vgsp: all-ones scrub release for registered waits; walk-page registration; GL frame publisher (CONT.416-417)

- gsp_sema_service: merge a scanner registration arriving after the
  stash record instead of dropping it (a dropped registration let the
  next authoring pass write the blocking pair into the page the guest
  was polling, so the wait never ended); author the finish word
  all-ones for every registered record while the sema word stays
  put-based once real submissions were observed (CONT.416/417).
- scrub_pin_qmp (host scanner): accept an unstable-but-live wait page
  instead of skipping it, and register the walk-resolved scrubber
  channel page - first the walk summary, then the per-VAS kernel row
  (h=0xc) when the summary path fails - so the release is armed before
  the guest polls (r1762/r1764 evidence).
- guest-diag: add gl_frames, the direct-GL (GLX pbuffer) frame
  publisher; with it the stack produced the first non-black
  GPU-rendered frames (40/40, 640x480, animated) through
  pbuffer -> sysmem -> ivshmem -> host frame_reader.
- gsp_rm_emulation: log the display-shim request words for triage.

1 files changed, 237 insertions, 0 deletions

diff --git a/tools/guest-diag/gl_frames.c b/tools/guest-diag/gl_frames.c
new file mode 100644
index 0000000..d442d66
--- /dev/null
+++ b/tools/guest-diag/gl_frames.c
@@ -0,0 +1,237 @@
+// SPDX-License-Identifier: BSD-3-Clause
+/*
+ * gl_frames.c - render animated frames with direct GL (GLX pbuffer) on the
+ * guest's NVIDIA vGPU and publish them into the ivshmem (kvmfr) BAR so the
+ * host can see what the GL client actually rendered.
+ *
+ * This is the GL sibling of tools/cuda-ab/frame_producer.cu: the transport
+ * (guest GPU -> guest sysmem -> ivshmem -> host frame_reader.py) is identical,
+ * but the pixels come from the direct-GL client channel instead of CUDA.
+ * The pbuffer avoids the X window draw path that wedges the X server.
+ *
+ * Layout at BAR offset 0x100000 (same as frame_producer.cu):
+ *   struct frame_hdr { magic, version, width, height, stride, format, seq }
+ *   then width*height*4 bytes of RGBA8.
+ *
+ * Usage: sudo ./gl_frames [frames] [width] [height] [fps]
+ * Prints: FRAME seq=<n> ms=<render+readback ms>
+ */
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <stdint.h>
+#include <time.h>
+#include <fcntl.h>
+#include <unistd.h>
+#include <sys/mman.h>
+#include <sys/stat.h>
+#include <glob.h>
+#include <math.h>
+#include <X11/Xlib.h>
+#include <GL/glx.h>
+#include <GL/gl.h>
+
+#define FRAME_OFFSET  0x100000UL
+#define HDR_MAGIC     0x4b564d46u	/* "FMVK" little-endian-ish */
+#define BAR_BYTES     (128UL << 20)
+#define BAR_PATH_DEF  "/sys/bus/pci/devices/0000:00:04.0/resource2"
+
+struct frame_hdr {
+	uint32_t magic;
+	uint32_t version;
+	uint32_t width;
+	uint32_t height;
+	uint32_t stride;
+	uint32_t format;	/* 0 = RGBA8 */
+	uint32_t seq;
+	uint32_t reserved;
+	uint64_t timestamp_ns;
+};
+
+/*
+ * The ivshmem kvmfr device moves with the QEMU topology (it was 00:03.0 in an
+ * older launch, 00:04.0 in the current one).  Discover the 128 MiB BAR instead
+ * of hardcoding a slot; GL_FRAMES_BAR overrides for debugging.
+ */
+static const char *find_bar(char *buf, size_t len)
+{
+	glob_t g;
+	size_t i;
+	const char *hit = NULL;
+
+	if (glob("/sys/bus/pci/devices/*/resource2", 0, NULL, &g) != 0)
+		return NULL;
+	for (i = 0; i < g.gl_pathc; i++) {
+		struct stat st;
+
+		if (stat(g.gl_pathv[i], &st) == 0 &&
+		    (unsigned long long)st.st_size == BAR_BYTES) {
+			snprintf(buf, len, "%s", g.gl_pathv[i]);
+			hit = buf;
+			break;
+		}
+	}
+	globfree(&g);
+	return hit;
+}
+
+static uint64_t now_ns(void)
+{
+	struct timespec ts;
+
+	clock_gettime(CLOCK_MONOTONIC, &ts);
+	return (uint64_t)ts.tv_sec * 1000000000ull + ts.tv_nsec;
+}
+
+static void draw_frame(int f, int w, int h)
+{
+	float a = (float)f * 0.15f;
+	int i;
+
+	glViewport(0, 0, w, h);
+	glClearColor(0.05f, 0.05f, 0.10f, 1.0f);
+	glClear(GL_COLOR_BUFFER_BIT);
+
+	glMatrixMode(GL_PROJECTION);
+	glLoadIdentity();
+	glOrtho(-1.0, 1.0, -1.0, 1.0, -1.0, 1.0);
+	glMatrixMode(GL_MODELVIEW);
+	glLoadIdentity();
+
+	for (i = 0; i < 8; i++) {
+		float t = a + (float)i * 0.78539816339f;
+
+		glBegin(GL_TRIANGLES);
+		glColor3f(1.0f, 0.2f + 0.09f * i, 0.1f);
+		glVertex2f(0.0f, 0.0f);
+		glColor3f(0.1f, 0.9f, 0.3f);
+		glVertex2f(0.75f * cosf(t), 0.75f * sinf(t));
+		glColor3f(0.2f, 0.3f, 1.0f);
+		glVertex2f(0.55f * cosf(t + 0.5f),
+			   0.55f * sinf(t + 0.5f));
+		glEnd();
+	}
+	glFinish();
+}
+
+int main(int argc, char **argv)
+{
+	int frames = argc > 1 ? atoi(argv[1]) : 30;
+	int w = argc > 2 ? atoi(argv[2]) : 640;
+	int h = argc > 3 ? atoi(argv[3]) : 480;
+	int fps = argc > 4 ? atoi(argv[4]) : 5;
+	static int attribs[] = {
+		GLX_DRAWABLE_TYPE, GLX_PBUFFER_BIT,
+		GLX_RENDER_TYPE, GLX_RGBA_BIT,
+		GLX_RED_SIZE, 8, GLX_GREEN_SIZE, 8, GLX_BLUE_SIZE, 8,
+		GLX_ALPHA_SIZE, 8,
+		None
+	};
+	GLXFBConfig *fb;
+	GLXPbuffer pb;
+	XVisualInfo *vi;
+	GLXContext ctx;
+	Display *d;
+	int scr, n = 0, fd, i;
+	unsigned char *bar, *px;
+	struct frame_hdr *hdr;
+	volatile uint32_t *seqp;
+
+	char found[256];
+	const char *bar_path = getenv("GL_FRAMES_BAR");
+
+	setvbuf(stdout, NULL, _IONBF, 0);
+	if (!bar_path)
+		bar_path = find_bar(found, sizeof(found));
+	if (!bar_path)
+		bar_path = BAR_PATH_DEF;
+	printf("BAR %s\n", bar_path);
+	fd = open(bar_path, O_RDWR | O_SYNC);
+	if (fd < 0) {
+		printf("OPEN_FAIL %s\n", bar_path);
+		return 1;
+	}
+	bar = (unsigned char *)mmap(NULL, BAR_BYTES, PROT_READ | PROT_WRITE,
+				    MAP_SHARED, fd, 0);
+	if (bar == MAP_FAILED) {
+		printf("MMAP_FAIL\n");
+		return 1;
+	}
+	hdr = (struct frame_hdr *)(bar + FRAME_OFFSET);
+	seqp = &hdr->seq;
+
+	d = XOpenDisplay(NULL);
+	if (!d) {
+		printf("NODISPLAY\n");
+		return 2;
+	}
+	scr = DefaultScreen(d);
+	fb = glXChooseFBConfig(d, scr, attribs, &n);
+	if (!fb || n < 1) {
+		printf("NOFBCONFIG\n");
+		return 3;
+	}
+	pb = glXCreatePbuffer(d, fb[0], (int[]){
+		GLX_PBUFFER_WIDTH, w, GLX_PBUFFER_HEIGHT, h, None });
+	if (!pb) {
+		printf("NOPBUFFER\n");
+		return 4;
+	}
+	vi = glXGetVisualFromFBConfig(d, fb[0]);
+	ctx = glXCreateContext(d, vi, NULL, True);
+	if (!ctx) {
+		printf("NOCTX\n");
+		return 5;
+	}
+	if (!glXMakeCurrent(d, pb, ctx)) {
+		printf("MAKECURRENT_FAIL\n");
+		return 6;
+	}
+	printf("GL %s | %s\n", (const char *)glGetString(GL_VENDOR),
+	       (const char *)glGetString(GL_RENDERER));
+
+	px = (unsigned char *)malloc((size_t)w * h * 4);
+	if (!px) {
+		printf("HOST_ALLOC_FAIL\n");
+		return 7;
+	}
+
+	hdr->magic = HDR_MAGIC;
+	hdr->version = 1;
+	hdr->width = (uint32_t)w;
+	hdr->height = (uint32_t)h;
+	hdr->stride = (uint32_t)w * 4;
+	hdr->format = 0;
+
+	for (i = 0; i < frames; i++) {
+		uint64_t t0 = now_ns();
+		uint32_t seq;
+
+		draw_frame(i, w, h);
+		glReadPixels(0, 0, w, h, GL_RGBA, GL_UNSIGNED_BYTE, px);
+		if (glGetError() != GL_NO_ERROR)
+			printf("GL_ERROR frame=%d\n", i);
+		memcpy(bar + FRAME_OFFSET + sizeof(struct frame_hdr), px,
+		       (size_t)w * h * 4);
+		hdr->timestamp_ns = now_ns();
+		__sync_synchronize();
+		seq = (uint32_t)(i + 1);
+		*seqp = seq;
+		__sync_synchronize();
+		printf("FRAME seq=%u ms=%.1f px0=%02x%02x%02x%02x\n", seq,
+		       (double)(now_ns() - t0) / 1e6,
+		       px[0], px[1], px[2], px[3]);
+		if (fps > 0)
+			usleep((useconds_t)(1000000 / fps));
+	}
+	printf("FRAME_DONE frames=%d %dx%d\n", frames, w, h);
+
+	glXMakeCurrent(d, None, NULL);
+	glXDestroyContext(d, ctx);
+	glXDestroyPbuffer(d, pb);
+	XCloseDisplay(d);
+	free(px);
+	munmap(bar, BAR_BYTES);
+	close(fd);
+	return 0;
+}