The pm_runtime_put() we were using immediately released power on the
device, which meant that we were generally turning the device off and
on once per frame. In many profiles I've looked at, that added up to
about 1% of CPU time, but this could get worse in the case of frequent
rendering and readback (as may happen in X rendering). By keeping the
device on until we've been idle for a couple of frames, we drop the
overhead of runtime PM down to sub-.1%.
Signed-off-by: Eric Anholt <eric@anholt.net>
If the allocation fails the current code returns success. If
copy_from_user() fails it returns the number of bytes remaining instead
of -EFAULT.
Fixes: d5b1a78a77 ("drm/vc4: Add support for drawing 3D frames.")
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
With the introduction of bin/render pipelining, the previous job may
not be completed when we start binning the next one. If the previous
job wrote our VBO, IB, or CS textures, then the binning stage might
get stale or uninitialized results.
Fixes the major rendering failure in glmark2 -b terrain.
Signed-off-by: Eric Anholt <eric@anholt.net>
Fixes: ca26d28bba ("drm/vc4: improve throughput by pipelining binning and rendering jobs")
Cc: stable@vger.kernel.org
-----BEGIN PGP SIGNATURE-----
iQEcBAABAgAGBQJX6H4uAAoJEHm+PkMAQRiG5sMH/3yzrMiUCSokdS+cvY+jgKAG
JS58JmRvBPz2mRaU3MRPBGRDeCz/Nc9LggL2ZcgM+E1ZYirlYyQfIED3lkqk5R07
kIN1wmb+kQhXyU4IY3fEX7joqyKC6zOy4DUChPkBQU0/0+VUmdVmcJvsuPlnMZtf
g95m0BdYTui+eDezASRqOEp3Lb5ONL4c3ao4yBP0LHF033ctj3VJQiyi5uERPZJ0
5e6Mo7Wxn78t9WqJLQAiEH46kTwT2plNlxf3XXqTenfIdbWhqE873HPGeSMa3VQV
VywXTpCpSPQsA8BYg66qIbebdKOhs9MOviHVfqDtwQlvwhjlBDya0gNHfI5fSy4=
=Y/L5
-----END PGP SIGNATURE-----
Merge tag 'v4.8-rc8' into drm-next
Linux 4.8-rc8
There was a lot of fallout in the imx/amdgpu/i915 drivers, so backmerge
it now to avoid troubles.
* tag 'v4.8-rc8': (1442 commits)
Linux 4.8-rc8
fault_in_multipages_readable() throws set-but-unused error
mm: check VMA flags to avoid invalid PROT_NONE NUMA balancing
radix tree: fix sibling entry handling in radix_tree_descend()
radix tree test suite: Test radix_tree_replace_slot() for multiorder entries
fix memory leaks in tracing_buffers_splice_read()
tracing: Move mutex to protect against resetting of seq data
MIPS: Fix delay slot emulation count in debugfs
MIPS: SMP: Fix possibility of deadlock when bringing CPUs online
mm: delete unnecessary and unsafe init_tlb_ubc()
huge tmpfs: fix Committed_AS leak
shmem: fix tmpfs to handle the huge= option properly
blk-mq: skip unmapped queues in blk_mq_alloc_request_hctx
MIPS: Fix pre-r6 emulation FPU initialisation
arm64: kgdb: handle read-only text / modules
arm64: Call numa_store_cpu_info() earlier.
locking/hung_task: Fix typo in CONFIG_DETECT_HUNG_TASK help text
nvme-rdma: only clear queue flags after successful connect
i2c: qup: skip qup_i2c_suspend if the device is already runtime suspended
perf/core: Limit matching exclusive events to one PMU
...
This came from the initial bringup code, which always idled the GPU
and always reset the overflow. That massively increases the size of
the working set when you're doing lots of small draws, though, as is
common on X desktops or piglit.
Signed-off-by: Eric Anholt <eric@anholt.net>
We'd end up NULL pointer dereferencing because we didn't take the
error path out in the parent. Fixes igt vc4_lookup_fail test.
Signed-off-by: Eric Anholt <eric@anholt.net>
Fixes: d5b1a78a77 ("drm/vc4: Add support for drawing 3D frames.")
Cc: stable@vger.kernel.org
The BO cache will complain if BOs are still allocated when we try to
destroy it (since freeing those BOs would try to hit the cache). You
could hit this if you were to unload the module after a GPU hang.
Signed-off-by: Eric Anholt <eric@anholt.net>
Fixes: 214613656b ("drm/vc4: Add an interface for capturing the GPU state after a hang.")
If you exceeded the size that kmalloc would return, you'd get a dmesg
warning and a return from the job submit. We can handle much
allocations with vmalloc, and drm_malloc_ab makes that decision.
Fixes failure in piglit's scissor-many.
Signed-off-by: Eric Anholt <eric@anholt.net>
If you managed to exceed the limit to switch to vmalloc, we'd use the
wrong free.
Signed-off-by: Eric Anholt <eric@anholt.net>
Fixes: d5b1a78a77 ("drm/vc4: Add support for drawing 3D frames.")
Cc: stable@vger.kernel.org
Since my last struct_mutex crusade someone escaped!
This already has the advantage that for the common case when someone
else holds a ref the unref won't even acquire dev->struct_mutex. And
I'm working on code to allow drivers to completely opt-out of any and
all dev->struct_mutex usage, but that only works if they use the
_unlocked variants everywhere.
v2: Drop comment too.
v3: Drop the other comment too.
Cc: Eric Anholt <eric@anholt.net>
Reviewed-by: Eric Anholt <eric@anholt.net>
Signed-off-by: Daniel Vetter <daniel.vetter@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1464630800-30786-15-git-send-email-daniel.vetter@ffwll.ch
drm_gem_object_lookup() has never required the drm_device for its file
local translation of the user handle to the GEM object. Let's remove the
unused parameter and save some space.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: dri-devel@lists.freedesktop.org
Cc: Dave Airlie <airlied@redhat.com>
Cc: Daniel Vetter <daniel.vetter@ffwll.ch>
[danvet: Fixup kerneldoc too.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
The hardware provides us with separate threads for binning and
rendering, and the existing model waits for them both to complete
before submitting the next job.
Splitting the binning and rendering submissions reduces idle time and
gives us approx 20-30% speedup with some x11perf tests such as -line10
and -tilerect1. Improves openarena performance by 1.01897% +/-
0.247857% (n=16).
Thanks to anholt for suggesting this.
v2: Rebase on the spurious resets fix (change by anholt).
Signed-off-by: Varad Gautam <varadgautam@gmail.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
Signed-off-by: Eric Anholt <eric@anholt.net>
This gets us functional GPU reset again, like we had until a refactor
at merge time. Tested with a little patch to stuff in a broken binner
job every 100 frames.
Signed-off-by: Eric Anholt <eric@anholt.net>
This may actually get us a feature that the closed driver didn't have:
turning off the GPU in between rendering jobs, while the V3D device is
still opened by the client.
There may be some tuning to be applied here to use autosuspend so that
we don't bounce the device's power so much, but in steady-state
GPU-bound rendering we keep the power on (since we keep multiple jobs
outstanding) and even if we power cycle on every job we can still
manage at least 680 fps.
More importantly, though, runtime PM will allow us to power off the
device to do a GPU reset.
v2: Switch #ifdef to CONFIG_PM not CONFIG_PM_SLEEP (caught by kbuild
test robot)
Signed-off-by: Eric Anholt <eric@anholt.net>
We were tracking the "where are the head pointers pointing" globally,
so if another job reused the same BOs and execution was at the same
point as last time we checked, we'd stop and trigger a reset even
though the GPU had made progress.
Signed-off-by: Eric Anholt <eric@anholt.net>
These ioctls end up getting exposed to fairly directly to GL users,
and having normal user operations print DRM errors is obviously wrong.
The message was originally to give us some idea of what happened when
a hang occurred, but we have a DRM_INFO from reset for that.
Signed-off-by: Eric Anholt <eric@anholt.net>
This caused the wait ioctls to claim that waiting had completed when
we actually got interrupted by a signal before it was done. Fixes
broken rendering throttling that produced serious lag in X window
dragging.
Signed-off-by: Eric Anholt <eric@anholt.net>
"exec->exec_bo" is NULL at this point so this code returns success. We
want to return -ENOMEM.
Fixes: d5b1a78a77 ('drm/vc4: Add support for drawing 3D frames.')
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
"state" is smaller than "kernel_state" so we end up corrupting memory.
Fixes: 214613656b ('drm/vc4: Add an interface for capturing the GPU state after a hang.')
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
The copy_to/from_user() functions return the number of bytes remaining
to be copied. We want to return error codes here.
Also it's a bad idea to print an error message if a copy from user fails
because users can use that to spam /var/log/messages which is annoying
so I removed those.
Fixes: 214613656b ('drm/vc4: Add an interface for capturing the GPU state after a hang.')
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
This can be parsed with vc4-gpu-tools tools for trying to figure out
what was going on.
v2: Use __u32-style types.
Signed-off-by: Eric Anholt <eric@anholt.net>
An async pageflip stores the modeset to be done and executes it once
the BOs are ready to be displayed. This gets us about 3x performance
in full screen rendering with pageflipping.
Signed-off-by: Eric Anholt <eric@anholt.net>
The user submission is basically a pointer to a command list and a
pointer to uniforms. We copy those in to the kernel, validate and
relocate them, and store the result in a GPU BO which we queue for
execution.
v2: Drop support for NV shader recs (not necessary for GL), simplify
vc4_use_bo(), improve bin flush/semaphore checks, use __u32 style
types.
Signed-off-by: Eric Anholt <eric@anholt.net>
Eric Anholt