drm/exynos: stop using sgtable in page fault handler
Already struct exynos_drm_gem_buf has pages of the buffer when buffer is
created, so just can use pages in page fault handler, we don't have to
make sgtable of the buffer. But this needs to construct pages of the
buffer that is imported from dma-buf prime.
struct exynos_drm_encoder was justing wrapping struct drm_encoder, it had
only a drm_encoder member and the internal exynos_drm_encoders ops that
was directly mapped to the drm_encoder helper funcs.
So now exynos DRM uses struct drm_encoder directly, this removes
completely the struct exynos_drm_encoder.
v2: add empty .mode_fixup() and .mode_set() to DSI and DPI to avoid null
pointer.
Gustavo Padovan [Wed, 5 Aug 2015 23:24:20 +0000 (20:24 -0300)]
drm/exynos: remove exynos_drm_create_enc_conn()
This functions was just hiding the encoder and connector creation in
a way that was less clean than if we get rid of it. For example,
exynos_encoder ops had .create_connector() defined only because we were
handing off the encoder and connector creation to
exynos_drm_create_enc_conn(). Without this function we can directly call
the create_connector function internally in the code, without the need of
any vtable access.
It also does some refactoring in the code like creating a bind function
for dpi devices.
Gustavo Padovan [Wed, 5 Aug 2015 23:24:18 +0000 (20:24 -0300)]
drm/exynos: remove extra call to exynos_dp_commit()
exynos_dp_commit() was getting called twice by exynos encoder core, once
inside the .enable() call and another time by .commit() itself.
The remove of the second call caused the wake of a bug, the operations
orders inside exynos_dp_commit was wrong and we had to move
exynos_dp_start_video() to be the last operation in there.
Gustavo Padovan [Tue, 11 Aug 2015 08:38:06 +0000 (17:38 +0900)]
drm/exynos: remove struct exynos_drm_display
This struct was just representing encoder information, it was a member of
struct exynos_drm_encoder, so any code trying to access encoder data would
have to go through the encoder struct, get the display struct and then get
the data it want.
During this patchset we also realized that the only data
exynos_drm_encoder needs to store is the drm_encoder parent and the
exynos_drm_encoder_ops.
Gustavo Padovan [Wed, 5 Aug 2015 23:24:15 +0000 (20:24 -0300)]
drm/exynos: simplify calculation of possible CRTCs
All CRTCs can only be LCD, HDMI or VIDI, so basically all CRTCs will be a
possible CRTCs. This patch removes an extra function with switch that was
only checking if the CRTC type was one of those three above.
Gustavo Padovan [Wed, 5 Aug 2015 23:24:13 +0000 (20:24 -0300)]
drm/exynos: remove wrappers for phy_power_{on,off}
phy_power_on() and phy_power_off() already checks for NULL pointer.
This patch removes the wrappers exynos_dp_phy_init() and
exynos_dp_phy_exit() since the only think they were doing was a check for
NULL phy.
drm/exynos: remove unused fields from struct exynos_drm_plane
Now after the move to use drm_plane_state directly struct drm_plane_state
has many unused fields, along with others that weren't used before the
plane state change. Thus remove them all.
Gustavo Padovan [Mon, 3 Aug 2015 05:40:44 +0000 (14:40 +0900)]
drm/exynos: use drm atomic state directly
For some fields the use of struct exynos_drm_plane filled with data from
the plane state just creates a source of duplicated information and
overhead. Here we change the crtc drivers to access the plane state
directly simplifying the code by not relying on a exynos internal struct.
Gustavo Padovan [Mon, 3 Aug 2015 05:39:36 +0000 (14:39 +0900)]
drm/exynos: pass struct exynos_drm_plane in update/enable
We already have the plane pointer in before calling .update_plane() or
disable_plane() so pass it directly to those calls avoiding a new
conversion from zpos to struct exynos_drm_plane.
v2: don't remove check for suspended in FIMD (comment by Joonyoung)
drm/exynos: clear channels only when iommu is enabled
This is simplest solution about reported problem[1]. It's no problem to
clear channel only when iommu is enabled, if we consider that we cannot
recognize iommu errors when iommu is disabled and it have been valid
until now. But this cannot be nice solution.
Already drm_iommu_attach_device checks whether support iommu internally.
It should clear channels always regardless iommu support. We didn't know
because we can detect the problem when iommu is enabled, so we don't
have to use drm_iommu_attach_device_if_possible and then we can remove
drm_iommu_attach_device_if_possible and clear_channels function pointer.
drm/exynos: move order to register vidi kms driver
The vidi is virtual kms driver and now it is registered earlier than
actual hw kms drivers, so it will occupy crtc index 0. Some users
assume the condition yet that actual hw kms driver has crtc index 0.
It may or may not be matter but let's arrange register order.
drm/exynos: remove unnecessary checking to support iommu
Already drm_iommu_attach_device and drm_iommu_detach_device check
whether support iommu internally, so we don't have to call
is_drm_iommu_supported before call them.
drm/exynos: remove to use ifdef CONFIG_ARM_DMA_USE_IOMMU
If CONFIG_ARM_DMA_USE_IOMMU is disable, CONFIG_DRM_EXYNOS_IOMMU also is
disable. When CONFIG_DRM_EXYNOS_IOMMU is disable,
is_drm_iommu_supported() returns always false, so we can remove to use
ifdef CONFIG_ARM_DMA_USE_IOMMU in is_drm_iommu_supported().
The patch removes redundant fields from hdmi_conf_regs. Their values
can be calculated from current_mode. This patch is the first step to remove
whole hdmi_conf_regs structure.
Andrzej Hajda [Thu, 9 Jul 2015 14:28:08 +0000 (16:28 +0200)]
drm/exynos/hdmi: remove private lock code
Most of the code is called by drm core framework, so it is already synchronized.
The only async function is irq routine which only calls drm framework so it
does not need to be synchronized.
Once pm_runtime_set_active() gets called, the kernel assumes that given
device has already enabled runtime pm and will call pm_runtime_suspend()
without matching pm_runtime_resume(). In case of DRM FIMC IPP driver,
this will result in calling clk_disable() without respective call to
clk_enable(). This patch removes call to pm_runtime_set_active() to
ensure that pm_runtime_suspend/resume calls will match.
Signed-off-by: Marek Szyprowski <m.szyprowski@samsung.com> Signed-off-by: Inki Dae <inki.dae@samsung.com>
drm/exynos: gsc: fix wrong bitwise operation for swap detection
The bits for rotation are not used as exclusively. So GSC_IN_ROT_270 can
not be used for swap detection. The definition of it is same with
GSC_IN_ROT_MASK. It is enough to check GSC_IN_ROT_90 bit is set or not to
check whether width / height size swapping is needed.
Dave Airlie [Fri, 14 Aug 2015 00:15:24 +0000 (10:15 +1000)]
Merge tag 'drm-amdkfd-next-fixes-2015-08-05' of git://people.freedesktop.org/~gabbayo/linux into drm-next
Two small bug fixes for the code you pulled for 4.3:
- Used a SHIFT define instead of a MASK define to check if a bit is turned on
when destroying hqd. Luckily, this is in gfx7 interface file with amdgpu,
which was used only for bring-up purposes of amdgpu, so no real effect on
a running system
- Used a logical AND instead of a bitwise AND operator, when initializing
sdma virtual memory when using SDMA queues
* tag 'drm-amdkfd-next-fixes-2015-08-05' of git://people.freedesktop.org/~gabbayo/linux:
drm/amdkfd: fix bug when initializing sdma vm
drm/amdgpu: fix bug when amdkfd destroys hqd
Dave Airlie [Fri, 14 Aug 2015 00:14:23 +0000 (10:14 +1000)]
Merge branch 'drm-sti-next-atomic-2015-08-11' of http://git.linaro.org/people/benjamin.gaignard/kernel into drm-next
This serie of patches fix minor bugs around how driver sub-components are
bind and planes z-ordering.
The main part is about atomic support: using more atomic helpers allow us
to simplify the code (~300 lines removed) and to ahve a better match between
drm concepts (planes and crtc) and hardware split.
[airlied: fixed up conflict in atomic code]
* 'drm-sti-next-atomic-2015-08-11' of http://git.linaro.org/people/benjamin.gaignard/kernel:
drm/sti: atomic crtc/plane update
drm/sti: rename files and functions
drm/sti: code clean up
drm/sti: fix dynamic z-ordering
drm: sti: fix sub-components bind
Dave Airlie [Fri, 14 Aug 2015 00:02:21 +0000 (10:02 +1000)]
Merge tag 'topic/drm-misc-2015-08-13' of git://anongit.freedesktop.org/drm-intel into drm-next
Final drm-misc pull for 4.3:
- fbdev emulation Kconfig option for everyone thanks to Archit. It's not
everything yet bit this is fairly tricky since it spawns all drivers.
- vgaarb & vgaswitcheroo polish from Thierry
- some drm_irq.c cleanups (Thierry)
- struct_mutex crusade from me
- more fbdev panic handling removal
- various things all over in drm core&helpers
* tag 'topic/drm-misc-2015-08-13' of git://anongit.freedesktop.org/drm-intel: (65 commits)
drm/atomic: Use KMS VBLANK API
drm/irq: Document return values more consistently
drm/irq: Make pipe unsigned and name consistent
drm/irq: Check for valid VBLANK before dereference
drm/irq: Remove negative CRTC index special-case
drm/plane: Remove redundant extern
drm/plane: Use consistent data types for format count
vga_switcheroo: Remove unnecessary checks
vga_switcheroo: Wrap overly long lines
vga_switcheroo: Use pr_fmt()
vga_switcheroo: Cleanup header comment
vga_switcheroo: Use pr_*() instead of printk()
vgaarb: Fix a few checkpatch errors and warnings
vgaarb: Use vgaarb: prefix consistently in messages
vgaarb: Stop complaining about absent devices
drm/atomic: fix null pointer access to mode_fixup callback
drm/i915: Use CONFIG_DRM_FBDEV_EMULATION
drm/core: Set mode to NULL when connectors in a set drops to 0.
drm/atomic: Call ww_acquire_done after check phase is complete
drm/atomic: Paper over locking WARN in default_state_clear
...
Thierry Reding [Wed, 12 Aug 2015 15:00:35 +0000 (17:00 +0200)]
drm/atomic: Use KMS VBLANK API
Instead of using the legacy VBLANK API, use the new KMS API. This is
part of an effort to convert all existing users so that the KMS API can
be changed to properly use per-CRTC data.
Signed-off-by: Thierry Reding <treding@nvidia.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Thierry Reding [Wed, 12 Aug 2015 15:00:31 +0000 (17:00 +0200)]
drm/irq: Make pipe unsigned and name consistent
Name all references to the pipe number (CRTC index) consistently to make
it easier to distinguish which is a pipe number and which is a pointer
to struct drm_crtc.
While at it also make all references to the pipe number unsigned because
there is no longer any reason why it should ever be negative.
Signed-off-by: Thierry Reding <treding@nvidia.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Thierry Reding [Wed, 12 Aug 2015 15:00:30 +0000 (17:00 +0200)]
drm/irq: Check for valid VBLANK before dereference
When accessing the array of per-CRTC VBLANK structures we must always
check that the index into the array is valid before dereferencing to
avoid crashing.
Signed-off-by: Thierry Reding <treding@nvidia.com>
[danvet: Squash in my own whitespace ocd fixup in drm_vblank_count.] Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Thierry Reding [Wed, 12 Aug 2015 15:00:24 +0000 (17:00 +0200)]
drm/irq: Remove negative CRTC index special-case
The drm_send_vblank_event() function treats negative CRTC indices as
meaning that a driver doesn't have proper VBLANK handling. This is the
only place where DRM needs negative CRTC indices, so in order to enable
subsequent cleanup, remove this special case and replace it by the more
obvious check for whether or not VBLANK support was initialized.
Signed-off-by: Thierry Reding <treding@nvidia.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Thierry Reding [Wed, 12 Aug 2015 14:32:11 +0000 (16:32 +0200)]
vga_switcheroo: Wrap overly long lines
Wrap overly long lines to make checkpatch happy. While at it, also add
blank lines after declarations to eliminate additional problems flagged
by checkpatch.
Signed-off-by: Thierry Reding <treding@nvidia.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Thierry Reding [Wed, 12 Aug 2015 14:32:10 +0000 (16:32 +0200)]
vga_switcheroo: Use pr_fmt()
Use pr_fmt() to define the "vga_switcheroo: " prefix that is prepended
to all output messages emitted by pr_*() functions. This allows making
existing strings much shorter and eliminates a bunch of warnings from
checkpatch about lines being overly long.
Signed-off-by: Thierry Reding <treding@nvidia.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Thierry Reding [Wed, 12 Aug 2015 14:32:07 +0000 (16:32 +0200)]
vgaarb: Fix a few checkpatch errors and warnings
Wrap overly long lines (offending lines were mostly comments, so trivial
to fix up) and a number of other coding style issues pointed out by the
checkpatch tool.
Signed-off-by: Thierry Reding <treding@nvidia.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Thierry Reding [Wed, 12 Aug 2015 14:32:05 +0000 (16:32 +0200)]
vgaarb: Stop complaining about absent devices
Some setups do not register a default VGA device, in which case the VGA
arbiter will still complain about the (non-existent) PCI device being a
non-VGA device.
Fix this by making the error message conditional on a default VGA device
having been set up. Note that the easy route of erroring out early isn't
going to work because otherwise priv->target won't be properly updated.
Signed-off-by: Thierry Reding <treding@nvidia.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Daniel Vetter [Mon, 10 Aug 2015 11:34:08 +0000 (13:34 +0200)]
drm/i915: Use CONFIG_DRM_FBDEV_EMULATION
Instead of our own duplicated one. This fixes a bug in the driver
unload code if DRM_FBDEV_EMULATION=n but DRM_I915_FBDEV=y because we
try to unregister the nonexistent fbdev drm_framebuffer.
drm/atomic: Only destroy connector states with connection mutex held
I tried to fix races of atomic commits against connector
hot-unplugging. The idea is to ensure lifetimes by holding the
connection_mutex long enough. That works for synchronous commits, but
not for async ones.
For async atomic commit we really need to fix up connector lifetimes
for real. But that's a much bigger task, so just add more duct-tape:
For cleaning up connector states we currently don't need the connector
itself. So NULL it out and remove the locking check. Of course that
check was to protect the entire sequence, but the modeset itself
should be save since currently DP MST hot-removal does a dpms-off. And
that should synchronize with any outstanding async atomic commit.
Or at least that's my hope, this is all a giant mess.
Daniel Vetter [Thu, 9 Jul 2015 21:32:46 +0000 (23:32 +0200)]
drm/qxl: Don't take dev->struct_mutex in bo_force_delete
It really doesn't protect anything which doesn't have other locks
already. It also doesn't seem to be wired up into the driver unload
code fwiw, but that's a different issue.
Reviewed-by: Thierry Reding <treding@nvidia.com> Signed-off-by: Daniel Vetter <daniel.vetter@intel.com>
Daniel Vetter [Thu, 9 Jul 2015 21:32:42 +0000 (23:32 +0200)]
drm/rockchip: Don't grab dev->struct_mutex for in mmap offset ioctl
Since David Herrmann's mmap vma manager rework we don't need to grab
dev->struct_mutex any more to prevent races when looking up the mmap
offset. Drop it and instead don't forget to use the unref_unlocked
variant (since the drm core still cares).
Aside: I stumbled over the mmap handler which directly does a
dma_mmap_attrs. But totally fails to grab a reference on the
underlying object and hence looks like it happily just leaks the ptes
since there's no guarantee the mmap isn't still around when
gem_free_object is called. Which the kerneldoc of dma_mmap_attrs
explicitly forbids.
v2: Fixup compile fail 0-day spotted.
Cc: Mark Yao <mark.yao@rock-chips.com> Reviewed-by: Thierry Reding <treding@nvidia.com> Signed-off-by: Daniel Vetter <daniel.vetter@intel.com>
Daniel Vetter [Thu, 9 Jul 2015 21:32:41 +0000 (23:32 +0200)]
drm/cma-helper: Don't grab dev->struct_mutex for in mmap offset ioctl
Since David Herrmann's mmap vma manager rework we don't need to grab
dev->struct_mutex any more to prevent races when looking up the mmap
offset. Drop it and instead don't forget to use the unref_unlocked
variant (since the drm core still cares).
Cc: Laurent Pinchart <laurent.pinchart@ideasonboard.com> Cc: Rob Clark <robdclark@gmail.com> Reviewed-by: Thierry Reding <treding@nvidia.com> Signed-off-by: Daniel Vetter <daniel.vetter@intel.com>
Daniel Vetter [Thu, 9 Jul 2015 21:32:40 +0000 (23:32 +0200)]
drm/cirrus: Don't grab dev->struct_mutex for in mmap offset ioctl
Since David Herrmann's mmap vma manager rework we don't need to grab
dev->struct_mutex any more to prevent races when looking up the mmap
offset. Drop it and instead don't forget to use the unref_unlocked
variant (since the drm core still cares).
Reviewed-by: Thierry Reding <treding@nvidia.com> Signed-off-by: Daniel Vetter <daniel.vetter@intel.com>
Daniel Vetter [Thu, 9 Jul 2015 21:32:39 +0000 (23:32 +0200)]
drm/mga200g: Hold a proper reference for cursor_set
Looking up an obj, immediate dropping the acquired reference and then
continuing to use it isn't how this is supposed to work. Fix this by
holding a reference for the entire function.
While at it stop grabbing dev->struct_mutex, it doesn't protect
anything here.
Reviewed-by: Thierry Reding <treding@nvidia.com> Signed-off-by: Daniel Vetter <daniel.vetter@intel.com>
Daniel Vetter [Thu, 9 Jul 2015 21:32:38 +0000 (23:32 +0200)]
drm/mga200g: Don't grab dev->struct_mutex for in mmap offset ioctl
Since David Herrmann's mmap vma manager rework we don't need to grab
dev->struct_mutex any more to prevent races when looking up the mmap
offset. Drop it and instead don't forget to use the unref_unlocked
variant (since the drm core still cares).
Reviewed-by: Thierry Reding <treding@nvidia.com> Signed-off-by: Daniel Vetter <daniel.vetter@intel.com>
Daniel Vetter [Thu, 9 Jul 2015 21:32:37 +0000 (23:32 +0200)]
drm/bochs: Don't grab dev->struct_mutex for in mmap offset ioctl
Since David Herrmann's mmap vma manager rework we don't need to grab
dev->struct_mutex any more to prevent races when looking up the mmap
offset. Drop it and instead don't forget to use the unref_unlocked
variant (since the drm core still cares).
Reviewed-by: Thierry Reding <treding@nvidia.com> Signed-off-by: Daniel Vetter <daniel.vetter@intel.com>
Daniel Vetter [Thu, 9 Jul 2015 21:32:36 +0000 (23:32 +0200)]
drm/ast: Don't grab dev->struct_mutex for in mmap offset ioctl
Since David Herrmann's mmap vma manager rework we don't need to grab
dev->struct_mutex any more to prevent races when looking up the mmap
offset. Drop it and instead don't forget to use the unref_unlocked
variant (since the drm core still cares).
Reviewed-by: Thierry Reding <treding@nvidia.com> Signed-off-by: Daniel Vetter <daniel.vetter@intel.com>
Archit Taneja [Wed, 5 Aug 2015 06:58:57 +0000 (12:28 +0530)]
drm: Add top level Kconfig option for DRM fbdev emulation
Legacy fbdev emulation support via DRM is achieved through KMS FB helpers.
Most modesetting drivers enable provide fbdev emulation by default by
selecting KMS FB helpers. A few provide a separate Kconfig option for the
user to enable or disbale fbdev emulation.
Enabling fbdev emulation is finally a distro-level decision. Having a top
level Kconfig option for fbdev emulation helps by providing a uniform way
to enable/disable fbdev emulation for any modesetting driver. It also lets
us remove unnecessary driver specific Kconfig options that causes bloat.
With a top level Kconfig in place, we can stub out the fb helper functions
when not needed without breaking functionality. Having stub functions also
prevents drivers to require wrapping fb helper function calls with #ifdefs.
DRM_FBDEV_EMULATION defaults to y since many drivers enable fbdev
emulation by default and majority of distributions expect the fbdev
interface in the kernel.
Signed-off-by: Archit Taneja <architt@codeaurora.org> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
drivers/gpu/drm/drm_fb_helper.c:390:13: warning: 'drm_fb_helper_force_kernel_mode' defined but not used [-Wunused-function]
static bool drm_fb_helper_force_kernel_mode(void)
^
Move drm_fb_helper_force_kernel_mode() inside the existing #ifdef to fix
this.
Signed-off-by: Geert Uytterhoeven <geert+renesas@glider.be> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
As of commit 5ea1f752ae04be40 ("drm: add
drm_fb_helper_restore_fbdev_mode_unlocked()"),
drm_fb_helper_restore_fbdev_mode() is no longer public, and drivers
should call drm_fb_helper_restore_fbdev_mode_unlocked() from their
->lastclose callbacks instead.
Update the documentation to reflect this, and absorb the one liner
drm_fb_helper_restore_fbdev_mode() into its single caller.
Signed-off-by: Geert Uytterhoeven <geert+renesas@glider.be> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Use the newly created wrapper drm_fb_helper functions instead of calling
core fbdev functions directly. They also simplify the fb_info creation.
v3:
- Don't touch remove_conflicting_framebuffers
v2:
- remove unused goto label 'out'
Cc: Daniel Vetter <daniel.vetter@ffwll.ch> Cc: David Airlie <airlied@linux.ie> Cc: Alex Deucher <alexander.deucher@amd.com> Signed-off-by: Archit Taneja <architt@codeaurora.org> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
we started checking the locking in drm_for_each_connector but somehow
I totally missed drm_mode_config_reset. There's no problem there since
this function should only be called in single-threaded contexts
(driver load or resume), so just wrap the loop with the right lock.
Daniel Vetter [Tue, 28 Jul 2015 11:18:41 +0000 (13:18 +0200)]
drm/fb-helper: Stop using trylocks in force_restore
Since the panic handling is gone this is only used for force-restoring
the fbdev/fbcon from sysrq, and that's done with a work item. No need
any more to do trylocks, we can just do normal locking.
Reviewed-by: Rob Clark <robdclark@gmail.com> Signed-off-by: Daniel Vetter <daniel.vetter@intel.com>
Daniel Vetter [Tue, 28 Jul 2015 11:18:40 +0000 (13:18 +0200)]
drm/fbdev: Return -EBUSY when oopsing
Trying to do anything with kms drivers when oopsing has become a
failing proposition. But since we can end up in the fbdev code simply
due to the console unblanking that's done unconditionally just
removing our panic handler isn't enough. We need to block all fbdev
callbacks when oopsing.
There was already one in the blank handler, but it failed silently.
That makes it impossible for drivers (like i915) who subclass these
functions to figure this out.
Instead consistently return -EBUSY so that everyone knows that we
really don't want to be bothered right now. This also allows us to
remove a pile of FIXMEs from the i915 fbdev code (since due to the
failure code they now won't attempt to grab dangerous locks any more).
Cc: Dave Airlie <airlied@gmail.com> Cc: Rodrigo Vivi <rodrigo.vivi@gmail.com> Reviewed-by: Rob Clark <robdclark@gmail.com> Signed-off-by: Daniel Vetter <daniel.vetter@intel.com>
projects / karo-tx-linux.git / log