]> www.pilppa.org Git - linux-2.6-omap-h63xx.git/log
linux-2.6-omap-h63xx.git
17 years ago[SCSI] ibmvfc: Fix log level filtering
Brian King [Wed, 29 Oct 2008 13:46:33 +0000 (08:46 -0500)]
[SCSI] ibmvfc: Fix log level filtering

The ibmvfc log level filtering logic was reversed. The log_level scsi
host parameter should result in more verbose logs when log_level is
larger, not smaller.

Signed-off-by: Brian King <brking@linux.vnet.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
17 years ago[SCSI] Clean up my email address and use a single standard address for everything
Alan Cox [Mon, 27 Oct 2008 15:16:36 +0000 (15:16 +0000)]
[SCSI] Clean up my email address and use a single standard address for everything

Signed-off-by: Alan Cox <alan@redhat.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
17 years ago[SCSI] aacraid: check pci_alloc_consistent errors
FUJITA Tomonori [Thu, 23 Oct 2008 08:36:08 +0000 (17:36 +0900)]
[SCSI] aacraid: check pci_alloc_consistent errors

We need to check the address that pci_alloc_consistent() returns since
it might fail.

When pci_alloc_consistent() fails, some IOMMUs set the dma_handle
argument to zero. So we can't use fibptr->hw_fib_pa directly here.

Signed-off-by: FUJITA Tomonori <fujita.tomonori@lab.ntt.co.jp>
Cc: Aacraid List <aacraid@adaptec.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
17 years ago[SCSI] advansys, arcmsr, ipr, nsp32, qla1280, stex: use pci_ioremap_bar()
Arjan van de Ven [Sun, 28 Sep 2008 23:18:02 +0000 (16:18 -0700)]
[SCSI] advansys, arcmsr, ipr, nsp32, qla1280, stex: use pci_ioremap_bar()

Use the newly introduced pci_ioremap_bar() function in drivers/scsi.
pci_ioremap_bar() just takes a pci device and a bar number, with the goal
of making it really hard to get wrong, while also having a central place
to stick sanity checks.

Signed-off-by: Arjan van de Ven <arjan@linux.intel.com>
Acked-by: Matthew Wilcox <willy@linux.intel.com>
Cc: Brian King <brking@us.ibm.com>
Cc: Ed Lin <ed.lin@promise.com>
Cc: Nick Cheng <nick.cheng@areca.com.tw>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
17 years ago[SCSI] scsi ioctl: fix kernel-doc warning
Randy Dunlap [Wed, 22 Oct 2008 23:37:13 +0000 (16:37 -0700)]
[SCSI] scsi ioctl: fix kernel-doc warning

Fix kernel-doc parameter warning and correct the function name:

Warning(linux-next-20081022//drivers/scsi/scsi_ioctl.c:281): No description found for parameter 'ndelay'

Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
17 years agox86, pci: move arch/x86/pci/pci.h to arch/x86/include/asm/pci_x86.h
Jaswinder Singh Rajput [Sat, 27 Dec 2008 13:02:28 +0000 (18:32 +0530)]
x86, pci: move arch/x86/pci/pci.h to arch/x86/include/asm/pci_x86.h

Impact: cleanup

Now that arch/x86/pci/pci.h is used in a number of other places as well,
move the lowlevel x86 pci definitions into the architecture include files.
(not to be confused with the existing arch/x86/include/asm/pci.h file,
which provides public details about x86 PCI)

Tested on: X86_32_UP, X86_32_SMP and X86_64_SMP

Signed-off-by: Jaswinder Singh Rajput <jaswinderrajput@gmail.com>
Acked-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
17 years agox86_64: pci-gart_64.c iommu_fullflush should be static
Jaswinder Singh Rajput [Mon, 29 Dec 2008 15:08:09 +0000 (20:38 +0530)]
x86_64: pci-gart_64.c iommu_fullflush should be static

Impact: cleanup, reduce kernel size a bit, avoid sparse warning

Fixes sparse warning:

  arch/x86/kernel/pci-gart_64.c:55:5: warning: symbol 'iommu_fullflush' was not declared. Should it be static?

Signed-off-by: Jaswinder Singh Rajput <jaswinderrajput@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
17 years agox86: efi.c declare add_efi_memmap before they get used
Jaswinder Singh Rajput [Mon, 29 Dec 2008 15:06:40 +0000 (20:36 +0530)]
x86: efi.c declare add_efi_memmap before they get used

Impact: cleanup, avoid sparse warning

Fixes this sparse warning:

  arch/x86/kernel/efi.c:67:5: warning: symbol 'add_efi_memmap' was not declared. Should it be static?

Signed-off-by: Jaswinder Singh Rajput <jaswinderrajput@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
17 years agox86: io_apic.c io_apic_sync should be static
Jaswinder Singh Rajput [Mon, 29 Dec 2008 15:04:35 +0000 (20:34 +0530)]
x86: io_apic.c io_apic_sync should be static

Impact: cleanup, reduce kernel size a bit, avoid sparse warning

Fixes sparse warning:

  arch/x86/kernel/io_apic.c:709:6: warning: symbol 'io_apic_sync' was not declared. Should it be static?

Signed-off-by: Jaswinder Singh Rajput <jaswinderrajput@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
17 years agox86: apic.c declare pic_mode before they get used
Jaswinder Singh Rajput [Mon, 29 Dec 2008 15:02:52 +0000 (20:32 +0530)]
x86: apic.c declare pic_mode before they get used

Impact: cleanup, avoid sparse warning

In asm/mpspec.h moved out pic_mode from CONFIG_X86_32 as it is common
for both 32 and 64 bit.

Fixes this sparse warning for x86_64:

  arch/x86/kernel/apic.c:128:5: warning: symbol 'pic_mode' was not declared. Should it be static?

Signed-off-by: Jaswinder Singh Rajput <jaswinderrajput@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
17 years agooprofile: reordering some code in cpu_buffer.c
Robert Richter [Tue, 23 Dec 2008 00:35:12 +0000 (01:35 +0100)]
oprofile: reordering some code in cpu_buffer.c

Reordering code to keep alloc/free functions together.

Signed-off-by: Robert Richter <robert.richter@amd.com>
17 years agohsmmc: Fix bus voltage reset, clean-up for checkpatch
Tony Lindgren [Mon, 29 Dec 2008 16:15:47 +0000 (18:15 +0200)]
hsmmc: Fix bus voltage reset, clean-up for checkpatch

The bus voltage was being reset to 3V even if card was not
removed. Move the reset code to happen after slot has been
powered down. Also clean-up for checkpatch.

Signed-off-by: Tony Lindgren <tony@atomide.com>
17 years agoRT: fix push_rt_task() to handle dequeue_pushable properly
Gregory Haskins [Mon, 29 Dec 2008 14:39:53 +0000 (09:39 -0500)]
RT: fix push_rt_task() to handle dequeue_pushable properly

A panic was discovered by Chirag Jog where a BUG_ON sanity check
in the new "pushable_task" logic would trigger a panic under
certain circumstances:

http://lkml.org/lkml/2008/9/25/189

Gilles Carry discovered that the root cause was attributed to the
pushable_tasks list getting corrupted in the push_rt_task logic.
This was the result of a dropped rq lock in double_lock_balance
allowing a task in the process of being pushed to potentially migrate
away, and thus corrupt the pushable_tasks() list.

I traced back the problem as introduced by the pushable_tasks patch
that went in recently.   There is a "retry" path in push_rt_task()
that actually had a compound conditional to decide whether to
retry or exit.  I missed the meaning behind the rationale for the
virtual "if(!task) goto out;" portion of the compound statement and
thus did not handle it properly.  The new pushable_tasks logic
actually creates three distinct conditions:

1) an untouched and unpushable task should be dequeued
2) a migrated task where more pushable tasks remain should be retried
3) a migrated task where no more pushable tasks exist should exit

The original logic mushed (1) and (3) together, resulting in the
system dequeuing a migrated task (against an unlocked foreign run-queue
nonetheless).

To fix this, we get rid of the notion of "paranoid" and we support the
three unique conditions properly.  The paranoid feature is no longer
relevant with the new pushable logic (since pushable naturally limits
the loop) anyway, so lets just remove it.

Reported-By: Chirag Jog <chirag@linux.vnet.ibm.com>
Found-by: Gilles Carry <gilles.carry@bull.net>
Signed-off-by: Gregory Haskins <ghaskins@novell.com>
17 years agosched: create "pushable_tasks" list to limit pushing to one attempt
Gregory Haskins [Mon, 29 Dec 2008 14:39:53 +0000 (09:39 -0500)]
sched: create "pushable_tasks" list to limit pushing to one attempt

The RT scheduler employs a "push/pull" design to actively balance tasks
within the system (on a per disjoint cpuset basis).  When a task is
awoken, it is immediately determined if there are any lower priority
cpus which should be preempted.  This is opposed to the way normal
SCHED_OTHER tasks behave, which will wait for a periodic rebalancing
operation to occur before spreading out load.

When a particular RQ has more than 1 active RT task, it is said to
be in an "overloaded" state.  Once this occurs, the system enters
the active balancing mode, where it will try to push the task away,
or persuade a different cpu to pull it over.  The system will stay
in this state until the system falls back below the <= 1 queued RT
task per RQ.

However, the current implementation suffers from a limitation in the
push logic.  Once overloaded, all tasks (other than current) on the
RQ are analyzed on every push operation, even if it was previously
unpushable (due to affinity, etc).  Whats more, the operation stops
at the first task that is unpushable and will not look at items
lower in the queue.  This causes two problems:

1) We can have the same tasks analyzed over and over again during each
   push, which extends out the fast path in the scheduler for no
   gain.  Consider a RQ that has dozens of tasks that are bound to a
   core.  Each one of those tasks will be encountered and skipped
   for each push operation while they are queued.

2) There may be lower-priority tasks under the unpushable task that
   could have been successfully pushed, but will never be considered
   until either the unpushable task is cleared, or a pull operation
   succeeds.  The net result is a potential latency source for mid
   priority tasks.

This patch aims to rectify these two conditions by introducing a new
priority sorted list: "pushable_tasks".  A task is added to the list
each time a task is activated or preempted.  It is removed from the
list any time it is deactivated, made current, or fails to push.

This works because a task only needs to be attempted to push once.
After an initial failure to push, the other cpus will eventually try to
pull the task when the conditions are proper.  This also solves the
problem that we don't completely analyze all tasks due to encountering
an unpushable tasks.  Now every task will have a push attempted (when
appropriate).

This reduces latency both by shorting the critical section of the
rq->lock for certain workloads, and by making sure the algorithm
considers all eligible tasks in the system.

[ rostedt: added a couple more BUG_ONs ]

Signed-off-by: Gregory Haskins <ghaskins@novell.com>
Acked-by: Steven Rostedt <srostedt@redhat.com>
17 years agoplist: fix PLIST_NODE_INIT to work with debug enabled
Gregory Haskins [Mon, 29 Dec 2008 14:39:53 +0000 (09:39 -0500)]
plist: fix PLIST_NODE_INIT to work with debug enabled

It seems that PLIST_NODE_INIT breaks if used and DEBUG_PI_LIST is defined.
Since there are no current users of PLIST_NODE_INIT, this has gone
undetected.  This patch fixes the build issue that enables the
DEBUG_PI_LIST later in the series when we use it in init_task.h

Signed-off-by: Gregory Haskins <ghaskins@novell.com>
17 years agosched: add sched_class->needs_post_schedule() member
Gregory Haskins [Mon, 29 Dec 2008 14:39:52 +0000 (09:39 -0500)]
sched: add sched_class->needs_post_schedule() member

We currently run class->post_schedule() outside of the rq->lock, which
means that we need to test for the need to post_schedule outside of
the lock to avoid a forced reacquistion.  This is currently not a problem
as we only look at rq->rt.overloaded.  However, we want to enhance this
going forward to look at more state to reduce the need to post_schedule to
a bare minimum set.  Therefore, we introduce a new member-func called
needs_post_schedule() which tests for the post_schedule condtion without
actually performing the work.  Therefore it is safe to call this
function before the rq->lock is released, because we are guaranteed not
to drop the lock at an intermediate point (such as what post_schedule()
may do).

We will use this later in the series

[ rostedt: removed paranoid BUG_ON ]

Signed-off-by: Gregory Haskins <ghaskins@novell.com>
17 years agosched: make double-lock-balance fair
Gregory Haskins [Mon, 29 Dec 2008 14:39:51 +0000 (09:39 -0500)]
sched: make double-lock-balance fair

double_lock balance() currently favors logically lower cpus since they
often do not have to release their own lock to acquire a second lock.
The result is that logically higher cpus can get starved when there is
a lot of pressure on the RQs.  This can result in higher latencies on
higher cpu-ids.

This patch makes the algorithm more fair by forcing all paths to have
to release both locks before acquiring them again.  Since callsites to
double_lock_balance already consider it a potential preemption/reschedule
point, they have the proper logic to recheck for atomicity violations.

Signed-off-by: Gregory Haskins <ghaskins@novell.com>
17 years agosched: pull only one task during NEWIDLE balancing to limit critical section
Gregory Haskins [Mon, 29 Dec 2008 14:39:50 +0000 (09:39 -0500)]
sched: pull only one task during NEWIDLE balancing to limit critical section

git-id c4acb2c0669c5c5c9b28e9d02a34b5c67edf7092 attempted to limit
newidle critical section length by stopping after at least one task
was moved.  Further investigation has shown that there are other
paths nested further inside the algorithm which still remain that allow
long latencies to occur with newidle balancing.  This patch applies
the same technique inside balance_tasks() to limit the duration of
this optional balancing operation.

Signed-off-by: Gregory Haskins <ghaskins@novell.com>
CC: Nick Piggin <npiggin@suse.de>
17 years agosched: only try to push a task on wakeup if it is migratable
Gregory Haskins [Mon, 29 Dec 2008 14:39:50 +0000 (09:39 -0500)]
sched: only try to push a task on wakeup if it is migratable

There is no sense in wasting time trying to push a task away that
cannot move anywhere else.  We gain no benefit from trying to push
other tasks at this point, so if the task being woken up is non
migratable, just skip the whole operation.  This reduces overhead
in the wakeup path for certain tasks.

Signed-off-by: Gregory Haskins <ghaskins@novell.com>
17 years agosched: use highest_prio.next to optimize pull operations
Gregory Haskins [Mon, 29 Dec 2008 14:39:50 +0000 (09:39 -0500)]
sched: use highest_prio.next to optimize pull operations

We currently take the rq->lock for every cpu in an overload state during
pull_rt_tasks().  However, we now have enough information via the
highest_prio.[curr|next] fields to determine if there is any tasks of
interest to warrant the overhead of the rq->lock, before we actually take
it.  So we use this information to reduce lock contention during the
pull for the case where the source-rq doesnt have tasks that preempt
the current task.

Signed-off-by: Gregory Haskins <ghaskins@novell.com>
17 years agosched: use highest_prio.curr for pull threshold
Gregory Haskins [Mon, 29 Dec 2008 14:39:49 +0000 (09:39 -0500)]
sched: use highest_prio.curr for pull threshold

highest_prio.curr is actually a more accurate way to keep track of
the pull_rt_task() threshold since it is always up to date, even
if the "next" task migrates during double_lock.  Therefore, stop
looking at the "next" task object and simply use the highest_prio.curr.

Signed-off-by: Gregory Haskins <ghaskins@novell.com>
17 years agosched: track the next-highest priority on each runqueue
Gregory Haskins [Mon, 29 Dec 2008 14:39:49 +0000 (09:39 -0500)]
sched: track the next-highest priority on each runqueue

We will use this later in the series to reduce the amount of rq-lock
contention during a pull operation

Signed-off-by: Gregory Haskins <ghaskins@novell.com>
17 years agosched: cleanup inc/dec_rt_tasks
Gregory Haskins [Mon, 29 Dec 2008 14:39:49 +0000 (09:39 -0500)]
sched: cleanup inc/dec_rt_tasks

Move some common definitions up to the function prologe to simplify the
body logic.

Signed-off-by: Gregory Haskins <ghaskins@novell.com>
17 years agohsmmc: Fix setting the voltage back to 3V
Tony Lindgren [Mon, 29 Dec 2008 14:24:25 +0000 (16:24 +0200)]
hsmmc: Fix setting the voltage back to 3V

We should not set it in mmc_omap_detect, as that does not necessarily
tell that card got removed. Reset the voltage back to 3V after powering
off the slot instead.

Signed-off-by: Tony Lindgren <tony@atomide.com>
17 years agox86/oprofile: fix pci_dev use count for AMD northbridge devices
Robert Richter [Mon, 15 Dec 2008 14:09:50 +0000 (15:09 +0100)]
x86/oprofile: fix pci_dev use count for AMD northbridge devices

This patch fixes the PCI device use count for AMD northbridge
devices. In case of an IBS LVT initialization failure, the PCI device
is released now by calling pci_dev_put().

If there are no initialization errors, the devices are released in
pci_get_device() while iterating.

Signed-off-by: Robert Richter <robert.richter@amd.com>
17 years agooprofile: remove ring buffer inline functions in cpu_buffer.h
Robert Richter [Tue, 16 Dec 2008 15:19:54 +0000 (16:19 +0100)]
oprofile: remove ring buffer inline functions in cpu_buffer.h

This patch moves ring buffer inline functions to cpu_buffer.c.

Cc: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Robert Richter <robert.richter@amd.com>
17 years agooprofile: rename cpu buffer functions
Robert Richter [Wed, 24 Dec 2008 15:53:53 +0000 (16:53 +0100)]
oprofile: rename cpu buffer functions

This patch renames cpu buffer functions to something more oprofile
specific names. Functions will be moved to the global name space.

Cc: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Robert Richter <robert.richter@amd.com>
17 years agoMerge branch 'topic/kmemtrace' of git://git.kernel.org/pub/scm/linux/kernel/git/penbe...
Ingo Molnar [Mon, 29 Dec 2008 14:16:24 +0000 (15:16 +0100)]
Merge branch 'topic/kmemtrace' of git://git.kernel.org/pub/scm/linux/kernel/git/penberg/slab-2.6 into tracing/kmemtrace

17 years agotracing/selftest: remove TRACE_CONT reference
Ingo Molnar [Mon, 29 Dec 2008 14:07:47 +0000 (15:07 +0100)]
tracing/selftest: remove TRACE_CONT reference

Impact: build fix

TRACE_CONT is gone - fix up the self-test too.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
17 years agokmemtrace: add missing newline
Vegard Nossum [Thu, 6 Nov 2008 16:42:18 +0000 (17:42 +0100)]
kmemtrace: add missing newline

This was causing artifacts in my dmesg.

Acked-by: Eduard - Gabriel Munteanu <eduard.munteanu@linux360.ro>
Signed-off-by: Vegard Nossum <vegard.nossum@gmail.com>
Signed-off-by: Pekka Enberg <penberg@cs.helsinki.fi>
17 years agokmemtrace: remove config option for enabling tracing at boot
Pekka Enberg [Fri, 10 Oct 2008 08:02:59 +0000 (11:02 +0300)]
kmemtrace: remove config option for enabling tracing at boot

Users can pass kmemtrace.enabled=yes as a kernel parameter to enable kmemtrace
at boot so remove the useless CONFIG_KMEMTRACE_DEFAULT_ENABLED config option.

Cc: Eduard - Gabriel Munteanu <eduard.munteanu@linux360.ro>
Signed-off-by: Pekka Enberg <penberg@cs.helsinki.fi>
17 years agokmemtrace: allow kmemtrace to be enabled after boot
Pekka Enberg [Fri, 10 Oct 2008 07:57:44 +0000 (10:57 +0300)]
kmemtrace: allow kmemtrace to be enabled after boot

The kmemtrace_init() function returns early if kmemtrace is disabled at boot
causing kmemtrace_setup_late() to also bail out on NULL channel. This has the
unfortunate side effect that none of the debugfs files needed to enable
kmemtrace after boot are created.

Cc: Eduard - Gabriel Munteanu <eduard.munteanu@linux360.ro>
Signed-off-by: Pekka Enberg <penberg@cs.helsinki.fi>
17 years agokmemtrace: remove unnecessary casts
Pekka Enberg [Mon, 1 Sep 2008 07:11:54 +0000 (10:11 +0300)]
kmemtrace: remove unnecessary casts

Now that we use _RET_IP_ there's no need to cast 'caller' to unsigned long.

Signed-off-by: Pekka Enberg <penberg@cs.helsinki.fi>
17 years agokmemtrace: SLUB hooks for caller-tracking functions.
Eduard - Gabriel Munteanu [Sun, 24 Aug 2008 17:49:35 +0000 (20:49 +0300)]
kmemtrace: SLUB hooks for caller-tracking functions.

This patch adds kmemtrace hooks for __kmalloc_track_caller() and
__kmalloc_node_track_caller(). Currently, they set the call site pointer
to the value recieved as a parameter. (This could change if we implement
stack trace exporting in kmemtrace.)

Signed-off-by: Eduard - Gabriel Munteanu <eduard.munteanu@linux360.ro>
Signed-off-by: Pekka Enberg <penberg@cs.helsinki.fi>
17 years agokmemtrace: Better alternative to "kmemtrace: fix printk format warnings".
Eduard - Gabriel Munteanu [Tue, 19 Aug 2008 17:43:24 +0000 (20:43 +0300)]
kmemtrace: Better alternative to "kmemtrace: fix printk format warnings".

Fix the problem "kmemtrace: fix printk format warnings" attempted to fix,
but resulted in marker-probe format mismatch warnings. Instead of carrying
size_t into probes, we get rid of it by casting to unsigned long, just as
we did with gfp_t.

This way, we don't need to change marker format strings and we don't have
to rely on other format specifiers like "%zu", making for consistent use
of more generic data types (since there are no format specifiers for
gfp_t, for example).

Signed-off-by: Eduard - Gabriel Munteanu <eduard.munteanu@linux360.ro>
Signed-off-by: Pekka Enberg <penberg@cs.helsinki.fi>
17 years agokmemtrace: Fix typos in documentation.
Eduard - Gabriel Munteanu [Tue, 19 Aug 2008 17:43:27 +0000 (20:43 +0300)]
kmemtrace: Fix typos in documentation.

Corrected the ABI description and the kmemtrace usage guide. Thanks to
Randy Dunlap for noticing these errors.

Signed-off-by: Eduard - Gabriel Munteanu <eduard.munteanu@linux360.ro>
Signed-off-by: Pekka Enberg <penberg@cs.helsinki.fi>
17 years agokmemtrace: SLUB hooks.
Eduard - Gabriel Munteanu [Tue, 19 Aug 2008 17:43:26 +0000 (20:43 +0300)]
kmemtrace: SLUB hooks.

This adds hooks for the SLUB allocator, to allow tracing with kmemtrace.

Signed-off-by: Eduard - Gabriel Munteanu <eduard.munteanu@linux360.ro>
Signed-off-by: Pekka Enberg <penberg@cs.helsinki.fi>
17 years agokmemtrace: SLOB hooks.
Eduard - Gabriel Munteanu [Sun, 10 Aug 2008 17:14:07 +0000 (20:14 +0300)]
kmemtrace: SLOB hooks.

This adds hooks for the SLOB allocator, to allow tracing with kmemtrace.

We also convert some inline functions to __always_inline to make sure
_RET_IP_, which expands to __builtin_return_address(0), always works
as expected.

Acked-by: Matt Mackall <mpm@selenic.com>
Signed-off-by: Eduard - Gabriel Munteanu <eduard.munteanu@linux360.ro>
Signed-off-by: Pekka Enberg <penberg@cs.helsinki.fi>
17 years agokmemtrace: SLAB hooks.
Eduard - Gabriel Munteanu [Sun, 10 Aug 2008 17:14:05 +0000 (20:14 +0300)]
kmemtrace: SLAB hooks.

This adds hooks for the SLAB allocator, to allow tracing with kmemtrace.

We also convert some inline functions to __always_inline to make sure
_RET_IP_, which expands to __builtin_return_address(0), always works
as expected.

Signed-off-by: Eduard - Gabriel Munteanu <eduard.munteanu@linux360.ro>
Signed-off-by: Pekka Enberg <penberg@cs.helsinki.fi>
17 years agokmemtrace: Additional documentation.
Eduard - Gabriel Munteanu [Sun, 10 Aug 2008 17:14:04 +0000 (20:14 +0300)]
kmemtrace: Additional documentation.

Documented kmemtrace's ABI, purpose and design. Also includes a short
usage guide, FAQ, as well as a link to the userspace application's Git
repository, which is currently hosted at repo.or.cz.

Signed-off-by: Eduard - Gabriel Munteanu <eduard.munteanu@linux360.ro>
Signed-off-by: Pekka Enberg <penberg@cs.helsinki.fi>
17 years agokmemtrace: Core implementation.
Eduard - Gabriel Munteanu [Sun, 10 Aug 2008 17:14:03 +0000 (20:14 +0300)]
kmemtrace: Core implementation.

kmemtrace provides tracing for slab allocator functions, such as kmalloc,
kfree, kmem_cache_alloc, kmem_cache_free etc.. Collected data is then fed
to the userspace application in order to analyse allocation hotspots,
internal fragmentation and so on, making it possible to see how well an
allocator performs, as well as debug and profile kernel code.

Signed-off-by: Eduard - Gabriel Munteanu <eduard.munteanu@linux360.ro>
Signed-off-by: Pekka Enberg <penberg@cs.helsinki.fi>
17 years agoSLUB: Replace __builtin_return_address(0) with _RET_IP_.
Eduard - Gabriel Munteanu [Tue, 19 Aug 2008 17:43:25 +0000 (20:43 +0300)]
SLUB: Replace __builtin_return_address(0) with _RET_IP_.

This patch replaces __builtin_return_address(0) with _RET_IP_, since a
previous patch moved _RET_IP_ and _THIS_IP_ to include/linux/kernel.h and
they're widely available now. This makes for shorter and easier to read
code.

[penberg@cs.helsinki.fi: remove _RET_IP_ casts to void pointer]
Signed-off-by: Eduard - Gabriel Munteanu <eduard.munteanu@linux360.ro>
Signed-off-by: Pekka Enberg <penberg@cs.helsinki.fi>
17 years agoInput: map_to_7segment.h - convert to __inline__ for userspace
Mike Frysinger [Mon, 29 Dec 2008 12:00:23 +0000 (04:00 -0800)]
Input: map_to_7segment.h - convert to __inline__ for userspace

Use __inline__ rather than inline for map_to_seg7() since it is exported
to userspace.

Signed-off-by: Mike Frysinger <vapier@gentoo.org>
Signed-off-by: Dmitry Torokhov <dtor@mail.ru>
17 years agoInput: add support for enhanced rotary controller on pxa930 and pxa935
Yong Yao [Mon, 29 Dec 2008 12:00:02 +0000 (04:00 -0800)]
Input: add support for enhanced rotary controller on pxa930 and pxa935

Signed-off-by: Yong Yao <yaoyong@marvell.com>
Signed-off-by: Eric Miao <eric.miao@marvell.com>
Signed-off-by: Dmitry Torokhov <dtor@mail.ru>
17 years agoInput: add support for trackball on pxa930 and pxa935
Yong Yao [Mon, 29 Dec 2008 11:59:59 +0000 (03:59 -0800)]
Input: add support for trackball on pxa930 and pxa935

Signed-off-by: Yong Yao <yaoyong@marvell.com>
Signed-off-by: Eric Miao <eric.miao@marvell.com>
Signed-off-by: Dmitry Torokhov <dtor@mail.ru>
17 years agolocking, percpu counters: introduce separate lock classes
Peter Zijlstra [Fri, 26 Dec 2008 14:08:55 +0000 (15:08 +0100)]
locking, percpu counters: introduce separate lock classes

Impact: fix lockdep false positives

Classify percpu_counter instances similar to regular lock objects --
that is, per instantiation site.

The networking code has increased its use of percpu_counters, which
leads to false positives if they are treated as a single class.

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
17 years agosata_sil: add Large Block Transfer support
Robert Hancock [Thu, 25 Dec 2008 01:06:06 +0000 (19:06 -0600)]
sata_sil: add Large Block Transfer support

This implements support for the Large Block Transfer feature found in Silicon
Image 311x controllers. This allows transferring bigger contiguous chunks of
data from system memory and avoids the 64KB boundary restriction of standard
SFF controllers.

This is based on a patch from Jeff Garzik (from the sii-lbt branch of
libata-dev) but includes a few bug fixes: Since the bmdma2 register does not
implement the status bits, the original bmdma register must be used except
where the bmdma2 register is required. As well the DMA boundary should be
31-bit instead of 32-bit since the top bit of the length field is still
required for the PRD end-of-table flag.

Signed-off-by: Robert Hancock <hancockr@shaw.ca>
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>
17 years ago[libata] ata_piix: cleanup dmi strings checking
Jiri Slaby [Wed, 10 Dec 2008 13:07:22 +0000 (14:07 +0100)]
[libata] ata_piix: cleanup dmi strings checking

Commit
ATA: piix, fix pointer deref on suspend
fixed a possible oops in an ugly manner. Use newly introduced dmi_match()
to make the code pretty again.

Signed-off-by: Jiri Slaby <jirislaby@gmail.com>
Cc: Alexandru Romanescu <a_romanescu@yahoo.co.uk>
Cc: Tejun Heo <tj@kernel.org>
Cc: Alan Cox <alan@lxorguk.ukuu.org.uk>
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>
17 years agoDMI: add dmi_match
Jiri Slaby [Wed, 10 Dec 2008 13:07:21 +0000 (14:07 +0100)]
DMI: add dmi_match

Add a wrapper for testing system_info which will handle also NULL
system infos.

This will be used by the ata PIIX driver.

Signed-off-by: Jiri Slaby <jirislaby@gmail.com>
Cc: Alexandru Romanescu <a_romanescu@yahoo.co.uk>
Cc: Tejun Heo <tj@kernel.org>
Cc: Alan Cox <alan@lxorguk.ukuu.org.uk>
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>
17 years agolibata: blacklist NCQ on OCZ CORE 2 SSD (resend)
Lubomir Bulej [Mon, 22 Dec 2008 10:35:22 +0000 (11:35 +0100)]
libata: blacklist NCQ on OCZ CORE 2 SSD (resend)

The patchlet below blacklists NCQ on OCZ CORE v2 SSD drive(s). Even
though the drive advertises NCQ support with queue depth 1, it responds
with all-zeroes FIS to NCQ commands which triggers ata error handling
several times before the kernel decides to disable NCQ on the drive.

Signed-off-by: Lubomir Bulej <lubomir.bulej@dsrg.mff.cuni.cz>
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>
17 years agox86: introducing asm/sys_ia32.h
Jaswinder Singh Rajput [Sat, 27 Dec 2008 16:07:10 +0000 (21:37 +0530)]
x86: introducing asm/sys_ia32.h

Impact: cleanup, avoid 44 sparse warnings, new file asm/sys_ia32.h

Fixes following sparse warnings:

  CHECK   arch/x86/ia32/sys_ia32.c
arch/x86/ia32/sys_ia32.c:53:17: warning: symbol 'sys32_truncate64' was not declared. Should it be static?
arch/x86/ia32/sys_ia32.c:60:17: warning: symbol 'sys32_ftruncate64' was not declared. Should it be static?
arch/x86/ia32/sys_ia32.c:98:17: warning: symbol 'sys32_stat64' was not declared. Should it be static?
arch/x86/ia32/sys_ia32.c:109:17: warning: symbol 'sys32_lstat64' was not declared. Should it be static?
arch/x86/ia32/sys_ia32.c:119:17: warning: symbol 'sys32_fstat64' was not declared. Should it be static?
arch/x86/ia32/sys_ia32.c:128:17: warning: symbol 'sys32_fstatat' was not declared. Should it be static?
arch/x86/ia32/sys_ia32.c:164:17: warning: symbol 'sys32_mmap' was not declared. Should it be static?
arch/x86/ia32/sys_ia32.c:195:17: warning: symbol 'sys32_mprotect' was not declared. Should it be static?
arch/x86/ia32/sys_ia32.c:201:17: warning: symbol 'sys32_pipe' was not declared. Should it be static?
arch/x86/ia32/sys_ia32.c:215:17: warning: symbol 'sys32_rt_sigaction' was not declared. Should it be static?
arch/x86/ia32/sys_ia32.c:291:17: warning: symbol 'sys32_sigaction' was not declared. Should it be static?
arch/x86/ia32/sys_ia32.c:330:17: warning: symbol 'sys32_rt_sigprocmask' was not declared. Should it be static?
arch/x86/ia32/sys_ia32.c:370:17: warning: symbol 'sys32_alarm' was not declared. Should it be static?
arch/x86/ia32/sys_ia32.c:383:17: warning: symbol 'sys32_old_select' was not declared. Should it be static?
arch/x86/ia32/sys_ia32.c:393:17: warning: symbol 'sys32_waitpid' was not declared. Should it be static?
arch/x86/ia32/sys_ia32.c:401:17: warning: symbol 'sys32_sysfs' was not declared. Should it be static?
arch/x86/ia32/sys_ia32.c:406:17: warning: symbol 'sys32_sched_rr_get_interval' was not declared. Should it be static?
arch/x86/ia32/sys_ia32.c:421:17: warning: symbol 'sys32_rt_sigpending' was not declared. Should it be static?
arch/x86/ia32/sys_ia32.c:445:17: warning: symbol 'sys32_rt_sigqueueinfo' was not declared. Should it be static?
arch/x86/ia32/sys_ia32.c:472:17: warning: symbol 'sys32_sysctl' was not declared. Should it be static?
arch/x86/ia32/sys_ia32.c:517:17: warning: symbol 'sys32_pread' was not declared. Should it be static?
arch/x86/ia32/sys_ia32.c:524:17: warning: symbol 'sys32_pwrite' was not declared. Should it be static?
arch/x86/ia32/sys_ia32.c:532:17: warning: symbol 'sys32_personality' was not declared. Should it be static?
arch/x86/ia32/sys_ia32.c:545:17: warning: symbol 'sys32_sendfile' was not declared. Should it be static?
arch/x86/ia32/sys_ia32.c:565:17: warning: symbol 'sys32_mmap2' was not declared. Should it be static?
arch/x86/ia32/sys_ia32.c:589:17: warning: symbol 'sys32_olduname' was not declared. Should it be static?
arch/x86/ia32/sys_ia32.c:626:6: warning: symbol 'sys32_uname' was not declared. Should it be static?
arch/x86/ia32/sys_ia32.c:641:6: warning: symbol 'sys32_ustat' was not declared. Should it be static?
arch/x86/ia32/sys_ia32.c:663:17: warning: symbol 'sys32_execve' was not declared. Should it be static?
arch/x86/ia32/sys_ia32.c:678:17: warning: symbol 'sys32_clone' was not declared. Should it be static?
arch/x86/ia32/sys_ia32.c:693:6: warning: symbol 'sys32_lseek' was not declared. Should it be static?
arch/x86/ia32/sys_ia32.c:698:6: warning: symbol 'sys32_kill' was not declared. Should it be static?
arch/x86/ia32/sys_ia32.c:703:6: warning: symbol 'sys32_fadvise64_64' was not declared. Should it be static?
arch/x86/ia32/sys_ia32.c:712:6: warning: symbol 'sys32_vm86_warning' was not declared. Should it be static?
arch/x86/ia32/sys_ia32.c:726:6: warning: symbol 'sys32_lookup_dcookie' was not declared. Should it be static?
arch/x86/ia32/sys_ia32.c:732:20: warning: symbol 'sys32_readahead' was not declared. Should it be static?
arch/x86/ia32/sys_ia32.c:738:17: warning: symbol 'sys32_sync_file_range' was not declared. Should it be static?
arch/x86/ia32/sys_ia32.c:746:17: warning: symbol 'sys32_fadvise64' was not declared. Should it be static?
arch/x86/ia32/sys_ia32.c:753:17: warning: symbol 'sys32_fallocate' was not declared. Should it be static?
  CHECK   arch/x86/ia32/ia32_signal.c
arch/x86/ia32/ia32_signal.c:126:17: warning: symbol 'sys32_sigsuspend' was not declared. Should it be static?
arch/x86/ia32/ia32_signal.c:141:17: warning: symbol 'sys32_sigaltstack' was not declared. Should it be static?
arch/x86/ia32/ia32_signal.c:249:17: warning: symbol 'sys32_sigreturn' was not declared. Should it be static?
arch/x86/ia32/ia32_signal.c:279:17: warning: symbol 'sys32_rt_sigreturn' was not declared. Should it be static?
  CHECK   arch/x86/ia32/ipc32.c
arch/x86/ia32/ipc32.c:12:17: warning: symbol 'sys32_ipc' was not declared. Should it be static?

Signed-off-by: Jaswinder Singh Rajput <jaswinderrajput@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
17 years agox86: mach-default setup.c cleanups
Cyrill Gorcunov [Sat, 27 Dec 2008 14:10:18 +0000 (17:10 +0300)]
x86: mach-default setup.c cleanups

Impact: cleanup

- Break long lines into shorter form.
- Use pr_ macros instead of plain printk.

Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
17 years agox86: mark get_cpu_leaves() with __cpuinit annotation
Sergio Luis [Sun, 28 Dec 2008 07:12:26 +0000 (04:12 -0300)]
x86: mark get_cpu_leaves() with __cpuinit annotation

Impact: fix section mismatch warning

Commit b2bb85549134c005e997e5a7ed303bda6a1ae738 ("x86: Remove cpumask games
in x86/kernel/cpu/intel_cacheinfo.c") introduced get_cpu_leaves(), which
references __cpuinit cpuid4_cache_lookup().

Mark get_cpu_leaves() with a __cpuinit annotation.

Signed-off-by: Sergio Luis <sergio@larces.uece.br>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
17 years agoMerge branch 'linus' into x86/cleanups
Ingo Molnar [Mon, 29 Dec 2008 12:09:40 +0000 (13:09 +0100)]
Merge branch 'linus' into x86/cleanups

17 years agotracing/ftrace: make trace_find_cmdline() generally available
Ingo Molnar [Mon, 29 Dec 2008 12:02:17 +0000 (13:02 +0100)]
tracing/ftrace: make trace_find_cmdline() generally available

Impact: build fix

On !CONFIG_CONTEXT_SWITCH_TRACER trace_find_cmdline() is not defined:

 kernel/trace/trace_output.c: In function 'trace_ctxwake_print':
 kernel/trace/trace_output.c:499: error: implicit declaration of function 'trace_find_cmdline'
 kernel/trace/trace_output.c:499: warning: assignment makes pointer from integer without a cast

Move it to the generic section in trace.h.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
17 years agotracing/branch-tracer: adapt to the stat tracing API
Frederic Weisbecker [Sat, 27 Dec 2008 22:25:38 +0000 (23:25 +0100)]
tracing/branch-tracer: adapt to the stat tracing API

Impact: refactor the branch tracer

This patch adapts the branch tracer to the tracing API.

This is a proof of concept because the branch tracer implements two
"stat tracing" that were split in two files.

So I added an option to the branch tracer: stat_all_branch.
If it is set, then trace_stat will output all of the branches
entries stats. Otherwise, it will print the annotated branches.

Its is a kind of quick trick, waiting for a better solution.

By default, the annotated branches stat are sorted by incorrect branch
prediction percentage.

Ie:

 correct incorrect  %        Function                  File              Line
 ------- ---------  -        --------                  ----              ----
       0        1 100 native_smp_prepare_cpus        smpboot.c            1228
       0        1 100 hpet_rtc_timer_reinit          hpet.c               1057
       0    18032 100 sched_info_queued              sched_stats.h        223
       0      684 100 yield_task_fair                sched_fair.c         984
       0      282 100 pre_schedule_rt                sched_rt.c           1263
       0    13414 100 sched_info_dequeued            sched_stats.h        178
       0    21724 100 sched_info_switch              sched_stats.h        270
       0        1 100 get_signal_to_deliver          signal.c             1820
       0        8 100 __cancel_work_timer            workqueue.c          560
       0      212 100 verify_export_symbols          module.c             1509
       0       17 100 __rmqueue_fallback             page_alloc.c         793
       0       43 100 clear_page_mlock               internal.h           129
       0      124 100 try_to_unmap_anon              rmap.c               1021
       0       53 100 try_to_unmap_anon              rmap.c               1013
       0        6 100 vma_address                    rmap.c               232
       0     3301 100 try_to_unmap_file              rmap.c               1082
       0      466 100 try_to_unmap_file              rmap.c               1077
       0        1 100 mem_cgroup_create              memcontrol.c         1090
       0        3 100 inotify_find_update_watch      inotify.c            726
       2    30163  99 perf_counter_task_sched_out    perf_counter.c       385
       1     2935  99 percpu_free                    allocpercpu.c        138
    1544   297672  99 dentry_lru_del_init            dcache.c             153
       8     1074  99 input_pass_event               input.c              86
    1390    76781  98 mapping_unevictable            pagemap.h            50
     280     6665  95 pick_next_task_rt              sched_rt.c           889
     750     4826  86 next_pidmap                    pid.c                194
       2        8  80 blocking_notifier_chain_regist notifier.c           220
      36      130  78 ioremap_pte_range              ioremap.c            22
    1093     3247  74 IS_ERR                         err.h                34
    1023     2908  73 sched_slice                    sched_fair.c         445
      22       60  73 disk_put_part                  genhd.h              206
[...]

It enables a developer to quickly address the source of incorrect branch
predictions.  Note that this sorting would be better with a second sort on
the number of incorrect predictions.

Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
17 years agotracing/ftrace: provide the base infrastructure for histogram tracing
Frederic Weisbecker [Mon, 29 Dec 2008 04:44:51 +0000 (20:44 -0800)]
tracing/ftrace: provide the base infrastructure for histogram tracing

Impact: extend the tracing API

The goal of this patch is to normalize and make more easy the
implementation of statistical (histogram) tracing.

It implements a trace_stat file into the /debugfs/tracing directory where
one can print a one-shot output of statistics/histogram entries.

A tracer has to provide two basic iterator callbacks:

  stat_start() => the first entry
  stat_next(prev, idx) => the next one.

Note that it is adapted for arrays or hash tables or lists.... since it
provides a pointer to the previous entry and the current index of the
iterator.

These two callbacks are called to get a snapshot of the statistics at each
opening of the trace_stat file because. The values are so updated between
two "cat trace_stat". And the tracer is free to lock its datas during the
iteration to keep consistent values.

Since it is almost always interesting to sort statisticals values to
address the problems by priority, this infrastructure provides a "sorting"
of the stat entries too if desired. A tracer has just to provide a
stat_cmp callback to compare two entries and the stat tracing
infrastructure will build a sorted list of the given entries.

A last callback, called stat_headers, can be implemented by a tracer to
output headers on its trace.

If one of these callbacks is changed on runtime, it just have to signal it
to the stat tracing API by calling the init_tracer_stat() helper.

Changes in V2:

- Fix a memory leak if the user opens multiple times the trace_stat file
  without closing it. Now we always free our list before rebuilding it.

Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
17 years agoftrace: change trace.c to use registered events
Steven Rostedt [Wed, 24 Dec 2008 04:24:13 +0000 (23:24 -0500)]
ftrace: change trace.c to use registered events

Impact: rework trace.c to use new event register API

Almost every ftrace event has to implement its output display in
trace.c through a different function. Some events did not handle
all the formats (trace, latency-trace, raw, hex, binary), and
this method does not scale well.

This patch converts the format functions to use the event API to
find the event and and print its format. Currently, we have
a print function for trace, latency_trace, raw, hex and binary.
A trace_nop_print is available if the event wants to avoid output
on a particular format.

Perhaps other tracers could use this in the future (like mmiotrace and
function_graph).

Signed-off-by: Steven Rostedt <srostedt@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
17 years agoftrace: set up trace event hash infrastructure
Steven Rostedt [Wed, 24 Dec 2008 04:24:12 +0000 (23:24 -0500)]
ftrace: set up trace event hash infrastructure

Impact: simplify/generalize/refactor trace.c

The trace.c file is becoming more difficult to maintain due to the
growing number of events. There is several formats that an event may
be printed. This patch sets up the infrastructure of an event hash to
allow for events to register how they should be printed.

Signed-off-by: Steven Rostedt <srostedt@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
17 years agoftrace: remove obsolete print continue functionality
Steven Rostedt [Wed, 24 Dec 2008 04:24:11 +0000 (23:24 -0500)]
ftrace: remove obsolete print continue functionality

Impact: cleanup, remove obsolete code

Now that the ring buffer used by ftrace allows for variable length
entries, we do not need the 'cont' feature of the buffer.  This code
makes other parts of ftrace more complex and by removing this it
simplifies the ftrace code.

Signed-off-by: Steven Rostedt <srostedt@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
17 years agogenirq: check chip->ack before calling
Wang Chen [Mon, 29 Dec 2008 05:35:11 +0000 (13:35 +0800)]
genirq: check chip->ack before calling

Impact: fix theoretical NULL dereference

The generic irq layer doesn't know whether irq_chip has ack routine on some
architectures or not. Upon that, before calling chip->ack, we should check
that it's not NULL.

Signed-off-by: Wang Chen <wangchen@cn.fujitsu.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
17 years agosparseirq: move __weak symbols into separate compilation unit
Yinghai Lu [Mon, 29 Dec 2008 00:01:13 +0000 (16:01 -0800)]
sparseirq: move __weak symbols into separate compilation unit

GCC has a bug with __weak alias functions: if the functions are in
the same compilation unit as their call site, GCC can decide to
inline them - and thus rob the linker of the opportunity to override
the weak alias with the real thing.

So move all the IRQ handling related __weak symbols to kernel/irq/chip.c.

Signed-off-by: Yinghai Lu <yinghai@kernel.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
17 years agoMerge branches 'tracing/docs', 'tracing/function-graph-tracer' and 'linus' into traci...
Ingo Molnar [Mon, 29 Dec 2008 10:25:11 +0000 (11:25 +0100)]
Merge branches 'tracing/docs', 'tracing/function-graph-tracer' and 'linus' into tracing/core

17 years ago[ARM] pxafb: add support for overlay1 and overlay2 as framebuffer devices
Eric Miao [Tue, 23 Dec 2008 09:49:43 +0000 (17:49 +0800)]
[ARM] pxafb: add support for overlay1 and overlay2 as framebuffer devices

PXA27x and later processors support overlay1 and overlay2 on-top of the
base framebuffer (although under-neath the base is also possible). They
support palette and no-palette RGB formats, as well as YUV formats (only
available on overlay2). These overlays have dedicated DMA channels and
behave in a similar way as a framebuffer.

This heavily simplified and re-structured work is based on the original
pxafb_overlay.c (which is pending for mainline merge for a long time).

The major problems with this pxafb_overlay.c are (if you are interested
in the history):

  1. heavily redundant (the control logics for overlay1 and overlay2 are
     actually identical except for some small operations,  which are now
     abstracted into a 'pxafb_layer_ops' structure)

  2. a lot of useless and un-tested code (two workarounds which are now
     fixed on mature silicons)

  3. cursorfb is actually useless, hardware cursor should not be used
     this way, and the code was actually un-tested for a long time.

The code in this patch should be self-explanatory, I tried to add minimum
comments. As said, this is basically simplified, there are several things
still on the pending list:

  1. palette mode is un-supported and un-tested (although re-using the
     palette code of the base framebuffer is actually very easy now with
     previous clean-up patches)

  2. fb_pan_display for overlay(s) is un-supported

  3. the base framebuffer can actually be abstracted by 'pxafb_layer' as
     well, which will help further re-use of the code and keep a better
     and consistent structure. (This is the reason I named it 'pxafb_layer'
     instead of 'pxafb_overlay' or something alike)

See Documentation/fb/pxafb.txt for additional usage information.

Signed-off-by: Eric Miao <eric.miao@marvell.com>
Cc: Rodolfo Giometti <giometti@linux.it>
Signed-off-by: Eric Miao <ycmiao@ycmiao-hp520.(none)>
17 years ago[ARM] pxafb: cleanup of the timing checking code
Eric Miao [Thu, 18 Dec 2008 14:51:54 +0000 (22:51 +0800)]
[ARM] pxafb: cleanup of the timing checking code

Signed-off-by: Eric Miao <eric.miao@marvell.com>
Signed-off-by: Eric Miao <ycmiao@ycmiao-hp520.(none)>
17 years ago[ARM] pxafb: cleanup of the color format manipulation code
Eric Miao [Thu, 18 Dec 2008 14:36:26 +0000 (22:36 +0800)]
[ARM] pxafb: cleanup of the color format manipulation code

1. introduce var_to_depth() to calculate the color depth including the
   transparency bit

2. the conversion from 'fb_var_screeninfo' to LCCR3 BPP bits can be re-
   used by overlays (in OVLxC1), thus an individual pxafb_var_to_bpp()
   has been separated out.

3. pxafb_setmode() should really set the color bitfields correctly at
   begining, introduce a pxafb_set_pixfmt() for this

4. allow user apps to specify color formats within fb_var_screeninfo,
   and checking of this in pxafb_check_var() has been simplified as
   below:

   a) pxafb_var_to_bpp() should pass - which means a basically correct
      bits_per_pixel and color depth setting
   b) the RGBT bitfields are then forced into supported values by
      pxafb_set_pixfmt()

Signed-off-by: Eric Miao <eric.miao@marvell.com>
Signed-off-by: Eric Miao <ycmiao@ycmiao-hp520.(none)>
17 years ago[ARM] pxafb: add palette format support for LCCR4_PAL_FOR_3
Eric Miao [Thu, 18 Dec 2008 14:10:00 +0000 (22:10 +0800)]
[ARM] pxafb: add palette format support for LCCR4_PAL_FOR_3

Add the palette format support for LCCR4_PAL_FOR_3, and fix the
issue of LCCR4 being never assigned.

Also remove the useless pxafb_set_truecolor().

Signed-off-by: Eric Miao <eric.miao@marvell.com>
Signed-off-by: Eric Miao <ycmiao@ycmiao-hp520.(none)>
17 years ago[ARM] pxafb: add support for FBIOPAN_DISPLAY by dma braching
Eric Miao [Wed, 17 Dec 2008 08:50:43 +0000 (16:50 +0800)]
[ARM] pxafb: add support for FBIOPAN_DISPLAY by dma braching

dma branching is enabled by extending the current setup_frame_dma()
function to allow a 2nd set of frame/palette dma descriptors to be
used.

As a result, pxafb_dma_buff.dma_desc[], pxafb_dma_buff.pal_desc[]
and pxafb_info.fdadr[] are doubled.

This allows maximum re-use of the current dma setup code, although
the pxafb_info.fdadr[xx] for FBRx register values looks a bit odd.

Signed-off-by: Eric Miao <eric.miao@marvell.com>
Signed-off-by: Eric Miao <ycmiao@ycmiao-hp520.(none)>
17 years ago[ARM] pxafb: allow pxafb_set_par() to start from arbitrary yoffset
Eric Miao [Wed, 17 Dec 2008 06:56:54 +0000 (14:56 +0800)]
[ARM] pxafb: allow pxafb_set_par() to start from arbitrary yoffset

Note the var->yres_virtual is only re-calculated from the fix.smem_len
when text mode acceleration is enabled (which is default), this is due
to the issue as Russell suggested below:

Previous experience of doing this with the X server and acornfb is that
it causes all sorts of problems - it seems to force the X server into
assuming that the framebuffer should be panned no matter what settings
you ask it for.

The recommended workaround (implemented in acornfb) is to only do these
kinds of adjustments if text mode acceleration is enabled.  IIRC, the X
server should be disabling text mode acceleration when it maps the
framebuffer.  I seem to remember that there are X servers which forget
to do that though.

Signed-off-by: Eric Miao <eric.miao@marvell.com>
Signed-off-by: Eric Miao <ycmiao@ycmiao-hp520.(none)>
17 years ago[ARM] pxafb: allow video memory size to be configurable
Eric Miao [Tue, 16 Dec 2008 03:54:34 +0000 (11:54 +0800)]
[ARM] pxafb: allow video memory size to be configurable

The amount of video memory size is decided according to the following
order:

1. <xres> x <yres> x <bits_per_pixel> by default, which is the backward
   compatible way

2. size specified in platform data

3. size specified in module parameter 'options' string or specified in
   kernel boot command line (see updated Documentation/fb/pxafb.txt)

And now since the memory is allocated from system memory, the pxafb_mmap
can be removed and the default fb_mmap() should be working all right.

Also, since we now have introduced the 'struct pxafb_dma_buff' for DMA
descriptors and palettes, the allocation can be separated cleanly.

NOTE: the LCD DMA actually supports chained transfer (i.e. page-based
transfers), to simplify the logic and keep the performance (with less
TLB misses when accessing from memory mapped user space), the memory
is allocated by alloc_pages_*() to ensures it's physical contiguous.

Signed-off-by: Eric Miao <eric.miao@marvell.com>
Signed-off-by: Eric Miao <ycmiao@ycmiao-hp520.(none)>
17 years ago[ARM] pxa: add document on the MFP design and how to use it
Eric Miao [Thu, 25 Dec 2008 09:19:02 +0000 (17:19 +0800)]
[ARM] pxa: add document on the MFP design and how to use it

Signed-off-by: Eric Miao <eric.miao@marvell.com>
17 years ago[ARM] sa1100_wdt: don't assume CLOCK_TICK_RATE to be a constant
Eric Miao [Wed, 24 Dec 2008 03:32:45 +0000 (11:32 +0800)]
[ARM] sa1100_wdt: don't assume CLOCK_TICK_RATE to be a constant

See description of commit:

   [ARM] rtc-sa1100: don't assume CLOCK_TICK_RATE to be a constant

for additional information.

Signed-off-by: Eric Miao <eric.miao@marvell.com>
17 years ago[ARM] rtc-sa1100: don't assume CLOCK_TICK_RATE to be a constant
Eric Miao [Thu, 18 Dec 2008 03:10:32 +0000 (11:10 +0800)]
[ARM] rtc-sa1100: don't assume CLOCK_TICK_RATE to be a constant

As Nicolas and Russell pointed out, CLOCK_TICK_RATE is no more
a constant on PXA when multiple processors and platforms are
selected, change TIMER_FREQ in rtc-sa1100.c into a variable.

Since the code to decide the clock tick rate is re-used from
timer.c, introduce a common get_clock_tick_rate() for this.

Signed-off-by: Eric Miao <eric.miao@marvell.com>
Acked-by: Nicolas Pitre <nico@marvell.com>
17 years ago[ARM] pxa/tavorevb: update board support (smartpanel LCD + keypad)
Eric Miao [Sat, 27 Dec 2008 16:40:29 +0000 (00:40 +0800)]
[ARM] pxa/tavorevb: update board support (smartpanel LCD + keypad)

Signed-off-by: Eric Miao <eric.miao@marvell.com>
17 years agoMerge branch 'topic/failslab' into for-linus
Pekka Enberg [Mon, 29 Dec 2008 09:47:05 +0000 (11:47 +0200)]
Merge branch 'topic/failslab' into for-linus

Conflicts:

mm/slub.c

Signed-off-by: Pekka Enberg <penberg@cs.helsinki.fi>
17 years agoMerge branches 'topic/fixes', 'topic/cleanups' and 'topic/documentation' into for...
Pekka Enberg [Mon, 29 Dec 2008 09:45:47 +0000 (11:45 +0200)]
Merge branches 'topic/fixes', 'topic/cleanups' and 'topic/documentation' into for-linus

17 years agoslub: avoid leaking caches or refcounts on sysfs error
David Rientjes [Thu, 18 Dec 2008 06:09:46 +0000 (22:09 -0800)]
slub: avoid leaking caches or refcounts on sysfs error

If a slab cache is mergeable and the sysfs alias cannot be added, the
target cache shall have its refcount decremented.  kmem_cache_create()
will return NULL, so if kmem_cache_destroy() is ever called on the target
cache, it will never be freed if the refcount has been leaked.

Likewise, if a slab cache is not mergeable and the sysfs link cannot be
added, the new cache shall be removed from the slab_caches list.
kmem_cache_create() will return NULL, so it will be impossible to call
kmem_cache_destroy() on it.

Both of these operations require slub_lock since refcount of all slab
caches and slab_caches are protected by the lock.

In the mergeable case, it would be better to restore objsize and offset
back to their original values, but this could race with another merge
since slub_lock was dropped.

Cc: Christoph Lameter <cl@linux-foundation.org>
Signed-off-by: David Rientjes <rientjes@google.com>
Signed-off-by: Pekka Enberg <penberg@cs.helsinki.fi>
17 years agoslab: Fix comment on #endif
Pascal Terjan [Tue, 25 Nov 2008 14:08:19 +0000 (15:08 +0100)]
slab: Fix comment on #endif

This #endif in slab.h is described as closing the inner block while it's for
the big CONFIG_NUMA one. That makes reading the code a bit harder.

This trivial patch fixes the comment.

Signed-off-by: Pascal Terjan <pterjan@mandriva.com>
Signed-off-by: Pekka Enberg <penberg@cs.helsinki.fi>
17 years agoslab: remove GFP_THISNODE clearing from alloc_slabmgmt()
Pekka Enberg [Wed, 26 Nov 2008 08:01:31 +0000 (10:01 +0200)]
slab: remove GFP_THISNODE clearing from alloc_slabmgmt()

Commit 6cb062296f73e74768cca2f3eaf90deac54de02d ("Categorize GFP flags")
left one call-site in alloc_slabmgmt() to clear GFP_THISNODE instead of
GFP_CONSTRAINT_MASK. Unfortunately, that ends up clearing __GFP_NOWARN
and __GFP_NORETRY as well which is not what we want. As the only caller
of alloc_slabmgmt() already clears GFP_CONSTRAINT_MASK before passing
local_flags to it, we can just remove the clearing of GFP_THISNODE.

This patch should fix spurious page allocation failure warnings on the
mempool_alloc() path. See the following URL for the original discussion
of the bug:

  http://lkml.org/lkml/2008/10/27/100

Acked-by: Christoph Lameter <cl@linux-foundation.org>
Reported-by: Miklos Szeredi <miklos@szeredi.hu>
Signed-off-by: Pekka Enberg <penberg@cs.helsinki.fi>
17 years agoslub: Add might_sleep_if() to slab_alloc()
OGAWA Hirofumi [Wed, 19 Nov 2008 12:23:59 +0000 (21:23 +0900)]
slub: Add might_sleep_if() to slab_alloc()

Currently SLUB doesn't warn about __GFP_WAIT. Add it into slab_alloc().

Acked-by: Christoph Lameter <cl@linux-foundation.org>
Signed-off-by: OGAWA Hirofumi <hirofumi@mail.parknet.co.jp>
Signed-off-by: Pekka Enberg <penberg@cs.helsinki.fi>
17 years agoMerge branch 'linus' into timers/hrtimers
Ingo Molnar [Mon, 29 Dec 2008 08:42:58 +0000 (09:42 +0100)]
Merge branch 'linus' into timers/hrtimers

Conflicts:
sound/drivers/pcsp/pcsp.c

Semantic conflict:

        sound/core/hrtimer.c

17 years agoSLUB: failslab support
Akinobu Mita [Tue, 23 Dec 2008 10:37:01 +0000 (19:37 +0900)]
SLUB: failslab support

Currently fault-injection capability for SLAB allocator is only
available to SLAB. This patch makes it available to SLUB, too.

[penberg@cs.helsinki.fi: unify slab and slub implementations]
Cc: Christoph Lameter <cl@linux-foundation.org>
Cc: Matt Mackall <mpm@selenic.com>
Signed-off-by: Akinobu Mita <akinobu.mita@gmail.com>
Signed-off-by: Pekka Enberg <penberg@cs.helsinki.fi>
17 years agoMerge current mainline tree into linux-omap tree
Tony Lindgren [Mon, 29 Dec 2008 08:51:06 +0000 (10:51 +0200)]
Merge current mainline tree into linux-omap tree

Merge branches 'master' and 'linus'

17 years agodrm/i915: fix modeset devname allocation + agp init return check.
Dave Airlie [Mon, 29 Dec 2008 06:35:02 +0000 (16:35 +1000)]
drm/i915: fix modeset devname allocation + agp init return check.

devname needs to be allocated before the irq is installed, so the
irq routines get the correct name in /proc.

Also check the return value from the AGP init function, and
fixup the exit points.

Signed-off-by: Dave Airlie <airlied@redhat.com>
17 years agodrm/i915: Remove redundant test in error path.
Julia Lawall [Sun, 21 Dec 2008 15:28:47 +0000 (16:28 +0100)]
drm/i915: Remove redundant test in error path.

The error path for object list being null is in the second goto target.

Signed-off-by: Julia Lawall <julia@diku.dk>
Signed-off-by: Eric Anholt <eric@anholt.net>
Signed-off-by: Dave Airlie <airlied@linux.ie>
17 years agodrm: Add a debug node for vblank state.
Eric Anholt [Sat, 20 Dec 2008 01:23:38 +0000 (17:23 -0800)]
drm: Add a debug node for vblank state.

Signed-off-by: Eric Anholt <eric@anholt.net>
Signed-off-by: Dave Airlie <airlied@linux.ie>
17 years agodrm: Avoid use-before-null-test on dev in drm_cleanup().
Eric Anholt [Fri, 19 Dec 2008 23:07:11 +0000 (15:07 -0800)]
drm: Avoid use-before-null-test on dev in drm_cleanup().

Signed-off-by: Eric Anholt <eric@anholt.net>
Signed-off-by: Dave Airlie <airlied@linux.ie>
17 years agodrm/i915: Don't print to dmesg when taking signal during object_pin.
Eric Anholt [Fri, 19 Dec 2008 22:47:48 +0000 (14:47 -0800)]
drm/i915: Don't print to dmesg when taking signal during object_pin.

This showed up in logs where people had a hung chip, so pinning was blocked
on the chip unpinning other buffers, and the X Server took its scheduler
signal during that time.

Signed-off-by: Eric Anholt <eric@anholt.net>
Signed-off-by: Dave Airlie <airlied@linux.ie>
17 years agodrm: pin new and unpin old buffer when setting a mode.
Kristian Høgsberg [Thu, 18 Dec 2008 03:14:46 +0000 (22:14 -0500)]
drm: pin new and unpin old buffer when setting a mode.

This removes the requirement for user space to pin a buffer before
setting a mode that is backed by the pixels from that buffer.

Signed-off-by: Kristian Høgsberg <krh@redhat.com>
Signed-off-by: Eric Anholt <eric@anholt.net>
Signed-off-by: Dave Airlie <airlied@linux.ie>
17 years agodrm/i915: un-EXPORT and make 'intelfb_panic' static
Hannes Eder [Fri, 19 Dec 2008 11:34:27 +0000 (12:34 +0100)]
drm/i915: un-EXPORT and make 'intelfb_panic' static

Fix this sparse warning:

  drivers/gpu/drm/i915/intel_fb.c:417:5: warning: symbol 'intelfb_panic' was not declared. Should it be static?

Signed-off-by: Hannes Eder <hannes@hanneseder.net>
Signed-off-by: Eric Anholt <eric@anholt.net>
Signed-off-by: Dave Airlie <airlied@linux.ie>
17 years agodrm/i915: Delete unused, pointless i915_driver_firstopen.
Eric Anholt [Fri, 19 Dec 2008 22:30:31 +0000 (14:30 -0800)]
drm/i915: Delete unused, pointless i915_driver_firstopen.

Thanks to Hannes Eder for pointing out that this code was dead according to
sparse.

Signed-off-by: Eric Anholt <eric@anholt.net>
Signed-off-by: Dave Airlie <airlied@linux.ie>
17 years agodrm/i915: fix sparse warnings: returning void-valued expression
Hannes Eder [Thu, 18 Dec 2008 20:24:18 +0000 (21:24 +0100)]
drm/i915: fix sparse warnings: returning void-valued expression

Signed-off-by: Hannes Eder <hannes@hanneseder.net>
Signed-off-by: Eric Anholt <eric@anholt.net>
Signed-off-by: Dave Airlie <airlied@linux.ie>
17 years agodrm/i915: fix sparse warnings: move 'extern' decls to header file
Hannes Eder [Thu, 18 Dec 2008 20:22:24 +0000 (21:22 +0100)]
drm/i915: fix sparse warnings: move 'extern' decls to header file

Move 'extern'-decls from "intel_dvo.c" to "dvo.h", as "dvo.h" is
included by and only by files where the symbols are either defined or
used.

Signed-off-by: Hannes Eder <hannes@hanneseder.net>
Signed-off-by: Eric Anholt <eric@anholt.net>
Signed-off-by: Dave Airlie <airlied@linux.ie>
17 years agodrm/i915: fix sparse warnings: make symbols static
Hannes Eder [Thu, 18 Dec 2008 20:18:47 +0000 (21:18 +0100)]
drm/i915: fix sparse warnings: make symbols static

Signed-off-by: Hannes Eder <hannes@hanneseder.net>
Signed-off-by: Eric Anholt <eric@anholt.net>
Signed-off-by: Dave Airlie <airlied@linux.ie>
17 years agodrm/i915: fix sparse warnings: declare one-bit bitfield as unsigned
Hannes Eder [Thu, 18 Dec 2008 14:09:00 +0000 (15:09 +0100)]
drm/i915: fix sparse warnings: declare one-bit bitfield as unsigned

Signed-off-by: Hannes Eder <hannes@hanneseder.net>
Signed-off-by: Eric Anholt <eric@anholt.net>
Acked-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Dave Airlie <airlied@linux.ie>
17 years agodrm/i915: Don't double-unpin buffers if we take a signal in evict_everything().
Eric Anholt [Wed, 10 Dec 2008 18:09:41 +0000 (10:09 -0800)]
drm/i915: Don't double-unpin buffers if we take a signal in evict_everything().

We haven't seen this in practice, but it was visible when looking at a bug
report from when i915_gem_evict_everything() was broken and would always
return error.

Signed-off-by: Eric Anholt <eric@anholt.net>
Signed-off-by: Dave Airlie <airlied@linux.ie>
17 years agodrm/i915: Fix fbcon setup to align display pitch to 64b.
Eric Anholt [Thu, 11 Dec 2008 01:23:00 +0000 (17:23 -0800)]
drm/i915: Fix fbcon setup to align display pitch to 64b.

This is required by the display plane, and fixes 1400x1050 laptop displays.

Signed-off-by: Eric Anholt <eric@anholt.net>
Signed-off-by: Dave Airlie <airlied@redhat.com>
17 years agodrm/i915: Add missing userland definitions for gem init/execbuffer.
Eric Anholt [Thu, 18 Dec 2008 06:32:14 +0000 (22:32 -0800)]
drm/i915: Add missing userland definitions for gem init/execbuffer.

fdo bug #19132.

Signed-off-by: Eric Anholt <eric@anholt.net>
Signed-off-by: Dave Airlie <airlied@redhat.com>
17 years agoi915/drm: provide compat defines for userspace for certain struct members.
Dave Airlie [Fri, 19 Dec 2008 05:07:46 +0000 (15:07 +1000)]
i915/drm: provide compat defines for userspace for certain struct members.

Painfully userspace started using new names that were never actually to be
used from the external repo.

Also fill out the gaps in the structure for old/new userspace compat

Add compat defines for these structs.

Signed-off-by: Dave Airlie <airlied@redhat.com>
17 years agodrm: drop DRM_IOCTL_MODE_REPLACEFB, add+remove works just as well.
Kristian H�gsberg [Thu, 18 Dec 2008 03:14:37 +0000 (13:14 +1000)]
drm: drop DRM_IOCTL_MODE_REPLACEFB, add+remove works just as well.

The replace fb ioctl replaces the backing buffer object for a modesetting
framebuffer object.  This can be acheived by just creating a new
framebuffer backed by the new buffer object, setting that for the crtcs
in question and then removing the old framebuffer object.

Signed-off-by: Kristian Hogsberg <krh@redhat.com>
Acked-by: Jakob Bornecrantz <jakob@vmware.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>