[Intel-gfx] [PATCH i-g-t 1/2] lib/i915: Report scheduler caps for timeslicing

[Intel-gfx] [PATCH i-g-t 1/2] lib/i915: Report scheduler caps for timeslicing

[Chris Wilson]

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
---
 include/drm-uapi/i915_drm.h |  8 +++++---
 lib/i915/gem_scheduler.c    | 15 +++++++++++++++
 lib/i915/gem_scheduler.h    |  1 +
 3 files changed, 21 insertions(+), 3 deletions(-)

diff --git a/include/drm-uapi/i915_drm.h b/include/drm-uapi/i915_drm.h
index 2b55af13a..a222b6bfb 100644
--- a/include/drm-uapi/i915_drm.h
+++ b/include/drm-uapi/i915_drm.h
@@ -523,6 +523,7 @@ typedef struct drm_i915_irq_wait {
 #define   I915_SCHEDULER_CAP_PREEMPTION	(1ul << 2)
 #define   I915_SCHEDULER_CAP_SEMAPHORES	(1ul << 3)
 #define   I915_SCHEDULER_CAP_ENGINE_BUSY_STATS	(1ul << 4)
+#define   I915_SCHEDULER_CAP_TIMESLICING	(1ul << 5)
 
 #define I915_PARAM_HUC_STATUS		 42
 
@@ -1040,9 +1041,10 @@ struct drm_i915_gem_exec_fence {
 	 */
 	__u32 handle;
 
-#define I915_EXEC_FENCE_WAIT            (1<<0)
-#define I915_EXEC_FENCE_SIGNAL          (1<<1)
-#define __I915_EXEC_FENCE_UNKNOWN_FLAGS (-(I915_EXEC_FENCE_SIGNAL << 1))
+#define I915_EXEC_FENCE_WAIT            (1u << 0)
+#define I915_EXEC_FENCE_SIGNAL          (1u << 1)
+#define I915_EXEC_FENCE_WAIT_SUBMIT     (1u << 2)
+#define __I915_EXEC_FENCE_UNKNOWN_FLAGS (-(I915_EXEC_FENCE_WAIT_SUBMIT << 1))
 	__u32 flags;
 };
 
diff --git a/lib/i915/gem_scheduler.c b/lib/i915/gem_scheduler.c
index 1beb85dec..a1dc694e5 100644
--- a/lib/i915/gem_scheduler.c
+++ b/lib/i915/gem_scheduler.c
@@ -131,6 +131,19 @@ bool gem_scheduler_has_engine_busy_stats(int fd)
 		I915_SCHEDULER_CAP_ENGINE_BUSY_STATS;
 }
 
+/**
+ * gem_scheduler_has_timeslicing:
+ * @fd: open i915 drm file descriptor
+ *
+ * Feature test macro to query whether the driver supports using HW preemption
+ * to implement timeslicing of userspace batches. This allows userspace to
+ * implement micro-level scheduling within their own batches.
+ */
+bool gem_scheduler_has_timeslicing(int fd)
+{
+	return gem_scheduler_capability(fd) & I915_SCHEDULER_CAP_TIMESLICING;
+}
+
 /**
  * gem_scheduler_print_capability:
  * @fd: open i915 drm file descriptor
@@ -151,6 +164,8 @@ void gem_scheduler_print_capability(int fd)
 		igt_info(" - With preemption enabled\n");
 	if (caps & I915_SCHEDULER_CAP_SEMAPHORES)
 		igt_info(" - With HW semaphores enabled\n");
+	if (caps & I915_SCHEDULER_CAP_TIMESLICING)
+		igt_info(" - With user timeslicing enabled\n");
 	if (caps & I915_SCHEDULER_CAP_ENGINE_BUSY_STATS)
 		igt_info(" - With engine busy statistics\n");
 }
diff --git a/lib/i915/gem_scheduler.h b/lib/i915/gem_scheduler.h
index 14bd4cac4..d43e84bd2 100644
--- a/lib/i915/gem_scheduler.h
+++ b/lib/i915/gem_scheduler.h
@@ -32,6 +32,7 @@ bool gem_scheduler_has_ctx_priority(int fd);
 bool gem_scheduler_has_preemption(int fd);
 bool gem_scheduler_has_semaphores(int fd);
 bool gem_scheduler_has_engine_busy_stats(int fd);
+bool gem_scheduler_has_timeslicing(int fd);
 void gem_scheduler_print_capability(int fd);
 
 #endif /* GEM_SCHEDULER_H */
-- 
2.26.2

_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

[igt-dev] [PATCH i-g-t 1/2] lib/i915: Report scheduler caps for timeslicing

[Chris Wilson]

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
---
 include/drm-uapi/i915_drm.h |  8 +++++---
 lib/i915/gem_scheduler.c    | 15 +++++++++++++++
 lib/i915/gem_scheduler.h    |  1 +
 3 files changed, 21 insertions(+), 3 deletions(-)

diff --git a/include/drm-uapi/i915_drm.h b/include/drm-uapi/i915_drm.h
index 2b55af13a..a222b6bfb 100644
--- a/include/drm-uapi/i915_drm.h
+++ b/include/drm-uapi/i915_drm.h
@@ -523,6 +523,7 @@ typedef struct drm_i915_irq_wait {
 #define   I915_SCHEDULER_CAP_PREEMPTION	(1ul << 2)
 #define   I915_SCHEDULER_CAP_SEMAPHORES	(1ul << 3)
 #define   I915_SCHEDULER_CAP_ENGINE_BUSY_STATS	(1ul << 4)
+#define   I915_SCHEDULER_CAP_TIMESLICING	(1ul << 5)
 
 #define I915_PARAM_HUC_STATUS		 42
 
@@ -1040,9 +1041,10 @@ struct drm_i915_gem_exec_fence {
 	 */
 	__u32 handle;
 
-#define I915_EXEC_FENCE_WAIT            (1<<0)
-#define I915_EXEC_FENCE_SIGNAL          (1<<1)
-#define __I915_EXEC_FENCE_UNKNOWN_FLAGS (-(I915_EXEC_FENCE_SIGNAL << 1))
+#define I915_EXEC_FENCE_WAIT            (1u << 0)
+#define I915_EXEC_FENCE_SIGNAL          (1u << 1)
+#define I915_EXEC_FENCE_WAIT_SUBMIT     (1u << 2)
+#define __I915_EXEC_FENCE_UNKNOWN_FLAGS (-(I915_EXEC_FENCE_WAIT_SUBMIT << 1))
 	__u32 flags;
 };
 
diff --git a/lib/i915/gem_scheduler.c b/lib/i915/gem_scheduler.c
index 1beb85dec..a1dc694e5 100644
--- a/lib/i915/gem_scheduler.c
+++ b/lib/i915/gem_scheduler.c
@@ -131,6 +131,19 @@ bool gem_scheduler_has_engine_busy_stats(int fd)
 		I915_SCHEDULER_CAP_ENGINE_BUSY_STATS;
 }
 
+/**
+ * gem_scheduler_has_timeslicing:
+ * @fd: open i915 drm file descriptor
+ *
+ * Feature test macro to query whether the driver supports using HW preemption
+ * to implement timeslicing of userspace batches. This allows userspace to
+ * implement micro-level scheduling within their own batches.
+ */
+bool gem_scheduler_has_timeslicing(int fd)
+{
+	return gem_scheduler_capability(fd) & I915_SCHEDULER_CAP_TIMESLICING;
+}
+
 /**
  * gem_scheduler_print_capability:
  * @fd: open i915 drm file descriptor
@@ -151,6 +164,8 @@ void gem_scheduler_print_capability(int fd)
 		igt_info(" - With preemption enabled\n");
 	if (caps & I915_SCHEDULER_CAP_SEMAPHORES)
 		igt_info(" - With HW semaphores enabled\n");
+	if (caps & I915_SCHEDULER_CAP_TIMESLICING)
+		igt_info(" - With user timeslicing enabled\n");
 	if (caps & I915_SCHEDULER_CAP_ENGINE_BUSY_STATS)
 		igt_info(" - With engine busy statistics\n");
 }
diff --git a/lib/i915/gem_scheduler.h b/lib/i915/gem_scheduler.h
index 14bd4cac4..d43e84bd2 100644
--- a/lib/i915/gem_scheduler.h
+++ b/lib/i915/gem_scheduler.h
@@ -32,6 +32,7 @@ bool gem_scheduler_has_ctx_priority(int fd);
 bool gem_scheduler_has_preemption(int fd);
 bool gem_scheduler_has_semaphores(int fd);
 bool gem_scheduler_has_engine_busy_stats(int fd);
+bool gem_scheduler_has_timeslicing(int fd);
 void gem_scheduler_print_capability(int fd);
 
 #endif /* GEM_SCHEDULER_H */
-- 
2.26.2

_______________________________________________
igt-dev mailing list
igt-dev@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/igt-dev

[Intel-gfx] [PATCH i-g-t 2/2] i915/gem_exec_fence: Teach invalid-wait about invalid future fences

[Chris Wilson]

When we allow a wait on a future future fence, it must autoexpire if the
fence is never signaled by userspace. Also put future fences to work, as
the intention is to use them, along with WAIT_SUBMIT and semaphores, for
userspace to perform its own fine-grained scheduling. Or simply run
concurrent clients without having to flush batches between context
switches.

v2: Verify deadlock detection

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
---
 tests/i915/gem_exec_fence.c | 558 +++++++++++++++++++++++++++++++++++-
 1 file changed, 555 insertions(+), 3 deletions(-)

diff --git a/tests/i915/gem_exec_fence.c b/tests/i915/gem_exec_fence.c
index 4b0d87e4d..e51b7452e 100644
--- a/tests/i915/gem_exec_fence.c
+++ b/tests/i915/gem_exec_fence.c
@@ -46,6 +46,15 @@ struct sync_merge_data {
 #define SYNC_IOC_MERGE _IOWR(SYNC_IOC_MAGIC, 3, struct sync_merge_data)
 #endif
 
+#define MI_SEMAPHORE_WAIT		(0x1c << 23)
+#define   MI_SEMAPHORE_POLL             (1 << 15)
+#define   MI_SEMAPHORE_SAD_GT_SDD       (0 << 12)
+#define   MI_SEMAPHORE_SAD_GTE_SDD      (1 << 12)
+#define   MI_SEMAPHORE_SAD_LT_SDD       (2 << 12)
+#define   MI_SEMAPHORE_SAD_LTE_SDD      (3 << 12)
+#define   MI_SEMAPHORE_SAD_EQ_SDD       (4 << 12)
+#define   MI_SEMAPHORE_SAD_NEQ_SDD      (5 << 12)
+
 static void store(int fd, const struct intel_execution_engine2 *e,
 		  int fence, uint32_t target, unsigned offset_value)
 {
@@ -907,11 +916,12 @@ static void test_syncobj_invalid_wait(int fd)
 	struct drm_i915_gem_exec_fence fence = {
 		.handle = syncobj_create(fd, 0),
 	};
+	int out;
 
 	memset(&execbuf, 0, sizeof(execbuf));
 	execbuf.buffers_ptr = to_user_pointer(&obj);
 	execbuf.buffer_count = 1;
-	execbuf.flags = I915_EXEC_FENCE_ARRAY;
+	execbuf.flags = I915_EXEC_FENCE_ARRAY | I915_EXEC_FENCE_OUT;
 	execbuf.cliprects_ptr = to_user_pointer(&fence);
 	execbuf.num_cliprects = 1;
 
@@ -919,14 +929,59 @@ static void test_syncobj_invalid_wait(int fd)
 	obj.handle = gem_create(fd, 4096);
 	gem_write(fd, obj.handle, 0, &bbe, sizeof(bbe));
 
-	/* waiting before the fence is set is invalid */
+	/* waiting before the fence is set is^W may be invalid */
 	fence.flags = I915_EXEC_FENCE_WAIT;
-	igt_assert_eq(__gem_execbuf(fd, &execbuf), -EINVAL);
+	if (__gem_execbuf_wr(fd, &execbuf)) {
+		igt_assert_eq(__gem_execbuf(fd, &execbuf), -EINVAL);
+		return;
+	}
+
+	/* If we do allow the wait on a future fence, it should autoexpire */
+	gem_sync(fd, obj.handle);
+	out = execbuf.rsvd2 >> 32;
+	igt_assert_eq(sync_fence_status(out), -ETIMEDOUT);
+	close(out);
 
 	gem_close(fd, obj.handle);
 	syncobj_destroy(fd, fence.handle);
 }
 
+static void test_syncobj_incomplete_wait_submit(int i915)
+{
+	struct drm_i915_gem_exec_object2 obj = {
+		.handle = batch_create(i915),
+	};
+	struct drm_i915_gem_exec_fence fence = {
+		.handle = syncobj_create(i915, 0),
+		.flags = I915_EXEC_FENCE_WAIT | I915_EXEC_FENCE_WAIT_SUBMIT,
+	};
+	struct drm_i915_gem_execbuffer2 execbuf = {
+		.buffers_ptr = to_user_pointer(&obj),
+		.buffer_count = 1,
+
+		.cliprects_ptr = to_user_pointer(&fence),
+		.num_cliprects = 1,
+
+		.flags = I915_EXEC_FENCE_ARRAY | I915_EXEC_FENCE_OUT,
+	};
+	int out;
+
+	/* waiting before the fence is set is^W may be invalid */
+	if (__gem_execbuf_wr(i915, &execbuf)) {
+		igt_assert_eq(__gem_execbuf(i915, &execbuf), -EINVAL);
+		return;
+	}
+
+	/* If we do allow the wait on a future fence, it should autoexpire */
+	gem_sync(i915, obj.handle);
+	out = execbuf.rsvd2 >> 32;
+	igt_assert_eq(sync_fence_status(out), -ETIMEDOUT);
+	close(out);
+
+	gem_close(i915, obj.handle);
+	syncobj_destroy(i915, fence.handle);
+}
+
 static void test_syncobj_invalid_flags(int fd)
 {
 	const uint32_t bbe = MI_BATCH_BUFFER_END;
@@ -1073,6 +1128,398 @@ static void test_syncobj_wait(int fd)
 	}
 }
 
+static uint32_t future_batch(int i915, uint32_t offset)
+{
+	uint32_t handle = gem_create(i915, 4096);
+	const int gen = intel_gen(intel_get_drm_devid(i915));
+	uint32_t cs[16];
+	int i = 0;
+
+	cs[i] = MI_STORE_DWORD_IMM | (gen < 6 ? 1 << 22 : 0);
+	if (gen >= 8) {
+		cs[++i] = offset + 4000;
+		cs[++i] = 0;
+	} else if (gen >= 4) {
+		cs[++i] = 0;
+		cs[++i] = offset + 4000;
+	} else {
+		cs[i]--;
+		cs[++i] = offset + 4000;
+	}
+	cs[++i] = 1;
+	cs[i + 1] = MI_BATCH_BUFFER_END;
+	gem_write(i915, handle, 0, cs, sizeof(cs));
+
+	cs[i] = 2;
+	gem_write(i915, handle, 64, cs, sizeof(cs));
+
+	return handle;
+}
+
+static void test_syncobj_future(int i915, unsigned int engine)
+{
+	struct drm_i915_gem_exec_object2 obj = {
+		.offset = 24 << 20,
+		.handle = future_batch(i915, 24 << 20),
+		.flags = EXEC_OBJECT_PINNED,
+	};
+	struct drm_i915_gem_exec_fence fence = {};
+	struct drm_i915_gem_execbuffer2 execbuf  = {
+		.buffers_ptr = to_user_pointer(&obj),
+		.buffer_count = 1,
+		.cliprects_ptr = to_user_pointer(&fence),
+		.num_cliprects = 1,
+	};
+	const struct intel_execution_engine2 *e;
+
+	/*
+	 * Client A is waiting on a future fence from B. So even though its
+	 * execbuf is called first, we need to hold it in a queue waiting on
+	 * B.
+	 */
+	igt_require(gem_scheduler_enabled(i915));
+
+	__for_each_physical_engine(i915, e) {
+		uint32_t result;
+
+		igt_debug("waiting on future %s\n", e->name);
+		fence.handle = syncobj_create(i915, 0);
+
+		fence.flags = I915_EXEC_FENCE_WAIT;
+		execbuf.batch_start_offset = 0;
+		execbuf.flags = engine | I915_EXEC_FENCE_ARRAY;
+		execbuf.rsvd1 = 0;
+		gem_execbuf(i915, &execbuf); /* writes 1 */
+
+		fence.flags = I915_EXEC_FENCE_SIGNAL;
+		execbuf.batch_start_offset = 64;
+		execbuf.flags = e->flags | I915_EXEC_FENCE_ARRAY;
+		execbuf.rsvd1 = gem_context_clone_with_engines(i915, 0);
+		gem_execbuf(i915, &execbuf); /* writes 2 */
+		gem_context_destroy(i915, execbuf.rsvd1);
+
+		syncobj_destroy(i915, fence.handle);
+		gem_sync(i915, obj.handle); /* write hazard lies */
+		gem_read(i915, obj.handle, 4000, &result, sizeof(result));
+		igt_assert_eq(result, 1);
+	}
+
+	gem_close(i915, obj.handle);
+}
+
+static uint32_t future_submit_batch(int i915, uint32_t offset)
+{
+	uint32_t handle = gem_create(i915, 4096);
+	const int gen = intel_gen(intel_get_drm_devid(i915));
+	uint32_t cs[16];
+	int i = 0;
+
+	cs[i] = MI_STORE_DWORD_IMM | (gen < 6 ? 1 << 22 : 0);
+	if (gen >= 8) {
+		cs[++i] = offset + 4000;
+		cs[++i] = 0;
+	} else if (gen >= 4) {
+		cs[++i] = 0;
+		cs[++i] = offset + 4000;
+	} else {
+		cs[i]--;
+		cs[++i] = offset + 4000;
+	}
+	cs[++i] = 1;
+	cs[i + 1] = MI_BATCH_BUFFER_END;
+	igt_assert(i + 1 < ARRAY_SIZE(cs));
+	gem_write(i915, handle, 0, cs, sizeof(cs));
+
+	i = 0;
+	cs[i++] =
+		MI_SEMAPHORE_WAIT |
+		MI_SEMAPHORE_POLL |
+		MI_SEMAPHORE_SAD_EQ_SDD |
+		(4 - 2);
+	cs[i++] = 1;
+	cs[i++] = offset + 4000;
+	cs[i++] = 0;
+
+	cs[i] = MI_STORE_DWORD_IMM | (gen < 6 ? 1 << 22 : 0);
+	if (gen >= 8) {
+		cs[++i] = offset + 4000;
+		cs[++i] = 0;
+	} else if (gen >= 4) {
+		cs[++i] = 0;
+		cs[++i] = offset + 4000;
+	} else {
+		cs[i]--;
+		cs[++i] = offset + 4000;
+	}
+	cs[++i] = 2;
+	cs[++i] = MI_BATCH_BUFFER_END;
+	igt_assert(i < ARRAY_SIZE(cs));
+
+	gem_write(i915, handle, 64, cs, sizeof(cs));
+
+	return handle;
+}
+
+static void test_syncobj_future_submit(int i915, unsigned int engine)
+{
+	struct drm_i915_gem_exec_object2 obj = {
+		.offset = 24 << 20,
+		.handle = future_submit_batch(i915, 24 << 20),
+		.flags = EXEC_OBJECT_PINNED,
+	};
+	const struct intel_execution_engine2 *e;
+
+	/*
+	 * Here we submit client A waiting on client B, but internally client
+	 * B has a semaphore that waits on client A. This relies on timeslicing
+	 * to reorder B before A, even though userspace has asked to submit
+	 * A & B simultaneously (and due to the sequence we will submit B
+	 * then A).
+	 */
+	igt_require(gem_scheduler_has_timeslicing(i915));
+
+	__for_each_physical_engine(i915, e) {
+		struct drm_i915_gem_exec_fence fence = {
+			.handle = syncobj_create(i915, 0),
+		};
+		struct drm_i915_gem_execbuffer2 execbuf  = {
+			.buffers_ptr = to_user_pointer(&obj),
+			.buffer_count = 1,
+			.cliprects_ptr = to_user_pointer(&fence),
+			.num_cliprects = 1,
+			.flags = engine | I915_EXEC_FENCE_ARRAY,
+		};
+		uint32_t result;
+		int out;
+
+		igt_debug("waiting on future %s\n", e->name);
+
+		execbuf.rsvd1 = gem_context_clone_with_engines(i915, 0);
+		fence.flags = I915_EXEC_FENCE_WAIT | I915_EXEC_FENCE_WAIT_SUBMIT;
+		execbuf.batch_start_offset = 0;
+		execbuf.flags |= I915_EXEC_FENCE_OUT;
+		igt_require(__gem_execbuf_wr(i915, &execbuf) == 0); /* writes 1 */
+		execbuf.flags &= ~I915_EXEC_FENCE_OUT;
+		gem_context_destroy(i915, execbuf.rsvd1);
+
+		execbuf.rsvd1 = gem_context_clone_with_engines(i915, 0);
+		fence.flags = I915_EXEC_FENCE_SIGNAL;
+		execbuf.batch_start_offset = 64;
+		execbuf.flags &= ~I915_EXEC_RING_MASK;
+		execbuf.flags |= e->flags;
+		gem_execbuf(i915, &execbuf); /* writes 2 */
+		gem_context_destroy(i915, execbuf.rsvd1);
+
+		syncobj_destroy(i915, fence.handle);
+		gem_sync(i915, obj.handle); /* write hazard lies */
+		gem_read(i915, obj.handle, 4000, &result, sizeof(result));
+		igt_assert_eq(result, 2);
+
+		/* check we didn't autotimeout */
+		out = execbuf.rsvd2 >> 32;
+		igt_assert_eq(sync_fence_status(out), 1);
+		close(out);
+	}
+
+	gem_close(i915, obj.handle);
+}
+
+static void test_syncobj_future_past(int i915, unsigned int engine, int flags)
+{
+	struct drm_i915_gem_exec_object2 obj = {
+		.offset = 24 << 20,
+		.handle = future_batch(i915, 24 << 20),
+		.flags = EXEC_OBJECT_PINNED,
+	};
+	struct drm_i915_gem_exec_fence fence = {
+		.handle = syncobj_create(i915, 0),
+	};
+	struct drm_i915_gem_execbuffer2 execbuf  = {
+		.buffers_ptr = to_user_pointer(&obj),
+		.buffer_count = 1,
+		.cliprects_ptr = to_user_pointer(&fence),
+		.num_cliprects = 1,
+		.flags = engine | I915_EXEC_FENCE_ARRAY,
+	};
+	uint32_t result;
+	int out;
+
+	fence.flags = I915_EXEC_FENCE_WAIT | I915_EXEC_FENCE_SIGNAL | flags;
+	execbuf.batch_start_offset = 0;
+	execbuf.flags |= I915_EXEC_FENCE_OUT;
+	igt_require(__gem_execbuf_wr(i915, &execbuf) == 0); /* writes 1 */
+	execbuf.flags &= ~I915_EXEC_FENCE_OUT;
+
+	gem_sync(i915, obj.handle); /* write hazard lies */
+	gem_read(i915, obj.handle, 4000, &result, sizeof(result));
+	igt_assert_eq(result, 1);
+
+	/* check we didn't autotimeout */
+	out = execbuf.rsvd2 >> 32;
+	igt_assert_eq(sync_fence_status(out), 1);
+	close(out);
+
+	gem_close(i915, obj.handle);
+	syncobj_destroy(i915, fence.handle);
+}
+
+static void test_syncobj_future_self(int i915, unsigned int engine, int flags)
+{
+	struct drm_i915_gem_exec_object2 obj = {
+		.offset = 24 << 20,
+		.handle = future_batch(i915, 24 << 20),
+		.flags = EXEC_OBJECT_PINNED,
+	};
+	struct drm_i915_gem_exec_fence fence = {
+		.handle = syncobj_create(i915, 0),
+	};
+	struct drm_i915_gem_execbuffer2 execbuf  = {
+		.buffers_ptr = to_user_pointer(&obj),
+		.buffer_count = 1,
+		.cliprects_ptr = to_user_pointer(&fence),
+		.num_cliprects = 1,
+		.flags = engine | I915_EXEC_FENCE_ARRAY,
+	};
+	int out;
+
+	fence.flags = I915_EXEC_FENCE_WAIT | flags;
+	execbuf.batch_start_offset = 0;
+	execbuf.flags |= I915_EXEC_FENCE_OUT;
+	igt_require(__gem_execbuf_wr(i915, &execbuf) == 0); /* writes 1 */
+	execbuf.flags &= ~I915_EXEC_FENCE_OUT;
+
+	fence.flags = I915_EXEC_FENCE_SIGNAL;
+	execbuf.batch_start_offset = 64;
+	gem_execbuf(i915, &execbuf); /* writes 2 */
+
+	gem_sync(i915, obj.handle); /* write hazard lies */
+
+	gem_close(i915, obj.handle);
+	syncobj_destroy(i915, fence.handle);
+
+	out = execbuf.rsvd2 >> 32;
+	igt_assert_eq(sync_fence_status(out), -EDEADLK);
+	close(out);
+}
+
+static void
+test_syncobj_future_deadlock(int i915, unsigned int engine, int flags)
+{
+	struct drm_i915_gem_exec_object2 obj = {
+		.offset = 24 << 20,
+		.handle = future_batch(i915, 24 << 20),
+		.flags = EXEC_OBJECT_PINNED,
+	};
+	struct drm_i915_gem_exec_fence fence = {};
+	struct drm_i915_gem_execbuffer2 execbuf  = {
+		.buffers_ptr = to_user_pointer(&obj),
+		.buffer_count = 1,
+		.cliprects_ptr = to_user_pointer(&fence),
+		.num_cliprects = 1,
+	};
+	const struct intel_execution_engine2 *e;
+
+	__for_each_physical_engine(i915, e) {
+		int out;
+
+		fence.handle = syncobj_create(i915, 0),
+
+		fence.flags = I915_EXEC_FENCE_WAIT | flags;
+		execbuf.batch_start_offset = 0;
+		execbuf.flags = engine | I915_EXEC_FENCE_ARRAY,
+		execbuf.flags |= I915_EXEC_FENCE_OUT;
+		execbuf.rsvd1 = 0;
+		gem_execbuf_wr(i915, &execbuf); /* writes 1 */
+
+		fence.flags = I915_EXEC_FENCE_SIGNAL;
+		execbuf.batch_start_offset = 64;
+		execbuf.flags = e->flags | I915_EXEC_FENCE_ARRAY,
+		execbuf.flags |= I915_EXEC_FENCE_OUT | I915_EXEC_FENCE_IN;
+		execbuf.rsvd1 = gem_context_clone_with_engines(i915, 0);
+		execbuf.rsvd2 >>= 32;
+		gem_execbuf_wr(i915, &execbuf); /* writes 2 */
+		gem_context_destroy(i915, execbuf.rsvd1);
+
+		syncobj_destroy(i915, fence.handle);
+		gem_sync(i915, obj.handle);
+
+		/* How should this deadlock be resolved? */
+		out = execbuf.rsvd2;
+		igt_assert_eq(sync_fence_status(out), -EDEADLK);
+		close(out);
+
+		out = execbuf.rsvd2 >> 32;
+		igt_assert_eq(sync_fence_status(out), -EDEADLK);
+		close(out);
+	}
+
+	gem_close(i915, obj.handle);
+}
+
+static void
+test_syncobj_future_cycle(int i915, unsigned int engine, int flags)
+{
+	struct drm_i915_gem_exec_object2 obj = {
+		.offset = 24 << 20,
+		.handle = future_batch(i915, 24 << 20),
+		.flags = EXEC_OBJECT_PINNED,
+	};
+	const struct intel_execution_engine2 *e1, *e2;
+
+	__for_each_physical_engine(i915, e1) {
+	__for_each_physical_engine(i915, e2) {
+		struct drm_i915_gem_exec_fence fence = {
+			.handle = syncobj_create(i915, 0),
+		};
+		struct drm_i915_gem_execbuffer2 execbuf  = {
+			.buffers_ptr = to_user_pointer(&obj),
+			.buffer_count = 1,
+			.cliprects_ptr = to_user_pointer(&fence),
+			.num_cliprects = 1,
+			.flags = engine | I915_EXEC_FENCE_ARRAY,
+		};
+		int out;
+
+		fence.flags = I915_EXEC_FENCE_WAIT | flags;
+		execbuf.batch_start_offset = 0;
+		execbuf.flags |= I915_EXEC_FENCE_OUT;
+		igt_require(__gem_execbuf_wr(i915, &execbuf) == 0);
+
+		fence.flags = 0;
+		execbuf.rsvd1 = gem_context_clone_with_engines(i915, 0);
+		execbuf.rsvd2 >>= 32;
+		execbuf.flags &= ~I915_EXEC_RING_MASK;
+		execbuf.flags |= e1->flags | I915_EXEC_FENCE_IN;
+		gem_execbuf_wr(i915, &execbuf);
+		gem_context_destroy(i915, execbuf.rsvd1);
+		close(execbuf.rsvd2);
+
+		fence.flags = I915_EXEC_FENCE_SIGNAL;
+		execbuf.rsvd1 = gem_context_clone_with_engines(i915, 0);
+		execbuf.rsvd2 >>= 32;
+		execbuf.flags &= ~I915_EXEC_RING_MASK;
+		execbuf.flags |= e2->flags;
+		execbuf.batch_start_offset = 64;
+		gem_execbuf_wr(i915, &execbuf); /* writes 2 */
+		gem_context_destroy(i915, execbuf.rsvd1);
+
+		syncobj_destroy(i915, fence.handle);
+		gem_sync(i915, obj.handle);
+
+
+		/* How should this deadlock be resolved? */
+		out = execbuf.rsvd2 >> 32;
+		igt_assert_eq(sync_fence_status(out), -EDEADLK);
+		close(out);
+
+		out = execbuf.rsvd2;
+		igt_assert_eq(sync_fence_status(out), -EDEADLK);
+		close(out);
+	}}
+
+	gem_close(i915, obj.handle);
+}
+
 static void test_syncobj_export(int fd)
 {
 	const uint32_t bbe = MI_BATCH_BUFFER_END;
@@ -1358,6 +1805,105 @@ static void test_syncobj_channel(int fd)
 		syncobj_destroy(fd, syncobj[i]);
 }
 
+static bool has_future_syncobj(int i915)
+{
+	struct drm_i915_gem_exec_object2 obj = {
+		.handle = batch_create(i915),
+	};
+	struct drm_i915_gem_exec_fence fence = {
+		.handle = syncobj_create(i915, 0),
+		.flags = I915_EXEC_FENCE_WAIT | I915_EXEC_FENCE_SIGNAL,
+	};
+	struct drm_i915_gem_execbuffer2 execbuf  = {
+		.buffers_ptr = to_user_pointer(&obj),
+		.buffer_count = 1,
+		.cliprects_ptr = to_user_pointer(&fence),
+		.num_cliprects = 1,
+		.flags = I915_EXEC_FENCE_ARRAY,
+	};
+	bool result;
+
+	result = __gem_execbuf(i915, &execbuf) == 0;
+	gem_close(i915, obj.handle);
+
+	return result;
+}
+
+static void syncobj_futures(int i915)
+{
+	const struct intel_execution_engine2 *e;
+
+	igt_fixture {
+		igt_require(gem_scheduler_enabled(i915));
+		igt_require(has_future_syncobj(i915));
+	}
+
+	igt_subtest_with_dynamic("syncobj-future")
+		__for_each_physical_engine(i915, e) {
+			igt_dynamic_f("%s", e->name)
+				test_syncobj_future(i915, e->flags);
+		}
+
+	igt_subtest_with_dynamic("syncobj-future-past")
+		__for_each_physical_engine(i915, e) {
+			igt_dynamic_f("%s", e->name)
+				test_syncobj_future_past(i915, e->flags, 0);
+		}
+
+
+	igt_subtest_with_dynamic("syncobj-future-submit")
+		__for_each_physical_engine(i915, e) {
+			igt_dynamic_f("%s", e->name)
+				test_syncobj_future_submit(i915, e->flags);
+		}
+
+	igt_subtest_with_dynamic("syncobj-future-submit-past")
+		__for_each_physical_engine(i915, e) {
+			igt_dynamic_f("%s", e->name)
+				test_syncobj_future_past(i915, e->flags,
+						I915_EXEC_FENCE_WAIT_SUBMIT);
+		}
+
+	igt_subtest_with_dynamic("syncobj-future-self")
+		__for_each_physical_engine(i915, e) {
+			igt_dynamic_f("%s", e->name)
+				test_syncobj_future_self(i915, e->flags, 0);
+		}
+
+	igt_subtest_with_dynamic("syncobj-future-self-submit")
+		__for_each_physical_engine(i915, e) {
+			igt_dynamic_f("%s", e->name)
+				test_syncobj_future_self(i915, e->flags,
+						I915_EXEC_FENCE_WAIT_SUBMIT);
+		}
+
+	igt_subtest_with_dynamic("syncobj-future-deadlock")
+		__for_each_physical_engine(i915, e) {
+			igt_dynamic_f("%s", e->name)
+				test_syncobj_future_deadlock(i915, e->flags, 0);
+		}
+
+	igt_subtest_with_dynamic("syncobj-future-submit-deadlock")
+		__for_each_physical_engine(i915, e) {
+			igt_dynamic_f("%s", e->name)
+				test_syncobj_future_deadlock(i915, e->flags,
+							     I915_EXEC_FENCE_WAIT_SUBMIT);
+		}
+
+	igt_subtest_with_dynamic("syncobj-future-cycle")
+		__for_each_physical_engine(i915, e) {
+			igt_dynamic_f("%s", e->name)
+				test_syncobj_future_cycle(i915, e->flags, 0);
+		}
+
+	igt_subtest_with_dynamic("syncobj-future-submit-cycle")
+		__for_each_physical_engine(i915, e) {
+			igt_dynamic_f("%s", e->name)
+				test_syncobj_future_cycle(i915, e->flags,
+							  I915_EXEC_FENCE_WAIT_SUBMIT);
+		}
+}
+
 igt_main
 {
 	const struct intel_execution_engine2 *e;
@@ -1537,6 +2083,9 @@ igt_main
 		igt_subtest("syncobj-invalid-wait")
 			test_syncobj_invalid_wait(i915);
 
+		igt_subtest("syncobj-incomplete-wait-submit")
+			test_syncobj_incomplete_wait_submit(i915);
+
 		igt_subtest("syncobj-invalid-flags")
 			test_syncobj_invalid_flags(i915);
 
@@ -1546,6 +2095,9 @@ igt_main
 		igt_subtest("syncobj-wait")
 			test_syncobj_wait(i915);
 
+		igt_subtest_group
+			syncobj_futures(i915);
+
 		igt_subtest("syncobj-export")
 			test_syncobj_export(i915);
 
-- 
2.26.2

_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

[igt-dev] [PATCH i-g-t 2/2] i915/gem_exec_fence: Teach invalid-wait about invalid future fences

[Chris Wilson]

When we allow a wait on a future future fence, it must autoexpire if the
fence is never signaled by userspace. Also put future fences to work, as
the intention is to use them, along with WAIT_SUBMIT and semaphores, for
userspace to perform its own fine-grained scheduling. Or simply run
concurrent clients without having to flush batches between context
switches.

v2: Verify deadlock detection

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
---
 tests/i915/gem_exec_fence.c | 558 +++++++++++++++++++++++++++++++++++-
 1 file changed, 555 insertions(+), 3 deletions(-)

diff --git a/tests/i915/gem_exec_fence.c b/tests/i915/gem_exec_fence.c
index 4b0d87e4d..e51b7452e 100644
--- a/tests/i915/gem_exec_fence.c
+++ b/tests/i915/gem_exec_fence.c
@@ -46,6 +46,15 @@ struct sync_merge_data {
 #define SYNC_IOC_MERGE _IOWR(SYNC_IOC_MAGIC, 3, struct sync_merge_data)
 #endif
 
+#define MI_SEMAPHORE_WAIT		(0x1c << 23)
+#define   MI_SEMAPHORE_POLL             (1 << 15)
+#define   MI_SEMAPHORE_SAD_GT_SDD       (0 << 12)
+#define   MI_SEMAPHORE_SAD_GTE_SDD      (1 << 12)
+#define   MI_SEMAPHORE_SAD_LT_SDD       (2 << 12)
+#define   MI_SEMAPHORE_SAD_LTE_SDD      (3 << 12)
+#define   MI_SEMAPHORE_SAD_EQ_SDD       (4 << 12)
+#define   MI_SEMAPHORE_SAD_NEQ_SDD      (5 << 12)
+
 static void store(int fd, const struct intel_execution_engine2 *e,
 		  int fence, uint32_t target, unsigned offset_value)
 {
@@ -907,11 +916,12 @@ static void test_syncobj_invalid_wait(int fd)
 	struct drm_i915_gem_exec_fence fence = {
 		.handle = syncobj_create(fd, 0),
 	};
+	int out;
 
 	memset(&execbuf, 0, sizeof(execbuf));
 	execbuf.buffers_ptr = to_user_pointer(&obj);
 	execbuf.buffer_count = 1;
-	execbuf.flags = I915_EXEC_FENCE_ARRAY;
+	execbuf.flags = I915_EXEC_FENCE_ARRAY | I915_EXEC_FENCE_OUT;
 	execbuf.cliprects_ptr = to_user_pointer(&fence);
 	execbuf.num_cliprects = 1;
 
@@ -919,14 +929,59 @@ static void test_syncobj_invalid_wait(int fd)
 	obj.handle = gem_create(fd, 4096);
 	gem_write(fd, obj.handle, 0, &bbe, sizeof(bbe));
 
-	/* waiting before the fence is set is invalid */
+	/* waiting before the fence is set is^W may be invalid */
 	fence.flags = I915_EXEC_FENCE_WAIT;
-	igt_assert_eq(__gem_execbuf(fd, &execbuf), -EINVAL);
+	if (__gem_execbuf_wr(fd, &execbuf)) {
+		igt_assert_eq(__gem_execbuf(fd, &execbuf), -EINVAL);
+		return;
+	}
+
+	/* If we do allow the wait on a future fence, it should autoexpire */
+	gem_sync(fd, obj.handle);
+	out = execbuf.rsvd2 >> 32;
+	igt_assert_eq(sync_fence_status(out), -ETIMEDOUT);
+	close(out);
 
 	gem_close(fd, obj.handle);
 	syncobj_destroy(fd, fence.handle);
 }
 
+static void test_syncobj_incomplete_wait_submit(int i915)
+{
+	struct drm_i915_gem_exec_object2 obj = {
+		.handle = batch_create(i915),
+	};
+	struct drm_i915_gem_exec_fence fence = {
+		.handle = syncobj_create(i915, 0),
+		.flags = I915_EXEC_FENCE_WAIT | I915_EXEC_FENCE_WAIT_SUBMIT,
+	};
+	struct drm_i915_gem_execbuffer2 execbuf = {
+		.buffers_ptr = to_user_pointer(&obj),
+		.buffer_count = 1,
+
+		.cliprects_ptr = to_user_pointer(&fence),
+		.num_cliprects = 1,
+
+		.flags = I915_EXEC_FENCE_ARRAY | I915_EXEC_FENCE_OUT,
+	};
+	int out;
+
+	/* waiting before the fence is set is^W may be invalid */
+	if (__gem_execbuf_wr(i915, &execbuf)) {
+		igt_assert_eq(__gem_execbuf(i915, &execbuf), -EINVAL);
+		return;
+	}
+
+	/* If we do allow the wait on a future fence, it should autoexpire */
+	gem_sync(i915, obj.handle);
+	out = execbuf.rsvd2 >> 32;
+	igt_assert_eq(sync_fence_status(out), -ETIMEDOUT);
+	close(out);
+
+	gem_close(i915, obj.handle);
+	syncobj_destroy(i915, fence.handle);
+}
+
 static void test_syncobj_invalid_flags(int fd)
 {
 	const uint32_t bbe = MI_BATCH_BUFFER_END;
@@ -1073,6 +1128,398 @@ static void test_syncobj_wait(int fd)
 	}
 }
 
+static uint32_t future_batch(int i915, uint32_t offset)
+{
+	uint32_t handle = gem_create(i915, 4096);
+	const int gen = intel_gen(intel_get_drm_devid(i915));
+	uint32_t cs[16];
+	int i = 0;
+
+	cs[i] = MI_STORE_DWORD_IMM | (gen < 6 ? 1 << 22 : 0);
+	if (gen >= 8) {
+		cs[++i] = offset + 4000;
+		cs[++i] = 0;
+	} else if (gen >= 4) {
+		cs[++i] = 0;
+		cs[++i] = offset + 4000;
+	} else {
+		cs[i]--;
+		cs[++i] = offset + 4000;
+	}
+	cs[++i] = 1;
+	cs[i + 1] = MI_BATCH_BUFFER_END;
+	gem_write(i915, handle, 0, cs, sizeof(cs));
+
+	cs[i] = 2;
+	gem_write(i915, handle, 64, cs, sizeof(cs));
+
+	return handle;
+}
+
+static void test_syncobj_future(int i915, unsigned int engine)
+{
+	struct drm_i915_gem_exec_object2 obj = {
+		.offset = 24 << 20,
+		.handle = future_batch(i915, 24 << 20),
+		.flags = EXEC_OBJECT_PINNED,
+	};
+	struct drm_i915_gem_exec_fence fence = {};
+	struct drm_i915_gem_execbuffer2 execbuf  = {
+		.buffers_ptr = to_user_pointer(&obj),
+		.buffer_count = 1,
+		.cliprects_ptr = to_user_pointer(&fence),
+		.num_cliprects = 1,
+	};
+	const struct intel_execution_engine2 *e;
+
+	/*
+	 * Client A is waiting on a future fence from B. So even though its
+	 * execbuf is called first, we need to hold it in a queue waiting on
+	 * B.
+	 */
+	igt_require(gem_scheduler_enabled(i915));
+
+	__for_each_physical_engine(i915, e) {
+		uint32_t result;
+
+		igt_debug("waiting on future %s\n", e->name);
+		fence.handle = syncobj_create(i915, 0);
+
+		fence.flags = I915_EXEC_FENCE_WAIT;
+		execbuf.batch_start_offset = 0;
+		execbuf.flags = engine | I915_EXEC_FENCE_ARRAY;
+		execbuf.rsvd1 = 0;
+		gem_execbuf(i915, &execbuf); /* writes 1 */
+
+		fence.flags = I915_EXEC_FENCE_SIGNAL;
+		execbuf.batch_start_offset = 64;
+		execbuf.flags = e->flags | I915_EXEC_FENCE_ARRAY;
+		execbuf.rsvd1 = gem_context_clone_with_engines(i915, 0);
+		gem_execbuf(i915, &execbuf); /* writes 2 */
+		gem_context_destroy(i915, execbuf.rsvd1);
+
+		syncobj_destroy(i915, fence.handle);
+		gem_sync(i915, obj.handle); /* write hazard lies */
+		gem_read(i915, obj.handle, 4000, &result, sizeof(result));
+		igt_assert_eq(result, 1);
+	}
+
+	gem_close(i915, obj.handle);
+}
+
+static uint32_t future_submit_batch(int i915, uint32_t offset)
+{
+	uint32_t handle = gem_create(i915, 4096);
+	const int gen = intel_gen(intel_get_drm_devid(i915));
+	uint32_t cs[16];
+	int i = 0;
+
+	cs[i] = MI_STORE_DWORD_IMM | (gen < 6 ? 1 << 22 : 0);
+	if (gen >= 8) {
+		cs[++i] = offset + 4000;
+		cs[++i] = 0;
+	} else if (gen >= 4) {
+		cs[++i] = 0;
+		cs[++i] = offset + 4000;
+	} else {
+		cs[i]--;
+		cs[++i] = offset + 4000;
+	}
+	cs[++i] = 1;
+	cs[i + 1] = MI_BATCH_BUFFER_END;
+	igt_assert(i + 1 < ARRAY_SIZE(cs));
+	gem_write(i915, handle, 0, cs, sizeof(cs));
+
+	i = 0;
+	cs[i++] =
+		MI_SEMAPHORE_WAIT |
+		MI_SEMAPHORE_POLL |
+		MI_SEMAPHORE_SAD_EQ_SDD |
+		(4 - 2);
+	cs[i++] = 1;
+	cs[i++] = offset + 4000;
+	cs[i++] = 0;
+
+	cs[i] = MI_STORE_DWORD_IMM | (gen < 6 ? 1 << 22 : 0);
+	if (gen >= 8) {
+		cs[++i] = offset + 4000;
+		cs[++i] = 0;
+	} else if (gen >= 4) {
+		cs[++i] = 0;
+		cs[++i] = offset + 4000;
+	} else {
+		cs[i]--;
+		cs[++i] = offset + 4000;
+	}
+	cs[++i] = 2;
+	cs[++i] = MI_BATCH_BUFFER_END;
+	igt_assert(i < ARRAY_SIZE(cs));
+
+	gem_write(i915, handle, 64, cs, sizeof(cs));
+
+	return handle;
+}
+
+static void test_syncobj_future_submit(int i915, unsigned int engine)
+{
+	struct drm_i915_gem_exec_object2 obj = {
+		.offset = 24 << 20,
+		.handle = future_submit_batch(i915, 24 << 20),
+		.flags = EXEC_OBJECT_PINNED,
+	};
+	const struct intel_execution_engine2 *e;
+
+	/*
+	 * Here we submit client A waiting on client B, but internally client
+	 * B has a semaphore that waits on client A. This relies on timeslicing
+	 * to reorder B before A, even though userspace has asked to submit
+	 * A & B simultaneously (and due to the sequence we will submit B
+	 * then A).
+	 */
+	igt_require(gem_scheduler_has_timeslicing(i915));
+
+	__for_each_physical_engine(i915, e) {
+		struct drm_i915_gem_exec_fence fence = {
+			.handle = syncobj_create(i915, 0),
+		};
+		struct drm_i915_gem_execbuffer2 execbuf  = {
+			.buffers_ptr = to_user_pointer(&obj),
+			.buffer_count = 1,
+			.cliprects_ptr = to_user_pointer(&fence),
+			.num_cliprects = 1,
+			.flags = engine | I915_EXEC_FENCE_ARRAY,
+		};
+		uint32_t result;
+		int out;
+
+		igt_debug("waiting on future %s\n", e->name);
+
+		execbuf.rsvd1 = gem_context_clone_with_engines(i915, 0);
+		fence.flags = I915_EXEC_FENCE_WAIT | I915_EXEC_FENCE_WAIT_SUBMIT;
+		execbuf.batch_start_offset = 0;
+		execbuf.flags |= I915_EXEC_FENCE_OUT;
+		igt_require(__gem_execbuf_wr(i915, &execbuf) == 0); /* writes 1 */
+		execbuf.flags &= ~I915_EXEC_FENCE_OUT;
+		gem_context_destroy(i915, execbuf.rsvd1);
+
+		execbuf.rsvd1 = gem_context_clone_with_engines(i915, 0);
+		fence.flags = I915_EXEC_FENCE_SIGNAL;
+		execbuf.batch_start_offset = 64;
+		execbuf.flags &= ~I915_EXEC_RING_MASK;
+		execbuf.flags |= e->flags;
+		gem_execbuf(i915, &execbuf); /* writes 2 */
+		gem_context_destroy(i915, execbuf.rsvd1);
+
+		syncobj_destroy(i915, fence.handle);
+		gem_sync(i915, obj.handle); /* write hazard lies */
+		gem_read(i915, obj.handle, 4000, &result, sizeof(result));
+		igt_assert_eq(result, 2);
+
+		/* check we didn't autotimeout */
+		out = execbuf.rsvd2 >> 32;
+		igt_assert_eq(sync_fence_status(out), 1);
+		close(out);
+	}
+
+	gem_close(i915, obj.handle);
+}
+
+static void test_syncobj_future_past(int i915, unsigned int engine, int flags)
+{
+	struct drm_i915_gem_exec_object2 obj = {
+		.offset = 24 << 20,
+		.handle = future_batch(i915, 24 << 20),
+		.flags = EXEC_OBJECT_PINNED,
+	};
+	struct drm_i915_gem_exec_fence fence = {
+		.handle = syncobj_create(i915, 0),
+	};
+	struct drm_i915_gem_execbuffer2 execbuf  = {
+		.buffers_ptr = to_user_pointer(&obj),
+		.buffer_count = 1,
+		.cliprects_ptr = to_user_pointer(&fence),
+		.num_cliprects = 1,
+		.flags = engine | I915_EXEC_FENCE_ARRAY,
+	};
+	uint32_t result;
+	int out;
+
+	fence.flags = I915_EXEC_FENCE_WAIT | I915_EXEC_FENCE_SIGNAL | flags;
+	execbuf.batch_start_offset = 0;
+	execbuf.flags |= I915_EXEC_FENCE_OUT;
+	igt_require(__gem_execbuf_wr(i915, &execbuf) == 0); /* writes 1 */
+	execbuf.flags &= ~I915_EXEC_FENCE_OUT;
+
+	gem_sync(i915, obj.handle); /* write hazard lies */
+	gem_read(i915, obj.handle, 4000, &result, sizeof(result));
+	igt_assert_eq(result, 1);
+
+	/* check we didn't autotimeout */
+	out = execbuf.rsvd2 >> 32;
+	igt_assert_eq(sync_fence_status(out), 1);
+	close(out);
+
+	gem_close(i915, obj.handle);
+	syncobj_destroy(i915, fence.handle);
+}
+
+static void test_syncobj_future_self(int i915, unsigned int engine, int flags)
+{
+	struct drm_i915_gem_exec_object2 obj = {
+		.offset = 24 << 20,
+		.handle = future_batch(i915, 24 << 20),
+		.flags = EXEC_OBJECT_PINNED,
+	};
+	struct drm_i915_gem_exec_fence fence = {
+		.handle = syncobj_create(i915, 0),
+	};
+	struct drm_i915_gem_execbuffer2 execbuf  = {
+		.buffers_ptr = to_user_pointer(&obj),
+		.buffer_count = 1,
+		.cliprects_ptr = to_user_pointer(&fence),
+		.num_cliprects = 1,
+		.flags = engine | I915_EXEC_FENCE_ARRAY,
+	};
+	int out;
+
+	fence.flags = I915_EXEC_FENCE_WAIT | flags;
+	execbuf.batch_start_offset = 0;
+	execbuf.flags |= I915_EXEC_FENCE_OUT;
+	igt_require(__gem_execbuf_wr(i915, &execbuf) == 0); /* writes 1 */
+	execbuf.flags &= ~I915_EXEC_FENCE_OUT;
+
+	fence.flags = I915_EXEC_FENCE_SIGNAL;
+	execbuf.batch_start_offset = 64;
+	gem_execbuf(i915, &execbuf); /* writes 2 */
+
+	gem_sync(i915, obj.handle); /* write hazard lies */
+
+	gem_close(i915, obj.handle);
+	syncobj_destroy(i915, fence.handle);
+
+	out = execbuf.rsvd2 >> 32;
+	igt_assert_eq(sync_fence_status(out), -EDEADLK);
+	close(out);
+}
+
+static void
+test_syncobj_future_deadlock(int i915, unsigned int engine, int flags)
+{
+	struct drm_i915_gem_exec_object2 obj = {
+		.offset = 24 << 20,
+		.handle = future_batch(i915, 24 << 20),
+		.flags = EXEC_OBJECT_PINNED,
+	};
+	struct drm_i915_gem_exec_fence fence = {};
+	struct drm_i915_gem_execbuffer2 execbuf  = {
+		.buffers_ptr = to_user_pointer(&obj),
+		.buffer_count = 1,
+		.cliprects_ptr = to_user_pointer(&fence),
+		.num_cliprects = 1,
+	};
+	const struct intel_execution_engine2 *e;
+
+	__for_each_physical_engine(i915, e) {
+		int out;
+
+		fence.handle = syncobj_create(i915, 0),
+
+		fence.flags = I915_EXEC_FENCE_WAIT | flags;
+		execbuf.batch_start_offset = 0;
+		execbuf.flags = engine | I915_EXEC_FENCE_ARRAY,
+		execbuf.flags |= I915_EXEC_FENCE_OUT;
+		execbuf.rsvd1 = 0;
+		gem_execbuf_wr(i915, &execbuf); /* writes 1 */
+
+		fence.flags = I915_EXEC_FENCE_SIGNAL;
+		execbuf.batch_start_offset = 64;
+		execbuf.flags = e->flags | I915_EXEC_FENCE_ARRAY,
+		execbuf.flags |= I915_EXEC_FENCE_OUT | I915_EXEC_FENCE_IN;
+		execbuf.rsvd1 = gem_context_clone_with_engines(i915, 0);
+		execbuf.rsvd2 >>= 32;
+		gem_execbuf_wr(i915, &execbuf); /* writes 2 */
+		gem_context_destroy(i915, execbuf.rsvd1);
+
+		syncobj_destroy(i915, fence.handle);
+		gem_sync(i915, obj.handle);
+
+		/* How should this deadlock be resolved? */
+		out = execbuf.rsvd2;
+		igt_assert_eq(sync_fence_status(out), -EDEADLK);
+		close(out);
+
+		out = execbuf.rsvd2 >> 32;
+		igt_assert_eq(sync_fence_status(out), -EDEADLK);
+		close(out);
+	}
+
+	gem_close(i915, obj.handle);
+}
+
+static void
+test_syncobj_future_cycle(int i915, unsigned int engine, int flags)
+{
+	struct drm_i915_gem_exec_object2 obj = {
+		.offset = 24 << 20,
+		.handle = future_batch(i915, 24 << 20),
+		.flags = EXEC_OBJECT_PINNED,
+	};
+	const struct intel_execution_engine2 *e1, *e2;
+
+	__for_each_physical_engine(i915, e1) {
+	__for_each_physical_engine(i915, e2) {
+		struct drm_i915_gem_exec_fence fence = {
+			.handle = syncobj_create(i915, 0),
+		};
+		struct drm_i915_gem_execbuffer2 execbuf  = {
+			.buffers_ptr = to_user_pointer(&obj),
+			.buffer_count = 1,
+			.cliprects_ptr = to_user_pointer(&fence),
+			.num_cliprects = 1,
+			.flags = engine | I915_EXEC_FENCE_ARRAY,
+		};
+		int out;
+
+		fence.flags = I915_EXEC_FENCE_WAIT | flags;
+		execbuf.batch_start_offset = 0;
+		execbuf.flags |= I915_EXEC_FENCE_OUT;
+		igt_require(__gem_execbuf_wr(i915, &execbuf) == 0);
+
+		fence.flags = 0;
+		execbuf.rsvd1 = gem_context_clone_with_engines(i915, 0);
+		execbuf.rsvd2 >>= 32;
+		execbuf.flags &= ~I915_EXEC_RING_MASK;
+		execbuf.flags |= e1->flags | I915_EXEC_FENCE_IN;
+		gem_execbuf_wr(i915, &execbuf);
+		gem_context_destroy(i915, execbuf.rsvd1);
+		close(execbuf.rsvd2);
+
+		fence.flags = I915_EXEC_FENCE_SIGNAL;
+		execbuf.rsvd1 = gem_context_clone_with_engines(i915, 0);
+		execbuf.rsvd2 >>= 32;
+		execbuf.flags &= ~I915_EXEC_RING_MASK;
+		execbuf.flags |= e2->flags;
+		execbuf.batch_start_offset = 64;
+		gem_execbuf_wr(i915, &execbuf); /* writes 2 */
+		gem_context_destroy(i915, execbuf.rsvd1);
+
+		syncobj_destroy(i915, fence.handle);
+		gem_sync(i915, obj.handle);
+
+
+		/* How should this deadlock be resolved? */
+		out = execbuf.rsvd2 >> 32;
+		igt_assert_eq(sync_fence_status(out), -EDEADLK);
+		close(out);
+
+		out = execbuf.rsvd2;
+		igt_assert_eq(sync_fence_status(out), -EDEADLK);
+		close(out);
+	}}
+
+	gem_close(i915, obj.handle);
+}
+
 static void test_syncobj_export(int fd)
 {
 	const uint32_t bbe = MI_BATCH_BUFFER_END;
@@ -1358,6 +1805,105 @@ static void test_syncobj_channel(int fd)
 		syncobj_destroy(fd, syncobj[i]);
 }
 
+static bool has_future_syncobj(int i915)
+{
+	struct drm_i915_gem_exec_object2 obj = {
+		.handle = batch_create(i915),
+	};
+	struct drm_i915_gem_exec_fence fence = {
+		.handle = syncobj_create(i915, 0),
+		.flags = I915_EXEC_FENCE_WAIT | I915_EXEC_FENCE_SIGNAL,
+	};
+	struct drm_i915_gem_execbuffer2 execbuf  = {
+		.buffers_ptr = to_user_pointer(&obj),
+		.buffer_count = 1,
+		.cliprects_ptr = to_user_pointer(&fence),
+		.num_cliprects = 1,
+		.flags = I915_EXEC_FENCE_ARRAY,
+	};
+	bool result;
+
+	result = __gem_execbuf(i915, &execbuf) == 0;
+	gem_close(i915, obj.handle);
+
+	return result;
+}
+
+static void syncobj_futures(int i915)
+{
+	const struct intel_execution_engine2 *e;
+
+	igt_fixture {
+		igt_require(gem_scheduler_enabled(i915));
+		igt_require(has_future_syncobj(i915));
+	}
+
+	igt_subtest_with_dynamic("syncobj-future")
+		__for_each_physical_engine(i915, e) {
+			igt_dynamic_f("%s", e->name)
+				test_syncobj_future(i915, e->flags);
+		}
+
+	igt_subtest_with_dynamic("syncobj-future-past")
+		__for_each_physical_engine(i915, e) {
+			igt_dynamic_f("%s", e->name)
+				test_syncobj_future_past(i915, e->flags, 0);
+		}
+
+
+	igt_subtest_with_dynamic("syncobj-future-submit")
+		__for_each_physical_engine(i915, e) {
+			igt_dynamic_f("%s", e->name)
+				test_syncobj_future_submit(i915, e->flags);
+		}
+
+	igt_subtest_with_dynamic("syncobj-future-submit-past")
+		__for_each_physical_engine(i915, e) {
+			igt_dynamic_f("%s", e->name)
+				test_syncobj_future_past(i915, e->flags,
+						I915_EXEC_FENCE_WAIT_SUBMIT);
+		}
+
+	igt_subtest_with_dynamic("syncobj-future-self")
+		__for_each_physical_engine(i915, e) {
+			igt_dynamic_f("%s", e->name)
+				test_syncobj_future_self(i915, e->flags, 0);
+		}
+
+	igt_subtest_with_dynamic("syncobj-future-self-submit")
+		__for_each_physical_engine(i915, e) {
+			igt_dynamic_f("%s", e->name)
+				test_syncobj_future_self(i915, e->flags,
+						I915_EXEC_FENCE_WAIT_SUBMIT);
+		}
+
+	igt_subtest_with_dynamic("syncobj-future-deadlock")
+		__for_each_physical_engine(i915, e) {
+			igt_dynamic_f("%s", e->name)
+				test_syncobj_future_deadlock(i915, e->flags, 0);
+		}
+
+	igt_subtest_with_dynamic("syncobj-future-submit-deadlock")
+		__for_each_physical_engine(i915, e) {
+			igt_dynamic_f("%s", e->name)
+				test_syncobj_future_deadlock(i915, e->flags,
+							     I915_EXEC_FENCE_WAIT_SUBMIT);
+		}
+
+	igt_subtest_with_dynamic("syncobj-future-cycle")
+		__for_each_physical_engine(i915, e) {
+			igt_dynamic_f("%s", e->name)
+				test_syncobj_future_cycle(i915, e->flags, 0);
+		}
+
+	igt_subtest_with_dynamic("syncobj-future-submit-cycle")
+		__for_each_physical_engine(i915, e) {
+			igt_dynamic_f("%s", e->name)
+				test_syncobj_future_cycle(i915, e->flags,
+							  I915_EXEC_FENCE_WAIT_SUBMIT);
+		}
+}
+
 igt_main
 {
 	const struct intel_execution_engine2 *e;
@@ -1537,6 +2083,9 @@ igt_main
 		igt_subtest("syncobj-invalid-wait")
 			test_syncobj_invalid_wait(i915);
 
+		igt_subtest("syncobj-incomplete-wait-submit")
+			test_syncobj_incomplete_wait_submit(i915);
+
 		igt_subtest("syncobj-invalid-flags")
 			test_syncobj_invalid_flags(i915);
 
@@ -1546,6 +2095,9 @@ igt_main
 		igt_subtest("syncobj-wait")
 			test_syncobj_wait(i915);
 
+		igt_subtest_group
+			syncobj_futures(i915);
+
 		igt_subtest("syncobj-export")
 			test_syncobj_export(i915);
 
-- 
2.26.2

_______________________________________________
igt-dev mailing list
igt-dev@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/igt-dev

[igt-dev] ✓ Fi.CI.BAT: success for series starting with [i-g-t,1/2] lib/i915: Report scheduler caps for timeslicing

[Patchwork]

== Series Details ==

Series: series starting with [i-g-t,1/2] lib/i915: Report scheduler caps for timeslicing
URL   : https://patchwork.freedesktop.org/series/76974/
State : success

== Summary ==

CI Bug Log - changes from IGT_5634 -> IGTPW_4536
====================================================

Summary
-------

  **SUCCESS**

  No regressions found.

  External URL: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_4536/index.html


Changes
-------

  No changes found


Participating hosts (51 -> 44)
------------------------------

  Missing    (7): fi-ilk-m540 fi-hsw-4200u fi-byt-squawks fi-bsw-cyan fi-ctg-p8600 fi-byt-clapper fi-bdw-samus 


Build changes
-------------

  * CI: CI-20190529 -> None
  * IGT: IGT_5634 -> IGTPW_4536

  CI-20190529: 20190529
  CI_DRM_8433: db68fed086f2ddcdc30e0d9ca5faaba5e55d0d01 @ git://anongit.freedesktop.org/gfx-ci/linux
  IGTPW_4536: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_4536/index.html
  IGT_5634: 3cff0b137bcb2f6b329a6cb0fc03962d8cc65117 @ git://anongit.freedesktop.org/xorg/app/intel-gpu-tools



== Testlist changes ==

+igt@gem_exec_fence@syncobj-future
+igt@gem_exec_fence@syncobj-future-cycle
+igt@gem_exec_fence@syncobj-future-deadlock
+igt@gem_exec_fence@syncobj-future-past
+igt@gem_exec_fence@syncobj-future-self
+igt@gem_exec_fence@syncobj-future-self-submit
+igt@gem_exec_fence@syncobj-future-submit
+igt@gem_exec_fence@syncobj-future-submit-cycle
+igt@gem_exec_fence@syncobj-future-submit-deadlock
+igt@gem_exec_fence@syncobj-future-submit-past
+igt@gem_exec_fence@syncobj-incomplete-wait-submit

== Logs ==

For more details see: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_4536/index.html
_______________________________________________
igt-dev mailing list
igt-dev@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/igt-dev

[igt-dev] ✗ Fi.CI.IGT: failure for series starting with [i-g-t,1/2] lib/i915: Report scheduler caps for timeslicing

[Patchwork]

== Series Details ==

Series: series starting with [i-g-t,1/2] lib/i915: Report scheduler caps for timeslicing
URL   : https://patchwork.freedesktop.org/series/76974/
State : failure

== Summary ==

CI Bug Log - changes from IGT_5634_full -> IGTPW_4536_full
====================================================

Summary
-------

  **FAILURE**

  Serious unknown changes coming with IGTPW_4536_full absolutely need to be
  verified manually.
  
  If you think the reported changes have nothing to do with the changes
  introduced in IGTPW_4536_full, please notify your bug team to allow them
  to document this new failure mode, which will reduce false positives in CI.

  External URL: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_4536/index.html

Possible new issues
-------------------

  Here are the unknown changes that may have been introduced in IGTPW_4536_full:

### IGT changes ###

#### Possible regressions ####

  * igt@gem_ctx_persistence@legacy-engines-hostile@vebox:
    - shard-iclb:         [PASS][1] -> [FAIL][2]
   [1]: https://intel-gfx-ci.01.org/tree/drm-tip/IGT_5634/shard-iclb4/igt@gem_ctx_persistence@legacy-engines-hostile@vebox.html
   [2]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_4536/shard-iclb6/igt@gem_ctx_persistence@legacy-engines-hostile@vebox.html

  * {igt@gem_exec_fence@syncobj-future} (NEW):
    - shard-iclb:         NOTRUN -> [SKIP][3] +9 similar issues
   [3]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_4536/shard-iclb1/igt@gem_exec_fence@syncobj-future.html

  * {igt@gem_exec_fence@syncobj-future-submit-cycle} (NEW):
    - shard-tglb:         NOTRUN -> [SKIP][4] +9 similar issues
   [4]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_4536/shard-tglb3/igt@gem_exec_fence@syncobj-future-submit-cycle.html

  
#### Suppressed ####

  The following results come from untrusted machines, tests, or statuses.
  They do not affect the overall result.

  * {igt@i915_selftest@perf@request}:
    - shard-tglb:         [PASS][5] -> [DMESG-FAIL][6]
   [5]: https://intel-gfx-ci.01.org/tree/drm-tip/IGT_5634/shard-tglb6/igt@i915_selftest@perf@request.html
   [6]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_4536/shard-tglb1/igt@i915_selftest@perf@request.html

  
New tests
---------

  New tests have been introduced between IGT_5634_full and IGTPW_4536_full:

### New IGT tests (11) ###

  * igt@gem_exec_fence@syncobj-future:
    - Statuses : 7 skip(s)
    - Exec time: [0.0] s

  * igt@gem_exec_fence@syncobj-future-cycle:
    - Statuses : 7 skip(s)
    - Exec time: [0.0] s

  * igt@gem_exec_fence@syncobj-future-deadlock:
    - Statuses : 6 skip(s)
    - Exec time: [0.0] s

  * igt@gem_exec_fence@syncobj-future-past:
    - Statuses : 7 skip(s)
    - Exec time: [0.0] s

  * igt@gem_exec_fence@syncobj-future-self:
    - Statuses : 7 skip(s)
    - Exec time: [0.0] s

  * igt@gem_exec_fence@syncobj-future-self-submit:
    - Statuses : 7 skip(s)
    - Exec time: [0.0] s

  * igt@gem_exec_fence@syncobj-future-submit:
    - Statuses : 6 skip(s)
    - Exec time: [0.0] s

  * igt@gem_exec_fence@syncobj-future-submit-cycle:
    - Statuses : 7 skip(s)
    - Exec time: [0.0] s

  * igt@gem_exec_fence@syncobj-future-submit-deadlock:
    - Statuses : 7 skip(s)
    - Exec time: [0.0] s

  * igt@gem_exec_fence@syncobj-future-submit-past:
    - Statuses : 7 skip(s)
    - Exec time: [0.0] s

  * igt@gem_exec_fence@syncobj-incomplete-wait-submit:
    - Statuses : 7 pass(s)
    - Exec time: [0.0, 0.00] s

  

Known issues
------------

  Here are the changes found in IGTPW_4536_full that come from known issues:

### IGT changes ###

#### Issues hit ####

  * igt@gen9_exec_parse@allowed-all:
    - shard-kbl:          [PASS][7] -> [DMESG-WARN][8] ([i915#1436] / [i915#716])
   [7]: https://intel-gfx-ci.01.org/tree/drm-tip/IGT_5634/shard-kbl2/igt@gen9_exec_parse@allowed-all.html
   [8]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_4536/shard-kbl6/igt@gen9_exec_parse@allowed-all.html

  * igt@kms_cursor_crc@pipe-a-cursor-256x256-offscreen:
    - shard-kbl:          [PASS][9] -> [FAIL][10] ([i915#54] / [i915#93] / [i915#95])
   [9]: https://intel-gfx-ci.01.org/tree/drm-tip/IGT_5634/shard-kbl6/igt@kms_cursor_crc@pipe-a-cursor-256x256-offscreen.html
   [10]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_4536/shard-kbl3/igt@kms_cursor_crc@pipe-a-cursor-256x256-offscreen.html

  * igt@kms_cursor_crc@pipe-c-cursor-suspend:
    - shard-kbl:          [PASS][11] -> [DMESG-WARN][12] ([i915#180])
   [11]: https://intel-gfx-ci.01.org/tree/drm-tip/IGT_5634/shard-kbl2/igt@kms_cursor_crc@pipe-c-cursor-suspend.html
   [12]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_4536/shard-kbl1/igt@kms_cursor_crc@pipe-c-cursor-suspend.html

  * igt@kms_hdr@bpc-switch-suspend:
    - shard-kbl:          [PASS][13] -> [INCOMPLETE][14] ([i915#155])
   [13]: https://intel-gfx-ci.01.org/tree/drm-tip/IGT_5634/shard-kbl7/igt@kms_hdr@bpc-switch-suspend.html
   [14]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_4536/shard-kbl2/igt@kms_hdr@bpc-switch-suspend.html

  * igt@kms_plane_alpha_blend@pipe-a-constant-alpha-min:
    - shard-kbl:          [PASS][15] -> [FAIL][16] ([fdo#108145] / [i915#265] / [i915#93] / [i915#95])
   [15]: https://intel-gfx-ci.01.org/tree/drm-tip/IGT_5634/shard-kbl6/igt@kms_plane_alpha_blend@pipe-a-constant-alpha-min.html
   [16]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_4536/shard-kbl7/igt@kms_plane_alpha_blend@pipe-a-constant-alpha-min.html
    - shard-apl:          [PASS][17] -> [FAIL][18] ([fdo#108145] / [i915#265] / [i915#95])
   [17]: https://intel-gfx-ci.01.org/tree/drm-tip/IGT_5634/shard-apl4/igt@kms_plane_alpha_blend@pipe-a-constant-alpha-min.html
   [18]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_4536/shard-apl7/igt@kms_plane_alpha_blend@pipe-a-constant-alpha-min.html

  * igt@kms_psr@psr2_sprite_render:
    - shard-iclb:         [PASS][19] -> [SKIP][20] ([fdo#109441])
   [19]: https://intel-gfx-ci.01.org/tree/drm-tip/IGT_5634/shard-iclb2/igt@kms_psr@psr2_sprite_render.html
   [20]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_4536/shard-iclb8/igt@kms_psr@psr2_sprite_render.html

  * igt@kms_setmode@basic:
    - shard-kbl:          [PASS][21] -> [FAIL][22] ([i915#31])
   [21]: https://intel-gfx-ci.01.org/tree/drm-tip/IGT_5634/shard-kbl1/igt@kms_setmode@basic.html
   [22]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_4536/shard-kbl1/igt@kms_setmode@basic.html

  
#### Possible fixes ####

  * igt@gem_exec_params@invalid-bsd-ring:
    - shard-iclb:         [SKIP][23] ([fdo#109276]) -> [PASS][24]
   [23]: https://intel-gfx-ci.01.org/tree/drm-tip/IGT_5634/shard-iclb7/igt@gem_exec_params@invalid-bsd-ring.html
   [24]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_4536/shard-iclb2/igt@gem_exec_params@invalid-bsd-ring.html

  * igt@i915_pm_dc@dc6-psr:
    - shard-iclb:         [FAIL][25] ([i915#454]) -> [PASS][26]
   [25]: https://intel-gfx-ci.01.org/tree/drm-tip/IGT_5634/shard-iclb2/igt@i915_pm_dc@dc6-psr.html
   [26]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_4536/shard-iclb1/igt@i915_pm_dc@dc6-psr.html

  * igt@kms_cursor_crc@pipe-a-cursor-64x21-offscreen:
    - shard-kbl:          [FAIL][27] ([i915#54] / [i915#93] / [i915#95]) -> [PASS][28] +1 similar issue
   [27]: https://intel-gfx-ci.01.org/tree/drm-tip/IGT_5634/shard-kbl3/igt@kms_cursor_crc@pipe-a-cursor-64x21-offscreen.html
   [28]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_4536/shard-kbl7/igt@kms_cursor_crc@pipe-a-cursor-64x21-offscreen.html

  * igt@kms_cursor_crc@pipe-b-cursor-suspend:
    - shard-kbl:          [DMESG-WARN][29] ([i915#180]) -> [PASS][30]
   [29]: https://intel-gfx-ci.01.org/tree/drm-tip/IGT_5634/shard-kbl1/igt@kms_cursor_crc@pipe-b-cursor-suspend.html
   [30]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_4536/shard-kbl1/igt@kms_cursor_crc@pipe-b-cursor-suspend.html

  * igt@kms_cursor_crc@pipe-c-cursor-dpms:
    - shard-apl:          [FAIL][31] ([i915#54]) -> [PASS][32]
   [31]: https://intel-gfx-ci.01.org/tree/drm-tip/IGT_5634/shard-apl6/igt@kms_cursor_crc@pipe-c-cursor-dpms.html
   [32]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_4536/shard-apl3/igt@kms_cursor_crc@pipe-c-cursor-dpms.html
    - shard-kbl:          [FAIL][33] ([i915#54]) -> [PASS][34]
   [33]: https://intel-gfx-ci.01.org/tree/drm-tip/IGT_5634/shard-kbl4/igt@kms_cursor_crc@pipe-c-cursor-dpms.html
   [34]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_4536/shard-kbl6/igt@kms_cursor_crc@pipe-c-cursor-dpms.html

  * igt@kms_cursor_legacy@all-pipes-torture-move:
    - shard-hsw:          [DMESG-WARN][35] ([i915#128]) -> [PASS][36]
   [35]: https://intel-gfx-ci.01.org/tree/drm-tip/IGT_5634/shard-hsw4/igt@kms_cursor_legacy@all-pipes-torture-move.html
   [36]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_4536/shard-hsw6/igt@kms_cursor_legacy@all-pipes-torture-move.html

  * {igt@kms_flip@flip-vs-suspend-interruptible@c-dp1}:
    - shard-apl:          [DMESG-WARN][37] ([i915#180]) -> [PASS][38] +3 similar issues
   [37]: https://intel-gfx-ci.01.org/tree/drm-tip/IGT_5634/shard-apl8/igt@kms_flip@flip-vs-suspend-interruptible@c-dp1.html
   [38]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_4536/shard-apl2/igt@kms_flip@flip-vs-suspend-interruptible@c-dp1.html

  * igt@kms_pipe_crc_basic@suspend-read-crc-pipe-b:
    - shard-kbl:          [INCOMPLETE][39] ([i915#155]) -> [PASS][40]
   [39]: https://intel-gfx-ci.01.org/tree/drm-tip/IGT_5634/shard-kbl2/igt@kms_pipe_crc_basic@suspend-read-crc-pipe-b.html
   [40]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_4536/shard-kbl7/igt@kms_pipe_crc_basic@suspend-read-crc-pipe-b.html

  * igt@kms_psr@psr2_sprite_mmap_cpu:
    - shard-iclb:         [SKIP][41] ([fdo#109441]) -> [PASS][42]
   [41]: https://intel-gfx-ci.01.org/tree/drm-tip/IGT_5634/shard-iclb1/igt@kms_psr@psr2_sprite_mmap_cpu.html
   [42]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_4536/shard-iclb2/igt@kms_psr@psr2_sprite_mmap_cpu.html

  
#### Warnings ####

  * igt@i915_pm_dc@dc6-psr:
    - shard-tglb:         [SKIP][43] ([i915#468]) -> [FAIL][44] ([i915#454]) +1 similar issue
   [43]: https://intel-gfx-ci.01.org/tree/drm-tip/IGT_5634/shard-tglb2/igt@i915_pm_dc@dc6-psr.html
   [44]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_4536/shard-tglb7/igt@i915_pm_dc@dc6-psr.html

  * igt@kms_content_protection@atomic-dpms:
    - shard-apl:          [TIMEOUT][45] ([i915#1319]) -> [FAIL][46] ([fdo#110321] / [fdo#110336])
   [45]: https://intel-gfx-ci.01.org/tree/drm-tip/IGT_5634/shard-apl7/igt@kms_content_protection@atomic-dpms.html
   [46]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_4536/shard-apl7/igt@kms_content_protection@atomic-dpms.html

  * igt@kms_fbcon_fbt@fbc-suspend:
    - shard-apl:          [FAIL][47] ([i915#1525]) -> [FAIL][48] ([i915#95])
   [47]: https://intel-gfx-ci.01.org/tree/drm-tip/IGT_5634/shard-apl4/igt@kms_fbcon_fbt@fbc-suspend.html
   [48]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_4536/shard-apl2/igt@kms_fbcon_fbt@fbc-suspend.html
    - shard-kbl:          [FAIL][49] ([i915#64]) -> [FAIL][50] ([i915#93] / [i915#95])
   [49]: https://intel-gfx-ci.01.org/tree/drm-tip/IGT_5634/shard-kbl6/igt@kms_fbcon_fbt@fbc-suspend.html
   [50]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_4536/shard-kbl4/igt@kms_fbcon_fbt@fbc-suspend.html

  
  {name}: This element is suppressed. This means it is ignored when computing
          the status of the difference (SUCCESS, WARNING, or FAILURE).

  [fdo#108145]: https://bugs.freedesktop.org/show_bug.cgi?id=108145
  [fdo#109276]: https://bugs.freedesktop.org/show_bug.cgi?id=109276
  [fdo#109441]: https://bugs.freedesktop.org/show_bug.cgi?id=109441
  [fdo#110321]: https://bugs.freedesktop.org/show_bug.cgi?id=110321
  [fdo#110336]: https://bugs.freedesktop.org/show_bug.cgi?id=110336
  [i915#128]: https://gitlab.freedesktop.org/drm/intel/issues/128
  [i915#1319]: https://gitlab.freedesktop.org/drm/intel/issues/1319
  [i915#1436]: https://gitlab.freedesktop.org/drm/intel/issues/1436
  [i915#1525]: https://gitlab.freedesktop.org/drm/intel/issues/1525
  [i915#1542]: https://gitlab.freedesktop.org/drm/intel/issues/1542
  [i915#155]: https://gitlab.freedesktop.org/drm/intel/issues/155
  [i915#180]: https://gitlab.freedesktop.org/drm/intel/issues/180
  [i915#265]: https://gitlab.freedesktop.org/drm/intel/issues/265
  [i915#31]: https://gitlab.freedesktop.org/drm/intel/issues/31
  [i915#454]: https://gitlab.freedesktop.org/drm/intel/issues/454
  [i915#468]: https://gitlab.freedesktop.org/drm/intel/issues/468
  [i915#54]: https://gitlab.freedesktop.org/drm/intel/issues/54
  [i915#64]: https://gitlab.freedesktop.org/drm/intel/issues/64
  [i915#716]: https://gitlab.freedesktop.org/drm/intel/issues/716
  [i915#93]: https://gitlab.freedesktop.org/drm/intel/issues/93
  [i915#95]: https://gitlab.freedesktop.org/drm/intel/issues/95


Participating hosts (8 -> 8)
------------------------------

  No changes in participating hosts


Build changes
-------------

  * CI: CI-20190529 -> None
  * IGT: IGT_5634 -> IGTPW_4536

  CI-20190529: 20190529
  CI_DRM_8433: db68fed086f2ddcdc30e0d9ca5faaba5e55d0d01 @ git://anongit.freedesktop.org/gfx-ci/linux
  IGTPW_4536: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_4536/index.html
  IGT_5634: 3cff0b137bcb2f6b329a6cb0fc03962d8cc65117 @ git://anongit.freedesktop.org/xorg/app/intel-gpu-tools

== Logs ==

For more details see: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_4536/index.html
_______________________________________________
igt-dev mailing list
igt-dev@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/igt-dev

[Intel-gfx] [PATCH i-g-t 2/2] i915/gem_exec_fence: Teach invalid-wait about invalid future fences

[Chris Wilson]

When we allow a wait on a future future fence, it must autoexpire if the
fence is never signaled by userspace. Also put future fences to work, as
the intention is to use them, along with WAIT_SUBMIT and semaphores, for
userspace to perform its own fine-grained scheduling. Or simply run
concurrent clients without having to flush batches between context
switches.

v2: Verify deadlock detection

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
---
 tests/i915/gem_exec_fence.c | 680 +++++++++++++++++++++++++++++++++++-
 1 file changed, 677 insertions(+), 3 deletions(-)

diff --git a/tests/i915/gem_exec_fence.c b/tests/i915/gem_exec_fence.c
index 4140bff24..eb1165080 100644
--- a/tests/i915/gem_exec_fence.c
+++ b/tests/i915/gem_exec_fence.c
@@ -1123,11 +1123,12 @@ static void test_syncobj_invalid_wait(int fd)
 	struct drm_i915_gem_exec_fence fence = {
 		.handle = syncobj_create(fd, 0),
 	};
+	int out;
 
 	memset(&execbuf, 0, sizeof(execbuf));
 	execbuf.buffers_ptr = to_user_pointer(&obj);
 	execbuf.buffer_count = 1;
-	execbuf.flags = I915_EXEC_FENCE_ARRAY;
+	execbuf.flags = I915_EXEC_FENCE_ARRAY | I915_EXEC_FENCE_OUT;
 	execbuf.cliprects_ptr = to_user_pointer(&fence);
 	execbuf.num_cliprects = 1;
 
@@ -1135,14 +1136,59 @@ static void test_syncobj_invalid_wait(int fd)
 	obj.handle = gem_create(fd, 4096);
 	gem_write(fd, obj.handle, 0, &bbe, sizeof(bbe));
 
-	/* waiting before the fence is set is invalid */
+	/* waiting before the fence is set is^W may be invalid */
 	fence.flags = I915_EXEC_FENCE_WAIT;
-	igt_assert_eq(__gem_execbuf(fd, &execbuf), -EINVAL);
+	if (__gem_execbuf_wr(fd, &execbuf)) {
+		igt_assert_eq(__gem_execbuf(fd, &execbuf), -EINVAL);
+		return;
+	}
+
+	/* If we do allow the wait on a future fence, it should autoexpire */
+	gem_sync(fd, obj.handle);
+	out = execbuf.rsvd2 >> 32;
+	igt_assert_eq(sync_fence_status(out), -ETIMEDOUT);
+	close(out);
 
 	gem_close(fd, obj.handle);
 	syncobj_destroy(fd, fence.handle);
 }
 
+static void test_syncobj_incomplete_wait_submit(int i915)
+{
+	struct drm_i915_gem_exec_object2 obj = {
+		.handle = batch_create(i915),
+	};
+	struct drm_i915_gem_exec_fence fence = {
+		.handle = syncobj_create(i915, 0),
+		.flags = I915_EXEC_FENCE_WAIT | I915_EXEC_FENCE_WAIT_SUBMIT,
+	};
+	struct drm_i915_gem_execbuffer2 execbuf = {
+		.buffers_ptr = to_user_pointer(&obj),
+		.buffer_count = 1,
+
+		.cliprects_ptr = to_user_pointer(&fence),
+		.num_cliprects = 1,
+
+		.flags = I915_EXEC_FENCE_ARRAY | I915_EXEC_FENCE_OUT,
+	};
+	int out;
+
+	/* waiting before the fence is set is^W may be invalid */
+	if (__gem_execbuf_wr(i915, &execbuf)) {
+		igt_assert_eq(__gem_execbuf(i915, &execbuf), -EINVAL);
+		return;
+	}
+
+	/* If we do allow the wait on a future fence, it should autoexpire */
+	gem_sync(i915, obj.handle);
+	out = execbuf.rsvd2 >> 32;
+	igt_assert_eq(sync_fence_status(out), -ETIMEDOUT);
+	close(out);
+
+	gem_close(i915, obj.handle);
+	syncobj_destroy(i915, fence.handle);
+}
+
 static void test_syncobj_invalid_flags(int fd)
 {
 	const uint32_t bbe = MI_BATCH_BUFFER_END;
@@ -1289,6 +1335,517 @@ static void test_syncobj_wait(int fd)
 	}
 }
 
+static uint32_t future_batch(int i915, uint32_t offset)
+{
+	uint32_t handle = gem_create(i915, 4096);
+	const int gen = intel_gen(intel_get_drm_devid(i915));
+	uint32_t cs[16];
+	int i = 0;
+
+	cs[i] = MI_STORE_DWORD_IMM | (gen < 6 ? 1 << 22 : 0);
+	if (gen >= 8) {
+		cs[++i] = offset + 4000;
+		cs[++i] = 0;
+	} else if (gen >= 4) {
+		cs[++i] = 0;
+		cs[++i] = offset + 4000;
+	} else {
+		cs[i]--;
+		cs[++i] = offset + 4000;
+	}
+	cs[++i] = 1;
+	cs[i + 1] = MI_BATCH_BUFFER_END;
+	gem_write(i915, handle, 0, cs, sizeof(cs));
+
+	cs[i] = 2;
+	gem_write(i915, handle, 64, cs, sizeof(cs));
+
+	return handle;
+}
+
+static void test_syncobj_future(int i915, unsigned int engine)
+{
+	struct drm_i915_gem_exec_object2 obj = {
+		.offset = 24 << 20,
+		.handle = future_batch(i915, 24 << 20),
+		.flags = EXEC_OBJECT_PINNED,
+	};
+	struct drm_i915_gem_exec_fence fence = {};
+	struct drm_i915_gem_execbuffer2 execbuf  = {
+		.buffers_ptr = to_user_pointer(&obj),
+		.buffer_count = 1,
+		.cliprects_ptr = to_user_pointer(&fence),
+		.num_cliprects = 1,
+	};
+	const struct intel_execution_engine2 *e;
+
+	/*
+	 * Client A is waiting on a future fence from B. So even though its
+	 * execbuf is called first, we need to hold it in a queue waiting on
+	 * B.
+	 */
+	igt_require(gem_scheduler_enabled(i915));
+
+	__for_each_physical_engine(i915, e) {
+		uint32_t result;
+
+		igt_debug("waiting on future %s\n", e->name);
+		fence.handle = syncobj_create(i915, 0);
+
+		fence.flags = I915_EXEC_FENCE_WAIT;
+		execbuf.batch_start_offset = 0;
+		execbuf.flags = engine | I915_EXEC_FENCE_ARRAY;
+		execbuf.rsvd1 = 0;
+		gem_execbuf(i915, &execbuf); /* writes 1 */
+
+		fence.flags = I915_EXEC_FENCE_SIGNAL;
+		execbuf.batch_start_offset = 64;
+		execbuf.flags = e->flags | I915_EXEC_FENCE_ARRAY;
+		execbuf.rsvd1 = gem_context_clone_with_engines(i915, 0);
+		gem_execbuf(i915, &execbuf); /* writes 2 */
+		gem_context_destroy(i915, execbuf.rsvd1);
+
+		syncobj_destroy(i915, fence.handle);
+		gem_sync(i915, obj.handle); /* write hazard lies */
+		gem_read(i915, obj.handle, 4000, &result, sizeof(result));
+		igt_assert_eq(result, 1);
+	}
+
+	gem_close(i915, obj.handle);
+}
+
+static uint32_t future_submit_batch(int i915, uint32_t offset)
+{
+	uint32_t handle = gem_create(i915, 4096);
+	const int gen = intel_gen(intel_get_drm_devid(i915));
+	uint32_t cs[16];
+	int i = 0;
+
+	cs[i] = MI_STORE_DWORD_IMM | (gen < 6 ? 1 << 22 : 0);
+	if (gen >= 8) {
+		cs[++i] = offset + 4000;
+		cs[++i] = 0;
+	} else if (gen >= 4) {
+		cs[++i] = 0;
+		cs[++i] = offset + 4000;
+	} else {
+		cs[i]--;
+		cs[++i] = offset + 4000;
+	}
+	cs[++i] = 1;
+	cs[i + 1] = MI_BATCH_BUFFER_END;
+	igt_assert(i + 1 < ARRAY_SIZE(cs));
+	gem_write(i915, handle, 0, cs, sizeof(cs));
+
+	i = 0;
+	cs[i++] =
+		MI_SEMAPHORE_WAIT |
+		MI_SEMAPHORE_POLL |
+		MI_SEMAPHORE_SAD_EQ_SDD |
+		(4 - 2);
+	cs[i++] = 1;
+	cs[i++] = offset + 4000;
+	cs[i++] = 0;
+
+	cs[i] = MI_STORE_DWORD_IMM | (gen < 6 ? 1 << 22 : 0);
+	if (gen >= 8) {
+		cs[++i] = offset + 4000;
+		cs[++i] = 0;
+	} else if (gen >= 4) {
+		cs[++i] = 0;
+		cs[++i] = offset + 4000;
+	} else {
+		cs[i]--;
+		cs[++i] = offset + 4000;
+	}
+	cs[++i] = 2;
+	cs[++i] = MI_BATCH_BUFFER_END;
+	igt_assert(i < ARRAY_SIZE(cs));
+
+	gem_write(i915, handle, 64, cs, sizeof(cs));
+
+	return handle;
+}
+
+static void test_syncobj_future_submit(int i915, unsigned int engine)
+{
+	struct drm_i915_gem_exec_object2 obj = {
+		.offset = 24 << 20,
+		.handle = future_submit_batch(i915, 24 << 20),
+		.flags = EXEC_OBJECT_PINNED,
+	};
+	const struct intel_execution_engine2 *e;
+
+	/*
+	 * Here we submit client A waiting on client B, but internally client
+	 * B has a semaphore that waits on client A. This relies on timeslicing
+	 * to reorder B before A, even though userspace has asked to submit
+	 * A & B simultaneously (and due to the sequence we will submit B
+	 * then A).
+	 */
+	igt_require(gem_scheduler_has_timeslicing(i915));
+
+	__for_each_physical_engine(i915, e) {
+		struct drm_i915_gem_exec_fence fence = {
+			.handle = syncobj_create(i915, 0),
+		};
+		struct drm_i915_gem_execbuffer2 execbuf  = {
+			.buffers_ptr = to_user_pointer(&obj),
+			.buffer_count = 1,
+			.cliprects_ptr = to_user_pointer(&fence),
+			.num_cliprects = 1,
+			.flags = engine | I915_EXEC_FENCE_ARRAY,
+		};
+		uint32_t result;
+		int out;
+
+		igt_debug("waiting on future %s\n", e->name);
+
+		execbuf.rsvd1 = gem_context_clone_with_engines(i915, 0);
+		fence.flags = I915_EXEC_FENCE_WAIT | I915_EXEC_FENCE_WAIT_SUBMIT;
+		execbuf.batch_start_offset = 0;
+		execbuf.flags |= I915_EXEC_FENCE_OUT;
+		igt_require(__gem_execbuf_wr(i915, &execbuf) == 0); /* writes 1 */
+		execbuf.flags &= ~I915_EXEC_FENCE_OUT;
+		gem_context_destroy(i915, execbuf.rsvd1);
+
+		execbuf.rsvd1 = gem_context_clone_with_engines(i915, 0);
+		fence.flags = I915_EXEC_FENCE_SIGNAL;
+		execbuf.batch_start_offset = 64;
+		execbuf.flags &= ~I915_EXEC_RING_MASK;
+		execbuf.flags |= e->flags;
+		gem_execbuf(i915, &execbuf); /* writes 2 */
+		gem_context_destroy(i915, execbuf.rsvd1);
+
+		syncobj_destroy(i915, fence.handle);
+		gem_sync(i915, obj.handle); /* write hazard lies */
+		gem_read(i915, obj.handle, 4000, &result, sizeof(result));
+		igt_assert_eq(result, 2);
+
+		/* check we didn't autotimeout */
+		out = execbuf.rsvd2 >> 32;
+		igt_assert_eq(sync_fence_status(out), 1);
+		close(out);
+	}
+
+	gem_close(i915, obj.handle);
+}
+
+static void test_syncobj_future_past(int i915, unsigned int engine, int flags)
+{
+	struct drm_i915_gem_exec_object2 obj = {
+		.offset = 24 << 20,
+		.handle = future_batch(i915, 24 << 20),
+		.flags = EXEC_OBJECT_PINNED,
+	};
+	struct drm_i915_gem_exec_fence fence = {
+		.handle = syncobj_create(i915, 0),
+	};
+	struct drm_i915_gem_execbuffer2 execbuf  = {
+		.buffers_ptr = to_user_pointer(&obj),
+		.buffer_count = 1,
+		.cliprects_ptr = to_user_pointer(&fence),
+		.num_cliprects = 1,
+		.flags = engine | I915_EXEC_FENCE_ARRAY,
+	};
+	uint32_t result;
+	int out;
+
+	fence.flags = I915_EXEC_FENCE_WAIT | I915_EXEC_FENCE_SIGNAL | flags;
+	execbuf.batch_start_offset = 0;
+	execbuf.flags |= I915_EXEC_FENCE_OUT;
+	igt_require(__gem_execbuf_wr(i915, &execbuf) == 0); /* writes 1 */
+	execbuf.flags &= ~I915_EXEC_FENCE_OUT;
+
+	gem_sync(i915, obj.handle); /* write hazard lies */
+	gem_read(i915, obj.handle, 4000, &result, sizeof(result));
+	igt_assert_eq(result, 1);
+
+	/* check we didn't autotimeout */
+	out = execbuf.rsvd2 >> 32;
+	igt_assert_eq(sync_fence_status(out), 1);
+	close(out);
+
+	gem_close(i915, obj.handle);
+	syncobj_destroy(i915, fence.handle);
+}
+
+static void test_syncobj_future_self(int i915, unsigned int engine, int flags)
+{
+	struct drm_i915_gem_exec_object2 obj = {
+		.offset = 24 << 20,
+		.handle = future_batch(i915, 24 << 20),
+		.flags = EXEC_OBJECT_PINNED,
+	};
+	struct drm_i915_gem_exec_fence fence = {
+		.handle = syncobj_create(i915, 0),
+	};
+	struct drm_i915_gem_execbuffer2 execbuf  = {
+		.buffers_ptr = to_user_pointer(&obj),
+		.buffer_count = 1,
+		.cliprects_ptr = to_user_pointer(&fence),
+		.num_cliprects = 1,
+		.flags = engine | I915_EXEC_FENCE_ARRAY,
+	};
+	int out;
+
+	fence.flags = I915_EXEC_FENCE_WAIT | flags;
+	execbuf.batch_start_offset = 0;
+	execbuf.flags |= I915_EXEC_FENCE_OUT;
+	igt_require(__gem_execbuf_wr(i915, &execbuf) == 0); /* writes 1 */
+	execbuf.flags &= ~I915_EXEC_FENCE_OUT;
+
+	fence.flags = I915_EXEC_FENCE_SIGNAL;
+	execbuf.batch_start_offset = 64;
+	gem_execbuf(i915, &execbuf); /* writes 2 */
+
+	gem_sync(i915, obj.handle); /* write hazard lies */
+
+	gem_close(i915, obj.handle);
+	syncobj_destroy(i915, fence.handle);
+
+	out = execbuf.rsvd2 >> 32;
+	igt_assert_eq(sync_fence_status(out), -EDEADLK);
+	close(out);
+}
+
+static void test_syncobj_future_pair(int i915, unsigned int engine)
+{
+	struct drm_i915_gem_exec_object2 obj = {
+		.offset = 24 << 20,
+		.handle = submitN_batches(i915, 24 << 20, 2),
+		.flags = EXEC_OBJECT_PINNED,
+	};
+	struct drm_i915_gem_exec_fence fence[2] = {
+		{ .handle = syncobj_create(i915, 0) },
+		{ .handle = syncobj_create(i915, 0) }
+	};
+	struct drm_i915_gem_execbuffer2 execbuf  = {
+		.buffers_ptr = to_user_pointer(&obj),
+		.buffer_count = 1,
+		.cliprects_ptr = to_user_pointer(fence),
+		.num_cliprects = 2,
+	};
+	const struct intel_execution_engine2 *e;
+
+	__for_each_physical_engine(i915, e) {
+		int out = 0;
+
+		gem_write(i915, obj.handle, 0, &out, sizeof(out));
+		fence[0].handle = syncobj_create(i915, 0);
+		fence[1].handle = syncobj_create(i915, 0);
+
+		fence[0].flags = I915_EXEC_FENCE_SIGNAL;
+		fence[1].flags =
+			I915_EXEC_FENCE_WAIT | I915_EXEC_FENCE_WAIT_SUBMIT;
+		execbuf.batch_start_offset = 1024;
+		execbuf.flags =
+			engine | I915_EXEC_FENCE_ARRAY | I915_EXEC_FENCE_OUT;
+		execbuf.rsvd1 = gem_context_create(i915);
+		igt_require(__gem_execbuf_wr(i915, &execbuf) == 0);
+		gem_context_destroy(i915, execbuf.rsvd1);
+		execbuf.rsvd2 >>= 32;
+
+		fence[0].flags =
+			I915_EXEC_FENCE_WAIT | I915_EXEC_FENCE_WAIT_SUBMIT;
+		fence[1].flags = I915_EXEC_FENCE_SIGNAL;
+		execbuf.batch_start_offset = 2048;
+		execbuf.flags =
+			e->flags | I915_EXEC_FENCE_ARRAY | I915_EXEC_FENCE_OUT;
+		execbuf.rsvd1 = gem_context_create(i915);
+		gem_execbuf_wr(i915, &execbuf);
+		gem_context_destroy(i915, execbuf.rsvd1);
+
+		syncobj_destroy(i915, fence[0].handle);
+		syncobj_destroy(i915, fence[1].handle);
+
+		gem_sync(i915, obj.handle); /* write hazard lies */
+
+		out = execbuf.rsvd2;
+		igt_assert_eq(sync_fence_status(out), 1);
+		close(out);
+
+		out = execbuf.rsvd2 >> 32;
+		igt_assert_eq(sync_fence_status(out), 1);
+		close(out);
+
+		gem_read(i915, obj.handle, 0, &out, sizeof(out));
+		igt_assert_eq(out, 16);
+	}
+
+	gem_close(i915, obj.handle);
+}
+
+static void test_syncobj_future_group(int i915, unsigned int engine, int count)
+{
+	struct drm_i915_gem_exec_object2 obj = {
+		.offset = 24 << 20,
+		.handle = submitN_batches(i915, 24 << 20, count),
+		.flags = EXEC_OBJECT_PINNED,
+	};
+	struct drm_i915_gem_exec_fence fence[count];
+	struct drm_i915_gem_execbuffer2 execbuf  = {
+		.buffers_ptr = to_user_pointer(&obj),
+		.buffer_count = 1,
+		.cliprects_ptr = to_user_pointer(fence),
+		.num_cliprects = count,
+		.flags = engine | I915_EXEC_FENCE_ARRAY | I915_EXEC_FENCE_OUT,
+	};
+	int out[count];
+	uint32_t result;
+
+	for (int i = 0; i < count; i++) {
+		fence[i].handle = syncobj_create(i915, 0);
+		fence[i].flags =
+			I915_EXEC_FENCE_WAIT | I915_EXEC_FENCE_WAIT_SUBMIT;
+	}
+
+	for (int i = 0; i < count; i++) {
+		fence[i].flags = I915_EXEC_FENCE_SIGNAL;
+
+		execbuf.batch_start_offset = 1024 * (i + 1);
+		execbuf.rsvd1 = gem_context_create(i915);
+		gem_execbuf_wr(i915, &execbuf);
+		gem_context_destroy(i915, execbuf.rsvd1);
+
+		out[i] = execbuf.rsvd2 >> 32;
+		fence[i].flags =
+			I915_EXEC_FENCE_WAIT | I915_EXEC_FENCE_WAIT_SUBMIT;
+	}
+	gem_sync(i915, obj.handle); /* write hazard lies */
+
+	/* As both batches were waiting for the other to start -- deadlock? */
+	for (int i = 0; i < count; i++)  {
+		syncobj_destroy(i915, fence[i].handle);
+		igt_assert_eq(sync_fence_status(out[i]), 1);
+		close(out[i]);
+	}
+
+	/* Nevertheless, we ignored^Wresolved the deadlock and let them run */
+	gem_read(i915, obj.handle, 0, &result, sizeof(result));
+	igt_assert_eq(result, 8 * count);
+	gem_close(i915, obj.handle);
+}
+
+
+static void
+test_syncobj_future_deadlock(int i915, unsigned int engine, int flags)
+{
+	struct drm_i915_gem_exec_object2 obj = {
+		.offset = 24 << 20,
+		.handle = future_batch(i915, 24 << 20),
+		.flags = EXEC_OBJECT_PINNED,
+	};
+	struct drm_i915_gem_exec_fence fence = {};
+	struct drm_i915_gem_execbuffer2 execbuf  = {
+		.buffers_ptr = to_user_pointer(&obj),
+		.buffer_count = 1,
+		.cliprects_ptr = to_user_pointer(&fence),
+		.num_cliprects = 1,
+	};
+	const struct intel_execution_engine2 *e;
+
+	__for_each_physical_engine(i915, e) {
+		int out;
+
+		fence.handle = syncobj_create(i915, 0),
+
+		fence.flags = I915_EXEC_FENCE_WAIT | flags;
+		execbuf.batch_start_offset = 0;
+		execbuf.flags = engine | I915_EXEC_FENCE_ARRAY,
+		execbuf.flags |= I915_EXEC_FENCE_OUT;
+		execbuf.rsvd1 = 0;
+		gem_execbuf_wr(i915, &execbuf); /* writes 1 */
+
+		fence.flags = I915_EXEC_FENCE_SIGNAL;
+		execbuf.batch_start_offset = 64;
+		execbuf.flags = e->flags | I915_EXEC_FENCE_ARRAY,
+		execbuf.flags |= I915_EXEC_FENCE_OUT | I915_EXEC_FENCE_IN;
+		execbuf.rsvd1 = gem_context_clone_with_engines(i915, 0);
+		execbuf.rsvd2 >>= 32;
+		gem_execbuf_wr(i915, &execbuf); /* writes 2 */
+		gem_context_destroy(i915, execbuf.rsvd1);
+
+		syncobj_destroy(i915, fence.handle);
+		gem_sync(i915, obj.handle);
+
+		/* How should this deadlock be resolved? */
+		out = execbuf.rsvd2;
+		igt_assert_eq(sync_fence_status(out), -EDEADLK);
+		close(out);
+
+		out = execbuf.rsvd2 >> 32;
+		igt_assert_eq(sync_fence_status(out), -EDEADLK);
+		close(out);
+	}
+
+	gem_close(i915, obj.handle);
+}
+
+static void
+test_syncobj_future_cycle(int i915, unsigned int engine, int flags)
+{
+	struct drm_i915_gem_exec_object2 obj = {
+		.offset = 24 << 20,
+		.handle = future_batch(i915, 24 << 20),
+		.flags = EXEC_OBJECT_PINNED,
+	};
+	const struct intel_execution_engine2 *e1, *e2;
+
+	__for_each_physical_engine(i915, e1) {
+	__for_each_physical_engine(i915, e2) {
+		struct drm_i915_gem_exec_fence fence = {
+			.handle = syncobj_create(i915, 0),
+		};
+		struct drm_i915_gem_execbuffer2 execbuf  = {
+			.buffers_ptr = to_user_pointer(&obj),
+			.buffer_count = 1,
+			.cliprects_ptr = to_user_pointer(&fence),
+			.num_cliprects = 1,
+			.flags = engine | I915_EXEC_FENCE_ARRAY,
+		};
+		int out;
+
+		fence.flags = I915_EXEC_FENCE_WAIT | flags;
+		execbuf.batch_start_offset = 0;
+		execbuf.flags |= I915_EXEC_FENCE_OUT;
+		igt_require(__gem_execbuf_wr(i915, &execbuf) == 0);
+
+		fence.flags = 0;
+		execbuf.rsvd1 = gem_context_clone_with_engines(i915, 0);
+		execbuf.rsvd2 >>= 32;
+		execbuf.flags &= ~I915_EXEC_RING_MASK;
+		execbuf.flags |= e1->flags | I915_EXEC_FENCE_IN;
+		gem_execbuf_wr(i915, &execbuf);
+		gem_context_destroy(i915, execbuf.rsvd1);
+		close(execbuf.rsvd2);
+
+		fence.flags = I915_EXEC_FENCE_SIGNAL;
+		execbuf.rsvd1 = gem_context_clone_with_engines(i915, 0);
+		execbuf.rsvd2 >>= 32;
+		execbuf.flags &= ~I915_EXEC_RING_MASK;
+		execbuf.flags |= e2->flags;
+		execbuf.batch_start_offset = 64;
+		gem_execbuf_wr(i915, &execbuf); /* writes 2 */
+		gem_context_destroy(i915, execbuf.rsvd1);
+
+		syncobj_destroy(i915, fence.handle);
+		gem_sync(i915, obj.handle);
+
+
+		/* How should this deadlock be resolved? */
+		out = execbuf.rsvd2 >> 32;
+		igt_assert_eq(sync_fence_status(out), -EDEADLK);
+		close(out);
+
+		out = execbuf.rsvd2;
+		igt_assert_eq(sync_fence_status(out), -EDEADLK);
+		close(out);
+	}}
+
+	gem_close(i915, obj.handle);
+}
+
 static void test_syncobj_export(int fd)
 {
 	const uint32_t bbe = MI_BATCH_BUFFER_END;
@@ -1574,6 +2131,117 @@ static void test_syncobj_channel(int fd)
 		syncobj_destroy(fd, syncobj[i]);
 }
 
+static bool has_future_syncobj(int i915)
+{
+	struct drm_i915_gem_exec_object2 obj = {
+		.handle = batch_create(i915),
+	};
+	struct drm_i915_gem_exec_fence fence = {
+		.handle = syncobj_create(i915, 0),
+		.flags = I915_EXEC_FENCE_WAIT | I915_EXEC_FENCE_SIGNAL,
+	};
+	struct drm_i915_gem_execbuffer2 execbuf  = {
+		.buffers_ptr = to_user_pointer(&obj),
+		.buffer_count = 1,
+		.cliprects_ptr = to_user_pointer(&fence),
+		.num_cliprects = 1,
+		.flags = I915_EXEC_FENCE_ARRAY,
+	};
+	bool result;
+
+	result = __gem_execbuf(i915, &execbuf) == 0;
+	gem_close(i915, obj.handle);
+
+	return result;
+}
+
+static void syncobj_futures(int i915)
+{
+	const struct intel_execution_engine2 *e;
+
+	igt_fixture {
+		igt_require(gem_scheduler_enabled(i915));
+		igt_require(has_future_syncobj(i915));
+	}
+
+	igt_subtest_with_dynamic("syncobj-future")
+		__for_each_physical_engine(i915, e) {
+			igt_dynamic_f("%s", e->name)
+				test_syncobj_future(i915, e->flags);
+		}
+
+	igt_subtest_with_dynamic("syncobj-future-past")
+		__for_each_physical_engine(i915, e) {
+			igt_dynamic_f("%s", e->name)
+				test_syncobj_future_past(i915, e->flags, 0);
+		}
+
+
+	igt_subtest_with_dynamic("syncobj-future-submit")
+		__for_each_physical_engine(i915, e) {
+			igt_dynamic_f("%s", e->name)
+				test_syncobj_future_submit(i915, e->flags);
+		}
+
+	igt_subtest_with_dynamic("syncobj-future-submit-past")
+		__for_each_physical_engine(i915, e) {
+			igt_dynamic_f("%s", e->name)
+				test_syncobj_future_past(i915, e->flags,
+						I915_EXEC_FENCE_WAIT_SUBMIT);
+		}
+
+	igt_subtest_with_dynamic("syncobj-future-self")
+		__for_each_physical_engine(i915, e) {
+			igt_dynamic_f("%s", e->name)
+				test_syncobj_future_self(i915, e->flags, 0);
+		}
+
+	igt_subtest_with_dynamic("syncobj-future-self-submit")
+		__for_each_physical_engine(i915, e) {
+			igt_dynamic_f("%s", e->name)
+				test_syncobj_future_self(i915, e->flags,
+						I915_EXEC_FENCE_WAIT_SUBMIT);
+		}
+
+	igt_subtest_with_dynamic("syncobj-future-pair")
+		__for_each_physical_engine(i915, e) {
+			igt_dynamic_f("%s", e->name)
+				test_syncobj_future_pair(i915, e->flags);
+		}
+
+	igt_subtest_with_dynamic("syncobj-future-group")
+		__for_each_physical_engine(i915, e) {
+			igt_dynamic_f("%s", e->name)
+				test_syncobj_future_group(i915, e->flags, 67);
+		}
+
+	igt_subtest_with_dynamic("syncobj-future-deadlock")
+		__for_each_physical_engine(i915, e) {
+			igt_dynamic_f("%s", e->name)
+				test_syncobj_future_deadlock(i915, e->flags, 0);
+		}
+
+	igt_subtest_with_dynamic("syncobj-future-submit-deadlock")
+		__for_each_physical_engine(i915, e) {
+			igt_dynamic_f("%s", e->name)
+				test_syncobj_future_deadlock(i915, e->flags,
+							     I915_EXEC_FENCE_WAIT_SUBMIT);
+		}
+
+	igt_subtest_with_dynamic("syncobj-future-cycle")
+		__for_each_physical_engine(i915, e) {
+			igt_dynamic_f("%s", e->name)
+				test_syncobj_future_cycle(i915, e->flags, 0);
+		}
+
+	igt_subtest_with_dynamic("syncobj-future-submit-cycle")
+		__for_each_physical_engine(i915, e) {
+			igt_dynamic_f("%s", e->name)
+				test_syncobj_future_cycle(i915, e->flags,
+							  I915_EXEC_FENCE_WAIT_SUBMIT);
+		}
+}
+
 igt_main
 {
 	const struct intel_execution_engine2 *e;
@@ -1786,6 +2454,9 @@ igt_main
 		igt_subtest("syncobj-invalid-wait")
 			test_syncobj_invalid_wait(i915);
 
+		igt_subtest("syncobj-incomplete-wait-submit")
+			test_syncobj_incomplete_wait_submit(i915);
+
 		igt_subtest("syncobj-invalid-flags")
 			test_syncobj_invalid_flags(i915);
 
@@ -1795,6 +2466,9 @@ igt_main
 		igt_subtest("syncobj-wait")
 			test_syncobj_wait(i915);
 
+		igt_subtest_group
+			syncobj_futures(i915);
+
 		igt_subtest("syncobj-export")
 			test_syncobj_export(i915);
 
-- 
2.26.2

_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

[Intel-gfx] [PATCH i-g-t 2/2] i915/gem_exec_fence: Teach invalid-wait about invalid future fences

[Chris Wilson]

When we allow a wait on a future future fence, it must autoexpire if the
fence is never signaled by userspace. Also put future fences to work, as
the intention is to use them, along with WAIT_SUBMIT and semaphores, for
userspace to perform its own fine-grained scheduling. Or simply run
concurrent clients without having to flush batches between context
switches.

v2: Verify deadlock detection

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
---
 tests/i915/gem_exec_fence.c | 430 +++++++++++++++++++++++++++++++++++-
 1 file changed, 427 insertions(+), 3 deletions(-)

diff --git a/tests/i915/gem_exec_fence.c b/tests/i915/gem_exec_fence.c
index 17fdaebd5..374b273e4 100644
--- a/tests/i915/gem_exec_fence.c
+++ b/tests/i915/gem_exec_fence.c
@@ -47,6 +47,15 @@ struct sync_merge_data {
 #define SYNC_IOC_MERGE _IOWR(SYNC_IOC_MAGIC, 3, struct sync_merge_data)
 #endif
 
+#define MI_SEMAPHORE_WAIT		(0x1c << 23)
+#define   MI_SEMAPHORE_POLL             (1 << 15)
+#define   MI_SEMAPHORE_SAD_GT_SDD       (0 << 12)
+#define   MI_SEMAPHORE_SAD_GTE_SDD      (1 << 12)
+#define   MI_SEMAPHORE_SAD_LT_SDD       (2 << 12)
+#define   MI_SEMAPHORE_SAD_LTE_SDD      (3 << 12)
+#define   MI_SEMAPHORE_SAD_EQ_SDD       (4 << 12)
+#define   MI_SEMAPHORE_SAD_NEQ_SDD      (5 << 12)
+
 static void store(int fd, const struct intel_execution_engine2 *e,
 		  int fence, uint32_t target, unsigned offset_value)
 {
@@ -913,11 +922,12 @@ static void test_syncobj_invalid_wait(int fd)
 	struct drm_i915_gem_exec_fence fence = {
 		.handle = syncobj_create(fd, 0),
 	};
+	int out;
 
 	memset(&execbuf, 0, sizeof(execbuf));
 	execbuf.buffers_ptr = to_user_pointer(&obj);
 	execbuf.buffer_count = 1;
-	execbuf.flags = I915_EXEC_FENCE_ARRAY;
+	execbuf.flags = I915_EXEC_FENCE_ARRAY | I915_EXEC_FENCE_OUT;
 	execbuf.cliprects_ptr = to_user_pointer(&fence);
 	execbuf.num_cliprects = 1;
 
@@ -925,14 +935,59 @@ static void test_syncobj_invalid_wait(int fd)
 	obj.handle = gem_create(fd, 4096);
 	gem_write(fd, obj.handle, 0, &bbe, sizeof(bbe));
 
-	/* waiting before the fence is set is invalid */
+	/* waiting before the fence is set is^W may be invalid */
 	fence.flags = I915_EXEC_FENCE_WAIT;
-	igt_assert_eq(__gem_execbuf(fd, &execbuf), -EINVAL);
+	if (__gem_execbuf_wr(fd, &execbuf)) {
+		igt_assert_eq(__gem_execbuf(fd, &execbuf), -EINVAL);
+		return;
+	}
+
+	/* If we do allow the wait on a future fence, it should autoexpire */
+	gem_sync(fd, obj.handle);
+	out = execbuf.rsvd2 >> 32;
+	igt_assert_eq(sync_fence_status(out), -ETIMEDOUT);
+	close(out);
 
 	gem_close(fd, obj.handle);
 	syncobj_destroy(fd, fence.handle);
 }
 
+static void test_syncobj_incomplete_wait_submit(int i915)
+{
+	struct drm_i915_gem_exec_object2 obj = {
+		.handle = batch_create(i915),
+	};
+	struct drm_i915_gem_exec_fence fence = {
+		.handle = syncobj_create(i915, 0),
+		.flags = I915_EXEC_FENCE_WAIT | I915_EXEC_FENCE_WAIT_SUBMIT,
+	};
+	struct drm_i915_gem_execbuffer2 execbuf = {
+		.buffers_ptr = to_user_pointer(&obj),
+		.buffer_count = 1,
+
+		.cliprects_ptr = to_user_pointer(&fence),
+		.num_cliprects = 1,
+
+		.flags = I915_EXEC_FENCE_ARRAY | I915_EXEC_FENCE_OUT,
+	};
+	int out;
+
+	/* waiting before the fence is set is^W may be invalid */
+	if (__gem_execbuf_wr(i915, &execbuf)) {
+		igt_assert_eq(__gem_execbuf(i915, &execbuf), -EINVAL);
+		return;
+	}
+
+	/* If we do allow the wait on a future fence, it should autoexpire */
+	gem_sync(i915, obj.handle);
+	out = execbuf.rsvd2 >> 32;
+	igt_assert_eq(sync_fence_status(out), -ETIMEDOUT);
+	close(out);
+
+	gem_close(i915, obj.handle);
+	syncobj_destroy(i915, fence.handle);
+}
+
 static void test_syncobj_invalid_flags(int fd)
 {
 	const uint32_t bbe = MI_BATCH_BUFFER_END;
@@ -1079,6 +1134,319 @@ static void test_syncobj_wait(int fd)
 	}
 }
 
+static uint32_t future_batch(int i915, uint32_t offset)
+{
+	uint32_t handle = gem_create(i915, 4096);
+	const int gen = intel_gen(intel_get_drm_devid(i915));
+	uint32_t cs[16];
+	int i = 0;
+
+	cs[i] = MI_STORE_DWORD_IMM | (gen < 6 ? 1 << 22 : 0);
+	if (gen >= 8) {
+		cs[++i] = offset + 4000;
+		cs[++i] = 0;
+	} else if (gen >= 4) {
+		cs[++i] = 0;
+		cs[++i] = offset + 4000;
+	} else {
+		cs[i]--;
+		cs[++i] = offset + 4000;
+	}
+	cs[++i] = 1;
+	cs[i + 1] = MI_BATCH_BUFFER_END;
+	gem_write(i915, handle, 0, cs, sizeof(cs));
+
+	cs[i] = 2;
+	gem_write(i915, handle, 64, cs, sizeof(cs));
+
+	return handle;
+}
+
+static void test_syncobj_future(int i915, unsigned int engine)
+{
+	struct drm_i915_gem_exec_object2 obj = {
+		.offset = 24 << 20,
+		.handle = future_batch(i915, 24 << 20),
+		.flags = EXEC_OBJECT_PINNED,
+	};
+	struct drm_i915_gem_exec_fence fence = {
+		.handle = syncobj_create(i915, 0),
+	};
+	struct drm_i915_gem_execbuffer2 execbuf  = {
+		.buffers_ptr = to_user_pointer(&obj),
+		.buffer_count = 1,
+		.cliprects_ptr = to_user_pointer(&fence),
+		.num_cliprects = 1,
+		.flags = engine | I915_EXEC_FENCE_ARRAY,
+	};
+	uint32_t result;
+
+	/*
+	 * Client A is waiting on a future fence from B. So even though its
+	 * execbuf is called first, we need to hold it in a queue waiting on
+	 * B.
+	 */
+	igt_require(gem_scheduler_enabled(i915));
+
+	execbuf.rsvd1 = gem_context_create(i915);
+	fence.flags = I915_EXEC_FENCE_WAIT;
+	execbuf.batch_start_offset = 0;
+	igt_require(__gem_execbuf(i915, &execbuf) == 0); /* writes 1 */
+	gem_context_destroy(i915, execbuf.rsvd1);
+
+	execbuf.rsvd1 = gem_context_create(i915);
+	fence.flags = I915_EXEC_FENCE_SIGNAL;
+	execbuf.batch_start_offset = 64;
+	gem_execbuf(i915, &execbuf); /* writes 2 */
+	gem_context_destroy(i915, execbuf.rsvd1);
+
+	gem_sync(i915, obj.handle); /* write hazard lies */
+	gem_read(i915, obj.handle, 4000, &result, sizeof(result));
+	igt_assert_eq(result, 1);
+
+	gem_close(i915, obj.handle);
+	syncobj_destroy(i915, fence.handle);
+}
+
+static uint32_t future_submit_batch(int i915, uint32_t offset)
+{
+	uint32_t handle = gem_create(i915, 4096);
+	const int gen = intel_gen(intel_get_drm_devid(i915));
+	uint32_t cs[16];
+	int i = 0;
+
+	cs[i] = MI_STORE_DWORD_IMM | (gen < 6 ? 1 << 22 : 0);
+	if (gen >= 8) {
+		cs[++i] = offset + 4000;
+		cs[++i] = 0;
+	} else if (gen >= 4) {
+		cs[++i] = 0;
+		cs[++i] = offset + 4000;
+	} else {
+		cs[i]--;
+		cs[++i] = offset + 4000;
+	}
+	cs[++i] = 1;
+	cs[i + 1] = MI_BATCH_BUFFER_END;
+	igt_assert(i + 1 < ARRAY_SIZE(cs));
+	gem_write(i915, handle, 0, cs, sizeof(cs));
+
+	i = 0;
+	cs[i++] =
+		MI_SEMAPHORE_WAIT |
+		MI_SEMAPHORE_POLL |
+		MI_SEMAPHORE_SAD_EQ_SDD |
+		(4 - 2);
+	cs[i++] = 1;
+	cs[i++] = offset + 4000;
+	cs[i++] = 0;
+
+	cs[i] = MI_STORE_DWORD_IMM | (gen < 6 ? 1 << 22 : 0);
+	if (gen >= 8) {
+		cs[++i] = offset + 4000;
+		cs[++i] = 0;
+	} else if (gen >= 4) {
+		cs[++i] = 0;
+		cs[++i] = offset + 4000;
+	} else {
+		cs[i]--;
+		cs[++i] = offset + 4000;
+	}
+	cs[++i] = 2;
+	cs[++i] = MI_BATCH_BUFFER_END;
+	igt_assert(i < ARRAY_SIZE(cs));
+
+	gem_write(i915, handle, 64, cs, sizeof(cs));
+
+	return handle;
+}
+
+static void test_syncobj_future_submit(int i915, unsigned int engine)
+{
+	struct drm_i915_gem_exec_object2 obj = {
+		.offset = 24 << 20,
+		.handle = future_submit_batch(i915, 24 << 20),
+		.flags = EXEC_OBJECT_PINNED,
+	};
+	struct drm_i915_gem_exec_fence fence = {
+		.handle = syncobj_create(i915, 0),
+	};
+	struct drm_i915_gem_execbuffer2 execbuf  = {
+		.buffers_ptr = to_user_pointer(&obj),
+		.buffer_count = 1,
+		.cliprects_ptr = to_user_pointer(&fence),
+		.num_cliprects = 1,
+		.flags = engine | I915_EXEC_FENCE_ARRAY,
+	};
+	uint32_t result;
+	int out;
+
+	/*
+	 * Here we submit client A waiting on client B, but internally client
+	 * B has a semaphore that waits on client A. This relies on timeslicing
+	 * to reorder B before A, even though userspace has asked to submit
+	 * A & B simultaneously (and due to the sequence we will submit B
+	 * then A).
+	 */
+	igt_require(gem_scheduler_has_timeslicing(i915));
+
+	execbuf.rsvd1 = gem_context_create(i915);
+	fence.flags = I915_EXEC_FENCE_WAIT | I915_EXEC_FENCE_WAIT_SUBMIT;
+	execbuf.batch_start_offset = 0;
+	execbuf.flags |= I915_EXEC_FENCE_OUT;
+	igt_require(__gem_execbuf_wr(i915, &execbuf) == 0); /* writes 1 */
+	execbuf.flags &= ~I915_EXEC_FENCE_OUT;
+	gem_context_destroy(i915, execbuf.rsvd1);
+
+	execbuf.rsvd1 = gem_context_create(i915);
+	fence.flags = I915_EXEC_FENCE_SIGNAL;
+	execbuf.batch_start_offset = 64;
+	gem_execbuf(i915, &execbuf); /* writes 2 */
+	gem_context_destroy(i915, execbuf.rsvd1);
+
+	gem_sync(i915, obj.handle); /* write hazard lies */
+	gem_read(i915, obj.handle, 4000, &result, sizeof(result));
+	igt_assert_eq(result, 2);
+
+	/* check we didn't autotimeout */
+	out = execbuf.rsvd2 >> 32;
+	igt_assert_eq(sync_fence_status(out), 1);
+	close(out);
+
+	gem_close(i915, obj.handle);
+	syncobj_destroy(i915, fence.handle);
+}
+
+static void test_syncobj_future_past(int i915, unsigned int engine, int flags)
+{
+	struct drm_i915_gem_exec_object2 obj = {
+		.offset = 24 << 20,
+		.handle = future_batch(i915, 24 << 20),
+		.flags = EXEC_OBJECT_PINNED,
+	};
+	struct drm_i915_gem_exec_fence fence = {
+		.handle = syncobj_create(i915, 0),
+	};
+	struct drm_i915_gem_execbuffer2 execbuf  = {
+		.buffers_ptr = to_user_pointer(&obj),
+		.buffer_count = 1,
+		.cliprects_ptr = to_user_pointer(&fence),
+		.num_cliprects = 1,
+		.flags = engine | I915_EXEC_FENCE_ARRAY,
+	};
+	uint32_t result;
+	int out;
+
+	fence.flags = I915_EXEC_FENCE_WAIT | I915_EXEC_FENCE_SIGNAL | flags;
+	execbuf.batch_start_offset = 0;
+	execbuf.flags |= I915_EXEC_FENCE_OUT;
+	igt_require(__gem_execbuf_wr(i915, &execbuf) == 0); /* writes 1 */
+	execbuf.flags &= ~I915_EXEC_FENCE_OUT;
+
+	gem_sync(i915, obj.handle); /* write hazard lies */
+	gem_read(i915, obj.handle, 4000, &result, sizeof(result));
+	igt_assert_eq(result, 1);
+
+	/* check we didn't autotimeout */
+	out = execbuf.rsvd2 >> 32;
+	igt_assert_eq(sync_fence_status(out), 1);
+	close(out);
+
+	gem_close(i915, obj.handle);
+	syncobj_destroy(i915, fence.handle);
+}
+
+static void test_syncobj_future_self(int i915, unsigned int engine, int flags)
+{
+	struct drm_i915_gem_exec_object2 obj = {
+		.offset = 24 << 20,
+		.handle = future_batch(i915, 24 << 20),
+		.flags = EXEC_OBJECT_PINNED,
+	};
+	struct drm_i915_gem_exec_fence fence = {
+		.handle = syncobj_create(i915, 0),
+	};
+	struct drm_i915_gem_execbuffer2 execbuf  = {
+		.buffers_ptr = to_user_pointer(&obj),
+		.buffer_count = 1,
+		.cliprects_ptr = to_user_pointer(&fence),
+		.num_cliprects = 1,
+		.flags = engine | I915_EXEC_FENCE_ARRAY,
+	};
+	uint32_t result;
+	int out;
+
+	fence.flags = I915_EXEC_FENCE_WAIT | flags;
+	execbuf.batch_start_offset = 0;
+	execbuf.flags |= I915_EXEC_FENCE_OUT;
+	igt_require(__gem_execbuf_wr(i915, &execbuf) == 0); /* writes 1 */
+	execbuf.flags &= ~I915_EXEC_FENCE_OUT;
+
+	fence.flags = I915_EXEC_FENCE_SIGNAL;
+	execbuf.batch_start_offset = 64;
+	gem_execbuf(i915, &execbuf); /* writes 2 */
+
+	gem_sync(i915, obj.handle); /* write hazard lies */
+	gem_read(i915, obj.handle, 4000, &result, sizeof(result));
+	igt_assert_eq(result, 2);
+
+	/* check we didn't autotimeout */
+	out = execbuf.rsvd2 >> 32;
+	igt_assert_eq(sync_fence_status(out), 1);
+	close(out);
+
+	gem_close(i915, obj.handle);
+	syncobj_destroy(i915, fence.handle);
+}
+
+static void
+test_syncobj_future_deadlock(int i915, unsigned int engine, int flags)
+{
+	struct drm_i915_gem_exec_object2 obj = {
+		.offset = 24 << 20,
+		.handle = future_batch(i915, 24 << 20),
+		.flags = EXEC_OBJECT_PINNED,
+	};
+	struct drm_i915_gem_exec_fence fence = {
+		.handle = syncobj_create(i915, 0),
+	};
+	struct drm_i915_gem_execbuffer2 execbuf  = {
+		.buffers_ptr = to_user_pointer(&obj),
+		.buffer_count = 1,
+		.cliprects_ptr = to_user_pointer(&fence),
+		.num_cliprects = 1,
+		.flags = engine | I915_EXEC_FENCE_ARRAY,
+	};
+	int out;
+
+	fence.flags = I915_EXEC_FENCE_WAIT | flags;
+	execbuf.batch_start_offset = 0;
+	execbuf.flags |= I915_EXEC_FENCE_OUT;
+	igt_require(__gem_execbuf_wr(i915, &execbuf) == 0); /* writes 1 */
+
+	fence.flags = I915_EXEC_FENCE_SIGNAL;
+	execbuf.rsvd1 = gem_context_create(i915);
+	execbuf.rsvd2 >>= 32;
+	execbuf.flags |= I915_EXEC_FENCE_IN;
+	execbuf.batch_start_offset = 64;
+	gem_execbuf_wr(i915, &execbuf); /* writes 2 */
+	gem_context_destroy(i915, execbuf.rsvd1);
+
+	gem_sync(i915, obj.handle);
+
+	gem_close(i915, obj.handle);
+	syncobj_destroy(i915, fence.handle);
+
+	/* How should this deadlock be resolved? */
+	out = execbuf.rsvd2;
+	igt_assert_eq(sync_fence_status(out), -EDEADLK);
+	close(out);
+
+	out = execbuf.rsvd2 >> 32;
+	igt_assert_eq(sync_fence_status(out), -EDEADLK);
+	close(out);
+}
+
 static void test_syncobj_export(int fd)
 {
 	const uint32_t bbe = MI_BATCH_BUFFER_END;
@@ -1548,6 +1916,9 @@ igt_main
 		igt_subtest("syncobj-invalid-wait")
 			test_syncobj_invalid_wait(i915);
 
+		igt_subtest("syncobj-incomplete-wait-submit")
+			test_syncobj_incomplete_wait_submit(i915);
+
 		igt_subtest("syncobj-invalid-flags")
 			test_syncobj_invalid_flags(i915);
 
@@ -1557,6 +1928,59 @@ igt_main
 		igt_subtest("syncobj-wait")
 			test_syncobj_wait(i915);
 
+		igt_subtest_with_dynamic("syncobj-future")
+			__for_each_physical_engine(i915, e) {
+				igt_dynamic_f("%s", e->name)
+					test_syncobj_future(i915, e->flags);
+			}
+
+		igt_subtest_with_dynamic("syncobj-future-past")
+			__for_each_physical_engine(i915, e) {
+				igt_dynamic_f("%s", e->name)
+					test_syncobj_future_past(i915, e->flags,
+								 0);
+			}
+
+
+		igt_subtest_with_dynamic("syncobj-future-submit")
+			__for_each_physical_engine(i915, e) {
+				igt_dynamic_f("%s", e->name)
+					test_syncobj_future_submit(i915, e->flags);
+			}
+
+		igt_subtest_with_dynamic("syncobj-future-submit-past")
+			__for_each_physical_engine(i915, e) {
+				igt_dynamic_f("%s", e->name)
+					test_syncobj_future_past(i915, e->flags,
+								 I915_EXEC_FENCE_WAIT_SUBMIT);
+			}
+
+		igt_subtest_with_dynamic("syncobj-future-self")
+			__for_each_physical_engine(i915, e) {
+				igt_dynamic_f("%s", e->name)
+					test_syncobj_future_self(i915, e->flags,
+								 0);
+			}
+
+		igt_subtest_with_dynamic("syncobj-future-self-submit")
+			__for_each_physical_engine(i915, e) {
+				igt_dynamic_f("%s", e->name)
+					test_syncobj_future_self(i915, e->flags,
+								 I915_EXEC_FENCE_WAIT_SUBMIT);
+			}
+
+		igt_subtest_with_dynamic("syncobj-future-deadlock")
+			__for_each_physical_engine(i915, e) {
+				igt_dynamic_f("%s", e->name)
+					test_syncobj_future_deadlock(i915, e->flags, 0);
+			}
+
+		igt_subtest_with_dynamic("syncobj-future-submit-deadlock")
+			__for_each_physical_engine(i915, e) {
+				igt_dynamic_f("%s", e->name)
+					test_syncobj_future_deadlock(i915, e->flags, I915_EXEC_FENCE_WAIT_SUBMIT);
+			}
+
 		igt_subtest("syncobj-export")
 			test_syncobj_export(i915);
 
-- 
2.26.2

_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx