| Age | Commit message (Collapse) | Author |
|---|
|
Given an implicit semaphore from one engine to the next, check that if
we skip the wait on that semaphore the following batch although
submitted early (as it depends along the single engine timeline) is not
executed ahead of its dependency.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Cc: Matthew Brost <matthew.brost@intel.com>
Reviewed-by: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
|
|
Check that we can reorder batches around userspace semaphore waits by
injecting a semaphore that is only released by a later context.
NB: This is expected to fail with the current execlists implementation.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Mika Kuoppala <mika.kuoppala@linux.intel.com>
|
|
There is no guarantee that spinners are and will be implemented
using batches. As we have igt_spin_t, manipulate it through
igt_spin_* functions consistently and hide the batch nature.
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
|
|
Instead of opencoding the poll into the spinner, use
a helper to check if spinner has started.
v2: use zero as presumed offset (Chris)
v3: cleanup the relocs (Chris)
v4: leave the domains to zero, avoid relocation (Chris)
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
|
|
When resubmitting the spinner, fill in the expected offset so that we
are not tempted to relocate it across contexts (as the spinner must
point back to itself for the MI_BB_START to work). In this case, these
should work correctly as they are reusing the same active vma, but for
pedagogy we should dtrt.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Reviewed-by: Mika Kuoppala <mika.kuoppala@linux.intel.com>
|
|
If we have two tasks running on xcs0 and xcs1 independently, but who
queue subsequent work onto rcs, we may insert semaphores before the rcs
work and pick unwisely which task to run first. To maximise throughput,
we want to run on rcs whichever task is ready first. Conversely, if we
pick wrongly that can be used to trigger a GPU hang with unaware
userspace.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
|
|
We may use HW semaphores to schedule nearly-ready work such that they
are already spinning on the GPU waiting for the completion on another
engine. However, we don't want for that spinning task to actually block
any real work should it be scheduled.
v2: No typeof autos
v3: Don't cheat, check gen8 as well
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
|
|
How much energy does spinning on a semaphore consume relative to plain
old spinning?
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
|
|
The write hazard lies extend also to the cache-dirty tracking; as we
purposefully do not tell the kernel we are writing to the bo, it fails
to note the CPU cache as dirty and so the gem_read() may not
sufficiently flush the caches prior to reading back from the GPU.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Antonio Argenziano <antonio.argenziano@intel.com>
Reviewed-by: Antonio Argenziano <antonio.argenziano@intel.com>
|
|
use IOCTLs gem_read for reading and, gem_sync to wait on a batch, this
will help testing platforms that might not expose gtt mapping.
v2:
- use a local helper for reading from BOs. (Chris)
- Always sync before reading. (Chris)
v3:
- Better helper naming. (Chris)
- Start read from actual offset. (Chris)
v4:
- Always use byte size in helper. (Chris)
v5:
- Fix byte<->count conversion. (Chris)
Signed-off-by: Antonio Argenziano <antonio.argenziano@intel.com>
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
|
|
Still the blighted
Starting subtest: pi-ringfull-bsd
child 0 died with signal 14, Alarm clock
Subtest pi-ringfull-bsd: FAIL (0.133s)
So let's make sure our timer cannot fire later on during exit.
References https://bugs.freedesktop.org/show_bug.cgi?id=109584
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
|
|
Make sure that the SIGARLM handler is captured so that we don't
accidentally shoot ourselves in the foot instead of merely waking up the
execbuf.
The test is hitting
child 0 died with signal 14, Alarm clock
when we expect it to be hitting the inherited signal handler across the
fork. Which is odd.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
|
|
We can already move all the tests with distinct prefixes: gem_, gen3_
and i915_.
pm_ and drv_ tests will follow in batches, so we can do the
adjustments in the reporting/filtering layer of the CI system.
v2: Fix test-list.txt generation with meson
v3: Fix docs build (Petri)
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Petri Latvala <petri.latvala@intel.com>
Cc: Martin Peres <martin.peres@linux.intel.com>
Signed-off-by: Arkadiusz Hiler <arkadiusz.hiler@intel.com>
Reviewed-by: Petri Latvala <petri.latvala@intel.com>
Tested-by: Petri Latvala <petri.latvala@intel.com>
Acked-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Acked-by: Chris Wilson <chris@chris-wilson.co.uk>
|
