[PATCH 051/109] drm/amd/ras: Add gpu rma exception handling

From: YiPeng Chai <[email protected]>

Add gpu rma exception handling.

Signed-off-by: YiPeng Chai <[email protected]>
Reviewed-by: Hawking Zhang <[email protected]>
Signed-off-by: Alex Deucher <[email protected]>
---
 drivers/gpu/drm/amd/ras/core/core.c        |  3 +++
 drivers/gpu/drm/amd/ras/core/ras_process.c | 17 +++++++++++++++--
 2 files changed, 18 insertions(+), 2 deletions(-)

diff --git a/drivers/gpu/drm/amd/ras/core/core.c b/drivers/gpu/drm/amd/ras/core/core.c
index 8a8f5134fdac7..f093f29635888 100644
--- a/drivers/gpu/drm/amd/ras/core/core.c
+++ b/drivers/gpu/drm/amd/ras/core/core.c
@@ -287,6 +287,9 @@ static int ras_core_eeprom_recovery(struct ras_core_context *ras_core)
 		return ret;
 	}
 
+	if (ras_eeprom_mgr_get_gpu_op_status(ras_core) == RAS_GPU_OP_STATUS_RMA)
+		return -EPERM;
+
 	return ret;
 }
 
diff --git a/drivers/gpu/drm/amd/ras/core/ras_process.c b/drivers/gpu/drm/amd/ras/core/ras_process.c
index dd4a4c5673bdd..e7f95a82fb901 100644
--- a/drivers/gpu/drm/amd/ras/core/ras_process.c
+++ b/drivers/gpu/drm/amd/ras/core/ras_process.c
@@ -173,6 +173,14 @@ int ras_process_handle_ras_event(struct ras_core_context *ras_core)
 	if (ras_eeprom_mgr_fw_record_enabled(ras_core))
 		ras_umc_dump_fw_records(ras_core);
 
+	if (ras_eeprom_mgr_get_gpu_op_status(ras_core) == RAS_GPU_OP_STATUS_RMA) {
+		uint32_t rma_reset = GPU_RESET_CAUSE_RMA;
+
+		RAS_DEV_ERR(ras_core->dev, "Device is RMA, stop gpu work!\n");
+		ras_core_event_notify(ras_core, RAS_EVENT_ID__RESET_GPU, &rma_reset);
+		return -ERESTART;
+	}
+
 	do {
 		umc_event_count = atomic_read(&ras_proc->umc_interrupt_count);
 		ret = ras_process_umc_event(ras_core, umc_event_count);
@@ -210,6 +218,7 @@ static int ras_process_thread(void *context)
 {
 	struct ras_core_context *ras_core = (struct ras_core_context *)context;
 	struct ras_process *ras_proc = &ras_core->ras_proc;
+	int ret;
 
 	while (!kthread_should_stop()) {
 		ras_wait_event_interruptible_timeout(&ras_proc->ras_process_wq,
@@ -228,9 +237,13 @@ static int ras_process_thread(void *context)
 			continue;
 
 		if (ras_core->sys_fn && ras_core->sys_fn->async_handle_ras_event)
-			ras_core->sys_fn->async_handle_ras_event(ras_core, NULL);
+			ret = ras_core->sys_fn->async_handle_ras_event(ras_core, NULL);
 		else
-			ras_process_handle_ras_event(ras_core);
+			ret = ras_process_handle_ras_event(ras_core);
+
+		/* GPU Device is RMA */
+		if (ret == -ERESTART)
+			break;
 	}
 
 	return 0;
-- 
2.55.0