Pre DCE4 hw doesn't have reliable pageflip completion
interrupts, so instead polling for flip completion is
used from within the vblank irq handler to complete
page flips.
This causes a race if pageflip ioctl is called close to
vblank:
1. pageflip ioctl queues execution of radeon_flip_work_func.
2. vblank irq fires, radeon_crtc_handle_vblank checks for
flip_status == FLIP_SUBMITTED finds none, no-ops.
3. radeon_flip_work_func runs inside vblank, decides to
set flip_status == FLIP_SUBMITTED and programs the
flip into hw.
4. hw executes flip immediately (because in vblank), but
as 2 already happened, the flip completion routine only
emits the flip completion event one refresh later ->
wrong vblank count/timestamp for completion and no
performance gain, as instead of delaying the flip until
next vblank, we now delay the next flip by 1 refresh
while waiting for the delayed flip completion event.
Given we often don't gain anything due to this race, but
lose precision, prevent the programmed flip from executing
in vblank on pre DCE4 asics to avoid this race.
On pre-AVIVO hw we can't program the hw for edge-triggered
flips, they always execute anywhere in vblank. Therefore delay
the actual flip programming until after vblank on pre-AVIVO.
Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>
Signed-off-by: Mario Kleiner <mario.kleiner.de@gmail.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Pre DCE4 hardware doesn't have (reliable) pageflip completion
irqs, therefore we have to use the old polling method for flip
completion handling in vblank irq.
As vblank irqs fire a bit before start of vblank (when the
linebuffer fifo read position reaches end of scanout), we
have some fudge for flip completion handling in the last
lines of active scanout. Old code assumed the threshold to
be 99% of active scanout height, a ballpark estimate which
worked ok. Since we know since a while how to calculate the
actual threshold from linebuffer size, lets make use of it
to get a more accurate threshold.
This completion path is still prone to some races in corner
cases, especially on pre-AVIVO hardware, so document them
a bit better in the code comments.
Acked-by: Michel Dänzer <michel.daenzer@amd.com>
Signed-off-by: Mario Kleiner <mario.kleiner.de@gmail.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Before commit a325725633 ("drm: Lobotomize set_busid nonsense for !pci
drivers"), several DRM drivers for platform devices used to expose an
explicit "drm_driver.set_busid" callback, invariably backed by
drm_platform_set_busid().
Commit a325725633 removed drm_platform_set_busid(), along with the
referring .set_busid field initializations. This was justified because
interchangeable functionality had been implemented in drm_dev_alloc() /
drm_dev_init(), which DRM_IOCTL_SET_VERSION would rely on going forward.
However, commit a325725633 also removed drm_virtio_set_busid(), for
which the same consolidation was not appropriate: this .set_busid callback
had been implemented with drm_pci_set_busid(), and not
drm_platform_set_busid(). The error regressed Xorg/xserver on QEMU's
"virtio-vga" card; the drmGetBusid() function from libdrm would no longer
return stable PCI identifiers like "pci:0000:00:02.0", but rather unstable
platform ones like "virtio0".
Reinstate drm_virtio_set_busid() with judicious use of
git checkout -p a325725633c2^ -- drivers/gpu/drm/virtio
Cc: Daniel Vetter <daniel.vetter@intel.com>
Cc: Emil Velikov <emil.l.velikov@gmail.com>
Cc: Gerd Hoffmann <kraxel@redhat.com>
Cc: Gustavo Padovan <gustavo.padovan@collabora.co.uk>
Cc: Hans de Goede <hdegoede@redhat.com>
Cc: Joachim Frieben <jfrieben@hotmail.com>
Cc: stable@vger.kernel.org # v4.8
Reported-by: Joachim Frieben <jfrieben@hotmail.com>
Fixes: a325725633
Ref: https://bugzilla.redhat.com/show_bug.cgi?id=1366842
Signed-off-by: Laszlo Ersek <lersek@redhat.com>
Reviewed-by: Emil Velikov <emil.l.velikov@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
This pull request includes,
- Code refactoring on HDMI DDC and PHY.
- Regression fixup on deadlock issue with G2D pm integration.
- Fixup on page fault issue with wait_for_vblank mechianism specific to Exynos drm.
- And some cleanups.
* 'exynos-drm-next' of git://git.kernel.org/pub/scm/linux/kernel/git/daeinki/drm-exynos:
drm/exynos: g2d: simplify g2d_free_runqueue_node()
drm/exynos: g2d: use autosuspend mode for PM runtime
drm/exynos: g2d: wait for engine to finish
drm/exynos: g2d: remove runqueue nodes in g2d_{close,remove}()
drm/exynos: g2d: move PM management to runqueue worker
Revert "drm/exynos: g2d: fix system and runtime pm integration"
drm/exynos: use drm core to handle page-flip event
drm/exynos: mark exynos_dp_crtc_clock_enable() static
drm/exynos/fimd: add clock rate checking
drm/exynos: fix pending update handling
drm/exynos/vidi: use timer for vblanks instead of sleeping worker
drm/exynos: g2d: beautify probing message
drm/exynos: mixer: simplify loop in vp_win_reset()
drm/exynos: mixer: convert booleans to flags in mixer context
gpu: drm: exynos_hdmi: Remove duplicate initialization of regulator bulk consumer
gpu: drm: exynos_hdmi: Move PHY logic into single function
gpu: drm: exynos_hdmi: Move DDC logic into single function
The runqueue worker currently issues a get() when a new
node is processed, and a put() once a node is completed.
The corresponding suspend and resume calls currently only
do clock gating, but with the upcoming introduction of
IOMMU runpm also the corresponding IOMMU domain gets
enabled (for get()) and disabled (for put()). This
introduces performance regressions with we mitigate here.
Switch PM runtime to autosuspend, such that clock gating
and IOMMU control only happens when the engine is idle for
a 'long' time.
Signed-off-by: Tobias Jakobi <tjakobi@math.uni-bielefeld.de>
Acked-by: Marek Szyprowski <m.szyprowski@samsung.com>
Signed-off-by: Inki Dae <inki.dae@samsung.com>
While the engine works on a runqueue node it does memory access to
the buffers associated with that node.
Make sure that the engine is idle when g2d_close() and/or
g2d_remove() are called, i.e. buffer associated with the process (for
g2d_close()), or all buffers (for g2d_remove()) can be safely be
unmapped.
We have to take into account that the engine might be in an undefined
state, i.e. it hangs and doesn't become idle. In this case, we issue
a hardware reset to return the hardware and the driver context into a
proper state.
Signed-off-by: Tobias Jakobi <tjakobi@math.uni-bielefeld.de>
Acked-by: Marek Szyprowski <m.szyprowski@samsung.com>
Signed-off-by: Inki Dae <inki.dae@samsung.com>
The driver might be closed (and/or removed) while there are still
nodes queued for processing.
Make sure to remove these nodes, which means all of them in
the case of g2d_remove() and only those belonging to the
corresponding process in g2d_close().
Signed-off-by: Tobias Jakobi <tjakobi@math.uni-bielefeld.de>
Acked-by: Marek Szyprowski <m.szyprowski@samsung.com>
Signed-off-by: Inki Dae <inki.dae@samsung.com>
This reverts commit b05984e21a.
The change, i.e. merging the sleep and runpm operations, produces
a deadlock situation:
(1) exynos_g2d_exec_ioctl() prepares a runqueue node and
calls g2d_exec_runqueue()
(2) g2d_exec_runqueue() calls g2d_dma_start() which gets
runtime PM sync
(3) runtime PM core calls g2d_runtime_resume()
(4) g2d_runtime_resume() calls g2d_exec_runqueue(), which
loops back to (2)
Due to mutexes that are in place, a deadlock situation is created.
Signed-off-by: Tobias Jakobi <tjakobi@math.uni-bielefeld.de>
Acked-by: Marek Szyprowski <m.szyprowski@samsung.com>
Signed-off-by: Inki Dae <inki.dae@samsung.com>
Exynos DRM framework handled page-flip event with custom code.
The patch replaces it with drm-core vblank queue.
Signed-off-by: Andrzej Hajda <a.hajda@samsung.com>
Signed-off-by: Inki Dae <inki.dae@samsung.com>
We get 1 warning when building kernel with W=1:
drivers/gpu/drm/exynos/exynos_dp.c:46:5: warning: no previous prototype for 'exynos_dp_crtc_clock_enable' [-Wmissing-prototypes]
In fact, this function is only used in the file in which it is
declared and don't need a declaration, but can be made static.
So this patch marks it 'static'.
Signed-off-by: Baoyou Xie <baoyou.xie@linaro.org>
Signed-off-by: Inki Dae <inki.dae@samsung.com>
In case of some platforms fimd clocks can be configured to
very low values, as a result refresh rate can be very low and
driver/drm-core will timeout waiting for vblanks, it will result
in premature removal of framebuffers and will cause oopses.
The patch adds atomic_check callback to fimd to prevent setting
such modes.
Reported-by: Tobias Jakobi <tjakobi@math.uni-bielefeld.de>
Signed-off-by: Andrzej Hajda <a.hajda@samsung.com>
Signed-off-by: Inki Dae <inki.dae@samsung.com>
Exynos DRM devices update their registers at vblank time. Exynos-DRM uses
custom mechanism to wait for vblank. This mechanism is error prone -
variables are not updated atomically. As a result in certain circumstances
user space can try to free buffers which are still in use by hardware,
in such cases IOMMU can throw OOPS.
The patch instead of fixing the mechanism replaces it with drm core helper.
Signed-off-by: Andrzej Hajda <a.hajda@samsung.com>
Reviewed-by: Gustavo Padovan <gustavo.padovan@collabora.co.uk>
Signed-off-by: Inki Dae <inki.dae@samsung.com>
VIDI driver uses fake vblank handler to generate vblank events.
It was implemented using worker which slept for vblank time, additionally
it did not work if there were no page flips. The patch replaces it with
timer, uses drm_crtc_vblank_(on|off) helpers to manage it and fixes
behavior for non-page-flip cases.
This change allows further improvements of vblank in exynos-drm framework.
Signed-off-by: Andrzej Hajda <a.hajda@samsung.com>
Reviewed-by: Gustavo Padovan <gustavo.padovan@collabora.co.uk>
Signed-off-by: Inki Dae <inki.dae@samsung.com>
The mixer context struct already has a 'flags' field, so
we can use it to store the 'interlace', 'vp_enabled' and
'has_sclk' booleans.
We use the non-atomic helper functions to access these bits.
Signed-off-by: Tobias Jakobi <tjakobi@math.uni-bielefeld.de>
Reviewed-by: Andrzej Hajda <a.hajda@samsung.com>
Signed-off-by: Inki Dae <inki.dae@samsung.com>
The helper, devm_regulator_bulk_get() initializes the consumer as NULL,
so this code can be ignored.
Signed-off-by: Milo Kim <woogyom.kim@gmail.com>
Reviewed-by: Andrzej Hajda <a.hajda@samsung.com>
Signed-off-by: Inki Dae <inki.dae@samsung.com>
Some additional fixes for 4.9:
- The rest of Christian's GTT rework which fixes a long standing bug
in the GPUVM code among other things
- Changes to the pci shutdown callbacks for certain hypervisors
- Fix hpd interrupt storms on eDP panels which have the hpd interrupt
enabled by the bios
- misc cleanups and bug fixes
* 'drm-next-4.9' of git://people.freedesktop.org/~agd5f/linux: (33 commits)
drm/radeon: always apply pci shutdown callbacks
drm/amdgpu: always apply pci shutdown callbacks (v2)
drm/amdgpu: improve VM PTE trace points
drm/amdgpu: fix GART_DEBUGFS define
drm/amdgpu: free userptrs even if GTT isn't bound
drm/amd/amdgpu: Various cleanups for DCEv6
drm/amdgpu: fix BO move offsets
drm/amdgpu: fix amdgpu_move_blit on 32bit systems
drm/amdgpu: fix gtt_mgr bo's offset
drm/amdgpu: fix initializing the VM BO shadow
drm/amdgpu: fix initializing the VM last eviction counter
drm/amdgpu: cleanup VM shadow BO unreferencing
drm/amdgpu: allocate GTT space for shadow VM page tables
drm/amdgpu: rename all rbo variable to abo v2
drm/amdgpu: remove unused member from struct amdgpu_bo
drm/amdgpu: add a custom GTT memory manager v2
drm/amdgpu/dce6: disable hpd on local panels
drm/amdgpu/dce8: disable hpd on local panels
drm/amdgpu/dce11: disable hpd on local panels
drm/amdgpu/dce10: disable hpd on local panels
...
drm/tegra: Changes for v4.9-rc1
One bugfix that avoids overwriting the Y plane base address when
displaying buffers with one of the YUV/YVU formats.
* tag 'drm/tegra/for-4.9-rc1' of git://anongit.freedesktop.org/tegra/linux:
drm/tegra: Fix window[0] base address corruption
drm/panel: Changes for v4.9-rc1
Adds support for one more panel to the simple-panel driver, fixes up a
couple of delays and flags for existing panels and finally adds a new
driver for the DSI panel found on Nexus 7 devices.
* tag 'drm/panel/for-4.9-rc1' of git://anongit.freedesktop.org/tegra/linux:
drm/panel: Add JDI LT070ME05000 WUXGA DSI Panel
dt-bindings: Add JDI LT070ME05000 panel bindings
drm/panel: simple: Fix bus_format for the Olimex LCD-OLinuXino-4.3TS
drm/panel: simple-panel: Add delay timings for Starry KR122EA0SRA
drm/panel: simple: Fix bus flags for Ortustech com43h4m85ulc
drm/panel: simple: Add Innolux G101ICE-L01 panel
drm/panel: simple: Add delay timing for Sharp LQ123P1JX31
drm/dsi: Implement DCS set/get display brightness
drm/dsi: Order DCS helpers by command code
We can't properly detect all hypervisors and we
need this to properly tear down the hardware.
v2: trivial warning fix
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Just to cleanup some radeon leftovers.
sed -i "s/rbo/abo/g" drivers/gpu/drm/amd/amdgpu/*.c
sed -i "s/rbo/abo/g" drivers/gpu/drm/amd/amdgpu/*.h
v2: rebased
Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
bugfixes for qemu (bochs, qxl and virtio-gpu) drm drivers
* tag 'drm-qemu-20160921' of git://git.kraxel.org/linux:
drm/virtio: add real fence context and seqno
drm/virtio: drop virtio_gpu_execbuffer_ioctl() wrapping
virtio-gpu: avoid possible NULL pointer dereference
drm/qxl: reapply cursor after SetCrtc calls
bochs: ignore device if there isn't enougth memory
