8 years agodebug stuff stuff/pipelined_fencing
Daniel Vetter [Thu, 11 Feb 2010 17:15:02 +0000 (18:15 +0100)]
debug stuff



8 years agodrm/i915: pipelined fencing, part 2: fence setup
Daniel Vetter [Fri, 12 Feb 2010 08:26:33 +0000 (09:26 +0100)]
drm/i915: pipelined fencing, part 2: fence setup

This wires the fence setup part up. I've also implemented pipelined
fence setup for 4th gen chips - this may be useful when the pageflip
code switches to pipelined fencing.

With this change, every batchbuffer can use all available fences (save
pinned and scanout, of course) without ever stalling the gpu!

v2: Code in intel_display.c tried to be clever and not uncondionally
call i915_gem_object_get_fence_reg, which screws up fence state.

v3: Work around hw oddities. I've found out that the official
intel docs don't tell the whole story about MI_LOAD_REGISTER_IMM.
I've documented my findings in i915_reg.h

v4: Adapt to Sandybridge support.

Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
8 years agodrm/i915: infrastructure to track pipelined fence setup
Daniel Vetter [Fri, 12 Feb 2010 08:55:54 +0000 (09:55 +0100)]
drm/i915: infrastructure to track pipelined fence setup

Fenced gtt access by the cpu needs to wait for the gpu to have updated
the fence reg if it was pipelined. This patch adds the necessary

I'm not too fond of the code in i915_gem_retire_requests, but I
think that a new list to track fences a la objects for retiring
is overkill.

v2: Retire fences before request and explain in a big comment why this
leads to setup_seqno inconsistencies. Also a few new BUG_ONs if used
to check consistency (and catch the above-mentioned problem).

v3: Like for pipelined fence use, ensure that a fence doesn't outlive
it's object under certain perculiar error-conditions in

v4: Add some more phat comments to explain what the changes in v3 actually
fixed. Also fix up a cache-coherency problem in case userspace ever mixes
fenced with non-fenced access in different batchbuffers.

v5: Adapt to Sandybridge support.

Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
8 years agodrm/i915: use gpu fence usage accounting for more accurate fence lru
Daniel Vetter [Mon, 8 Feb 2010 15:10:52 +0000 (16:10 +0100)]
drm/i915: use gpu fence usage accounting for more accurate fence lru

Now that we don't superflously wait for the gpu when stealing a fence,
only update the fence lru when the gpu actually needs the fence.

Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
8 years agodrm/i915: optimize fence-stealing for execbuf2
Daniel Vetter [Fri, 12 Feb 2010 08:36:10 +0000 (09:36 +0100)]
drm/i915: optimize fence-stealing for execbuf2

Only wait for the gpu when the gpu is actually using this fence reg.
This should help with mixed execbuf2/old execbuf userspace in the
face of gtt access (cpu fallbacks).

v2: This was too buggy to describe in words. Fix it.

Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
8 years agodrm/i915: pipelined fencing, part 1: fence stealing
Daniel Vetter [Fri, 12 Feb 2010 08:10:04 +0000 (09:10 +0100)]
drm/i915: pipelined fencing, part 1: fence stealing

This is mostly just infrastructure to thread the pipelined
argument through all relevant functions. Nothing is really
pipelined, yet, because i915_gem_object_get_fence_reg waits
for the fenced gpu access when pipelined == 1.

Pipelined fence clearing is a no-op because we'll overwrite
the fence reg anyway in i915_gem_object_get_fence_reg. The only
other caller of i915_gem_clear_fence_reg is the unbind code,
which is always synchronous.

Only mark execbuffer fences as pipelined for now and but a FIXME
comment for the pageflip stuff. Doing pipelined fencing for pageflip
needs some more usage tracking and I'm not sure this is worth it
for the fence will be pinned for an indetermined long time.

v2: Change clear_fence in i915_gem_object_unbind to put_fence.
With the pipelining it 'll be no longer sufficient to simply clear
the fence reg, some more synchronization well be needed later.

v3: Adapt to Sandybridge support.

Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
8 years agodrm/i915: track gpu fence usage more precisely
Daniel Vetter [Thu, 11 Feb 2010 17:11:25 +0000 (18:11 +0100)]
drm/i915: track gpu fence usage more precisely

With execbuf2, not every buffer on the active list that has a
fence reg allocated necessarily needs it. Track this usage more

Instead of wasting another 30 bits in struct drm_i915_gem_object,
I've decided to put all flags and small integer fields together
and convert them into C bitfields.

v2: Fixes and small cleanups for the fence state tracking.
v3: Add current_execbuffer_needs_fencing to correctly recover
from errors in i915_gem_do_execbuffer. Also clarify the usage
of fenced_gpu_access.

Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
8 years agodrm/i915: drop i915_add_request right in front of i915_wait_request
Daniel Vetter [Thu, 11 Feb 2010 21:53:20 +0000 (22:53 +0100)]
drm/i915: drop i915_add_request right in front of i915_wait_request

... take advantage of the new implicit request issuing of

Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
8 years agodrm/i915: move the wait_rendering call into flush_gpu_write_domain
Daniel Vetter [Thu, 11 Feb 2010 21:37:04 +0000 (22:37 +0100)]
drm/i915: move the wait_rendering call into flush_gpu_write_domain

One caller (for the pageflip support) want's a purely pipelined flush.
Distinguish this case by a new parameter. This will also be useful
later on for pipelined fencing.

v2: Simplify the code by depending upon the implicit request emitting
of i915_wait_request.

Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
8 years agodrm/i915: drop seqno argument from i915_gem_object_move_to_active
Daniel Vetter [Thu, 11 Feb 2010 21:16:02 +0000 (22:16 +0100)]
drm/i915: drop seqno argument from i915_gem_object_move_to_active

By moving one i915_add_request we can soley depend on the new
auto-seqno-numbering behaviour.

Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
8 years agodrm/i915: kill an no longer necessary BUG_ON
Daniel Vetter [Tue, 2 Feb 2010 21:54:18 +0000 (22:54 +0100)]
drm/i915: kill an no longer necessary BUG_ON

i915_gem_object_move_to_active can handle zero seqno's for us now.
And not emitting a request is not fatal here - we'll try to emit
a new one if we have to wait for some rendering to complete.

In case this assumption ever gets accidently broken, there's already
a BUG_ON to catch it in i915_do_wait_request.

So just silently ignore ENOMEM here instead of screwing up the whole

Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
8 years agodrm/i915: move flushing list processing to i915_retire_commands
Daniel Vetter [Thu, 11 Feb 2010 21:29:04 +0000 (22:29 +0100)]
drm/i915: move flushing list processing to i915_retire_commands

... instead of threading flush_domains through the execbuf code to

With this change 2 small cleanups are possible (linewise the majority
of the patch):

- The flush_domains parameter of i915_add_request is always 0. Drop it
  and the corresponding logic.
- Dito for the seqno param of i915_gem_process_flushing_list.

Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
8 years agodrm/i915: only one interrupt per batchbuffer is not enough!
Daniel Vetter [Tue, 2 Feb 2010 16:08:37 +0000 (17:08 +0100)]
drm/i915: only one interrupt per batchbuffer is not enough!

Previously I thought that one interrupt per batchbuffer should be
enough. Now tedious benchmarking showed this to be wrong.

Therefore track whether any commands have been isssued with a future
seqno (like pipelined fencing changes or flushes). If this is the case
emit a request before issueing the batchbuffer.

Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
8 years agodrm/i915: move flushing list processing to i915_gem_flush
Daniel Vetter [Thu, 11 Feb 2010 21:19:40 +0000 (22:19 +0100)]
drm/i915: move flushing list processing to i915_gem_flush

Now that we can move objects to the active list without already having
emitted a request, move the flushing list handling into i915_gem_flush.
This makes more sense and allows to drop a few i915_add_request calls
that are not strictly necessary.

Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
8 years agodrm/i915: allow lazy emitting of requests
Daniel Vetter [Thu, 11 Feb 2010 21:13:59 +0000 (22:13 +0100)]
drm/i915: allow lazy emitting of requests

Sometimes (like when flushing in preparation of batchbuffer execution)
we know that we'll emit a request but haven't yet done so. Allow this
case by simply taking the next seqno by default. Ensure that a request
is eventually emitted before waiting for an request by issuing it
in i915_wait_request iff this is not yet done.

Also replace one open-coded version of i915_gem_object_wait_rendering,
to prevent future code-diversion.

Chris Wilson asked me to explain and clarify what this patch does and why.
Here it goes:

Old way of moving objects onto the active list and associating them with a

1. i915_add_request + store the returned seqno somewhere
2. i915_gem_object_move_to_active (with the stored seqno as parameter)

For the current users, this is all fine. But I'd like to associate objects
(and fence regs) with the batchbuffer request deep down in the execbuf
call-chain. I thought about three ways of implementing this.

a) Don't care, just emit request when we need a new seqno. When heavily
pipelining fence reg changes, this would have caused tons of superflous
request (and corresponding irqs).

b) Thread all changed fences, objects, whatever through the execbuf-maze,
so that when we emit a request, we can store the new seqno at all the right

c) Kill that seqno-threading-around business by simply storing the next
seqno, i.e. allow 2. to be done before 1. in the above sequence.

I've decided to implement c) (in this patch). The following patches are
just fall-out that resulted from this small conceptual change.

* We can handle the flushing list processing where we actually emit a flush
  (i915_gem_flush and i915_retire_commands) instead of in i915_add_request.
  The code makes IMHO more sense this way (and i915_add_request looses the
  flush_domains parameter, obviously).

* We can avoid emitting unnecessary requests. IMHO there's no point in
  emitting more than one request per batchbuffer (with or without an
  corresponding irq).

* By enforcing 2. before 1. ordering in the above sequence the seqno
  argument of i915_gem_object_move_to_active is redundant and can be

v2: Now i915_wait_request issues request if it is not yet emitted.
Also introduce i915_gem_next_request_seqno(dev) just in case we ever
need to do some prep work before using a new seqno.

Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
8 years agodrm/i915: move fence lru to struct drm_i915_fence_reg
Daniel Vetter [Mon, 1 Feb 2010 16:35:46 +0000 (17:35 +0100)]
drm/i915: move fence lru to struct drm_i915_fence_reg

This lru tracks fences, not objects, so move it where it belongs. Later
changes will make that disdinction more clear, atm this is just cosmetics.

Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
8 years agoEnable distcc
Daniel Vetter [Tue, 2 Feb 2010 22:03:53 +0000 (23:03 +0100)]
Enable distcc

8 years agodrm/i915: Use a dmi quirk to skip a broken SDVO TV output.
Zhao Yakui [Mon, 8 Feb 2010 13:35:12 +0000 (21:35 +0800)]
drm/i915: Use a dmi quirk to skip a broken SDVO TV output.

This IBM system has a multi-function SDVO card that reports both VGA
and TV, but the system has no TV connector.  The TV connector always
reported as connected, which would lead to poor modesetting choices.


Signed-off-by: Zhao Yakui <yakui.zhao@intel.com>
Tested-by: Vance <liangghv@sg.ibm.com>
Cc: stable@kernel.org
Signed-off-by: Eric Anholt <eric@anholt.net>
8 years agodrm/i915: enable/disable LVDS port at DPMS time
Jesse Barnes [Thu, 11 Feb 2010 20:41:05 +0000 (12:41 -0800)]
drm/i915: enable/disable LVDS port at DPMS time

It would be good to disable the LVDS port when we shut down the panel
to save power.  We haven't done so until now because we had trouble
getting the right LVDS parameters from the BIOS.  I think we're past
that now, so enabling and disabling the port should be safe, though it
would probably be made cleaner with some additional changes to the
display code, where we also bang on the LVDS reg to set the pairing
correctly etc.

Seems to save a bit of power (up to 300mW in my basic wattsup
meter testing).

Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Eric Anholt <eric@anholt.net>
8 years agodrm/i915: check for multiple write domains in pin_and_relocate
Daniel Vetter [Fri, 19 Feb 2010 10:52:02 +0000 (11:52 +0100)]
drm/i915: check for multiple write domains in pin_and_relocate

The assumption that an object has only ever one write domain is deeply
threaded into gem (it's even encoded the the singular of the variable
name). Don't let userspace screw us over.

Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Eric Anholt <eric@anholt.net>
8 years agodrm/i915: clean-up i915_gem_flush_gpu_write_domain
Daniel Vetter [Fri, 19 Feb 2010 10:52:01 +0000 (11:52 +0100)]
drm/i915: clean-up i915_gem_flush_gpu_write_domain

Now that we have an exact gpu write domain tracking, we don't need
to move objects to the active list ourself. i915_add_request will
take care of that under all circumstances.

Idea stolen from a patch by Chris Wilson <chris@chris-wilson.co.uk>.

Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Eric Anholt <eric@anholt.net>
8 years agodrm/i915: reuse i915_gpu_idle helper
Daniel Vetter [Fri, 19 Feb 2010 10:52:00 +0000 (11:52 +0100)]
drm/i915: reuse i915_gpu_idle helper

We have it, so use it. This required moving the function to avoid
a forward declaration.

Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Eric Anholt <eric@anholt.net>
8 years agodrm/i915: ensure lru ordering of fence_list
Daniel Vetter [Fri, 19 Feb 2010 10:51:59 +0000 (11:51 +0100)]
drm/i915: ensure lru ordering of fence_list

The fence_list should be lru ordered for otherwise we might try
to steal a fence reg from an active object even though there are
fences from inactive objects available. lru ordering was obeyed
for gpu access everywhere save when moving dirty objects from
flushing_list to active_list.

Fixing this cause the code to indent way to much, so I've extracted
the flushing_list processing logic into its on function.

Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Eric Anholt <eric@anholt.net>
8 years agodrm/i915: extract fence stealing code
Daniel Vetter [Fri, 19 Feb 2010 10:51:58 +0000 (11:51 +0100)]
drm/i915: extract fence stealing code

The spaghetti logic in there tripped up my brain's code parser for a
few secs. Prevent this from happening again by extracting the fence
stealing code into a seperate functions. IMHO this slightly clears up
the code flow.

v2: Beautified according to ickle's comments.
v3: ickle forgot to flush his comment queue ... Now there's also a
we-are-paranoid BUG_ON in there.
v4: I've forgotten to switch on my brain when doing v3. Now the BUG_ON
actually checks something useful.
v5: Clean up a stale comment as noted by Eric Anholt.

Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Eric Anholt <eric@anholt.net>
8 years agodrm/i915: fixup active list locking in object_unbind
Daniel Vetter [Fri, 19 Feb 2010 10:51:57 +0000 (11:51 +0100)]
drm/i915: fixup active list locking in object_unbind

All other accesses take this spinlock, so do this here, too.

Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Eric Anholt <eric@anholt.net>
8 years agodrm/i915: reuse i915_gem_object_put_fence_reg for fence stealing code
Daniel Vetter [Fri, 19 Feb 2010 10:51:56 +0000 (11:51 +0100)]
drm/i915: reuse i915_gem_object_put_fence_reg for fence stealing code

This has a few functional changes against the old code:

* a few more unnecessary loads and stores to the drm_i915_fence_reg
  objects. Also an unnecessary store to the hw fence register.

* zaps any userspace mappings before doing other flushes. Only changes
  anything when userspace does racy stuff against itself.

* also flush GTT domain. This is a noop, but still try to keep the
  bookkeeping correct.

Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Eric Anholt <eric@anholt.net>
8 years agodrm/i915: Add dependency on the intel agp module
Zhenyu Wang [Tue, 23 Feb 2010 06:05:24 +0000 (14:05 +0800)]
drm/i915: Add dependency on the intel agp module

See http://bugzilla.kernel.org/show_bug.cgi?id=15021

Make sure that the appropriate AGP module is loaded and probed before
trying to set up the DRM.  The DRM already depends on the AGP core,
but in this case we know the specific AGP driver we need too, and can
help users avoid the trap of loading the AGP driver after the DRM

Signed-off-by: Zhenyu Wang <zhenyuw@linux.intel.com>
Signed-off-by: Eric Anholt <eric@anholt.net>
8 years agodrm/i915: More s/IS_IRONLAKE/HAS_PCH_SPLIT for Sandybridge.
Eric Anholt [Fri, 29 Jan 2010 00:45:52 +0000 (16:45 -0800)]
drm/i915: More s/IS_IRONLAKE/HAS_PCH_SPLIT for Sandybridge.

I think this is pretty much correct.  Not really tested.

Signed-off-by: Eric Anholt <eric@anholt.net>
8 years agodrm/i915: Correct the Sandybridge chipset info structs.
Eric Anholt [Fri, 29 Jan 2010 00:13:29 +0000 (16:13 -0800)]
drm/i915: Correct the Sandybridge chipset info structs.

Disables CXSR until it's done, and sets the mobile bit on mobile.

Signed-off-by: Eric Anholt <eric@anholt.net>
8 years agodrm/i915: Disable the hangcheck reset on Sandybridge until we add support.
Eric Anholt [Fri, 8 Jan 2010 22:25:16 +0000 (14:25 -0800)]
drm/i915: Disable the hangcheck reset on Sandybridge until we add support.

Signed-off-by: Eric Anholt <eric@anholt.net>
8 years agodrm/i915: Add a new mobile Sandybridge PCI ID.
Eric Anholt [Thu, 7 Jan 2010 23:08:18 +0000 (15:08 -0800)]
drm/i915: Add a new mobile Sandybridge PCI ID.

Signed-off-by: Eric Anholt <eric@anholt.net>
8 years agoagp/intel: Add a new Sandybridge HB/IG PCI ID combo.
Eric Anholt [Fri, 8 Jan 2010 00:21:46 +0000 (16:21 -0800)]
agp/intel: Add a new Sandybridge HB/IG PCI ID combo.

Signed-off-by: Eric Anholt <eric@anholt.net>
8 years agodrm/i915, agp/intel: Fix stolen memory size on Sandybridge
Zhenyu Wang [Tue, 10 Nov 2009 17:25:25 +0000 (01:25 +0800)]
drm/i915, agp/intel: Fix stolen memory size on Sandybridge

New memory control config reg at 0x50 should be used for stolen
memory size detection on Sandybridge.

Signed-off-by: Zhenyu Wang <zhenyuw@linux.intel.com>
Signed-off-by: Eric Anholt <eric@anholt.net>
8 years agodrm/i915: Correct locking in the modesetting failure path, fixing a BUG_ON.
Eric Anholt [Mon, 9 Nov 2009 22:57:34 +0000 (14:57 -0800)]
drm/i915: Correct locking in the modesetting failure path, fixing a BUG_ON.

Signed-off-by: Eric Anholt <eric@anholt.net>
8 years agodrm/i915: Disable the surface tile swizzling on Sandybridge.
Eric Anholt [Thu, 5 Nov 2009 23:30:35 +0000 (15:30 -0800)]
drm/i915: Disable the surface tile swizzling on Sandybridge.

I can't explain this, except that it makes my display correct.

Signed-off-by: Eric Anholt <eric@anholt.net>
8 years agoagp/intel: Use a non-reserved value for the cache field of the PTEs.
Eric Anholt [Mon, 2 Nov 2009 23:33:05 +0000 (15:33 -0800)]
agp/intel: Use a non-reserved value for the cache field of the PTEs.

I don't know if this is what we'll want to be using long term, we'll see.

Signed-off-by: Eric Anholt <eric@anholt.net>
8 years agodrm/i915: Fix sandybridge status page setup.
Eric Anholt [Mon, 2 Nov 2009 20:08:22 +0000 (12:08 -0800)]
drm/i915: Fix sandybridge status page setup.

The register's moved to the same location as the one for the BCS, it seems.

Signed-off-by: Eric Anholt <eric@anholt.net>
8 years agodrm/i915: Set up fence registers on sandybridge.
Eric Anholt [Mon, 26 Oct 2009 23:44:17 +0000 (16:44 -0700)]
drm/i915: Set up fence registers on sandybridge.

Signed-off-by: Eric Anholt <eric@anholt.net>
8 years agodrm/i915: Add initial bits for VGA modesetting bringup on Sandybridge.
Eric Anholt [Thu, 22 Oct 2009 23:11:14 +0000 (16:11 -0700)]
drm/i915: Add initial bits for VGA modesetting bringup on Sandybridge.

Signed-off-by: Eric Anholt <eric@anholt.net>
8 years agoagp/intel: Add support for Sandybridge.
Eric Anholt [Thu, 22 Oct 2009 23:10:52 +0000 (16:10 -0700)]
agp/intel: Add support for Sandybridge.

Signed-off-by: Eric Anholt <eric@anholt.net>
8 years agoMerge remote branch 'korg/drm-radeon-testing' into drm-next-stage
Dave Airlie [Thu, 25 Feb 2010 03:44:04 +0000 (13:44 +1000)]
Merge remote branch 'korg/drm-radeon-testing' into drm-next-stage

* korg/drm-radeon-testing: (62 commits)
  drm/radeon/kms: update new pll algo
  drm/radeon/kms: add support for square microtiles on r3xx-r5xx
  drm/radeon/kms: force pinning buffer into visible VRAM
  drm/radeon/kms/evergreen: fix typo in cursor code
  drm/radeon/kms: implement reading active PCIE lanes on R600+
  drm/radeon/kms: for downclocking non-mobility check PERFORMANCE state
  drm/radeon/kms: simplify storing current and requested PM mode
  drm/radeon: fixes for r6xx/r7xx gfx init
  drm/radeon/rv740: fix backend setup
  drm/radeon/kms: fix R3XX/R4XX memory controller initialization
  [rfc] drm/radeon/kms: pm debugging check for vbl.
  drm/radeon: Fix memory allocation failures in the preKMS command stream checking.
  drm: Add generic multipart buffer.
  drm/radeon/kms: simplify memory controller setup V2
  drm/radeon: Add asic hook for dma copy to r200 cards.
  drm/radeon/kms: Create asic structure for r300 pcie cards.
  drm/radeon/kms: remove unused r600_gart_clear_page
  drm/radeon/kms: remove HDP flushes from fence emit (v2)
  drm/radeon/kms: add LVDS pll quirk for Dell Studio 15
  drm/radeon/kms: simplify picking power state


8 years agoMerge remote branch 'anholt/drm-intel-next' into drm-next-stage
Dave Airlie [Thu, 25 Feb 2010 03:39:36 +0000 (13:39 +1000)]
Merge remote branch 'anholt/drm-intel-next' into drm-next-stage

* anholt/drm-intel-next:
  drm/i915: Record batch buffer following GPU error
  drm/i915: give up on 8xx lid status
  drm/i915: reduce some of the duplication of tiling checking
  drm/i915: blow away userspace mappings before fence change
  drm/i915: move a gtt flush to the correct place
  agp/intel: official names for Pineview and Ironlake
  drm/i915: overlay: drop superflous gpu flushes
  drm/i915: overlay: nuke readback to flush wc caches
  drm/i915: provide self-refresh status in debugfs
  drm/i915: provide FBC status in debugfs
  drm/i915: fix drps disable so unload & re-load works
  drm/i915: Fix OGLC performance regression on 945
  drm/i915: Deobfuscate the render p-state obfuscation
  drm/i915: add dynamic performance control support for Ironlake
  drm/i915: enable memory self refresh on 9xx
  drm/i915: Don't reserve compatibility fence regs in KMS mode.
  drm/i915: Keep MCHBAR always enabled
  drm/i915: Replace open-coded eviction in i915_gem_idle()

8 years agoMerge remote branch 'korg/drm-core-next' into drm-next-stage
Dave Airlie [Thu, 25 Feb 2010 03:39:29 +0000 (13:39 +1000)]
Merge remote branch 'korg/drm-core-next' into drm-next-stage

* korg/drm-core-next:
  drm/ttm: handle OOM in ttm_tt_swapout
  drm/radeon/kms/atom: fix shr/shl ops
  drm/kms: fix spelling of "CLOCK"
  drm/kms: fix fb_changed = true else statement
  drivers/gpu/drm/drm_fb_helper.c: don't use private implementation of atoi()
  drm: switch all GEM/KMS ioctls to unlocked ioctl status.
  Use drm_gem_object_[handle_]unreference_unlocked where possible
  drm: introduce drm_gem_object_[handle_]unreference_unlocked

8 years agodrm/ttm: handle OOM in ttm_tt_swapout
Maarten Maathuis [Sat, 20 Feb 2010 02:22:21 +0000 (03:22 +0100)]
drm/ttm: handle OOM in ttm_tt_swapout

- Without this change I get a general protection fault.
- Also use PTR_ERR where applicable.

Signed-off-by: Maarten Maathuis <madman2003@gmail.com>
Reviewed-by: Dave Airlie <airlied@redhat.com>
Acked-by: Thomas Hellstrom <thellstrom@vmware.com>
Cc: stable@kernel.org
Signed-off-by: Dave Airlie <airlied@redhat.com>
8 years agodrm/radeon/kms/atom: fix shr/shl ops
Alex Deucher [Wed, 24 Feb 2010 02:56:12 +0000 (21:56 -0500)]
drm/radeon/kms/atom: fix shr/shl ops

The whole attribute table is valid for
shr/shl ops.

Fixes fdo bug 26668

Signed-off-by: Alex Deucher <alexdeucher@gmail.com>
Cc: stable@kernel.org
Signed-off-by: Dave Airlie <airlied@redhat.com>
8 years agodrm/radeon/kms: update new pll algo
Alex Deucher [Tue, 23 Feb 2010 08:24:38 +0000 (03:24 -0500)]
drm/radeon/kms: update new pll algo

- add support for pre-avivo chips
- add support for fixed post/ref dividers
- add support for non-fractional fb dividers

By default avivo chips use the new algo and
pre-avivo chips use the old algo. Use the "new_pll"
module option to toggle between them.

Signed-off-by: Alex Deucher <alexdeucher@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
8 years agodrm/radeon/kms: add support for square microtiles on r3xx-r5xx
Marek Olšák [Sun, 14 Feb 2010 06:10:10 +0000 (07:10 +0100)]
drm/radeon/kms: add support for square microtiles on r3xx-r5xx

Signed-off-by: Marek Olšák <maraeo@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
8 years agodrm/radeon/kms: force pinning buffer into visible VRAM
Jerome Glisse [Fri, 19 Feb 2010 14:33:54 +0000 (14:33 +0000)]
drm/radeon/kms: force pinning buffer into visible VRAM

This patch properly set visible VRAM and enforce any pinned buffer
to be into visible VRAM. We might later add a flag to release this
constraint for some newer hw more clever than previous.

Signed-off-by: Jerome Glisse <jglisse@redhat.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
8 years agodrm/radeon/kms/evergreen: fix typo in cursor code
Alex Deucher [Tue, 23 Feb 2010 18:12:43 +0000 (13:12 -0500)]
drm/radeon/kms/evergreen: fix typo in cursor code

cursor x/y are surface relative.

fixes fdo bug 26551

Signed-off-by: Alex Deucher <alexdeucher@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
8 years agoLinux 2.6.33
Linus Torvalds [Wed, 24 Feb 2010 18:52:17 +0000 (10:52 -0800)]
Linux 2.6.33

8 years agoMerge branch 'urgent' of git://git.kernel.org/pub/scm/linux/kernel/git/kyle/parisc-2.6
Linus Torvalds [Wed, 24 Feb 2010 18:51:21 +0000 (10:51 -0800)]
Merge branch 'urgent' of git://git./linux/kernel/git/kyle/parisc-2.6

* 'urgent' of git://git.kernel.org/pub/scm/linux/kernel/git/kyle/parisc-2.6:
  parisc: Set PCI CLS early in boot.

8 years agoMerge branch 'release' of git://git.kernel.org/pub/scm/linux/kernel/git/aegl/linux-2.6
Linus Torvalds [Wed, 24 Feb 2010 18:51:04 +0000 (10:51 -0800)]
Merge branch 'release' of git://git./linux/kernel/git/aegl/linux-2.6

* 'release' of git://git.kernel.org/pub/scm/linux/kernel/git/aegl/linux-2.6:
  [IA64] Fix broken sn2 build

8 years agoparisc: Set PCI CLS early in boot.
Carlos O'Donell [Mon, 22 Feb 2010 23:25:59 +0000 (23:25 +0000)]
parisc: Set PCI CLS early in boot.

Set the PCI CLS early in the boot process to prevent
device failures. In pcibios_set_master use the new
pci_cache_line_size instead of a hard-coded value.

Signed-off-by: Carlos O'Donell <carlos@codesourcery.com>
Reviewed-by: Grant Grundler <grundler@google.com>
Signed-off-by: Kyle McMartin <kyle@redhat.com>
8 years agoMerge branch 'for-linus' of git://git.monstr.eu/linux-2.6-microblaze
Linus Torvalds [Wed, 24 Feb 2010 15:43:02 +0000 (07:43 -0800)]
Merge branch 'for-linus' of git://git.monstr.eu/linux-2.6-microblaze

* 'for-linus' of git://git.monstr.eu/linux-2.6-microblaze:
  microblaze: Fix out_le32() macro
  microblaze: Fix cache loop function for cache range

8 years agoMerge branch 'for-linus' of git://git.kernel.dk/linux-2.6-block
Linus Torvalds [Wed, 24 Feb 2010 15:42:42 +0000 (07:42 -0800)]
Merge branch 'for-linus' of git://git.kernel.dk/linux-2.6-block

* 'for-linus' of git://git.kernel.dk/linux-2.6-block:
  Revert "block: improve queue_should_plug() by looking at IO depths"

8 years agomicroblaze: Fix out_le32() macro
Steven J. Magnani [Mon, 22 Feb 2010 15:25:42 +0000 (09:25 -0600)]
microblaze: Fix out_le32() macro

Trailing semicolon causes compilation involving out_le32() to fail.

Signed-off-by: Steven J. Magnani <steve@digidescorp.com>
Signed-off-by: Michal Simek <monstr@monstr.eu>
8 years agomicroblaze: Fix cache loop function for cache range
Michal Simek [Mon, 15 Feb 2010 09:50:42 +0000 (10:50 +0100)]
microblaze: Fix cache loop function for cache range

I create wrong asm code but none test shows that this part of code is wrong.
I am not convinces that were good idea to create asm optimized macros
for caches. The reason is that there is not optimization with previous code
that's why make sense to add old code and do some benchmarking which
functions are faster.

Signed-off-by: Michal Simek <monstr@monstr.eu>
8 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net-2.6
Linus Torvalds [Wed, 24 Feb 2010 03:44:07 +0000 (19:44 -0800)]
Merge git://git./linux/kernel/git/davem/net-2.6

* git://git.kernel.org/pub/scm/linux/kernel/git/davem/net-2.6:
  net: bug fix for vlan + gro issue
  tc35815: Remove a wrong netif_wake_queue() call which triggers BUG_ON
  cdc_ether: new PID for Ericsson C3607w to the whitelist (resubmit)
  IPv6: better document max_addresses parameter
  MAINTAINERS: update mv643xx_eth maintenance status
  e1000: Fix DMA mapping error handling on RX
  iwlwifi: sanity check before counting number of tfds can be free
  iwlwifi: error checking for number of tfds in queue
  iwlwifi: set HT flags after channel in rxon

8 years agonet: bug fix for vlan + gro issue
Ajit Khaparde [Tue, 16 Feb 2010 20:25:43 +0000 (20:25 +0000)]
net: bug fix for vlan + gro issue

Traffic (tcp) doesnot start on a vlan interface when gro is enabled.
Even the tcp handshake was not taking place.
This is because, the eth_type_trans call before the netif_receive_skb
in napi_gro_finish() resets the skb->dev to napi->dev from the previously
set vlan netdev interface. This causes the ip_route_input to drop the
incoming packet considering it as a packet coming from a martian source.

I could repro this on (stable) and 2.6.33-rc7.
With this fix, the traffic starts and the test runs fine on both vlan
and non-vlan interfaces.

CC: Herbert Xu <herbert@gondor.apana.org.au>
CC: Patrick McHardy <kaber@trash.net>
Signed-off-by: Ajit Khaparde <ajitk@serverengines.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoMerge branch 'release' of git://git.kernel.org/pub/scm/linux/kernel/git/lenb/linux...
Linus Torvalds [Wed, 24 Feb 2010 02:15:05 +0000 (18:15 -0800)]
Merge branch 'release' of git://git./linux/kernel/git/lenb/linux-acpi-2.6

* 'release' of git://git.kernel.org/pub/scm/linux/kernel/git/lenb/linux-acpi-2.6:
  ACPI: Be in TS_POLLING state during mwait based C-state entry
  ACPI: Fix regression where _PPC is not read at boot even when ignore_ppc=0
  acer-wmi: Respect current backlight level when loading

8 years agoMerge branch 'drm-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/airlied...
Linus Torvalds [Wed, 24 Feb 2010 02:13:34 +0000 (18:13 -0800)]
Merge branch 'drm-linus' of git://git./linux/kernel/git/airlied/drm-2.6

* 'drm-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/airlied/drm-2.6:
  drm/vmwgfx: Fix queries if no dma buffer thrashing is occuring.
  drm/nv50: fix vram ptes on IGPs to point at stolen system memory
  drm/nv50: fix instmem binding on IGPs to point at stolen system memory
  drm/nv50: improve vram page table construction
  drm/nv50: more efficient clearing of gpu page table entries
  drm/nv50: make nv50_mem_vm_{bind,unbind} operate only on vram
  drm/nouveau: Fix up pre-nv17 analog load detection.

8 years ago[IA64] Fix broken sn2 build
Hedi Berriche [Tue, 23 Feb 2010 23:58:49 +0000 (23:58 +0000)]
[IA64] Fix broken sn2 build

Revert the change made to arch/ia64/sn/kernel/setup.c by commit
204fba4aa303ea4a7bb726a539bf4a5b9e3203d0 as it breaks the build.

Fixing the build the b94b08081fcecf83fa690d6c5664f6316fe72208 way
breaks xpc because genksyms then fails to generate an CRC for
per_cpu____sn_cnodeid_to_nasid because of limitations in the
generic genksyms code.

Signed-off-by: Hedi Berriche <hedi@sgi.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>
8 years agoMerge branch 'master' of git://git.kernel.org/pub/scm/linux/kernel/git/linville/wirel...
David S. Miller [Tue, 23 Feb 2010 09:27:05 +0000 (01:27 -0800)]
Merge branch 'master' of git://git./linux/kernel/git/linville/wireless-2.6

8 years agotc35815: Remove a wrong netif_wake_queue() call which triggers BUG_ON
Atsushi Nemoto [Fri, 19 Feb 2010 05:13:58 +0000 (05:13 +0000)]
tc35815: Remove a wrong netif_wake_queue() call which triggers BUG_ON

The netif_wake_queue() is called correctly (i.e. only on !txfull
condition) from txdone routine.  So Unconditional call to the
netif_wake_queue() here is wrong.  This might cause calling of
start_xmit routine on txfull state and trigger BUG_ON.

This bug does not happen when NAPI disabled.  After txdone there
must be at least one free tx slot.  But with NAPI, this is not
true anymore and the BUG_ON can hits on heavy load.

In this driver NAPI was enabled on 2.6.33-rc1 so this is
regression from 2.6.32 kernel.

Reported-by: Ralf Roesch <ralf.roesch@rw-gmbh.de>
Signed-off-by: Atsushi Nemoto <anemo@mba.ocn.ne.jp>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agocdc_ether: new PID for Ericsson C3607w to the whitelist (resubmit)
Torgny Johansson [Fri, 19 Feb 2010 01:59:15 +0000 (01:59 +0000)]
cdc_ether: new PID for Ericsson C3607w to the whitelist (resubmit)

This patch adds a new vid/pid to the cdc_ether whitelist.

Device added:
- Ericsson Mobile Broadband variant C3607w

Signed-off-by: Torgny Johansson <torgny.johansson@gmail.com>
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoIPv6: better document max_addresses parameter
Brian Haley [Mon, 22 Feb 2010 12:27:21 +0000 (12:27 +0000)]
IPv6: better document max_addresses parameter

Andrew Morton wrote:
>> >From ip-sysctl.txt file in kernel documentation I can see following description
>> for max_addresses:
>> max_addresses - INTEGER
>>         Number of maximum addresses per interface.  0 disables limitation.
>>         It is recommended not set too large value (or 0) because it would
>>         be too easy way to crash kernel to allow to create too much of
>>         autoconfigured addresses.

>> If this parameter applies only for auto-configured IP addressed, please state
>> it more clearly in docs or rename the parameter to show that it refers to
>> auto-configuration.

It did mention autoconfigured in the text, but the below makes it more obvious.

More clearly document IPv6 max_addresses parameter.

Signed-off-by: Brian Haley <brian.haley@hp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoMAINTAINERS: update mv643xx_eth maintenance status
Lennert Buytenhek [Mon, 22 Feb 2010 22:34:54 +0000 (22:34 +0000)]
MAINTAINERS: update mv643xx_eth maintenance status

I am no longer with Marvell.

Signed-off-by: Lennert Buytenhek <buytenh@wantstofly.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoe1000: Fix DMA mapping error handling on RX
Anton Blanchard [Fri, 19 Feb 2010 17:54:53 +0000 (17:54 +0000)]
e1000: Fix DMA mapping error handling on RX

Check for error return from pci_map_single/pci_map_page and clean up.

With this and the previous patch the driver was able to handle a significant
percentage of errors (I set the fault injection rate to 10% and could still
download large files at a reasonable speed).

Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoRevert "block: improve queue_should_plug() by looking at IO depths"
Jens Axboe [Tue, 23 Feb 2010 07:40:43 +0000 (08:40 +0100)]
Revert "block: improve queue_should_plug() by looking at IO depths"

This reverts commit fb1e75389bd06fd5987e9cda1b4e0305c782f854.

"Benjamin S." <sbenni@gmx.de> reports that the patch in question
causes a big drop in sequential throughput for him, dropping from
200MB/sec down to only 70MB/sec.

Needs to be investigated more fully, for now lets just revert the
offending commit.



Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
8 years agodrm/vmwgfx: Fix queries if no dma buffer thrashing is occuring.
Thomas Hellstrom [Sun, 21 Feb 2010 14:54:55 +0000 (14:54 +0000)]
drm/vmwgfx: Fix queries if no dma buffer thrashing is occuring.

Intercept query commands and apply relocations to their guest pointers.

Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
8 years agoMerge remote branch 'nouveau/for-airlied' of ../drm-nouveau-next into drm-linus
Dave Airlie [Tue, 23 Feb 2010 05:42:18 +0000 (15:42 +1000)]
Merge remote branch 'nouveau/for-airlied' of ../drm-nouveau-next into drm-linus

* 'nouveau/for-airlied' of ../drm-nouveau-next:
  drm/nv50: fix vram ptes on IGPs to point at stolen system memory
  drm/nv50: fix instmem binding on IGPs to point at stolen system memory
  drm/nv50: improve vram page table construction
  drm/nv50: more efficient clearing of gpu page table entries
  drm/nv50: make nv50_mem_vm_{bind,unbind} operate only on vram
  drm/nouveau: Fix up pre-nv17 analog load detection.

8 years agoMerge branch 'pcc' into release
Len Brown [Tue, 23 Feb 2010 05:39:00 +0000 (00:39 -0500)]
Merge branch 'pcc' into release

8 years agoMerge branches 'bugzilla-14207' and 'idle' into release
Len Brown [Tue, 23 Feb 2010 05:19:48 +0000 (00:19 -0500)]
Merge branches 'bugzilla-14207' and 'idle' into release

8 years agoMerge branch 'upstream' of git://ftp.linux-mips.org/pub/scm/upstream-linus
Linus Torvalds [Tue, 23 Feb 2010 03:51:39 +0000 (19:51 -0800)]
Merge branch 'upstream' of git://ftp.linux-mips.org/upstream-linus

* 'upstream' of git://ftp.linux-mips.org/pub/scm/upstream-linus:
  MIPS: BCM47xx: Fix 128MB RAM support
  MIPS: Highmem: Fix build error

8 years agoMerge branch 'parisc/tracehook' of git://git.kernel.org/pub/scm/linux/kernel/git...
Linus Torvalds [Tue, 23 Feb 2010 03:51:13 +0000 (19:51 -0800)]
Merge branch 'parisc/tracehook' of git://git./linux/kernel/git/frob/linux-2.6-roland

* 'parisc/tracehook' of git://git.kernel.org/pub/scm/linux/kernel/git/frob/linux-2.6-roland:
  Revert "parisc: HAVE_ARCH_TRACEHOOK"

8 years agofs/exec.c: fix initial stack reservation
Michael Neuling [Mon, 22 Feb 2010 20:44:24 +0000 (12:44 -0800)]
fs/exec.c: fix initial stack reservation

803bf5ec259941936262d10ecc84511b76a20921 ("fs/exec.c: restrict initial
stack space expansion to rlimit") attempts to limit the initial stack to
20*PAGE_SIZE.  Unfortunately, in attempting ensure the stack is not
reduced in size, we ended up not changing the stack at all.

This size reduction check is not necessary as the expand_stack call does
this already.

This caused a regression in UML resulting in most guest processes being

Signed-off-by: Michael Neuling <mikey@neuling.org>
Reviewed-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Acked-by: WANG Cong <xiyou.wangcong@gmail.com>
Cc: Anton Blanchard <anton@samba.org>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: James Morris <jmorris@namei.org>
Cc: Serge Hallyn <serue@us.ibm.com>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Jouni Malinen <j@w1.fi>
Cc: <stable@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
8 years agoefifb: fix framebuffer handoff
Marcin Slusarz [Mon, 22 Feb 2010 20:44:22 +0000 (12:44 -0800)]
efifb: fix framebuffer handoff

Commit 4410f3910947dcea8672280b3adecd53cec4e85e ("fbdev: add support for
handoff from firmware to hw framebuffers") didn't add fb_destroy
operation to efifb.  Fix it and change aperture_size to match size
passed to request_mem_region.

Addresses http://bugzilla.kernel.org/show_bug.cgi?id=15151

Signed-off-by: Marcin Slusarz <marcin.slusarz@gmail.com>
Reported-by: Alex Zhavnerchik <alex.vizor@gmail.com>
Tested-by: Alex Zhavnerchik <alex.vizor@gmail.com>
Acked-by: Peter Jones <pjones@redhat.com>
Cc: Huang Ying <ying.huang@intel.com>
Cc: Dave Airlie <airlied@redhat.com>
Cc: "Rafael J. Wysocki" <rjw@sisk.pl>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
8 years agogeode-mfgpt: restore previous behavior for selecting IRQ
Jens Rottmann [Mon, 22 Feb 2010 20:44:20 +0000 (12:44 -0800)]
geode-mfgpt: restore previous behavior for selecting IRQ

geode-mfgpt: restore previous behavior for selecting IRQ

The MFGPT IRQ used to be, in order of decreasing priority,
 * IRQ supplied by the user as a boot-time parameter,
 * IRQ previously set by the BIOS or another driver,
 * default IRQ given at compile time.

Return to this behavior, which got broken when splitting the
MFGPT/clocksource driver for 2.6.33-rc1.

Signed-off-by: Jens Rottmann <JRottmann@LiPPERTEmbedded.de>
Acked-by: Andres Salomon <dilinger@collabora.co.uk>
Cc: Jordan Crouse <jordan.crouse@amd.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: john stultz <johnstul@us.ibm.com>
Acked-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
8 years agoidr: fix a critical misallocation bug, take#2
Tejun Heo [Mon, 22 Feb 2010 20:44:19 +0000 (12:44 -0800)]
idr: fix a critical misallocation bug, take#2

This is retry of reverted 859ddf09743a8cc680af33f7259ccd0fd36bfe9d
("idr: fix a critical misallocation bug") which contained two bugs.

* pa[idp->layers] should be cleared even if it's not used by
  sub_alloc() because it's used by mark idr_mark_full().

* The original condition check also assigned pa[l] to p which the new
  code didn't do thus leaving p pointing at the wrong layer.

Both problems have been fixed and the idr code has received good amount
testing using userland testing setup where simple bitmap allocator is
run parallel to verify the result of idr allocation.

The bug this patch fixes is caused by sub_alloc() optimization path
bypassing out-of-room condition check and restarting allocation loop
with starting value higher than maximum allowed value.  For detailed
description, please read commit message of 859ddf09.

Signed-off-by: Tejun Heo <tj@kernel.org>
Based-on-patch-from: Eric Paris <eparis@redhat.com>
Reported-by: Eric Paris <eparis@redhat.com>
Tested-by: Stefan Lippers-Hollmann <s.l-h@gmx.de>
Tested-by: Serge Hallyn <serue@us.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
8 years agokernel/sys.c: fix missing rcu protection for sys_getpriority()
Tetsuo Handa [Mon, 22 Feb 2010 20:44:16 +0000 (12:44 -0800)]
kernel/sys.c: fix missing rcu protection for sys_getpriority()

find_task_by_vpid() is not safe without rcu_read_lock().  2.6.33-rc7 got
RCU protection for sys_setpriority() but missed it for sys_getpriority().

Signed-off-by: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: "Paul E. McKenney" <paulmck@us.ibm.com>
Acked-by: Serge Hallyn <serue@us.ibm.com>
Acked-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
8 years agomemcg: fix oom killing a child process in an other cgroup
KAMEZAWA Hiroyuki [Mon, 22 Feb 2010 20:44:14 +0000 (12:44 -0800)]
memcg: fix oom killing a child process in an other cgroup

Presently the oom-killer is memcg aware and it finds the worst process
from processes under memcg(s) in oom.  Then, it kills victim's child

It may kill a child in another cgroup and may not be any help for
recovery.  And it will break the assumption users have.

This patch fixes it.

Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Reviewed-by: Minchan Kim <minchan.kim@gmail.com>
Cc: Balbir Singh <balbir@linux.vnet.ibm.com>
Reviewed-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Acked-by: David Rientjes <rientjes@google.com>
Cc: <stable@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
8 years agodrm/nv50: fix vram ptes on IGPs to point at stolen system memory
Ben Skeggs [Fri, 19 Feb 2010 22:10:11 +0000 (08:10 +1000)]
drm/nv50: fix vram ptes on IGPs to point at stolen system memory

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
8 years agodrm/nv50: fix instmem binding on IGPs to point at stolen system memory
Ben Skeggs [Fri, 19 Feb 2010 22:06:36 +0000 (08:06 +1000)]
drm/nv50: fix instmem binding on IGPs to point at stolen system memory

This also modifies the unused PRAMIN PT entries to be all zeroes, can't
really recall why I used 9/0 initially, just that it didn't work for
some reason.  It was likely masking a bug elsewhere that's since been

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
8 years agodrm/nv50: improve vram page table construction
Ben Skeggs [Thu, 11 Feb 2010 01:31:44 +0000 (11:31 +1000)]
drm/nv50: improve vram page table construction

This commit changes nouveau to construct PTEs which look very much like
the ones the binary driver creates.

I presume that filling multiple PTEs identically with length flags and
the physical address of the start of a block of VRAM is a hint to the
memory controller that it need not perform additional page table lookups
for that range of addresses.

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
8 years agodrm/nv50: more efficient clearing of gpu page table entries
Ben Skeggs [Thu, 11 Feb 2010 00:25:53 +0000 (10:25 +1000)]
drm/nv50: more efficient clearing of gpu page table entries

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
8 years agodrm/nv50: make nv50_mem_vm_{bind,unbind} operate only on vram
Ben Skeggs [Thu, 11 Feb 2010 00:23:30 +0000 (10:23 +1000)]
drm/nv50: make nv50_mem_vm_{bind,unbind} operate only on vram

GART is handled elsewhere, no reason to have the code for it here too.

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
8 years agodrm/nouveau: Fix up pre-nv17 analog load detection.
Francisco Jerez [Wed, 10 Feb 2010 18:25:58 +0000 (19:25 +0100)]
drm/nouveau: Fix up pre-nv17 analog load detection.

Signed-off-by: Francisco Jerez <currojerez@riseup.net>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
8 years agodrm/radeon/kms: implement reading active PCIE lanes on R600+
Rafał Miłecki [Thu, 18 Feb 2010 20:24:28 +0000 (20:24 +0000)]
drm/radeon/kms: implement reading active PCIE lanes on R600+

Signed-off-by: Rafał Miłecki <zajec5@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
8 years agodrm/radeon/kms: for downclocking non-mobility check PERFORMANCE state
Rafał Miłecki [Sun, 21 Feb 2010 22:46:30 +0000 (22:46 +0000)]
drm/radeon/kms: for downclocking non-mobility check PERFORMANCE state

AtomBIOS tables on non-mobility GPU do not contain POWERSAVE/BATTERY.

Signed-off-by: Rafał Miłecki <zajec5@gmail.com>
Reviewed-by: Alex Deucher <alexdeucher@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
8 years agodrm/radeon/kms: simplify storing current and requested PM mode
Rafał Miłecki [Sat, 20 Feb 2010 23:15:04 +0000 (23:15 +0000)]
drm/radeon/kms: simplify storing current and requested PM mode

We kept pointers to requested and current clock modes in every power state.
That was useless, more /global/ pointers in power struct are enough.

Signed-off-by: Rafał Miłecki <zajec5@gmail.com>
Reviewed-by: Alex Deucher <alexdeucher@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
8 years agodrm/radeon: fixes for r6xx/r7xx gfx init
Alex Deucher [Fri, 19 Feb 2010 21:22:31 +0000 (16:22 -0500)]
drm/radeon: fixes for r6xx/r7xx gfx init

- updated swizzle modes for backend map setup
- fix programming of a few gfx regs
- properly handle pipe/backend setup on LE cards

Signed-off-by: Alex Deucher <alexdeucher@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
8 years agodrm/radeon/rv740: fix backend setup
Alex Deucher [Fri, 19 Feb 2010 21:07:02 +0000 (16:07 -0500)]
drm/radeon/rv740: fix backend setup

This patch fixes occlusion queries and rendering errors
on rv740 boards. Hardcoding the backend map is not an optimal
solution, but a better fix is being worked on.

Signed-off-by: Alex Deucher <alexdeucher@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
8 years agodrm/radeon/kms: fix R3XX/R4XX memory controller initialization
Jerome Glisse [Thu, 18 Feb 2010 14:23:49 +0000 (14:23 +0000)]
drm/radeon/kms: fix R3XX/R4XX memory controller initialization

Version 2 of memory controller did break the initialization for
R3XX/R4XX hardware. This patch fix it.

Signed-off-by: Jerome Glisse <jglisse@redhat.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
8 years ago[rfc] drm/radeon/kms: pm debugging check for vbl.
Dave Airlie [Thu, 18 Feb 2010 05:58:36 +0000 (15:58 +1000)]
[rfc] drm/radeon/kms: pm debugging check for vbl.

This patch adds a check on avivo chips to see if we are in the VBL
region for the active crtcs when we trigger the engine change.

I appear to have glitches locally on pm transistion (not sure all
fixes are in yet) and this at least seems to be correct here,
maybe others can test on systems with no glitches.

8 years agodrm/radeon: Fix memory allocation failures in the preKMS command stream checking.
Pauli Nieminen [Mon, 1 Feb 2010 17:11:16 +0000 (19:11 +0200)]
drm/radeon: Fix memory allocation failures in the preKMS command stream checking.

Allocation of single large block of memory may fail under memory
presure. drm_buffer object can hold one large block of data in
multiple independ pages which preents alloation failures.

This patch converts all access to command stream to use drm_buffer
interface. All direct access to array has to go tough drm_buffer
functions to get correct pointer.

Outputting the command stream to ring buffer needs to be awear of
the split nature of drm_buffer. The output operation requires the

Signed-off-by: Pauli Nieminen <suokkos@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
8 years agodrm: Add generic multipart buffer.
Pauli Nieminen [Mon, 1 Feb 2010 17:11:15 +0000 (19:11 +0200)]
drm: Add generic multipart buffer.

Allocating multiple pages of memory for data that is coming
from user space may fail. To fix memory allocation failures
the buffer object should be split to multiple independ pages.

drm buffer provides generic interface to copy and process
large data arrays from user space.

Interface includes allocation and free functions to allocate
the buffer object and data storage pages.

All access operations are performed relative to a internal
pointer which is advanced with drm_buffer_advance function.

The buffer can be accessed using drm_buffer_pointer_to_XXX
functions if it is known that requested object doesn't split
over a page boundary. These functions don't do any error
checking to maximize performance.

If there is large object which could be split there is special
drm_buffer_read_object function. drm_buffer_read_object takes
a pointer as argument which is used as temporary store for
data if it is split over boundary in the buffer.

Signed-off-by: Pauli Nieminen <suokkos@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
8 years agoMIPS: BCM47xx: Fix 128MB RAM support
Hauke Mehrtens [Sat, 20 Feb 2010 18:51:20 +0000 (19:51 +0100)]
MIPS: BCM47xx: Fix 128MB RAM support

Ignoring the last page when ddr size is 128M. Cached accesses to last page
is causing the processor to prefetch using address above 128M stepping out
of the DDR address space.

Signed-off-by: Hauke Mehrtens <hauke@hauke-m.de>
Cc: linux-mips@linux-mips.org
Patchwork: http://patchwork.linux-mips.org/patch/981/
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
8 years agoMIPS: Highmem: Fix build error
Yoichi Yuasa [Sat, 20 Feb 2010 12:23:22 +0000 (21:23 +0900)]
MIPS: Highmem: Fix build error

arch/mips/mm/highmem.c: In function 'kmap_init':
arch/mips/mm/highmem.c:130: error: 'init_mm' undeclared (first use in this function)
arch/mips/mm/highmem.c:130: error: (Each undeclared identifier is reported only once
arch/mips/mm/highmem.c:130: error: for each function it appears in.)

Signed-off-by: Yoichi Yuasa <yuasa@linux-mips.org>
Cc: linux-mips <linux-mips@linux-mips.org>
Patchwork: http://patchwork.linux-mips.org/patch/980/
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
8 years agoRevert "parisc: HAVE_ARCH_TRACEHOOK"
Roland McGrath [Mon, 22 Feb 2010 18:43:43 +0000 (10:43 -0800)]
Revert "parisc: HAVE_ARCH_TRACEHOOK"

This reverts commit 81bf550d9cdfe0325eb1504b06c9f6511b442c1a.

HAVE_ARCH_TRACEHOOK requires defining the user_regset interfaces,
including task_user_regset_view().  parisc doesn't do that yet,
so don't lie about it.

Signed-off-by: Roland McGrath <roland@redhat.com>
8 years agoACPI: Be in TS_POLLING state during mwait based C-state entry
Pallipadi, Venkatesh [Wed, 10 Feb 2010 18:35:31 +0000 (10:35 -0800)]
ACPI: Be in TS_POLLING state during mwait based C-state entry

ACPI deep C-state entry had a long standing bug/missing feature, wherein we were sending
resched IPIs when an idle CPU is in mwait based deep C-state. Only mwait based C1 was using
the write to the monitored address to wake up mwait'ing CPU.

This patch changes the code to retain TS_POLLING bit if we are entering an mwait based
deep C-state.

The patch has been verified to reduce the number of resched IPIs in general and also
improves the performance/power on workloads with low system utilization (i.e., when mwait based
deep C-states are being used).

Fixes "netperf ~50% regression with 2.6.33-rc1, bisect to 1b9508f"

Reported-by: Lin Ming <ming.m.lin@intel.com>
Tested-by: Alex Shi <alex.shi@intel.com>
Signed-off-by: Venkatesh Pallipadi <venkatesh.pallipadi@intel.com>
Signed-off-by: Len Brown <len.brown@intel.com>