* [PATCH] drm/amdgpu: Use driver mode reset for data poison handling
@ 2024-04-16 4:34 Hawking Zhang
2024-04-16 5:08 ` Zhou1, Tao
0 siblings, 1 reply; 5+ messages in thread
From: Hawking Zhang @ 2024-04-16 4:34 UTC (permalink / raw)
To: amd-gfx, Tao Zhou; +Cc: Hawking Zhang
mode-2 reset is the only reliable method that can get
GC/SDMA back when poison is consumed. mmhub requires
mode-1 reset.
Signed-off-by: Hawking Zhang <Hawking.Zhang@amd.com>
---
drivers/gpu/drm/amd/amdkfd/kfd_int_process_v9.c | 8 ++------
1 file changed, 2 insertions(+), 6 deletions(-)
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_int_process_v9.c b/drivers/gpu/drm/amd/amdkfd/kfd_int_process_v9.c
index c368c70df3f4a..b6caf6eda8a0c 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_int_process_v9.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_int_process_v9.c
@@ -163,17 +163,13 @@ static void event_interrupt_poison_consumption_v9(struct kfd_node *dev,
case SOC15_IH_CLIENTID_SE2SH:
case SOC15_IH_CLIENTID_SE3SH:
case SOC15_IH_CLIENTID_UTCL2:
- ret = kfd_dqm_evict_pasid(dev->dqm, pasid);
block = AMDGPU_RAS_BLOCK__GFX;
- if (ret)
- reset = AMDGPU_RAS_GPU_RESET_MODE2_RESET;
+ reset = AMDGPU_RAS_GPU_RESET_MODE2_RESET;
break;
case SOC15_IH_CLIENTID_VMC:
case SOC15_IH_CLIENTID_VMC1:
- ret = kfd_dqm_evict_pasid(dev->dqm, pasid);
block = AMDGPU_RAS_BLOCK__MMHUB;
- if (ret)
- reset = AMDGPU_RAS_GPU_RESET_MODE1_RESET;
+ reset = AMDGPU_RAS_GPU_RESET_MODE1_RESET;
break;
case SOC15_IH_CLIENTID_SDMA0:
case SOC15_IH_CLIENTID_SDMA1:
--
2.17.1
^ permalink raw reply related [flat|nested] 5+ messages in thread
* RE: [PATCH] drm/amdgpu: Use driver mode reset for data poison handling
2024-04-16 4:34 [PATCH] drm/amdgpu: Use driver mode reset for data poison handling Hawking Zhang
@ 2024-04-16 5:08 ` Zhou1, Tao
2024-04-16 5:51 ` Zhang, Hawking
0 siblings, 1 reply; 5+ messages in thread
From: Zhou1, Tao @ 2024-04-16 5:08 UTC (permalink / raw)
To: Zhang, Hawking, amd-gfx; +Cc: Zhang, Hawking
[AMD Official Use Only - General]
Reviewed-by: Tao Zhou <tao.zhou1@amd.com>
> -----Original Message-----
> From: Hawking Zhang <Hawking.Zhang@amd.com>
> Sent: Tuesday, April 16, 2024 12:34 PM
> To: amd-gfx@lists.freedesktop.org; Zhou1, Tao <Tao.Zhou1@amd.com>
> Cc: Zhang, Hawking <Hawking.Zhang@amd.com>
> Subject: [PATCH] drm/amdgpu: Use driver mode reset for data poison handling
>
> mode-2 reset is the only reliable method that can get GC/SDMA back when
> poison is consumed. mmhub requires
> mode-1 reset.
>
> Signed-off-by: Hawking Zhang <Hawking.Zhang@amd.com>
> ---
> drivers/gpu/drm/amd/amdkfd/kfd_int_process_v9.c | 8 ++------
> 1 file changed, 2 insertions(+), 6 deletions(-)
>
> diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_int_process_v9.c
> b/drivers/gpu/drm/amd/amdkfd/kfd_int_process_v9.c
> index c368c70df3f4a..b6caf6eda8a0c 100644
> --- a/drivers/gpu/drm/amd/amdkfd/kfd_int_process_v9.c
> +++ b/drivers/gpu/drm/amd/amdkfd/kfd_int_process_v9.c
> @@ -163,17 +163,13 @@ static void
> event_interrupt_poison_consumption_v9(struct kfd_node *dev,
> case SOC15_IH_CLIENTID_SE2SH:
> case SOC15_IH_CLIENTID_SE3SH:
> case SOC15_IH_CLIENTID_UTCL2:
> - ret = kfd_dqm_evict_pasid(dev->dqm, pasid);
> block = AMDGPU_RAS_BLOCK__GFX;
> - if (ret)
> - reset = AMDGPU_RAS_GPU_RESET_MODE2_RESET;
> + reset = AMDGPU_RAS_GPU_RESET_MODE2_RESET;
> break;
> case SOC15_IH_CLIENTID_VMC:
> case SOC15_IH_CLIENTID_VMC1:
> - ret = kfd_dqm_evict_pasid(dev->dqm, pasid);
> block = AMDGPU_RAS_BLOCK__MMHUB;
> - if (ret)
> - reset = AMDGPU_RAS_GPU_RESET_MODE1_RESET;
> + reset = AMDGPU_RAS_GPU_RESET_MODE1_RESET;
> break;
> case SOC15_IH_CLIENTID_SDMA0:
> case SOC15_IH_CLIENTID_SDMA1:
> --
> 2.17.1
^ permalink raw reply [flat|nested] 5+ messages in thread
* RE: [PATCH] drm/amdgpu: Use driver mode reset for data poison handling
2024-04-16 5:08 ` Zhou1, Tao
@ 2024-04-16 5:51 ` Zhang, Hawking
0 siblings, 0 replies; 5+ messages in thread
From: Zhang, Hawking @ 2024-04-16 5:51 UTC (permalink / raw)
To: Zhou1, Tao, amd-gfx
[AMD Official Use Only - General]
Please ignore this one, will send out a new one
-----Original Message-----
From: Zhou1, Tao <Tao.Zhou1@amd.com>
Sent: Tuesday, April 16, 2024 01:08
To: Zhang, Hawking <Hawking.Zhang@amd.com>; amd-gfx@lists.freedesktop.org
Cc: Zhang, Hawking <Hawking.Zhang@amd.com>
Subject: RE: [PATCH] drm/amdgpu: Use driver mode reset for data poison handling
[AMD Official Use Only - General]
Reviewed-by: Tao Zhou <tao.zhou1@amd.com>
> -----Original Message-----
> From: Hawking Zhang <Hawking.Zhang@amd.com>
> Sent: Tuesday, April 16, 2024 12:34 PM
> To: amd-gfx@lists.freedesktop.org; Zhou1, Tao <Tao.Zhou1@amd.com>
> Cc: Zhang, Hawking <Hawking.Zhang@amd.com>
> Subject: [PATCH] drm/amdgpu: Use driver mode reset for data poison
> handling
>
> mode-2 reset is the only reliable method that can get GC/SDMA back
> when poison is consumed. mmhub requires
> mode-1 reset.
>
> Signed-off-by: Hawking Zhang <Hawking.Zhang@amd.com>
> ---
> drivers/gpu/drm/amd/amdkfd/kfd_int_process_v9.c | 8 ++------
> 1 file changed, 2 insertions(+), 6 deletions(-)
>
> diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_int_process_v9.c
> b/drivers/gpu/drm/amd/amdkfd/kfd_int_process_v9.c
> index c368c70df3f4a..b6caf6eda8a0c 100644
> --- a/drivers/gpu/drm/amd/amdkfd/kfd_int_process_v9.c
> +++ b/drivers/gpu/drm/amd/amdkfd/kfd_int_process_v9.c
> @@ -163,17 +163,13 @@ static void
> event_interrupt_poison_consumption_v9(struct kfd_node *dev,
> case SOC15_IH_CLIENTID_SE2SH:
> case SOC15_IH_CLIENTID_SE3SH:
> case SOC15_IH_CLIENTID_UTCL2:
> - ret = kfd_dqm_evict_pasid(dev->dqm, pasid);
> block = AMDGPU_RAS_BLOCK__GFX;
> - if (ret)
> - reset = AMDGPU_RAS_GPU_RESET_MODE2_RESET;
> + reset = AMDGPU_RAS_GPU_RESET_MODE2_RESET;
> break;
> case SOC15_IH_CLIENTID_VMC:
> case SOC15_IH_CLIENTID_VMC1:
> - ret = kfd_dqm_evict_pasid(dev->dqm, pasid);
> block = AMDGPU_RAS_BLOCK__MMHUB;
> - if (ret)
> - reset = AMDGPU_RAS_GPU_RESET_MODE1_RESET;
> + reset = AMDGPU_RAS_GPU_RESET_MODE1_RESET;
> break;
> case SOC15_IH_CLIENTID_SDMA0:
> case SOC15_IH_CLIENTID_SDMA1:
> --
> 2.17.1
^ permalink raw reply [flat|nested] 5+ messages in thread
* Re: [PATCH] drm/amdgpu: Use driver mode reset for data poison handling
2024-04-16 5:56 Hawking Zhang
@ 2024-04-17 18:59 ` Deucher, Alexander
0 siblings, 0 replies; 5+ messages in thread
From: Deucher, Alexander @ 2024-04-17 18:59 UTC (permalink / raw)
To: Zhang, Hawking, amd-gfx, Zhou1, Tao
[-- Attachment #1: Type: text/plain, Size: 3158 bytes --]
[Public]
Acked-by: Alex Deucher <alexander.deucher@amd.com>
________________________________
From: amd-gfx <amd-gfx-bounces@lists.freedesktop.org> on behalf of Hawking Zhang <Hawking.Zhang@amd.com>
Sent: Tuesday, April 16, 2024 1:56 AM
To: amd-gfx@lists.freedesktop.org <amd-gfx@lists.freedesktop.org>; Zhou1, Tao <Tao.Zhou1@amd.com>
Cc: Zhang, Hawking <Hawking.Zhang@amd.com>
Subject: [PATCH] drm/amdgpu: Use driver mode reset for data poison handling
mode-2 reset is the only reliable method that can get
GC/SDMA back when poison is consumed. mmhub requires
mode-1 reset.
Signed-off-by: Hawking Zhang <Hawking.Zhang@amd.com>
---
.../gpu/drm/amd/amdkfd/kfd_int_process_v9.c | 22 +++----------------
1 file changed, 3 insertions(+), 19 deletions(-)
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_int_process_v9.c b/drivers/gpu/drm/amd/amdkfd/kfd_int_process_v9.c
index c368c70df3f4a..94eb2493103ef 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_int_process_v9.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_int_process_v9.c
@@ -144,7 +144,7 @@ static void event_interrupt_poison_consumption_v9(struct kfd_node *dev,
uint16_t pasid, uint16_t client_id)
{
enum amdgpu_ras_block block = 0;
- int old_poison, ret = -EINVAL;
+ int old_poison;
uint32_t reset = 0;
struct kfd_process *p = kfd_lookup_process_by_pasid(pasid);
@@ -163,17 +163,13 @@ static void event_interrupt_poison_consumption_v9(struct kfd_node *dev,
case SOC15_IH_CLIENTID_SE2SH:
case SOC15_IH_CLIENTID_SE3SH:
case SOC15_IH_CLIENTID_UTCL2:
- ret = kfd_dqm_evict_pasid(dev->dqm, pasid);
block = AMDGPU_RAS_BLOCK__GFX;
- if (ret)
- reset = AMDGPU_RAS_GPU_RESET_MODE2_RESET;
+ reset = AMDGPU_RAS_GPU_RESET_MODE2_RESET;
break;
case SOC15_IH_CLIENTID_VMC:
case SOC15_IH_CLIENTID_VMC1:
- ret = kfd_dqm_evict_pasid(dev->dqm, pasid);
block = AMDGPU_RAS_BLOCK__MMHUB;
- if (ret)
- reset = AMDGPU_RAS_GPU_RESET_MODE1_RESET;
+ reset = AMDGPU_RAS_GPU_RESET_MODE1_RESET;
break;
case SOC15_IH_CLIENTID_SDMA0:
case SOC15_IH_CLIENTID_SDMA1:
@@ -189,18 +185,6 @@ static void event_interrupt_poison_consumption_v9(struct kfd_node *dev,
kfd_signal_poison_consumed_event(dev, pasid);
- /* resetting queue passes, do page retirement without gpu reset
- * resetting queue fails, fallback to gpu reset solution
- */
- if (!ret)
- dev_warn(dev->adev->dev,
- "RAS poison consumption, unmap queue flow succeeded: client id %d\n",
- client_id);
- else
- dev_warn(dev->adev->dev,
- "RAS poison consumption, fall back to gpu reset flow: client id %d\n",
- client_id);
-
amdgpu_amdkfd_ras_poison_consumption_handler(dev->adev, block, reset);
}
--
2.17.1
[-- Attachment #2: Type: text/html, Size: 6894 bytes --]
^ permalink raw reply related [flat|nested] 5+ messages in thread
* [PATCH] drm/amdgpu: Use driver mode reset for data poison handling
@ 2024-04-16 5:56 Hawking Zhang
2024-04-17 18:59 ` Deucher, Alexander
0 siblings, 1 reply; 5+ messages in thread
From: Hawking Zhang @ 2024-04-16 5:56 UTC (permalink / raw)
To: amd-gfx, Tao Zhou; +Cc: Hawking Zhang
mode-2 reset is the only reliable method that can get
GC/SDMA back when poison is consumed. mmhub requires
mode-1 reset.
Signed-off-by: Hawking Zhang <Hawking.Zhang@amd.com>
---
.../gpu/drm/amd/amdkfd/kfd_int_process_v9.c | 22 +++----------------
1 file changed, 3 insertions(+), 19 deletions(-)
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_int_process_v9.c b/drivers/gpu/drm/amd/amdkfd/kfd_int_process_v9.c
index c368c70df3f4a..94eb2493103ef 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_int_process_v9.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_int_process_v9.c
@@ -144,7 +144,7 @@ static void event_interrupt_poison_consumption_v9(struct kfd_node *dev,
uint16_t pasid, uint16_t client_id)
{
enum amdgpu_ras_block block = 0;
- int old_poison, ret = -EINVAL;
+ int old_poison;
uint32_t reset = 0;
struct kfd_process *p = kfd_lookup_process_by_pasid(pasid);
@@ -163,17 +163,13 @@ static void event_interrupt_poison_consumption_v9(struct kfd_node *dev,
case SOC15_IH_CLIENTID_SE2SH:
case SOC15_IH_CLIENTID_SE3SH:
case SOC15_IH_CLIENTID_UTCL2:
- ret = kfd_dqm_evict_pasid(dev->dqm, pasid);
block = AMDGPU_RAS_BLOCK__GFX;
- if (ret)
- reset = AMDGPU_RAS_GPU_RESET_MODE2_RESET;
+ reset = AMDGPU_RAS_GPU_RESET_MODE2_RESET;
break;
case SOC15_IH_CLIENTID_VMC:
case SOC15_IH_CLIENTID_VMC1:
- ret = kfd_dqm_evict_pasid(dev->dqm, pasid);
block = AMDGPU_RAS_BLOCK__MMHUB;
- if (ret)
- reset = AMDGPU_RAS_GPU_RESET_MODE1_RESET;
+ reset = AMDGPU_RAS_GPU_RESET_MODE1_RESET;
break;
case SOC15_IH_CLIENTID_SDMA0:
case SOC15_IH_CLIENTID_SDMA1:
@@ -189,18 +185,6 @@ static void event_interrupt_poison_consumption_v9(struct kfd_node *dev,
kfd_signal_poison_consumed_event(dev, pasid);
- /* resetting queue passes, do page retirement without gpu reset
- * resetting queue fails, fallback to gpu reset solution
- */
- if (!ret)
- dev_warn(dev->adev->dev,
- "RAS poison consumption, unmap queue flow succeeded: client id %d\n",
- client_id);
- else
- dev_warn(dev->adev->dev,
- "RAS poison consumption, fall back to gpu reset flow: client id %d\n",
- client_id);
-
amdgpu_amdkfd_ras_poison_consumption_handler(dev->adev, block, reset);
}
--
2.17.1
^ permalink raw reply related [flat|nested] 5+ messages in thread
end of thread, other threads:[~2024-04-17 19:00 UTC | newest]
Thread overview: 5+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2024-04-16 4:34 [PATCH] drm/amdgpu: Use driver mode reset for data poison handling Hawking Zhang
2024-04-16 5:08 ` Zhou1, Tao
2024-04-16 5:51 ` Zhang, Hawking
2024-04-16 5:56 Hawking Zhang
2024-04-17 18:59 ` Deucher, Alexander
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).