kernel: coherent GSP-RM replies for engine-context creation

- GSP_RM_ALLOC: echo the request payload back with status forced to NV_OK.
  The kernel driver copies the reply params over the caller's params buffer
  using the class-expected size, so a zeroed reply smashed libcuda's stack
  (SIGSEGV at the engine-registration continuation); echoing makes the
  copy-back idempotent.
- CE_GET_PHYSICAL_CAPS / CE_GET_ALL(_PHYSICAL)_CAPS: report sysmem, P2P and
  BL copy caps and mark the GR-synchronous CE (GRCE, COPY0). Without a GRCE
  the client's context walk ends with an empty secondary-engine record and
  the engine RM alloc is issued with class 0 (rejected -> cuCtxCreate fails).
- TU104 engine table: CE0 shares the GR PBDMA ids (the driver's GR/GRCE
  runqueue validation requires it), plus an entry for the async copy engine
  (COPY14) advertised in the GSP static info engine caps.
- tools/guest-diag/ioctl_trace: decode the RM status fields of alloc,
  control and free so the first failing driver call is visible in traces.

gitignore: keep the CUDA analysis scratch tree (incl. WORKING-HERE.md) local.

1 files changed, 25 insertions, 10 deletions

diff --git a/tools/guest-diag/ioctl_trace.c b/tools/guest-diag/ioctl_trace.c
index fd1802a..04fea05 100644
--- a/tools/guest-diag/ioctl_trace.c
+++ b/tools/guest-diag/ioctl_trace.c
@@ -38,17 +38,29 @@ static int read_child(pid_t pid, unsigned long addr, void *buf, size_t len)
 	return 0;
 }
 
-static void dump_alloc(pid_t pid, unsigned long arg)
+static void dump_alloc(pid_t pid, unsigned long arg, unsigned int size)
 {
 	unsigned int v[12];
+	unsigned int st;
 
 	if (read_child(pid, arg, v, sizeof(v)))
 		return;
-	/* NVOS21_PARAMETERS: hRoot, hObjectParent, hObjectNew, hClass,
-	 * pAllocParms, pRightsRequested, paramsSize, flags, status */
+	/* NVOS21_PARAMETERS (0x20): ... paramsSize@0x14, status@0x10
+	 * NVOS64_PARAMETERS (0x30): ... status@0x28 */
+	st = (size == 0x30) ? v[10] : v[4];
 	fprintf(stderr,
-		"[iot]   alloc root=0x%x parent=0x%x new=0x%x class=0x%x params=%u flags=0x%x\n",
-		v[0], v[1], v[2], v[3], v[6], v[7]);
+		"[iot]   alloc root=0x%x parent=0x%x new=0x%x class=0x%x st=0x%x\n",
+		v[0], v[1], v[2], v[3], st);
+}
+
+static void dump_free(pid_t pid, unsigned long arg)
+{
+	unsigned int v[4];
+
+	if (read_child(pid, arg, v, sizeof(v)))
+		return;
+	/* NVOS00_PARAMETERS: hRoot, hObjectParent, hObjectOld, status */
+	fprintf(stderr, "[iot]   free old=0x%x st=0x%x\n", v[2], v[3]);
 }
 
 static void dump_control(pid_t pid, unsigned long arg)
@@ -59,12 +71,13 @@ static void dump_control(pid_t pid, unsigned long arg)
 
 	if (read_child(pid, arg, v, sizeof(v)))
 		return;
-	/* NVOS54_PARAMETERS: hClient, hObject, cmd, flags, params, paramsSize, ... */
+	/* NVOS54_PARAMETERS: hClient, hObject, cmd, flags, params, paramsSize,
+	 * status@0x1c */
 	cmd = v[2];
 	memcpy(&pParams, &v[4], sizeof(pParams));
 	fprintf(stderr,
-		"[iot]   control client=0x%x object=0x%x cmd=0x%x params=0x%llx size=%u\n",
-		v[0], v[1], cmd, pParams, v[5]);
+		"[iot]   control client=0x%x object=0x%x cmd=0x%x params=0x%llx size=%u st=0x%x\n",
+		v[0], v[1], cmd, pParams, v[5], v[7]);
 }
 
 int main(int argc, char **argv)
@@ -148,8 +161,10 @@ int main(int argc, char **argv)
 					fprintf(stderr,
 						"[iot] req=0x%llx arg=0x%llx ret=0x%llx\n",
 						req, a2, ret);
-					if (req == 0xc030462bULL)
-						dump_alloc(pid, a2);
+					if ((req & 0xff) == 0x2b)
+						dump_alloc(pid, a2, (req >> 16) & 0x3fff);
+					else if ((req & 0xff) == 0x29)
+						dump_free(pid, a2);
 					else if ((req & 0xff) == 0x2a)
 						dump_control(pid, a2);
 				}