[Intel-gfx] [PATCH i-g-t] lib: Support writing arbitrary data from the start of a busy spinner
Chris Wilson
chris at chris-wilson.co.uk
Fri Dec 18 14:25:20 UTC 2020
Allow the caller to specify a dword, or an arbitrary payload, to be
written by the busy spinner, just prior to starting its infinite loop.
This is similar to the dependency method, that makes a target busy
without writing anything.
Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>
---
lib/igt_dummyload.c | 71 +++++++++++++++++++++++++++++++++++--
lib/igt_dummyload.h | 12 +++++++
tests/i915/gem_spin_batch.c | 46 ++++++++++++++++++++++++
3 files changed, 127 insertions(+), 2 deletions(-)
diff --git a/lib/igt_dummyload.c b/lib/igt_dummyload.c
index 28fcbf81f..d5a68a46c 100644
--- a/lib/igt_dummyload.c
+++ b/lib/igt_dummyload.c
@@ -63,7 +63,7 @@
#define MI_ARB_CHK (0x5 << 23)
static const int BATCH_SIZE = 4096;
-static const int LOOP_START_OFFSET = 64;
+static const int LOOP_START_OFFSET = 256;
static IGT_LIST_HEAD(spin_list);
static pthread_mutex_t list_lock = PTHREAD_MUTEX_INITIALIZER;
@@ -132,6 +132,11 @@ emit_recursive_batch(igt_spin_t *spin,
!gem_class_can_store_dword(fd, engine->class))
continue;
+ if (opts->flags & (IGT_SPIN_STORE_DWORD |
+ IGT_SPIN_STORE_DATA) &&
+ !gem_class_can_store_dword(fd, engine->class))
+ continue;
+
flags[nengine++] = engine->flags;
}
} else {
@@ -160,6 +165,8 @@ emit_recursive_batch(igt_spin_t *spin,
if (opts->dependency) {
igt_assert(!(opts->flags & IGT_SPIN_POLL_RUN));
+ igt_assert(!(opts->flags & (IGT_SPIN_STORE_DWORD |
+ IGT_SPIN_STORE_DATA)));
obj[SCRATCH].handle = opts->dependency;
obj[SCRATCH].offset = addr;
@@ -178,6 +185,9 @@ emit_recursive_batch(igt_spin_t *spin,
execbuf->buffer_count++;
} else if (opts->flags & IGT_SPIN_POLL_RUN) {
+ igt_assert(!(opts->flags & (IGT_SPIN_STORE_DWORD |
+ IGT_SPIN_STORE_DATA)));
+
r = &relocs[obj[BATCH].relocation_count++];
igt_assert(!opts->dependency);
@@ -230,6 +240,63 @@ emit_recursive_batch(igt_spin_t *spin,
*cs++ = 1;
+ execbuf->buffer_count++;
+ } else if (opts->flags & (IGT_SPIN_STORE_DWORD | IGT_SPIN_STORE_DATA)) {
+ int len, cmd;
+
+ igt_assert(opts->store_handle);
+ igt_assert((opts->store_offset & 3) == 0);
+ if (opts->flags & IGT_SPIN_STORE_DATA) {
+ igt_assert(!(opts->flags & IGT_SPIN_STORE_DWORD));
+ igt_assert(opts->store_length < LOOP_START_OFFSET - 16);
+ igt_assert(opts->store_data);
+ len = opts->store_length;
+ } else {
+ len = sizeof(uint32_t);
+ }
+
+ addr += 4096; /* guard page */
+ obj[SCRATCH].offset = addr;
+ obj[SCRATCH].handle = opts->store_handle;
+ obj[SCRATCH].flags = EXEC_OBJECT_WRITE;
+
+ r = &relocs[obj[BATCH].relocation_count++];
+ r->read_domains = I915_GEM_DOMAIN_RENDER;
+ r->write_domain = I915_GEM_DOMAIN_RENDER;
+
+ if (gen == 4 || gen == 5) {
+ execbuf->flags |= I915_EXEC_SECURE;
+ igt_require(__igt_device_set_master(fd) == 0);
+ }
+
+ r->presumed_offset = obj[SCRATCH].offset;
+ r->target_handle = obj[SCRATCH].handle;
+ r->offset = sizeof(uint32_t) * 1;
+ r->delta = opts->store_offset;
+
+ cmd = len / sizeof(uint32_t) + 1;
+ if (gen >= 4)
+ cmd++;
+ *cs++ = 0x20 << 23 | (gen < 6 ? 1 << 22 : 0) | cmd;
+
+ if (gen >= 8) {
+ *cs++ = r->presumed_offset + r->delta;
+ *cs++ = 0;
+ } else if (gen >= 4) {
+ *cs++ = 0;
+ *cs++ = r->presumed_offset + r->delta;
+ r->offset += sizeof(uint32_t);
+ } else {
+ *cs++ = r->presumed_offset + r->delta;
+ }
+
+ if (opts->flags & IGT_SPIN_STORE_DWORD) {
+ *cs++ = opts->store_dw;
+ } else {
+ memcpy(cs, opts->store_data, len);
+ cs += len / sizeof(*cs);
+ }
+
execbuf->buffer_count++;
}
@@ -258,7 +325,7 @@ emit_recursive_batch(igt_spin_t *spin,
* trouble. See https://bugs.freedesktop.org/show_bug.cgi?id=102262
*/
if (!(opts->flags & IGT_SPIN_FAST))
- cs += 960;
+ cs = spin->batch + 1000;
/*
* When using a cmdparser, the batch is copied into a read only location
diff --git a/lib/igt_dummyload.h b/lib/igt_dummyload.h
index 9e19ffabc..bc8e8ebb3 100644
--- a/lib/igt_dummyload.h
+++ b/lib/igt_dummyload.h
@@ -62,6 +62,16 @@ struct igt_spin_factory {
unsigned int engine;
unsigned int flags;
int fence;
+
+ uint32_t store_handle;
+ uint32_t store_offset;
+ union {
+ struct {
+ void *store_data;
+ uint32_t store_length;
+ };
+ uint32_t store_dw;
+ };
};
#define IGT_SPIN_FENCE_IN (1 << 0)
@@ -73,6 +83,8 @@ struct igt_spin_factory {
#define IGT_SPIN_INVALID_CS (1 << 6)
#define IGT_SPIN_USERPTR (1 << 7)
#define IGT_SPIN_SOFTDEP (1 << 8)
+#define IGT_SPIN_STORE_DWORD (1 << 9)
+#define IGT_SPIN_STORE_DATA (1 << 10)
igt_spin_t *
__igt_spin_factory(int fd, const struct igt_spin_factory *opts);
diff --git a/tests/i915/gem_spin_batch.c b/tests/i915/gem_spin_batch.c
index 19bc4638d..1ce865d6d 100644
--- a/tests/i915/gem_spin_batch.c
+++ b/tests/i915/gem_spin_batch.c
@@ -71,6 +71,45 @@ static void spin(int fd,
assert_within_epsilon(timeout_100ms * loops, elapsed, MAX_ERROR);
}
+static void store(int fd, const struct intel_execution_engine2 *e2)
+{
+ igt_spin_t *spin;
+ uint32_t handle;
+ uint32_t *map;
+ int i;
+
+ handle = gem_create(fd, 4096);
+ map = gem_mmap__device_coherent(fd, handle, 0, 4096, PROT_WRITE);
+
+ spin = igt_spin_new(fd,
+ .engine = e2->flags,
+ .store_handle = handle,
+ .store_dw = 0xdeadbeef,
+ .flags = IGT_SPIN_STORE_DWORD);
+ igt_spin_end(spin);
+ gem_sync(fd, spin->handle);
+ igt_spin_free(fd, spin);
+ igt_assert_eq(map[0], 0xdeadbeef);
+
+ for (i = 0; i < 16; i++)
+ map[i] = i;
+ spin = igt_spin_new(fd,
+ .engine = e2->flags,
+ .store_handle = handle,
+ .store_offset = sizeof(uint32_t),
+ .store_length = 16 * sizeof(uint32_t),
+ .store_data = map,
+ .flags = IGT_SPIN_STORE_DATA);
+ igt_spin_end(spin);
+ gem_sync(fd, spin->handle);
+ igt_spin_free(fd, spin);
+ for (i = 0; i < 16; i++)
+ igt_assert_eq(map[i + 1], i);
+
+ munmap(map, 4096);
+ gem_close(fd, handle);
+}
+
#define RESUBMIT_NEW_CTX (1 << 0)
#define RESUBMIT_ALL_ENGINES (1 << 1)
@@ -204,6 +243,13 @@ igt_main
igt_subtest("spin-all-new")
spin_all(fd, PARALLEL_SPIN_NEW_CTX);
+ igt_subtest_with_dynamic("store"){
+ __for_each_physical_engine(fd, e2) {
+ igt_dynamic_f("%s", e2->name)
+ store(fd, e2);
+ }
+ }
+
__for_each_physical_engine(fd, e2) {
igt_subtest_f("%s", e2->name)
spin(fd, e2, 0, 3);
--
2.29.2
More information about the Intel-gfx
mailing list