openwrt/staging/blogic.git
11 years agodrm/i915: add pipe_config->has_pch_encoder
Daniel Vetter [Tue, 26 Mar 2013 23:44:55 +0000 (00:44 +0100)]
drm/i915: add pipe_config->has_pch_encoder

This is used way too often in the enable/disable paths. And will
be even more useful in the future.

Note that correct semantics of this change highly depend upon
correct updating of intel_crtc->config: Like with all other
modeset state, we need to call ->disable with the old config,
but ->mode_set and ->enable with the new config.

v2: Do not yet use the flag in the ->disable callbacks - atm we don't
yet have support for the information stored in the pipe_config in the
hw state readout code, so this will be wrong at boot-up/resume.

v3: Rebased on top of the hdmi/dp ddi encoder merging.

v4: Fixup stupid rebase error which lead to a NULL vfunc deref.

v5: On haswell the VGA port is on the PCH!

v6: s/IS_HASWELL/HAS_DDI/, spotted by Paulo Zanoni. Also add a missing
parameter name in a function declaration.

v7: Don't forget to git add ...

Reviewed-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: drop helper vtable for sdvo encoder
Daniel Vetter [Tue, 26 Mar 2013 23:44:54 +0000 (00:44 +0100)]
drm/i915: drop helper vtable for sdvo encoder

Completely unused by now. Separate patch in case I've missed a
place somewhere which dereferences the helper vtable but actually
shouldn't do so.

v2: Resolve rebase conflict with Egbert Eich's hpd infrastructure
rework.

Reviewed-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: add pipe_config->pixel_multiplier
Daniel Vetter [Tue, 26 Mar 2013 23:44:53 +0000 (00:44 +0100)]
drm/i915: add pipe_config->pixel_multiplier

Used by SDVO (and hopefully, eventually HDMI, if we ever get around
to fixing up the low dotclock CEA modes ...).

This required adding a new encoder->mode_set callback to be able to
pass around the intel_crtc_config.

Reviewed-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: add pipe_config->timings_set
Daniel Vetter [Tue, 26 Mar 2013 23:44:52 +0000 (00:44 +0100)]
drm/i915: add pipe_config->timings_set

Only used by the lvds encoder. Note that we shouldn't do the same
simple conversion with the FORCE_6BPC flag, since that's much better
handled by moving all the pipe_bpc computation around.

This requires that we pass the pipe config around to encoders, so
that they can set special attributes and set constraints. To do so
introduce a new ->compute_config encoder callback, which is called in
stead of the drm crtc helper's ->mode_fixup.

To avoid massive churn all over the codebase we don't want to convert
all existing ->mode_fixup functions. Instead I've opted to convert
them on an as-needed basis (mostly to cut down on rebase conflicts and
to have more freedom to experiment around while developing the
patches).

Reviewed-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: compute pipe_config earlier
Daniel Vetter [Tue, 26 Mar 2013 23:44:51 +0000 (00:44 +0100)]
drm/i915: compute pipe_config earlier

To make decent modeset state checking possible (e.g. for the check
mode with atomic modesetting) we want to have the full pipe
configuration and state checks done before we touch the hw.

To ensure that all the little bits&pieces that are now moved to the
pipe_config handle this correctly, move its computation to the right
spot now, before we touch the hw in the disable_pipes step.

Reviewed-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Reviewed-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: introduce struct intel_crtc_config
Daniel Vetter [Tue, 26 Mar 2013 23:44:50 +0000 (00:44 +0100)]
drm/i915: introduce struct intel_crtc_config

Currently only containing the requested and the adjusted mode. And
only crtc callbacks are converted somewhat to it, encoders will be
done on a as-needed basis (simply too much churn in one patch
otherwise).

Future patches will add tons more useful stuff to this struct,
starting with the very simple.

v2: Store the pipe_config in the intel_crtc, so that the ->mode-set,
->enable and also ->disable have easy access to it.

v3: Store the pipe config in the right crtc ...

v4: Rebased.

v5: Fixup an OOPS when trying to kfree an ERR_PTR.

v6: Used drm_moode_copy and some other small cleanups as suggested
by Ville Syrjälä.

v7: drm_mode_copy preserves the mode id of the destination, so no need
to clear it again (Ville).

v8: Break a long line spotted by Paulo.

Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Reviewed-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: return actual brightness to .get_brightness callback
Jani Nikula [Tue, 12 Mar 2013 09:44:16 +0000 (11:44 +0200)]
drm/i915: return actual brightness to .get_brightness callback

The backlight device .get_brightness callback is supposed to return the
actual brightness by querying the hardware, not the cached value. See
Documentation/ABI/stable/sysfs-class-backlight. The callback is there to
support the actual_brightness sysfs file.

With the backlight_level and backlight device brightness now in sync, one
can reliably get the cached value through the brightness sysfs file.

Signed-off-by: Jani Nikula <jani.nikula@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: keep backlight_level and backlight device brightness in sync
Jani Nikula [Tue, 12 Mar 2013 09:44:15 +0000 (11:44 +0200)]
drm/i915: keep backlight_level and backlight device brightness in sync

A single point of truth would be better than two, but achieving that would
require more abstractions for CONFIG_BACKLIGHT_CLASS_DEVICE=n with not a
whole lot of real benefits. Take the short route and just keep the
backlight levels in sync. In particular, update backlight device brightness
on opregion brightness changes.

Signed-off-by: Jani Nikula <jani.nikula@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: wire up SDVO hpd support on cpt/ppt
Daniel Vetter [Tue, 26 Mar 2013 21:38:43 +0000 (22:38 +0100)]
drm/i915: wire up SDVO hpd support on cpt/ppt

Now with Egbert Eich's hpd infrastructure rework merged this is dead
simple. And we need this to make output detection work on SDVO - with
the cleaned-up drm polling helpers outputs which claim to have hpd
support are no longer polled.

Now SDVO claims to do that, but it's not actually wired up. So just do
it.

Reviewed-by: Rodrigo Vivi <rodrigo.vivi@gmail.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: Wait for vblank between disabling a sprite and unpinning the fb
Ville Syrjälä [Wed, 27 Mar 2013 15:49:13 +0000 (17:49 +0200)]
drm/i915: Wait for vblank between disabling a sprite and unpinning the fb

When disabling a sprite, wait for the sprite to stop fetching data
from memory before unpinning the fb.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: clear crt hotplug compare voltage field before setting
Daniel Vetter [Wed, 27 Mar 2013 14:47:11 +0000 (15:47 +0100)]
drm/i915: clear crt hotplug compare voltage field before setting

Noticed while reviewing the hotplug irq setup code. Just looks better.

Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agoRevert "drm/i915: set dummy page for stolen objects"
Imre Deak [Tue, 26 Mar 2013 13:14:19 +0000 (15:14 +0200)]
Revert "drm/i915: set dummy page for stolen objects"

Since for_each_sg_page supports already memory w/o backing pages we can
revert the corresponding workaround.

This reverts commit 5bd4687e57bbacec20930f580d025aee9fa1f4d8.

Signed-off-by: Imre Deak <imre.deak@intel.com>
Reviewed-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agolib/scatterlist: sg_page_iter: support sg lists w/o backing pages
Imre Deak [Tue, 26 Mar 2013 13:14:18 +0000 (15:14 +0200)]
lib/scatterlist: sg_page_iter: support sg lists w/o backing pages

The i915 driver uses sg lists for memory without backing 'struct page'
pages, similarly to other IO memory regions, setting only the DMA
address for these. It does this, so that it can program the HW MMU
tables in a uniform way both for sg lists with and without backing pages.

Without a valid page pointer we can't call nth_page to get the current
page in __sg_page_iter_next, so add a helper that relevant users can
call separately. Also add a helper to get the DMA address of the current
page (idea from Daniel).

Convert all places in i915, to use the new API.

Signed-off-by: Imre Deak <imre.deak@intel.com>
Reviewed-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: Apply alignment restrictions on scanout surfaces for VT-d
Chris Wilson [Tue, 5 Mar 2013 14:52:39 +0000 (14:52 +0000)]
drm/i915: Apply alignment restrictions on scanout surfaces for VT-d

From the w/a database:

'To prevent false VT-d type 6 error:

  The primary display plane must be 256KiB aligned, and require an extra
  128 PTEs of padding afterward;

  The sprites planes must be 128KiB aligned, and require an extra 64 PTEs
  of padding afterward;

  The cursors must be 64KiB aligned, and require an extra 2 PTEs of
  padding afterward.'

As we use the same function to pin the primary and sprite planes, we can
simply use the more strict requirements for scanouts for both.

Instead of using explicit padding PTEs following the scanout objects, we
should be able to use the scratch page that is always mapped into the
unused PTEs to avoid the VT-d error.

References: https://bugs.freedesktop.org/show_bug.cgi?id=59626
References: https://bugs.freedesktop.org/show_bug.cgi?id=59627
References: https://bugs.freedesktop.org/show_bug.cgi?id=59631
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Damien Lespiau <damien.lespiau@intel.com>
[danvet: Apply s/vtd_wa/vtd_scanout_wa/ bikeshed since Damien likes
it, too.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agoDRM/i915: Get rid if the 'hotplug_supported_mask' in struct drm_i915_private.
Egbert Eich [Mon, 25 Feb 2013 17:06:52 +0000 (12:06 -0500)]
DRM/i915: Get rid if the 'hotplug_supported_mask' in struct drm_i915_private.

Now since we have replaced the bits to show interest in hotplug IRQs
we can go and nuke the 'hotplug_supported_mask'.

Signed-off-by: Egbert Eich <eich@suse.de>
Acked-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agoDRM/i915: Remove i965_hpd_irq_setup.
Egbert Eich [Mon, 25 Feb 2013 17:06:51 +0000 (12:06 -0500)]
DRM/i915: Remove i965_hpd_irq_setup.

After
"Convert HPD interrupts to make use of HPD pin assignment in encoders."
This function is now basically the same as i915_hpd_irq_setup().

Consolidating both functions in one requires one more check for
I915_HAS_HOTPLUG(dev) in the i965 code path and one more check for
IS_G4X(dev) in the i915 code path. These are considered harmless.

Signed-off-by: Egbert Eich <eich@suse.de>
Acked-by: Chris Wilson <chris@chris-wilson.co.uk>
[danvet: Fixup patch conflict and make it compile.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: implement ibx_hpd_irq_setup
Daniel Vetter [Wed, 27 Mar 2013 14:55:01 +0000 (15:55 +0100)]
drm/i915: implement ibx_hpd_irq_setup

This fixes a regression introduced in

commit e5868a318d1ae28f760f77bb91ce5deb751733fd
Author: Egbert Eich <eich@suse.de>
Date:   Thu Feb 28 04:17:12 2013 -0500

    DRM/i915: Convert HPD interrupts to make use of HPD pin assignment in encode

Due to the irq setup rework in 3.9, see

commit 20afbda209d708be66944907966486d0c1331cb8
Author: Daniel Vetter <daniel.vetter@ffwll.ch>
Date:   Tue Dec 11 14:05:07 2012 +0100

    drm/i915: Fixup hpd irq register setup ordering

Egbert Eich's hpd rework blows up on pch-split platforms - it walks
the encoder list before that has been set up completely. The new init
sequence is:

1. irq enabling
2. modeset init
3. hpd setup

We need to move around the ibx setup a bit to fix this.

Ville Syrjälä pointed out in his review that we can't touch SDEIER
after the interrupt handler is set up, since that'll race with Paulo
Zanoni's PCH interrupt race fix:

commit 44498aea293b37af1d463acd9658cdce1ecdf427
Author: Paulo Zanoni <paulo.r.zanoni@intel.com>
Date:   Fri Feb 22 17:05:28 2013 -0300

    drm/i915: also disable south interrupts when handling them

We fix that by unconditionally enabling all interrupts in SDEIER, but
masking them as-needed in SDEIMR. Since only the single-threaded
setup/teardown (or suspend/resume) code touches that, no further
locking is required.

While at it also simplify the mask handling - we start out with all
interrupts cleared in the postinstall hook, and never enable a hpd
interrupt before hpd_irq_setup is called.

And finally, for consistency rename the ibx hpd setup function to
ibx_hpd_irq_setup.

v2: Fix race around SDEIER writes (Ville).

v3: Remove the superflous posting read for SDEIER, spotted by Ville.

Ville also wondered whether we shouldn't clear SDEIIR, since now
SDE interrupts are enabled before we have an irq handler installed.
But the master interrupt control bit in DEIER is still cleared, so we
should be fine.

Cc: Egbert Eich <eich@suse.de>
Cc: Jesse Barnes <jbarnes@virtuousgeek.org>
Cc: Paulo Zanoni <paulo.r.zanoni@intel.com>
Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=62798
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agoDRM/i915: Convert HPD interrupts to make use of HPD pin assignment in encoders (v2)
Egbert Eich [Thu, 28 Feb 2013 09:17:12 +0000 (04:17 -0500)]
DRM/i915: Convert HPD interrupts to make use of HPD pin assignment in encoders (v2)

This allows to enable HPD interrupts for individual pins to only receive
hotplug events from lines which are connected and working.

v2: Restructured initailization of const arrays following a suggstion
    by Chris Wilson <chris@chris-wilson.co.uk>

Signed-off-by: Egbert Eich <eich@suse.de>
Acked-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Jesse Barnes <jbarnes@virtuousgeek.org> (v1)
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agoDRM/I915: Add enum hpd_pin to intel_encoder.
Egbert Eich [Mon, 25 Feb 2013 17:06:49 +0000 (12:06 -0500)]
DRM/I915: Add enum hpd_pin to intel_encoder.

To clean up hotplug support we add a new enum to intel_encoder:
enum hpd_pin. It allows the encoder to request a hpd line but leave
the details which IRQ is responsible on which chipset generation
to i915_irq.c.
This way requesting hotplug support will become really simple on
the encoder/connector level.

Signed-off-by: Egbert Eich <eich@suse.de>
Acked-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agoDRM/i915: Remove valleyview_hpd_irq_setup.
Egbert Eich [Mon, 25 Feb 2013 17:06:48 +0000 (12:06 -0500)]
DRM/i915: Remove valleyview_hpd_irq_setup.

It's basically identical to i915_hpd_irq_setup().

Signed-off-by: Egbert Eich <eich@suse.de>
Reviewed-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Acked-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: emit a hotplug event on resume
Jesse Barnes [Tue, 26 Mar 2013 16:25:46 +0000 (09:25 -0700)]
drm/i915: emit a hotplug event on resume

This will poke userspace into probing for configuration changes that may
have occurred across suspend/resume.

Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@gmail.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: enable VT switchless resume v3
Jesse Barnes [Tue, 26 Mar 2013 16:25:45 +0000 (09:25 -0700)]
drm/i915: enable VT switchless resume v3

With the other bits in place, we can do this safely.

v2: disable backlight on suspend to prevent premature enablement on resume
v3: disable CRTCs on suspend to allow RTD3 (Kristen)

Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@gmail.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: restore cursor and sprite state when forcing a config restore v2
Jesse Barnes [Tue, 26 Mar 2013 20:25:27 +0000 (13:25 -0700)]
drm/i915: restore cursor and sprite state when forcing a config restore v2

Needed for VT switchless resume.

v2: cursor state is now handled correctly in crtc_enable (Daniel)

Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@gmail.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: add sprite restore function v3
Jesse Barnes [Tue, 26 Mar 2013 16:25:43 +0000 (09:25 -0700)]
drm/i915: add sprite restore function v3

To be used to restore sprite state on resume.

v2: move sprite tracking bits up so we don't track modified sprite state
v3: use src_x/y in sprite suspend/resume code (Ville)

Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@gmail.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm: add initial_config function to fb helper
Jesse Barnes [Tue, 19 Feb 2013 21:31:39 +0000 (13:31 -0800)]
drm: add initial_config function to fb helper

Rather than building a config which may or may not work, let the driver
build an initial fb config.  This allows the driver to use the BIOS boot
configuration for example, displaying kernel messages and the initial fb
console on the same outputs the BIOS lit up at boot time.  If that
fails, the driver can still fall back the same way as the core.

Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: dri-devel@lists.freedesktop.org
Acked-by: Dave Airlie <airlied@gmail.com>
Reviewed-by: Imre Deak <imre.deak@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: Introduce i915_gem_object_create_stolen_for_preallocated
Chris Wilson [Tue, 19 Feb 2013 21:31:37 +0000 (13:31 -0800)]
drm/i915: Introduce i915_gem_object_create_stolen_for_preallocated

Wrap a preallocated region of stolen memory within an ordinary GEM
object, for example the BIOS framebuffer.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Imre Deak <imre.deak@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: Always call fence-lost prior to removing the fence
Chris Wilson [Tue, 26 Mar 2013 11:29:27 +0000 (11:29 +0000)]
drm/i915: Always call fence-lost prior to removing the fence

There is a minute window for a race between put-fence removing the fence
and for a new transaction by an external party on the GTT mmap. That is
we must zap the mmap prior to removing the fence and not afterwards.

Fixes regression from
commit 61050808bb019ebea966b7b5bfd357aaf219fb51
Author: Chris Wilson <chris@chris-wilson.co.uk>
Date:   Tue Apr 17 15:31:31 2012 +0100

    drm/i915: Refactor put_fence() to use the common fence writing routine

v2: Remember the fence to remove with a local variable (gcc)

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Daniel Vetter <daniel.vetter@ffwll.ch>
Reviewed-by: Imre Deak <imre.deak@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: HSW PM Frequency bits fix
Rodrigo Vivi [Mon, 25 Mar 2013 20:55:49 +0000 (17:55 -0300)]
drm/i915: HSW PM Frequency bits fix

According to HSW PM programming guide, frequency bits starts at
24 instead of 25.

v2: Paulo Zanoni noticed that only frequency bits can be set at
GEN6_RPNSWREQ. All others are read only.

CC: Ben Widawsky <ben@bwidawsk.net>
CC: Paulo Zanoni <paulo.r.zanoni@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@gmail.com>
Reviewed-by: Ben Widawsky <ben@bwidawsk.net>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: Warn if a pipe is enabled with a bogus port
Damien Lespiau [Mon, 25 Mar 2013 15:16:14 +0000 (15:16 +0000)]
drm/i915: Warn if a pipe is enabled with a bogus port

If TRANS_DDI_FUNC_CTL has been wrongly programmed with an incorrect
port, we are currently trying to read PORT_CLK_SEL(port) with an
uninitialized value.

Handle that case by returning PORT_CLK_SEL_NONE and warning about it.

v2: Move the warning inside intel_ddi_get_crtc_pll (Paulo Zanoni)

Signed-off-by: Damien Lespiau <damien.lespiau@intel.com>
Reviewed-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: Don't overclock on Haswell
Ben Widawsky [Sun, 24 Mar 2013 00:46:31 +0000 (17:46 -0700)]
drm/i915: Don't overclock on Haswell

HSW doesn't overclock the same way as IVB or SNB. I do not know about
VLV, so I've kept that off as well. I'm still working on getting the doc
updates to explain how we overclock on Haswell.

Cc: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Reviewed-by: Jesse Barnes <jbarnes@virtuousgeek.org>
[danvet: Add missing () spotted by Wu Fengguang's kernel build robot.
Acked by Ben.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: there's no PIPESTAT on HAS_PCH_SPLIT platforms
Paulo Zanoni [Fri, 22 Mar 2013 17:24:16 +0000 (14:24 -0300)]
drm/i915: there's no PIPESTAT on HAS_PCH_SPLIT platforms

So don't read it when capturing the error state. This solves
"unclaimed register" messages on Haswell when we have a GPU hang.

V2: Check for HAS_PCH_SPLIT instead of Gen5+ because VLV still has
this register.

Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: there's no DSPPOS register on gen4+
Paulo Zanoni [Fri, 22 Mar 2013 17:20:57 +0000 (14:20 -0300)]
drm/i915: there's no DSPPOS register on gen4+

So don't read it when capturing the error state. This solves some
"unclaimed register" messages on Haswell when we hang the GPU.

Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: fix DSPADDR Gen check
Paulo Zanoni [Fri, 22 Mar 2013 17:19:21 +0000 (14:19 -0300)]
drm/i915: fix DSPADDR Gen check

The first version of commit "drm/i915: there's no DSPADDR register on
Haswell" added 2 "!IS_HASWELL" checks. When reviewing the patch, Ben
suggested to make these checks more future-proof, so when Daniel
applied the patch he fixed the first check but not the second. This
commit makes the second check also "future-proof".

Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: Use BUG() in a case of a programming error
Damien Lespiau [Thu, 7 Mar 2013 15:30:28 +0000 (15:30 +0000)]
drm/i915: Use BUG() in a case of a programming error

The port number should always be correctly set. Do the same thing as the
switch above and use BUG() to signal that branch is not supposed to be
taken.

Signed-off-by: Damien Lespiau <damien.lespiau@intel.com>
Reviewed-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: Rename intel_ddi_enable_pipe_func() to transcoder_func()
Damien Lespiau [Thu, 7 Mar 2013 15:30:27 +0000 (15:30 +0000)]
drm/i915: Rename intel_ddi_enable_pipe_func() to transcoder_func()

We are really talking about the transcoder function here and the disable
version uses trancoder in its name already, so let's try to be
consistent.

Signed-off-by: Damien Lespiau <damien.lespiau@intel.com>
Reviewed-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: Cleanup if the EDP transcoder has a bobug input value
Damien Lespiau [Thu, 7 Mar 2013 15:30:26 +0000 (15:30 +0000)]
drm/i915: Cleanup if the EDP transcoder has a bobug input value

In the case where the hardware has been wrongly programmed and the EDP
TRANS_DDI_FUNC_CTL register has a bogus value in its EDP Input field, we
were using the pipe variable uninitialized.

In this case, shutdown the transcoder. It will be programmed correctly
the next time we try to enabled eDP.

Note from Paulo's review: Wrong modeset sequence can easily lead to
frozen machines hence the disable_ddi call might be risky. But since
things are awry already, doesn't matter too much.

Signed-off-by: Damien Lespiau <damien.lespiau@intel.com>
Reviewed-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
[danvet: Add note about Paulo's caution about potential hangs.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: Error out if we are trying to use VGA with SPLL already in use
Damien Lespiau [Thu, 7 Mar 2013 15:30:25 +0000 (15:30 +0000)]
drm/i915: Error out if we are trying to use VGA with SPLL already in use

Our static analysis tool noticed that 'reg' could be used uninitialized if
we are trying to get a PLL to drive VGA and SPLL is already in use
(plls->spll_refcoung != 0).

In the (error) case above, let's return false to the caller and emit an
error.

Signed-off-by: Damien Lespiau <damien.lespiau@intel.com>
Reviewed-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: Implement WaSwitchSolVfFArbitrationPriority
Ben Widawsky [Wed, 20 Mar 2013 21:49:14 +0000 (14:49 -0700)]
drm/i915: Implement WaSwitchSolVfFArbitrationPriority

Bspec mentions this for HSW+. I can't quite tell what the effects are,
and I don't easily have a way to test this.

Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: Set the VIC in AVI infoframe for SDVO
Ville Syrjälä [Wed, 20 Mar 2013 16:10:07 +0000 (18:10 +0200)]
drm/i915: Set the VIC in AVI infoframe for SDVO

We do this for HDMI already, so I don't know why we wouldn't do
it for SDVO as well.

This is completely untested due to lack of hardware.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: Kill a strange comment about DPMS functions
Ville Syrjälä [Wed, 20 Mar 2013 15:05:09 +0000 (17:05 +0200)]
drm/i915: Kill a strange comment about DPMS functions

This comment looks like some historical leftover. Get rid of it.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: Correct sandybrige overclocking
Ben Widawsky [Wed, 20 Mar 2013 03:19:56 +0000 (20:19 -0700)]
drm/i915: Correct sandybrige overclocking

Change the gen6+ max delay if the pcode read was successful (not the
inverse).

The previous code was all sorts of wrong and has existed since I broke
it:
commit 42c0526c930523425ff6edc95b7235ce7ab9308d
Author: Ben Widawsky <ben@bwidawsk.net>
Date:   Wed Sep 26 10:34:00 2012 -0700

    drm/i915: Extract PCU communication

I added some parentheses for clarity, and I also corrected the debug
message message to use the mask (wrong before I came along) and added a
print to show the value we're changing from.

Looking over the code, I'm not actually sure what we're trying to do. I
introduced the bug simply by extracting the function not implementing
anything new. We already set max_delay based on the capabilities
register (which is what we use elsewhere to determine min and max).
This would potentially increase it, I suppose? Jesse, I can't find the
document which explains the definitions of the pcode commands, maybe you
have it around.

Based on Jesse's response, this could potentially be for -fixes, or
stable, or maybe lead to us dropping it entirely. As the current code is
is, things won't completely break because of the aforementioned
capabilities register, and in my experimentation, enabling this has no
effect, it goes from 1100->1100.

I found this while reviewing Jesse's VLV patches.

Cc: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Reviewed-by: Jesse Barnes <jbarnes@virtuousgeek.org>
[danvet: Bikeshed-away the redudant parens spotted by Chris Wilson.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: Introduce GEN7_FEATURES for device info
Ben Widawsky [Fri, 15 Mar 2013 18:17:54 +0000 (11:17 -0700)]
drm/i915: Introduce GEN7_FEATURES for device info

Recommended by Chris.

v2: Make it GEN7_FEATURES, and use it for vlv and hsw also (Ben)

Cc: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: Move num_pipes to intel info
Ben Widawsky [Wed, 13 Mar 2013 21:05:41 +0000 (14:05 -0700)]
drm/i915: Move num_pipes to intel info

Requested by Daniel.

v2: Fix incorrect num_pipe settings. (Chris)

Cc: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Reviewed-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: fixup pd vs pt confusion in gen6 ppgtt code
Daniel Vetter [Tue, 19 Mar 2013 22:48:39 +0000 (23:48 +0100)]
drm/i915: fixup pd vs pt confusion in gen6 ppgtt code

The index variable points at a page table, not a page directory or a
pde. Ben Widawsky fix this up correctly in his ppgtt cleanup, but I've
botched the job and copy&pasted the old confusion from the original
gen6 ppgtt code in

commit def886c3768d24c4e0aa56ff98b5a468c2b5c9bf
Author: Daniel Vetter <daniel.vetter@ffwll.ch>
Date:   Thu Jan 24 14:44:56 2013 -0800

    drm/i915: vfuncs for ppgtt

Reviewed-by: Ben Widawsky <ben@bwidawsk.net>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agostyle nit: Align function parameter continuation properly.
Daniel Vetter [Tue, 19 Mar 2013 22:37:08 +0000 (23:37 +0100)]
style nit: Align function parameter continuation properly.

11 years agodrm/i915: VLV doesn't have HDMI on port C
Jesse Barnes [Fri, 8 Mar 2013 18:46:03 +0000 (10:46 -0800)]
drm/i915: VLV doesn't have HDMI on port C

Port C is for eDP.  Port B is shared between HDMI and DP.

Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: DSPFW and BLC regs are in the display offset range
Jesse Barnes [Fri, 8 Mar 2013 18:45:59 +0000 (10:45 -0800)]
drm/i915: DSPFW and BLC regs are in the display offset range

Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: set conservative clock gating values on VLV v2
Jesse Barnes [Fri, 8 Mar 2013 18:45:54 +0000 (10:45 -0800)]
drm/i915: set conservative clock gating values on VLV v2

We'll re-enable select bits as needed after testing and power measurement.

v2: split out wake handling bits (Jani)

Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: fix WaDisablePSDDualDispatchEnable on VLV v2
Jesse Barnes [Fri, 8 Mar 2013 18:45:51 +0000 (10:45 -0800)]
drm/i915: fix WaDisablePSDDualDispatchEnable on VLV v2

Can prevent a hang when we get to tessellation.  We need to set bit 15
as well for this workaround.

v2: update changelog with accurate info

Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: add more VLV IDs
Jesse Barnes [Fri, 8 Mar 2013 18:45:50 +0000 (10:45 -0800)]
drm/i915: add more VLV IDs

Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: use VLV DIP routines on VLV v2
Jesse Barnes [Fri, 8 Mar 2013 18:46:01 +0000 (10:46 -0800)]
drm/i915: use VLV DIP routines on VLV v2

This fixes up broken logic introduced in

commit 90b107c8f7ea75ef55db4e0515dda86b245f8978
Author: Shobhit Kumar <shobhit.kumar@intel.com>
Date:   Wed Mar 28 13:39:32 2012 -0700

    drm/i915: Enable HDMI on ValleyView

That one was probably a rebase fail along the way.

v2: clean up init ordering (Daniel)

Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
[danvet: Pimp commit message a bit.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: add media well to VLV force wake routines v2
Jesse Barnes [Fri, 8 Mar 2013 18:45:57 +0000 (10:45 -0800)]
drm/i915: add media well to VLV force wake routines v2

We could split this out into a separate routine at some point as an
optimization.

v2: use FORCEWAKE_KERNEL (Ville)

Note: Ville mentioned in his review that he declines to be responsible
if this blows up due to the lack of "readback a register != FW_ACK,
but from the same cacheline" magic we have in other forcewake
implementations.

Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
[danvet: Bikeshed overtly long lines according to checkpatch.pl. Nope,
this time around I didn't screw up printk message since I've left
those alone.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: don't use plane pipe select on VLV
Jesse Barnes [Fri, 8 Mar 2013 18:46:00 +0000 (10:46 -0800)]
drm/i915: don't use plane pipe select on VLV

Planes are fixed to pipes in VLV.

Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm: modify pages_to_sg prime helper to create optimized SG table
Rahul Sharma [Mon, 28 Jan 2013 13:38:48 +0000 (08:38 -0500)]
drm: modify pages_to_sg prime helper to create optimized SG table

It fixes the issue arises due to passing 'nr_pages' in place of 'nents' to
sg_alloc_table. When ARM_HAS_SG_CHAIN is disabled, it is causing failure in
creating SG table for the buffers having more than 204 physical pages i.e.
equal to SG_MAX_SINGLE_ALLOC.

When using sg_alloc_table_from_pages interface, in place of sg_alloc_table,
page list will be passes to get each contiguous section which is represented
by a single entry in the table. For a Contiguous Buffer, number of entries
should be equal to 1.

Following check is causing the failure which is not applicable for Non-Contig
buffers:

if (WARN_ON_ONCE(nents > max_ents))
return -EINVAL;

Above patch is well tested for EXYNOS4 and EXYNOS5 for with/wihtout IOMMU
supprot. NOUVEAU and RADEON platforms also depends on drm_prime_pages_to_sg
helper function.

This set is base on "exynos-drm-fixes" branch at
http://git.kernel.org/?p=linux/kernel/git/daeinki/drm-exynos.git

Signed-off-by: Rahul Sharma <rahul.sharma@samsung.com>
Acked-by: Dave Airlie <airlied@gmail.com>
Reviewed-by: Aaron Plattner <aplattner@nvidia.com>
Tested-by: Aaron Plattner <aplattner@nvidia.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: use for_each_sg_page for setting up the gtt ptes
Imre Deak [Mon, 18 Feb 2013 17:28:04 +0000 (19:28 +0200)]
drm/i915: use for_each_sg_page for setting up the gtt ptes

The existing gtt setup code is correct - and so doesn't need to be fixed to
handle compact dma scatter lists similarly to the previous patches. Still,
take the for_each_sg_page macro into use, to get somewhat simpler code.

Signed-off-by: Imre Deak <imre.deak@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: create compact dma scatter lists for gem objects
Imre Deak [Mon, 18 Feb 2013 17:28:03 +0000 (19:28 +0200)]
drm/i915: create compact dma scatter lists for gem objects

So far we created a sparse dma scatter list for gem objects, where each
scatter list entry represented only a single page. In the future we'll
have to handle compact scatter lists too where each entry can consist of
multiple pages, for example for objects imported through PRIME.

The previous patches have already fixed up all other places where the
i915 driver _walked_ these lists. Here we have the corresponding fix to
_create_ compact lists. It's not a performance or memory footprint
improvement, but it helps to better exercise the new logic.

Reference: http://www.spinics.net/lists/dri-devel/msg33917.html
Signed-off-by: Imre Deak <imre.deak@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: handle walking compact dma scatter lists
Imre Deak [Mon, 18 Feb 2013 17:28:02 +0000 (19:28 +0200)]
drm/i915: handle walking compact dma scatter lists

So far the assumption was that each dma scatter list entry contains only
a single page. This might not hold in the future, when we'll introduce
compact scatter lists, so prepare for this everywhere in the i915 code
where we walk such a list.

We'll fix the place _creating_ these lists separately in the next patch
to help the reviewing/bisectability.

Reference: http://www.spinics.net/lists/dri-devel/msg33917.html
Signed-off-by: Imre Deak <imre.deak@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: set dummy page for stolen objects
Imre Deak [Fri, 22 Mar 2013 21:10:44 +0000 (23:10 +0200)]
drm/i915: set dummy page for stolen objects

This is needed since currently sg_for_each_page assumes that we have
a valid page in each sg item. It is only a real problem for
CONFIG_SPARSEMEM where the page is dereferenced, in other cases the
iterator works ok with an invalid page pointer.

We can remove this workaround when we have fixed sg_page_iter to work on
scatterlists without backing pages.

Signed-off-by: Imre Deak <imre.deak@intel.com>
11 years agodrm: handle compact dma scatter lists in drm_clflush_sg()
Imre Deak [Mon, 18 Feb 2013 17:28:01 +0000 (19:28 +0200)]
drm: handle compact dma scatter lists in drm_clflush_sg()

So far the assumption was that each scatter list entry contains a single
page. This might not hold in the future, when we'll introduce compact
scatter lists, so prepare for this here.

Reference: http://www.spinics.net/lists/dri-devel/msg33917.html
Signed-off-by: Imre Deak <imre.deak@intel.com>
Acked-by: Dave Airlie <airlied@gmail.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agoMerge tag 'v3.9-rc3' into drm-intel-next-queued
Daniel Vetter [Tue, 19 Mar 2013 08:47:30 +0000 (09:47 +0100)]
Merge tag 'v3.9-rc3' into drm-intel-next-queued

Backmerge so that I can merge Imre Deak's coalesced sg entries fixes,
which depend upon the new for_each_sg_page introduce in

commit a321e91b6d73ed011ffceed384c40d2785cf723b
Author: Imre Deak <imre.deak@intel.com>
Date:   Wed Feb 27 17:02:56 2013 -0800

    lib/scatterlist: add simple page iterator

The merge itself is just two trivial conflicts:

Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: allow force wake at init time on VLV v2
Jesse Barnes [Fri, 8 Mar 2013 18:45:53 +0000 (10:45 -0800)]
drm/i915: allow force wake at init time on VLV v2

We need to set the 'allow force wake' bit to enable forcewake handling
later on.

v2: split from clock gating patch (Jani)
    check for allowwakeack (Ville)

Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: Remove unneeded dev argument
Ben Widawsky [Thu, 14 Mar 2013 00:21:06 +0000 (17:21 -0700)]
drm/i915: Remove unneeded dev argument

Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Acked-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: Remove unused file arg from execbuf
Ben Widawsky [Thu, 14 Mar 2013 00:21:05 +0000 (17:21 -0700)]
drm/i915: Remove unused file arg from execbuf

Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Acked-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agoLinux 3.9-rc3
Linus Torvalds [Sun, 17 Mar 2013 22:59:32 +0000 (15:59 -0700)]
Linux 3.9-rc3

11 years agoperf,x86: fix link failure for non-Intel configs
David Rientjes [Sun, 17 Mar 2013 22:49:10 +0000 (15:49 -0700)]
perf,x86: fix link failure for non-Intel configs

Commit 1d9d8639c063 ("perf,x86: fix kernel crash with PEBS/BTS after
suspend/resume") introduces a link failure since
perf_restore_debug_store() is only defined for CONFIG_CPU_SUP_INTEL:

arch/x86/power/built-in.o: In function `restore_processor_state':
(.text+0x45c): undefined reference to `perf_restore_debug_store'

Fix it by defining the dummy function appropriately.

Signed-off-by: David Rientjes <rientjes@google.com>
Cc: stable@vger.kernel.org
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agoperf,x86: fix wrmsr_on_cpu() warning on suspend/resume
Linus Torvalds [Sun, 17 Mar 2013 22:44:43 +0000 (15:44 -0700)]
perf,x86: fix wrmsr_on_cpu() warning on suspend/resume

Commit 1d9d8639c063 ("perf,x86: fix kernel crash with PEBS/BTS after
suspend/resume") fixed a crash when doing PEBS performance profiling
after resuming, but in using init_debug_store_on_cpu() to restore the
DS_AREA mtrr it also resulted in a new WARN_ON() triggering.

init_debug_store_on_cpu() uses "wrmsr_on_cpu()", which in turn uses CPU
cross-calls to do the MSR update.  Which is not really valid at the
early resume stage, and the warning is quite reasonable.  Now, it all
happens to _work_, for the simple reason that smp_call_function_single()
ends up just doing the call directly on the CPU when the CPU number
matches, but we really should just do the wrmsr() directly instead.

This duplicates the wrmsr() logic, but hopefully we can just remove the
wrmsr_on_cpu() version eventually.

Reported-and-tested-by: Parag Warudkar <parag.lkml@gmail.com>
Cc: stable@vger.kernel.org
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agodrm/i915: Resurrect ring kicking for semaphores, selectively
Chris Wilson [Thu, 14 Mar 2013 15:52:05 +0000 (17:52 +0200)]
drm/i915: Resurrect ring kicking for semaphores, selectively

Once we thought we got semaphores working, we disabled kicking the ring
if hangcheck fired whilst waiting upon a ring as it was doing more harm
than good:

commit 4e0e90dcb8a7df1229c69e30abebb59b0b3c2a1f
Author: Daniel Vetter <daniel.vetter@ffwll.ch>
Date:   Wed Dec 14 13:56:58 2011 +0100

    drm/i915: kicking rings stuck on semaphores considered harmful

However, life is never that easy and semaphores are still causing
problems whereby the value written by one ring (bcs) is not being
propagated to the waiter (rcs). Thus the waiter never wakes up and we
declare the GPU hung, which often has unfortunate consequences, even if
we successfully reset the GPU.

But the GPU is idle as it has completed the work, just didn't notify its
clients. So we can detect the incomplete wait during hang check and
probe the target ring to see if has indeed emitted the breadcrumb seqno
following the work and then and only then kick the waiter.

Based on a suggestion by Ben Widawsky.

v2: cross-check wait with iphdr. fix signaller calculation.

References: https://bugs.freedesktop.org/show_bug.cgi?id=54226
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Mika Kuoppala <mika.kuoppala@intel.com>
Cc: Daniel Vetter <daniel.vetter@ffwll.ch>
Cc: Ben Widawsky <ben@bwidawsk.net>
Acked-by: Ben Widawsky <ben@bwidawsk.net>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: add missing space in error message
Paulo Zanoni [Wed, 6 Mar 2013 23:03:22 +0000 (20:03 -0300)]
drm/i915: add missing space in error message

To avoid this:
[  256.798060] [drm] capturing error event; look for more information
in/sys/kernel/debug/dri/0/i915_error_state

Ben Widawsky identified that this regression has been introduced in

commit 2f86f1916504525a6fdd6b412374b4ebf1102cbe
Author: Ben Widawsky <ben@bwidawsk.net>
Date:   Mon Jan 28 15:32:15 2013 -0800

    drm/i915: Error state should print /sys/kernel/debug
        ...
    [danvet: split up long line.] <----- he did it
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Reviewed-by: Ben Widawsky <ben@bwidawsk.net>
[danvet: Pimp commit message with the regression note. Also, order
more brown paper bags, I've run out.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: don't save/restore PCH_LVDS on LPT
Paulo Zanoni [Wed, 6 Mar 2013 23:03:20 +0000 (20:03 -0300)]
drm/i915: don't save/restore PCH_LVDS on LPT

Because the register does not exist on LPT. The interesting fact is
that reading/writing PCH_LVDS on LPT does *not* give us "unclaimed
register" messages, but the register value is always 0.

Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: reorganize intel_lvds_supported
Paulo Zanoni [Wed, 6 Mar 2013 23:03:19 +0000 (20:03 -0300)]
drm/i915: reorganize intel_lvds_supported

Now it returns false for all platforms unless they're explicitly
listed on the function. There should be no real difference, except for
the fact that it now returns false on Haswell.

Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Reviewed-by: Ben Widawsky <ben@bwidawsk.net>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: add HAS_POWER_WELL
Paulo Zanoni [Wed, 6 Mar 2013 23:03:18 +0000 (20:03 -0300)]
drm/i915: add HAS_POWER_WELL

We're starting to add many IS_HASWELL checks for the power well code,
so add a HAS_POWER_WELL macro to properly document that we're checking
for hardware that has the power down well.

Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
[danvet: Resolve conflicts since some converted code was added by
not-yet merged patches.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: there's no DSPADDR register on Haswell
Paulo Zanoni [Wed, 6 Mar 2013 23:03:14 +0000 (20:03 -0300)]
drm/i915: there's no DSPADDR register on Haswell

So don't read it when we hang the GPU. This solves "unclaimed
register" messages.

Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Reviewed-by: Ben Widawsky <ben@bwidawsk.net>
[danvet: Future-proof by adding a gen >= 7 check in addition to the
!IS_HSW check from Paulo's original patch, suggested by Ben.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: there's no DSPSIZE register on gen4+
Paulo Zanoni [Wed, 6 Mar 2013 23:03:13 +0000 (20:03 -0300)]
drm/i915: there's no DSPSIZE register on gen4+

So don't read it when capturing the error state. This solves some
"unclaimed register" messages on Haswell when we hang the GPU.

Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Reviewed-by: Ben Widawsky <ben@bwidawsk.net>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: capture the correct cursor registers on IVB
Paulo Zanoni [Wed, 6 Mar 2013 23:03:12 +0000 (20:03 -0300)]
drm/i915: capture the correct cursor registers on IVB

This solves some "unclaimed register" messages when there's a GPU hang
on Haswell.

Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Reviewed-by: Ben Widawsky <ben@bwidawsk.net>
[danvet: Add missing IS_VLV check as spotted by Ville Syrjälä.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: disable sound first on intel_disable_ddi
Paulo Zanoni [Wed, 6 Mar 2013 23:03:09 +0000 (20:03 -0300)]
drm/i915: disable sound first on intel_disable_ddi

Our mode set sequence documentation says audio must be disabled first.

Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Reviewed-by: Ben Widawsky <ben@bwidawsk.net>
[danvet: Resolve conflict since the first patch in this series isn't
applied yet. Also bikeshed commit message as suggested by Ben.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agoMerge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/mason/linux...
Linus Torvalds [Sun, 17 Mar 2013 18:04:14 +0000 (11:04 -0700)]
Merge branch 'for-linus' of git://git./linux/kernel/git/mason/linux-btrfs

Pull btrfs fixes from Chris Mason:
 "Eric's rcu barrier patch fixes a long standing problem with our
  unmount code hanging on to devices in workqueue helpers.  Liu Bo
  nailed down a difficult assertion for in-memory extent mappings."

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/mason/linux-btrfs:
  Btrfs: fix warning of free_extent_map
  Btrfs: fix warning when creating snapshots
  Btrfs: return as soon as possible when edquot happens
  Btrfs: return EIO if we have extent tree corruption
  btrfs: use rcu_barrier() to wait for bdev puts at unmount
  Btrfs: remove btrfs_try_spin_lock
  Btrfs: get better concurrency for snapshot-aware defrag work

11 years agoBtrfs: fix warning of free_extent_map
Liu Bo [Fri, 15 Mar 2013 14:46:39 +0000 (08:46 -0600)]
Btrfs: fix warning of free_extent_map

Users report that an extent map's list is still linked when it's actually
going to be freed from cache.

The story is that

a) when we're going to drop an extent map and may split this large one into
smaller ems, and if this large one is flagged as EXTENT_FLAG_LOGGING which means
that it's on the list to be logged, then the smaller ems split from it will also
be flagged as EXTENT_FLAG_LOGGING, and this is _not_ expected.

b) we'll keep ems from unlinking the list and freeing when they are flagged with
EXTENT_FLAG_LOGGING, because the log code holds one reference.

The end result is the warning, but the truth is that we set the flag
EXTENT_FLAG_LOGGING only during fsync.

So clear flag EXTENT_FLAG_LOGGING for extent maps split from a large one.

Reported-by: Johannes Hirte <johannes.hirte@fem.tu-ilmenau.de>
Reported-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Liu Bo <bo.li.liu@oracle.com>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
11 years agoMerge branch 'kbuild' of git://git.kernel.org/pub/scm/linux/kernel/git/mmarek/kbuild
Linus Torvalds [Sat, 16 Mar 2013 01:06:55 +0000 (18:06 -0700)]
Merge branch 'kbuild' of git://git./linux/kernel/git/mmarek/kbuild

Pull kbuild fix from Michal Marek:
 "One fix for for make headers_install/headers_check to not require make
  3.81.  The requirement has been accidentally introduced in 3.7."

* 'kbuild' of git://git.kernel.org/pub/scm/linux/kernel/git/mmarek/kbuild:
  kbuild: fix make headers_check with make 3.80

11 years agoMerge tag 'for-3.9-rc3' of git://openrisc.net/jonas/linux
Linus Torvalds [Sat, 16 Mar 2013 01:05:37 +0000 (18:05 -0700)]
Merge tag 'for-3.9-rc3' of git://openrisc.net/jonas/linux

Pull OpenRISC bug fixes from Jonas Bonn:

 - The GPIO descriptor work has exposed how broken the non-GPIOLIB bits
   for OpenRISC were.  We now require GPIOLIB as this is the preferred
   way forward.

 - The system.h split introduced a bug in llist.h for arches using
   asm-generic/cmpxchg.h directly, which is currently only OpenRISC.
   The patch here moves two defines from asm-generic/atomic.h to
   asm-generic/cmpxchg.h to make things work as they should.

 - The VIRT_TO_BUS selector was added for OpenRISC, but OpenRISC does
   not have the virt_to_bus methods, so there's a patch to remove it
   again.

* tag 'for-3.9-rc3' of git://openrisc.net/jonas/linux:
  openrisc: remove HAVE_VIRT_TO_BUS
  asm-generic: move cmpxchg*_local defs to cmpxchg.h
  openrisc: require gpiolib

11 years agoMerge tag 'char-misc-3.9-rc2' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh...
Linus Torvalds [Sat, 16 Mar 2013 01:04:38 +0000 (18:04 -0700)]
Merge tag 'char-misc-3.9-rc2' of git://git./linux/kernel/git/gregkh/char-misc

Pull char/misc fixes from Greg Kroah-Hartman:
 "Here are some tiny fixes for the w1 drivers and the final removal
  patch for getting rid of CONFIG_EXPERIMENTAL (all users of it are now
  gone from your tree, this just drops the Kconfig item itself.)

  All have been in the linux-next tree for a while"

* tag 'char-misc-3.9-rc2' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/char-misc:
  final removal of CONFIG_EXPERIMENTAL
  w1: fix oops when w1_search is called from netlink connector
  w1-gpio: fix unused variable warning
  w1-gpio: remove erroneous __exit and __exit_p()
  ARM: w1-gpio: fix erroneous gpio requests

11 years agoMerge tag 'sound-3.9' of git://git.kernel.org/pub/scm/linux/kernel/git/tiwai/sound
Linus Torvalds [Sat, 16 Mar 2013 00:35:49 +0000 (17:35 -0700)]
Merge tag 'sound-3.9' of git://git./linux/kernel/git/tiwai/sound

Pull sound fixes from Takashi Iwai:
 "A collection of small fixes, as expected for the middle rc:
   - A couple of fixes for potential NULL dereferences and out-of-range
     array accesses revealed by static code parsers
   - A fix for the wrong error handling detected by trinity
   - A regression fix for missing audio on some MacBooks
   - CA0132 DSP loader fixes
   - Fix for EAPD control of IDT codecs on machines w/o speaker
   - Fix a regression in the HD-audio widget list parser code
   - Workaround for the NuForce UDH-100 USB audio"

* tag 'sound-3.9' of git://git.kernel.org/pub/scm/linux/kernel/git/tiwai/sound:
  ALSA: hda - Fix missing EAPD/GPIO setup for Cirrus codecs
  sound: sequencer: cap array index in seq_chn_common_event()
  ALSA: hda/ca0132 - Remove extra setting of dsp_state.
  ALSA: hda/ca0132 - Check download state of DSP.
  ALSA: hda/ca0132 - Check if dspload_image succeeded.
  ALSA: hda - Disable IDT eapd_switch if there are no internal speakers
  ALSA: hda - Fix snd_hda_get_num_raw_conns() to return a correct value
  ALSA: usb-audio: add a workaround for the NuForce UDH-100
  ALSA: asihpi - fix potential NULL pointer dereference
  ALSA: seq: Fix missing error handling in snd_seq_timer_open()

11 years agoMerge branch 'fixes-for-3.9' of git://git.linaro.org/people/mszyprowski/linux-dma...
Linus Torvalds [Sat, 16 Mar 2013 00:35:03 +0000 (17:35 -0700)]
Merge branch 'fixes-for-3.9' of git://git.linaro.org/people/mszyprowski/linux-dma-mapping

Pull DMA-mapping fix from Marek Szyprowski:
 "An important fix for all ARM architectures which use ZONE_DMA.
  Without it dma_alloc_* calls with GFP_ATOMIC flag might have allocated
  buffers outsize DMA zone."

* 'fixes-for-3.9' of git://git.linaro.org/people/mszyprowski/linux-dma-mapping:
  ARM: DMA-mapping: add missing GFP_DMA flag for atomic buffer allocation

11 years agoMerge tag 'mfd-fixes-3.9-1' of git://git.kernel.org/pub/scm/linux/kernel/git/sameo...
Linus Torvalds [Sat, 16 Mar 2013 00:34:01 +0000 (17:34 -0700)]
Merge tag 'mfd-fixes-3.9-1' of git://git./linux/kernel/git/sameo/mfd-fixes

Pull MFD fixes from Samuel Ortiz:
 "This is the first batch of MFD fixes for 3.9.

  With this one we have:

   - An ab8500 build failure fix.
   - An ab8500 device tree parsing fix.
   - A fix for twl4030_madc remove routine to work properly (when
     built-in).
   - A fix for properly registering palmas interrupt handler.
   - A fix for omap-usb init routine to actually write into the
     hostconfig register.
   - A couple of warning fixes for ab8500-gpadc and tps65912"

* tag 'mfd-fixes-3.9-1' of git://git.kernel.org/pub/scm/linux/kernel/git/sameo/mfd-fixes:
  mfd: twl4030-madc: Remove __exit_p annotation
  mfd: ab8500: Kill "reg" property from binding
  mfd: ab8500-gpadc: Complain if we fail to enable vtvout LDO
  mfd: wm831x: Don't forward declare enum wm831x_auxadc
  mfd: twl4030-audio: Fix argument type for twl4030_audio_disable_resource()
  mfd: tps65912: Declare and use tps65912_irq_exit()
  mfd: palmas: Provide irq flags through DT/platform data
  mfd: Make AB8500_CORE select POWER_SUPPLY to fix build error
  mfd: omap-usb-host: Actually update hostconfig

11 years agoMerge tag 'hwmon-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/groeck...
Linus Torvalds [Sat, 16 Mar 2013 00:33:13 +0000 (17:33 -0700)]
Merge tag 'hwmon-for-linus' of git://git./linux/kernel/git/groeck/linux-staging

Pull hwmon fixes from Guenter Roeck:
 "Bug fixes for pmbus, ltc2978, and lineage-pem drivers

  Added specific maintainer for some hwmon drivers"

* tag 'hwmon-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/groeck/linux-staging:
  hwmon: (pmbus/ltc2978) Fix temperature reporting
  hwmon: (pmbus) Fix krealloc() misuse in pmbus_add_attribute()
  hwmon: (lineage-pem) Add missing terminating entry for pem_[input|fan]_attributes
  MAINTAINERS: Add maintainer for MAX6697, INA209, and INA2XX drivers

11 years agoperf,x86: fix kernel crash with PEBS/BTS after suspend/resume
Stephane Eranian [Fri, 15 Mar 2013 13:26:07 +0000 (14:26 +0100)]
perf,x86: fix kernel crash with PEBS/BTS after suspend/resume

This patch fixes a kernel crash when using precise sampling (PEBS)
after a suspend/resume. Turns out the CPU notifier code is not invoked
on CPU0 (BP). Therefore, the DS_AREA (used by PEBS) is not restored properly
by the kernel and keeps it power-on/resume value of 0 causing any PEBS
measurement to crash when running on CPU0.

The workaround is to add a hook in the actual resume code to restore
the DS Area MSR value. It is invoked for all CPUS. So for all but CPU0,
the DS_AREA will be restored twice but this is harmless.

Reported-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Stephane Eranian <eranian@google.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agoALSA: hda - Fix missing EAPD/GPIO setup for Cirrus codecs
Takashi Iwai [Fri, 15 Mar 2013 13:23:32 +0000 (14:23 +0100)]
ALSA: hda - Fix missing EAPD/GPIO setup for Cirrus codecs

During the transition to the generic parser, the hook to the codec
specific automute function was forgotten.  This resulted in the silent
output on some MacBooks.

Signed-off-by: Takashi Iwai <tiwai@suse.de>
11 years agosound: sequencer: cap array index in seq_chn_common_event()
Dan Carpenter [Fri, 15 Mar 2013 06:14:22 +0000 (09:14 +0300)]
sound: sequencer: cap array index in seq_chn_common_event()

"chn" here is a number between 0 and 255, but ->chn_info[] only has
16 elements so there is a potential write beyond the end of the
array.

If the seq_mode isn't SEQ_2 then we let the individual drivers
(either opl3.c or midi_synth.c) handle it.  Those functions all
do a bounds check on "chn" so I haven't changed anything here.
The opl3.c driver has up to 18 channels and not 16.

Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
11 years agomfd: twl4030-madc: Remove __exit_p annotation
Arnd Bergmann [Thu, 14 Mar 2013 21:56:38 +0000 (22:56 +0100)]
mfd: twl4030-madc: Remove __exit_p annotation

4740f73fe5 "mfd: remove use of __devexit" removed the __devexit annotation
on the twl4030_madc_remove function, but left an __exit_p() present on the
pointer to this function. Using __exit_p was as wrong with the devexit in
place as it is now, but now we get a gcc warning about an unused function.

In order for the twl4030_madc_remove to work correctly in built-in code, we
have to remove the __exit_p.

Cc: Bill Pemberton <wfp5p@virginia.edu>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Samuel Ortiz <sameo@linux.intel.com>
11 years agoALSA: hda/ca0132 - Remove extra setting of dsp_state.
Dylan Reid [Fri, 15 Mar 2013 00:27:46 +0000 (17:27 -0700)]
ALSA: hda/ca0132 - Remove extra setting of dsp_state.

spec->dsp_state is initialized to DSP_DOWNLOAD_INIT, no need to reset
and check it in ca0132_download_dsp().

Signed-off-by: Dylan Reid <dgreid@chromium.org>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
11 years agoALSA: hda/ca0132 - Check download state of DSP.
Dylan Reid [Fri, 15 Mar 2013 00:27:45 +0000 (17:27 -0700)]
ALSA: hda/ca0132 - Check download state of DSP.

Instead of using the dspload_is_loaded() function, check the dsp_state
that is kept in the spec.  The dspload_is_loaded() function returns
true if the DSP transfer was never started.  This false-positive leads
to multiple second delays when ca0132_setup_efaults() times out on
each write.

Signed-off-by: Dylan Reid <dgreid@chromium.org>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
11 years agoALSA: hda/ca0132 - Check if dspload_image succeeded.
Dylan Reid [Fri, 15 Mar 2013 00:27:44 +0000 (17:27 -0700)]
ALSA: hda/ca0132 - Check if dspload_image succeeded.

If dspload_image() fails, it was ignored and dspload_wait_loaded() was
still called.  dsp_loaded should never be set to true in this case,
skip it.  The check in dspload_wait_loaded() return true if the DSP is
loaded or if it never started.

Signed-off-by: Dylan Reid <dgreid@chromium.org>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
11 years agomm/fremap.c: fix possible oops on error path
Michel Lespinasse [Thu, 14 Mar 2013 23:50:02 +0000 (16:50 -0700)]
mm/fremap.c: fix possible oops on error path

The vm_flags introduced in 6d7825b10dbe ("mm/fremap.c: fix oops on error
path") is supposed to avoid a compiler warning about unitialized
vm_flags without changing the generated code.

However I am concerned that this is going to be very brittle, and fail
with some compiler versions. The failure could be either of:

- compiler could actually load vma->vm_flags before checking for the
  !vma condition, thus reintroducing the oops

- compiler could optimize out the !vma check, since the pointer just got
  dereferenced shortly before (so the compiler knows it can't be NULL!)

I propose reversing this part of the change and initializing vm_flags to 0
just to avoid the bogus uninitialized use warning.

Signed-off-by: Michel Lespinasse <walken@google.com>
Cc: Tommi Rantala <tt.rantala@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agoMerge branch 'rcu/urgent' of git://git.kernel.org/pub/scm/linux/kernel/git/paulmck...
Linus Torvalds [Thu, 14 Mar 2013 21:53:07 +0000 (14:53 -0700)]
Merge branch 'rcu/urgent' of git://git./linux/kernel/git/paulmck/linux-rcu

Pull fix for hlist_entry_safe() regression from Paul McKenney:
 "This contains a single commit that fixes a regression in
  hlist_entry_safe().  This macro references its argument twice, which
  can cause NULL-pointer errors.  This commit applies a gcc statement
  expression, creating a temporary variable to avoid the double
  reference.  This has been posted to LKML at

    https://lkml.org/lkml/2013/3/9/75.

  Kudos to CAI Qian, whose testing uncovered this, to Eric Dumazet, who
  spotted root cause, and to Li Zefan, who tested this commit."

* 'rcu/urgent' of git://git.kernel.org/pub/scm/linux/kernel/git/paulmck/linux-rcu:
  list: Fix double fetch of pointer in hlist_entry_safe()

11 years agolist: Fix double fetch of pointer in hlist_entry_safe()
Paul E. McKenney [Sat, 9 Mar 2013 15:38:41 +0000 (07:38 -0800)]
list: Fix double fetch of pointer in hlist_entry_safe()

The current version of hlist_entry_safe() fetches the pointer twice,
once to test for NULL and the other to compute the offset back to the
enclosing structure.  This is OK for normal lock-based use because in
that case, the pointer cannot change.  However, when the pointer is
protected by RCU (as in "rcu_dereference(p)"), then the pointer can
change at any time.  This use case can result in the following sequence
of events:

1. CPU 0 invokes hlist_entry_safe(), fetches the RCU-protected
pointer as sees that it is non-NULL.

2. CPU 1 invokes hlist_del_rcu(), deleting the entry that CPU 0
just fetched a pointer to.  Because this is the last entry
in the list, the pointer fetched by CPU 0 is now NULL.

3. CPU 0 refetches the pointer, obtains NULL, and then gets a
NULL-pointer crash.

This commit therefore applies gcc's "({ })" statement expression to
create a temporary variable so that the specified pointer is fetched
only once, avoiding the above sequence of events.  Please note that
it is the caller's responsibility to use rcu_dereference() as needed.
This allows RCU-protected uses to work correctly without imposing
any additional overhead on the non-RCU case.

Many thanks to Eric Dumazet for spotting root cause!

Reported-by: CAI Qian <caiqian@redhat.com>
Reported-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Tested-by: Li Zefan <lizefan@huawei.com>
11 years agoMerge branch 'for_linus' of git://git.kernel.org/pub/scm/linux/kernel/git/jack/linux-fs
Linus Torvalds [Thu, 14 Mar 2013 19:11:28 +0000 (12:11 -0700)]
Merge branch 'for_linus' of git://git./linux/kernel/git/jack/linux-fs

Pull ext2, ext3, reiserfs, quota fixes from Jan Kara:
 "A fix for regression in ext2, and a format string issue in ext3.  The
  rest isn't too serious."

* 'for_linus' of git://git.kernel.org/pub/scm/linux/kernel/git/jack/linux-fs:
  ext2: Fix BUG_ON in evict() on inode deletion
  reiserfs: Use kstrdup instead of kmalloc/strcpy
  ext3: Fix format string issues
  quota: add missing use of dq_data_lock in __dquot_initialize

11 years agoBtrfs: fix warning when creating snapshots
Liu Bo [Wed, 13 Mar 2013 13:43:03 +0000 (07:43 -0600)]
Btrfs: fix warning when creating snapshots

Creating snapshot passes extent_root to commit its transaction,
but it can lead to the warning of checking root for quota in
the __btrfs_end_transaction() when someone else is committing
the current transaction.  Since we've recorded the needed root
in trans_handle, just use it to get rid of the warning.

Signed-off-by: Liu Bo <bo.li.liu@oracle.com>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
11 years agoBtrfs: return as soon as possible when edquot happens
Wang Shilong [Wed, 6 Mar 2013 11:51:47 +0000 (11:51 +0000)]
Btrfs: return as soon as possible when edquot happens

If one of qgroup fails to reserve firstly, we should return immediately,
it is unnecessary to continue check.

Signed-off-by: Wang Shilong <wangsl-fnst@cn.fujitsu.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
11 years agoBtrfs: return EIO if we have extent tree corruption
Josef Bacik [Fri, 8 Mar 2013 20:41:02 +0000 (15:41 -0500)]
Btrfs: return EIO if we have extent tree corruption

The callers of lookup_inline_extent_info all handle getting an error back
properly, so return an error if we have corruption instead of being a jerk and
panicing.  Still WARN_ON() since this is kind of crucial and I've been seeing it
a bit too much recently for my taste, I think we're doing something wrong
somewhere.  Thanks,

Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
11 years agobtrfs: use rcu_barrier() to wait for bdev puts at unmount
Eric Sandeen [Sat, 9 Mar 2013 15:18:39 +0000 (15:18 +0000)]
btrfs: use rcu_barrier() to wait for bdev puts at unmount

Doing this would reliably fail with -EBUSY for me:

# mount /dev/sdb2 /mnt/scratch; umount /mnt/scratch; mkfs.btrfs -f /dev/sdb2
...
unable to open /dev/sdb2: Device or resource busy

because mkfs.btrfs tries to open the device O_EXCL, and somebody still has it.

Using systemtap to track bdev gets & puts shows a kworker thread doing a
blkdev put after mkfs attempts a get; this is left over from the unmount
path:

btrfs_close_devices
__btrfs_close_devices
call_rcu(&device->rcu, free_device);
free_device
INIT_WORK(&device->rcu_work, __free_device);
schedule_work(&device->rcu_work);

so unmount might complete before __free_device fires & does its blkdev_put.

Adding an rcu_barrier() to btrfs_close_devices() causes unmount to wait
until all blkdev_put()s are done, and the device is truly free once
unmount completes.

Cc: stable@vger.kernel.org
Signed-off-by: Eric Sandeen <sandeen@redhat.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
11 years agoBtrfs: remove btrfs_try_spin_lock
Liu Bo [Mon, 11 Mar 2013 09:37:45 +0000 (09:37 +0000)]
Btrfs: remove btrfs_try_spin_lock

Remove a useless function declaration

Signed-off-by: Liu Bo <bo.li.liu@oracle.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>