The drm driver .load() operation is prone to race conditions as it
initializes the driver after registering the device nodes. Its usage is
deprecated, inline it in the probe function and call drm_dev_alloc() and
drm_dev_register() explicitly.
For consistency inline the .unload() handler in the remove function as
well.
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Reviewed-by: Tomi Valkeinen <tomi.valkeinen@ti.com>
Calling drm_vblank_cleanup() in drm_dev_unregister() causes issues with
drivers that have moved away from the .load() and .unload() midlayer.
Those drivers call drm_dev_unregister() as the first operation at unbind
time, before shutting down the device. This results in warnings due to
drm_vblank_cleanup() being called with vblank interrupts still active,
and then to vblank events being sent after cleanup.
Fix the problem by moving vblank cleanup from drm_dev_unregister() to
drm_dev_release() that is guaranteed to be called after drivers shut
down the device.
Suggested-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Tested-by: Tomi Valkeinen <tomi.valkeinen@ti.com>
Reviewed-by: Lucas Stach <l.stach@pengutronix.de>
Move the list of pending IRQ wait instances to the omap_drm_private
structure and the wait queue head to the IRQ wait structure.
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Reviewed-by: Tomi Valkeinen <tomi.valkeinen@ti.com>
Now that the IRQ list is used for IRQ wait only we can merge
omap_drm_irq and omap_irq_wait and simplify the implementation.
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Reviewed-by: Tomi Valkeinen <tomi.valkeinen@ti.com>
The function is only used in omap_irq.c and is just a wrapper around
dispc_mgr_get_vsync_irq(). Remove it and call the dispc function
directly.
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Reviewed-by: Tomi Valkeinen <tomi.valkeinen@ti.com>
The IRQ wait functions are called from the DSS enable and disable
operations only, where the DISPC is guaranteed to be enabled. There's no
need for manual DISPC power management there.
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Reviewed-by: Tomi Valkeinen <tomi.valkeinen@ti.com>
The only omap_drm_irq handler doesn't use the irqstatus parameter passed
to the function. Remove it.
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Reviewed-by: Tomi Valkeinen <tomi.valkeinen@ti.com>
The IRQ registration functions are not used outside of their compilation
unit, make them static. As the __omap_irq_(un)register() functions are
only called by their omap_irq_(un)register() counterparts, merge them
together.
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Reviewed-by: Tomi Valkeinen <tomi.valkeinen@ti.com>
Instead of going through a complicated private IRQ registration
mechanism, handle the vblank interrupt activation with the standard
drm_crtc_vblank_get() and drm_crtc_vblank_put() mechanism. This will let
the DRM core keep the vblank interrupt enabled as long as needed to
update the frame counter.
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Reviewed-by: Tomi Valkeinen <tomi.valkeinen@ti.com>
The CRTC pending flag will need to be accessed atomically in the vblank
interrupt handler, memory barriers won't be enough to protect it. Use a
spinlock instead.
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Reviewed-by: Tomi Valkeinen <tomi.valkeinen@ti.com>
The vblank interrupt is disabled after one occurrence, preventing the
atomic update event from being processed twice. However, this also
prevents the software frame counter from being updated correctly that
would require vblank interrupts to be kept enabled while the CRTC is
active.
In preparation for vblank interrupt fixes, make sure that the atomic
update event will be processed once only when the vblank interrupt will
be kept enabled.
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Reviewed-by: Tomi Valkeinen <tomi.valkeinen@ti.com>
The omapdrm DSS manager enable/disable operations check the DSS manager
state to avoid double enabling/disabling. Check the CRTC software state
instead to decrease the dependency of the DRM layer to the DSS layer.
The dispc_mgr_is_enabled() function then be turned into a static
function, but needs to be moved up in its compilation unit to avoid a
forward declaration.
Add a WARN_ON to catch double enable or disable that should be prevented
by the DRM core and would be a clear sign of a bug. The warning should
eventually be removed.
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Reviewed-by: Tomi Valkeinen <tomi.valkeinen@ti.com>
The DRM core supports skipping plane update for inactive CRTCs for
hardware that don't need it or can't cope with it. That's our case, and
the driver already skips flushing planes on inactice CRTCs.
We can't remove the check from the driver, as active CRTCs are disabled
at the hardware level when an atomic flush is performed if a mode set is
pending. There's however no need to forward the plane commit calls to
the driver, so use the DRM core infrastructure to skip them with a
detailed comment to explain why the check must still be kept in the
driver.
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Reviewed-by: Tomi Valkeinen <tomi.valkeinen@ti.com>
Instead of conditioning planes update based on the DSS manager hardware
state, use the enabled field newly added to the omap_crtc structure.
This reduces the dependency from the DRM layer to the DSS layer.
The enabled field is a transitory measure, the implementation should use
the CRTC atomic state instead. However, given that CRTCs are currently
not enabled/disabled through their .enable() and .disable() operations
but through a convoluted code paths starting at the associated encoder
operations, there is not clear guarantee that the atomic state always
matches the hardware state. This will be refactored later, at which
point the enabled field will be removed.
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Reviewed-by: Tomi Valkeinen <tomi.valkeinen@ti.com>
Instead of going through a complicated registration mechanism, just
call the OCP error IRQ handler directly from the main IRQ handler.
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Reviewed-by: Tomi Valkeinen <tomi.valkeinen@ti.com>
Instead of going through a complicated registration mechanism, just
expose the CRTC error IRQ function and call it directly from the main
IRQ handler.
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Reviewed-by: Tomi Valkeinen <tomi.valkeinen@ti.com>
As the FIFO underflow IRQ handler just prints an error message to the
kernel log, simplify the code by not registering one IRQ handler per
plane but print the messages directly from the main IRQ handler.
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Reviewed-by: Tomi Valkeinen <tomi.valkeinen@ti.com>
Don't print userspace parameters validation failures as error messages
to avoid giving userspace the ability to flood the kernel log.
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Reviewed-by: Tomi Valkeinen <tomi.valkeinen@ti.com>
The hardware requires all planes to have an identical pitch in number of
pixels. Given that all supported formats use the same number of bytes
per pixel in all planes, framebuffer creation checks can be simplified.
The implementations assumes that no format use more than two planes
which is true with the existing hardware.
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Reviewed-by: Tomi Valkeinen <tomi.valkeinen@ti.com>
Merge the single-user objects_lookup inline function into its caller,
allowing reuse of the error code path.
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Reviewed-by: Tomi Valkeinen <tomi.valkeinen@ti.com>
The driver stores in a custom structure named format several pieces of
information about the format that are available in the DRM core. Remove
them and get the information from the DRM core instead.
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Reviewed-by: Tomi Valkeinen <tomi.valkeinen@ti.com>
The only multi-planar format supported by the driver is NV12, there will
thus never be more than two planes per framebuffer.
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Reviewed-by: Tomi Valkeinen <tomi.valkeinen@ti.com>
PIN_HIGH is an expensive operation (in comparison to allocating from the
hole stack) unsuitable for frequent use (such as switching between
contexts). However, the kernel context should be pinned just once for
the lifetime of the driver, and here it is appropriate to keep it out of
the mappable range (in order to maximise mappable space for users).
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20161218153724.8439-6-chris@chris-wilson.co.uk
The requests conversion introduced a nasty bug where we could generate a
new request in the middle of constructing a request if we needed to idle
the system in order to evict space for a context. The request to idle
would be executed (and waited upon) before the current one, creating a
minor havoc in the seqno accounting, as we will consider the current
request to already be completed (prior to deferred seqno assignment) but
ring->last_retired_head would have been updated and still could allow
us to overwrite the current request before execution.
We also employed two different mechanisms to track the active context
until it was switched out. The legacy method allowed for waiting upon an
active context (it could forcibly evict any vma, including context's),
but the execlists method took a step backwards by pinning the vma for
the entire active lifespan of the context (the only way to evict was to
idle the entire GPU, not individual contexts). However, to circumvent
the tricky issue of locking (i.e. we cannot take struct_mutex at the
time of i915_gem_request_submit(), where we would want to move the
previous context onto the active tracker and unpin it), we take the
execlists approach and keep the contexts pinned until retirement.
The benefit of the execlists approach, more important for execlists than
legacy, was the reduction in work in pinning the context for each
request - as the context was kept pinned until idle, it could short
circuit the pinning for all active contexts.
We introduce new engine vfuncs to pin and unpin the context
respectively. The context is pinned at the start of the request, and
only unpinned when the following request is retired (this ensures that
the context is idle and coherent in main memory before we unpin it). We
move the engine->last_context tracking into the retirement itself
(rather than during request submission) in order to allow the submission
to be reordered or unwound without undue difficultly.
And finally an ulterior motive for unifying context handling was to
prepare for mock requests.
v2: Rename to last_retired_context, split out legacy_context tracking
for MI_SET_CONTEXT.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20161218153724.8439-3-chris@chris-wilson.co.uk
ast_get_dram_info() configures a window in order to access BMC memory.
A BMC register can be configured to disallow this, and if so, causes
an infinite loop in the ast driver which renders the system unusable.
Fix this by erroring out if an error is detected. On powerpc systems with
EEH, this leads to the device being fenced and the system continuing to
operate.
Cc: <stable@vger.kernel.org> # 3.10+
Signed-off-by: Russell Currey <ruscur@russell.cc>
Reviewed-by: Joel Stanley <joel@jms.id.au>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Link: http://patchwork.freedesktop.org/patch/msgid/20161215051241.20815-1-ruscur@russell.cc
- Modeset state needs mode_config->connection mutex, that covers
figuring out the encoder, and reading properties (since in the
atomic case those need to look at connector->state).
- Don't hold any locks for stuff that's invariant (i.e. possible
connectors).
- Same for connector lookup and unref, those don't need any locks.
- And finally the probe stuff is only protected by mode_config->mutex.
While at it updated the kerneldoc for these fields in drm_connector
and add docs explaining what's protected by which locks.
Reviewed-by: Sean Paul <seanpaul@chromium.org>
Signed-off-by: Daniel Vetter <daniel.vetter@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20161213230814.19598-10-daniel.vetter@ffwll.ch
Only static connectors should be left at this point, and we should be
able to clean them out by simply dropping that last reference still
around from drm_connector_init.
If that leaves anything behind then we have a driver bug.
Doing the final cleanup this way also allows us to use
drm_connector_iter, removing the very last place where we walk
connector_list explicitly in drm core&helpers.
Reviewed-by: Harry Wentland <harry.wentland@amd.com>
Reviewed-by: Sean Paul <seanpaul@chromium.org>
Signed-off-by: Daniel Vetter <daniel.vetter@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20161213230814.19598-8-daniel.vetter@ffwll.ch
The requirements for connector_list locking are a bit tricky:
- We need to be able to jump over zombie conectors (i.e. with refcount
== 0, but not yet removed from the list). If instead we require that
there's no zombies on the list then the final kref_put must happen
under the list protection lock, which means that locking context
leaks all over the place. Not pretty - better to deal with zombies
and wrap the locking just around the list_del in the destructor.
- When we walk the list we must _not_ hold the connector list lock. We
walk the connector list at an absolutely massive amounts of places,
if all those places can't ever call drm_connector_unreference the
code would get unecessarily complicated.
- connector_list needs it own lock, again too many places that walk it
that we could reuse e.g. mode_config.mutex without resulting in
inversions.
- Lots of code uses these loops to look-up a connector, i.e. they want
to be able to call drm_connector_reference. But on the other hand we
want connectors to stay on that list until they're dead (i.e.
connector_list can't hold a full reference), which means despite the
"can't hold lock for the loop body" rule we need to make sure a
connector doesn't suddenly become a zombie.
At first Dave&I discussed various horror-show approaches using srcu,
but turns out it's fairly easy:
- For the loop body we always hold an additional reference to the
current connector. That means it can't zombify, and it also means
it'll stay on the list, which means we can use it as our iterator to
find the next connector.
- When we try to find the next connector we only have to jump over
zombies. To make sure we don't chase bad pointers that entire loop
is protected with the new connect_list_lock spinlock. And because we
know that we're starting out with a non-zombie (need to drop our
reference for the old connector only after we have our new one),
we're guranteed to still be on the connector_list and either find
the next non-zombie or complete the iteration.
- Only downside is that we need to make sure that the temporary
reference for the loop body doesn't leak. iter_get/put() functions +
lockdep make sure that's the case.
- To avoid a flag day the new iterator macro has an _iter postfix. We
can rename it back once all the users of the unsafe version are gone
(there's about 100 list walkers for the connector_list).
For now this patch only converts all the list walking in the core,
leaving helpers and drivers for later patches. The nice thing is that
we can now finally remove 2 FIXME comments from the
register/unregister functions.
v2:
- use irqsafe spinlocks, so that we can use this in drm_state_dump
too.
- nuke drm_modeset_lock_all from drm_connector_init, now entirely
cargo-culted nonsense.
v3:
- do {} while (!kref_get_unless_zero), makes for a tidier loop (Dave).
- pretty kerneldoc
- add EXPORT_SYMBOL, helpers&drivers are supposed to use this.
v4: Change lockdep annotations to only check whether we release the
iter fake lock again (i.e. make sure that iter_put is called), but
not check any locking dependecies itself. That seams to require a
recursive read lock in trylock mode.
Cc: Dave Airlie <airlied@gmail.com>
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Sean Paul <seanpaul@chromium.org>
Signed-off-by: Daniel Vetter <daniel.vetter@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20161213230814.19598-6-daniel.vetter@ffwll.ch
<drm/drm_crtc.h> used to define most of the in-kernel KMS API. It has
now been split into separate files for each object type, but still
includes most other KMS headers to avoid breaking driver compilation.
As a step towards fixing that problem, remove the inclusion of
<drm/drm_encoder.h> from <drm/drm_crtc.h> and include it instead where
appropriate. Also remove the forward declarations of the drm_encoder and
drm_encoder_helper_funcs structures from <drm/drm_crtc.h> as they're not
needed in the header.
<drm/drm_encoder.h> now has to include <drm/drm_mode.h> and contain a
forward declaration of struct drm_encoder in order to allow including it
as the first header in a compilation unit.
Signed-off-by: Laurent Pinchart <laurent.pinchart+renesas@ideasonboard.com>
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Reviewed-by: Sinclair Yeh <syeh@vmware.com> # For vmwgfx
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Archit Taneja <architt@codeaurora.org>
Link: http://patchwork.freedesktop.org/patch/msgid/1481709550-29226-2-git-send-email-laurent.pinchart+renesas@ideasonboard.com
Pull i915/gvt KVMGT updates from Zhenyu Wang:
"KVMGT support depending on the VFIO/mdev framework"
* tag 'kvmgt-vfio-mdev-for-v4.10-rc1' of git://github.com/01org/gvt-linux:
drm/i915/gvt/kvmgt: add vfio/mdev support to KVMGT
drm/i915/gvt/kvmgt: read/write GPA via KVM API
drm/i915/gvt/kvmgt: replace kmalloc() by kzalloc()
