Eager CAPTURE format probe in RequestInit

Chromium's vaapi_video_decoder may call vaCreateContext with surfaces=NULL,
surfaces_count=0 and then create surfaces afterwards via vaCreateSurfaces2.
In that order driver_data->video_format is still NULL when CreateContext
runs and our early `video_format == NULL` guard returns OPERATION_FAILED.
Confirmed via temporary request_log() — Brave hits exactly that path on
ohm.

Move the probe out of RequestCreateSurfaces into a new
video_format_probe() helper in video.c, and call it eagerly from
RequestInit. RequestCreateSurfaces still re-probes if init came up NULL,
which preserves the original lazy behaviour for any caller that needs it.

Also small clean-up of surface.c since the probe block moved out: drop
the now-dead `bool found` local.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

src/request.c

@@ -39,6 +39,7 @@
 #include "request.h"
 #include "utils.h"
 #include "v4l2.h"
+#include "video.h"
 
 #include <assert.h>
 #include <stdio.h>
@@ -179,6 +180,19 @@ VAStatus VA_DRIVER_INIT_FUNC(VADriverContextP context)
 	driver_data->video_fd = video_fd;
 	driver_data->media_fd = media_fd;
 
+	/* Probe the CAPTURE pixel format eagerly so vaCreateContext doesn't
+	 * race ahead of vaCreateSurfaces. Fine if this returns NULL — we
+	 * still let the driver init succeed and let surface creation try
+	 * again later (preserves the original lazy behaviour for any caller
+	 * that needs it). */
+	driver_data->video_format = video_format_probe(video_fd);
+	if (driver_data->video_format != NULL)
+		request_log("Init: detected CAPTURE format %s (mplane=%d)\n",
+			    driver_data->video_format->description,
+			    driver_data->video_format->v4l2_mplane);
+	else
+		request_log("Init: no CAPTURE format detected at probe time\n");
+
 	status = VA_STATUS_SUCCESS;
 	goto complete;
 

src/surface.c

@@ -65,54 +65,31 @@ VAStatus RequestCreateSurfaces2(VADriverContextP context, unsigned int format,
 	unsigned int index;
 	unsigned int i, j;
 	VASurfaceID id;
-	bool found;
 	int rc;
 
 	if (format != VA_RT_FORMAT_YUV420)
 		return VA_STATUS_ERROR_UNSUPPORTED_RT_FORMAT;
 
 
-        if (!driver_data->video_format) {
-		/* Single-plane CAPTURE first (the original sunxi-cedrus path) */
-		found = v4l2_find_format(driver_data->video_fd,
-					 V4L2_BUF_TYPE_VIDEO_CAPTURE,
-					 V4L2_PIX_FMT_SUNXI_TILED_NV12);
-		if (found)
-			video_format = video_format_find(V4L2_PIX_FMT_SUNXI_TILED_NV12, false);
-
-		if (video_format == NULL) {
-			found = v4l2_find_format(driver_data->video_fd,
-						 V4L2_BUF_TYPE_VIDEO_CAPTURE,
-						 V4L2_PIX_FMT_NV12);
-			if (found)
-				video_format = video_format_find(V4L2_PIX_FMT_NV12, false);
-		}
-
-		/* Multi-plane CAPTURE fallback (Rockchip hantro / RK3588 VDPU381) */
-		if (video_format == NULL) {
-			found = v4l2_find_format(driver_data->video_fd,
-						 V4L2_BUF_TYPE_VIDEO_CAPTURE_MPLANE,
-						 V4L2_PIX_FMT_NV12);
-			if (found)
-				video_format = video_format_find(V4L2_PIX_FMT_NV12, true);
-		}
-
-		if (video_format == NULL)
+	if (!driver_data->video_format) {
+		/* Could happen if RequestInit's eager probe came back NULL —
+		 * try one more time here in case the device only became
+		 * format-enumerable after streaming setup. */
+		driver_data->video_format = video_format_probe(driver_data->video_fd);
+		if (driver_data->video_format == NULL)
 			return VA_STATUS_ERROR_OPERATION_FAILED;
-
-		driver_data->video_format = video_format;
-
-		capture_type = v4l2_type_video_capture(video_format->v4l2_mplane);
-
-		rc = v4l2_set_format(driver_data->video_fd, capture_type,
-				     video_format->v4l2_format, width, height);
-		if (rc < 0)
-			return VA_STATUS_ERROR_OPERATION_FAILED;
-        } else {
-		video_format = driver_data->video_format;
-		capture_type = v4l2_type_video_capture(video_format->v4l2_mplane);
 	}
 
+	video_format = driver_data->video_format;
+	capture_type = v4l2_type_video_capture(video_format->v4l2_mplane);
+
+	/* Set the CAPTURE format on first surface creation so the size is
+	 * pinned to the picture dimensions the caller passed in. */
+	rc = v4l2_set_format(driver_data->video_fd, capture_type,
+			     video_format->v4l2_format, width, height);
+	if (rc < 0)
+		return VA_STATUS_ERROR_OPERATION_FAILED;
+
 	rc = v4l2_get_format(driver_data->video_fd, capture_type, &format_width,
 			     &format_height, destination_bytesperlines,
 			     destination_sizes, NULL);

src/video.c

@@ -32,6 +32,7 @@
 #include <linux/videodev2.h>
 
 #include "utils.h"
+#include "v4l2.h"
 #include "video.h"
 
 static struct video_format formats[] = {
@@ -81,6 +82,40 @@ struct video_format *video_format_find(unsigned int pixelformat, bool mplane)
 	return NULL;
 }
 
+/*
+ * Probe the V4L2 video device for a supported CAPTURE pixel format and
+ * return the matching video_format entry. Tries single-plane CAPTURE first
+ * (the original sunxi-cedrus path: SUNXI_TILED_NV12 or NV12), then falls
+ * back to multi-plane CAPTURE (Rockchip hantro / RK3588 VDPU381). Returns
+ * NULL if no match.
+ *
+ * Centralised here so RequestInit can populate driver_data->video_format
+ * before any vaCreateContext call. The original library set it lazily in
+ * RequestCreateSurfaces, but Chromium's vaapi_video_decoder may call
+ * vaCreateContext first (with surfaces=NULL, surfaces_count=0) and only
+ * create surfaces afterwards via vaCreateSurfaces2.
+ */
+struct video_format *video_format_probe(int video_fd)
+{
+	struct video_format *format = NULL;
+
+	if (v4l2_find_format(video_fd, V4L2_BUF_TYPE_VIDEO_CAPTURE,
+			     V4L2_PIX_FMT_SUNXI_TILED_NV12))
+		format = video_format_find(V4L2_PIX_FMT_SUNXI_TILED_NV12, false);
+
+	if (format == NULL &&
+	    v4l2_find_format(video_fd, V4L2_BUF_TYPE_VIDEO_CAPTURE,
+			     V4L2_PIX_FMT_NV12))
+		format = video_format_find(V4L2_PIX_FMT_NV12, false);
+
+	if (format == NULL &&
+	    v4l2_find_format(video_fd, V4L2_BUF_TYPE_VIDEO_CAPTURE_MPLANE,
+			     V4L2_PIX_FMT_NV12))
+		format = video_format_find(V4L2_PIX_FMT_NV12, true);
+
+	return format;
+}
+
 bool video_format_is_linear(struct video_format *format)
 {
 	if (format == NULL)

src/video.h

@@ -39,6 +39,7 @@ struct video_format {
 };
 
 struct video_format *video_format_find(unsigned int pixelformat, bool mplane);
+struct video_format *video_format_probe(int video_fd);
 bool video_format_is_linear(struct video_format *format);
 
 #endif