[Intel-gfx] [igt-dev] [PATCH i-g-t] i915/gem_request_retire: Switch from random blitter loads to dummy
Chris Wilson
chris at chris-wilson.co.uk
Mon Nov 30 16:24:44 UTC 2020
Quoting Tvrtko Ursulin (2020-11-30 16:20:18)
>
> On 30/11/2020 10:36, Chris Wilson wrote:
> > Use the spinners to provide exactly the right amount of background
> > busyness.
> >
> > Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>
> > Cc: Tvrtko Ursulin <tvrtko.ursulin at linux.intel.com>
> > ---
> > lib/igt_dummyload.c | 23 ++---
> > lib/igt_dummyload.h | 1 +
> > tests/i915/gem_request_retire.c | 170 ++++----------------------------
> > 3 files changed, 34 insertions(+), 160 deletions(-)
> >
> > diff --git a/lib/igt_dummyload.c b/lib/igt_dummyload.c
> > index d58f73108..891e4eaab 100644
> > --- a/lib/igt_dummyload.c
> > +++ b/lib/igt_dummyload.c
> > @@ -161,19 +161,20 @@ emit_recursive_batch(igt_spin_t *spin,
> > if (opts->dependency) {
> > igt_assert(!(opts->flags & IGT_SPIN_POLL_RUN));
> >
> > - r = &relocs[obj[BATCH].relocation_count++];
> > -
> > - /* dummy write to dependency */
> > obj[SCRATCH].handle = opts->dependency;
> > obj[SCRATCH].offset = addr;
> > - obj[SCRATCH].flags = EXEC_OBJECT_WRITE;
> > -
> > - r->presumed_offset = obj[SCRATCH].offset;
> > - r->target_handle = obj[SCRATCH].handle;
> > - r->offset = sizeof(uint32_t) * 1020;
> > - r->delta = 0;
> > - r->read_domains = I915_GEM_DOMAIN_RENDER;
> > - r->write_domain = I915_GEM_DOMAIN_RENDER;
> > + if (!(opts->flags & IGT_SPIN_SOFTDEP)) {
> > + obj[SCRATCH].flags = EXEC_OBJECT_WRITE;
> > +
> > + /* dummy write to dependency */
> > + r = &relocs[obj[BATCH].relocation_count++];
> > + r->presumed_offset = obj[SCRATCH].offset;
> > + r->target_handle = obj[SCRATCH].handle;
> > + r->offset = sizeof(uint32_t) * 1020;
> > + r->delta = 0;
> > + r->read_domains = I915_GEM_DOMAIN_RENDER;
> > + r->write_domain = I915_GEM_DOMAIN_RENDER;
> > + }
> >
> > execbuf->buffer_count++;
> > } else if (opts->flags & IGT_SPIN_POLL_RUN) {
> > diff --git a/lib/igt_dummyload.h b/lib/igt_dummyload.h
> > index 6d3e65ce2..b8baaa6b4 100644
> > --- a/lib/igt_dummyload.h
> > +++ b/lib/igt_dummyload.h
> > @@ -71,6 +71,7 @@ struct igt_spin_factory {
> > #define IGT_SPIN_NO_PREEMPTION (1 << 4)
> > #define IGT_SPIN_INVALID_CS (1 << 5)
> > #define IGT_SPIN_USERPTR (1 << 6)
> > +#define IGT_SPIN_SOFTDEP (1 << 7)
> >
> > igt_spin_t *
> > __igt_spin_factory(int fd, const struct igt_spin_factory *opts);
> > diff --git a/tests/i915/gem_request_retire.c b/tests/i915/gem_request_retire.c
> > index 31fb41987..c23ddfb7b 100644
> > --- a/tests/i915/gem_request_retire.c
> > +++ b/tests/i915/gem_request_retire.c
> > @@ -52,130 +52,6 @@
> > IGT_TEST_DESCRIPTION("Collection of tests targeting request retirement code"
> > " paths.");
> >
> > -#define WIDTH 4096
> > -#define HEIGHT 4096
> > -#define BO_SIZE (WIDTH * HEIGHT * sizeof(uint32_t))
> > -
> > -static uint32_t
> > -blit(int fd, uint32_t dst, uint32_t src, uint32_t ctx_id)
> > -{
> > - const unsigned int copies = 1000;
> > - uint32_t batch[12 * copies + 5];
> > - struct drm_i915_gem_relocation_entry reloc[2 * copies];
> > - struct drm_i915_gem_exec_object2 obj[3];
> > - struct drm_i915_gem_execbuffer2 exec;
> > - uint32_t handle;
> > - unsigned int i = 0, j, r = 0;
> > -
> > - for (j = 0; j < copies; j++) {
> > - reloc[r].target_handle = dst;
> > - reloc[r].delta = 0;
> > - reloc[r].offset = (i + 4) * sizeof(uint32_t);
> > - reloc[r].presumed_offset = 0;
> > - reloc[r].read_domains = I915_GEM_DOMAIN_RENDER;
> > - reloc[r].write_domain = I915_GEM_DOMAIN_RENDER;
> > -
> > - r++;
> > -
> > - reloc[r].target_handle = src;
> > - reloc[r].delta = 0;
> > - reloc[r].offset = (i + 7) * sizeof(uint32_t);
> > - if (intel_gen(intel_get_drm_devid(fd)) >= 8)
> > - reloc[r].offset += sizeof(uint32_t);
> > - reloc[r].presumed_offset = 0;
> > - reloc[r].read_domains = I915_GEM_DOMAIN_RENDER;
> > - reloc[r].write_domain = 0;
> > -
> > - r++;
> > -
> > - batch[i++] = XY_SRC_COPY_BLT_CMD |
> > - XY_SRC_COPY_BLT_WRITE_ALPHA |
> > - XY_SRC_COPY_BLT_WRITE_RGB;
> > - if (intel_gen(intel_get_drm_devid(fd)) >= 8)
> > - batch[i - 1] |= 8;
> > - else
> > - batch[i - 1] |= 6;
> > -
> > - batch[i++] = (3 << 24) | /* 32 bits */
> > - (0xcc << 16) | /* copy ROP */
> > - WIDTH*4;
> > - batch[i++] = 0; /* dst x1,y1 */
> > - batch[i++] = (HEIGHT << 16) | WIDTH; /* dst x2,y2 */
> > - batch[i++] = 0; /* dst reloc */
> > - if (intel_gen(intel_get_drm_devid(fd)) >= 8)
> > - batch[i++] = 0;
> > - batch[i++] = 0; /* src x1,y1 */
> > - batch[i++] = WIDTH*4;
> > - batch[i++] = 0; /* src reloc */
> > - if (intel_gen(intel_get_drm_devid(fd)) >= 8)
> > - batch[i++] = 0;
> > - }
> > -
> > - batch[i++] = MI_BATCH_BUFFER_END;
> > -
> > - while (i % 4)
> > - batch[i++] = MI_NOOP;
> > -
> > - handle = gem_create(fd, sizeof(batch));
> > - gem_write(fd, handle, 0, batch, sizeof(batch));
> > -
> > - memset(obj, 0, sizeof(obj));
> > - memset(&exec, 0, sizeof(exec));
> > -
> > - obj[exec.buffer_count++].handle = dst;
> > - if (src != dst)
> > - obj[exec.buffer_count++].handle = src;
> > - obj[exec.buffer_count].handle = handle;
> > - obj[exec.buffer_count].relocation_count = 2 * copies;
> > - obj[exec.buffer_count].relocs_ptr = to_user_pointer(reloc);
> > - exec.buffer_count++;
> > - exec.buffers_ptr = to_user_pointer(obj);
> > -
> > - exec.batch_len = i * sizeof(uint32_t);
> > - exec.flags = I915_EXEC_BLT;
> > - i915_execbuffer2_set_context_id(exec, ctx_id);
> > -
> > - gem_execbuf(fd, &exec);
> > -
> > - return handle;
> > -}
> > -
> > -static uint32_t
> > -noop(int fd, uint32_t src, uint32_t ctx_id)
> > -{
> > - uint32_t batch[4];
> > - struct drm_i915_gem_exec_object2 obj[2];
> > - struct drm_i915_gem_execbuffer2 exec;
> > - uint32_t handle;
> > - unsigned int i = 0;
> > -
> > - batch[i++] = MI_NOOP;
> > - batch[i++] = MI_BATCH_BUFFER_END;
> > - batch[i++] = MI_NOOP;
> > - batch[i++] = MI_NOOP;
> > -
> > - handle = gem_create(fd, 4096);
> > - gem_write(fd, handle, 0, batch, sizeof(batch));
> > -
> > - memset(obj, 0, sizeof(obj));
> > - memset(&exec, 0, sizeof(exec));
> > -
> > - obj[exec.buffer_count++].handle = src;
> > - obj[exec.buffer_count].handle = handle;
> > - obj[exec.buffer_count].relocation_count = 0;
> > - obj[exec.buffer_count].relocs_ptr = to_user_pointer(0);
> > - exec.buffer_count++;
> > - exec.buffers_ptr = to_user_pointer(obj);
> > -
> > - exec.batch_len = i * sizeof(uint32_t);
> > - exec.flags = I915_EXEC_RENDER;
> > - i915_execbuffer2_set_context_id(exec, ctx_id);
> > -
> > - gem_execbuf(fd, &exec);
> > -
> > - return handle;
> > -}
> > -
> > /*
> > * A single bo is operated from batchbuffers submitted from two contexts and on
> > * different rings.
> > @@ -185,37 +61,33 @@ noop(int fd, uint32_t src, uint32_t ctx_id)
> > static void
> > test_retire_vma_not_inactive(int fd)
> > {
> > - uint32_t ctx_id;
> > - uint32_t src, dst;
> > - uint32_t blit_bb, noop_bb;
> > -
> > - igt_require(HAS_BLT_RING(intel_get_drm_devid(fd)));
> > + struct intel_execution_engine2 *e;
> >
> > - ctx_id = gem_context_create(fd);
> > + igt_spin_t *bg = NULL;
> >
> > - /* Create some bos batch buffers will operate on. */
> > - src = gem_create(fd, BO_SIZE);
> > - dst = gem_create(fd, BO_SIZE);
> > + __for_each_physical_engine(fd, e) {
> > + igt_spin_t *spin;
> > + uint32_t ctx;
> >
> > - /* Submit a long running batch. */
> > - blit_bb = blit(fd, dst, src, 0);
> > + if (!bg) {
> > + bg = igt_spin_new(fd, .engine = e->flags);
> > + continue;
> > + }
> >
> > - /* Submit a quick batch referencing the same object. */
> > - noop_bb = noop(fd, src, ctx_id);
> > + ctx = gem_context_clone_with_engines(fd, 0);
> > + spin = igt_spin_new(fd, ctx,
> > + .engine = e->flags,
> > + .dependency = bg->handle,
> > + .flags = IGT_SPIN_SOFTDEP);
> > + gem_context_destroy(fd, ctx);
> > + igt_spin_end(spin);
> >
> > - /* Wait for the quick batch to complete. */
> > - gem_sync(fd, noop_bb);
> > - gem_close(fd, noop_bb);
> > -
> > - /* Now destroy the context in which the quick batch was submitted. */
> > - gem_context_destroy(fd, ctx_id);
> > -
> > - /* Wait for the slow batch to finish and clean up. */
> > - gem_sync(fd, blit_bb);
> > - gem_close(fd, blit_bb);
> > + gem_sync(fd, spin->handle);
> > + igt_spin_free(fd, spin);
> > + }
> >
> > - gem_close(fd, src);
> > - gem_close(fd, dst);
> > + igt_drop_caches_set(fd, DROP_RETIRE);
> > + igt_spin_free(fd, bg);
> > }
> >
> > int fd;
> >
>
> I wonder how relevant this test is today but anyway doesn't harm. It is
> not exactly like-for-like replacement but I *think* the key was context
> close before involved object is idled so I think is is fine.
Who can say which test will find the next unexpected bug :)
[kcov, kcov, kcov!]
It's mostly like for like; close before vma idle looked to be the goal
of the test, certainly worth poking at.
-Chris
More information about the Intel-gfx
mailing list