As the clflush operates on cache lines, and we can flush any byte
address, in order to flush all bytes given in the range we issue an
extra clflush on the last byte to ensure the last cacheline is flushed.
We can can the iteration to be over the actual cache lines to avoid this
double clflush on the last byte.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Imre Deak <imre.deak@intel.com>
Reviewed-by: Imre Deak <imre.deak@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
We use the same check already in the atomic core, so might as well
make this official. And it's also reused in e.g. i915.
Motivated by Maarten's idea to extract a connector_changed state out
of mode_changed.
Cc: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Reviewed-By: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@intel.com>
i915 fixes for stuff in next
* tag 'drm-intel-next-fixes-2015-06-18' of git://anongit.freedesktop.org/drm-intel:
drm/i915: Don't set enabled value of all CRTCs when restoring the mode
drm/i915: Don't update staged config during force restore modesets
drm/i915: Don't check modeset state in the hw state force restore path
drm/i915: Add SCRATCH1 and ROW_CHICKEN3 to the register whitelist.
drm/i915: Extend the parser to check register writes against a mask/value pair.
drm/i915: Fix command parser to validate multiple register access with the same command.
drm/i915: Don't skip request retirement if the active list is empty
one fix, one revert
* tag 'drm-intel-fixes-2015-06-18' of git://anongit.freedesktop.org/drm-intel:
Revert "drm/i915: Don't skip request retirement if the active list is empty"
drm/i915: Always reset vma->ggtt_view.pages cache on unbinding
This results in a nice cleanup, as we can replace the complicated logic
from should_ignore_backlight_request() with a simple check for the type
being native.
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Acked-by: Darren Hart <dvhart@linux.intel.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
If you do radeon.mst=1 on a gpu without mst hw, and then
plug some mst hw it will oops instead of falling back.
So check we have DCE5 at least before proceeding.
Signed-off-by: Dave Airlie <airlied@redhat.com>
Signed-off-by: Christian König <christian.koenig@amd.com>
This tells userspace that it's safe to use the RADEON_VA_UNMAP operation
of the DRM_RADEON_GEM_VA ioctl.
Cc: stable@vger.kernel.org
(NOTE: Backporting this commit requires at least backports of commits
26d4d129b6042197b4cbc8341c0618f99231af2f,
48afbd70ac7b6aa62e8d452091023941d8085f8a and
c29c0876ec05d51a93508a39b90b92c29ba6423d as well, otherwise using
RADEON_VA_UNMAP runs into trouble)
Signed-off-by: Michel Dänzer <michel.daenzer@amd.com>
Signed-off-by: Christian König <christian.koenig@amd.com>
The IPU code was installing its chained interrupt handler (which enables
the interrupt) before it was setting its data, which provokes an oops on
kexec. Fix this by converting to irq_set_chained_handler_and_data().
[drm] Initialized drm 1.1.0 20060810
imx-drm display-subsystem: parent device of /soc/aips-bus@02000000/ldb@020e0008/lvds-channel@1 is not available
imx-drm display-subsystem: parent device of /soc/aips-bus@02000000/ldb@020e0008/lvds-channel@1 is not available
Unable to handle kernel NULL pointer dereference at virtual address 00000070
pgd = c0004000
[00000070] *pgd=00000000
Internal error: Oops: 5 [#1] SMP ARM
Modules linked in:
CPU: 0 PID: 1 Comm: swapper/0 Not tainted 4.1.0-rc6+ #1693
Hardware name: Freescale i.MX6 Quad/DualLite (Device Tree)
task: d74c0000 ti: d74aa000 task.ti: d74aa000
PC is at ipu_irq_handle+0x28/0xd8
LR is at ipu_irq_handler+0x6c/0xc0
pc : [<c03c56d8>] lr : [<c03c58a4>] psr: 200001d3
sp : d74abbd0 ip : d74abc00 fp : d74abbfc
r10: 000001e0 r9 : c0085154 r8 : 00000009
r7 : 00000000 r6 : 00000000 r5 : d74abc04 r4 : c0a6b6a8
r3 : 00000000 r2 : 00000009 r1 : d74abc04 r0 : 00000000
Flags: nzCv IRQs off FIQs off Mode SVC_32 ISA ARM Segment kernel
Control: 10c5387d Table: 10004059 DAC: 00000015
Process swapper/0 (pid: 1, stack limit = 0xd74aa210)
Stack: (0xd74abbd0 to 0xd74ac000)
Backtrace:
[<c03c56b0>] (ipu_irq_handle) from [<c03c58a4>] (ipu_irq_handler+0x6c/0xc0)
[<c03c5838>] (ipu_irq_handler) from [<c0080154>] (generic_handle_irq+0x28/0x38)
[<c008012c>] (generic_handle_irq) from [<c0080288>] (__handle_domain_irq+0x5c/0xb8)
[<c008022c>] (__handle_domain_irq) from [<c0009428>] (gic_handle_irq+0x28/0x68)
[<c0009400>] (gic_handle_irq) from [<c0013dc4>] (__irq_svc+0x44/0x5c)
[<c07638fc>] (_raw_spin_unlock_irqrestore) from [<c00803bc>] (__irq_put_desc_unlock+0x1c/0x40)
[<c00803a0>] (__irq_put_desc_unlock) from [<c00841f4>] (__irq_set_handler+0x54/0x5c)
[<c00841a0>] (__irq_set_handler) from [<c03c5f48>] (ipu_probe+0x29c/0x708)
[<c03c5cac>] (ipu_probe) from [<c03d3848>] (platform_drv_probe+0x50/0xac)
[<c03d37f8>] (platform_drv_probe) from [<c03d1f3c>] (driver_probe_device+0x1d4/0x278)
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
Cc: Alexandre Courbot <gnurou@gmail.com>
Cc: Hans Ulli Kroll <ulli.kroll@googlemail.com>
Cc: Jason Cooper <jason@lakedaemon.net>
Cc: Lee Jones <lee.jones@linaro.org>
Cc: Linus Walleij <linus.walleij@linaro.org>
Cc: Thierry Reding <thierry.reding@gmail.com>
Cc: linux-arm-kernel@lists.infradead.org
Link: http://lkml.kernel.org/r/E1Z4z02-0002SI-Br@rmk-PC.arm.linux.org.uk
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
fixes and virtio-vga support.
* 'virtio-gpu-drm-next' of git://git.kraxel.org/linux:
virtio-gpu: add locking for vbuf pool
drm/virtgpu: initialise fbdev after getting initial display info
Add virtio-vga bits.
This contains fixes for the long-standing build issues that some of the
bridge drivers were exposing. Other than that it's mostly cleanup and a
couple of new simple panels that are supported.
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v2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=4rJg
-----END PGP SIGNATURE-----
Merge tag 'drm/panel/for-4.2-rc1' of git://anongit.freedesktop.org/tegra/linux into drm-next
drm/panel: Changes for v4.2-rc1
This contains fixes for the long-standing build issues that some of the
bridge drivers were exposing. Other than that it's mostly cleanup and a
couple of new simple panels that are supported.
* tag 'drm/panel/for-4.2-rc1' of git://anongit.freedesktop.org/tegra/linux:
drm/panel: simple: Add bus format for HannStar HSD100PXN1
drm/panel: simple: Add display timing for HannStar HSD100PXN1
drm/panel: ld9040: Remove useless padding
drm/panel: Constify OF match tables
drm/bridge: Remove stale ptn3460.h include
drm/bridge: ps8622: Include linux/gpio/consumer.h
drm/bridge: ptn3460: Include linux/gpio/consumer.h
drm/bridge: dw-hdmi: Return number of EDID modes
drm/panel: simple: Add support for LG LB070WV8 800x480 7" panel
drm/bridge: ptn3460: Pass flags to devm_gpiod_get()
drm/bridge: ps8622: Pass flags to devm_gpiod_get()
drm/bridge: ptn3460: Fix I2C ID table to match the reported modalias
drm/bridge: dw-hdmi: Staticize dw_hdmi_bridge_funcs
This contains a couple of mostly fixes for issues that have crept up in
recent versions of linux-next. One issue is that DP AUX transactions of
more than 4 bytes will access the wrong FIFO registers and hence become
corrupt. Another fix is required to restore functionality of Tegra20 if
using the GART. The current code expects the IOMMU aperture to be the
complete 4 GiB address space, whereas the GART on Tegra20 only provides
a 128 MiB aperture. One more issue with IOMMU support is that on 64-bit
ARM, swiotlb is the default IOMMU implementation backing the DMA API. A
side-effect of that is that when dma_map_sg() is called to flush caches
(yes, this is a bit of a hack, but ARM does not provide a better API),
swiotlb will immediately run out of memory because its bounce buffer is
too small to make a framebuffer.
Finally I've included a mostly cosmetic fix that stores register values
in u32 rather than unsigned long to avoid sign-extension issues on 64-
bit ARM. This is only a precaution since it hasn't caused any issues
(yet).
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v2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=3vlA
-----END PGP SIGNATURE-----
Merge tag 'drm/tegra/for-4.2-rc1' of git://anongit.freedesktop.org/tegra/linux into drm-next
drm/tegra: Changes for v4.2-rc1
This contains a couple of mostly fixes for issues that have crept up in
recent versions of linux-next. One issue is that DP AUX transactions of
more than 4 bytes will access the wrong FIFO registers and hence become
corrupt. Another fix is required to restore functionality of Tegra20 if
using the GART. The current code expects the IOMMU aperture to be the
complete 4 GiB address space, whereas the GART on Tegra20 only provides
a 128 MiB aperture. One more issue with IOMMU support is that on 64-bit
ARM, swiotlb is the default IOMMU implementation backing the DMA API. A
side-effect of that is that when dma_map_sg() is called to flush caches
(yes, this is a bit of a hack, but ARM does not provide a better API),
swiotlb will immediately run out of memory because its bounce buffer is
too small to make a framebuffer.
Finally I've included a mostly cosmetic fix that stores register values
in u32 rather than unsigned long to avoid sign-extension issues on 64-
bit ARM. This is only a precaution since it hasn't caused any issues
(yet).
* tag 'drm/tegra/for-4.2-rc1' of git://anongit.freedesktop.org/tegra/linux:
drm/tegra: dpaux: Registers are 32-bit
drm/tegra: gem: Flush pages after allocation
drm/tegra: gem: Take into account IOMMU aperture
drm/tegra: dpaux: Fix transfers larger than 4 bytes
rcar-du fixes
* 'drm/next/du' of git://linuxtv.org/pinchartl/fbdev:
drm: rcar-du: Use the drm atomic state duplication helpers for planes
drm: rcar-du: Clean up planes in the error paths of .atomic_commit()
drm: rcar-du: Convert rcar_du_encoders_init_one() return value to 0/<0
drm: rcar-du: Clarify error message when encoder initialization fails
drm: rcar-du: Fix crash with groups that have less than 9 planes
drm: rcar-du: Disable all planes when stopping the CRTC
drm: rcar-du: Print the error value when DRM/KMS init fails
The code in intel_crtc_restore_mode() sets the enabled value of all the
CRTCs when restoring the mode after a suspend/resume cycle. When more
than one CRTC is enabled, that causes drm_atomic_helper_check_modeset()
to fail if there is more than one pipe enabled, since all but one CRTC
has valid connector data. Instead, set only the enabled value for the
CRTC passed as an argument.
v2: Don't leak atomic state. (Matt)
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=90468
References: https://bugs.freedesktop.org/show_bug.cgi?id=90396
Signed-off-by: Ander Conselvan de Oliveira <ander.conselvan.de.oliveira@intel.com>
Reviewed-by: Matt Roper <matthew.d.roper@intel.com>
The force restore path relies on the staged config to preserve the
configuration used before a suspend/resume cycle. The update done to it
in intel_modeset_fixup_state() would cause that information to be lost
after the first modeset, making it impossible to restore the modes for
pipes B and C.
References: https://bugs.freedesktop.org/show_bug.cgi?id=90468
Signed-off-by: Ander Conselvan de Oliveira <ander.conselvan.de.oliveira@intel.com>
Reviewed-by: Matt Roper <matthew.d.roper@intel.com>
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
Since the force restore logic will restore the CRTCs state one at a
time, it is possible that the state will be inconsistent until the whole
operation finishes. A call to intel_modeset_check_state() is done once
it's over, so don't check the state multiple times in between. This
regression was introduced in:
commit 7f27126ea3db6ade886f18fd39caf0ff0cd1d37f
Author: Jesse Barnes <jbarnes@virtuousgeek.org>
Date: Wed Nov 5 14:26:06 2014 -0800
drm/i915: factor out compute_config from __intel_set_mode v3
v2: Rename check parameter to force_restore. (Matt)
Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=94431
Cc: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Ander Conselvan de Oliveira <ander.conselvan.de.oliveira@intel.com>
Reviewed-by: Matt Roper <matthew.d.roper@intel.com>
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
These files are built off of a tristate Kconfig option and also contain
modular function calls so they should explicitly include module.h to
avoid compile breakage during header shuffles done in the future.
Cc: David Airlie <airlied@linux.ie>
Cc: Mark Yao <mark.yao@rock-chips.com>
Cc: dri-devel@lists.freedesktop.org
Signed-off-by: Paul Gortmaker <paul.gortmaker@windriver.com>
buf_size_in_bytes must be large enough to hold ->num_watch_points and
watch_mode so I have added a sizeof(int) * 2 to the minimum size.
Also we have to subtract sizeof(*args) from the max args_idx limit so
that it matches the allocation. Also I changed a > to >= for the last
compare.
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
Ensure that the duplicate and destroy plane state operations will always
be in sync with the DRM core implementation of the plane state by using
the __drm_atomic_helper_plane_duplicate_state() and
__drm_atomic_helper_plane_destroy_state() functions designed especially
for this purpose.
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
When the .atomic_commit() handler fails, clean up planes previoulsy
prepared by drm_atomic_helper_prepare_planes() with a call to
drm_atomic_helper_cleanup_planes().
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
The function returns 1 on success, and either 0 or a negative error code
on failure. As the 0 and negative values don't need to be differentiated
by the caller, convert it to the usual scheme of returning 0 on success
and a negative error code on failure.
Signed-off-by: Laurent Pinchart <laurent.pinchart+renesas@ideasonboard.com>
A failure to initialize an encoder currently prints an error message in
the kernel log without mentioning which encoder failed to initialize. To
help debugging initialization issues print the encoder DT node name.
This requires moving the error message to the rcar_du_encoders_init_one
function and refactoring it slightly.
Signed-off-by: Laurent Pinchart <laurent.pinchart+renesas@ideasonboard.com>
Commit 917de180379d ("drm: rcar-du: Implement universal plane support")
made the number of planes per group dynamic, but didn't update all loops
over the planes array, resulting in out-of-bound accesses on DU
instances that have an odd number of CRTCs (such as the R8A7790). Fix
it.
Fixes: 917de180379d ("drm: rcar-du: Implement universal plane support")
Cc: stable@vger.kernel.org
Signed-off-by: Laurent Pinchart <laurent.pinchart+renesas@ideasonboard.com>
The DSnPR plane configuration registers are updated on vblank, and no
vblank will occur once the CRTC is stopped. We thus can't only disable
planes right before starting the CRTC as it would start scanning out
immediately from old frame buffers until the next vblank.
Fix the problem by disabling all planes when stopping the CRTC and wait
for the change to take effect. This increases the CRTC stop delay,
especially when multiple CRTCs are stopped in one operation as we now
wait for one vblank per CRTC. Whether this can be improved needs to be
researched.
Signed-off-by: Laurent Pinchart <laurent.pinchart+renesas@ideasonboard.com>
This should avoid issues with the fbdev path trying to render
before we've gotten the display info.
Signed-off-by: Dave Airlie <airlied@redhat.com>
[ kraxel: wait for display-info reply ]
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
Atomic modesetting support for omapdrm.
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1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=PlyA
-----END PGP SIGNATURE-----
Merge tag 'omapdrm-4.2-atomic' of git://git.kernel.org/pub/scm/linux/kernel/git/tomba/linux into drm-next
omapdrm atomic modesetting support
Atomic modesetting support for omapdrm.
" we've had issues with omapdrm for years,
which we've not been able to fix properly (like warnings/crashes when unloading
modules, page-flips tearing, race issues with fbs). All those problems seem to
be gone after this rewrite of omapdrm for atomic modesetting, and the resulting
code is much cleaner and more maintainable."
* tag 'omapdrm-4.2-atomic' of git://git.kernel.org/pub/scm/linux/kernel/git/tomba/linux: (43 commits)
drm: omapdrm: new vblank and event handling
drm: omapdrm: merge omap_crtc_flush and omap_crtc_atomic_flush
drm: omapdrm: add lock for fb pinning
drm: omapdrm: if omap_plane_atomic_update fails, disable plane
drm: omapdrm: inline omap_plane_setup into update/disable
drm: omapdrm: omap_plane_setup() cannot fail, use WARN
drm: omapdrm: Don't setup planes manually from CRTC .enable()/.disable()
drm: omapdrm: Don't flush CRTC when enabling or disabling it
drm: omapdrm: Move encoder setup to encoder operations
drm: omapdrm: Simplify DSS power management
drm: omapdrm: Remove nested PM get/sync when configuring encoders
drm: omapdrm: Support unlinking page flip events prematurely
drm: omapdrm: omap_crtc_flush() isn't called with modeset locked
drm: omapdrm: Don't get/put dispc in omap_crtc_flush()
drm: omapdrm: Make the omap_crtc_flush function static
drm: omapdrm: Remove omap_plane enabled field
drm: omapdrm: Remove omap_crtc enabled field
drm: omapdrm: Move crtc info out of the crtc structure
drm: omapdrm: Move plane info and win out of the plane structure
drm: omapdrm: Switch crtc and plane set_property to atomic helpers
...
Turns out 1366x768 does not in fact work on this hardware.
Signed-off-by: Adam Jackson <ajax@redhat.com>
Cc: stable@vger.kernel.org
Signed-off-by: Dave Airlie <airlied@redhat.com>
Only bit 27 of SCRATCH1 and bit 6 of ROW_CHICKEN3 are allowed to be
set because of security-sensitive bits we don't want userspace to mess
with. On HSW hardware the whitelisted bits control whether atomic
read-modify-write operations are performed on L3 or on GTI, and when
set to L3 (which can be 10x-30x better performing than on GTI,
depending on the application) require great care to avoid a system
hang, so we currently program them to be handled on GTI by default.
Beignet can immediately start taking advantage of this change to
enable L3 atomics. Mesa should eventually switch to L3 atomics too,
but a number of non-trivial changes are still required so it will
continue using GTI atomics for now.
Signed-off-by: Francisco Jerez <currojerez@riseup.net>
Reviewed-by: Zhigang Gong <zhigang.gong@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
In some cases it might be unnecessary or dangerous to give userspace
the right to write arbitrary values to some register, even though it
might be desirable to give it control of some of its bits. This patch
extends the register whitelist entries to contain a mask/value pair in
addition to the register offset. For registers with non-zero mask,
any LRM writes and LRI writes where the bits of the immediate given by
the mask don't match the specified value will be rejected.
This will be used in my next patch to grant userspace partial write
access to some sensitive registers.
Signed-off-by: Francisco Jerez <currojerez@riseup.net>
Reviewed-by: Zhigang Gong <zhigang.gong@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
Until now the software command checker assumed that commands could
read or write at most a single register per packet. This is not
necessarily the case, MI_LOAD_REGISTER_IMM expects a variable-length
list of offset/value pairs and writes them in sequence. The previous
code would only check whether the first entry was valid, effectively
allowing userspace to write unrestricted registers of the MMIO space
by sending a multi-register write with a legal first register, with
potential security implications on Gen6 and 7 hardware.
Fix it by extending the drm_i915_cmd_descriptor table to represent
multi-register access and making validate_cmd() iterate for all
register offsets present in the command packet.
Signed-off-by: Francisco Jerez <currojerez@riseup.net>
Reviewed-by: Zhigang Gong <zhigang.gong@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
This reverts commit 0aedb1626566efd72b369c01992ee7413c82a0c5.
I messed things up while applying [1] to drm-intel-fixes. Rectify.
[1] http://mid.gmane.org/1432827156-9605-1-git-send-email-ville.syrjala@linux.intel.com
Fixes: 0aedb1626566 ("drm/i915: Don't skip request retirement if the active list is empty")
Cc: stable@vger.kernel.org
Acked-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
Apparently we can have requests even if though the active list is empty,
so do the request retirement regardless of whether there's anything
on the active list.
The way it happened here is that during suspend intel_ring_idle()
notices the olr hanging around and then proceeds to get rid of it by
adding a request. However since there was nothing on the active lists
i915_gem_retire_requests() didn't clean those up, and so the idle work
never runs, and we leave the GPU "busy" during suspend resulting in a
WARN later.
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
With the introduction of multiple views of an obj in the same vm, each
vma was taught to cache its copy of the pages (so that different views
could have different page arrangements). However, this missed decoupling
those vma->ggtt_view.pages when the vma released its reference on the
obj->pages. As we don't always free the vma, this leads to a possible
scenario (e.g. execbuffer interrupted by the shrinker) where the vma
points to a stale obj->pages, and explodes.
Fixes regression from commit fe14d5f4e5468c5b80a24f1a64abcbe116143670
Author: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Date: Wed Dec 10 17:27:58 2014 +0000
drm/i915: Infrastructure for supporting different GGTT views per object
Tvrtko says, if someone else will be confused how this can happen, key
is the reservation execbuffer path. That puts the VMA on the exec_list
which prevents i915_vma_unbind and i915_gem_vma_destroy from fully
destroying the VMA. So the VMA is left existing as an empty object in
the list - unbound and disassociated with the backing store. Kind of a
cached memory object. And then re-using it needs to clear the cached
pages pointer which is fixed above.
Bugzilla: https://bugzilla.redhat.com/show_bug.cgi?id=1227892
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Cc: Daniel Vetter <daniel.vetter@ffwll.ch>
Cc: Michel Thierry <michel.thierry@intel.com>
Cc: stable@vger.kernel.org
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
[Jani: Added Tvrtko's explanation to commit message.]
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
Fix warnings related to size_t when building for 64-bit architectures:
drivers/gpu/drm/drm_gem_cma_helper.c: In function ‘drm_gem_cma_create’:
drivers/gpu/drm/drm_gem_cma_helper.c:114:4: warning: format ‘%d’ expects argument of type ‘int’, but argument 3 has type ‘size_t’ [-Wformat=]
size);
^
drivers/gpu/drm/drm_gem_cma_helper.c: In function ‘drm_gem_cma_describe’:
drivers/gpu/drm/drm_gem_cma_helper.c:393:4: warning: format ‘%d’ expects argument of type ‘int’, but argument 8 has type ‘size_t’ [-Wformat=]
off, &cma_obj->paddr, cma_obj->vaddr, obj->size);
Signed-off-by: Magnus Damm <damm+renesas@opensource.se>
Signed-off-by: Geert Uytterhoeven <geert+renesas@glider.be>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Rework the crtc event/flip_wait system as follows:
- If we enable a crtc (full modeset), we set omap_crtc->pending and
register vblank irq.
- If we need to set GO bit (page flip), we do the same but also set the
GO bit.
- On vblank we unregister the irq, clear the 'pending' flag, send vblank
event to userspace if crtc->state->event != NULL, and wake up
'pending_wait' wq.
- In omap_atomic_complete() we wait for the 'pending' flag to get reset
for all enabled crtcs using 'pending_wait' wq.
The above ensures that we send the events to userspace in vblank, and
that after the wait in omap_atomic_complete() everything for the
affected crtcs has been completed.
Signed-off-by: Tomi Valkeinen <tomi.valkeinen@ti.com>
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
omap_crtc_atomic_flush() is the only user of omap_crtc_flush(), so just
move the code from omap_crtc_flush() to omap_crtc_atomic_flush().
Signed-off-by: Tomi Valkeinen <tomi.valkeinen@ti.com>
Acked-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Before atomic modesetting omap_framebuffer_pin() and
omap_framebuffer_unpin() were always called with modesetting locks
taken. With atomic modesetting support this is no longer the case, and
we need locking to protect the pin_count and the paddr, as multiple
threads may pin the same fb concurrently.
This patch adds a mutex to struct omap_framebuffer, and uses it in
omap_framebuffer_pin() and omap_framebuffer_unpin().
Signed-off-by: Tomi Valkeinen <tomi.valkeinen@ti.com>
Acked-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
omap_plane_atomic_update() calls dispc_ovl_setup(), which can fail (but
shouldn't). To make the code a bit more robust, make sure the plane gets
disabled if dispc_ovl_setup() fails, as otherwise we might get illegal
HW configuration leading to error interrupts.
Signed-off-by: Tomi Valkeinen <tomi.valkeinen@ti.com>
Acked-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
At the moment we have omap_plane_setup() function which handles both
enabling (and configuring) and disabling the plane. With atomic
modesetting we have separate hooks for plane enable/config and disable.
This patch moves the code from omap_plane_setup() to
omap_plane_atomic_update() and omap_plane_atomic_disable().
Signed-off-by: Tomi Valkeinen <tomi.valkeinen@ti.com>
Acked-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
With atomic modesetting, omap_plane_setup()'s return value is ignored as
the functions using it cannot fail. dispc_ovl_setup(), called by
omap_plane_setup(), can fail (but shouldn't).
Instead of returning an error from omap_plane_setup() which gets
ignored, return void and use WARN if dispc_ovl_setup() fails.
Signed-off-by: Tomi Valkeinen <tomi.valkeinen@ti.com>
Acked-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Planes setup is handled by the DRM core through the atomic helpers,
there's no need to duplicate the code in the CRTC .enable() and
.disable() operations.
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Signed-off-by: Tomi Valkeinen <tomi.valkeinen@ti.com>
The omap_crtc_flush() call in omap_crtc_enable() and omap_crtc_disable()
is a no-op, as the display manager is always disabled at this point. Just
remove the function call.
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Signed-off-by: Tomi Valkeinen <tomi.valkeinen@ti.com>
Now that the driver is fully converted to atomic operations, and that
the atomic helpers call the operations in the right order, we can move
encoder setup to where it belongs, in the encoder operations.
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Signed-off-by: Tomi Valkeinen <tomi.valkeinen@ti.com>
Instead of sprinkling dispc_runtime_get() and dispc_runtime_put() calls
in various CRTC operations, move all power management code to the atomic
commit function.
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Signed-off-by: Tomi Valkeinen <tomi.valkeinen@ti.com>
The omap_crtc_encoder_setup() function is always called with the DSS
enabled. Remove the dispc_runtime_get() and dispc_runtime_put() calls.
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Signed-off-by: Tomi Valkeinen <tomi.valkeinen@ti.com>
DRM page flip vblank events requested by page flips or atomic commits
are created by the DRM core and then passed to driver through CRTC
states (for atomic commit) or directly to the page flip handler (for
legacy page flips). The events are then kept aside until the page flip
completes, at which point drivers queue them for delivery with a call to
drm_send_vblank_event().
When a DRM file handle is closed events pending for delivery are cleaned
up automatically by the DRM core. Events that have been passed to the
driver but haven't completed yet, however, are not handled by the DRM
core. Drivers are responsible for destroying them and must not attempt
to queue them for delivery. This is usually handled by drivers'
preclose() handlers that cancel and destroy page flip events that
reference the file handle being closed.
With asynchronous atomic updates the story becomes more complex. Several
asynchronous atomic updates can be pending, each of them carrying
per-CRTC events. As the atomic_commit() operation doesn't receive a file
handle context, drivers can't know which file handle a pending update
refers to, making it difficult to cancel or wait for completion of
updates related to the file handle being closed.
It should be noted that cancelling page flips or waiting for atomic
updates completion isn't required by the DRM core when closing a file
handle. The only requirement is that no event gets queued for delivery
after the preclose() operation returns. This can easily be achieved by
storing events for atomic commits in a list, unlinking events from the
file handle being closed by setting the file_priv field to NULL, and
skipping delivery of unlinked events.
This logic replaces the page flip cancellation completely.
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Signed-off-by: Tomi Valkeinen <tomi.valkeinen@ti.com>
