GitHub/MotorolaMobilityLLC/kernel-slsi.git
11 years agodrm/i915: don't send DP "idle" pattern before "normal" on HSW PORT_A
Paulo Zanoni [Tue, 29 Jan 2013 18:35:18 +0000 (16:35 -0200)]
drm/i915: don't send DP "idle" pattern before "normal" on HSW PORT_A

The DP_TP_STATUS register for PORT_A doesn't exist. Our documentation
will be fixed soon, so the code does not match it for now.

This solves "Timed out waiting for DP idle patterns" and "unclaimed
register" messages on eDP.

V1: Was called "drm/i915: don't read DP_TP_STATUS(PORT_A)"
V2: Was called "drm/i915: don't send DP idle pattern before normal
pattern on HSW"
V3: Only change the code that touches PORT_A.

Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Reviewed-by: Jani Nikula <jani.nikula@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: don't run hsw power well code on !hsw
Daniel Vetter [Wed, 30 Jan 2013 14:59:56 +0000 (15:59 +0100)]
drm/i915: don't run hsw power well code on !hsw

Dumps annoying noise into the dmesg:

[drm:intel_set_power_well] *ERROR* Timeout enabling power well

Reported-by: Sedat Dilek <sedat.dilek@gmail.com>
Cc: Sedat Dilek <sedat.dilek@gmail.com>
Cc: Paulo Zanoni <paulo.r.zanoni@intel.com>
Tested-by: Sedat Dilek <sedat.dilek@gmail.com>
Reviewed-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: kill cargo-culted locking from power well code
Daniel Vetter [Wed, 30 Jan 2013 14:59:57 +0000 (15:59 +0100)]
drm/i915: kill cargo-culted locking from power well code

We may not concurrently change the power wells code. Which
is already guaranteed since modesets aren't concurrent. That
leaves races against setup/teardown/suspend/resume, and for
those we already (try) rather hard not to hit concurrent
modesets.

No debug WARN_ON added since that would require us to grab the
modeset locks in init/suspend code. Which is again just cargo
culting since just grabbing the locks in those paths isn't good
enough, we need the right order of operations, too.

Cc: Paulo Zanoni <paulo.r.zanoni@intel.com>
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@gmail.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: Only run idle processing from i915_gem_retire_requests_worker
Chris Wilson [Tue, 8 Jan 2013 11:02:57 +0000 (11:02 +0000)]
drm/i915: Only run idle processing from i915_gem_retire_requests_worker

When adding the fb idle detection to mark-inactive, it was forgotten
that userspace can drive the processing of retire-requests. We assumed
that it would be principally driven by the retire requests worker,
running once every second whilst active and so we would get the deferred
timer for free. Instead we spend too many CPU cycles reclocking the LVDS
preventing real work from being done.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reported-and-tested-by: Alexander Lam <lambchop468@gmail.com>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=58843
Cc: stable@vger.kernel.org
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@gmail.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: Fix CAGF for HSW
Ben Widawsky [Tue, 29 Jan 2013 20:00:15 +0000 (12:00 -0800)]
drm/i915: Fix CAGF for HSW

The shift changed, hurray.

Reported-by: Kenneth Graunke <kenneth@whitecape.org>
Cc: Paulo Zanoni <przanoni@gmail.com>
Reviewed-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Tested-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Cc: stable@vger.kernel.org
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: Reclaim GTT space for failed PPGTT
Ben Widawsky [Sat, 26 Jan 2013 00:41:04 +0000 (16:41 -0800)]
drm/i915: Reclaim GTT space for failed PPGTT

When the PPGTT init fails, we may as well reuse the space that we were
reserving for the PPGTT PDEs.

This also fixes an extraneous mutex_unlock.

Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Reviewed-by: Jani Nikula <jani.nikula@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: remove intel_gtt structure
Ben Widawsky [Thu, 24 Jan 2013 22:45:00 +0000 (14:45 -0800)]
drm/i915: remove intel_gtt structure

With the probe call in our dispatch table, we can now cut away the
last three remaining members in the intel_gtt shared struct and so
remove it completely.

v2: Rebased on top of Daniel's series

Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Reviewed-by: Damien Lespiau <damien.lespiau@intel.com>
[danvet: bikeshed commit message a bit.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: Add probe and remove to the gtt ops
Ben Widawsky [Thu, 24 Jan 2013 21:49:57 +0000 (13:49 -0800)]
drm/i915: Add probe and remove to the gtt ops

The idea, and much of the code came originally from:

commit 0712f0249c3148d8cf42a3703403c278590d4de5
Author: Ben Widawsky <ben@bwidawsk.net>
Date:   Fri Jan 18 17:23:16 2013 -0800

    drm/i915: Create a vtable for i915 gtt

Daniel didn't like the color of that patch series, and so I asked him to
start something which appealed to his sense of color. The preceding
patches are those, and now this is going on top of that.

[extracted from the original commit message]

One immediately obvious thing to implement is our gmch probing. The init
function was getting massively bloated. Fundamentally, all that's needed
from GMCH probing is the GTT size, and the stolen size. It makes design
sense to put the mappable calculation in there as well, but the code
turns out a bit nicer without it (IMO)

The intel_gtt bridge thing is still here, but the subsequent patches
will finish ripping that out.

Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
[danvet: Bikeshedded one comment (GMADR is just the PCI aperture, we
use it for other things than just accessing tiled surfaces through a
linear view) and cut the newly added long lines a bit. Also one
checkpatch error.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: extract hw ppgtt setup/cleanup code
Daniel Vetter [Thu, 24 Jan 2013 21:49:56 +0000 (13:49 -0800)]
drm/i915: extract hw ppgtt setup/cleanup code

At the moment only cosmetics, but being able to initialize/cleanup
arbitrary ppgtt address spaces paves the way to have more than one of
them ... Just in case we ever get around to implementing real
per-process address spaces. Note that in that case another vfunc for
ppgtt would be beneficial though. But that can wait until the code
grows a second place which initializes ppgtts.

Reviewed-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: pte_encode is gen6+
Daniel Vetter [Thu, 24 Jan 2013 22:44:57 +0000 (14:44 -0800)]
drm/i915: pte_encode is gen6+

All the other gen6+ hw code has the gen6_ prefix, so be consistent
about it.

Reviewed-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: vfuncs for ppgtt
Daniel Vetter [Thu, 24 Jan 2013 22:44:56 +0000 (14:44 -0800)]
drm/i915: vfuncs for ppgtt

Like for the global gtt we want a notch more flexibility here. Only
big change (besides a few tiny function parameter adjustments) was to
move gen6_ppgtt_insert_entries up (and remove _sg_ from its name, we
only have one kind of insert_entries since the last gtt cleanup).

We could also extract the platform ppgtt setup/teardown code a bit
better, but I don't care that much.

With this we have the hw details of pte writing nicely hidden away
behind a bit of abstraction. Which should pave the way for
different/multiple ppgtts (e.g. what we need for real ppgtt support).

Reviewed-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: vfuncs for gtt_clear_range/insert_entries
Daniel Vetter [Thu, 24 Jan 2013 22:44:55 +0000 (14:44 -0800)]
drm/i915: vfuncs for gtt_clear_range/insert_entries

We have a few too many differences here, so finally take the prepared
abstraction and run with it. A few smaller changes are required to get
things into shape:

- move i915_cache_level up since we need it in the gt funcs
- split up i915_ggtt_clear_range and move the two functions down to
  where the relevant insert_entries functions are
- adjustments to a few function parameter lists

Now we have 2 functions which deal with the gen6+ global gtt
(gen6_ggtt_ prefix) and 2 functions which deal with the legacy gtt
code in the intel-gtt.c fake agp driver (i915_ggtt_ prefix).

Init is still a bit a mess, but honestly I don't care about that.

One thing I've thought about while deciding on the exact interfaces is
a flag parameter for ->clear_range: We could use that to decide
between writing invalid pte entries or scratch pte entries. In case we
ever get around to fixing all our bugs which currently prevent us from
filling the gtt with empty ptes for the truly unused ranges ...

Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
[bwidawsk: Moved functions to the gtt struct]
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Reviewed-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: Error state should print /sys/kernel/debug
Ben Widawsky [Mon, 28 Jan 2013 23:32:15 +0000 (15:32 -0800)]
drm/i915: Error state should print /sys/kernel/debug

/sys/kernel/debug has more or less been the standard location of debugfs
for several years now. Other parts of DRM already use this location, so
we should as well.

Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Reviewed-by: Carl Worth <cworth@cworth.org>
Reviewed-by: Damien Lespiau <damien.lespiau@intel.com>
[danvet: split up long line.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: move DP save/restore into i915_ums.c
Daniel Vetter [Fri, 25 Jan 2013 16:53:22 +0000 (17:53 +0100)]
drm/i915: move DP save/restore into i915_ums.c

Note that this slightly changes the order, but we only move it within
the block of registers that restore encoder state. Specifically LVDS
is now restored after DP, whereas previously it was done before.

Legacy vga is still restored afterwards, which seems to be the
important thing (if there's anything important in this restore
ordering at all).

Reviewed-by: Imre Deak <imre.deak@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: dont save/restore VGA state for kms
Daniel Vetter [Fri, 25 Jan 2013 16:53:21 +0000 (17:53 +0100)]
drm/i915: dont save/restore VGA state for kms

The only thing we really care about that it is off. To do so, reuse
the recently created i915_redisable_vga function, which is already
used to put obnoxious firmware into check on lid reopening.

Reviewed-by: Imre Deak <imre.deak@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: extract ums suspend/resume into i915_ums.c
Daniel Vetter [Fri, 25 Jan 2013 16:53:20 +0000 (17:53 +0100)]
drm/i915: extract ums suspend/resume into i915_ums.c

Similarly to how i915_dma.c is shaping up to be the dungeon hole for
all things supporting dri1, create a new one to hide all the crazy
things which are only really useful for ums support. Biggest part is
the register suspend/resume support.

Unfortunately a lot of it is still intermingled with bits and pieces
we might still need, so needs more analysis and needs to stay in
i915_suspend.c for now.

Reviewed-by: Imre Deak <imre.deak@intel.com>
v2: s/modeset_reg/display_reg/ as suggested by Imre, to avoid
confusion between the kernel modeset code and display save/restore to
support ums.

v3: Fixup alphabetical order in the Makefile, spotted by Chris Wilson.

Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: move modeset checks out of save/restore_modeset_reg
Daniel Vetter [Fri, 25 Jan 2013 16:53:19 +0000 (17:53 +0100)]
drm/i915: move modeset checks out of save/restore_modeset_reg

That way the control flow is clearer, and it prepares the stage
to extract these ums functions and hide them somewhere.

There's still tons of display stuff outside of these, but that
requires more work.

Reviewed-by: Imre Deak <imre.deak@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: Implement WaVSRefCountFullforceMissDisable
Ben Widawsky [Sat, 26 Jan 2013 19:52:00 +0000 (11:52 -0800)]
drm/i915: Implement WaVSRefCountFullforceMissDisable

Implements WaVSRefCountFullforceMissDisable as documented in the BSpec
3D workarounds chapter.

Cc: Paulo Zanoni <przanoni@gmail.com>
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: turn on the power well before suspending
Paulo Zanoni [Fri, 25 Jan 2013 18:59:15 +0000 (16:59 -0200)]
drm/i915: turn on the power well before suspending

Our suspend code touches a lot of registers all over the place, so we
need to enable the power well before suspending.

Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Reviewed-by: Jani Nikula <jani.nikula@intel.com>
[danvet: Fixup compilation by stealing the header decl from the
dynamic power wells patch.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: set TRANSCODER_EDP even earlier
Paulo Zanoni [Fri, 25 Jan 2013 18:59:16 +0000 (16:59 -0200)]
drm/i915: set TRANSCODER_EDP even earlier

Instead of setting it at the beginning of haswell_crtc_mode_set, let's
set it at the beginning of intel_crtc_mode_set. When
intel_crt_mode_set calls drm_vblank_pre_modeset we already need to
have the transcoder_edp correctly set, because eventually
drm_vblank_pre_modeset calls functions that call i915_pipe_enabled
from i915_irq.c, which will read PIPECONF(cpu_transcoder).

This is a bug that affects us since we added support for
TRANSCODER_EDP, but I was only able to see the problem after
suspending a machine with the power well disabled (got an "unclaimed
register" error.

Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Reviewed-by: Jani Nikula <jani.nikula@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: only disable enabled planes on intel_fb_restore_mode
Paulo Zanoni [Fri, 25 Jan 2013 18:59:13 +0000 (16:59 -0200)]
drm/i915: only disable enabled planes on intel_fb_restore_mode

We should avoid touching registers that are on the power down well
when we don't need to, because if we touch these registers when the
power well is disabled we'll get tons of "unclaimed register"
messages. This commit fixes some of these messages.

Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: fix intel_init_power_wells
Paulo Zanoni [Fri, 25 Jan 2013 18:59:11 +0000 (16:59 -0200)]
drm/i915: fix intel_init_power_wells

The current code was wrong in many different ways, so this is a full
rewrite. We don't have "different power wells for different parts of
the GPU", we have a single power well, but we have multiple registers
that can be used to request enabling/disabling the power well. So
let's be a good citizen and only use the register we're suppose to
use, except when we're loading the driver, where we clear the request
made by the BIOS.

If any of the registers is requesting the power well to be enabled, it
will be enabled. If none of the registers is requesting the power well
to be enabled, it will be disabled.

For now we're just forcing the power well to be enabled, but in the
next commits we'll change this.

V2:
  - Remove debug messages that could be misleading due to possible
    race conditions with KVMr, Debug and BIOS.
  - Don't wait on disabling: after a conversaion with a hardware
    engineer we discovered that the "restriction" on bit 31 is just
    for the "enable" case, and we don't even need to wait on the
    "disable" case.

Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: SWF screatch registers need an offset on VLV
Ville Syrjälä [Thu, 24 Jan 2013 13:29:33 +0000 (15:29 +0200)]
drm/i915: SWF screatch registers need an offset on VLV

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: Include display_mmio_offset in sequencer index/data registers
Ville Syrjälä [Fri, 25 Jan 2013 19:44:45 +0000 (21:44 +0200)]
drm/i915: Include display_mmio_offset in sequencer index/data registers

SR01 needs to be touched to disable VGA on non-UMS setups too.
So the sequencer registers need to include the appripriate offset
on VLV.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: Pass VLV_DISPLAY_BASE + reg to intel_{hdmi, dp}_init on VLV
Ville Syrjälä [Fri, 25 Jan 2013 19:44:44 +0000 (21:44 +0200)]
drm/i915: Pass VLV_DISPLAY_BASE + reg to intel_{hdmi, dp}_init on VLV

When passing the DP/HDMI/SDVO registers to the encoder init functions,
include the VLV specific offset in the value.

v2: Resolved conflicts w/ VLV SDVO elimination

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: VLV doesn't have SDVO
Ville Syrjälä [Fri, 25 Jan 2013 19:44:43 +0000 (21:44 +0200)]
drm/i915: VLV doesn't have SDVO

Don't call intel_sdvo_init() for VLV.

Preserve the same behaviour as when intel_sdvo_init() would
have returned false.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: Always use adpa_reg
Ville Syrjälä [Fri, 25 Jan 2013 19:44:42 +0000 (21:44 +0200)]
drm/i915: Always use adpa_reg

Instead of using ADPA/VLV_ADPA/PCH_ADPA in various parts of
intel_crt code, just use adpa_reg which always contains the
correct value for the platform.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: PLL registers need an offset on VLV
Ville Syrjälä [Fri, 25 Jan 2013 19:44:41 +0000 (21:44 +0200)]
drm/i915: PLL registers need an offset on VLV

v2: Dropped the clock gating registers

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: Set display_mmio_offset for VLV
Ville Syrjälä [Thu, 24 Jan 2013 13:29:56 +0000 (15:29 +0200)]
drm/i915: Set display_mmio_offset for VLV

This will cause display registers to include the correct
offset on VLV.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: GPIO/GMBUS registers need an offset on VLV
Ville Syrjälä [Thu, 24 Jan 2013 13:29:55 +0000 (15:29 +0200)]
drm/i915: GPIO/GMBUS registers need an offset on VLV

GPIO/GMBUS registers must be offset on VLV, so simply
adjust gpio_mmio_base to include the correct offset.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: DPIO registers are VLV only and need an offset
Ville Syrjälä [Thu, 24 Jan 2013 13:29:53 +0000 (15:29 +0200)]
drm/i915: DPIO registers are VLV only and need an offset

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: Spell out VLV_DISPLAY_BASE for interrupt registers
Ville Syrjälä [Thu, 24 Jan 2013 13:29:52 +0000 (15:29 +0200)]
drm/i915: Spell out VLV_DISPLAY_BASE for interrupt registers

Instead of 0x18xxxx use (VLV_DISPLAY_BASE + xxxx).

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: Make VLV_GUNIT_CLOCK_GATE register value more readable
Ville Syrjälä [Thu, 24 Jan 2013 13:29:51 +0000 (15:29 +0200)]
drm/i915: Make VLV_GUNIT_CLOCK_GATE register value more readable

Instead of 0x18xxxx use (VLV_DISPLAY_BASE + xxxx).

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: FB_BLC_SELF_VLV is VLV only and needs an offset
Ville Syrjälä [Thu, 24 Jan 2013 13:29:48 +0000 (15:29 +0200)]
drm/i915: FB_BLC_SELF_VLV is VLV only and needs an offset

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: Pipe palette registers need an offset on VLV
Ville Syrjälä [Thu, 24 Jan 2013 13:29:47 +0000 (15:29 +0200)]
drm/i915: Pipe palette registers need an offset on VLV

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: Pipe timing registers need an offset on VLV
Ville Syrjälä [Thu, 24 Jan 2013 13:29:46 +0000 (15:29 +0200)]
drm/i915: Pipe timing registers need an offset on VLV

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: PORT_HOTPLUG registers need an offset on VLV
Ville Syrjälä [Thu, 24 Jan 2013 13:29:44 +0000 (15:29 +0200)]
drm/i915: PORT_HOTPLUG registers need an offset on VLV

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: Panel fitter registers need an offset on VLV
Ville Syrjälä [Thu, 24 Jan 2013 13:29:43 +0000 (15:29 +0200)]
drm/i915: Panel fitter registers need an offset on VLV

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: DPFLIPSTAT and DPINVGTT registers are VLV only and need an offset
Ville Syrjälä [Thu, 24 Jan 2013 13:29:41 +0000 (15:29 +0200)]
drm/i915: DPFLIPSTAT and DPINVGTT registers are VLV only and need an offset

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: DSPFW registers need an offset on VLV
Ville Syrjälä [Thu, 24 Jan 2013 13:29:39 +0000 (15:29 +0200)]
drm/i915: DSPFW registers need an offset on VLV

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: VLV_DDL is VLV only and needs an offset
Ville Syrjälä [Thu, 24 Jan 2013 13:29:38 +0000 (15:29 +0200)]
drm/i915: VLV_DDL is VLV only and needs an offset

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: Cursor registers need an offset on VLV
Ville Syrjälä [Thu, 24 Jan 2013 13:29:37 +0000 (15:29 +0200)]
drm/i915: Cursor registers need an offset on VLV

CURSIZE is not present on VLV, so it was left out, as were the IVB
specific cursor B registers.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: Pipe registers need an offset on VLV
Ville Syrjälä [Thu, 24 Jan 2013 13:29:36 +0000 (15:29 +0200)]
drm/i915: Pipe registers need an offset on VLV

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: Primary plane registers need an offset on VLV
Ville Syrjälä [Thu, 24 Jan 2013 13:29:35 +0000 (15:29 +0200)]
drm/i915: Primary plane registers need an offset on VLV

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: PIPE M/N registers need an offset on VLV
Ville Syrjälä [Thu, 24 Jan 2013 13:29:32 +0000 (15:29 +0200)]
drm/i915: PIPE M/N registers need an offset on VLV

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: VLV_VIDEO_DIP_CTL is for VLV only
Ville Syrjälä [Thu, 24 Jan 2013 13:29:31 +0000 (15:29 +0200)]
drm/i915: VLV_VIDEO_DIP_CTL is for VLV only

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: Per-pipe PP registers are for VLV only
Ville Syrjälä [Thu, 24 Jan 2013 13:29:30 +0000 (15:29 +0200)]
drm/i915: Per-pipe PP registers are for VLV only

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: AUD_VID_DID needs an offset on VLV
Ville Syrjälä [Thu, 24 Jan 2013 13:29:29 +0000 (15:29 +0200)]
drm/i915: AUD_VID_DID needs an offset on VLV

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: Add display_display_mmio_offset to intel_device_info
Ville Syrjälä [Thu, 24 Jan 2013 13:29:28 +0000 (15:29 +0200)]
drm/i915: Add display_display_mmio_offset to intel_device_info

Add an optional offset to intel_device_info, which will added
to most display register offsets.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: Convert intel_dp to enum port
Ville Syrjälä [Thu, 24 Jan 2013 13:29:27 +0000 (15:29 +0200)]
drm/i915: Convert intel_dp to enum port

Use intel_dig_port->port rather than intel_dp->output_reg.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: Convert intel_hdmi to enum port
Ville Syrjälä [Thu, 24 Jan 2013 13:29:26 +0000 (15:29 +0200)]
drm/i915: Convert intel_hdmi to enum port

Use intel_dig_port->port rather than intel_hdmi->sdvox_erg.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.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 save/restore DSPARB on gen5+
Paulo Zanoni [Fri, 18 Jan 2013 20:29:03 +0000 (18:29 -0200)]
drm/i915: don't save/restore DSPARB on gen5+

Because the register does not exist in gen5+.

This patch solves "unclaimed register" messages on Haswell after
suspend/resume.

Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Reviewed-by: Jani Nikula <jani.nikula@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: fixup sbi_read/write locking
Daniel Vetter [Tue, 22 Jan 2013 14:33:27 +0000 (15:33 +0100)]
drm/i915: fixup sbi_read/write locking

commit 09153000b8ca32a539a1207edebabd0d40b6c61b
Author: Daniel Vetter <daniel.vetter@ffwll.ch>
Date:   Wed Dec 12 14:06:44 2012 +0100

    drm/i915: rework locking for intel_dpio|sbi_read|write

reworked the locking around sbi_read/write functions for 3.8-fixes.
But

commit dde86e2db54545ef981b64805097a7b4c3156d6e
Author: Paulo Zanoni <paulo.r.zanoni@intel.com>
Date:   Sat Dec 1 12:04:25 2012 -0200

    drm/i915: add lpt_init_pch_refcl

Added new use-cases in the -next tree which has not been updated in
the merge. Fix it up.

Reported-by: Ben Widawsky <ben@bwidawsk.net>
Reviewed-by: Ben Widawsky <ben@bwidawsk.net>
Tested-by: Ben Widawsky <ben@bwidawsk.net>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: HDMI/DP - ELD info refresh support for Haswell
Wang Xingchao [Tue, 22 Jan 2013 15:25:25 +0000 (23:25 +0800)]
drm/i915: HDMI/DP - ELD info refresh support for Haswell

ELD info should be updated dynamically according to hot plug event.
For haswell chip, clear/set the eld valid bit and output enable bit
from callback intel_disable/eanble_ddi().

Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@gmail.com>
Signed-off-by: Wang Xingchao <xingchao.wang@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: use gem_set_seqno() on hardware init
Mika Kuoppala [Tue, 22 Jan 2013 12:12:17 +0000 (14:12 +0200)]
drm/i915: use gem_set_seqno() on hardware init

When machine was rebooted or module was reloaded,
gem_hw_init() set last_seqno to be identical to next_seqno.
This lead to situation that waits for first ever request
always passed immediately regardless if it was actually
executed.

Use gem_set_seqno() to be consistent how hw is
initialized on init, wrap and on resume.

Signed-off-by: Mika Kuoppala <mika.kuoppala@intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: add quirk to invert brightness on Packard Bell NCL20
Jani Nikula [Tue, 22 Jan 2013 10:50:36 +0000 (12:50 +0200)]
drm/i915: add quirk to invert brightness on Packard Bell NCL20

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=44156
Reported-by: Alan Zimmerman <alan.zimm@gmail.com>
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: add quirk to invert brightness on eMachines e725
Jani Nikula [Tue, 22 Jan 2013 10:50:35 +0000 (12:50 +0200)]
drm/i915: add quirk to invert brightness on eMachines e725

Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=31522#c35
[Note: There are more than one broken setups in the bug. This fixes one.]
Reported-by: Martins <andrissr@inbox.lv>
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: add quirk to invert brightness on eMachines G725
Jani Nikula [Tue, 22 Jan 2013 10:50:34 +0000 (12:50 +0200)]
drm/i915: add quirk to invert brightness on eMachines G725

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=59628
Reported-by: Roland Gruber <post@rolandgruber.de>
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: clarify concurrent hang detect/gpu reset consistency
Daniel Vetter [Thu, 6 Dec 2012 15:23:37 +0000 (16:23 +0100)]
drm/i915: clarify concurrent hang detect/gpu reset consistency

Damien Lespiau wondered how race the gpu reset/hang detection code is
against concurrent gpu resets/hang detections or combinations thereof.
Luckily the single work item is guranteed to never run concurrently,
so reset handling is already single-threaded.

Hence we only have to worry about concurrent hang detections, or a
hang detection firing off while we're still processing an older gpu
reset request. Due to the new mechanism of setting the reset in
progress flag and the ordering guaranteed by the schedule_work
function there's nothing to do but add a comment explaining why we're
safe.

The only thing I've noticed is that we still try to reset the gpu now,
even when it is declared terminally wedged. Add a check for that to
avoid continous warnings about failed resets, in case the hangcheck
timer ever gets stuck.

Reviewed-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: create a race-free reset detection
Daniel Vetter [Thu, 6 Dec 2012 08:01:42 +0000 (09:01 +0100)]
drm/i915: create a race-free reset detection

With the previous patch the state transition handling of the reset
code itself is now (hopefully) race free and solid. But that still
leaves out everyone else - with the various lock-free wait paths
we have there's the possibility that the reset happens between the
point where we read the seqno we should wait on and the actual wait.

And if __wait_seqno then never sees the RESET_IN_PROGRESS state, we'll
happily wait for a seqno which will in all likelyhood never signal.

In practice this is not a big problem since the X server gets
constantly interrupted, and can then submit more work (hopefully) to
unblock everyone else: As soon as a new seqno write lands, all waiters
will unblock. But running the i-g-t reset testcase ZZ_hangman can
expose this race, especially on slower hw with fewer cpu cores.

Now looking forward to ARB_robustness and friends that's not the best
possible behaviour, hence this patch adds a reset_counter to be able
to detect any reset, even if a given thread never observed the
in-progress state.

The important part is to correctly order things:
- The write side needs to increment the counter after any seqno gets
  reset.  Hence we need to do that at the end of the reset work, and
  again wake everyone up. We also need to place a barrier in between
  any possible seqno changes and the counter increment, since any
  unlock operations only guarantee that nothing leaks out, but not
  that at later load operation gets moved ahead.
- On the read side we need to ensure that no reset can sneak in and
  invalidate the seqno. In all cases we can use the one-sided barrier
  that unlock operations guarantee (of the lock protecting the
  respective seqno/ring pair) to ensure correct ordering. Hence it is
  sufficient to place the atomic read before the mutex/spin_unlock and
  no additional barriers are required.

The end-result of all this is that we need to wake up everyone twice
in a reset operation:
- First, before the reset starts, to get any lockholders of the locks,
  so that the reset can proceed.
- Second, after the reset is completed, to allow waiters to properly
  and reliably detect the reset condition and bail out.

I admit that this entire reset_counter thing smells a bit like
overkill, but I think it's justified since it makes it really explicit
what the bail-out condition is. And we need a reset counter anyway to
implement ARB_robustness, and imo with finer-grained locking on the
horizont this is the most resilient scheme I could think of.

v2: Drop spurious change in the wait_for_error EXIT_COND - we only
need to wait until we leave the reset-in-progress wedged state.

v3: Don't play tricks with barriers in the throttle ioctl, the
spin_unlock is barrier enough.

I've also considered using a little helper to grab the current
reset_counter, but then decided that hiding the atomic_read isn't a
great idea, since having it explicitly show up in the code is a nice
remainder to reviews to check the memory barriers.

v4: Add a comment to explain why we need to fall through in
__wait_seqno in the end variable assignments.

v5: Review from Damien:
- s/smb/smp/ in a comment
- don't increment the reset counter after we've set it to WEDGED. Now
  we (again) properly wedge the gpu when the reset fails.

Reviewed-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: Only apply the mb() when flushing the GTT domain during a finish
Chris Wilson [Tue, 9 Oct 2012 18:24:38 +0000 (19:24 +0100)]
drm/i915: Only apply the mb() when flushing the GTT domain during a finish

Now that we seem to have brought order to the GTT barriers, the last one
to review is the terminal barrier before we unbind the buffer from the
GTT. This needs to only be performed if the buffer still resides in the
GTT domain, and so we can skip some needless barriers otherwise.

Signed-off-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: Only insert the mb() before updating the fence parameter
Chris Wilson [Tue, 9 Oct 2012 18:24:37 +0000 (19:24 +0100)]
drm/i915: Only insert the mb() before updating the fence parameter

With a fence, we only need to insert a memory barrier around the actual
fence alteration for CPU accesses through the GTT. Performing the
barrier in flush-fence was inserting unnecessary and expensive barriers
for never fenced objects.

Note removing the barriers from flush-fence, which was effectively a
barrier before every direct access through the GTT, revealed that we
where missing a barrier before the first access through the GTT. Lack of
that barrier was sufficient to cause GPU hangs.

v2: Add a couple more comments to explain the new barriers

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Reviewed-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: clear up wedged transitions
Daniel Vetter [Thu, 15 Nov 2012 16:17:22 +0000 (17:17 +0100)]
drm/i915: clear up wedged transitions

We have two important transitions of the wedged state in the current
code:

- 0 -> 1: This means a hang has been detected, and signals to everyone
  that they please get of any locks, so that the reset work item can
  do its job.

- 1 -> 0: The reset handler has completed.

Now the last transition mixes up two states: "Reset completed and
successful" and "Reset failed". To distinguish these two we do some
tricks with the reset completion, but I simply could not convince
myself that this doesn't race under odd circumstances.

Hence split this up, and add a new terminal state indicating that the
hw is gone for good.

Also add explicit #defines for both states, update comments.

v2: Split out the reset handling bugfix for the throttle ioctl.

v3: s/tmp/wedged/ sugested by Chris Wilson. Also fixup up a rebase
error which prevented this patch from actually compiling.

v4: To unify the wedged state with the reset counter, keep the
reset-in-progress state just as a flag. The terminally-wedged state is
now denoted with a big number.

v5: Add a comment to the reset_counter special values explaining that
WEDGED & RESET_IN_PROGRESS needs to be true for the code to be
correct.

v6: Fixup logic errors introduced with the wedged+reset_counter
unification. Since WEDGED implies reset-in-progress (in a way we're
terminally stuck in the dead-but-reset-not-completed state), we need
ensure that we check for this everywhere. The specific bug was in
wait_for_error, which would simply have timed out.

v7: Extract an inline i915_reset_in_progress helper to make the code
more readable. Also annote the reset-in-progress case with an
unlikely, to help the compiler optimize the fastpath. Do the same for
the terminally wedged case with i915_terminally_wedged.

Reviewed-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-Off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: fix reset handling in the throttle ioctl
Daniel Vetter [Wed, 14 Nov 2012 16:14:06 +0000 (17:14 +0100)]
drm/i915: fix reset handling in the throttle ioctl

While auditing the code I've noticed one place (the throttle ioctl)
which does not yet wait for the reset handler to complete and doesn't
properly decode the wedge state into -EAGAIN/-EIO. Fix this up by
calling the right helpers. This might explain the oddball "my
compositor just died in a successfull gpu reset" reports. Or maybe not, since
current mesa doesn't use this ioctl to throttle command submission.

The throttle ioctl doesn't take the struct_mutex, so to avoid busy-looping
with -EAGAIN while a reset is in process, check for errors first and wait
for the handler to complete if a reset is pending by calling
i915_gem_wait_for_error.

Reviewed-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: move wedged to the other gpu error handling stuff
Daniel Vetter [Wed, 14 Nov 2012 16:14:05 +0000 (17:14 +0100)]
drm/i915: move wedged to the other gpu error handling stuff

And to make Ben Widawsky happier, use the gpu_error instead of
the entire device as the argument in some functions.

Drop the outdated comment on ->wedged for now, a follow-up patch will
change the semantics and add a proper comment again.

Reviewed-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: extract hangcheck/reset/error_state state into substruct
Daniel Vetter [Wed, 14 Nov 2012 16:14:04 +0000 (17:14 +0100)]
drm/i915: extract hangcheck/reset/error_state state into substruct

This has been sprinkled all over the place in dev_priv. I think
it'd be good to also move all the code into a separate file like
i915_gem_error.c, but that's for another patch.

Reviewed-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: move dev_priv->mm out of line
Daniel Vetter [Wed, 14 Nov 2012 16:14:03 +0000 (17:14 +0100)]
drm/i915: move dev_priv->mm out of line

Tha one is really big, since it contains tons of comments explaining
how things work. Which is nice ;-)

Reviewed-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agoagp/intel: Add gma_bus_addr
Ben Widawsky [Fri, 18 Jan 2013 20:30:34 +0000 (12:30 -0800)]
agp/intel: Add gma_bus_addr

It is no longer used in the i915 code, so isolate it from the shared
struct.

This was originally part of:
commit 0e275518f325418d559c05327775bff894b237f7
Author: Ben Widawsky <ben@bwidawsk.net>
Date:   Mon Jan 14 13:35:33 2013 -0800

    agp/intel: decouple more of the agp-i915 sharing

Reviewed-by: Rodrigo Vivi <rodrigo.vivi@gmail.com>
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
That commit had some other hunks which can't be used due to issues
Daniel found in previous commits.

Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
[danvet: drop squash notice from the commit since it's imo ok to keep
this one separate.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: Needs_dmar, not
Ben Widawsky [Fri, 18 Jan 2013 20:30:33 +0000 (12:30 -0800)]
drm/i915: Needs_dmar, not

The reasoning behind our code taking two paths depending upon whether or
not we may have been configured for IOMMU isn't clear to me. It should
always be safe to use the pci mapping functions as they are designed to
abstract the decision we were handling in i915.

Aside from simpler code, removing another member for the intel_gtt
struct is a nice motivation.

I ran this by Chris, and he wasn't concerned about the extra kzalloc,
and memory references vs. page_to_phys calculation in the case without
IOMMU.

v2: Update commit message

v3: Remove needs_dmar addition from Zhenyu upstream

This reverts (and then other stuff)
commit 20652097dadd9a7fb4d652f25466299974bc78f9
Author: Zhenyu Wang <zhenyuw@linux.intel.com>
Date:   Thu Dec 13 23:47:47 2012 +0800

    drm/i915: Fix missed needs_dmar setting

Reviewed-by: Rodrigo Vivi <rodrigo.vivi@gmail.com> (v2)
Cc: Zhenyu Wang <zhenyuw@linux.intel.com>
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
[danvet: Squash in follow-up fix to remove the bogus hunk which
deleted the dma_mask configuration for gen6+.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: Remove scratch page from shared
Ben Widawsky [Fri, 18 Jan 2013 20:30:32 +0000 (12:30 -0800)]
drm/i915: Remove scratch page from shared

We already had a mapping in both (minus the phys_addr in AGP).

Reviewed-by: Rodrigo Vivi <rodrigo.vivi@gmail.com>
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: Cut out the infamous ILK w/a from AGP layer
Ben Widawsky [Fri, 18 Jan 2013 20:30:31 +0000 (12:30 -0800)]
drm/i915: Cut out the infamous ILK w/a from AGP layer

And, move it to where the rest of the logic is.

There is some slight functionality changes. There was extra paranoid
checks in AGP code making sure we never do idle maps on gen2 parts. That
was not duplicated as the simple PCI id check should do the right thing.

v2: use IS_GEN5 && IS_MOBILE check instead. For now, this is the same as
IS_IRONLAKE_M but is more future proof. The workaround docs hint that
more than one platform may be effected, but we've never seen such a
platform in the wild. (Rodrigo, Daniel)

Reviewed-by: Rodrigo Vivi <rodrigo.vivi@gmail.com> (v1)
Cc: Dave Airlie <airlied@redhat.com>
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: Provide the quantization range in the AVI infoframe
Ville Syrjälä [Thu, 17 Jan 2013 14:31:31 +0000 (16:31 +0200)]
drm/i915: Provide the quantization range in the AVI infoframe

The AVI infoframe is able to inform the display whether the source is
sending full or limited range RGB data.

As per CEA-861 [1] we must first check whether the display reports the
quantization range as selectable, and if so we can set the approriate
bits in the AVI inforframe.

[1] CEA-861-E - 6.4 Format of Version 2 AVI InfoFrame

v2: Give the Q bits better names, add spec chapter information

Reviewed-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/edid: Add drm_rgb_quant_range_selectable()
Ville Syrjälä [Thu, 17 Jan 2013 14:31:30 +0000 (16:31 +0200)]
drm/edid: Add drm_rgb_quant_range_selectable()

drm_rgb_quant_range_selectable() will report whether the monitor
claims to support for RGB quantization range selection.

The information can be found in the CEA Video capability block.

v2: s/quantzation/quantization/ in the comment

Reviewed-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Acked-by: David Airlie <airlied@linux.ie>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: Add "Automatic" mode for the "Broadcast RGB" property
Ville Syrjälä [Thu, 17 Jan 2013 14:31:29 +0000 (16:31 +0200)]
drm/i915: Add "Automatic" mode for the "Broadcast RGB" property

Add a new "Automatic" mode to the "Broadcast RGB" range property.
When selected the driver automagically selects between full range and
limited range output.

Based on CEA-861 [1] guidelines, limited range output is selected if the
mode is a CEA mode, except 640x480. Otherwise full range output is used.
Additionally DVI monitors should most likely default to full range
always.

As per DP1.2a [2] DisplayPort should always use full range for 18bpp, and
otherwise will follow CEA-861 rules.

NOTE: The default value for the property will now be "Automatic"
so some people may be affected in case they're relying on the
current full range default.

[1] CEA-861-E - 5.1 Default Encoding Parameters
[2] VESA DisplayPort Ver.1.2a - 5.1.1.1 Video Colorimetry

v2: Use has_hdmi_sink to check if a HDMI monitor is present
v3: Add information about relevant spec chapters

Reviewed-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: Fix RGB color range property for PCH platforms
Ville Syrjälä [Thu, 17 Jan 2013 14:31:28 +0000 (16:31 +0200)]
drm/i915: Fix RGB color range property for PCH platforms

The RGB color range select bit on the DP/SDVO/HDMI registers
disappeared when PCH was introduced, and instead a new PIPECONF bit
was added that performs the same function.

Add a new INTEL_MODE_LIMITED_COLOR_RANGE private mode flag, and set
it in the encoder mode_fixup if limited color range is requested.
Set the the PIPECONF bit 13 based on the flag.

Experimentation showed that simply toggling the bit while the pipe is
active doesn't work. We need to restart the pipe, which luckily already
happens.

The DP/SDVO/HDMI bit 8 is marked MBZ in the docs, so avoid setting it,
although it doesn't seem to do any harm in practice.

TODO:
- the PIPECONF bit too seems to have disappeared from HSW. Need a
  volunteer to test if it's just a documentation issue or if it's really
  gone. If the bit is gone and no easy replacement is found, then I suppose
  we may need to use the pipe CSC unit to perform the range compression.

v2: Use mode private_flags instead of intel_encoder virtual functions
v3: Moved the intel_dp color_range handling after bpc check to help
    later patches

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=46800
Reviewed-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: Remove use of gtt_mappable_entries
Ben Widawsky [Thu, 17 Jan 2013 20:45:17 +0000 (12:45 -0800)]
drm/i915: Remove use of gtt_mappable_entries

Mappable_end, ie. size is almost always what you want as opposed to the
number of entries. Since we already have that information, we can scrap
the number of entries and only calculate it when needed.

If gtt_start is !0, this will have slightly different behavior. This
difference can only occur in DRI1, and exists when we try to kick out
the firmware fb. The new code seems like a bugfix to me.

The other case where we've changed the behavior is during init we check
the mappable region against our current known upper and lower limits
(64MB, and 512MB). This now matches the comment, and makes things more
convenient after removing gtt_mappable_entries.

Also worth noting is the setting of mappable_end is taken out of setup
because we do it earlier now in the DRI2 case and therefore need to add
that tiny hunk to support the DRI1 IOCTL.

v2: Move up mappable end to before legacy AGP init

v3: Add the dev_priv inclusion here from previous rebase error in patch
5

Reviewed-by: Rodrigo Vivi <rodrigo.vivi@gmail.com> (v2)
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
[danvet: squash in fix for a printk format flag mismatch warning.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: Remove use on gma_bus_addr on gen6+
Ben Widawsky [Thu, 17 Jan 2013 20:45:16 +0000 (12:45 -0800)]
drm/i915: Remove use on gma_bus_addr on gen6+

We have enough info to not use the intel_gtt bridge stuff.

v2: Move setup of mappable_base above the legacy init stuff because we
still need that on older platforms. (Daniel)

v3: Remove the dev_priv hunk which was rebased in by accident

Reviewed-by: Rodrigo Vivi <rodrigo.vivi@gmail.com> (v2)
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: Create a gtt structure
Ben Widawsky [Thu, 17 Jan 2013 20:45:15 +0000 (12:45 -0800)]
drm/i915: Create a gtt structure

The purpose of the gtt structure is to help isolate our gtt specific
properties from the rest of the code (in doing so it help us finish the
isolation from the AGP connection).

The following members are pulled out (and renamed):
gtt_start
gtt_total
gtt_mappable_end
gtt_mappable
gtt_base_addr
gsm

The gtt structure will serve as a nice place to put gen specific gtt
routines in upcoming patches. As far as what else I feel belongs in this
structure: it is meant to encapsulate the GTT's physical properties.
This is why I've not added fields which track various drm_mm properties,
or things like gtt_mtrr (which is itself a pretty transient field).

Reviewed-by: Rodrigo Vivi <rodrigo.vivi@gmail.com>
[Ben modified commit messages]
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: Remove gtt_mappable_total
Ben Widawsky [Thu, 17 Jan 2013 20:45:14 +0000 (12:45 -0800)]
drm/i915: Remove gtt_mappable_total

With the assertion from the previous patch in place, it should be safe
to get rid gtt_mappable_total. Keeps things saner to not have to track
the same info in two places.

In order to keep the diff as simple as possible and keep with the
existing gtt_setup semantics we opt to keep gtt_mappable_end. It's not
as consistent with the 'total' used in the previous patch, but that can
be fixed later.

Reviewed-by: Rodrigo Vivi <rodrigo.vivi@gmail.com>
[Ben modified commit message]
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: Mappable_end can't ever be > end
Ben Widawsky [Thu, 17 Jan 2013 20:45:13 +0000 (12:45 -0800)]
drm/i915: Mappable_end can't ever be > end

Both DRI1 and DRI2 can never specify a mappable size which goes past the
GTT size.  Don't pretend otherwise.

Reviewed-by: Rodrigo Vivi <rodrigo.vivi@gmail.com>
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: Kill gtt_end
Ben Widawsky [Thu, 17 Jan 2013 20:45:12 +0000 (12:45 -0800)]
drm/i915: Kill gtt_end

It's duplicated in the more useful gtt_total.

Reviewed-by: Rodrigo Vivi <rodrigo.vivi@gmail.com>
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: Fix SPRITE0_FLIP_DONE_INT_EN_VLV and SPRITE0_FLIPDONE_INT_STATUS_VLV
Ville Syrjälä [Wed, 16 Jan 2013 17:59:03 +0000 (19:59 +0200)]
drm/i915: Fix SPRITE0_FLIP_DONE_INT_EN_VLV and SPRITE0_FLIPDONE_INT_STATUS_VLV

Fix up some copypaste errors in the PIPESTAT register for VLV.

SPRITE0_FLIP_DONE_INT_EN_VLV is bit 22, not bit 26.

SPRITE0_FLIPDONE_INT_STATUS_VLV is bit 14, not bit 15.

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: Use the reloc.handle as an index into the execbuffer array
Chris Wilson [Tue, 8 Jan 2013 10:53:17 +0000 (10:53 +0000)]
drm/i915: Use the reloc.handle as an index into the execbuffer array

Using copywinwin10 as an example that is dependent upon emitting a lot
of relocations (2 per operation), we see improvements of:

c2d/gm45: 618000.0/sec to 623000.0/sec.
i3-330m: 748000.0/sec to 789000.0/sec.

(measured relative to a baseline with neither optimisations applied).

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: Allow userspace to hint that the relocations were known
Daniel Vetter [Thu, 17 Jan 2013 21:23:36 +0000 (22:23 +0100)]
drm/i915: Allow userspace to hint that the relocations were known

Userspace is able to hint to the kernel that its command stream and
auxiliary state buffers already hold the correct presumed addresses and
so the relocation process may be skipped if the kernel does not need to
move any buffers in preparation for the execbuffer. Thus for the common
case where the allotment of buffers is static between batches, we can
avoid the overhead of individually checking the relocation entries.

Note that this requires userspace to supply the domain tracking and
requests for workarounds itself that would otherwise be computed based
upon the relocation entries.

Using copywinwin10 as an example that is dependent upon emitting a lot
of relocations (2 per operation), we see improvements of:

c2d/gm45: 618000.0/sec to 632000.0/sec.
i3-330m: 748000.0/sec to 830000.0/sec.

(measured relative to a baseline with neither optimisations applied).

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Imre Deak <imre.deak@intel.com>
[danvet: Fixup merge conflict in userspace header due to different
baseline trees.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: Move the execbuffer objects list from the stack into the tracker
Chris Wilson [Tue, 8 Jan 2013 10:53:15 +0000 (10:53 +0000)]
drm/i915: Move the execbuffer objects list from the stack into the tracker

Instead of passing around the eb-objects hashtable and a separate object
list, we can include the object list into the eb-objects structure for
convenience.

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: Take the handle idr spinlock once for looking up the exec objects
Chris Wilson [Tue, 8 Jan 2013 10:53:14 +0000 (10:53 +0000)]
drm/i915: Take the handle idr spinlock once for looking up the exec objects

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: Mark a temporary allocation for copy-from-user as such
Chris Wilson [Tue, 8 Jan 2013 10:53:13 +0000 (10:53 +0000)]
drm/i915: Mark a temporary allocation for copy-from-user as such

The difference is that the kernel will then know that this memory will
be reclaimable in the near future.

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: Bail if we attempt to allocate pages for a purged object
Chris Wilson [Tue, 8 Jan 2013 10:53:09 +0000 (10:53 +0000)]
drm/i915: Bail if we attempt to allocate pages for a purged object

Move the existing checking inside bind_to_gtt() to the more appropriate
layer in order to prevent recreation of the pages after they have been
explicitly truncated.

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: Add a debug interface to forcibly evict and shrink our object caches
Chris Wilson [Tue, 15 Jan 2013 12:39:35 +0000 (12:39 +0000)]
drm/i915: Add a debug interface to forcibly evict and shrink our object caches

As a means to investigate some bad system behaviour related to the
purging of the active, inactive and unbound lists, it is useful to be
able to manually control when those lists should be cleared.

v2: use _safe list iterators as we kick objects from the list as we
walk.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
[danvet: Add a small comment explaining why we don't need to check and
wait for gpu resets, acked by Chris on irc.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: use gtt_get_size() instead of open coding it
Imre Deak [Mon, 7 Jan 2013 19:47:35 +0000 (21:47 +0200)]
drm/i915: use gtt_get_size() instead of open coding it

Signed-off-by: Imre Deak <imre.deak@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: merge {i965, sandybridge}_write_fence_reg()
Imre Deak [Mon, 7 Jan 2013 19:47:34 +0000 (21:47 +0200)]
drm/i915: merge {i965, sandybridge}_write_fence_reg()

The two functions are rather similar, so merge them.

Signed-off-by: Imre Deak <imre.deak@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: merge get_gtt_alignment/get_unfenced_gtt_alignment()
Imre Deak [Mon, 7 Jan 2013 19:47:33 +0000 (21:47 +0200)]
drm/i915: merge get_gtt_alignment/get_unfenced_gtt_alignment()

The two functions are rather similar, so merge them.

Signed-off-by: Imre Deak <imre.deak@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: Remove pch_rq_mask from struct drm_i915_private.
Egbert Eich [Thu, 10 Jan 2013 15:02:39 +0000 (10:02 -0500)]
drm/i915: Remove pch_rq_mask from struct drm_i915_private.

This variable is only used locally in the irq postinstall
functions for ivybridge and ironlake.

Signed-off-by: Egbert Eich <eich@suse.de>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: wake up all pageflip waiters
Daniel Vetter [Thu, 20 Dec 2012 20:24:07 +0000 (21:24 +0100)]
drm/i915: wake up all pageflip waiters

Otherwise it seems like we can get stuck with concurrent waiters.
Right now this /shouldn't/ be a problem, since all pending pageflip
waiters are serialized by the one mode_config.mutex, so there's at
most on waiter. But better paranoid than sorry, since this is tricky
code.

v2: WARN_ON(waitqueue_active) before waiting, as suggested by Chris
Wilson.

Acked-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agoMerge tag 'drm-intel-next-2012-12-21' of git://people.freedesktop.org/~danvet/drm...
Dave Airlie [Thu, 17 Jan 2013 10:34:08 +0000 (20:34 +1000)]
Merge tag 'drm-intel-next-2012-12-21' of git://people.freedesktop.org/~danvet/drm-intel into drm-next

Daniel writes:
- seqno wrap fixes and debug infrastructure from Mika Kuoppala and Chris
  Wilson
- some leftover kill-agp on gen6+ patches from Ben
- hotplug improvements from Damien
- clear fb when allocated from stolen, avoids dirt on the fbcon (Chris)
- Stolen mem support from Chris Wilson, one of the many steps to get to
  real fastboot support.
- Some DDI code cleanups from Paulo.
- Some refactorings around lvds and dp code.
- some random little bits&pieces

* tag 'drm-intel-next-2012-12-21' of git://people.freedesktop.org/~danvet/drm-intel: (93 commits)
  drm/i915: Return the real error code from intel_set_mode()
  drm/i915: Make GSM void
  drm/i915: Move GSM mapping into dev_priv
  drm/i915: Move even more gtt code to i915_gem_gtt
  drm/i915: Make next_seqno debugs entry to use i915_gem_set_seqno
  drm/i915: Introduce i915_gem_set_seqno()
  drm/i915: Always clear semaphore mboxes on seqno wrap
  drm/i915: Initialize hardware semaphore state on ring init
  drm/i915: Introduce ring set_seqno
  drm/i915: Missed conversion to gtt_pte_t
  drm/i915: Bug on unsupported swizzled platforms
  drm/i915: BUG() if fences are used on unsupported platform
  drm/i915: fixup overlay stolen memory leak
  drm/i915: clean up PIPECONF bpc #defines
  drm/i915: add intel_dp_set_signal_levels
  drm/i915: remove leftover display.update_wm assignment
  drm/i915: check for the PCH when setting pch_transcoder
  drm/i915: Clear the stolen fb before enabling
  drm/i915: Access to snooped system memory through the GTT is incoherent
  drm/i915: Remove stale comment about intel_dp_detect()
  ...

Conflicts:
drivers/gpu/drm/i915/intel_display.c

11 years agoLinux 3.8-rc3
Linus Torvalds [Thu, 10 Jan 2013 02:59:55 +0000 (18:59 -0800)]
Linux 3.8-rc3

11 years agoMerge branch 'fixes' of git://git.linaro.org/people/rmk/linux-arm
Linus Torvalds [Wed, 9 Jan 2013 16:58:57 +0000 (08:58 -0800)]
Merge branch 'fixes' of git://git.linaro.org/people/rmk/linux-arm

Pull ARM fixes from Russell King.

* 'fixes' of git://git.linaro.org/people/rmk/linux-arm:
  ARM: 7616/1: cache-l2x0: aurora: Use writel_relaxed instead of writel
  ARM: 7615/1: cache-l2x0: aurora: Invalidate during clean operation with WT enable
  ARM: 7614/1: mm: fix wrong branch from Cortex-A9 to PJ4b
  ARM: 7612/1: imx: Do not select some errata that depends on !ARCH_MULTIPLATFORM
  ARM: 7611/1: VIC: fix bug in VIC irqdomain code
  ARM: 7610/1: versatile: bump IRQ numbers
  ARM: 7609/1: disable errata work-arounds which access secure registers
  ARM: 7608/1: l2x0: Only set .set_debug on PL310 r3p0 and earlier

11 years agoMerge tag 'edac_fixes_for_3.8' of git://git.kernel.org/pub/scm/linux/kernel/git/bp/bp
Linus Torvalds [Wed, 9 Jan 2013 16:43:56 +0000 (08:43 -0800)]
Merge tag 'edac_fixes_for_3.8' of git://git./linux/kernel/git/bp/bp

Pull EDAC fixes from Borislav Petkov:
 "Two error path fixes causing a crash and a Kconfig fix for an issue
  which spilled all EDAC suboptions into the 'Device Drivers' menu."

* tag 'edac_fixes_for_3.8' of git://git.kernel.org/pub/scm/linux/kernel/git/bp/bp:
  EDAC: Cleanup device deregistering path
  EDAC: Fix EDAC Kconfig menu
  EDAC: Fix kernel panic on module unloading

11 years agomm: reinstante dropped pmd_trans_splitting() check
Linus Torvalds [Wed, 9 Jan 2013 16:36:54 +0000 (08:36 -0800)]
mm: reinstante dropped pmd_trans_splitting() check

The check for a pmd being in the process of being split was dropped by
mistake by commit d10e63f29488 ("mm: numa: Create basic numa page
hinting infrastructure"). Put it back.

Reported-by: Dave Jones <davej@redhat.com>
Debugged-by: Hillf Danton <dhillf@gmail.com>
Acked-by: Andrea Arcangeli <aarcange@redhat.com>
Acked-by: Mel Gorman <mgorman@suse.de>
Cc: Kirill Shutemov <kirill@shutemov.name>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agocred: Remove tgcred pointer from struct cred
Marc Dionne [Wed, 9 Jan 2013 14:16:30 +0000 (14:16 +0000)]
cred: Remove tgcred pointer from struct cred

Commit 3a50597de863 ("KEYS: Make the session and process keyrings
per-thread") removed the definition of the thread_group_cred structure,
but left a now unused pointer in struct cred.

Signed-off-by: Marc Dionne <marc.c.dionne@gmail.com>
Signed-off-by: David Howells <dhowells@redhat.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>