[Intel-gfx] [PATCH igt v3] igt/gem_busy: Replace arbitrary busy batch with indefinite spinbatch
Chris Wilson
chris at chris-wilson.co.uk
Sat Dec 2 08:47:17 UTC 2017
In CI, we were observing situations where the busy blt would complete
before the very next instruction (in userspace) to assert that it was
busy. This is entirely possible if the process was scheduled away and
slept for longer than the arbitrary batch. Instead replace arbitrariness
with a precise infinity.
v2: Be respectful to UP!
v3: Move spinbatch to owning process to avoid serialisation delays.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=103829
Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>
---
tests/gem_busy.c | 263 +++++++++++++++++++++++++++++++++----------------------
1 file changed, 157 insertions(+), 106 deletions(-)
diff --git a/tests/gem_busy.c b/tests/gem_busy.c
index 2b88270b..4ba23241 100644
--- a/tests/gem_busy.c
+++ b/tests/gem_busy.c
@@ -21,10 +21,16 @@
* IN THE SOFTWARE.
*/
+#include <sched.h>
+#include <signal.h>
+#include <sys/ioctl.h>
+
#include "igt.h"
#include "igt_rand.h"
+#include "igt_vgem.h"
#define LOCAL_EXEC_NO_RELOC (1<<11)
+#define PAGE_ALIGN(x) ALIGN(x, 4096)
/* Exercise the busy-ioctl, ensuring the ABI is never broken */
IGT_TEST_DESCRIPTION("Basic check of busy-ioctl ABI.");
@@ -59,80 +65,6 @@ static void __gem_busy(int fd,
*read = busy.busy >> 16;
}
-static uint32_t busy_blt(int fd)
-{
- const int gen = intel_gen(intel_get_drm_devid(fd));
- const int has_64bit_reloc = gen >= 8;
- struct drm_i915_gem_execbuffer2 execbuf;
- struct drm_i915_gem_exec_object2 object[2];
- struct drm_i915_gem_relocation_entry reloc[200], *r;
- uint32_t *map;
- int factor = 100;
- int i = 0;
-
- memset(object, 0, sizeof(object));
- object[0].handle = gem_create(fd, 1024*1024);
- object[1].handle = gem_create(fd, 4096);
-
- r = memset(reloc, 0, sizeof(reloc));
- map = gem_mmap__cpu(fd, object[1].handle, 0, 4096, PROT_WRITE);
- gem_set_domain(fd, object[1].handle,
- I915_GEM_DOMAIN_CPU, I915_GEM_DOMAIN_CPU);
-
-#define COPY_BLT_CMD (2<<29|0x53<<22|0x6)
-#define BLT_WRITE_ALPHA (1<<21)
-#define BLT_WRITE_RGB (1<<20)
- while (factor--) {
- /* XY_SRC_COPY */
- map[i++] = COPY_BLT_CMD | BLT_WRITE_ALPHA | BLT_WRITE_RGB;
- if (has_64bit_reloc)
- map[i-1] += 2;
- map[i++] = 0xcc << 16 | 1 << 25 | 1 << 24 | (4*1024);
- map[i++] = 0;
- map[i++] = 256 << 16 | 1024;
-
- r->offset = i * sizeof(uint32_t);
- r->target_handle = object[0].handle;
- r->read_domains = I915_GEM_DOMAIN_RENDER;
- r->write_domain = I915_GEM_DOMAIN_RENDER;
- r++;
- map[i++] = 0;
- if (has_64bit_reloc)
- map[i++] = 0;
-
- map[i++] = 0;
- map[i++] = 4096;
-
- r->offset = i * sizeof(uint32_t);
- r->target_handle = object[0].handle;
- r->read_domains = I915_GEM_DOMAIN_RENDER;
- r->write_domain = 0;
- r++;
- map[i++] = 0;
- if (has_64bit_reloc)
- map[i++] = 0;
- }
- map[i++] = MI_BATCH_BUFFER_END;
- igt_assert(i <= 4096/sizeof(uint32_t));
- igt_assert(r - reloc <= ARRAY_SIZE(reloc));
- munmap(map, 4096);
-
- object[1].relocs_ptr = to_user_pointer(reloc);
- object[1].relocation_count = r - reloc;
-
- memset(&execbuf, 0, sizeof(execbuf));
- execbuf.buffers_ptr = to_user_pointer(object);
- execbuf.buffer_count = 2;
- if (gen >= 6)
- execbuf.flags = I915_EXEC_BLT;
- gem_execbuf(fd, &execbuf);
- igt_assert(gem_bo_busy(fd, object[0].handle));
-
- igt_debug("Created busy handle %d\n", object[0].handle);
- gem_close(fd, object[1].handle);
- return object[0].handle;
-}
-
static bool exec_noop(int fd,
uint32_t *handles,
unsigned ring,
@@ -167,6 +99,7 @@ static bool still_busy(int fd, uint32_t handle)
static void semaphore(int fd, unsigned ring, uint32_t flags)
{
uint32_t bbe = MI_BATCH_BUFFER_END;
+ igt_spin_t *spin;
uint32_t handle[3];
uint32_t read, write;
uint32_t active;
@@ -179,7 +112,8 @@ static void semaphore(int fd, unsigned ring, uint32_t flags)
gem_write(fd, handle[BATCH], 0, &bbe, sizeof(bbe));
/* Create a long running batch which we can use to hog the GPU */
- handle[BUSY] = busy_blt(fd);
+ handle[BUSY] = gem_create(fd, 4096);
+ spin = igt_spin_batch_new(fd, 0, ring, handle[BUSY]);
/* Queue a batch after the busy, it should block and remain "busy" */
igt_assert(exec_noop(fd, handle, ring | flags, false));
@@ -208,6 +142,7 @@ static void semaphore(int fd, unsigned ring, uint32_t flags)
/* Check that our long batch was long enough */
igt_assert(still_busy(fd, handle[BUSY]));
+ igt_spin_batch_free(fd, spin);
/* And make sure it becomes idle again */
gem_sync(fd, handle[TEST]);
@@ -364,47 +299,146 @@ static void xchg_u32(void *array, unsigned i, unsigned j)
u32[j] = tmp;
}
+struct cork {
+ int device;
+ uint32_t handle;
+ uint32_t fence;
+};
+
+static void plug(int fd, struct cork *c)
+{
+ struct vgem_bo bo;
+ int dmabuf;
+
+ c->device = drm_open_driver(DRIVER_VGEM);
+
+ bo.width = bo.height = 1;
+ bo.bpp = 4;
+ vgem_create(c->device, &bo);
+ c->fence = vgem_fence_attach(c->device, &bo, VGEM_FENCE_WRITE);
+
+ dmabuf = prime_handle_to_fd(c->device, bo.handle);
+ c->handle = prime_fd_to_handle(fd, dmabuf);
+ close(dmabuf);
+}
+
+static void unplug(struct cork *c)
+{
+ vgem_fence_signal(c->device, c->fence);
+ close(c->device);
+}
+
+static void alarm_handler(int sig)
+{
+}
+
+static int __execbuf(int fd, struct drm_i915_gem_execbuffer2 *execbuf)
+{
+ return ioctl(fd, DRM_IOCTL_I915_GEM_EXECBUFFER2, execbuf);
+}
+
+static unsigned int measure_ring_size(int fd)
+{
+ struct sigaction sa = { .sa_handler = alarm_handler };
+ struct drm_i915_gem_exec_object2 obj[2];
+ struct drm_i915_gem_execbuffer2 execbuf;
+ const uint32_t bbe = MI_BATCH_BUFFER_END;
+ unsigned int count, last;
+ struct itimerval itv;
+ struct cork c;
+
+ memset(obj, 0, sizeof(obj));
+ obj[1].handle = gem_create(fd, 4096);
+ gem_write(fd, obj[1].handle, 0, &bbe, sizeof(bbe));
+
+ memset(&execbuf, 0, sizeof(execbuf));
+ execbuf.buffers_ptr = to_user_pointer(obj + 1);
+ execbuf.buffer_count = 1;
+ gem_execbuf(fd, &execbuf);
+ gem_sync(fd, obj[1].handle);
+
+ plug(fd, &c);
+ obj[0].handle = c.handle;
+
+ execbuf.buffers_ptr = to_user_pointer(obj);
+ execbuf.buffer_count = 2;
+
+ sigaction(SIGALRM, &sa, NULL);
+ itv.it_interval.tv_sec = 0;
+ itv.it_interval.tv_usec = 100;
+ itv.it_value.tv_sec = 0;
+ itv.it_value.tv_usec = 1000;
+ setitimer(ITIMER_REAL, &itv, NULL);
+
+ last = -1;
+ count = 0;
+ do {
+ if (__execbuf(fd, &execbuf) == 0) {
+ count++;
+ continue;
+ }
+
+ if (last == count)
+ break;
+
+ last = count;
+ } while (1);
+
+ memset(&itv, 0, sizeof(itv));
+ setitimer(ITIMER_REAL, &itv, NULL);
+
+ unplug(&c);
+ gem_close(fd, obj[1].handle);
+ gem_quiescent_gpu(fd);
+
+ return count;
+}
+
static void close_race(int fd)
{
-#define N_HANDLES 4096
- const int ncpus = sysconf(_SC_NPROCESSORS_ONLN);
- uint32_t *handles;
+ const unsigned int ncpus = sysconf(_SC_NPROCESSORS_ONLN);
+ const unsigned int nhandles = measure_ring_size(fd) / 2;
+ unsigned int engines[16], nengine;
unsigned long *control;
- unsigned long count = 0;
+ uint32_t *handles;
int i;
- intel_require_memory(N_HANDLES, 4096, CHECK_RAM);
+ igt_require(ncpus > 1);
+ intel_require_memory(nhandles, 4096, CHECK_RAM);
- /* One thread spawning work and randomly closing fd.
+ /*
+ * One thread spawning work and randomly closing handles.
* One background thread per cpu checking busyness.
*/
+ nengine = 0;
+ for_each_engine(fd, i)
+ engines[nengine++] = i;
+ igt_require(nengine);
+
control = mmap(NULL, 4096, PROT_WRITE, MAP_SHARED | MAP_ANON, -1, 0);
igt_assert(control != MAP_FAILED);
- handles = mmap(NULL, N_HANDLES*sizeof(*handles),
+ handles = mmap(NULL, PAGE_ALIGN(nhandles*sizeof(*handles)),
PROT_WRITE, MAP_SHARED | MAP_ANON, -1, 0);
igt_assert(handles != MAP_FAILED);
- for (i = 0; i < N_HANDLES; i++) {
- handles[i] = gem_create(fd, 4096);
- }
-
- igt_fork(child, ncpus) {
+ igt_fork(child, ncpus - 1) {
struct drm_i915_gem_busy busy;
- uint32_t indirection[N_HANDLES];
+ uint32_t indirection[nhandles];
+ unsigned long count = 0;
- for (i = 0; i < N_HANDLES; i++)
+ for (i = 0; i < nhandles; i++)
indirection[i] = i;
hars_petruska_f54_1_random_perturb(child);
memset(&busy, 0, sizeof(busy));
do {
- igt_permute_array(indirection, N_HANDLES, xchg_u32);
+ igt_permute_array(indirection, nhandles, xchg_u32);
__sync_synchronize();
- for (i = 0; i < N_HANDLES; i++) {
- busy.handle = indirection[handles[i]];
+ for (i = 0; i < nhandles; i++) {
+ busy.handle = handles[indirection[i]];
/* Check that the busy computation doesn't
* explode in the face of random gem_close().
*/
@@ -414,30 +448,47 @@ static void close_race(int fd)
} while(*(volatile long *)control == 0);
igt_debug("child[%d]: count = %lu\n", child, count);
- control[child] = count;
+ control[child + 1] = count;
}
- igt_until_timeout(20) {
- int j = rand() % N_HANDLES;
+ igt_fork(child, 1) {
+ struct sched_param rt = {.sched_priority = 99 };
+ igt_spin_t *spin[nhandles];
+ unsigned long count = 0;
+
+ igt_assert(sched_setscheduler(getpid(), SCHED_RR, &rt) == 0);
+
+ for (i = 0; i < nhandles; i++) {
+ spin[i] = igt_spin_batch_new(fd, 0,
+ engines[rand() % nengine], 0);
+ handles[i] = spin[i]->handle;
+ }
- gem_close(fd, handles[j]);
+ igt_until_timeout(20) {
+ for (i = 0; i < nhandles; i++) {
+ igt_spin_batch_free(fd, spin[i]);
+ spin[i] = igt_spin_batch_new(fd, 0,
+ engines[rand() % nengine],
+ 0);
+ handles[i] = spin[i]->handle;
+ __sync_synchronize();
+ }
+ count += nhandles;
+ }
+ control[0] = count;
__sync_synchronize();
- handles[j] = busy_blt(fd);
- count++;
+ for (i = 0; i < nhandles; i++)
+ igt_spin_batch_free(fd, spin[i]);
}
- control[0] = 1;
igt_waitchildren();
- for (i = 0; i < ncpus; i++)
- control[ncpus + 1] += control[i + 1];
+ for (i = 0; i < ncpus - 1; i++)
+ control[ncpus] += control[i + 1];
igt_info("Total execs %lu, busy-ioctls %lu\n",
- count, control[ncpus + 1] * N_HANDLES);
-
- for (i = 0; i < N_HANDLES; i++)
- gem_close(fd, handles[i]);
+ control[0], control[ncpus] * nhandles);
- munmap(handles, N_HANDLES * sizeof(*handles));
+ munmap(handles, PAGE_ALIGN(nhandles * sizeof(*handles)));
munmap(control, 4096);
gem_quiescent_gpu(fd);
--
2.15.1
More information about the Intel-gfx
mailing list