diff mbox series

[i-g-t,2/2] tests/gem_exec_reloc: Calculate softpin offsets from batch size

Message ID 20191023090752.23566-2-janusz.krzysztofik@linux.intel.com (mailing list archive)
State New, archived
Headers show
Series [i-g-t,1/2] tests/gem_exec_reloc: Don't filter out addresses when on PPGTT | expand

Commit Message

Janusz Krzysztofik Oct. 23, 2019, 9:07 a.m. UTC
From: Janusz Krzysztofik <janusz.krzysztofik@intel.com>

The basic-range subtest assumes 4kB minimum batch size.  On future
backends with possibly bigger minimum batch sizes this subtest will
fail as buffer objects may overlap.  To avoid object overlapping,
offsets need to be calculated with actual minimum batch size in mind.

Replace hardcoded constants corresponding to the assumed 4kB minimum
batch size with values calculated from a variable supposed to represent
actual batch size.  For now, the variable is still initialized with the
4kB value, suitable for backends currently supported by IGT.

Signed-off-by: Janusz Krzysztofik <janusz.krzysztofik@intel.com>
Cc: Katarzyna Dec <katarzyna.dec@intel.com>
Cc: Stuart Summers <stuart.summers@intel.com>
---
 tests/i915/gem_exec_reloc.c | 21 +++++++++++++--------
 1 file changed, 13 insertions(+), 8 deletions(-)

Comments

Chris Wilson Oct. 23, 2019, 9:13 a.m. UTC | #1
Quoting Janusz Krzysztofik (2019-10-23 10:07:52)
> From: Janusz Krzysztofik <janusz.krzysztofik@intel.com>
> 
> The basic-range subtest assumes 4kB minimum batch size.  On future
> backends with possibly bigger minimum batch sizes this subtest will
> fail as buffer objects may overlap.  To avoid object overlapping,
> offsets need to be calculated with actual minimum batch size in mind.
> 
> Replace hardcoded constants corresponding to the assumed 4kB minimum
> batch size with values calculated from a variable supposed to represent
> actual batch size.  For now, the variable is still initialized with the
> 4kB value, suitable for backends currently supported by IGT.

You could do this entirely within the reloc API, no? By probing whether
a 4KiB offset is legal?
-Chris
Janusz Krzysztofik Oct. 23, 2019, 9:16 a.m. UTC | #2
Hi Chris,

On Wednesday, October 23, 2019 11:13:02 AM CEST Chris Wilson wrote:
> Quoting Janusz Krzysztofik (2019-10-23 10:07:52)
> > From: Janusz Krzysztofik <janusz.krzysztofik@intel.com>
> > 
> > The basic-range subtest assumes 4kB minimum batch size.  On future
> > backends with possibly bigger minimum batch sizes this subtest will
> > fail as buffer objects may overlap.  To avoid object overlapping,
> > offsets need to be calculated with actual minimum batch size in mind.
> > 
> > Replace hardcoded constants corresponding to the assumed 4kB minimum
> > batch size with values calculated from a variable supposed to represent
> > actual batch size.  For now, the variable is still initialized with the
> > 4kB value, suitable for backends currently supported by IGT.
> 
> You could do this entirely within the reloc API, no? By probing whether
> a 4KiB offset is legal?

Thanks for the hint, I'll try to take this path.
Janusz

> -Chris
>
diff mbox series

Patch

diff --git a/tests/i915/gem_exec_reloc.c b/tests/i915/gem_exec_reloc.c
index 8f88826e..8d2f10ac 100644
--- a/tests/i915/gem_exec_reloc.c
+++ b/tests/i915/gem_exec_reloc.c
@@ -518,16 +518,21 @@  static void basic_range(int fd, unsigned flags)
 	struct drm_i915_gem_execbuffer2 execbuf;
 	uint64_t address_mask = has_64b_reloc(fd) ? ~(uint64_t)0 : ~(uint32_t)0;
 	uint64_t gtt_size = gem_aperture_size(fd);
+	/* FIXME: switch to a suitable library function when available */
+	uint64_t batch_size = 4096;
 	const uint32_t bbe = MI_BATCH_BUFFER_END;
 	igt_spin_t *spin = NULL;
-	int count, n;
+	int batch_order, count, n;
 
 	igt_require(gem_has_softpin(fd));
 
-	for (count = 12; gtt_size >> (count + 1); count++)
+	for (batch_order = 0; batch_size >> (batch_order + 1); batch_order++)
 		;
 
-	count -= 12;
+	for (count = batch_order; gtt_size >> (count + 1); count++)
+		;
+
+	count -= batch_order;
 
 	memset(obj, 0, sizeof(obj));
 	memset(reloc, 0, sizeof(reloc));
@@ -535,8 +540,8 @@  static void basic_range(int fd, unsigned flags)
 
 	n = 0;
 	for (int i = 0; i <= count; i++) {
-		obj[n].handle = gem_create(fd, 4096);
-		obj[n].offset = (1ull << (i + 12)) - 4096;
+		obj[n].handle = gem_create(fd, batch_size);
+		obj[n].offset = (1ull << (i + batch_order)) - batch_size;
 		obj[n].offset = gen8_canonical_address(obj[n].offset);
 		obj[n].flags = EXEC_OBJECT_PINNED | EXEC_OBJECT_SUPPORTS_48B_ADDRESS;
 		if (!gem_uses_ppgtt(fd)) {
@@ -556,8 +561,8 @@  static void basic_range(int fd, unsigned flags)
 		n++;
 	}
 	for (int i = 1; i < count; i++) {
-		obj[n].handle = gem_create(fd, 4096);
-		obj[n].offset = 1ull << (i + 12);
+		obj[n].handle = gem_create(fd, batch_size);
+		obj[n].offset = 1ull << (i + batch_order);
 		obj[n].offset = gen8_canonical_address(obj[n].offset);
 		obj[n].flags = EXEC_OBJECT_PINNED | EXEC_OBJECT_SUPPORTS_48B_ADDRESS;
 		if (!gem_uses_ppgtt(fd)) {
@@ -578,7 +583,7 @@  static void basic_range(int fd, unsigned flags)
 	}
 	igt_require(n);
 
-	obj[n].handle = gem_create(fd, 4096);
+	obj[n].handle = gem_create(fd, batch_size);
 	obj[n].relocs_ptr = to_user_pointer(reloc);
 	obj[n].relocation_count = n;
 	gem_write(fd, obj[n].handle, 0, &bbe, sizeof(bbe));