drm/amd/display: Add extra DMUB logging to track message timeout

[Description]
- Add logging for first DMUB inbox message that timed out to diagnostic
  data
- It is useful to track the first failed message for debug purposes
  because once DMUB becomes hung (typically on a message), it will
  remain hung and all subsequent messages. In these cases we're
  interested in knowing which is the first message that failed.
Reviewed-by: Josip Pavic <josip.pavic@amd.com>
Acked-by: Roman Li <roman.li@amd.com>
Signed-off-by: Alvin Lee <alvin.lee2@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drivers/gpu/drm/amd/display/dc/dc_dmub_srv.c

@@ -23,6 +23,7 @@
  *
  */
 
+#include "dm_services.h"
 #include "dc.h"
 #include "dc_dmub_srv.h"
 #include "../dmub/dmub_srv.h"
@@ -198,6 +199,11 @@ bool dc_dmub_srv_wait_for_idle(struct dc_dmub_srv *dc_dmub_srv,
 
 		if (status != DMUB_STATUS_OK) {
 			DC_LOG_DEBUG("No reply for DMUB command: status=%d\n", status);
+			if (!dmub->debug.timeout_occured) {
+				dmub->debug.timeout_occured = true;
+				dmub->debug.timeout_cmd = *cmd_list;
+				dmub->debug.timestamp = dm_get_timestamp(dc_dmub_srv->ctx);
+			}
 			dc_dmub_srv_log_diagnostic_data(dc_dmub_srv);
 			return false;
 		}
@@ -904,6 +910,7 @@ bool dc_dmub_srv_get_diagnostic_data(struct dc_dmub_srv *dc_dmub_srv, struct dmu
 void dc_dmub_srv_log_diagnostic_data(struct dc_dmub_srv *dc_dmub_srv)
 {
 	struct dmub_diagnostic_data diag_data = {0};
+	uint32_t i;
 
 	if (!dc_dmub_srv || !dc_dmub_srv->dmub) {
 		DC_LOG_ERROR("%s: invalid parameters.", __func__);
@@ -935,7 +942,8 @@ void dc_dmub_srv_log_diagnostic_data(struct dc_dmub_srv *dc_dmub_srv)
 	DC_LOG_DEBUG("    scratch [13]       : %08x", diag_data.scratch[13]);
 	DC_LOG_DEBUG("    scratch [14]       : %08x", diag_data.scratch[14]);
 	DC_LOG_DEBUG("    scratch [15]       : %08x", diag_data.scratch[15]);
-	DC_LOG_DEBUG("    pc                 : %08x", diag_data.pc);
+	for (i = 0; i < DMUB_PC_SNAPSHOT_COUNT; i++)
+		DC_LOG_DEBUG("    pc[%d]             : %08x", i, diag_data.pc[i]);
 	DC_LOG_DEBUG("    unk_fault_addr     : %08x", diag_data.undefined_address_fault_addr);
 	DC_LOG_DEBUG("    inst_fault_addr    : %08x", diag_data.inst_fetch_fault_addr);
 	DC_LOG_DEBUG("    data_fault_addr    : %08x", diag_data.data_write_fault_addr);

drivers/gpu/drm/amd/display/dmub/dmub_srv.h

@@ -71,6 +71,8 @@
 extern "C" {
 #endif
 
+#define DMUB_PC_SNAPSHOT_COUNT 10
+
 /* Forward declarations */
 struct dmub_srv;
 struct dmub_srv_common_regs;
@@ -299,6 +301,17 @@ struct dmub_srv_hw_params {
 	enum dmub_ips_disable_type disable_ips;
 };
 
+/**
+ * struct dmub_srv_debug - Debug info for dmub_srv
+ * @timeout_occured: Indicates a timeout occured on any message from driver to dmub
+ * @timeout_cmd: first cmd sent from driver that timed out - subsequent timeouts are not stored
+ */
+struct dmub_srv_debug {
+	bool timeout_occured;
+	union dmub_rb_cmd timeout_cmd;
+	unsigned long long timestamp;
+};
+
 /**
  * struct dmub_diagnostic_data - Diagnostic data retrieved from DMCUB for
  * debugging purposes, including logging, crash analysis, etc.
@@ -306,7 +319,7 @@ struct dmub_srv_hw_params {
 struct dmub_diagnostic_data {
 	uint32_t dmcub_version;
 	uint32_t scratch[17];
-	uint32_t pc;
+	uint32_t pc[DMUB_PC_SNAPSHOT_COUNT];
 	uint32_t undefined_address_fault_addr;
 	uint32_t inst_fetch_fault_addr;
 	uint32_t data_write_fault_addr;
@@ -317,6 +330,7 @@ struct dmub_diagnostic_data {
 	uint32_t inbox0_wptr;
 	uint32_t inbox0_size;
 	uint32_t gpint_datain0;
+	struct dmub_srv_debug timeout_info;
 	uint8_t is_dmcub_enabled : 1;
 	uint8_t is_dmcub_soft_reset : 1;
 	uint8_t is_dmcub_secure_reset : 1;
@@ -506,6 +520,7 @@ struct dmub_srv {
 	struct dmub_visual_confirm_color visual_confirm_color;
 
 	enum dmub_srv_power_state_type power_state;
+	struct dmub_srv_debug debug;
 };
 
 /**

drivers/gpu/drm/amd/display/dmub/src/dmub_dcn20.c

@@ -472,4 +472,5 @@ void dmub_dcn20_get_diagnostic_data(struct dmub_srv *dmub, struct dmub_diagnosti
 
 	REG_GET(DMCUB_REGION3_CW6_TOP_ADDRESS, DMCUB_REGION3_CW6_ENABLE, &is_cw6_enabled);
 	diag_data->is_cw6_enabled = is_cw6_enabled;
+	diag_data->timeout_info = dmub->debug;
 }

drivers/gpu/drm/amd/display/dmub/src/dmub_dcn31.c

@@ -466,6 +466,7 @@ void dmub_dcn31_get_diagnostic_data(struct dmub_srv *dmub, struct dmub_diagnosti
 
 	REG_GET(DMCUB_REGION3_CW6_TOP_ADDRESS, DMCUB_REGION3_CW6_ENABLE, &is_cw6_enabled);
 	diag_data->is_cw6_enabled = is_cw6_enabled;
+	diag_data->timeout_info = dmub->debug;
 }
 
 bool dmub_dcn31_should_detect(struct dmub_srv *dmub)

drivers/gpu/drm/amd/display/dmub/src/dmub_dcn32.c

@@ -478,6 +478,8 @@ void dmub_dcn32_get_diagnostic_data(struct dmub_srv *dmub, struct dmub_diagnosti
 	diag_data->is_cw6_enabled = is_cw6_enabled;
 
 	diag_data->gpint_datain0 = REG_READ(DMCUB_GPINT_DATAIN0);
+
+	diag_data->timeout_info = dmub->debug;
 }
 void dmub_dcn32_configure_dmub_in_system_memory(struct dmub_srv *dmub)
 {

drivers/gpu/drm/amd/display/dmub/src/dmub_dcn35.c

@@ -516,6 +516,7 @@ void dmub_dcn35_get_diagnostic_data(struct dmub_srv *dmub, struct dmub_diagnosti
 	diag_data->is_cw6_enabled = is_cw6_enabled;
 
 	diag_data->gpint_datain0 = REG_READ(DMCUB_GPINT_DATAIN0);
+	diag_data->timeout_info = dmub->debug;
 }
 void dmub_dcn35_configure_dmub_in_system_memory(struct dmub_srv *dmub)
 {