Skip to contentSkip to conversation

Revert "drm/amdgpu: add debugfs amdgpu_reset_level"

PATCH In review Victor Zhao 3 participants 5 messages Oct 13, 2022

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

PATCH 1/3 Revert "drm/amdgpu: add debugfs amdgpu_reset_level"

This reverts commit 3ae992d5e1194a16e3d977076eb5722fa6e410d8.

This commit breaks the reset logic for aldebaran, revert it for now. Will move the mask inside the reset handler.

---
 drivers/gpu/drm/amd/amdgpu/amdgpu.h         | 4 ----
 drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c | 2 --
 drivers/gpu/drm/amd/amdgpu/amdgpu_reset.c   | 8 --------
 drivers/gpu/drm/amd/amdgpu/amdgpu_ring.c    | 3 ---
 4 files changed, 17 deletions(-)
drivers/gpu/drm/amd/amdgpu/amdgpu.h +0 -4
index 3ce91f660c3f..0e6ddf05c23c 100644
@@ -274,9 +274,6 @@ extern int amdgpu_vcnfw_log;
 #define AMDGPU_RESET_VCE			(1 << 13)
 #define AMDGPU_RESET_VCE1			(1 << 14)
 
-#define AMDGPU_RESET_LEVEL_SOFT_RECOVERY (1 << 0)
-#define AMDGPU_RESET_LEVEL_MODE2 (1 << 1)
-
 /* max cursor sizes (in pixels) */
 #define CIK_CURSOR_WIDTH 128
 #define CIK_CURSOR_HEIGHT 128
@@ -1065,7 +1062,6 @@ struct amdgpu_device {
 
 	struct work_struct		reset_work;
 
-	uint32_t						amdgpu_reset_level_mask;
 	bool                            job_hang;
 };
 
drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c +0 -2
index 6066aebf491c..de61a85c4b02 100644
@@ -1954,8 +1954,6 @@ int amdgpu_debugfs_init(struct amdgpu_device *adev)
 		return PTR_ERR(ent);
 	}
 
-	debugfs_create_u32("amdgpu_reset_level", 0600, root, &adev->amdgpu_reset_level_mask);
-
 	/* Register debugfs entries for amdgpu_ttm */
 	amdgpu_ttm_debugfs_init(adev);
 	amdgpu_debugfs_pm_init(adev);
drivers/gpu/drm/amd/amdgpu/amdgpu_reset.c +0 -8
index 9da5ead50c90..831fb222139c 100644
@@ -37,8 +37,6 @@ int amdgpu_reset_init(struct amdgpu_device *adev)
 {
 	int ret = 0;
 
-	adev->amdgpu_reset_level_mask = 0x1;
-
 	switch (adev->ip_versions[MP1_HWIP][0]) {
 	case IP_VERSION(13, 0, 2):
 		ret = aldebaran_reset_init(adev);
@@ -76,9 +74,6 @@ int amdgpu_reset_prepare_hwcontext(struct amdgpu_device *adev,
 {
 	struct amdgpu_reset_handler *reset_handler = NULL;
 
-	if (!(adev->amdgpu_reset_level_mask & AMDGPU_RESET_LEVEL_MODE2))
-		return -ENOSYS;
-
 	if (test_bit(AMDGPU_SKIP_MODE2_RESET, &reset_context->flags))
 		return -ENOSYS;
 
@@ -98,9 +93,6 @@ int amdgpu_reset_perform_reset(struct amdgpu_device *adev,
 	int ret;
 	struct amdgpu_reset_handler *reset_handler = NULL;
 
-	if (!(adev->amdgpu_reset_level_mask & AMDGPU_RESET_LEVEL_MODE2))
-		return -ENOSYS;
-
 	if (test_bit(AMDGPU_SKIP_MODE2_RESET, &reset_context->flags))
 		return -ENOSYS;
 
drivers/gpu/drm/amd/amdgpu/amdgpu_ring.c +0 -3
index 3e316b013fd9..d3558c34d406 100644
@@ -405,9 +405,6 @@ bool amdgpu_ring_soft_recovery(struct amdgpu_ring *ring, unsigned int vmid,
 {
 	ktime_t deadline = ktime_add_us(ktime_get(), 10000);
 
-	if (!(ring->adev->amdgpu_reset_level_mask & AMDGPU_RESET_LEVEL_SOFT_RECOVERY))
-		return false;
-
 	if (amdgpu_sriov_vf(ring->adev) || !ring->funcs->soft_recovery || !fence)
 		return false;
 
Signature · 1 line
2.25.1
#1 lore raw

PATCH 2/3 Revert "drm/amdgpu: let mode2 reset fallback to default when failure"

This reverts commit 3efc702897c54c95c332632157ab042e942512c7.

This commit reverted the AMDGPU_SKIP_MODE2_RESET as it conflicts with the original design of reset handler. Will redesign it.

---
 drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c | 1 -
 drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 7 +------
 drivers/gpu/drm/amd/amdgpu/amdgpu_job.c    | 1 -
 drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c    | 1 -
 drivers/gpu/drm/amd/amdgpu/amdgpu_reset.c  | 6 ------
 drivers/gpu/drm/amd/amdgpu/amdgpu_reset.h  | 3 +--
 drivers/gpu/drm/amd/amdgpu/mxgpu_ai.c      | 1 -
 drivers/gpu/drm/amd/amdgpu/mxgpu_nv.c      | 1 -
 drivers/gpu/drm/amd/amdgpu/mxgpu_vi.c      | 1 -
 9 files changed, 2 insertions(+), 20 deletions(-)
drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c +0 -1
index 03bbfaa51cbc..0561812aa0a4 100644
@@ -134,7 +134,6 @@ static void amdgpu_amdkfd_reset_work(struct work_struct *work)
 	reset_context.method = AMD_RESET_METHOD_NONE;
 	reset_context.reset_req_dev = adev;
 	clear_bit(AMDGPU_NEED_FULL_RESET, &reset_context.flags);
-	clear_bit(AMDGPU_SKIP_MODE2_RESET, &reset_context.flags);
 
 	amdgpu_device_gpu_recover(adev, NULL, &reset_context);
 }
drivers/gpu/drm/amd/amdgpu/amdgpu_device.c +1 -6
index ab8f970b2849..bb73fb420ffc 100644
@@ -5210,7 +5210,6 @@ int amdgpu_device_gpu_recover(struct amdgpu_device *adev,
 
 	reset_context->job = job;
 	reset_context->hive = hive;
-
 	/*
 	 * Build list of devices to reset.
 	 * In case we are in XGMI hive mode, resort the device list
@@ -5337,11 +5336,8 @@ int amdgpu_device_gpu_recover(struct amdgpu_device *adev,
 			amdgpu_ras_resume(adev);
 	} else {
 		r = amdgpu_do_asic_reset(device_list_handle, reset_context);
-		if (r && r == -EAGAIN) {
-			set_bit(AMDGPU_SKIP_MODE2_RESET, &reset_context->flags);
-			adev->asic_reset_res = 0;
+		if (r && r == -EAGAIN)
 			goto retry;
-		}
 
 		if (!r && gpu_reset_for_dev_remove)
 			goto recover_end;
@@ -5777,7 +5773,6 @@ pci_ers_result_t amdgpu_pci_slot_reset(struct pci_dev *pdev)
 	reset_context.reset_req_dev = adev;
 	set_bit(AMDGPU_NEED_FULL_RESET, &reset_context.flags);
 	set_bit(AMDGPU_SKIP_HW_RESET, &reset_context.flags);
-	set_bit(AMDGPU_SKIP_MODE2_RESET, &reset_context.flags);
 
 	adev->no_hw_access = true;
 	r = amdgpu_device_pre_asic_reset(adev, &reset_context);
drivers/gpu/drm/amd/amdgpu/amdgpu_job.c +0 -1
index 46c99331d7f1..cd968e781077 100644
@@ -72,7 +72,6 @@ static enum drm_gpu_sched_stat amdgpu_job_timedout(struct drm_sched_job *s_job)
 		reset_context.method = AMD_RESET_METHOD_NONE;
 		reset_context.reset_req_dev = adev;
 		clear_bit(AMDGPU_NEED_FULL_RESET, &reset_context.flags);
-		clear_bit(AMDGPU_SKIP_MODE2_RESET, &reset_context.flags);
 
 		r = amdgpu_device_gpu_recover(ring->adev, job, &reset_context);
 		if (r)
drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c +0 -1
index 2dad7aa9a03b..75f1402101f4 100644
@@ -1950,7 +1950,6 @@ static void amdgpu_ras_do_recovery(struct work_struct *work)
 		reset_context.method = AMD_RESET_METHOD_NONE;
 		reset_context.reset_req_dev = adev;
 		clear_bit(AMDGPU_NEED_FULL_RESET, &reset_context.flags);
-		clear_bit(AMDGPU_SKIP_MODE2_RESET, &reset_context.flags);
 
 		amdgpu_device_gpu_recover(ras->adev, NULL, &reset_context);
 	}
drivers/gpu/drm/amd/amdgpu/amdgpu_reset.c +0 -6
index 831fb222139c..f778466bb9db 100644
@@ -74,9 +74,6 @@ int amdgpu_reset_prepare_hwcontext(struct amdgpu_device *adev,
 {
 	struct amdgpu_reset_handler *reset_handler = NULL;
 
-	if (test_bit(AMDGPU_SKIP_MODE2_RESET, &reset_context->flags))
-		return -ENOSYS;
-
 	if (adev->reset_cntl && adev->reset_cntl->get_reset_handler)
 		reset_handler = adev->reset_cntl->get_reset_handler(
 			adev->reset_cntl, reset_context);
@@ -93,9 +90,6 @@ int amdgpu_reset_perform_reset(struct amdgpu_device *adev,
 	int ret;
 	struct amdgpu_reset_handler *reset_handler = NULL;
 
-	if (test_bit(AMDGPU_SKIP_MODE2_RESET, &reset_context->flags))
-		return -ENOSYS;
-
 	if (adev->reset_cntl)
 		reset_handler = adev->reset_cntl->get_reset_handler(
 			adev->reset_cntl, reset_context);
drivers/gpu/drm/amd/amdgpu/amdgpu_reset.h +1 -2
index f5318fedf2f0..f4a501ff87d9 100644
@@ -30,8 +30,7 @@ enum AMDGPU_RESET_FLAGS {
 
 	AMDGPU_NEED_FULL_RESET = 0,
 	AMDGPU_SKIP_HW_RESET = 1,
-	AMDGPU_SKIP_MODE2_RESET = 2,
-	AMDGPU_RESET_FOR_DEVICE_REMOVE = 3,
+	AMDGPU_RESET_FOR_DEVICE_REMOVE = 2,
 };
 
 struct amdgpu_reset_context {
drivers/gpu/drm/amd/amdgpu/mxgpu_ai.c +0 -1
index a2f04b249132..12906ba74462 100644
@@ -290,7 +290,6 @@ static void xgpu_ai_mailbox_flr_work(struct work_struct *work)
 		reset_context.method = AMD_RESET_METHOD_NONE;
 		reset_context.reset_req_dev = adev;
 		clear_bit(AMDGPU_NEED_FULL_RESET, &reset_context.flags);
-		clear_bit(AMDGPU_SKIP_MODE2_RESET, &reset_context.flags);
 
 		amdgpu_device_gpu_recover(adev, NULL, &reset_context);
 	}
drivers/gpu/drm/amd/amdgpu/mxgpu_nv.c +0 -1
index a977f0027928..e07757eea7ad 100644
@@ -317,7 +317,6 @@ static void xgpu_nv_mailbox_flr_work(struct work_struct *work)
 		reset_context.method = AMD_RESET_METHOD_NONE;
 		reset_context.reset_req_dev = adev;
 		clear_bit(AMDGPU_NEED_FULL_RESET, &reset_context.flags);
-		clear_bit(AMDGPU_SKIP_MODE2_RESET, &reset_context.flags);
 
 		amdgpu_device_gpu_recover(adev, NULL, &reset_context);
 	}
drivers/gpu/drm/amd/amdgpu/mxgpu_vi.c +0 -1
index fd14fa9b9cd7..288c414babdf 100644
@@ -529,7 +529,6 @@ static void xgpu_vi_mailbox_flr_work(struct work_struct *work)
 		reset_context.method = AMD_RESET_METHOD_NONE;
 		reset_context.reset_req_dev = adev;
 		clear_bit(AMDGPU_NEED_FULL_RESET, &reset_context.flags);
-		clear_bit(AMDGPU_SKIP_MODE2_RESET, &reset_context.flags);
 
 		amdgpu_device_gpu_recover(adev, NULL, &reset_context);
 	}
Signature · 1 line
2.25.1
#2 lore raw

PATCH 3/3 drm/amdgpu: Refactor mode2 reset logic for v11.0.7

In reply to Victor Zhao

- refactor mode2 on v11.0.7 to align with aldebaran
- comment out using mode2 reset as default for now, will introduce another controller to replace previous reset_level_mask

---
 drivers/gpu/drm/amd/amdgpu/sienna_cichlid.c | 23 ++++++++++++++-------
 1 file changed, 16 insertions(+), 7 deletions(-)
drivers/gpu/drm/amd/amdgpu/sienna_cichlid.c +16 -7
index 7aa570c1ce4a..a096f0b740ce 100644
@@ -31,6 +31,17 @@
 #include "amdgpu_psp.h"
 #include "amdgpu_xgmi.h"
 
+static bool sienna_cichlid_is_mode2_default(struct amdgpu_reset_control *reset_ctl)
+{
+	struct amdgpu_device *adev = (struct amdgpu_device *)reset_ctl->handle;
+#if 0
+	if (adev->ip_versions[MP1_HWIP][0] == IP_VERSION(11, 0, 7) &&
+	    adev->pm.fw_version >= 0x3a5500 && !amdgpu_sriov_vf(adev))
+		return true;
+#endif
+	return false;
+}
+
 static struct amdgpu_reset_handler *
 sienna_cichlid_get_reset_handler(struct amdgpu_reset_control *reset_ctl,
 			    struct amdgpu_reset_context *reset_context)
@@ -44,15 +55,13 @@ sienna_cichlid_get_reset_handler(struct amdgpu_reset_control *reset_ctl,
 			if (handler->reset_method == reset_context->method)
 				return handler;
 		}
-	} else {
-		list_for_each_entry(handler, &reset_ctl->reset_handlers,
+	}
+
+	if (sienna_cichlid_is_mode2_default(reset_ctl)) {
+		list_for_each_entry (handler, &reset_ctl->reset_handlers,
 				     handler_list) {
-			if (handler->reset_method == AMD_RESET_METHOD_MODE2 &&
-			    adev->pm.fw_version >= 0x3a5500 &&
-			    !amdgpu_sriov_vf(adev)) {
-				reset_context->method = AMD_RESET_METHOD_MODE2;
+			if (handler->reset_method == AMD_RESET_METHOD_MODE2)
 				return handler;
-			}
 		}
 	}
 
Signature · 1 line
2.25.1
Reviewed-by on v1 1/3
#3 lore raw
In reply to Victor Zhao
Quoted 4 lines · Victor Zhao

This reverts commit 3ae992d5e1194a16e3d977076eb5722fa6e410d8.

This commit breaks the reset logic for aldebaran, revert it for now. Will move the mask inside the reset handler.

Thanks for reverting. Please cc me also for refactored patches for SMU 11.0.7.

Series is

Thanks,
Lijo

Quoted 89 lines
---
  drivers/gpu/drm/amd/amdgpu/amdgpu.h         | 4 ----
  drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c | 2 --
  drivers/gpu/drm/amd/amdgpu/amdgpu_reset.c   | 8 --------
  drivers/gpu/drm/amd/amdgpu/amdgpu_ring.c    | 3 ---
  4 files changed, 17 deletions(-)
drivers/gpu/drm/amd/amdgpu/amdgpu.h +0 -4
index 3ce91f660c3f..0e6ddf05c23c 100644
@@ -274,9 +274,6 @@ extern int amdgpu_vcnfw_log;
  #define AMDGPU_RESET_VCE			(1 << 13)
  #define AMDGPU_RESET_VCE1			(1 << 14)
 
-#define AMDGPU_RESET_LEVEL_SOFT_RECOVERY (1 << 0)
-#define AMDGPU_RESET_LEVEL_MODE2 (1 << 1)
-
  /* max cursor sizes (in pixels) */
  #define CIK_CURSOR_WIDTH 128
  #define CIK_CURSOR_HEIGHT 128
@@ -1065,7 +1062,6 @@ struct amdgpu_device {
 
  	struct work_struct		reset_work;
 
-	uint32_t						amdgpu_reset_level_mask;
  	bool                            job_hang;
  };
 
drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c +0 -2
index 6066aebf491c..de61a85c4b02 100644
@@ -1954,8 +1954,6 @@ int amdgpu_debugfs_init(struct amdgpu_device *adev)
  		return PTR_ERR(ent);
  	}
 
-	debugfs_create_u32("amdgpu_reset_level", 0600, root, &adev->amdgpu_reset_level_mask);
-
  	/* Register debugfs entries for amdgpu_ttm */
  	amdgpu_ttm_debugfs_init(adev);
  	amdgpu_debugfs_pm_init(adev);
drivers/gpu/drm/amd/amdgpu/amdgpu_reset.c +0 -8
index 9da5ead50c90..831fb222139c 100644
@@ -37,8 +37,6 @@ int amdgpu_reset_init(struct amdgpu_device *adev)
  {
  	int ret = 0;
 
-	adev->amdgpu_reset_level_mask = 0x1;
-
  	switch (adev->ip_versions[MP1_HWIP][0]) {
  	case IP_VERSION(13, 0, 2):
  		ret = aldebaran_reset_init(adev);
@@ -76,9 +74,6 @@ int amdgpu_reset_prepare_hwcontext(struct amdgpu_device *adev,
  {
  	struct amdgpu_reset_handler *reset_handler = NULL;
 
-	if (!(adev->amdgpu_reset_level_mask & AMDGPU_RESET_LEVEL_MODE2))
-		return -ENOSYS;
-
  	if (test_bit(AMDGPU_SKIP_MODE2_RESET, &reset_context->flags))
  		return -ENOSYS;
 
@@ -98,9 +93,6 @@ int amdgpu_reset_perform_reset(struct amdgpu_device *adev,
  	int ret;
  	struct amdgpu_reset_handler *reset_handler = NULL;
 
-	if (!(adev->amdgpu_reset_level_mask & AMDGPU_RESET_LEVEL_MODE2))
-		return -ENOSYS;
-
  	if (test_bit(AMDGPU_SKIP_MODE2_RESET, &reset_context->flags))
  		return -ENOSYS;
 
drivers/gpu/drm/amd/amdgpu/amdgpu_ring.c +0 -3
index 3e316b013fd9..d3558c34d406 100644
@@ -405,9 +405,6 @@ bool amdgpu_ring_soft_recovery(struct amdgpu_ring *ring, unsigned int vmid,
  {
  	ktime_t deadline = ktime_add_us(ktime_get(), 10000);
 
-	if (!(ring->adev->amdgpu_reset_level_mask & AMDGPU_RESET_LEVEL_SOFT_RECOVERY))
-		return false;
-
  	if (amdgpu_sriov_vf(ring->adev) || !ring->funcs->soft_recovery || !fence)
  		return false;
 
#4 lore raw
In reply to Victor Zhao

[Public]

This patch is missing your signed-off-by.

Alex

________________________________
From: amd-gfx <[email protected]> on behalf of Victor Zhao <[email protected]>
Sent: Thursday, October 13, 2022 4:27 AM
To: [email protected] <[email protected]>
Cc: Grodzovsky, Andrey <[email protected]>; Lazar, Lijo <[email protected]>; Zhao, Victor <[email protected]>
Subject: [PATCH 1/3] Revert "drm/amdgpu: add debugfs amdgpu_reset_level"

This reverts commit 3ae992d5e1194a16e3d977076eb5722fa6e410d8.

This commit breaks the reset logic for aldebaran, revert it for now. Will move the mask inside the reset handler.

---
 drivers/gpu/drm/amd/amdgpu/amdgpu.h         | 4 ----
 drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c | 2 --
 drivers/gpu/drm/amd/amdgpu/amdgpu_reset.c   | 8 --------
 drivers/gpu/drm/amd/amdgpu/amdgpu_ring.c    | 3 ---
 4 files changed, 17 deletions(-)
drivers/gpu/drm/amd/amdgpu/amdgpu.h +0 -4
index 3ce91f660c3f..0e6ddf05c23c 100644
@@ -274,9 +274,6 @@ extern int amdgpu_vcnfw_log;
 #define AMDGPU_RESET_VCE                        (1 << 13)
 #define AMDGPU_RESET_VCE1                       (1 << 14)
 
-#define AMDGPU_RESET_LEVEL_SOFT_RECOVERY (1 << 0)
-#define AMDGPU_RESET_LEVEL_MODE2 (1 << 1)
-
 /* max cursor sizes (in pixels) */
 #define CIK_CURSOR_WIDTH 128
 #define CIK_CURSOR_HEIGHT 128
@@ -1065,7 +1062,6 @@ struct amdgpu_device {
 
         struct work_struct              reset_work;
 
-       uint32_t                                                amdgpu_reset_level_mask;
         bool                            job_hang;
 };
 
drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c +0 -2
index 6066aebf491c..de61a85c4b02 100644
@@ -1954,8 +1954,6 @@ int amdgpu_debugfs_init(struct amdgpu_device *adev)
                 return PTR_ERR(ent);
         }
 
-       debugfs_create_u32("amdgpu_reset_level", 0600, root, &adev->amdgpu_reset_level_mask);
-
         /* Register debugfs entries for amdgpu_ttm */
         amdgpu_ttm_debugfs_init(adev);
         amdgpu_debugfs_pm_init(adev);
drivers/gpu/drm/amd/amdgpu/amdgpu_reset.c +0 -8
index 9da5ead50c90..831fb222139c 100644
@@ -37,8 +37,6 @@ int amdgpu_reset_init(struct amdgpu_device *adev)
 {
         int ret = 0;
 
-       adev->amdgpu_reset_level_mask = 0x1;
-
         switch (adev->ip_versions[MP1_HWIP][0]) {
         case IP_VERSION(13, 0, 2):
                 ret = aldebaran_reset_init(adev);
@@ -76,9 +74,6 @@ int amdgpu_reset_prepare_hwcontext(struct amdgpu_device *adev,
 {
         struct amdgpu_reset_handler *reset_handler = NULL;
 
-       if (!(adev->amdgpu_reset_level_mask & AMDGPU_RESET_LEVEL_MODE2))
-               return -ENOSYS;
-
         if (test_bit(AMDGPU_SKIP_MODE2_RESET, &reset_context->flags))
                 return -ENOSYS;
 
@@ -98,9 +93,6 @@ int amdgpu_reset_perform_reset(struct amdgpu_device *adev,
         int ret;
         struct amdgpu_reset_handler *reset_handler = NULL;
 
-       if (!(adev->amdgpu_reset_level_mask & AMDGPU_RESET_LEVEL_MODE2))
-               return -ENOSYS;
-
         if (test_bit(AMDGPU_SKIP_MODE2_RESET, &reset_context->flags))
                 return -ENOSYS;
 
drivers/gpu/drm/amd/amdgpu/amdgpu_ring.c +0 -3
index 3e316b013fd9..d3558c34d406 100644
@@ -405,9 +405,6 @@ bool amdgpu_ring_soft_recovery(struct amdgpu_ring *ring, unsigned int vmid,
 {
         ktime_t deadline = ktime_add_us(ktime_get(), 10000);
 
-       if (!(ring->adev->amdgpu_reset_level_mask & AMDGPU_RESET_LEVEL_SOFT_RECOVERY))
-               return false;
-
         if (amdgpu_sriov_vf(ring->adev) || !ring->funcs->soft_recovery || !fence)
                 return false;
 

--
2.25.1