Skip to contentSkip to conversation

igt: Add support for testing writeback connectors

PATCH Superseded Rodrigo Siqueira 4 participants 36 messages Jun 13, 2019 – Jul 19, 2019

Subsystems (2)

v6 of 9 Superseded Latest is v9 →

lore
Share & tools
lore
#0 lore raw

v6 cover igt: Add support for testing writeback connectors

DRM already have writeback connectors support, which is a way to expose in DRM the hardware functionality from display engines that allows writing back into memory the result of the DE's composition of supported planes.

These days, I'm working to add the writeback support into the vkms[1], and I used the kms_writeback to validate my implementation. As a result, I had to rebase the v5 version of Liviu's series, and I also fixed a small issue in the first path (lib/igt_kms: Add writeback support).

Patches have been originally implemented by Brian, Liviu has done the v3 and v4 updates to them. I’m just sending a v6.

1. https://patchwork.freedesktop.org/series/61738/

Brian Starkey (6):
lib/igt_kms: Add writeback support
kms_writeback: Add initial writeback tests
lib: Add function to hash a framebuffer
kms_writeback: Add writeback-check-output lib/igt_kms: Add igt_output_clone_pipe for cloning
kms_writeback: Add tests using a cloned output

 lib/igt_fb.c           |  66 ++++++
 lib/igt_fb.h           |   3 +
 lib/igt_kms.c          | 157 +++++++++----
 lib/igt_kms.h          |  10 +
 tests/Makefile.sources |   1 +
 tests/kms_writeback.c  | 492 +++++++++++++++++++++++++++++++++++++++++
 tests/meson.build      |   1 +
 7 files changed, 692 insertions(+), 38 deletions(-)
 create mode 100644 tests/kms_writeback.c
Signature · 6 lines
2.21.0


-- 
Rodrigo Siqueira
https://siqueira.tech

_______________________________________________
igt-dev mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/igt-dev

#1 lore raw

PATCH v6 1/6 lib/igt_kms: Add writeback support

Add support in igt_kms for writeback connectors, with the ability to attach framebuffers.

v5: Rebase and add DRM_CLIENT_CAP_WRITEBACK_CONNECTORS before drmModeGetResources()

[rebased and updated to the latest igt style]
---
 lib/igt_kms.c | 57 +++++++++++++++++++++++++++++++++++++++++++++++++++
 lib/igt_kms.h |  6 ++++++
 2 files changed, 63 insertions(+)
lib/igt_kms.c +57 -0
index da188a39..140db346 100644
@@ -325,6 +325,9 @@ const char * const igt_connector_prop_names[IGT_NUM_CONNECTOR_PROPS] = {
 	[IGT_CONNECTOR_BROADCAST_RGB] = "Broadcast RGB",
 	[IGT_CONNECTOR_CONTENT_PROTECTION] = "Content Protection",
 	[IGT_CONNECTOR_VRR_CAPABLE] = "vrr_capable",
+	[IGT_CONNECTOR_WRITEBACK_PIXEL_FORMATS] = "WRITEBACK_PIXEL_FORMATS",
+	[IGT_CONNECTOR_WRITEBACK_FB_ID] = "WRITEBACK_FB_ID",
+	[IGT_CONNECTOR_WRITEBACK_OUT_FENCE_PTR] = "WRITEBACK_OUT_FENCE_PTR",
 };
 
 /*
@@ -557,6 +560,7 @@ static const struct type_name connector_type_names[] = {
 	{ DRM_MODE_CONNECTOR_VIRTUAL, "Virtual" },
 	{ DRM_MODE_CONNECTOR_DSI, "DSI" },
 	{ DRM_MODE_CONNECTOR_DPI, "DPI" },
+	{ DRM_MODE_CONNECTOR_WRITEBACK, "Writeback" },
 	{}
 };
 
@@ -1889,6 +1893,12 @@ static void igt_output_reset(igt_output_t *output)
 	if (igt_output_has_prop(output, IGT_CONNECTOR_BROADCAST_RGB))
 		igt_output_set_prop_value(output, IGT_CONNECTOR_BROADCAST_RGB,
 					  BROADCAST_RGB_FULL);
+	if (igt_output_has_prop(output, IGT_CONNECTOR_WRITEBACK_FB_ID))
+		igt_output_set_prop_value(output, IGT_CONNECTOR_WRITEBACK_FB_ID, 0);
+	if (igt_output_has_prop(output, IGT_CONNECTOR_WRITEBACK_OUT_FENCE_PTR)) {
+		igt_output_clear_prop_changed(output, IGT_CONNECTOR_WRITEBACK_OUT_FENCE_PTR);
+		output->writeback_out_fence_fd = -1;
+	}
 }
 
 /**
@@ -1901,6 +1911,8 @@ static void igt_output_reset(igt_output_t *output)
  * For outputs:
  * - %IGT_CONNECTOR_CRTC_ID
  * - %IGT_CONNECTOR_BROADCAST_RGB (if applicable)
+ * - %IGT_CONNECTOR_WRITEBACK_FB_ID
+ * - %IGT_CONNECTOR_WRITEBACK_OUT_FENCE_PTR
  * - igt_output_override_mode() to default.
  *
  * For pipes:
@@ -1970,6 +1982,8 @@ void igt_display_require(igt_display_t *display, int drm_fd)
 
 	display->drm_fd = drm_fd;
 
+	drmSetClientCap(drm_fd, DRM_CLIENT_CAP_WRITEBACK_CONNECTORS, 1);
+
 	resources = drmModeGetResources(display->drm_fd);
 	if (!resources)
 		goto out;
@@ -2263,6 +2277,11 @@ static void igt_output_fini(igt_output_t *output)
 	kmstest_free_connector_config(&output->config);
 	free(output->name);
 	output->name = NULL;
+
+	if (output->writeback_out_fence_fd != -1) {
+		close(output->writeback_out_fence_fd);
+		output->writeback_out_fence_fd = -1;
+	}
 }
 
 /**
@@ -3325,6 +3344,11 @@ static void igt_atomic_prepare_connector_commit(igt_output_t *output, drmModeAto
 					  output->props[i],
 					  output->values[i]));
 	}
+
+	if (output->writeback_out_fence_fd != -1) {
+		close(output->writeback_out_fence_fd);
+		output->writeback_out_fence_fd = -1;
+	}
 }
 
 /*
@@ -3447,6 +3471,16 @@ display_commit_changed(igt_display_t *display, enum igt_commit_style s)
 		else
 			/* no modeset in universal commit, no change to crtc. */
 			output->changed &= 1 << IGT_CONNECTOR_CRTC_ID;
+
+		if (s == COMMIT_ATOMIC) {
+			if (igt_output_is_prop_changed(output, IGT_CONNECTOR_WRITEBACK_OUT_FENCE_PTR))
+				igt_assert(output->writeback_out_fence_fd >= 0);
+
+			output->values[IGT_CONNECTOR_WRITEBACK_OUT_FENCE_PTR] = 0;
+			output->values[IGT_CONNECTOR_WRITEBACK_FB_ID] = 0;
+			igt_output_clear_prop_changed(output, IGT_CONNECTOR_WRITEBACK_FB_ID);
+			igt_output_clear_prop_changed(output, IGT_CONNECTOR_WRITEBACK_OUT_FENCE_PTR);
+		}
 	}
 
 	if (display->first_commit) {
@@ -4119,6 +4153,29 @@ void igt_pipe_request_out_fence(igt_pipe_t *pipe)
 	igt_pipe_obj_set_prop_value(pipe, IGT_CRTC_OUT_FENCE_PTR, (ptrdiff_t)&pipe->out_fence_fd);
 }
 
+/**
+ * igt_output_set_writeback_fb:
+ * @output: Target output
+ * @fb: Target framebuffer
+ *
+ * This function sets the given @fb to be used as the target framebuffer for the
+ * writeback engine at the next atomic commit. It will also request a writeback
+ * out fence that will contain the fd number of the out fence created by KMS if
+ * the given @fb is valid.
+ */
+void igt_output_set_writeback_fb(igt_output_t *output, struct igt_fb *fb)
+{
+	igt_display_t *display = output->display;
+
+	LOG(display, "%s: output_set_writeback_fb(%d)\n", output->name, fb ? fb->fb_id : 0);
+
+	igt_output_set_prop_value(output, IGT_CONNECTOR_WRITEBACK_FB_ID, fb ? fb->fb_id : 0);
+	/* only request a writeback out fence if the framebuffer is valid */
+	if (fb)
+		igt_output_set_prop_value(output, IGT_CONNECTOR_WRITEBACK_OUT_FENCE_PTR,
+					  (ptrdiff_t)&output->writeback_out_fence_fd);
+}
+
 /**
  * igt_wait_for_vblank_count:
  * @drm_fd: A drm file descriptor
lib/igt_kms.h +6 -0
index a448a003..cacc6b90 100644
@@ -123,6 +123,9 @@ enum igt_atomic_connector_properties {
        IGT_CONNECTOR_BROADCAST_RGB,
        IGT_CONNECTOR_CONTENT_PROTECTION,
        IGT_CONNECTOR_VRR_CAPABLE,
+       IGT_CONNECTOR_WRITEBACK_PIXEL_FORMATS,
+       IGT_CONNECTOR_WRITEBACK_FB_ID,
+       IGT_CONNECTOR_WRITEBACK_OUT_FENCE_PTR,
        IGT_NUM_CONNECTOR_PROPS
 };
 
@@ -364,6 +367,8 @@ typedef struct {
 	bool use_override_mode;
 	drmModeModeInfo override_mode;
 
+	int32_t writeback_out_fence_fd;
+
 	/* bitmask of changed properties */
 	uint64_t changed;
 
@@ -414,6 +419,7 @@ igt_plane_t *igt_output_get_plane_type_index(igt_output_t *output,
 igt_output_t *igt_output_from_connector(igt_display_t *display,
     drmModeConnector *connector);
 const drmModeModeInfo *igt_std_1024_mode_get(void);
+void igt_output_set_writeback_fb(igt_output_t *output, struct igt_fb *fb);
 
 igt_plane_t *igt_pipe_get_plane_type(igt_pipe_t *pipe, int plane_type);
 int igt_pipe_count_plane_type(igt_pipe_t *pipe, int plane_type);
Signature · 6 lines
2.21.0


-- 
Rodrigo Siqueira
https://siqueira.tech

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

#9 lore raw

PATCH v6 2/6 kms_writeback: Add initial writeback tests

In reply to Rodrigo Siqueira

Add tests for the WRITEBACK_PIXEL_FORMATS, WRITEBACK_OUT_FENCE_PTR and WRITEBACK_FB_ID properties on writeback connectors, ensuring their behaviour is correct.

[rebased and updated do_writeback_test() function to address feedback]
---
 tests/Makefile.sources |   1 +
 tests/kms_writeback.c  | 314 +++++++++++++++++++++++++++++++++++++++++
 tests/meson.build      |   1 +
 3 files changed, 316 insertions(+)
 create mode 100644 tests/kms_writeback.c
tests/Makefile.sources +1 -0
index 027ed82f..03cc8efa 100644
@@ -77,6 +77,7 @@ TESTS_progs = \
 	kms_universal_plane \
 	kms_vblank \
 	kms_vrr \
+	kms_writeback \
 	meta_test \
 	perf \
 	perf_pmu \
tests/kms_writeback.c new file +314 -0

314 lines not on this page · raw message

tests/meson.build +1 -0
index f168fbba..9c77cfcd 100644
@@ -63,6 +63,7 @@ test_progs = [
 	'kms_universal_plane',
 	'kms_vblank',
 	'kms_vrr',
+	'kms_writeback',
 	'meta_test',
 	'panfrost_get_param',
 	'panfrost_gem_new',
Signature · 1 line
2.21.0

_______________________________________________
igt-dev mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/igt-dev

#22 lore raw

PATCH v6 3/6 lib: Add function to hash a framebuffer

In reply to Rodrigo Siqueira

To use writeback buffers as a CRC source, we need to be able to hash them. Implement a simple FVA-1a hashing routine for this purpose.

Doing a bytewise hash on the framebuffer directly can be very slow if the memory is noncached. By making a copy of each line in the FB first (which can take advantage of word-access speedup), we can do the hash on a cached copy, which is much faster (10x speedup on my platform).

v6: use igt_memcpy_from_wc() instead of plain memcpy, as suggested by
Chris Wilson

[rebased and updated to the most recent API]
---
 lib/igt_fb.c | 66 ++++++++++++++++++++++++++++++++++++++++++++++++++++
 lib/igt_fb.h |  3 +++
 2 files changed, 69 insertions(+)
lib/igt_fb.c +66 -0
index 9d4f905e..d07dae39 100644
@@ -3256,6 +3256,72 @@ bool igt_fb_supported_format(uint32_t drm_format)
 	return false;
 }
 
+/*
+ * This implements the FNV-1a hashing algorithm instead of CRC, for
+ * simplicity
+ * http://www.isthe.com/chongo/tech/comp/fnv/index.html
+ *
+ * hash = offset_basis
+ * for each octet_of_data to be hashed
+ *         hash = hash xor octet_of_data
+ *         hash = hash * FNV_prime
+ * return hash
+ *
+ * 32 bit offset_basis = 2166136261
+ * 32 bit FNV_prime = 224 + 28 + 0x93 = 16777619
+ */
+int igt_fb_get_crc(struct igt_fb *fb, igt_crc_t *crc)
+{
+#define FNV1a_OFFSET_BIAS 2166136261
+#define FNV1a_PRIME 16777619
+	uint32_t hash;
+	void *map;
+	char *ptr, *line = NULL;
+	int x, y, cpp = igt_drm_format_to_bpp(fb->drm_format) / 8;
+	uint32_t stride = calc_plane_stride(fb, 0);
+
+	if (fb->is_dumb)
+		map = kmstest_dumb_map_buffer(fb->fd, fb->gem_handle, fb->size,
+					      PROT_READ);
+	else
+		map = gem_mmap__gtt(fb->fd, fb->gem_handle, fb->size,
+				    PROT_READ);
+	ptr = map;
+
+	/*
+	 * Framebuffers are often uncached, which can make byte-wise accesses
+	 * very slow. We copy each line of the FB into a local buffer to speed
+	 * up the hashing.
+	 */
+	line = malloc(stride);
+	if (!line) {
+		munmap(map, fb->size);
+		return -ENOMEM;
+	}
+
+	hash = FNV1a_OFFSET_BIAS;
+
+	for (y = 0; y < fb->height; y++, ptr += stride) {
+
+		igt_memcpy_from_wc(line, ptr, stride);
+
+		for (x = 0; x < fb->width * cpp; x++) {
+			hash ^= line[x];
+			hash *= FNV1a_PRIME;
+		}
+	}
+
+	crc->n_words = 1;
+	crc->crc[0] = hash;
+
+	free(line);
+	munmap(map, fb->size);
+
+	return 0;
+#undef FNV1a_OFFSET_BIAS
+#undef FNV1a_PRIME
+}
+
 /**
  * igt_format_is_yuv:
  * @drm_format: drm fourcc
lib/igt_fb.h +3 -0
index adefebe1..a2741c05 100644
@@ -37,6 +37,7 @@
 #include <i915_drm.h>
 
 #include "igt_color_encoding.h"
+#include "igt_debugfs.h"
 
 /*
  * Internal format to denote a buffer compatible with pixman's
@@ -194,5 +195,7 @@ int igt_format_plane_bpp(uint32_t drm_format, int plane);
 void igt_format_array_fill(uint32_t **formats_array, unsigned int *count,
 			   bool allow_yuv);
 
+int igt_fb_get_crc(struct igt_fb *fb, igt_crc_t *crc);
+
 #endif /* __IGT_FB_H__ */
 
Signature · 1 line
2.21.0

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

#28 lore raw

PATCH v6 4/6 kms_writeback: Add writeback-check-output

In reply to Rodrigo Siqueira

Add a test which makes commits using the writeback connector, and checks the output buffer hash to make sure it is/isn't written as appropriate.

---
 tests/kms_writeback.c | 124 ++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 124 insertions(+)
tests/kms_writeback.c +124 -0
index 66ef48a6..0f20dadd 100644
@@ -30,6 +30,7 @@
 #include "igt.h"
 #include "igt_core.h"
 #include "igt_fb.h"
+#include "sw_sync.h"
 
 static drmModePropertyBlobRes *get_writeback_formats_blob(igt_output_t *output)
 {
@@ -221,6 +222,116 @@ static void writeback_fb_id(igt_output_t *output, igt_fb_t *valid_fb, igt_fb_t *
 	igt_assert(ret == 0);
 }
 
+static void fill_fb(igt_fb_t *fb, double color[3])
+{
+	cairo_t *cr = igt_get_cairo_ctx(fb->fd, fb);
+	igt_assert(cr);
+
+	igt_paint_color(cr, 0, 0, fb->width, fb->height,
+			color[0], color[1], color[2]);
+}
+
+static void get_and_wait_out_fence(igt_output_t *output)
+{
+	int ret;
+
+	igt_assert(output->writeback_out_fence_fd >= 0);
+
+	ret = sync_fence_wait(output->writeback_out_fence_fd, 1000);
+	igt_assert(ret == 0);
+	close(output->writeback_out_fence_fd);
+	output->writeback_out_fence_fd = -1;
+}
+
+static void writeback_sequence(igt_output_t *output, igt_plane_t *plane,
+				igt_fb_t *in_fb, igt_fb_t *out_fbs[], int n_commits)
+{
+	int i, color_idx = 0;
+	double in_fb_colors[2][3] = {
+		{ 1.0, 0.0, 0.0 },
+		{ 0.0, 1.0, 0.0 },
+	};
+	double clear_color[3] = { 1.0, 1.0, 1.0 };
+	igt_crc_t cleared_crc, out_expected;
+
+	for (i = 0; i < n_commits; i++, color_idx++) {
+		/* Change the input color each time */
+		fill_fb(in_fb, in_fb_colors[color_idx % 2]);
+
+		if (out_fbs[i]) {
+			igt_crc_t out_before;
+
+			/* Get the expected CRC */
+			fill_fb(out_fbs[i], in_fb_colors[color_idx % 2]);
+			igt_fb_get_crc(out_fbs[i], &out_expected);
+
+			fill_fb(out_fbs[i], clear_color);
+			if (i == 0)
+				igt_fb_get_crc(out_fbs[i], &cleared_crc);
+			igt_fb_get_crc(out_fbs[i], &out_before);
+			igt_assert_crc_equal(&cleared_crc, &out_before);
+		}
+
+		/* Commit */
+		igt_plane_set_fb(plane, in_fb);
+		igt_output_set_writeback_fb(output, out_fbs[i]);
+
+		igt_display_commit_atomic(output->display,
+					  DRM_MODE_ATOMIC_ALLOW_MODESET,
+					  NULL);
+		if (out_fbs[i])
+			get_and_wait_out_fence(output);
+
+		/* Make sure the old output buffer is untouched */
+		if (i > 0 && out_fbs[i - 1] && (out_fbs[i] != out_fbs[i - 1])) {
+			igt_crc_t out_prev;
+			igt_fb_get_crc(out_fbs[i - 1], &out_prev);
+			igt_assert_crc_equal(&cleared_crc, &out_prev);
+		}
+
+		/* Make sure this output buffer is written */
+		if (out_fbs[i]) {
+			igt_crc_t out_after;
+			igt_fb_get_crc(out_fbs[i], &out_after);
+			igt_assert_crc_equal(&out_expected, &out_after);
+
+			/* And clear it, for the next time */
+			fill_fb(out_fbs[i], clear_color);
+		}
+	}
+}
+
+static void writeback_check_output(igt_output_t *output, igt_plane_t *plane,
+				   igt_fb_t *input_fb, igt_fb_t *output_fb)
+{
+	igt_fb_t *out_fbs[2] = { 0 };
+	igt_fb_t second_out_fb;
+	int ret;
+
+	/* One commit, with a writeback. */
+	writeback_sequence(output, plane, input_fb, &output_fb, 1);
+
+	/* Two commits, the second with no writeback */
+	out_fbs[0] = output_fb;
+	writeback_sequence(output, plane, input_fb, out_fbs, 2);
+
+	/* Two commits, both with writeback */
+	out_fbs[1] = output_fb;
+	writeback_sequence(output, plane, input_fb, out_fbs, 2);
+
+	ret = igt_create_fb(output_fb->fd, output_fb->width, output_fb->height,
+			    DRM_FORMAT_XRGB8888,
+			    igt_fb_mod_to_tiling(0),
+			    &second_out_fb);
+	igt_require(ret > 0);
+
+	/* Two commits, with different writeback buffers */
+	out_fbs[1] = &second_out_fb;
+	writeback_sequence(output, plane, input_fb, out_fbs, 2);
+
+	igt_remove_fb(output_fb->fd, &second_out_fb);
+}
+
 igt_main
 {
 	igt_display_t display;
@@ -307,6 +418,19 @@ igt_main
 		igt_remove_fb(display.drm_fd, &output_fb);
 	}
 
+	igt_subtest("writeback-check-output") {
+		igt_fb_t output_fb;
+		ret = igt_create_fb(display.drm_fd, mode.hdisplay, mode.vdisplay,
+				    DRM_FORMAT_XRGB8888,
+				    igt_fb_mod_to_tiling(0),
+				    &output_fb);
+		igt_require(ret > 0);
+
+		writeback_check_output(output, plane, &input_fb, &output_fb);
+
+		igt_remove_fb(display.drm_fd, &output_fb);
+	}
+
 	igt_fixture {
 		igt_remove_fb(display.drm_fd, &input_fb);
 		igt_display_fini(&display);
Signature · 1 line
2.21.0

_______________________________________________
igt-dev mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/igt-dev

#30 lore raw

PATCH v6 5/6 lib/igt_kms: Add igt_output_clone_pipe for cloning

In reply to Rodrigo Siqueira

An output can be added as a clone of any other output(s) attached to a pipe using igt_output_clone_pipe()

v5: Drop field out_fence_requested from struct igt_pipe (Brian Starkey)

---
 lib/igt_kms.c | 100 +++++++++++++++++++++++++++++++-------------------
 lib/igt_kms.h |   4 ++
 2 files changed, 66 insertions(+), 38 deletions(-)
lib/igt_kms.c +62 -38
index 140db346..b85a0404 100644
@@ -1765,6 +1765,17 @@ static void igt_display_log_shift(igt_display_t *display, int shift)
 	igt_assert(display->log_shift >= 0);
 }
 
+static int igt_output_idx(igt_output_t *output)
+{
+	int i;
+
+	for (i = 0; i < output->display->n_outputs; i++)
+		if (&output->display->outputs[i] == output)
+			return i;
+
+	return -1;
+}
+
 static void igt_output_refresh(igt_output_t *output)
 {
 	igt_display_t *display = output->display;
@@ -2317,42 +2328,6 @@ void igt_display_fini(igt_display_t *display)
 	display->planes = NULL;
 }
 
-static void igt_display_refresh(igt_display_t *display)
-{
-	igt_output_t *output;
-	int i;
-
-	unsigned long pipes_in_use = 0;
-
-       /* Check that two outputs aren't trying to use the same pipe */
-	for (i = 0; i < display->n_outputs; i++) {
-		output = &display->outputs[i];
-
-		if (output->pending_pipe != PIPE_NONE) {
-			if (pipes_in_use & (1 << output->pending_pipe))
-				goto report_dup;
-
-			pipes_in_use |= 1 << output->pending_pipe;
-		}
-
-		if (output->force_reprobe)
-			igt_output_refresh(output);
-	}
-
-	return;
-
-report_dup:
-	for (; i > 0; i--) {
-		igt_output_t *b = &display->outputs[i - 1];
-
-		igt_assert_f(output->pending_pipe !=
-			     b->pending_pipe,
-			     "%s and %s are both trying to use pipe %s\n",
-			     igt_output_name(output), igt_output_name(b),
-			     kmstest_pipe_name(output->pending_pipe));
-	}
-}
-
 static igt_pipe_t *igt_output_get_driving_pipe(igt_output_t *output)
 {
 	igt_display_t *display = output->display;
@@ -2376,6 +2351,40 @@ static igt_pipe_t *igt_output_get_driving_pipe(igt_output_t *output)
 	return &display->pipes[pipe];
 }
 
+static void igt_display_refresh(igt_display_t *display)
+{
+	igt_output_t *output;
+	igt_pipe_t *pipe;
+	int i;
+
+	unsigned long pipes_in_use = 0;
+	unsigned long pending_crtc_idx_mask;
+
+	/* Check that outputs and pipes agree wrt. cloning */
+	for (i = 0; i < display->n_outputs; i++) {
+		output = &display->outputs[i];
+		pending_crtc_idx_mask = 1 << output->pending_pipe;
+
+		pipe = igt_output_get_driving_pipe(output);
+		if (pipe) {
+			igt_assert_f(pipe->outputs & (1 << igt_output_idx(output)),
+				     "Output %s not expected to be using pipe %s\n",
+				     igt_output_name(output),
+				     kmstest_pipe_name(pipe->pipe));
+
+			if (pipes_in_use & pending_crtc_idx_mask)
+				LOG(display, "Output %s clones pipe %s\n",
+				    igt_output_name(output),
+				    kmstest_pipe_name(pipe->pipe));
+		}
+
+		pipes_in_use |= pending_crtc_idx_mask;
+
+		if (output->force_reprobe)
+			igt_output_refresh(output);
+	}
+}
+
 static igt_plane_t *igt_pipe_get_plane(igt_pipe_t *pipe, int plane_idx)
 {
 	igt_require_f(plane_idx >= 0 && plane_idx < pipe->n_planes,
@@ -3766,6 +3775,7 @@ void igt_output_override_mode(igt_output_t *output, const drmModeModeInfo *mode)
 	output->use_override_mode = !!mode;
 
 	if (pipe) {
+		igt_debug("overriding pipe mode in %s way\n", output->display->is_atomic ? "atomic" : "legacy");
 		if (output->display->is_atomic)
 			igt_pipe_obj_replace_prop_blob(pipe, IGT_CRTC_MODE_ID, igt_output_get_mode(output), sizeof(*mode));
 		else
@@ -3773,6 +3783,16 @@ void igt_output_override_mode(igt_output_t *output, const drmModeModeInfo *mode)
 	}
 }
 
+void igt_output_clone_pipe(igt_output_t *output, enum pipe pipe)
+{
+	igt_display_t *display = output->display;
+	uint32_t current_clones = display->pipes[pipe].outputs;
+
+	igt_output_set_pipe(output, pipe);
+
+	display->pipes[pipe].outputs |= current_clones;
+}
+
 /*
  * igt_output_set_pipe:
  * @output: Target output for which the pipe is being set to
@@ -3789,11 +3809,15 @@ void igt_output_set_pipe(igt_output_t *output, enum pipe pipe)
 
 	igt_assert(output->name);
 
-	if (output->pending_pipe != PIPE_NONE)
+	if (output->pending_pipe != PIPE_NONE) {
 		old_pipe = igt_output_get_driving_pipe(output);
+		old_pipe->outputs &= ~(1 << igt_output_idx(output));
+	}
 
-	if (pipe != PIPE_NONE)
+	if (pipe != PIPE_NONE) {
 		pipe_obj = &display->pipes[pipe];
+		pipe_obj->outputs = (1 << igt_output_idx(output));
+	}
 
 	LOG(display, "%s: set_pipe(%s)\n", igt_output_name(output),
 	    kmstest_pipe_name(pipe));
lib/igt_kms.h +4 -0
index cacc6b90..676839bb 100644
@@ -354,6 +354,8 @@ struct igt_pipe {
 	uint32_t crtc_id;
 
 	int32_t out_fence_fd;
+
+	uint32_t outputs;
 };
 
 typedef struct {
@@ -411,6 +413,8 @@ const char *igt_output_name(igt_output_t *output);
 drmModeModeInfo *igt_output_get_mode(igt_output_t *output);
 void igt_output_override_mode(igt_output_t *output, const drmModeModeInfo *mode);
 void igt_output_set_pipe(igt_output_t *output, enum pipe pipe);
+void igt_output_clone_pipe(igt_output_t *output, enum pipe pipe);
+
 igt_plane_t *igt_output_get_plane(igt_output_t *output, int plane_idx);
 igt_plane_t *igt_output_get_plane_type(igt_output_t *output, int plane_type);
 int igt_output_count_plane_type(igt_output_t *output, int plane_type);
Signature · 1 line
2.21.0

_______________________________________________
igt-dev mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/igt-dev

#34 lore raw

PATCH v6 6/6 kms_writeback: Add tests using a cloned output

In reply to Rodrigo Siqueira

Update the connector search to also optionally attempt to find a non-writeback connector to clone to.

Add a subtest which is the same as writeback-check-output, but also clones to the second connector.

[rebased and addressed comments by Maarten]
---
 tests/kms_writeback.c | 64 +++++++++++++++++++++++++++++++++++++++----
 1 file changed, 59 insertions(+), 5 deletions(-)
tests/kms_writeback.c +59 -5
index 0f20dadd..ae536bbf 100644
@@ -51,7 +51,8 @@ static drmModePropertyBlobRes *get_writeback_formats_blob(igt_output_t *output)
 	return blob;
 }
 
-static bool check_writeback_config(igt_display_t *display, igt_output_t *output)
+static bool check_writeback_config(igt_display_t *display, igt_output_t *output,
+				   int pipe, igt_output_t **clone)
 {
 	igt_fb_t input_fb, output_fb;
 	igt_plane_t *plane;
@@ -91,6 +92,30 @@ static bool check_writeback_config(igt_display_t *display, igt_output_t *output)
 
 	ret = igt_display_try_commit_atomic(display, DRM_MODE_ATOMIC_TEST_ONLY |
 					    DRM_MODE_ATOMIC_ALLOW_MODESET, NULL);
+	if (!ret && clone) {
+		/* Try and find a clone */
+		int i, newret;
+		*clone = NULL;
+
+		for (i = 0; i < display->n_outputs; i++) {
+			igt_output_t *second_output = &display->outputs[i];
+			if (output != second_output &&
+			    igt_pipe_connector_valid(pipe, second_output)) {
+
+				igt_output_clone_pipe(second_output, pipe);
+				igt_output_override_mode(output, &override_mode);
+				newret = igt_display_try_commit_atomic(display, DRM_MODE_ATOMIC_TEST_ONLY |
+								       DRM_MODE_ATOMIC_ALLOW_MODESET, NULL);
+				igt_output_set_pipe(second_output, PIPE_NONE);
+				igt_debug("try_commit_atomic clone returned %d\n", newret);
+				if (!newret) {
+					*clone = second_output;
+					igt_debug("Selected clone %s\n", (*clone)->name);
+					break;
+				}
+			}
+		}
+	}
 	igt_plane_set_fb(plane, NULL);
 	igt_remove_fb(display->drm_fd, &input_fb);
 	igt_remove_fb(display->drm_fd, &output_fb);
@@ -98,7 +123,8 @@ static bool check_writeback_config(igt_display_t *display, igt_output_t *output)
 	return !ret;
 }
 
-static igt_output_t *kms_writeback_get_output(igt_display_t *display)
+static igt_output_t *kms_writeback_get_output(igt_display_t *display, enum pipe *pipe,
+					      igt_output_t **clone)
 {
 	int i;
 
@@ -114,10 +140,16 @@ static igt_output_t *kms_writeback_get_output(igt_display_t *display)
 		for (j = 0; j < igt_display_get_n_pipes(display); j++) {
 			igt_output_set_pipe(output, j);
 
-			if (check_writeback_config(display, output)) {
+			if (check_writeback_config(display, output, j, clone)) {
 				igt_debug("Using connector %u:%s on pipe %d\n",
 					  output->config.connector->connector_id,
 					  output->name, j);
+				if (clone && *clone)
+					igt_debug("Cloning to connector %u:%s\n",
+						  (*clone)->config.connector->connector_id,
+						  (*clone)->name);
+				if (pipe)
+					*pipe = j;
 				return output;
 			}
 		}
@@ -335,10 +367,11 @@ static void writeback_check_output(igt_output_t *output, igt_plane_t *plane,
 igt_main
 {
 	igt_display_t display;
-	igt_output_t *output;
+	igt_output_t *output, *clone;
 	igt_plane_t *plane;
 	igt_fb_t input_fb;
 	drmModeModeInfo mode;
+	enum pipe pipe;
 	int ret;
 
 	memset(&display, 0, sizeof(display));
@@ -353,7 +386,7 @@ igt_main
 
 		igt_require(display.is_atomic);
 
-		output = kms_writeback_get_output(&display);
+		output = kms_writeback_get_output(&display, &pipe, &clone);
 		igt_require(output);
 
 		if (output->use_override_mode)
@@ -431,6 +464,27 @@ igt_main
 		igt_remove_fb(display.drm_fd, &output_fb);
 	}
 
+	igt_subtest("writeback-check-output-clone") {
+		igt_fb_t output_fb;
+
+		igt_require(clone);
+
+		ret = igt_create_fb(display.drm_fd, mode.hdisplay, mode.vdisplay,
+				    DRM_FORMAT_XRGB8888,
+				    igt_fb_mod_to_tiling(0),
+				    &output_fb);
+		igt_require(ret > 0);
+
+		igt_output_clone_pipe(clone, pipe);
+		igt_output_override_mode(clone, &mode);
+
+		writeback_check_output(output, plane, &input_fb, &output_fb);
+
+		igt_output_set_pipe(clone, PIPE_NONE);
+
+		igt_remove_fb(display.drm_fd, &output_fb);
+	}
+
 	igt_fixture {
 		igt_remove_fb(display.drm_fd, &input_fb);
 		igt_display_fini(&display);
Signature · 1 line
2.21.0

_______________________________________________
igt-dev mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/igt-dev

#35 lore raw

[igt-dev] ✗ Fi.CI.BAT: failure for igt: Add support for testing writeback connectors (rev6)

In reply to Rodrigo Siqueira
== Series Details ==

== Series Details ==

Series: igt: Add support for testing writeback connectors (rev6)
URL : https://patchwork.freedesktop.org/series/39229/
State : failure

== Summary ==

CI Bug Log - changes from IGT_5055 -> IGTPW_3143
====================================================

Summary
-------

**FAILURE**

Serious unknown changes coming with IGTPW_3143 absolutely need to be verified manually.

If you think the reported changes have nothing to do with the changes introduced in IGTPW_3143, please notify your bug team to allow them to document this new failure mode, which will reduce false positives in CI.

External URL: https://patchwork.freedesktop.org/api/1.0/series/39229/revisions/6/mbox/

Possible new issues
-------------------

Here are the unknown changes that may have been introduced in IGTPW_3143:

### IGT changes ###

#### Possible regressions ####

* igt@kms_chamelium@vga-hpd-fast:
- fi-cml-u2: NOTRUN -> [SKIP][1] +1 similar issue
[1]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_3143/fi-cml-u2/igt@[email protected]

#### Warnings ####

* igt@kms_chamelium@vga-hpd-fast:
- fi-icl-u2: [SKIP][2] ([fdo#109309]) -> [SKIP][3] +1 similar issue
[2]: https://intel-gfx-ci.01.org/tree/drm-tip/IGT_5055/fi-icl-u2/igt@[email protected]
[3]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_3143/fi-icl-u2/igt@[email protected]

Known issues
------------

Here are the changes found in IGTPW_3143 that come from known issues:

### IGT changes ###

#### Issues hit ####

* igt@gem_ctx_create@basic-files:
- fi-icl-dsi: [PASS][4] -> [INCOMPLETE][5] ([fdo#107713] / [fdo#109100])
[4]: https://intel-gfx-ci.01.org/tree/drm-tip/IGT_5055/fi-icl-dsi/igt@[email protected]
[5]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_3143/fi-icl-dsi/igt@[email protected]

* igt@kms_busy@basic-flip-c:
- fi-skl-6770hq: [PASS][6] -> [SKIP][7] ([fdo#109271] / [fdo#109278]) +2 similar issues
[6]: https://intel-gfx-ci.01.org/tree/drm-tip/IGT_5055/fi-skl-6770hq/igt@[email protected]
[7]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_3143/fi-skl-6770hq/igt@[email protected]

* igt@kms_flip@basic-flip-vs-dpms:
- fi-skl-6770hq: [PASS][8] -> [SKIP][9] ([fdo#109271]) +23 similar issues
[8]: https://intel-gfx-ci.01.org/tree/drm-tip/IGT_5055/fi-skl-6770hq/igt@[email protected]
[9]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_3143/fi-skl-6770hq/igt@[email protected]
- fi-bxt-dsi: [PASS][10] -> [INCOMPLETE][11] ([fdo#103927])
[10]: https://intel-gfx-ci.01.org/tree/drm-tip/IGT_5055/fi-bxt-dsi/igt@[email protected]
[11]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_3143/fi-bxt-dsi/igt@[email protected]

#### Possible fixes ####

* igt@gem_ctx_create@basic-files:
- fi-icl-y: [INCOMPLETE][12] ([fdo#107713] / [fdo#109100]) -> [PASS][13]
[12]: https://intel-gfx-ci.01.org/tree/drm-tip/IGT_5055/fi-icl-y/igt@[email protected]
[13]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_3143/fi-icl-y/igt@[email protected]

* igt@gem_ctx_switch@basic-default:
- fi-icl-guc: [INCOMPLETE][14] ([fdo#107713] / [fdo#108569]) -> [PASS][15]
[14]: https://intel-gfx-ci.01.org/tree/drm-tip/IGT_5055/fi-icl-guc/igt@[email protected]
[15]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_3143/fi-icl-guc/igt@[email protected]

* igt@gem_exec_basic@basic-all:
- fi-cml-u2: [INCOMPLETE][16] ([fdo#110566]) -> [PASS][17]
[16]: https://intel-gfx-ci.01.org/tree/drm-tip/IGT_5055/fi-cml-u2/igt@[email protected]
[17]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_3143/fi-cml-u2/igt@[email protected]

* igt@kms_frontbuffer_tracking@basic:
- fi-hsw-peppy: [DMESG-WARN][18] ([fdo#102614]) -> [PASS][19]
[18]: https://intel-gfx-ci.01.org/tree/drm-tip/IGT_5055/fi-hsw-peppy/igt@[email protected]
[19]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_3143/fi-hsw-peppy/igt@[email protected]

[fdo#102614]: https://bugs.freedesktop.org/show_bug.cgi?id=102614
[fdo#103927]: https://bugs.freedesktop.org/show_bug.cgi?id=103927
[fdo#107713]: https://bugs.freedesktop.org/show_bug.cgi?id=107713
[fdo#108569]: https://bugs.freedesktop.org/show_bug.cgi?id=108569
[fdo#109100]: https://bugs.freedesktop.org/show_bug.cgi?id=109100
[fdo#109271]: https://bugs.freedesktop.org/show_bug.cgi?id=109271
[fdo#109278]: https://bugs.freedesktop.org/show_bug.cgi?id=109278
[fdo#109309]: https://bugs.freedesktop.org/show_bug.cgi?id=109309
[fdo#110566]: https://bugs.freedesktop.org/show_bug.cgi?id=110566

Participating hosts (53 -> 47)
------------------------------

Additional (1): fi-skl-6600u
Missing (7): fi-kbl-soraka fi-ilk-m540 fi-hsw-4200u fi-byt-squawks fi-bsw-cyan fi-byt-clapper fi-bdw-samus

Build changes
-------------

* IGT: IGT_5055 -> IGTPW_3143

CI_DRM_6253: 83fdc69645c5c6b511e36e171f1c75a6132f007c @ git://anongit.freedesktop.org/gfx-ci/linux
IGTPW_3143: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_3143/
IGT_5055: 495287320225e7f180d384cad7b207b77154438f @ git://anongit.freedesktop.org/xorg/app/intel-gpu-tools

== Testlist changes ==

+igt@kms_writeback@writeback-check-output
+igt@kms_writeback@writeback-check-output-clone
+igt@kms_writeback@writeback-fb-id
+igt@kms_writeback@writeback-invalid-out-fence
+igt@kms_writeback@writeback-pixel-formats

== Logs ==

For more details see: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_3143/
_______________________________________________
igt-dev mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/igt-dev

Reviewed-by on v6 1/6
#2 lore raw

Re: [igt-dev] [PATCH V6 i-g-t 1/6] lib/igt_kms: Add writeback support

In reply to Brian Starkey
Quoted 5 lines · Brian Starkey

Add support in igt_kms for writeback connectors, with the ability to attach framebuffers.

v5: Rebase and add DRM_CLIENT_CAP_WRITEBACK_CONNECTORS before drmModeGetResources()

Thanks for updating this! Given that I have done the last changes and this is mostly a refresh, not sure if I should add more Reviewed-by's to the other patches.

Best regards,
Liviu

Quoted 171 lines
[rebased and updated to the latest igt style]
---
 lib/igt_kms.c | 57 +++++++++++++++++++++++++++++++++++++++++++++++++++
 lib/igt_kms.h |  6 ++++++
 2 files changed, 63 insertions(+)
lib/igt_kms.c +57 -0
index da188a39..140db346 100644
@@ -325,6 +325,9 @@ const char * const igt_connector_prop_names[IGT_NUM_CONNECTOR_PROPS] = {
 	[IGT_CONNECTOR_BROADCAST_RGB] = "Broadcast RGB",
 	[IGT_CONNECTOR_CONTENT_PROTECTION] = "Content Protection",
 	[IGT_CONNECTOR_VRR_CAPABLE] = "vrr_capable",
+	[IGT_CONNECTOR_WRITEBACK_PIXEL_FORMATS] = "WRITEBACK_PIXEL_FORMATS",
+	[IGT_CONNECTOR_WRITEBACK_FB_ID] = "WRITEBACK_FB_ID",
+	[IGT_CONNECTOR_WRITEBACK_OUT_FENCE_PTR] = "WRITEBACK_OUT_FENCE_PTR",
 };
 
 /*
@@ -557,6 +560,7 @@ static const struct type_name connector_type_names[] = {
 	{ DRM_MODE_CONNECTOR_VIRTUAL, "Virtual" },
 	{ DRM_MODE_CONNECTOR_DSI, "DSI" },
 	{ DRM_MODE_CONNECTOR_DPI, "DPI" },
+	{ DRM_MODE_CONNECTOR_WRITEBACK, "Writeback" },
 	{}
 };
 
@@ -1889,6 +1893,12 @@ static void igt_output_reset(igt_output_t *output)
 	if (igt_output_has_prop(output, IGT_CONNECTOR_BROADCAST_RGB))
 		igt_output_set_prop_value(output, IGT_CONNECTOR_BROADCAST_RGB,
 					  BROADCAST_RGB_FULL);
+	if (igt_output_has_prop(output, IGT_CONNECTOR_WRITEBACK_FB_ID))
+		igt_output_set_prop_value(output, IGT_CONNECTOR_WRITEBACK_FB_ID, 0);
+	if (igt_output_has_prop(output, IGT_CONNECTOR_WRITEBACK_OUT_FENCE_PTR)) {
+		igt_output_clear_prop_changed(output, IGT_CONNECTOR_WRITEBACK_OUT_FENCE_PTR);
+		output->writeback_out_fence_fd = -1;
+	}
 }
 
 /**
@@ -1901,6 +1911,8 @@ static void igt_output_reset(igt_output_t *output)
  * For outputs:
  * - %IGT_CONNECTOR_CRTC_ID
  * - %IGT_CONNECTOR_BROADCAST_RGB (if applicable)
+ * - %IGT_CONNECTOR_WRITEBACK_FB_ID
+ * - %IGT_CONNECTOR_WRITEBACK_OUT_FENCE_PTR
  * - igt_output_override_mode() to default.
  *
  * For pipes:
@@ -1970,6 +1982,8 @@ void igt_display_require(igt_display_t *display, int drm_fd)
 
 	display->drm_fd = drm_fd;
 
+	drmSetClientCap(drm_fd, DRM_CLIENT_CAP_WRITEBACK_CONNECTORS, 1);
+
 	resources = drmModeGetResources(display->drm_fd);
 	if (!resources)
 		goto out;
@@ -2263,6 +2277,11 @@ static void igt_output_fini(igt_output_t *output)
 	kmstest_free_connector_config(&output->config);
 	free(output->name);
 	output->name = NULL;
+
+	if (output->writeback_out_fence_fd != -1) {
+		close(output->writeback_out_fence_fd);
+		output->writeback_out_fence_fd = -1;
+	}
 }
 
 /**
@@ -3325,6 +3344,11 @@ static void igt_atomic_prepare_connector_commit(igt_output_t *output, drmModeAto
 					  output->props[i],
 					  output->values[i]));
 	}
+
+	if (output->writeback_out_fence_fd != -1) {
+		close(output->writeback_out_fence_fd);
+		output->writeback_out_fence_fd = -1;
+	}
 }
 
 /*
@@ -3447,6 +3471,16 @@ display_commit_changed(igt_display_t *display, enum igt_commit_style s)
 		else
 			/* no modeset in universal commit, no change to crtc. */
 			output->changed &= 1 << IGT_CONNECTOR_CRTC_ID;
+
+		if (s == COMMIT_ATOMIC) {
+			if (igt_output_is_prop_changed(output, IGT_CONNECTOR_WRITEBACK_OUT_FENCE_PTR))
+				igt_assert(output->writeback_out_fence_fd >= 0);
+
+			output->values[IGT_CONNECTOR_WRITEBACK_OUT_FENCE_PTR] = 0;
+			output->values[IGT_CONNECTOR_WRITEBACK_FB_ID] = 0;
+			igt_output_clear_prop_changed(output, IGT_CONNECTOR_WRITEBACK_FB_ID);
+			igt_output_clear_prop_changed(output, IGT_CONNECTOR_WRITEBACK_OUT_FENCE_PTR);
+		}
 	}
 
 	if (display->first_commit) {
@@ -4119,6 +4153,29 @@ void igt_pipe_request_out_fence(igt_pipe_t *pipe)
 	igt_pipe_obj_set_prop_value(pipe, IGT_CRTC_OUT_FENCE_PTR, (ptrdiff_t)&pipe->out_fence_fd);
 }
 
+/**
+ * igt_output_set_writeback_fb:
+ * @output: Target output
+ * @fb: Target framebuffer
+ *
+ * This function sets the given @fb to be used as the target framebuffer for the
+ * writeback engine at the next atomic commit. It will also request a writeback
+ * out fence that will contain the fd number of the out fence created by KMS if
+ * the given @fb is valid.
+ */
+void igt_output_set_writeback_fb(igt_output_t *output, struct igt_fb *fb)
+{
+	igt_display_t *display = output->display;
+
+	LOG(display, "%s: output_set_writeback_fb(%d)\n", output->name, fb ? fb->fb_id : 0);
+
+	igt_output_set_prop_value(output, IGT_CONNECTOR_WRITEBACK_FB_ID, fb ? fb->fb_id : 0);
+	/* only request a writeback out fence if the framebuffer is valid */
+	if (fb)
+		igt_output_set_prop_value(output, IGT_CONNECTOR_WRITEBACK_OUT_FENCE_PTR,
+					  (ptrdiff_t)&output->writeback_out_fence_fd);
+}
+
 /**
  * igt_wait_for_vblank_count:
  * @drm_fd: A drm file descriptor
lib/igt_kms.h +6 -0
index a448a003..cacc6b90 100644
@@ -123,6 +123,9 @@ enum igt_atomic_connector_properties {
        IGT_CONNECTOR_BROADCAST_RGB,
        IGT_CONNECTOR_CONTENT_PROTECTION,
        IGT_CONNECTOR_VRR_CAPABLE,
+       IGT_CONNECTOR_WRITEBACK_PIXEL_FORMATS,
+       IGT_CONNECTOR_WRITEBACK_FB_ID,
+       IGT_CONNECTOR_WRITEBACK_OUT_FENCE_PTR,
        IGT_NUM_CONNECTOR_PROPS
 };
 
@@ -364,6 +367,8 @@ typedef struct {
 	bool use_override_mode;
 	drmModeModeInfo override_mode;
 
+	int32_t writeback_out_fence_fd;
+
 	/* bitmask of changed properties */
 	uint64_t changed;
 
@@ -414,6 +419,7 @@ igt_plane_t *igt_output_get_plane_type_index(igt_output_t *output,
 igt_output_t *igt_output_from_connector(igt_display_t *display,
     drmModeConnector *connector);
 const drmModeModeInfo *igt_std_1024_mode_get(void);
+void igt_output_set_writeback_fb(igt_output_t *output, struct igt_fb *fb);
 
 igt_plane_t *igt_pipe_get_plane_type(igt_pipe_t *pipe, int plane_type);
 int igt_pipe_count_plane_type(igt_pipe_t *pipe, int plane_type);
Signature · 6 lines
2.21.0


-- 
Rodrigo Siqueira
https://siqueira.tech
Signature · 12 lines
====================
| I would like to |
| fix the world,  |
| but they're not |
| giving me the   |
 \ source code!  /
  ---------------
    ¯\_(ツ)_/¯
_______________________________________________
igt-dev mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/igt-dev
#3 lore raw

Re: [igt-dev] [PATCH V6 i-g-t 1/6] lib/igt_kms: Add writeback support

Quoted 5 lines · Liviu Dudau
Quoted 3 lines · Brian Starkey

Add support in igt_kms for writeback connectors, with the ability to attach framebuffers.

Quoted 8 lines
Quoted 2 lines

v5: Rebase and add DRM_CLIENT_CAP_WRITEBACK_CONNECTORS before drmModeGetResources()

Thanks for updating this! Given that I have done the last changes and this is mostly a refresh, not sure if I should add more Reviewed-by's to the other patches.

Thanks Liviu!

I just forgot to add my SoB, and for some reason, gmail does not allow me to send an email on someone behalf. Btw, I can fix it after everybody agrees that the kms_writeback is ready for landing.

Quoted 186 lines

Best regards,
Liviu

Quoted 168 lines
[rebased and updated to the latest igt style]
---
 lib/igt_kms.c | 57 +++++++++++++++++++++++++++++++++++++++++++++++++++
 lib/igt_kms.h |  6 ++++++
 2 files changed, 63 insertions(+)
lib/igt_kms.c +57 -0
index da188a39..140db346 100644
@@ -325,6 +325,9 @@ const char * const igt_connector_prop_names[IGT_NUM_CONNECTOR_PROPS] = {
      [IGT_CONNECTOR_BROADCAST_RGB] = "Broadcast RGB",
      [IGT_CONNECTOR_CONTENT_PROTECTION] = "Content Protection",
      [IGT_CONNECTOR_VRR_CAPABLE] = "vrr_capable",
+     [IGT_CONNECTOR_WRITEBACK_PIXEL_FORMATS] = "WRITEBACK_PIXEL_FORMATS",
+     [IGT_CONNECTOR_WRITEBACK_FB_ID] = "WRITEBACK_FB_ID",
+     [IGT_CONNECTOR_WRITEBACK_OUT_FENCE_PTR] = "WRITEBACK_OUT_FENCE_PTR",
 };
 
 /*
@@ -557,6 +560,7 @@ static const struct type_name connector_type_names[] = {
      { DRM_MODE_CONNECTOR_VIRTUAL, "Virtual" },
      { DRM_MODE_CONNECTOR_DSI, "DSI" },
      { DRM_MODE_CONNECTOR_DPI, "DPI" },
+     { DRM_MODE_CONNECTOR_WRITEBACK, "Writeback" },
      {}
 };
 
@@ -1889,6 +1893,12 @@ static void igt_output_reset(igt_output_t *output)
      if (igt_output_has_prop(output, IGT_CONNECTOR_BROADCAST_RGB))
              igt_output_set_prop_value(output, IGT_CONNECTOR_BROADCAST_RGB,
                                        BROADCAST_RGB_FULL);
+     if (igt_output_has_prop(output, IGT_CONNECTOR_WRITEBACK_FB_ID))
+             igt_output_set_prop_value(output, IGT_CONNECTOR_WRITEBACK_FB_ID, 0);
+     if (igt_output_has_prop(output, IGT_CONNECTOR_WRITEBACK_OUT_FENCE_PTR)) {
+             igt_output_clear_prop_changed(output, IGT_CONNECTOR_WRITEBACK_OUT_FENCE_PTR);
+             output->writeback_out_fence_fd = -1;
+     }
 }
 
 /**
@@ -1901,6 +1911,8 @@ static void igt_output_reset(igt_output_t *output)
  * For outputs:
  * - %IGT_CONNECTOR_CRTC_ID
  * - %IGT_CONNECTOR_BROADCAST_RGB (if applicable)
+ * - %IGT_CONNECTOR_WRITEBACK_FB_ID
+ * - %IGT_CONNECTOR_WRITEBACK_OUT_FENCE_PTR
  * - igt_output_override_mode() to default.
  *
  * For pipes:
@@ -1970,6 +1982,8 @@ void igt_display_require(igt_display_t *display, int drm_fd)
 
      display->drm_fd = drm_fd;
 
+     drmSetClientCap(drm_fd, DRM_CLIENT_CAP_WRITEBACK_CONNECTORS, 1);
+
      resources = drmModeGetResources(display->drm_fd);
      if (!resources)
              goto out;
@@ -2263,6 +2277,11 @@ static void igt_output_fini(igt_output_t *output)
      kmstest_free_connector_config(&output->config);
      free(output->name);
      output->name = NULL;
+
+     if (output->writeback_out_fence_fd != -1) {
+             close(output->writeback_out_fence_fd);
+             output->writeback_out_fence_fd = -1;
+     }
 }
 
 /**
@@ -3325,6 +3344,11 @@ static void igt_atomic_prepare_connector_commit(igt_output_t *output, drmModeAto
                                        output->props[i],
                                        output->values[i]));
      }
+
+     if (output->writeback_out_fence_fd != -1) {
+             close(output->writeback_out_fence_fd);
+             output->writeback_out_fence_fd = -1;
+     }
 }
 
 /*
@@ -3447,6 +3471,16 @@ display_commit_changed(igt_display_t *display, enum igt_commit_style s)
              else
                      /* no modeset in universal commit, no change to crtc. */
                      output->changed &= 1 << IGT_CONNECTOR_CRTC_ID;
+
+             if (s == COMMIT_ATOMIC) {
+                     if (igt_output_is_prop_changed(output, IGT_CONNECTOR_WRITEBACK_OUT_FENCE_PTR))
+                             igt_assert(output->writeback_out_fence_fd >= 0);
+
+                     output->values[IGT_CONNECTOR_WRITEBACK_OUT_FENCE_PTR] = 0;
+                     output->values[IGT_CONNECTOR_WRITEBACK_FB_ID] = 0;
+                     igt_output_clear_prop_changed(output, IGT_CONNECTOR_WRITEBACK_FB_ID);
+                     igt_output_clear_prop_changed(output, IGT_CONNECTOR_WRITEBACK_OUT_FENCE_PTR);
+             }
      }
 
      if (display->first_commit) {
@@ -4119,6 +4153,29 @@ void igt_pipe_request_out_fence(igt_pipe_t *pipe)
      igt_pipe_obj_set_prop_value(pipe, IGT_CRTC_OUT_FENCE_PTR, (ptrdiff_t)&pipe->out_fence_fd);
 }
 
+/**
+ * igt_output_set_writeback_fb:
+ * @output: Target output
+ * @fb: Target framebuffer
+ *
+ * This function sets the given @fb to be used as the target framebuffer for the
+ * writeback engine at the next atomic commit. It will also request a writeback
+ * out fence that will contain the fd number of the out fence created by KMS if
+ * the given @fb is valid.
+ */
+void igt_output_set_writeback_fb(igt_output_t *output, struct igt_fb *fb)
+{
+     igt_display_t *display = output->display;
+
+     LOG(display, "%s: output_set_writeback_fb(%d)\n", output->name, fb ? fb->fb_id : 0);
+
+     igt_output_set_prop_value(output, IGT_CONNECTOR_WRITEBACK_FB_ID, fb ? fb->fb_id : 0);
+     /* only request a writeback out fence if the framebuffer is valid */
+     if (fb)
+             igt_output_set_prop_value(output, IGT_CONNECTOR_WRITEBACK_OUT_FENCE_PTR,
+                                       (ptrdiff_t)&output->writeback_out_fence_fd);
+}
+
 /**
  * igt_wait_for_vblank_count:
  * @drm_fd: A drm file descriptor
lib/igt_kms.h +6 -1
index a448a003..cacc6b90 100644
@@ -123,6 +123,9 @@ enum igt_atomic_connector_properties {
        IGT_CONNECTOR_BROADCAST_RGB,
        IGT_CONNECTOR_CONTENT_PROTECTION,
        IGT_CONNECTOR_VRR_CAPABLE,
+       IGT_CONNECTOR_WRITEBACK_PIXEL_FORMATS,
+       IGT_CONNECTOR_WRITEBACK_FB_ID,
+       IGT_CONNECTOR_WRITEBACK_OUT_FENCE_PTR,
        IGT_NUM_CONNECTOR_PROPS
 };
 
@@ -364,6 +367,8 @@ typedef struct {
      bool use_override_mode;
      drmModeModeInfo override_mode;
 
+     int32_t writeback_out_fence_fd;
+
      /* bitmask of changed properties */
      uint64_t changed;
 
@@ -414,6 +419,7 @@ igt_plane_t *igt_output_get_plane_type_index(igt_output_t *output,
 igt_output_t *igt_output_from_connector(igt_display_t *display,
     drmModeConnector *connector);
 const drmModeModeInfo *igt_std_1024_mode_get(void);
+void igt_output_set_writeback_fb(igt_output_t *output, struct igt_fb *fb);
 
 igt_plane_t *igt_pipe_get_plane_type(igt_pipe_t *pipe, int plane_type);
 int igt_pipe_count_plane_type(igt_pipe_t *pipe, int plane_type);
--

2.21.0

Quoted 3 lines

--
Rodrigo Siqueira
https://siqueira.tech

--
====================

| I would like to |
| fix the world,  |
| but they're not |
| giving me the   |
 \ source code!  /
  ---------------
    ¯\_(ツ)_/¯
Signature · 6 lines
Rodrigo Siqueira
https://siqueira.tech
_______________________________________________
igt-dev mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/igt-dev
#4 lore raw

Re: [igt-dev] [PATCH V6 i-g-t 1/6] lib/igt_kms: Add writeback support

Quoted 10 lines · Rodrigo Siqueira
Quoted 9 lines · Liviu Dudau
Quoted 5 lines · Brian Starkey

Add support in igt_kms for writeback connectors, with the ability to attach framebuffers.

v5: Rebase and add DRM_CLIENT_CAP_WRITEBACK_CONNECTORS before drmModeGetResources()

Quoted 8 lines
Quoted 3 lines

Thanks for updating this! Given that I have done the last changes and this is mostly a refresh, not sure if I should add more Reviewed-by's to the other patches.

Thanks Liviu!

I just forgot to add my SoB, and for some reason, gmail does not allow me to send an email on someone behalf.

FWIW, that's a good thing, and it's required to pass DMARC checks.

Instead, git-send-email should add a From line at the beginning of the message when sending a patch on behalf of someone else. I wonder what happened here.

Quoted 190 lines

Btw, I can fix it after
everybody agrees that the kms_writeback is ready for landing.

Quoted 184 lines

Best regards,
Liviu

Quoted 167 lines
[rebased and updated to the latest igt style]
---
 lib/igt_kms.c | 57 +++++++++++++++++++++++++++++++++++++++++++++++++++
 lib/igt_kms.h |  6 ++++++
 2 files changed, 63 insertions(+)
lib/igt_kms.c +57 -0
index da188a39..140db346 100644
@@ -325,6 +325,9 @@ const char * const igt_connector_prop_names[IGT_NUM_CONNECTOR_PROPS] = {
      [IGT_CONNECTOR_BROADCAST_RGB] = "Broadcast RGB",
      [IGT_CONNECTOR_CONTENT_PROTECTION] = "Content Protection",
      [IGT_CONNECTOR_VRR_CAPABLE] = "vrr_capable",
+     [IGT_CONNECTOR_WRITEBACK_PIXEL_FORMATS] = "WRITEBACK_PIXEL_FORMATS",
+     [IGT_CONNECTOR_WRITEBACK_FB_ID] = "WRITEBACK_FB_ID",
+     [IGT_CONNECTOR_WRITEBACK_OUT_FENCE_PTR] = "WRITEBACK_OUT_FENCE_PTR",
 };
 
 /*
@@ -557,6 +560,7 @@ static const struct type_name connector_type_names[] = {
      { DRM_MODE_CONNECTOR_VIRTUAL, "Virtual" },
      { DRM_MODE_CONNECTOR_DSI, "DSI" },
      { DRM_MODE_CONNECTOR_DPI, "DPI" },
+     { DRM_MODE_CONNECTOR_WRITEBACK, "Writeback" },
      {}
 };
 
@@ -1889,6 +1893,12 @@ static void igt_output_reset(igt_output_t *output)
      if (igt_output_has_prop(output, IGT_CONNECTOR_BROADCAST_RGB))
              igt_output_set_prop_value(output, IGT_CONNECTOR_BROADCAST_RGB,
                                        BROADCAST_RGB_FULL);
+     if (igt_output_has_prop(output, IGT_CONNECTOR_WRITEBACK_FB_ID))
+             igt_output_set_prop_value(output, IGT_CONNECTOR_WRITEBACK_FB_ID, 0);
+     if (igt_output_has_prop(output, IGT_CONNECTOR_WRITEBACK_OUT_FENCE_PTR)) {
+             igt_output_clear_prop_changed(output, IGT_CONNECTOR_WRITEBACK_OUT_FENCE_PTR);
+             output->writeback_out_fence_fd = -1;
+     }
 }
 
 /**
@@ -1901,6 +1911,8 @@ static void igt_output_reset(igt_output_t *output)
  * For outputs:
  * - %IGT_CONNECTOR_CRTC_ID
  * - %IGT_CONNECTOR_BROADCAST_RGB (if applicable)
+ * - %IGT_CONNECTOR_WRITEBACK_FB_ID
+ * - %IGT_CONNECTOR_WRITEBACK_OUT_FENCE_PTR
  * - igt_output_override_mode() to default.
  *
  * For pipes:
@@ -1970,6 +1982,8 @@ void igt_display_require(igt_display_t *display, int drm_fd)
 
      display->drm_fd = drm_fd;
 
+     drmSetClientCap(drm_fd, DRM_CLIENT_CAP_WRITEBACK_CONNECTORS, 1);
+
      resources = drmModeGetResources(display->drm_fd);
      if (!resources)
              goto out;
@@ -2263,6 +2277,11 @@ static void igt_output_fini(igt_output_t *output)
      kmstest_free_connector_config(&output->config);
      free(output->name);
      output->name = NULL;
+
+     if (output->writeback_out_fence_fd != -1) {
+             close(output->writeback_out_fence_fd);
+             output->writeback_out_fence_fd = -1;
+     }
 }
 
 /**
@@ -3325,6 +3344,11 @@ static void igt_atomic_prepare_connector_commit(igt_output_t *output, drmModeAto
                                        output->props[i],
                                        output->values[i]));
      }
+
+     if (output->writeback_out_fence_fd != -1) {
+             close(output->writeback_out_fence_fd);
+             output->writeback_out_fence_fd = -1;
+     }
 }
 
 /*
@@ -3447,6 +3471,16 @@ display_commit_changed(igt_display_t *display, enum igt_commit_style s)
              else
                      /* no modeset in universal commit, no change to crtc. */
                      output->changed &= 1 << IGT_CONNECTOR_CRTC_ID;
+
+             if (s == COMMIT_ATOMIC) {
+                     if (igt_output_is_prop_changed(output, IGT_CONNECTOR_WRITEBACK_OUT_FENCE_PTR))
+                             igt_assert(output->writeback_out_fence_fd >= 0);
+
+                     output->values[IGT_CONNECTOR_WRITEBACK_OUT_FENCE_PTR] = 0;
+                     output->values[IGT_CONNECTOR_WRITEBACK_FB_ID] = 0;
+                     igt_output_clear_prop_changed(output, IGT_CONNECTOR_WRITEBACK_FB_ID);
+                     igt_output_clear_prop_changed(output, IGT_CONNECTOR_WRITEBACK_OUT_FENCE_PTR);
+             }
      }
 
      if (display->first_commit) {
@@ -4119,6 +4153,29 @@ void igt_pipe_request_out_fence(igt_pipe_t *pipe)
      igt_pipe_obj_set_prop_value(pipe, IGT_CRTC_OUT_FENCE_PTR, (ptrdiff_t)&pipe->out_fence_fd);
 }
 
+/**
+ * igt_output_set_writeback_fb:
+ * @output: Target output
+ * @fb: Target framebuffer
+ *
+ * This function sets the given @fb to be used as the target framebuffer for the
+ * writeback engine at the next atomic commit. It will also request a writeback
+ * out fence that will contain the fd number of the out fence created by KMS if
+ * the given @fb is valid.
+ */
+void igt_output_set_writeback_fb(igt_output_t *output, struct igt_fb *fb)
+{
+     igt_display_t *display = output->display;
+
+     LOG(display, "%s: output_set_writeback_fb(%d)\n", output->name, fb ? fb->fb_id : 0);
+
+     igt_output_set_prop_value(output, IGT_CONNECTOR_WRITEBACK_FB_ID, fb ? fb->fb_id : 0);
+     /* only request a writeback out fence if the framebuffer is valid */
+     if (fb)
+             igt_output_set_prop_value(output, IGT_CONNECTOR_WRITEBACK_OUT_FENCE_PTR,
+                                       (ptrdiff_t)&output->writeback_out_fence_fd);
+}
+
 /**
  * igt_wait_for_vblank_count:
  * @drm_fd: A drm file descriptor
lib/igt_kms.h +6 -1
index a448a003..cacc6b90 100644
@@ -123,6 +123,9 @@ enum igt_atomic_connector_properties {
        IGT_CONNECTOR_BROADCAST_RGB,
        IGT_CONNECTOR_CONTENT_PROTECTION,
        IGT_CONNECTOR_VRR_CAPABLE,
+       IGT_CONNECTOR_WRITEBACK_PIXEL_FORMATS,
+       IGT_CONNECTOR_WRITEBACK_FB_ID,
+       IGT_CONNECTOR_WRITEBACK_OUT_FENCE_PTR,
        IGT_NUM_CONNECTOR_PROPS
 };
 
@@ -364,6 +367,8 @@ typedef struct {
      bool use_override_mode;
      drmModeModeInfo override_mode;
 
+     int32_t writeback_out_fence_fd;
+
      /* bitmask of changed properties */
      uint64_t changed;
 
@@ -414,6 +419,7 @@ igt_plane_t *igt_output_get_plane_type_index(igt_output_t *output,
 igt_output_t *igt_output_from_connector(igt_display_t *display,
     drmModeConnector *connector);
 const drmModeModeInfo *igt_std_1024_mode_get(void);
+void igt_output_set_writeback_fb(igt_output_t *output, struct igt_fb *fb);
 
 igt_plane_t *igt_pipe_get_plane_type(igt_pipe_t *pipe, int plane_type);
 int igt_pipe_count_plane_type(igt_pipe_t *pipe, int plane_type);
--

2.21.0

Quoted 3 lines

--
Rodrigo Siqueira
https://siqueira.tech

--
====================

Quoted 4 lines
I would like to |
fix the world,  |
but they're not |
giving me the   |

\ source code! /
---------------

    ¯\_(ツ)_/¯

_______________________________________________
igt-dev mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/igt-dev

#5 lore raw

Re: [igt-dev] [PATCH V6 i-g-t 1/6] lib/igt_kms: Add writeback support

Quoted 18 lines · Simon
Quoted 16 lines · Rodrigo Siqueira
Quoted 4 lines · Liviu Dudau
Quoted 3 lines · Brian Starkey

Add support in igt_kms for writeback connectors, with the ability to attach framebuffers.

Quoted 8 lines
Quoted 2 lines

v5: Rebase and add DRM_CLIENT_CAP_WRITEBACK_CONNECTORS before drmModeGetResources()

Thanks for updating this! Given that I have done the last changes and this is mostly a refresh, not sure if I should add more Reviewed-by's to the other patches.

Thanks Liviu!

Quoted 8 lines
Quoted 2 lines

I just forgot to add my SoB, and for some reason, gmail does not allow me to send an email on someone behalf.

FWIW, that's a good thing, and it's required to pass DMARC checks.

Instead, git-send-email should add a From line at the beginning of the message when sending a patch on behalf of someone else. I wonder what happened here.

Thank you for your help.

I’m using neomutt for sending patches, and I’ll take a look at git-send-email. Additionally, it looks like Gmail requires that I add a new account in order to allow me to send patches on someone behalf. I’ll take some time to read about this issue, and I will try to resend the patchset again.

Btw, if you have time, could you take a look in this series?

Best regards

Quoted 190 lines
Quoted 190 lines

Btw, I can fix it after
everybody agrees that the kms_writeback is ready for landing.

Quoted 184 lines

Best regards,
Liviu

Quoted 167 lines
[rebased and updated to the latest igt style]
---
 lib/igt_kms.c | 57 +++++++++++++++++++++++++++++++++++++++++++++++++++
 lib/igt_kms.h |  6 ++++++
 2 files changed, 63 insertions(+)
lib/igt_kms.c +57 -0
index da188a39..140db346 100644
@@ -325,6 +325,9 @@ const char * const igt_connector_prop_names[IGT_NUM_CONNECTOR_PROPS] = {
      [IGT_CONNECTOR_BROADCAST_RGB] = "Broadcast RGB",
      [IGT_CONNECTOR_CONTENT_PROTECTION] = "Content Protection",
      [IGT_CONNECTOR_VRR_CAPABLE] = "vrr_capable",
+     [IGT_CONNECTOR_WRITEBACK_PIXEL_FORMATS] = "WRITEBACK_PIXEL_FORMATS",
+     [IGT_CONNECTOR_WRITEBACK_FB_ID] = "WRITEBACK_FB_ID",
+     [IGT_CONNECTOR_WRITEBACK_OUT_FENCE_PTR] = "WRITEBACK_OUT_FENCE_PTR",
 };
 
 /*
@@ -557,6 +560,7 @@ static const struct type_name connector_type_names[] = {
      { DRM_MODE_CONNECTOR_VIRTUAL, "Virtual" },
      { DRM_MODE_CONNECTOR_DSI, "DSI" },
      { DRM_MODE_CONNECTOR_DPI, "DPI" },
+     { DRM_MODE_CONNECTOR_WRITEBACK, "Writeback" },
      {}
 };
 
@@ -1889,6 +1893,12 @@ static void igt_output_reset(igt_output_t *output)
      if (igt_output_has_prop(output, IGT_CONNECTOR_BROADCAST_RGB))
              igt_output_set_prop_value(output, IGT_CONNECTOR_BROADCAST_RGB,
                                        BROADCAST_RGB_FULL);
+     if (igt_output_has_prop(output, IGT_CONNECTOR_WRITEBACK_FB_ID))
+             igt_output_set_prop_value(output, IGT_CONNECTOR_WRITEBACK_FB_ID, 0);
+     if (igt_output_has_prop(output, IGT_CONNECTOR_WRITEBACK_OUT_FENCE_PTR)) {
+             igt_output_clear_prop_changed(output, IGT_CONNECTOR_WRITEBACK_OUT_FENCE_PTR);
+             output->writeback_out_fence_fd = -1;
+     }
 }
 
 /**
@@ -1901,6 +1911,8 @@ static void igt_output_reset(igt_output_t *output)
  * For outputs:
  * - %IGT_CONNECTOR_CRTC_ID
  * - %IGT_CONNECTOR_BROADCAST_RGB (if applicable)
+ * - %IGT_CONNECTOR_WRITEBACK_FB_ID
+ * - %IGT_CONNECTOR_WRITEBACK_OUT_FENCE_PTR
  * - igt_output_override_mode() to default.
  *
  * For pipes:
@@ -1970,6 +1982,8 @@ void igt_display_require(igt_display_t *display, int drm_fd)
 
      display->drm_fd = drm_fd;
 
+     drmSetClientCap(drm_fd, DRM_CLIENT_CAP_WRITEBACK_CONNECTORS, 1);
+
      resources = drmModeGetResources(display->drm_fd);
      if (!resources)
              goto out;
@@ -2263,6 +2277,11 @@ static void igt_output_fini(igt_output_t *output)
      kmstest_free_connector_config(&output->config);
      free(output->name);
      output->name = NULL;
+
+     if (output->writeback_out_fence_fd != -1) {
+             close(output->writeback_out_fence_fd);
+             output->writeback_out_fence_fd = -1;
+     }
 }
 
 /**
@@ -3325,6 +3344,11 @@ static void igt_atomic_prepare_connector_commit(igt_output_t *output, drmModeAto
                                        output->props[i],
                                        output->values[i]));
      }
+
+     if (output->writeback_out_fence_fd != -1) {
+             close(output->writeback_out_fence_fd);
+             output->writeback_out_fence_fd = -1;
+     }
 }
 
 /*
@@ -3447,6 +3471,16 @@ display_commit_changed(igt_display_t *display, enum igt_commit_style s)
              else
                      /* no modeset in universal commit, no change to crtc. */
                      output->changed &= 1 << IGT_CONNECTOR_CRTC_ID;
+
+             if (s == COMMIT_ATOMIC) {
+                     if (igt_output_is_prop_changed(output, IGT_CONNECTOR_WRITEBACK_OUT_FENCE_PTR))
+                             igt_assert(output->writeback_out_fence_fd >= 0);
+
+                     output->values[IGT_CONNECTOR_WRITEBACK_OUT_FENCE_PTR] = 0;
+                     output->values[IGT_CONNECTOR_WRITEBACK_FB_ID] = 0;
+                     igt_output_clear_prop_changed(output, IGT_CONNECTOR_WRITEBACK_FB_ID);
+                     igt_output_clear_prop_changed(output, IGT_CONNECTOR_WRITEBACK_OUT_FENCE_PTR);
+             }
      }
 
      if (display->first_commit) {
@@ -4119,6 +4153,29 @@ void igt_pipe_request_out_fence(igt_pipe_t *pipe)
      igt_pipe_obj_set_prop_value(pipe, IGT_CRTC_OUT_FENCE_PTR, (ptrdiff_t)&pipe->out_fence_fd);
 }
 
+/**
+ * igt_output_set_writeback_fb:
+ * @output: Target output
+ * @fb: Target framebuffer
+ *
+ * This function sets the given @fb to be used as the target framebuffer for the
+ * writeback engine at the next atomic commit. It will also request a writeback
+ * out fence that will contain the fd number of the out fence created by KMS if
+ * the given @fb is valid.
+ */
+void igt_output_set_writeback_fb(igt_output_t *output, struct igt_fb *fb)
+{
+     igt_display_t *display = output->display;
+
+     LOG(display, "%s: output_set_writeback_fb(%d)\n", output->name, fb ? fb->fb_id : 0);
+
+     igt_output_set_prop_value(output, IGT_CONNECTOR_WRITEBACK_FB_ID, fb ? fb->fb_id : 0);
+     /* only request a writeback out fence if the framebuffer is valid */
+     if (fb)
+             igt_output_set_prop_value(output, IGT_CONNECTOR_WRITEBACK_OUT_FENCE_PTR,
+                                       (ptrdiff_t)&output->writeback_out_fence_fd);
+}
+
 /**
  * igt_wait_for_vblank_count:
  * @drm_fd: A drm file descriptor
lib/igt_kms.h +6 -1
index a448a003..cacc6b90 100644
@@ -123,6 +123,9 @@ enum igt_atomic_connector_properties {
        IGT_CONNECTOR_BROADCAST_RGB,
        IGT_CONNECTOR_CONTENT_PROTECTION,
        IGT_CONNECTOR_VRR_CAPABLE,
+       IGT_CONNECTOR_WRITEBACK_PIXEL_FORMATS,
+       IGT_CONNECTOR_WRITEBACK_FB_ID,
+       IGT_CONNECTOR_WRITEBACK_OUT_FENCE_PTR,
        IGT_NUM_CONNECTOR_PROPS
 };
 
@@ -364,6 +367,8 @@ typedef struct {
      bool use_override_mode;
      drmModeModeInfo override_mode;
 
+     int32_t writeback_out_fence_fd;
+
      /* bitmask of changed properties */
      uint64_t changed;
 
@@ -414,6 +419,7 @@ igt_plane_t *igt_output_get_plane_type_index(igt_output_t *output,
 igt_output_t *igt_output_from_connector(igt_display_t *display,
     drmModeConnector *connector);
 const drmModeModeInfo *igt_std_1024_mode_get(void);
+void igt_output_set_writeback_fb(igt_output_t *output, struct igt_fb *fb);
 
 igt_plane_t *igt_pipe_get_plane_type(igt_pipe_t *pipe, int plane_type);
 int igt_pipe_count_plane_type(igt_pipe_t *pipe, int plane_type);
--

2.21.0

Quoted 3 lines

--
Rodrigo Siqueira
https://siqueira.tech

--
====================

Quoted 4 lines
I would like to |
fix the world,  |
but they're not |
giving me the   |

\ source code! /
---------------

    ¯\_(ツ)_/¯
Signature · 6 lines
Rodrigo Siqueira
https://siqueira.tech
_______________________________________________
igt-dev mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/igt-dev
#6 lore raw

Re: [igt-dev] [PATCH V6 i-g-t 1/6] lib/igt_kms: Add writeback support

Quoted 27 lines · Rodrigo Siqueira
Quoted 17 lines · Simon
Quoted 16 lines · Rodrigo Siqueira
Quoted 4 lines · Liviu Dudau
Quoted 3 lines · Brian Starkey

Add support in igt_kms for writeback connectors, with the ability to attach framebuffers.

Quoted 8 lines
Quoted 2 lines

v5: Rebase and add DRM_CLIENT_CAP_WRITEBACK_CONNECTORS before drmModeGetResources()

Thanks for updating this! Given that I have done the last changes and this is mostly a refresh, not sure if I should add more Reviewed-by's to the other patches.

Thanks Liviu!

Quoted 8 lines
Quoted 2 lines

I just forgot to add my SoB, and for some reason, gmail does not allow me to send an email on someone behalf.

FWIW, that's a good thing, and it's required to pass DMARC checks.

Instead, git-send-email should add a From line at the beginning of the message when sending a patch on behalf of someone else. I wonder what happened here.

Quoted 7 lines

Thank you for your help.

I’m using neomutt for sending patches, and I’ll take a look at git-send-email. Additionally, it looks like Gmail requires that I add a new account in order to allow me to send patches on someone behalf. I’ll take some time to read about this issue, and I will try to resend the patchset again.

When sending a patch on someone else's behalf, you shouldn't send the e-mail with the sender (From header) set to someone else. You should use your normal address. The From line in the body of the message will make Git understand the author is someone else.

So no new account setup required.

I really recommend using git-send-email for patches. There are too many ways to make a mistake if you try sending emails manually. Here is a tutorial:
https://git-send-email.io/

Quoted 1 line

Btw, if you have time, could you take a look in this series?

Sure, I've already began looking at the series. I'll continue soon.

Thanks for your work!

Quoted 191 lines

Best regards

Quoted 187 lines
Quoted 187 lines

Btw, I can fix it after
everybody agrees that the kms_writeback is ready for landing.

Quoted 183 lines

Best regards,
Liviu

Quoted 167 lines
[rebased and updated to the latest igt style]
---
 lib/igt_kms.c | 57 +++++++++++++++++++++++++++++++++++++++++++++++++++
 lib/igt_kms.h |  6 ++++++
 2 files changed, 63 insertions(+)
lib/igt_kms.c +57 -0
index da188a39..140db346 100644
@@ -325,6 +325,9 @@ const char * const igt_connector_prop_names[IGT_NUM_CONNECTOR_PROPS] = {
      [IGT_CONNECTOR_BROADCAST_RGB] = "Broadcast RGB",
      [IGT_CONNECTOR_CONTENT_PROTECTION] = "Content Protection",
      [IGT_CONNECTOR_VRR_CAPABLE] = "vrr_capable",
+     [IGT_CONNECTOR_WRITEBACK_PIXEL_FORMATS] = "WRITEBACK_PIXEL_FORMATS",
+     [IGT_CONNECTOR_WRITEBACK_FB_ID] = "WRITEBACK_FB_ID",
+     [IGT_CONNECTOR_WRITEBACK_OUT_FENCE_PTR] = "WRITEBACK_OUT_FENCE_PTR",
 };
 
 /*
@@ -557,6 +560,7 @@ static const struct type_name connector_type_names[] = {
      { DRM_MODE_CONNECTOR_VIRTUAL, "Virtual" },
      { DRM_MODE_CONNECTOR_DSI, "DSI" },
      { DRM_MODE_CONNECTOR_DPI, "DPI" },
+     { DRM_MODE_CONNECTOR_WRITEBACK, "Writeback" },
      {}
 };
 
@@ -1889,6 +1893,12 @@ static void igt_output_reset(igt_output_t *output)
      if (igt_output_has_prop(output, IGT_CONNECTOR_BROADCAST_RGB))
              igt_output_set_prop_value(output, IGT_CONNECTOR_BROADCAST_RGB,
                                        BROADCAST_RGB_FULL);
+     if (igt_output_has_prop(output, IGT_CONNECTOR_WRITEBACK_FB_ID))
+             igt_output_set_prop_value(output, IGT_CONNECTOR_WRITEBACK_FB_ID, 0);
+     if (igt_output_has_prop(output, IGT_CONNECTOR_WRITEBACK_OUT_FENCE_PTR)) {
+             igt_output_clear_prop_changed(output, IGT_CONNECTOR_WRITEBACK_OUT_FENCE_PTR);
+             output->writeback_out_fence_fd = -1;
+     }
 }
 
 /**
@@ -1901,6 +1911,8 @@ static void igt_output_reset(igt_output_t *output)
  * For outputs:
  * - %IGT_CONNECTOR_CRTC_ID
  * - %IGT_CONNECTOR_BROADCAST_RGB (if applicable)
+ * - %IGT_CONNECTOR_WRITEBACK_FB_ID
+ * - %IGT_CONNECTOR_WRITEBACK_OUT_FENCE_PTR
  * - igt_output_override_mode() to default.
  *
  * For pipes:
@@ -1970,6 +1982,8 @@ void igt_display_require(igt_display_t *display, int drm_fd)
 
      display->drm_fd = drm_fd;
 
+     drmSetClientCap(drm_fd, DRM_CLIENT_CAP_WRITEBACK_CONNECTORS, 1);
+
      resources = drmModeGetResources(display->drm_fd);
      if (!resources)
              goto out;
@@ -2263,6 +2277,11 @@ static void igt_output_fini(igt_output_t *output)
      kmstest_free_connector_config(&output->config);
      free(output->name);
      output->name = NULL;
+
+     if (output->writeback_out_fence_fd != -1) {
+             close(output->writeback_out_fence_fd);
+             output->writeback_out_fence_fd = -1;
+     }
 }
 
 /**
@@ -3325,6 +3344,11 @@ static void igt_atomic_prepare_connector_commit(igt_output_t *output, drmModeAto
                                        output->props[i],
                                        output->values[i]));
      }
+
+     if (output->writeback_out_fence_fd != -1) {
+             close(output->writeback_out_fence_fd);
+             output->writeback_out_fence_fd = -1;
+     }
 }
 
 /*
@@ -3447,6 +3471,16 @@ display_commit_changed(igt_display_t *display, enum igt_commit_style s)
              else
                      /* no modeset in universal commit, no change to crtc. */
                      output->changed &= 1 << IGT_CONNECTOR_CRTC_ID;
+
+             if (s == COMMIT_ATOMIC) {
+                     if (igt_output_is_prop_changed(output, IGT_CONNECTOR_WRITEBACK_OUT_FENCE_PTR))
+                             igt_assert(output->writeback_out_fence_fd >= 0);
+
+                     output->values[IGT_CONNECTOR_WRITEBACK_OUT_FENCE_PTR] = 0;
+                     output->values[IGT_CONNECTOR_WRITEBACK_FB_ID] = 0;
+                     igt_output_clear_prop_changed(output, IGT_CONNECTOR_WRITEBACK_FB_ID);
+                     igt_output_clear_prop_changed(output, IGT_CONNECTOR_WRITEBACK_OUT_FENCE_PTR);
+             }
      }
 
      if (display->first_commit) {
@@ -4119,6 +4153,29 @@ void igt_pipe_request_out_fence(igt_pipe_t *pipe)
      igt_pipe_obj_set_prop_value(pipe, IGT_CRTC_OUT_FENCE_PTR, (ptrdiff_t)&pipe->out_fence_fd);
 }
 
+/**
+ * igt_output_set_writeback_fb:
+ * @output: Target output
+ * @fb: Target framebuffer
+ *
+ * This function sets the given @fb to be used as the target framebuffer for the
+ * writeback engine at the next atomic commit. It will also request a writeback
+ * out fence that will contain the fd number of the out fence created by KMS if
+ * the given @fb is valid.
+ */
+void igt_output_set_writeback_fb(igt_output_t *output, struct igt_fb *fb)
+{
+     igt_display_t *display = output->display;
+
+     LOG(display, "%s: output_set_writeback_fb(%d)\n", output->name, fb ? fb->fb_id : 0);
+
+     igt_output_set_prop_value(output, IGT_CONNECTOR_WRITEBACK_FB_ID, fb ? fb->fb_id : 0);
+     /* only request a writeback out fence if the framebuffer is valid */
+     if (fb)
+             igt_output_set_prop_value(output, IGT_CONNECTOR_WRITEBACK_OUT_FENCE_PTR,
+                                       (ptrdiff_t)&output->writeback_out_fence_fd);
+}
+
 /**
  * igt_wait_for_vblank_count:
  * @drm_fd: A drm file descriptor
lib/igt_kms.h +6 -1
index a448a003..cacc6b90 100644
@@ -123,6 +123,9 @@ enum igt_atomic_connector_properties {
        IGT_CONNECTOR_BROADCAST_RGB,
        IGT_CONNECTOR_CONTENT_PROTECTION,
        IGT_CONNECTOR_VRR_CAPABLE,
+       IGT_CONNECTOR_WRITEBACK_PIXEL_FORMATS,
+       IGT_CONNECTOR_WRITEBACK_FB_ID,
+       IGT_CONNECTOR_WRITEBACK_OUT_FENCE_PTR,
        IGT_NUM_CONNECTOR_PROPS
 };
 
@@ -364,6 +367,8 @@ typedef struct {
      bool use_override_mode;
      drmModeModeInfo override_mode;
 
+     int32_t writeback_out_fence_fd;
+
      /* bitmask of changed properties */
      uint64_t changed;
 
@@ -414,6 +419,7 @@ igt_plane_t *igt_output_get_plane_type_index(igt_output_t *output,
 igt_output_t *igt_output_from_connector(igt_display_t *display,
     drmModeConnector *connector);
 const drmModeModeInfo *igt_std_1024_mode_get(void);
+void igt_output_set_writeback_fb(igt_output_t *output, struct igt_fb *fb);
 
 igt_plane_t *igt_pipe_get_plane_type(igt_pipe_t *pipe, int plane_type);
 int igt_pipe_count_plane_type(igt_pipe_t *pipe, int plane_type);
--

2.21.0

Quoted 3 lines

--
Rodrigo Siqueira
https://siqueira.tech

--
====================

Quoted 4 lines
I would like to |
fix the world,  |
but they're not |
giving me the   |

\ source code! /
---------------

    ¯\_(ツ)_/¯

_______________________________________________
igt-dev mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/igt-dev

#7 lore raw

Re: [igt-dev] [PATCH V6 i-g-t 1/6] lib/igt_kms: Add writeback support

Quoted 42 lines · Simon
Quoted 27 lines · Rodrigo Siqueira
Quoted 17 lines · Simon
Quoted 16 lines · Rodrigo Siqueira
Quoted 4 lines · Liviu Dudau
Quoted 3 lines · Brian Starkey

Add support in igt_kms for writeback connectors, with the ability to attach framebuffers.

Quoted 8 lines
Quoted 2 lines

v5: Rebase and add DRM_CLIENT_CAP_WRITEBACK_CONNECTORS before drmModeGetResources()

Thanks for updating this! Given that I have done the last changes and this is mostly a refresh, not sure if I should add more Reviewed-by's to the other patches.

Thanks Liviu!

Quoted 8 lines
Quoted 2 lines

I just forgot to add my SoB, and for some reason, gmail does not allow me to send an email on someone behalf.

FWIW, that's a good thing, and it's required to pass DMARC checks.

Instead, git-send-email should add a From line at the beginning of the message when sending a patch on behalf of someone else. I wonder what happened here.

Quoted 7 lines

Thank you for your help.

I’m using neomutt for sending patches, and I’ll take a look at git-send-email. Additionally, it looks like Gmail requires that I add a new account in order to allow me to send patches on someone behalf. I’ll take some time to read about this issue, and I will try to resend the patchset again.

When sending a patch on someone else's behalf, you shouldn't send the e-mail with the sender (From header) set to someone else. You should use your normal address. The From line in the body of the message will make Git understand the author is someone else.

Quoted 6 lines

So no new account setup required.

I really recommend using git-send-email for patches. There are too many ways to make a mistake if you try sending emails manually. Here is a tutorial:
https://git-send-email.io/

Thank you very much for the tutorial! I'll read it and prepare to migrate from 'neomutt -H' to git-send-mail.

Quoted 3 lines
Quoted 1 line

Btw, if you have time, could you take a look in this series?

Sure, I've already began looking at the series. I'll continue soon.

Thank you again for your help.

Quoted 193 lines

Thanks for your work!

Quoted 191 lines

Best regards

Quoted 187 lines
Quoted 187 lines

Btw, I can fix it after
everybody agrees that the kms_writeback is ready for landing.

Quoted 183 lines

Best regards,
Liviu

Quoted 167 lines
[rebased and updated to the latest igt style]
---
 lib/igt_kms.c | 57 +++++++++++++++++++++++++++++++++++++++++++++++++++
 lib/igt_kms.h |  6 ++++++
 2 files changed, 63 insertions(+)
lib/igt_kms.c +57 -0
index da188a39..140db346 100644
@@ -325,6 +325,9 @@ const char * const igt_connector_prop_names[IGT_NUM_CONNECTOR_PROPS] = {
      [IGT_CONNECTOR_BROADCAST_RGB] = "Broadcast RGB",
      [IGT_CONNECTOR_CONTENT_PROTECTION] = "Content Protection",
      [IGT_CONNECTOR_VRR_CAPABLE] = "vrr_capable",
+     [IGT_CONNECTOR_WRITEBACK_PIXEL_FORMATS] = "WRITEBACK_PIXEL_FORMATS",
+     [IGT_CONNECTOR_WRITEBACK_FB_ID] = "WRITEBACK_FB_ID",
+     [IGT_CONNECTOR_WRITEBACK_OUT_FENCE_PTR] = "WRITEBACK_OUT_FENCE_PTR",
 };
 
 /*
@@ -557,6 +560,7 @@ static const struct type_name connector_type_names[] = {
      { DRM_MODE_CONNECTOR_VIRTUAL, "Virtual" },
      { DRM_MODE_CONNECTOR_DSI, "DSI" },
      { DRM_MODE_CONNECTOR_DPI, "DPI" },
+     { DRM_MODE_CONNECTOR_WRITEBACK, "Writeback" },
      {}
 };
 
@@ -1889,6 +1893,12 @@ static void igt_output_reset(igt_output_t *output)
      if (igt_output_has_prop(output, IGT_CONNECTOR_BROADCAST_RGB))
              igt_output_set_prop_value(output, IGT_CONNECTOR_BROADCAST_RGB,
                                        BROADCAST_RGB_FULL);
+     if (igt_output_has_prop(output, IGT_CONNECTOR_WRITEBACK_FB_ID))
+             igt_output_set_prop_value(output, IGT_CONNECTOR_WRITEBACK_FB_ID, 0);
+     if (igt_output_has_prop(output, IGT_CONNECTOR_WRITEBACK_OUT_FENCE_PTR)) {
+             igt_output_clear_prop_changed(output, IGT_CONNECTOR_WRITEBACK_OUT_FENCE_PTR);
+             output->writeback_out_fence_fd = -1;
+     }
 }
 
 /**
@@ -1901,6 +1911,8 @@ static void igt_output_reset(igt_output_t *output)
  * For outputs:
  * - %IGT_CONNECTOR_CRTC_ID
  * - %IGT_CONNECTOR_BROADCAST_RGB (if applicable)
+ * - %IGT_CONNECTOR_WRITEBACK_FB_ID
+ * - %IGT_CONNECTOR_WRITEBACK_OUT_FENCE_PTR
  * - igt_output_override_mode() to default.
  *
  * For pipes:
@@ -1970,6 +1982,8 @@ void igt_display_require(igt_display_t *display, int drm_fd)
 
      display->drm_fd = drm_fd;
 
+     drmSetClientCap(drm_fd, DRM_CLIENT_CAP_WRITEBACK_CONNECTORS, 1);
+
      resources = drmModeGetResources(display->drm_fd);
      if (!resources)
              goto out;
@@ -2263,6 +2277,11 @@ static void igt_output_fini(igt_output_t *output)
      kmstest_free_connector_config(&output->config);
      free(output->name);
      output->name = NULL;
+
+     if (output->writeback_out_fence_fd != -1) {
+             close(output->writeback_out_fence_fd);
+             output->writeback_out_fence_fd = -1;
+     }
 }
 
 /**
@@ -3325,6 +3344,11 @@ static void igt_atomic_prepare_connector_commit(igt_output_t *output, drmModeAto
                                        output->props[i],
                                        output->values[i]));
      }
+
+     if (output->writeback_out_fence_fd != -1) {
+             close(output->writeback_out_fence_fd);
+             output->writeback_out_fence_fd = -1;
+     }
 }
 
 /*
@@ -3447,6 +3471,16 @@ display_commit_changed(igt_display_t *display, enum igt_commit_style s)
              else
                      /* no modeset in universal commit, no change to crtc. */
                      output->changed &= 1 << IGT_CONNECTOR_CRTC_ID;
+
+             if (s == COMMIT_ATOMIC) {
+                     if (igt_output_is_prop_changed(output, IGT_CONNECTOR_WRITEBACK_OUT_FENCE_PTR))
+                             igt_assert(output->writeback_out_fence_fd >= 0);
+
+                     output->values[IGT_CONNECTOR_WRITEBACK_OUT_FENCE_PTR] = 0;
+                     output->values[IGT_CONNECTOR_WRITEBACK_FB_ID] = 0;
+                     igt_output_clear_prop_changed(output, IGT_CONNECTOR_WRITEBACK_FB_ID);
+                     igt_output_clear_prop_changed(output, IGT_CONNECTOR_WRITEBACK_OUT_FENCE_PTR);
+             }
      }
 
      if (display->first_commit) {
@@ -4119,6 +4153,29 @@ void igt_pipe_request_out_fence(igt_pipe_t *pipe)
      igt_pipe_obj_set_prop_value(pipe, IGT_CRTC_OUT_FENCE_PTR, (ptrdiff_t)&pipe->out_fence_fd);
 }
 
+/**
+ * igt_output_set_writeback_fb:
+ * @output: Target output
+ * @fb: Target framebuffer
+ *
+ * This function sets the given @fb to be used as the target framebuffer for the
+ * writeback engine at the next atomic commit. It will also request a writeback
+ * out fence that will contain the fd number of the out fence created by KMS if
+ * the given @fb is valid.
+ */
+void igt_output_set_writeback_fb(igt_output_t *output, struct igt_fb *fb)
+{
+     igt_display_t *display = output->display;
+
+     LOG(display, "%s: output_set_writeback_fb(%d)\n", output->name, fb ? fb->fb_id : 0);
+
+     igt_output_set_prop_value(output, IGT_CONNECTOR_WRITEBACK_FB_ID, fb ? fb->fb_id : 0);
+     /* only request a writeback out fence if the framebuffer is valid */
+     if (fb)
+             igt_output_set_prop_value(output, IGT_CONNECTOR_WRITEBACK_OUT_FENCE_PTR,
+                                       (ptrdiff_t)&output->writeback_out_fence_fd);
+}
+
 /**
  * igt_wait_for_vblank_count:
  * @drm_fd: A drm file descriptor
lib/igt_kms.h +6 -1
index a448a003..cacc6b90 100644
@@ -123,6 +123,9 @@ enum igt_atomic_connector_properties {
        IGT_CONNECTOR_BROADCAST_RGB,
        IGT_CONNECTOR_CONTENT_PROTECTION,
        IGT_CONNECTOR_VRR_CAPABLE,
+       IGT_CONNECTOR_WRITEBACK_PIXEL_FORMATS,
+       IGT_CONNECTOR_WRITEBACK_FB_ID,
+       IGT_CONNECTOR_WRITEBACK_OUT_FENCE_PTR,
        IGT_NUM_CONNECTOR_PROPS
 };
 
@@ -364,6 +367,8 @@ typedef struct {
      bool use_override_mode;
      drmModeModeInfo override_mode;
 
+     int32_t writeback_out_fence_fd;
+
      /* bitmask of changed properties */
      uint64_t changed;
 
@@ -414,6 +419,7 @@ igt_plane_t *igt_output_get_plane_type_index(igt_output_t *output,
 igt_output_t *igt_output_from_connector(igt_display_t *display,
     drmModeConnector *connector);
 const drmModeModeInfo *igt_std_1024_mode_get(void);
+void igt_output_set_writeback_fb(igt_output_t *output, struct igt_fb *fb);
 
 igt_plane_t *igt_pipe_get_plane_type(igt_pipe_t *pipe, int plane_type);
 int igt_pipe_count_plane_type(igt_pipe_t *pipe, int plane_type);
--

2.21.0

Quoted 3 lines

--
Rodrigo Siqueira
https://siqueira.tech

--
====================

Quoted 4 lines
I would like to |
fix the world,  |
but they're not |
giving me the   |

\ source code! /
---------------

    ¯\_(ツ)_/¯
Signature · 6 lines
Rodrigo Siqueira
https://siqueira.tech
_______________________________________________
igt-dev mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/igt-dev
Reviewed-by on v6 1/6
#8 lore raw

Re: [igt-dev] [PATCH V6 i-g-t 1/6] lib/igt_kms: Add writeback support

In reply to Brian Starkey
Quoted 9 lines · Brian Starkey

Add support in igt_kms for writeback connectors, with the ability to attach framebuffers.

v5: Rebase and add DRM_CLIENT_CAP_WRITEBACK_CONNECTORS before drmModeGetResources()

[rebased and updated to the latest igt style]

This patch LGTM.

Quoted 168 lines
---
 lib/igt_kms.c | 57 +++++++++++++++++++++++++++++++++++++++++++++++++++
 lib/igt_kms.h |  6 ++++++
 2 files changed, 63 insertions(+)
lib/igt_kms.c +57 -0
index da188a39..140db346 100644
@@ -325,6 +325,9 @@ const char * const igt_connector_prop_names[IGT_NUM_CONNECTOR_PROPS] = {
 	[IGT_CONNECTOR_BROADCAST_RGB] = "Broadcast RGB",
 	[IGT_CONNECTOR_CONTENT_PROTECTION] = "Content Protection",
 	[IGT_CONNECTOR_VRR_CAPABLE] = "vrr_capable",
+	[IGT_CONNECTOR_WRITEBACK_PIXEL_FORMATS] = "WRITEBACK_PIXEL_FORMATS",
+	[IGT_CONNECTOR_WRITEBACK_FB_ID] = "WRITEBACK_FB_ID",
+	[IGT_CONNECTOR_WRITEBACK_OUT_FENCE_PTR] = "WRITEBACK_OUT_FENCE_PTR",
 };
 
 /*
@@ -557,6 +560,7 @@ static const struct type_name connector_type_names[] = {
 	{ DRM_MODE_CONNECTOR_VIRTUAL, "Virtual" },
 	{ DRM_MODE_CONNECTOR_DSI, "DSI" },
 	{ DRM_MODE_CONNECTOR_DPI, "DPI" },
+	{ DRM_MODE_CONNECTOR_WRITEBACK, "Writeback" },
 	{}
 };
 
@@ -1889,6 +1893,12 @@ static void igt_output_reset(igt_output_t *output)
 	if (igt_output_has_prop(output, IGT_CONNECTOR_BROADCAST_RGB))
 		igt_output_set_prop_value(output, IGT_CONNECTOR_BROADCAST_RGB,
 					  BROADCAST_RGB_FULL);
+	if (igt_output_has_prop(output, IGT_CONNECTOR_WRITEBACK_FB_ID))
+		igt_output_set_prop_value(output, IGT_CONNECTOR_WRITEBACK_FB_ID, 0);
+	if (igt_output_has_prop(output, IGT_CONNECTOR_WRITEBACK_OUT_FENCE_PTR)) {
+		igt_output_clear_prop_changed(output, IGT_CONNECTOR_WRITEBACK_OUT_FENCE_PTR);
+		output->writeback_out_fence_fd = -1;
+	}
 }
 
 /**
@@ -1901,6 +1911,8 @@ static void igt_output_reset(igt_output_t *output)
  * For outputs:
  * - %IGT_CONNECTOR_CRTC_ID
  * - %IGT_CONNECTOR_BROADCAST_RGB (if applicable)
+ * - %IGT_CONNECTOR_WRITEBACK_FB_ID
+ * - %IGT_CONNECTOR_WRITEBACK_OUT_FENCE_PTR
  * - igt_output_override_mode() to default.
  *
  * For pipes:
@@ -1970,6 +1982,8 @@ void igt_display_require(igt_display_t *display, int drm_fd)
 
 	display->drm_fd = drm_fd;
 
+	drmSetClientCap(drm_fd, DRM_CLIENT_CAP_WRITEBACK_CONNECTORS, 1);
+
 	resources = drmModeGetResources(display->drm_fd);
 	if (!resources)
 		goto out;
@@ -2263,6 +2277,11 @@ static void igt_output_fini(igt_output_t *output)
 	kmstest_free_connector_config(&output->config);
 	free(output->name);
 	output->name = NULL;
+
+	if (output->writeback_out_fence_fd != -1) {
+		close(output->writeback_out_fence_fd);
+		output->writeback_out_fence_fd = -1;
+	}
 }
 
 /**
@@ -3325,6 +3344,11 @@ static void igt_atomic_prepare_connector_commit(igt_output_t *output, drmModeAto
 					  output->props[i],
 					  output->values[i]));
 	}
+
+	if (output->writeback_out_fence_fd != -1) {
+		close(output->writeback_out_fence_fd);
+		output->writeback_out_fence_fd = -1;
+	}
 }
 
 /*
@@ -3447,6 +3471,16 @@ display_commit_changed(igt_display_t *display, enum igt_commit_style s)
 		else
 			/* no modeset in universal commit, no change to crtc. */
 			output->changed &= 1 << IGT_CONNECTOR_CRTC_ID;
+
+		if (s == COMMIT_ATOMIC) {
+			if (igt_output_is_prop_changed(output, IGT_CONNECTOR_WRITEBACK_OUT_FENCE_PTR))
+				igt_assert(output->writeback_out_fence_fd >= 0);
+
+			output->values[IGT_CONNECTOR_WRITEBACK_OUT_FENCE_PTR] = 0;
+			output->values[IGT_CONNECTOR_WRITEBACK_FB_ID] = 0;
+			igt_output_clear_prop_changed(output, IGT_CONNECTOR_WRITEBACK_FB_ID);
+			igt_output_clear_prop_changed(output, IGT_CONNECTOR_WRITEBACK_OUT_FENCE_PTR);
+		}
 	}
 
 	if (display->first_commit) {
@@ -4119,6 +4153,29 @@ void igt_pipe_request_out_fence(igt_pipe_t *pipe)
 	igt_pipe_obj_set_prop_value(pipe, IGT_CRTC_OUT_FENCE_PTR, (ptrdiff_t)&pipe->out_fence_fd);
 }
 
+/**
+ * igt_output_set_writeback_fb:
+ * @output: Target output
+ * @fb: Target framebuffer
+ *
+ * This function sets the given @fb to be used as the target framebuffer for the
+ * writeback engine at the next atomic commit. It will also request a writeback
+ * out fence that will contain the fd number of the out fence created by KMS if
+ * the given @fb is valid.
+ */
+void igt_output_set_writeback_fb(igt_output_t *output, struct igt_fb *fb)
+{
+	igt_display_t *display = output->display;
+
+	LOG(display, "%s: output_set_writeback_fb(%d)\n", output->name, fb ? fb->fb_id : 0);
+
+	igt_output_set_prop_value(output, IGT_CONNECTOR_WRITEBACK_FB_ID, fb ? fb->fb_id : 0);
+	/* only request a writeback out fence if the framebuffer is valid */
+	if (fb)
+		igt_output_set_prop_value(output, IGT_CONNECTOR_WRITEBACK_OUT_FENCE_PTR,
+					  (ptrdiff_t)&output->writeback_out_fence_fd);
+}
+
 /**
  * igt_wait_for_vblank_count:
  * @drm_fd: A drm file descriptor
lib/igt_kms.h +6 -0
index a448a003..cacc6b90 100644
@@ -123,6 +123,9 @@ enum igt_atomic_connector_properties {
        IGT_CONNECTOR_BROADCAST_RGB,
        IGT_CONNECTOR_CONTENT_PROTECTION,
        IGT_CONNECTOR_VRR_CAPABLE,
+       IGT_CONNECTOR_WRITEBACK_PIXEL_FORMATS,
+       IGT_CONNECTOR_WRITEBACK_FB_ID,
+       IGT_CONNECTOR_WRITEBACK_OUT_FENCE_PTR,
        IGT_NUM_CONNECTOR_PROPS
 };
 
@@ -364,6 +367,8 @@ typedef struct {
 	bool use_override_mode;
 	drmModeModeInfo override_mode;
 
+	int32_t writeback_out_fence_fd;
+
 	/* bitmask of changed properties */
 	uint64_t changed;
 
@@ -414,6 +419,7 @@ igt_plane_t *igt_output_get_plane_type_index(igt_output_t *output,
 igt_output_t *igt_output_from_connector(igt_display_t *display,
     drmModeConnector *connector);
 const drmModeModeInfo *igt_std_1024_mode_get(void);
+void igt_output_set_writeback_fb(igt_output_t *output, struct igt_fb *fb);
 
 igt_plane_t *igt_pipe_get_plane_type(igt_pipe_t *pipe, int plane_type);
 int igt_pipe_count_plane_type(igt_pipe_t *pipe, int plane_type);
Signature · 7 lines
2.21.0


_______________________________________________
igt-dev mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/igt-dev

_______________________________________________
igt-dev mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/igt-dev

#10 lore raw

Re: [igt-dev] [PATCH V6 i-g-t 2/6] kms_writeback: Add initial writeback tests

In reply to Brian Starkey

Hi,

Thanks for the patch! Here are a few comments.

For bonus points, it would be nice to add igt_describe descriptions of each sub-test.

Quoted 114 lines · Brian Starkey

Add tests for the WRITEBACK_PIXEL_FORMATS, WRITEBACK_OUT_FENCE_PTR and WRITEBACK_FB_ID properties on writeback connectors, ensuring their behaviour is correct.

[rebased and updated do_writeback_test() function to address feedback]
---
 tests/Makefile.sources |   1 +
 tests/kms_writeback.c  | 314 +++++++++++++++++++++++++++++++++++++++++
 tests/meson.build      |   1 +
 3 files changed, 316 insertions(+)
 create mode 100644 tests/kms_writeback.c
tests/Makefile.sources +1 -0
index 027ed82f..03cc8efa 100644
@@ -77,6 +77,7 @@ TESTS_progs = \
 	kms_universal_plane \
 	kms_vblank \
 	kms_vrr \
+	kms_writeback \
 	meta_test \
 	perf \
 	perf_pmu \
tests/kms_writeback.c new file +82 -0
new file mode 100644
index 00000000..66ef48a6
@@ -0,0 +1,314 @@
+/*
+ * (C) COPYRIGHT 2017 ARM Limited. All rights reserved.
+ *
+ * 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.
+ *
+ */
+
+#include <errno.h>
+#include <stdbool.h>
+#include <stdio.h>
+#include <string.h>
+
+#include "igt.h"
+#include "igt_core.h"
+#include "igt_fb.h"
+
+static drmModePropertyBlobRes *get_writeback_formats_blob(igt_output_t *output)
+{
+	drmModePropertyBlobRes *blob = NULL;
+	uint64_t blob_id;
+	int ret;
+
+	ret = kmstest_get_property(output->display->drm_fd,
+				   output->config.connector->connector_id,
+				   DRM_MODE_OBJECT_CONNECTOR,
+				   igt_connector_prop_names[IGT_CONNECTOR_WRITEBACK_PIXEL_FORMATS],
+				   NULL, &blob_id, NULL);
+	if (ret)
+		blob = drmModeGetPropertyBlob(output->display->drm_fd, blob_id);
+
+	igt_assert(blob);
+
+	return blob;
+}
+
+static bool check_writeback_config(igt_display_t *display, igt_output_t *output)
+{
+	igt_fb_t input_fb, output_fb;
+	igt_plane_t *plane;
+	uint32_t writeback_format = DRM_FORMAT_XRGB8888;
+	uint64_t tiling = igt_fb_mod_to_tiling(0);
+	int width, height, ret;
+	drmModeModeInfo override_mode = {
+		.clock = 25175,
+		.hdisplay = 640,
+		.hsync_start = 656,
+		.hsync_end = 752,
+		.htotal = 800,
+		.hskew = 0,
+		.vdisplay = 480,
+		.vsync_start = 490,
+		.vsync_end = 492,
+		.vtotal = 525,
+		.vscan = 0,
+		.vrefresh = 60,
+		.flags = DRM_MODE_FLAG_NHSYNC | DRM_MODE_FLAG_NVSYNC,
+		.name = {"640x480-60"},
+	};
+	igt_output_override_mode(output, &override_mode);
+
+	width = override_mode.hdisplay;
+	height = override_mode.vdisplay;
+
+	ret = igt_create_fb(display->drm_fd, width, height, DRM_FORMAT_XRGB8888, tiling, &input_fb);
+	igt_assert(ret >= 0);

tests/kms_writeback.c line 92

Quoted 10 lines
tests/kms_writeback.c new file +10 -0
+
+	ret = igt_create_fb(display->drm_fd, width, height, writeback_format, tiling, &output_fb);
+	igt_assert(ret >= 0);
+
+	plane = igt_output_get_plane_type(output, DRM_PLANE_TYPE_PRIMARY);
+	igt_plane_set_fb(plane, &input_fb);
+	igt_output_set_writeback_fb(output, &output_fb);
+
+	ret = igt_display_try_commit_atomic(display, DRM_MODE_ATOMIC_TEST_ONLY |
+					    DRM_MODE_ATOMIC_ALLOW_MODESET, NULL);

Okay, we're using atomic test-only mode to know if we can perform tests with the writeback output. It's probably fine, but we don't use WRITEBACK_PIXEL_FORMATS, which makes me a little bit sad.

Quoted 9 lines
tests/kms_writeback.c new file +9 -0
+	igt_plane_set_fb(plane, NULL);
+	igt_remove_fb(display->drm_fd, &input_fb);
+	igt_remove_fb(display->drm_fd, &output_fb);
+
+	return !ret;
+}
+
+static igt_output_t *kms_writeback_get_output(igt_display_t *display)
+{

tests/kms_writeback.c line 111

Quoted 10 lines
tests/kms_writeback.c new file +10 -0
+	int i;
+
+	for (i = 0; i < display->n_outputs; i++) {
+		igt_output_t *output = &display->outputs[i];
+		int j;
+
+		if (output->config.connector->connector_type != DRM_MODE_CONNECTOR_WRITEBACK)
+			continue;
+
+		kmstest_force_connector(display->drm_fd, output->config.connector, FORCE_CONNECTOR_ON);

Hmm. Is this really necessary? "Real" userspace won't do this, so I don't think we need to either.

tests/kms_writeback.c line 121

Quoted 9 lines
tests/kms_writeback.c new file +9 -0
+		for (j = 0; j < igt_display_get_n_pipes(display); j++) {
+			igt_output_set_pipe(output, j);
+
+			if (check_writeback_config(display, output)) {
+				igt_debug("Using connector %u:%s on pipe %d\n",
+					  output->config.connector->connector_id,
+					  output->name, j);
+				return output;
+			}

We could probably add an igt_debug statement in case we don't use this writeback output.

Quoted 10 lines
tests/kms_writeback.c new file +10 -0
+		}
+
+		/* Restore any connectors we don't use, so we don't trip on them later */
+		kmstest_force_connector(display->drm_fd, output->config.connector, FORCE_CONNECTOR_UNSPECIFIED);
+	}
+
+	return NULL;
+}
+
+static void check_writeback_fb_id(igt_output_t *output)

tests/kms_writeback.c line 141

Quoted 10 lines
tests/kms_writeback.c new file +10 -0
+{
+	uint64_t check_fb_id;
+
+	check_fb_id = igt_output_get_prop(output, IGT_CONNECTOR_WRITEBACK_FB_ID);
+	igt_assert(check_fb_id == 0);
+}
+
+static int do_writeback_test(igt_output_t *output, uint32_t flags,
+			      uint32_t fb_id, int32_t *out_fence_ptr,
+			      bool ptr_valid)

flags seems to always be set to DRM_MODE_ATOMIC_ALLOW_MODESET. We can probably remove the parameter from this function.

Quoted 6 lines
tests/kms_writeback.c new file +6 -0
+{
+	int ret;
+	igt_display_t *display = output->display;
+	struct kmstest_connector_config *config = &output->config;
+
+	igt_output_set_prop_value(output, IGT_CONNECTOR_CRTC_ID, config->crtc->crtc_id);

tests/kms_writeback.c line 157

Quoted 10 lines
tests/kms_writeback.c new file +10 -0
+	igt_output_set_prop_value(output, IGT_CONNECTOR_WRITEBACK_FB_ID, fb_id);
+	igt_output_set_prop_value(output, IGT_CONNECTOR_WRITEBACK_OUT_FENCE_PTR, (uint64_t)out_fence_ptr);
+
+	if (ptr_valid)
+		*out_fence_ptr = 0;
+
+	ret = igt_display_try_commit_atomic(display, flags, NULL);
+
+	if (ptr_valid)
+		igt_assert(*out_fence_ptr == -1);

I'm confused. Why should this be -1 even if we
igt_display_try_commit_atomic succeeds?

Quoted 17 lines
tests/kms_writeback.c new file +17 -0
+	/* WRITEBACK_FB_ID must always read as zero */
+	check_writeback_fb_id(output);
+
+	return ret;
+}
+
+static void invalid_out_fence(igt_output_t *output, igt_fb_t *valid_fb, igt_fb_t *invalid_fb)
+{
+	int i, ret;
+	int32_t out_fence;
+	struct {
+		uint32_t fb_id;
+		bool ptr_valid;
+		int32_t *out_fence_ptr;
+	} invalid_tests[] = {
+		{
+			/* No output buffer, but the WRITEBACK_OUT_FENCE_PTR set. */

tests/kms_writeback.c line 185

Quoted 10 lines
tests/kms_writeback.c new file +10 -0
+			.fb_id = 0,
+			.ptr_valid = true,
+			.out_fence_ptr = &out_fence,
+		},
+		{
+			/* Invalid output buffer. */
+			.fb_id = invalid_fb->fb_id,
+			.ptr_valid = true,
+			.out_fence_ptr = &out_fence,
+		},

This doesn't belong in this function (invalid_out_fence), since this checks an invalid framebuffer, not an invalid fence. We should probably move it to writeback_fb_id (and rename that function to test_fb?).

Quoted 4 lines
tests/kms_writeback.c new file +4 -0
+		{
+			/* Invalid WRITEBACK_OUT_FENCE_PTR. */
+			.fb_id = valid_fb->fb_id,
+			.ptr_valid = false,

tests/kms_writeback.c line 199

Quoted 10 lines
tests/kms_writeback.c new file +10 -0
+			.out_fence_ptr = (int32_t *)0x8,
+		},
+	};
+
+	for (i = 0; i < ARRAY_SIZE(invalid_tests); i++) {
+		ret = do_writeback_test(output, DRM_MODE_ATOMIC_ALLOW_MODESET,
+					invalid_tests[i].fb_id,
+					invalid_tests[i].out_fence_ptr,
+					invalid_tests[i].ptr_valid);
+		igt_assert(ret != 0);

Maybe we can check for -ret == EINVAL?

tests/kms_writeback.c line 203

Quoted 4 lines
tests/kms_writeback.c new file +4 -0
+	}
+}
+
+static void writeback_fb_id(igt_output_t *output, igt_fb_t *valid_fb, igt_fb_t *invalid_fb)

invalid_fb doesn't seem to be used here. valid_fb seems to be set to the input framebuffer. It's probably not a good idea to use the same FB for input and writeback output.

Quoted 41 lines
tests/kms_writeback.c new file +41 -0
+{
+
+	int ret;
+
+	/* Valid output buffer */
+	ret = do_writeback_test(output, DRM_MODE_ATOMIC_ALLOW_MODESET,
+				valid_fb->fb_id, NULL, false);
+	igt_assert(ret == 0);
+
+	/* Invalid object for WRITEBACK_FB_ID */
+	ret = do_writeback_test(output, DRM_MODE_ATOMIC_ALLOW_MODESET,
+				output->id, NULL, false);
+	igt_assert(ret == -EINVAL);
+
+	/* Zero WRITEBACK_FB_ID */
+	ret = do_writeback_test(output, DRM_MODE_ATOMIC_ALLOW_MODESET,
+				0, NULL, false);
+	igt_assert(ret == 0);
+}
+
+igt_main
+{
+	igt_display_t display;
+	igt_output_t *output;
+	igt_plane_t *plane;
+	igt_fb_t input_fb;
+	drmModeModeInfo mode;
+	int ret;
+
+	memset(&display, 0, sizeof(display));
+
+	igt_fixture {
+		display.drm_fd = drm_open_driver_master(DRIVER_ANY);
+		igt_display_require(&display, display.drm_fd);
+
+		kmstest_set_vt_graphics_mode();
+
+		igt_display_require(&display, display.drm_fd);
+
+		igt_require(display.is_atomic);
+

tests/kms_writeback.c line 254

Quoted 10 lines
tests/kms_writeback.c new file +10 -0
+		output = kms_writeback_get_output(&display);
+		igt_require(output);
+
+		if (output->use_override_mode)
+			memcpy(&mode, &output->override_mode, sizeof(mode));
+		else
+			memcpy(&mode, &output->config.default_mode, sizeof(mode));
+
+		plane = igt_output_get_plane_type(output, DRM_PLANE_TYPE_PRIMARY);
+		igt_require(plane);

Maybe we can assert on this?

tests/kms_writeback.c line 259

Quoted 4 lines
tests/kms_writeback.c new file +4 -0
+		ret = igt_create_fb(display.drm_fd, mode.hdisplay,
+				    mode.vdisplay,
+				    DRM_FORMAT_XRGB8888,
+				    igt_fb_mod_to_tiling(0),

This is supposed to take a modifier. DRM_FORMAT_MOD_LINEAR would be better to make this clear.

(Applies to other lines of this patch)

tests/kms_writeback.c line 266

Quoted 7 lines
tests/kms_writeback.c new file +7 -0
+				    &input_fb);
+		igt_assert(ret >= 0);
+		igt_plane_set_fb(plane, &input_fb);
+	}
+
+	igt_subtest("writeback-pixel-formats") {
+		drmModePropertyBlobRes *formats_blob = get_writeback_formats_blob(output);

Need to drmModeFreePropertyBlob this.

Quoted 4 lines
tests/kms_writeback.c new file +4 -0
+		const char *valid_chars = "0123456 ABCGNRUVXY";
+		unsigned int i;
+		char *c;
+

tests/kms_writeback.c line 280

Quoted 10 lines
tests/kms_writeback.c new file +10 -0
+		/*
+		 * We don't have a comprehensive list of formats, so just check
+		 * that the blob length is sensible and that it doesn't contain
+		 * any outlandish characters
+		 */
+		igt_assert(!(formats_blob->length % 4));
+		c = formats_blob->data;
+		for (i = 0; i < formats_blob->length; i++)
+			igt_assert_f(strchr(valid_chars, c[i]),
+				     "Unexpected character %c\n", c[i]);

Honestly, I'm not a fan of this check. There's no guarantee that fourcc codes will be made from ASCII characters, in fact some formats already have non-printable chars in them. I don't want to have to update this test when a new fourcc format is added.

We currently don't have a test for IN_FORMATS. If we really want to check formats, we could have a big array of known formats and add a bool is_valid_format(uint32_t fmt) function.

tests/kms_writeback.c line 284

Quoted 4 lines
tests/kms_writeback.c new file +4 -0
+	}
+
+	igt_subtest("writeback-invalid-out-fence") {
+		igt_fb_t invalid_fb;

invalid_output_fb would be a better name IMHO.

tests/kms_writeback.c line 290

Quoted 6 lines
tests/kms_writeback.c new file +6 -0
+		ret = igt_create_fb(display.drm_fd, mode.hdisplay / 2,
+				    mode.vdisplay / 2,
+				    DRM_FORMAT_XRGB8888,
+				    igt_fb_mod_to_tiling(0),
+				    &invalid_fb);
+		igt_require(ret > 0);

We should probably use a different variable: ret is signed, igt_create_fb isn't. Also ret doesn't make it clear that the return value is the KMS framebuffer ID.

(Applies to other subtests)

tests/kms_writeback.c line 292

Quoted 1 line
tests/kms_writeback.c new file +1 -0
+		invalid_out_fence(output, &input_fb, &invalid_fb);

(Still not sure why we provide the input FB to this function.)

Quoted 33 lines
tests/kms_writeback.c new file +21 -0
+		igt_remove_fb(display.drm_fd, &invalid_fb);
+	}
+
+	igt_subtest("writeback-fb-id") {
+		igt_fb_t output_fb;
+		ret = igt_create_fb(display.drm_fd, mode.hdisplay, mode.vdisplay,
+				    DRM_FORMAT_XRGB8888,
+				    igt_fb_mod_to_tiling(0),
+				    &output_fb);
+		igt_require(ret > 0);
+
+		writeback_fb_id(output, &input_fb, &output_fb);
+
+		igt_remove_fb(display.drm_fd, &output_fb);
+	}
+
+	igt_fixture {
+		igt_remove_fb(display.drm_fd, &input_fb);
+		igt_display_fini(&display);
+	}
+}
tests/meson.build +1 -0
index f168fbba..9c77cfcd 100644
@@ -63,6 +63,7 @@ test_progs = [
 	'kms_universal_plane',
 	'kms_vblank',
 	'kms_vrr',
+	'kms_writeback',
 	'meta_test',
 	'panfrost_get_param',
 	'panfrost_gem_new',
Quoted 4 lines

_______________________________________________
igt-dev mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/igt-dev

_______________________________________________
igt-dev mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/igt-dev

#23 lore raw

Re: [igt-dev] [PATCH V6 i-g-t 3/6] lib: Add function to hash a framebuffer

In reply to Brian Starkey

Mostly LGTM, here are a few nits.

Quoted 35 lines · Brian Starkey

To use writeback buffers as a CRC source, we need to be able to hash them. Implement a simple FVA-1a hashing routine for this purpose.

Doing a bytewise hash on the framebuffer directly can be very slow if the memory is noncached. By making a copy of each line in the FB first (which can take advantage of word-access speedup), we can do the hash on a cached copy, which is much faster (10x speedup on my platform).

v6: use igt_memcpy_from_wc() instead of plain memcpy, as suggested by
Chris Wilson

[rebased and updated to the most recent API]
---
 lib/igt_fb.c | 66 ++++++++++++++++++++++++++++++++++++++++++++++++++++
 lib/igt_fb.h |  3 +++
 2 files changed, 69 insertions(+)
lib/igt_fb.c +8 -0
index 9d4f905e..d07dae39 100644
@@ -3256,6 +3256,72 @@ bool igt_fb_supported_format(uint32_t drm_format)
 	return false;
 }
 
+/*
+ * This implements the FNV-1a hashing algorithm instead of CRC, for
+ * simplicity
+ * http://www.isthe.com/chongo/tech/comp/fnv/index.html
+ *
+ * hash = offset_basis
+ * for each octet_of_data to be hashed
+ *         hash = hash xor octet_of_data

lib/igt_fb.c bool igt_fb_supported_format(uint32_t drm_format) · line 3276

Quoted 10 lines
lib/igt_fb.c +10 -0
+ *         hash = hash * FNV_prime
+ * return hash
+ *
+ * 32 bit offset_basis = 2166136261
+ * 32 bit FNV_prime = 224 + 28 + 0x93 = 16777619
+ */
+int igt_fb_get_crc(struct igt_fb *fb, igt_crc_t *crc)
+{
+#define FNV1a_OFFSET_BIAS 2166136261
+#define FNV1a_PRIME 16777619

I'd just use plain uint32_t variables for those, but no big deal.

lib/igt_fb.c bool igt_fb_supported_format(uint32_t drm_format) · line 3281

Quoted 5 lines
lib/igt_fb.c +5 -0
+	uint32_t hash;
+	void *map;
+	char *ptr, *line = NULL;
+	int x, y, cpp = igt_drm_format_to_bpp(fb->drm_format) / 8;
+	uint32_t stride = calc_plane_stride(fb, 0);

We could return -EINVAL in case fb->num_planes != 1.

lib/igt_fb.c bool igt_fb_supported_format(uint32_t drm_format) · line 3289

Quoted 7 lines
lib/igt_fb.c +7 -0
+	if (fb->is_dumb)
+		map = kmstest_dumb_map_buffer(fb->fd, fb->gem_handle, fb->size,
+					      PROT_READ);
+	else
+		map = gem_mmap__gtt(fb->fd, fb->gem_handle, fb->size,
+				    PROT_READ);
+	ptr = map;

Nit: no need for this, can assign the result of mmap directly to ptr.

Quoted 7 lines
lib/igt_fb.c +7 -0
+
+	/*
+	 * Framebuffers are often uncached, which can make byte-wise accesses
+	 * very slow. We copy each line of the FB into a local buffer to speed
+	 * up the hashing.
+	 */
+	line = malloc(stride);

lib/igt_fb.c bool igt_fb_supported_format(uint32_t drm_format) · line 3306

Quoted 10 lines
lib/igt_fb.c +10 -0
+	if (!line) {
+		munmap(map, fb->size);
+		return -ENOMEM;
+	}
+
+	hash = FNV1a_OFFSET_BIAS;
+
+	for (y = 0; y < fb->height; y++, ptr += stride) {
+
+		igt_memcpy_from_wc(line, ptr, stride);

Nit: no need to copy the whole stride actually, we can just copy fb->width * cpp since we're only going to read that.

Quoted 41 lines
lib/igt_fb.c +18 -0
+
+		for (x = 0; x < fb->width * cpp; x++) {
+			hash ^= line[x];
+			hash *= FNV1a_PRIME;
+		}
+	}
+
+	crc->n_words = 1;
+	crc->crc[0] = hash;
+
+	free(line);
+	munmap(map, fb->size);
+
+	return 0;
+#undef FNV1a_OFFSET_BIAS
+#undef FNV1a_PRIME
+}
+
 /**
  * igt_format_is_yuv:
  * @drm_format: drm fourcc
lib/igt_fb.h +3 -0
index adefebe1..a2741c05 100644
@@ -37,6 +37,7 @@
 #include <i915_drm.h>
 
 #include "igt_color_encoding.h"
+#include "igt_debugfs.h"
 
 /*
  * Internal format to denote a buffer compatible with pixman's
@@ -194,5 +195,7 @@ int igt_format_plane_bpp(uint32_t drm_format, int plane);
 void igt_format_array_fill(uint32_t **formats_array, unsigned int *count,
 			   bool allow_yuv);
 
+int igt_fb_get_crc(struct igt_fb *fb, igt_crc_t *crc);
+
 #endif /* __IGT_FB_H__ */
 
Quoted 4 lines

_______________________________________________
igt-dev mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/igt-dev

_______________________________________________
igt-dev mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/igt-dev

Reviewed-by
#24 lore raw

Re: [igt-dev] [PATCH V6 i-g-t 3/6] lib: Add function to hash a framebuffer

Quoted 51 lines · Simon

Mostly LGTM, here are a few nits.

Quoted 35 lines · Brian Starkey

To use writeback buffers as a CRC source, we need to be able to hash them. Implement a simple FVA-1a hashing routine for this purpose.

Doing a bytewise hash on the framebuffer directly can be very slow if the memory is noncached. By making a copy of each line in the FB first (which can take advantage of word-access speedup), we can do the hash on a cached copy, which is much faster (10x speedup on my platform).

v6: use igt_memcpy_from_wc() instead of plain memcpy, as suggested by
Chris Wilson

[rebased and updated to the most recent API]
---
 lib/igt_fb.c | 66 ++++++++++++++++++++++++++++++++++++++++++++++++++++
 lib/igt_fb.h |  3 +++
 2 files changed, 69 insertions(+)
lib/igt_fb.c +8 -0
index 9d4f905e..d07dae39 100644
@@ -3256,6 +3256,72 @@ bool igt_fb_supported_format(uint32_t drm_format)
 	return false;
 }
 
+/*
+ * This implements the FNV-1a hashing algorithm instead of CRC, for
+ * simplicity
+ * http://www.isthe.com/chongo/tech/comp/fnv/index.html
+ *
+ * hash = offset_basis
+ * for each octet_of_data to be hashed
+ *         hash = hash xor octet_of_data

lib/igt_fb.c bool igt_fb_supported_format(uint32_t drm_format) · line 3276

Quoted 10 lines
lib/igt_fb.c +10 -0
+ *         hash = hash * FNV_prime
+ * return hash
+ *
+ * 32 bit offset_basis = 2166136261
+ * 32 bit FNV_prime = 224 + 28 + 0x93 = 16777619
+ */
+int igt_fb_get_crc(struct igt_fb *fb, igt_crc_t *crc)
+{
+#define FNV1a_OFFSET_BIAS 2166136261
+#define FNV1a_PRIME 16777619

I'd just use plain uint32_t variables for those, but no big deal.

Quoted 7 lines

lib/igt_fb.c bool igt_fb_supported_format(uint32_t drm_format) · line 3281

Quoted 5 lines
lib/igt_fb.c +5 -0
+	uint32_t hash;
+	void *map;
+	char *ptr, *line = NULL;
+	int x, y, cpp = igt_drm_format_to_bpp(fb->drm_format) / 8;
+	uint32_t stride = calc_plane_stride(fb, 0);

We could return -EINVAL in case fb->num_planes != 1.

Let's not waste cycles. With this ^ fixed, this patch is:

Other nits are optional.

Quoted 72 lines

lib/igt_fb.c bool igt_fb_supported_format(uint32_t drm_format) · line 3289

Quoted 7 lines
lib/igt_fb.c +7 -0
+	if (fb->is_dumb)
+		map = kmstest_dumb_map_buffer(fb->fd, fb->gem_handle, fb->size,
+					      PROT_READ);
+	else
+		map = gem_mmap__gtt(fb->fd, fb->gem_handle, fb->size,
+				    PROT_READ);
+	ptr = map;

Nit: no need for this, can assign the result of mmap directly to ptr.

Quoted 7 lines
lib/igt_fb.c +7 -0
+
+	/*
+	 * Framebuffers are often uncached, which can make byte-wise accesses
+	 * very slow. We copy each line of the FB into a local buffer to speed
+	 * up the hashing.
+	 */
+	line = malloc(stride);

lib/igt_fb.c bool igt_fb_supported_format(uint32_t drm_format) · line 3306

Quoted 10 lines
lib/igt_fb.c +10 -0
+	if (!line) {
+		munmap(map, fb->size);
+		return -ENOMEM;
+	}
+
+	hash = FNV1a_OFFSET_BIAS;
+
+	for (y = 0; y < fb->height; y++, ptr += stride) {
+
+		igt_memcpy_from_wc(line, ptr, stride);

Nit: no need to copy the whole stride actually, we can just copy fb->width * cpp since we're only going to read that.

Quoted 41 lines
lib/igt_fb.c +18 -0
+
+		for (x = 0; x < fb->width * cpp; x++) {
+			hash ^= line[x];
+			hash *= FNV1a_PRIME;
+		}
+	}
+
+	crc->n_words = 1;
+	crc->crc[0] = hash;
+
+	free(line);
+	munmap(map, fb->size);
+
+	return 0;
+#undef FNV1a_OFFSET_BIAS
+#undef FNV1a_PRIME
+}
+
 /**
  * igt_format_is_yuv:
  * @drm_format: drm fourcc
lib/igt_fb.h +3 -0
index adefebe1..a2741c05 100644
@@ -37,6 +37,7 @@
 #include <i915_drm.h>
 
 #include "igt_color_encoding.h"
+#include "igt_debugfs.h"
 
 /*
  * Internal format to denote a buffer compatible with pixman's
@@ -194,5 +195,7 @@ int igt_format_plane_bpp(uint32_t drm_format, int plane);
 void igt_format_array_fill(uint32_t **formats_array, unsigned int *count,
 			   bool allow_yuv);
 
+int igt_fb_get_crc(struct igt_fb *fb, igt_crc_t *crc);
+
 #endif /* __IGT_FB_H__ */
 
Quoted 8 lines
Quoted 4 lines

_______________________________________________
igt-dev mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/igt-dev

_______________________________________________
igt-dev mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/igt-dev

_______________________________________________
igt-dev mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/igt-dev

#11 lore raw

Re: [igt-dev] [PATCH V6 i-g-t 2/6] kms_writeback: Add initial writeback tests

In reply to Ser, Simon
Quoted 6 lines · Simon

Hi,

Thanks for the patch! Here are a few comments.

For bonus points, it would be nice to add igt_describe descriptions of each sub-test.

Hi Simon,

First of all, thanks for your feedback; I already applied most of your suggestions. I just have some inline comments/questions.

Quoted 121 lines · Brian Starkey
Quoted 120 lines

Add tests for the WRITEBACK_PIXEL_FORMATS, WRITEBACK_OUT_FENCE_PTR and WRITEBACK_FB_ID properties on writeback connectors, ensuring their behaviour is correct.

[rebased and updated do_writeback_test() function to address feedback]
---
 tests/Makefile.sources |   1 +
 tests/kms_writeback.c  | 314 +++++++++++++++++++++++++++++++++++++++++
 tests/meson.build      |   1 +
 3 files changed, 316 insertions(+)
 create mode 100644 tests/kms_writeback.c
tests/Makefile.sources +1 -0
index 027ed82f..03cc8efa 100644
@@ -77,6 +77,7 @@ TESTS_progs = \
 	kms_universal_plane \
 	kms_vblank \
 	kms_vrr \
+	kms_writeback \
 	meta_test \
 	perf \
 	perf_pmu \
tests/kms_writeback.c new file +88 -0
new file mode 100644
index 00000000..66ef48a6
@@ -0,0 +1,314 @@
+/*
+ * (C) COPYRIGHT 2017 ARM Limited. All rights reserved.
+ *
+ * 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.
+ *
+ */
+
+#include <errno.h>
+#include <stdbool.h>
+#include <stdio.h>
+#include <string.h>
+
+#include "igt.h"
+#include "igt_core.h"
+#include "igt_fb.h"
+
+static drmModePropertyBlobRes *get_writeback_formats_blob(igt_output_t *output)
+{
+	drmModePropertyBlobRes *blob = NULL;
+	uint64_t blob_id;
+	int ret;
+
+	ret = kmstest_get_property(output->display->drm_fd,
+				   output->config.connector->connector_id,
+				   DRM_MODE_OBJECT_CONNECTOR,
+				   igt_connector_prop_names[IGT_CONNECTOR_WRITEBACK_PIXEL_FORMATS],
+				   NULL, &blob_id, NULL);
+	if (ret)
+		blob = drmModeGetPropertyBlob(output->display->drm_fd, blob_id);
+
+	igt_assert(blob);
+
+	return blob;
+}
+
+static bool check_writeback_config(igt_display_t *display, igt_output_t *output)
+{
+	igt_fb_t input_fb, output_fb;
+	igt_plane_t *plane;
+	uint32_t writeback_format = DRM_FORMAT_XRGB8888;
+	uint64_t tiling = igt_fb_mod_to_tiling(0);
+	int width, height, ret;
+	drmModeModeInfo override_mode = {
+		.clock = 25175,
+		.hdisplay = 640,
+		.hsync_start = 656,
+		.hsync_end = 752,
+		.htotal = 800,
+		.hskew = 0,
+		.vdisplay = 480,
+		.vsync_start = 490,
+		.vsync_end = 492,
+		.vtotal = 525,
+		.vscan = 0,
+		.vrefresh = 60,
+		.flags = DRM_MODE_FLAG_NHSYNC | DRM_MODE_FLAG_NVSYNC,
+		.name = {"640x480-60"},
+	};
+	igt_output_override_mode(output, &override_mode);
+
+	width = override_mode.hdisplay;
+	height = override_mode.vdisplay;
+
+	ret = igt_create_fb(display->drm_fd, width, height, DRM_FORMAT_XRGB8888, tiling, &input_fb);
+	igt_assert(ret >= 0);
+
+	ret = igt_create_fb(display->drm_fd, width, height, writeback_format, tiling, &output_fb);
+	igt_assert(ret >= 0);
+
+	plane = igt_output_get_plane_type(output, DRM_PLANE_TYPE_PRIMARY);
+	igt_plane_set_fb(plane, &input_fb);
Quoted 8 lines

tests/kms_writeback.c line 92

Quoted 4 lines
tests/kms_writeback.c new file +4 -0
+	igt_output_set_writeback_fb(output, &output_fb);
+
+	ret = igt_display_try_commit_atomic(display, DRM_MODE_ATOMIC_TEST_ONLY |
+					    DRM_MODE_ATOMIC_ALLOW_MODESET, NULL);

Okay, we're using atomic test-only mode to know if we can perform tests with the writeback output. It's probably fine, but we don't use WRITEBACK_PIXEL_FORMATS, which makes me a little bit sad.

Sorry, I did not fully understand part. Did you expect to see something like the below code before igt_display_try_commit_atomic()?

igt_output_set_prop_value(output, WRITEBACK_PIXEL_FORMATS,

                           DRM_FORMAT_XRGB8888);
Quoted 14 lines
Quoted 14 lines
tests/kms_writeback.c new file +14 -0
+	igt_plane_set_fb(plane, NULL);
+	igt_remove_fb(display->drm_fd, &input_fb);
+	igt_remove_fb(display->drm_fd, &output_fb);
+
+	return !ret;
+}
+
+static igt_output_t *kms_writeback_get_output(igt_display_t *display)
+{
+	int i;
+
+	for (i = 0; i < display->n_outputs; i++) {
+		igt_output_t *output = &display->outputs[i];
+		int j;
Quoted 8 lines

tests/kms_writeback.c line 111

Quoted 5 lines
tests/kms_writeback.c new file +5 -0
+
+		if (output->config.connector->connector_type != DRM_MODE_CONNECTOR_WRITEBACK)
+			continue;
+
+		kmstest_force_connector(display->drm_fd, output->config.connector, FORCE_CONNECTOR_ON);

Hmm. Is this really necessary? "Real" userspace won't do this, so I don't think we need to either.

Hmmm, I have a little experience with userspace; however, I tested kms_writeback on top of vkms without this line, and everything worked well. If I remove this line, should I also drop the line that force connector to FORCE_CONNECTOR_UNSPECIFIED?

Another question, if FORCE_CONNECTOR_ON is something that userspace won't want to do, why do we have it?

Quoted 48 lines

tests/kms_writeback.c line 121

Quoted 9 lines
tests/kms_writeback.c new file +9 -0
+		for (j = 0; j < igt_display_get_n_pipes(display); j++) {
+			igt_output_set_pipe(output, j);
+
+			if (check_writeback_config(display, output)) {
+				igt_debug("Using connector %u:%s on pipe %d\n",
+					  output->config.connector->connector_id,
+					  output->name, j);
+				return output;
+			}

We could probably add an igt_debug statement in case we don't use this writeback output.

Quoted 10 lines
tests/kms_writeback.c new file +10 -0
+		}
+
+		/* Restore any connectors we don't use, so we don't trip on them later */
+		kmstest_force_connector(display->drm_fd, output->config.connector, FORCE_CONNECTOR_UNSPECIFIED);
+	}
+
+	return NULL;
+}
+
+static void check_writeback_fb_id(igt_output_t *output)

tests/kms_writeback.c line 141

Quoted 10 lines
tests/kms_writeback.c new file +10 -0
+{
+	uint64_t check_fb_id;
+
+	check_fb_id = igt_output_get_prop(output, IGT_CONNECTOR_WRITEBACK_FB_ID);
+	igt_assert(check_fb_id == 0);
+}
+
+static int do_writeback_test(igt_output_t *output, uint32_t flags,
+			      uint32_t fb_id, int32_t *out_fence_ptr,
+			      bool ptr_valid)

flags seems to always be set to DRM_MODE_ATOMIC_ALLOW_MODESET. We can probably remove the parameter from this function.

Quoted 11 lines
tests/kms_writeback.c new file +11 -0
+{
+	int ret;
+	igt_display_t *display = output->display;
+	struct kmstest_connector_config *config = &output->config;
+
+	igt_output_set_prop_value(output, IGT_CONNECTOR_CRTC_ID, config->crtc->crtc_id);
+	igt_output_set_prop_value(output, IGT_CONNECTOR_WRITEBACK_FB_ID, fb_id);
+	igt_output_set_prop_value(output, IGT_CONNECTOR_WRITEBACK_OUT_FENCE_PTR, (uint64_t)out_fence_ptr);
+
+	if (ptr_valid)
+		*out_fence_ptr = 0;
Quoted 8 lines

tests/kms_writeback.c line 157

Quoted 5 lines
tests/kms_writeback.c new file +5 -0
+
+	ret = igt_display_try_commit_atomic(display, flags, NULL);
+
+	if (ptr_valid)
+		igt_assert(*out_fence_ptr == -1);

I'm confused. Why should this be -1 even if we
igt_display_try_commit_atomic succeeds?

I inspected the drm_mode_atomic_ioctl() function and I noticed that It calls complete_signaling() in its turn this function assign -1 to out_fence_ptr. Since all of this happens at the end of atomic_commit, I believe that we don’t need it. I already removed it.

Quoted 152 lines
Quoted 17 lines
tests/kms_writeback.c new file +17 -0
+	/* WRITEBACK_FB_ID must always read as zero */
+	check_writeback_fb_id(output);
+
+	return ret;
+}
+
+static void invalid_out_fence(igt_output_t *output, igt_fb_t *valid_fb, igt_fb_t *invalid_fb)
+{
+	int i, ret;
+	int32_t out_fence;
+	struct {
+		uint32_t fb_id;
+		bool ptr_valid;
+		int32_t *out_fence_ptr;
+	} invalid_tests[] = {
+		{
+			/* No output buffer, but the WRITEBACK_OUT_FENCE_PTR set. */

tests/kms_writeback.c line 185

Quoted 10 lines
tests/kms_writeback.c new file +10 -0
+			.fb_id = 0,
+			.ptr_valid = true,
+			.out_fence_ptr = &out_fence,
+		},
+		{
+			/* Invalid output buffer. */
+			.fb_id = invalid_fb->fb_id,
+			.ptr_valid = true,
+			.out_fence_ptr = &out_fence,
+		},

This doesn't belong in this function (invalid_out_fence), since this checks an invalid framebuffer, not an invalid fence. We should probably move it to writeback_fb_id (and rename that function to test_fb?).

Quoted 4 lines
tests/kms_writeback.c new file +4 -0
+		{
+			/* Invalid WRITEBACK_OUT_FENCE_PTR. */
+			.fb_id = valid_fb->fb_id,
+			.ptr_valid = false,

tests/kms_writeback.c line 199

Quoted 10 lines
tests/kms_writeback.c new file +10 -0
+			.out_fence_ptr = (int32_t *)0x8,
+		},
+	};
+
+	for (i = 0; i < ARRAY_SIZE(invalid_tests); i++) {
+		ret = do_writeback_test(output, DRM_MODE_ATOMIC_ALLOW_MODESET,
+					invalid_tests[i].fb_id,
+					invalid_tests[i].out_fence_ptr,
+					invalid_tests[i].ptr_valid);
+		igt_assert(ret != 0);

Maybe we can check for -ret == EINVAL?

tests/kms_writeback.c line 203

Quoted 4 lines
tests/kms_writeback.c new file +4 -0
+	}
+}
+
+static void writeback_fb_id(igt_output_t *output, igt_fb_t *valid_fb, igt_fb_t *invalid_fb)

invalid_fb doesn't seem to be used here. valid_fb seems to be set to the input framebuffer. It's probably not a good idea to use the same FB for input and writeback output.

Quoted 41 lines
tests/kms_writeback.c new file +41 -0
+{
+
+	int ret;
+
+	/* Valid output buffer */
+	ret = do_writeback_test(output, DRM_MODE_ATOMIC_ALLOW_MODESET,
+				valid_fb->fb_id, NULL, false);
+	igt_assert(ret == 0);
+
+	/* Invalid object for WRITEBACK_FB_ID */
+	ret = do_writeback_test(output, DRM_MODE_ATOMIC_ALLOW_MODESET,
+				output->id, NULL, false);
+	igt_assert(ret == -EINVAL);
+
+	/* Zero WRITEBACK_FB_ID */
+	ret = do_writeback_test(output, DRM_MODE_ATOMIC_ALLOW_MODESET,
+				0, NULL, false);
+	igt_assert(ret == 0);
+}
+
+igt_main
+{
+	igt_display_t display;
+	igt_output_t *output;
+	igt_plane_t *plane;
+	igt_fb_t input_fb;
+	drmModeModeInfo mode;
+	int ret;
+
+	memset(&display, 0, sizeof(display));
+
+	igt_fixture {
+		display.drm_fd = drm_open_driver_master(DRIVER_ANY);
+		igt_display_require(&display, display.drm_fd);
+
+		kmstest_set_vt_graphics_mode();
+
+		igt_display_require(&display, display.drm_fd);
+
+		igt_require(display.is_atomic);
+

tests/kms_writeback.c line 254

Quoted 10 lines
tests/kms_writeback.c new file +10 -0
+		output = kms_writeback_get_output(&display);
+		igt_require(output);
+
+		if (output->use_override_mode)
+			memcpy(&mode, &output->override_mode, sizeof(mode));
+		else
+			memcpy(&mode, &output->config.default_mode, sizeof(mode));
+
+		plane = igt_output_get_plane_type(output, DRM_PLANE_TYPE_PRIMARY);
+		igt_require(plane);

Maybe we can assert on this?

tests/kms_writeback.c line 259

Quoted 4 lines
tests/kms_writeback.c new file +4 -0
+		ret = igt_create_fb(display.drm_fd, mode.hdisplay,
+				    mode.vdisplay,
+				    DRM_FORMAT_XRGB8888,
+				    igt_fb_mod_to_tiling(0),

This is supposed to take a modifier. DRM_FORMAT_MOD_LINEAR would be better to make this clear.

(Applies to other lines of this patch)

tests/kms_writeback.c line 266

Quoted 7 lines
tests/kms_writeback.c new file +7 -0
+				    &input_fb);
+		igt_assert(ret >= 0);
+		igt_plane_set_fb(plane, &input_fb);
+	}
+
+	igt_subtest("writeback-pixel-formats") {
+		drmModePropertyBlobRes *formats_blob = get_writeback_formats_blob(output);

Need to drmModeFreePropertyBlob this.

Quoted 4 lines
tests/kms_writeback.c new file +4 -0
+		const char *valid_chars = "0123456 ABCGNRUVXY";
+		unsigned int i;
+		char *c;
+

tests/kms_writeback.c line 280

Quoted 10 lines
tests/kms_writeback.c new file +10 -0
+		/*
+		 * We don't have a comprehensive list of formats, so just check
+		 * that the blob length is sensible and that it doesn't contain
+		 * any outlandish characters
+		 */
+		igt_assert(!(formats_blob->length % 4));
+		c = formats_blob->data;
+		for (i = 0; i < formats_blob->length; i++)
+			igt_assert_f(strchr(valid_chars, c[i]),
+				     "Unexpected character %c\n", c[i]);

Honestly, I'm not a fan of this check. There's no guarantee that fourcc codes will be made from ASCII characters, in fact some formats already have non-printable chars in them. I don't want to have to update this test when a new fourcc format is added.

Quoted 3 lines

We currently don't have a test for IN_FORMATS. If we really want to check formats, we could have a big array of known formats and add a bool is_valid_format(uint32_t fmt) function.

Agree with you. How about remove this test?

Thanks
Best Regards

Quoted 61 lines

tests/kms_writeback.c line 284

Quoted 4 lines
tests/kms_writeback.c new file +4 -0
+	}
+
+	igt_subtest("writeback-invalid-out-fence") {
+		igt_fb_t invalid_fb;

invalid_output_fb would be a better name IMHO.

tests/kms_writeback.c line 290

Quoted 6 lines
tests/kms_writeback.c new file +6 -0
+		ret = igt_create_fb(display.drm_fd, mode.hdisplay / 2,
+				    mode.vdisplay / 2,
+				    DRM_FORMAT_XRGB8888,
+				    igt_fb_mod_to_tiling(0),
+				    &invalid_fb);
+		igt_require(ret > 0);

We should probably use a different variable: ret is signed, igt_create_fb isn't. Also ret doesn't make it clear that the return value is the KMS framebuffer ID.

(Applies to other subtests)

tests/kms_writeback.c line 292

Quoted 1 line
tests/kms_writeback.c new file +1 -0
+		invalid_out_fence(output, &input_fb, &invalid_fb);

(Still not sure why we provide the input FB to this function.)

Quoted 37 lines
tests/kms_writeback.c new file +21 -0
+		igt_remove_fb(display.drm_fd, &invalid_fb);
+	}
+
+	igt_subtest("writeback-fb-id") {
+		igt_fb_t output_fb;
+		ret = igt_create_fb(display.drm_fd, mode.hdisplay, mode.vdisplay,
+				    DRM_FORMAT_XRGB8888,
+				    igt_fb_mod_to_tiling(0),
+				    &output_fb);
+		igt_require(ret > 0);
+
+		writeback_fb_id(output, &input_fb, &output_fb);
+
+		igt_remove_fb(display.drm_fd, &output_fb);
+	}
+
+	igt_fixture {
+		igt_remove_fb(display.drm_fd, &input_fb);
+		igt_display_fini(&display);
+	}
+}
tests/meson.build +1 -0
index f168fbba..9c77cfcd 100644
@@ -63,6 +63,7 @@ test_progs = [
 	'kms_universal_plane',
 	'kms_vblank',
 	'kms_vrr',
+	'kms_writeback',
 	'meta_test',
 	'panfrost_get_param',
 	'panfrost_gem_new',

_______________________________________________
igt-dev mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/igt-dev

Signature · 2 lines
Rodrigo Siqueira
https://siqueira.tech

_______________________________________________
igt-dev mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/igt-dev

#25 lore raw

Re: [igt-dev] [PATCH V6 i-g-t 3/6] lib: Add function to hash a framebuffer

In reply to Ser, Simon
Quoted 60 lines · Simon
Quoted 51 lines · Simon

Mostly LGTM, here are a few nits.

Quoted 35 lines · Brian Starkey

To use writeback buffers as a CRC source, we need to be able to hash them. Implement a simple FVA-1a hashing routine for this purpose.

Doing a bytewise hash on the framebuffer directly can be very slow if the memory is noncached. By making a copy of each line in the FB first (which can take advantage of word-access speedup), we can do the hash on a cached copy, which is much faster (10x speedup on my platform).

v6: use igt_memcpy_from_wc() instead of plain memcpy, as suggested by
Chris Wilson

[rebased and updated to the most recent API]
---
 lib/igt_fb.c | 66 ++++++++++++++++++++++++++++++++++++++++++++++++++++
 lib/igt_fb.h |  3 +++
 2 files changed, 69 insertions(+)
lib/igt_fb.c +8 -0
index 9d4f905e..d07dae39 100644
@@ -3256,6 +3256,72 @@ bool igt_fb_supported_format(uint32_t drm_format)
 	return false;
 }
 
+/*
+ * This implements the FNV-1a hashing algorithm instead of CRC, for
+ * simplicity
+ * http://www.isthe.com/chongo/tech/comp/fnv/index.html
+ *
+ * hash = offset_basis
+ * for each octet_of_data to be hashed
+ *         hash = hash xor octet_of_data

lib/igt_fb.c bool igt_fb_supported_format(uint32_t drm_format) · line 3276

Quoted 10 lines
lib/igt_fb.c +10 -0
+ *         hash = hash * FNV_prime
+ * return hash
+ *
+ * 32 bit offset_basis = 2166136261
+ * 32 bit FNV_prime = 224 + 28 + 0x93 = 16777619
+ */
+int igt_fb_get_crc(struct igt_fb *fb, igt_crc_t *crc)
+{
+#define FNV1a_OFFSET_BIAS 2166136261
+#define FNV1a_PRIME 16777619

I'd just use plain uint32_t variables for those, but no big deal.

Quoted 7 lines

lib/igt_fb.c bool igt_fb_supported_format(uint32_t drm_format) · line 3281

Quoted 5 lines
lib/igt_fb.c +5 -0
+	uint32_t hash;
+	void *map;
+	char *ptr, *line = NULL;
+	int x, y, cpp = igt_drm_format_to_bpp(fb->drm_format) / 8;
+	uint32_t stride = calc_plane_stride(fb, 0);

We could return -EINVAL in case fb->num_planes != 1.

Quoted 5 lines

Let's not waste cycles. With this ^ fixed, this patch is:

Other nits are optional.

I agreed with all your suggestions, and I already applied all of them. Should I wait for the other patches review, or should I resend the new version?

Thanks for all the feedback
Best Regards

Quoted 80 lines
Quoted 80 lines

lib/igt_fb.c bool igt_fb_supported_format(uint32_t drm_format) · line 3289

Quoted 7 lines
lib/igt_fb.c +7 -0
+	if (fb->is_dumb)
+		map = kmstest_dumb_map_buffer(fb->fd, fb->gem_handle, fb->size,
+					      PROT_READ);
+	else
+		map = gem_mmap__gtt(fb->fd, fb->gem_handle, fb->size,
+				    PROT_READ);
+	ptr = map;

Nit: no need for this, can assign the result of mmap directly to ptr.

Quoted 7 lines
lib/igt_fb.c +7 -0
+
+	/*
+	 * Framebuffers are often uncached, which can make byte-wise accesses
+	 * very slow. We copy each line of the FB into a local buffer to speed
+	 * up the hashing.
+	 */
+	line = malloc(stride);

lib/igt_fb.c bool igt_fb_supported_format(uint32_t drm_format) · line 3306

Quoted 10 lines
lib/igt_fb.c +10 -0
+	if (!line) {
+		munmap(map, fb->size);
+		return -ENOMEM;
+	}
+
+	hash = FNV1a_OFFSET_BIAS;
+
+	for (y = 0; y < fb->height; y++, ptr += stride) {
+
+		igt_memcpy_from_wc(line, ptr, stride);

Nit: no need to copy the whole stride actually, we can just copy fb->width * cpp since we're only going to read that.

Quoted 41 lines
lib/igt_fb.c +18 -0
+
+		for (x = 0; x < fb->width * cpp; x++) {
+			hash ^= line[x];
+			hash *= FNV1a_PRIME;
+		}
+	}
+
+	crc->n_words = 1;
+	crc->crc[0] = hash;
+
+	free(line);
+	munmap(map, fb->size);
+
+	return 0;
+#undef FNV1a_OFFSET_BIAS
+#undef FNV1a_PRIME
+}
+
 /**
  * igt_format_is_yuv:
  * @drm_format: drm fourcc
lib/igt_fb.h +3 -0
index adefebe1..a2741c05 100644
@@ -37,6 +37,7 @@
 #include <i915_drm.h>
 
 #include "igt_color_encoding.h"
+#include "igt_debugfs.h"
 
 /*
  * Internal format to denote a buffer compatible with pixman's
@@ -194,5 +195,7 @@ int igt_format_plane_bpp(uint32_t drm_format, int plane);
 void igt_format_array_fill(uint32_t **formats_array, unsigned int *count,
 			   bool allow_yuv);
 
+int igt_fb_get_crc(struct igt_fb *fb, igt_crc_t *crc);
+
 #endif /* __IGT_FB_H__ */
 
Quoted 4 lines

_______________________________________________
igt-dev mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/igt-dev

_______________________________________________
igt-dev mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/igt-dev

Signature · 2 lines
Rodrigo Siqueira
https://siqueira.tech

_______________________________________________
igt-dev mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/igt-dev

#12 lore raw

Re: [igt-dev] [PATCH V6 i-g-t 2/6] kms_writeback: Add initial writeback tests

In reply to Rodrigo Siqueira
Quoted 143 lines · Rodrigo Siqueira
Quoted 6 lines · Simon

Hi,

Thanks for the patch! Here are a few comments.

For bonus points, it would be nice to add igt_describe descriptions of each sub-test.

Hi Simon,

First of all, thanks for your feedback; I already applied most of your suggestions. I just have some inline comments/questions.

Quoted 121 lines · Brian Starkey
Quoted 120 lines

Add tests for the WRITEBACK_PIXEL_FORMATS, WRITEBACK_OUT_FENCE_PTR and WRITEBACK_FB_ID properties on writeback connectors, ensuring their behaviour is correct.

[rebased and updated do_writeback_test() function to address feedback]
---
 tests/Makefile.sources |   1 +
 tests/kms_writeback.c  | 314 +++++++++++++++++++++++++++++++++++++++++
 tests/meson.build      |   1 +
 3 files changed, 316 insertions(+)
 create mode 100644 tests/kms_writeback.c
tests/Makefile.sources +1 -0
index 027ed82f..03cc8efa 100644
@@ -77,6 +77,7 @@ TESTS_progs = \
 	kms_universal_plane \
 	kms_vblank \
 	kms_vrr \
+	kms_writeback \
 	meta_test \
 	perf \
 	perf_pmu \
tests/kms_writeback.c new file +88 -0
new file mode 100644
index 00000000..66ef48a6
@@ -0,0 +1,314 @@
+/*
+ * (C) COPYRIGHT 2017 ARM Limited. All rights reserved.
+ *
+ * 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.
+ *
+ */
+
+#include <errno.h>
+#include <stdbool.h>
+#include <stdio.h>
+#include <string.h>
+
+#include "igt.h"
+#include "igt_core.h"
+#include "igt_fb.h"
+
+static drmModePropertyBlobRes *get_writeback_formats_blob(igt_output_t *output)
+{
+	drmModePropertyBlobRes *blob = NULL;
+	uint64_t blob_id;
+	int ret;
+
+	ret = kmstest_get_property(output->display->drm_fd,
+				   output->config.connector->connector_id,
+				   DRM_MODE_OBJECT_CONNECTOR,
+				   igt_connector_prop_names[IGT_CONNECTOR_WRITEBACK_PIXEL_FORMATS],
+				   NULL, &blob_id, NULL);
+	if (ret)
+		blob = drmModeGetPropertyBlob(output->display->drm_fd, blob_id);
+
+	igt_assert(blob);
+
+	return blob;
+}
+
+static bool check_writeback_config(igt_display_t *display, igt_output_t *output)
+{
+	igt_fb_t input_fb, output_fb;
+	igt_plane_t *plane;
+	uint32_t writeback_format = DRM_FORMAT_XRGB8888;
+	uint64_t tiling = igt_fb_mod_to_tiling(0);
+	int width, height, ret;
+	drmModeModeInfo override_mode = {
+		.clock = 25175,
+		.hdisplay = 640,
+		.hsync_start = 656,
+		.hsync_end = 752,
+		.htotal = 800,
+		.hskew = 0,
+		.vdisplay = 480,
+		.vsync_start = 490,
+		.vsync_end = 492,
+		.vtotal = 525,
+		.vscan = 0,
+		.vrefresh = 60,
+		.flags = DRM_MODE_FLAG_NHSYNC | DRM_MODE_FLAG_NVSYNC,
+		.name = {"640x480-60"},
+	};
+	igt_output_override_mode(output, &override_mode);
+
+	width = override_mode.hdisplay;
+	height = override_mode.vdisplay;
+
+	ret = igt_create_fb(display->drm_fd, width, height, DRM_FORMAT_XRGB8888, tiling, &input_fb);
+	igt_assert(ret >= 0);
+
+	ret = igt_create_fb(display->drm_fd, width, height, writeback_format, tiling, &output_fb);
+	igt_assert(ret >= 0);
+
+	plane = igt_output_get_plane_type(output, DRM_PLANE_TYPE_PRIMARY);
+	igt_plane_set_fb(plane, &input_fb);
Quoted 8 lines

tests/kms_writeback.c line 92

Quoted 4 lines
tests/kms_writeback.c new file +4 -0
+	igt_output_set_writeback_fb(output, &output_fb);
+
+	ret = igt_display_try_commit_atomic(display, DRM_MODE_ATOMIC_TEST_ONLY |
+					    DRM_MODE_ATOMIC_ALLOW_MODESET, NULL);

Okay, we're using atomic test-only mode to know if we can perform tests with the writeback output. It's probably fine, but we don't use WRITEBACK_PIXEL_FORMATS, which makes me a little bit sad.

Quoted 5 lines

Sorry, I did not fully understand part. Did you expect to see something like the below code before igt_display_try_commit_atomic()?

igt_output_set_prop_value(output, WRITEBACK_PIXEL_FORMATS,

                           DRM_FORMAT_XRGB8888);

Hmm, no, we cannot change the list of writeback formats (it's immutable).

This comment doesn't require any action, it's just me thinking aloud :P

I'm just thinking that it would be nice to choose our format depending on the WRITEBACK_PIXEL_FORMATS property. And probably run tests with each supported format (or, alternatively, choose a random one). That way we can make sure WRITEBACK_PIXEL_FORMATS isn't broken.

However, this can be left for a later patch.

Quoted 23 lines
Quoted 14 lines
Quoted 14 lines
tests/kms_writeback.c new file +14 -0
+	igt_plane_set_fb(plane, NULL);
+	igt_remove_fb(display->drm_fd, &input_fb);
+	igt_remove_fb(display->drm_fd, &output_fb);
+
+	return !ret;
+}
+
+static igt_output_t *kms_writeback_get_output(igt_display_t *display)
+{
+	int i;
+
+	for (i = 0; i < display->n_outputs; i++) {
+		igt_output_t *output = &display->outputs[i];
+		int j;
Quoted 8 lines

tests/kms_writeback.c line 111

Quoted 5 lines
tests/kms_writeback.c new file +5 -0
+
+		if (output->config.connector->connector_type != DRM_MODE_CONNECTOR_WRITEBACK)
+			continue;
+
+		kmstest_force_connector(display->drm_fd, output->config.connector, FORCE_CONNECTOR_ON);

Hmm. Is this really necessary? "Real" userspace won't do this, so I don't think we need to either.

Quoted 4 lines

Hmmm, I have a little experience with userspace; however, I tested kms_writeback on top of vkms without this line, and everything worked well. If I remove this line, should I also drop the line that force connector to FORCE_CONNECTOR_UNSPECIFIED?

I believe so.

Quoted 2 lines

Another question, if FORCE_CONNECTOR_ON is something that userspace won't want to do, why do we have it?

We use kmstest_force_connector in injection tests: those pick a disconnected HDMI connector and trick the kernel into thinking it's connected. We generally also force an EDID and this is used to emulate a screen (e.g. a screen that supports audio, 4K or 3D).

This is only meant to be used for testing though, hence "real userspace shouldn't use it".

Quoted 57 lines
Quoted 48 lines

tests/kms_writeback.c line 121

Quoted 9 lines
tests/kms_writeback.c new file +9 -0
+		for (j = 0; j < igt_display_get_n_pipes(display); j++) {
+			igt_output_set_pipe(output, j);
+
+			if (check_writeback_config(display, output)) {
+				igt_debug("Using connector %u:%s on pipe %d\n",
+					  output->config.connector->connector_id,
+					  output->name, j);
+				return output;
+			}

We could probably add an igt_debug statement in case we don't use this writeback output.

Quoted 10 lines
tests/kms_writeback.c new file +10 -0
+		}
+
+		/* Restore any connectors we don't use, so we don't trip on them later */
+		kmstest_force_connector(display->drm_fd, output->config.connector, FORCE_CONNECTOR_UNSPECIFIED);
+	}
+
+	return NULL;
+}
+
+static void check_writeback_fb_id(igt_output_t *output)

tests/kms_writeback.c line 141

Quoted 10 lines
tests/kms_writeback.c new file +10 -0
+{
+	uint64_t check_fb_id;
+
+	check_fb_id = igt_output_get_prop(output, IGT_CONNECTOR_WRITEBACK_FB_ID);
+	igt_assert(check_fb_id == 0);
+}
+
+static int do_writeback_test(igt_output_t *output, uint32_t flags,
+			      uint32_t fb_id, int32_t *out_fence_ptr,
+			      bool ptr_valid)

flags seems to always be set to DRM_MODE_ATOMIC_ALLOW_MODESET. We can probably remove the parameter from this function.

Quoted 11 lines
tests/kms_writeback.c new file +11 -0
+{
+	int ret;
+	igt_display_t *display = output->display;
+	struct kmstest_connector_config *config = &output->config;
+
+	igt_output_set_prop_value(output, IGT_CONNECTOR_CRTC_ID, config->crtc->crtc_id);
+	igt_output_set_prop_value(output, IGT_CONNECTOR_WRITEBACK_FB_ID, fb_id);
+	igt_output_set_prop_value(output, IGT_CONNECTOR_WRITEBACK_OUT_FENCE_PTR, (uint64_t)out_fence_ptr);
+
+	if (ptr_valid)
+		*out_fence_ptr = 0;
Quoted 8 lines

tests/kms_writeback.c line 157

Quoted 5 lines
tests/kms_writeback.c new file +5 -0
+
+	ret = igt_display_try_commit_atomic(display, flags, NULL);
+
+	if (ptr_valid)
+		igt_assert(*out_fence_ptr == -1);

I'm confused. Why should this be -1 even if we
igt_display_try_commit_atomic succeeds?

Quoted 4 lines

I inspected the drm_mode_atomic_ioctl() function and I noticed that It calls complete_signaling() in its turn this function assign -1 to out_fence_ptr. Since all of this happens at the end of atomic_commit, I believe that we don’t need it. I already removed it.

I think I'm still confused :)

I'm probably missing something. As far as I understand, we are doing an atomic commit, sometimes with a valid WRITEBACK_FB_ID and WRITEBACK_OUT_FENCE_PTR. In this case it should start the writeback process in the kernel and we should get a valid out_fence_ptr?

Why do we always get -1?

Quoted 152 lines
Quoted 152 lines
Quoted 17 lines
tests/kms_writeback.c new file +17 -0
+	/* WRITEBACK_FB_ID must always read as zero */
+	check_writeback_fb_id(output);
+
+	return ret;
+}
+
+static void invalid_out_fence(igt_output_t *output, igt_fb_t *valid_fb, igt_fb_t *invalid_fb)
+{
+	int i, ret;
+	int32_t out_fence;
+	struct {
+		uint32_t fb_id;
+		bool ptr_valid;
+		int32_t *out_fence_ptr;
+	} invalid_tests[] = {
+		{
+			/* No output buffer, but the WRITEBACK_OUT_FENCE_PTR set. */

tests/kms_writeback.c line 185

Quoted 10 lines
tests/kms_writeback.c new file +10 -0
+			.fb_id = 0,
+			.ptr_valid = true,
+			.out_fence_ptr = &out_fence,
+		},
+		{
+			/* Invalid output buffer. */
+			.fb_id = invalid_fb->fb_id,
+			.ptr_valid = true,
+			.out_fence_ptr = &out_fence,
+		},

This doesn't belong in this function (invalid_out_fence), since this checks an invalid framebuffer, not an invalid fence. We should probably move it to writeback_fb_id (and rename that function to test_fb?).

Quoted 4 lines
tests/kms_writeback.c new file +4 -0
+		{
+			/* Invalid WRITEBACK_OUT_FENCE_PTR. */
+			.fb_id = valid_fb->fb_id,
+			.ptr_valid = false,

tests/kms_writeback.c line 199

Quoted 10 lines
tests/kms_writeback.c new file +10 -0
+			.out_fence_ptr = (int32_t *)0x8,
+		},
+	};
+
+	for (i = 0; i < ARRAY_SIZE(invalid_tests); i++) {
+		ret = do_writeback_test(output, DRM_MODE_ATOMIC_ALLOW_MODESET,
+					invalid_tests[i].fb_id,
+					invalid_tests[i].out_fence_ptr,
+					invalid_tests[i].ptr_valid);
+		igt_assert(ret != 0);

Maybe we can check for -ret == EINVAL?

tests/kms_writeback.c line 203

Quoted 4 lines
tests/kms_writeback.c new file +4 -0
+	}
+}
+
+static void writeback_fb_id(igt_output_t *output, igt_fb_t *valid_fb, igt_fb_t *invalid_fb)

invalid_fb doesn't seem to be used here. valid_fb seems to be set to the input framebuffer. It's probably not a good idea to use the same FB for input and writeback output.

Quoted 41 lines
tests/kms_writeback.c new file +41 -0
+{
+
+	int ret;
+
+	/* Valid output buffer */
+	ret = do_writeback_test(output, DRM_MODE_ATOMIC_ALLOW_MODESET,
+				valid_fb->fb_id, NULL, false);
+	igt_assert(ret == 0);
+
+	/* Invalid object for WRITEBACK_FB_ID */
+	ret = do_writeback_test(output, DRM_MODE_ATOMIC_ALLOW_MODESET,
+				output->id, NULL, false);
+	igt_assert(ret == -EINVAL);
+
+	/* Zero WRITEBACK_FB_ID */
+	ret = do_writeback_test(output, DRM_MODE_ATOMIC_ALLOW_MODESET,
+				0, NULL, false);
+	igt_assert(ret == 0);
+}
+
+igt_main
+{
+	igt_display_t display;
+	igt_output_t *output;
+	igt_plane_t *plane;
+	igt_fb_t input_fb;
+	drmModeModeInfo mode;
+	int ret;
+
+	memset(&display, 0, sizeof(display));
+
+	igt_fixture {
+		display.drm_fd = drm_open_driver_master(DRIVER_ANY);
+		igt_display_require(&display, display.drm_fd);
+
+		kmstest_set_vt_graphics_mode();
+
+		igt_display_require(&display, display.drm_fd);
+
+		igt_require(display.is_atomic);
+

tests/kms_writeback.c line 254

Quoted 10 lines
tests/kms_writeback.c new file +10 -0
+		output = kms_writeback_get_output(&display);
+		igt_require(output);
+
+		if (output->use_override_mode)
+			memcpy(&mode, &output->override_mode, sizeof(mode));
+		else
+			memcpy(&mode, &output->config.default_mode, sizeof(mode));
+
+		plane = igt_output_get_plane_type(output, DRM_PLANE_TYPE_PRIMARY);
+		igt_require(plane);

Maybe we can assert on this?

tests/kms_writeback.c line 259

Quoted 4 lines
tests/kms_writeback.c new file +4 -0
+		ret = igt_create_fb(display.drm_fd, mode.hdisplay,
+				    mode.vdisplay,
+				    DRM_FORMAT_XRGB8888,
+				    igt_fb_mod_to_tiling(0),

This is supposed to take a modifier. DRM_FORMAT_MOD_LINEAR would be better to make this clear.

(Applies to other lines of this patch)

tests/kms_writeback.c line 266

Quoted 7 lines
tests/kms_writeback.c new file +7 -0
+				    &input_fb);
+		igt_assert(ret >= 0);
+		igt_plane_set_fb(plane, &input_fb);
+	}
+
+	igt_subtest("writeback-pixel-formats") {
+		drmModePropertyBlobRes *formats_blob = get_writeback_formats_blob(output);

Need to drmModeFreePropertyBlob this.

Quoted 4 lines
tests/kms_writeback.c new file +4 -0
+		const char *valid_chars = "0123456 ABCGNRUVXY";
+		unsigned int i;
+		char *c;
+

tests/kms_writeback.c line 280

Quoted 10 lines
tests/kms_writeback.c new file +10 -0
+		/*
+		 * We don't have a comprehensive list of formats, so just check
+		 * that the blob length is sensible and that it doesn't contain
+		 * any outlandish characters
+		 */
+		igt_assert(!(formats_blob->length % 4));
+		c = formats_blob->data;
+		for (i = 0; i < formats_blob->length; i++)
+			igt_assert_f(strchr(valid_chars, c[i]),
+				     "Unexpected character %c\n", c[i]);

Honestly, I'm not a fan of this check. There's no guarantee that fourcc codes will be made from ASCII characters, in fact some formats already have non-printable chars in them. I don't want to have to update this test when a new fourcc format is added.

Quoted 5 lines
Quoted 3 lines

We currently don't have a test for IN_FORMATS. If we really want to check formats, we could have a big array of known formats and add a bool is_valid_format(uint32_t fmt) function.

Agree with you. How about remove this test?

I guess we could always keep the length % 4 test, even if it's just a sanity test. At least this one won't ever need to be changed.

I don't feel strongly about it.

Quoted 56 lines

Thanks
Best Regards

Quoted 53 lines

tests/kms_writeback.c line 284

Quoted 4 lines
tests/kms_writeback.c new file +4 -0
+	}
+
+	igt_subtest("writeback-invalid-out-fence") {
+		igt_fb_t invalid_fb;

invalid_output_fb would be a better name IMHO.

tests/kms_writeback.c line 290

Quoted 6 lines
tests/kms_writeback.c new file +6 -0
+		ret = igt_create_fb(display.drm_fd, mode.hdisplay / 2,
+				    mode.vdisplay / 2,
+				    DRM_FORMAT_XRGB8888,
+				    igt_fb_mod_to_tiling(0),
+				    &invalid_fb);
+		igt_require(ret > 0);

We should probably use a different variable: ret is signed, igt_create_fb isn't. Also ret doesn't make it clear that the return value is the KMS framebuffer ID.

(Applies to other subtests)

tests/kms_writeback.c line 292

Quoted 1 line
tests/kms_writeback.c new file +1 -0
+		invalid_out_fence(output, &input_fb, &invalid_fb);

(Still not sure why we provide the input FB to this function.)

Quoted 29 lines
tests/kms_writeback.c new file +21 -0
+		igt_remove_fb(display.drm_fd, &invalid_fb);
+	}
+
+	igt_subtest("writeback-fb-id") {
+		igt_fb_t output_fb;
+		ret = igt_create_fb(display.drm_fd, mode.hdisplay, mode.vdisplay,
+				    DRM_FORMAT_XRGB8888,
+				    igt_fb_mod_to_tiling(0),
+				    &output_fb);
+		igt_require(ret > 0);
+
+		writeback_fb_id(output, &input_fb, &output_fb);
+
+		igt_remove_fb(display.drm_fd, &output_fb);
+	}
+
+	igt_fixture {
+		igt_remove_fb(display.drm_fd, &input_fb);
+		igt_display_fini(&display);
+	}
+}
tests/meson.build +0 -0
index f168fbba..9c77cfcd 100644
@@ -63,6 +63,7 @@ test_progs = [
 	'kms_universal_plane',
 	'kms_vblank',
 	'kms_vrr',
Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
tests/meson.build +1 -0
+	'kms_writeback',
 	'meta_test',
 	'panfrost_get_param',
 	'panfrost_gem_new',

_______________________________________________
igt-dev mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/igt-dev

_______________________________________________
igt-dev mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/igt-dev

#26 lore raw

Re: [igt-dev] [PATCH V6 i-g-t 3/6] lib: Add function to hash a framebuffer

In reply to Rodrigo Siqueira
Quoted 63 lines · Rodrigo Siqueira
Quoted 62 lines · Simon
Quoted 51 lines · Simon

Mostly LGTM, here are a few nits.

Quoted 35 lines · Brian Starkey

To use writeback buffers as a CRC source, we need to be able to hash them. Implement a simple FVA-1a hashing routine for this purpose.

Doing a bytewise hash on the framebuffer directly can be very slow if the memory is noncached. By making a copy of each line in the FB first (which can take advantage of word-access speedup), we can do the hash on a cached copy, which is much faster (10x speedup on my platform).

v6: use igt_memcpy_from_wc() instead of plain memcpy, as suggested by
Chris Wilson

[rebased and updated to the most recent API]
---
 lib/igt_fb.c | 66 ++++++++++++++++++++++++++++++++++++++++++++++++++++
 lib/igt_fb.h |  3 +++
 2 files changed, 69 insertions(+)
lib/igt_fb.c +8 -0
index 9d4f905e..d07dae39 100644
@@ -3256,6 +3256,72 @@ bool igt_fb_supported_format(uint32_t drm_format)
 	return false;
 }
 
+/*
+ * This implements the FNV-1a hashing algorithm instead of CRC, for
+ * simplicity
+ * http://www.isthe.com/chongo/tech/comp/fnv/index.html
+ *
+ * hash = offset_basis
+ * for each octet_of_data to be hashed
+ *         hash = hash xor octet_of_data

lib/igt_fb.c bool igt_fb_supported_format(uint32_t drm_format) · line 3276

Quoted 10 lines
lib/igt_fb.c +10 -0
+ *         hash = hash * FNV_prime
+ * return hash
+ *
+ * 32 bit offset_basis = 2166136261
+ * 32 bit FNV_prime = 224 + 28 + 0x93 = 16777619
+ */
+int igt_fb_get_crc(struct igt_fb *fb, igt_crc_t *crc)
+{
+#define FNV1a_OFFSET_BIAS 2166136261
+#define FNV1a_PRIME 16777619

I'd just use plain uint32_t variables for those, but no big deal.

Quoted 7 lines

lib/igt_fb.c bool igt_fb_supported_format(uint32_t drm_format) · line 3281

Quoted 5 lines
lib/igt_fb.c +5 -0
+	uint32_t hash;
+	void *map;
+	char *ptr, *line = NULL;
+	int x, y, cpp = igt_drm_format_to_bpp(fb->drm_format) / 8;
+	uint32_t stride = calc_plane_stride(fb, 0);

We could return -EINVAL in case fb->num_planes != 1.

Let's not waste cycles. With this ^ fixed, this patch is:

Quoted 7 lines
Quoted 3 lines

Other nits are optional.

I agreed with all your suggestions, and I already applied all of them. Should I wait for the other patches review, or should I resend the new version?

I'm fine with waiting for the full review before a new version of the whole patchset, but you can also send an updated version of a single patch with:

    git send-email --in-reply-to="<[email protected]>" -1 <commit hash>

where In-Reply-To is the Message-Id of the patch you want to update. I agree it's a little tedious since you need to extract the Message-Id from the message header.

Quoted 1 line

Thanks for all the feedback

:)

Quoted 82 lines

Best Regards

Quoted 80 lines
Quoted 80 lines

lib/igt_fb.c bool igt_fb_supported_format(uint32_t drm_format) · line 3289

Quoted 7 lines
lib/igt_fb.c +7 -0
+	if (fb->is_dumb)
+		map = kmstest_dumb_map_buffer(fb->fd, fb->gem_handle, fb->size,
+					      PROT_READ);
+	else
+		map = gem_mmap__gtt(fb->fd, fb->gem_handle, fb->size,
+				    PROT_READ);
+	ptr = map;

Nit: no need for this, can assign the result of mmap directly to ptr.

Quoted 7 lines
lib/igt_fb.c +7 -0
+
+	/*
+	 * Framebuffers are often uncached, which can make byte-wise accesses
+	 * very slow. We copy each line of the FB into a local buffer to speed
+	 * up the hashing.
+	 */
+	line = malloc(stride);

lib/igt_fb.c bool igt_fb_supported_format(uint32_t drm_format) · line 3306

Quoted 10 lines
lib/igt_fb.c +10 -0
+	if (!line) {
+		munmap(map, fb->size);
+		return -ENOMEM;
+	}
+
+	hash = FNV1a_OFFSET_BIAS;
+
+	for (y = 0; y < fb->height; y++, ptr += stride) {
+
+		igt_memcpy_from_wc(line, ptr, stride);

Nit: no need to copy the whole stride actually, we can just copy fb->width * cpp since we're only going to read that.

Quoted 41 lines
lib/igt_fb.c +18 -0
+
+		for (x = 0; x < fb->width * cpp; x++) {
+			hash ^= line[x];
+			hash *= FNV1a_PRIME;
+		}
+	}
+
+	crc->n_words = 1;
+	crc->crc[0] = hash;
+
+	free(line);
+	munmap(map, fb->size);
+
+	return 0;
+#undef FNV1a_OFFSET_BIAS
+#undef FNV1a_PRIME
+}
+
 /**
  * igt_format_is_yuv:
  * @drm_format: drm fourcc
lib/igt_fb.h +3 -0
index adefebe1..a2741c05 100644
@@ -37,6 +37,7 @@
 #include <i915_drm.h>
 
 #include "igt_color_encoding.h"
+#include "igt_debugfs.h"
 
 /*
  * Internal format to denote a buffer compatible with pixman's
@@ -194,5 +195,7 @@ int igt_format_plane_bpp(uint32_t drm_format, int plane);
 void igt_format_array_fill(uint32_t **formats_array, unsigned int *count,
 			   bool allow_yuv);
 
+int igt_fb_get_crc(struct igt_fb *fb, igt_crc_t *crc);
+
 #endif /* __IGT_FB_H__ */
 
Quoted 4 lines

_______________________________________________
igt-dev mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/igt-dev

_______________________________________________
igt-dev mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/igt-dev

_______________________________________________
igt-dev mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/igt-dev

#29 lore raw

Re: [igt-dev] [PATCH V6 i-g-t 4/6] kms_writeback: Add writeback-check-output

In reply to Brian Starkey

Mostly looks good, here are a few comments.

Quoted 32 lines · Brian Starkey

Add a test which makes commits using the writeback connector, and checks the output buffer hash to make sure it is/isn't written as appropriate.

---
 tests/kms_writeback.c | 124 ++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 124 insertions(+)
tests/kms_writeback.c +8 -0
index 66ef48a6..0f20dadd 100644
@@ -30,6 +30,7 @@
 #include "igt.h"
 #include "igt_core.h"
 #include "igt_fb.h"
+#include "sw_sync.h"
 
 static drmModePropertyBlobRes *get_writeback_formats_blob(igt_output_t *output)
 {
@@ -221,6 +222,116 @@ static void writeback_fb_id(igt_output_t *output, igt_fb_t *valid_fb, igt_fb_t *
 	igt_assert(ret == 0);
 }
 
+static void fill_fb(igt_fb_t *fb, double color[3])
+{
+	cairo_t *cr = igt_get_cairo_ctx(fb->fd, fb);
+	igt_assert(cr);
+
+	igt_paint_color(cr, 0, 0, fb->width, fb->height,
+			color[0], color[1], color[2]);

tests/kms_writeback.c static void writeback_fb_id(igt_output_t *output, igt_fb_t *valid_fb, igt_fb_t * · line 241

Quoted 10 lines
tests/kms_writeback.c +10 -0
+}
+
+static void get_and_wait_out_fence(igt_output_t *output)
+{
+	int ret;
+
+	igt_assert(output->writeback_out_fence_fd >= 0);
+
+	ret = sync_fence_wait(output->writeback_out_fence_fd, 1000);
+	igt_assert(ret == 0);

When filtering failures in CI Bug Log, "ret == 0" is not very helpful and may not be unique. Can we add a descriptive error message, with something among the lines of:

    igt_assert_f(ret == 0, "sync_fence_wait failed: %s\n", strerror(-ret));
Quoted 8 lines
tests/kms_writeback.c +8 -0
+	close(output->writeback_out_fence_fd);
+	output->writeback_out_fence_fd = -1;
+}
+
+static void writeback_sequence(igt_output_t *output, igt_plane_t *plane,
+				igt_fb_t *in_fb, igt_fb_t *out_fbs[], int n_commits)
+{
+	int i, color_idx = 0;

tests/kms_writeback.c static void writeback_fb_id(igt_output_t *output, igt_fb_t *valid_fb, igt_fb_t * · line 259

Quoted 10 lines
tests/kms_writeback.c +10 -0
+	double in_fb_colors[2][3] = {
+		{ 1.0, 0.0, 0.0 },
+		{ 0.0, 1.0, 0.0 },
+	};
+	double clear_color[3] = { 1.0, 1.0, 1.0 };
+	igt_crc_t cleared_crc, out_expected;
+
+	for (i = 0; i < n_commits; i++, color_idx++) {
+		/* Change the input color each time */
+		fill_fb(in_fb, in_fb_colors[color_idx % 2]);

Why do we need two indexes (i and color_idx)? I think [i % 2] should work just fine.

tests/kms_writeback.c static void writeback_fb_id(igt_output_t *output, igt_fb_t *valid_fb, igt_fb_t * · line 266

Quoted 6 lines
tests/kms_writeback.c +6 -0
+		if (out_fbs[i]) {
+			igt_crc_t out_before;
+
+			/* Get the expected CRC */
+			fill_fb(out_fbs[i], in_fb_colors[color_idx % 2]);
+			igt_fb_get_crc(out_fbs[i], &out_expected);

Using out_fbs[i] to compute the expected CRC is a little bit confusing. I think we can just use in_fb here (which is already filled with the proper color).

Quoted 9 lines
tests/kms_writeback.c +9 -0
+			fill_fb(out_fbs[i], clear_color);
+			if (i == 0)
+				igt_fb_get_crc(out_fbs[i], &cleared_crc);
+			igt_fb_get_crc(out_fbs[i], &out_before);
+			igt_assert_crc_equal(&cleared_crc, &out_before);
+		}
+
+		/* Commit */
+		igt_plane_set_fb(plane, in_fb);

tests/kms_writeback.c static void writeback_fb_id(igt_output_t *output, igt_fb_t *valid_fb, igt_fb_t * · line 286

Quoted 10 lines
tests/kms_writeback.c +10 -0
+		igt_output_set_writeback_fb(output, out_fbs[i]);
+
+		igt_display_commit_atomic(output->display,
+					  DRM_MODE_ATOMIC_ALLOW_MODESET,
+					  NULL);
+		if (out_fbs[i])
+			get_and_wait_out_fence(output);
+
+		/* Make sure the old output buffer is untouched */
+		if (i > 0 && out_fbs[i - 1] && (out_fbs[i] != out_fbs[i - 1])) {

Nit: unnecessary parentheses here

Quoted 30 lines
tests/kms_writeback.c +30 -0
+			igt_crc_t out_prev;
+			igt_fb_get_crc(out_fbs[i - 1], &out_prev);
+			igt_assert_crc_equal(&cleared_crc, &out_prev);
+		}
+
+		/* Make sure this output buffer is written */
+		if (out_fbs[i]) {
+			igt_crc_t out_after;
+			igt_fb_get_crc(out_fbs[i], &out_after);
+			igt_assert_crc_equal(&out_expected, &out_after);
+
+			/* And clear it, for the next time */
+			fill_fb(out_fbs[i], clear_color);
+		}
+	}
+}
+
+static void writeback_check_output(igt_output_t *output, igt_plane_t *plane,
+				   igt_fb_t *input_fb, igt_fb_t *output_fb)
+{
+	igt_fb_t *out_fbs[2] = { 0 };
+	igt_fb_t second_out_fb;
+	int ret;
+
+	/* One commit, with a writeback. */
+	writeback_sequence(output, plane, input_fb, &output_fb, 1);
+
+	/* Two commits, the second with no writeback */
+	out_fbs[0] = output_fb;
+	writeback_sequence(output, plane, input_fb, out_fbs, 2);

tests/kms_writeback.c static void writeback_fb_id(igt_output_t *output, igt_fb_t *valid_fb, igt_fb_t * · line 326

Quoted 10 lines
tests/kms_writeback.c +10 -0
+
+	/* Two commits, both with writeback */
+	out_fbs[1] = output_fb;
+	writeback_sequence(output, plane, input_fb, out_fbs, 2);
+
+	ret = igt_create_fb(output_fb->fd, output_fb->width, output_fb->height,
+			    DRM_FORMAT_XRGB8888,
+			    igt_fb_mod_to_tiling(0),
+			    &second_out_fb);
+	igt_require(ret > 0);

unsigned int cast to int here

Quoted 11 lines
tests/kms_writeback.c +7 -0
+	/* Two commits, with different writeback buffers */
+	out_fbs[1] = &second_out_fb;
+	writeback_sequence(output, plane, input_fb, out_fbs, 2);
+
+	igt_remove_fb(output_fb->fd, &second_out_fb);
+}
+
 igt_main
 {
 	igt_display_t display;

tests/kms_writeback.c igt_main · line 427

Quoted 10 lines
tests/kms_writeback.c +7 -0
@@ -307,6 +418,19 @@ igt_main
 		igt_remove_fb(display.drm_fd, &output_fb);
 	}
 
+	igt_subtest("writeback-check-output") {
+		igt_fb_t output_fb;
+		ret = igt_create_fb(display.drm_fd, mode.hdisplay, mode.vdisplay,
+				    DRM_FORMAT_XRGB8888,
+				    igt_fb_mod_to_tiling(0),
+				    &output_fb);
+		igt_require(ret > 0);

unsigned int cast to int here

Quoted 8 lines
tests/kms_writeback.c +5 -0
+		writeback_check_output(output, plane, &input_fb, &output_fb);
+
+		igt_remove_fb(display.drm_fd, &output_fb);
+	}
+
 	igt_fixture {
 		igt_remove_fb(display.drm_fd, &input_fb);
 		igt_display_fini(&display);
Quoted 4 lines

_______________________________________________
igt-dev mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/igt-dev

_______________________________________________
igt-dev mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/igt-dev

#31 lore raw

Re: [igt-dev] [PATCH V6 i-g-t 5/6] lib/igt_kms: Add igt_output_clone_pipe for cloning

In reply to Brian Starkey

So, to test these last two patches we'd need specific hardware right? Because VKMS doesn't support cloning yet (does it?).

What kind of hardware supports cloned writeback outputs? I have a Raspberry Pi which supports writeback via VC4, but I don't think it has writeback cloning. I'm also not willing to install any proprietary driver.

I guess we could land the first part of the series, and wait for VKMS to support cloned outputs to land the last two patches.

Any other ideas?

Quoted 182 lines · Brian Starkey

An output can be added as a clone of any other output(s) attached to a pipe using igt_output_clone_pipe()

v5: Drop field out_fence_requested from struct igt_pipe (Brian Starkey)

---
 lib/igt_kms.c | 100 +++++++++++++++++++++++++++++++-------------------
 lib/igt_kms.h |   4 ++
 2 files changed, 66 insertions(+), 38 deletions(-)
lib/igt_kms.c +62 -38
index 140db346..b85a0404 100644
@@ -1765,6 +1765,17 @@ static void igt_display_log_shift(igt_display_t *display, int shift)
 	igt_assert(display->log_shift >= 0);
 }
 
+static int igt_output_idx(igt_output_t *output)
+{
+	int i;
+
+	for (i = 0; i < output->display->n_outputs; i++)
+		if (&output->display->outputs[i] == output)
+			return i;
+
+	return -1;
+}
+
 static void igt_output_refresh(igt_output_t *output)
 {
 	igt_display_t *display = output->display;
@@ -2317,42 +2328,6 @@ void igt_display_fini(igt_display_t *display)
 	display->planes = NULL;
 }
 
-static void igt_display_refresh(igt_display_t *display)
-{
-	igt_output_t *output;
-	int i;
-
-	unsigned long pipes_in_use = 0;
-
-       /* Check that two outputs aren't trying to use the same pipe */
-	for (i = 0; i < display->n_outputs; i++) {
-		output = &display->outputs[i];
-
-		if (output->pending_pipe != PIPE_NONE) {
-			if (pipes_in_use & (1 << output->pending_pipe))
-				goto report_dup;
-
-			pipes_in_use |= 1 << output->pending_pipe;
-		}
-
-		if (output->force_reprobe)
-			igt_output_refresh(output);
-	}
-
-	return;
-
-report_dup:
-	for (; i > 0; i--) {
-		igt_output_t *b = &display->outputs[i - 1];
-
-		igt_assert_f(output->pending_pipe !=
-			     b->pending_pipe,
-			     "%s and %s are both trying to use pipe %s\n",
-			     igt_output_name(output), igt_output_name(b),
-			     kmstest_pipe_name(output->pending_pipe));
-	}
-}
-
 static igt_pipe_t *igt_output_get_driving_pipe(igt_output_t *output)
 {
 	igt_display_t *display = output->display;
@@ -2376,6 +2351,40 @@ static igt_pipe_t *igt_output_get_driving_pipe(igt_output_t *output)
 	return &display->pipes[pipe];
 }
 
+static void igt_display_refresh(igt_display_t *display)
+{
+	igt_output_t *output;
+	igt_pipe_t *pipe;
+	int i;
+
+	unsigned long pipes_in_use = 0;
+	unsigned long pending_crtc_idx_mask;
+
+	/* Check that outputs and pipes agree wrt. cloning */
+	for (i = 0; i < display->n_outputs; i++) {
+		output = &display->outputs[i];
+		pending_crtc_idx_mask = 1 << output->pending_pipe;
+
+		pipe = igt_output_get_driving_pipe(output);
+		if (pipe) {
+			igt_assert_f(pipe->outputs & (1 << igt_output_idx(output)),
+				     "Output %s not expected to be using pipe %s\n",
+				     igt_output_name(output),
+				     kmstest_pipe_name(pipe->pipe));
+
+			if (pipes_in_use & pending_crtc_idx_mask)
+				LOG(display, "Output %s clones pipe %s\n",
+				    igt_output_name(output),
+				    kmstest_pipe_name(pipe->pipe));
+		}
+
+		pipes_in_use |= pending_crtc_idx_mask;
+
+		if (output->force_reprobe)
+			igt_output_refresh(output);
+	}
+}
+
 static igt_plane_t *igt_pipe_get_plane(igt_pipe_t *pipe, int plane_idx)
 {
 	igt_require_f(plane_idx >= 0 && plane_idx < pipe->n_planes,
@@ -3766,6 +3775,7 @@ void igt_output_override_mode(igt_output_t *output, const drmModeModeInfo *mode)
 	output->use_override_mode = !!mode;
 
 	if (pipe) {
+		igt_debug("overriding pipe mode in %s way\n", output->display->is_atomic ? "atomic" : "legacy");
 		if (output->display->is_atomic)
 			igt_pipe_obj_replace_prop_blob(pipe, IGT_CRTC_MODE_ID, igt_output_get_mode(output), sizeof(*mode));
 		else
@@ -3773,6 +3783,16 @@ void igt_output_override_mode(igt_output_t *output, const drmModeModeInfo *mode)
 	}
 }
 
+void igt_output_clone_pipe(igt_output_t *output, enum pipe pipe)
+{
+	igt_display_t *display = output->display;
+	uint32_t current_clones = display->pipes[pipe].outputs;
+
+	igt_output_set_pipe(output, pipe);
+
+	display->pipes[pipe].outputs |= current_clones;
+}
+
 /*
  * igt_output_set_pipe:
  * @output: Target output for which the pipe is being set to
@@ -3789,11 +3809,15 @@ void igt_output_set_pipe(igt_output_t *output, enum pipe pipe)
 
 	igt_assert(output->name);
 
-	if (output->pending_pipe != PIPE_NONE)
+	if (output->pending_pipe != PIPE_NONE) {
 		old_pipe = igt_output_get_driving_pipe(output);
+		old_pipe->outputs &= ~(1 << igt_output_idx(output));
+	}
 
-	if (pipe != PIPE_NONE)
+	if (pipe != PIPE_NONE) {
 		pipe_obj = &display->pipes[pipe];
+		pipe_obj->outputs = (1 << igt_output_idx(output));
+	}
 
 	LOG(display, "%s: set_pipe(%s)\n", igt_output_name(output),
 	    kmstest_pipe_name(pipe));
lib/igt_kms.h +4 -0
index cacc6b90..676839bb 100644
@@ -354,6 +354,8 @@ struct igt_pipe {
 	uint32_t crtc_id;
 
 	int32_t out_fence_fd;
+
+	uint32_t outputs;
 };
 
 typedef struct {
@@ -411,6 +413,8 @@ const char *igt_output_name(igt_output_t *output);
 drmModeModeInfo *igt_output_get_mode(igt_output_t *output);
 void igt_output_override_mode(igt_output_t *output, const drmModeModeInfo *mode);
 void igt_output_set_pipe(igt_output_t *output, enum pipe pipe);
+void igt_output_clone_pipe(igt_output_t *output, enum pipe pipe);
+
 igt_plane_t *igt_output_get_plane(igt_output_t *output, int plane_idx);
 igt_plane_t *igt_output_get_plane_type(igt_output_t *output, int plane_type);
 int igt_output_count_plane_type(igt_output_t *output, int plane_type);
Quoted 4 lines

_______________________________________________
igt-dev mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/igt-dev

_______________________________________________
igt-dev mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/igt-dev

#15 lore raw

Re: [igt-dev] [PATCH V6 i-g-t 2/6] kms_writeback: Add initial writeback tests

In reply to Rodrigo Siqueira

Hi Rodrigo,

Quoted 172 lines · Rodrigo Siqueira
Quoted 6 lines · Simon

Hi,

Thanks for the patch! Here are a few comments.

For bonus points, it would be nice to add igt_describe descriptions of each sub-test.

Hi Simon,

First of all, thanks for your feedback; I already applied most of your suggestions. I just have some inline comments/questions.

Quoted 121 lines · Brian Starkey
Quoted 120 lines

Add tests for the WRITEBACK_PIXEL_FORMATS, WRITEBACK_OUT_FENCE_PTR and WRITEBACK_FB_ID properties on writeback connectors, ensuring their behaviour is correct.

[rebased and updated do_writeback_test() function to address feedback]
---
 tests/Makefile.sources |   1 +
 tests/kms_writeback.c  | 314 +++++++++++++++++++++++++++++++++++++++++
 tests/meson.build      |   1 +
 3 files changed, 316 insertions(+)
 create mode 100644 tests/kms_writeback.c
tests/Makefile.sources +1 -0
index 027ed82f..03cc8efa 100644
@@ -77,6 +77,7 @@ TESTS_progs = \
 	kms_universal_plane \
 	kms_vblank \
 	kms_vrr \
+	kms_writeback \
 	meta_test \
 	perf \
 	perf_pmu \
tests/kms_writeback.c new file +88 -0
new file mode 100644
index 00000000..66ef48a6
@@ -0,0 +1,314 @@
+/*
+ * (C) COPYRIGHT 2017 ARM Limited. All rights reserved.
+ *
+ * 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.
+ *
+ */
+
+#include <errno.h>
+#include <stdbool.h>
+#include <stdio.h>
+#include <string.h>
+
+#include "igt.h"
+#include "igt_core.h"
+#include "igt_fb.h"
+
+static drmModePropertyBlobRes *get_writeback_formats_blob(igt_output_t *output)
+{
+	drmModePropertyBlobRes *blob = NULL;
+	uint64_t blob_id;
+	int ret;
+
+	ret = kmstest_get_property(output->display->drm_fd,
+				   output->config.connector->connector_id,
+				   DRM_MODE_OBJECT_CONNECTOR,
+				   igt_connector_prop_names[IGT_CONNECTOR_WRITEBACK_PIXEL_FORMATS],
+				   NULL, &blob_id, NULL);
+	if (ret)
+		blob = drmModeGetPropertyBlob(output->display->drm_fd, blob_id);
+
+	igt_assert(blob);
+
+	return blob;
+}
+
+static bool check_writeback_config(igt_display_t *display, igt_output_t *output)
+{
+	igt_fb_t input_fb, output_fb;
+	igt_plane_t *plane;
+	uint32_t writeback_format = DRM_FORMAT_XRGB8888;
+	uint64_t tiling = igt_fb_mod_to_tiling(0);
+	int width, height, ret;
+	drmModeModeInfo override_mode = {
+		.clock = 25175,
+		.hdisplay = 640,
+		.hsync_start = 656,
+		.hsync_end = 752,
+		.htotal = 800,
+		.hskew = 0,
+		.vdisplay = 480,
+		.vsync_start = 490,
+		.vsync_end = 492,
+		.vtotal = 525,
+		.vscan = 0,
+		.vrefresh = 60,
+		.flags = DRM_MODE_FLAG_NHSYNC | DRM_MODE_FLAG_NVSYNC,
+		.name = {"640x480-60"},
+	};
+	igt_output_override_mode(output, &override_mode);
+
+	width = override_mode.hdisplay;
+	height = override_mode.vdisplay;
+
+	ret = igt_create_fb(display->drm_fd, width, height, DRM_FORMAT_XRGB8888, tiling, &input_fb);
+	igt_assert(ret >= 0);
+
+	ret = igt_create_fb(display->drm_fd, width, height, writeback_format, tiling, &output_fb);
+	igt_assert(ret >= 0);
+
+	plane = igt_output_get_plane_type(output, DRM_PLANE_TYPE_PRIMARY);
+	igt_plane_set_fb(plane, &input_fb);
Quoted 8 lines

tests/kms_writeback.c line 92

Quoted 4 lines
tests/kms_writeback.c new file +4 -0
+	igt_output_set_writeback_fb(output, &output_fb);
+
+	ret = igt_display_try_commit_atomic(display, DRM_MODE_ATOMIC_TEST_ONLY |
+					    DRM_MODE_ATOMIC_ALLOW_MODESET, NULL);

Okay, we're using atomic test-only mode to know if we can perform tests with the writeback output. It's probably fine, but we don't use WRITEBACK_PIXEL_FORMATS, which makes me a little bit sad.

Sorry, I did not fully understand part. Did you expect to see something like the below code before igt_display_try_commit_atomic()?

igt_output_set_prop_value(output, WRITEBACK_PIXEL_FORMATS,

                           DRM_FORMAT_XRGB8888);
Quoted 14 lines
Quoted 14 lines
tests/kms_writeback.c new file +14 -0
+	igt_plane_set_fb(plane, NULL);
+	igt_remove_fb(display->drm_fd, &input_fb);
+	igt_remove_fb(display->drm_fd, &output_fb);
+
+	return !ret;
+}
+
+static igt_output_t *kms_writeback_get_output(igt_display_t *display)
+{
+	int i;
+
+	for (i = 0; i < display->n_outputs; i++) {
+		igt_output_t *output = &display->outputs[i];
+		int j;
Quoted 8 lines

tests/kms_writeback.c line 111

Quoted 5 lines
tests/kms_writeback.c new file +5 -0
+
+		if (output->config.connector->connector_type != DRM_MODE_CONNECTOR_WRITEBACK)
+			continue;
+
+		kmstest_force_connector(display->drm_fd, output->config.connector, FORCE_CONNECTOR_ON);

Hmm. Is this really necessary? "Real" userspace won't do this, so I don't think we need to either.

Quoted 7 lines

Hmmm, I have a little experience with userspace; however, I tested kms_writeback on top of vkms without this line, and everything worked well. If I remove this line, should I also drop the line that force connector to FORCE_CONNECTOR_UNSPECIFIED?

Another question, if FORCE_CONNECTOR_ON is something that userspace won't want to do, why do we have it?

This is probably a left-over from the times when the design of the writeback connectors called for them to be always disconnected, in order not to trip old userspace. Last iteration before upstreaming the writeback into DRM we switched to an earlier design where DRM_CLIENT_CAP_WRITEBACK_CONNECTORS is needed in order to expose the writeback connector, and then it is reported as connected. You can drop this from the patch.

Quoted 58 lines
Quoted 49 lines

tests/kms_writeback.c line 121

Quoted 10 lines
tests/kms_writeback.c new file +9 -0
+		for (j = 0; j < igt_display_get_n_pipes(display); j++) {
+			igt_output_set_pipe(output, j);
+
+			if (check_writeback_config(display, output)) {
+				igt_debug("Using connector %u:%s on pipe %d\n",
+					  output->config.connector->connector_id,
+					  output->name, j);
+				return output;
+			}

We could probably add an igt_debug statement in case we don't use this writeback output.

Quoted 10 lines
tests/kms_writeback.c new file +10 -0
+		}
+
+		/* Restore any connectors we don't use, so we don't trip on them later */
+		kmstest_force_connector(display->drm_fd, output->config.connector, FORCE_CONNECTOR_UNSPECIFIED);
+	}
+
+	return NULL;
+}
+
+static void check_writeback_fb_id(igt_output_t *output)

tests/kms_writeback.c line 141

Quoted 10 lines
tests/kms_writeback.c new file +10 -0
+{
+	uint64_t check_fb_id;
+
+	check_fb_id = igt_output_get_prop(output, IGT_CONNECTOR_WRITEBACK_FB_ID);
+	igt_assert(check_fb_id == 0);
+}
+
+static int do_writeback_test(igt_output_t *output, uint32_t flags,
+			      uint32_t fb_id, int32_t *out_fence_ptr,
+			      bool ptr_valid)

flags seems to always be set to DRM_MODE_ATOMIC_ALLOW_MODESET. We can probably remove the parameter from this function.

Quoted 11 lines
tests/kms_writeback.c new file +11 -0
+{
+	int ret;
+	igt_display_t *display = output->display;
+	struct kmstest_connector_config *config = &output->config;
+
+	igt_output_set_prop_value(output, IGT_CONNECTOR_CRTC_ID, config->crtc->crtc_id);
+	igt_output_set_prop_value(output, IGT_CONNECTOR_WRITEBACK_FB_ID, fb_id);
+	igt_output_set_prop_value(output, IGT_CONNECTOR_WRITEBACK_OUT_FENCE_PTR, (uint64_t)out_fence_ptr);
+
+	if (ptr_valid)
+		*out_fence_ptr = 0;
Quoted 8 lines

tests/kms_writeback.c line 157

Quoted 5 lines
tests/kms_writeback.c new file +5 -0
+
+	ret = igt_display_try_commit_atomic(display, flags, NULL);
+
+	if (ptr_valid)
+		igt_assert(*out_fence_ptr == -1);

I'm confused. Why should this be -1 even if we
igt_display_try_commit_atomic succeeds?

Quoted 4 lines

I inspected the drm_mode_atomic_ioctl() function and I noticed that It calls complete_signaling() in its turn this function assign -1 to out_fence_ptr. Since all of this happens at the end of atomic_commit, I believe that we don’t need it. I already removed it.

Yeah, I think the reasons for that have been lost into the mist of time.

Quoted 24 lines
Quoted 24 lines
Quoted 24 lines
tests/kms_writeback.c new file +23 -0
+	/* WRITEBACK_FB_ID must always read as zero */
+	check_writeback_fb_id(output);
+
+	return ret;
+}
+
+static void invalid_out_fence(igt_output_t *output, igt_fb_t *valid_fb, igt_fb_t *invalid_fb)
+{
+	int i, ret;
+	int32_t out_fence;
+	struct {
+		uint32_t fb_id;
+		bool ptr_valid;
+		int32_t *out_fence_ptr;
+	} invalid_tests[] = {
+		{
+			/* No output buffer, but the WRITEBACK_OUT_FENCE_PTR set. */
+			.fb_id = 0,
+			.ptr_valid = true,
+			.out_fence_ptr = &out_fence,
+		},
+		{
+			/* Invalid output buffer. */
Quoted 8 lines
Quoted 8 lines

tests/kms_writeback.c line 185

Quoted 4 lines
tests/kms_writeback.c new file +4 -0
+			.fb_id = invalid_fb->fb_id,
+			.ptr_valid = true,
+			.out_fence_ptr = &out_fence,
+		},

This doesn't belong in this function (invalid_out_fence), since this checks an invalid framebuffer, not an invalid fence. We should probably move it to writeback_fb_id (and rename that function to test_fb?).

It tries to test that you can't trick the driver to do any work on a fence if the framebuffer is invalid, so the set of tests tries: no fb with valid fence, invalid fb with valid fence, valid fb but invalid fence and assumes that no fb with invalid fence is a NOP anyway.

Quoted 81 lines
Quoted 81 lines
Quoted 5 lines
tests/kms_writeback.c new file +4 -0
+		{
+			/* Invalid WRITEBACK_OUT_FENCE_PTR. */
+			.fb_id = valid_fb->fb_id,
+			.ptr_valid = false,

tests/kms_writeback.c line 199

Quoted 10 lines
tests/kms_writeback.c new file +10 -0
+			.out_fence_ptr = (int32_t *)0x8,
+		},
+	};
+
+	for (i = 0; i < ARRAY_SIZE(invalid_tests); i++) {
+		ret = do_writeback_test(output, DRM_MODE_ATOMIC_ALLOW_MODESET,
+					invalid_tests[i].fb_id,
+					invalid_tests[i].out_fence_ptr,
+					invalid_tests[i].ptr_valid);
+		igt_assert(ret != 0);

Maybe we can check for -ret == EINVAL?

tests/kms_writeback.c line 203

Quoted 4 lines
tests/kms_writeback.c new file +4 -0
+	}
+}
+
+static void writeback_fb_id(igt_output_t *output, igt_fb_t *valid_fb, igt_fb_t *invalid_fb)

invalid_fb doesn't seem to be used here. valid_fb seems to be set to the input framebuffer. It's probably not a good idea to use the same FB for input and writeback output.

Quoted 41 lines
tests/kms_writeback.c new file +41 -0
+{
+
+	int ret;
+
+	/* Valid output buffer */
+	ret = do_writeback_test(output, DRM_MODE_ATOMIC_ALLOW_MODESET,
+				valid_fb->fb_id, NULL, false);
+	igt_assert(ret == 0);
+
+	/* Invalid object for WRITEBACK_FB_ID */
+	ret = do_writeback_test(output, DRM_MODE_ATOMIC_ALLOW_MODESET,
+				output->id, NULL, false);
+	igt_assert(ret == -EINVAL);
+
+	/* Zero WRITEBACK_FB_ID */
+	ret = do_writeback_test(output, DRM_MODE_ATOMIC_ALLOW_MODESET,
+				0, NULL, false);
+	igt_assert(ret == 0);
+}
+
+igt_main
+{
+	igt_display_t display;
+	igt_output_t *output;
+	igt_plane_t *plane;
+	igt_fb_t input_fb;
+	drmModeModeInfo mode;
+	int ret;
+
+	memset(&display, 0, sizeof(display));
+
+	igt_fixture {
+		display.drm_fd = drm_open_driver_master(DRIVER_ANY);
+		igt_display_require(&display, display.drm_fd);
+
+		kmstest_set_vt_graphics_mode();
+
+		igt_display_require(&display, display.drm_fd);
+
+		igt_require(display.is_atomic);
+

tests/kms_writeback.c line 254

Quoted 10 lines
tests/kms_writeback.c new file +10 -0
+		output = kms_writeback_get_output(&display);
+		igt_require(output);
+
+		if (output->use_override_mode)
+			memcpy(&mode, &output->override_mode, sizeof(mode));
+		else
+			memcpy(&mode, &output->config.default_mode, sizeof(mode));
+
+		plane = igt_output_get_plane_type(output, DRM_PLANE_TYPE_PRIMARY);
+		igt_require(plane);

Maybe we can assert on this?

Quoted 7 lines
Quoted 7 lines

tests/kms_writeback.c line 259

Quoted 4 lines
tests/kms_writeback.c new file +4 -0
+		ret = igt_create_fb(display.drm_fd, mode.hdisplay,
+				    mode.vdisplay,
+				    DRM_FORMAT_XRGB8888,
+				    igt_fb_mod_to_tiling(0),

This is supposed to take a modifier. DRM_FORMAT_MOD_LINEAR would be better to make this clear.

Agree. The patchset pre-dates the modifiers support (or has the same age, I forgot)

Quoted 24 lines
Quoted 24 lines

(Applies to other lines of this patch)

tests/kms_writeback.c line 266

Quoted 7 lines
tests/kms_writeback.c new file +7 -0
+				    &input_fb);
+		igt_assert(ret >= 0);
+		igt_plane_set_fb(plane, &input_fb);
+	}
+
+	igt_subtest("writeback-pixel-formats") {
+		drmModePropertyBlobRes *formats_blob = get_writeback_formats_blob(output);

Need to drmModeFreePropertyBlob this.

Quoted 11 lines
tests/kms_writeback.c new file +11 -0
+		const char *valid_chars = "0123456 ABCGNRUVXY";
+		unsigned int i;
+		char *c;
+
+		/*
+		 * We don't have a comprehensive list of formats, so just check
+		 * that the blob length is sensible and that it doesn't contain
+		 * any outlandish characters
+		 */
+		igt_assert(!(formats_blob->length % 4));
+		c = formats_blob->data;
Quoted 8 lines
Quoted 8 lines

tests/kms_writeback.c line 280

Quoted 3 lines
tests/kms_writeback.c new file +3 -0
+		for (i = 0; i < formats_blob->length; i++)
+			igt_assert_f(strchr(valid_chars, c[i]),
+				     "Unexpected character %c\n", c[i]);

Honestly, I'm not a fan of this check. There's no guarantee that fourcc codes will be made from ASCII characters, in fact some formats already have non-printable chars in them. I don't want to have to update this test when a new fourcc format is added.

Like the comments says, we don't have a full list of formats to check against. Suggestions on how to improve are welcome, but I don't think we should delay (any longer) the patchset due to this.

Best regards,
Liviu

Quoted 75 lines
Quoted 4 lines

We currently don't have a test for IN_FORMATS. If we really want to check formats, we could have a big array of known formats and add a bool is_valid_format(uint32_t fmt) function.

Agree with you. How about remove this test?

Thanks
Best Regards

Quoted 61 lines

tests/kms_writeback.c line 284

Quoted 4 lines
tests/kms_writeback.c new file +4 -0
+	}
+
+	igt_subtest("writeback-invalid-out-fence") {
+		igt_fb_t invalid_fb;

invalid_output_fb would be a better name IMHO.

tests/kms_writeback.c line 290

Quoted 6 lines
tests/kms_writeback.c new file +6 -0
+		ret = igt_create_fb(display.drm_fd, mode.hdisplay / 2,
+				    mode.vdisplay / 2,
+				    DRM_FORMAT_XRGB8888,
+				    igt_fb_mod_to_tiling(0),
+				    &invalid_fb);
+		igt_require(ret > 0);

We should probably use a different variable: ret is signed, igt_create_fb isn't. Also ret doesn't make it clear that the return value is the KMS framebuffer ID.

(Applies to other subtests)

tests/kms_writeback.c line 292

Quoted 1 line
tests/kms_writeback.c new file +1 -0
+		invalid_out_fence(output, &input_fb, &invalid_fb);

(Still not sure why we provide the input FB to this function.)

Quoted 37 lines
tests/kms_writeback.c new file +21 -0
+		igt_remove_fb(display.drm_fd, &invalid_fb);
+	}
+
+	igt_subtest("writeback-fb-id") {
+		igt_fb_t output_fb;
+		ret = igt_create_fb(display.drm_fd, mode.hdisplay, mode.vdisplay,
+				    DRM_FORMAT_XRGB8888,
+				    igt_fb_mod_to_tiling(0),
+				    &output_fb);
+		igt_require(ret > 0);
+
+		writeback_fb_id(output, &input_fb, &output_fb);
+
+		igt_remove_fb(display.drm_fd, &output_fb);
+	}
+
+	igt_fixture {
+		igt_remove_fb(display.drm_fd, &input_fb);
+		igt_display_fini(&display);
+	}
+}
tests/meson.build +1 -0
index f168fbba..9c77cfcd 100644
@@ -63,6 +63,7 @@ test_progs = [
 	'kms_universal_plane',
 	'kms_vblank',
 	'kms_vrr',
+	'kms_writeback',
 	'meta_test',
 	'panfrost_get_param',
 	'panfrost_gem_new',

_______________________________________________
igt-dev mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/igt-dev

Signature · 2 lines
Rodrigo Siqueira
https://siqueira.tech
Signature · 12 lines
====================
| I would like to |
| fix the world,  |
| but they're not |
| giving me the   |
 \ source code!  /
  ---------------
    ¯\_(ツ)_/¯
_______________________________________________
igt-dev mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/igt-dev
#13 lore raw

Re: [igt-dev] [PATCH V6 i-g-t 2/6] kms_writeback: Add initial writeback tests

In reply to Ser, Simon
Quoted 195 lines · Simon
Quoted 143 lines · Rodrigo Siqueira
Quoted 6 lines · Simon

Hi,

Thanks for the patch! Here are a few comments.

For bonus points, it would be nice to add igt_describe descriptions of each sub-test.

Hi Simon,

First of all, thanks for your feedback; I already applied most of your suggestions. I just have some inline comments/questions.

Quoted 121 lines · Brian Starkey
Quoted 120 lines

Add tests for the WRITEBACK_PIXEL_FORMATS, WRITEBACK_OUT_FENCE_PTR and WRITEBACK_FB_ID properties on writeback connectors, ensuring their behaviour is correct.

[rebased and updated do_writeback_test() function to address feedback]
---
 tests/Makefile.sources |   1 +
 tests/kms_writeback.c  | 314 +++++++++++++++++++++++++++++++++++++++++
 tests/meson.build      |   1 +
 3 files changed, 316 insertions(+)
 create mode 100644 tests/kms_writeback.c
tests/Makefile.sources +1 -0
index 027ed82f..03cc8efa 100644
@@ -77,6 +77,7 @@ TESTS_progs = \
 	kms_universal_plane \
 	kms_vblank \
 	kms_vrr \
+	kms_writeback \
 	meta_test \
 	perf \
 	perf_pmu \
tests/kms_writeback.c new file +88 -0
new file mode 100644
index 00000000..66ef48a6
@@ -0,0 +1,314 @@
+/*
+ * (C) COPYRIGHT 2017 ARM Limited. All rights reserved.
+ *
+ * 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.
+ *
+ */
+
+#include <errno.h>
+#include <stdbool.h>
+#include <stdio.h>
+#include <string.h>
+
+#include "igt.h"
+#include "igt_core.h"
+#include "igt_fb.h"
+
+static drmModePropertyBlobRes *get_writeback_formats_blob(igt_output_t *output)
+{
+	drmModePropertyBlobRes *blob = NULL;
+	uint64_t blob_id;
+	int ret;
+
+	ret = kmstest_get_property(output->display->drm_fd,
+				   output->config.connector->connector_id,
+				   DRM_MODE_OBJECT_CONNECTOR,
+				   igt_connector_prop_names[IGT_CONNECTOR_WRITEBACK_PIXEL_FORMATS],
+				   NULL, &blob_id, NULL);
+	if (ret)
+		blob = drmModeGetPropertyBlob(output->display->drm_fd, blob_id);
+
+	igt_assert(blob);
+
+	return blob;
+}
+
+static bool check_writeback_config(igt_display_t *display, igt_output_t *output)
+{
+	igt_fb_t input_fb, output_fb;
+	igt_plane_t *plane;
+	uint32_t writeback_format = DRM_FORMAT_XRGB8888;
+	uint64_t tiling = igt_fb_mod_to_tiling(0);
+	int width, height, ret;
+	drmModeModeInfo override_mode = {
+		.clock = 25175,
+		.hdisplay = 640,
+		.hsync_start = 656,
+		.hsync_end = 752,
+		.htotal = 800,
+		.hskew = 0,
+		.vdisplay = 480,
+		.vsync_start = 490,
+		.vsync_end = 492,
+		.vtotal = 525,
+		.vscan = 0,
+		.vrefresh = 60,
+		.flags = DRM_MODE_FLAG_NHSYNC | DRM_MODE_FLAG_NVSYNC,
+		.name = {"640x480-60"},
+	};
+	igt_output_override_mode(output, &override_mode);
+
+	width = override_mode.hdisplay;
+	height = override_mode.vdisplay;
+
+	ret = igt_create_fb(display->drm_fd, width, height, DRM_FORMAT_XRGB8888, tiling, &input_fb);
+	igt_assert(ret >= 0);
+
+	ret = igt_create_fb(display->drm_fd, width, height, writeback_format, tiling, &output_fb);
+	igt_assert(ret >= 0);
+
+	plane = igt_output_get_plane_type(output, DRM_PLANE_TYPE_PRIMARY);
+	igt_plane_set_fb(plane, &input_fb);
Quoted 8 lines

tests/kms_writeback.c line 92

Quoted 4 lines
tests/kms_writeback.c new file +4 -0
+	igt_output_set_writeback_fb(output, &output_fb);
+
+	ret = igt_display_try_commit_atomic(display, DRM_MODE_ATOMIC_TEST_ONLY |
+					    DRM_MODE_ATOMIC_ALLOW_MODESET, NULL);

Okay, we're using atomic test-only mode to know if we can perform tests with the writeback output. It's probably fine, but we don't use WRITEBACK_PIXEL_FORMATS, which makes me a little bit sad.

Quoted 5 lines

Sorry, I did not fully understand part. Did you expect to see something like the below code before igt_display_try_commit_atomic()?

igt_output_set_prop_value(output, WRITEBACK_PIXEL_FORMATS,

                           DRM_FORMAT_XRGB8888);

Hmm, no, we cannot change the list of writeback formats (it's immutable).

This comment doesn't require any action, it's just me thinking aloud :P

I'm just thinking that it would be nice to choose our format depending on the WRITEBACK_PIXEL_FORMATS property. And probably run tests with each supported format (or, alternatively, choose a random one). That way we can make sure WRITEBACK_PIXEL_FORMATS isn't broken.

However, this can be left for a later patch.

Quoted 23 lines
Quoted 14 lines
Quoted 14 lines
tests/kms_writeback.c new file +14 -0
+	igt_plane_set_fb(plane, NULL);
+	igt_remove_fb(display->drm_fd, &input_fb);
+	igt_remove_fb(display->drm_fd, &output_fb);
+
+	return !ret;
+}
+
+static igt_output_t *kms_writeback_get_output(igt_display_t *display)
+{
+	int i;
+
+	for (i = 0; i < display->n_outputs; i++) {
+		igt_output_t *output = &display->outputs[i];
+		int j;
Quoted 8 lines

tests/kms_writeback.c line 111

Quoted 5 lines
tests/kms_writeback.c new file +5 -0
+
+		if (output->config.connector->connector_type != DRM_MODE_CONNECTOR_WRITEBACK)
+			continue;
+
+		kmstest_force_connector(display->drm_fd, output->config.connector, FORCE_CONNECTOR_ON);

Hmm. Is this really necessary? "Real" userspace won't do this, so I don't think we need to either.

Quoted 4 lines

Hmmm, I have a little experience with userspace; however, I tested kms_writeback on top of vkms without this line, and everything worked well. If I remove this line, should I also drop the line that force connector to FORCE_CONNECTOR_UNSPECIFIED?

I believe so.

Quoted 2 lines

Another question, if FORCE_CONNECTOR_ON is something that userspace won't want to do, why do we have it?

Quoted 8 lines

We use kmstest_force_connector in injection tests: those pick a disconnected HDMI connector and trick the kernel into thinking it's connected. We generally also force an EDID and this is used to emulate a screen (e.g. a screen that supports audio, 4K or 3D).

This is only meant to be used for testing though, hence "real userspace shouldn't use it".

Ahhh, I see. Thanks for the explanation

Quoted 62 lines
Quoted 57 lines
Quoted 48 lines

tests/kms_writeback.c line 121

Quoted 9 lines
tests/kms_writeback.c new file +9 -0
+		for (j = 0; j < igt_display_get_n_pipes(display); j++) {
+			igt_output_set_pipe(output, j);
+
+			if (check_writeback_config(display, output)) {
+				igt_debug("Using connector %u:%s on pipe %d\n",
+					  output->config.connector->connector_id,
+					  output->name, j);
+				return output;
+			}

We could probably add an igt_debug statement in case we don't use this writeback output.

Quoted 10 lines
tests/kms_writeback.c new file +10 -0
+		}
+
+		/* Restore any connectors we don't use, so we don't trip on them later */
+		kmstest_force_connector(display->drm_fd, output->config.connector, FORCE_CONNECTOR_UNSPECIFIED);
+	}
+
+	return NULL;
+}
+
+static void check_writeback_fb_id(igt_output_t *output)

tests/kms_writeback.c line 141

Quoted 10 lines
tests/kms_writeback.c new file +10 -0
+{
+	uint64_t check_fb_id;
+
+	check_fb_id = igt_output_get_prop(output, IGT_CONNECTOR_WRITEBACK_FB_ID);
+	igt_assert(check_fb_id == 0);
+}
+
+static int do_writeback_test(igt_output_t *output, uint32_t flags,
+			      uint32_t fb_id, int32_t *out_fence_ptr,
+			      bool ptr_valid)

flags seems to always be set to DRM_MODE_ATOMIC_ALLOW_MODESET. We can probably remove the parameter from this function.

Quoted 11 lines
tests/kms_writeback.c new file +11 -0
+{
+	int ret;
+	igt_display_t *display = output->display;
+	struct kmstest_connector_config *config = &output->config;
+
+	igt_output_set_prop_value(output, IGT_CONNECTOR_CRTC_ID, config->crtc->crtc_id);
+	igt_output_set_prop_value(output, IGT_CONNECTOR_WRITEBACK_FB_ID, fb_id);
+	igt_output_set_prop_value(output, IGT_CONNECTOR_WRITEBACK_OUT_FENCE_PTR, (uint64_t)out_fence_ptr);
+
+	if (ptr_valid)
+		*out_fence_ptr = 0;
Quoted 8 lines

tests/kms_writeback.c line 157

Quoted 5 lines
tests/kms_writeback.c new file +5 -0
+
+	ret = igt_display_try_commit_atomic(display, flags, NULL);
+
+	if (ptr_valid)
+		igt_assert(*out_fence_ptr == -1);

I'm confused. Why should this be -1 even if we
igt_display_try_commit_atomic succeeds?

Quoted 4 lines

I inspected the drm_mode_atomic_ioctl() function and I noticed that It calls complete_signaling() in its turn this function assign -1 to out_fence_ptr. Since all of this happens at the end of atomic_commit, I believe that we don’t need it. I already removed it.

Quoted 8 lines

I think I'm still confused :)

I'm probably missing something. As far as I understand, we are doing an atomic commit, sometimes with a valid WRITEBACK_FB_ID and WRITEBACK_OUT_FENCE_PTR. In this case it should start the writeback process in the kernel and we should get a valid out_fence_ptr?

Why do we always get -1?

FWIU userspace uses the WRITEBACK_OUT_FENCE_PTR to provide a pointer for the kernel to fill with a sync_file file descriptor, which will signal once the writeback is finished. If the signal was correctly handled, the out_fence_ptr would be set to -1, because of this we always get -1.

Thanks for your feedback, I'll prepare a new version soon.

Quoted 227 lines
Quoted 152 lines
Quoted 152 lines
Quoted 17 lines
tests/kms_writeback.c new file +17 -0
+	/* WRITEBACK_FB_ID must always read as zero */
+	check_writeback_fb_id(output);
+
+	return ret;
+}
+
+static void invalid_out_fence(igt_output_t *output, igt_fb_t *valid_fb, igt_fb_t *invalid_fb)
+{
+	int i, ret;
+	int32_t out_fence;
+	struct {
+		uint32_t fb_id;
+		bool ptr_valid;
+		int32_t *out_fence_ptr;
+	} invalid_tests[] = {
+		{
+			/* No output buffer, but the WRITEBACK_OUT_FENCE_PTR set. */

tests/kms_writeback.c line 185

Quoted 10 lines
tests/kms_writeback.c new file +10 -0
+			.fb_id = 0,
+			.ptr_valid = true,
+			.out_fence_ptr = &out_fence,
+		},
+		{
+			/* Invalid output buffer. */
+			.fb_id = invalid_fb->fb_id,
+			.ptr_valid = true,
+			.out_fence_ptr = &out_fence,
+		},

This doesn't belong in this function (invalid_out_fence), since this checks an invalid framebuffer, not an invalid fence. We should probably move it to writeback_fb_id (and rename that function to test_fb?).

Quoted 4 lines
tests/kms_writeback.c new file +4 -0
+		{
+			/* Invalid WRITEBACK_OUT_FENCE_PTR. */
+			.fb_id = valid_fb->fb_id,
+			.ptr_valid = false,

tests/kms_writeback.c line 199

Quoted 10 lines
tests/kms_writeback.c new file +10 -0
+			.out_fence_ptr = (int32_t *)0x8,
+		},
+	};
+
+	for (i = 0; i < ARRAY_SIZE(invalid_tests); i++) {
+		ret = do_writeback_test(output, DRM_MODE_ATOMIC_ALLOW_MODESET,
+					invalid_tests[i].fb_id,
+					invalid_tests[i].out_fence_ptr,
+					invalid_tests[i].ptr_valid);
+		igt_assert(ret != 0);

Maybe we can check for -ret == EINVAL?

tests/kms_writeback.c line 203

Quoted 4 lines
tests/kms_writeback.c new file +4 -0
+	}
+}
+
+static void writeback_fb_id(igt_output_t *output, igt_fb_t *valid_fb, igt_fb_t *invalid_fb)

invalid_fb doesn't seem to be used here. valid_fb seems to be set to the input framebuffer. It's probably not a good idea to use the same FB for input and writeback output.

Quoted 41 lines
tests/kms_writeback.c new file +41 -0
+{
+
+	int ret;
+
+	/* Valid output buffer */
+	ret = do_writeback_test(output, DRM_MODE_ATOMIC_ALLOW_MODESET,
+				valid_fb->fb_id, NULL, false);
+	igt_assert(ret == 0);
+
+	/* Invalid object for WRITEBACK_FB_ID */
+	ret = do_writeback_test(output, DRM_MODE_ATOMIC_ALLOW_MODESET,
+				output->id, NULL, false);
+	igt_assert(ret == -EINVAL);
+
+	/* Zero WRITEBACK_FB_ID */
+	ret = do_writeback_test(output, DRM_MODE_ATOMIC_ALLOW_MODESET,
+				0, NULL, false);
+	igt_assert(ret == 0);
+}
+
+igt_main
+{
+	igt_display_t display;
+	igt_output_t *output;
+	igt_plane_t *plane;
+	igt_fb_t input_fb;
+	drmModeModeInfo mode;
+	int ret;
+
+	memset(&display, 0, sizeof(display));
+
+	igt_fixture {
+		display.drm_fd = drm_open_driver_master(DRIVER_ANY);
+		igt_display_require(&display, display.drm_fd);
+
+		kmstest_set_vt_graphics_mode();
+
+		igt_display_require(&display, display.drm_fd);
+
+		igt_require(display.is_atomic);
+

tests/kms_writeback.c line 254

Quoted 10 lines
tests/kms_writeback.c new file +10 -0
+		output = kms_writeback_get_output(&display);
+		igt_require(output);
+
+		if (output->use_override_mode)
+			memcpy(&mode, &output->override_mode, sizeof(mode));
+		else
+			memcpy(&mode, &output->config.default_mode, sizeof(mode));
+
+		plane = igt_output_get_plane_type(output, DRM_PLANE_TYPE_PRIMARY);
+		igt_require(plane);

Maybe we can assert on this?

tests/kms_writeback.c line 259

Quoted 4 lines
tests/kms_writeback.c new file +4 -0
+		ret = igt_create_fb(display.drm_fd, mode.hdisplay,
+				    mode.vdisplay,
+				    DRM_FORMAT_XRGB8888,
+				    igt_fb_mod_to_tiling(0),

This is supposed to take a modifier. DRM_FORMAT_MOD_LINEAR would be better to make this clear.

(Applies to other lines of this patch)

tests/kms_writeback.c line 266

Quoted 7 lines
tests/kms_writeback.c new file +7 -0
+				    &input_fb);
+		igt_assert(ret >= 0);
+		igt_plane_set_fb(plane, &input_fb);
+	}
+
+	igt_subtest("writeback-pixel-formats") {
+		drmModePropertyBlobRes *formats_blob = get_writeback_formats_blob(output);

Need to drmModeFreePropertyBlob this.

Quoted 4 lines
tests/kms_writeback.c new file +4 -0
+		const char *valid_chars = "0123456 ABCGNRUVXY";
+		unsigned int i;
+		char *c;
+

tests/kms_writeback.c line 280

Quoted 10 lines
tests/kms_writeback.c new file +10 -0
+		/*
+		 * We don't have a comprehensive list of formats, so just check
+		 * that the blob length is sensible and that it doesn't contain
+		 * any outlandish characters
+		 */
+		igt_assert(!(formats_blob->length % 4));
+		c = formats_blob->data;
+		for (i = 0; i < formats_blob->length; i++)
+			igt_assert_f(strchr(valid_chars, c[i]),
+				     "Unexpected character %c\n", c[i]);

Honestly, I'm not a fan of this check. There's no guarantee that fourcc codes will be made from ASCII characters, in fact some formats already have non-printable chars in them. I don't want to have to update this test when a new fourcc format is added.

Quoted 5 lines
Quoted 3 lines

We currently don't have a test for IN_FORMATS. If we really want to check formats, we could have a big array of known formats and add a bool is_valid_format(uint32_t fmt) function.

Agree with you. How about remove this test?

I guess we could always keep the length % 4 test, even if it's just a sanity test. At least this one won't ever need to be changed.

I don't feel strongly about it.

Quoted 64 lines

Thanks
Best Regards

Quoted 61 lines

tests/kms_writeback.c line 284

Quoted 4 lines
tests/kms_writeback.c new file +4 -0
+	}
+
+	igt_subtest("writeback-invalid-out-fence") {
+		igt_fb_t invalid_fb;

invalid_output_fb would be a better name IMHO.

tests/kms_writeback.c line 290

Quoted 6 lines
tests/kms_writeback.c new file +6 -0
+		ret = igt_create_fb(display.drm_fd, mode.hdisplay / 2,
+				    mode.vdisplay / 2,
+				    DRM_FORMAT_XRGB8888,
+				    igt_fb_mod_to_tiling(0),
+				    &invalid_fb);
+		igt_require(ret > 0);

We should probably use a different variable: ret is signed, igt_create_fb isn't. Also ret doesn't make it clear that the return value is the KMS framebuffer ID.

(Applies to other subtests)

tests/kms_writeback.c line 292

Quoted 1 line
tests/kms_writeback.c new file +1 -0
+		invalid_out_fence(output, &input_fb, &invalid_fb);

(Still not sure why we provide the input FB to this function.)

Quoted 37 lines
tests/kms_writeback.c new file +21 -0
+		igt_remove_fb(display.drm_fd, &invalid_fb);
+	}
+
+	igt_subtest("writeback-fb-id") {
+		igt_fb_t output_fb;
+		ret = igt_create_fb(display.drm_fd, mode.hdisplay, mode.vdisplay,
+				    DRM_FORMAT_XRGB8888,
+				    igt_fb_mod_to_tiling(0),
+				    &output_fb);
+		igt_require(ret > 0);
+
+		writeback_fb_id(output, &input_fb, &output_fb);
+
+		igt_remove_fb(display.drm_fd, &output_fb);
+	}
+
+	igt_fixture {
+		igt_remove_fb(display.drm_fd, &input_fb);
+		igt_display_fini(&display);
+	}
+}
tests/meson.build +1 -0
index f168fbba..9c77cfcd 100644
@@ -63,6 +63,7 @@ test_progs = [
 	'kms_universal_plane',
 	'kms_vblank',
 	'kms_vrr',
+	'kms_writeback',
 	'meta_test',
 	'panfrost_get_param',
 	'panfrost_gem_new',

_______________________________________________
igt-dev mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/igt-dev

Signature · 2 lines
Rodrigo Siqueira
https://siqueira.tech

_______________________________________________
igt-dev mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/igt-dev

#27 lore raw

Re: [igt-dev] [PATCH V6 i-g-t 3/6] lib: Add function to hash a framebuffer

In reply to Ser, Simon
Quoted 76 lines · Simon
Quoted 63 lines · Rodrigo Siqueira
Quoted 62 lines · Simon
Quoted 51 lines · Simon

Mostly LGTM, here are a few nits.

Quoted 35 lines · Brian Starkey

To use writeback buffers as a CRC source, we need to be able to hash them. Implement a simple FVA-1a hashing routine for this purpose.

Doing a bytewise hash on the framebuffer directly can be very slow if the memory is noncached. By making a copy of each line in the FB first (which can take advantage of word-access speedup), we can do the hash on a cached copy, which is much faster (10x speedup on my platform).

v6: use igt_memcpy_from_wc() instead of plain memcpy, as suggested by
Chris Wilson

[rebased and updated to the most recent API]
---
 lib/igt_fb.c | 66 ++++++++++++++++++++++++++++++++++++++++++++++++++++
 lib/igt_fb.h |  3 +++
 2 files changed, 69 insertions(+)
lib/igt_fb.c +8 -0
index 9d4f905e..d07dae39 100644
@@ -3256,6 +3256,72 @@ bool igt_fb_supported_format(uint32_t drm_format)
 	return false;
 }
 
+/*
+ * This implements the FNV-1a hashing algorithm instead of CRC, for
+ * simplicity
+ * http://www.isthe.com/chongo/tech/comp/fnv/index.html
+ *
+ * hash = offset_basis
+ * for each octet_of_data to be hashed
+ *         hash = hash xor octet_of_data

lib/igt_fb.c bool igt_fb_supported_format(uint32_t drm_format) · line 3276

Quoted 10 lines
lib/igt_fb.c +10 -0
+ *         hash = hash * FNV_prime
+ * return hash
+ *
+ * 32 bit offset_basis = 2166136261
+ * 32 bit FNV_prime = 224 + 28 + 0x93 = 16777619
+ */
+int igt_fb_get_crc(struct igt_fb *fb, igt_crc_t *crc)
+{
+#define FNV1a_OFFSET_BIAS 2166136261
+#define FNV1a_PRIME 16777619

I'd just use plain uint32_t variables for those, but no big deal.

Quoted 7 lines

lib/igt_fb.c bool igt_fb_supported_format(uint32_t drm_format) · line 3281

Quoted 5 lines
lib/igt_fb.c +5 -0
+	uint32_t hash;
+	void *map;
+	char *ptr, *line = NULL;
+	int x, y, cpp = igt_drm_format_to_bpp(fb->drm_format) / 8;
+	uint32_t stride = calc_plane_stride(fb, 0);

We could return -EINVAL in case fb->num_planes != 1.

Let's not waste cycles. With this ^ fixed, this patch is:

Quoted 7 lines
Quoted 3 lines

Other nits are optional.

I agreed with all your suggestions, and I already applied all of them. Should I wait for the other patches review, or should I resend the new version?

I'm fine with waiting for the full review before a new version of the whole patchset, but you can also send an updated version of a single patch with:

Quoted 5 lines
    git send-email --in-reply-to="<[email protected]>" -1 <commit hash>

where In-Reply-To is the Message-Id of the patch you want to update. I agree it's a little tedious since you need to extract the Message-Id from the message header.

Thanks for the tip with git-send-mail. Since I already applied most of your suggestions, I'll send a full version soon.

Quoted 86 lines
Quoted 1 line

Thanks for all the feedback

:)

Quoted 82 lines

Best Regards

Quoted 80 lines
Quoted 80 lines

lib/igt_fb.c bool igt_fb_supported_format(uint32_t drm_format) · line 3289

Quoted 7 lines
lib/igt_fb.c +7 -0
+	if (fb->is_dumb)
+		map = kmstest_dumb_map_buffer(fb->fd, fb->gem_handle, fb->size,
+					      PROT_READ);
+	else
+		map = gem_mmap__gtt(fb->fd, fb->gem_handle, fb->size,
+				    PROT_READ);
+	ptr = map;

Nit: no need for this, can assign the result of mmap directly to ptr.

Quoted 7 lines
lib/igt_fb.c +7 -0
+
+	/*
+	 * Framebuffers are often uncached, which can make byte-wise accesses
+	 * very slow. We copy each line of the FB into a local buffer to speed
+	 * up the hashing.
+	 */
+	line = malloc(stride);

lib/igt_fb.c bool igt_fb_supported_format(uint32_t drm_format) · line 3306

Quoted 10 lines
lib/igt_fb.c +10 -0
+	if (!line) {
+		munmap(map, fb->size);
+		return -ENOMEM;
+	}
+
+	hash = FNV1a_OFFSET_BIAS;
+
+	for (y = 0; y < fb->height; y++, ptr += stride) {
+
+		igt_memcpy_from_wc(line, ptr, stride);

Nit: no need to copy the whole stride actually, we can just copy fb->width * cpp since we're only going to read that.

Quoted 41 lines
lib/igt_fb.c +18 -0
+
+		for (x = 0; x < fb->width * cpp; x++) {
+			hash ^= line[x];
+			hash *= FNV1a_PRIME;
+		}
+	}
+
+	crc->n_words = 1;
+	crc->crc[0] = hash;
+
+	free(line);
+	munmap(map, fb->size);
+
+	return 0;
+#undef FNV1a_OFFSET_BIAS
+#undef FNV1a_PRIME
+}
+
 /**
  * igt_format_is_yuv:
  * @drm_format: drm fourcc
lib/igt_fb.h +3 -0
index adefebe1..a2741c05 100644
@@ -37,6 +37,7 @@
 #include <i915_drm.h>
 
 #include "igt_color_encoding.h"
+#include "igt_debugfs.h"
 
 /*
  * Internal format to denote a buffer compatible with pixman's
@@ -194,5 +195,7 @@ int igt_format_plane_bpp(uint32_t drm_format, int plane);
 void igt_format_array_fill(uint32_t **formats_array, unsigned int *count,
 			   bool allow_yuv);
 
+int igt_fb_get_crc(struct igt_fb *fb, igt_crc_t *crc);
+
 #endif /* __IGT_FB_H__ */
 
Quoted 4 lines

_______________________________________________
igt-dev mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/igt-dev

_______________________________________________
igt-dev mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/igt-dev

Signature · 2 lines
Rodrigo Siqueira
https://siqueira.tech

_______________________________________________
igt-dev mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/igt-dev

#32 lore raw

Re: [igt-dev] [PATCH V6 i-g-t 5/6] lib/igt_kms: Add igt_output_clone_pipe for cloning

In reply to Ser, Simon
Quoted 2 lines · Simon

So, to test these last two patches we'd need specific hardware right? Because VKMS doesn't support cloning yet (does it?).

hmmm... actually, VKMS successfully pass in this test. However, if you compare "writeback-check-output" and "writeback-check-output-clone", you will notice they are very similar. Maybe, this test does not correctly validating cloning feature?

Quoted 9 lines

What kind of hardware supports cloned writeback outputs? I have a Raspberry Pi which supports writeback via VC4, but I don't think it has writeback cloning. I'm also not willing to install any proprietary driver.

I guess we could land the first part of the series, and wait for VKMS to support cloned outputs to land the last two patches.

Any other ideas?

btw, I'm totally comfortable with the idea of focusing on the first part of this series.

Thanks

Quoted 187 lines · Brian Starkey
Quoted 186 lines

An output can be added as a clone of any other output(s) attached to a pipe using igt_output_clone_pipe()

v5: Drop field out_fence_requested from struct igt_pipe (Brian Starkey)

---
 lib/igt_kms.c | 100 +++++++++++++++++++++++++++++++-------------------
 lib/igt_kms.h |   4 ++
 2 files changed, 66 insertions(+), 38 deletions(-)
lib/igt_kms.c +62 -38
index 140db346..b85a0404 100644
@@ -1765,6 +1765,17 @@ static void igt_display_log_shift(igt_display_t *display, int shift)
 	igt_assert(display->log_shift >= 0);
 }
 
+static int igt_output_idx(igt_output_t *output)
+{
+	int i;
+
+	for (i = 0; i < output->display->n_outputs; i++)
+		if (&output->display->outputs[i] == output)
+			return i;
+
+	return -1;
+}
+
 static void igt_output_refresh(igt_output_t *output)
 {
 	igt_display_t *display = output->display;
@@ -2317,42 +2328,6 @@ void igt_display_fini(igt_display_t *display)
 	display->planes = NULL;
 }
 
-static void igt_display_refresh(igt_display_t *display)
-{
-	igt_output_t *output;
-	int i;
-
-	unsigned long pipes_in_use = 0;
-
-       /* Check that two outputs aren't trying to use the same pipe */
-	for (i = 0; i < display->n_outputs; i++) {
-		output = &display->outputs[i];
-
-		if (output->pending_pipe != PIPE_NONE) {
-			if (pipes_in_use & (1 << output->pending_pipe))
-				goto report_dup;
-
-			pipes_in_use |= 1 << output->pending_pipe;
-		}
-
-		if (output->force_reprobe)
-			igt_output_refresh(output);
-	}
-
-	return;
-
-report_dup:
-	for (; i > 0; i--) {
-		igt_output_t *b = &display->outputs[i - 1];
-
-		igt_assert_f(output->pending_pipe !=
-			     b->pending_pipe,
-			     "%s and %s are both trying to use pipe %s\n",
-			     igt_output_name(output), igt_output_name(b),
-			     kmstest_pipe_name(output->pending_pipe));
-	}
-}
-
 static igt_pipe_t *igt_output_get_driving_pipe(igt_output_t *output)
 {
 	igt_display_t *display = output->display;
@@ -2376,6 +2351,40 @@ static igt_pipe_t *igt_output_get_driving_pipe(igt_output_t *output)
 	return &display->pipes[pipe];
 }
 
+static void igt_display_refresh(igt_display_t *display)
+{
+	igt_output_t *output;
+	igt_pipe_t *pipe;
+	int i;
+
+	unsigned long pipes_in_use = 0;
+	unsigned long pending_crtc_idx_mask;
+
+	/* Check that outputs and pipes agree wrt. cloning */
+	for (i = 0; i < display->n_outputs; i++) {
+		output = &display->outputs[i];
+		pending_crtc_idx_mask = 1 << output->pending_pipe;
+
+		pipe = igt_output_get_driving_pipe(output);
+		if (pipe) {
+			igt_assert_f(pipe->outputs & (1 << igt_output_idx(output)),
+				     "Output %s not expected to be using pipe %s\n",
+				     igt_output_name(output),
+				     kmstest_pipe_name(pipe->pipe));
+
+			if (pipes_in_use & pending_crtc_idx_mask)
+				LOG(display, "Output %s clones pipe %s\n",
+				    igt_output_name(output),
+				    kmstest_pipe_name(pipe->pipe));
+		}
+
+		pipes_in_use |= pending_crtc_idx_mask;
+
+		if (output->force_reprobe)
+			igt_output_refresh(output);
+	}
+}
+
 static igt_plane_t *igt_pipe_get_plane(igt_pipe_t *pipe, int plane_idx)
 {
 	igt_require_f(plane_idx >= 0 && plane_idx < pipe->n_planes,
@@ -3766,6 +3775,7 @@ void igt_output_override_mode(igt_output_t *output, const drmModeModeInfo *mode)
 	output->use_override_mode = !!mode;
 
 	if (pipe) {
+		igt_debug("overriding pipe mode in %s way\n", output->display->is_atomic ? "atomic" : "legacy");
 		if (output->display->is_atomic)
 			igt_pipe_obj_replace_prop_blob(pipe, IGT_CRTC_MODE_ID, igt_output_get_mode(output), sizeof(*mode));
 		else
@@ -3773,6 +3783,16 @@ void igt_output_override_mode(igt_output_t *output, const drmModeModeInfo *mode)
 	}
 }
 
+void igt_output_clone_pipe(igt_output_t *output, enum pipe pipe)
+{
+	igt_display_t *display = output->display;
+	uint32_t current_clones = display->pipes[pipe].outputs;
+
+	igt_output_set_pipe(output, pipe);
+
+	display->pipes[pipe].outputs |= current_clones;
+}
+
 /*
  * igt_output_set_pipe:
  * @output: Target output for which the pipe is being set to
@@ -3789,11 +3809,15 @@ void igt_output_set_pipe(igt_output_t *output, enum pipe pipe)
 
 	igt_assert(output->name);
 
-	if (output->pending_pipe != PIPE_NONE)
+	if (output->pending_pipe != PIPE_NONE) {
 		old_pipe = igt_output_get_driving_pipe(output);
+		old_pipe->outputs &= ~(1 << igt_output_idx(output));
+	}
 
-	if (pipe != PIPE_NONE)
+	if (pipe != PIPE_NONE) {
 		pipe_obj = &display->pipes[pipe];
+		pipe_obj->outputs = (1 << igt_output_idx(output));
+	}
 
 	LOG(display, "%s: set_pipe(%s)\n", igt_output_name(output),
 	    kmstest_pipe_name(pipe));
lib/igt_kms.h +4 -0
index cacc6b90..676839bb 100644
@@ -354,6 +354,8 @@ struct igt_pipe {
 	uint32_t crtc_id;
 
 	int32_t out_fence_fd;
+
+	uint32_t outputs;
 };
 
 typedef struct {
@@ -411,6 +413,8 @@ const char *igt_output_name(igt_output_t *output);
 drmModeModeInfo *igt_output_get_mode(igt_output_t *output);
 void igt_output_override_mode(igt_output_t *output, const drmModeModeInfo *mode);
 void igt_output_set_pipe(igt_output_t *output, enum pipe pipe);
+void igt_output_clone_pipe(igt_output_t *output, enum pipe pipe);
+
 igt_plane_t *igt_output_get_plane(igt_output_t *output, int plane_idx);
 igt_plane_t *igt_output_get_plane_type(igt_output_t *output, int plane_type);
 int igt_output_count_plane_type(igt_output_t *output, int plane_type);

_______________________________________________
igt-dev mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/igt-dev

Signature · 2 lines
Rodrigo Siqueira
https://siqueira.tech

_______________________________________________
igt-dev mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/igt-dev

#16 lore raw

Re: [Intel-gfx] [igt-dev] [PATCH V6 i-g-t 2/6] kms_writeback: Add initial writeback tests

In reply to [email protected]

Thanks for the clarification!

Quoted 26 lines · On Tue, 2019-07-16 at 16:22 +0100, [email protected] wrote:
Quoted 17 lines
Quoted 17 lines
Quoted 17 lines
tests/kms_writeback.c new file +17 -0
+static void invalid_out_fence(igt_output_t *output, igt_fb_t *valid_fb, igt_fb_t *invalid_fb)
+{
+	int i, ret;
+	int32_t out_fence;
+	struct {
+		uint32_t fb_id;
+		bool ptr_valid;
+		int32_t *out_fence_ptr;
+	} invalid_tests[] = {
+		{
+			/* No output buffer, but the WRITEBACK_OUT_FENCE_PTR set. */
+			.fb_id = 0,
+			.ptr_valid = true,
+			.out_fence_ptr = &out_fence,
+		},
+		{
+			/* Invalid output buffer. */
Quoted 8 lines
Quoted 8 lines

tests/kms_writeback.c line 185

Quoted 4 lines
tests/kms_writeback.c new file +4 -0
+			.fb_id = invalid_fb->fb_id,
+			.ptr_valid = true,
+			.out_fence_ptr = &out_fence,
+		},

This doesn't belong in this function (invalid_out_fence), since this checks an invalid framebuffer, not an invalid fence. We should probably move it to writeback_fb_id (and rename that function to test_fb?).

Quoted 4 lines

It tries to test that you can't trick the driver to do any work on a fence if the framebuffer is invalid, so the set of tests tries: no fb with valid fence, invalid fb with valid fence, valid fb but invalid fence and assumes that no fb with invalid fence is a NOP anyway.

Yeah, that makes sense, it's just confusing to put it in a function named invalid_out_fence. Here the out fence is valid, but the output buffer isn't, so it should probably be moved away (or this function should be renamed).

Quoted 123 lines
Quoted 80 lines
Quoted 80 lines
Quoted 4 lines
tests/kms_writeback.c new file +4 -0
+		{
+			/* Invalid WRITEBACK_OUT_FENCE_PTR. */
+			.fb_id = valid_fb->fb_id,
+			.ptr_valid = false,

tests/kms_writeback.c line 199

Quoted 10 lines
tests/kms_writeback.c new file +10 -0
+			.out_fence_ptr = (int32_t *)0x8,
+		},
+	};
+
+	for (i = 0; i < ARRAY_SIZE(invalid_tests); i++) {
+		ret = do_writeback_test(output, DRM_MODE_ATOMIC_ALLOW_MODESET,
+					invalid_tests[i].fb_id,
+					invalid_tests[i].out_fence_ptr,
+					invalid_tests[i].ptr_valid);
+		igt_assert(ret != 0);

Maybe we can check for -ret == EINVAL?

tests/kms_writeback.c line 203

Quoted 4 lines
tests/kms_writeback.c new file +4 -0
+	}
+}
+
+static void writeback_fb_id(igt_output_t *output, igt_fb_t *valid_fb, igt_fb_t *invalid_fb)

invalid_fb doesn't seem to be used here. valid_fb seems to be set to the input framebuffer. It's probably not a good idea to use the same FB for input and writeback output.

Quoted 41 lines
tests/kms_writeback.c new file +41 -0
+{
+
+	int ret;
+
+	/* Valid output buffer */
+	ret = do_writeback_test(output, DRM_MODE_ATOMIC_ALLOW_MODESET,
+				valid_fb->fb_id, NULL, false);
+	igt_assert(ret == 0);
+
+	/* Invalid object for WRITEBACK_FB_ID */
+	ret = do_writeback_test(output, DRM_MODE_ATOMIC_ALLOW_MODESET,
+				output->id, NULL, false);
+	igt_assert(ret == -EINVAL);
+
+	/* Zero WRITEBACK_FB_ID */
+	ret = do_writeback_test(output, DRM_MODE_ATOMIC_ALLOW_MODESET,
+				0, NULL, false);
+	igt_assert(ret == 0);
+}
+
+igt_main
+{
+	igt_display_t display;
+	igt_output_t *output;
+	igt_plane_t *plane;
+	igt_fb_t input_fb;
+	drmModeModeInfo mode;
+	int ret;
+
+	memset(&display, 0, sizeof(display));
+
+	igt_fixture {
+		display.drm_fd = drm_open_driver_master(DRIVER_ANY);
+		igt_display_require(&display, display.drm_fd);
+
+		kmstest_set_vt_graphics_mode();
+
+		igt_display_require(&display, display.drm_fd);
+
+		igt_require(display.is_atomic);
+

tests/kms_writeback.c line 254

Quoted 10 lines
tests/kms_writeback.c new file +10 -0
+		output = kms_writeback_get_output(&display);
+		igt_require(output);
+
+		if (output->use_override_mode)
+			memcpy(&mode, &output->override_mode, sizeof(mode));
+		else
+			memcpy(&mode, &output->config.default_mode, sizeof(mode));
+
+		plane = igt_output_get_plane_type(output, DRM_PLANE_TYPE_PRIMARY);
+		igt_require(plane);

Maybe we can assert on this?

Quoted 7 lines
Quoted 7 lines

tests/kms_writeback.c line 259

Quoted 4 lines
tests/kms_writeback.c new file +4 -0
+		ret = igt_create_fb(display.drm_fd, mode.hdisplay,
+				    mode.vdisplay,
+				    DRM_FORMAT_XRGB8888,
+				    igt_fb_mod_to_tiling(0),

This is supposed to take a modifier. DRM_FORMAT_MOD_LINEAR would be better to make this clear.

Agree. The patchset pre-dates the modifiers support (or has the same age, I forgot)

Quoted 23 lines
Quoted 23 lines

(Applies to other lines of this patch)

tests/kms_writeback.c line 266

Quoted 7 lines
tests/kms_writeback.c new file +7 -0
+				    &input_fb);
+		igt_assert(ret >= 0);
+		igt_plane_set_fb(plane, &input_fb);
+	}
+
+	igt_subtest("writeback-pixel-formats") {
+		drmModePropertyBlobRes *formats_blob = get_writeback_formats_blob(output);

Need to drmModeFreePropertyBlob this.

Quoted 11 lines
tests/kms_writeback.c new file +11 -0
+		const char *valid_chars = "0123456 ABCGNRUVXY";
+		unsigned int i;
+		char *c;
+
+		/*
+		 * We don't have a comprehensive list of formats, so just check
+		 * that the blob length is sensible and that it doesn't contain
+		 * any outlandish characters
+		 */
+		igt_assert(!(formats_blob->length % 4));
+		c = formats_blob->data;
Quoted 8 lines
Quoted 8 lines

tests/kms_writeback.c line 280

Quoted 3 lines
tests/kms_writeback.c new file +3 -0
+		for (i = 0; i < formats_blob->length; i++)
+			igt_assert_f(strchr(valid_chars, c[i]),
+				     "Unexpected character %c\n", c[i]);

Honestly, I'm not a fan of this check. There's no guarantee that fourcc codes will be made from ASCII characters, in fact some formats already have non-printable chars in them. I don't want to have to update this test when a new fourcc format is added.

Quoted 3 lines

Like the comments says, we don't have a full list of formats to check against. Suggestions on how to improve are welcome, but I don't think we should delay (any longer) the patchset due to this.

Agreed.

Quoted 2 lines

Best regards,
Liviu

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

#14 lore raw

Re: [igt-dev] [PATCH V6 i-g-t 2/6] kms_writeback: Add initial writeback tests

In reply to Rodrigo Siqueira
Quoted 276 lines · Rodrigo Siqueira
Quoted 195 lines · Simon
Quoted 143 lines · Rodrigo Siqueira
Quoted 6 lines · Simon

Hi,

Thanks for the patch! Here are a few comments.

For bonus points, it would be nice to add igt_describe descriptions of each sub-test.

Hi Simon,

First of all, thanks for your feedback; I already applied most of your suggestions. I just have some inline comments/questions.

Quoted 121 lines · Brian Starkey
Quoted 120 lines

Add tests for the WRITEBACK_PIXEL_FORMATS, WRITEBACK_OUT_FENCE_PTR and WRITEBACK_FB_ID properties on writeback connectors, ensuring their behaviour is correct.

[rebased and updated do_writeback_test() function to address feedback]
---
 tests/Makefile.sources |   1 +
 tests/kms_writeback.c  | 314 +++++++++++++++++++++++++++++++++++++++++
 tests/meson.build      |   1 +
 3 files changed, 316 insertions(+)
 create mode 100644 tests/kms_writeback.c
tests/Makefile.sources +1 -0
index 027ed82f..03cc8efa 100644
@@ -77,6 +77,7 @@ TESTS_progs = \
 	kms_universal_plane \
 	kms_vblank \
 	kms_vrr \
+	kms_writeback \
 	meta_test \
 	perf \
 	perf_pmu \
tests/kms_writeback.c new file +88 -0
new file mode 100644
index 00000000..66ef48a6
@@ -0,0 +1,314 @@
+/*
+ * (C) COPYRIGHT 2017 ARM Limited. All rights reserved.
+ *
+ * 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.
+ *
+ */
+
+#include <errno.h>
+#include <stdbool.h>
+#include <stdio.h>
+#include <string.h>
+
+#include "igt.h"
+#include "igt_core.h"
+#include "igt_fb.h"
+
+static drmModePropertyBlobRes *get_writeback_formats_blob(igt_output_t *output)
+{
+	drmModePropertyBlobRes *blob = NULL;
+	uint64_t blob_id;
+	int ret;
+
+	ret = kmstest_get_property(output->display->drm_fd,
+				   output->config.connector->connector_id,
+				   DRM_MODE_OBJECT_CONNECTOR,
+				   igt_connector_prop_names[IGT_CONNECTOR_WRITEBACK_PIXEL_FORMATS],
+				   NULL, &blob_id, NULL);
+	if (ret)
+		blob = drmModeGetPropertyBlob(output->display->drm_fd, blob_id);
+
+	igt_assert(blob);
+
+	return blob;
+}
+
+static bool check_writeback_config(igt_display_t *display, igt_output_t *output)
+{
+	igt_fb_t input_fb, output_fb;
+	igt_plane_t *plane;
+	uint32_t writeback_format = DRM_FORMAT_XRGB8888;
+	uint64_t tiling = igt_fb_mod_to_tiling(0);
+	int width, height, ret;
+	drmModeModeInfo override_mode = {
+		.clock = 25175,
+		.hdisplay = 640,
+		.hsync_start = 656,
+		.hsync_end = 752,
+		.htotal = 800,
+		.hskew = 0,
+		.vdisplay = 480,
+		.vsync_start = 490,
+		.vsync_end = 492,
+		.vtotal = 525,
+		.vscan = 0,
+		.vrefresh = 60,
+		.flags = DRM_MODE_FLAG_NHSYNC | DRM_MODE_FLAG_NVSYNC,
+		.name = {"640x480-60"},
+	};
+	igt_output_override_mode(output, &override_mode);
+
+	width = override_mode.hdisplay;
+	height = override_mode.vdisplay;
+
+	ret = igt_create_fb(display->drm_fd, width, height, DRM_FORMAT_XRGB8888, tiling, &input_fb);
+	igt_assert(ret >= 0);
+
+	ret = igt_create_fb(display->drm_fd, width, height, writeback_format, tiling, &output_fb);
+	igt_assert(ret >= 0);
+
+	plane = igt_output_get_plane_type(output, DRM_PLANE_TYPE_PRIMARY);
+	igt_plane_set_fb(plane, &input_fb);
Quoted 8 lines

tests/kms_writeback.c line 92

Quoted 4 lines
tests/kms_writeback.c new file +4 -0
+	igt_output_set_writeback_fb(output, &output_fb);
+
+	ret = igt_display_try_commit_atomic(display, DRM_MODE_ATOMIC_TEST_ONLY |
+					    DRM_MODE_ATOMIC_ALLOW_MODESET, NULL);

Okay, we're using atomic test-only mode to know if we can perform tests with the writeback output. It's probably fine, but we don't use WRITEBACK_PIXEL_FORMATS, which makes me a little bit sad.

Quoted 5 lines

Sorry, I did not fully understand part. Did you expect to see something like the below code before igt_display_try_commit_atomic()?

igt_output_set_prop_value(output, WRITEBACK_PIXEL_FORMATS,

                           DRM_FORMAT_XRGB8888);

Hmm, no, we cannot change the list of writeback formats (it's immutable).

This comment doesn't require any action, it's just me thinking aloud :P

I'm just thinking that it would be nice to choose our format depending on the WRITEBACK_PIXEL_FORMATS property. And probably run tests with each supported format (or, alternatively, choose a random one). That way we can make sure WRITEBACK_PIXEL_FORMATS isn't broken.

However, this can be left for a later patch.

Quoted 23 lines
Quoted 14 lines
Quoted 14 lines
tests/kms_writeback.c new file +14 -0
+	igt_plane_set_fb(plane, NULL);
+	igt_remove_fb(display->drm_fd, &input_fb);
+	igt_remove_fb(display->drm_fd, &output_fb);
+
+	return !ret;
+}
+
+static igt_output_t *kms_writeback_get_output(igt_display_t *display)
+{
+	int i;
+
+	for (i = 0; i < display->n_outputs; i++) {
+		igt_output_t *output = &display->outputs[i];
+		int j;
Quoted 8 lines

tests/kms_writeback.c line 111

Quoted 5 lines
tests/kms_writeback.c new file +5 -0
+
+		if (output->config.connector->connector_type != DRM_MODE_CONNECTOR_WRITEBACK)
+			continue;
+
+		kmstest_force_connector(display->drm_fd, output->config.connector, FORCE_CONNECTOR_ON);

Hmm. Is this really necessary? "Real" userspace won't do this, so I don't think we need to either.

Quoted 4 lines

Hmmm, I have a little experience with userspace; however, I tested kms_writeback on top of vkms without this line, and everything worked well. If I remove this line, should I also drop the line that force connector to FORCE_CONNECTOR_UNSPECIFIED?

I believe so.

Quoted 2 lines

Another question, if FORCE_CONNECTOR_ON is something that userspace won't want to do, why do we have it?

Quoted 8 lines

We use kmstest_force_connector in injection tests: those pick a disconnected HDMI connector and trick the kernel into thinking it's connected. We generally also force an EDID and this is used to emulate a screen (e.g. a screen that supports audio, 4K or 3D).

This is only meant to be used for testing though, hence "real userspace shouldn't use it".

Ahhh, I see. Thanks for the explanation

Quoted 69 lines
Quoted 57 lines
Quoted 48 lines

tests/kms_writeback.c line 121

Quoted 9 lines
tests/kms_writeback.c new file +9 -0
+		for (j = 0; j < igt_display_get_n_pipes(display); j++) {
+			igt_output_set_pipe(output, j);
+
+			if (check_writeback_config(display, output)) {
+				igt_debug("Using connector %u:%s on pipe %d\n",
+					  output->config.connector->connector_id,
+					  output->name, j);
+				return output;
+			}

We could probably add an igt_debug statement in case we don't use this writeback output.

Quoted 10 lines
tests/kms_writeback.c new file +10 -0
+		}
+
+		/* Restore any connectors we don't use, so we don't trip on them later */
+		kmstest_force_connector(display->drm_fd, output->config.connector, FORCE_CONNECTOR_UNSPECIFIED);
+	}
+
+	return NULL;
+}
+
+static void check_writeback_fb_id(igt_output_t *output)

tests/kms_writeback.c line 141

Quoted 10 lines
tests/kms_writeback.c new file +10 -0
+{
+	uint64_t check_fb_id;
+
+	check_fb_id = igt_output_get_prop(output, IGT_CONNECTOR_WRITEBACK_FB_ID);
+	igt_assert(check_fb_id == 0);
+}
+
+static int do_writeback_test(igt_output_t *output, uint32_t flags,
+			      uint32_t fb_id, int32_t *out_fence_ptr,
+			      bool ptr_valid)

flags seems to always be set to DRM_MODE_ATOMIC_ALLOW_MODESET. We can probably remove the parameter from this function.

Quoted 11 lines
tests/kms_writeback.c new file +11 -0
+{
+	int ret;
+	igt_display_t *display = output->display;
+	struct kmstest_connector_config *config = &output->config;
+
+	igt_output_set_prop_value(output, IGT_CONNECTOR_CRTC_ID, config->crtc->crtc_id);
+	igt_output_set_prop_value(output, IGT_CONNECTOR_WRITEBACK_FB_ID, fb_id);
+	igt_output_set_prop_value(output, IGT_CONNECTOR_WRITEBACK_OUT_FENCE_PTR, (uint64_t)out_fence_ptr);
+
+	if (ptr_valid)
+		*out_fence_ptr = 0;
Quoted 8 lines

tests/kms_writeback.c line 157

Quoted 5 lines
tests/kms_writeback.c new file +5 -0
+
+	ret = igt_display_try_commit_atomic(display, flags, NULL);
+
+	if (ptr_valid)
+		igt_assert(*out_fence_ptr == -1);

I'm confused. Why should this be -1 even if we
igt_display_try_commit_atomic succeeds?

Quoted 4 lines

I inspected the drm_mode_atomic_ioctl() function and I noticed that It calls complete_signaling() in its turn this function assign -1 to out_fence_ptr. Since all of this happens at the end of atomic_commit, I believe that we don’t need it. I already removed it.

I think I'm still confused :)

I'm probably missing something. As far as I understand, we are doing an atomic commit, sometimes with a valid WRITEBACK_FB_ID and WRITEBACK_OUT_FENCE_PTR. In this case it should start the writeback process in the kernel and we should get a valid out_fence_ptr?

Quoted 6 lines
Quoted 1 line

Why do we always get -1?

FWIU userspace uses the WRITEBACK_OUT_FENCE_PTR to provide a pointer for the kernel to fill with a sync_file file descriptor, which will signal once the writeback is finished. If the signal was correctly handled, the out_fence_ptr would be set to -1, because of this we always get -1.

But in case a writeback operation has been successfully started, it won't be -1 (since it will be a valid FD). As I've understood it, this function will always be called without triggering a writeback operation.

Maybe we should rename do_writeback_test to do_noop_writeback (or a better name) and add a comment explaining that *out_fence_ptr will be -1 because:

- Either ret != 0 and we submitted a bad commit
- Either ret == 0 (valid commit) but out_fence_ptr == NULL (so no writeback operation started)

Maybe we could igt_assert(ret != 0 || out_fence_ptr == NULL) to make sure we don't call this function and successfully trigger a writeback operation.

What do you think?

Quoted 221 lines

Thanks for your feedback, I'll prepare a new version soon.

Quoted 219 lines
Quoted 152 lines
Quoted 152 lines
Quoted 17 lines
tests/kms_writeback.c new file +17 -0
+	/* WRITEBACK_FB_ID must always read as zero */
+	check_writeback_fb_id(output);
+
+	return ret;
+}
+
+static void invalid_out_fence(igt_output_t *output, igt_fb_t *valid_fb, igt_fb_t *invalid_fb)
+{
+	int i, ret;
+	int32_t out_fence;
+	struct {
+		uint32_t fb_id;
+		bool ptr_valid;
+		int32_t *out_fence_ptr;
+	} invalid_tests[] = {
+		{
+			/* No output buffer, but the WRITEBACK_OUT_FENCE_PTR set. */

tests/kms_writeback.c line 185

Quoted 10 lines
tests/kms_writeback.c new file +10 -0
+			.fb_id = 0,
+			.ptr_valid = true,
+			.out_fence_ptr = &out_fence,
+		},
+		{
+			/* Invalid output buffer. */
+			.fb_id = invalid_fb->fb_id,
+			.ptr_valid = true,
+			.out_fence_ptr = &out_fence,
+		},

This doesn't belong in this function (invalid_out_fence), since this checks an invalid framebuffer, not an invalid fence. We should probably move it to writeback_fb_id (and rename that function to test_fb?).

Quoted 4 lines
tests/kms_writeback.c new file +4 -0
+		{
+			/* Invalid WRITEBACK_OUT_FENCE_PTR. */
+			.fb_id = valid_fb->fb_id,
+			.ptr_valid = false,

tests/kms_writeback.c line 199

Quoted 10 lines
tests/kms_writeback.c new file +10 -0
+			.out_fence_ptr = (int32_t *)0x8,
+		},
+	};
+
+	for (i = 0; i < ARRAY_SIZE(invalid_tests); i++) {
+		ret = do_writeback_test(output, DRM_MODE_ATOMIC_ALLOW_MODESET,
+					invalid_tests[i].fb_id,
+					invalid_tests[i].out_fence_ptr,
+					invalid_tests[i].ptr_valid);
+		igt_assert(ret != 0);

Maybe we can check for -ret == EINVAL?

tests/kms_writeback.c line 203

Quoted 4 lines
tests/kms_writeback.c new file +4 -0
+	}
+}
+
+static void writeback_fb_id(igt_output_t *output, igt_fb_t *valid_fb, igt_fb_t *invalid_fb)

invalid_fb doesn't seem to be used here. valid_fb seems to be set to the input framebuffer. It's probably not a good idea to use the same FB for input and writeback output.

Quoted 41 lines
tests/kms_writeback.c new file +41 -0
+{
+
+	int ret;
+
+	/* Valid output buffer */
+	ret = do_writeback_test(output, DRM_MODE_ATOMIC_ALLOW_MODESET,
+				valid_fb->fb_id, NULL, false);
+	igt_assert(ret == 0);
+
+	/* Invalid object for WRITEBACK_FB_ID */
+	ret = do_writeback_test(output, DRM_MODE_ATOMIC_ALLOW_MODESET,
+				output->id, NULL, false);
+	igt_assert(ret == -EINVAL);
+
+	/* Zero WRITEBACK_FB_ID */
+	ret = do_writeback_test(output, DRM_MODE_ATOMIC_ALLOW_MODESET,
+				0, NULL, false);
+	igt_assert(ret == 0);
+}
+
+igt_main
+{
+	igt_display_t display;
+	igt_output_t *output;
+	igt_plane_t *plane;
+	igt_fb_t input_fb;
+	drmModeModeInfo mode;
+	int ret;
+
+	memset(&display, 0, sizeof(display));
+
+	igt_fixture {
+		display.drm_fd = drm_open_driver_master(DRIVER_ANY);
+		igt_display_require(&display, display.drm_fd);
+
+		kmstest_set_vt_graphics_mode();
+
+		igt_display_require(&display, display.drm_fd);
+
+		igt_require(display.is_atomic);
+

tests/kms_writeback.c line 254

Quoted 10 lines
tests/kms_writeback.c new file +10 -0
+		output = kms_writeback_get_output(&display);
+		igt_require(output);
+
+		if (output->use_override_mode)
+			memcpy(&mode, &output->override_mode, sizeof(mode));
+		else
+			memcpy(&mode, &output->config.default_mode, sizeof(mode));
+
+		plane = igt_output_get_plane_type(output, DRM_PLANE_TYPE_PRIMARY);
+		igt_require(plane);

Maybe we can assert on this?

tests/kms_writeback.c line 259

Quoted 4 lines
tests/kms_writeback.c new file +4 -0
+		ret = igt_create_fb(display.drm_fd, mode.hdisplay,
+				    mode.vdisplay,
+				    DRM_FORMAT_XRGB8888,
+				    igt_fb_mod_to_tiling(0),

This is supposed to take a modifier. DRM_FORMAT_MOD_LINEAR would be better to make this clear.

(Applies to other lines of this patch)

tests/kms_writeback.c line 266

Quoted 7 lines
tests/kms_writeback.c new file +7 -0
+				    &input_fb);
+		igt_assert(ret >= 0);
+		igt_plane_set_fb(plane, &input_fb);
+	}
+
+	igt_subtest("writeback-pixel-formats") {
+		drmModePropertyBlobRes *formats_blob = get_writeback_formats_blob(output);

Need to drmModeFreePropertyBlob this.

Quoted 4 lines
tests/kms_writeback.c new file +4 -0
+		const char *valid_chars = "0123456 ABCGNRUVXY";
+		unsigned int i;
+		char *c;
+

tests/kms_writeback.c line 280

Quoted 10 lines
tests/kms_writeback.c new file +10 -0
+		/*
+		 * We don't have a comprehensive list of formats, so just check
+		 * that the blob length is sensible and that it doesn't contain
+		 * any outlandish characters
+		 */
+		igt_assert(!(formats_blob->length % 4));
+		c = formats_blob->data;
+		for (i = 0; i < formats_blob->length; i++)
+			igt_assert_f(strchr(valid_chars, c[i]),
+				     "Unexpected character %c\n", c[i]);

Honestly, I'm not a fan of this check. There's no guarantee that fourcc codes will be made from ASCII characters, in fact some formats already have non-printable chars in them. I don't want to have to update this test when a new fourcc format is added.

Quoted 5 lines
Quoted 3 lines

We currently don't have a test for IN_FORMATS. If we really want to check formats, we could have a big array of known formats and add a bool is_valid_format(uint32_t fmt) function.

Agree with you. How about remove this test?

I guess we could always keep the length % 4 test, even if it's just a sanity test. At least this one won't ever need to be changed.

I don't feel strongly about it.

Quoted 56 lines

Thanks
Best Regards

Quoted 53 lines

tests/kms_writeback.c line 284

Quoted 4 lines
tests/kms_writeback.c new file +4 -0
+	}
+
+	igt_subtest("writeback-invalid-out-fence") {
+		igt_fb_t invalid_fb;

invalid_output_fb would be a better name IMHO.

tests/kms_writeback.c line 290

Quoted 6 lines
tests/kms_writeback.c new file +6 -0
+		ret = igt_create_fb(display.drm_fd, mode.hdisplay / 2,
+				    mode.vdisplay / 2,
+				    DRM_FORMAT_XRGB8888,
+				    igt_fb_mod_to_tiling(0),
+				    &invalid_fb);
+		igt_require(ret > 0);

We should probably use a different variable: ret is signed, igt_create_fb isn't. Also ret doesn't make it clear that the return value is the KMS framebuffer ID.

(Applies to other subtests)

tests/kms_writeback.c line 292

Quoted 1 line
tests/kms_writeback.c new file +1 -0
+		invalid_out_fence(output, &input_fb, &invalid_fb);

(Still not sure why we provide the input FB to this function.)

Quoted 29 lines
tests/kms_writeback.c new file +21 -0
+		igt_remove_fb(display.drm_fd, &invalid_fb);
+	}
+
+	igt_subtest("writeback-fb-id") {
+		igt_fb_t output_fb;
+		ret = igt_create_fb(display.drm_fd, mode.hdisplay, mode.vdisplay,
+				    DRM_FORMAT_XRGB8888,
+				    igt_fb_mod_to_tiling(0),
+				    &output_fb);
+		igt_require(ret > 0);
+
+		writeback_fb_id(output, &input_fb, &output_fb);
+
+		igt_remove_fb(display.drm_fd, &output_fb);
+	}
+
+	igt_fixture {
+		igt_remove_fb(display.drm_fd, &input_fb);
+		igt_display_fini(&display);
+	}
+}
tests/meson.build +0 -0
index f168fbba..9c77cfcd 100644
@@ -63,6 +63,7 @@ test_progs = [
 	'kms_universal_plane',
 	'kms_vblank',
 	'kms_vrr',
Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
tests/meson.build +1 -0
+	'kms_writeback',
 	'meta_test',
 	'panfrost_get_param',
 	'panfrost_gem_new',

_______________________________________________
igt-dev mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/igt-dev

_______________________________________________
igt-dev mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/igt-dev

#33 lore raw

Re: [igt-dev] [PATCH V6 i-g-t 5/6] lib/igt_kms: Add igt_output_clone_pipe for cloning

In reply to Rodrigo Siqueira
Quoted 8 lines · Rodrigo Siqueira
Quoted 2 lines · Simon

So, to test these last two patches we'd need specific hardware right? Because VKMS doesn't support cloning yet (does it?).

hmmm... actually, VKMS successfully pass in this test. However, if you compare "writeback-check-output" and "writeback-check-output-clone", you will notice they are very similar. Maybe, this test does not correctly validating cloning feature?

Hmm. Are you sure it doesn't skip at igt_require(clone)? It should, because VKMS only exposes one connector… Might be a bug in the test.

Quoted 5 lines
Quoted 5 lines

What kind of hardware supports cloned writeback outputs? I have a Raspberry Pi which supports writeback via VC4, but I don't think it has writeback cloning. I'm also not willing to install any proprietary driver.

Quoted 7 lines
Quoted 4 lines

I guess we could land the first part of the series, and wait for VKMS to support cloned outputs to land the last two patches.

Any other ideas?

btw, I'm totally comfortable with the idea of focusing on the first part of this series.

Yeah, let's land the first part and keep cloning for later :)

Quoted 175 lines

Thanks

Quoted 173 lines · Brian Starkey
Quoted 172 lines

An output can be added as a clone of any other output(s) attached to a pipe using igt_output_clone_pipe()

v5: Drop field out_fence_requested from struct igt_pipe (Brian Starkey)

---
 lib/igt_kms.c | 100 +++++++++++++++++++++++++++++++-------------------
 lib/igt_kms.h |   4 ++
 2 files changed, 66 insertions(+), 38 deletions(-)
lib/igt_kms.c +62 -38
index 140db346..b85a0404 100644
@@ -1765,6 +1765,17 @@ static void igt_display_log_shift(igt_display_t *display, int shift)
 	igt_assert(display->log_shift >= 0);
 }
 
+static int igt_output_idx(igt_output_t *output)
+{
+	int i;
+
+	for (i = 0; i < output->display->n_outputs; i++)
+		if (&output->display->outputs[i] == output)
+			return i;
+
+	return -1;
+}
+
 static void igt_output_refresh(igt_output_t *output)
 {
 	igt_display_t *display = output->display;
@@ -2317,42 +2328,6 @@ void igt_display_fini(igt_display_t *display)
 	display->planes = NULL;
 }
 
-static void igt_display_refresh(igt_display_t *display)
-{
-	igt_output_t *output;
-	int i;
-
-	unsigned long pipes_in_use = 0;
-
-       /* Check that two outputs aren't trying to use the same pipe */
-	for (i = 0; i < display->n_outputs; i++) {
-		output = &display->outputs[i];
-
-		if (output->pending_pipe != PIPE_NONE) {
-			if (pipes_in_use & (1 << output->pending_pipe))
-				goto report_dup;
-
-			pipes_in_use |= 1 << output->pending_pipe;
-		}
-
-		if (output->force_reprobe)
-			igt_output_refresh(output);
-	}
-
-	return;
-
-report_dup:
-	for (; i > 0; i--) {
-		igt_output_t *b = &display->outputs[i - 1];
-
-		igt_assert_f(output->pending_pipe !=
-			     b->pending_pipe,
-			     "%s and %s are both trying to use pipe %s\n",
-			     igt_output_name(output), igt_output_name(b),
-			     kmstest_pipe_name(output->pending_pipe));
-	}
-}
-
 static igt_pipe_t *igt_output_get_driving_pipe(igt_output_t *output)
 {
 	igt_display_t *display = output->display;
@@ -2376,6 +2351,40 @@ static igt_pipe_t *igt_output_get_driving_pipe(igt_output_t *output)
 	return &display->pipes[pipe];
 }
 
+static void igt_display_refresh(igt_display_t *display)
+{
+	igt_output_t *output;
+	igt_pipe_t *pipe;
+	int i;
+
+	unsigned long pipes_in_use = 0;
+	unsigned long pending_crtc_idx_mask;
+
+	/* Check that outputs and pipes agree wrt. cloning */
+	for (i = 0; i < display->n_outputs; i++) {
+		output = &display->outputs[i];
+		pending_crtc_idx_mask = 1 << output->pending_pipe;
+
+		pipe = igt_output_get_driving_pipe(output);
+		if (pipe) {
+			igt_assert_f(pipe->outputs & (1 << igt_output_idx(output)),
+				     "Output %s not expected to be using pipe %s\n",
+				     igt_output_name(output),
+				     kmstest_pipe_name(pipe->pipe));
+
+			if (pipes_in_use & pending_crtc_idx_mask)
+				LOG(display, "Output %s clones pipe %s\n",
+				    igt_output_name(output),
+				    kmstest_pipe_name(pipe->pipe));
+		}
+
+		pipes_in_use |= pending_crtc_idx_mask;
+
+		if (output->force_reprobe)
+			igt_output_refresh(output);
+	}
+}
+
 static igt_plane_t *igt_pipe_get_plane(igt_pipe_t *pipe, int plane_idx)
 {
 	igt_require_f(plane_idx >= 0 && plane_idx < pipe->n_planes,
@@ -3766,6 +3775,7 @@ void igt_output_override_mode(igt_output_t *output, const drmModeModeInfo *mode)
 	output->use_override_mode = !!mode;
 
 	if (pipe) {
+		igt_debug("overriding pipe mode in %s way\n", output->display->is_atomic ? "atomic" : "legacy");
 		if (output->display->is_atomic)
 			igt_pipe_obj_replace_prop_blob(pipe, IGT_CRTC_MODE_ID, igt_output_get_mode(output), sizeof(*mode));
 		else
@@ -3773,6 +3783,16 @@ void igt_output_override_mode(igt_output_t *output, const drmModeModeInfo *mode)
 	}
 }
 
+void igt_output_clone_pipe(igt_output_t *output, enum pipe pipe)
+{
+	igt_display_t *display = output->display;
+	uint32_t current_clones = display->pipes[pipe].outputs;
+
+	igt_output_set_pipe(output, pipe);
+
+	display->pipes[pipe].outputs |= current_clones;
+}
+
 /*
  * igt_output_set_pipe:
  * @output: Target output for which the pipe is being set to
@@ -3789,11 +3809,15 @@ void igt_output_set_pipe(igt_output_t *output, enum pipe pipe)
 
 	igt_assert(output->name);
 
-	if (output->pending_pipe != PIPE_NONE)
+	if (output->pending_pipe != PIPE_NONE) {
 		old_pipe = igt_output_get_driving_pipe(output);
+		old_pipe->outputs &= ~(1 << igt_output_idx(output));
+	}
 
-	if (pipe != PIPE_NONE)
+	if (pipe != PIPE_NONE) {
 		pipe_obj = &display->pipes[pipe];
+		pipe_obj->outputs = (1 << igt_output_idx(output));
+	}
 
 	LOG(display, "%s: set_pipe(%s)\n", igt_output_name(output),
 	    kmstest_pipe_name(pipe));
lib/igt_kms.h +2 -0
index cacc6b90..676839bb 100644
@@ -354,6 +354,8 @@ struct igt_pipe {
 	uint32_t crtc_id;
 
 	int32_t out_fence_fd;
+
+	uint32_t outputs;
 };
 
Quoted 14 lines
Quoted 6 lines
Quoted 6 lines
lib/igt_kms.h +1 -0
 typedef struct {
@@ -411,6 +413,8 @@ const char *igt_output_name(igt_output_t *output);
 drmModeModeInfo *igt_output_get_mode(igt_output_t *output);
 void igt_output_override_mode(igt_output_t *output, const drmModeModeInfo *mode);
 void igt_output_set_pipe(igt_output_t *output, enum pipe pipe);
+void igt_output_clone_pipe(igt_output_t *output, enum pipe pipe);
Quoted 8 lines
Quoted 8 lines
lib/igt_kms.h +1 -0
+
 igt_plane_t *igt_output_get_plane(igt_output_t *output, int plane_idx);
 igt_plane_t *igt_output_get_plane_type(igt_output_t *output, int plane_type);
 int igt_output_count_plane_type(igt_output_t *output, int plane_type);

_______________________________________________
igt-dev mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/igt-dev

_______________________________________________
igt-dev mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/igt-dev

#17 lore raw

Re: [igt-dev] [PATCH V6 i-g-t 2/6] kms_writeback: Add initial writeback tests

In reply to Ser, Simon
Quoted 34 lines · Simon

Thanks for the clarification!

Quoted 26 lines · On Tue, 2019-07-16 at 16:22 +0100, [email protected] wrote:
Quoted 17 lines
Quoted 17 lines
Quoted 17 lines
tests/kms_writeback.c new file +17 -0
+static void invalid_out_fence(igt_output_t *output, igt_fb_t *valid_fb, igt_fb_t *invalid_fb)
+{
+	int i, ret;
+	int32_t out_fence;
+	struct {
+		uint32_t fb_id;
+		bool ptr_valid;
+		int32_t *out_fence_ptr;
+	} invalid_tests[] = {
+		{
+			/* No output buffer, but the WRITEBACK_OUT_FENCE_PTR set. */
+			.fb_id = 0,
+			.ptr_valid = true,
+			.out_fence_ptr = &out_fence,
+		},
+		{
+			/* Invalid output buffer. */
Quoted 8 lines
Quoted 8 lines

tests/kms_writeback.c line 185

Quoted 4 lines
tests/kms_writeback.c new file +4 -0
+			.fb_id = invalid_fb->fb_id,
+			.ptr_valid = true,
+			.out_fence_ptr = &out_fence,
+		},

This doesn't belong in this function (invalid_out_fence), since this checks an invalid framebuffer, not an invalid fence. We should probably move it to writeback_fb_id (and rename that function to test_fb?).

Quoted 4 lines

It tries to test that you can't trick the driver to do any work on a fence if the framebuffer is invalid, so the set of tests tries: no fb with valid fence, invalid fb with valid fence, valid fb but invalid fence and assumes that no fb with invalid fence is a NOP anyway.

Quoted 4 lines

Yeah, that makes sense, it's just confusing to put it in a function named invalid_out_fence. Here the out fence is valid, but the output buffer isn't, so it should probably be moved away (or this function should be renamed).

Don't want to offend or anything, but this does sound like bikeshedding. You have a couple of parameters that you want to have a test for because they are linked together (output framebuffer and fence) and you go through the combination of possible bad options in the test. Not sure what name we can use for the function, other than maybe 'test_invalid_parameters'? Given that 2/3 tests an invalid out fence, the name was thought to be relevant.

Having invalid out buffer test separate into its own test brings no benefits, IMHO.

Best regards,
Liviu

Quoted 132 lines
Quoted 124 lines
Quoted 81 lines
Quoted 81 lines
Quoted 5 lines
tests/kms_writeback.c new file +4 -0
+		{
+			/* Invalid WRITEBACK_OUT_FENCE_PTR. */
+			.fb_id = valid_fb->fb_id,
+			.ptr_valid = false,

tests/kms_writeback.c line 199

Quoted 10 lines
tests/kms_writeback.c new file +10 -0
+			.out_fence_ptr = (int32_t *)0x8,
+		},
+	};
+
+	for (i = 0; i < ARRAY_SIZE(invalid_tests); i++) {
+		ret = do_writeback_test(output, DRM_MODE_ATOMIC_ALLOW_MODESET,
+					invalid_tests[i].fb_id,
+					invalid_tests[i].out_fence_ptr,
+					invalid_tests[i].ptr_valid);
+		igt_assert(ret != 0);

Maybe we can check for -ret == EINVAL?

tests/kms_writeback.c line 203

Quoted 4 lines
tests/kms_writeback.c new file +4 -0
+	}
+}
+
+static void writeback_fb_id(igt_output_t *output, igt_fb_t *valid_fb, igt_fb_t *invalid_fb)

invalid_fb doesn't seem to be used here. valid_fb seems to be set to the input framebuffer. It's probably not a good idea to use the same FB for input and writeback output.

Quoted 41 lines
tests/kms_writeback.c new file +41 -0
+{
+
+	int ret;
+
+	/* Valid output buffer */
+	ret = do_writeback_test(output, DRM_MODE_ATOMIC_ALLOW_MODESET,
+				valid_fb->fb_id, NULL, false);
+	igt_assert(ret == 0);
+
+	/* Invalid object for WRITEBACK_FB_ID */
+	ret = do_writeback_test(output, DRM_MODE_ATOMIC_ALLOW_MODESET,
+				output->id, NULL, false);
+	igt_assert(ret == -EINVAL);
+
+	/* Zero WRITEBACK_FB_ID */
+	ret = do_writeback_test(output, DRM_MODE_ATOMIC_ALLOW_MODESET,
+				0, NULL, false);
+	igt_assert(ret == 0);
+}
+
+igt_main
+{
+	igt_display_t display;
+	igt_output_t *output;
+	igt_plane_t *plane;
+	igt_fb_t input_fb;
+	drmModeModeInfo mode;
+	int ret;
+
+	memset(&display, 0, sizeof(display));
+
+	igt_fixture {
+		display.drm_fd = drm_open_driver_master(DRIVER_ANY);
+		igt_display_require(&display, display.drm_fd);
+
+		kmstest_set_vt_graphics_mode();
+
+		igt_display_require(&display, display.drm_fd);
+
+		igt_require(display.is_atomic);
+

tests/kms_writeback.c line 254

Quoted 10 lines
tests/kms_writeback.c new file +10 -0
+		output = kms_writeback_get_output(&display);
+		igt_require(output);
+
+		if (output->use_override_mode)
+			memcpy(&mode, &output->override_mode, sizeof(mode));
+		else
+			memcpy(&mode, &output->config.default_mode, sizeof(mode));
+
+		plane = igt_output_get_plane_type(output, DRM_PLANE_TYPE_PRIMARY);
+		igt_require(plane);

Maybe we can assert on this?

Quoted 7 lines
Quoted 7 lines

tests/kms_writeback.c line 259

Quoted 4 lines
tests/kms_writeback.c new file +4 -0
+		ret = igt_create_fb(display.drm_fd, mode.hdisplay,
+				    mode.vdisplay,
+				    DRM_FORMAT_XRGB8888,
+				    igt_fb_mod_to_tiling(0),

This is supposed to take a modifier. DRM_FORMAT_MOD_LINEAR would be better to make this clear.

Agree. The patchset pre-dates the modifiers support (or has the same age, I forgot)

Quoted 23 lines
Quoted 23 lines

(Applies to other lines of this patch)

tests/kms_writeback.c line 266

Quoted 7 lines
tests/kms_writeback.c new file +7 -0
+				    &input_fb);
+		igt_assert(ret >= 0);
+		igt_plane_set_fb(plane, &input_fb);
+	}
+
+	igt_subtest("writeback-pixel-formats") {
+		drmModePropertyBlobRes *formats_blob = get_writeback_formats_blob(output);

Need to drmModeFreePropertyBlob this.

Quoted 11 lines
tests/kms_writeback.c new file +11 -0
+		const char *valid_chars = "0123456 ABCGNRUVXY";
+		unsigned int i;
+		char *c;
+
+		/*
+		 * We don't have a comprehensive list of formats, so just check
+		 * that the blob length is sensible and that it doesn't contain
+		 * any outlandish characters
+		 */
+		igt_assert(!(formats_blob->length % 4));
+		c = formats_blob->data;
Quoted 8 lines
Quoted 8 lines

tests/kms_writeback.c line 280

Quoted 3 lines
tests/kms_writeback.c new file +3 -0
+		for (i = 0; i < formats_blob->length; i++)
+			igt_assert_f(strchr(valid_chars, c[i]),
+				     "Unexpected character %c\n", c[i]);

Honestly, I'm not a fan of this check. There's no guarantee that fourcc codes will be made from ASCII characters, in fact some formats already have non-printable chars in them. I don't want to have to update this test when a new fourcc format is added.

Quoted 3 lines

Like the comments says, we don't have a full list of formats to check against. Suggestions on how to improve are welcome, but I don't think we should delay (any longer) the patchset due to this.

Agreed.

Quoted 2 lines

Best regards,
Liviu

Signature · 12 lines
====================
| I would like to |
| fix the world,  |
| but they're not |
| giving me the   |
 \ source code!  /
  ---------------
    ¯\_(ツ)_/¯
_______________________________________________
igt-dev mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/igt-dev
#18 lore raw

Re: [igt-dev] [PATCH V6 i-g-t 2/6] kms_writeback: Add initial writeback tests

Quoted 40 lines · On Thu, 2019-07-18 at 10:49 +0100, [email protected] wrote:
Quoted 34 lines · Simon

Thanks for the clarification!

Quoted 26 lines · On Tue, 2019-07-16 at 16:22 +0100, [email protected] wrote:
Quoted 17 lines
Quoted 17 lines
Quoted 17 lines
tests/kms_writeback.c new file +17 -0
+static void invalid_out_fence(igt_output_t *output, igt_fb_t *valid_fb, igt_fb_t *invalid_fb)
+{
+	int i, ret;
+	int32_t out_fence;
+	struct {
+		uint32_t fb_id;
+		bool ptr_valid;
+		int32_t *out_fence_ptr;
+	} invalid_tests[] = {
+		{
+			/* No output buffer, but the WRITEBACK_OUT_FENCE_PTR set. */
+			.fb_id = 0,
+			.ptr_valid = true,
+			.out_fence_ptr = &out_fence,
+		},
+		{
+			/* Invalid output buffer. */
Quoted 8 lines
Quoted 8 lines

tests/kms_writeback.c line 185

Quoted 4 lines
tests/kms_writeback.c new file +4 -0
+			.fb_id = invalid_fb->fb_id,
+			.ptr_valid = true,
+			.out_fence_ptr = &out_fence,
+		},

This doesn't belong in this function (invalid_out_fence), since this checks an invalid framebuffer, not an invalid fence. We should probably move it to writeback_fb_id (and rename that function to test_fb?).

Quoted 4 lines

It tries to test that you can't trick the driver to do any work on a fence if the framebuffer is invalid, so the set of tests tries: no fb with valid fence, invalid fb with valid fence, valid fb but invalid fence and assumes that no fb with invalid fence is a NOP anyway.

Quoted 4 lines

Yeah, that makes sense, it's just confusing to put it in a function named invalid_out_fence. Here the out fence is valid, but the output buffer isn't, so it should probably be moved away (or this function should be renamed).

Quoted 8 lines

Don't want to offend or anything, but this does sound like bikeshedding. You have a couple of parameters that you want to have a test for because they are linked together (output framebuffer and fence) and you go through the combination of possible bad options in the test. Not sure what name we can use for the function, other than maybe 'test_invalid_parameters'? Given that 2/3 tests an invalid out fence, the name was thought to be relevant.

Having invalid out buffer test separate into its own test brings no benefits, IMHO.

Well, the issue is that I've been confused when reviewing the patch series. I had trouble understanding what the test does and why. I also had trouble to identify that do_writeback_test never submits a writeback operation (see other e-mail).

A name that is relevant "all the time, most of the time", is not relevant at all in my opinion. It just tricks the reader into thinking the test does one thing, while it also does something else.

If it would be obvious, I wouldn't mind. But here IMHO it hurts readability. So I'd prefer to rename the function.

If you think it's obvious, then maybe it's just me. I'd love to hear from others if they have a different opinion.

(As a side note, I agree I have a tendency to bikeshed, I try to mark my bikesheddings behind "nit:" flags.)

Quoted 134 lines

Best regards,
Liviu

Quoted 131 lines
Quoted 123 lines
Quoted 80 lines
Quoted 80 lines
Quoted 4 lines
tests/kms_writeback.c new file +4 -0
+		{
+			/* Invalid WRITEBACK_OUT_FENCE_PTR. */
+			.fb_id = valid_fb->fb_id,
+			.ptr_valid = false,

tests/kms_writeback.c line 199

Quoted 10 lines
tests/kms_writeback.c new file +10 -0
+			.out_fence_ptr = (int32_t *)0x8,
+		},
+	};
+
+	for (i = 0; i < ARRAY_SIZE(invalid_tests); i++) {
+		ret = do_writeback_test(output, DRM_MODE_ATOMIC_ALLOW_MODESET,
+					invalid_tests[i].fb_id,
+					invalid_tests[i].out_fence_ptr,
+					invalid_tests[i].ptr_valid);
+		igt_assert(ret != 0);

Maybe we can check for -ret == EINVAL?

tests/kms_writeback.c line 203

Quoted 4 lines
tests/kms_writeback.c new file +4 -0
+	}
+}
+
+static void writeback_fb_id(igt_output_t *output, igt_fb_t *valid_fb, igt_fb_t *invalid_fb)

invalid_fb doesn't seem to be used here. valid_fb seems to be set to the input framebuffer. It's probably not a good idea to use the same FB for input and writeback output.

Quoted 41 lines
tests/kms_writeback.c new file +41 -0
+{
+
+	int ret;
+
+	/* Valid output buffer */
+	ret = do_writeback_test(output, DRM_MODE_ATOMIC_ALLOW_MODESET,
+				valid_fb->fb_id, NULL, false);
+	igt_assert(ret == 0);
+
+	/* Invalid object for WRITEBACK_FB_ID */
+	ret = do_writeback_test(output, DRM_MODE_ATOMIC_ALLOW_MODESET,
+				output->id, NULL, false);
+	igt_assert(ret == -EINVAL);
+
+	/* Zero WRITEBACK_FB_ID */
+	ret = do_writeback_test(output, DRM_MODE_ATOMIC_ALLOW_MODESET,
+				0, NULL, false);
+	igt_assert(ret == 0);
+}
+
+igt_main
+{
+	igt_display_t display;
+	igt_output_t *output;
+	igt_plane_t *plane;
+	igt_fb_t input_fb;
+	drmModeModeInfo mode;
+	int ret;
+
+	memset(&display, 0, sizeof(display));
+
+	igt_fixture {
+		display.drm_fd = drm_open_driver_master(DRIVER_ANY);
+		igt_display_require(&display, display.drm_fd);
+
+		kmstest_set_vt_graphics_mode();
+
+		igt_display_require(&display, display.drm_fd);
+
+		igt_require(display.is_atomic);
+

tests/kms_writeback.c line 254

Quoted 10 lines
tests/kms_writeback.c new file +10 -0
+		output = kms_writeback_get_output(&display);
+		igt_require(output);
+
+		if (output->use_override_mode)
+			memcpy(&mode, &output->override_mode, sizeof(mode));
+		else
+			memcpy(&mode, &output->config.default_mode, sizeof(mode));
+
+		plane = igt_output_get_plane_type(output, DRM_PLANE_TYPE_PRIMARY);
+		igt_require(plane);

Maybe we can assert on this?

Quoted 7 lines
Quoted 7 lines

tests/kms_writeback.c line 259

Quoted 4 lines
tests/kms_writeback.c new file +4 -0
+		ret = igt_create_fb(display.drm_fd, mode.hdisplay,
+				    mode.vdisplay,
+				    DRM_FORMAT_XRGB8888,
+				    igt_fb_mod_to_tiling(0),

This is supposed to take a modifier. DRM_FORMAT_MOD_LINEAR would be better to make this clear.

Agree. The patchset pre-dates the modifiers support (or has the same age, I forgot)

Quoted 23 lines
Quoted 23 lines

(Applies to other lines of this patch)

tests/kms_writeback.c line 266

Quoted 7 lines
tests/kms_writeback.c new file +7 -0
+				    &input_fb);
+		igt_assert(ret >= 0);
+		igt_plane_set_fb(plane, &input_fb);
+	}
+
+	igt_subtest("writeback-pixel-formats") {
+		drmModePropertyBlobRes *formats_blob = get_writeback_formats_blob(output);

Need to drmModeFreePropertyBlob this.

Quoted 11 lines
tests/kms_writeback.c new file +11 -0
+		const char *valid_chars = "0123456 ABCGNRUVXY";
+		unsigned int i;
+		char *c;
+
+		/*
+		 * We don't have a comprehensive list of formats, so just check
+		 * that the blob length is sensible and that it doesn't contain
+		 * any outlandish characters
+		 */
+		igt_assert(!(formats_blob->length % 4));
+		c = formats_blob->data;
Quoted 8 lines
Quoted 8 lines

tests/kms_writeback.c line 280

Quoted 3 lines
tests/kms_writeback.c new file +3 -0
+		for (i = 0; i < formats_blob->length; i++)
+			igt_assert_f(strchr(valid_chars, c[i]),
+				     "Unexpected character %c\n", c[i]);

Honestly, I'm not a fan of this check. There's no guarantee that fourcc codes will be made from ASCII characters, in fact some formats already have non-printable chars in them. I don't want to have to update this test when a new fourcc format is added.

Quoted 3 lines

Like the comments says, we don't have a full list of formats to check against. Suggestions on how to improve are welcome, but I don't think we should delay (any longer) the patchset due to this.

Agreed.

Quoted 2 lines

Best regards,
Liviu

_______________________________________________
igt-dev mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/igt-dev

#19 lore raw

Re: [igt-dev] [PATCH V6 i-g-t 2/6] kms_writeback: Add initial writeback tests

Quoted 55 lines · Simon
Quoted 40 lines · On Thu, 2019-07-18 at 10:49 +0100, [email protected] wrote:
Quoted 34 lines · Simon

Thanks for the clarification!

Quoted 26 lines · On Tue, 2019-07-16 at 16:22 +0100, [email protected] wrote:
Quoted 17 lines
Quoted 17 lines
Quoted 17 lines
(quoted diff) +17 -0
+static void invalid_out_fence(igt_output_t *output, igt_fb_t *valid_fb, igt_fb_t *invalid_fb)
+{
+	int i, ret;
+	int32_t out_fence;
+	struct {
+		uint32_t fb_id;
+		bool ptr_valid;
+		int32_t *out_fence_ptr;
+	} invalid_tests[] = {
+		{
+			/* No output buffer, but the WRITEBACK_OUT_FENCE_PTR set. */
+			.fb_id = 0,
+			.ptr_valid = true,
+			.out_fence_ptr = &out_fence,
+		},
+		{
+			/* Invalid output buffer. */
Quoted 8 lines
Quoted 8 lines

Review

Quoted 4 lines
(quoted diff) +4 -0
+			.fb_id = invalid_fb->fb_id,
+			.ptr_valid = true,
+			.out_fence_ptr = &out_fence,
+		},

This doesn't belong in this function (invalid_out_fence), since this checks an invalid framebuffer, not an invalid fence. We should probably move it to writeback_fb_id (and rename that function to test_fb?).

Quoted 4 lines

It tries to test that you can't trick the driver to do any work on a fence if the framebuffer is invalid, so the set of tests tries: no fb with valid fence, invalid fb with valid fence, valid fb but invalid fence and assumes that no fb with invalid fence is a NOP anyway.

Quoted 4 lines

Yeah, that makes sense, it's just confusing to put it in a function named invalid_out_fence. Here the out fence is valid, but the output buffer isn't, so it should probably be moved away (or this function should be renamed).

Quoted 8 lines

Don't want to offend or anything, but this does sound like bikeshedding. You have a couple of parameters that you want to have a test for because they are linked together (output framebuffer and fence) and you go through the combination of possible bad options in the test. Not sure what name we can use for the function, other than maybe 'test_invalid_parameters'? Given that 2/3 tests an invalid out fence, the name was thought to be relevant.

Having invalid out buffer test separate into its own test brings no benefits, IMHO.

Well, the issue is that I've been confused when reviewing the patch series. I had trouble understanding what the test does and why. I also had trouble to identify that do_writeback_test never submits a writeback operation (see other e-mail).

Quoted 6 lines

A name that is relevant "all the time, most of the time", is not relevant at all in my opinion. It just tricks the reader into thinking the test does one thing, while it also does something else.

If it would be obvious, I wouldn't mind. But here IMHO it hurts readability. So I'd prefer to rename the function.

I take your comments as a valid point.

Does "test_invalid_parameters" sound like a good name for the function? Is so, Rodrigo, can you please use that name in the next revision of the patch?

Quoted 6 lines

If you think it's obvious, then maybe it's just me. I'd love to hear from others if they have a different opinion.

(As a side note, I agree I have a tendency to bikeshed, I try to mark my bikesheddings behind "nit:" flags.)

I've only said "it sounded like" :)

Best regards,
Liviu

Quoted 135 lines
Quoted 135 lines

Best regards,
Liviu

Quoted 131 lines
Quoted 123 lines
Quoted 80 lines
Quoted 80 lines
Quoted 4 lines
(quoted diff) +4 -0
+		{
+			/* Invalid WRITEBACK_OUT_FENCE_PTR. */
+			.fb_id = valid_fb->fb_id,
+			.ptr_valid = false,

Review

Quoted 10 lines
(quoted diff) +10 -0
+			.out_fence_ptr = (int32_t *)0x8,
+		},
+	};
+
+	for (i = 0; i < ARRAY_SIZE(invalid_tests); i++) {
+		ret = do_writeback_test(output, DRM_MODE_ATOMIC_ALLOW_MODESET,
+					invalid_tests[i].fb_id,
+					invalid_tests[i].out_fence_ptr,
+					invalid_tests[i].ptr_valid);
+		igt_assert(ret != 0);

Maybe we can check for -ret == EINVAL?

Review

Quoted 4 lines
(quoted diff) +4 -0
+	}
+}
+
+static void writeback_fb_id(igt_output_t *output, igt_fb_t *valid_fb, igt_fb_t *invalid_fb)

invalid_fb doesn't seem to be used here. valid_fb seems to be set to the input framebuffer. It's probably not a good idea to use the same FB for input and writeback output.

Quoted 41 lines
(quoted diff) +41 -0
+{
+
+	int ret;
+
+	/* Valid output buffer */
+	ret = do_writeback_test(output, DRM_MODE_ATOMIC_ALLOW_MODESET,
+				valid_fb->fb_id, NULL, false);
+	igt_assert(ret == 0);
+
+	/* Invalid object for WRITEBACK_FB_ID */
+	ret = do_writeback_test(output, DRM_MODE_ATOMIC_ALLOW_MODESET,
+				output->id, NULL, false);
+	igt_assert(ret == -EINVAL);
+
+	/* Zero WRITEBACK_FB_ID */
+	ret = do_writeback_test(output, DRM_MODE_ATOMIC_ALLOW_MODESET,
+				0, NULL, false);
+	igt_assert(ret == 0);
+}
+
+igt_main
+{
+	igt_display_t display;
+	igt_output_t *output;
+	igt_plane_t *plane;
+	igt_fb_t input_fb;
+	drmModeModeInfo mode;
+	int ret;
+
+	memset(&display, 0, sizeof(display));
+
+	igt_fixture {
+		display.drm_fd = drm_open_driver_master(DRIVER_ANY);
+		igt_display_require(&display, display.drm_fd);
+
+		kmstest_set_vt_graphics_mode();
+
+		igt_display_require(&display, display.drm_fd);
+
+		igt_require(display.is_atomic);
+

Review

Quoted 10 lines
(quoted diff) +10 -0
+		output = kms_writeback_get_output(&display);
+		igt_require(output);
+
+		if (output->use_override_mode)
+			memcpy(&mode, &output->override_mode, sizeof(mode));
+		else
+			memcpy(&mode, &output->config.default_mode, sizeof(mode));
+
+		plane = igt_output_get_plane_type(output, DRM_PLANE_TYPE_PRIMARY);
+		igt_require(plane);

Maybe we can assert on this?

Quoted 7 lines
Quoted 7 lines

Review

Quoted 4 lines
(quoted diff) +4 -0
+		ret = igt_create_fb(display.drm_fd, mode.hdisplay,
+				    mode.vdisplay,
+				    DRM_FORMAT_XRGB8888,
+				    igt_fb_mod_to_tiling(0),

This is supposed to take a modifier. DRM_FORMAT_MOD_LINEAR would be better to make this clear.

Agree. The patchset pre-dates the modifiers support (or has the same age, I forgot)

Quoted 23 lines
Quoted 23 lines

(Applies to other lines of this patch)

Review

Quoted 7 lines
(quoted diff) +7 -0
+				    &input_fb);
+		igt_assert(ret >= 0);
+		igt_plane_set_fb(plane, &input_fb);
+	}
+
+	igt_subtest("writeback-pixel-formats") {
+		drmModePropertyBlobRes *formats_blob = get_writeback_formats_blob(output);

Need to drmModeFreePropertyBlob this.

Quoted 11 lines
(quoted diff) +11 -0
+		const char *valid_chars = "0123456 ABCGNRUVXY";
+		unsigned int i;
+		char *c;
+
+		/*
+		 * We don't have a comprehensive list of formats, so just check
+		 * that the blob length is sensible and that it doesn't contain
+		 * any outlandish characters
+		 */
+		igt_assert(!(formats_blob->length % 4));
+		c = formats_blob->data;
Quoted 8 lines
Quoted 8 lines

Review

Quoted 3 lines
(quoted diff) +3 -0
+		for (i = 0; i < formats_blob->length; i++)
+			igt_assert_f(strchr(valid_chars, c[i]),
+				     "Unexpected character %c\n", c[i]);

Honestly, I'm not a fan of this check. There's no guarantee that fourcc codes will be made from ASCII characters, in fact some formats already have non-printable chars in them. I don't want to have to update this test when a new fourcc format is added.

Quoted 3 lines

Like the comments says, we don't have a full list of formats to check against. Suggestions on how to improve are welcome, but I don't think we should delay (any longer) the patchset due to this.

Agreed.

Quoted 2 lines

Best regards,
Liviu

Signature · 12 lines
====================
| I would like to |
| fix the world,  |
| but they're not |
| giving me the   |
 \ source code!  /
  ---------------
    ¯\_(ツ)_/¯
_______________________________________________
igt-dev mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/igt-dev
#20 lore raw

Re: [igt-dev] [PATCH V6 i-g-t 2/6] kms_writeback: Add initial writeback tests

Quoted 60 lines · On Thu, 2019-07-18 at 12:15 +0100, [email protected] wrote:
Quoted 59 lines · Simon
Quoted 40 lines · On Thu, 2019-07-18 at 10:49 +0100, [email protected] wrote:
Quoted 34 lines · Simon

Thanks for the clarification!

Quoted 26 lines · On Tue, 2019-07-16 at 16:22 +0100, [email protected] wrote:
Quoted 17 lines
Quoted 17 lines
Quoted 17 lines
(quoted diff) +17 -0
+static void invalid_out_fence(igt_output_t *output, igt_fb_t *valid_fb, igt_fb_t *invalid_fb)
+{
+	int i, ret;
+	int32_t out_fence;
+	struct {
+		uint32_t fb_id;
+		bool ptr_valid;
+		int32_t *out_fence_ptr;
+	} invalid_tests[] = {
+		{
+			/* No output buffer, but the WRITEBACK_OUT_FENCE_PTR set. */
+			.fb_id = 0,
+			.ptr_valid = true,
+			.out_fence_ptr = &out_fence,
+		},
+		{
+			/* Invalid output buffer. */
Quoted 8 lines
Quoted 8 lines

Review

Quoted 4 lines
(quoted diff) +4 -0
+			.fb_id = invalid_fb->fb_id,
+			.ptr_valid = true,
+			.out_fence_ptr = &out_fence,
+		},

This doesn't belong in this function (invalid_out_fence), since this checks an invalid framebuffer, not an invalid fence. We should probably move it to writeback_fb_id (and rename that function to test_fb?).

Quoted 4 lines

It tries to test that you can't trick the driver to do any work on a fence if the framebuffer is invalid, so the set of tests tries: no fb with valid fence, invalid fb with valid fence, valid fb but invalid fence and assumes that no fb with invalid fence is a NOP anyway.

Quoted 4 lines

Yeah, that makes sense, it's just confusing to put it in a function named invalid_out_fence. Here the out fence is valid, but the output buffer isn't, so it should probably be moved away (or this function should be renamed).

Quoted 8 lines

Don't want to offend or anything, but this does sound like bikeshedding. You have a couple of parameters that you want to have a test for because they are linked together (output framebuffer and fence) and you go through the combination of possible bad options in the test. Not sure what name we can use for the function, other than maybe 'test_invalid_parameters'? Given that 2/3 tests an invalid out fence, the name was thought to be relevant.

Having invalid out buffer test separate into its own test brings no benefits, IMHO.

Well, the issue is that I've been confused when reviewing the patch series. I had trouble understanding what the test does and why. I also had trouble to identify that do_writeback_test never submits a writeback operation (see other e-mail).

A name that is relevant "all the time, most of the time", is not relevant at all in my opinion. It just tricks the reader into thinking the test does one thing, while it also does something else.

Quoted 7 lines
Quoted 2 lines

If it would be obvious, I wouldn't mind. But here IMHO it hurts readability. So I'd prefer to rename the function.

I take your comments as a valid point.

Does "test_invalid_parameters" sound like a good name for the function? Is so, Rodrigo, can you please use that name in the next revision of the patch?

Yes, this name sounds perfectly fine :)

Quoted 7 lines
Quoted 5 lines

If you think it's obvious, then maybe it's just me. I'd love to hear from others if they have a different opinion.

(As a side note, I agree I have a tendency to bikeshed, I try to mark my bikesheddings behind "nit:" flags.)

I've only said "it sounded like" :)

Eheh :P

Quoted 137 lines

Best regards,
Liviu

Quoted 134 lines
Quoted 134 lines

Best regards,
Liviu

Quoted 131 lines
Quoted 123 lines
Quoted 80 lines
Quoted 80 lines
Quoted 4 lines
(quoted diff) +4 -0
+		{
+			/* Invalid WRITEBACK_OUT_FENCE_PTR. */
+			.fb_id = valid_fb->fb_id,
+			.ptr_valid = false,

Review

Quoted 10 lines
(quoted diff) +10 -0
+			.out_fence_ptr = (int32_t *)0x8,
+		},
+	};
+
+	for (i = 0; i < ARRAY_SIZE(invalid_tests); i++) {
+		ret = do_writeback_test(output, DRM_MODE_ATOMIC_ALLOW_MODESET,
+					invalid_tests[i].fb_id,
+					invalid_tests[i].out_fence_ptr,
+					invalid_tests[i].ptr_valid);
+		igt_assert(ret != 0);

Maybe we can check for -ret == EINVAL?

Review

Quoted 4 lines
(quoted diff) +4 -0
+	}
+}
+
+static void writeback_fb_id(igt_output_t *output, igt_fb_t *valid_fb, igt_fb_t *invalid_fb)

invalid_fb doesn't seem to be used here. valid_fb seems to be set to the input framebuffer. It's probably not a good idea to use the same FB for input and writeback output.

Quoted 41 lines
(quoted diff) +41 -0
+{
+
+	int ret;
+
+	/* Valid output buffer */
+	ret = do_writeback_test(output, DRM_MODE_ATOMIC_ALLOW_MODESET,
+				valid_fb->fb_id, NULL, false);
+	igt_assert(ret == 0);
+
+	/* Invalid object for WRITEBACK_FB_ID */
+	ret = do_writeback_test(output, DRM_MODE_ATOMIC_ALLOW_MODESET,
+				output->id, NULL, false);
+	igt_assert(ret == -EINVAL);
+
+	/* Zero WRITEBACK_FB_ID */
+	ret = do_writeback_test(output, DRM_MODE_ATOMIC_ALLOW_MODESET,
+				0, NULL, false);
+	igt_assert(ret == 0);
+}
+
+igt_main
+{
+	igt_display_t display;
+	igt_output_t *output;
+	igt_plane_t *plane;
+	igt_fb_t input_fb;
+	drmModeModeInfo mode;
+	int ret;
+
+	memset(&display, 0, sizeof(display));
+
+	igt_fixture {
+		display.drm_fd = drm_open_driver_master(DRIVER_ANY);
+		igt_display_require(&display, display.drm_fd);
+
+		kmstest_set_vt_graphics_mode();
+
+		igt_display_require(&display, display.drm_fd);
+
+		igt_require(display.is_atomic);
+

Review

Quoted 10 lines
(quoted diff) +10 -0
+		output = kms_writeback_get_output(&display);
+		igt_require(output);
+
+		if (output->use_override_mode)
+			memcpy(&mode, &output->override_mode, sizeof(mode));
+		else
+			memcpy(&mode, &output->config.default_mode, sizeof(mode));
+
+		plane = igt_output_get_plane_type(output, DRM_PLANE_TYPE_PRIMARY);
+		igt_require(plane);

Maybe we can assert on this?

Quoted 7 lines
Quoted 7 lines

Review

Quoted 4 lines
(quoted diff) +4 -0
+		ret = igt_create_fb(display.drm_fd, mode.hdisplay,
+				    mode.vdisplay,
+				    DRM_FORMAT_XRGB8888,
+				    igt_fb_mod_to_tiling(0),

This is supposed to take a modifier. DRM_FORMAT_MOD_LINEAR would be better to make this clear.

Agree. The patchset pre-dates the modifiers support (or has the same age, I forgot)

Quoted 23 lines
Quoted 23 lines

(Applies to other lines of this patch)

Review

Quoted 7 lines
(quoted diff) +7 -0
+				    &input_fb);
+		igt_assert(ret >= 0);
+		igt_plane_set_fb(plane, &input_fb);
+	}
+
+	igt_subtest("writeback-pixel-formats") {
+		drmModePropertyBlobRes *formats_blob = get_writeback_formats_blob(output);

Need to drmModeFreePropertyBlob this.

Quoted 11 lines
(quoted diff) +11 -0
+		const char *valid_chars = "0123456 ABCGNRUVXY";
+		unsigned int i;
+		char *c;
+
+		/*
+		 * We don't have a comprehensive list of formats, so just check
+		 * that the blob length is sensible and that it doesn't contain
+		 * any outlandish characters
+		 */
+		igt_assert(!(formats_blob->length % 4));
+		c = formats_blob->data;
Quoted 8 lines
Quoted 8 lines

Review

Quoted 3 lines
(quoted diff) +3 -0
+		for (i = 0; i < formats_blob->length; i++)
+			igt_assert_f(strchr(valid_chars, c[i]),
+				     "Unexpected character %c\n", c[i]);

Honestly, I'm not a fan of this check. There's no guarantee that fourcc codes will be made from ASCII characters, in fact some formats already have non-printable chars in them. I don't want to have to update this test when a new fourcc format is added.

Quoted 3 lines

Like the comments says, we don't have a full list of formats to check against. Suggestions on how to improve are welcome, but I don't think we should delay (any longer) the patchset due to this.

Agreed.

Quoted 2 lines

Best regards,
Liviu

_______________________________________________
igt-dev mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/igt-dev

#21 lore raw

Re: [igt-dev] [PATCH V6 i-g-t 2/6] kms_writeback: Add initial writeback tests

In reply to [email protected]
Quoted 61 lines · On Thu, Jul 18, 2019 at 8:15 AM [email protected] <[email protected]> wrote:
Quoted 59 lines · Simon
Quoted 40 lines · On Thu, 2019-07-18 at 10:49 +0100, [email protected] wrote:
Quoted 34 lines · Simon

Thanks for the clarification!

Quoted 26 lines · On Tue, 2019-07-16 at 16:22 +0100, [email protected] wrote:
Quoted 17 lines
Quoted 17 lines
Quoted 17 lines
(quoted diff) +17 -0
+static void invalid_out_fence(igt_output_t *output, igt_fb_t *valid_fb, igt_fb_t *invalid_fb)
+{
+ int i, ret;
+ int32_t out_fence;
+ struct {
+         uint32_t fb_id;
+         bool ptr_valid;
+         int32_t *out_fence_ptr;
+ } invalid_tests[] = {
+         {
+                 /* No output buffer, but the WRITEBACK_OUT_FENCE_PTR set. */
+                 .fb_id = 0,
+                 .ptr_valid = true,
+                 .out_fence_ptr = &out_fence,
+         },
+         {
+                 /* Invalid output buffer. */
Quoted 8 lines
Quoted 8 lines

Review

Quoted 4 lines
(quoted diff) +4 -0
+                 .fb_id = invalid_fb->fb_id,
+                 .ptr_valid = true,
+                 .out_fence_ptr = &out_fence,
+         },

This doesn't belong in this function (invalid_out_fence), since this checks an invalid framebuffer, not an invalid fence. We should probably move it to writeback_fb_id (and rename that function to test_fb?).

Quoted 4 lines

It tries to test that you can't trick the driver to do any work on a fence if the framebuffer is invalid, so the set of tests tries: no fb with valid fence, invalid fb with valid fence, valid fb but invalid fence and assumes that no fb with invalid fence is a NOP anyway.

Quoted 4 lines

Yeah, that makes sense, it's just confusing to put it in a function named invalid_out_fence. Here the out fence is valid, but the output buffer isn't, so it should probably be moved away (or this function should be renamed).

Quoted 8 lines

Don't want to offend or anything, but this does sound like bikeshedding. You have a couple of parameters that you want to have a test for because they are linked together (output framebuffer and fence) and you go through the combination of possible bad options in the test. Not sure what name we can use for the function, other than maybe 'test_invalid_parameters'? Given that 2/3 tests an invalid out fence, the name was thought to be relevant.

Having invalid out buffer test separate into its own test brings no benefits, IMHO.

Well, the issue is that I've been confused when reviewing the patch series. I had trouble understanding what the test does and why. I also had trouble to identify that do_writeback_test never submits a writeback operation (see other e-mail).

A name that is relevant "all the time, most of the time", is not relevant at all in my opinion. It just tricks the reader into thinking the test does one thing, while it also does something else.

Quoted 7 lines
Quoted 2 lines

If it would be obvious, I wouldn't mind. But here IMHO it hurts readability. So I'd prefer to rename the function.

I take your comments as a valid point.

Does "test_invalid_parameters" sound like a good name for the function? Is so, Rodrigo, can you please use that name in the next revision of the patch?

Sure, I'll do that.

Thanks

Quoted 157 lines
Quoted 6 lines

If you think it's obvious, then maybe it's just me. I'd love to hear from others if they have a different opinion.

(As a side note, I agree I have a tendency to bikeshed, I try to mark my bikesheddings behind "nit:" flags.)

I've only said "it sounded like" :)

Best regards,
Liviu

Quoted 127 lines
Quoted 127 lines

Best regards,
Liviu

Quoted 123 lines
Quoted 123 lines
Quoted 80 lines
Quoted 80 lines
Quoted 4 lines
(quoted diff) +4 -0
+         {
+                 /* Invalid WRITEBACK_OUT_FENCE_PTR. */
+                 .fb_id = valid_fb->fb_id,
+                 .ptr_valid = false,

Review

Quoted 10 lines
(quoted diff) +10 -0
+                 .out_fence_ptr = (int32_t *)0x8,
+         },
+ };
+
+ for (i = 0; i < ARRAY_SIZE(invalid_tests); i++) {
+         ret = do_writeback_test(output, DRM_MODE_ATOMIC_ALLOW_MODESET,
+                                 invalid_tests[i].fb_id,
+                                 invalid_tests[i].out_fence_ptr,
+                                 invalid_tests[i].ptr_valid);
+         igt_assert(ret != 0);

Maybe we can check for -ret == EINVAL?

Review

Quoted 4 lines

+ }

(quoted diff) +3 -0
+}
+
+static void writeback_fb_id(igt_output_t *output, igt_fb_t *valid_fb, igt_fb_t *invalid_fb)

invalid_fb doesn't seem to be used here. valid_fb seems to be set to the input framebuffer. It's probably not a good idea to use the same FB for input and writeback output.

Quoted 41 lines
(quoted diff) +41 -0
+{
+
+ int ret;
+
+ /* Valid output buffer */
+ ret = do_writeback_test(output, DRM_MODE_ATOMIC_ALLOW_MODESET,
+                         valid_fb->fb_id, NULL, false);
+ igt_assert(ret == 0);
+
+ /* Invalid object for WRITEBACK_FB_ID */
+ ret = do_writeback_test(output, DRM_MODE_ATOMIC_ALLOW_MODESET,
+                         output->id, NULL, false);
+ igt_assert(ret == -EINVAL);
+
+ /* Zero WRITEBACK_FB_ID */
+ ret = do_writeback_test(output, DRM_MODE_ATOMIC_ALLOW_MODESET,
+                         0, NULL, false);
+ igt_assert(ret == 0);
+}
+
+igt_main
+{
+ igt_display_t display;
+ igt_output_t *output;
+ igt_plane_t *plane;
+ igt_fb_t input_fb;
+ drmModeModeInfo mode;
+ int ret;
+
+ memset(&display, 0, sizeof(display));
+
+ igt_fixture {
+         display.drm_fd = drm_open_driver_master(DRIVER_ANY);
+         igt_display_require(&display, display.drm_fd);
+
+         kmstest_set_vt_graphics_mode();
+
+         igt_display_require(&display, display.drm_fd);
+
+         igt_require(display.is_atomic);
+

Review

Quoted 10 lines
(quoted diff) +10 -0
+         output = kms_writeback_get_output(&display);
+         igt_require(output);
+
+         if (output->use_override_mode)
+                 memcpy(&mode, &output->override_mode, sizeof(mode));
+         else
+                 memcpy(&mode, &output->config.default_mode, sizeof(mode));
+
+         plane = igt_output_get_plane_type(output, DRM_PLANE_TYPE_PRIMARY);
+         igt_require(plane);

Maybe we can assert on this?

Quoted 7 lines
Quoted 7 lines

Review

Quoted 4 lines
(quoted diff) +4 -0
+         ret = igt_create_fb(display.drm_fd, mode.hdisplay,
+                             mode.vdisplay,
+                             DRM_FORMAT_XRGB8888,
+                             igt_fb_mod_to_tiling(0),

This is supposed to take a modifier. DRM_FORMAT_MOD_LINEAR would be better to make this clear.

Agree. The patchset pre-dates the modifiers support (or has the same age, I forgot)

Quoted 31 lines
Quoted 31 lines

(Applies to other lines of this patch)

Review

Quoted 7 lines
(quoted diff) +7 -0
+                             &input_fb);
+         igt_assert(ret >= 0);
+         igt_plane_set_fb(plane, &input_fb);
+ }
+
+ igt_subtest("writeback-pixel-formats") {
+         drmModePropertyBlobRes *formats_blob = get_writeback_formats_blob(output);

Need to drmModeFreePropertyBlob this.

Quoted 4 lines
(quoted diff) +4 -0
+         const char *valid_chars = "0123456 ABCGNRUVXY";
+         unsigned int i;
+         char *c;
+

Review

Quoted 10 lines
(quoted diff) +10 -0
+         /*
+          * We don't have a comprehensive list of formats, so just check
+          * that the blob length is sensible and that it doesn't contain
+          * any outlandish characters
+          */
+         igt_assert(!(formats_blob->length % 4));
+         c = formats_blob->data;
+         for (i = 0; i < formats_blob->length; i++)
+                 igt_assert_f(strchr(valid_chars, c[i]),
+                              "Unexpected character %c\n", c[i]);

Honestly, I'm not a fan of this check. There's no guarantee that fourcc codes will be made from ASCII characters, in fact some formats already have non-printable chars in them. I don't want to have to update this test when a new fourcc format is added.

Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
Quoted 3 lines

Like the comments says, we don't have a full list of formats to check against. Suggestions on how to improve are welcome, but I don't think we should delay (any longer) the patchset due to this.

Agreed.

Quoted 2 lines

Best regards,
Liviu

--
====================

| I would like to |
| fix the world,  |
| but they're not |
| giving me the   |
 \ source code!  /
  ---------------
    ¯\_(ツ)_/¯
Signature · 6 lines
Rodrigo Siqueira
https://siqueira.tech
_______________________________________________
igt-dev mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/igt-dev