On allocating a request, we apply some backpressure if we fail to allocate a request. The backpressure we apply involves waiting for the device to idle, causing a stall on the clients (trying to throttle heavy allocators) and as we may be inside a plugged critical section, the only way to idle the GPU is by waiting for hangcheck to kick in. Try to avoid the allocation failure by trimming the VM caches between iterations. References: https://bugs.freedesktop.org/show_bug.cgi?id=106680 Signed-off-by: Chris Wilson <chris@xxxxxxxxxxxxxxxxxx> --- tests/gem_exec_await.c | 7 +++++++ 1 file changed, 7 insertions(+) diff --git a/tests/gem_exec_await.c b/tests/gem_exec_await.c index 5cfeb8ec8..b9a2c6105 100644 --- a/tests/gem_exec_await.c +++ b/tests/gem_exec_await.c @@ -147,6 +147,13 @@ static void wide(int fd, int ring_size, int timeout, unsigned int flags) count = 0; igt_until_timeout(timeout) { struct timespec start, now; + + /* + * Flush all caches between runs, we do not want to encounter + * reclaim in the middle of the plugged critical section. + */ + intel_purge_vm_caches(fd); + for (unsigned e = 0; e < nengine; e++) { uint64_t address; int i; -- 2.19.0 _______________________________________________ Intel-gfx mailing list Intel-gfx@xxxxxxxxxxxxxxxxxxxxx https://lists.freedesktop.org/mailman/listinfo/intel-gfx
