mali_kbase: Add more gpu error events

Bug: 312206733
Change-Id: I154b927152d768944e02fe6083e66bdbb0f47e3a
Signed-off-by: Varad Gautam <varadgautam@google.com>

6 files changed, 31 insertions, 4 deletions

diff --git a/mali_kbase/csf/mali_kbase_csf.c b/mali_kbase/csf/mali_kbase_csf.c
index 94bd2d0..7fcd45e 100644
--- a/mali_kbase/csf/mali_kbase_csf.c
+++ b/mali_kbase/csf/mali_kbase_csf.c
@@ -40,6 +40,7 @@
 #include <tl/mali_kbase_tracepoints.h>
 #include "mali_kbase_csf_mcu_shared_reg.h"
 #include <linux/version_compat_defs.h>
+#include "mali_kbase_config_defaults.h"
 
 #define CS_REQ_EXCEPTION_MASK (CS_REQ_FAULT_MASK | CS_REQ_FATAL_MASK)
 #define CS_ACK_EXCEPTION_MASK (CS_ACK_FAULT_MASK | CS_ACK_FATAL_MASK)
@@ -1922,6 +1923,7 @@ static void report_tiler_oom_error(struct kbase_queue_group *group)
 
 	kbase_csf_event_add_error(group->kctx, &group->error_fatal, &error);
 	kbase_event_wakeup_sync(group->kctx);
+	pixel_gpu_uevent_kmd_error_send(group->kctx->kbdev, GPU_UEVENT_INFO_TILER_OOM);
 }
 
 static void flush_gpu_cache_on_fatal_error(struct kbase_device *kbdev)
@@ -2108,6 +2110,7 @@ static void timer_event_worker(struct work_struct *data)
 	struct kbase_device *const kbdev = kctx->kbdev;
 	bool reset_prevented = false;
 	int err = kbase_reset_gpu_prevent_and_wait(kbdev);
+	pixel_gpu_uevent_kmd_error_send(kbdev, GPU_UEVENT_INFO_PROGRESS_TIMER);
 
 	if (err)
 		dev_warn(
@@ -2366,6 +2369,7 @@ static void cs_error_worker(struct work_struct *const data)
 	bool reset_prevented = false;
 	int err;
 
+	pixel_gpu_uevent_kmd_error_send(kbdev, GPU_UEVENT_INFO_CS_ERROR);
 	kbase_debug_csf_fault_wait_completion(kbdev);
 	err = kbase_reset_gpu_prevent_and_wait(kbdev);
 
diff --git a/mali_kbase/csf/mali_kbase_csf_firmware.c b/mali_kbase/csf/mali_kbase_csf_firmware.c
index 7b9aefd..d6d2456 100644
--- a/mali_kbase/csf/mali_kbase_csf_firmware.c
+++ b/mali_kbase/csf/mali_kbase_csf_firmware.c
@@ -31,6 +31,7 @@
 #include "mali_kbase_reset_gpu.h"
 #include "mali_kbase_ctx_sched.h"
 #include "mali_kbase_csf_scheduler.h"
+#include "mali_kbase_config_defaults.h"
 #include <mali_kbase_hwaccess_time.h>
 #include "device/mali_kbase_device.h"
 #include "backend/gpu/mali_kbase_pm_internal.h"
@@ -1558,6 +1559,7 @@ static void firmware_error_worker(struct work_struct *const data)
 {
 	struct kbase_device *const kbdev =
 		container_of(data, struct kbase_device, csf.fw_error_work);
+	pixel_gpu_uevent_kmd_error_send(kbdev, GPU_UEVENT_INFO_FW_ERROR);
 
 	handle_internal_firmware_fatal(kbdev);
 }
diff --git a/mali_kbase/csf/mali_kbase_csf_scheduler.c b/mali_kbase/csf/mali_kbase_csf_scheduler.c
index fcb972f..8d0fff6 100644
--- a/mali_kbase/csf/mali_kbase_csf_scheduler.c
+++ b/mali_kbase/csf/mali_kbase_csf_scheduler.c
@@ -1114,6 +1114,7 @@ static bool scheduler_protm_wait_quit(struct kbase_device *kbdev)
 		dev_warn(kbdev->dev, "[%llu] Timeout (%d ms), protm_quit wait skipped",
 			 kbase_backend_get_cycle_cnt(kbdev), fw_timeout_ms);
 		schedule_actions_trigger_df(kbdev, kctx, DF_PROTECTED_MODE_EXIT_TIMEOUT);
+		pixel_gpu_uevent_kmd_error_send(kbdev, GPU_UEVENT_INFO_PMODE_EXIT_TIMEOUT);
 		success = false;
 	}
 
@@ -4504,10 +4505,12 @@ static void scheduler_group_check_protm_enter(struct kbase_device *const kbdev,
 				err = kbase_csf_wait_protected_mode_enter(kbdev);
 				up_write(&kbdev->csf.pmode_sync_sem);
 
-				if (err)
+				if (err) {
+					pixel_gpu_uevent_kmd_error_send(kbdev, GPU_UEVENT_INFO_PMODE_ENTRY_FAILURE);
 					schedule_actions_trigger_df(
 						kbdev, input_grp->kctx,
 						DF_PROTECTED_MODE_ENTRY_FAILURE);
+				}
 
 				scheduler->protm_enter_time = ktime_get_raw();
 
@@ -5944,7 +5947,7 @@ static void evict_lru_or_blocked_csg(struct kbase_device *kbdev)
 		suspend_queue_group(lru_idle_group);
 		if (wait_csg_slots_suspend(kbdev, &slot_mask)) {
 			enum dumpfault_error_type error_type = DF_CSG_SUSPEND_TIMEOUT;
-
+			pixel_gpu_uevent_kmd_error_send(kbdev, GPU_UEVENT_INFO_CSG_SUSPEND);
 			dev_warn(
 				kbdev->dev,
 				"[%llu] LRU idle group %d of context %d_%d failed to suspend on slot %d (timeout %d ms)",
diff --git a/mali_kbase/mmu/mali_kbase_mmu_hw_direct.c b/mali_kbase/mmu/mali_kbase_mmu_hw_direct.c
index 4cfa496..bfe6527 100644
--- a/mali_kbase/mmu/mali_kbase_mmu_hw_direct.c
+++ b/mali_kbase/mmu/mali_kbase_mmu_hw_direct.c
@@ -28,6 +28,7 @@
 #include <mmu/mali_kbase_mmu_hw.h>
 #include <tl/mali_kbase_tracepoints.h>
 #include <linux/delay.h>
+#include "mali_kbase_config_defaults.h"
 
 #if MALI_USE_CSF
 /**
@@ -179,6 +180,7 @@ static int wait_ready(struct kbase_device *kbdev, unsigned int as_nr)
 	if (!err)
 		return 0;
 
+	pixel_gpu_uevent_kmd_error_send(kbdev, GPU_UEVENT_INFO_MMU_AS_ACTIVE_STUCK);
 	dev_err(kbdev->dev,
 		"AS_ACTIVE bit stuck for as %u. Might be caused by unstable GPU clk/pwr or faulty system",
 		as_nr);
diff --git a/mali_kbase/platform/pixel/pixel_gpu_uevent.c b/mali_kbase/platform/pixel/pixel_gpu_uevent.c
index 4a5c1d3..9f0589c 100644
--- a/mali_kbase/platform/pixel/pixel_gpu_uevent.c
+++ b/mali_kbase/platform/pixel/pixel_gpu_uevent.c
@@ -36,9 +36,17 @@ static bool gpu_uevent_check_valid(const struct gpu_uevent *evt)
         case GPU_UEVENT_INFO_CSG_SLOT_READY:
         case GPU_UEVENT_INFO_L2_PM_TIMEOUT:
         case GPU_UEVENT_INFO_PM_TIMEOUT:
+        case GPU_UEVENT_INFO_TILER_OOM:
+        case GPU_UEVENT_INFO_PROGRESS_TIMER:
+        case GPU_UEVENT_INFO_CS_ERROR:
+        case GPU_UEVENT_INFO_FW_ERROR:
+        case GPU_UEVENT_INFO_PMODE_EXIT_TIMEOUT:
+        case GPU_UEVENT_INFO_PMODE_ENTRY_FAILURE:
+        case GPU_UEVENT_INFO_GPU_PAGE_FAULT:
+        case GPU_UEVENT_INFO_MMU_AS_ACTIVE_STUCK:
             return true;
         default:
-            break;
+            return false;
         }
     case GPU_UEVENT_TYPE_GPU_RESET:
         switch (evt->info) {
@@ -46,7 +54,7 @@ static bool gpu_uevent_check_valid(const struct gpu_uevent *evt)
         case GPU_UEVENT_INFO_CSF_RESET_FAILED:
             return true;
         default:
-            break;
+            return false;
         }
     default:
         break;
diff --git a/mali_kbase/platform/pixel/pixel_gpu_uevent.h b/mali_kbase/platform/pixel/pixel_gpu_uevent.h
index b6756ec..b477262 100644
--- a/mali_kbase/platform/pixel/pixel_gpu_uevent.h
+++ b/mali_kbase/platform/pixel/pixel_gpu_uevent.h
@@ -48,6 +48,14 @@ static inline const char *gpu_uevent_type_str(enum gpu_uevent_type type) {
     GPU_UEVENT_INFO(PM_TIMEOUT)                 \
     GPU_UEVENT_INFO(CSF_RESET_OK)               \
     GPU_UEVENT_INFO(CSF_RESET_FAILED)           \
+    GPU_UEVENT_INFO(TILER_OOM)                  \
+    GPU_UEVENT_INFO(PROGRESS_TIMER)             \
+    GPU_UEVENT_INFO(CS_ERROR)                   \
+    GPU_UEVENT_INFO(FW_ERROR)                   \
+    GPU_UEVENT_INFO(PMODE_EXIT_TIMEOUT)         \
+    GPU_UEVENT_INFO(PMODE_ENTRY_FAILURE)        \
+    GPU_UEVENT_INFO(GPU_PAGE_FAULT)             \
+    GPU_UEVENT_INFO(MMU_AS_ACTIVE_STUCK)        \
     GPU_UEVENT_INFO(MAX)
 
 #define GPU_UEVENT_INFO(info) GPU_UEVENT_INFO_##info,