GitHub/moto-9609/android_kernel_motorola_exynos9610.git
9 years agodrm/atomic: Clear crtc_state->active in drm_atomic_helper_set_config.
Maarten Lankhorst [Mon, 1 Jun 2015 06:59:53 +0000 (08:59 +0200)]
drm/atomic: Clear crtc_state->active in drm_atomic_helper_set_config.

This fixes some regressions in i915 when converting to atomic.
set_config failed with -EINVAL, and I received the following warning
in dmesg:

[drm:drm_atomic_crtc_check] [CRTC:20] active without enabled

Solve this by clearing active when a crtc is disabled.

Because crtc_state->enable implies that connectors are active the
change from disabled->enabled can only happen for the crtc that's
being set_config'd, and checking for !crtc_state->enable is sufficient
here.

Cc: dri-devel@lists.freedesktop.org
Reviewed-by: Rob Clark <robdclark@gmail.com>
Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
9 years agoMerge tag 'topic/drm-misc-2015-05-27' of git://anongit.freedesktop.org/drm-intel...
Dave Airlie [Wed, 3 Jun 2015 23:17:45 +0000 (09:17 +1000)]
Merge tag 'topic/drm-misc-2015-05-27' of git://anongit.freedesktop.org/drm-intel into drm-next

One more round of drm-misc, again mostly atomic. Big thing is the
userspace blob code from Daniel Stone, with support for the mode_id blob
now added to the atomic ioctl. Finally we can do atomic modesets!

Note that the atomic ioctl is still behind the module knob since the
weston patches aren't quite ready yet imo - they lack TEST_ONLY support,
which is a fairly crucial bit of the atomic api. But besides that I think
it's all good to go. That's also why we didn't bother to hide the new blob
ioctls behind the knob, that part won't need to change. And if weston
patches get in shape in time we could throw the "atomic by default patch"
on top for 4.2.

* tag 'topic/drm-misc-2015-05-27' of git://anongit.freedesktop.org/drm-intel:
  drm: Fix off-by-one in vblank hardware counter wraparound handling
  drm/atomic: fix out of bounds read in for_each_*_in_state helpers
  drm/atomic: Add MODE_ID property
  drm/atomic: Add current-mode blob to CRTC state
  drm: Add drm_atomic_set_mode_for_crtc
  drm: check for garbage in unused addfb2 fields
  drm: Retain reference to blob properties in lookup
  drm/mode: Add user blob-creation ioctl
  drm: Return error value from blob creation
  drm: Allow creating blob properties without copy
  drm/mode: Unstatic kernel-userspace mode conversion
  drm/mode: Validate modes inside drm_crtc_convert_umode
  drm/crtc_helper: Replace open-coded CRTC state helpers
  drm: kerneldoc fixes for blob properties
  drm/DocBook: Add more drm_bridge documentation
  drm: bridge: Allow daisy chaining of bridges
  drm/atomic: add all affected planes in drm_atomic_helper_check_modeset
  drm/atomic: add drm_atomic_add_affected_planes
  drm/atomic: add commit_planes_on_crtc helper

9 years agoMerge tag 'drm-amdkfd-next-2015-06-03' of git://people.freedesktop.org/~gabbayo/linux...
Dave Airlie [Wed, 3 Jun 2015 23:15:39 +0000 (09:15 +1000)]
Merge tag 'drm-amdkfd-next-2015-06-03' of git://people.freedesktop.org/~gabbayo/linux into drm-next

drm-amdkfd-next-2015-06-03:

- Add the H/W debugger support module, including new IOCTLs to:
  - register/unregister a process as a debugged process
  - Set address watch-point in the debugged process's GPU kernel
  - Do a wave control operation in the debugged process's waves
  See the commit messages for more details on the available operations.

  The debugged process can only perform debug operation on itself. It is
  blocked by the amdkfd+H/W from performing operations on other processes's
  waves or GPU kernels. The blocking is done by setting the VMID and PASID of
  the debugged process in the packets that are sent to the CP with the debug
  instructions.

- Add support for static user-mode queues. These queues are regular queues,
  but because they belong to the debugged process, we need to make sure the CP
  doesn't preempt them during a debug operation. Therefore, we mark them as
  static for the CP ignore them during preemption.

- Support killing all the waves when a process is terminated. This is needed
  in case a process is terminated but we can't UNMAP its queues (can occur due
  to several reasons). In that case, the CP could be stuck unless we kill all
  its waves. This function is *very* important as it provides the kernel a high
  level of control over the GPU. The reason we didn't upstream this function
  so far, is because it is implemented using the H/W debugger module functions,
  so we had to wait until we can upstream the H/W debugger module.

- Replace declaration of bitmap from unsigned long to standard DECLARE_BITMAP

* tag 'drm-amdkfd-next-2015-06-03' of git://people.freedesktop.org/~gabbayo/linux:
  drm/amdkfd: Enforce kill all waves on process termination
  drm/radeon: Add ATC VMID<-->PASID functions to kfd->kgd
  drm/amdkfd: Implement address watch debugger IOCTL
  drm/amdkfd: Implement wave control debugger IOCTL
  drm/amdkfd: Implement (un)register debugger IOCTLs
  drm/amdkfd: Add address watch operation to debugger
  drm/amdkfd: Add wave control operation to debugger
  drm/amdkfd: Add skeleton H/W debugger module support
  drm/amdkfd: Add static user-mode queues support
  drm/amdkfd: add H/W debugger IOCTL set definitions
  drm/radeon: Add H/W debugger kfd->kgd functions
  drm/amdkfd: Use DECLARE_BITMAP

9 years agodrm/amdkfd: Enforce kill all waves on process termination
Ben Goz [Wed, 20 May 2015 15:05:44 +0000 (18:05 +0300)]
drm/amdkfd: Enforce kill all waves on process termination

This commit makes sure that on process termination, after
we're destroying all the active queues, we're killing all the
existing wave front of the current process.

By doing this we're making sure that if any of the CUs were blocked
by infinite loop we're enforcing it to end the shader explicitly.

Signed-off-by: Ben Goz <ben.goz@amd.com>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
9 years agodrm/radeon: Add ATC VMID<-->PASID functions to kfd->kgd
Alexey Skidanov [Tue, 19 May 2015 16:25:01 +0000 (19:25 +0300)]
drm/radeon: Add ATC VMID<-->PASID functions to kfd->kgd

This patch adds three new interfaces to kfd2kgd interface file of radeon.

The interfaces are:

- Check if a specific VMID has a valid PASID mapping
- Retrieve the PASID which is mapped to a specific VMID
- Issue a VMID invalidation request to the ATC

Signed-off-by: Alexey Skidanov <Alexey.Skidanov@amd.com>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
9 years agodrm/amdkfd: Implement address watch debugger IOCTL
Yair Shachar [Wed, 20 May 2015 11:09:39 +0000 (14:09 +0300)]
drm/amdkfd: Implement address watch debugger IOCTL

v2:

- rename get_dbgmgr_mutex to kfd_get_dbgmgr_mutex to namespace it
- change void* to uint64_t inside ioctl arguments
- use kmalloc instead of kzalloc because we use copy_from_user
  immediately after it

Signed-off-by: Yair Shachar <yair.shachar@amd.com>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
9 years agodrm/amdkfd: Implement wave control debugger IOCTL
Yair Shachar [Wed, 20 May 2015 11:09:24 +0000 (14:09 +0300)]
drm/amdkfd: Implement wave control debugger IOCTL

v2:

- rename get_dbgmgr_mutex to kfd_get_dbgmgr_mutex to namespace it
- change void* to uint64_t inside ioctl arguments
- use kmalloc instead of kzalloc because we use copy_from_user
  immediately after it

Signed-off-by: Yair Shachar <yair.shachar@amd.com>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
9 years agodrm/amdkfd: Implement (un)register debugger IOCTLs
Yair Shachar [Wed, 20 May 2015 11:08:55 +0000 (14:08 +0300)]
drm/amdkfd: Implement (un)register debugger IOCTLs

v2: rename get_dbgmgr_mutex to kfd_get_dbgmgr_mutex to namespace it

Signed-off-by: Yair Shachar <yair.shachar@amd.com>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
9 years agodrm/amdkfd: Add address watch operation to debugger
Yair Shachar [Wed, 20 May 2015 10:59:17 +0000 (13:59 +0300)]
drm/amdkfd: Add address watch operation to debugger

The address watch operation gives the ability to specify watch points
which will generate a shader breakpoint, based on a specified single
address or range of addresses.

There is support for read/write/any access modes.

Signed-off-by: Yair Shachar <yair.shachar@amd.com>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
9 years agodrm/amdkfd: Add wave control operation to debugger
Yair Shachar [Wed, 20 May 2015 10:58:12 +0000 (13:58 +0300)]
drm/amdkfd: Add wave control operation to debugger

The wave control operation supports several command types executed upon
existing wave fronts that belong to the currently debugged process.

The available commands are:

HALT   - Freeze wave front(s) execution
RESUME - Resume freezed wave front(s) execution
KILL   - Kill existing wave front(s)

Signed-off-by: Yair Shachar <yair.shachar@amd.com>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
9 years agodrm/amdkfd: Add skeleton H/W debugger module support
Yair Shachar [Wed, 20 May 2015 10:48:26 +0000 (13:48 +0300)]
drm/amdkfd: Add skeleton H/W debugger module support

This patch adds the skeleton H/W debugger module support. This code
enables registration and unregistration of a single HSA process at a
time.

The module saves the process's pasid and use it to verify that only the
registered process is allowed to execute debugger operations through the
kernel driver.

v2: rename get_dbgmgr_mutex to kfd_get_dbgmgr_mutex to namespace it

Signed-off-by: Yair Shachar <yair.shachar@amd.com>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
9 years agodrm/amdkfd: Add static user-mode queues support
Yair Shachar [Wed, 20 May 2015 10:43:04 +0000 (13:43 +0300)]
drm/amdkfd: Add static user-mode queues support

This patch adds support for static user-mode queues in QCM.
Queues which are designated as static can NOT be preempted by
the CP microcode when it is executing its scheduling algorithm.

This is needed for supporting the debugger feature, because we
can't allow the CP to preempt queues which are currently being debugged.

The number of queues that can be designated as static is limited by the
number of HQDs (Hardware Queue Descriptors).

Signed-off-by: Yair Shachar <yair.shachar@amd.com>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
9 years agodrm/amdkfd: add H/W debugger IOCTL set definitions
Yair Shachar [Sun, 7 Dec 2014 15:05:22 +0000 (17:05 +0200)]
drm/amdkfd: add H/W debugger IOCTL set definitions

This patch adds four new IOCTLs to amdkfd. These IOCTLs expose a H/W
debugger functionality to the userspace.

The IOCTLs are:

- AMDKFD_IOC_DBG_REGISTER:

The purpose of this IOCTL is to notify amdkfd that a process wants to use
GPU debugging facilities on itself only.
It is expected that this IOCTL would be called before any other H/W
debugger requests are sent to amdkfd and for each GPU where the H/W
debugging needs to be enabled. The use of this IOCTL ensures that only
one instance of a debugger is active in the system.

- AMDKFD_IOC_DBG_UNREGISTER:

This IOCTL detaches the debugger/debugged process from the H/W
Debug which was established by the AMDKFD_IOC_DBG_REGISTER IOCTL.

- AMDKFD_IOC_DBG_ADDRESS_WATCH:

This IOCTL allows to set different watchpoints with various conditions as
indicated by the IOCTL's arguments. The available number of watchpoints
is retrieved from topology. This operation is confined to the current
debugged process, which was registered through AMDKFD_IOC_DBG_REGISTER.

- AMDKFD_IOC_DBG_WAVE_CONTROL:

This IOCTL allows to control a wavefront as indicated by the IOCTL's
arguments. For example, you can halt/resume or kill either a
single wavefront or a set of wavefronts. This operation is confined to
the current debugged process, which was registered through
AMDKFD_IOC_DBG_REGISTER.

Because the arguments for the address watch IOCTL and wave control IOCTL
are dynamic, meaning that they could vary in size, the userspace passes a
pointer to a structure (in userspace) that contains the value of the
arguments. The kernel driver is responsible to parse this structure and
validate its contents.

v2: change void* to uint64_t inside ioctl arguments

Signed-off-by: Yair Shachar <yair.shachar@amd.com>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
9 years agodrm/radeon: Add H/W debugger kfd->kgd functions
Yair Shachar [Sun, 28 Sep 2014 08:51:15 +0000 (11:51 +0300)]
drm/radeon: Add H/W debugger kfd->kgd functions

This patch adds new interface functions to the kfd2kgd interface file. The
new functions allow to perform H/W debugger operations by writing to GPU
registers.

Signed-off-by: Yair Shachar <yair.shachar@amd.com>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
9 years agodrm/amdkfd: Use DECLARE_BITMAP
Joe Perches [Wed, 20 May 2015 01:37:51 +0000 (18:37 -0700)]
drm/amdkfd: Use DECLARE_BITMAP

Use the generic mechanism to declare a bitmap instead of unsigned long.

It seems that "struct kfd_process.allocated_queue_bitmap" is unused.
Maybe it could be deleted instead.

Signed-off-by: Joe Perches <joe@perches.com>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
9 years agoMerge tag 'drm-intel-next-fixes-2015-05-29' of git://anongit.freedesktop.org/drm...
Dave Airlie [Tue, 2 Jun 2015 08:10:50 +0000 (18:10 +1000)]
Merge tag 'drm-intel-next-fixes-2015-05-29' of git://anongit.freedesktop.org/drm-intel into drm-next

Fixes for 4.2. Nothing too serious (given that it's still pre merge
window). With that it's off for 2 weeks of vacation for me and taking care
of 4.2 fixes for Jani.

* tag 'drm-intel-next-fixes-2015-05-29' of git://anongit.freedesktop.org/drm-intel:
  drm/i915: limit PPGTT size to 2GB in 32-bit platforms
  drm/i915: Another fbdev hack to avoid PSR on fbcon.
  drm/i915: Return the frontbuffer flip to enable intel_crtc_enable_planes.
  drm/i915: disable IPS while getting the sink CRCs
  drm/i915: Disable 12bpc hdmi for now
  drm/i915: Adjust sideband locking a bit for CHV/VLV
  drm/i915: s/dpio_lock/sb_lock/
  drm/i915: Kill intel_flush_primary_plane()
  drm/i915: Throw out WIP CHV power well definitions
  drm/i915: Use the default 600ns LDO programming sequence delay
  drm/i915: Remove unnecessary null check in execlists_context_unqueue
  drm/i915: Use spinlocks for checking when to waitboost
  drm/i915: Fix the confusing comment about the ioctl limits
  Revert "drm/i915: Force clean compilation with -Werror"

9 years agodrm/ttm: dma: Don't crash on memory in the vmalloc range
Alexandre Courbot [Fri, 15 May 2015 07:09:54 +0000 (16:09 +0900)]
drm/ttm: dma: Don't crash on memory in the vmalloc range

dma_alloc_coherent() can return memory in the vmalloc range.
virt_to_page() cannot handle such addresses and crashes. This
patch detects such cases and obtains the struct page * using
vmalloc_to_page() instead.

Signed-off-by: Alexandre Courbot <acourbot@nvidia.com>
Acked-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
9 years agodrm/i915: limit PPGTT size to 2GB in 32-bit platforms
Michel Thierry [Fri, 29 May 2015 13:15:05 +0000 (14:15 +0100)]
drm/i915: limit PPGTT size to 2GB in 32-bit platforms

We already set this limit for the GGTT.

This is a temporary patch until a full replacement of size_t variables
(inadequate in 32-bit kernel) is in place.

Regression from:
commit a4e0bedca678c81eea4cd79a4bd502335639f73a
Author: Michel Thierry <michel.thierry@intel.com>
Date:   Wed Apr 8 12:13:35 2015 +0100

drm/i915: Use complete address space in true PPGTT

v2: Prettify code and explain why this is needed. (Chris)
v3: Don't hide the compilation warning in 32-bit. (Chris)

Suggested-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Cc: Mika Kuoppala <mika.kuoppala@intel.com>
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Michel Thierry <michel.thierry@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/i915: Another fbdev hack to avoid PSR on fbcon.
Rodrigo Vivi [Thu, 28 May 2015 17:26:58 +0000 (10:26 -0700)]
drm/i915: Another fbdev hack to avoid PSR on fbcon.

With unified modeset and flip paths introduced recently when switching
to fbcon PSR was being disabled on fb_set_par path but re-enabled on
fb_pan_display one, causing missed screen updates and un unusable
console.

Regression introduced with:

commit bb54662350662815b4bfc2ff4464330a2dbd7041
Author: Ander Conselvan de Oliveira <ander.conselvan.de.oliveira@intel.com>
Date:   Tue Apr 21 17:13:13 2015 +0300

    drm/i915: Unify modeset and flip paths of intel_crtc_set_config()

Cc: Ander Conselvan de Oliveira <ander.conselvan.de.oliveira@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/i915: Return the frontbuffer flip to enable intel_crtc_enable_planes.
Rodrigo Vivi [Thu, 28 May 2015 17:21:16 +0000 (10:21 -0700)]
drm/i915: Return the frontbuffer flip to enable intel_crtc_enable_planes.

Without this frontbuffer flip when enabling planes PSR got compromised
and wasn't being enabled waiting forever on the flush that never
arrived.

Another solution would to create a enable_cursor function and split this
frontbuffer flip among the different plane enable and disable functions.
But if necessary this can be done in a follow up work. For now let's
just fix the regression.

It was removed by:

commit 87d4300a7dbc19634018e147b4753f3c9bb5f471
Author: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Date:   Tue Apr 21 17:12:54 2015 +0300

    drm/i915: Move intel_(pre_disable/post_enable)_primary to intel_display.c, and use it there.

Cc: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agoMerge branch 'drm-tda998x-devel' of git://ftp.arm.linux.org.uk/~rmk/linux-arm into...
Dave Airlie [Thu, 28 May 2015 23:19:59 +0000 (09:19 +1000)]
Merge branch 'drm-tda998x-devel' of git://ftp.arm.linux.org.uk/~rmk/linux-arm into drm-next

warning fix for tda998x

* 'drm-tda998x-devel' of git://ftp.arm.linux.org.uk/~rmk/linux-arm:
  drm/i2c: tda998x: fix compiler warning for ssize_t

9 years agodrm: clean up drm_mm debugfs output
Russell King [Thu, 28 May 2015 09:36:27 +0000 (10:36 +0100)]
drm: clean up drm_mm debugfs output

The drm_mm debugfs output is difficult to read as two different formats
are used for the addresses:

0x00000080000000-0x0000008000b000: 45056: used
0x8000b000-0x80016000: 45056: free
0x00000080016000-0x0000008001b000: 20480: used
0x8001b000-0x817a1000: 24666112: free
0x000000817a1000-0x000000817a8000: 28672: used
0x000000817a8000-0x00000081ba8000: 4194304: used

Fix this by using %#018llx for all addresses, thus making the output:

0x0000000080000000-0x000000008000b000: 45056: used
0x000000008000b000-0x0000000080016000: 45056: free
0x0000000080016000-0x000000008001b000: 20480: used
0x000000008001b000-0x00000000817a1000: 24666112: free
0x00000000817a1000-0x00000000817a8000: 28672: used
0x00000000817a8000-0x0000000081ba8000: 4194304: used

Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
9 years agoMerge tag 'drm-intel-next-2015-05-22' of git://anongit.freedesktop.org/drm-intel...
Dave Airlie [Thu, 28 May 2015 23:11:49 +0000 (09:11 +1000)]
Merge tag 'drm-intel-next-2015-05-22' of git://anongit.freedesktop.org/drm-intel into drm-next

- cpt modeset sequence fixes from Ville
- more rps boosting tuning from Chris
- S3 support for skl (Damien)
- a pile of w/a for bxt from various people
- cleanup of primary plane pixel formats (Damien)
- a big pile of small patches with fixes and cleanups all over

* tag 'drm-intel-next-2015-05-22' of git://anongit.freedesktop.org/drm-intel: (90 commits)
  drm/i915: Update DRIVER_DATE to 20150522
  drm/i915: Introduce DRM_I915_THROTTLE_JIFFIES
  drm/i915: Use the correct destructor for freeing requests on error
  drm/i915/skl: don't fail colorkey + scaler request
  drm/i915: Enable GTT caching on gen8
  drm/i915: Move WaProgramL3SqcReg1Default:bdw to init_clock_gating()
  drm/i915: Use ilk_init_lp_watermarks() on BDW
  drm/i915: Disable FDI RX/TX before the ports
  drm/i915: Disable CRT port after pipe on PCH platforms
  drm/i915: Disable SDVO port after the pipe on PCH platforms
  drm/i915: Disable HDMI port after the pipe on PCH platforms
  drm/i915: Fix the IBX transcoder B workarounds
  drm/i915: Write the SDVO reg twice on IBX
  drm/i915: Fix DP enhanced framing for CPT
  drm/i915: Clean up the CPT DP .get_hw_state() port readout
  drm/i915: Clarfify the DP code platform checks
  drm/i915: Remove the double register write from intel_disable_hdmi()
  drm/i915: Remove a bogus 12bpc "toggle" from intel_disable_hdmi()
  drm/i915/skl: Deinit/init the display at suspend/resume
  drm/i915: Free RPS boosts for all laggards
  ...

9 years agoMerge branch 'drm-next-4.2' of git://people.freedesktop.org/~agd5f/linux into drm...
Dave Airlie [Thu, 28 May 2015 23:10:54 +0000 (09:10 +1000)]
Merge branch 'drm-next-4.2' of git://people.freedesktop.org/~agd5f/linux into drm-next

for amdgpu separately next week.  Highlights for radeon:
- VCE1 support
- Bug fixes and misc cleanups

* 'drm-next-4.2' of git://people.freedesktop.org/~agd5f/linux:
  radeon: Deinline indirect register accessor functions
  drm/radeon: Fix max_vblank_count value for current display engines
  drm/radeon: stop using addr to check for BO move
  drm/radeon: clean up radeon_audio_enable
  drm/radeon: take the mode_config mutex when dealing with hpds (v2)
  drm/radeon: make dpcd parameters const
  drm/radeon: Use DECLARE_BITMAP
  drm/radeon/tn/si: enable/disable vce cg when encoding v2
  drm/radeon: add support for vce 1.0 clock gating
  drm/radeon: add VCE 1.0 support v4
  drm/radeon/dpm: add vce support for SI
  drm/radeon/dpm: add vce dpm support for TN
  drm/radeon: implement tn_set_vce_clocks
  drm/radeon: implement si_set_vce_clocks v2
  drm/radeon: allow some more VCE firmware versions
  drm/radeon: rework VCE FW size calculation
  drm/radeon: add a GPU reset counter queryable by userspace

9 years agoradeon: Deinline indirect register accessor functions
Denys Vlasenko [Wed, 20 May 2015 11:02:37 +0000 (13:02 +0200)]
radeon: Deinline indirect register accessor functions

This patch deinlines indirect register accessor functions.

These functions perform two mmio accesses, framed by spin lock/unlock.
Spin lock/unlock by itself takes more than 50 cycles in ideal case
(if lock is exclusively cached on current CPU).

With this .config: http://busybox.net/~vda/kernel_config,
after uninlining these functions have sizes and callsite counts
as follows:

r600_uvd_ctx_rreg: 111 bytes, 4 callsites
r600_uvd_ctx_wreg: 113 bytes, 5 callsites
eg_pif_phy0_rreg: 106 bytes, 13 callsites
eg_pif_phy0_wreg: 108 bytes, 13 callsites
eg_pif_phy1_rreg: 107 bytes, 13 callsites
eg_pif_phy1_wreg: 108 bytes, 13 callsites
rv370_pcie_rreg: 111 bytes, 21 callsites
rv370_pcie_wreg: 113 bytes, 24 callsites
r600_rcu_rreg: 111 bytes, 16 callsites
r600_rcu_wreg: 113 bytes, 25 callsites
cik_didt_rreg: 106 bytes, 10 callsites
cik_didt_wreg: 107 bytes, 10 callsites
tn_smc_rreg: 106 bytes, 126 callsites
tn_smc_wreg: 107 bytes, 116 callsites
eg_cg_rreg: 107 bytes, 20 callsites
eg_cg_wreg: 108 bytes, 52 callsites

Functions r100_mm_rreg() and r100_mm_rreg() have a fast path and
a locked (slow) path. This patch deinlines only slow path.

r100_mm_rreg_slow: 78 bytes, 2083 callsites
r100_mm_wreg_slow: 81 bytes, 3570 callsites

Reduction in code size is more than 65,000 bytes:

    text     data      bss       dec     hex filename
85740176 22294680 20627456 128662312 7ab3b28 vmlinux.before
85674192 22294776 20627456 128598664 7aa4288 vmlinux

Signed-off-by: Denys Vlasenko <dvlasenk@redhat.com>
Cc: Christian König <christian.koenig@amd.com>
Cc: Alex Deucher <alexander.deucher@amd.com>
Cc: linux-kernel@vger.kernel.org
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
9 years agodrm/i915: disable IPS while getting the sink CRCs
Paulo Zanoni [Mon, 25 May 2015 21:52:29 +0000 (18:52 -0300)]
drm/i915: disable IPS while getting the sink CRCs

This commit is the "sink CRC" version of:

commit 8c740dcea254a1472df2c0ac5ac585412a2507ec
Author: Paulo Zanoni <paulo.r.zanoni@intel.com>
Date:   Fri Oct 17 18:42:03 2014 -0300
    drm/i915: disable IPS while getting the pipe CRCs.

For some unknown reason, when IPS gets enabled, the sink CRC changes.
Since hsw_enable_ips() doesn't really guarantee to enable IPS (it
depends on package C-states), we can't really predict if IPS is
enabled or disabled while running our CRC tests, so let's just
completely disable IPS while sink CRCs are being used.

If we find a way to make IPS not change the pipe CRC result, we may
want to fix IPS and then revert this patch (and 8c740dcea too). While
this doesn't happen, let's merge this patch, so the IGT tests relying
on sink CRCs can work properly.

This was discovered while developing a new IGT test, which will
probably be called kms_frontbuffer_tracking.

Testcase: igt/kms_frontbuffer_tracking (not on upstream IGT yet)
Cc: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/i915: Disable 12bpc hdmi for now
Daniel Vetter [Thu, 28 May 2015 07:38:45 +0000 (09:38 +0200)]
drm/i915: Disable 12bpc hdmi for now

It's totally broken, and since

commit d328c9d78d64ca11e744fe227096990430a88477
Author: Daniel Vetter <daniel.vetter@ffwll.ch>
Date:   Fri Apr 10 16:22:37 2015 +0200

    drm/i915: Select starting pipe bpp irrespective or the primary plane

the kernel will try to use it even for the common rgb888 framebuffers.
Ville has patches to fix it all up properly, but unfortunately they're
stuck in review limbo. And since the 4.2 feature cutoff has passed we
need to somehow  handle this regression.

Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
Cc: Paulo Zanoni <przanoni@gmail.com>
Signed-off-by: Daniel Vetter <daniel.vetter@intel.com>
9 years agodrm/i915: Adjust sideband locking a bit for CHV/VLV
Ville Syrjälä [Tue, 26 May 2015 17:42:31 +0000 (20:42 +0300)]
drm/i915: Adjust sideband locking a bit for CHV/VLV

chv_enable_pll() doesn't need to hold sb_lock for the entire duration of
the function. Drop the lock as soon as possible.

valleyview_set_cdclk() does a potential lock+unlock+lock+unlock cycle
with sb_lock. Grab the lock a few lines earlier so we can make do
with a single lock+unlock cycle always.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Jani Nikula <jani.nikula@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/i915: s/dpio_lock/sb_lock/
Ville Syrjälä [Tue, 26 May 2015 17:42:30 +0000 (20:42 +0300)]
drm/i915: s/dpio_lock/sb_lock/

Rename dpio_lock to sb_lock to inform the reader that its primary
purpose is to protect the sideband mailbox rather than some DPIO
state.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Jani Nikula <jani.nikula@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/i915: Kill intel_flush_primary_plane()
Ville Syrjälä [Tue, 26 May 2015 17:27:23 +0000 (20:27 +0300)]
drm/i915: Kill intel_flush_primary_plane()

The primary plane frobbing was removed from the sprite code in
 commit ecce87ea3ab55ad0dc64460e6422c357d158a55e
 Author: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
 Date:   Tue Apr 21 17:12:50 2015 +0300

    drm/i915: Remove implicitly disabling primary plane for now

but the intel_flush_primary_plane() calls were left behind. Replace them
with straight forward POSTING_READ() of the sprite surface address
register.

The other user of intel_flush_primary_plane() is g4x_disable_trickle_feed()
where we can just inline the steps directly.

This allows intel_flush_primary_plane() to be killed off.

Cc: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/i915: Throw out WIP CHV power well definitions
Ville Syrjälä [Tue, 26 May 2015 17:22:39 +0000 (20:22 +0300)]
drm/i915: Throw out WIP CHV power well definitions

Expecting CHV power wells to be just an extended versions of the VLV
power wells, a bunch of commented out power wells were added in
anticipation when Punit folks would implement it all. Turns out they
never did, and instead CHV has fewer power wells than VLV. Rip out all
the #if 0'ed junk that's not needed.

v2: Rename the "pipe-a" well to "display" to match VLV
    Clarify the pipe A power well relationship to pipes B and C (Deepak)

Reviewed-by: Deepak S <deepak.s@linux.intel.com>
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/i915: Use the default 600ns LDO programming sequence delay
Ville Syrjälä [Tue, 26 May 2015 17:22:38 +0000 (20:22 +0300)]
drm/i915: Use the default 600ns LDO programming sequence delay

Not sure which LDO programming sequence delay should be used for the CHV
PHY, but the spec says that 600ns is "Used by default for initial
bringup", and the BIOS seems to use that, so let's do the same.

Reviewed-by: Deepak S <deepak.s@linux.intel.com>
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/radeon: Fix max_vblank_count value for current display engines
Michel Dänzer [Tue, 26 May 2015 08:53:39 +0000 (17:53 +0900)]
drm/radeon: Fix max_vblank_count value for current display engines

The value was much too low, which could cause the userspace visible
vblank counter to move backwards when the hardware counter wrapped
around.

Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Michel Dänzer <michel.daenzer@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
9 years agodrm/radeon: stop using addr to check for BO move
Christian König [Tue, 26 May 2015 10:24:15 +0000 (12:24 +0200)]
drm/radeon: stop using addr to check for BO move

It is theoretically possible that a swapped out BO gets the
same GTT address, but different backing pages while being swapped in.

Instead just use another VA state to note updated areas.

Signed-off-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
9 years agodrm/radeon: clean up radeon_audio_enable
Alex Deucher [Mon, 18 May 2015 15:11:48 +0000 (11:11 -0400)]
drm/radeon: clean up radeon_audio_enable

- make it static
- fix mask/bool handling for last param

Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Cc: stable@vger.kernel.org
9 years agodrm/i915: Remove unnecessary null check in execlists_context_unqueue
Michel Thierry [Mon, 27 Apr 2015 11:31:44 +0000 (12:31 +0100)]
drm/i915: Remove unnecessary null check in execlists_context_unqueue

commit 53292cdb066950611e5bc2e0eb109c7edb42af78 ("drm/i915: Workaround
to avoid lite restore with HEAD==TAIL") added a check for req0 != null
which is unnecessary.

The only way req0 could be null is if the list was empty, and this is
already addressed at the beginning of execlists_context_unqueue().

Reported-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Michel Thierry <michel.thierry@intel.com>
Reviewed-by: Jani Nikula <jani.nikula@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm: Fix off-by-one in vblank hardware counter wraparound handling
Michel Dänzer [Tue, 26 May 2015 08:53:38 +0000 (17:53 +0900)]
drm: Fix off-by-one in vblank hardware counter wraparound handling

dev->max_vblank_count contains the largest value that can be represented
by the hardware counter. When the hardware counter wraps around, we have
to add that value + 1 to get the same value as if the hardware counter
didn't wrap around.

Signed-off-by: Michel Dänzer <michel.daenzer@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/i915: Use spinlocks for checking when to waitboost
Chris Wilson [Thu, 21 May 2015 20:01:47 +0000 (21:01 +0100)]
drm/i915: Use spinlocks for checking when to waitboost

In commit 1854d5ca0dd7a9fc11243ff220a3e93fce2b4d3e
Author: Chris Wilson <chris@chris-wilson.co.uk>
Date:   Tue Apr 7 16:20:32 2015 +0100

    drm/i915: Deminish contribution of wait-boosting from clients

we removed an atomic timer based check for allowing waitboosting and
moved it below the mutex taken during RPS. However, that mutex can be
held for long periods of time on Vallyview/Cherryview as communication
with the PCU is slow. As clients may frequently wait for results (e.g.
such as tranform feedback) we introduced contention between the client
and the RPS worker. We can take advantage of the RPS worker, by
switching the wait boost decision to use spin locks and defer the
actual reclocking to the worker.

Fixes a regression of up to 45% on Baytrail and Baswell!

v2 (Daniel):
- Use max_freq_softlimit instead of the not-yet-merged boost
  frequency.
- Don't inject a fake irq into the boost work, instead treat
  client_boost as just another legit waker.

v3: Drop the now unused mask (Chris).

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=90112
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> (v1)
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/atomic: fix out of bounds read in for_each_*_in_state helpers
Andrey Ryabinin [Mon, 25 May 2015 10:29:44 +0000 (13:29 +0300)]
drm/atomic: fix out of bounds read in for_each_*_in_state helpers

for_each_*_in_state validate array index after
access to array elements, thus perform out of bounds read.

Fix this by validating index in the first place and read
array element iff validation was successful.

Fixes: df63b9994eaf ("drm/atomic: Add for_each_{connector,crtc,plane}_in_state helper macros")
Signed-off-by: Andrey Ryabinin <a.ryabinin@samsung.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/i915: Fix the confusing comment about the ioctl limits
Damien Lespiau [Tue, 26 May 2015 13:57:19 +0000 (14:57 +0100)]
drm/i915: Fix the confusing comment about the ioctl limits

It was reported that this comment was confusing, and indeed it is.

v2: (one year later!) Add the range for the DRM_I915_* iotcl defines
    (Daniel)

Signed-off-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/radeon: take the mode_config mutex when dealing with hpds (v2)
Alex Deucher [Fri, 15 May 2015 15:48:52 +0000 (11:48 -0400)]
drm/radeon: take the mode_config mutex when dealing with hpds (v2)

Since we are messing with state in the worker.

v2: drop the changes in the mst worker

Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Cc: stable@vger.kernel.org
9 years agodrm/radeon: make dpcd parameters const
Alex Deucher [Thu, 14 May 2015 16:47:45 +0000 (12:47 -0400)]
drm/radeon: make dpcd parameters const

Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
9 years agodrm/radeon: Use DECLARE_BITMAP
Joe Perches [Wed, 20 May 2015 01:37:52 +0000 (18:37 -0700)]
drm/radeon: Use DECLARE_BITMAP

Use the generic mechanism to declare a bitmap instead of unsigned long.

Signed-off-by: Joe Perches <joe@perches.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
9 years agodrm/radeon/tn/si: enable/disable vce cg when encoding v2
Alex Deucher [Mon, 11 May 2015 20:01:55 +0000 (22:01 +0200)]
drm/radeon/tn/si: enable/disable vce cg when encoding v2

Some of the vce clocks are automatic, others need to
be manually enabled.  For ease, just disable cg when
vce is active.

v2: rebased, call vce_v1_0_enable_mgcg directly

Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
9 years agodrm/radeon: add support for vce 1.0 clock gating
Alex Deucher [Mon, 11 May 2015 20:01:54 +0000 (22:01 +0200)]
drm/radeon: add support for vce 1.0 clock gating

Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
9 years agodrm/radeon: add VCE 1.0 support v4
Christian König [Mon, 11 May 2015 20:01:53 +0000 (22:01 +0200)]
drm/radeon: add VCE 1.0 support v4

Initial support for VCE 1.0 using newest firmware.

v2: rebased
v3: fix for TN
v4: fix FW size calculation

Signed-off-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
9 years agodrm/radeon/dpm: add vce support for SI
Alex Deucher [Mon, 11 May 2015 20:01:52 +0000 (22:01 +0200)]
drm/radeon/dpm: add vce support for SI

Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
9 years agodrm/radeon/dpm: add vce dpm support for TN
Alex Deucher [Mon, 11 May 2015 20:01:51 +0000 (22:01 +0200)]
drm/radeon/dpm: add vce dpm support for TN

Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
9 years agodrm/radeon: implement tn_set_vce_clocks
Alex Deucher [Mon, 11 May 2015 20:01:50 +0000 (22:01 +0200)]
drm/radeon: implement tn_set_vce_clocks

This implements the function to set the vce clocks
on TN hardware.

Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
9 years agodrm/radeon: implement si_set_vce_clocks v2
Christian König [Mon, 11 May 2015 20:01:49 +0000 (22:01 +0200)]
drm/radeon: implement si_set_vce_clocks v2

For setting clocks with VCE v1.0

v2: (chk) rebased on current tree

Signed-off-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
9 years agodrm/radeon: allow some more VCE firmware versions
Christian König [Mon, 11 May 2015 20:01:48 +0000 (22:01 +0200)]
drm/radeon: allow some more VCE firmware versions

They seem to work fine with the kernel interface.

Signed-off-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
9 years agodrm/radeon: rework VCE FW size calculation
Christian König [Mon, 11 May 2015 20:01:47 +0000 (22:01 +0200)]
drm/radeon: rework VCE FW size calculation

Previously we were completely over allocating, fix this
by actually implementing the size calculation.

Signed-off-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
9 years agodrm/radeon: add a GPU reset counter queryable by userspace
Marek Olšák [Wed, 29 Apr 2015 17:40:33 +0000 (19:40 +0200)]
drm/radeon: add a GPU reset counter queryable by userspace

Userspace will be able to tell whether a GPU reset occured by comparing
an old referece value of the counter with a new value.

Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Marek Olšák <marek.olsak@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
9 years agodrm/atomic: Add MODE_ID property
Daniel Stone [Mon, 25 May 2015 18:11:52 +0000 (19:11 +0100)]
drm/atomic: Add MODE_ID property

Atomic modesetting: now with modesetting support.

v2: Moved drm_atomic_set_mode_prop_for_crtc from previous patch; removed
    state->active fiddling, documented return code. Changed property
    type to DRM_MODE_PROP_BLOB.

Signed-off-by: Daniel Stone <daniels@collabora.com>
Tested-by: Sean Paul <seanpaul@chromium.org>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/atomic: Add current-mode blob to CRTC state
Daniel Stone [Mon, 25 May 2015 18:11:51 +0000 (19:11 +0100)]
drm/atomic: Add current-mode blob to CRTC state

Add a blob property tracking the current mode to the CRTC state, and
ensure it is properly updated and referenced.

v2: Continue using crtc_state->mode inside getcrtc, instead of reading
    out the mode blob. Use IS_ERR and PTR_ERR from create_blob. Move
    set_mode_prop_for_crtc to later patch where it actually gets used.
    Enforce !!state->enable == !!state->mode_blob inside
    drm_atomic_crtc_check.

Signed-off-by: Daniel Stone <daniels@collabora.com>
Tested-by: Sean Paul <seanpaul@chromium.org>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm: Add drm_atomic_set_mode_for_crtc
Daniel Stone [Tue, 26 May 2015 13:36:48 +0000 (14:36 +0100)]
drm: Add drm_atomic_set_mode_for_crtc

Add a new helper, to be used later for blob property management, that
sets the mode for a CRTC state, as well as updating the CRTC enable/active
state at the same time.

v2: Do not touch active/mode_changed in CRTC state. Document return
    value. Remove stray drm_atomic_set_mode_prop_for_crtc declaration.

v3: Remove i915 changes, and leave it directly bashing crtc_state->mode
    for the meantime.

Signed-off-by: Daniel Stone <daniels@collabora.com>
Tested-by: Sean Paul <seanpaul@chromium.org>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm: check for garbage in unused addfb2 fields
Daniel Vetter [Wed, 20 May 2015 14:53:53 +0000 (16:53 +0200)]
drm: check for garbage in unused addfb2 fields

Unfortunately old userspace didn't clear this properly, but since
we've added fb modifiers that's fixed. Checking properly that unused
fields is important for abi extensions, and just right now there's a
bunch of discussions going on about how exactly the additional aux
planes for render compression should be specified.

So let's first make sure that the values in those additional fields
can be indeed used.

Cc: Thierry Reding <thierry.reding@gmail.com>
Testcase: igt/kms_addfb/unused-*
Reviewed-by: Thierry Reding <treding@nvidia.com>
Signed-off-by: Daniel Vetter <daniel.vetter@intel.com>
9 years agodrm: Retain reference to blob properties in lookup
Daniel Stone [Mon, 25 May 2015 18:11:49 +0000 (19:11 +0100)]
drm: Retain reference to blob properties in lookup

When we look up a blob property, make sure we retain a reference to the
blob for the lifetime.

v2: Use DRM_MODE_PROP_BLOB, not PROP_OBJECT + OBJECT_BLOB.

Signed-off-by: Daniel Stone <daniels@collabora.com>
Tested-by: Sean Paul <seanpaul@chromium.org>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agoMerge branch 'drm/next/adv7511' of git://linuxtv.org/pinchartl/fbdev into drm-next
Dave Airlie [Tue, 26 May 2015 06:20:26 +0000 (16:20 +1000)]
Merge branch 'drm/next/adv7511' of git://linuxtv.org/pinchartl/fbdev into drm-next

adv7511 fix.

* 'drm/next/adv7511' of git://linuxtv.org/pinchartl/fbdev:
  drm: adv7511: Fix crash in IRQ handler when no encoder is associated

9 years agoMerge branch 'drm/next/rcar-du' of git://linuxtv.org/pinchartl/fbdev into drm-next
Dave Airlie [Tue, 26 May 2015 06:19:53 +0000 (16:19 +1000)]
Merge branch 'drm/next/rcar-du' of git://linuxtv.org/pinchartl/fbdev into drm-next

rcar-du fixes
* 'drm/next/rcar-du' of git://linuxtv.org/pinchartl/fbdev:
  drm: rcar-du: Split planes pre-association 4/4 between CRTCs
  drm: rcar-du: Store the number of CRTCs per group in the group structure
  drm: rcar-du: Consider plane to CRTC associations in the plane allocator
  drm: rcar-du: Keep plane to CRTC associations when disabling a plane
  drm: rcar-du: Add plane allocation debugging
  drm: rcar-du: Rename to_rcar_du_plane_state to to_rcar_plane_state
  drm: rcar-du: Embed rcar_du_planes structure into rcar_du_group
  drm: rcar-du: Move properties from rcar_du_planes to rcar_du_device
  drm: rcar-du: Document the rcar_du_plane_state structure
  drm: rcar-du: Document the rcar_du_crtc structure

9 years agoRevert "drm/i915: Force clean compilation with -Werror"
Daniel Vetter [Tue, 26 May 2015 05:46:21 +0000 (07:46 +0200)]
Revert "drm/i915: Force clean compilation with -Werror"

This reverts commit 118182e9d7d5afa0c7c10f568afb46ab78b462e9.

It's causing too much trouble when compile-testing for non-i915 folks.

Cc: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Jani Nikula <jani.nikula@intel.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Daniel Vetter <daniel.vetter@intel.com>
9 years agodrm: rcar-du: Split planes pre-association 4/4 between CRTCs
Laurent Pinchart [Tue, 28 Apr 2015 19:00:28 +0000 (22:00 +0300)]
drm: rcar-du: Split planes pre-association 4/4 between CRTCs

If we have more than one CRTCs in a group pre-associate planes 0-3 with
CRTC 0 and planes 4-7 with CRTC 1 to minimize flicker occurring when the
association is changed. The pre-association could be controlled by a
module parameter if needed.

Signed-off-by: Laurent Pinchart <laurent.pinchart+renesas@ideasonboard.com>
9 years agodrm: rcar-du: Store the number of CRTCs per group in the group structure
Laurent Pinchart [Tue, 28 Apr 2015 14:36:33 +0000 (17:36 +0300)]
drm: rcar-du: Store the number of CRTCs per group in the group structure

The number of CRTCs in a group is only used to implement plane
initialization for now, but is also needed to implement pre-association
of planes to CRTCs. Store it in the group structure instead of computing
it on demand.

Signed-off-by: Laurent Pinchart <laurent.pinchart+renesas@ideasonboard.com>
9 years agodrm: rcar-du: Consider plane to CRTC associations in the plane allocator
Laurent Pinchart [Tue, 28 Apr 2015 21:51:01 +0000 (00:51 +0300)]
drm: rcar-du: Consider plane to CRTC associations in the plane allocator

Hardware planes are driven by the timing generator of the CRTC they are
associated to. Changing the association requires restarting the CRTC
group that the plane belongs to, resulting in flicker on the other CRTC.

To avoid flicker as much as possible, try to allocate planes first from
the free planes already associated with the target CRTC. If allocation
fails then fall back to allocation from all free planes.

Signed-off-by: Laurent Pinchart <laurent.pinchart+renesas@ideasonboard.com>
9 years agodrm: rcar-du: Keep plane to CRTC associations when disabling a plane
Laurent Pinchart [Tue, 28 Apr 2015 15:01:45 +0000 (18:01 +0300)]
drm: rcar-du: Keep plane to CRTC associations when disabling a plane

Changing the plane to CRTC associations requires restarting the CRTC
group, creating visible flicker. Mitigate the issue by changing plane
association only when a plane becomes enabled, not when it get disabled.

Signed-off-by: Laurent Pinchart <laurent.pinchart+renesas@ideasonboard.com>
9 years agodrm: rcar-du: Add plane allocation debugging
Laurent Pinchart [Wed, 25 Feb 2015 20:31:52 +0000 (22:31 +0200)]
drm: rcar-du: Add plane allocation debugging

Plane allocation is a complex process, add debugging statements to help
finding out what could might wrong.

Signed-off-by: Laurent Pinchart <laurent.pinchart+renesas@ideasonboard.com>
9 years agodrm: rcar-du: Rename to_rcar_du_plane_state to to_rcar_plane_state
Laurent Pinchart [Tue, 28 Apr 2015 21:48:17 +0000 (00:48 +0300)]
drm: rcar-du: Rename to_rcar_du_plane_state to to_rcar_plane_state

All other cast functions are named without using "du", make the plane
state cast consistent with them.

Signed-off-by: Laurent Pinchart <laurent.pinchart+renesas@ideasonboard.com>
9 years agodrm: rcar-du: Embed rcar_du_planes structure into rcar_du_group
Laurent Pinchart [Tue, 28 Apr 2015 21:05:56 +0000 (00:05 +0300)]
drm: rcar-du: Embed rcar_du_planes structure into rcar_du_group

The rcar_du_planes structure contains a single field and is only
instantiated in the rcar_du_group structure. Embed it directly and
remove the rcar_du_planes structure.

Signed-off-by: Laurent Pinchart <laurent.pinchart+renesas@ideasonboard.com>
9 years agodrm: rcar-du: Move properties from rcar_du_planes to rcar_du_device
Laurent Pinchart [Tue, 28 Apr 2015 20:59:29 +0000 (23:59 +0300)]
drm: rcar-du: Move properties from rcar_du_planes to rcar_du_device

The plane property objects are instantiated once per CRTC group, while
they should be instantiated once globally for the device. Fix this and
move them to the rcar_du_device structure.

Signed-off-by: Laurent Pinchart <laurent.pinchart+renesas@ideasonboard.com>
9 years agodrm: rcar-du: Document the rcar_du_plane_state structure
Laurent Pinchart [Tue, 28 Apr 2015 21:12:40 +0000 (00:12 +0300)]
drm: rcar-du: Document the rcar_du_plane_state structure

Document the structure fields using kerneldoc.

Signed-off-by: Laurent Pinchart <laurent.pinchart+renesas@ideasonboard.com>
9 years agodrm: rcar-du: Document the rcar_du_crtc structure
Laurent Pinchart [Tue, 28 Apr 2015 22:11:57 +0000 (01:11 +0300)]
drm: rcar-du: Document the rcar_du_crtc structure

Document the structure fields using kerneldoc.

Signed-off-by: Laurent Pinchart <laurent.pinchart+renesas@ideasonboard.com>
9 years agodrm: adv7511: Fix crash in IRQ handler when no encoder is associated
Laurent Pinchart [Wed, 13 May 2015 21:31:07 +0000 (00:31 +0300)]
drm: adv7511: Fix crash in IRQ handler when no encoder is associated

The ADV7511 is probed before its slave encoder init function associates
it with an encoder. This creates a time window during which hot plug
detection interrupts can occur with an encoder, resulting in a crash in
the IRQ handler.

Fix this by ignoring hot plug detection IRQs when no encoder is
associated yet.

Signed-off-by: Laurent Pinchart <laurent.pinchart+renesas@ideasonboard.com>
Acked-by: Lars-Peter Clausen <lars@metafoo.de>
9 years agodrm/i915: Update DRIVER_DATE to 20150522
Daniel Vetter [Fri, 22 May 2015 17:45:27 +0000 (19:45 +0200)]
drm/i915: Update DRIVER_DATE to 20150522

Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/mode: Add user blob-creation ioctl
Daniel Stone [Fri, 22 May 2015 12:34:51 +0000 (13:34 +0100)]
drm/mode: Add user blob-creation ioctl

Add an ioctl which allows users to create blob properties from supplied
data. Currently this only supports modes, creating a drm_display_mode from
the userspace drm_mode_modeinfo.

v2: Removed size/type checks.
    Rebased on new patches to allow error propagation from create_blob,
    as well as avoiding double-allocation.

Signed-off-by: Daniel Stone <daniels@collabora.com>
Reviewed-by: Maarten Lankhorst <maarten.lankhorst@intel.com>
Tested-by: Sean Paul <seanpaul@chromium.org>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm: Return error value from blob creation
Daniel Stone [Fri, 22 May 2015 12:34:50 +0000 (13:34 +0100)]
drm: Return error value from blob creation

Change drm_property_create_blob to return an ERR_PTR-encoded error on
failure, so we can pass the failure reason down.

Signed-off-by: Daniel Stone <daniels@collabora.com>
Cc: Maarten Lankhorst <maarten.lankhorst@intel.com>
Tested-by: Sean Paul <seanpaul@chromium.org>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm: Allow creating blob properties without copy
Daniel Stone [Fri, 22 May 2015 12:34:49 +0000 (13:34 +0100)]
drm: Allow creating blob properties without copy

Make the data parameter to drm_property_create_blob optional; if
omitted, the copy will be skipped and the data will be empty.

Signed-off-by: Daniel Stone <daniels@collabora.com>
Reviewed-by: Maarten Lankhorst <maarten.lankhorst@intel.com>
Tested-by: Sean Paul <seanpaul@chromium.org>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/mode: Unstatic kernel-userspace mode conversion
Daniel Stone [Fri, 22 May 2015 12:34:48 +0000 (13:34 +0100)]
drm/mode: Unstatic kernel-userspace mode conversion

Move the drm_display_mode <-> drm_mode_modeinfo conversion functions
from drm_crtc.c to drm_modes.c, and make them non-static so that others
can use them.

Signed-off-by: Daniel Stone <daniels@collabora.com>
Tested-by: Sean Paul <seanpaul@chromium.org>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/mode: Validate modes inside drm_crtc_convert_umode
Daniel Stone [Fri, 22 May 2015 12:34:47 +0000 (13:34 +0100)]
drm/mode: Validate modes inside drm_crtc_convert_umode

The only user of convert_umode was also performing mode validation, so
do that in the same place.

Signed-off-by: Daniel Stone <daniels@collabora.com>
Tested-by: Sean Paul <seanpaul@chromium.org>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/crtc_helper: Replace open-coded CRTC state helpers
Daniel Stone [Fri, 22 May 2015 12:34:45 +0000 (13:34 +0100)]
drm/crtc_helper: Replace open-coded CRTC state helpers

Rather than open-coding our own CRTC state helpers, use the atomic helpers
added in f5e7840b0c, and make our freeing behaviour consistent as well.

Signed-off-by: Daniel Stone <daniels@collabora.com>
Tested-by: Sean Paul <seanpaul@chromium.org>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm: kerneldoc fixes for blob properties
Daniel Stone [Fri, 22 May 2015 12:34:44 +0000 (13:34 +0100)]
drm: kerneldoc fixes for blob properties

Change '@param foo' to '@foo:' to fit kerneldoc style.

672cb1d6ae mistakenly added an extra parameter to the kerneldoc for
drm_property_unreference_blob which wasn't actually present.

Signed-off-by: Daniel Stone <daniels@collabora.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/i915: Introduce DRM_I915_THROTTLE_JIFFIES
Chris Wilson [Thu, 21 May 2015 20:01:48 +0000 (21:01 +0100)]
drm/i915: Introduce DRM_I915_THROTTLE_JIFFIES

As Daniel commented on

commit b7ffe1362c5f468b853223acc9268804aa92afc8
Author: Chris Wilson <chris@chris-wilson.co.uk>
Date:   Mon Apr 27 13:41:24 2015 +0100

    drm/i915: Free RPS boosts for all laggards

it is better to be explicit when sharing hardcoded values such as
throttle/boost timeouts. Make it so!

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/i915: Use the correct destructor for freeing requests on error
Chris Wilson [Thu, 21 May 2015 20:01:45 +0000 (21:01 +0100)]
drm/i915: Use the correct destructor for freeing requests on error

After allocating from the slab cache, we then need to free the request
back into the slab cache upon error (and not call kfree as that leads
to eventual memory corruption).

Fixes regression from
commit efab6d8dd158fdccbe6a030f89fbf9ca0a9564e4
Author: Chris Wilson <chris@chris-wilson.co.uk>
Date:   Tue Apr 7 16:20:57 2015 +0100

    drm/i915: Use a separate slab for requests

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/i915/skl: don't fail colorkey + scaler request
Chandra Konduru [Mon, 18 May 2015 23:18:44 +0000 (16:18 -0700)]
drm/i915/skl: don't fail colorkey + scaler request

There is a mplayer video failure reported with xv.
This is because there is a request to do both plane scaling
and colorkey. Because skl hw doesn't support plane scaling
and colorkey at the same time, request is failed which is expected
behavior.

To make xv operate, this patch allows colorkey continue to work
without using scaler. Then behavior would be similar to platforms
without plane scaler support.

Signed-off-by: Chandra Konduru <chandra.konduru@intel.com>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=90449
[danvet: change can_scale to bool as requested by Ville.]
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/i915: Enable GTT caching on gen8
Ville Syrjälä [Tue, 19 May 2015 17:32:57 +0000 (20:32 +0300)]
drm/i915: Enable GTT caching on gen8

GTT caching was disabled by default on gen8 due to not working with
big pages. Some information suggests that it got fixed, but still
GTT caching has been left disabled by default. Or could be it just
meant that the default was changed to off, and hence the problem
got solved.

Enable GTT caching in the hopes of some performance increase.
Whether or not the big pages issue has been fixed is irrelevant
at this stage since we don't use big pages.

This gives me a 1-2% improvement in xonotic on my BSW. Haven't tried
BDW, but supposedly it has larger TLBs so might not benefit as much.
On HSW GTT caching is enabled by default.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/i915: Move WaProgramL3SqcReg1Default:bdw to init_clock_gating()
Ville Syrjälä [Tue, 19 May 2015 17:32:56 +0000 (20:32 +0300)]
drm/i915: Move WaProgramL3SqcReg1Default:bdw to init_clock_gating()

GEN8_L3SQCREG1 isn't saved in the context (verified by going through
a context dump), and so we shouldn't be using the ring w/a code to
initialize it. Also Bspec explicitly talks about MMIO and writing it
with the CPU.

Additionally there's another w/a WaTempDisableDOPClkGating:bdw which
tells us to disable DOP clock gating around the GEN8_L3SQCREG1 write
to make sure everyone notices the change. So let's do that as well.

Cc: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/i915: Use ilk_init_lp_watermarks() on BDW
Ville Syrjälä [Tue, 19 May 2015 17:32:55 +0000 (20:32 +0300)]
drm/i915: Use ilk_init_lp_watermarks() on BDW

We're not using ilk_init_lp_watermarks() on BDW for some reason.
Probably due to the BDW patches and the relevant WM patches landing
roughlly at the same time. Fix it up.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/i915: Disable FDI RX/TX before the ports
Ville Syrjälä [Tue, 5 May 2015 14:17:38 +0000 (17:17 +0300)]
drm/i915: Disable FDI RX/TX before the ports

Bspec says we should disable the FDI RX/TX before disabling the PCH
ports. Do so.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/i915: Disable CRT port after pipe on PCH platforms
Ville Syrjälä [Tue, 5 May 2015 14:17:37 +0000 (17:17 +0300)]
drm/i915: Disable CRT port after pipe on PCH platforms

Follow the BSpec sequence for the CRT port as well on PCH platforms,
ie. disable the pipe before the port.

Didn't bother looking at DDI in detail yet, so leave that one be even
though the CRT is a PCH port there.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/i915: Disable SDVO port after the pipe on PCH platforms
Ville Syrjälä [Tue, 5 May 2015 14:17:36 +0000 (17:17 +0300)]
drm/i915: Disable SDVO port after the pipe on PCH platforms

While at it also remove the redundant/unneeded w/a like done for hdmi
already.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Jesse Barnes <jbarnes@virtuousgeek.org>
[danvet: Mention that this also removes the unneeded w/a, as suggested
by Jesse.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/i915: Disable HDMI port after the pipe on PCH platforms
Ville Syrjälä [Tue, 5 May 2015 14:17:35 +0000 (17:17 +0300)]
drm/i915: Disable HDMI port after the pipe on PCH platforms

BSpec says we should disable all ports after the pipe on PCH
platforms. Do so. Fixes a pipe off timeout on ILK now caused by
the transcoder B workaround.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/i915: Fix the IBX transcoder B workarounds
Ville Syrjälä [Tue, 5 May 2015 14:17:34 +0000 (17:17 +0300)]
drm/i915: Fix the IBX transcoder B workarounds

Currently the IBX transcoder B workarounds are not working correctly.
Well, the HDMI one seems to be working somewhat, but the DP one is
definitely busted.

After a bit of experimentation it looks like the best way to make this
work is first disable the port on transcoder B, and then re-enable it
transcoder A, and immediately disable it again.

We can also clean up the code by noting that we can't be called without
a valid crtc. And also note that port A on ILK does not need the
workaround, so let's check for that one too.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/i915: Write the SDVO reg twice on IBX
Ville Syrjälä [Tue, 5 May 2015 14:17:33 +0000 (17:17 +0300)]
drm/i915: Write the SDVO reg twice on IBX

On IBX the SDVO/HDMI register write may be masked when enabling the
port, so it may need to written twice. The HDMI code does this, but
the SDVO code does not. Add the workaround to the SDVO code as well.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/i915: Fix DP enhanced framing for CPT
Ville Syrjälä [Tue, 5 May 2015 14:17:31 +0000 (17:17 +0300)]
drm/i915: Fix DP enhanced framing for CPT

Currently we're always enabling enhanced framing on CPT even if the sink
doesn't support it. Fix this up by actaully looking at what the sink
tells us.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/i915: Clean up the CPT DP .get_hw_state() port readout
Ville Syrjälä [Tue, 5 May 2015 14:17:30 +0000 (17:17 +0300)]
drm/i915: Clean up the CPT DP .get_hw_state() port readout

Define a TRANS_DP_PIPE_TO_PORT() to make the CPT DP .get_hw_state()
pipe readout neater.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/i915: Clarfify the DP code platform checks
Ville Syrjälä [Tue, 5 May 2015 14:17:29 +0000 (17:17 +0300)]
drm/i915: Clarfify the DP code platform checks

intel_dp.c is a mess with all the checks for different
platform/PCH variants and ports. Try to clean it up by recognizing
the following facts:
- IVB port A, and CPT port B/C/D are always the special cases
- VLV/CHV don't have port A
- Using the same kind of logic everywhere makes things much easier to
  parse

So let's move the IVB port A and PCH port B/C/D checks to be done first,
and let the other cases fall through, and always check for these things
using the same logic.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/i915: Remove the double register write from intel_disable_hdmi()
Ville Syrjälä [Tue, 5 May 2015 14:17:28 +0000 (17:17 +0300)]
drm/i915: Remove the double register write from intel_disable_hdmi()

IBX can have problems with the first write to the port register getting
masked when enabling the port. We are trying to apply the workaround
also when disabling the port where it's not needed, and we also try
to apply it for CPT/PPT as well which don't need it. Just kill it.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Jesse Barnes <jbarnes@virtuousgeek.org>
[danvet: Resolve conflict with the remove CHV if block.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/i915: Remove a bogus 12bpc "toggle" from intel_disable_hdmi()
Ville Syrjälä [Tue, 5 May 2015 14:17:27 +0000 (17:17 +0300)]
drm/i915: Remove a bogus 12bpc "toggle" from intel_disable_hdmi()

The IBX 12bpc port enable toggle is only relevant when enabling
the port, not when disabling it. Also this code doesn't actually
toggle anything, and essentially just writes the port register
one extra time. Furthermore CPT/PPT don't need such workarounds
and yet we include them. Just kill it.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/i915/skl: Deinit/init the display at suspend/resume
Damien Lespiau [Thu, 21 May 2015 15:37:48 +0000 (16:37 +0100)]
drm/i915/skl: Deinit/init the display at suspend/resume

We need to re-init the display hardware when going out of suspend. This
includes:

  - Hooking the PCH to the reset logic
  - Restoring CDCDLK
  - Enabling the DDB power

Among those, only the CDCDLK one is a bit tricky. There's some
complexity in that:

  - DPLL0 (which is the source for CDCLK) has two VCOs, each with a set
    of supported frequencies. As eDP also uses DPLL0 for its link rate,
    once DPLL0 is on, we restrict the possible eDP link rates the chosen
    VCO.
  - CDCLK also limits the bandwidth available to push pixels.

So, as a first step, this commit restore what the BIOS set, until I can
do more testing.

In case that's of interest for the reviewer, I've unit tested the
function that derives the decimal frequency field:

  #include <stdio.h>
  #include <stdint.h>
  #include <assert.h>

  #define ARRAY_SIZE(x) (sizeof(x) / sizeof(*(x)))

  static const struct dpll_freq {
          unsigned int freq;
          unsigned int decimal;
  } freqs[] = {
          { .freq = 308570, .decimal = 0b01001100111},
          { .freq = 337500, .decimal = 0b01010100001},
          { .freq = 432000, .decimal = 0b01101011110},
          { .freq = 450000, .decimal = 0b01110000010},
          { .freq = 540000, .decimal = 0b10000110110},
          { .freq = 617140, .decimal = 0b10011010000},
          { .freq = 675000, .decimal = 0b10101000100},
  };

  static void intbits(unsigned int v)
  {
          int i;

          for(i = 10; i >= 0; i--)
                  putchar('0' + ((v >> i) & 1));
  }

  static unsigned int freq_decimal(unsigned int freq /* in kHz */)
  {
          return (freq - 1000) / 500;
  }

  static void test_freq(const struct dpll_freq *entry)
  {
          unsigned int decimal = freq_decimal(entry->freq);

          printf("freq: %d, expected: ", entry->freq);
          intbits(entry->decimal);
          printf(", got: ");
          intbits(decimal);
          putchar('\n');

          assert(decimal == entry->decimal);
  }

  int main(int argc, char **argv)
  {
          int i;

          for (i = 0; i < ARRAY_SIZE(freqs); i++)
                  test_freq(&freqs[i]);

          return 0;
  }

v2:
  - Rebase on top of -nightly
  - Use (freq - 1000) / 500 for the decimal frequency (Ville)
  - Fix setting the enable bit of HSW_NDE_RSTWRN_OPT (Ville)
  - Rename skl_display_{resume,suspend} to skl_{init,uninit}_cdclk to
    be consistent with the BXT code (Ville)
  - Store boot CDCLK in ddi_pll_init (Ville)
  - Merge dev_priv's skl_boot_cdclk into cdclk_freq
  - Use LCPLL_PLL_LOCK instead of (1 << 30) (Ville)
  - Replace various '0' by SKL_DPLL0 to be a bit more explicit that
    we're programming DPLL0
  - Busy poll the PCU before doing the frequency change. It takes about
    3/4 cycles, each separated by 10us, to get the ACK from the CPU
    (Ville)

v3:
  - Restore dev_priv->skl_boot_cdclk, leaving unification with
    dev_priv->cdclk_freq for a later patch (Daniel, Ville)

Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/i915: Free RPS boosts for all laggards
Chris Wilson [Mon, 27 Apr 2015 12:41:24 +0000 (13:41 +0100)]
drm/i915: Free RPS boosts for all laggards

If the client stalls on a congested request, chosen to be 20ms old to
match throttling, allow the client a free RPS boost.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
[danvet: s/rq/req/]
[danvet: s/0/NULL/ reported by 0-day build]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
9 years agodrm/i915: Don't downclock whilst we have clients waiting for GPU results
Chris Wilson [Mon, 27 Apr 2015 12:41:23 +0000 (13:41 +0100)]
drm/i915: Don't downclock whilst we have clients waiting for GPU results

If we have clients stalled waiting for requests, ignore the GPU if it
signals that it should downclock due to low load. This helps prevent
the automatic timeout from causing extremely long running batches from
taking even longer.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>