Merge drm-fixes into drm-next.
Both i915 and radeon need this done for later patches.
Conflicts:
drivers/gpu/drm/drm_crtc_helper.c
drivers/gpu/drm/i915/i915_drv.h
drivers/gpu/drm/i915/i915_gem.c
drivers/gpu/drm/i915/i915_gem_execbuffer.c
drivers/gpu/drm/i915/i915_gem_gtt.c
This should avoid races between connector probing and HPD
irqs in the future, currently mode_config.mutex blocks this
possibility.
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Dave Airlie <airlied@redhat.com>
After the split-out of crtc locks from the big mode_config.mutex
there's still two major areas it protects:
- Various connector probe states, like connector->status, EDID
properties, probed mode lists and similar information.
- The links from connector->encoder and encoder->crtc and other
modeset-relevant connector state (e.g. properties which control the
panel fitter).
The later is used by modeset operations. But they don't really care
about the former since it's allowed to e.g. enable a disconnected VGA
output or with a mode not in the probed list.
Thus far this hasn't been a problem, but for the atomic modeset
conversion Rob Clark needs to convert all modeset relevant locks into
w/w locks. This is required because the order of acquisition is
determined by how userspace supplies the atomic modeset data. This has
run into troubles in the detect path since the i915 load detect code
needs _both_ protections offered by the mode_config.mutex: It updates
probe state and it needs to change the modeset configuration to enable
the temporary load detect pipe.
The big deal here is that for the probe/detect users of this lock a
plain mutex fits best, but for atomic modesets we really want a w/w
mutex. To fix this lets split out a new connection_mutex lock for the
modeset relevant parts.
For simplicity I've decided to only add one additional lock for all
connector/encoder links and modeset configuration states. We have
piles of different modeset objects in addition to those (like bridges
or panels), so adding per-object locks would be much more effort.
Also, we're guaranteed (at least for now) to do a full modeset if we
need to acquire this lock. Which means that fine-grained locking is
fairly irrelevant compared to the amount of time the full modeset will
take.
I've done a full audit, and there's just a few things that justify
special focus:
- Locking in drm_sysfs.c is almost completely absent. We should
sprinkle mode_config.connection_mutex over this file a bit, but
since it already lacks mode_config.mutex this patch wont make the
situation any worse. This is material for a follow-up patch.
- omap has a omap_framebuffer_flush function which walks the
connector->encoder->crtc links and is called from many contexts.
Some look like they don't acquire mode_config.mutex, so this is
already racy. Again fixing this is material for a separate patch.
- The radeon hot_plug function to retrain DP links looks at
connector->dpms. Currently this happens without any locking, so is
already racy. I think radeon_hotplug_work_func should gain
mutex_lock/unlock calls for the mode_config.connection_mutex.
- Same applies to i915's intel_dp_hot_plug. But again, this is already
racy.
- i915 load_detect code needs to acquire this lock. Which means the
w/w dance due to Rob's work will be nicely contained to _just_ this
function.
I've added fixme comments everywhere where it looks suspicious but in
the sysfs code. After a quick irc discussion with Dave Airlie it
sounds like the lack of locking in there is due to sysfs cleanup fun
at module unload.
v1: original (only compile tested)
v2: missing mutex_init(), etc (from Rob Clark)
v3: i915 needs more care in the conversion:
- Protect the edp pp logic with the connection_mutex.
- Use connection_mutex in the backlight code due to
get_pipe_from_connector.
- Use drm_modeset_lock_all in suspend/resume paths.
- Update lock checks in the overlay code.
Cc: Alex Deucher <alexdeucher@gmail.com>
Cc: Rob Clark <robdclark@gmail.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Reviewed-by: Rob Clark <robdclark@gmail.com>
Instead of trying to flip inside the vblank period when
the buffer is idle, offload blocking for idle to a kernel
thread and program the flip directly into the hardware.
v2: add error handling, fix EBUSY handling
v3: add proper exclusive_lock handling
v4: update crtc->primary->fb when the flip actually happens
Signed-off-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
DCE-4/5/6 can't support more than 12 bpc deep color over hdmi,
so clamp to 12 bpc when a hdmi deep color capable display is
connected. This even makes sense on DCE-8+, which could do up
to 16 bpc, as driving with more than 12 bpc would only waste
video bandwidth as long as we don't support framebuffers with
more than 12 bpc depth.
On pre-DCE4 we clamp hdmi bit depth to 8 bpc, as those asics
don't support hdmi deep color.
Signed-off-by: Mario Kleiner <mario.kleiner.de@gmail.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Program HDMI_CONTROL to send general control packets
for hdmi deep color mode signalling at every video
frame if bpc > 8.
This is only supported on evergreen / DCE-4 and later.
v2: rebase
Signed-off-by: Mario Kleiner <mario.kleiner.de@gmail.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Need to adjust the pll up for deep color modes.
Additionally, the atom bpc defines were wrong in certain
cases.
v2: set the adjusted clock to the pll clock for hdmi deep
color. This fixes display and audio issues with deep color
as reported by Andy Furniss <adf.lists@gmail.com>
v3: set crtc_clock as well
v4: setcrtcinfo on the adjusted mode
v5: just use the adjusted clock for setting the pll
v6: only use the adjusted clock for hdmi
v7: only DCE5 and DCE6 and bpc > 8
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
I'm not really sure how these should be calculated
for deep color. The hw generated values seem to work.
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Fill VM page tables from the GART page table if applicable.
v2: fix copy&paste error
Signed-off-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Fill VM page tables from the GART page table if applicable.
Signed-off-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Instead of trying to flip inside the vblank period when
the buffer is idle, offload blocking for idle to a kernel
thread and program the flip directly into the hardware.
v2: add error handling, fix EBUSY handling
v3: add proper exclusive_lock handling
Signed-off-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
We activate the VBLANK irq manually anyway, so this is unnecessary.
Signed-off-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
They are doing the same on all generations anyway.
Signed-off-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
We were checking the ext clock rather than the display clock.
Noticed by ArtForz on IRC.
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Cc: stable@vger.kernel.org
Removes useless quirk a7f465f73363fce409870f62173d518b1bc02ae6 introduced with
fdo #7770 as a failed attempt to minimize stability issues with hp zx1 chipset/
ATI FireGL X1 graphics adapter configuration
(see http://marc.info/?l=linux-ia64&m=140077543819871&w=2 for details/reason)
Signed-off-by: Émeric MASCHINO <emeric.maschino@gmail.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Thanks to advanced RE of fglrx we finally know what exactly needs to be
handled of AFMT change.
This has been tested for possible regressions on:
1) DCE2 HD2400 (RV610)
2) DCE3 HD3470 (RV620)
For a reference and details see:
https://bugzilla.kernel.org/show_bug.cgi?id=76231
Signed-off-by: Rafał Miłecki <zajec5@gmail.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Recent RE efforts revealed ops performed by fglrx during HDMI setup.
This mostly adds masks to r/w ops plus few single missing bits.
This has been tested for possible regressions on:
1) DCE2 HD2400 (RV610)
2) DCE3 HD3470 (RV620)
For a reference and details see:
https://bugzilla.kernel.org/show_bug.cgi?id=76231
Signed-off-by: Rafał Miłecki <zajec5@gmail.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
What initially seemed to be a typo in fglrx (using register 0x740c
instead of 0x74dc) appeared to be a correct behavior. DCE3 has ACR and
CRC registers swapped which explains why we needed
WREG32(HDMI0_AUDIO_CRC_CONTROL + offset, 0x1000);
This has been tested for possible regressions on DCE3 HD3470 (RV620).
Signed-off-by: Rafał Miłecki <zajec5@gmail.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
DCE 3.1 and 3.2 should be programmed in a different way than DCE 2 and
DCE 3. The order of setting registers and sets of registers are
different.
It's still unsure how we will handle DCE 3.1 vs. DCE 3.2, since they
have few differences as well.
For now separate DCE 2 and DCE 3 path, so we can work on it without a
risk of breaking DCE 3.1+.
This has been tested for possible regressions on DCE32 HD4550 (RV710).
Signed-off-by: Rafał Miłecki <zajec5@gmail.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
This patch makes it possible to decide how many address
bits are spend on the page directory vs the page tables.
v2: remove unintended change
Signed-off-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
This patch implements support for VRAM page table entry compression.
PTE construction is enhanced to identify physically contiguous page
ranges and mark them in the PTE fragment field. L1/L2 TLB support is
enabled for 64KB (SI/CIK) and 256KB (NI) PTE fragments, significantly
improving TLB utilization for VRAM allocations.
Linear store bandwidth is improved from 60GB/s to 125GB/s on Pitcairn.
Unigine Heaven 3.0 sees an average improvement from 24.7 to 27.7 FPS
on default settings at 1920x1200 resolution with vsync disabled.
See main comment in radeon_vm.c for a technical description.
v2 (chk): rebased and simplified.
v3 (chk): add missing hw setup
v4 (chk): rebased on current drm-fixes-3.15
v5 (chk): fix comments and commit text
Signed-off-by: Jay Cornwall <jay@jcornwall.me>
Signed-off-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
The i2c and aux buses use the same pads so add
a mutex to protect access to the pads.
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Christian König <christian.koenig@amd.com>
Setting the power state prior to restoring the display
hardware leads to blank screens on some systems. Drop
the power state set from dpm resume. The power state
will get set as part of the mode set sequence. Also
add an explicit power state set after mode set resume
to cover PX and headless systems.
bug:
https://bugzilla.kernel.org/show_bug.cgi?id=76761
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Cc: stable@vger.kernel.org
Signed-off-by: Dave Airlie <airlied@redhat.com>
So a few people complained that
commit 177cf92de4
Author: Daniel Vetter <daniel.vetter@ffwll.ch>
Date: Tue Apr 1 22:14:59 2014 +0200
drm/crtc-helpers: fix dpms on logic
which was merged into 3.15-rc1, broke resume on radeons. Strangely git
bisect lead everyone to
commit 25f397a429
Author: Daniel Vetter <daniel.vetter@ffwll.ch>
Date: Fri Jul 19 18:57:11 2013 +0200
drm/crtc-helper: explicit DPMS on after modeset
which was merged long ago and actually part of 3.14.
Digging deeper I've noticed (again) that the call to
drm_helper_resume_force_mode in the radeon resume handlers was a no-op
previously because everything gets shut down on suspend. radeon does
this with explicit calls to drm_helper_connector_dpms with DPMS_OFF.
But with 177c we now force the dpms state to ON, so suddenly
resume_force_mode actually forced the crtcs back on.
This is the intention of the change after all, the problem is that
radeon resumes the fbdev console layer _before_ restoring the display,
through calling fb_set_suspend. And fbcon does an immediate ->set_par,
which in turn causes the same forced mode restore to happen.
Two concurrent modeset operations didn't lead to happiness. Fix this
by delaying the fbcon resume until the end of the readeon resum
functions.
v2: Fix up a bit of the spelling fail.
References: https://lkml.org/lkml/2014/5/29/1043
References: https://lkml.org/lkml/2014/5/2/388
Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=74751
Tested-by: Ken Moffat <zarniwhoop@ntlworld.com>
Cc: Alex Deucher <alexdeucher@gmail.com>
Cc: Ken Moffat <zarniwhoop@ntlworld.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Dave Airlie <airlied@gmail.com>
It hangs the hardware.
Signed-off-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Cc: stable@vger.kernel.org
When the PX card is off don't try and access it. Avoid hw access
to the card while it's off (e.g., reading back invalid temperature).
v2: be less strict
bug:
https://bugzilla.kernel.org/show_bug.cgi?id=76321
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Cc: stable@vger.kernel.org
Signed-off-by: Christian König <christian.koenig@amd.com>
When accel is not working on device with virtual address space radeon
segfault because the ib buffer is NULL and trying to map it inside the
virtual address space trigger segfault. This patch only map the ib
buffer if accel is working.
Cc: <stable@vger.kernel.org>
Signed-off-by: Jérôme Glisse <jglisse@redhat.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Christian König <christian.koenig@amd.com>
Probably a copy paste typo.
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Christian König <christian.koenig@amd.com>
Cc: stable@vger.kernel.org
Some buffers (UVD/VM page tables) must be placed in VRAM,
but the byte restriction for moving buffers didn't took this
into account.
v2: keep closer to the original code
Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Take padding into account as well.
Fixes: https://bugzilla.kernel.org/show_bug.cgi?id=75651
Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Newer PX systems have non-VGA pci class dGPUs. Update
the ATRM fetch method to handle those cases.
bug:
https://bugzilla.kernel.org/show_bug.cgi?id=75401
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Christian König <christian.koenig@amd.com>
Cc: stable@vger.kernel.org
Mullins is DCE83 just like Kabini. Set the proper number
of endpoints on mullins.
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Christian König <christian.koenig@amd.com>
v2 (chk): fix image size storage
v3 (chk): fix UV size calculation
Signed-off-by: Leo Liu <leo.liu@amd.com>
Signed-off-by: Christian König <christian.koenig@amd.com>
