Move object release into a separate worker. Releasing objects requires
sending commands to the host. Doing that in the dequeue worker will
cause deadlocks in case the command queue gets filled up, because the
dequeue worker is also the one which will free up slots in the command
queue.
Reported-by: Chia-I Wu <olvaffe@gmail.com>
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
Reviewed-by: Chia-I Wu <olvaffe@gmail.com>
Tested-by: Chia-I Wu <olvaffe@gmail.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20190830060116.10476-1-kraxel@redhat.com
virtio-gpu basically needs a sg_table for the bo, to tell the host where
the backing pages for the object are. So the gem shmem helpers are a
perfect fit. Some drm_gem_object_funcs need thin wrappers to update the
host state, but otherwise the helpers handle everything just fine.
Once the fencing was sorted the switch was surprisingly easy and for the
most part just removing the ttm code.
v4: fix drm_gem_object_funcs name.
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
Acked-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Reviewed-by: Chia-I Wu <olvaffe@gmail.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20190829103301.3539-15-kraxel@redhat.com
Rework fencing workflow. Stop using ttm helpers, use the
virtio_gpu_array_* helpers instead.
Due to using the gem reservation object it is initialized and ready for
use before calling ttm_bo_init. So we can simply use the standard
fencing workflow and drop the tricky logic which checks whenever the
command is in flight still.
v6: rewrite most of the patch.
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
Reviewed-by: Chia-I Wu <olvaffe@gmail.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20190829103301.3539-10-kraxel@redhat.com
Rework fencing workflow, starting with virtio_gpu_execbuffer_ioctl.
Stop using ttm helpers, use the virtio_gpu_array_* helpers (which work
on the reservation objects directly) instead.
Also store the object array in struct virtio_gpu_vbuffer, so we
explicitly keep a reference of all buffers used instead of depending
on ttm_bo_put() checking whenever the object is actually idle before
releasing it.
New workflow:
(1) All gem objects needed by a command are added to a
virtio_gpu_object_array.
(2) All reservation objects will be locked (virtio_gpu_array_lock_resv).
(3) virtio_gpu_fence_emit() completes fence initialization.
(4) fence gets added to the objects, reservation objects are unlocked
(virtio_gpu_array_add_fence, virtio_gpu_array_unlock_resv).
(5) virtio command is submitted to the host.
(6) The completion callback (virtio_gpu_dequeue_ctrl_func)
will drop object references and free virtio_gpu_object_array.
v6: rewrite most of the patch.
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
Reviewed-by: Chia-I Wu <olvaffe@gmail.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20190829103301.3539-9-kraxel@redhat.com
Split virtqueue_kick() call into virtqueue_kick_prepare(), which
requires serialization, and virtqueue_notify(), which does not. Move
the virtqueue_notify() call out of the critical section protected by the
queue lock. This avoids triggering a vmexit while holding the lock and
thereby fixes a rather bad spinlock contention.
Suggested-by: Chia-I Wu <olvaffe@gmail.com>
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
Reviewed-by: Chia-I Wu <olvaffe@gmail.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20190813082509.29324-3-kraxel@redhat.com
drm_connector_update_edid_property can sleep, we must not
call it while holding a spinlock. Move the callsite.
Fixes: b4b01b4995 ("drm/virtio: add edid support")
Reported-by: Max Filippov <jcmvbkbc@gmail.com>
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
Tested-by: Max Filippov <jcmvbkbc@gmail.com>
Tested-by: Cornelia Huck <cohuck@redhat.com>
Acked-by: Cornelia Huck <cohuck@redhat.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20190405044602.2334-1-kraxel@redhat.com
After data is copied to the cache entry, atomic_set is used indicate
that the data is the entry is valid without appropriate memory barriers.
Similarly the read side was missing the corresponding memory barriers.
Signed-off-by: David Riley <davidriley@chromium.org>
Link: http://patchwork.freedesktop.org/patch/msgid/20190610211810.253227-5-davidriley@chromium.org
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
virtio_gpu_cmd_get_capset would check for the existence of an entry
under lock. If it was not found, it would unlock and call
virtio_gpu_cmd_get_capset to create a new entry. The new entry would
be added it to the list without checking if it was added by another
task during the period where the lock was not held resulting in
duplicate entries.
Compounding this issue, virtio_gpu_cmd_capset_cb would stop iterating
after find the first matching entry. Multiple callbacks would modify
the first entry, but any subsequent entries and their associated waiters
would eventually timeout since they don't become valid, also wasting
memory along the way.
Signed-off-by: David Riley <davidriley@chromium.org>
Link: http://patchwork.freedesktop.org/patch/msgid/20190605234423.11348-3-davidriley@chromium.org
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
This patch moves the virtio_gpu_cmd_create_resource() call (which
notifies the host about the new resource created) into the
virtio_gpu_object_create() function. That way we can call
virtio_gpu_cmd_create_resource() before ttm_bo_init(), so the host
already knows about the object when ttm initializes the object and calls
our driver callbacks.
Specifically the object is already created when the
virtio_gpu_ttm_tt_bind() callback invokes virtio_gpu_object_attach(),
so the extra virtio_gpu_object_attach() calls done after
virtio_gpu_object_create() are not needed any more.
The fence support for the create ioctl becomes a bit more tricky though.
The code moved into virtio_gpu_object_create() too. We first submit the
(fenced) virtio_gpu_cmd_create_resource() command, then initialize the
ttm object, and finally attach just created object to the fence for the
command in case it didn't finish yet.
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
Acked-by: Noralf Trønnes <noralf@tronnes.org>
Link: http://patchwork.freedesktop.org/patch/msgid/20190318113332.10900-6-kraxel@redhat.com
Add format, width and height fields to the virtio_gpu_object_params
struct. With that in place we can use the parameter struct for
virtio_gpu_cmd_create_resource() calls too.
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
Acked-by: Noralf Trønnes <noralf@tronnes.org>
Link: http://patchwork.freedesktop.org/patch/msgid/20190318113332.10900-4-kraxel@redhat.com
Since commit "9fdd90c0f4 drm/virtio: add virtio_gpu_alloc_fence()"
fences are not allocated any more by virtio_gpu_fence_emit(). So there
is no need to pass down a reference to the fence pointer, a plain
pointer is enough now.
Convert virtio_gpu_fence_emit() and callers.
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
Reviewed-by: Robert Foss <robert.foss@collabora.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20181128151021.29565-2-kraxel@redhat.com
Refactor fence creation, add fences to relevant GPU
operations and add cursor helper functions.
This removes the potential for allocation failures from the
cmd_submit and atomic_commit paths.
Now a fence will be allocated first and only after that
will we proceed with the rest of the execution.
Signed-off-by: Gustavo Padovan <gustavo.padovan@collabora.com>
Signed-off-by: Robert Foss <robert.foss@collabora.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20181112165157.32765-2-robert.foss@collabora.com
Suggested-by: Rob Herring <robh@kernel.org>
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
Move virtio_gpu_resource_id_{get,put} to virtgpu_object.c and make them
static. Allocate and free the id on creation and destroy, drop all
other calls. That way objects have a valid handle for the whole
lifetime of the object.
Also fixes ids leaking. Worst offender are dumb buffers, and I think
some error paths too.
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
Reviewed-by: Dave Airlie <airlied@redhat.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20181019061847.18958-7-kraxel@redhat.com
Track whenever the virtio_gpu_object is already created (i.e. host knows
about it) in a new variable. Add checks to virtio_gpu_object_attach()
to do nothing on objects not created yet.
Make virtio_gpu_ttm_bo_destroy() use the new variable too, instead of
expecting hw_res_handle indicating the object state.
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
Reviewed-by: Dave Airlie <airlied@redhat.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20181019061847.18958-2-kraxel@redhat.com
Pass virtio_gpu_object down to virtio_gpu_cmd_transfer_to_host_2d and
virtio_gpu_cmd_transfer_to_host_3d functions, instead of passing just
the virtio resource handle.
This is needed to lookup the scatter list of the object, for dma sync.
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
Reviewed-by: Jiandi An <jiandi.an@amd.com>
Tested-by: Jiandi An <jiandi.an@amd.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20180920062924.6514-1-kraxel@redhat.com
The new function balances virtio_gpu_object_attach().
Also make virtio_gpu_cmd_resource_inval_backing() static and switch
call sites to the new virtio_gpu_object_attach() function.
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
Reviewed-by: Dave Airlie <airlied@redhat.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20180829122026.27012-2-kraxel@redhat.com
This doesn't affect runtime because in the current code "idx" is always
valid.
First, we read from "vgdev->capsets[idx].max_size" before checking
whether "idx" is within bounds. And secondly the bounds check is off by
one so we could end up reading one element beyond the end of the
vgdev->capsets[] array.
Fixes: 62fb7a5e10 ("virtio-gpu: add 3d/virgl support")
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20180704094250.m7sgvvzg3dhcvv3h@kili.mountain
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
Wait until we have enough space in the virt queue to actually queue up
our request. Avoids the guest spinning in case we have a non-zero
amount of free entries but not enough for the request.
Cc: stable@vger.kernel.org
Reported-by: Alain Magloire <amagloire@blackberry.com>
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
Reviewed-by: Dave Airlie <airlied@redhat.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20180403095904.11152-1-kraxel@redhat.com
Signed-off-by: Sean Paul <seanpaul@chromium.org>
virtio_gpu_queue_ctrl_buffer_locked is called with ctrlq.qlock taken, it
releases and acquires this lock. This causes a sparse warning. Add
appropriate annotations for sparse context checking.
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
When virtio_gpu_free_vbufs exits due to list empty, it does not
drop the free_vbufs lock that it took.
list empty is not expected to happen anyway, but it can't hurt to fix
this and drop the lock.
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
virtio-gpu guest driver appearently can run out of buffers.
allocate some extra buffers, as quick stopgap for 4.9.
analyzing root cause and fixing it properly is TBD.
Reported-by: Jiri Slaby <jslaby@suse.cz>
Tested-by: Jiri Slaby <jslaby@suse.cz>
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
Add helper function to handle the submission of fenced control requests.
Make sure we initialize the fence while holding the virtqueue lock, so
requests can't be reordered.
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
