Skip to contentSkip to conversation

kms_flip_event_leak and kms_vblank fixes for

PATCH Superseded [email protected] 2 participants 13 messages May 16, 2016 – May 24, 2016 Library code

v5 of 6 Superseded Latest is v6 →

lore
Share & tools
lore
#0 lore raw

v5 cover kms_flip_event_leak and kms_vblank fixes for

From: Robert Foss <[email protected]>

In addition to the changes made in v4, danvet suggested that the argument for the (newly renamed) helper function kmstest_get_vbl_flag should be changed.

I investigated it and found that all of the places where the helper function currently is used, would require more of a rework with changed arguments than with unchanged arguments, so I decided to not change the arguments.

Changes since v1:
- kms_vblank: Removed un-used members of data_t struct.
- Rename plane_counter to n_planes.
- Removed un-needed handling CURSOR plane location.
- Added names for additional planes in update kmstest_plane_name.

Changes since v2:
- Rebased onto trunk which already contains kms_flip_event_leak changes.

Changes since v3:
- kms_vblank.c: Removed drm_fd from data_t, to decrease the number of unrelated changes in the patch.
- kms_vblank.c: Changed "int valid_tests" to "unsigned valid_tests".
- kms_vblank.c: Changed crtc_id_flag/crtc_id_to_flag to pipe_id_flag/pipe_id_to_flag.
- kms_vblank.c: Changed to "vbl.request.type |= crtc_id_flag;".
- kms_vblank.c: Moved crtc_id_to_flag to igt_kms.c.
- kms_vblank.c: Renamed crtc_id_to_flag to pipe_id_to_vbl_flag.
- kms_flip.c: Changed test busy_mode boolean into mode_busy bitfield in struct data.
- igt_kms.h: Moved IGT_PLANE_CURSOR position comment.

Changes since v4:
- kms_vblank.c: Removed whitespace at end of file.
- igt_kms: Changed name of helper function to kmstest_get_vbl_flag
- igt_kms: Changed kmstest_get_vbl_flag to not set DRM_VBLANK_RELATIVE flag

Robert Foss (7):
lib/igt_kms: Add support for up to 10 planes. lib/igt_kms: Fix plane counting in igt_display_init. lib/igt_kms: Switch to verbose assert.
lib/igt_kms: Added pipe_id_to_vbl_flag() to igt_kms.
kms_vblank: Switch from using crtc0 statically to explicitly setting
mode.
igt_kms: Change igt_wait_for_vblank to use helper function.
kms_flip: Change __wait_for_vblank to use helper function.

 lib/igt_kms.c      |  36 ++++++++++--
 lib/igt_kms.h      |   9 ++-
 tests/kms_flip.c   |   8 +--
 tests/kms_vblank.c | 161 ++++++++++++++++++++++++++++++++++++++++++++---------
 4 files changed, 179 insertions(+), 35 deletions(-)
Signature · 6 lines
2.7.4

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

PATCH v5 1/7 lib/igt_kms: Add support for up to 10 planes.

From: Robert Foss <[email protected]>

Increase the number of planes supported to 10.

kmstest_plane_name only previously supported 4 planes, this patch adds support for up to 10 planes.

---
 lib/igt_kms.c | 6 ++++++
 lib/igt_kms.h | 8 +++++++-
 2 files changed, 13 insertions(+), 1 deletion(-)
lib/igt_kms.c +6 -0
index f85be1e..c2283fd 100644
@@ -320,6 +320,12 @@ const char *kmstest_plane_name(enum igt_plane plane)
 		[IGT_PLANE_1] = "plane1",
 		[IGT_PLANE_2] = "plane2",
 		[IGT_PLANE_3] = "plane3",
+		[IGT_PLANE_4] = "plane4",
+		[IGT_PLANE_5] = "plane5",
+		[IGT_PLANE_6] = "plane6",
+		[IGT_PLANE_7] = "plane7",
+		[IGT_PLANE_8] = "plane8",
+		[IGT_PLANE_9] = "plane9",
 		[IGT_PLANE_CURSOR] = "cursor",
 	};
 
lib/igt_kms.h +7 -1
index 5c83401..d9a4e11 100644
@@ -54,7 +54,13 @@ enum igt_plane {
 	IGT_PLANE_PRIMARY = IGT_PLANE_1,
 	IGT_PLANE_2,
 	IGT_PLANE_3,
-	IGT_PLANE_CURSOR,
+	IGT_PLANE_4,
+	IGT_PLANE_5,
+	IGT_PLANE_6,
+	IGT_PLANE_7,
+	IGT_PLANE_8,
+	IGT_PLANE_9,
+	IGT_PLANE_CURSOR, /* IGT_PLANE_CURSOR is always the last plane. */
 	IGT_MAX_PLANES,
 };
 
Signature · 6 lines
2.7.4

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

PATCH v5 2/7 lib/igt_kms: Fix plane counting in igt_display_init.

In reply to [email protected]

From: Robert Foss <[email protected]>

Fix issue where the plane counting fails due to the number and configuration of planes being unlike the intel configuration.

---
 lib/igt_kms.c | 6 ++++--
 1 file changed, 4 insertions(+), 2 deletions(-)
lib/igt_kms.c +4 -2
index c2283fd..876f65c 100644
@@ -1289,6 +1289,7 @@ void igt_display_init(igt_display_t *display, int drm_fd)
 		igt_plane_t *plane;
 		int p = IGT_PLANE_2;
 		int j, type;
+		uint8_t n_planes = 0;
 
 		pipe->crtc_id = resources->crtcs[i];
 		pipe->display = display;
@@ -1336,8 +1337,10 @@ void igt_display_init(igt_display_t *display, int drm_fd)
 				break;
 			}
 
+			n_planes++;
 			plane->pipe = pipe;
 			plane->drm_plane = drm_plane;
+
 			if (is_atomic == 0) {
 				display->is_atomic = 1;
 				igt_atomic_fill_plane_props(display, plane, IGT_NUM_PLANE_PROPS, igt_plane_prop_names);
@@ -1386,8 +1389,7 @@ void igt_display_init(igt_display_t *display, int drm_fd)
 			plane->is_cursor = true;
 		}
 
-		/* planes = 1 primary, (p-1) sprites, 1 cursor */
-		pipe->n_planes = p + 1;
+		pipe->n_planes = n_planes;
 
 		/* make sure we don't overflow the plane array */
 		igt_assert(pipe->n_planes <= IGT_MAX_PLANES);
Signature · 6 lines
2.7.4

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

PATCH v5 3/7 lib/igt_kms: Switch to verbose assert.

In reply to [email protected]

From: Robert Foss <[email protected]>

Switch igt_assert to igt_assert_lte to provide more diagnostic information.

---
 lib/igt_kms.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
lib/igt_kms.c +1 -1
index 876f65c..9af9603 100644
@@ -1392,7 +1392,7 @@ void igt_display_init(igt_display_t *display, int drm_fd)
 		pipe->n_planes = n_planes;
 
 		/* make sure we don't overflow the plane array */
-		igt_assert(pipe->n_planes <= IGT_MAX_PLANES);
+		igt_assert_lte(pipe->n_planes, IGT_MAX_PLANES);
 	}
 
 	/*
Signature · 6 lines
2.7.4

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

PATCH v5 4/7 lib/igt_kms: Added pipe_id_to_vbl_flag() to igt_kms.

In reply to [email protected]

From: Robert Foss <[email protected]>

Added function pipe_id_to_vbl_flag() to igt_kms. This function formats a pipe_id into the format expected by DRM while processing DRM_IOCTL_WAIT_VBLANK.

---
 lib/igt_kms.c | 16 ++++++++++++++++
 lib/igt_kms.h |  1 +
 2 files changed, 17 insertions(+)
lib/igt_kms.c +16 -0
index 9af9603..5ef73c8 100644
@@ -2557,3 +2557,19 @@ void igt_reset_connectors(void)
 		close(fd);
 	}
 }
+
+/**
+ * kmstest_get_vbl_flag:
+ *
+ * Convert a pipe id into the flag representation
+ * expected in DRM while processing DRM_IOCTL_WAIT_VBLANK.
+ */
+uint32_t kmstest_get_vbl_flag(uint32_t pipe_id)
+{
+	if (pipe_id == 0)
+		return 0;
+	else if (pipe_id == 1)
+		return _DRM_VBLANK_SECONDARY;
+	else
+		return pipe_id << 1;
+}
lib/igt_kms.h +1 -0
index d9a4e11..9fecf2b 100644
@@ -400,6 +400,7 @@ void igt_wait_for_vblank(int drm_fd, enum pipe pipe);
 
 void igt_enable_connectors(void);
 void igt_reset_connectors(void);
+uint32_t kmstest_get_vbl_flag(uint32_t pipe_id);
 
 #define EDID_LENGTH 128
 const unsigned char* igt_kms_get_base_edid(void);
Signature · 6 lines
2.7.4

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

PATCH v5 5/7 kms_vblank: Switch from using crtc0 statically to explicitly setting mode.

In reply to [email protected]

From: Robert Foss <[email protected]>

Previously crtc0 was statically used for VBLANK tests, but that assumption is not valid for the VC4 platform. Instead we're now explicitly setting the mode.

Also add support for testing all connected connectors during the same test.

---
 tests/kms_vblank.c | 161 ++++++++++++++++++++++++++++++++++++++++++++---------
 1 file changed, 136 insertions(+), 25 deletions(-)
tests/kms_vblank.c +136 -25
index 40ab6fd..9bc4929 100644
@@ -44,6 +44,14 @@
 
 IGT_TEST_DESCRIPTION("Test speed of WaitVblank.");
 
+typedef struct {
+	igt_display_t display;
+	struct igt_fb primary_fb;
+	igt_output_t *output;
+	enum pipe pipe;
+	uint8_t mode_busy:1;
+} data_t;
+
 static double elapsed(const struct timespec *start,
 		      const struct timespec *end,
 		      int loop)
@@ -51,38 +59,119 @@ static double elapsed(const struct timespec *start,
 	return (1e6*(end->tv_sec - start->tv_sec) + (end->tv_nsec - start->tv_nsec)/1000)/loop;
 }
 
-static bool crtc0_active(int fd)
+static bool prepare_crtc(data_t *data, int fd, igt_output_t *output)
 {
-	union drm_wait_vblank vbl;
+	drmModeModeInfo *mode;
+	igt_display_t *display = &data->display;
+	igt_plane_t *primary;
+
+	/* select the pipe we want to use */
+	igt_output_set_pipe(output, data->pipe);
+	igt_display_commit(display);
+
+	if (!output->valid) {
+		igt_output_set_pipe(output, PIPE_ANY);
+		igt_display_commit(display);
+		return false;
+	}
 
-	memset(&vbl, 0, sizeof(vbl));
-	vbl.request.type = DRM_VBLANK_RELATIVE;
-	return drmIoctl(fd, DRM_IOCTL_WAIT_VBLANK, &vbl) == 0;
+	/* create and set the primary plane fb */
+	mode = igt_output_get_mode(output);
+	igt_create_color_fb(fd, mode->hdisplay, mode->vdisplay,
+			    DRM_FORMAT_XRGB8888,
+			    LOCAL_DRM_FORMAT_MOD_NONE,
+			    0.0, 0.0, 0.0,
+			    &data->primary_fb);
+
+	primary = igt_output_get_plane(output, IGT_PLANE_PRIMARY);
+	igt_plane_set_fb(primary, &data->primary_fb);
+
+	igt_display_commit(display);
+
+	igt_wait_for_vblank(fd, data->pipe);
+
+	return true;
+}
+
+static void cleanup_crtc(data_t *data, int fd, igt_output_t *output)
+{
+	igt_display_t *display = &data->display;
+	igt_plane_t *primary;
+
+	igt_remove_fb(fd, &data->primary_fb);
+
+	primary = igt_output_get_plane(output, IGT_PLANE_PRIMARY);
+	igt_plane_set_fb(primary, NULL);
+
+	igt_output_set_pipe(output, PIPE_ANY);
+	igt_display_commit(display);
+}
+
+static void run_test(data_t *data, int fd, void (*testfunc)(data_t *, int))
+{
+	igt_display_t *display = &data->display;
+	igt_output_t *output;
+	enum pipe p;
+	unsigned int valid_tests = 0;
+
+	for_each_connected_output(display, output) {
+		for_each_pipe(display, p) {
+			data->pipe = p;
+
+			if (!prepare_crtc(data, fd, output))
+				continue;
+
+			valid_tests++;
+
+			igt_info("Beginning %s on pipe %s, connector %s\n",
+				 igt_subtest_name(),
+				 kmstest_pipe_name(data->pipe),
+				 igt_output_name(output));
+
+			testfunc(data, fd);
+
+			igt_info("\n%s on pipe %s, connector %s: PASSED\n\n",
+				 igt_subtest_name(),
+				 kmstest_pipe_name(data->pipe),
+				 igt_output_name(output));
+
+			/* cleanup what prepare_crtc() has done */
+			cleanup_crtc(data, fd, output);
+		}
+	}
+
+	igt_require_f(valid_tests, "no valid crtc/connector combinations found\n");
 }
 
-static void accuracy(int fd)
+static void accuracy(data_t *data, int fd)
 {
 	union drm_wait_vblank vbl;
 	unsigned long target;
+	uint32_t pipe_id_flag;
 	int n;
 
 	memset(&vbl, 0, sizeof(vbl));
+	pipe_id_flag = kmstest_get_vbl_flag(data->pipe);
 
 	vbl.request.type = DRM_VBLANK_RELATIVE;
+	vbl.request.type |= pipe_id_flag;
 	vbl.request.sequence = 1;
 	do_ioctl(fd, DRM_IOCTL_WAIT_VBLANK, &vbl);
 
 	target = vbl.reply.sequence + 60;
 	for (n = 0; n < 60; n++) {
 		vbl.request.type = DRM_VBLANK_RELATIVE;
+		vbl.request.type |= pipe_id_flag;
 		vbl.request.sequence = 1;
 		do_ioctl(fd, DRM_IOCTL_WAIT_VBLANK, &vbl);
 
 		vbl.request.type = DRM_VBLANK_ABSOLUTE | DRM_VBLANK_EVENT;
+		vbl.request.type |= pipe_id_flag;
 		vbl.request.sequence = target;
 		do_ioctl(fd, DRM_IOCTL_WAIT_VBLANK, &vbl);
 	}
 	vbl.request.type = DRM_VBLANK_RELATIVE;
+	vbl.request.type |= pipe_id_flag;
 	vbl.request.sequence = 0;
 	do_ioctl(fd, DRM_IOCTL_WAIT_VBLANK, &vbl);
 	igt_assert_eq(vbl.reply.sequence, target);
@@ -94,22 +183,26 @@ static void accuracy(int fd)
 	}
 }
 
-static void vblank_query(int fd, bool busy)
+static void vblank_query(data_t *data, int fd)
 {
 	union drm_wait_vblank vbl;
 	struct timespec start, end;
 	unsigned long sq, count = 0;
 	struct drm_event_vblank buf;
+	uint32_t pipe_id_flag;
 
 	memset(&vbl, 0, sizeof(vbl));
+	pipe_id_flag = kmstest_get_vbl_flag(data->pipe);
 
-	if (busy) {
+	if (data->mode_busy) {
 		vbl.request.type = DRM_VBLANK_RELATIVE | DRM_VBLANK_EVENT;
+		vbl.request.type |= pipe_id_flag;
 		vbl.request.sequence = 72;
 		do_ioctl(fd, DRM_IOCTL_WAIT_VBLANK, &vbl);
 	}
 
 	vbl.request.type = DRM_VBLANK_RELATIVE;
+	vbl.request.type |= pipe_id_flag;
 	vbl.request.sequence = 0;
 	do_ioctl(fd, DRM_IOCTL_WAIT_VBLANK, &vbl);
 
@@ -118,6 +211,7 @@ static void vblank_query(int fd, bool busy)
 	clock_gettime(CLOCK_MONOTONIC, &start);
 	do {
 		vbl.request.type = DRM_VBLANK_RELATIVE;
+		vbl.request.type |= pipe_id_flag;
 		vbl.request.sequence = 0;
 		do_ioctl(fd, DRM_IOCTL_WAIT_VBLANK, &vbl);
 		count++;
@@ -125,28 +219,32 @@ static void vblank_query(int fd, bool busy)
 	clock_gettime(CLOCK_MONOTONIC, &end);
 
 	igt_info("Time to query current counter (%s):		%7.3fµs\n",
-		 busy ? "busy" : "idle", elapsed(&start, &end, count));
+		 data->mode_busy ? "busy" : "idle", elapsed(&start, &end, count));
 
-	if (busy)
+	if (data->mode_busy)
 		igt_assert_eq(read(fd, &buf, sizeof(buf)), sizeof(buf));
 }
 
-static void vblank_wait(int fd, bool busy)
+static void vblank_wait(data_t *data, int fd)
 {
 	union drm_wait_vblank vbl;
 	struct timespec start, end;
 	unsigned long sq, count = 0;
 	struct drm_event_vblank buf;
+	uint32_t pipe_id_flag;
 
 	memset(&vbl, 0, sizeof(vbl));
+	pipe_id_flag = kmstest_get_vbl_flag(data->pipe);
 
-	if (busy) {
+	if (data->mode_busy) {
 		vbl.request.type = DRM_VBLANK_RELATIVE | DRM_VBLANK_EVENT;
+		vbl.request.type |= pipe_id_flag;
 		vbl.request.sequence = 72;
 		do_ioctl(fd, DRM_IOCTL_WAIT_VBLANK, &vbl);
 	}
 
 	vbl.request.type = DRM_VBLANK_RELATIVE;
+	vbl.request.type |= pipe_id_flag;
 	vbl.request.sequence = 0;
 	do_ioctl(fd, DRM_IOCTL_WAIT_VBLANK, &vbl);
 
@@ -155,6 +253,7 @@ static void vblank_wait(int fd, bool busy)
 	clock_gettime(CLOCK_MONOTONIC, &start);
 	do {
 		vbl.request.type = DRM_VBLANK_RELATIVE;
+		vbl.request.type |= pipe_id_flag;
 		vbl.request.sequence = 1;
 		do_ioctl(fd, DRM_IOCTL_WAIT_VBLANK, &vbl);
 		count++;
@@ -163,36 +262,48 @@ static void vblank_wait(int fd, bool busy)
 
 	igt_info("Time to wait for %ld/%d vblanks (%s):		%7.3fµs\n",
 		 count, (int)(vbl.reply.sequence - sq),
-		 busy ? "busy" : "idle",
+		 data->mode_busy ? "busy" : "idle",
 		 elapsed(&start, &end, count));
 
-	if (busy)
+	if (data->mode_busy)
 		igt_assert_eq(read(fd, &buf, sizeof(buf)), sizeof(buf));
 }
 
 igt_main
 {
 	int fd;
+	data_t data;
 
 	igt_skip_on_simulation();
 
 	igt_fixture {
 		fd = drm_open_driver(DRIVER_ANY);
-		igt_require(crtc0_active(fd));
+		kmstest_set_vt_graphics_mode();
+		igt_display_init(&data.display, fd);
 	}
 
-	igt_subtest("accuracy")
-		accuracy(fd);
+	igt_subtest("accuracy") {
+		data.mode_busy = 0;
+		run_test(&data, fd, accuracy);
+	}
 
-	igt_subtest("query-idle")
-		vblank_query(fd, false);
+	igt_subtest("query-idle") {
+		data.mode_busy = 0;
+		run_test(&data, fd, vblank_query);
+	}
 
-	igt_subtest("query-busy")
-		vblank_query(fd, true);
+	igt_subtest("query-busy") {
+		data.mode_busy = 1;
+		run_test(&data, fd, vblank_query);
+	}
 
-	igt_subtest("wait-idle")
-		vblank_wait(fd, false);
+	igt_subtest("wait-idle") {
+		data.mode_busy = 0;
+		run_test(&data, fd, vblank_wait);
+	}
 
-	igt_subtest("wait-busy")
-		vblank_wait(fd, true);
+	igt_subtest("wait-busy") {
+		data.mode_busy = 1;
+		run_test(&data, fd, vblank_wait);
+	}
 }
Signature · 6 lines
2.7.4

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

PATCH v5 6/7 igt_kms: Change igt_wait_for_vblank to use helper function.

In reply to [email protected]

From: Robert Foss <[email protected]>

Change igt_wait_for_vblank() to use kmstest_get_vbl_flag() helper function.

---
 lib/igt_kms.c | 6 ++++--
 1 file changed, 4 insertions(+), 2 deletions(-)
lib/igt_kms.c +4 -2
index 5ef73c8..bc022c8 100644
@@ -2488,11 +2488,13 @@ void igt_crtc_set_background(igt_pipe_t *pipe, uint64_t background)
 void igt_wait_for_vblank(int drm_fd, enum pipe pipe)
 {
 	drmVBlank wait_vbl;
+	uint32_t pipe_id_flag;
 
 	memset(&wait_vbl, 0, sizeof(wait_vbl));
+	pipe_id_flag = kmstest_get_vbl_flag(pipe);
 
-	wait_vbl.request.type = pipe << DRM_VBLANK_HIGH_CRTC_SHIFT |
-				DRM_VBLANK_RELATIVE;
+	wait_vbl.request.type = DRM_VBLANK_RELATIVE;
+	wait_vbl.request.type |= pipe_id_flag;
 	wait_vbl.request.sequence = 1;
 
 	igt_assert(drmWaitVBlank(drm_fd, &wait_vbl) == 0);
Signature · 6 lines
2.7.4

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

PATCH v5 7/7 kms_flip: Change __wait_for_vblank to use helper function.

In reply to [email protected]

From: Robert Foss <[email protected]>

Change __wait_for_vblank() to use kmstest_get_vbl_flag() helper function.

---
 tests/kms_flip.c | 8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)
tests/kms_flip.c +4 -4
index eda2fcc..0ecca07 100644
@@ -481,15 +481,15 @@ static int __wait_for_vblank(unsigned int flags, int crtc_idx,
 {
 	drmVBlank wait_vbl;
 	int ret;
-	unsigned crtc_idx_mask;
+	uint32_t pipe_id_flag;
 	bool event = !(flags & TEST_VBLANK_BLOCK);
 
 	memset(&wait_vbl, 0, sizeof(wait_vbl));
 
-	crtc_idx_mask = crtc_idx << DRM_VBLANK_HIGH_CRTC_SHIFT;
-	igt_assert(!(crtc_idx_mask & ~DRM_VBLANK_HIGH_CRTC_MASK));
+	pipe_id_flag = kmstest_get_vbl_flag(crtc_idx);
+	igt_assert(!(pipe_id_flag & ~DRM_VBLANK_HIGH_CRTC_MASK));
 
-	wait_vbl.request.type = crtc_idx_mask;
+	wait_vbl.request.type = pipe_id_flag;
 	if (flags & TEST_VBLANK_ABSOLUTE)
 		wait_vbl.request.type |= DRM_VBLANK_ABSOLUTE;
 	else
Signature · 6 lines
2.7.4

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

Re: [PATCH i-g-t v5 7/7] kms_flip: Change __wait_for_vblank to use helper function.

Quoted 15 lines · On Mon, May 16, 2016 at 09:38:32AM -0400, [email protected] wrote:

From: Robert Foss <[email protected]>

Change __wait_for_vblank() to use kmstest_get_vbl_flag() helper function.

---
 tests/kms_flip.c | 8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)
tests/kms_flip.c +0 -0
index eda2fcc..0ecca07 100644
@@ -481,15 +481,15 @@ static int __wait_for_vblank(unsigned int flags, int crtc_idx,
 {

tests/kms_flip.c static int __wait_for_vblank(unsigned int flags, int crtc_idx, · line 490

Quoted 10 lines
tests/kms_flip.c +1 -3
 	drmVBlank wait_vbl;
 	int ret;
-	unsigned crtc_idx_mask;
+	uint32_t pipe_id_flag;
 	bool event = !(flags & TEST_VBLANK_BLOCK);
 
 	memset(&wait_vbl, 0, sizeof(wait_vbl));
 
-	crtc_idx_mask = crtc_idx << DRM_VBLANK_HIGH_CRTC_SHIFT;
-	igt_assert(!(crtc_idx_mask & ~DRM_VBLANK_HIGH_CRTC_MASK));

If crtc_idx is 1 (pipe B), crtc_idx_mask = (1 << DRM_VBLANK_HIGH_CRTC_SHIFT) = 2.

tests/kms_flip.c static int __wait_for_vblank(unsigned int flags, int crtc_idx, · line 490

Quoted 2 lines
tests/kms_flip.c +2 -0
+	pipe_id_flag = kmstest_get_vbl_flag(crtc_idx);
+	igt_assert(!(pipe_id_flag & ~DRM_VBLANK_HIGH_CRTC_MASK));

If crtc_idx is 1 (pipe B), kmstest_get_vbl_flag(crtc_idx) = DRM_VBLANK_SECONDARY = 0x20000000

And the assertion fails as !(0x20000000 & ~0x0000003e) = 0.

Should kmstest_get_vbl_flag() always return pipe_id << 1?

Quoted 9 lines
tests/kms_flip.c +1 -1
-	wait_vbl.request.type = crtc_idx_mask;
+	wait_vbl.request.type = pipe_id_flag;
 	if (flags & TEST_VBLANK_ABSOLUTE)
 		wait_vbl.request.type |= DRM_VBLANK_ABSOLUTE;
 	else
Signature · 1 line
2.7.4

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

#9 lore raw

Re: [PATCH i-g-t v5 7/7] kms_flip: Change __wait_for_vblank to use helper function.

Quoted 27 lines · Marius Vlad
Quoted 15 lines · On Mon, May 16, 2016 at 09:38:32AM -0400, [email protected] wrote:

From: Robert Foss <[email protected]>

Change __wait_for_vblank() to use kmstest_get_vbl_flag() helper function.

---
  tests/kms_flip.c | 8 ++++----
  1 file changed, 4 insertions(+), 4 deletions(-)
tests/kms_flip.c +0 -0
index eda2fcc..0ecca07 100644
@@ -481,15 +481,15 @@ static int __wait_for_vblank(unsigned int flags, int crtc_idx,
  {

tests/kms_flip.c static int __wait_for_vblank(unsigned int flags, int crtc_idx, · line 490

Quoted 10 lines
tests/kms_flip.c +1 -3
  	drmVBlank wait_vbl;
  	int ret;
-	unsigned crtc_idx_mask;
+	uint32_t pipe_id_flag;
  	bool event = !(flags & TEST_VBLANK_BLOCK);
 
  	memset(&wait_vbl, 0, sizeof(wait_vbl));
 
-	crtc_idx_mask = crtc_idx << DRM_VBLANK_HIGH_CRTC_SHIFT;
-	igt_assert(!(crtc_idx_mask & ~DRM_VBLANK_HIGH_CRTC_MASK));

If crtc_idx is 1 (pipe B), crtc_idx_mask = (1 << DRM_VBLANK_HIGH_CRTC_SHIFT) = 2.

Quoted 7 lines

tests/kms_flip.c static int __wait_for_vblank(unsigned int flags, int crtc_idx, · line 490

Quoted 2 lines
tests/kms_flip.c +2 -0
+	pipe_id_flag = kmstest_get_vbl_flag(crtc_idx);
+	igt_assert(!(pipe_id_flag & ~DRM_VBLANK_HIGH_CRTC_MASK));

If crtc_idx is 1 (pipe B), kmstest_get_vbl_flag(crtc_idx) = DRM_VBLANK_SECONDARY = 0x20000000

And the assertion fails as !(0x20000000 & ~0x0000003e) = 0.

Should kmstest_get_vbl_flag() always return pipe_id << 1?

From re-reading the pipe id parsing code in drm_irq.c drm_wait_vblank()
it would seem to me that the assertion is incorrect, specifically for the case of _DRM_VBLANK_SECONDARY.

Supplying the flag _DRM_VBLANK_SECONDARY and setting the high crtc id field to 1 fboth seem to be valid ways to communicate the same thing.

	flags = vblwait->request.type & _DRM_VBLANK_FLAGS_MASK;
	high_pipe = (vblwait->request.type &_DRM_VBLANK_HIGH_CRTC_MASK);
	if (high_pipe)
		pipe = high_pipe >> _DRM_VBLANK_HIGH_CRTC_SHIFT;
	else
		pipe = flags & _DRM_VBLANK_SECONDARY ? 1 : 0;

Maybe adding correct and more thorough assertion to kmstest_get_vbl_flag() and removing the failing assertion is the way forward.

Quoted 10 lines
Quoted 10 lines
tests/kms_flip.c +1 -2
-	wait_vbl.request.type = crtc_idx_mask;
+	wait_vbl.request.type = pipe_id_flag;
  	if (flags & TEST_VBLANK_ABSOLUTE)
  		wait_vbl.request.type |= DRM_VBLANK_ABSOLUTE;
  	else
--

2.7.4

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

#10 lore raw

Re: [PATCH i-g-t v5 7/7] kms_flip: Change __wait_for_vblank to use helper function.

Quoted 36 lines · Robert Foss
Quoted 33 lines · Marius Vlad
Quoted 15 lines · On Mon, May 16, 2016 at 09:38:32AM -0400, [email protected] wrote:

From: Robert Foss <[email protected]>

Change __wait_for_vblank() to use kmstest_get_vbl_flag() helper function.

---
 tests/kms_flip.c | 8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)
tests/kms_flip.c +0 -0
index eda2fcc..0ecca07 100644
@@ -481,15 +481,15 @@ static int __wait_for_vblank(unsigned int flags, int crtc_idx,
 {

tests/kms_flip.c static int __wait_for_vblank(unsigned int flags, int crtc_idx, · line 490

Quoted 10 lines
tests/kms_flip.c +1 -3
 	drmVBlank wait_vbl;
 	int ret;
-	unsigned crtc_idx_mask;
+	uint32_t pipe_id_flag;
 	bool event = !(flags & TEST_VBLANK_BLOCK);
 
 	memset(&wait_vbl, 0, sizeof(wait_vbl));
 
-	crtc_idx_mask = crtc_idx << DRM_VBLANK_HIGH_CRTC_SHIFT;
-	igt_assert(!(crtc_idx_mask & ~DRM_VBLANK_HIGH_CRTC_MASK));

If crtc_idx is 1 (pipe B), crtc_idx_mask = (1 << DRM_VBLANK_HIGH_CRTC_SHIFT) = 2.

tests/kms_flip.c static int __wait_for_vblank(unsigned int flags, int crtc_idx, · line 490

Quoted 2 lines
tests/kms_flip.c +2 -0
+	pipe_id_flag = kmstest_get_vbl_flag(crtc_idx);
+	igt_assert(!(pipe_id_flag & ~DRM_VBLANK_HIGH_CRTC_MASK));

If crtc_idx is 1 (pipe B), kmstest_get_vbl_flag(crtc_idx) = DRM_VBLANK_SECONDARY = 0x20000000

And the assertion fails as !(0x20000000 & ~0x0000003e) = 0.

Quoted 8 lines
Quoted 1 line

Should kmstest_get_vbl_flag() always return pipe_id << 1?

From re-reading the pipe id parsing code in drm_irq.c drm_wait_vblank() it would seem to me that the assertion is incorrect, specifically for the case of _DRM_VBLANK_SECONDARY.

Supplying the flag _DRM_VBLANK_SECONDARY and setting the high crtc id field to 1 fboth seem to be valid ways to communicate the same thing.

Indeed.

Quoted 11 lines
	flags = vblwait->request.type & _DRM_VBLANK_FLAGS_MASK;
	high_pipe = (vblwait->request.type &_DRM_VBLANK_HIGH_CRTC_MASK);
	if (high_pipe)
		pipe = high_pipe >> _DRM_VBLANK_HIGH_CRTC_SHIFT;
	else
		pipe = flags & _DRM_VBLANK_SECONDARY ? 1 : 0;

Maybe adding correct and more thorough assertion to kmstest_get_vbl_flag() and removing the failing assertion is the way forward.

Sounds reasonable to me, though I assume that the assertion had/has some point in determining that you have a valid crtc_idx.

Quoted 15 lines
Quoted 11 lines
Quoted 11 lines
tests/kms_flip.c +1 -2
-	wait_vbl.request.type = crtc_idx_mask;
+	wait_vbl.request.type = pipe_id_flag;
 	if (flags & TEST_VBLANK_ABSOLUTE)
 		wait_vbl.request.type |= DRM_VBLANK_ABSOLUTE;
 	else
--

2.7.4

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

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

#11 lore raw
In reply to [email protected]

Hi,

I'm getting mixed results w/ this series applied. The kernel seems to trip in different ways either from suspend or on reload (running the BAT suite on a HSW machine), and though it seems totally unrelated I can't reproduce the same behaviour without the series applied. Haven't managed to get some valid output as the machine dead-locks/freezes, so please be patient until I try to isolate the problem. It could be I stumbled something totally unrelated.

Quoted 46 lines · On Mon, May 16, 2016 at 09:38:25AM -0400, [email protected] wrote:

From: Robert Foss <[email protected]>

In addition to the changes made in v4, danvet suggested that the argument for the (newly renamed) helper function kmstest_get_vbl_flag should be changed.

I investigated it and found that all of the places where the helper function currently is used, would require more of a rework with changed arguments than with unchanged arguments, so I decided to not change the arguments.

Changes since v1:
- kms_vblank: Removed un-used members of data_t struct.
- Rename plane_counter to n_planes.
- Removed un-needed handling CURSOR plane location.
- Added names for additional planes in update kmstest_plane_name.

Changes since v2:
- Rebased onto trunk which already contains kms_flip_event_leak changes.

Changes since v3:
- kms_vblank.c: Removed drm_fd from data_t, to decrease the number of unrelated changes in the patch.
- kms_vblank.c: Changed "int valid_tests" to "unsigned valid_tests".
- kms_vblank.c: Changed crtc_id_flag/crtc_id_to_flag to pipe_id_flag/pipe_id_to_flag.
- kms_vblank.c: Changed to "vbl.request.type |= crtc_id_flag;".
- kms_vblank.c: Moved crtc_id_to_flag to igt_kms.c.
- kms_vblank.c: Renamed crtc_id_to_flag to pipe_id_to_vbl_flag.
- kms_flip.c: Changed test busy_mode boolean into mode_busy bitfield in struct data.
- igt_kms.h: Moved IGT_PLANE_CURSOR position comment.

Changes since v4:
- kms_vblank.c: Removed whitespace at end of file.
- igt_kms: Changed name of helper function to kmstest_get_vbl_flag
- igt_kms: Changed kmstest_get_vbl_flag to not set DRM_VBLANK_RELATIVE flag

Robert Foss (7):
lib/igt_kms: Add support for up to 10 planes. lib/igt_kms: Fix plane counting in igt_display_init. lib/igt_kms: Switch to verbose assert.
lib/igt_kms: Added pipe_id_to_vbl_flag() to igt_kms.
kms_vblank: Switch from using crtc0 statically to explicitly setting
mode.
igt_kms: Change igt_wait_for_vblank to use helper function.
kms_flip: Change __wait_for_vblank to use helper function.

Quoted 9 lines
 lib/igt_kms.c      |  36 ++++++++++--
 lib/igt_kms.h      |   9 ++-
 tests/kms_flip.c   |   8 +--
 tests/kms_vblank.c | 161 ++++++++++++++++++++++++++++++++++++++++++++---------
 4 files changed, 179 insertions(+), 35 deletions(-)
Signature · 1 line
2.7.4

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

#12 lore raw

Hey,

Thanks for being thorough, give me a shout if there is anything specific I can help you look into.

Rob.

Quoted 11 lines · Marius Vlad

Hi,

I'm getting mixed results w/ this series applied. The kernel seems to trip in different ways either from suspend or on reload (running the BAT suite on a HSW machine), and though it seems totally unrelated I can't reproduce the same behaviour without the series applied. Haven't managed to get some valid output as the machine dead-locks/freezes, so please be patient until I try to isolate the problem. It could be I stumbled something totally unrelated.

Quoted 56 lines
Quoted 55 lines · On Mon, May 16, 2016 at 09:38:25AM -0400, [email protected] wrote:

From: Robert Foss <[email protected]>

In addition to the changes made in v4, danvet suggested that the argument for the (newly renamed) helper function kmstest_get_vbl_flag should be changed.

I investigated it and found that all of the places where the helper function currently is used, would require more of a rework with changed arguments than with unchanged arguments, so I decided to not change the arguments.

Changes since v1:
- kms_vblank: Removed un-used members of data_t struct.
- Rename plane_counter to n_planes.
- Removed un-needed handling CURSOR plane location.
- Added names for additional planes in update kmstest_plane_name.

Changes since v2:
- Rebased onto trunk which already contains kms_flip_event_leak changes.

Changes since v3:
- kms_vblank.c: Removed drm_fd from data_t, to decrease the number of unrelated changes in the patch.
- kms_vblank.c: Changed "int valid_tests" to "unsigned valid_tests".
- kms_vblank.c: Changed crtc_id_flag/crtc_id_to_flag to pipe_id_flag/pipe_id_to_flag.
- kms_vblank.c: Changed to "vbl.request.type |= crtc_id_flag;".
- kms_vblank.c: Moved crtc_id_to_flag to igt_kms.c.
- kms_vblank.c: Renamed crtc_id_to_flag to pipe_id_to_vbl_flag.
- kms_flip.c: Changed test busy_mode boolean into mode_busy bitfield in struct data.
- igt_kms.h: Moved IGT_PLANE_CURSOR position comment.

Changes since v4:
- kms_vblank.c: Removed whitespace at end of file.
- igt_kms: Changed name of helper function to kmstest_get_vbl_flag
- igt_kms: Changed kmstest_get_vbl_flag to not set DRM_VBLANK_RELATIVE flag

Robert Foss (7):
lib/igt_kms: Add support for up to 10 planes. lib/igt_kms: Fix plane counting in igt_display_init. lib/igt_kms: Switch to verbose assert.
lib/igt_kms: Added pipe_id_to_vbl_flag() to igt_kms.
kms_vblank: Switch from using crtc0 statically to explicitly setting
mode.
igt_kms: Change igt_wait_for_vblank to use helper function.
kms_flip: Change __wait_for_vblank to use helper function.

  lib/igt_kms.c      |  36 ++++++++++--
  lib/igt_kms.h      |   9 ++-
  tests/kms_flip.c   |   8 +--
  tests/kms_vblank.c | 161 ++++++++++++++++++++++++++++++++++++++++++++---------
  4 files changed, 179 insertions(+), 35 deletions(-)

--
2.7.4

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