Skip to contentSkip to conversation

drm/i915: Switch to fake context on older gens

PATCH Applied Mika Kuoppala 3 participants 8 messages Mar 14, 2014 – Mar 24, 2014

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH 1/2 drm/i915: Switch to fake context on older gens

We used to have per file descriptor hang stats for the i915_get_reset_stats_ioctl() and for default context banning.

commit 0eea67eb26000657079b7fc41079097942339452

Author: Ben Widawsky <[email protected]>
Date:   Fri Dec 6 14:11:19 2013 -0800

drm/i915: Create a per file_priv default context

made having separate hangstats in file_private redundant as i915_hw_context already contained hangstats. So

commit c482972a086e03e6a6d27e4f7af2d868bf659648

Author: Ben Widawsky <[email protected]>
Date:   Fri Dec 6 14:11:20 2013 -0800

drm/i915: Piggy back hangstats off of contexts

consolidated the hangstats and enabled further improvements.

commit 44e2c0705a19e09d7b0f30a591f92e473e5ef89e

Author: Mika Kuoppala <[email protected]>
Date:   Thu Jan 30 16:01:15 2014 +0200

drm/i915: Use i915_hw_context to set reset stats

tried to reap full benefits of consolidation but fell short as we never 'switch' to the fake private context on gens that don't have hw_contexts, so request->ctx remained NULL on those.

Fix this by 'switching' to fake context so that when request is submitted to ring, proper context gets assigned to it.

References: https://bugs.freedesktop.org/show_bug.cgi?id=76055

---
 drivers/gpu/drm/i915/i915_gem_context.c |    6 ++++--
 1 file changed, 4 insertions(+), 2 deletions(-)
drivers/gpu/drm/i915/i915_gem_context.c +4 -2
index ce41cff..b5a5837 100644
@@ -775,9 +775,11 @@ int i915_switch_context(struct intel_ring_buffer *ring,
 
 	BUG_ON(file && to == NULL);
 
-	/* We have the fake context, but don't supports switching. */
-	if (!HAS_HW_CONTEXTS(ring->dev))
+	/* We have the fake context */
+	if (!HAS_HW_CONTEXTS(ring->dev)) {
+		ring->last_context = to;
 		return 0;
+	}
 
 	return do_switch(ring, to);
 }
Signature · 1 line
1.7.9.5
#1 lore raw

PATCH 2/2 drm/i915: Return -ENOENT for unknown contexts

If hw_contexts are disabled, we always return the per file descriptor default context stats. Make sure that the context is correctly given and fail accordingly if not.

---
 drivers/gpu/drm/i915/i915_gem_context.c |    6 +++++-
 1 file changed, 5 insertions(+), 1 deletion(-)
drivers/gpu/drm/i915/i915_gem_context.c +5 -1
index b5a5837..a8e625d 100644
@@ -573,8 +573,12 @@ i915_gem_context_get(struct drm_i915_file_private *file_priv, u32 id)
 {
 	struct i915_hw_context *ctx;
 
-	if (!HAS_HW_CONTEXTS(file_priv->dev_priv->dev))
+	if (!HAS_HW_CONTEXTS(file_priv->dev_priv->dev)) {
+		if (id != DEFAULT_CONTEXT_ID)
+			return ERR_PTR(-ENOENT);
+
 		return file_priv->private_default_ctx;
+	}
 
 	ctx = (struct i915_hw_context *)idr_find(&file_priv->context_idr, id);
 	if (!ctx)
Signature · 1 line
1.7.9.5
#2 lore raw

Re: [PATCH 2/2] drm/i915: Return -ENOENT for unknown contexts

Quoted 5 lines · Mika Kuoppala

If hw_contexts are disabled, we always return the per file descriptor default context stats. Make sure that the context is correctly given and fail accordingly if not.

Hm, does this fix a testcase somewhere? If sou can you specify which one? Of there's no such thing I guess we should try to add a new testcase for this for pre-gen6 ...
-Daniel

Quoted 29 lines
---
 drivers/gpu/drm/i915/i915_gem_context.c |    6 +++++-
 1 file changed, 5 insertions(+), 1 deletion(-)
drivers/gpu/drm/i915/i915_gem_context.c +5 -1
index b5a5837..a8e625d 100644
@@ -573,8 +573,12 @@ i915_gem_context_get(struct drm_i915_file_private *file_priv, u32 id)
 {
 	struct i915_hw_context *ctx;
 
-	if (!HAS_HW_CONTEXTS(file_priv->dev_priv->dev))
+	if (!HAS_HW_CONTEXTS(file_priv->dev_priv->dev)) {
+		if (id != DEFAULT_CONTEXT_ID)
+			return ERR_PTR(-ENOENT);
+
 		return file_priv->private_default_ctx;
+	}
 
 	ctx = (struct i915_hw_context *)idr_find(&file_priv->context_idr, id);
 	if (!ctx)
Signature · 6 lines
1.7.9.5

_______________________________________________
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

PATCH 1/1 tests/gem_reset_stats: run non hw context tests also on older gens

To gain more coverage on interface, default context and banning. As there is no proper reset support for gen <= 3, we only do limited interface testing on those.

---
 tests/gem_reset_stats.c |   91 ++++++++++++++++++++++++++++++++++++++---------
 1 file changed, 75 insertions(+), 16 deletions(-)
tests/gem_reset_stats.c +75 -16
index 3719f40..2bb4681 100644
@@ -51,6 +51,9 @@
 #define RS_BATCH_PENDING (1 << 1)
 #define RS_UNKNOWN       (1 << 2)
 
+static uint32_t devid;
+static bool hw_contexts;
+
 struct local_drm_i915_reset_stats {
 	__u32 ctx_id;
 	__u32 flags;
@@ -101,6 +104,9 @@ static const struct target_ring {
 
 static bool has_context(const struct target_ring *ring)
 {
+	if (!hw_contexts)
+		return false;
+
 	if(ring->exec == I915_EXEC_RENDER)
 		return true;
 
@@ -277,7 +283,7 @@ static int inject_hang_ring(int fd, int ctx, int ring)
 
 	srandom(time(NULL));
 
-	if (intel_gen(intel_get_drm_devid(fd)) >= 8)
+	if (intel_gen(devid) >= 8)
 		cmd_len = 3;
 
 	buf = malloc(BUFSIZE);
@@ -960,7 +966,7 @@ static int _test_params(int fd, int ctx, uint32_t flags, uint32_t pad)
 
 typedef enum { root = 0, user } cap_t;
 
-static void test_param_ctx(const int fd, const int ctx, const cap_t cap)
+static void _check_param_ctx(const int fd, const int ctx, const cap_t cap)
 {
 	const uint32_t bad = rand() + 1;
 
@@ -981,8 +987,7 @@ static void check_params(const int fd, const int ctx, cap_t cap)
 	igt_assert(ioctl(fd, GET_RESET_STATS_IOCTL, 0) == -1);
 	igt_assert(_test_params(fd, 0xbadbad, 0, 0) == -ENOENT);
 
-	test_param_ctx(fd, 0, cap);
-	test_param_ctx(fd, ctx, cap);
+	_check_param_ctx(fd, ctx, cap);
 }
 
 static void _test_param(const int fd, const int ctx)
@@ -1002,7 +1007,7 @@ static void _test_param(const int fd, const int ctx)
 	igt_waitchildren();
 }
 
-static void test_params(void)
+static void test_params_ctx(void)
 {
 	int fd, ctx;
 
@@ -1015,29 +1020,76 @@ static void test_params(void)
 	close(fd);
 }
 
-#define RING_HAS_CONTEXTS current_ring->contexts(current_ring)
-#define RUN_CTX_TEST(...) do { igt_skip_on(RING_HAS_CONTEXTS == false); __VA_ARGS__; } while (0)
+static void test_params(void)
+{
+	int fd;
 
-int fd;
+	fd = drm_open_any();
+	igt_assert(fd >= 0);
 
-igt_main
+	_test_param(fd, 0);
+
+	close(fd);
+
+}
+
+static bool gem_has_hw_contexts(int fd)
 {
 	struct local_drm_i915_gem_context_create create;
-	uint32_t devid;
 	int ret;
 
+	memset(&create, 0, sizeof(create));
+	ret = drmIoctl(fd, CONTEXT_CREATE_IOCTL, &create);
+
+	if (ret == 0) {
+		drmIoctl(fd, CONTEXT_DESTROY_IOCTL, &create);
+		return true;
+	}
+
+	return false;
+}
+
+static bool gem_has_reset_stats(int fd)
+{
+	struct local_drm_i915_reset_stats rs;
+	int ret;
+
+	/* Carefully set flags and pad to zero, otherwise
+	   we get -EINVAL
+	*/
+	memset(&rs, 0, sizeof(rs));
+
+	ret = drmIoctl(fd, GET_RESET_STATS_IOCTL, &rs);
+	if (ret == 0)
+		return true;
+
+	/* If we get EPERM, we have support but did not
+	   have CAP_SYSADM */
+	if (ret == -1 && errno == EPERM)
+		return true;
+
+	return false;
+}
+
+#define RING_HAS_CONTEXTS (current_ring->contexts(current_ring))
+#define RUN_CTX_TEST(...) do { igt_skip_on(RING_HAS_CONTEXTS == false); __VA_ARGS__; } while (0)
+
+static int fd;
+
+igt_main
+{
 	igt_skip_on_simulation();
 
 	igt_fixture {
+		bool has_reset_stats;
 		fd = drm_open_any();
 		devid = intel_get_drm_devid(fd);
-		igt_require_f(intel_gen(devid) >= 4,
-			      "Architecture %d too old\n", intel_gen(devid));
 
-		ret = drmIoctl(fd, CONTEXT_CREATE_IOCTL, &create);
-		igt_skip_on_f(ret != 0 && (errno == ENODEV || errno == EINVAL),
-			      "Kernel is too old, or contexts not supported: %s\n",
-			      strerror(errno));
+		hw_contexts = gem_has_hw_contexts(fd);
+		has_reset_stats = gem_has_reset_stats(fd);
+
+		igt_require_f(has_reset_stats,
+			      "No reset stats ioctl support. Too old kernel?\n");
 	}
 
 	igt_subtest("params")
@@ -1052,6 +1104,13 @@ igt_main
 		igt_fixture
 			gem_require_ring(fd, current_ring->exec);
 
+		igt_fixture
+			igt_require_f(intel_gen(devid) >= 4,
+				      "gen %d doesn't support reset\n", intel_gen(devid));
+
+		igt_subtest_f("params-ctx-%s", name)
+			RUN_CTX_TEST(test_params_ctx());
+
 		igt_subtest_f("reset-stats-%s", name)
 			test_rs(4, 1, 0);
 
Signature · 1 line
1.7.9.5
#4 lore raw

Re: [PATCH] tests/gem_reset_stats: run non hw context tests also on older gens

Quoted 5 lines · Mika Kuoppala

To gain more coverage on interface, default context and banning. As there is no proper reset support for gen <= 3, we only do limited interface testing on those.

Applied, thanks for the test. And can you please list all your outstanding reset patches on our review board and sign someone up?

Thanks, Daniel

Quoted 174 lines
---
 tests/gem_reset_stats.c |   91 ++++++++++++++++++++++++++++++++++++++---------
 1 file changed, 75 insertions(+), 16 deletions(-)
tests/gem_reset_stats.c +75 -16
index 3719f40..2bb4681 100644
@@ -51,6 +51,9 @@
 #define RS_BATCH_PENDING (1 << 1)
 #define RS_UNKNOWN       (1 << 2)
 
+static uint32_t devid;
+static bool hw_contexts;
+
 struct local_drm_i915_reset_stats {
 	__u32 ctx_id;
 	__u32 flags;
@@ -101,6 +104,9 @@ static const struct target_ring {
 
 static bool has_context(const struct target_ring *ring)
 {
+	if (!hw_contexts)
+		return false;
+
 	if(ring->exec == I915_EXEC_RENDER)
 		return true;
 
@@ -277,7 +283,7 @@ static int inject_hang_ring(int fd, int ctx, int ring)
 
 	srandom(time(NULL));
 
-	if (intel_gen(intel_get_drm_devid(fd)) >= 8)
+	if (intel_gen(devid) >= 8)
 		cmd_len = 3;
 
 	buf = malloc(BUFSIZE);
@@ -960,7 +966,7 @@ static int _test_params(int fd, int ctx, uint32_t flags, uint32_t pad)
 
 typedef enum { root = 0, user } cap_t;
 
-static void test_param_ctx(const int fd, const int ctx, const cap_t cap)
+static void _check_param_ctx(const int fd, const int ctx, const cap_t cap)
 {
 	const uint32_t bad = rand() + 1;
 
@@ -981,8 +987,7 @@ static void check_params(const int fd, const int ctx, cap_t cap)
 	igt_assert(ioctl(fd, GET_RESET_STATS_IOCTL, 0) == -1);
 	igt_assert(_test_params(fd, 0xbadbad, 0, 0) == -ENOENT);
 
-	test_param_ctx(fd, 0, cap);
-	test_param_ctx(fd, ctx, cap);
+	_check_param_ctx(fd, ctx, cap);
 }
 
 static void _test_param(const int fd, const int ctx)
@@ -1002,7 +1007,7 @@ static void _test_param(const int fd, const int ctx)
 	igt_waitchildren();
 }
 
-static void test_params(void)
+static void test_params_ctx(void)
 {
 	int fd, ctx;
 
@@ -1015,29 +1020,76 @@ static void test_params(void)
 	close(fd);
 }
 
-#define RING_HAS_CONTEXTS current_ring->contexts(current_ring)
-#define RUN_CTX_TEST(...) do { igt_skip_on(RING_HAS_CONTEXTS == false); __VA_ARGS__; } while (0)
+static void test_params(void)
+{
+	int fd;
 
-int fd;
+	fd = drm_open_any();
+	igt_assert(fd >= 0);
 
-igt_main
+	_test_param(fd, 0);
+
+	close(fd);
+
+}
+
+static bool gem_has_hw_contexts(int fd)
 {
 	struct local_drm_i915_gem_context_create create;
-	uint32_t devid;
 	int ret;
 
+	memset(&create, 0, sizeof(create));
+	ret = drmIoctl(fd, CONTEXT_CREATE_IOCTL, &create);
+
+	if (ret == 0) {
+		drmIoctl(fd, CONTEXT_DESTROY_IOCTL, &create);
+		return true;
+	}
+
+	return false;
+}
+
+static bool gem_has_reset_stats(int fd)
+{
+	struct local_drm_i915_reset_stats rs;
+	int ret;
+
+	/* Carefully set flags and pad to zero, otherwise
+	   we get -EINVAL
+	*/
+	memset(&rs, 0, sizeof(rs));
+
+	ret = drmIoctl(fd, GET_RESET_STATS_IOCTL, &rs);
+	if (ret == 0)
+		return true;
+
+	/* If we get EPERM, we have support but did not
+	   have CAP_SYSADM */
+	if (ret == -1 && errno == EPERM)
+		return true;
+
+	return false;
+}
+
+#define RING_HAS_CONTEXTS (current_ring->contexts(current_ring))
+#define RUN_CTX_TEST(...) do { igt_skip_on(RING_HAS_CONTEXTS == false); __VA_ARGS__; } while (0)
+
+static int fd;
+
+igt_main
+{
 	igt_skip_on_simulation();
 
 	igt_fixture {
+		bool has_reset_stats;
 		fd = drm_open_any();
 		devid = intel_get_drm_devid(fd);
-		igt_require_f(intel_gen(devid) >= 4,
-			      "Architecture %d too old\n", intel_gen(devid));
 
-		ret = drmIoctl(fd, CONTEXT_CREATE_IOCTL, &create);
-		igt_skip_on_f(ret != 0 && (errno == ENODEV || errno == EINVAL),
-			      "Kernel is too old, or contexts not supported: %s\n",
-			      strerror(errno));
+		hw_contexts = gem_has_hw_contexts(fd);
+		has_reset_stats = gem_has_reset_stats(fd);
+
+		igt_require_f(has_reset_stats,
+			      "No reset stats ioctl support. Too old kernel?\n");
 	}
 
 	igt_subtest("params")
@@ -1052,6 +1104,13 @@ igt_main
 		igt_fixture
 			gem_require_ring(fd, current_ring->exec);
 
+		igt_fixture
+			igt_require_f(intel_gen(devid) >= 4,
+				      "gen %d doesn't support reset\n", intel_gen(devid));
+
+		igt_subtest_f("params-ctx-%s", name)
+			RUN_CTX_TEST(test_params_ctx());
+
 		igt_subtest_f("reset-stats-%s", name)
 			test_rs(4, 1, 0);
 
Signature · 6 lines
1.7.9.5

_______________________________________________
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
maintainer Tested-by Lu Hua on v1 1/2
#5 lore raw
In reply to Mika Kuoppala
Quoted 31 lines · Mika Kuoppala

We used to have per file descriptor hang stats for the i915_get_reset_stats_ioctl() and for default context banning.

commit 0eea67eb26000657079b7fc41079097942339452

Author: Ben Widawsky <[email protected]>
Date:   Fri Dec 6 14:11:19 2013 -0800

drm/i915: Create a per file_priv default context

made having separate hangstats in file_private redundant as i915_hw_context already contained hangstats. So

commit c482972a086e03e6a6d27e4f7af2d868bf659648

Author: Ben Widawsky <[email protected]>
Date:   Fri Dec 6 14:11:20 2013 -0800

drm/i915: Piggy back hangstats off of contexts

consolidated the hangstats and enabled further improvements.

commit 44e2c0705a19e09d7b0f30a591f92e473e5ef89e

Author: Mika Kuoppala <[email protected]>
Date:   Thu Jan 30 16:01:15 2014 +0200

drm/i915: Use i915_hw_context to set reset stats

tried to reap full benefits of consolidation but fell short as we never 'switch' to the fake private context on gens that don't have hw_contexts, so request->ctx remained NULL on those.

Quoted 5 lines

Fix this by 'switching' to fake context so that when request is submitted to ring, proper context gets assigned to it.

References: https://bugs.freedesktop.org/show_bug.cgi?id=76055

I'd say this still applies:

Quoted 30 lines
---
 drivers/gpu/drm/i915/i915_gem_context.c |    6 ++++--
 1 file changed, 4 insertions(+), 2 deletions(-)
drivers/gpu/drm/i915/i915_gem_context.c +4 -2
index ce41cff..b5a5837 100644
@@ -775,9 +775,11 @@ int i915_switch_context(struct intel_ring_buffer *ring,
 
 	BUG_ON(file && to == NULL);
 
-	/* We have the fake context, but don't supports switching. */
-	if (!HAS_HW_CONTEXTS(ring->dev))
+	/* We have the fake context */
+	if (!HAS_HW_CONTEXTS(ring->dev)) {
+		ring->last_context = to;
 		return 0;
+	}
 
 	return do_switch(ring, to);
 }
Signature · 6 lines
1.7.9.5

_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx
Signature · 1 line
Jani Nikula, Intel Open Source Technology Center
#6 lore raw
Quoted 36 lines · Jani Nikula
Quoted 35 lines · Mika Kuoppala

We used to have per file descriptor hang stats for the i915_get_reset_stats_ioctl() and for default context banning.

commit 0eea67eb26000657079b7fc41079097942339452

Author: Ben Widawsky <[email protected]>
Date:   Fri Dec 6 14:11:19 2013 -0800

drm/i915: Create a per file_priv default context

made having separate hangstats in file_private redundant as i915_hw_context already contained hangstats. So

commit c482972a086e03e6a6d27e4f7af2d868bf659648

Author: Ben Widawsky <[email protected]>
Date:   Fri Dec 6 14:11:20 2013 -0800

drm/i915: Piggy back hangstats off of contexts

consolidated the hangstats and enabled further improvements.

commit 44e2c0705a19e09d7b0f30a591f92e473e5ef89e

Author: Mika Kuoppala <[email protected]>
Date:   Thu Jan 30 16:01:15 2014 +0200

drm/i915: Use i915_hw_context to set reset stats

tried to reap full benefits of consolidation but fell short as we never 'switch' to the fake private context on gens that don't have hw_contexts, so request->ctx remained NULL on those.

Fix this by 'switching' to fake context so that when request is submitted to ring, proper context gets assigned to it.

Quoted 5 lines
Quoted 1 line

References: https://bugs.freedesktop.org/show_bug.cgi?id=76055

I'd say this still applies:

Also if your patch intends to fix a bug, please use a Bugzilla: reference, I tend to use References only when there's an issue seen in a report, but it's not the main issue.
-Daniel

Quoted 38 lines
Quoted 31 lines
---
 drivers/gpu/drm/i915/i915_gem_context.c |    6 ++++--
 1 file changed, 4 insertions(+), 2 deletions(-)
drivers/gpu/drm/i915/i915_gem_context.c +4 -2
index ce41cff..b5a5837 100644
@@ -775,9 +775,11 @@ int i915_switch_context(struct intel_ring_buffer *ring,
 
 	BUG_ON(file && to == NULL);
 
-	/* We have the fake context, but don't supports switching. */
-	if (!HAS_HW_CONTEXTS(ring->dev))
+	/* We have the fake context */
+	if (!HAS_HW_CONTEXTS(ring->dev)) {
+		ring->last_context = to;
 		return 0;
+	}
 
 	return do_switch(ring, to);
 }
Signature · 6 lines
1.7.9.5

_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx
Signature · 5 lines
Jani Nikula, Intel Open Source Technology Center
_______________________________________________
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
#7 lore raw
Quoted 39 lines · Daniel Vetter
Quoted 38 lines · Jani Nikula
Quoted 36 lines · Mika Kuoppala

We used to have per file descriptor hang stats for the i915_get_reset_stats_ioctl() and for default context banning.

commit 0eea67eb26000657079b7fc41079097942339452

Author: Ben Widawsky <[email protected]>
Date:   Fri Dec 6 14:11:19 2013 -0800

drm/i915: Create a per file_priv default context

made having separate hangstats in file_private redundant as i915_hw_context already contained hangstats. So

commit c482972a086e03e6a6d27e4f7af2d868bf659648

Author: Ben Widawsky <[email protected]>
Date:   Fri Dec 6 14:11:20 2013 -0800

drm/i915: Piggy back hangstats off of contexts

consolidated the hangstats and enabled further improvements.

commit 44e2c0705a19e09d7b0f30a591f92e473e5ef89e

Author: Mika Kuoppala <[email protected]>
Date:   Thu Jan 30 16:01:15 2014 +0200

drm/i915: Use i915_hw_context to set reset stats

tried to reap full benefits of consolidation but fell short as we never 'switch' to the fake private context on gens that don't have hw_contexts, so request->ctx remained NULL on those.

Fix this by 'switching' to fake context so that when request is submitted to ring, proper context gets assigned to it.

References: https://bugs.freedesktop.org/show_bug.cgi?id=76055

Quoted 7 lines
Quoted 3 lines

I'd say this still applies:

Also if your patch intends to fix a bug, please use a Bugzilla: reference, I tend to use References only when there's an issue seen in a report, but it's not the main issue.

Queued for -next, thanks for the patch.
-Daniel

Signature · 3 lines
Daniel Vetter
Software Engineer, Intel Corporation
+41 (0) 79 365 57 48 - http://blog.ffwll.ch