[v3,2/2] drm/amdgpu: Add work function for GPU reset event

Commit Message

André Almeida Nov. 25, 2022, 5:52 p.m. UTC

Add a work function to send a GPU reset uevent and scheduled it during
a GPU reset.

Co-developed-by: Shashank Sharma <[e-mail]>
Signed-off-by: Shashank Sharma <[e-mail]>
Signed-off-by: André Almeida <[e-mail]>
---
V3:
   - Merge two last commits

V2: Addressed review comments from Christian
   - Changed the name of the work to gpu_reset_event_work
   - Added a structure to accommodate some additional information
     (like a PID and some flags)
   - Do not add new structure in amdgpu.h
---
 drivers/gpu/drm/amd/amdgpu/amdgpu.h        |  4 +++
 drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 30 ++++++++++++++++++++++
 2 files changed, 34 insertions(+)

Patch

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
index 6b74df446694..88cb5b739c5d 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
@@ -60,6 +60,8 @@ 
 #include <drm/amdgpu_drm.h>
 #include <drm/drm_gem.h>
 #include <drm/drm_ioctl.h>
+#include <drm/gpu_scheduler.h>
+#include <drm/drm_sysfs.h>
 
 #include <kgd_kfd_interface.h>
 #include "dm_pp_interface.h"
@@ -1003,6 +1005,7 @@  struct amdgpu_device {
 
 	int asic_reset_res;
 	struct work_struct		xgmi_reset_work;
+	struct work_struct		gpu_reset_event_work;
 	struct list_head		reset_list;
 
 	long				gfx_timeout;
@@ -1036,6 +1039,7 @@  struct amdgpu_device {
 	pci_channel_state_t		pci_channel_state;
 
 	struct amdgpu_reset_control     *reset_cntl;
+	struct drm_reset_event_info	reset_event_info;
 	uint32_t                        ip_versions[MAX_HWIP][HWIP_MAX_INSTANCE];
 
 	bool				ram_is_direct_mapped;
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
index b2b1c66bfe39..d04541fdb606 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
@@ -77,6 +77,7 @@ 
 #include <linux/pm_runtime.h>
 
 #include <drm/drm_drv.h>
+#include <drm/drm_sysfs.h>
 
 MODULE_FIRMWARE("amdgpu/vega10_gpu_info.bin");
 MODULE_FIRMWARE("amdgpu/vega12_gpu_info.bin");
@@ -3365,6 +3366,19 @@  bool amdgpu_device_has_dc_support(struct amdgpu_device *adev)
 	return amdgpu_device_asic_has_dc_support(adev->asic_type);
 }
 
+static void amdgpu_device_reset_event_func(struct work_struct *__work)
+{
+	struct amdgpu_device *adev = container_of(__work, struct amdgpu_device,
+						   gpu_reset_event_work);
+	/*
+	 * A GPU reset has happened, inform the userspace and pass the reset
+	 * related information
+	 */
+	drm_sysfs_reset_event(&adev->ddev, &adev->reset_event_info);
+
+	put_pid(adev->reset_event_info.pid);
+}
+
 static void amdgpu_device_xgmi_reset_func(struct work_struct *__work)
 {
 	struct amdgpu_device *adev =
@@ -3616,6 +3630,7 @@  int amdgpu_device_init(struct amdgpu_device *adev,
 			  amdgpu_device_delay_enable_gfx_off);
 
 	INIT_WORK(&adev->xgmi_reset_work, amdgpu_device_xgmi_reset_func);
+	INIT_WORK(&adev->gpu_reset_event_work, amdgpu_device_reset_event_func);
 
 	adev->gfx.gfx_off_req_count = 1;
 	adev->gfx.gfx_off_residency = 0;
@@ -4920,6 +4935,21 @@  int amdgpu_do_asic_reset(struct list_head *device_list_handle,
 					goto out;
 
 				vram_lost = amdgpu_device_check_vram_lost(tmp_adev);
+
+				if (reset_context->job && reset_context->job->vm) {
+					tmp_adev->reset_event_info.pid =
+						find_get_pid(reset_context->job->vm->task_info.pid);
+				} else {
+					tmp_adev->reset_event_info.pid = NULL;
+				}
+
+				if (vram_lost)
+					tmp_adev->reset_event_info.flags |=
+						DRM_RESET_EVENT_VRAM_LOST;
+
+				/* Send GPU reset event */
+				schedule_work(&tmp_adev->gpu_reset_event_work);
+
 #ifdef CONFIG_DEV_COREDUMP
 				tmp_adev->reset_vram_lost = vram_lost;
 				memset(&tmp_adev->reset_task_info, 0,