platform/kernel/linux-arm64.git
11 years agodrm/crtc-helper: explicit DPMS on after modeset
Daniel Vetter [Fri, 14 Jun 2013 22:13:13 +0000 (00:13 +0200)]
drm/crtc-helper: explicit DPMS on after modeset

Atm the crtc helper implementation of set_config has really
inconsisten semantics: If just an fb update is good enough, dpms state
will be left as-is, but if we do a full modeset we force everything to
dpms on.

This change has already been applied to the i915 modeset code in

commit e3de42b68478a8c95dd27520e9adead2af9477a5
Author: Imre Deak <imre.deak@intel.com>
Date:   Fri May 3 19:44:07 2013 +0200

    drm/i915: force full modeset if the connector is in DPMS OFF mode

which according to Greg KH seems to aim for a new record in most
Bugzilla: links in a commit message.

The history of this dpms forcing is pretty interesting. This patch
here is an almost-revert of

commit 811aaa55ba21ab37407018cfc01770d6b037d3fb
Author: Keith Packard <keithp@keithp.com>
Date:   Thu Feb 3 16:57:28 2011 -0800

    drm: Only set DPMS ON when actually configuring a mode

which fixed the bug of trying to dpms on disabled outputs, but
introduced the new discrepancy between an fb update only and full
modesets. The actual introduction of this goes back to

commit bf9dc102e284a5aa78c73fc9d72e11d5ccd8669f
Author: Keith Packard <keithp@keithp.com>
Date:   Fri Nov 26 10:45:58 2010 -0800

    drm: Set connector DPMS status to ON in drm_crtc_helper_set_config

And if you'd dig around in the i915 driver code there's even more fun
around forcing dpms on and losing our heads and temper of the
resulting inconsistencies. Especially the DP re-training code had tons
of funny stuff in it.

Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Dave Airlie <airlied@redhat.com>
11 years agodrm/crtc-helper: no need to check for fb->depth/bpp
Daniel Vetter [Fri, 14 Jun 2013 22:13:12 +0000 (00:13 +0200)]
drm/crtc-helper: no need to check for fb->depth/bpp

... since we already check for fb->pixel_format, which encodes all
this. The other two fields are only for backwards compat of older
drivers (and we might want to look into eventually just killing them).

Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Dave Airlie <airlied@redhat.com>
11 years agodrm/crtc-helpers: Enforce sane set_config api
Daniel Vetter [Fri, 14 Jun 2013 22:13:11 +0000 (00:13 +0200)]
drm/crtc-helpers: Enforce sane set_config api

There's no point in trying to clean up after driver-bugs, so just blow
up. Furthermore it's an interface abuse to set no mode but have an fb
and aslo to try to set an fb without enough connectors. These two
spefici cases of interface abuse have been committed by the fb helper,
but that's been fixed meanwhile in

commit 7e53f3a423146745a4e4bb93362d488dfad502a8
Author: Daniel Vetter <daniel.vetter@ffwll.ch>
Date:   Mon Jan 21 10:52:17 2013 +0100

    drm/fb-helper: fixup set_config semantics

The i915 driver has been shipping since a while with these BUGs with
no reports, so should be save.

Note that this drops an ugly case where we clear crtc->fb behind the
upper levels back and so cause a refcounting mayhem, which Russell
Kins spotted while trying to hunt down a drm framebuffer leak.

Reported-by: Russell King <linux@arm.linux.org.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Dave Airlie <airlied@redhat.com>
11 years agodrm_vm: drop explicit VM_IO setting
Al Viro [Sun, 23 Jun 2013 20:26:50 +0000 (06:26 +1000)]
drm_vm: drop explicit VM_IO setting

io_remap_pfn_range already sets this internally.

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Dave Airlie <airlied@redhat.com>
11 years agogpu:drm:tilcdc: get preferred_bpp value from DT
Benoit Parrot [Tue, 18 Jun 2013 22:18:31 +0000 (17:18 -0500)]
gpu:drm:tilcdc: get preferred_bpp value from DT

The preferred_bpp value in currently hard-coded to 16.
This causes color corruption on the am335x-evm lcd panel which
requires 32 bpp instead. This changes attempts to use the configured
bpp value from the DT or built-in panel-info struct.

Signed-off-by: Benoit Parrot <bparrot@ti.com>
Acked-by: Rob Clark <robdclark@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
11 years agodrm/shmobile: Drop usage of removed drm_plane enabled field
Laurent Pinchart [Wed, 19 Jun 2013 14:45:50 +0000 (16:45 +0200)]
drm/shmobile: Drop usage of removed drm_plane enabled field

The enabled field has been removed from struct drm_plane. Don't use it
in the driver.

Signed-off-by: Laurent Pinchart <laurent.pinchart+renesas@ideasonboard.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
11 years agodrm/mgag200: Don't do full cleanup if mgag200_device_init fails
Christopher Harvey [Wed, 5 Jun 2013 15:29:57 +0000 (11:29 -0400)]
drm/mgag200: Don't do full cleanup if mgag200_device_init fails

Running mgag200_driver_unload when the driver init fails early on
causes functions like drm_mode_config_cleanup to be called. The
problem is, drm_mode_config_cleanup crashes because the corresponding
init hasn't happend yet. There really isn't anything to cleanup after
mgag200_device_init, so we can just pass the error code upwards.

Acked-by: Julia Lemire <jlemire@matrox.com>
Signed-off-by: Christopher Harvey <charvey@matrox.com>
Acked-by: Paul Menzel <paulepanter@users.sourceforge.net>
Signed-off-by: Dave Airlie <airlied@gmail.com>
11 years agodrm/mgag200: Hardware cursor support
Christopher Harvey [Wed, 5 Jun 2013 19:24:26 +0000 (15:24 -0400)]
drm/mgag200: Hardware cursor support

G200 cards support, at best, 16 colour palleted images for the cursor
so we do a conversion in the cursor_set function, and reject cursors
with more than 16 colours, or cursors with partial transparency. Xorg
falls back gracefully to software cursors in this case.

We can't disable/enable the cursor hardware without causing momentary
corruption around the cursor. Instead, once the cursor is on we leave
it on, and simulate turning the cursor off by moving it
offscreen. This works well.

Since we can't disable -> update -> enable the cursors, we double
buffer cursor icons, then just move the base address that points to
the old cursor, to the new. This also works well, but uses an extra
page of memory.

The cursor buffers are lazily-allocated on first cursor_set. This is
to make sure they don't take priority over any framebuffers in case of
limited memory.

Here is a representation of how the bitmap for the cursor is mapped in G200 memory :

  Each line of color cursor use 6 Slices of 8 bytes. Slices 0 to 3
  are used for the 4bpp bitmap, slice 4 for XOR mask and slice 5 for
  AND mask. Each line has the following format:

      //      Byte 0  Byte 1  Byte 2  Byte 3  Byte 4  Byte 5  Byte 6 Byte 7
      //
      // S0:  P00-01  P02-03  P04-05  P06-07  P08-09  P10-11  P12-13 P14-15
      // S1:  P16-17  P18-19  P20-21  P22-23  P24-25  P26-27  P28-29 P30-31
      // S2:  P32-33  P34-35  P36-37  P38-39  P40-41  P42-43  P44-45 P46-47
      // S3:  P48-49  P50-51  P52-53  P54-55  P56-57  P58-59  P60-61 P62-63
      // S4:  X63-56  X55-48  X47-40  X39-32  X31-24  X23-16  X15-08 X07-00
      // S5:  A63-56  A55-48  A47-40  A39-32  A31-24  A23-16  A15-08 A07-00
      //
      //       S0 to S5      = Slices 0 to 5
      //       P00 to P63    = Bitmap - pixels 0 to 63
      //       X00 to X63    = always 0 - pixels 0 to 63
      //       A00 to A63    = transparent markers - pixels 0 to 63
      //                       1 means colour, 0 means transparent

Signed-off-by: Christopher Harvey <charvey@matrox.com>
Signed-off-by: Mathieu Larouche <mathieu.larouche@matrox.com>
Acked-by: Julia Lemire <jlemire@matrox.com>
Tested-by: Julia Lemire <jlemire@matrox.com>
Signed-off-by: Dave Airlie <airlied@gmail.com>
11 years agodrm: Drop all the stub gamma_get, gamma_set, load_lut functions from drivers
Ville Syrjälä [Mon, 27 May 2013 17:19:58 +0000 (20:19 +0300)]
drm: Drop all the stub gamma_get, gamma_set, load_lut functions from drivers

Many of the drivers didn't implement palette/gamma handling, but were forced
to provide stubs for the hooks to avoid drm_fb_helper from oopsing. Now that
the hooks are optional, we can eliminate all the stubs.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Dave Airlie <airlied@gmail.com>
11 years agodrm/fb-helper: Make load_lut and gamma_set/gamma_get hooks optional
Ville Syrjälä [Mon, 27 May 2013 17:19:57 +0000 (20:19 +0300)]
drm/fb-helper: Make load_lut and gamma_set/gamma_get hooks optional

Check whether the crtc provides the load_lut callback before calling it.
This allows the driver to provide the hook only for those CRTCs that
actually have the hardware support for it.

Also check whether the driver provided the fb_helper gamma_set/gamma_get
hooks. It's a driver bug if it allows non-truecolor fbdev visuals w/o
these hooks, but auditing all the drivers is too tedious. So just slap
a big WARN_ON() there and bail out before things start to explode.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Dave Airlie <airlied@gmail.com>
11 years agodrm/fb-helper: Don't clobber the display palette when fbdev isn't bound
Ville Syrjälä [Mon, 27 May 2013 17:19:56 +0000 (20:19 +0300)]
drm/fb-helper: Don't clobber the display palette when fbdev isn't bound

Perform the drm_fb_helper_is_bound() check to avoid clobbering the
display palette of some other KMS client.

While at it, fix up the locking by grabbing all modeset locks for the
duration of the fb_setcmap operation.

v2: Make a note of the locking changes in the commit message

Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Dave Airlie <airlied@gmail.com>
11 years agodrm: Remove some unused stuff from drm_plane
Ville Syrjälä [Mon, 3 Jun 2013 13:11:42 +0000 (16:11 +0300)]
drm: Remove some unused stuff from drm_plane

There's a bunch of unused members inside drm_plane, bloating the size of
the structure needlessly. Eliminate them.

v2: Remove all of it from kernel-doc too

Reviewed-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Dave Airlie <airlied@gmail.com>
11 years agodrm/i915: Drop bogus fbdev sprite disable code
Ville Syrjälä [Mon, 3 Jun 2013 13:11:41 +0000 (16:11 +0300)]
drm/i915: Drop bogus fbdev sprite disable code

plane->enabled is never set, so this code didn't do anything.

Also drm_fb_helper_restore_fbdev_mode() will now disable all cursors
and sprites for us, so we don't have to bother anymore.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Dave Airlie <airlied@gmail.com>
11 years agodrm: Add kernel-doc for plane functions
Ville Syrjälä [Wed, 5 Jun 2013 12:39:56 +0000 (15:39 +0300)]
drm: Add kernel-doc for plane functions

v2: Follow the drm_crtc documentation fixes

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Dave Airlie <airlied@gmail.com>
11 years agodrm/fb-helper: Disable cursors and planes when restoring fbdev mode
Ville Syrjälä [Mon, 3 Jun 2013 13:10:42 +0000 (16:10 +0300)]
drm/fb-helper: Disable cursors and planes when restoring fbdev mode

Cursors and plane can obscure whatever fbdev wants to show the user.
Disable them all in drm_fb_helper_restore_fbdev_mode.

After the cursors and planes have been disabled, user space needs to
explicitly re-enable them to make them visible again.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Dave Airlie <airlied@gmail.com>
11 years agodrm: Add drm_plane_force_disable()
Ville Syrjälä [Mon, 3 Jun 2013 13:10:40 +0000 (16:10 +0300)]
drm: Add drm_plane_force_disable()

drm_plane_force_disable() will forcibly disable the plane even if user
had previously requested the plane to be enabled.

This can be used to force planes to be off when restoring the fbdev
mode.

The code was simply pulled from drm_framebuffer_remove(), which now
calls the new function as well.

v2: Check plane->fb in drm_plane_force_disable(), drop bogus comment
    about disabling crtc

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Dave Airlie <airlied@gmail.com>
11 years agodrm/edid: Add both 60Hz and 59.94Hz CEA modes to connector's mode list
Ville Syrjälä [Fri, 31 May 2013 12:23:41 +0000 (15:23 +0300)]
drm/edid: Add both 60Hz and 59.94Hz CEA modes to connector's mode list

Having both modes can be beneficial for video playback cases. If you can
match the video framerate exactly, and the audio and video clocks come
from the same source, you should be able to avoid dropped/repeated
frames without expensive operations such as resampling the audio to
match video output rate.

Rather than add both variants based on the CEA extension short video
descriptors in do_cea_modes(), add only one variant there. Once all
the EDID has been fully probed, do a loop over the entire probed mode
list, during which we add the other variants for all modes that match
CEA modes. This allows us to match modes that didn't come via the CEA
short video descriptors. For example one Samsung TV here doesn't have
the 640x480-60 mode as a SVD, but instead it's specified via a detailed
timing descriptor.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Dave Airlie <airlied@redhat.com>
11 years agodrm/vmwgfx: Don't access file_priv in cursor_set when handle==0
Ville Syrjälä [Mon, 3 Jun 2013 13:10:41 +0000 (16:10 +0300)]
drm/vmwgfx: Don't access file_priv in cursor_set when handle==0

We want to disable the cursor by calling ->cursor_set() with handle=0
from places where we don't have a file_priv, so don't try to access it
unless necessary.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Jakob Bornecrantz <jakob@vmware.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
11 years agodrm: Improve drm_crtc documentation
Ville Syrjälä [Wed, 5 Jun 2013 12:39:55 +0000 (12:39 +0000)]
drm: Improve drm_crtc documentation

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
11 years agoMerge tag 'drm-intel-next-2013-06-01' of git://people.freedesktop.org/~danvet/drm...
Dave Airlie [Mon, 10 Jun 2013 22:38:56 +0000 (08:38 +1000)]
Merge tag 'drm-intel-next-2013-06-01' of git://people.freedesktop.org/~danvet/drm-intel into drm-next

Daniel writes:
Another round of drm-intel-next for 3.11. Highlights:
- Haswell IPS support (Paulo Zanoni)
- VECS support on Haswell (Ben Widawsky, Xiang Haihao, ...)
- Haswell watermark fixes (Paulo Zanoni)
- "Make the gun bigger again" multithread fence fix from Chris.
- i915_error_state finnally no longer fails with -ENOMEM! Big thanks to
  Mika for tackling this.
- vlv sideband locking fixes from Jani
- Hangcheck prep work for arb_robustness support (Mika&Chris)
- edp vs cpu port confusion clean-up from Imre
- pile of smaller fixes and cleanups all over.

* tag 'drm-intel-next-2013-06-01' of git://people.freedesktop.org/~danvet/drm-intel: (70 commits)
  drm/i915: add i915_ips_status debugfs entry
  drm/i915: add enable_ips module option
  drm/i915: implement IPS feature
  drm/i915: fix up the edp power well check
  drm/i915: add I915_PARAM_HAS_VEBOX to i915_getparam
  drm/i915: add I915_EXEC_VEBOX to i915_gem_do_execbuffer()
  drm/i915: add VEBOX into debugfs
  drm/i915: Enable vebox interrupts
  drm/i915: vebox interrupt get/put
  drm/i915: consolidate interrupt naming scheme
  drm/i915: Convert irq_refounct to struct
  drm/i915: make PM interrupt writes non-destructive
  drm/i915: Add PM regs to pre/post install
  drm/i915: Create an ivybridge_irq_preinstall
  drm/i915: Create a more generic pm handler for hsw+
  drm/i915: add support for 5/6 data buffer partitioning on Haswell
  drm/i915: properly set HSW WM_LP watermarks
  drm/i915: properly set HSW WM_PIPE registers
  drm/i915: fix pch_nop support
  drm/i915: Vebox ringbuffer init
  ...

11 years agodrm: Sort connector modes based on vrefresh
Ville Syrjälä [Fri, 31 May 2013 12:17:08 +0000 (12:17 +0000)]
drm: Sort connector modes based on vrefresh

Keeping the modes sorted by vrefresh before the pixel clock makes the
mode list somehow more pleasing to the eye.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
11 years agodrm: Add probed modes in probe order
Ville Syrjälä [Fri, 31 May 2013 12:17:07 +0000 (12:17 +0000)]
drm: Add probed modes in probe order

Keeping the modes in the same order as we probe them makes it a bit
easier to track what's happening.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
11 years agodrm: Preserve the list head in drm_mode_copy
Ville Syrjälä [Fri, 31 May 2013 12:17:06 +0000 (12:17 +0000)]
drm: Preserve the list head in drm_mode_copy

Preserve the destination mode's list head in drm_mode_copy. Just
in case someone decides that it's a good idea to overwrite a mode which
happens to be on some list,

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
11 years agodrm/cma: Cocci spatch "ptr_ret.spatch"
Thomas Meyer [Sat, 1 Jun 2013 09:56:46 +0000 (09:56 +0000)]
drm/cma: Cocci spatch "ptr_ret.spatch"

Signed-off-by: Thomas Meyer <thomas@m3y3r.de>
Signed-off-by: Dave Airlie <airlied@redhat.com>
11 years agodrm/prime: Cocci spatch "err_cast.spatch"
Thomas Meyer [Sat, 1 Jun 2013 10:09:27 +0000 (10:09 +0000)]
drm/prime: Cocci spatch "err_cast.spatch"

Signed-off-by: Thomas Meyer <thomas@m3y3r.de>
Signed-off-by: Dave Airlie <airlied@redhat.com>
11 years agodrm: Cocci spatch "memdup.spatch"
Thomas Meyer [Wed, 22 May 2013 21:06:30 +0000 (21:06 +0000)]
drm: Cocci spatch "memdup.spatch"

Signed-off-by: Thomas Meyer <thomas@m3y3r.de>
Signed-off-by: Dave Airlie <airlied@redhat.com>
11 years agoMerge branch 'drm/next' of git://linuxtv.org/pinchartl/fbdev into drm-next
Dave Airlie [Mon, 10 Jun 2013 22:14:21 +0000 (08:14 +1000)]
Merge branch 'drm/next' of git://linuxtv.org/pinchartl/fbdev into drm-next

GEM CMA PRIME support from Laurent.

* 'drm/next' of git://linuxtv.org/pinchartl/fbdev:
  drm: GEM CMA: Add DRM PRIME support
  drm: GEM CMA: Split object mapping into GEM mapping and CMA mapping
  drm: GEM CMA: Split object creation into object alloc and DMA memory alloc
  drm/omap: Use drm_gem_mmap_obj() to implement dma-buf mmap
  drm/gem: Split drm_gem_mmap() into object search and object mapping

11 years agodrm: Constify the pretty-print functions
Ville Syrjälä [Fri, 7 Jun 2013 15:43:07 +0000 (15:43 +0000)]
drm: Constify the pretty-print functions

The structures and strings involved with various pretty-print functions
aren't meant to be modified, so make them all const. The exception is
drm_connector_enum_list which does get modified in drm_connector_init().

While at it move the drm_get_connector_status_name() prototype from
drmP.h to drm_crtc.h where it belongs.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Dave Airlie <airlied@redhat.com>
11 years agodrm/i915: Print pretty names for pixel formats
Ville Syrjälä [Fri, 7 Jun 2013 15:43:05 +0000 (15:43 +0000)]
drm/i915: Print pretty names for pixel formats

Use drm_get_format_name to print more readable pixel format names
in debug output.

Also unify the debug messages to say "unsupported pixel format",
which better describes what is going on.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Dave Airlie <airlied@redhat.com>
11 years agodrm: Print pretty names for pixel formats
Ville Syrjälä [Mon, 10 Jun 2013 08:15:10 +0000 (11:15 +0300)]
drm: Print pretty names for pixel formats

Rather than just printing the pixel format as a hex number, decode the
fourcc into human readable form, and also decode the LE vs. BE flag.

Keep printing the raw hex number too in case it contains non-printable
characters.

Some examples what the new drm_get_format_name() produces:
DRM_FORMAT_XRGB8888: "XR24 little-endian (0x34325258)"
DRM_FORMAT_YUYV: "YUYV little-endian (0x56595559)"
DRM_FORMAT_RGB565|DRM_FORMAT_BIG_ENDIAN: "RG16 big-endian (0xb6314752)"
Unprintable characters: "D??? big-endian (0xff7f0244)"

v2: Fix patch author

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
11 years agodrm: move pinning/unpinning to buffer attach
Maarten Lankhorst [Tue, 9 Apr 2013 07:52:54 +0000 (09:52 +0200)]
drm: move pinning/unpinning to buffer attach

This allows importing bo's to own device to work without requiring that the buffer is pinned in GART.

Signed-off-by: Maarten Lankhorst <maarten.lankhorst@canonical.com>
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Dave Airlie <airlied@redhat.com>
11 years agodrm: add unpin function to prime helpers
Maarten Lankhorst [Tue, 9 Apr 2013 07:18:44 +0000 (09:18 +0200)]
drm: add unpin function to prime helpers

Prevents buffers from being pinned forever.

Signed-off-by: Maarten Lankhorst <maarten.lankhorst@canonical.com>
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Dave Airlie <airlied@redhat.com>
11 years agodrm: GEM CMA: Add DRM PRIME support
Laurent Pinchart [Sun, 17 Feb 2013 00:57:30 +0000 (01:57 +0100)]
drm: GEM CMA: Add DRM PRIME support

Signed-off-by: Laurent Pinchart <laurent.pinchart+renesas@ideasonboard.com>
Reviewed-by: Rob Clark <robdclark@gmail.com>
11 years agodrm: GEM CMA: Split object mapping into GEM mapping and CMA mapping
Laurent Pinchart [Tue, 16 Apr 2013 12:32:34 +0000 (14:32 +0200)]
drm: GEM CMA: Split object mapping into GEM mapping and CMA mapping

The CMA-specific mapping code will be used to implement dma-buf mmap
support.

Signed-off-by: Laurent Pinchart <laurent.pinchart+renesas@ideasonboard.com>
Signed-off-by: Rob Clark <robdclark@gmail.com>
11 years agodrm: GEM CMA: Split object creation into object alloc and DMA memory alloc
Laurent Pinchart [Sun, 17 Feb 2013 00:54:26 +0000 (01:54 +0100)]
drm: GEM CMA: Split object creation into object alloc and DMA memory alloc

This allows creating a GEM CMA object without an associated DMA memory
buffer, and will be used to implement DRM PRIME support.

Signed-off-by: Laurent Pinchart <laurent.pinchart+renesas@ideasonboard.com>
Reviewed-by: Rob Clark <robdclark@gmail.com>
11 years agodrm/omap: Use drm_gem_mmap_obj() to implement dma-buf mmap
Laurent Pinchart [Tue, 16 Apr 2013 12:21:23 +0000 (14:21 +0200)]
drm/omap: Use drm_gem_mmap_obj() to implement dma-buf mmap

The dma-buf mmap code was copied from the GEM mmap implementation.
Replace it with the new drm_gem_mmap_obj() function.

Signed-off-by: Laurent Pinchart <laurent.pinchart+renesas@ideasonboard.com>
Reviewed-by: Rob Clark <robdclark@gmail.com>
11 years agodrm/gem: Split drm_gem_mmap() into object search and object mapping
Laurent Pinchart [Tue, 16 Apr 2013 12:14:52 +0000 (14:14 +0200)]
drm/gem: Split drm_gem_mmap() into object search and object mapping

The drm_gem_mmap() function first finds the GEM object to be mapped
based on the fake mmap offset and then maps the object. Split the object
mapping code into a standalone drm_gem_mmap_obj() function that can be
used to implement dma-buf mmap() operations.

Signed-off-by: Laurent Pinchart <laurent.pinchart+renesas@ideasonboard.com>
Reviewed-by: Rob Clark <robdclark@gmail.com>
11 years agodrm_modes: videomode: add doubleclk flag
Steffen Trumtrar [Mon, 27 May 2013 12:33:35 +0000 (12:33 +0000)]
drm_modes: videomode: add doubleclk flag

Signed-off-by: Steffen Trumtrar <s.trumtrar@pengutronix.de>
Signed-off-by: Dave Airlie <airlied@redhat.com>
11 years agovideo: of: display_timing: add doubleclk flag
Steffen Trumtrar [Mon, 27 May 2013 12:33:34 +0000 (12:33 +0000)]
video: of: display_timing: add doubleclk flag

Signed-off-by: Steffen Trumtrar <s.trumtrar@pengutronix.de>
Signed-off-by: Dave Airlie <airlied@redhat.com>
11 years agovideo: display_timing: add doubleclk flag
Steffen Trumtrar [Mon, 27 May 2013 12:33:33 +0000 (12:33 +0000)]
video: display_timing: add doubleclk flag

Signed-off-by: Steffen Trumtrar <s.trumtrar@pengutronix.de>
Acked-by: Jean-Christophe PLAGNIOL-VILLARD <plagnioj@jcrosoft.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
11 years agovideo: display_timing: make parameter const
Steffen Trumtrar [Mon, 27 May 2013 12:33:05 +0000 (12:33 +0000)]
video: display_timing: make parameter const

As the device_node pointer is not changed in of_get_display_timing and
parse_timing_property it can be a const pointer.

Signed-off-by: Steffen Trumtrar <s.trumtrar@pengutronix.de>
Signed-off-by: Dave Airlie <airlied@redhat.com>
11 years agodrm/i915: add i915_ips_status debugfs entry
Paulo Zanoni [Fri, 31 May 2013 19:33:24 +0000 (16:33 -0300)]
drm/i915: add i915_ips_status debugfs entry

It just prints whether it's supported/enabled/disabled. Feature
requested by the power management team.

v2: Checkpatch started complaining about seq_printf with 1 argument.

Requested-by: Kristen Accardi <kristen.c.accardi@intel.com>
Signed-off-by: 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: add enable_ips module option
Paulo Zanoni [Fri, 31 May 2013 19:33:23 +0000 (16:33 -0300)]
drm/i915: add enable_ips module option

IPS is still enabled by default. Feature requested by the power
management team.

This should also help testing the feature on some early pre-production
hardware where there were relationship problems between IPS and PSR.

v2: Rebase on top of the newest IPS implementation.
v3: Check i915_enable_ips at compute_config, not supports_ips, so the
    kernel parameter will be ignored at haswell_get_pipe_config.

Requested-by: Kristen Accardi <kristen.c.accardi@intel.com>
Signed-off-by: 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: implement IPS feature
Paulo Zanoni [Fri, 31 May 2013 19:33:22 +0000 (16:33 -0300)]
drm/i915: implement IPS feature

Intermediate Pixel Storage is a feature that should reduce the number
of times the display engine wakes up memory to read pixels, so it
should allow deeper PC states. IPS can only be enabled on ULT pipe A
with 8:8:8 pipe pixel formats.

With eDP 1920x1080 and correct watermarks but without FBC this moves
my PC7 residency from 2.5% to around 38%.

v2: - It's tied to pipe A, not port A
    - Add pipe_config support (Chris)
    - Add some assertions (Chris)
    - Rebase against latest dinq
v3: - Don't ever set ips_enabled to false (Daniel)
    - Only check for ips_enabled at hsw_disable_ips (Daniel)
v4: - Add hsw_compute_ips_config (Daniel)
    - Use the new dump_pipe_config (Daniel)

Signed-off-by: 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: fix up the edp power well check
Daniel Vetter [Fri, 31 May 2013 15:49:17 +0000 (17:49 +0200)]
drm/i915: fix up the edp power well check

Now that we track the cpu transcoder we need accurately in the pipe
config we can finally fix up the transcoder check. With the current
code eDP on port D will be broken since we'd errornously cut the
power.

For reference see

commit 2124b72e6283c4e84a55e71077fee91793f4c801
Author: Paulo Zanoni <paulo.r.zanoni@intel.com>
Date:   Fri Mar 22 14:07:23 2013 -0300

    drm/i915: don't disable the power well yet

v2:
- Kill the now outdated comment (Paulo)
- Add the missing crtc->base.enabled check and consolidate it (Paulo)
- Smash all checks together, looks neater that way.

v3: Kill the unused encoder variable.

Cc: Takashi Iwai <tiwai@suse.de>
Cc: Paulo Zanoni <paulo.r.zanoni@intel.com>
Reviewed-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: add I915_PARAM_HAS_VEBOX to i915_getparam
Xiang, Haihao [Wed, 29 May 2013 02:22:34 +0000 (19:22 -0700)]
drm/i915: add I915_PARAM_HAS_VEBOX to i915_getparam

This will let userland only try to use the new ring
when the appropriate kernel is present

Signed-off-by: Xiang, Haihao <haihao.xiang@intel.com>
Reviewed-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: add I915_EXEC_VEBOX to i915_gem_do_execbuffer()
Xiang, Haihao [Wed, 29 May 2013 02:22:33 +0000 (19:22 -0700)]
drm/i915: add I915_EXEC_VEBOX to i915_gem_do_execbuffer()

A user can run batchbuffer via VEBOX ring.

Signed-off-by: Xiang, Haihao <haihao.xiang@intel.com>
Reviewed-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: add VEBOX into debugfs
Xiang, Haihao [Wed, 29 May 2013 16:22:36 +0000 (09:22 -0700)]
drm/i915: add VEBOX into debugfs

v2: Removed rebase relic VECS ring from i915_gem_request_info (Damien)

v3: s/hsw/hws in debugfs which I introduced in v2 (Jon)

Signed-off-by: Xiang, Haihao <haihao.xiang@intel.com>
[Order changed, and modified by]
CC: "Bloomfield, Jon" <jon.bloomfield@intel.com>
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: Enable vebox interrupts
Ben Widawsky [Wed, 29 May 2013 02:22:31 +0000 (19:22 -0700)]
drm/i915: Enable vebox interrupts

Similar to a patch originally written by:

v2: Reversed the meanings of masked and enabled (Haihao)
Made non-destructive writes in case enable/disabler rps runs first
(Haihao)

v3: Reword error message (Damien)
Modify postinstall to do the right thing based on previous fixup. (Ben)

CC: Xiang, Haihao <haihao.xiang@intel.com>
Reviewed-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: vebox interrupt get/put
Ben Widawsky [Wed, 29 May 2013 02:22:30 +0000 (19:22 -0700)]
drm/i915: vebox interrupt get/put

v2: Use the correct lock to protect PM interrupt regs, this was
accidentally lost from earlier (Haihao)
Fix return types (Ben)

Reviewed-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: consolidate interrupt naming scheme
Ben Widawsky [Wed, 29 May 2013 02:22:29 +0000 (19:22 -0700)]
drm/i915: consolidate interrupt naming scheme

The motivation here is we're going to add some new interrupt definitions
and handling outside of the GT interrupts which is all we've managed so
far (with some RPS exceptions). By consolidating the names in the future
we can make thing a bit cleaner as we don't need to define register
names twice, and we can leverage pretty decent overlap in HW registers
since ILK.

To explain briefly what is in the comments: there are two sets of
interrupt masking/enabling registers. At least so far, the definitions
of the two sets overlap. The old code setup distinct names for
interrupts in each set, ie. one for global, and one for ring. This made
things confusing when using the wrong defines in the wrong places.

rebase: Modified VLV bits

v2: Renamed GT_RENDER_MASTER to GT_RENDER_CS_MASTER (Damien)

Reviewed-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: Convert irq_refounct to struct
Ben Widawsky [Wed, 29 May 2013 02:22:28 +0000 (19:22 -0700)]
drm/i915: Convert irq_refounct to struct

It's overkill on older gens, but it's useful for newer gens.

Reviewed-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: make PM interrupt writes non-destructive
Ben Widawsky [Wed, 29 May 2013 02:22:27 +0000 (19:22 -0700)]
drm/i915: make PM interrupt writes non-destructive

PM interrupts have an expanded role on HSW. It helps route the EBOX
interrupts. This patch is necessary to make the existing code which
touches the mask, and enable registers more friendly to other code paths
that also will need these registers.

To be more explicit:
At preinstall all interrupts are masked and disabled. This implies that
preinstall should always happen before any enabling/disabling of RPS or
other interrupts.

The PMIMR is touched by the workqueue, so enable/disable touch IER and
IIR. Similarly, the code currently expects IMR has no use outside of the
RPS related interrupts so they unconditionally set 0, or ~0. We could
use IER in the workqueue, and IMR elsewhere, but since the workqueue
use-case is more transient the existing usage makes sense.

Disable RPS events:
IER := IER & ~GEN6_PM_RPS_EVENTS // Disable RPS related interrupts
IIR := GEN6_PM_RPS_EVENTS // Disable any outstanding interrupts

Enable RPS events:
IER := IER | GEN6_PM_RPS_EVENTS // Enable the RPS related interrupts
IIR := GEN6_PM_RPS_EVENTS // Make sure there were no leftover events
(really shouldn't happen)

v2: Shouldn't destroy PMIIR or PMIMR VEBOX interrupt state in
enable/disable rps functions (Haihao)

v3: Bug found by Chris where we were clearing the wrong bits at rps
disable.
    expanded commit message

v4: v3 was based off the wrong branch

v5: Added the setting of PMIMR because of previous patch update

CC: Chris Wilson <chris@chris-wilson.co.uk>
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: Add PM regs to pre/post install
Ben Widawsky [Wed, 29 May 2013 02:22:26 +0000 (19:22 -0700)]
drm/i915: Add PM regs to pre/post install

At the moment, these values are wiped out anyway by the rps
enable/disable. That will be changed in the next patch though.

v2: Add post install setup to address issue found by Damien in the next
patch.
replaced
WARN_ON(dev_priv->rps.pm_iir != 0);
with rps.pm_iir = 0;

With the v2 of this patch and the deferred pm enabling (which changed
since the original patches) we're now able to get PM interrupts before
we've brought up enabled rps. At this point in boot, we don't want to do
anything about it, so we simply ignore it. Since writing the original
assertion, the code has changed quite a bit, and I believe removing this
assertion is perfectly safe.

Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Reviewed-by: Damien Lespiau <damien.lespiau@intel.com>
[danvet: I don't agree with the justification to drop the WARN and
added a FIXME to that effect.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: Create an ivybridge_irq_preinstall
Ben Widawsky [Wed, 29 May 2013 02:22:25 +0000 (19:22 -0700)]
drm/i915: Create an ivybridge_irq_preinstall

Just duplicates ironlake_irq_preinstall for now.

v2: Add new PCH_NOP check (Damien)
Add SDEIMR comment (Damien)

Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Reviewed-by: Damien Lespiau <damien.lespiau@intel.com>
[danvet: Update now outdated comment.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: Create a more generic pm handler for hsw+
Ben Widawsky [Wed, 29 May 2013 02:22:24 +0000 (19:22 -0700)]
drm/i915: Create a more generic pm handler for hsw+

HSW has some special requirements for the VEBOX. Splitting out the
interrupt handler will make the code a bit nicer and less error prone
when we begin to handle those.

The slight functional change in this patch (queueing work while holding
the spinlock) is intentional as it makes a subsequent patch a bit nicer.
The change should also only effect HSW platforms.

Reviewed-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: add support for 5/6 data buffer partitioning on Haswell
Paulo Zanoni [Fri, 31 May 2013 13:19:21 +0000 (10:19 -0300)]
drm/i915: add support for 5/6 data buffer partitioning on Haswell

Now we compute the results for both 1/2 and 5/6 partitioning and then
use hsw_find_best_result to choose which one to use.

With this patch, Haswell watermarks support should be in good shape.
The only improvement we're missing is the case where the primary plane
is disabled: we always assume it's enabled, so we take it into
consideration when calculating the watermarks.

v2: - Check the latency when finding the best result

Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: properly set HSW WM_LP watermarks
Paulo Zanoni [Fri, 31 May 2013 14:45:06 +0000 (11:45 -0300)]
drm/i915: properly set HSW WM_LP watermarks

We were previously only setting the WM_PIPE registers, now we are
setting the LP watermark registers. This should allow deeper PC
states, resulting in power savings.

We're only using 1/2 data buffer partitioning for now.

v2: Merge both hsw_compute_pri_wm_* functions (Ville)
v3: - Simplify hsw_compute_wm_results (Ville)
    - Rebase due to changes on the previous patch
v4: Unconfuse wm_lp/level (Ville)

Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: properly set HSW WM_PIPE registers
Paulo Zanoni [Fri, 31 May 2013 13:08:35 +0000 (10:08 -0300)]
drm/i915: properly set HSW WM_PIPE registers

We were previously calling sandybridge_update_wm on HSW, but the SNB
function didn't really match the HSW specification, so we were just
writing the wrong values.

With this patch, the haswell_update_wm function will set the correct
values for the WM_PIPE registers, but it will still keep all the LP
watermarks disabled.

The patch may look a little bit over-complicated for now, but it's
because much of the infrastructure for setting the LP watermarks is
already in place, so we won't have too much code churn on the patch
that sets the LP watermarks.

v2: - Fix pixel_rate on panel fitter case (Ville)
    - Try to not overflow (Ville)
    - Remove useless variable (Ville)
    - Fix p->pri_horiz_pixels (Paulo)
v3: - Fix rounding errors on hsw_wm_method2 (Ville)
v4: - Fix memcmp bug (Paulo)

Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: fix pch_nop support
Daniel Vetter [Wed, 29 May 2013 19:43:05 +0000 (21:43 +0200)]
drm/i915: fix pch_nop support

This was accidentally broken in the south error interrupt handling
work:

commit 8664281b64c457705db72fc60143d03827e75ca9
Author: Paulo Zanoni <paulo.r.zanoni@intel.com>
Date:   Fri Apr 12 17:57:57 2013 -0300

    drm/i915: report Gen5+ CPU and PCH FIFO underruns

Cc: Paulo Zanoni <paulo.r.zanoni@intel.com>
Cc: Ben Widawsky <ben@bwidawsk.net>
Reviewed-by: Ben Widawsky <ben@bwidawsk.net>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: Vebox ringbuffer init
Ben Widawsky [Wed, 29 May 2013 02:22:23 +0000 (19:22 -0700)]
drm/i915: Vebox ringbuffer init

v2: Add set_seqno which didn't exist before rebase (Haihao)

Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Reviewed-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Xiang, Haihao <haihao.xiang@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: add HAS_VEBOX
Xiang, Haihao [Wed, 29 May 2013 02:22:22 +0000 (19:22 -0700)]
drm/i915: add HAS_VEBOX

The flag will be useful to help share code between IVB, and HSW as the
programming is similar in many places with this as one of the major
differences.

Signed-off-by: Xiang, Haihao <haihao.xiang@intel.com>
[Commit message + small fix by]
Reviewed-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: Rename ring flush functions
Ben Widawsky [Wed, 29 May 2013 02:22:21 +0000 (19:22 -0700)]
drm/i915: Rename ring flush functions

Historically we considered the render ring to have special flush
semantics and everything else to fall under a more general umbrella.
Probably by coincidence more than anything we decided to make the bsd
ring have the default *other* flush. As the new vebox ring exposes, the
bsd ring is actually the weird one. Doing this allows us to call
gen6_ring_flush for the vebox because calling blt_ring_flush would be
weird...

This patch should have no functional change.

Reviewed-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: Add VECS semaphore bits
Ben Widawsky [Wed, 29 May 2013 02:22:20 +0000 (19:22 -0700)]
drm/i915: Add VECS semaphore bits

Like the other rings, the VECS supports semaphores. The semaphore stuff
is a bit wonky so this patch on it's own should be nice for review.

This patch should have no functional impact.

v2: Fix the English parts of clarification (again, register names were
right, text was reversed) (Damien)
Restore the still valid invariant. (Damien)
The bsd semaphore register should be MI_SEMAPHORE_SYNC_VVE (Damien)

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: Introduce VECS: the 4th ring
Ben Widawsky [Wed, 29 May 2013 02:22:19 +0000 (19:22 -0700)]
drm/i915: Introduce VECS: the 4th ring

The video enhancement command streamer is a new ring on HSW which does
what it sounds like it does. This patch provides the most minimal
inception of the ring.

In order to support a new ring, we need to bump the number. The patch
may look trivial to the untrained eye, but bumping the number of rings
is a bit scary. As such the patch is not terribly useful by itself, but
a pretty nice place to find issues during a bisection.

Reviewed-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: Semaphore MBOX update generalization
Ben Widawsky [Wed, 29 May 2013 02:22:18 +0000 (19:22 -0700)]
drm/i915: Semaphore MBOX update generalization

This replaces the existing MBOX update code with a more generalized
calculation for emitting mbox updates. We also create a sentinel for
doing the updates so we can more abstractly deal with the rings.

When doing MBOX updates the code must be aware of the /other/ rings.
Until now the platforms which supported semaphores had a fixed number of
rings and so it made sense for the code to be very specialized
(hardcoded).

The patch does contain a functional change, but should have no
behavioral changes.

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: Comments for semaphore clarification
Ben Widawsky [Wed, 29 May 2013 02:22:17 +0000 (19:22 -0700)]
drm/i915: Comments for semaphore clarification

Semaphores are tied very closely to the rings in the GPU. Trivial patch
adds comments to the existing code so that when we add new rings we can
include comments there as well. It also helps distinguish the ring to
semaphore mailbox interactions by using the ringname in the semaphore
data structures.

This patch should have no functional impact.

v2: The English parts (as opposed to register names) of the comments
were reversed. (Damien)

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: release cursor when crtc is destroyed
Mika Kuoppala [Tue, 23 Apr 2013 14:27:08 +0000 (17:27 +0300)]
drm/i915: release cursor when crtc is destroyed

crtc is holding a reference to a cursor bo and it needs
to be released when crtc is destroyed so that we don't leak
the cursor bo.

v2: Enhance set and move cursor so that disabled
cursor is handled correctly (Ville Syrjälä)

Signed-off-by: Mika Kuoppala <mika.kuoppala@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: Avoid promoting a simulated hang to 'wedged'
Chris Wilson [Tue, 28 May 2013 09:38:44 +0000 (10:38 +0100)]
drm/i915: Avoid promoting a simulated hang to 'wedged'

It appears that a beneficial side-effect of Mika's more accurate hangman
work is to speed up hang detection and execution. This exposes a bug in
the reset code that then treats repeated simulated hangs as an
indication that the machine is wedged. Jiggle the code around so that we
only do the simulation processing from the hangcheck and avoid confusing
it with a real hang.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=65060
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: drop a few really redundant WARNs in hsw mode_set
Daniel Vetter [Tue, 28 May 2013 14:28:55 +0000 (16:28 +0200)]
drm/i915: drop a few really redundant WARNs in hsw mode_set

- Correct cpu->pch display matching is already check when we detect
  the PCH type at driver load.
- Plane/pipe state is already checked both when a) enabling, b)
  disabling and in c) the modeset state checker. No need to go
  overboard and also check it in in between a) and b).

Cc: Paulo Zanoni <przanoni@gmail.com>
Reviewed-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: add basic pipe config dump support
Daniel Vetter [Tue, 28 May 2013 10:05:54 +0000 (12:05 +0200)]
drm/i915: add basic pipe config dump support

All this pipe config abstraction adds another layer of complexity, so
it's good to have better visibility into what's going on exactly.
Doesn't dump out everything yet, and some bits are a bit duplicated
but this should be a good start.

Note that at boot-up a lot of the fields are 0 even for enabled pipes,
this is simply because our hw state readout code doesn't support
everything.

v2: Remove a few more now redudant debug output lines.

v3: Review from Paulo
- use transcoder_name
- fix up format specifiers
- add missing ':' in debug output

Cc: Paulo Zanoni <przanoni@gmail.com>
Reviewed-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: fix error return code in init_pipe_control()
Wei Yongjun [Tue, 28 May 2013 09:51:44 +0000 (17:51 +0800)]
drm/i915: fix error return code in init_pipe_control()

Fix to return -ENOMEM in the kmap() error handling case
instead of 0, as done elsewhere in this function.

Signed-off-by: Wei Yongjun <yongjun_wei@trendmicro.com.cn>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: document why dvo/sdvo/crt need a special dpms function
Jani Nikula [Tue, 28 May 2013 09:35:02 +0000 (12:35 +0300)]
drm/i915: document why dvo/sdvo/crt need a special dpms function

In the cloned case, changing just one output but keeping the other, the
pipe state won't change and intel_crtc_update_dpms will be a nop, but we
still need to update the dpms state of the output being changed.

Only dvo, sdvo and crt are cloneable, so only those three have special
dpms functions.

Signed-off-by: Jani Nikula <jani.nikula@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: hw state readout&check support for cpu_transcoder
Daniel Vetter [Tue, 21 May 2013 22:50:22 +0000 (00:50 +0200)]
drm/i915: hw state readout&check support for cpu_transcoder

This allows us to drop a bunch of ugly hacks and finally implement
what

commit cc464b2a17c59adedbdc02cc54341d630354edc3
Author: Paulo Zanoni <paulo.r.zanoni@intel.com>
Date:   Fri Jan 25 16:59:16 2013 -0200

    drm/i915: set TRANSCODER_EDP even earlier

tried to achieve, but that was reverted again in

commit bba2181c49f1dddf8b592804a1b53cc1a3cf408a
Author: Daniel Vetter <daniel.vetter@ffwll.ch>
Date:   Fri Mar 22 10:53:40 2013 +0100

    Revert "drm/i915: set TRANSCODER_EDP even earlier"

Now we should always have a consistent cpu_transcoder in the
pipe_config.

v2: Fix up the code as spotted by Paulo:
- read the register for real
- assign the right pipes
- break out if the hw state doesn't make sense

v3: Shut up gcc.

Cc: Paulo Zanoni <paulo.r.zanoni@intel.com>
Reviewed-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: fixup i915_pipe_enabled check in i915_irq.c
Daniel Vetter [Tue, 21 May 2013 22:50:23 +0000 (00:50 +0200)]
drm/i915: fixup i915_pipe_enabled check in i915_irq.c

Well, as well as we can without completely revamping the drm vblank
code. The issue are that
- The vblank code needs to work on both ums and kms.
- It deals always deals with pipes.
- It doesn't take any of the kms locks.

The last part is not really fixable without revamping the drm vblank
code, since the drm core <-> driver interactions is a veritable pile
of spaghettis. But the other pieces can be fixed by switching on the
MODESET driver flag and either checking the hw state directly (ums
case) or just querying our sw tracking (with broken locking, but
that's not worse than what we've had).

Note that this essentially reverts

commit 702e7a56af3780d8b3a717f698209bef44187bb0
Author: Paulo Zanoni <paulo.r.zanoni@intel.com>
Date:   Tue Oct 23 18:29:59 2012 -0200

    drm/i915: convert PIPECONF to use transcoder instead of pipe

for the ums case, which will fix a NULL deref (since we really don't
have any crtcs set up).

But the real reason to do this is to drop our reliance on the
cpu_transcoder: By only checking intel_crtc->active we don't need to
make sure that the pipe_config (or at least the cpu_transcoder)
contain safe values even when the pipe is off.

Cc: Paulo Zanoni <paulo.r.zanoni@intel.com>
Cc: 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: remove unused is_cpu_edp()
Imre Deak [Thu, 16 May 2013 11:40:37 +0000 (14:40 +0300)]
drm/i915: remove unused is_cpu_edp()

Signed-off-by: Imre Deak <imre.deak@intel.com>
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@gmail.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: replace is_cpu_edp() with a check for port A
Imre Deak [Thu, 16 May 2013 11:40:36 +0000 (14:40 +0300)]
drm/i915: replace is_cpu_edp() with a check for port A

The patch changes all remaining is_cpu_edp() check with a check for port
A. We can do this, since in all these cases ValleyView is handled
separately and port A is always a CPU side eDP port.

Signed-off-by: Imre Deak <imre.deak@intel.com>
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@gmail.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: merge VLV eDP and DP AUX clock divider calculation
Imre Deak [Thu, 16 May 2013 11:40:35 +0000 (14:40 +0300)]
drm/i915: merge VLV eDP and DP AUX clock divider calculation

On ValleyView for both eDP and DP the AUX input clock is 200MHz, so we
can calculate for both the clock divider for the 2MHz target rate at the
same place. Afterwards we can also replace the is_cpu_edp() check with a
check for port A.

Signed-off-by: Imre Deak <imre.deak@intel.com>
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@gmail.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: stop using is_cpu_edp() in intel_disable/post_disable_dp
Imre Deak [Thu, 23 May 2013 16:39:40 +0000 (19:39 +0300)]
drm/i915: stop using is_cpu_edp() in intel_disable/post_disable_dp

Based on 3739850b46f - "drm/i915: disable the cpu edp port after the
cpu pipe" and the bspec disabling sequence for IVB and older it seems we
have to distinguish only the CPU vs. PCH port case, whether it's a DP or
eDP doesn't seem to matter. For IVB and older on the CPU side we can
only have eDP on port A, DP ports can only be on the PCH side. On VLV we
have only CPU side eDP/DP ports, no PCH. So the condition for the
disabling sequence we need for CPU ports is port == A || IS_VLV.

This allows us to remove is_cpu_edp() completely in a later patch.

v2:
- simplify (and fix) the condition for CPU side ports and adjust the
  commit message accordingly (Daniel)

Signed-off-by: Imre Deak <imre.deak@intel.com>
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@gmail.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: release scratch page at module unload
Imre Deak [Wed, 22 May 2013 14:47:13 +0000 (17:47 +0300)]
drm/i915: release scratch page at module unload

Signed-off-by: Imre Deak <imre.deak@intel.com>
Acked-by: Mika Kuoppala <mika.kuoppala@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: Call context fini at cleanup
Ben Widawsky [Sat, 25 May 2013 19:26:39 +0000 (12:26 -0700)]
drm/i915: Call context fini at cleanup

If contexts were actually initialized, and we fail somewhere later during
init this would possibly leak memory, and lead to some error messages
about unclean takedown. As the odds of this occurring, and someone
actually caring/noticing are pretty slim, the patch isn't terribly
important.

Found by code inspection while working on something else.

Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: context debug messages
Ben Widawsky [Sat, 25 May 2013 19:26:38 +0000 (12:26 -0700)]
drm/i915: context debug messages

Add some debug messages to help figure out what goes wrong on context
initialization.

Later in the PPGTT series, I ended up having a lot of failures after
reset. In many cases it was extra difficult to debug because I hadn't
even realized that contexts failed to reinitialize after reset (again an
artifact of some later patches).

This fairly benign patch does help debug some potential issues which
arise later.

Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: use drm_mm_takedown
Ben Widawsky [Sat, 25 May 2013 19:26:37 +0000 (12:26 -0700)]
drm/i915: use drm_mm_takedown

I noticed this while doing the VMA abstraction. AFAICT, it won't
actually fix anything, but it is the correct thing to do.

Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: use mappable size for fb kickout
Ben Widawsky [Sat, 25 May 2013 19:26:36 +0000 (12:26 -0700)]
drm/i915: use mappable size for fb kickout

The GTT start is either 0 in the KMS case, or some value which is set
only after the init IOCTL in the UMS case. In both cases, we don't have
this information until after we've tried to kick out the firmware fb.

This patch should have no functional change since we kzalloc the GTT
struct anyway. It only clarifies the situation for people who end up
having to look at that code.

This weirdness was introduced in:

commit 93d187993b783c68383a884091a600d9ad499ea6
Author: Ben Widawsky <ben@bwidawsk.net>
Date:   Thu Jan 17 12:45:17 2013 -0800

    drm/i915: Remove use of gtt_mappable_entries

Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: pre-fixes for checkpatch
Ben Widawsky [Sat, 25 May 2013 19:26:35 +0000 (12:26 -0700)]
drm/i915: pre-fixes for checkpatch

Since I'll need to modify i915_gem_object_bind_to_gtt(), fix the errors
now to get checkpatch to not complain.

Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
[danvet: Resolve conflict with Chris' improved debug output, and
bikeshed the new variable with s/max/gtt_max/ a bit while at it.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: Fix error state memory leaks
Ben Widawsky [Sat, 25 May 2013 21:42:54 +0000 (14:42 -0700)]
drm/i915: Fix error state memory leaks

Found with kmemleak.

Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: introduce i915_hangcheck_ring_hung
Mika Kuoppala [Mon, 13 May 2013 13:32:11 +0000 (16:32 +0300)]
drm/i915: introduce i915_hangcheck_ring_hung

In preparation to track per ring progress in hangcheck,
add i915_hangcheck_ring_hung.

v2: omit dev parameter (Ben Widawsky)

Signed-off-by: Mika Kuoppala <mika.kuoppala@intel.com>
Reviewed-by: Ben Widawsky <ben@bwidawsk.net>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: track ring progression using seqnos
Mika Kuoppala [Fri, 24 May 2013 14:16:07 +0000 (17:16 +0300)]
drm/i915: track ring progression using seqnos

Instead of relying in acthd, track ring seqno progression
to detect if ring has hung.

v2: put hangcheck stuff inside struct (Chris Wilson)

v3: initialize hangcheck.seqno (Ben Widawsky)

Signed-off-by: Mika Kuoppala <mika.kuoppala@intel.com>
Reviewed-by: Ben Widawsky <ben@bwidawsk.net>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: pass seqno to i915_hangcheck_ring_idle
Mika Kuoppala [Mon, 13 May 2013 13:32:09 +0000 (16:32 +0300)]
drm/i915: pass seqno to i915_hangcheck_ring_idle

In preparation for next commit, pass seqno as a parameter
to i915_hangcheck_ring_idle as it will be used inside
i915_hangcheck_elapsed.

Signed-off-by: Mika Kuoppala <mika.kuoppala@intel.com>
Reviewed-by: Ben Widawsky <ben@bwidawsk.net>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: add haswell_update_sprite_wm
Paulo Zanoni [Fri, 24 May 2013 14:59:18 +0000 (11:59 -0300)]
drm/i915: add haswell_update_sprite_wm

On Haswell, whenever we change the sprites we need to completely
recalculate all the watermarks, because the sprites are one of the
parameters to the LP watermarks, so a change on the sprites may
trigger a change on which LP levels are enabled.

So on this commit we store all the parameters we need to store for
proper recalculation of the Haswell WMs and then call
haswell_update_wm.

Notice that for now our haswell_update_wm function is not really using
these parameters we're storing, but on the next commits we'll use
these parameters.

Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm/i915: add "enable" argument to intel_update_sprite_watermarks
Paulo Zanoni [Fri, 24 May 2013 14:59:17 +0000 (11:59 -0300)]
drm/i915: add "enable" argument to intel_update_sprite_watermarks

Because we want to call it from the "sprite disable" paths, since on
Haswell we need to update the sprite watermarks when we disable
sprites.

For now, all this patch does is to add the "enable" argument and call
intel_update_sprite_watermarks from inside ivb_disable_plane. This
shouldn't change how the code behaves because on
sandybridge_update_sprite_wm we just ignore the "!enable" case. The
patches that implement Haswell watermarks will make use of the changes
introduced by this patch.

Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agodrm: Don't leak phys_wc "handles" to userspace
Andy Lutomirski [Mon, 13 May 2013 23:58:48 +0000 (23:58 +0000)]
drm: Don't leak phys_wc "handles" to userspace

I didn't fix this in the earlier patch -- it would have broken the
build due to the now-deleted garbage in drm_os_linux.h.

Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Andy Lutomirski <luto@amacapital.net>
Signed-off-by: Dave Airlie <airlied@redhat.com>
11 years agodrm: Remove mtrr_add and mtrr_del fallback hack for non-MTRR systems
Andy Lutomirski [Mon, 13 May 2013 23:58:47 +0000 (23:58 +0000)]
drm: Remove mtrr_add and mtrr_del fallback hack for non-MTRR systems

There are no users left in drivers/gpu.

Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Andy Lutomirski <luto@amacapital.net>
Signed-off-by: Dave Airlie <airlied@redhat.com>
11 years agouvesafb: Clean up MTRR code
Andy Lutomirski [Mon, 13 May 2013 23:58:46 +0000 (23:58 +0000)]
uvesafb: Clean up MTRR code

The old code allowed very strange memory types.  Now it works like
all the other video drivers: ioremap_wc is used unconditionally,
and MTRRs are set if PAT is unavailable (unless MTRR is disabled
by a module parameter).

UC, WB, and WT support is gone.  If there are MTRR conflicts that prevent
addition of a WC MTRR, adding a non-conflicting MTRR is pointless; it's
better to just turn off MTRR support entirely.

As an added bonus, any MTRR added is freed on unload.

Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Andy Lutomirski <luto@amacapital.net>
Signed-off-by: Dave Airlie <airlied@redhat.com>
11 years agoradeon: Switch to arch_phys_wc_add and add a missing ..._del
Andy Lutomirski [Mon, 13 May 2013 23:58:45 +0000 (23:58 +0000)]
radeon: Switch to arch_phys_wc_add and add a missing ..._del

Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Andy Lutomirski <luto@amacapital.net>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
11 years agoi915: Use arch_phys_wc_{add,del}
Andy Lutomirski [Mon, 13 May 2013 23:58:44 +0000 (23:58 +0000)]
i915: Use arch_phys_wc_{add,del}

i915 open-coded logic that was essentially equivalent to the new API.

Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Andy Lutomirski <luto@amacapital.net>
Signed-off-by: Dave Airlie <airlied@redhat.com>
11 years agodrm, agpgart: Use pgprot_writecombine for AGP maps and make the MTRR optional
Andy Lutomirski [Mon, 13 May 2013 23:58:43 +0000 (23:58 +0000)]
drm, agpgart: Use pgprot_writecombine for AGP maps and make the MTRR optional

I'm not sure I understand the intent of the previous behavior.  mmap
on /dev/agpgart and DRM_AGP maps had no cache flags set, so they
would be fully cacheable.  But the DRM code (most of the time) would
add a write-combining MTRR that would change the effective memory
type to WC.

The new behavior just requests WC explicitly for all AGP maps.

If there is any code out there that expects cacheable access to the
AGP aperture (because the drm driver doesn't request an MTRR or
because it's using /dev/agpgart directly), then it will now end up
with a UC or WC mapping, depending on the architecture and PAT
availability.  But cacheable access to the aperture seems like it's
asking for trouble, because, AIUI, the aperture is an alias of RAM.

Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Andy Lutomirski <luto@amacapital.net>
Signed-off-by: Dave Airlie <airlied@redhat.com>
11 years agodrm: Update drm_addmap and drm_mmap to use PAT WC instead of MTRRs
Andy Lutomirski [Mon, 13 May 2013 23:58:42 +0000 (23:58 +0000)]
drm: Update drm_addmap and drm_mmap to use PAT WC instead of MTRRs

Previously, DRM_FRAME_BUFFER mappings, as well as DRM_REGISTERS
mappings with DRM_WRITE_COMBINING set, resulted in an unconditional
MTRR being added but the actual mappings being created as UC-.

Now these mappings have the MTRR added only if needed, but they will
be mapped with pgprot_writecombine.

The non-WC DRM_REGISTERS case now uses pgprot_noncached instead of
hardcoding the bit twiddling.

The DRM_AGP case is unchanged for now.

[airlied: fix ppc build]
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Andy Lutomirski <luto@amacapital.net>
Signed-off-by: Dave Airlie <airlied@redhat.com>
11 years agodrm (ast, cirrus, mgag200, nouveau, savage, vmwgfx): Remove drm_mtrr_{add, del}
Andy Lutomirski [Mon, 13 May 2013 23:58:41 +0000 (23:58 +0000)]
drm (ast, cirrus, mgag200, nouveau, savage, vmwgfx): Remove drm_mtrr_{add, del}

This replaces drm_mtrr_{add,del} with arch_phys_wc_{add,del}.  The
interface is simplified (because the base and size parameters to
drm_mtrr_del never did anything), and it no longer adds MTRRs on
systems that don't need them.

Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Andy Lutomirski <luto@amacapital.net>
Signed-off-by: Dave Airlie <airlied@redhat.com>
11 years agoAdd arch_phys_wc_{add, del} to manipulate WC MTRRs if needed
Andy Lutomirski [Mon, 13 May 2013 23:58:40 +0000 (23:58 +0000)]
Add arch_phys_wc_{add, del} to manipulate WC MTRRs if needed

Several drivers currently use mtrr_add through various #ifdef guards
and/or drm wrappers.  The vast majority of them want to add WC MTRRs
on x86 systems and don't actually need the MTRR if PAT (i.e.
ioremap_wc, etc) are working.

arch_phys_wc_add and arch_phys_wc_del are new functions, available
on all architectures and configurations, that add WC MTRRs on x86 if
needed (and handle errors) and do nothing at all otherwise.  They're
also easier to use than mtrr_add and mtrr_del, so the call sites can
be simplified.

As an added benefit, this will avoid wasting MTRRs and possibly
warning pointlessly on PAT-supporting systems.

Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Andy Lutomirski <luto@amacapital.net>
Signed-off-by: Dave Airlie <airlied@redhat.com>