Skip to contentSkip to conversation

tests/gem_exec_parse: fix batch_len setting for cmd-crossing-page

PATCH In review [email protected] 2 participants 4 messages Oct 15, 2014 – Oct 22, 2014 Library code

lore
Share & tools
lore
#0 lore raw

PATCH 1/2 tests/gem_exec_parse: fix batch_len setting for cmd-crossing-page

From: Brad Volkin <[email protected]>

The size of the batch buffer passed to the kernel is significantly larger than the size of the batch buffer passed to the function. A proposed optimization as part of the batch copy kernel series is to use batch_len for the copy and parse operations, which leads to a false "batch without MI_BATCH_BUFFER_END" failure for this test.

---
 tests/gem_exec_parse.c | 5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)
tests/gem_exec_parse.c +3 -2
index 05f271c..568bd4a 100644
@@ -144,9 +144,10 @@ static void exec_split_batch(int fd, uint32_t *cmds,
 	struct drm_i915_gem_exec_object2 objs[1];
 	uint32_t cmd_bo;
 	uint32_t noop[1024] = { 0 };
+	const int alloc_size = 4096 * 2;
 
 	// Allocate and fill a 2-page batch with noops
-	cmd_bo = gem_create(fd, 4096 * 2);
+	cmd_bo = gem_create(fd, alloc_size);
 	gem_write(fd, cmd_bo, 0, noop, sizeof(noop));
 	gem_write(fd, cmd_bo, 4096, noop, sizeof(noop));
 
@@ -167,7 +168,7 @@ static void exec_split_batch(int fd, uint32_t *cmds,
 	execbuf.buffers_ptr = (uintptr_t)objs;
 	execbuf.buffer_count = 1;
 	execbuf.batch_start_offset = 0;
-	execbuf.batch_len = size;
+	execbuf.batch_len = alloc_size;
 	execbuf.cliprects_ptr = 0;
 	execbuf.num_cliprects = 0;
 	execbuf.DR1 = 0;
Signature · 1 line
1.9.1
#1 lore raw

PATCH 2/2 tests/gem_exec_parse: test for chained batch buffers

From: Brad Volkin <[email protected]>

libva makes extensive use of chained batch buffers. The batch buffer copy portion of the command parser has the potential to break chained batches, so add a simple test to make sure that doesn't happen.

---
 lib/intel_reg.h        |   1 +
 tests/gem_exec_parse.c | 105 +++++++++++++++++++++++++++++++++++++++++++++++++
 2 files changed, 106 insertions(+)
lib/intel_reg.h +1 -0
index f0fc5fd..fcc9d7c 100644
@@ -2571,6 +2571,7 @@ SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 #define MI_BATCH_BUFFER_END	(0xA << 23)
 #define MI_BATCH_NON_SECURE		(1)
 #define MI_BATCH_NON_SECURE_I965	(1 << 8)
+#define MI_BATCH_NON_SECURE_HSW		(1<<13) /* Additional bit for RCS */
 
 #define MAX_DISPLAY_PIPES	2
 
tests/gem_exec_parse.c +105 -0

117 lines not on this page · raw message

Signature · 1 line
1.9.1
#2 lore raw
In reply to [email protected]
Quoted 30 lines · On Wed, Oct 15, 2014 at 02:52:41PM -0700, [email protected] wrote:

From: Brad Volkin <[email protected]>

The size of the batch buffer passed to the kernel is significantly larger than the size of the batch buffer passed to the function. A proposed optimization as part of the batch copy kernel series is to use batch_len for the copy and parse operations, which leads to a false "batch without MI_BATCH_BUFFER_END" failure for this test.

---
 tests/gem_exec_parse.c | 5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)
tests/gem_exec_parse.c +2 -1
index 05f271c..568bd4a 100644
@@ -144,9 +144,10 @@ static void exec_split_batch(int fd, uint32_t *cmds,
 	struct drm_i915_gem_exec_object2 objs[1];
 	uint32_t cmd_bo;
 	uint32_t noop[1024] = { 0 };
+	const int alloc_size = 4096 * 2;
 
 	// Allocate and fill a 2-page batch with noops
-	cmd_bo = gem_create(fd, 4096 * 2);
+	cmd_bo = gem_create(fd, alloc_size);
 	gem_write(fd, cmd_bo, 0, noop, sizeof(noop));
 	gem_write(fd, cmd_bo, 4096, noop, sizeof(noop));
 

tests/gem_exec_parse.c static void exec_split_batch(int fd, uint32_t *cmds, · line 170

Quoted 3 lines
tests/gem_exec_parse.c +0 -0
@@ -167,7 +168,7 @@ static void exec_split_batch(int fd, uint32_t *cmds,
 	execbuf.buffers_ptr = (uintptr_t)objs;
 	execbuf.buffer_count = 1;
 	execbuf.batch_start_offset = 0;

Shouldn't we simply adjust batch_start_offset to be 4096-4, i.e. where we've placed the batch? Would have the benifit of also testing offset batches ;-)
-Daniel

Quoted 12 lines
tests/gem_exec_parse.c +1 -1
-	execbuf.batch_len = size;
+	execbuf.batch_len = alloc_size;
 	execbuf.cliprects_ptr = 0;
 	execbuf.num_cliprects = 0;
 	execbuf.DR1 = 0;
Signature · 6 lines
1.9.1

_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx
Signature · 3 lines
Daniel Vetter
Software Engineer, Intel Corporation
+41 (0) 79 365 57 48 - http://blog.ffwll.ch
#3 lore raw
Quoted 30 lines · Daniel Vetter
Quoted 29 lines · On Wed, Oct 15, 2014 at 02:52:41PM -0700, [email protected] wrote:

From: Brad Volkin <[email protected]>

The size of the batch buffer passed to the kernel is significantly larger than the size of the batch buffer passed to the function. A proposed optimization as part of the batch copy kernel series is to use batch_len for the copy and parse operations, which leads to a false "batch without MI_BATCH_BUFFER_END" failure for this test.

---
 tests/gem_exec_parse.c | 5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)
tests/gem_exec_parse.c +2 -1
index 05f271c..568bd4a 100644
@@ -144,9 +144,10 @@ static void exec_split_batch(int fd, uint32_t *cmds,
 	struct drm_i915_gem_exec_object2 objs[1];
 	uint32_t cmd_bo;
 	uint32_t noop[1024] = { 0 };
+	const int alloc_size = 4096 * 2;
 
 	// Allocate and fill a 2-page batch with noops
-	cmd_bo = gem_create(fd, 4096 * 2);
+	cmd_bo = gem_create(fd, alloc_size);
 	gem_write(fd, cmd_bo, 0, noop, sizeof(noop));
 	gem_write(fd, cmd_bo, 4096, noop, sizeof(noop));
 
Quoted 9 lines

tests/gem_exec_parse.c static void exec_split_batch(int fd, uint32_t *cmds, · line 170

Quoted 4 lines
tests/gem_exec_parse.c +0 -0
@@ -167,7 +168,7 @@ static void exec_split_batch(int fd, uint32_t *cmds,
 	execbuf.buffers_ptr = (uintptr_t)objs;
 	execbuf.buffer_count = 1;
 	execbuf.batch_start_offset = 0;

Shouldn't we simply adjust batch_start_offset to be 4096-4, i.e. where we've placed the batch? Would have the benifit of also testing offset batches ;-)
-Daniel

Good suggestion. I'll go with that approach instead.

Brad

Quoted 18 lines
Quoted 13 lines
tests/gem_exec_parse.c +1 -1
-	execbuf.batch_len = size;
+	execbuf.batch_len = alloc_size;
 	execbuf.cliprects_ptr = 0;
 	execbuf.num_cliprects = 0;
 	execbuf.DR1 = 0;
Signature · 6 lines
1.9.1

_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx
Signature · 3 lines
Daniel Vetter
Software Engineer, Intel Corporation
+41 (0) 79 365 57 48 - http://blog.ffwll.ch