gsp_sema_service: add content rescue for unmapped push-buffer pool VAs

- Add expected_sem_payload tracking to svc_channel for CE channels
- Implement svc_rescue_push_page() to scan guest RAM for push with
  expected semaphore release when VA translation fails
- Add svc_init_expected_payload() to initialize expected payload from pool
- Add rate-limited logging of distinct stalled VAs (svc_log_stalled_va)
- Advance expected_sem_payload on successful push decode
- Pushes SEMA_MAP_MAX to 65536, SVC_PIN_SLOTS to 32768 for scale

1 files changed, 105 insertions, 0 deletions

diff --git a/tools/guest-diag/scan.c b/tools/guest-diag/scan.c
new file mode 100644
index 0000000..d229dc8
--- /dev/null
+++ b/tools/guest-diag/scan.c
@@ -0,0 +1,105 @@
+/* scan.c - find the libcuda tracker objects in a process without ptrace.
+   Tracker layout (object p): p+0x9418 submitted, p+0x9420 completed,
+   p+0x9428 mirror, p+0x9430 sem object; sem object +0x10 -> sem VA (a
+   /dev/nvidia* mapping) whose u32 is small.
+
+   usage: scan PID   (prints candidate p and its fields) */
+#define _GNU_SOURCE
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <stdint.h>
+#include <fcntl.h>
+#include <unistd.h>
+#include <sys/stat.h>
+
+#define MAXR 4096
+static struct { uint64_t lo, hi; } wr[MAXR], nv[MAXR];
+static int nwr, nnv;
+
+static int in_nv(uint64_t a)
+{
+    for (int i = 0; i < nnv; i++)
+        if (a >= nv[i].lo && a < nv[i].hi) return 1;
+    return 0;
+}
+
+static int in_wr(uint64_t a)
+{
+    for (int i = 0; i < nwr; i++)
+        if (a >= wr[i].lo && a < wr[i].hi) return 1;
+    return 0;
+}
+
+static void load_maps(int pid)
+{
+    char path[64], line[1024];
+    snprintf(path, sizeof path, "/proc/%d/maps", pid);
+    FILE *f = fopen(path, "r");
+    if (!f) { perror("maps"); exit(1); }
+    while (fgets(line, sizeof line, f)) {
+        uint64_t lo, hi;
+        char perms[16], rest[768];
+        if (sscanf(line, "%lx-%lx %15s %*s %*s %*s %767[^\n]", &lo, &hi, perms, rest) < 3)
+            continue;
+        if (strchr(perms, 'w') && nwr < MAXR) { wr[nwr].lo = lo; wr[nwr].hi = hi; nwr++; }
+        if (strstr(line, "nvidia") && nnv < MAXR) { nv[nnv].lo = lo; nv[nnv].hi = hi; nnv++; }
+    }
+    fclose(f);
+}
+
+int main(int argc, char **argv)
+{
+    if (argc < 2) { fprintf(stderr, "usage: scan PID\n"); return 1; }
+    int pid = atoi(argv[1]);
+    load_maps(pid);
+    printf("writable regions=%d nvidia regions=%d\n", nwr, nnv);
+
+    char mp[64];
+    snprintf(mp, sizeof mp, "/proc/%d/mem", pid);
+    int fd = open(mp, O_RDONLY);
+    if (fd < 0) { perror("mem"); return 1; }
+
+    size_t CH = 4 << 20;
+    static uint8_t *buf;
+    buf = malloc(CH + 64);
+    int hits = 0;
+    for (int r = 0; r < nwr; r++) {
+        uint64_t a = wr[r].lo;
+        while (a < wr[r].hi) {
+            size_t n = (size_t)((wr[r].hi - a) < CH ? (wr[r].hi - a) : CH);
+            ssize_t got = pread(fd, buf, n, (off_t)a);
+            if (got <= 0) {  /* try smaller granularity */
+                size_t n2 = 256 << 10;
+                if (n > n2) { n = n2; continue; }
+                a += 4096; continue;
+            }
+            if ((size_t)got < n) {
+                /* partial read: scan what we got, then continue */
+                n = got;
+            }
+            n -= n % 8;
+            uint64_t *w = (uint64_t *)buf;
+            for (size_t i = 0; i + 4 <= n / 8; i++) {
+                uint64_t sub = w[i], last = w[i + 1], cur = w[i + 2], sem = w[i + 3];
+                if (sub == 0 || sub > 1000) continue;
+                if (last == 0 || last > 1000) continue;
+                if (cur > 0x200000000ULL) continue;
+                if (sem < 0x10000 || sem > 0x100000000ULL) continue;
+                uint64_t vaptr = 0;
+                if (pread(fd, &vaptr, 8, (off_t)(sem + 0x10)) != 8) continue;
+                if (!in_nv(vaptr)) continue;
+                uint32_t val = 0;
+                if (pread(fd, &val, 4, (off_t)vaptr) != 4) continue;
+                if (val > 1000) continue;
+                uint64_t p = a + i * 8 - 0x9418;
+                printf("P=0x%lx sub=%lu last=%lu cur=0x%lx sem_obj=0x%lx sem_va=0x%lx sem_val=%u\n",
+                       p, sub, last, cur, sem, vaptr, val);
+                hits++;
+            }
+            a += n;
+        }
+    }
+    printf("hits=%d\n", hits);
+    return 0;
+}
\ No newline at end of file