vgsp: m83 - host-readable sysmem error notifier for the RM's channel-error records

NV_PGRAPH_EXCEPTION reads 0x01000000 (GPC) with EXCEPTION1 GPC1:0=0b11 at the
QMD activation; the fault record the engine writes goes to the channel's error
notifier, which is server FBMEM today and unreadable. gsp_phys_rmcall_gr_notifier_sys=1
registers a host sysmem chunk as the hObjectError memory object and carries it
in the channel's errorNotifierMem with the SYS aperture, so the RM's
asynchronous channel-error report is dumped in the launch post-mortem
(the m35 lane still uses the captured substitute object shape).

1 files changed, 20 insertions, 0 deletions

diff --git a/kernel/gsp_phys_gpu.c b/kernel/gsp_phys_gpu.c
index a71ab97..f094d36 100644
--- a/kernel/gsp_phys_gpu.c
+++ b/kernel/gsp_phys_gpu.c
@@ -796,6 +796,26 @@ unsigned int gsp_phys_rmcall_gr_lmem_tpc_size_value(void)
 EXPORT_SYMBOL_GPL(gsp_phys_rmcall_gr_lmem_tpc_size_value);
 
 /*
+ * m83 sysmem error notifier (gsp_phys_rmcall_gr_notifier_sys; default 0).
+ * The mode-35 V7 error notifier is normally a server FBMEM class-0x40 memory
+ * object (unreadable). With 1 it is a host sysmem chunk registered as the
+ * hObjectError memory object and carried in the channel's errorNotifierMem
+ * with the SYS aperture, so the RM's asynchronous channel-error records for
+ * the GPC exception are readable (dumped in the launch post-mortem).
+ */
+static int gsp_phys_rmcall_gr_notifier_sys_param;
+module_param_named(gsp_phys_rmcall_gr_notifier_sys,
+		   gsp_phys_rmcall_gr_notifier_sys_param, int, 0644);
+MODULE_PARM_DESC(gsp_phys_rmcall_gr_notifier_sys,
+		 "Mode-35 error notifier as host sysmem (RM fault records readable; default 0)");
+
+int gsp_phys_rmcall_gr_notifier_sys_value(void)
+{
+	return gsp_phys_rmcall_gr_notifier_sys_param;
+}
+EXPORT_SYMBOL_GPL(gsp_phys_rmcall_gr_notifier_sys_value);
+
+/*
  * m78 lane-enable skip mask (gsp_phys_rmcall_gr_lane_skip; default 0). Bit 0
  * skips R2 #1059 GR_SET_CTXSW_PREEMPTION_MODE (CILP), bit 1 skips #1061
  * SET_TIMESLICE, bit 2 skips #1063 device-sched. Used to bisect which