@@ -65,17 +65,6 @@ static const int BATCH_SIZE = 4096;
static IGT_LIST(spin_list);
static pthread_mutex_t list_lock = PTHREAD_MUTEX_INITIALIZER;
-static void
-fill_reloc(struct drm_i915_gem_relocation_entry *reloc,
- uint32_t gem_handle, uint32_t offset,
- uint32_t read_domains, uint32_t write_domains)
-{
- reloc->target_handle = gem_handle;
- reloc->offset = offset * sizeof(uint32_t);
- reloc->read_domains = read_domains;
- reloc->write_domain = write_domains;
-}
-
static int
emit_recursive_batch(igt_spin_t *spin,
int fd, const struct igt_spin_factory *opts)
@@ -128,15 +117,20 @@ 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;
- fill_reloc(&relocs[obj[BATCH].relocation_count++],
- opts->dependency, 1020,
- I915_GEM_DOMAIN_RENDER,
- I915_GEM_DOMAIN_RENDER);
+ r->presumed_offset = 0;
+ 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) {
- unsigned int offset;
+ r = &relocs[obj[BATCH].relocation_count++];
igt_assert(!opts->dependency);
@@ -146,39 +140,43 @@ emit_recursive_batch(igt_spin_t *spin,
}
spin->poll_handle = gem_create(fd, 4096);
+ obj[SCRATCH].handle = spin->poll_handle;
if (__gem_set_caching(fd, spin->poll_handle,
I915_CACHING_CACHED) == 0)
- spin->running = gem_mmap__cpu(fd, spin->poll_handle,
- 0, 4096,
- PROT_READ | PROT_WRITE);
+ spin->poll = gem_mmap__cpu(fd, spin->poll_handle,
+ 0, 4096,
+ PROT_READ | PROT_WRITE);
else
- spin->running = gem_mmap__wc(fd, spin->poll_handle,
- 0, 4096,
- PROT_READ | PROT_WRITE);
- igt_assert_eq(*spin->running, 0);
+ spin->poll = gem_mmap__wc(fd, spin->poll_handle,
+ 0, 4096,
+ PROT_READ | PROT_WRITE);
+
+ igt_assert_eq(spin->poll[SPIN_POLL_START_IDX], 0);
+
+ r->presumed_offset = 0;
+ r->target_handle = obj[SCRATCH].handle;
+ r->offset = sizeof(uint32_t) * 1;
+ r->delta = sizeof(uint32_t) * SPIN_POLL_START_IDX;
+ r->read_domains = I915_GEM_DOMAIN_INSTRUCTION;
+ r->write_domain = I915_GEM_DOMAIN_INSTRUCTION;
*batch++ = MI_STORE_DWORD_IMM | (gen < 6 ? 1 << 22 : 0);
if (gen >= 8) {
- offset = 1;
- *batch++ = 0;
+ *batch++ = r->delta;
*batch++ = 0;
} else if (gen >= 4) {
- offset = 2;
- *batch++ = 0;
*batch++ = 0;
+ *batch++ = r->delta;
+ r->offset += sizeof(uint32_t);
} else {
- offset = 1;
batch[-1]--;
- *batch++ = 0;
+ *batch++ = r->delta;
}
*batch++ = 1;
- obj[SCRATCH].handle = spin->poll_handle;
- fill_reloc(&relocs[obj[BATCH].relocation_count++],
- spin->poll_handle, offset, 0, 0);
execbuf->buffer_count++;
}
@@ -408,8 +406,8 @@ void igt_spin_batch_free(int fd, igt_spin_t *spin)
gem_munmap((void *)((unsigned long)spin->batch & (~4095UL)),
BATCH_SIZE);
- if (spin->running) {
- gem_munmap(spin->running, 4096);
+ if (spin->poll) {
+ gem_munmap(spin->poll, 4096);
gem_close(fd, spin->poll_handle);
}
@@ -41,7 +41,8 @@ typedef struct igt_spin {
struct drm_i915_gem_exec_object2 obj[2];
struct drm_i915_gem_execbuffer2 execbuf;
uint32_t poll_handle;
- bool *running;
+ uint32_t *poll;
+#define SPIN_POLL_START_IDX 0
} igt_spin_t;
struct igt_spin_factory {
@@ -70,9 +71,19 @@ void igt_spin_batch_set_timeout(igt_spin_t *spin, int64_t ns);
void igt_spin_batch_end(igt_spin_t *spin);
void igt_spin_batch_free(int fd, igt_spin_t *spin);
-static inline void igt_spin_busywait_until_running(igt_spin_t *spin)
+static inline bool igt_spin_has_poll(const igt_spin_t *spin)
{
- while (!READ_ONCE(*spin->running))
+ return spin->poll;
+}
+
+static inline bool igt_spin_has_started(igt_spin_t *spin)
+{
+ return READ_ONCE(spin->poll[SPIN_POLL_START_IDX]);
+}
+
+static inline void igt_spin_busywait_until_started(igt_spin_t *spin)
+{
+ while (!igt_spin_has_started(spin))
;
}
@@ -181,10 +181,8 @@ static void norecovery(int i915)
spin = __igt_spin_batch_new(i915,
.ctx = param.ctx_id,
.flags = IGT_SPIN_POLL_RUN);
- igt_assert(spin->running);
+ igt_spin_busywait_until_started(spin);
- while (!READ_ONCE(*spin->running))
- ;
igt_force_gpu_reset(i915);
igt_spin_batch_end(spin);
@@ -704,8 +704,8 @@ static void inject_reset_context(int fd, unsigned int engine)
spin = __igt_spin_batch_factory(fd, &opts);
- if (spin->running)
- igt_spin_busywait_until_running(spin);
+ if (igt_spin_has_poll(spin))
+ igt_spin_busywait_until_started(spin);
else
usleep(1000); /* better than nothing */
@@ -186,8 +186,8 @@ static igt_spin_t * __spin_poll(int fd, uint32_t ctx, unsigned long flags)
static void __spin_wait(int fd, igt_spin_t *spin)
{
- if (spin->running) {
- igt_spin_busywait_until_running(spin);
+ if (igt_spin_has_poll(spin)) {
+ igt_spin_busywait_until_started(spin);
} else {
igt_debug("__spin_wait - usleep mode\n");
usleep(500e3); /* Better than nothing! */
@@ -79,29 +79,29 @@ poll_ring(int fd, unsigned ring, const char *name)
igt_require(gem_can_store_dword(fd, ring));
spin[0] = __igt_spin_batch_factory(fd, &opts);
- igt_assert(spin[0]->running);
+ igt_assert(igt_spin_has_poll(spin[0]));
cmd = *spin[0]->batch;
spin[1] = __igt_spin_batch_factory(fd, &opts);
- igt_assert(spin[1]->running);
+ igt_assert(igt_spin_has_poll(spin[1]));
+
igt_assert(cmd == *spin[1]->batch);
igt_spin_batch_end(spin[0]);
- while (!READ_ONCE(*spin[1]->running))
- ;
+ igt_spin_busywait_until_started(spin[1]);
+
igt_assert(!gem_bo_busy(fd, spin[0]->handle));
cycles = 0;
while ((elapsed = igt_nsec_elapsed(&tv)) < 2ull << 30) {
- unsigned int idx = cycles++ & 1;
+ const unsigned int idx = cycles++ & 1;
*spin[idx]->batch = cmd;
- *spin[idx]->running = 0;
+ spin[idx]->poll[SPIN_POLL_START_IDX] = 0;
gem_execbuf(fd, &spin[idx]->execbuf);
igt_spin_batch_end(spin[!idx]);
- while (!READ_ONCE(*spin[idx]->running))
- ;
+ igt_spin_busywait_until_started(spin[idx]);
}
igt_info("%s completed %ld cycles: %.3f us\n",
@@ -419,7 +419,7 @@ static void __rearm_spin_batch(igt_spin_t *spin)
const uint32_t mi_arb_chk = 0x5 << 23;
*spin->batch = mi_arb_chk;
- *spin->running = 0;
+ spin->poll[SPIN_POLL_START_IDX] = 0;
__sync_synchronize();
}
@@ -441,7 +441,7 @@ struct rt_pkt {
static bool __spin_wait(int fd, igt_spin_t *spin)
{
- while (!READ_ONCE(*spin->running)) {
+ while (!igt_spin_has_started(spin)) {
if (!gem_bo_busy(fd, spin->handle))
return false;
}
@@ -537,7 +537,7 @@ rthog_latency_on_ring(int fd, unsigned int engine, const char *name, unsigned in
passname[pass]);
break;
}
- igt_spin_busywait_until_running(spin);
+ igt_spin_busywait_until_started(spin);
igt_until_timeout(pass > 0 ? 5 : 2) {
struct timespec ts = { };
@@ -436,7 +436,7 @@ static void semaphore_codependency(int i915)
.ctx = ctx,
.engine = engine,
.flags = IGT_SPIN_POLL_RUN);
- igt_spin_busywait_until_running(task[i].xcs);
+ igt_spin_busywait_until_started(task[i].xcs);
/* Common rcs tasks will be queued in FIFO */
task[i].rcs =
@@ -1361,8 +1361,7 @@ static void measure_semaphore_power(int i915)
.engine = signaler,
.flags = IGT_SPIN_POLL_RUN);
gem_wait(i915, spin->handle, &jiffie); /* waitboost */
- igt_assert(spin->running);
- igt_spin_busywait_until_running(spin);
+ igt_spin_busywait_until_started(spin);
gpu_power_read(&power, &s_spin[0]);
usleep(100*1000);
@@ -225,7 +225,7 @@ wakeup_ring(int fd, unsigned ring, int timeout, int wlen)
.engine = execbuf.flags,
.flags = (IGT_SPIN_POLL_RUN |
IGT_SPIN_FAST));
- igt_assert(spin->running);
+ igt_assert(igt_spin_has_poll(spin));
cmd = *spin->batch;
gem_execbuf(fd, &execbuf);
@@ -239,10 +239,9 @@ wakeup_ring(int fd, unsigned ring, int timeout, int wlen)
cycles = 0;
do {
*spin->batch = cmd;
- *spin->running = 0;
+ spin->poll[SPIN_POLL_START_IDX] = 0;
gem_execbuf(fd, &spin->execbuf);
- while (!READ_ONCE(*spin->running))
- ;
+ igt_spin_busywait_until_started(spin);
this = gettime();
igt_spin_batch_end(spin);
@@ -264,10 +263,9 @@ wakeup_ring(int fd, unsigned ring, int timeout, int wlen)
cycles = 0;
do {
*spin->batch = cmd;
- *spin->running = 0;
+ spin->poll[SPIN_POLL_START_IDX] = 0;
gem_execbuf(fd, &spin->execbuf);
- while (!READ_ONCE(*spin->running))
- ;
+ igt_spin_busywait_until_started(spin);
for (int n = 0; n < wlen; n++)
gem_execbuf(fd, &execbuf);
@@ -410,7 +408,7 @@ active_wakeup_ring(int fd, unsigned ring, int timeout, int wlen)
.engine = execbuf.flags,
.flags = (IGT_SPIN_POLL_RUN |
IGT_SPIN_FAST));
- igt_assert(spin[0]->running);
+ igt_assert(igt_spin_has_poll(spin[0]));
cmd = *spin[0]->batch;
spin[1] = __igt_spin_batch_new(fd,
@@ -426,18 +424,17 @@ active_wakeup_ring(int fd, unsigned ring, int timeout, int wlen)
for (int warmup = 0; warmup <= 1; warmup++) {
*spin[0]->batch = cmd;
- *spin[0]->running = 0;
+ spin[0]->poll[SPIN_POLL_START_IDX] = 0;
gem_execbuf(fd, &spin[0]->execbuf);
end = gettime() + timeout/10.;
elapsed = 0;
cycles = 0;
do {
- while (!READ_ONCE(*spin[0]->running))
- ;
+ igt_spin_busywait_until_started(spin[0]);
*spin[1]->batch = cmd;
- *spin[1]->running = 0;
+ spin[1]->poll[SPIN_POLL_START_IDX] = 0;
gem_execbuf(fd, &spin[1]->execbuf);
this = gettime();
@@ -458,21 +455,20 @@ active_wakeup_ring(int fd, unsigned ring, int timeout, int wlen)
cycles, elapsed*1e6/cycles);
*spin[0]->batch = cmd;
- *spin[0]->running = 0;
+ spin[0]->poll[SPIN_POLL_START_IDX] = 0;
gem_execbuf(fd, &spin[0]->execbuf);
end = gettime() + timeout;
elapsed = 0;
cycles = 0;
do {
- while (!READ_ONCE(*spin[0]->running))
- ;
+ igt_spin_busywait_until_started(spin[0]);
for (int n = 0; n < wlen; n++)
gem_execbuf(fd, &execbuf);
*spin[1]->batch = cmd;
- *spin[1]->running = 0;
+ spin[1]->poll[SPIN_POLL_START_IDX] = 0;
gem_execbuf(fd, &spin[1]->execbuf);
this = gettime();
@@ -189,10 +189,10 @@ static unsigned long __spin_wait(int fd, igt_spin_t *spin)
igt_nsec_elapsed(&start);
- if (spin->running) {
+ if (igt_spin_has_poll(spin)) {
unsigned long timeout = 0;
- while (!READ_ONCE(*spin->running)) {
+ while (!igt_spin_has_started(spin)) {
unsigned long t = igt_nsec_elapsed(&start);
if ((t - timeout) > 250e6) {
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) Cc: Chris Wilson <chris@chris-wilson.co.uk> Signed-off-by: Mika Kuoppala <mika.kuoppala@linux.intel.com> --- lib/igt_dummyload.c | 66 +++++++++++++++++----------------- lib/igt_dummyload.h | 17 +++++++-- tests/i915/gem_ctx_exec.c | 4 +-- tests/i915/gem_ctx_isolation.c | 4 +-- tests/i915/gem_eio.c | 4 +-- tests/i915/gem_exec_latency.c | 22 ++++++------ tests/i915/gem_exec_schedule.c | 5 ++- tests/i915/gem_sync.c | 28 +++++++-------- tests/perf_pmu.c | 4 +-- 9 files changed, 78 insertions(+), 76 deletions(-)