Skip to contentSkip to conversation

test/gem_mocs_settings: Testing MOCS register settings

PATCH Superseded Peter Antoine 2 participants 5 messages Apr 11, 2016 Library code

v3 of 4 Superseded Latest is v4 →

lore
Share & tools
lore
#0 lore raw

PATCH v3 1/1 test/gem_mocs_settings: Testing MOCS register settings

The MOCS registers were added in Gen9 and define the caching policy. The registers are split into two sets. The first set controls the EDRAM policy and have a set for each engine, the second set controls the L3 policy. The two sets use the same index.

The RCS registers and the L3CC registers are stored in the RCS context.

The test checks that the registers are correct by checking the values by directly reading them via MMIO, then again it tests them by reading them from within a batch buffer. RCS engine is tested last as it programs the registers via a batch buffer and this will invalidate the test for workloads that don't use the render ring or don't run a render batch first.

v2: Reorganised the structure.
Added more tests. (Chris Wilson)
v3: Fixed a few bugs. (Chris Wilson)

---
 lib/ioctl_wrappers.c      |  13 ++
 lib/ioctl_wrappers.h      |   1 +
 tests/Makefile.sources    |   1 +
 tests/gem_mocs_settings.c | 559 ++++++++++++++++++++++++++++++++++++++++++++++
 4 files changed, 574 insertions(+)
 create mode 100644 tests/gem_mocs_settings.c
lib/ioctl_wrappers.c +13 -0
index 076bce8..8d74128 100644
@@ -1382,6 +1382,19 @@ void gem_require_ring(int fd, unsigned ring)
 	igt_require(gem_has_ring(fd, ring));
 }
 
+/**
+ * gem_has_mocs:
+ * @fd: open i915 drm file descriptor
+ *
+ * Feature test macro to query whether the device has MOCS registers.
+ * These exist gen 9+.
+ */
+void gem_has_mocs(int fd)
+{
+	const int gen = intel_gen(intel_get_drm_devid(fd));
+	igt_require(gen >= 9);
+}
+
 /* prime */
 
 /**
lib/ioctl_wrappers.h +1 -0
index d986f61..ebedf1e 100644
@@ -154,6 +154,7 @@ bool gem_has_softpin(int fd);
 void gem_require_caching(int fd);
 bool gem_has_ring(int fd, unsigned ring);
 void gem_require_ring(int fd, unsigned ring);
+void gem_has_mocs(int fd);
 
 /* prime */
 struct local_dma_buf_sync {
tests/Makefile.sources +1 -0
index 43f232f..d483c9e 100644
@@ -148,6 +148,7 @@ TESTS_progs = \
 	gem_lut_handle \
 	gem_mmap_offset_exhaustion \
 	gem_media_fill \
+	gem_mocs_settings \
 	gem_gpgpu_fill \
 	gem_pin \
 	gem_reg_read \
tests/gem_mocs_settings.c new file +559 -0

559 lines not on this page · raw message

Signature · 6 lines
1.9.1

_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
#1 lore raw
Quoted 33 lines · Peter Antoine

The MOCS registers were added in Gen9 and define the caching policy. The registers are split into two sets. The first set controls the EDRAM policy and have a set for each engine, the second set controls the L3 policy. The two sets use the same index.

The RCS registers and the L3CC registers are stored in the RCS context.

The test checks that the registers are correct by checking the values by directly reading them via MMIO, then again it tests them by reading them from within a batch buffer. RCS engine is tested last as it programs the registers via a batch buffer and this will invalidate the test for workloads that don't use the render ring or don't run a render batch first.

v2: Reorganised the structure.
Added more tests. (Chris Wilson)
v3: Fixed a few bugs. (Chris Wilson)

---
 lib/ioctl_wrappers.c      |  13 ++
 lib/ioctl_wrappers.h      |   1 +
 tests/Makefile.sources    |   1 +
 tests/gem_mocs_settings.c | 559 ++++++++++++++++++++++++++++++++++++++++++++++
 4 files changed, 574 insertions(+)
 create mode 100644 tests/gem_mocs_settings.c
lib/ioctl_wrappers.c +0 -0
index 076bce8..8d74128 100644
@@ -1382,6 +1382,19 @@ void gem_require_ring(int fd, unsigned ring)
 	igt_require(gem_has_ring(fd, ring));

lib/ioctl_wrappers.c void gem_require_ring(int fd, unsigned ring) · line 1392

Quoted 10 lines
lib/ioctl_wrappers.c +8 -0
 }
 
+/**
+ * gem_has_mocs:
+ * @fd: open i915 drm file descriptor
+ *
+ * Feature test macro to query whether the device has MOCS registers.
+ * These exist gen 9+.
+ */
+void gem_has_mocs(int fd)

gem_has_mocs() returns bool
gem_require_mocs() calls igt_require() and doesn't return if not available.

Quoted 93 lines
lib/ioctl_wrappers.c +5 -0
+{
+	const int gen = intel_gen(intel_get_drm_devid(fd));
+	igt_require(gen >= 9);
+}
+
 /* prime */
 
 /**
lib/ioctl_wrappers.h +1 -0
index d986f61..ebedf1e 100644
@@ -154,6 +154,7 @@ bool gem_has_softpin(int fd);
 void gem_require_caching(int fd);
 bool gem_has_ring(int fd, unsigned ring);
 void gem_require_ring(int fd, unsigned ring);
+void gem_has_mocs(int fd);
 
 /* prime */
 struct local_dma_buf_sync {
tests/Makefile.sources +1 -0
index 43f232f..d483c9e 100644
@@ -148,6 +148,7 @@ TESTS_progs = \
 	gem_lut_handle \
 	gem_mmap_offset_exhaustion \
 	gem_media_fill \
+	gem_mocs_settings \
 	gem_gpgpu_fill \
 	gem_pin \
 	gem_reg_read \
tests/gem_mocs_settings.c new file +55 -0
new file mode 100644
index 0000000..03c5089
@@ -0,0 +1,559 @@
+/*
+ * Copyright © 2016 Intel Corporation
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the next
+ * paragraph) shall be included in all copies or substantial portions of the
+ * Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+ * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
+ * IN THE SOFTWARE.
+ *
+ */
+
+/** @file gem_mocs_settings.c
+ *
+ * Check that the MOCs cache settings are valid.
+ */
+
+#include "igt.h"
+#include "igt_gt.h"
+
+#define MAX_NUMBER_MOCS_REGISTERS	(64)
+
+enum {
+	NONE,
+	RESET,
+	SUSPEND,
+	HIBERNATE
+};
+
+#define GEN9_LNCFCMOCS0		(0xB020)	/* L3 Cache Control base */
+#define GEN9_GFX_MOCS_0		(0xc800)	/* Graphics MOCS base register*/
+#define GEN9_MFX0_MOCS_0	(0xc900)	/* Media 0 MOCS base register*/
+#define GEN9_MFX1_MOCS_0	(0xcA00)	/* Media 1 MOCS base register*/
+#define GEN9_VEBOX_MOCS_0	(0xcB00)	/* Video MOCS base register*/
+#define GEN9_BLT_MOCS_0		(0xcc00)	/* Blitter MOCS base register*/
+
+#define ENGINE_GFX	(I915_EXEC_RENDER)
+#define ENGINE_MFX0	(I915_EXEC_BSD)
+#define ENGINE_MFX0_ALT	(I915_EXEC_BSD | 1 << 13)
+#define ENGINE_MFX1	(I915_EXEC_BSD | 2 << 13)
+#define ENGINE_BLT	(I915_EXEC_BLT)
+#define ENGINE_VEBOX	(I915_EXEC_VEBOX)
+

tests/gem_mocs_settings.c line 65

Quoted 10 lines
tests/gem_mocs_settings.c new file +10 -0
+struct mocs_entry {
+	uint32_t	control_value;
+	uint16_t	l3cc_value;
+};
+
+struct mocs_table {
+	uint32_t		size;
+	const struct mocs_entry	*table;
+};
+

/* The first entries in the MOCS tables are defined by uABI */

Quoted 22 lines
tests/gem_mocs_settings.c new file +22 -0
+static const struct mocs_entry skylake_mocs_table[] = {
+	{ 0x00000009, 0x0010 },
+	{ 0x00000038, 0x0030 },
+	{ 0x0000003b, 0x0030 },
+};
+
+static const struct mocs_entry broxton_mocs_table[] = {
+	{ 0x00000009, 0x0010 },
+	{ 0x00000038, 0x0030 },
+	{ 0x0000003b, 0x0030 },
+};
+
+static const struct mocs_entry dirty_table_mocs_table[] = {
+	{ 0x00007FFF, 0x003F },
+	{ 0x00007FFF, 0x003F },
+	{ 0x00007FFF, 0x003F },
+};
+
+static const uint32_t write_values[] = {
+	0xFFFFFFFF,
+	0xFFFFFFFF,
+	0xFFFFFFFF,

tests/gem_mocs_settings.c

Quoted 10 lines
tests/gem_mocs_settings.c new file +9 -0
+	0xFFFFFFFF
+};
 
+static void test_mocs_values(int fd)
+{
+	const struct intel_execution_engine *e;
+
+	for (e = intel_execution_engines; e->name; e++)
+		if (e->exec_id != I915_EXEC_DEFAULT &&
+		    gem_has_ring(fd, e->exec_id | e->flags))

You could use

unsigned engine;

for_each_engine(fd, engine)

	if (engine != 0) test_mocs_control_values(fd, engine);

Just to hide the gem_has_ring().

Quoted 17 lines
tests/gem_mocs_settings.c new file +17 -0
+		
+
+	test_mocs_l3cc_values(fd);
+}
+
+#define MI_STORE_REGISTER_MEM_64_BIT_ADDR	((0x24 << 23) | 2)
+
+static int create_read_batch(struct drm_i915_gem_relocation_entry *reloc,
+			     uint32_t *batch,
+			     uint32_t dst_handle,
+			     uint32_t size,
+			     uint32_t reg_base)
+{
+	unsigned int index;
+	unsigned int offset = 0;
+
+	for (index = 0, offset = 0; index < size; index++, offset += 4)

tests/gem_mocs_settings.c line 211

Quoted 10 lines
tests/gem_mocs_settings.c new file +10 -0
+	{
+		batch[offset]   = MI_STORE_REGISTER_MEM_64_BIT_ADDR;
+		batch[offset+1] = reg_base + (index * sizeof(uint32_t));
+		batch[offset+2] = index * sizeof(uint32_t);	/* reloc */
+		batch[offset+3] = 0;
+
+		reloc[index].offset = (offset + 2) * sizeof(uint32_t);
+		reloc[index].delta = index * sizeof(uint32_t);
+		reloc[index].target_handle = dst_handle;
+		reloc[index].write_domain = I915_GEM_DOMAIN_RENDER;

Will cause an -EINVAL as you are specifing a write_domain but not the corresponding read_domain.

tests/gem_mocs_settings.c line 261

Quoted 9 lines
tests/gem_mocs_settings.c new file +9 -0
+static void do_read_registers(int fd,
+			      uint32_t ctx_id,
+			      uint32_t dst_handle,
+			      uint32_t reg_base,
+			      uint32_t size,
+			      uint32_t engine_id)
+{
+	gem_execbuf(fd, &execbuf);
+	gem_wait(fd, dst_handle, &timeout_ns);

As mentioned earlier, tried to avoid manual wait/sync so that we stress the kernel harder to do the right thing. Sometimes the sync is required as part of the test, but only rarely and should be well documented.

Quoted 13 lines
tests/gem_mocs_settings.c new file +12 -0
+	gem_close(fd, handle);
+}
 
+static void check_control_registers(int fd,
+				    const struct intel_execution_engine *engine,
+				    uint32_t ctx_id,
+				    bool dirty)
+{
+	uint32_t dst_handle = gem_create(fd, 4096);
+	uint32_t reg_base  = get_engine_base(engine->exec_id | engine->flags);
+	uint32_t *read_regs;
+	struct mocs_table table;
+

tests/gem_mocs_settings.c line 340

Quoted 10 lines
tests/gem_mocs_settings.c new file +10 -0
+	igt_assert(get_mocs_settings(intel_get_drm_devid(fd), &table, dirty));
+
+	do_read_registers(fd,
+			  ctx_id,
+			  dst_handle,
+			  reg_base,
+			  table.size,
+			  engine->exec_id | engine->flags);
+
+	read_regs = gem_mmap__cpu(fd, dst_handle, 0, 4096, PROT_READ);

Here you need the corresponding
gem_set_comdin(fd, dst_handle, DOMAIN_CPU, 0);
and in check_l3cc_registers.

tests/gem_mocs_settings.c line 348

Quoted 8 lines
tests/gem_mocs_settings.c new file +8 -0
+
+	for (int index = 0; index < table.size; index++)
+		igt_assert_eq(read_regs[index],
+				table.table[index].control_value);
+
+	munmap(read_regs, 4096);
+	gem_close(fd, dst_handle);
+}

Otherwise, looking good. (Could use snooping or gem_read to minimise the clflush on bxt, but I don't think anyone will notice over 4k). -Chris

Signature · 5 lines
Chris Wilson, Intel Open Source Technology Centre
_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
#2 lore raw
Quoted 36 lines · Chris Wilson
Quoted 35 lines · Peter Antoine

The MOCS registers were added in Gen9 and define the caching policy. The registers are split into two sets. The first set controls the EDRAM policy and have a set for each engine, the second set controls the L3 policy. The two sets use the same index.

The RCS registers and the L3CC registers are stored in the RCS context.

The test checks that the registers are correct by checking the values by directly reading them via MMIO, then again it tests them by reading them from within a batch buffer. RCS engine is tested last as it programs the registers via a batch buffer and this will invalidate the test for workloads that don't use the render ring or don't run a render batch first.

v2: Reorganised the structure.
Added more tests. (Chris Wilson)
v3: Fixed a few bugs. (Chris Wilson)

---
 lib/ioctl_wrappers.c      |  13 ++
 lib/ioctl_wrappers.h      |   1 +
 tests/Makefile.sources    |   1 +
 tests/gem_mocs_settings.c | 559 ++++++++++++++++++++++++++++++++++++++++++++++
 4 files changed, 574 insertions(+)
 create mode 100644 tests/gem_mocs_settings.c
lib/ioctl_wrappers.c +0 -0
index 076bce8..8d74128 100644
@@ -1382,6 +1382,19 @@ void gem_require_ring(int fd, unsigned ring)
 	igt_require(gem_has_ring(fd, ring));
 }
 
Quoted 13 lines

lib/ioctl_wrappers.c void gem_require_ring(int fd, unsigned ring) · line 1392

Quoted 8 lines
lib/ioctl_wrappers.c +8 -0
+/**
+ * gem_has_mocs:
+ * @fd: open i915 drm file descriptor
+ *
+ * Feature test macro to query whether the device has MOCS registers.
+ * These exist gen 9+.
+ */
+void gem_has_mocs(int fd)

gem_has_mocs() returns bool
gem_require_mocs() calls igt_require() and doesn't return if not available.

Done.

Quoted 97 lines
Quoted 97 lines
lib/ioctl_wrappers.c +5 -0
+{
+	const int gen = intel_gen(intel_get_drm_devid(fd));
+	igt_require(gen >= 9);
+}
+
 /* prime */
 
 /**
lib/ioctl_wrappers.h +1 -0
index d986f61..ebedf1e 100644
@@ -154,6 +154,7 @@ bool gem_has_softpin(int fd);
 void gem_require_caching(int fd);
 bool gem_has_ring(int fd, unsigned ring);
 void gem_require_ring(int fd, unsigned ring);
+void gem_has_mocs(int fd);
 
 /* prime */
 struct local_dma_buf_sync {
tests/Makefile.sources +1 -0
index 43f232f..d483c9e 100644
@@ -148,6 +148,7 @@ TESTS_progs = \
 	gem_lut_handle \
 	gem_mmap_offset_exhaustion \
 	gem_media_fill \
+	gem_mocs_settings \
 	gem_gpgpu_fill \
 	gem_pin \
 	gem_reg_read \
tests/gem_mocs_settings.c new file +59 -0
new file mode 100644
index 0000000..03c5089
@@ -0,0 +1,559 @@
+/*
+ * Copyright © 2016 Intel Corporation
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the next
+ * paragraph) shall be included in all copies or substantial portions of the
+ * Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+ * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
+ * IN THE SOFTWARE.
+ *
+ */
+
+/** @file gem_mocs_settings.c
+ *
+ * Check that the MOCs cache settings are valid.
+ */
+
+#include "igt.h"
+#include "igt_gt.h"
+
+#define MAX_NUMBER_MOCS_REGISTERS	(64)
+
+enum {
+	NONE,
+	RESET,
+	SUSPEND,
+	HIBERNATE
+};
+
+#define GEN9_LNCFCMOCS0		(0xB020)	/* L3 Cache Control base */
+#define GEN9_GFX_MOCS_0		(0xc800)	/* Graphics MOCS base register*/
+#define GEN9_MFX0_MOCS_0	(0xc900)	/* Media 0 MOCS base register*/
+#define GEN9_MFX1_MOCS_0	(0xcA00)	/* Media 1 MOCS base register*/
+#define GEN9_VEBOX_MOCS_0	(0xcB00)	/* Video MOCS base register*/
+#define GEN9_BLT_MOCS_0		(0xcc00)	/* Blitter MOCS base register*/
+
+#define ENGINE_GFX	(I915_EXEC_RENDER)
+#define ENGINE_MFX0	(I915_EXEC_BSD)
+#define ENGINE_MFX0_ALT	(I915_EXEC_BSD | 1 << 13)
+#define ENGINE_MFX1	(I915_EXEC_BSD | 2 << 13)
+#define ENGINE_BLT	(I915_EXEC_BLT)
+#define ENGINE_VEBOX	(I915_EXEC_VEBOX)
+
+struct mocs_entry {
+	uint32_t	control_value;
+	uint16_t	l3cc_value;
+};
Quoted 9 lines

tests/gem_mocs_settings.c line 65

Quoted 6 lines
tests/gem_mocs_settings.c new file +6 -0
+
+struct mocs_table {
+	uint32_t		size;
+	const struct mocs_entry	*table;
+};
+

/* The first entries in the MOCS tables are defined by uABI */

Added.

Quoted 35 lines
Quoted 22 lines
tests/gem_mocs_settings.c new file +22 -0
+static const struct mocs_entry skylake_mocs_table[] = {
+	{ 0x00000009, 0x0010 },
+	{ 0x00000038, 0x0030 },
+	{ 0x0000003b, 0x0030 },
+};
+
+static const struct mocs_entry broxton_mocs_table[] = {
+	{ 0x00000009, 0x0010 },
+	{ 0x00000038, 0x0030 },
+	{ 0x0000003b, 0x0030 },
+};
+
+static const struct mocs_entry dirty_table_mocs_table[] = {
+	{ 0x00007FFF, 0x003F },
+	{ 0x00007FFF, 0x003F },
+	{ 0x00007FFF, 0x003F },
+};
+
+static const uint32_t write_values[] = {
+	0xFFFFFFFF,
+	0xFFFFFFFF,
+	0xFFFFFFFF,

tests/gem_mocs_settings.c

Quoted 10 lines
tests/gem_mocs_settings.c new file +9 -0
+	0xFFFFFFFF
+};
 
+static void test_mocs_values(int fd)
+{
+	const struct intel_execution_engine *e;
+
+	for (e = intel_execution_engines; e->name; e++)
+		if (e->exec_id != I915_EXEC_DEFAULT &&
+		    gem_has_ring(fd, e->exec_id | e->flags))

You could use

Quoted 6 lines

unsigned engine;

for_each_engine(fd, engine)

	if (engine != 0) test_mocs_control_values(fd, engine);

Just to hide the gem_has_ring().

Done.

Quoted 23 lines
Quoted 23 lines
tests/gem_mocs_settings.c new file +22 -0
+
+
+	test_mocs_l3cc_values(fd);
+}
+
+#define MI_STORE_REGISTER_MEM_64_BIT_ADDR	((0x24 << 23) | 2)
+
+static int create_read_batch(struct drm_i915_gem_relocation_entry *reloc,
+			     uint32_t *batch,
+			     uint32_t dst_handle,
+			     uint32_t size,
+			     uint32_t reg_base)
+{
+	unsigned int index;
+	unsigned int offset = 0;
+
+	for (index = 0, offset = 0; index < size; index++, offset += 4)
+	{
+		batch[offset]   = MI_STORE_REGISTER_MEM_64_BIT_ADDR;
+		batch[offset+1] = reg_base + (index * sizeof(uint32_t));
+		batch[offset+2] = index * sizeof(uint32_t);	/* reloc */
+		batch[offset+3] = 0;
Quoted 8 lines

tests/gem_mocs_settings.c line 211

Quoted 5 lines
tests/gem_mocs_settings.c new file +5 -0
+
+		reloc[index].offset = (offset + 2) * sizeof(uint32_t);
+		reloc[index].delta = index * sizeof(uint32_t);
+		reloc[index].target_handle = dst_handle;
+		reloc[index].write_domain = I915_GEM_DOMAIN_RENDER;

Will cause an -EINVAL as you are specifing a write_domain but not the corresponding read_domain.

Removed.
(Did not cause an error when tested)

Quoted 14 lines

tests/gem_mocs_settings.c line 261

Quoted 10 lines
tests/gem_mocs_settings.c new file +9 -0
+static void do_read_registers(int fd,
+			      uint32_t ctx_id,
+			      uint32_t dst_handle,
+			      uint32_t reg_base,
+			      uint32_t size,
+			      uint32_t engine_id)
+{
+	gem_execbuf(fd, &execbuf);
+	gem_wait(fd, dst_handle, &timeout_ns);

As mentioned earlier, tried to avoid manual wait/sync so that we stress the kernel harder to do the right thing. Sometimes the sync is required as part of the test, but only rarely and should be well documented.

replaced with set_domain.

Quoted 20 lines
Quoted 20 lines
tests/gem_mocs_settings.c new file +18 -0
+	gem_close(fd, handle);
+}
 
+static void check_control_registers(int fd,
+				    const struct intel_execution_engine *engine,
+				    uint32_t ctx_id,
+				    bool dirty)
+{
+	uint32_t dst_handle = gem_create(fd, 4096);
+	uint32_t reg_base  = get_engine_base(engine->exec_id | engine->flags);
+	uint32_t *read_regs;
+	struct mocs_table table;
+
+	igt_assert(get_mocs_settings(intel_get_drm_devid(fd), &table, dirty));
+
+	do_read_registers(fd,
+			  ctx_id,
+			  dst_handle,
+			  reg_base,
Quoted 8 lines

tests/gem_mocs_settings.c line 340

Quoted 4 lines
tests/gem_mocs_settings.c new file +4 -0
+			  table.size,
+			  engine->exec_id | engine->flags);
+
+	read_regs = gem_mmap__cpu(fd, dst_handle, 0, 4096, PROT_READ);

Here you need the corresponding
gem_set_comdin(fd, dst_handle, DOMAIN_CPU, 0);
and in check_l3cc_registers.

Added into do_read_registers() so should handle both.

Quoted 15 lines

tests/gem_mocs_settings.c line 348

Quoted 9 lines
tests/gem_mocs_settings.c new file +8 -0
+
+	for (int index = 0; index < table.size; index++)
+		igt_assert_eq(read_regs[index],
+				table.table[index].control_value);
+
+	munmap(read_regs, 4096);
+	gem_close(fd, dst_handle);
+}

Otherwise, looking good. (Could use snooping or gem_read to minimise the clflush on bxt, but I don't think anyone will notice over 4k). -Chris

Ok.

Will run it on something not a BXT and then push again. So the other patched are good as well?

Thanks for the review.
Peter.

--
Peter Antoine (Android Graphics Driver Software Engineer)
---------------------------------------------------------------------
Intel Corporation (UK) Limited
Registered No. 1134945 (England)
Registered Office: Pipers Way, Swindon SN3 1RJ VAT No: 860 2173 47

_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

#3 lore raw
Quoted 14 lines · Peter Antoine
Quoted 6 lines
Quoted 6 lines
tests/gem_mocs_settings.c new file +6 -0
+	for (index = 0, offset = 0; index < size; index++, offset += 4)
+	{
+		batch[offset]   = MI_STORE_REGISTER_MEM_64_BIT_ADDR;
+		batch[offset+1] = reg_base + (index * sizeof(uint32_t));
+		batch[offset+2] = index * sizeof(uint32_t);	/* reloc */
+		batch[offset+3] = 0;
Quoted 8 lines

tests/gem_mocs_settings.c line 211

Quoted 5 lines
tests/gem_mocs_settings.c new file +5 -0
+
+		reloc[index].offset = (offset + 2) * sizeof(uint32_t);
+		reloc[index].delta = index * sizeof(uint32_t);
+		reloc[index].target_handle = dst_handle;
+		reloc[index].write_domain = I915_GEM_DOMAIN_RENDER;

Will cause an -EINVAL as you are specifing a write_domain but not the corresponding read_domain.

Quoted 1 line

Removed.

I hope you meant added the missing read_domains!

Quoted 1 line

(Did not cause an error when tested)

Hmm, ok we don't actually warn about that!

Quoted 43 lines
Quoted 5 lines
Quoted 5 lines
tests/gem_mocs_settings.c new file +5 -0
+static void do_read_registers(int fd,
+			      uint32_t ctx_id,
+			      uint32_t dst_handle,
+			      uint32_t reg_base,
+			      uint32_t size,
Quoted 8 lines

tests/gem_mocs_settings.c line 261

Quoted 4 lines
tests/gem_mocs_settings.c new file +4 -0
+			      uint32_t engine_id)
+{
+	gem_execbuf(fd, &execbuf);
+	gem_wait(fd, dst_handle, &timeout_ns);

As mentioned earlier, tried to avoid manual wait/sync so that we stress the kernel harder to do the right thing. Sometimes the sync is required as part of the test, but only rarely and should be well documented.

replaced with set_domain.

Quoted 20 lines
Quoted 20 lines
tests/gem_mocs_settings.c new file +18 -0
+	gem_close(fd, handle);
+}
 
+static void check_control_registers(int fd,
+				    const struct intel_execution_engine *engine,
+				    uint32_t ctx_id,
+				    bool dirty)
+{
+	uint32_t dst_handle = gem_create(fd, 4096);
+	uint32_t reg_base  = get_engine_base(engine->exec_id | engine->flags);
+	uint32_t *read_regs;
+	struct mocs_table table;
+
+	igt_assert(get_mocs_settings(intel_get_drm_devid(fd), &table, dirty));
+
+	do_read_registers(fd,
+			  ctx_id,
+			  dst_handle,
+			  reg_base,
Quoted 8 lines

tests/gem_mocs_settings.c line 340

Quoted 4 lines
tests/gem_mocs_settings.c new file +4 -0
+			  table.size,
+			  engine->exec_id | engine->flags);
+
+	read_regs = gem_mmap__cpu(fd, dst_handle, 0, 4096, PROT_READ);

Here you need the corresponding
gem_set_comdin(fd, dst_handle, DOMAIN_CPU, 0);
and in check_l3cc_registers.

Quoted 1 line

Added into do_read_registers() so should handle both.

But less explicit and doesn't document the code as well. Move the set-domain next to the usage of dst_handle in the CPU domain. -Chris

Signature · 5 lines
Chris Wilson, Intel Open Source Technology Centre
_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
#4 lore raw
Quoted 15 lines · Chris Wilson
Quoted 14 lines · Peter Antoine
Quoted 14 lines

tests/gem_mocs_settings.c line 211

Quoted 11 lines
tests/gem_mocs_settings.c new file +11 -0
+	for (index = 0, offset = 0; index < size; index++, offset += 4)
+	{
+		batch[offset]   = MI_STORE_REGISTER_MEM_64_BIT_ADDR;
+		batch[offset+1] = reg_base + (index * sizeof(uint32_t));
+		batch[offset+2] = index * sizeof(uint32_t);	/* reloc */
+		batch[offset+3] = 0;
+
+		reloc[index].offset = (offset + 2) * sizeof(uint32_t);
+		reloc[index].delta = index * sizeof(uint32_t);
+		reloc[index].target_handle = dst_handle;
+		reloc[index].write_domain = I915_GEM_DOMAIN_RENDER;

Will cause an -EINVAL as you are specifing a write_domain but not the corresponding read_domain.

Quoted 8 lines
Quoted 1 line

Removed.

I hope you meant added the missing read_domains!

Quoted 1 line

(Did not cause an error when tested)

Hmm, ok we don't actually warn about that!

BXT no error, SKL error (different versions of the kernel). Added missing read_domains.

Quoted 43 lines
Quoted 43 lines
Quoted 5 lines
Quoted 5 lines
tests/gem_mocs_settings.c new file +5 -0
+static void do_read_registers(int fd,
+			      uint32_t ctx_id,
+			      uint32_t dst_handle,
+			      uint32_t reg_base,
+			      uint32_t size,
Quoted 8 lines

tests/gem_mocs_settings.c line 261

Quoted 4 lines
tests/gem_mocs_settings.c new file +4 -0
+			      uint32_t engine_id)
+{
+	gem_execbuf(fd, &execbuf);
+	gem_wait(fd, dst_handle, &timeout_ns);

As mentioned earlier, tried to avoid manual wait/sync so that we stress the kernel harder to do the right thing. Sometimes the sync is required as part of the test, but only rarely and should be well documented.

replaced with set_domain.

Quoted 28 lines
Quoted 14 lines
tests/gem_mocs_settings.c new file +12 -0
+	gem_close(fd, handle);
+}
 
+static void check_control_registers(int fd,
+				    const struct intel_execution_engine *engine,
+				    uint32_t ctx_id,
+				    bool dirty)
+{
+	uint32_t dst_handle = gem_create(fd, 4096);
+	uint32_t reg_base  = get_engine_base(engine->exec_id | engine->flags);
+	uint32_t *read_regs;
+	struct mocs_table table;
+

tests/gem_mocs_settings.c line 340

Quoted 10 lines
tests/gem_mocs_settings.c new file +10 -0
+	igt_assert(get_mocs_settings(intel_get_drm_devid(fd), &table, dirty));
+
+	do_read_registers(fd,
+			  ctx_id,
+			  dst_handle,
+			  reg_base,
+			  table.size,
+			  engine->exec_id | engine->flags);
+
+	read_regs = gem_mmap__cpu(fd, dst_handle, 0, 4096, PROT_READ);

Here you need the corresponding
gem_set_comdin(fd, dst_handle, DOMAIN_CPU, 0);
and in check_l3cc_registers.

Quoted 7 lines
Quoted 1 line

Added into do_read_registers() so should handle both.

But less explicit and doesn't document the code as well. Move the set-domain next to the usage of dst_handle in the CPU domain. -Chris

Moved out of do_ream_domain()

Found an issue on SKL, it does not have the snoop bit that the BXT has, changed test so that the dirty check takes account of the different settable bits.

Version 4, coming now.

Peter.

--
Peter Antoine (Android Graphics Driver Software Engineer)
---------------------------------------------------------------------
Intel Corporation (UK) Limited
Registered No. 1134945 (England)
Registered Office: Pipers Way, Swindon SN3 1RJ VAT No: 860 2173 47
_______________________________________________
Intel-gfx mailing list [email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx