickens <hugh@veritas.com>
Cc: Michael Ellerman <michael@ellerman.id.au>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Paul Mackerras <paulus@samba.org>
17 years agoMerge branch 'for_paulus' of master.kernel.org:/pub/scm/linux/kernel/git/galak/powerpc
Paul Mackerras [Mon, 11 Dec 2006 05:31:42 +0000 (16:31 +1100)]
Merge branch 'for_paulus' of /linux/kernel/git/galak/powerpc

17 years ago[PPC] Fix compile failure do to introduction of PHY_POLL
Kumar Gala [Mon, 11 Dec 2006 05:26:16 +0000 (23:26 -0600)]
[PPC] Fix compile failure do to introduction of PHY_POLL

PHY_POLL is defined in <linux/phy.h> include it in <linux/fsl_devices.h> so
board code will have it defined.

Signed-off-by: Kumar Gala <galak@kernel.crashing.org>
17 years agoMerge branch '85xx' into for_paulus
Kumar Gala [Mon, 11 Dec 2006 05:16:08 +0000 (23:16 -0600)]
Merge branch '85xx' into for_paulus

17 years ago[POWERPC] Only export __mtdcr/__mfdcr if CONFIG_PPC_DCR is set
Kumar Gala [Mon, 11 Dec 2006 05:15:47 +0000 (23:15 -0600)]
[POWERPC] Only export __mtdcr/__mfdcr if CONFIG_PPC_DCR is set

On 85xx we don't build in dcr support because the core doesn't implement the
instructions.  This caused problems when building an 85xx kernel.  Additionally
made it so we only build __mtdcr/__mfdcr if we are CONFIG_PPC_DCR_NATIVE.

The 85xx build issue wasPointed out by Dai Haruki.

Signed-off-by: Kumar Gala <galak@kernel.crashing.org>
17 years ago[POWERPC] Remove old dcr.S
Benjamin Herrenschmidt [Mon, 11 Dec 2006 04:08:40 +0000 (15:08 +1100)]
[POWERPC] Remove old dcr.S

When I renamed dcr.S to dcr_low.S (and added dcr.c) it looks like the
old dcr.S file didn't properly get removed.  This fixes it.

Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: Paul Mackerras <paulus@samba.org>
17 years ago[POWERPC] Fix SPU coredump code for max_fdset removal
Paul Mackerras [Mon, 11 Dec 2006 04:13:37 +0000 (15:13 +1100)]
[POWERPC] Fix SPU coredump code for max_fdset removal

Commit bbea9f69668a3d0cf9feba15a724cd02896f8675 removed the max_fdset
element of struct fdtable.  It appears that checking max_fds is
sufficient now.

Signed-off-by: Paul Mackerras <paulus@samba.org>
17 years ago[POWERPC] Fix irq routing on some 32-bit PowerMacs
Benjamin Herrenschmidt [Mon, 11 Dec 2006 03:09:07 +0000 (14:09 +1100)]
[POWERPC] Fix irq routing on some 32-bit PowerMacs

The changes to use pci_read_irq_line() broke interrupt parsing
on some 32-bit powermacs (oops).  The reason is a bit obscure.
The code to parse interrupts happens earlier now, during
pcibios_fixup() as the PCI bus is being probed.  However, the
current implementation pci_device_to_OF_node() for 32-bit
powerpc relies, on machines like PowerMac which renumber PCI buses,
on a table called pci_OF_bus_map containing a map of bus numbers
between the kernel and the firmware which is setup only later.
Thus, it fails to match the device node.  In addition, some of
Apple internal PCI devices lack a proper PCI_INTERRUPT_PIN, thus
preventing the fallback mapping code to work.

This patch fixes it by making pci_device_to_OF_node() 32-bit
implementation use a different algorithm that works without
using the pci_OF_bus_map thing (which I intend to deprecate
anyway). It's a bit slower but that function isn't called in
any hot path hopefully.

Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
17 years ago[POWERPC] ps3: Add vuart support
Geoff Levand [Sat, 9 Dec 2006 02:27:47 +0000 (18:27 -0800)]
[POWERPC] ps3: Add vuart support

Adds support for the PS3 virtual UART (vuart).  The vuart provides a
bi-directional byte stream data link between logical partitions.

This is needed for the ps3 graphics driver and the ps3 power
control support to be able to communicate with the lv1 policy
module.

Signed-off-by: Geoff Levand <geoffrey.levand@am.sony.com>
Acked-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Paul Mackerras <paulus@samba.org>
17 years ago[POWERPC] Support ibm,dynamic-reconfiguration-memory nodes
Paul Mackerras [Wed, 29 Nov 2006 11:27:42 +0000 (22:27 +1100)]
[POWERPC] Support ibm,dynamic-reconfiguration-memory nodes

For PAPR partitions with large amounts of memory, the firmware has an
alternative, more compact representation for the information about the
memory in the partition and its NUMA associativity information.  This
adds the code to the kernel to parse this alternative representation.

The other part of this patch is telling the firmware that we can
handle the alternative representation.  There is however a subtlety
here, because the firmware will invoke a reboot if the memory
representation we request is different from the representation that
firmware is currently using.  This is because firmware can't change
the representation on the fly.  Further, some firmware versions used
on POWER5+ machines have a bug where this reboot leaves the machine
with an altered value of load-base, which will prevent any kernel
booting until it is reset to the normal value (0x4000).  Because of
this bug, we do NOT set fake_elf.rpanote.new_mem_def = 1, and thus we
do not request the new representation on POWER5+ and earlier machines.
We do request the new representation on POWER6, which uses the
ibm,client-architecture-support call.

Signed-off-by: Paul Mackerras <paulus@samba.org>
17 years ago[MIPS] Export local_flush_data_cache_page for sake of IDE.
Ralf Baechle [Sun, 10 Dec 2006 18:43:59 +0000 (18:43 +0000)]
[MIPS] Export local_flush_data_cache_page for sake of IDE.

On a CPU with aliases the IDE core needs to flush caches in the special
IDE variants of insw, insl etc.  If IDE support is built as a module this
will only work if local_flush_data_cache_page happens is exported as a
module.

As per policy export local_flush_data_cache_page as GPL symbol only.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
17 years ago[MIPS] Export pm_power_off
Ralf Baechle [Sun, 10 Dec 2006 15:09:38 +0000 (15:09 +0000)]
[MIPS] Export pm_power_off

This is required for ipmi_poweroff.c to work as a module.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
17 years ago[MIPS] Export csum_partial_copy_nocheck.
Ralf Baechle [Sun, 10 Dec 2006 15:05:11 +0000 (15:05 +0000)]
[MIPS] Export csum_partial_copy_nocheck.

ibmtr.c and typhoon.c use it.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
17 years ago[MIPS] Move die and die_if_kernel() from system.h to ptrace.h
Ralf Baechle [Sun, 10 Dec 2006 15:02:17 +0000 (15:02 +0000)]
[MIPS] Move die and die_if_kernel() from system.h to ptrace.h

This eleminates the need to include ptrace.h into system.h and fixes a
harmless namespace conflict on the PC symbol in bpck.c.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
17 years ago[MIPS] Discard .exit.text at linktime.
Ralf Baechle [Sun, 10 Dec 2006 14:57:28 +0000 (14:57 +0000)]
[MIPS] Discard .exit.text at linktime.

This fixes fairly unobvious breakage of various drivers.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
17 years ago[MIPS] Fix build of several IDE drivers by providing pci_get_legacy_ide_irq
Ralf Baechle [Sat, 9 Dec 2006 16:12:18 +0000 (16:12 +0000)]
[MIPS] Fix build of several IDE drivers by providing pci_get_legacy_ide_irq

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
17 years ago[CRYPTO] dm-crypt: Select CRYPTO_CBC
Herbert Xu [Sat, 9 Dec 2006 22:50:36 +0000 (09:50 +1100)]
[CRYPTO] dm-crypt: Select CRYPTO_CBC

As CBC is the default chaining method for cryptoloop, we should select
it from cryptoloop to ease the transition.  Spotted by Rene Herman.

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] add MODULE_* attributes to bit reversal library
Cal Peake [Sun, 10 Dec 2006 11:22:05 +0000 (06:22 -0500)]
[PATCH] add MODULE_* attributes to bit reversal library

Add MODULE_* attributes to the new bit reversal library. Most notably
MODULE_LICENSE which prevents superfluous kernel tainting.

Signed-off-by: Cal Peake <cp@absolutedigital.net>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years agoMerge master.kernel.org:/pub/scm/linux/kernel/git/davem/sparc-2.6
Linus Torvalds [Sun, 10 Dec 2006 18:00:00 +0000 (10:00 -0800)]
Merge /pub/scm/linux/kernel/git/davem/sparc-2.6

* master.kernel.org:/pub/scm/linux/kernel/git/davem/sparc-2.6:
  [SPARC64]: Fix several kprobes bugs.
  [SPARC64]: Update defconfig.
  [SPARC64]: dma remove extra brackets
  [SPARC{32,64}]: Propagate ptrace_traceme() return value.
  [SPARC64]: Replace kmalloc+memset with kzalloc
  [SPARC]: Check kzalloc() return value in SUN4D irq/iommu init.
  [SPARC]: Replace kmalloc+memset with kzalloc
  [SPARC64]: Run ctrl-alt-del action for sun4v powerdown request.
  [SPARC64]: Unaligned accesses to userspace are hard errors.
  [SPARC64]: Call do_mathemu on illegal instruction traps too.
  [SPARC64]: Update defconfig.
  [SPARC64]: Add irqtrace/stacktrace/lockdep support.

17 years agoMerge branch 'master' of master.kernel.org:/pub/scm/linux/kernel/git/mchehab/v4l-dvb
Linus Torvalds [Sun, 10 Dec 2006 17:59:18 +0000 (09:59 -0800)]
Merge branch 'master' of /linux/kernel/git/mchehab/v4l-dvb

* 'master' of master.kernel.org:/pub/scm/linux/kernel/git/mchehab/v4l-dvb: (132 commits)
  V4L/DVB 4949b: Fix container_of pointer retreival
  V4L/DVB (4949a): Fix INIT_WORK
  V4L/DVB (4949): Cxusb: codingstyle cleanups
  V4L/DVB (4948): Cxusb: Convert tuner functions to use dvb_pll_attach
  V4L/DVB (4947): Cx88: trivial cleanups
  V4L/DVB (4946): Cx88: Move cx88_dvb_bus_ctrl out of the card-specific area
  V4L/DVB (4945): Cx88: consolidate cx22702_config structs
  V4L/DVB (4944): Cx88: Convert DViCO FusionHDTV Hybrid to use dvb_pll_attach
  V4L/DVB (4943): Cx88: cleanup dvb_pll_attach for lgdt3302 tuners
  V4L/DVB (4953): Usbvision minor fixes
  V4L/DVB (4951): Add version.h, since it is required for VIDIOC_QUERYCAP
  V4L/DVB (4940): Or51211: Changed SNR and signal strength calculations
  V4L/DVB (4939): Or51132: Changed SNR and signal strength reporting
  V4L/DVB (4938): Cx88: Convert lgdt3302 tuning function to use dvb_pll_attach
  V4L/DVB (4941): Remove LINUX_VERSION_CODE and fix identations
  V4L/DVB (4942): Whitespace cleanups
  V4L/DVB (4937): Usbvision cleanup and code reorganization
  V4L/DVB (4936): Make MT4049FM5 tuner to set FM Gain to Normal
  V4L/DVB (4935): Added the capability of selecting fm gain by tuner
  V4L/DVB (4934): Usbvision radio requires GainNormal at e register
  ...

17 years ago[PATCH] kvm: userspace interface
Avi Kivity [Sun, 10 Dec 2006 10:21:36 +0000 (02:21 -0800)]
[PATCH] kvm: userspace interface

web site: http://kvm.sourceforge.net

mailing list: kvm-devel@lists.sourceforge.net
  (http://lists.sourceforge.net/lists/listinfo/kvm-devel)

The following patchset adds a driver for Intel's hardware virtualization
extensions to the x86 architecture.  The driver adds a character device
(/dev/kvm) that exposes the virtualization capabilities to userspace.  Using
this driver, a process can run a virtual machine (a "guest") in a fully
virtualized PC containing its own virtual hard disks, network adapters, and
display.

Using this driver, one can start multiple virtual machines on a host.

Each virtual machine is a process on the host; a virtual cpu is a thread in
that process.  kill(1), nice(1), top(1) work as expected.  In effect, the
driver adds a third execution mode to the existing two: we now have kernel
mode, user mode, and guest mode.  Guest mode has its own address space mapping
guest physical memory (which is accessible to user mode by mmap()ing
/dev/kvm).  Guest mode has no access to any I/O devices; any such access is
intercepted and directed to user mode for emulation.

The driver supports i386 and x86_64 hosts and guests.  All combinations are
allowed except x86_64 guest on i386 host.  For i386 guests and hosts, both pae
and non-pae paging modes are supported.

SMP hosts and UP guests are supported.  At the moment only Intel
hardware is supported, but AMD virtualization support is being worked on.

Performance currently is non-stellar due to the naive implementation of the
mmu virtualization, which throws away most of the shadow page table entries
every context switch.  We plan to address this in two ways:

- cache shadow page tables across tlb flushes
- wait until AMD and Intel release processors with nested page tables

Currently a virtual desktop is responsive but consumes a lot of CPU.  Under
Windows I tried playing pinball and watching a few flash movies; with a recent
CPU one can hardly feel the virtualization.  Linux/X is slower, probably due
to X being in a separate process.

In addition to the driver, you need a slightly modified qemu to provide I/O
device emulation and the BIOS.

Caveats (akpm: might no longer be true):

- The Windows install currently bluescreens due to a problem with the
  virtual APIC.  We are working on a fix.  A temporary workaround is to
  use an existing image or install through qemu
- Windows 64-bit does not work.  That's also true for qemu, so it's
  probably a problem with the device model.

[bero@arklinux.org: build fix]
[simon.kagstrom@bth.se: build fix, other fixes]
[uril@qumranet.com: KVM: Expose interrupt bitmap]
[akpm@osdl.org: i386 build fix]
[mingo@elte.hu: i386 fixes]
[rdreier@cisco.com: add log levels to all printks]
[randy.dunlap@oracle.com: Fix sparse NULL and C99 struct init warnings]
[anthony@codemonkey.ws: KVM: AMD SVM: 32-bit host support]
Signed-off-by: Yaniv Kamay <yaniv@qumranet.com>
Signed-off-by: Avi Kivity <avi@qumranet.com>
Cc: Simon Kagstrom <simon.kagstrom@bth.se>
Cc: Bernhard Rosenkraenzer <bero@arklinux.org>
Signed-off-by: Uri Lublin <uril@qumranet.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Roland Dreier <rolandd@cisco.com>
Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com>
Signed-off-by: Anthony Liguori <anthony@codemonkey.ws>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] clocksource: small cleanup
Daniel Walker [Sun, 10 Dec 2006 10:21:33 +0000 (02:21 -0800)]
[PATCH] clocksource: small cleanup

Mostly changing alignment.  Just some general cleanup.

[akpm@osdl.org: build fix]
Signed-off-by: Daniel Walker <dwalker@mvista.com>
Acked-by: John Stultz <johnstul@us.ibm.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] clocksource: add usage of CONFIG_SYSFS
Daniel Walker [Sun, 10 Dec 2006 10:21:30 +0000 (02:21 -0800)]
[PATCH] clocksource: add usage of CONFIG_SYSFS

Simply adds some ifdefs to remove clocksoure sysfs code when CONFIG_SYSFS
isn't turn on.

Signed-off-by: Daniel Walker <dwalker@mvista.com>
Acked-by: John Stultz <johnstul@us.ibm.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] user of the jiffies rounding patch: Slab
Arjan van de Ven [Sun, 10 Dec 2006 10:21:28 +0000 (02:21 -0800)]
[PATCH] user of the jiffies rounding patch: Slab

This patch introduces users of the round_jiffies() function in the slab code.

The slab code has a few "run every second" timers for background work; these
are obviously not timing critical as long as they happen roughly at the right
frequency.

Signed-off-by: Arjan van de Ven <arjan@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] user of the jiffies rounding code: JBD
Arjan van de Ven [Sun, 10 Dec 2006 10:21:26 +0000 (02:21 -0800)]
[PATCH] user of the jiffies rounding code: JBD

This patch introduces a user: of the round_jiffies() function; the "5 second"
ext3/jbd wakeup.

While "every 5 seconds" doesn't sound as a problem, there can be many of these
(and these timers do add up over all the kernel).  The "5 second" wakeup isn't
really timing sensitive; in addition even with rounding it'll still happen
every 5 seconds (with the exception of the very first time, which is likely to
be rounded up to somewhere closer to 6 seconds)

Signed-off-by: Arjan van de Ven <arjan@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] round_jiffies infrastructure
Arjan van de Ven [Sun, 10 Dec 2006 10:21:24 +0000 (02:21 -0800)]
[PATCH] round_jiffies infrastructure

Introduce a round_jiffies() function as well as a round_jiffies_relative()
function.  These functions round a jiffies value to the next whole second.
The primary purpose of this rounding is to cause all "we don't care exactly
when" timers to happen at the same jiffy.

This avoids multiple timers firing within the second for no real reason;
with dynamic ticks these extra timers cause wakeups from deep sleep CPU
sleep states and thus waste power.

The exact wakeup moment is skewed by the cpu number, to avoid all cpus from
waking up at the exact same time (and hitting the same lock/cachelines
there)

[akpm@osdl.org: fix variable type]
Signed-off-by: Arjan van de Ven <arjan@linux.intel.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] fdtable: Implement new pagesize-based fdtable allocator
Vadim Lobanov [Sun, 10 Dec 2006 10:21:22 +0000 (02:21 -0800)]
[PATCH] fdtable: Implement new pagesize-based fdtable allocator

This patch provides an improved fdtable allocation scheme, useful for
expanding fdtable file descriptor entries.  The main focus is on the fdarray,
as its memory usage grows 128 times faster than that of an fdset.

The allocation algorithm sizes the fdarray in such a way that its memory usage
increases in easy page-sized chunks. The overall algorithm expands the allowed
size in powers of two, in order to amortize the cost of invoking vmalloc() for
larger allocation sizes. Namely, the following sizes for the fdarray are
considered, and the smallest that accommodates the requested fd count is
chosen:

    pagesize / 4
    pagesize / 2
    pagesize      <- memory allocator switch point
    pagesize * 2
    pagesize * 4
    ...etc...

Unlike the current implementation, this allocation scheme does not require a
loop to compute the optimal fdarray size, and can be done in efficient
straightline code.

Furthermore, since the fdarray overflows the pagesize boundary long before any
of the fdsets do, it makes sense to optimize run-time by allocating both
fdsets in a single swoop.  Even together, they will still be, by far, smaller
than the fdarray.  The fdtable->open_fds is now used as the anchor for the
fdset memory allocation.

Signed-off-by: Vadim Lobanov <vlobanov@speakeasy.net>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Dipankar Sarma <dipankar@in.ibm.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] fdtable: Remove the free_files field
Vadim Lobanov [Sun, 10 Dec 2006 10:21:17 +0000 (02:21 -0800)]
[PATCH] fdtable: Remove the free_files field

An fdtable can either be embedded inside a files_struct or standalone (after
being expanded).  When an fdtable is being discarded after all RCU references
to it have expired, we must either free it directly, in the standalone case,
or free the files_struct it is contained within, in the embedded case.

Currently the free_files field controls this behavior, but we can get rid of
it entirely, as all the necessary information is already recorded.  We can
distinguish embedded and standalone fdtables using max_fds, and if it is
embedded we can divine the relevant files_struct using container_of().

Signed-off-by: Vadim Lobanov <vlobanov@speakeasy.net>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Dipankar Sarma <dipankar@in.ibm.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] fdtable: Make fdarray and fdsets equal in size
Vadim Lobanov [Sun, 10 Dec 2006 10:21:12 +0000 (02:21 -0800)]
[PATCH] fdtable: Make fdarray and fdsets equal in size

Currently, each fdtable supports three dynamically-sized arrays of data: the
fdarray and two fdsets.  The code allows the number of fds supported by the
fdarray (fdtable->max_fds) to differ from the number of fds supported by each
of the fdsets (fdtable->max_fdset).

In practice, it is wasteful for these two sizes to differ: whenever we hit a
limit on the smaller-capacity structure, we will reallocate the entire fdtable
and all the dynamic arrays within it, so any delta in the memory used by the
larger-capacity structure will never be touched at all.

Rather than hogging this excess, we shouldn't even allocate it in the first
place, and keep the capacities of the fdarray and the fdsets equal.  This
patch removes fdtable->max_fdset.  As an added bonus, most of the supporting
code becomes simpler.

Signed-off-by: Vadim Lobanov <vlobanov@speakeasy.net>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Dipankar Sarma <dipankar@in.ibm.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] fdtable: Delete pointless code in dup_fd()
Vadim Lobanov [Sun, 10 Dec 2006 10:21:09 +0000 (02:21 -0800)]
[PATCH] fdtable: Delete pointless code in dup_fd()

The dup_fd() function creates a new files_struct and fdtable embedded inside
that files_struct, and then possibly expands the fdtable using expand_files().

The out_release error path is invoked when expand_files() returns an error
code.  However, when this attempt to expand fails, the fdtable is left in its
original embedded form, so it is pointless to try to free the associated
fdarray and fdsets.

Signed-off-by: Vadim Lobanov <vlobanov@speakeasy.net>
Cc: Dipankar Sarma <dipankar@in.ibm.com>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] dio: lock refcount operations
Zach Brown [Sun, 10 Dec 2006 10:21:07 +0000 (02:21 -0800)]
[PATCH] dio: lock refcount operations

The wait_for_more_bios() function name was poorly chosen.  While looking to
clean it up it I noticed that the dio struct refcounting between the bio
completion and dio submission paths was racey.

The bio submission path was simply freeing the dio struct if
atomic_dec_and_test() indicated that it dropped the final reference.

The aio bio completion path was dereferencing its dio struct pointer *after
dropping its reference* based on the remaining number of references.

These two paths could race and result in the aio bio completion path
dereferencing a freed dio, though this was not observed in the wild.

This moves the refcount under the bio lock so that bio completion can drop
its reference and decide to wake all in one atomic step.

Once testing and waking is locked dio_await_one() can test its sleeping
condition and mark itself uninterruptible under the lock.  It gets simpler
and wait_for_more_bios() disappears.

The addition of the interrupt masking spin lock acquiry in dio_bio_submit()
looks alarming.  This lock acquiry existed in that path before the recent
dio completion patch set.  We shouldn't expect significant performance
regression from returning to the behaviour that existed before the
completion clean up work.

This passed 4k block ext3 O_DIRECT fsx and aio-stress on an SMP machine.

Signed-off-by: Zach Brown <zach.brown@oracle.com>
Cc: Badari Pulavarty <pbadari@us.ibm.com>
Cc: Suparna Bhattacharya <suparna@in.ibm.com>
Cc: Jeff Moyer <jmoyer@redhat.com>
Cc: <xfs-masters@oss.sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] dio: only call aio_complete() after returning -EIOCBQUEUED
Zach Brown [Sun, 10 Dec 2006 10:21:05 +0000 (02:21 -0800)]
[PATCH] dio: only call aio_complete() after returning -EIOCBQUEUED

The only time it is safe to call aio_complete() is when the ->ki_retry
function returns -EIOCBQUEUED to the AIO core.  direct_io_worker() has
historically done this by relying on its caller to translate positive return
codes into -EIOCBQUEUED for the aio case.  It did this by trying to keep
conditionals in sync.  direct_io_worker() knew when finished_one_bio() was
going to call aio_complete().  It would reverse the test and wait and free the
dio in the cases it thought that finished_one_bio() wasn't going to.

Not surprisingly, it ended up getting it wrong.  'ret' could be a negative
errno from the submission path but it failed to communicate this to
finished_one_bio().  direct_io_worker() would return < 0, it's callers
wouldn't raise -EIOCBQUEUED, and aio_complete() would be called.  In the
future finished_one_bio()'s tests wouldn't reflect this and aio_complete()
would be called for a second time which can manifest as an oops.

The previous cleanups have whittled the sync and async completion paths down
to the point where we can collapse them and clearly reassert the invariant
that we must only call aio_complete() after returning -EIOCBQUEUED.
direct_io_worker() will only return -EIOCBQUEUED when it is not the last to
drop the dio refcount and the aio bio completion path will only call
aio_complete() when it is the last to drop the dio refcount.
direct_io_worker() can ensure that it is the last to drop the reference count
by waiting for bios to drain.  It does this for sync ops, of course, and for
partial dio writes that must fall back to buffered and for aio ops that saw
errors during submission.

This means that operations that end up waiting, even if they were issued as
aio ops, will not call aio_complete() from dio.  Instead we return the return
code of the operation and let the aio core call aio_complete().  This is
purposely done to fix a bug where AIO DIO file extensions would call
aio_complete() before their callers have a chance to update i_size.

Now that direct_io_worker() is explicitly returning -EIOCBQUEUED its callers
no longer have to translate for it.  XFS needs to be careful not to free
resources that will be used during AIO completion if -EIOCBQUEUED is returned.
 We maintain the previous behaviour of trying to write fs metadata for O_SYNC
aio+dio writes.

Signed-off-by: Zach Brown <zach.brown@oracle.com>
Cc: Badari Pulavarty <pbadari@us.ibm.com>
Cc: Suparna Bhattacharya <suparna@in.ibm.com>
Acked-by: Jeff Moyer <jmoyer@redhat.com>
Cc: <xfs-masters@oss.sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] dio: remove duplicate bio wait code
Zach Brown [Sun, 10 Dec 2006 10:21:01 +0000 (02:21 -0800)]
[PATCH] dio: remove duplicate bio wait code

Now that we have a single refcount and waiting path we can reuse it in the
async 'should_wait' path.  It continues to rely on the fragile link between
the conditional in dio_complete_aio() which decides to complete the AIO and
the conditional in direct_io_worker() which decides to wait and free.

By waiting before dropping the reference we stop dio_bio_end_aio() from
calling dio_complete_aio() which used to wake up the waiter after seeing the
reference count drop to 0.  We hoist this wake up into dio_bio_end_aio() which
now notices when it's left a single remaining reference that is held by the
waiter.

Signed-off-by: Zach Brown <zach.brown@oracle.com>
Cc: Badari Pulavarty <pbadari@us.ibm.com>
Cc: Suparna Bhattacharya <suparna@in.ibm.com>
Acked-by: Jeff Moyer <jmoyer@redhat.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] dio: formalize bio counters as a dio reference count
Zach Brown [Sun, 10 Dec 2006 10:20:59 +0000 (02:20 -0800)]
[PATCH] dio: formalize bio counters as a dio reference count

Previously we had two confusing counts of bio progress.  'bio_count' was
decremented as bios were processed and freed by the dio core.  It was used to
indicate final completion of the dio operation.  'bios_in_flight' reflected
how many bios were between submit_bio() and bio->end_io.  It was used by the
sync path to decide when to wake up and finish completing bios and was ignored
by the async path.

This patch collapses the two notions into one notion of a dio reference count.
 bios hold a dio reference when they're between submit_bio and bio->end_io.

Since bios_in_flight was only used in the sync path it is now equivalent to
dio->refcount - 1 which accounts for direct_io_worker() holding a reference
for the duration of the operation.

dio_bio_complete() -> finished_one_bio() was called from the sync path after
finding bios on the list that the bio->end_io function had deposited.
finished_one_bio() can not drop the dio reference on behalf of these bios now
because bio->end_io already has.  The is_async test in finished_one_bio()
meant that it never actually did anything other than drop the bio_count for
sync callers.  So we remove its refcount decrement, don't call it from
dio_bio_complete(), and hoist its call up into the async dio_bio_complete()
caller after an explicit refcount decrement.  It is renamed dio_complete_aio()
to reflect the remaining work it actually does.

Signed-off-by: Zach Brown <zach.brown@oracle.com>
Cc: Badari Pulavarty <pbadari@us.ibm.com>
Cc: Suparna Bhattacharya <suparna@in.ibm.com>
Acked-by: Jeff Moyer <jmoyer@redhat.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] dio: call blk_run_address_space() once per op
Zach Brown [Sun, 10 Dec 2006 10:20:56 +0000 (02:20 -0800)]
[PATCH] dio: call blk_run_address_space() once per op

We only need to call blk_run_address_space() once after all the bios for the
direct IO op have been submitted.  This removes the chance of calling
blk_run_address_space() after spurious wake ups as the sync path waits for
bios to drain.  It's also one less difference betwen the sync and async paths.

In the process we remove a redundant dio_bio_submit() that its caller had
already performed.

Signed-off-by: Zach Brown <zach.brown@oracle.com>
Cc: Badari Pulavarty <pbadari@us.ibm.com>
Cc: Suparna Bhattacharya <suparna@in.ibm.com>
Acked-by: Jeff Moyer <jmoyer@redhat.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] dio: centralize completion in dio_complete()
Zach Brown [Sun, 10 Dec 2006 10:20:54 +0000 (02:20 -0800)]
[PATCH] dio: centralize completion in dio_complete()

There have been a lot of bugs recently due to the way direct_io_worker() tries
to decide how to finish direct IO operations.  In the worst examples it has
failed to call aio_complete() at all (hang) or called it too many times
(oops).

This set of patches cleans up the completion phase with the goal of removing
the complexity that lead to these bugs.  We end up with one path that
calculates the result of the operation after all off the bios have completed.
We decide when to generate a result of the operation using that path based on
the final release of a refcount on the dio structure.

I tried to progress towards the final state in steps that were relatively easy
to understand.  Each step should compile but I only tested the final result of
having all the patches applied.

I've tested these on low end PC drives with aio-stress, the direct IO tests I
could manage to get running in LTP, orasim, and some home-brew functional
tests.

In http://lkml.org/lkml/2006/9/21/103 IBM reports success with ext2 and ext3
running DIO LTP tests.  They found that XFS bug which has since been addressed
in the patch series.

This patch:

The mechanics which decide the result of a direct IO operation were duplicated
in the sync and async paths.

The async path didn't check page_errors which can manifest as silently
returning success when the final pointer in an operation faults and its
matching file region is filled with zeros.

The sync path and async path differed in whether they passed errors to the
caller's dio->end_io operation.  The async path was passing errors to it which
trips an assertion in XFS, though it is apparently harmless.

This centralizes the completion phase of dio ops in one place.  AIO will now
return EFAULT consistently and all paths fall back to the previously sync
behaviour of passing the number of bytes 'transferred' to the dio->end_io
callback, regardless of errors.

dio_await_completion() doesn't have to propogate EIO from non-uptodate bios
now that it's being propogated through dio_complete() via dio->io_error.  This
lets it return void which simplifies its sole caller.

Signed-off-by: Zach Brown <zach.brown@oracle.com>
Cc: Badari Pulavarty <pbadari@us.ibm.com>
Cc: Suparna Bhattacharya <suparna@in.ibm.com>
Acked-by: Jeff Moyer <jmoyer@redhat.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] md: assorted md and raid1 one-liners
NeilBrown [Sun, 10 Dec 2006 10:20:52 +0000 (02:20 -0800)]
[PATCH] md: assorted md and raid1 one-liners

Fix few bugs that meant that:
  - superblocks weren't alway written at exactly the right time (this
    could show up if the array was not written to - writting to the array
    causes lots of superblock updates and so hides these errors).

  - restarting device recovery after a clean shutdown (version-1 metadata
    only) didn't work as intended (or at all).

1/ Ensure superblock is updated when a new device is added.
2/ Remove an inappropriate test on MD_RECOVERY_SYNC in md_do_sync.
   The body of this if takes one of two branches depending on whether
   MD_RECOVERY_SYNC is set, so testing it in the clause of the if
   is wrong.
3/ Flag superblock for updating after a resync/recovery finishes.
4/ If we find the neeed to restart a recovery in the middle (version-1
   metadata only) make sure a full recovery (not just as guided by
   bitmaps) does get done.

Signed-off-by: Neil Brown <neilb@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] md: return a non-zero error to bi_end_io as appropriate in raid5
NeilBrown [Sun, 10 Dec 2006 10:20:51 +0000 (02:20 -0800)]
[PATCH] md: return a non-zero error to bi_end_io as appropriate in raid5

Currently raid5 depends on clearing the BIO_UPTODATE flag to signal an error
to higher levels.  While this should be sufficient, it is safer to explicitly
set the error code as well - less room for confusion.

Signed-off-by: Neil Brown <neilb@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] md: remove some old ifdefed-out code from raid5.c
NeilBrown [Sun, 10 Dec 2006 10:20:50 +0000 (02:20 -0800)]
[PATCH] md: remove some old ifdefed-out code from raid5.c

There are some vestiges of old code that was used for bypassing the stripe
cache on reads in raid5.c.  This was never updated after the change from
buffer_heads to bios, but was left as a reminder.

That functionality has nowe been implemented in a completely different way, so
the old code can go.

Signed-off-by: Neil Brown <neilb@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] MD: conditionalize some code
Jeff Garzik [Sun, 10 Dec 2006 10:20:50 +0000 (02:20 -0800)]
[PATCH] MD: conditionalize some code

The autorun code is only used if this module is built into the static
kernel image.  Adjust #ifdefs accordingly.

Signed-off-by: Jeff Garzik <jeff@garzik.org>
Acked-by: NeilBrown <neilb@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] md: fix innocuous bug in raid6 stripe_to_pdidx
NeilBrown [Sun, 10 Dec 2006 10:20:49 +0000 (02:20 -0800)]
[PATCH] md: fix innocuous bug in raid6 stripe_to_pdidx

stripe_to_pdidx finds the index of the parity disk for a given stripe.  It
assumes raid5 in that it uses "disks-1" to determine the number of data disks.

This is incorrect for raid6 but fortunately the two usages cancel each other
out.  The only way that 'data_disks' affects the calculation of pd_idx in
raid5_compute_sector is when it is divided into the sector number.  But as
that sector number is calculated by multiplying in the wrong value of
'data_disks' the division produces the right value.

So it is innocuous but needs to be fixed.

Also change the calculation of raid_disks in compute_blocknr to make it
more obviously correct (it seems at first to always use disks-1 too).

Signed-off-by: Neil Brown <neilb@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] md: enable bypassing cache for reads
Raz Ben-Jehuda(caro) [Sun, 10 Dec 2006 10:20:48 +0000 (02:20 -0800)]
[PATCH] md: enable bypassing cache for reads

Call the chunk_aligned_read where appropriate.

Signed-off-by: Neil Brown <neilb@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] md: allow reads that have bypassed the cache to be retried on failure
Raz Ben-Jehuda(caro) [Sun, 10 Dec 2006 10:20:47 +0000 (02:20 -0800)]
[PATCH] md: allow reads that have bypassed the cache to be retried on failure

If a bypass-the-cache read fails, we simply try again through the cache.  If
it fails again it will trigger normal recovery precedures.

update 1:

From: NeilBrown <neilb@suse.de>

1/
  chunk_aligned_read and retry_aligned_read assume that
      data_disks == raid_disks - 1
  which is not true for raid6.
  So when an aligned read request bypasses the cache, we can get the wrong data.

2/ The cloned bio is being used-after-free in raid5_align_endio
   (to test BIO_UPTODATE).

3/ We forgot to add rdev->data_offset when submitting
   a bio for aligned-read

4/ clone_bio calls blk_recount_segments and then we change bi_bdev,
   so we need to invalidate the segment counts.

5/ We don't de-reference the rdev when the read completes.
   This means we need to record the rdev to so it is still
   available in the end_io routine.  Fortunately
   bi_next in the original bio is unused at this point so
   we can stuff it in there.

6/ We leak a cloned bio if the target rdev is not usable.

From: NeilBrown <neilb@suse.de>

update 2:

1/ When aligned requests fail (read error) they need to be retried
   via the normal method (stripe cache).  As we cannot be sure that
   we can process a single read in one go (we may not be able to
   allocate all the stripes needed) we store a bio-being-retried
   and a list of bioes-that-still-need-to-be-retried.
   When find a bio that needs to be retried, we should add it to
   the list, not to single-bio...

2/ We were never incrementing 'scnt' when resubmitting failed
   aligned requests.

[akpm@osdl.org: build fix]
Signed-off-by: Neil Brown <neilb@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] md: handle bypassing the read cache (assuming nothing fails)
Raz Ben-Jehuda(caro) [Sun, 10 Dec 2006 10:20:46 +0000 (02:20 -0800)]
[PATCH] md: handle bypassing the read cache (assuming nothing fails)

Signed-off-by: Neil Brown <neilb@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] md: define raid5_mergeable_bvec
Raz Ben-Jehuda(caro) [Sun, 10 Dec 2006 10:20:45 +0000 (02:20 -0800)]
[PATCH] md: define raid5_mergeable_bvec

This will encourage read request to be on only one device, so we will often be
able to bypass the cache for read requests.

Signed-off-by: Neil Brown <neilb@suse.de>
Cc: Jens Axboe <jens.axboe@oracle.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] md: tidy up device-change notification when an md array is stopped
NeilBrown [Sun, 10 Dec 2006 10:20:44 +0000 (02:20 -0800)]
[PATCH] md: tidy up device-change notification when an md array is stopped

An md array can be stopped leaving all the setting still in place, or it can
torn down and destroyed.  set_capacity and other change notifications only
happen in the latter case, but should happen in both.

Signed-off-by: Neil Brown <neilb@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] Fbdev driver for IBM GXT4500P videocards
Paul Mackerras [Sun, 10 Dec 2006 10:20:42 +0000 (02:20 -0800)]
[PATCH] Fbdev driver for IBM GXT4500P videocards

This is an fbdev driver for the IBM GXT4500P display card found in some IBM
System P (pSeries) machines.  These cards have hardware 2D and 3D
capabilities, but the driver does not use them; it just exports a dumb
framebuffer.

Signed-off-by: Paul Mackerras <paulus@samba.org>
Acked-by: James Simmons <jsimmons@infradead.org>
Cc: "Antonino A. Daplas" <adaplas@pol.net>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] ide-cd: Handle strange interrupt on the Intel ESB2
Alan Cox [Sun, 10 Dec 2006 10:20:39 +0000 (02:20 -0800)]
[PATCH] ide-cd: Handle strange interrupt on the Intel ESB2

The ESB2 appears to emit spurious DMA interrupts when configured for native
mode and handling ATAPI devices.  Stratus were able to pin this bug down and
produce a patch.  This is a rework which applies the fixup only to the ESB2
(for now).  We can apply it to other chips later if the same problem is found.

This code has been tested and confirmed to fix the problem on the tested
systems.

Signed-off-by: Alan Cox <alan@redhat.com>
(Most of the hard work done by Stratus however)
Cc: Jens Axboe <axboe@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] kernel/sched.c: whitespace cleanups
Miguel Ojeda Sandonis [Sun, 10 Dec 2006 10:20:38 +0000 (02:20 -0800)]
[PATCH] kernel/sched.c: whitespace cleanups

[akpm@osdl.org: additional cleanups]
Signed-off-by: Miguel Ojeda Sandonis <maxextreme@gmail.com>
Acked-by: Ingo Molnar <mingo@elte.hu>
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] sched: optimize activate_task for RT task
Chen, Kenneth W [Sun, 10 Dec 2006 10:20:36 +0000 (02:20 -0800)]
[PATCH] sched: optimize activate_task for RT task

RT task does not participate in interactiveness priority and thus shouldn't
be bothered with timestamp and p->sleep_type manipulation when task is
being put on run queue.  Bypass all of the them with a single if (rt_task)
test.

Signed-off-by: Ken Chen <kenneth.w.chen@intel.com>
Acked-by: Ingo Molnar <mingo@elte.hu>
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Cc: "Siddha, Suresh B" <suresh.b.siddha@intel.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] sched: remove lb_stopbalance counter
Chen, Kenneth W [Sun, 10 Dec 2006 10:20:35 +0000 (02:20 -0800)]
[PATCH] sched: remove lb_stopbalance counter

Remove scheduler stats lb_stopbalance counter.  This counter can be
calculated by: lb_balanced - lb_nobusyg - lb_nobusyq.  There is no need to
create gazillion counters while we can derive the value.

Signed-off-by: Ken Chen <kenneth.w.chen@intel.com>
Signed-off-by: Suresh Siddha <suresh.b.siddha@intel.com>
Acked-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] sched: decrease number of load balances
Siddha, Suresh B [Sun, 10 Dec 2006 10:20:33 +0000 (02:20 -0800)]
[PATCH] sched: decrease number of load balances

Currently at a particular domain, each cpu in the sched group will do a
load balance at the frequency of balance_interval.  More the cores and
threads, more the cpus will be in each sched group at SMP and NUMA domain.
And we endup spending quite a bit of time doing load balancing in those
domains.

Fix this by making only one cpu(first idle cpu or first cpu in the group if
all the cpus are busy) in the sched group do the load balance at that
particular sched domain and this load will slowly percolate down to the
other cpus with in that group(when they do load balancing at lower
domains).

Signed-off-by: Suresh Siddha <suresh.b.siddha@intel.com>
Cc: Christoph Lameter <clameter@engr.sgi.com>
Cc: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] sched: improve migration accuracy
Mike Galbraith [Sun, 10 Dec 2006 10:20:31 +0000 (02:20 -0800)]
[PATCH] sched: improve migration accuracy

Co-opt rq->timestamp_last_tick to maintain a cache_hot_time evaluation
reference timestamp at both tick and sched times to prevent said reference,
formerly rq->timestamp_last_tick, from being behind task->last_ran at
evaluation time, and to move said reference closer to current time on the
remote processor, intent being to improve cache hot evaluation and
timestamp adjustment accuracy for task migration.

Fix minor sched_time double accounting error which occurs when a task
passing through schedule() does not schedule off, and takes the next timer
tick.

[kenneth.w.chen@intel.com: cleanup]
Signed-off-by: Mike Galbraith <efault@gmx.de>
Acked-by: Ingo Molnar <mingo@elte.hu>
Acked-by: Ken Chen <kenneth.w.chen@intel.com>
Cc: Don Mullis <dwm@meer.net>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] sched: add option to serialize load balancing
Christoph Lameter [Sun, 10 Dec 2006 10:20:29 +0000 (02:20 -0800)]
[PATCH] sched: add option to serialize load balancing

Large sched domains can be very expensive to scan.  Add an option SD_SERIALIZE
to the sched domain flags.  If that flag is set then we make sure that no
other such domain is being balanced.

[akpm@osdl.org: build fix]
Signed-off-by: Christoph Lameter <clameter@sgi.com>
Cc: Peter Williams <pwil3058@bigpond.net.au>
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Cc: Christoph Lameter <clameter@sgi.com>
Cc: "Siddha, Suresh B" <suresh.b.siddha@intel.com>
Cc: "Chen, Kenneth W" <kenneth.w.chen@intel.com>
Acked-by: Ingo Molnar <mingo@elte.hu>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] sched: call tasklet less frequently
Christoph Lameter [Sun, 10 Dec 2006 10:20:27 +0000 (02:20 -0800)]
[PATCH] sched: call tasklet less frequently

Trigger softirq less frequently

We trigger the softirq before this patch using offset of sd->interval.
However, if the queue is busy then it is sufficient to schedule the softirq
with sd->interval * busy_factor.

So we modify the calculation of the next time to balance by taking
the interval added to last_balance again. This is only the
right value if the idle/busy situation continues as is.

There are two potential trouble spots:
- If the queue was idle and now gets busy then we call rebalance
  early. However, that is not a problem because we will then use
  the longer interval for the next period.

- If the queue was busy and becomes idle then we potentially
  wait too long before rebalancing. However, when the task
  goes idle then idle_balance is called. We add another calculation
  of the next balance time based on sd->interval in idle_balance
  so that we will rebalance soon.

V2->V3:
- Calculate rebalance time based on current jiffies and not
  based on the jiffies at the last time we load balanced.
  We no longer rely on staggering and therefore we can
  affort to do this now.

V3->V4:
- Use functions to do jiffy comparisons.

Signed-off-by: Christoph Lameter <clameter@sgi.com>
Cc: Peter Williams <pwil3058@bigpond.net.au>
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Cc: Christoph Lameter <clameter@sgi.com>
Cc: "Siddha, Suresh B" <suresh.b.siddha@intel.com>
Cc: "Chen, Kenneth W" <kenneth.w.chen@intel.com>
Acked-by: Ingo Molnar <mingo@elte.hu>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] sched: use softirq for load balancing
Christoph Lameter [Sun, 10 Dec 2006 10:20:25 +0000 (02:20 -0800)]
[PATCH] sched: use softirq for load balancing

Call rebalance_tick (renamed to run_rebalance_domains) from a newly introduced
softirq.

We calculate the earliest time for each layer of sched domains to be rescanned
(this is the rescan time for idle) and use the earliest of those to schedule
the softirq via a new field "next_balance" added to struct rq.

Signed-off-by: Christoph Lameter <clameter@sgi.com>
Cc: Peter Williams <pwil3058@bigpond.net.au>
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Cc: Christoph Lameter <clameter@sgi.com>
Cc: "Siddha, Suresh B" <suresh.b.siddha@intel.com>
Cc: "Chen, Kenneth W" <kenneth.w.chen@intel.com>
Acked-by: Ingo Molnar <mingo@elte.hu>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] sched: move idle status calculation into rebalance_tick()
Christoph Lameter [Sun, 10 Dec 2006 10:20:23 +0000 (02:20 -0800)]
[PATCH] sched: move idle status calculation into rebalance_tick()

Perform the idle state determination in rebalance_tick.

If we separate balancing from sched_tick then we also need to determine the
idle state in rebalance_tick.

V2->V3
Remove useless idlle != 0 check. Checking nr_running seems
to be sufficient. Thanks Suresh.

Signed-off-by: Christoph Lameter <clameter@sgi.com>
Cc: Peter Williams <pwil3058@bigpond.net.au>
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Cc: Christoph Lameter <clameter@sgi.com>
Cc: "Siddha, Suresh B" <suresh.b.siddha@intel.com>
Cc: "Chen, Kenneth W" <kenneth.w.chen@intel.com>
Acked-by: Ingo Molnar <mingo@elte.hu>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] sched: extract load calculation from rebalance_tick
Christoph Lameter [Sun, 10 Dec 2006 10:20:22 +0000 (02:20 -0800)]
[PATCH] sched: extract load calculation from rebalance_tick

A load calculation is always done in rebalance_tick() in addition to the real
load balancing activities that only take place when certain jiffie counts have
been reached.  Move that processing into a separate function and call it
directly from scheduler_tick().

Also extract the time slice handling from scheduler_tick and put it into a
separate function.  Then we can clean up scheduler_tick significantly.  It
will no longer have any gotos.

Signed-off-by: Christoph Lameter <clameter@sgi.com>
Cc: Peter Williams <pwil3058@bigpond.net.au>
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Cc: Christoph Lameter <clameter@sgi.com>
Cc: "Siddha, Suresh B" <suresh.b.siddha@intel.com>
Cc: "Chen, Kenneth W" <kenneth.w.chen@intel.com>
Acked-by: Ingo Molnar <mingo@elte.hu>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] sched: disable interrupts for locking in load_balance()
Christoph Lameter [Sun, 10 Dec 2006 10:20:21 +0000 (02:20 -0800)]
[PATCH] sched: disable interrupts for locking in load_balance()

Interrupts must be disabled for request queue locks if we want to run
load_balance() with interrupts enabled.

Signed-off-by: Christoph Lameter <clameter@sgi.com>
Cc: Peter Williams <pwil3058@bigpond.net.au>
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Cc: Christoph Lameter <clameter@sgi.com>
Cc: "Siddha, Suresh B" <suresh.b.siddha@intel.com>
Cc: "Chen, Kenneth W" <kenneth.w.chen@intel.com>
Acked-by: Ingo Molnar <mingo@elte.hu>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] sched: remove staggering of load balancing
Christoph Lameter [Sun, 10 Dec 2006 10:20:19 +0000 (02:20 -0800)]
[PATCH] sched: remove staggering of load balancing

Timer interrupts already are staggered.  We do not need an additional layer of
time staggering for short load balancing actions that take a reasonably small
portion of the time slice.

For load balancing on large sched_domains we will add a serialization later
that avoids concurrent load balance operations and thus has the same effect as
load staggering.

Signed-off-by: Christoph Lameter <clameter@sgi.com>
Cc: Peter Williams <pwil3058@bigpond.net.au>
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Cc: Christoph Lameter <clameter@sgi.com>
Cc: "Siddha, Suresh B" <suresh.b.siddha@intel.com>
Cc: "Chen, Kenneth W" <kenneth.w.chen@intel.com>
Acked-by: Ingo Molnar <mingo@elte.hu>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] sched: avoid taking rq lock in wake_priority_sleeper
Christoph Lameter [Sun, 10 Dec 2006 10:20:13 +0000 (02:20 -0800)]
[PATCH] sched: avoid taking rq lock in wake_priority_sleeper

Avoid taking the request queue lock in wake_priority_sleeper if there are no
running processes.

Signed-off-by: Christoph Lameter <clameter@sgi.com>
Cc: Peter Williams <pwil3058@bigpond.net.au>
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Cc: Christoph Lameter <clameter@sgi.com>
Cc: "Siddha, Suresh B" <suresh.b.siddha@intel.com>
Cc: "Chen, Kenneth W" <kenneth.w.chen@intel.com>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Acked-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] sched domain: increase the SMT busy rebalance interval
Siddha, Suresh B [Sun, 10 Dec 2006 10:20:12 +0000 (02:20 -0800)]
[PATCH] sched domain: increase the SMT busy rebalance interval

With SMT, if the logical processor is busy, load balance happens for every
8msec(min)-16msec(max).  There is no need to do this often, as this is just
for fairness(to maintain uniform runqueue lengths) and default time slice
anyhow is 100msec.

Appended patch increases this interval to 64msec(min)-128msec(max) when the
logical processor is busy.

Signed-off-by: Suresh Siddha <suresh.b.siddha@intel.com>
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Acked-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] move_task_off_dead_cpu() should be called with disabled ints
Kirill Korotaev [Sun, 10 Dec 2006 10:20:11 +0000 (02:20 -0800)]
[PATCH] move_task_off_dead_cpu() should be called with disabled ints

move_task_off_dead_cpu() requires interrupts to be disabled, while
migrate_dead() calls it with enabled interrupts.  Added appropriate
comments to functions and added BUG_ON(!irqs_disabled()) into
double_rq_lock() and double_lock_balance() which are the origin sources of
such bugs.

Signed-off-by: Kirill Korotaev <dev@openvz.org>
Acked-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] ched domain: move sched group allocations to percpu area
Siddha, Suresh B [Sun, 10 Dec 2006 10:20:07 +0000 (02:20 -0800)]
[PATCH] ched domain: move sched group allocations to percpu area

Move the sched group allocations to percpu area.  This will minimize cross
node memory references and also cleans up the sched groups allocation for
allnodes sched domain.

Signed-off-by: Suresh Siddha <suresh.b.siddha@intel.com>
Acked-by: Ingo Molnar <mingo@elte.hu>
Acked-by: Christoph Lameter <clameter@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>