userspace/cuda: libcuda analysis tree (debug-focused), guest ioctl tracer

Analysis tooling for debugging CUDA execution in Open-IOV guests:
- userspace/cuda/570.144/: fetch a library from a running guest (nc),
  objdump helpers, crash ip -> file vaddr + symbol mapping, and Ghidra
  headless import/analysis/decompile scripts (project kept so single
  functions decompile in seconds). The purpose is debugging correctness
  and performance of CUDA in Open-IOV VMs (not a reimplementation);
  extracted binaries and decompiled output are gitignored.
- tools/guest-diag/ioctl_trace.c: ptrace tracker for the NVIDIA driver
  ioctls of a child (libcuda issues raw syscalls, so LD_PRELOAD sees
  nothing). Decodes RM_ALLOC (class) and RM_CONTROL (cmd) arguments;
  this is how the cuCtxCreate crash sequence was captured.
- tools/test/cuda_drv_test.c: unbuffered + per-call markers so a crash
  identifies the CUDA call it happened in.

1 files changed, 14 insertions, 0 deletions

diff --git a/tools/test/cuda_drv_test.c b/tools/test/cuda_drv_test.c
index 6868a2e..20ad88a 100644
--- a/tools/test/cuda_drv_test.c
+++ b/tools/test/cuda_drv_test.c
@@ -30,8 +30,13 @@ typedef uint64_t CUdeviceptr;
     printf("%-24s OK\n", #expr); \
 } while (0)
 
+/* Print before entering libcuda so a crash identifies the call. */
+#define MARK(name) do { printf("-> %s\n", name); } while (0)
+
 int main(void)
 {
+    setvbuf(stdout, NULL, _IONBF, 0);
+
     void *h = dlopen("libcuda.so.1", RTLD_NOW);
     if (!h) { printf("dlopen libcuda.so.1 failed: %s\n", dlerror()); return 1; }
 
@@ -59,28 +64,37 @@ int main(void)
     unsigned int back[64];
     size_t i;
 
+    MARK("cuInit");
     CHECK(cuInit(0));
+    MARK("cuDeviceGetCount");
     CHECK(cuDeviceGetCount(&count));
     printf("device count = %d\n", count);
     if (count < 1) return 1;
 
+    MARK("cuDeviceGet");
     CHECK(cuDeviceGet(&dev, 0));
     printf("device 0 = %d\n", dev);
 
+    MARK("cuCtxCreate");
     CHECK(cuCtxCreate(&ctx, 0, dev));
     printf("context = %p\n", ctx);
 
+    MARK("cuMemAlloc");
     CHECK(cuMemAlloc(&dptr, sizeof(buf)));
     printf("device ptr = 0x%llx (%zu bytes)\n", (unsigned long long)dptr, sizeof(buf));
 
     for (i = 0; i < 64; i++) buf[i] = (unsigned int)(i * 3 + 1);
+    MARK("cuMemcpyHtoD");
     CHECK(cuMemcpyHtoD(dptr, buf, sizeof(buf)));
     memset(back, 0, sizeof(back));
+    MARK("cuMemcpyDtoH");
     CHECK(cuMemcpyDtoH(back, dptr, sizeof(buf)));
     printf("roundtrip %s (back[7]=%u expected 22)\n",
            memcmp(buf, back, sizeof(buf)) ? "MISMATCH" : "match", back[7]);
 
+    MARK("cuMemFree");
     CHECK(cuMemFree(dptr));
+    MARK("cuCtxDestroy");
     CHECK(cuCtxDestroy(ctx));
     printf("ALL OK\n");
     return 0;