]> git.karo-electronics.de Git - linux-beck.git/log
linux-beck.git
14 years agosh: free-without-alloc fix for sh_mobile_lcdcfb
Magnus Damm [Wed, 13 Oct 2010 07:17:45 +0000 (07:17 +0000)]
sh: free-without-alloc fix for sh_mobile_lcdcfb

Without this fix the LCDC driver will try to free
framebuffer memory even though the allocation failed.

Signed-off-by: Magnus Damm <damm@opensource.se>
Signed-off-by: Paul Mundt <lethal@linux-sh.org>
14 years agosh: perf: Set up perf_max_events.
Paul Mundt [Wed, 13 Oct 2010 08:52:14 +0000 (17:52 +0900)]
sh: perf: Set up perf_max_events.

Presently this is uninitialized in the architecture code, so it's
artificlally capped to the default initialization value. Set it up at
registration time.

Signed-off-by: Paul Mundt <lethal@linux-sh.org>
14 years agosh: perf: Support SH-X3 hardware counters.
Paul Mundt [Tue, 12 Oct 2010 22:17:03 +0000 (07:17 +0900)]
sh: perf: Support SH-X3 hardware counters.

The PMCAT location has conveniently moved on newer SH-X3 parts, special
case this for now with a note. This will probably want to be redone in a
less visually offensive way when/if more information becomes available.

Signed-off-by: Paul Mundt <lethal@linux-sh.org>
14 years agosh: wire up perf alignment and emulation faults.
Paul Mundt [Tue, 12 Oct 2010 21:55:26 +0000 (06:55 +0900)]
sh: wire up perf alignment and emulation faults.

This plugs in the alignment and emulation fault reporting for perf sw
events.

Signed-off-by: Paul Mundt <lethal@linux-sh.org>
14 years agosh: boards/mach-x3proto: gpio: fix error handling code
Kulikov Vasiliy [Sun, 10 Oct 2010 17:28:27 +0000 (17:28 +0000)]
sh: boards/mach-x3proto: gpio: fix error handling code

Checks for (irq < 0) and (ilsel < 0) didn't make sense since they were
unsigned.  Made them signed.

Signed-off-by: Vasiliy Kulikov <segooon@gmail.com>
Signed-off-by: Paul Mundt <lethal@linux-sh.org>
14 years agosh: Fix up the SH-3 build.
Paul Mundt [Wed, 6 Oct 2010 17:57:39 +0000 (02:57 +0900)]
sh: Fix up the SH-3 build.

SH-3 lacks an MMUCR_TI definition for global TLB flushes. As SH-3 parts
lack a split TLB, the same global flush behaviour is accomplished
through the flush bit, which just happens to be the same as on SH-4.

This fixes up the build for all SH-3 MMU parts.

Signed-off-by: Paul Mundt <lethal@linux-sh.org>
14 years agosh: Fix address calculation of Initrd
Nobuhiro Iwamatsu [Tue, 5 Oct 2010 08:16:08 +0000 (08:16 +0000)]
sh: Fix address calculation of Initrd

Signed-off-by: Nobuhiro Iwamatsu <nobuhiro.iwamatsu.yj@renesas.com>
Signed-off-by: Paul Mundt <lethal@linux-sh.org>
14 years agosh: fix uninitialized spinlock
Akinobu Mita [Tue, 5 Oct 2010 15:54:00 +0000 (15:54 +0000)]
sh: fix uninitialized spinlock

The spinlock in traps_64.c is used without initialization.
This fixes it by declaring DEFINE_SPINLOCK() and makes the spinlock static
variable.

Signed-off-by: Akinobu Mita <akinobu.mita@gmail.com>
Cc: Paul Mundt <lethal@linux-sh.org>
Cc: linux-sh@vger.kernel.org
Signed-off-by: Paul Mundt <lethal@linux-sh.org>
14 years agosh: intc: Fix build with IRQ balancing disabled.
Paul Mundt [Wed, 6 Oct 2010 07:44:10 +0000 (16:44 +0900)]
sh: intc: Fix build with IRQ balancing disabled.

The balancing stubs obviously need to be static inline..

Signed-off-by: Paul Mundt <lethal@linux-sh.org>
14 years agosh: intc: Add missing files.
Paul Mundt [Wed, 6 Oct 2010 06:38:16 +0000 (15:38 +0900)]
sh: intc: Add missing files.

The Kconfig and Makefile were overlooked, add those in now to improve
odds of building.

Signed-off-by: Paul Mundt <lethal@linux-sh.org>
14 years agosh: Wire up INTC subgroup splitting for SH7786 SCIF1.
Paul Mundt [Tue, 5 Oct 2010 14:26:42 +0000 (23:26 +0900)]
sh: Wire up INTC subgroup splitting for SH7786 SCIF1.

SH7786 is the big user for subgroup splitting, mostly for the PCIe block,
but those will follow later. For now we simply split up SCIF1, as used by
the serial console on SDK7786 and others.

Signed-off-by: Paul Mundt <lethal@linux-sh.org>
14 years agosh: intc: Split up the INTC code.
Paul Mundt [Tue, 5 Oct 2010 13:10:30 +0000 (22:10 +0900)]
sh: intc: Split up the INTC code.

This splits up the sh intc core in to something more vaguely resembling
a subsystem. Most of the functionality was alread fairly well
compartmentalized, and there were only a handful of interdependencies
that needed to be resolved in the process.

This also serves as future-proofing for the genirq and sparseirq rework,
which will make some of the split out functionality wholly generic,
allowing things to be killed off in place with minimal migration pain.

Signed-off-by: Paul Mundt <lethal@linux-sh.org>
14 years agosh: intc: Handle early lookups of subgroup IRQs.
Paul Mundt [Tue, 5 Oct 2010 09:13:23 +0000 (18:13 +0900)]
sh: intc: Handle early lookups of subgroup IRQs.

If lookups happen while the radix node still points to a subgroup
mapping, an IRQ hasn't yet been made available for the specified id, so
error out accordingly. Once the slot is replaced with an IRQ mapping and
the tag is discarded, lookup can commence as normal.

Signed-off-by: Paul Mundt <lethal@linux-sh.org>
14 years agosh: intc: Support virtual mappings for IRQ subgroups.
Paul Mundt [Mon, 4 Oct 2010 19:47:03 +0000 (04:47 +0900)]
sh: intc: Support virtual mappings for IRQ subgroups.

Many interrupts that share a single mask source but are on different
hardware vectors will have an associated register tied to an INTEVT that
denotes the precise cause for the interrupt exception being triggered.

This introduces the concept of IRQ subgroups in the intc core, where
a virtual IRQ map is constructed for each of the pre-defined cause bits,
and a higher level chained handler takes control of the parent INTEVT.
This enables CPUs with heavily muxed IRQ vectors (especially across
disjoint blocks) to break things out in to a series of managed chained
handlers while being able to dynamically lookup and adopt the IRQs
created for them.

This is largely an opt-in interface, requiring CPUs to manually submit
IRQs for subgroup splitting, in addition to providing identifiers in
their enum maps that can be used for lazy lookup via the radix tree.

Signed-off-by: Paul Mundt <lethal@linux-sh.org>
14 years agosh: intc: Implement reverse mapping for IRQs to per-controller IDs.
Paul Mundt [Thu, 23 Sep 2010 11:09:38 +0000 (20:09 +0900)]
sh: intc: Implement reverse mapping for IRQs to per-controller IDs.

This implements a scheme roughly analogous to the PowerPC virtual to
hardware IRQ mapping, which we use for IRQ to per-controller ID mapping.
This makes it possible for drivers to use the IDs directly for lookup
instead of hardcoding the vector.

The main motivation for this work is as a building block for dynamically
allocating virtual IRQs for demuxing INTC events sharing a single INTEVT
in addition to a common masking source.

Signed-off-by: Paul Mundt <lethal@linux-sh.org>
14 years agosh: pfc: Fix up BUG() triggered by gpiolib debugfs lookups.
Paul Mundt [Sun, 3 Oct 2010 20:15:20 +0000 (05:15 +0900)]
sh: pfc: Fix up BUG() triggered by gpiolib debugfs lookups.

The gpiolib debugfs entry takes a hammer approach and iterates over all
of the potential GPIOs, regardless of their type. The SH PFC code on the
other hand contains a variable mismash of input/output/function types
spread out sparsely, leading to situations where the debug code can
trigger an out of range enum for the type. Since we already have an error
path for out of range enums, we can just hand that up to the higher level
instead of the current BUG() behaviour.

Signed-off-by: Paul Mundt <lethal@linux-sh.org>
14 years agosh: Drop __initdata for SH-X3 pinmux tables.
Paul Mundt [Sun, 3 Oct 2010 20:00:19 +0000 (05:00 +0900)]
sh: Drop __initdata for SH-X3 pinmux tables.

The gpio sysfs support needs to get at these later, so drop the
__initdata annotations.

Signed-off-by: Paul Mundt <lethal@linux-sh.org>
14 years agosh: pfc: support pinmux deregistration.
Paul Mundt [Sun, 3 Oct 2010 18:54:56 +0000 (03:54 +0900)]
sh: pfc: support pinmux deregistration.

Presently the pinmux code is a one-way thing, but there's nothing
preventing an unregistration if no one has grabbed any of the pins.
This will permit us to save a bit of memory on systems that require pin
demux for certain peripherals in the case where registration of those
peripherals fails, or they are otherwise not attached to the system.

Signed-off-by: Paul Mundt <lethal@linux-sh.org>
14 years agosh: mach-x3proto: Improve ILSEL debugging.
Paul Mundt [Sun, 3 Oct 2010 18:07:49 +0000 (03:07 +0900)]
sh: mach-x3proto: Improve ILSEL debugging.

At the moment ILSEL blows up with a BUG when aliased sets are handed in,
but as the enable call is able to hand back errors we opt for that path
instead. None of the ILSEL peripherals are vital to the board's
operation, so trapping a BUG is a bit excessive.

Signed-off-by: Paul Mundt <lethal@linux-sh.org>
14 years agosh: mach-x3proto: gpio-keys support.
Paul Mundt [Sun, 3 Oct 2010 18:00:49 +0000 (03:00 +0900)]
sh: mach-x3proto: gpio-keys support.

This adds gpio-keys mappings for the button matrix on the baseboard,
now that we have support for the pin controller.

Signed-off-by: Paul Mundt <lethal@linux-sh.org>
14 years agosh: mach-x3proto: Support for baseboard GPIOs.
Paul Mundt [Sun, 3 Oct 2010 17:59:29 +0000 (02:59 +0900)]
sh: mach-x3proto: Support for baseboard GPIOs.

This adds trivial support for the GPIOs implemented through the baseboard
CPLD, used for driving the button matrix.

Signed-off-by: Paul Mundt <lethal@linux-sh.org>
14 years agosh: mach-x3proto: Move the ilsel header to a better place.
Paul Mundt [Sun, 3 Oct 2010 17:50:32 +0000 (02:50 +0900)]
sh: mach-x3proto: Move the ilsel header to a better place.

We'll be adding more headers for this board, so move this over to its own
directory.

Signed-off-by: Paul Mundt <lethal@linux-sh.org>
14 years agosh: Allow GPIO chips to register IRQ mappings.
Paul Mundt [Sat, 2 Oct 2010 18:55:39 +0000 (03:55 +0900)]
sh: Allow GPIO chips to register IRQ mappings.

As non-PFC chips are added that may support IRQs, pass through to the
generic helper instead of triggering the WARN_ON().

Signed-off-by: Paul Mundt <lethal@linux-sh.org>
14 years agosh: Support early IRQ vector map reservation for delayed controllers.
Paul Mundt [Sat, 2 Oct 2010 13:02:07 +0000 (22:02 +0900)]
sh: Support early IRQ vector map reservation for delayed controllers.

Some controllers will need to be initialized lazily due to pinmux
constraints, while others may simply have no need to be brought online if
there are no backing devices for them attached. In this case it's still
necessary to be able to reserve their hardware vector map before dynamic
IRQs get a hold of them.

Signed-off-by: Paul Mundt <lethal@linux-sh.org>
14 years agosh: Handle pinmux for SH-X3 proto IRQ/IRL modes.
Paul Mundt [Sat, 2 Oct 2010 10:43:40 +0000 (19:43 +0900)]
sh: Handle pinmux for SH-X3 proto IRQ/IRL modes.

The SH-X3 proto CPU has all of the external IRQ and IRL pins muxed, make
sure that we're able to grab them before attempting to register their
respective IRQ controllers.

Signed-off-by: Paul Mundt <lethal@linux-sh.org>
14 years agosh: pinmux support for SH-X3 proto CPUs.
Paul Mundt [Sat, 2 Oct 2010 10:20:51 +0000 (19:20 +0900)]
sh: pinmux support for SH-X3 proto CPUs.

This adds in support for GPIO/pinmux on the SH-X3 proto CPUs. This will
subsequently be used by the x3proto board.

Signed-off-by: Paul Mundt <lethal@linux-sh.org>
14 years agosh: Support IRQ balancing for SH-X3 proto cores, too.
Paul Mundt [Fri, 1 Oct 2010 16:04:30 +0000 (01:04 +0900)]
sh: Support IRQ balancing for SH-X3 proto cores, too.

This adds in hardware IRQ auto-distribution support for SH-X3 proto CPUs,
following the SH7786 support.

Signed-off-by: Paul Mundt <lethal@linux-sh.org>
14 years agosh: Support userimask for all SH-X3 interrupt controllers.
Paul Mundt [Fri, 1 Oct 2010 15:43:43 +0000 (00:43 +0900)]
sh: Support userimask for all SH-X3 interrupt controllers.

This shuffles some of the shared bits out of the 7786 code and in to a
shared SH-X3 support file. Presently just for userimask, but also a good
place for the IRQ balancing wrappers.

Signed-off-by: Paul Mundt <lethal@linux-sh.org>
14 years agosh: Update SH-X3 subtype for clkdev lookups.
Paul Mundt [Fri, 1 Oct 2010 14:49:56 +0000 (23:49 +0900)]
sh: Update SH-X3 subtype for clkdev lookups.

Rewrite the SH-X3 proto CPU clock framework for clkdev.

Signed-off-by: Paul Mundt <lethal@linux-sh.org>
14 years agosh: boot kernel with SR.BL set
Magnus Damm [Fri, 24 Sep 2010 09:05:38 +0000 (09:05 +0000)]
sh: boot kernel with SR.BL set

Update the SH kernel to keep SR.BL set until the VBR
register has been initialized. Useful to allow boot
of the kernel even though exceptions are pending.

Without this patch there is a window of time when
exceptions such as NMI are enabled but no exception
handlers are installed.

This patch modifies both the zImage loader and the
actual kernel to boot with BL=1, but the zImage
loader is modfied in such a way that the init_sr
value is unchanged to not break the zImage loader
provided by kexec.

Tested on sh7724 Ecovec and on the SH4AL-DSP core
included in sh7372.

Signed-off-by: Magnus Damm <damm@opensource.se>
Signed-off-by: Paul Mundt <lethal@linux-sh.org>
14 years agosh: fix a kfree address in clkdev code
Guennadi Liakhovetski [Fri, 17 Sep 2010 09:25:10 +0000 (09:25 +0000)]
sh: fix a kfree address in clkdev code

kfree() in clkdev_drop() function should actually be called with an address of
a struct clk_lookup_alloc object, and not struct clk_lookup, as presently done.
This just happens to work, because "struct clk_lookup cl" is the first
member in struct clk_lookup_alloc.

Signed-off-by: Guennadi Liakhovetski <g.liakhovetski@gmx.de>
Signed-off-by: Paul Mundt <lethal@linux-sh.org>
14 years agosh: provide generic arch_debugfs_dir.
Paul Mundt [Thu, 23 Sep 2010 19:04:26 +0000 (04:04 +0900)]
sh: provide generic arch_debugfs_dir.

While sh previously had its own debugfs root, there now exists a
common arch_debugfs_dir prototype, so we switch everything over to
that.  Presumably once more architectures start making use of this
we'll be able to just kill off the stub kdebugfs wrapper.

Signed-off-by: Paul Mundt <lethal@linux-sh.org>
14 years agosh: change to new flag variable
matt mooney [Thu, 23 Sep 2010 06:51:21 +0000 (06:51 +0000)]
sh: change to new flag variable

Replace EXTRA_CFLAGS with ccflags-y.

Signed-off-by: matt mooney <mfm@muteddisk.com>
Signed-off-by: Paul Mundt <lethal@linux-sh.org>
14 years agosh: pci: Use a generic raw spinlock for PCI config access locking.
Paul Mundt [Mon, 20 Sep 2010 09:56:13 +0000 (18:56 +0900)]
sh: pci: Use a generic raw spinlock for PCI config access locking.

This copies the pci_config_lock idea from x86 over, allowing us to kill
off a couple of existing private locks. At the same time, these need to
be converted to raw spinlocks for -rt kernels, so we make that change at
the same time. This should make it easier for future parts to get the
locking right instead of inevitable ending up with lock type mismatches.

Signed-off-by: Paul Mundt <lethal@linux-sh.org>
14 years agosh: pci: clock framework support for SH7786 PCIe.
Paul Mundt [Mon, 20 Sep 2010 09:45:11 +0000 (18:45 +0900)]
sh: pci: clock framework support for SH7786 PCIe.

This gets each port handling its MSTP bit, as well as moving the PHY
clock management in to the clock framework.

Signed-off-by: Paul Mundt <lethal@linux-sh.org>
14 years agosh: pci: Use I/O accessors consistently in SH7786 PCIe init code.
Paul Mundt [Mon, 20 Sep 2010 08:10:02 +0000 (17:10 +0900)]
sh: pci: Use I/O accessors consistently in SH7786 PCIe init code.

Some of the existing code is flipping between __raw_xxx() and
pci_{read,write}_reg(). As the latter are just wrappers for the former,
flip over to using them consistently.

Signed-off-by: Paul Mundt <lethal@linux-sh.org>
14 years agosh: pci: Support ports with disabled links on SH7786 PCIe.
Paul Mundt [Mon, 20 Sep 2010 07:12:58 +0000 (16:12 +0900)]
sh: pci: Support ports with disabled links on SH7786 PCIe.

Presently we error out if a link is disabled and simply drop the port
registration outright. This follows the PPC changes and simply reports on
the link state on boot, leaving the port registered, in order to more
easily deal with hotplug on future parts.

Signed-off-by: Paul Mundt <lethal@linux-sh.org>
14 years agosh: pci: Discard initial PCICONF4/5 settings for SH7786 PCIe.
Paul Mundt [Mon, 20 Sep 2010 07:00:42 +0000 (16:00 +0900)]
sh: pci: Discard initial PCICONF4/5 settings for SH7786 PCIe.

These settings are properly propagated by the hardware already, so
there's no need to bother with them manually.

Signed-off-by: Paul Mundt <lethal@linux-sh.org>
14 years agosh: pci: Support root complex config accesses on SH7786 PCIe.
Paul Mundt [Mon, 20 Sep 2010 06:39:54 +0000 (15:39 +0900)]
sh: pci: Support root complex config accesses on SH7786 PCIe.

The SH7786 PCIe is presently unable to enumerate itself in root complex
mode, and has no visibility through either type 0 or type 1 accesses,
despite having a mostly sensible extended config space for each port.
Attempts to generate type 0 or type 1 config cycles result in completer
aborts, so we're ultimately forced to use SuperHyway transactions
instead.

As each port has a single port <-> device mapping that resolves for any
PCI_SLOT definition, we simply hijack devfn 0 for the SuperHyway
transaction and bump up the devfn limit.

With enumeration of the root complex now possible, we also need to insert
an early fixup to hide the BARs from the kernel. With all of that done,
it's now possible to use the pcieport services with all of the PCIe
ports, which is the first step to power management support.

Signed-off-by: Paul Mundt <lethal@linux-sh.org>
14 years agosh: pci: Move Renesas PCI IDs to a better place.
Paul Mundt [Mon, 20 Sep 2010 06:37:25 +0000 (15:37 +0900)]
sh: pci: Move Renesas PCI IDs to a better place.

Previously these IDs were only used by one driver, so there was not much
need for having them generically defined. Now that this will no longer
hold true, move them over.

Signed-off-by: Paul Mundt <lethal@linux-sh.org>
14 years agosh: pci: Give SH7786 PHY some time to settle.
Paul Mundt [Sun, 19 Sep 2010 04:57:51 +0000 (13:57 +0900)]
sh: pci: Give SH7786 PHY some time to settle.

The spec suggests waiting up to 500ms for the PHY to settle before
testing link state, but practice shows that 100ms is sufficient (this is
the delay value we also use on the other SH-4A PCI controllers, too).
This makes device detection much more reliable, although in the future it
should be a bit faster to simply serialize with a TLP IRQ.

Signed-off-by: Paul Mundt <lethal@linux-sh.org>
14 years agosh: pci: Toggle configuration accesses on SH7786.
Paul Mundt [Sun, 19 Sep 2010 04:54:50 +0000 (13:54 +0900)]
sh: pci: Toggle configuration accesses on SH7786.

After configuration accesses have been completed deassert the
configuration access enable cleanly.

Signed-off-by: Paul Mundt <lethal@linux-sh.org>
14 years agosh: pci: Use generic pci_enable_resources() for pcibios_enable_device().
Paul Mundt [Sun, 19 Sep 2010 04:51:15 +0000 (13:51 +0900)]
sh: pci: Use generic pci_enable_resources() for pcibios_enable_device().

Signed-off-by: Paul Mundt <lethal@linux-sh.org>
14 years agousb: Fix up r8a66597-hcd section mismatches.
Paul Mundt [Thu, 16 Sep 2010 08:16:31 +0000 (17:16 +0900)]
usb: Fix up r8a66597-hcd section mismatches.

The _remove() routine is flagged __init_or_module, despite only being
used in a __devexit context. As the rest of the driver is already
balanced out with __devinit, switch to __devexit and __devexit_p()
wrapping.

Signed-off-by: Paul Mundt <lethal@linux-sh.org>
14 years agomfd: Fix up section mismatches in SH SDHI.
Paul Mundt [Thu, 16 Sep 2010 08:14:40 +0000 (17:14 +0900)]
mfd: Fix up section mismatches in SH SDHI.

The current probe/remove definitions are split between __init and
__devexit, make them consistent by switching to __devinit.

Signed-off-by: Paul Mundt <lethal@linux-sh.org>
14 years agosh: kill big kernel lock
Arnd Bergmann [Tue, 14 Sep 2010 19:35:00 +0000 (19:35 +0000)]
sh: kill big kernel lock

The only BKL user in arch/sh protects a single bit,
so we can trivially replace it with test_and_set_bit.

Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Cc: Paul Mundt <lethal@linux-sh.org>
Cc: linux-sh@vger.kernel.org
Signed-off-by: Paul Mundt <lethal@linux-sh.org>
14 years agoRevert "sh: ecovec24: modify tsc2007 platform settings"
Paul Mundt [Tue, 14 Sep 2010 09:26:38 +0000 (18:26 +0900)]
Revert "sh: ecovec24: modify tsc2007 platform settings"

According to Morimoto-san, this is no longer needed. Revert it.

This reverts commit e0009b0a44f28227571d8cddebc5ccdae86027a6.

Signed-off-by: Paul Mundt <lethal@linux-sh.org>
14 years agosh: Provide a non-multiplexed sys_recvmmsg path.
Paul Mundt [Tue, 14 Sep 2010 08:43:11 +0000 (17:43 +0900)]
sh: Provide a non-multiplexed sys_recvmmsg path.

Now that the rest of the socket calls are provided through their own
paths, do the same for sys_recvmmsg. It's unlikely we'll ever be able to
kill off the socketcall path, but this at least permits userspace to
gradually begin migrating.

Signed-off-by: Paul Mundt <lethal@linux-sh.org>
14 years agosh: Add syscall entries for non multiplexed socket calls
Carmelo AMOROSO [Wed, 1 Sep 2010 01:49:29 +0000 (01:49 +0000)]
sh: Add syscall entries for non multiplexed socket calls

Linux kernel already has socket syscalls that can be invoked
without the multiplexing sys_socketcall wrapper.
C library wrappers are ready to use them directly. It needs just
to define the missing syscall numbers and provide the related entries
into the syscalls table, like sh64 aleady does.

Signed-off-by: Francesco Rundo <francesco.rundo@st.com>
Signed-off-by: Carmelo Amoroso <carmelo.amoroso@st.com>
Signed-off-by: Paul Mundt <lethal@linux-sh.org>
14 years agosh: ecovec24: modify tsc2007 platform settings
Kuninori Morimoto [Fri, 27 Aug 2010 09:49:30 +0000 (09:49 +0000)]
sh: ecovec24: modify tsc2007 platform settings

This patch modify x_plate_ohms to correct value for tsc2007 panel,
and removed un-necessary ts_get_pendown_state()

Signed-off-by: Kuninori Morimoto <kuninori.morimoto.gx@renesas.com>
Signed-off-by: Paul Mundt <lethal@linux-sh.org>
14 years agosh: Set CONFIG_SYSFS_DEPRECATED_V2=n
Matt Fleming [Sat, 21 Aug 2010 14:49:38 +0000 (14:49 +0000)]
sh: Set CONFIG_SYSFS_DEPRECATED_V2=n

As the help for the config option suggests, this option really shouldn't
be set by default for any recent distribution as it changes the layout
of sysfs. I spotted this while running debian when udev got very
confused by the sysfs layout and failed to create some device nodes.

Signed-off-by: Matt Fleming <matt@console-pimps.org>
Signed-off-by: Paul Mundt <lethal@linux-sh.org>
14 years agoMerge branch 'sh/pci-express-integration'
Paul Mundt [Tue, 7 Sep 2010 08:56:27 +0000 (17:56 +0900)]
Merge branch 'sh/pci-express-integration'

14 years agosh: Hook up 3rd memory window for all SH7786 PCIe channels.
Paul Mundt [Tue, 7 Sep 2010 08:07:05 +0000 (17:07 +0900)]
sh: Hook up 3rd memory window for all SH7786 PCIe channels.

Now that the resource assignment issues are resolved, we can finally wire
up the small third memory window -- in the future we may reclaim this for
MSI.

Signed-off-by: Paul Mundt <lethal@linux-sh.org>
14 years agosh: Properly wire up channel 2's I/O window on SH7786 PCIe.
Paul Mundt [Tue, 7 Sep 2010 08:05:08 +0000 (17:05 +0900)]
sh: Properly wire up channel 2's I/O window on SH7786 PCIe.

An IORESOURCE_IO was missing here, which meant that we weren't properly
establishing the I/O window for this particular slot. With this
corrected, cards with I/O BARs have them actually assigned and
accessible.

Signed-off-by: Paul Mundt <lethal@linux-sh.org>
14 years agosh: Ignore 32-bit windows in 29-bit mode for SH7786 PCIe.
Paul Mundt [Tue, 7 Sep 2010 08:03:10 +0000 (17:03 +0900)]
sh: Ignore 32-bit windows in 29-bit mode for SH7786 PCIe.

Certain memory windows are only available for 32-bit space, so skip over
these in 29-bit mode. This will severely restrict the amount of memory
that can be mapped, but since a boot loader bug makes booting in 29-bit
mode close to impossible anyways, everything is ok.

Signed-off-by: Paul Mundt <lethal@linux-sh.org>
14 years agosh: Establish a SuperHyway<->PCIe window mapping on SH7786 PCIe.
Paul Mundt [Tue, 7 Sep 2010 07:12:26 +0000 (16:12 +0900)]
sh: Establish a SuperHyway<->PCIe window mapping on SH7786 PCIe.

This bumps up the low address to match the physical memory windows for
SHway<->PCIe transfers. The previous implementation was banking on a 1:1
virt<->phys SHway mapping, which doesn't apply here.

Signed-off-by: Paul Mundt <lethal@linux-sh.org>
14 years agosh: Make SH7786 PCIe port reset logic more aggressive.
Paul Mundt [Tue, 7 Sep 2010 07:11:04 +0000 (16:11 +0900)]
sh: Make SH7786 PCIe port reset logic more aggressive.

This attempts a more complete port reset, building on top of the existing
approach.

Signed-off-by: Paul Mundt <lethal@linux-sh.org>
14 years agosh: Additional register definitions for SH7786 PCIe.
Matt Fleming [Tue, 7 Sep 2010 07:09:14 +0000 (16:09 +0900)]
sh: Additional register definitions for SH7786 PCIe.

Signed-off-by: Matt Fleming <matt@console-pimps.org>
Signed-off-by: Paul Mundt <lethal@linux-sh.org>
14 years agoMerge branch 'master' of git://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux-2.6
Paul Mundt [Fri, 20 Aug 2010 11:52:23 +0000 (20:52 +0900)]
Merge branch 'master' of git://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux-2.6

Conflicts:
arch/sh/kernel/process_32.c

Signed-off-by: Paul Mundt <lethal@linux-sh.org>
14 years agoMerge branch 'sh/pci-express-integration'
Paul Mundt [Fri, 20 Aug 2010 11:39:22 +0000 (20:39 +0900)]
Merge branch 'sh/pci-express-integration'

14 years agosh: sh2007/sh7757lcr defconfig reduction.
Paul Mundt [Fri, 20 Aug 2010 11:38:24 +0000 (20:38 +0900)]
sh: sh2007/sh7757lcr defconfig reduction.

These were newly added while the defconfig reduction work was ongoing,
strip them down now.

Signed-off-by: Paul Mundt <lethal@linux-sh.org>
14 years agosh: Relax devfn constraints for SH7786 PCIe.
Paul Mundt [Fri, 20 Aug 2010 11:26:41 +0000 (20:26 +0900)]
sh: Relax devfn constraints for SH7786 PCIe.

SH7786 PCIe has 1 slot per port, but no specific restriction on function.
Relax the devfn restriction and look to the slot number instead when
configured as a root complex.

Signed-off-by: Paul Mundt <lethal@linux-sh.org>
14 years agosh: reinstate clock framework rate rounding.
Paul Mundt [Fri, 20 Aug 2010 10:10:38 +0000 (19:10 +0900)]
sh: reinstate clock framework rate rounding.

This was killed off by a simplification patch previously that failed to
take the cpufreq use case in to account, so reinstate the old bounding
logic. The lowest rate bounding on the other hand was broken in that it
never actually got assigned a rate and the best fit rate was instead just
getting lucky based on the ordering of the rate table, fix this up so the
code actually does what it was intended to do originally.

Signed-off-by: Paul Mundt <lethal@linux-sh.org>
14 years agosh: Fix up SH7786 PCIe PHY initialization.
Paul Mundt [Fri, 20 Aug 2010 07:04:59 +0000 (16:04 +0900)]
sh: Fix up SH7786 PCIe PHY initialization.

This brings the clocking and register setting in line with the somewhat
factually ambiguous specification.

Signed-off-by: Paul Mundt <lethal@linux-sh.org>
14 years agosh: Support type 1 accesses for SH7786 PCI.
Paul Mundt [Fri, 20 Aug 2010 06:59:40 +0000 (15:59 +0900)]
sh: Support type 1 accesses for SH7786 PCI.

This enables support for type 1 config space accesses on the SH7786
PCI controller. At the same time, add in some extra sanity checks for
controller asserted errors.

Signed-off-by: Paul Mundt <lethal@linux-sh.org>
14 years agoMerge branch 'bugfixes' of git://git.linux-nfs.org/projects/trondmy/nfs-2.6
Linus Torvalds [Wed, 18 Aug 2010 22:45:23 +0000 (15:45 -0700)]
Merge branch 'bugfixes' of git://git.linux-nfs.org/projects/trondmy/nfs-2.6

* 'bugfixes' of git://git.linux-nfs.org/projects/trondmy/nfs-2.6:
  NFS: Fix an Oops in the NFSv4 atomic open code
  NFS: Fix the selection of security flavours in Kconfig
  NFS: fix the return value of nfs_file_fsync()
  rpcrdma: Fix SQ size calculation when memreg is FRMR
  xprtrdma: Do not truncate iova_start values in frmr registrations.
  nfs: Remove redundant NULL check upon kfree()
  nfs: Add "lookupcache" to displayed mount options
  NFS: allow close-to-open cache semantics to apply to root of NFS filesystem
  SUNRPC: fix NFS client over TCP hangs due to packet loss (Bug 16494)

14 years agoMerge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/jikos/hid
Linus Torvalds [Wed, 18 Aug 2010 22:29:38 +0000 (15:29 -0700)]
Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/jikos/hid

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/jikos/hid:
  USB HID: Add ID for eGalax Multitouch used in JooJoo tablet
  HID: hiddev: fix memory corruption due to invalid intfdata
  HID: hiddev: protect against disconnect/NULL-dereference race
  HID: picolcd: correct ordering of framebuffer freeing
  HID: picolcd: testing the wrong variable

14 years agoMerge branch 'release' of git://git.kernel.org/pub/scm/linux/kernel/git/aegl/linux-2.6
Linus Torvalds [Wed, 18 Aug 2010 20:27:41 +0000 (13:27 -0700)]
Merge branch 'release' of git://git.kernel.org/pub/scm/linux/kernel/git/aegl/linux-2.6

* 'release' of git://git.kernel.org/pub/scm/linux/kernel/git/aegl/linux-2.6:
  [IA64] Fix build error: conflicting types for â€˜sys_execve’

14 years agoFix the declaration of sys_execve() in asm-generic/syscalls.h
David Howells [Wed, 18 Aug 2010 17:55:33 +0000 (18:55 +0100)]
Fix the declaration of sys_execve() in asm-generic/syscalls.h

Fix the declaration of sys_execve() in asm-generic/syscalls.h to have
various consts applied to its pointers.

Signed-off-by: David Howells <dhowells@redhat.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
14 years ago[IA64] Fix build error: conflicting types for ‘sys_execve’
Tony Luck [Wed, 18 Aug 2010 17:17:44 +0000 (10:17 -0700)]
[IA64] Fix build error: conflicting types for â€˜sys_execve’

arch/ia64/kernel/process.c:636: error: conflicting types for â€˜sys_execve’

commit d7627467b7a8dd6944885290a03a07ceb28c10eb
Make do_execve() take a const filename pointer

Missed the declaration of sys_execve in the ia64 asm/unistd.h (perhaps
because there is no reason for it to be there ... it might be a left over
from the COMPAT code?). Just delete the conflicting version.

Signed-off-by: Tony Luck <tony.luck@intel.com>
14 years agoMerge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs-2.6
Linus Torvalds [Wed, 18 Aug 2010 16:35:08 +0000 (09:35 -0700)]
Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs-2.6

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs-2.6:
  fs: brlock vfsmount_lock
  fs: scale files_lock
  lglock: introduce special lglock and brlock spin locks
  tty: fix fu_list abuse
  fs: cleanup files_lock locking
  fs: remove extra lookup in __lookup_hash
  fs: fs_struct rwlock to spinlock
  apparmor: use task path helpers
  fs: dentry allocation consolidation
  fs: fix do_lookup false negative
  mbcache: Limit the maximum number of cache entries
  hostfs ->follow_link() braino
  hostfs: dumb (and usually harmless) tpyo - strncpy instead of strlcpy
  remove SWRITE* I/O types
  kill BH_Ordered flag
  vfs: update ctime when changing the file's permission by setfacl
  cramfs: only unlock new inodes
  fix reiserfs_evict_inode end_writeback second call

14 years agommc: build fix: mmc_pm_notify is only available with CONFIG_PM=y
Uwe Kleine-König [Wed, 18 Aug 2010 16:25:38 +0000 (09:25 -0700)]
mmc: build fix: mmc_pm_notify is only available with CONFIG_PM=y

This fixes a build breakage introduced by commit 4c2ef25fe0b8 ("mmc: fix
all hangs related to mmc/sd card insert/removal during suspend/resume")

Cc: David Brownell <david-b@pacbell.net>
Cc: Alan Stern <stern@rowland.harvard.edu>
Cc: linux-mmc@vger.kernel.org
Cc: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Uwe Kleine-König <u.kleine-koenig@pengutronix.de>
Acked-by: Kukjin Kim <kgene.kim@samsung.com>
Acked-by: Maxim Levitsky <maximlevitsky@gmail.com>
Acked-by: Randy Dunlap <randy.dunlap@oracle.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
14 years agoMerge branch 'perf-fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Wed, 18 Aug 2010 16:32:13 +0000 (09:32 -0700)]
Merge branch 'perf-fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip

* 'perf-fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip:
  perf tools: Fix build on POSIX shells
  latencytop: Fix kconfig dependency warnings
  perf annotate tui: Fix exit and RIGHT keys handling
  tracing: Sanitize value returned from write(trace_marker, "...", len)
  tracing/events: Convert format output to seq_file
  tracing: Extend recordmcount to better support Blackfin mcount
  tracing: Fix ring_buffer_read_page reading out of page boundary
  tracing: Fix an unallocated memory access in function_graph

14 years agoMerge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tiwai/sound-2.6
Linus Torvalds [Wed, 18 Aug 2010 16:30:08 +0000 (09:30 -0700)]
Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tiwai/sound-2.6

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tiwai/sound-2.6:
  ALSA: emu10k1 - delay the PCM interrupts (add pcm_irq_delay parameter)
  ALSA: hda - Fix ALC680 base model capture
  ASoC: Remove DSP mode support for WM8776
  ALSA: hda - Add quirk for Dell Vostro 1220
  ALSA: riptide - Fix detection / load of firmware files

14 years agoMerge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/gerg/m68knommu
Linus Torvalds [Wed, 18 Aug 2010 16:27:10 +0000 (09:27 -0700)]
Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/gerg/m68knommu

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/gerg/m68knommu:
  m68knommu: include sched.h in ColdFire/SPI driver
  m68knommu: formatting of pointers in printk()
  m68knommu: arch/m68k/include/asm/ide.h fix for nommu

14 years agoMerge branch 'for-linus' of git://neil.brown.name/md
Linus Torvalds [Wed, 18 Aug 2010 16:26:42 +0000 (09:26 -0700)]
Merge branch 'for-linus' of git://neil.brown.name/md

* 'for-linus' of git://neil.brown.name/md:
  md raid-1/10 Fix bio_rw bit manipulations again
  md: provide appropriate return value for spare_active functions.
  md: Notify sysfs when RAID1/5/10 disk is In_sync.
  Update recovery_offset even when external metadata is used.

14 years agoMerge branch 'merge-devicetree' of git://git.secretlab.ca/git/linux-2.6
Linus Torvalds [Wed, 18 Aug 2010 16:26:17 +0000 (09:26 -0700)]
Merge branch 'merge-devicetree' of git://git.secretlab.ca/git/linux-2.6

* 'merge-devicetree' of git://git.secretlab.ca/git/linux-2.6:
  spi.h: missing kernel-doc notation, please fix
  of: fix missing headers for of_address_to_resource() in MTD and SysACE drivers
  of: Fix missing includes
  ata: update for of_device to platform_device replacement
  microblaze: Fix of: eliminate of_device->node and dev_archdata->{of,prom}_node
  microblaze: Fix of/address: Merge all of the bus translation code
  booting-without-of: Remove nonexistent chapters from TOC, fix numbering

14 years agoNFS: Fix an Oops in the NFSv4 atomic open code
Trond Myklebust [Wed, 18 Aug 2010 13:25:42 +0000 (09:25 -0400)]
NFS: Fix an Oops in the NFSv4 atomic open code

Adam Lackorzynski reports:

with 2.6.35.2 I'm getting this reproducible Oops:

[  110.825396] BUG: unable to handle kernel NULL pointer dereference at
(null)
[  110.828638] IP: [<ffffffff811247b7>] encode_attrs+0x1a/0x2a4
[  110.828638] PGD be89f067 PUD bf18f067 PMD 0
[  110.828638] Oops: 0000 [#1] SMP
[  110.828638] last sysfs file: /sys/class/net/lo/operstate
[  110.828638] CPU 2
[  110.828638] Modules linked in: rtc_cmos rtc_core rtc_lib amd64_edac_mod
i2c_amd756 edac_core i2c_core dm_mirror dm_region_hash dm_log dm_snapshot
sg sr_mod usb_storage ohci_hcd mptspi tg3 mptscsih mptbase usbcore nls_base
[last unloaded: scsi_wait_scan]
[  110.828638]
[  110.828638] Pid: 11264, comm: setchecksum Not tainted 2.6.35.2 #1
[  110.828638] RIP: 0010:[<ffffffff811247b7>]  [<ffffffff811247b7>]
encode_attrs+0x1a/0x2a4
[  110.828638] RSP: 0000:ffff88003bf5b878  EFLAGS: 00010296
[  110.828638] RAX: ffff8800bddb48a8 RBX: ffff88003bf5bb18 RCX:
0000000000000000
[  110.828638] RDX: ffff8800be258800 RSI: 0000000000000000 RDI:
ffff88003bf5b9f8
[  110.828638] RBP: 0000000000000000 R08: ffff8800bddb48a8 R09:
0000000000000004
[  110.828638] R10: 0000000000000003 R11: ffff8800be779000 R12:
ffff8800be258800
[  110.828638] R13: ffff88003bf5b9f8 R14: ffff88003bf5bb20 R15:
ffff8800be258800
[  110.828638] FS:  0000000000000000(0000) GS:ffff880041e00000(0063)
knlGS:00000000556bd6b0
[  110.828638] CS:  0010 DS: 002b ES: 002b CR0: 000000008005003b
[  110.828638] CR2: 0000000000000000 CR3: 00000000be8ef000 CR4:
00000000000006e0
[  110.828638] DR0: 0000000000000000 DR1: 0000000000000000 DR2:
0000000000000000
[  110.828638] DR3: 0000000000000000 DR6: 00000000ffff0ff0 DR7:
0000000000000400
[  110.828638] Process setchecksum (pid: 11264, threadinfo
ffff88003bf5a000, task ffff88003f232210)
[  110.828638] Stack:
[  110.828638]  0000000000000000 ffff8800bfbcf920 0000000000000000
0000000000000ffe
[  110.828638] <0> 0000000000000000 0000000000000000 0000000000000000
0000000000000000
[  110.828638] <0> 0000000000000000 0000000000000000 0000000000000000
0000000000000000
[  110.828638] Call Trace:
[  110.828638]  [<ffffffff81124c1f>] ? nfs4_xdr_enc_setattr+0x90/0xb4
[  110.828638]  [<ffffffff81371161>] ? call_transmit+0x1c3/0x24a
[  110.828638]  [<ffffffff813774d9>] ? __rpc_execute+0x78/0x22a
[  110.828638]  [<ffffffff81371a91>] ? rpc_run_task+0x21/0x2b
[  110.828638]  [<ffffffff81371b7e>] ? rpc_call_sync+0x3d/0x5d
[  110.828638]  [<ffffffff8111e284>] ? _nfs4_do_setattr+0x11b/0x147
[  110.828638]  [<ffffffff81109466>] ? nfs_init_locked+0x0/0x32
[  110.828638]  [<ffffffff810ac521>] ? ifind+0x4e/0x90
[  110.828638]  [<ffffffff8111e2fb>] ? nfs4_do_setattr+0x4b/0x6e
[  110.828638]  [<ffffffff8111e634>] ? nfs4_do_open+0x291/0x3a6
[  110.828638]  [<ffffffff8111ed81>] ? nfs4_open_revalidate+0x63/0x14a
[  110.828638]  [<ffffffff811056c4>] ? nfs_open_revalidate+0xd7/0x161
[  110.828638]  [<ffffffff810a2de4>] ? do_lookup+0x1a4/0x201
[  110.828638]  [<ffffffff810a4733>] ? link_path_walk+0x6a/0x9d5
[  110.828638]  [<ffffffff810a42b6>] ? do_last+0x17b/0x58e
[  110.828638]  [<ffffffff810a5fbe>] ? do_filp_open+0x1bd/0x56e
[  110.828638]  [<ffffffff811cd5e0>] ? _atomic_dec_and_lock+0x30/0x48
[  110.828638]  [<ffffffff810a9b1b>] ? dput+0x37/0x152
[  110.828638]  [<ffffffff810ae063>] ? alloc_fd+0x69/0x10a
[  110.828638]  [<ffffffff81099f39>] ? do_sys_open+0x56/0x100
[  110.828638]  [<ffffffff81027a22>] ? ia32_sysret+0x0/0x5
[  110.828638] Code: 83 f1 01 e8 f5 ca ff ff 48 83 c4 50 5b 5d 41 5c c3 41
57 41 56 41 55 49 89 fd 41 54 49 89 d4 55 48 89 f5 53 48 81 ec 18 01 00 00
<8b> 06 89 c2 83 e2 08 83 fa 01 19 db 83 e3 f8 83 c3 18 a8 01 8d
[  110.828638] RIP  [<ffffffff811247b7>] encode_attrs+0x1a/0x2a4
[  110.828638]  RSP <ffff88003bf5b878>
[  110.828638] CR2: 0000000000000000
[  112.840396] ---[ end trace 95282e83fd77358f ]---

We need to ensure that the O_EXCL flag is turned off if the user doesn't
set O_CREAT.

Cc: stable@kernel.org
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
14 years agoMerge branch 'fix/asoc' into for-linus
Takashi Iwai [Wed, 18 Aug 2010 13:22:18 +0000 (15:22 +0200)]
Merge branch 'fix/asoc' into for-linus

14 years agoMerge branch 'fix/hda' into for-linus
Takashi Iwai [Wed, 18 Aug 2010 13:22:15 +0000 (15:22 +0200)]
Merge branch 'fix/hda' into for-linus

14 years agoALSA: emu10k1 - delay the PCM interrupts (add pcm_irq_delay parameter)
Jaroslav Kysela [Wed, 18 Aug 2010 12:08:17 +0000 (14:08 +0200)]
ALSA: emu10k1 - delay the PCM interrupts (add pcm_irq_delay parameter)

With some hardware combinations, the PCM interrupts are acknowledged
before the period boundary from the emu10k1 chip. The midlevel PCM code
gets confused and the playback stream is interrupted.

It seems that the interrupt processing shift by 2 samples is enough
to fix this issue. This default value does not harm other,
non-affected hardware.

More information: Kernel bugzilla bug#16300

[A copmile warning fixed by tiwai]

Signed-off-by: Jaroslav Kysela <perex@perex.cz>
Cc: <stable@kernel.org>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
14 years agofs: brlock vfsmount_lock
Nick Piggin [Tue, 17 Aug 2010 18:37:39 +0000 (04:37 +1000)]
fs: brlock vfsmount_lock

fs: brlock vfsmount_lock

Use a brlock for the vfsmount lock. It must be taken for write whenever
modifying the mount hash or associated fields, and may be taken for read when
performing mount hash lookups.

A new lock is added for the mnt-id allocator, so it doesn't need to take
the heavy vfsmount write-lock.

The number of atomics should remain the same for fastpath rlock cases, though
code would be slightly slower due to per-cpu access. Scalability is not not be
much improved in common cases yet, due to other locks (ie. dcache_lock) getting
in the way. However path lookups crossing mountpoints should be one case where
scalability is improved (currently requiring the global lock).

The slowpath is slower due to use of brlock. On a 64 core, 64 socket, 32 node
Altix system (high latency to remote nodes), a simple umount microbenchmark
(mount --bind mnt mnt2 ; umount mnt2 loop 1000 times), before this patch it
took 6.8s, afterwards took 7.1s, about 5% slower.

Cc: Al Viro <viro@ZenIV.linux.org.uk>
Signed-off-by: Nick Piggin <npiggin@kernel.dk>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
14 years agofs: scale files_lock
Nick Piggin [Tue, 17 Aug 2010 18:37:38 +0000 (04:37 +1000)]
fs: scale files_lock

fs: scale files_lock

Improve scalability of files_lock by adding per-cpu, per-sb files lists,
protected with an lglock. The lglock provides fast access to the per-cpu lists
to add and remove files. It also provides a snapshot of all the per-cpu lists
(although this is very slow).

One difficulty with this approach is that a file can be removed from the list
by another CPU. We must track which per-cpu list the file is on with a new
variale in the file struct (packed into a hole on 64-bit archs). Scalability
could suffer if files are frequently removed from different cpu's list.

However loads with frequent removal of files imply short interval between
adding and removing the files, and the scheduler attempts to avoid moving
processes too far away. Also, even in the case of cross-CPU removal, the
hardware has much more opportunity to parallelise cacheline transfers with N
cachelines than with 1.

A worst-case test of 1 CPU allocating files subsequently being freed by N CPUs
degenerates to contending on a single lock, which is no worse than before. When
more than one CPU are allocating files, even if they are always freed by
different CPUs, there will be more parallelism than the single-lock case.

Testing results:

On a 2 socket, 8 core opteron, I measure the number of times the lock is taken
to remove the file, the number of times it is removed by the same CPU that
added it, and the number of times it is removed by the same node that added it.

Booting:    locks=  25049 cpu-hits=  23174 (92.5%) node-hits=  23945 (95.6%)
kbuild -j16 locks=2281913 cpu-hits=2208126 (96.8%) node-hits=2252674 (98.7%)
dbench 64   locks=4306582 cpu-hits=4287247 (99.6%) node-hits=4299527 (99.8%)

So a file is removed from the same CPU it was added by over 90% of the time.
It remains within the same node 95% of the time.

Tim Chen ran some numbers for a 64 thread Nehalem system performing a compile.

                throughput
2.6.34-rc2      24.5
+patch          24.9

                us      sys     idle    IO wait (in %)
2.6.34-rc2      51.25   28.25   17.25   3.25
+patch          53.75   18.5    19      8.75

So significantly less CPU time spent in kernel code, higher idle time and
slightly higher throughput.

Single threaded performance difference was within the noise of microbenchmarks.
That is not to say penalty does not exist, the code is larger and more memory
accesses required so it will be slightly slower.

Cc: linux-kernel@vger.kernel.org
Cc: Tim Chen <tim.c.chen@linux.intel.com>
Cc: Andi Kleen <ak@linux.intel.com>
Signed-off-by: Nick Piggin <npiggin@kernel.dk>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
14 years agolglock: introduce special lglock and brlock spin locks
Nick Piggin [Tue, 17 Aug 2010 18:37:37 +0000 (04:37 +1000)]
lglock: introduce special lglock and brlock spin locks

lglock: introduce special lglock and brlock spin locks

This patch introduces "local-global" locks (lglocks). These can be used to:

- Provide fast exclusive access to per-CPU data, with exclusive access to
  another CPU's data allowed but possibly subject to contention, and to provide
  very slow exclusive access to all per-CPU data.
- Or to provide very fast and scalable read serialisation, and to provide
  very slow exclusive serialisation of data (not necessarily per-CPU data).

Brlocks are also implemented as a short-hand notation for the latter use
case.

Thanks to Paul for local/global naming convention.

Cc: linux-kernel@vger.kernel.org
Cc: Al Viro <viro@ZenIV.linux.org.uk>
Cc: "Paul E. McKenney" <paulmck@linux.vnet.ibm.com>
Signed-off-by: Nick Piggin <npiggin@kernel.dk>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
14 years agotty: fix fu_list abuse
Nick Piggin [Tue, 17 Aug 2010 18:37:36 +0000 (04:37 +1000)]
tty: fix fu_list abuse

tty: fix fu_list abuse

tty code abuses fu_list, which causes a bug in remount,ro handling.

If a tty device node is opened on a filesystem, then the last link to the inode
removed, the filesystem will be allowed to be remounted readonly. This is
because fs_may_remount_ro does not find the 0 link tty inode on the file sb
list (because the tty code incorrectly removed it to use for its own purpose).
This can result in a filesystem with errors after it is marked "clean".

Taking idea from Christoph's initial patch, allocate a tty private struct
at file->private_data and put our required list fields in there, linking
file and tty. This makes tty nodes behave the same way as other device nodes
and avoid meddling with the vfs, and avoids this bug.

The error handling is not trivial in the tty code, so for this bugfix, I take
the simple approach of using __GFP_NOFAIL and don't worry about memory errors.
This is not a problem because our allocator doesn't fail small allocs as a rule
anyway. So proper error handling is left as an exercise for tty hackers.

[ Arguably filesystem's device inode would ideally be divorced from the
driver's pseudo inode when it is opened, but in practice it's not clear whether
that will ever be worth implementing. ]

Cc: linux-kernel@vger.kernel.org
Cc: Christoph Hellwig <hch@infradead.org>
Cc: Alan Cox <alan@lxorguk.ukuu.org.uk>
Cc: Greg Kroah-Hartman <gregkh@suse.de>
Signed-off-by: Nick Piggin <npiggin@kernel.dk>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
14 years agofs: cleanup files_lock locking
Nick Piggin [Tue, 17 Aug 2010 18:37:35 +0000 (04:37 +1000)]
fs: cleanup files_lock locking

fs: cleanup files_lock locking

Lock tty_files with a new spinlock, tty_files_lock; provide helpers to
manipulate the per-sb files list; unexport the files_lock spinlock.

Cc: linux-kernel@vger.kernel.org
Cc: Christoph Hellwig <hch@infradead.org>
Cc: Alan Cox <alan@lxorguk.ukuu.org.uk>
Acked-by: Andi Kleen <ak@linux.intel.com>
Acked-by: Greg Kroah-Hartman <gregkh@suse.de>
Signed-off-by: Nick Piggin <npiggin@kernel.dk>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
14 years agofs: remove extra lookup in __lookup_hash
Nick Piggin [Tue, 17 Aug 2010 18:37:34 +0000 (04:37 +1000)]
fs: remove extra lookup in __lookup_hash

fs: remove extra lookup in __lookup_hash

Optimize lookup for create operations, where no dentry should often be
common-case. In cases where it is not, such as unlink, the added overhead
is much smaller than the removed.

Also, move comments about __d_lookup racyness to the __d_lookup call site.
d_lookup is intuitive; __d_lookup is what needs commenting. So in that same
vein, add kerneldoc comments to __d_lookup and clean up some of the comments:

- We are interested in how the RCU lookup works here, particularly with
  renames. Make that explicit, and point to the document where it is explained
  in more detail.
- RCU is pretty standard now, and macros make implementations pretty mindless.
  If we want to know about RCU barrier details, we look in RCU code.
- Delete some boring legacy comments because we don't care much about how the
  code used to work, more about the interesting parts of how it works now. So
  comments about lazy LRU may be interesting, but would better be done in the
  LRU or refcount management code.

Signed-off-by: Nick Piggin <npiggin@kernel.dk>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
14 years agofs: fs_struct rwlock to spinlock
Nick Piggin [Tue, 17 Aug 2010 18:37:33 +0000 (04:37 +1000)]
fs: fs_struct rwlock to spinlock

fs: fs_struct rwlock to spinlock

struct fs_struct.lock is an rwlock with the read-side used to protect root and
pwd members while taking references to them. Taking a reference to a path
typically requires just 2 atomic ops, so the critical section is very small.
Parallel read-side operations would have cacheline contention on the lock, the
dentry, and the vfsmount cachelines, so the rwlock is unlikely to ever give a
real parallelism increase.

Replace it with a spinlock to avoid one or two atomic operations in typical
path lookup fastpath.

Signed-off-by: Nick Piggin <npiggin@kernel.dk>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
14 years agoapparmor: use task path helpers
Nick Piggin [Tue, 17 Aug 2010 18:37:32 +0000 (04:37 +1000)]
apparmor: use task path helpers

apparmor: use task path helpers

Signed-off-by: Nick Piggin <npiggin@kernel.dk>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
14 years agofs: dentry allocation consolidation
Nick Piggin [Tue, 17 Aug 2010 18:37:31 +0000 (04:37 +1000)]
fs: dentry allocation consolidation

fs: dentry allocation consolidation

There are 2 duplicate copies of code in dentry allocation in path lookup.
Consolidate them into a single function.

Signed-off-by: Nick Piggin <npiggin@kernel.dk>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
14 years agofs: fix do_lookup false negative
Nick Piggin [Tue, 17 Aug 2010 18:37:30 +0000 (04:37 +1000)]
fs: fix do_lookup false negative

fs: fix do_lookup false negative

In do_lookup, if we initially find no dentry, we take the directory i_mutex and
re-check the lookup. If we find a dentry there, then we revalidate it if
needed. However if that revalidate asks for the dentry to be invalidated, we
return -ENOENT from do_lookup. What should happen instead is an attempt to
allocate and lookup a new dentry.

This is probably not noticed because it is rare. It is only reached if a
concurrent create races in first (in which case, the dentry probably won't be
invalidated anyway), or if the racy __d_lookup has failed due to a
false-negative (which is very rare).

Fix this by removing code and have it use the normal reval path.

Signed-off-by: Nick Piggin <npiggin@kernel.dk>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
14 years agombcache: Limit the maximum number of cache entries
Andreas Gruenbacher [Mon, 16 Aug 2010 17:05:23 +0000 (19:05 +0200)]
mbcache: Limit the maximum number of cache entries

Limit the maximum number of mb_cache entries depending on the number of
hash buckets: if the only limit to the number of cache entries is the
available memory the hash chains can grow very long, taking a long time
to search.

At least partially solves https://bugzilla.lustre.org/show_bug.cgi?id=22771.

Signed-off-by: Andreas Gruenbacher <agruen@suse.de>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
14 years agohostfs ->follow_link() braino
Al Viro [Wed, 18 Aug 2010 10:21:10 +0000 (06:21 -0400)]
hostfs ->follow_link() braino

we want the assignment to err done inside the if () to be
visible after it, so (re)declaring err inside if () body
is wrong.

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
14 years agohostfs: dumb (and usually harmless) tpyo - strncpy instead of strlcpy
Al Viro [Wed, 18 Aug 2010 10:18:57 +0000 (06:18 -0400)]
hostfs: dumb (and usually harmless) tpyo - strncpy instead of strlcpy

... not harmless in this case - we have a string in the end of buffer
already.

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
14 years agoMerge branch 'perf/urgent' of git://git.kernel.org/pub/scm/linux/kernel/git/acme...
Ingo Molnar [Wed, 18 Aug 2010 07:53:36 +0000 (09:53 +0200)]
Merge branch 'perf/urgent' of git://git.kernel.org/pub/scm/linux/kernel/git/acme/linux-2.6 into perf/urgent

14 years agomd raid-1/10 Fix bio_rw bit manipulations again
NeilBrown [Wed, 18 Aug 2010 06:16:05 +0000 (16:16 +1000)]
md raid-1/10 Fix bio_rw bit manipulations again

commit 7b6d91daee5cac6402186ff224c3af39d79f4a0e changed the behaviour
of a few variables in raid1 and raid10 from flags to bit-sets, but
left them as type 'bool' so they did not work.

Change them (back) to unsigned long.
(historical note: see 1ef04fefe2241087d9db7e9615c3f11b516e36cf)

Signed-off-by: NeilBrown <neilb@suse.de>
Reported-by: Jiri Slaby <jslaby@suse.cz> and many others
14 years agoremove SWRITE* I/O types
Christoph Hellwig [Wed, 11 Aug 2010 15:06:24 +0000 (17:06 +0200)]
remove SWRITE* I/O types

These flags aren't real I/O types, but tell ll_rw_block to always
lock the buffer instead of giving up on a failed trylock.

Instead add a new write_dirty_buffer helper that implements this semantic
and use it from the existing SWRITE* callers.  Note that the ll_rw_block
code had a bug where it didn't promote WRITE_SYNC_PLUG properly, which
this patch fixes.

In the ufs code clean up the helper that used to call ll_rw_block
to mirror sync_dirty_buffer, which is the function it implements for
compound buffers.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
14 years agokill BH_Ordered flag
Christoph Hellwig [Wed, 11 Aug 2010 15:05:45 +0000 (17:05 +0200)]
kill BH_Ordered flag

Instead of abusing a buffer_head flag just add a variant of
sync_dirty_buffer which allows passing the exact type of write
flag required.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
14 years agovfs: update ctime when changing the file's permission by setfacl
Jan Kara [Tue, 17 Aug 2010 10:42:13 +0000 (12:42 +0200)]
vfs: update ctime when changing the file's permission by setfacl

generic_acl_set didn't update the ctime of the file when its permission was
changed.

Steps to reproduce:
 # touch aaa
 # stat -c %Z aaa
 1275289822
 # setfacl -m  'u::x,g::x,o::x' aaa
 # stat -c %Z aaa
 1275289822                         <- unchanged

But, according to the spec of the ctime, vfs must update it.

Port of ext3 patch by Miao Xie <miaox@cn.fujitsu.com>.

CC: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
14 years agocramfs: only unlock new inodes
Alexander Shishkin [Tue, 17 Aug 2010 14:08:35 +0000 (17:08 +0300)]
cramfs: only unlock new inodes

Commit 77b8a75f5bb introduced a warning at fs/inode.c:692 unlock_new_inode(),
caused by unlock_new_inode() being called on existing inodes as well.

This patch changes setup_inode() to only call unlock_new_inode() for I_NEW
inodes.

Signed-off-by: Alexander Shishkin <virtuoso@slind.org>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>