vgsp: m66 - host-readable TSG fault buffers + post-mortem dump

m65 proved the scheduled QMD never writes its release anywhere (module sem
stays 0, guest page stays 5) - the engine dies or is cancelled before
completion. The TSG fault-method buffers (0xa06c010a) are the hardware's
record of why; force them to host sysmem (the captured aperture, param
gsp_phys_rmcall_faultbuf_sys=1) and dump both buffers' first 8 words in the
post-mortem so an engine fault entry (faulting address/engine) is visible.

1 files changed, 20 insertions, 0 deletions

diff --git a/kernel/gsp_phys_gpu.c b/kernel/gsp_phys_gpu.c
index f323c6c..3429507 100644
--- a/kernel/gsp_phys_gpu.c
+++ b/kernel/gsp_phys_gpu.c
@@ -726,6 +726,26 @@ int gsp_phys_rmcall_gr_rel_mod_value(void)
 EXPORT_SYMBOL_GPL(gsp_phys_rmcall_gr_rel_mod_value);
 
 /*
+ * m66 fault-buffer aperture (gsp_phys_rmcall_faultbuf_sys; default 0). When
+ * set, the mode-35 TSG fault-method buffers are registered host sysmem chunks
+ * (the aperture the capture itself used) instead of server FBMEM, so engine
+ * fault entries written by the hardware after a launch are host-readable and
+ * can be dumped in the post-mortem. Diagnostic only; the promote shape is
+ * otherwise identical.
+ */
+static int gsp_phys_rmcall_faultbuf_sys_param;
+module_param_named(gsp_phys_rmcall_faultbuf_sys,
+		   gsp_phys_rmcall_faultbuf_sys_param, int, 0644);
+MODULE_PARM_DESC(gsp_phys_rmcall_faultbuf_sys,
+		 "Register the mode-35 TSG fault buffers as host sysmem (captured aperture; default 0 = FBMEM first)");
+
+int gsp_phys_rmcall_faultbuf_sys_value(void)
+{
+	return gsp_phys_rmcall_faultbuf_sys_param;
+}
+EXPORT_SYMBOL_GPL(gsp_phys_rmcall_faultbuf_sys_value);
+
+/*
  * Ladder VAS ownership selector (gsp_phys_rmcall_vas_shared; default 0).
  * The H1 GR promote wall is a direct consequence of the ladder's
  * externally-owned FERMI_VASPACE_A: with IS_EXTERNALLY_OWNED the RM treats