[Docs] (Anland): Chrome decodes video in hardware through the YUV shared images, with probes to check it

.claude/skills/anland-mobilegl-plasma/SKILL.md

@@ -126,9 +126,9 @@ timeout 25 mgrun offscreen glmark2-es2-wayland -s 1280x720 --run-forever > /tmp/
 sleep 8; grep -a "presented through" /tmp/mgl-run.client.log
 ```
 Chrome: `bash scripts/chrome-launch.sh <tag> [flags]` (log `/tmp/mgl-chrome-<tag>.client.log`, screenshot `ch-<tag>.png`; `CHROME_GPU=in-process` puts the GPU back in the browser, which cannot recover from a device loss).
-Video: the container has a VA-API driver on the render node the GPU process gets (`/usr/lib/dri/msm_drm_drv_video.so`, V4L2 decode into its own NV12 dma-bufs); MobileGL cannot import those (only its own RGBA shared images; no dma-buf import in the phone's Vulkan, no `GL_OES_EGL_image_external`), and Chrome's GPU process aborts (SIGTRAP, "GPU process exited unexpectedly: exit_code=5") on the first decoded frame - window flicker + stalled video in a loop. `mobilegl-startup.sh chrome` therefore passes `--disable-accelerated-video-decode`. Diagnose a video with a CDP `Media.enable` dump (`kVideoDecoderName`); remote debugging needs a non-default `--user-data-dir`, and a fresh profile also needs `--password-store=basic` (else page loads wait forever on KWallet).
+Video: Chrome decodes in hardware. The GPU process's render node has a VA-API driver in the container (`/usr/lib/dri/msm_drm_drv_video.so`, DroidSpaces' V4L2 driver): Chrome allocates its frames with `vaCreateSurfaces` and exports them (`vaExportSurfaceHandle`, one fd, R8 + GR88 layers) - the driver's own buffers, never GBM ones - and imports each as an NV12 (P010 for 10-bit) EGLImage bound to `GL_TEXTURE_EXTERNAL_OES`. MobileGL takes such a FOREIGN YUV dma-buf through its labelled CPU-copy fallback: the server maps it and copies it into a YUV AHardwareBuffer of its own (`AHardwareBuffer_lockPlanes`) at the reader's first use in each frame (any present, publishing flush or native fence), then converts that into the texture's RGBA level on the GPU (Espryt: `GL_EXT_YUV_target` with the import's colour hints, `MOBILEGL_YUV_DRIVER_CONVERSION=1` for the driver's external sampler; Magma: `VkSamplerYcbcrConversion` on the AHB's external format). A server-allocated NV12/P010 buffer (GBM `gbm_bo_create(NV12)`) needs no copy. `MOBILEGL_CHROME_VIDEO_DECODE=software` in `mobilegl-startup.sh chrome` brings back `--disable-accelerated-video-decode`. Check a video with `scripts/probes/cdpmedia.py <port> <s>` in the container (CDP `Media.enable`: `kVideoDecoderName = VaapiVideoDecoder`, `kIsPlatformVideoDecoder = true`; `CDP_NAVIGATE=<url>` / `CDP_EXPR=<js>`), which needs Chrome started with `--remote-debugging-port` and a non-default `--user-data-dir` (a fresh profile also needs `--password-store=basic`, else page loads wait forever on KWallet), and grep Chrome's log for `exit_code` (a GPU-process restart). `scripts/probes/yuvprobe.c` pixel-checks foreign NV12/P010 imports, their per-frame refresh and a GBM NV12 buffer (`mgrun offscreen mgl-yuvprobe`, exit 0 = pass). KWin's own NV12 path (it splits a YUV buffer into R8 + GR88 imports) is not served: MobileGL imports no single-channel dma-bufs, so KWin advertises no YUV format over linux-dmabuf.
 Device loss: `setprop debug.mobilegl.inject_device_lost_pid <pid>` (as root) loses the server session of the client whose pid - as it sees itself in the container - is `<pid>`, once per value, on any backend and whether or not it presents; `debug.mobilegl.inject_device_lost_at N` instead loses whichever session makes the N-th device check (Espryt presents and read-backs only). `bash scripts/chrome-recover.sh <in-process|process> [rounds]` checks that Chrome repaints after such a loss without a restart (battery-guarded, stops below 40%). `bash scripts/kwin-recover.sh [rounds]` loses KWin's own session (default twice) and checks that KWin restarts compositing on a fresh session - same KWin, plasmashell and client processes, the screen still animating (glmark2 on Wayland, glxgears on X11, `CHROME=1` adds Chrome) - which needs the anland backend + kwin.patch reset handling deployed (`kwin-stage.sh build`).
-Probes: `scripts/probes/texfmt.c` (texture formats keep alpha?), `dmahash.c` (hash a process's dma-bufs); build in the container, e.g. `gcc -O1 /root/texfmt.c -o /usr/local/bin/mgl-texfmt -lEGL`, run with `mgrun offscreen`.
+Probes: `scripts/probes/texfmt.c` (texture formats keep alpha?), `dmahash.c` (hash a process's dma-bufs), `yuvprobe.c` (YUV dma-buf sampling, `gcc -O1 /root/yuvprobe.c -o /usr/local/bin/mgl-yuvprobe -lEGL -lGLESv2 -lgbm -lm`), `cdpmedia.py` (which decoder a Chrome page uses); build in the container, e.g. `gcc -O1 /root/texfmt.c -o /usr/local/bin/mgl-texfmt -lEGL`, run with `mgrun offscreen`.
 GPU hangs: a shader that never ends stops the WHOLE GPU (every process, MobileGL or not) until the kernel resets it - ~2 s, or ~10 s ("Preemption Fault" in dmesg) when a higher-priority context waits; nothing in user space shortens one stall. What the server does is end the guilty session so it cannot hang the GPU again (Magma: event-bracketed submissions + `mgl-gpu-watch`, budget `MOBILEGL_GPU_HANG_BUDGET_MS` / `debug.mobilegl.gpu_hang_budget_ms`, default 2000; Espryt: the driver's GUILTY reset status, which it reports after the ~2 s reset only). Repro: `scripts/probes/hangclient.c` (Wayland + GL 3.3 core; `-H 60 -N 1 -E 60` hangs every 60 frames, `-H -1` is a frame-gap observer; build with wayland-scanner's xdg-shell files, `-lwayland-client -lwayland-egl -lEGL -lOpenGL -lm`), and `gpuprobe.c` (NDK, plain adb shell; `gpuprobe 30 high` is a high-priority GLES context outside MobileGL - it both measures the device-wide stall and forces the 10 s preemption path).
 
 A result counts only if ALL hold:

.claude/skills/anland-mobilegl-plasma/scripts/probes/cdpmedia.py

@@ -0,0 +1,94 @@
+#!/usr/bin/env python3
+# cdpmedia.py [port] [seconds] - ask a Chrome started with --remote-debugging-port=<port> (and a
+# non-default --user-data-dir) which video decoder its first page's players use, and evaluate
+# `window.__mglStats` / the element `#t` text of the page. Standard library only (no websocket
+# module in the container): a minimal RFC 6455 client.
+#   python3 cdpmedia.py 9333 6
+# Prints kVideoDecoderName / kIsPlatformVideoDecoder / kVideoTracks properties and media errors as
+# they arrive, then the page's stats text.
+import base64, json, os, socket, struct, sys, time, urllib.request
+
+port = int(sys.argv[1]) if len(sys.argv) > 1 else 9333
+seconds = float(sys.argv[2]) if len(sys.argv) > 2 else 6.0
+targets = json.load(urllib.request.urlopen(f"http://127.0.0.1:{port}/json"))
+page = next(t for t in targets if t.get("type") == "page")
+url = page["webSocketDebuggerUrl"]
+path = url.split(f"127.0.0.1:{port}", 1)[1]
+sock = socket.create_connection(("127.0.0.1", port))
+key = base64.b64encode(os.urandom(16)).decode()
+sock.sendall((f"GET {path} HTTP/1.1\r\nHost: 127.0.0.1:{port}\r\nUpgrade: websocket\r\nConnection: Upgrade\r\n"
+              f"Sec-WebSocket-Key: {key}\r\nSec-WebSocket-Version: 13\r\n\r\n").encode())
+response = b""
+while b"\r\n\r\n" not in response:
+    response += sock.recv(4096)
+buffer = response.split(b"\r\n\r\n", 1)[1]
+
+
+def send(message):
+    data = json.dumps(message).encode()
+    header = bytes([0x81])
+    mask = os.urandom(4)
+    if len(data) < 126:
+        header += bytes([0x80 | len(data)])
+    elif len(data) < 65536:
+        header += bytes([0x80 | 126]) + struct.pack(">H", len(data))
+    else:
+        header += bytes([0x80 | 127]) + struct.pack(">Q", len(data))
+    sock.sendall(header + mask + bytes(b ^ mask[i % 4] for i, b in enumerate(data)))
+
+
+def recv_exact(n):
+    global buffer
+    while len(buffer) < n:
+        chunk = sock.recv(65536)
+        if not chunk:
+            raise EOFError
+        buffer += chunk
+    out, buffer = buffer[:n], buffer[n:]
+    return out
+
+
+def recv():
+    first, second = recv_exact(2)
+    length = second & 0x7F
+    if length == 126:
+        length = struct.unpack(">H", recv_exact(2))[0]
+    elif length == 127:
+        length = struct.unpack(">Q", recv_exact(8))[0]
+    payload = recv_exact(length)
+    return json.loads(payload) if first & 0x0F == 1 else None
+
+
+send({"id": 1, "method": "Media.enable"})
+# Players created before the domain was enabled report nothing: reload so the page makes new ones.
+if os.environ.get("CDP_NAVIGATE"):
+    send({"id": 3, "method": "Page.navigate", "params": {"url": os.environ["CDP_NAVIGATE"]}})
+elif os.environ.get("CDP_RELOAD", "1") == "1":
+    send({"id": 3, "method": "Page.reload"})
+deadline = time.time() + seconds
+sock.settimeout(0.5)
+seen = set()
+while time.time() < deadline:
+    try:
+        message = recv()
+    except socket.timeout:
+        continue
+    if not message or "method" not in message:
+        continue
+    if message["method"] == "Media.playerPropertiesChanged":
+        for prop in message["params"]["properties"]:
+            if prop["name"] in ("kVideoDecoderName", "kIsPlatformVideoDecoder", "kVideoTracks", "kResolution"):
+                line = f"{prop['name']} = {prop['value']}"
+                if line not in seen:
+                    seen.add(line)
+                    print(line)
+    elif message["method"] in ("Media.playerErrorsRaised",):
+        print("media error:", json.dumps(message["params"])[:300])
+send({"id": 2, "method": "Runtime.evaluate",
+      "params": {"expression": os.environ.get("CDP_EXPR", "(document.getElementById('t')||{}).textContent"), "returnByValue": True}})
+sock.settimeout(3)
+while True:
+    message = recv()
+    if message and message.get("id") == 2:
+        print("page:", message.get("result", {}).get("result", {}).get("value"))
+        break

.claude/skills/anland-mobilegl-plasma/scripts/probes/yuvprobe.c

@@ -0,0 +1,410 @@
+// yuvprobe.c - pixel check of YUV dma-buf sampling through MobileGL (GL_OES_EGL_image_external).
+//
+//   gcc -O1 yuvprobe.c -o /usr/local/bin/mgl-yuvprobe -lEGL -lGLESv2 -lgbm
+//   mgrun offscreen mgl-yuvprobe            # exit 0 = every check passed
+//
+// What it does, in an ES 3.0 context on a pbuffer:
+//   1. FOREIGN NV12 (and P010): a dma-buf from /dev/dma_heap/system - a buffer the MobileGL server
+//      did not allocate, as a video decoder's would be - with padded pitches and a gap before the
+//      chroma plane, filled with four vertical bars (red, green, blue, white) in BT.709 narrow
+//      range. Imported with eglCreateImage(EGL_LINUX_DMA_BUF_EXT) and its colour hints, bound to
+//      GL_TEXTURE_EXTERNAL_OES, drawn through a samplerExternalOES shader into an RGBA8 FBO and read
+//      back: each bar must come out its colour. The buffer is then rewritten with the bars in
+//      reverse order and, after a native fence (what ends a frame for an offscreen producer), drawn
+//      again: the new order must show (the server refreshes a foreign buffer at each frame).
+//   2. GBM NV12: gbm_bo_create(NV12) through the mobilegl GBM backend (a server-allocated YUV
+//      buffer, zero-copy), its planes printed, imported and sampled once (content undefined; the
+//      check is that every call succeeds).
+// Also prints the dma-buf formats/modifiers EGL lists for NV12/P010.
+#define _GNU_SOURCE
+#include <EGL/egl.h>
+#include <EGL/eglext.h>
+#include <GLES3/gl3.h>
+#include <GLES2/gl2ext.h>
+#include <fcntl.h>
+#include <gbm.h>
+#include <linux/dma-buf.h>
+#include <linux/dma-heap.h>
+#include <math.h>
+#include <stdint.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <sys/ioctl.h>
+#include <sys/mman.h>
+#include <unistd.h>
+
+#ifndef DRM_FORMAT_NV12
+#define DRM_FORMAT_NV12 0x3231564E
+#endif
+#ifndef DRM_FORMAT_P010
+#define DRM_FORMAT_P010 0x30313050
+#endif
+
+static int g_failures = 0;
+#define CHECK(cond, ...)                                                                                              \
+    do {                                                                                                               \
+        if (!(cond)) {                                                                                                 \
+            printf("FAIL: " __VA_ARGS__);                                                                              \
+            printf("\n");                                                                                              \
+            ++g_failures;                                                                                              \
+        }                                                                                                              \
+    } while (0)
+
+static PFNEGLCREATEIMAGEKHRPROC pCreateImage;
+static PFNEGLDESTROYIMAGEKHRPROC pDestroyImage;
+static PFNGLEGLIMAGETARGETTEXTURE2DOESPROC pTargetTexture;
+static PFNEGLQUERYDMABUFFORMATSEXTPROC pQueryFormats;
+static PFNEGLQUERYDMABUFMODIFIERSEXTPROC pQueryModifiers;
+static PFNEGLCREATESYNCKHRPROC pCreateSync;
+static PFNEGLDESTROYSYNCKHRPROC pDestroySync;
+static EGLDisplay g_display;
+
+enum { W = 256, H = 128, BARS = 4 };
+static const unsigned char kBars[BARS][3] = {{255, 0, 0}, {0, 255, 0}, {0, 0, 255}, {255, 255, 255}};
+
+// RGB -> BT.709 narrow-range codes at `bits` (8 or 10).
+static void Encode709(const unsigned char rgb[3], int bits, int out[3]) {
+    const double kr = 0.2126, kb = 0.0722, kg = 1.0 - kr - kb;
+    const double r = rgb[0] / 255.0, g = rgb[1] / 255.0, b = rgb[2] / 255.0;
+    const double y = kr * r + kg * g + kb * b;
+    const double cb = (b - y) / (2.0 * (1.0 - kb)), cr = (r - y) / (2.0 * (1.0 - kr));
+    const double scale = (double)(1 << (bits - 8));
+    out[0] = (int)lround((16.0 + 219.0 * y) * scale);
+    out[1] = (int)lround((128.0 + 224.0 * cb) * scale);
+    out[2] = (int)lround((128.0 + 224.0 * cr) * scale);
+}
+
+// Writes the bars (`reverse`d or not) into a semi-planar buffer: 1 byte samples, or P010's 16-bit
+// samples with the 10 significant bits high.
+static void FillBars(uint8_t* base, uint32_t pitchY, uint32_t offsetC, uint32_t pitchC, int bytes, int reverse) {
+    const int bits = bytes == 1 ? 8 : 10;
+    for (int y = 0; y < H; ++y) {
+        for (int x = 0; x < W; ++x) {
+            int bar = x * BARS / W;
+            if (reverse) bar = BARS - 1 - bar;
+            int code[3];
+            Encode709(kBars[bar], bits, code);
+            uint8_t* py = base + (size_t)y * pitchY + (size_t)x * bytes;
+            if (bytes == 1) py[0] = (uint8_t)code[0];
+            else *(uint16_t*)py = (uint16_t)(code[0] << 6);
+            if ((x & 1) == 0 && (y & 1) == 0) {
+                uint8_t* pc = base + offsetC + (size_t)(y / 2) * pitchC + (size_t)(x / 2) * 2 * bytes;
+                if (bytes == 1) {
+                    pc[0] = (uint8_t)code[1];
+                    pc[1] = (uint8_t)code[2];
+                } else {
+                    ((uint16_t*)pc)[0] = (uint16_t)(code[1] << 6);
+                    ((uint16_t*)pc)[1] = (uint16_t)(code[2] << 6);
+                }
+            }
+        }
+    }
+}
+
+static GLuint Compile(GLenum type, const char* source) {
+    GLuint shader = glCreateShader(type);
+    glShaderSource(shader, 1, &source, NULL);
+    glCompileShader(shader);
+    GLint ok = 0;
+    glGetShaderiv(shader, GL_COMPILE_STATUS, &ok);
+    if (!ok) {
+        char log[2048];
+        glGetShaderInfoLog(shader, sizeof(log), NULL, log);
+        printf("shader: %s\n", log);
+    }
+    return shader;
+}
+
+static GLuint g_program, g_fbo, g_rgba;
+
+static void SetUpDraw(void) {
+    const char* vs = "#version 300 es\n"
+                     "out vec2 uv;\n"
+                     "void main() {\n"
+                     "  vec2 p = vec2(float((gl_VertexID << 1) & 2), float(gl_VertexID & 2));\n"
+                     "  gl_Position = vec4(p * 2.0 - 1.0, 0.0, 1.0);\n"
+                     "  uv = p;\n"
+                     "}\n";
+    const char* fs = "#version 300 es\n"
+                     "#extension GL_OES_EGL_image_external_essl3 : require\n"
+                     "precision mediump float;\n"
+                     "uniform samplerExternalOES tex;\n"
+                     "in vec2 uv;\n"
+                     "out vec4 color;\n"
+                     "void main() { color = texture(tex, uv); }\n";
+    g_program = glCreateProgram();
+    glAttachShader(g_program, Compile(GL_VERTEX_SHADER, vs));
+    glAttachShader(g_program, Compile(GL_FRAGMENT_SHADER, fs));
+    glLinkProgram(g_program);
+    GLint linked = 0;
+    glGetProgramiv(g_program, GL_LINK_STATUS, &linked);
+    CHECK(linked, "the samplerExternalOES program did not link");
+    GLint type = 0, size = 0;
+    char name[64];
+    glGetActiveUniform(g_program, 0, sizeof(name), NULL, &size, (GLenum*)&type, name);
+    printf("uniform 0: %s type 0x%04x\n", name, type);
+    CHECK(type == GL_SAMPLER_EXTERNAL_OES, "the sampler reports type 0x%04x, not GL_SAMPLER_EXTERNAL_OES", type);
+    glGenTextures(1, &g_rgba);
+    glBindTexture(GL_TEXTURE_2D, g_rgba);
+    glTexStorage2D(GL_TEXTURE_2D, 1, GL_RGBA8, W, H);
+    glGenFramebuffers(1, &g_fbo);
+    glBindFramebuffer(GL_FRAMEBUFFER, g_fbo);
+    glFramebufferTexture2D(GL_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_TEXTURE_2D, g_rgba, 0);
+    CHECK(glCheckFramebufferStatus(GL_FRAMEBUFFER) == GL_FRAMEBUFFER_COMPLETE, "the readback FBO is incomplete");
+    GLuint vao;
+    glGenVertexArrays(1, &vao);
+    glBindVertexArray(vao);
+}
+
+// Draws the external texture `texture` and checks each bar's center (row `y` of the image).
+static void DrawAndCheck(GLuint texture, const char* what, int reverse, int tolerance) {
+    glBindFramebuffer(GL_FRAMEBUFFER, g_fbo);
+    glViewport(0, 0, W, H);
+    glUseProgram(g_program);
+    glActiveTexture(GL_TEXTURE0);
+    glBindTexture(GL_TEXTURE_EXTERNAL_OES, texture);
+    glUniform1i(glGetUniformLocation(g_program, "tex"), 0);
+    glDrawArrays(GL_TRIANGLES, 0, 3);
+    static unsigned char pixels[W * H * 4];
+    glReadPixels(0, 0, W, H, GL_RGBA, GL_UNSIGNED_BYTE, pixels);
+    GLenum error = glGetError();
+    CHECK(error == GL_NO_ERROR, "%s: GL error 0x%04x", what, error);
+    if (tolerance < 0) {
+        printf("%s: center pixel %d %d %d %d (content undefined)\n", what, pixels[(H / 2 * W + W / 2) * 4],
+               pixels[(H / 2 * W + W / 2) * 4 + 1], pixels[(H / 2 * W + W / 2) * 4 + 2],
+               pixels[(H / 2 * W + W / 2) * 4 + 3]);
+        return;
+    }
+    // Rows: GL row 0 is the image's first row (the uv mapping above), and the bars are vertical,
+    // so any row does; take two to catch a broken chroma row pitch.
+    const int rows[2] = {H / 4, (3 * H) / 4};
+    for (int r = 0; r < 2; ++r) {
+        for (int bar = 0; bar < BARS; ++bar) {
+            const int x = bar * (W / BARS) + W / (2 * BARS);
+            const unsigned char* p = &pixels[(rows[r] * W + x) * 4];
+            const unsigned char* want = kBars[reverse ? BARS - 1 - bar : bar];
+            int worst = 0;
+            for (int c = 0; c < 3; ++c) {
+                int d = abs((int)p[c] - (int)want[c]);
+                if (d > worst) worst = d;
+            }
+            printf("%s: row %d bar %d -> %3d %3d %3d %3d (want %3d %3d %3d)%s\n", what, rows[r], bar, p[0], p[1], p[2],
+                   p[3], want[0], want[1], want[2], worst > tolerance ? "  <-- off" : "");
+            CHECK(worst <= tolerance, "%s: bar %d is off by %d", what, bar, worst);
+        }
+    }
+}
+
+static int AllocateDmaHeap(size_t size) {
+    int heap = open("/dev/dma_heap/system", O_RDWR | O_CLOEXEC);
+    if (heap < 0) {
+        perror("open /dev/dma_heap/system");
+        return -1;
+    }
+    struct dma_heap_allocation_data data = {.len = size, .fd_flags = O_RDWR | O_CLOEXEC};
+    if (ioctl(heap, DMA_HEAP_IOCTL_ALLOC, &data) != 0) {
+        perror("DMA_HEAP_IOCTL_ALLOC");
+        close(heap);
+        return -1;
+    }
+    close(heap);
+    return (int)data.fd;
+}
+
+static void Sync(int fd, int start) {
+    struct dma_buf_sync sync = {.flags = DMA_BUF_SYNC_WRITE | (start ? DMA_BUF_SYNC_START : DMA_BUF_SYNC_END)};
+    ioctl(fd, DMA_BUF_IOCTL_SYNC, &sync);
+}
+
+static EGLImageKHR Import(int fd, uint32_t fourcc, uint32_t pitchY, uint32_t offsetC, uint32_t pitchC, int modifier) {
+    EGLint attribs[64];
+    int n = 0;
+#define A(k, v) (attribs[n++] = (k), attribs[n++] = (EGLint)(v))
+    A(EGL_WIDTH, W);
+    A(EGL_HEIGHT, H);
+    A(EGL_LINUX_DRM_FOURCC_EXT, fourcc);
+    A(EGL_DMA_BUF_PLANE0_FD_EXT, fd);
+    A(EGL_DMA_BUF_PLANE0_OFFSET_EXT, 0);
+    A(EGL_DMA_BUF_PLANE0_PITCH_EXT, pitchY);
+    A(EGL_DMA_BUF_PLANE1_FD_EXT, fd);
+    A(EGL_DMA_BUF_PLANE1_OFFSET_EXT, offsetC);
+    A(EGL_DMA_BUF_PLANE1_PITCH_EXT, pitchC);
+    if (modifier) {
+        A(EGL_DMA_BUF_PLANE0_MODIFIER_LO_EXT, 0);
+        A(EGL_DMA_BUF_PLANE0_MODIFIER_HI_EXT, 0);
+        A(EGL_DMA_BUF_PLANE1_MODIFIER_LO_EXT, 0);
+        A(EGL_DMA_BUF_PLANE1_MODIFIER_HI_EXT, 0);
+    }
+    A(EGL_YUV_COLOR_SPACE_HINT_EXT, EGL_ITU_REC709_EXT);
+    A(EGL_SAMPLE_RANGE_HINT_EXT, EGL_YUV_NARROW_RANGE_EXT);
+    attribs[n] = EGL_NONE;
+#undef A
+    EGLImageKHR image = pCreateImage(g_display, EGL_NO_CONTEXT, EGL_LINUX_DMA_BUF_EXT, NULL, attribs);
+    if (image == EGL_NO_IMAGE_KHR) printf("eglCreateImage failed: 0x%04x\n", eglGetError());
+    return image;
+}
+
+static GLuint ExternalTexture(EGLImageKHR image) {
+    GLuint texture;
+    glGenTextures(1, &texture);
+    glBindTexture(GL_TEXTURE_EXTERNAL_OES, texture);
+    pTargetTexture(GL_TEXTURE_EXTERNAL_OES, image);
+    GLenum error = glGetError();
+    CHECK(error == GL_NO_ERROR, "glEGLImageTargetTexture2DOES(GL_TEXTURE_EXTERNAL_OES) -> 0x%04x", error);
+    GLint bound = 0;
+    glGetIntegerv(GL_TEXTURE_BINDING_EXTERNAL_OES, &bound);
+    CHECK((GLuint)bound == texture, "GL_TEXTURE_BINDING_EXTERNAL_OES reads %d", bound);
+    return texture;
+}
+
+// A frame boundary as an offscreen producer makes one: a native fence.
+static void EndFrame(void) {
+    EGLint attribs[] = {EGL_NONE};
+    EGLSyncKHR sync = pCreateSync(g_display, EGL_SYNC_NATIVE_FENCE_ANDROID, attribs);
+    CHECK(sync != EGL_NO_SYNC_KHR, "eglCreateSync(native fence) failed");
+    if (sync != EGL_NO_SYNC_KHR) pDestroySync(g_display, sync);
+}
+
+static void ForeignTest(uint32_t fourcc, int bytes, int tolerance) {
+    const char* what = fourcc == DRM_FORMAT_NV12 ? "foreign NV12" : "foreign P010";
+    // Padded pitches and a gap before the chroma plane, as decoders lay buffers out.
+    const uint32_t pitchY = W * bytes + 64, pitchC = W * bytes + 128;
+    const uint32_t offsetC = pitchY * (H + 16);
+    const size_t size = offsetC + (size_t)pitchC * (H / 2) + 4096;
+    int fd = AllocateDmaHeap(size);
+    if (fd < 0) {
+        CHECK(0, "%s: no dma-buf to test with", what);
+        return;
+    }
+    uint8_t* map = mmap(NULL, size, PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0);
+    CHECK(map != MAP_FAILED, "%s: mmap failed", what);
+    if (map == MAP_FAILED) return;
+    Sync(fd, 1);
+    FillBars(map, pitchY, offsetC, pitchC, bytes, 0);
+    Sync(fd, 0);
+    EGLImageKHR image = Import(fd, fourcc, pitchY, offsetC, pitchC, fourcc == DRM_FORMAT_NV12);
+    CHECK(image != EGL_NO_IMAGE_KHR, "%s: import refused", what);
+    if (image == EGL_NO_IMAGE_KHR) return;
+    GLuint texture = ExternalTexture(image);
+    DrawAndCheck(texture, what, 0, tolerance);
+    EndFrame();
+    // New content, same buffer: shows after the frame boundary.
+    Sync(fd, 1);
+    FillBars(map, pitchY, offsetC, pitchC, bytes, 1);
+    Sync(fd, 0);
+    EndFrame();
+    DrawAndCheck(texture, what, 1, tolerance);
+    glDeleteTextures(1, &texture);
+    pDestroyImage(g_display, image);
+    munmap(map, size);
+    close(fd);
+}
+
+static void GbmTest(void) {
+    int node = open("/dev/dri/renderD128", O_RDWR | O_CLOEXEC);
+    if (node < 0) node = open("/dev/null", O_RDWR | O_CLOEXEC);
+    struct gbm_device* gbm = gbm_create_device(node);
+    if (gbm == NULL) {
+        CHECK(0, "gbm_create_device failed");
+        return;
+    }
+    printf("gbm backend: %s\n", gbm_device_get_backend_name(gbm));
+    const int supported = gbm_device_is_format_supported(gbm, DRM_FORMAT_NV12, 0);
+    printf("gbm NV12 supported: %d, P010: %d\n", supported,
+           gbm_device_is_format_supported(gbm, DRM_FORMAT_P010, 0));
+    CHECK(supported == 1, "gbm does not support NV12");
+    struct gbm_bo* bo = gbm_bo_create(gbm, W, H, DRM_FORMAT_NV12, 0);
+    CHECK(bo != NULL, "gbm_bo_create(NV12) failed");
+    if (bo != NULL) {
+        const int planes = gbm_bo_get_plane_count(bo);
+        printf("gbm NV12 bo: planes %d, stride %u/%u, offset %u/%u, modifier 0x%llx\n", planes,
+               gbm_bo_get_stride_for_plane(bo, 0), gbm_bo_get_stride_for_plane(bo, 1), gbm_bo_get_offset(bo, 0),
+               gbm_bo_get_offset(bo, 1), (unsigned long long)gbm_bo_get_modifier(bo));
+        CHECK(planes == 2, "the NV12 bo has %d planes", planes);
+        int fd = gbm_bo_get_fd(bo);
+        EGLImageKHR image =
+            Import(fd, DRM_FORMAT_NV12, gbm_bo_get_stride_for_plane(bo, 0), gbm_bo_get_offset(bo, 1),
+                   gbm_bo_get_stride_for_plane(bo, 1), 0);
+        CHECK(image != EGL_NO_IMAGE_KHR, "the server's own NV12 buffer was not imported");
+        if (image != EGL_NO_IMAGE_KHR) {
+            GLuint texture = ExternalTexture(image);
+            DrawAndCheck(texture, "gbm NV12", 0, -1);
+            glDeleteTextures(1, &texture);
+            pDestroyImage(g_display, image);
+        }
+        close(fd);
+        gbm_bo_destroy(bo);
+    }
+    gbm_device_destroy(gbm);
+    close(node);
+}
+
+int main(void) {
+    setvbuf(stdout, NULL, _IONBF, 0);
+    g_display = eglGetDisplay(EGL_DEFAULT_DISPLAY);
+    EGLint major, minor;
+    if (!eglInitialize(g_display, &major, &minor)) {
+        printf("FAIL: eglInitialize\n");
+        return 2;
+    }
+    const char* eglExtensions = eglQueryString(g_display, EGL_EXTENSIONS);
+    printf("EGL %d.%d dma_buf_import: %s\n", major, minor,
+           strstr(eglExtensions, "EGL_EXT_image_dma_buf_import") ? "yes" : "no");
+    eglBindAPI(EGL_OPENGL_ES_API);
+    const EGLint configAttribs[] = {EGL_SURFACE_TYPE, EGL_PBUFFER_BIT, EGL_RENDERABLE_TYPE, EGL_OPENGL_ES3_BIT,
+                                    EGL_RED_SIZE, 8, EGL_GREEN_SIZE, 8, EGL_BLUE_SIZE, 8, EGL_NONE};
+    EGLConfig config;
+    EGLint count = 0;
+    eglChooseConfig(g_display, configAttribs, &config, 1, &count);
+    if (count < 1) {
+        printf("FAIL: no config\n");
+        return 2;
+    }
+    const EGLint pbufferAttribs[] = {EGL_WIDTH, 16, EGL_HEIGHT, 16, EGL_NONE};
+    EGLSurface surface = eglCreatePbufferSurface(g_display, config, pbufferAttribs);
+    const EGLint contextAttribs[] = {EGL_CONTEXT_MAJOR_VERSION, 3, EGL_NONE};
+    EGLContext context = eglCreateContext(g_display, config, EGL_NO_CONTEXT, contextAttribs);
+    if (context == EGL_NO_CONTEXT || !eglMakeCurrent(g_display, surface, surface, context)) {
+        printf("FAIL: no ES 3 context\n");
+        return 2;
+    }
+    pCreateImage = (PFNEGLCREATEIMAGEKHRPROC)eglGetProcAddress("eglCreateImageKHR");
+    pDestroyImage = (PFNEGLDESTROYIMAGEKHRPROC)eglGetProcAddress("eglDestroyImageKHR");
+    pTargetTexture = (PFNGLEGLIMAGETARGETTEXTURE2DOESPROC)eglGetProcAddress("glEGLImageTargetTexture2DOES");
+    pQueryFormats = (PFNEGLQUERYDMABUFFORMATSEXTPROC)eglGetProcAddress("eglQueryDmaBufFormatsEXT");
+    pQueryModifiers = (PFNEGLQUERYDMABUFMODIFIERSEXTPROC)eglGetProcAddress("eglQueryDmaBufModifiersEXT");
+    pCreateSync = (PFNEGLCREATESYNCKHRPROC)eglGetProcAddress("eglCreateSyncKHR");
+    pDestroySync = (PFNEGLDESTROYSYNCKHRPROC)eglGetProcAddress("eglDestroySyncKHR");
+    printf("GL_RENDERER: %s\n", glGetString(GL_RENDERER));
+    const char* glExtensions = (const char*)glGetString(GL_EXTENSIONS);
+    const int external = strstr(glExtensions, "GL_OES_EGL_image_external_essl3") != NULL;
+    printf("GL_OES_EGL_image_external_essl3: %s\n", external ? "yes" : "no");
+    CHECK(external, "GL_OES_EGL_image_external_essl3 is not advertised");
+
+    EGLint formats[32], formatCount = 0;
+    pQueryFormats(g_display, 32, formats, &formatCount);
+    int nv12 = 0;
+    for (int i = 0; i < formatCount; ++i) {
+        printf("dma-buf format %.4s", (const char*)&formats[i]);
+        EGLuint64KHR modifiers[4];
+        EGLBoolean externalOnly[4];
+        EGLint modifierCount = 0;
+        pQueryModifiers(g_display, formats[i], 4, modifiers, externalOnly, &modifierCount);
+        for (int m = 0; m < modifierCount; ++m)
+            printf(" [modifier 0x%llx%s]", (unsigned long long)modifiers[m], externalOnly[m] ? " external-only" : "");
+        printf("\n");
+        nv12 |= formats[i] == (EGLint)DRM_FORMAT_NV12;
+    }
+    CHECK(nv12, "NV12 is not a listed dma-buf format");
+
+    SetUpDraw();
+    ForeignTest(DRM_FORMAT_NV12, 1, 12);
+    ForeignTest(DRM_FORMAT_P010, 2, 16);
+    GbmTest();
+
+    printf("%s (%d failure(s))\n", g_failures ? "FAILED" : "PASSED", g_failures);
+    eglMakeCurrent(g_display, EGL_NO_SURFACE, EGL_NO_SURFACE, EGL_NO_CONTEXT);
+    eglTerminate(g_display);
+    return g_failures ? 1 : 0;
+}

README.md

@@ -34,7 +34,7 @@ The repository is organized into following top-level modules:
 2. **MG_Impl** — front-end implementations of Graphics APIs that interact with `MG_State` and `MG_Backend`: OpenGL, plus the window-system APIs EGL, GLX, WGL, CGL and NSOpenGL.
 3. **MG_Backend** — per-backend translation layer that maps front-end Graphics APIs' semantics and state into concrete backend API calls (e.g. OpenGL ES, Vulkan).
 4. **MG_Pipe** and **MG_Remote** — the command stream between front-end and backend, and the client/server transport that lets the backend run in a separate process (see `docs/Disaggregated`).
-5. **MG_Gbm** — a GBM backend whose buffers are images shared with the MobileGL server, so Wayland clients and compositors can exchange them as dma-bufs.
+5. **MG_Gbm** — a GBM backend whose buffers are images shared with the MobileGL server, so Wayland clients and compositors can exchange them as dma-bufs (RGBA, and NV12/P010 YUV where the server's platform allocates them). YUV dma-bufs - the server's own or a video decoder's - import through `EGL_EXT_image_dma_buf_import` and are sampled through `GL_OES_EGL_image_external(_essl3)`: the backend converts the image into the texture's RGBA level at its first use in each frame (zero-copy for a server image; a foreign buffer, which the platform cannot import, is first copied into one by the CPU).
 6. **MG_Util** and other utility modules.
 
 ## Third-party components
@@ -173,6 +173,7 @@ MobileGL supports runtime configuration via environment variables.
 | `MOBILEGL_ESPRYT_AVOID_SAMPLER_MIPMAP_MIN_FILTER` | Avoid sampler mipmap minification filters. | `0`, `1` | `0` |
 | `MOBILEGL_COHERENT_AS_FLUSH` | Treat persistent `GL_MAP_FLUSH_EXPLICIT_BIT` maps as coherent (app-compat for engines like Flywheel that never flush them). | `0`, `1` | `0` |
 | `MOBILEGL_ESPRYT_FORCE_DS_READBACK_EMULATION` | Always emulate depth/stencil `glReadPixels`/`glGetTexImage` by shader sampling on Espryt, instead of using the driver's own depth/stencil readback where it has one. | `0`, `1` | `0` |
+| `MOBILEGL_YUV_DRIVER_CONVERSION` | Espryt: convert YUV images with the driver's external-sampler conversion (its own colour space) instead of `GL_EXT_YUV_target` with the import's `EGL_YUV_COLOR_SPACE_HINT_EXT` / `EGL_SAMPLE_RANGE_HINT_EXT`. | `0`, `1` | `0` |
 | `MOBILEGL_SERVER_FRAME_STATS` | Log one `P65ServerFrame` timing line per presented frame on the server (apply vs. wall time, socket read rate). On Android also `debug.mobilegl.server_frame_stats`. | `0`, `1` | `0` |
 | `VK_ICD_FILENAMES`      | Select the Vulkan ICD used by the Vulkan loader. | Path to an ICD JSON file             | Loader default |
 

docs/Disaggregated/notes/anland/runbook-plasma.md

@@ -124,6 +124,8 @@ sleep 8; grep -a "presented through" /tmp/mgl-run.client.log
 
 Chrome：`bash scripts/chrome-launch.sh <tag> [参数]`（日志 `/tmp/mgl-chrome-<tag>.client.log`，截图 `ch-<tag>.png`）。探针：`scripts/probes/texfmt.c`（各纹理格式是否保住 alpha）、`dmahash.c`（对某进程的 dma-buf 取样哈希）；放进容器 `/root` 编，例如 `gcc -O1 /root/texfmt.c -o /usr/local/bin/mgl-texfmt -lEGL`，用 `mgrun offscreen` 跑（之前的确切编译参数未验证）。
 
+视频：Chrome 走硬解（VaapiVideoDecoder）。容器的 VA-API 驱动（`msm_drm_drv_video.so`，V4L2 解码）自己分配 surface 并导出 NV12 dma-buf（单 fd，R8 + GR88 两层），Chrome 以 NV12/P010 EGLImage 绑到 `GL_TEXTURE_EXTERNAL_OES`。这种**外来** YUV dma-buf 平台导入不了，MobileGL 走明确标注的 CPU 拷贝兜底：server 映射它、每帧首次使用时拷进自己的 YUV AHB（`AHardwareBuffer_lockPlanes`），再在 GPU 上转换进纹理的 RGBA 层（Espryt：`GL_EXT_YUV_target` + 导入时的色彩 hint，`MOBILEGL_YUV_DRIVER_CONVERSION=1` 改用驱动的 external sampler；Magma：AHB external format 上的 `VkSamplerYcbcrConversion`）。server 自己分配的 NV12/P010（GBM）零拷贝。`MOBILEGL_CHROME_VIDEO_DECODE=software` 恢复 `--disable-accelerated-video-decode`。检查：容器里 `python3 scripts/probes/cdpmedia.py <port> <秒>`（CDP `Media.enable`，`kVideoDecoderName`），`scripts/probes/yuvprobe.c` 像素校验外来 NV12/P010 导入、逐帧刷新和 GBM NV12。KWin 自己的 NV12 路径（拆成 R8 + GR88 两个导入）不支持，KWin 不通告 YUV 格式。
+
 结论成立要同时满足：
 
 - client 日志是 `presented through linux-dmabuf shared images`，不是 `wl_shm`；出现 `could not be presented through a linux-dmabuf shared image` 说明退回了；