GitHub/MotorolaMobilityLLC/kernel-slsi.git
9 years agoMerge branch 'amdkfd-next-3.19' of git://people.freedesktop.org/~gabbayo/linux into...
Dave Airlie [Fri, 5 Dec 2014 01:10:59 +0000 (11:10 +1000)]
Merge branch 'amdkfd-next-3.19' of git://people.freedesktop.org/~gabbayo/linux into drm-next

Just three more fixes for 3.19.
This is the last request until -rc1. I will have our QA team run a full
HSA stack test on 3.19-rc1 and if we find problems, I will send a fixes pull
request.

* 'amdkfd-next-3.19' of git://people.freedesktop.org/~gabbayo/linux:
  amdkfd: Set *buffer_ptr to NULL in case of error
  amdkfd: use atomic allocations within srcu callbacks
  amdkfd: use sizeof(long) granularity for the pasid bitmask

9 years agoMerge branch 'drm-next-3.19-wip' of git://people.freedesktop.org/~agd5f/linux into...
Dave Airlie [Fri, 5 Dec 2014 01:08:31 +0000 (11:08 +1000)]
Merge branch 'drm-next-3.19-wip' of git://people.freedesktop.org/~agd5f/linux into drm-next

- More cursor and hotspot handling fixes
- Fix some typos in the new smc fan control code and enable on CI
- VM and CS cleanups

* 'drm-next-3.19-wip' of git://people.freedesktop.org/~agd5f/linux:
  drm/radeon: enable smc fan control on CI
  drm/radeon: use pointers instead of indexes for CS chunks
  drm/radeon: remove duplicates check
  drm/ttm: optionally move duplicates to a separate list
  drm/radeon: check the right ring in radeon_evict_flags()
  drm/radeon: fix copy paste typos in fan control for si/ci
  drm/radeon: Hide cursor on CRTCs used by fbdev (v2)
  drm/radeon: add spinlock for BO_VA status protection (v2)
  drm/radeon: fence PT updates as shared
  drm/radeon: rename radeon_cs_reloc to radeon_bo_list
  drm/radeon: drop the handle from radeon_cs_reloc
  drm/radeon drop gobj from radeon_cs_reloc
  drm/radeon: fix typo in new fan control registers for SI/CI
  drm/radeon: sync all BOs involved in a CS
  drm/radeon: Move hotspot handling out of radeon_set_cursor
  drm/radeon: Re-show the cursor after a modeset

9 years agoamdkfd: Set *buffer_ptr to NULL in case of error
Oded Gabbay [Thu, 4 Dec 2014 12:09:02 +0000 (14:09 +0200)]
amdkfd: Set *buffer_ptr to NULL in case of error

In function acquire_packet_buffer() we may return -ENOMEM. In that case, we
should set the *buffer_ptr to NULL, so that calling functions which check the
*buffer_ptr value as a criteria for success, will know that
acquire_packet_buffer() failed.

Reviewed-by: Alexey Skidanov <alexey.skidanov@amd.com>
Signed-off-by: Oded Gabbay <oded.gabbay@amd.com>
9 years agodrm/radeon: enable smc fan control on CI
Alex Deucher [Thu, 4 Dec 2014 00:05:19 +0000 (19:05 -0500)]
drm/radeon: enable smc fan control on CI

This seems to work well on CI boards after fixing the
last few bugs noticed by Chernovsky Oleg.

On boards with a high default fan speed this should
reduce fan noise.  Manual fan control is not enabled
yet.

Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
9 years agodrm/radeon: use pointers instead of indexes for CS chunks
Christian König [Wed, 3 Dec 2014 14:53:24 +0000 (15:53 +0100)]
drm/radeon: use pointers instead of indexes for CS chunks

Nobody is interested at which index the chunk is. What's needed is
a pointer to the chunk. Remove unused chunk_id field as well.

Signed-off-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
9 years agodrm/radeon: remove duplicates check
Christian König [Wed, 3 Dec 2014 14:46:49 +0000 (15:46 +0100)]
drm/radeon: remove duplicates check

Completely unnecessary since the ww_mutex used to reserve a buffer
can detect double reservations from the same thread anyway.

Signed-off-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
9 years agodrm/ttm: optionally move duplicates to a separate list
Christian König [Wed, 3 Dec 2014 14:46:48 +0000 (15:46 +0100)]
drm/ttm: optionally move duplicates to a separate list

This patch adds an optional list_head parameter to ttm_eu_reserve_buffers.
If specified duplicates in the execbuf list are no longer reported as errors,
but moved to this list instead.

Reviewed-by: Thomas Hellstrom <thellstrom@vmware.com>
Signed-off-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
9 years agodrm/radeon: check the right ring in radeon_evict_flags()
Alex Deucher [Wed, 3 Dec 2014 05:03:49 +0000 (00:03 -0500)]
drm/radeon: check the right ring in radeon_evict_flags()

Check the that ring we are using for copies is functional
rather than the GFX ring.  On newer asics we use the DMA
ring for bo moves.

Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Cc: stable@vger.kernel.org
9 years agodrm/radeon: fix copy paste typos in fan control for si/ci
Alex Deucher [Mon, 1 Dec 2014 22:18:53 +0000 (17:18 -0500)]
drm/radeon: fix copy paste typos in fan control for si/ci

Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
9 years agodrm/radeon: Hide cursor on CRTCs used by fbdev (v2)
Michel Dänzer [Fri, 28 Nov 2014 02:48:48 +0000 (11:48 +0900)]
drm/radeon: Hide cursor on CRTCs used by fbdev (v2)

Since we are now preserving the cursor across modesets, the cursor could
be left over in console if e.g. X crashed.

v2: add comment about universal plane support

Signed-off-by: Michel Dänzer <michel.daenzer@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
9 years agodrm/radeon: add spinlock for BO_VA status protection (v2)
Christian König [Thu, 27 Nov 2014 13:48:44 +0000 (14:48 +0100)]
drm/radeon: add spinlock for BO_VA status protection (v2)

Stop using the VM mutex for this

v2: fix typo in comment

Signed-off-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
9 years agodrm/radeon: fence PT updates as shared
Christian König [Thu, 27 Nov 2014 13:48:43 +0000 (14:48 +0100)]
drm/radeon: fence PT updates as shared

The BO_VA contains everything necessary.

Signed-off-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
9 years agodrm/radeon: rename radeon_cs_reloc to radeon_bo_list
Christian König [Thu, 27 Nov 2014 13:48:42 +0000 (14:48 +0100)]
drm/radeon: rename radeon_cs_reloc to radeon_bo_list

Better match what it is actually doing.

Signed-off-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
9 years agodrm/radeon: drop the handle from radeon_cs_reloc
Christian König [Thu, 27 Nov 2014 13:48:41 +0000 (14:48 +0100)]
drm/radeon: drop the handle from radeon_cs_reloc

It's only used for duplicate check and that
can be done on the original as well.

Signed-off-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
9 years agodrm/radeon drop gobj from radeon_cs_reloc
Christian König [Thu, 27 Nov 2014 13:48:40 +0000 (14:48 +0100)]
drm/radeon drop gobj from radeon_cs_reloc

It's only used once after initializing and that
ptr can be calculated from the BO as well.

Signed-off-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
9 years agodrm/radeon: fix typo in new fan control registers for SI/CI
Alex Deucher [Mon, 1 Dec 2014 03:03:57 +0000 (22:03 -0500)]
drm/radeon: fix typo in new fan control registers for SI/CI

Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
9 years agodrm/radeon: sync all BOs involved in a CS
Christian König [Wed, 26 Nov 2014 15:29:33 +0000 (16:29 +0100)]
drm/radeon: sync all BOs involved in a CS

Not just the userspace relocs, otherwise we won't wait
for a swapped out page tables to be swapped in again.

Signed-off-by: Christian König <christian.koenig@amd.com>
Cc: stable@vger.kernel.org
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
9 years agodrm/radeon: Move hotspot handling out of radeon_set_cursor
Michel Dänzer [Fri, 21 Nov 2014 02:48:58 +0000 (11:48 +0900)]
drm/radeon: Move hotspot handling out of radeon_set_cursor

It's only needed in radeon_crtc_cursor_set2.

Signed-off-by: Michel Dänzer <michel.daenzer@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
9 years agodrm/radeon: Re-show the cursor after a modeset
Michel Dänzer [Fri, 21 Nov 2014 02:48:57 +0000 (11:48 +0900)]
drm/radeon: Re-show the cursor after a modeset

Setting a mode seems to clear the cursor registers, so we need to
re-program them to make sure the cursor is visible.

Signed-off-by: Michel Dänzer <michel.daenzer@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
9 years agoamdkfd: use atomic allocations within srcu callbacks
Sasha Levin [Wed, 3 Dec 2014 15:19:36 +0000 (10:19 -0500)]
amdkfd: use atomic allocations within srcu callbacks

srcu callbacks are running in atomic context, we can't allocate using
__GFP_WAIT.

Signed-off-by: Sasha Levin <sasha.levin@oracle.com>
Signed-off-by: Oded Gabbay <oded.gabbay@amd.com>
9 years agoamdkfd: use sizeof(long) granularity for the pasid bitmask
Sasha Levin [Wed, 3 Dec 2014 14:26:25 +0000 (09:26 -0500)]
amdkfd: use sizeof(long) granularity for the pasid bitmask

All the bit operations (such as find_first_zero_bit()) read sizeof(long) bytes
at a time. If we allocated less than sizeof(long) bytes for the bitmask we
would be accessing invalid memory when working with the bitmask.

Change the allocator to allocate sizeof(long) multiples for the bitmask.

Signed-off-by: Sasha Levin <sasha.levin@oracle.com>
Signed-off-by: Oded Gabbay <oded.gabbay@amd.com>
9 years agoMerge branch 'vmwgfx-next' of git://people.freedesktop.org/~thomash/linux into drm...
Dave Airlie [Wed, 3 Dec 2014 09:35:05 +0000 (19:35 +1000)]
Merge branch 'vmwgfx-next' of git://people.freedesktop.org/~thomash/linux into drm-next

some vmware fixes.

* 'vmwgfx-next' of git://people.freedesktop.org/~thomash/linux:
  drm/vmwgfx: (Re)bind shaders to MOBs with the correct offset
  drm/vmwgfx: Fix fence event code
  drm/vmwgfx: Don't use memory accounting for kernel-side fence objects
  drm/vmwgfx: Fix error printout on signals pending

9 years agodrm/vmwgfx: (Re)bind shaders to MOBs with the correct offset
Thomas Hellstrom [Tue, 2 Dec 2014 11:41:01 +0000 (03:41 -0800)]
drm/vmwgfx: (Re)bind shaders to MOBs with the correct offset

This codepath is mostly hit when rebinding after a backup buffer swapout. It's
amazing that this error hasn't been more obvious but probably the shaders are
not reread from guest memory that often..

Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Reviewed-by: Jakob Bornecrantz <jakob@vmware.com>
Reviewed-by: Sinclair Yeh <syeh@vmware.com>
9 years agodrm/vmwgfx: Fix fence event code
Thomas Hellstrom [Tue, 2 Dec 2014 11:36:57 +0000 (03:36 -0800)]
drm/vmwgfx: Fix fence event code

The commit "vmwgfx: Rework fence event action" introduced a number of bugs
that are fixed with this commit:

a) A forgotten return stateemnt.
b) An if statement with identical branches.

Cc: <stable@vger.kernel.org>
Reported-by: Rob Clark <robdclark@gmail.com>
Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Reviewed-by: Jakob Bornecrantz <jakob@vmware.com>
Reviewed-by: Sinclair Yeh <syeh@vmware.com>
9 years agodrm/vmwgfx: Don't use memory accounting for kernel-side fence objects
Thomas Hellstrom [Tue, 2 Dec 2014 11:32:24 +0000 (03:32 -0800)]
drm/vmwgfx: Don't use memory accounting for kernel-side fence objects

Kernel side fence objects are used when unbinding resources and may thus be
created as part of a memory reclaim operation. This might trigger recursive
memory reclaims and result in the kernel running out of stack space.

So a simple way out is to avoid accounting of these fence objects.
In principle this is OK since while user-space can trigger the creation of
such objects, it can't really hold on to them. However, their lifetime is
quite long, so some form of accounting should perhaps be implemented in the
future.

Fixes kernel crashes when running, for example viewperf11 ensight-04 test 3
with low system memory settings.

Cc: <stable@vger.kernel.org>
Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Reviewed-by: Jakob Bornecrantz <jakob@vmware.com>
Reviewed-by: Sinclair Yeh <syeh@vmware.com>
9 years agodrm/vmwgfx: Fix error printout on signals pending
Thomas Hellstrom [Tue, 25 Nov 2014 07:20:05 +0000 (08:20 +0100)]
drm/vmwgfx: Fix error printout on signals pending

The function vmw_master_check() might return -ERESTARTSYS if there is a
signal pending, indicating that the IOCTL should be rerun, potentially from
user-space. At that point we shouldn't print out an error message since that
is not an error condition. In short, avoid bloating the kernel log when a
process refuses to die on SIGTERM.

Cc: <stable@vger.kernel.org>
Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Reviewed-by: Jakob Bornecrantz <jakob@vmware.com>
9 years agodrm: rcar-du: Fix NULL encoder pointer dereference
Laurent Pinchart [Thu, 27 Nov 2014 12:49:07 +0000 (14:49 +0200)]
drm: rcar-du: Fix NULL encoder pointer dereference

The DRM connector's encoder pointer is managed internally by the DRM
core and set to NULL when the DRM connector is disconnected from the
CRTC it was attached to. This results in a NULL pointer dereference in
the HDMI connector functions when trying to call the associated slave
encoder's operations.

Fix this by retrieving the slave encoder pointer from the R-Car
connector structure instead of the DRM connector structure.

Signed-off-by: Laurent Pinchart <laurent.pinchart+renesas@ideasonboard.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
9 years agoMerge tag 'drm-intel-next-2014-11-21-fixed' of git://anongit.freedesktop.org/drm...
Dave Airlie [Tue, 2 Dec 2014 22:25:59 +0000 (08:25 +1000)]
Merge tag 'drm-intel-next-2014-11-21-fixed' of git://anongit.freedesktop.org/drm-intel into drm-next

drm-intel-next-2014-11-21:
- infoframe tracking (for fastboot) from Jesse
- start of the dri1/ums support removal
- vlv forcewake timeout fixes (Imre)
- bunch of patches to polish the rps code (Imre) and improve it on bdw (Tom
  O'Rourke)
- on-demand pinning for execlist contexts
- vlv/chv backlight improvements (Ville)
- gen8+ render ctx w/a work from various people
- skl edp programming (Satheeshakrishna et al.)
- psr docbook (Rodrigo)
- piles of little fixes and improvements all over, as usual

* tag 'drm-intel-next-2014-11-21-fixed' of git://anongit.freedesktop.org/drm-intel: (117 commits)
  drm/i915: Don't pin LRC in GGTT when dumping in debugfs
  drm/i915: Update DRIVER_DATE to 20141121
  drm/i915/g4x: fix g4x infoframe readout
  drm/i915: Only call mod_timer() if not already pending
  drm/i915: Don't rely upon encoder->type for infoframe hw state readout
  drm/i915: remove the IRQs enabled WARN from intel_disable_gt_powersave
  drm/i915: Use ggtt error obj capture helper for gen8 semaphores
  drm/i915: vlv: increase timeout when setting idle GPU freq
  drm/i915: vlv: fix cdclk setting during modeset while suspended
  drm/i915: Dump hdmi pipe_config state
  drm/i915: Gen9 shadowed registers
  drm/i915/skl: Gen9 multi-engine forcewake
  drm/i915: Read power well status before other registers for drpc info
  drm/i915: Pin tiled objects for L-shaped configs
  drm/i915: Update ring freq for full gpu freq range
  drm/i915: change initial rps frequency for gen8
  drm/i915: Keep min freq above floor on HSW/BDW
  drm/i915: Use efficient frequency for HSW/BDW
  drm/i915: Can i915_gem_init_ioctl
  drm/i915: Sanitize ->lastclose
  ...

9 years agodrm/i915: Don't pin LRC in GGTT when dumping in debugfs
Thomas Daniel [Tue, 2 Dec 2014 13:21:18 +0000 (13:21 +0000)]
drm/i915: Don't pin LRC in GGTT when dumping in debugfs

LRC object does not need to be mapped into the GGTT when dumping. A side-effect
of this patch is that a compiler warning goes away (not checking return value
of i915_gem_obj_ggtt_pin).

v2: Broke out individual context dumping into a new function as the indentation
was getting a bit crazy.  Added notification of contexts with no gem object for
debugging purposes.  Removed unnecessary pin_pages and unpin_pages, replaced
with explicit get_pages for the context object as there may be no backing store
allocated at this time (Comment for get_pages says "Ensure that the associated
pages are gathered from the backing storage and pinned into our object").
Improved error checking - get_pages and get_page are checked for failure.

Signed-off-by: Thomas Daniel <thomas.daniel@intel.com>
[danvet: Align paramter continuation lines properly. Also add some
braces to the nested loops again for readability.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agoMerge branch 'linux-3.19' of git://anongit.freedesktop.org/git/nouveau/linux-2.6...
Dave Airlie [Tue, 2 Dec 2014 06:13:12 +0000 (16:13 +1000)]
Merge branch 'linux-3.19' of git://anongit.freedesktop.org/git/nouveau/linux-2.6 into drm-next

- Tegra K1 voltage support, and coherency improvements
- GM204 support (modesetting, still waiting on NVIDIA for signed fw to
proceed further), and a lot of bios/i2c/devinit adjustments needed to
support it
- GT21x memory reclocking work
- Various other bits and pieces, most of which are prep-work for a
couple of bigger projects I didn't get finished in time

* 'linux-3.19' of git://anongit.freedesktop.org/git/nouveau/linux-2.6: (73 commits)
  drm/nv50/kms: drop requirement that framebuffer bos be contig up-front
  drm/nv50/kms: directly use cursor image from userspace buffer
  drm/nouveau/kms: when pinning display-related buffers, force contig vram
  drm/nouveau: teach nouveau_bo_pin() how to force a contig vram allocation
  drm/nouveau/volt: add support for GK20A
  drm/nouveau/platform: add GPU speedo information to nouveau platform
  drm/nouveau/volt: allow non-bios voltage scaling
  drm/gf100-/gr: return non-fatal error code when fw not present
  drm/nouveau/devinit: bump priv ring timeouts before executing scripts
  drm/nouveau/bios: translate ramcfg strap through M0203
  drm/nouveau/fb: make use of M0203 routines for ram type determination
  drm/nouveau/bios: add parsing of BIT M(v2) +0x03 table
  drm/nouveau/core: allow vbios parsing without knowing chipset type
  drm/nouveau/lib: add null backend
  drm/nouveau/device: store revision
  drm/nouveau/core: add some forgotten subdevs to disable mask
  drm/gk20a/clk: fix max VCO value
  drm/nouveau: we need pin_refcnt for nouveau_bo_placement_set()
  drm/nv50-/kms: add some evo tracing ability for debugging
  drm/nv50/kms: use sclass() instead of trial-and-error
  ...

9 years agodrm/nv50/kms: drop requirement that framebuffer bos be contig up-front
Ben Skeggs [Mon, 10 Nov 2014 02:36:44 +0000 (12:36 +1000)]
drm/nv50/kms: drop requirement that framebuffer bos be contig up-front

We'll move them at pin() time if necessary.

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
9 years agodrm/nv50/kms: directly use cursor image from userspace buffer
Ben Skeggs [Mon, 10 Nov 2014 05:52:02 +0000 (15:52 +1000)]
drm/nv50/kms: directly use cursor image from userspace buffer

Preparation for transition to planes, which use framebuffers for the
cursor image.  We've always done copies from the userspace buffer up
until now for legacy reasons, there's no good reason to do so on the
chipsets this code covers.

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
9 years agodrm/nouveau/kms: when pinning display-related buffers, force contig vram
Ben Skeggs [Mon, 10 Nov 2014 02:35:06 +0000 (12:35 +1000)]
drm/nouveau/kms: when pinning display-related buffers, force contig vram

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
9 years agodrm/nouveau: teach nouveau_bo_pin() how to force a contig vram allocation
Ben Skeggs [Mon, 10 Nov 2014 01:24:27 +0000 (11:24 +1000)]
drm/nouveau: teach nouveau_bo_pin() how to force a contig vram allocation

We have the ability to move buffers around in the kernel if necessary,
and should probably use it rather than failing if userspace passes us
a non-contig buffer for a plane.

The NOUVEAU_GEM_TILE_NONCONTIG flag from userspace will become a mere
initial placement hint once all the relevant paths have been updated.

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
9 years agodrm/nouveau/volt: add support for GK20A
Vince Hsu [Tue, 2 Dec 2014 04:50:35 +0000 (12:50 +0800)]
drm/nouveau/volt: add support for GK20A

The voltage value are calculated by the hardware characterized
result.

Signed-off-by: Vince Hsu <vinceh@nvidia.com>
Reviewed-by: Alexandre Courbot <acourbot@nvidia.com>
Acked-by: Martin Peres <martin.peres@free.fr>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
9 years agodrm/nouveau/platform: add GPU speedo information to nouveau platform
Vince Hsu [Tue, 2 Dec 2014 04:50:34 +0000 (12:50 +0800)]
drm/nouveau/platform: add GPU speedo information to nouveau platform

For GK20A we need the GPU speedo value to calculate voltage levels.

Signed-off-by: Vince Hsu <vinceh@nvidia.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
9 years agodrm/nouveau/volt: allow non-bios voltage scaling
Vince Hsu [Tue, 2 Dec 2014 04:50:33 +0000 (12:50 +0800)]
drm/nouveau/volt: allow non-bios voltage scaling

Move the vbios parsing out of init() and call it conditionally if the
platform has a vbios. Non-vbios platforms can use the ctor() to init the
data structures.

Signed-off-by: Vince Hsu <vinceh@nvidia.com>
Acked-by: Alexandre Courbot <acourbot@nvidia.com>
Acked-by: Martin Peres <martin.peres@free.fr>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
9 years agodrm/gf100-/gr: return non-fatal error code when fw not present
Ben Skeggs [Tue, 25 Nov 2014 23:03:15 +0000 (09:03 +1000)]
drm/gf100-/gr: return non-fatal error code when fw not present

This allows the module to load without acceleration.

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
9 years agodrm/nouveau/devinit: bump priv ring timeouts before executing scripts
Ben Skeggs [Tue, 25 Nov 2014 22:51:10 +0000 (08:51 +1000)]
drm/nouveau/devinit: bump priv ring timeouts before executing scripts

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
9 years agodrm/nouveau/bios: translate ramcfg strap through M0203
Ben Skeggs [Tue, 4 Nov 2014 05:13:30 +0000 (15:13 +1000)]
drm/nouveau/bios: translate ramcfg strap through M0203

A machine has been spotted where the ramcfg strap is "8", and the ramcfg
xlat table goes 0-7,0-7, resulting in us selecting config 0 for memory
items.  On this particular system, config "8" is available and supposed
to be used.  It appears that starting from GT21x (where Mv2 appears),
we're supposed to use the value in this table instead.

One concern here is that not all the places we currently use ramcfg xlat
are supposed to be treated the same now.  The strap xlat table wasn't
removed from the vbios either, presumably for some kind of good reason.

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
9 years agodrm/nouveau/fb: make use of M0203 routines for ram type determination
Ben Skeggs [Tue, 4 Nov 2014 03:04:35 +0000 (13:04 +1000)]
drm/nouveau/fb: make use of M0203 routines for ram type determination

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
9 years agodrm/nouveau/bios: add parsing of BIT M(v2) +0x03 table
Ben Skeggs [Tue, 4 Nov 2014 02:06:25 +0000 (12:06 +1000)]
drm/nouveau/bios: add parsing of BIT M(v2) +0x03 table

We only support one kind of matching here (ramcfg strap), but it appears
alternate methods are possible.  I wrote a tool to scan our vbios repo
for other types, but did not see any used.  Hopefully this means there
aren't any in the wild that will now break.

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
9 years agodrm/nouveau/core: allow vbios parsing without knowing chipset type
Ben Skeggs [Tue, 18 Nov 2014 00:51:19 +0000 (10:51 +1000)]
drm/nouveau/core: allow vbios parsing without knowing chipset type

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
9 years agodrm/nouveau/lib: add null backend
Ben Skeggs [Tue, 18 Nov 2014 00:36:59 +0000 (10:36 +1000)]
drm/nouveau/lib: add null backend

For the moment, just used to speed up vbios-only testing.  Have some
ideas for extending in the future.

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
9 years agodrm/nouveau/device: store revision
Ben Skeggs [Mon, 17 Nov 2014 12:56:37 +0000 (22:56 +1000)]
drm/nouveau/device: store revision

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
9 years agodrm/nouveau/core: add some forgotten subdevs to disable mask
Ben Skeggs [Mon, 17 Nov 2014 12:52:11 +0000 (22:52 +1000)]
drm/nouveau/core: add some forgotten subdevs to disable mask

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
9 years agodrm/gk20a/clk: fix max VCO value
Alexandre Courbot [Thu, 13 Nov 2014 09:07:27 +0000 (18:07 +0900)]
drm/gk20a/clk: fix max VCO value

For some reason max_vco was set to a lower value that it can support,
which prevented some clock states to be applied. Fix this by setting it
to the same value as downstream.

Signed-off-by: Alexandre Courbot <acourbot@nvidia.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
9 years agodrm/nouveau: we need pin_refcnt for nouveau_bo_placement_set()
Ben Skeggs [Mon, 10 Nov 2014 01:12:17 +0000 (11:12 +1000)]
drm/nouveau: we need pin_refcnt for nouveau_bo_placement_set()

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
9 years agodrm/nv50-/kms: add some evo tracing ability for debugging
Ben Skeggs [Mon, 3 Nov 2014 06:43:59 +0000 (16:43 +1000)]
drm/nv50-/kms: add some evo tracing ability for debugging

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
9 years agodrm/nv50/kms: use sclass() instead of trial-and-error
Ben Skeggs [Mon, 3 Nov 2014 05:01:33 +0000 (15:01 +1000)]
drm/nv50/kms: use sclass() instead of trial-and-error

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
9 years agodrm/nv50/kms: remove a couple of cursor-related stub functions
Ben Skeggs [Mon, 3 Nov 2014 04:33:01 +0000 (14:33 +1000)]
drm/nv50/kms: remove a couple of cursor-related stub functions

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
9 years agodrm/nouveau: fix pin refcnt leak in failure path
Ben Skeggs [Thu, 6 Nov 2014 04:34:31 +0000 (14:34 +1000)]
drm/nouveau: fix pin refcnt leak in failure path

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
9 years agodrm/nouveau: synchronize BOs when required
Alexandre Courbot [Mon, 27 Oct 2014 09:49:19 +0000 (18:49 +0900)]
drm/nouveau: synchronize BOs when required

On architectures for which access to GPU memory is non-coherent,
caches need to be flushed and invalidated explicitly when BO control
changes between CPU and GPU.

This patch adds buffer synchronization functions which invokes the
correct API (PCI or DMA) to ensure synchronization is effective.

Based on the TTM DMA cache helper patches by Lucas Stach.

Signed-off-by: Lucas Stach <dev@lynxeye.de>
Signed-off-by: Alexandre Courbot <acourbot@nvidia.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
9 years agodrm/nouveau: allocate GPFIFOs and fences coherently
Alexandre Courbot [Mon, 27 Oct 2014 09:49:18 +0000 (18:49 +0900)]
drm/nouveau: allocate GPFIFOs and fences coherently

Specify TTM_PL_FLAG_UNCACHED when allocating GPFIFOs and fences to
allow them to be safely accessed by the kernel without being synced
on non-coherent architectures.

Signed-off-by: Alexandre Courbot <acourbot@nvidia.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
9 years agodrm/nouveau: implement explicitly coherent BOs
Alexandre Courbot [Mon, 27 Oct 2014 09:49:17 +0000 (18:49 +0900)]
drm/nouveau: implement explicitly coherent BOs

Allow nouveau_bo_new() to recognize the TTM_PL_FLAG_UNCACHED flag, which
means that we want the allocated BO to be perfectly coherent between the
CPU and GPU. This is useful on non-coherent architectures for which we
do not want to manually sync some rarely-accessed buffers: typically,
fences and pushbuffers.

A TTM BO allocated with the TTM_PL_FLAG_UNCACHED on a non-coherent
architecture will be populated using the DMA API, and accesses to it
performed using the coherent mapping performed by dma_alloc_coherent().

Signed-off-by: Alexandre Courbot <acourbot@nvidia.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
9 years agodrm/nouveau: introduce nv_device_is_cpu_coherent()
Alexandre Courbot [Mon, 27 Oct 2014 09:49:16 +0000 (18:49 +0900)]
drm/nouveau: introduce nv_device_is_cpu_coherent()

Add a function allowing us to know whether a device is CPU-coherent,
i.e. accesses performed by the CPU on GPU-mapped buffers will
be immediately visible on the GPU side and vice-versa.

For now, a device is considered to be coherent if it uses the PCI bus on
a non-ARM architecture.

Signed-off-by: Alexandre Courbot <acourbot@nvidia.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
9 years agodrm/nouveau: warn when moving a pinned object
Alexandre Courbot [Mon, 27 Oct 2014 09:11:52 +0000 (18:11 +0900)]
drm/nouveau: warn when moving a pinned object

Pinned BOs are supposed to remain in their current location until
unpinned. Display a warning for the supposedly-erroneous case where we
are trying to move such objects.

Signed-off-by: Alexandre Courbot <acourbot@nvidia.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
9 years agodrm/nouveau/disp: clear notify intr status when enabling, to prevent races
Ben Skeggs [Tue, 28 Oct 2014 01:21:44 +0000 (11:21 +1000)]
drm/nouveau/disp: clear notify intr status when enabling, to prevent races

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
9 years agodrm/nv50-/disp: rename class members to match nvidia channel names
Ben Skeggs [Thu, 2 Oct 2014 05:25:25 +0000 (15:25 +1000)]
drm/nv50-/disp: rename class members to match nvidia channel names

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
9 years agodrm/nouveau/core: remove some dead code that got forgotten
Ben Skeggs [Tue, 21 Oct 2014 05:23:16 +0000 (15:23 +1000)]
drm/nouveau/core: remove some dead code that got forgotten

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
9 years agodrm/nouveau/pwr/fuc: Fix thinko in nouveau_memx_wait()
Roy Spliet [Thu, 2 Oct 2014 16:01:58 +0000 (18:01 +0200)]
drm/nouveau/pwr/fuc: Fix thinko in nouveau_memx_wait()

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
9 years agodrm/nva3/clock: Allow user reclocking
Roy Spliet [Thu, 2 Oct 2014 16:01:57 +0000 (18:01 +0200)]
drm/nva3/clock: Allow user reclocking

Signed-off-by: Roy Spliet <rspliet@eclipso.eu>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
9 years agodrm/nouveau/fb/ramnva3: Reclocking script for GDDR3
Roy Spliet [Thu, 2 Oct 2014 16:01:56 +0000 (18:01 +0200)]
drm/nouveau/fb/ramnva3: Reclocking script for GDDR3

Signed-off-by: Roy Spliet <rspliet@eclipso.eu>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
9 years agodrm/nouveau/fb/ramnva3: Reclocking script for DDR2
Roy Spliet [Thu, 2 Oct 2014 16:01:55 +0000 (18:01 +0200)]
drm/nouveau/fb/ramnva3: Reclocking script for DDR2

Signed-off-by: Roy Spliet <rspliet@eclipso.eu>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
9 years agodrm/nouveau/fb/ramnva3: Reclocking script for DDR3
Roy Spliet [Thu, 2 Oct 2014 16:01:54 +0000 (18:01 +0200)]
drm/nouveau/fb/ramnva3: Reclocking script for DDR3

Signed-off-by: Roy Spliet <rspliet@eclipso.eu>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
9 years agodrm/nouveau/fb/ramnva3: Ressurect timing calculation code
Roy Spliet [Thu, 2 Oct 2014 16:01:53 +0000 (18:01 +0200)]
drm/nouveau/fb/ramnva3: Ressurect timing calculation code

Signed-off-by: Roy Spliet <rspliet@eclipso.eu>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
9 years agodrm/nouveau/fb/ramnva3: Link training for DDR3
Roy Spliet [Thu, 2 Oct 2014 16:01:52 +0000 (18:01 +0200)]
drm/nouveau/fb/ramnva3: Link training for DDR3

V2: fix whitespace errors in memx.fuc

Signed-off-by: Roy Spliet <rspliet@eclipso.eu>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
9 years agodrm/nouveau/fb/gddr3: Generate MR values
Roy Spliet [Thu, 2 Oct 2014 16:01:51 +0000 (18:01 +0200)]
drm/nouveau/fb/gddr3: Generate MR values

Signed-off-by: Roy Spliet <rspliet@eclipso.eu>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
9 years agodrm/nouveau/bios/ramcfg: 10_02_40 -> DLLoff
Roy Spliet [Thu, 2 Oct 2014 16:01:50 +0000 (18:01 +0200)]
drm/nouveau/bios/ramcfg: 10_02_40 -> DLLoff

Signed-off-by: Roy Spliet <rspliet@eclipso.eu>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
9 years agodrm/nouveau: move the (far too many...) different s/r paths to the same place
Ben Skeggs [Thu, 2 Oct 2014 03:31:00 +0000 (13:31 +1000)]
drm/nouveau: move the (far too many...) different s/r paths to the same place

No code changes.

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
9 years agodrm/gk104-/fifo: handle copy engine class errors
Ben Skeggs [Fri, 26 Sep 2014 05:33:24 +0000 (15:33 +1000)]
drm/gk104-/fifo: handle copy engine class errors

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
9 years agodrm/nouveau/bios: stop after NV+NPDS+ISBN image
Ben Skeggs [Thu, 25 Sep 2014 02:11:16 +0000 (12:11 +1000)]
drm/nouveau/bios: stop after NV+NPDS+ISBN image

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
9 years agodrm/nouveau/bios: add some more signatures as seen on my gtx660
Ben Skeggs [Thu, 25 Sep 2014 01:37:51 +0000 (11:37 +1000)]
drm/nouveau/bios: add some more signatures as seen on my gtx660

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
9 years agodrm/nouveau: add support for gm204
Ben Skeggs [Tue, 19 Aug 2014 00:23:47 +0000 (10:23 +1000)]
drm/nouveau: add support for gm204

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
9 years agodrm/nouveau/device: recognise GM204
Ben Skeggs [Mon, 18 Aug 2014 04:02:14 +0000 (14:02 +1000)]
drm/nouveau/device: recognise GM204

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
9 years agodrm/gm204/disp: some magic that fixes bringup of uninitialised outputs
Ben Skeggs [Thu, 28 Aug 2014 03:00:30 +0000 (13:00 +1000)]
drm/gm204/disp: some magic that fixes bringup of uninitialised outputs

Probably missing something here, doesn't make a lot of sense to write
or+link data into a register whose offset is calculated by the same
or+link info..

This is the all I've witnessed the binary driver and vbios doing so
far, so it'll do.

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
9 years agodrm/gm204/disp: initial support
Ben Skeggs [Mon, 18 Aug 2014 22:45:27 +0000 (08:45 +1000)]
drm/gm204/disp: initial support

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
9 years agodrm/gf110-/disp: magic that might help some tmds issues
Ben Skeggs [Thu, 28 Aug 2014 01:52:12 +0000 (11:52 +1000)]
drm/gf110-/disp: magic that might help some tmds issues

The binary driver has been doing this since GF119, and we've somehow
gotten away with it.  But, TMDS that hasn't been initialised already
by the x86 vbios code is distorted without it on GM204.

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
9 years agodrm/nouveau/bios: support for opcodes 0x47/0x48
Ben Skeggs [Wed, 20 Aug 2014 22:22:03 +0000 (08:22 +1000)]
drm/nouveau/bios: support for opcodes 0x47/0x48

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
9 years agodrm/nouveau/bios: add support for udisp 2.2
Ben Skeggs [Tue, 19 Aug 2014 00:58:39 +0000 (10:58 +1000)]
drm/nouveau/bios: add support for udisp 2.2

Not entirely sure why this got bumped at all yet.

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
9 years agodrm/nouveau/bios: add support for dp 4.1
Ben Skeggs [Mon, 18 Aug 2014 22:38:12 +0000 (08:38 +1000)]
drm/nouveau/bios: add support for dp 4.1

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
9 years agodrm/nouveau/i2c: add support for DCB_I2C_PMGR port type
Ben Skeggs [Mon, 18 Aug 2014 22:25:40 +0000 (08:25 +1000)]
drm/nouveau/i2c: add support for DCB_I2C_PMGR port type

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
9 years agodrm/nouveau/bios: add support for ccb 4.1
Ben Skeggs [Mon, 18 Aug 2014 22:20:26 +0000 (08:20 +1000)]
drm/nouveau/bios: add support for ccb 4.1

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
9 years agodrm/nouveau/bios: add support for dcb 4.1
Ben Skeggs [Mon, 18 Aug 2014 04:09:00 +0000 (14:09 +1000)]
drm/nouveau/bios: add support for dcb 4.1

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
9 years agodrm/gm204/i2c: add aux channel driver
Ben Skeggs [Mon, 18 Aug 2014 21:26:08 +0000 (07:26 +1000)]
drm/gm204/i2c: add aux channel driver

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
9 years agodrm/gm204/i2c: add pad driver
Ben Skeggs [Mon, 18 Aug 2014 21:18:01 +0000 (07:18 +1000)]
drm/gm204/i2c: add pad driver

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
9 years agodrm/nouveau/i2c: segregate aux channel adapter indices from bit-banged i2c
Ben Skeggs [Wed, 24 Sep 2014 00:41:50 +0000 (10:41 +1000)]
drm/nouveau/i2c: segregate aux channel adapter indices from bit-banged i2c

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
9 years agodrm/nouveau/bios: store aux addr independently of i2c
Ben Skeggs [Mon, 18 Aug 2014 22:14:08 +0000 (08:14 +1000)]
drm/nouveau/bios: store aux addr independently of i2c

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
9 years agodrm/nv50-/i2c: kill some unused struct members
Ben Skeggs [Mon, 18 Aug 2014 21:22:32 +0000 (07:22 +1000)]
drm/nv50-/i2c: kill some unused struct members

Left-over from before a rework a while back.

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
9 years agodrm/nouveau/bios: log if auxch accesses fail, also return 0x00 from rd when it does
Ben Skeggs [Thu, 21 Aug 2014 00:39:01 +0000 (10:39 +1000)]
drm/nouveau/bios: log if auxch accesses fail, also return 0x00 from rd when it does

Logging at trace level, rather than as en error, as it seems conceivable
that failure could be normal under certain circumstances (new bios,
older sink that doesn't support a particular DPCD address)

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
9 years agodrm/gm204/devinit: initial implementation
Ben Skeggs [Tue, 23 Sep 2014 05:42:45 +0000 (15:42 +1000)]
drm/gm204/devinit: initial implementation

Starting from GM204, certain registers are no longer accessible by the host
(or unsigned PMU firmware).

This commit implements devinit on PMU, using a signed microcode image, and
devinit data, from the VBIOS.

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
9 years agodrm/nouveau/devinit: allow impl to select its own cold-boot method
Ben Skeggs [Tue, 23 Sep 2014 05:39:26 +0000 (15:39 +1000)]
drm/nouveau/devinit: allow impl to select its own cold-boot method

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
9 years agodrm/nouveau/bios: add parsing of pmu image tables
Ben Skeggs [Mon, 22 Sep 2014 02:30:57 +0000 (12:30 +1000)]
drm/nouveau/bios: add parsing of pmu image tables

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
9 years agodrm/nouveau/bios: recognise nv-specific rom/pcir signatures
Ben Skeggs [Mon, 22 Sep 2014 01:28:21 +0000 (11:28 +1000)]
drm/nouveau/bios: recognise nv-specific rom/pcir signatures

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
9 years agodrm/nouveau/bios: use NPDE to locate images beyond those defined by PCIR
Ben Skeggs [Mon, 22 Sep 2014 01:21:57 +0000 (11:21 +1000)]
drm/nouveau/bios: use NPDE to locate images beyond those defined by PCIR

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
9 years agodrm/nouveau/bios: add NPDE parsing
Ben Skeggs [Sat, 20 Sep 2014 23:35:16 +0000 (09:35 +1000)]
drm/nouveau/bios: add NPDE parsing

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
9 years agodrm/nouveau/bios: fetch images beyond the first one in the rom
Ben Skeggs [Mon, 22 Sep 2014 01:08:01 +0000 (11:08 +1000)]
drm/nouveau/bios: fetch images beyond the first one in the rom

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
9 years agodrm/nouveau/bios: use size/type from pci data structure
Ben Skeggs [Mon, 22 Sep 2014 00:51:37 +0000 (10:51 +1000)]
drm/nouveau/bios: use size/type from pci data structure

The field at +0x2 is technically processor specific, though I don't know
that it's ever mattered in practice (yet).

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
9 years agodrm/nouveau/bios: add pci data structure parsing
Ben Skeggs [Sat, 20 Sep 2014 23:22:48 +0000 (09:22 +1000)]
drm/nouveau/bios: add pci data structure parsing

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
9 years agodrm/nouveau/bios: split out shadow methods
Ben Skeggs [Wed, 17 Sep 2014 23:24:10 +0000 (09:24 +1000)]
drm/nouveau/bios: split out shadow methods

We're about to need to be able to fetch additional chunks of data beyond
the primary bios image, which makes fetching a lot more complicated.

This splits out the verious shadowing routines to be nothing more than
very dumb "fetch this much data from this offset" routines, and leaves
the logic of what and how much to fetch in common code.

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>