[PATCH] drm/amdgpu: Refactor mode2 reset logic for v13.0.2

Ma, Le Le.Ma at amd.com
Thu Mar 3 03:39:34 UTC 2022


Reviewed-by: Le Ma <Le.Ma at amd.com>

> -----Original Message-----
> From: amd-gfx <amd-gfx-bounces at lists.freedesktop.org> On Behalf Of Lazar,
> Lijo
> Sent: Thursday, March 3, 2022 11:33 AM
> To: Lazar, Lijo <Lijo.Lazar at amd.com>; amd-gfx at lists.freedesktop.org
> Cc: Deucher, Alexander <Alexander.Deucher at amd.com>; Zhang, Hawking
> <Hawking.Zhang at amd.com>
> Subject: RE: [PATCH] drm/amdgpu: Refactor mode2 reset logic for v13.0.2
> 
> [Public]
> 
> <Ping>
> 
> Thanks,
> Lijo
> 
> -----Original Message-----
> From: amd-gfx <amd-gfx-bounces at lists.freedesktop.org> On Behalf Of Lijo
> Lazar
> Sent: Monday, February 28, 2022 3:27 PM
> To: amd-gfx at lists.freedesktop.org
> Cc: Deucher, Alexander <Alexander.Deucher at amd.com>; Zhang, Hawking
> <Hawking.Zhang at amd.com>
> Subject: [PATCH] drm/amdgpu: Refactor mode2 reset logic for v13.0.2
> 
> Use IP version and refactor reset logic to apply to a list of devices.
> 
> Signed-off-by: Lijo Lazar <lijo.lazar at amd.com>
> Reviewed-by: Hawking Zhang <Hawking.Zhang at amd.com>
> ---
>  drivers/gpu/drm/amd/amdgpu/aldebaran.c    | 66 +++++++++++++++++------
>  drivers/gpu/drm/amd/amdgpu/amdgpu_reset.c |  8 +--
>  2 files changed, 54 insertions(+), 20 deletions(-)
> 
> diff --git a/drivers/gpu/drm/amd/amdgpu/aldebaran.c
> b/drivers/gpu/drm/amd/amdgpu/aldebaran.c
> index a545df4efce1..c6cc493a5486 100644
> --- a/drivers/gpu/drm/amd/amdgpu/aldebaran.c
> +++ b/drivers/gpu/drm/amd/amdgpu/aldebaran.c
> @@ -31,6 +31,17 @@
>  #include "amdgpu_psp.h"
>  #include "amdgpu_xgmi.h"
> 
> +static bool aldebaran_is_mode2_default(struct amdgpu_reset_control
> +*reset_ctl) {
> +	struct amdgpu_device *adev = (struct amdgpu_device
> +*)reset_ctl->handle;
> +
> +	if ((adev->ip_versions[MP1_HWIP][0] == IP_VERSION(13, 0, 2) &&
> +	     adev->gmc.xgmi.connected_to_cpu))
> +		return true;
> +
> +	return false;
> +}
> +
>  static struct amdgpu_reset_handler *
>  aldebaran_get_reset_handler(struct amdgpu_reset_control *reset_ctl,
>  			    struct amdgpu_reset_context *reset_context) @@ -
> 48,7 +59,7 @@ aldebaran_get_reset_handler(struct amdgpu_reset_control
> *reset_ctl,
>  		}
>  	}
> 
> -	if (adev->gmc.xgmi.connected_to_cpu) {
> +	if (aldebaran_is_mode2_default(reset_ctl)) {
>  		list_for_each_entry(handler, &reset_ctl->reset_handlers,
>  				     handler_list) {
>  			if (handler->reset_method ==
> AMD_RESET_METHOD_MODE2) { @@ -136,18 +147,31 @@ static int
> aldebaran_mode2_perform_reset(struct amdgpu_reset_control *reset_ctl,
>  			      struct amdgpu_reset_context *reset_context)  {
> -	struct amdgpu_device *tmp_adev = NULL;
>  	struct amdgpu_device *adev = (struct amdgpu_device *)reset_ctl-
> >handle;
> +	struct amdgpu_device *tmp_adev = NULL;
> +	struct list_head reset_device_list;
>  	int r = 0;
> 
>  	dev_dbg(adev->dev, "aldebaran perform hw reset\n");
> -	if (reset_context->hive == NULL) {
> +	if (adev->ip_versions[MP1_HWIP][0] == IP_VERSION(13, 0, 2) &&
> +	    reset_context->hive == NULL) {
>  		/* Wrong context, return error */
>  		return -EINVAL;
>  	}
> 
> -	list_for_each_entry(tmp_adev, &reset_context->hive->device_list,
> -			     gmc.xgmi.head) {
> +	INIT_LIST_HEAD(&reset_device_list);
> +	if (reset_context->hive) {
> +		list_for_each_entry (tmp_adev,
> +				     &reset_context->hive->device_list,
> +				     gmc.xgmi.head)
> +			list_add_tail(&tmp_adev->reset_list,
> +				      &reset_device_list);
> +	} else {
> +		list_add_tail(&reset_context->reset_req_dev->reset_list,
> +			      &reset_device_list);
> +	}
> +
> +	list_for_each_entry (tmp_adev, &reset_device_list, reset_list) {
>  		mutex_lock(&tmp_adev->reset_cntl->reset_lock);
>  		tmp_adev->reset_cntl->active_reset =
> AMD_RESET_METHOD_MODE2;
>  	}
> @@ -155,8 +179,7 @@ aldebaran_mode2_perform_reset(struct
> amdgpu_reset_control *reset_ctl,
>  	 * Mode2 reset doesn't need any sync between nodes in XGMI hive,
> instead launch
>  	 * them together so that they can be completed asynchronously on
> multiple nodes
>  	 */
> -	list_for_each_entry(tmp_adev, &reset_context->hive->device_list,
> -			     gmc.xgmi.head) {
> +	list_for_each_entry (tmp_adev, &reset_device_list, reset_list) {
>  		/* For XGMI run all resets in parallel to speed up the process */
>  		if (tmp_adev->gmc.xgmi.num_physical_nodes > 1) {
>  			if (!queue_work(system_unbound_wq,
> @@ -174,9 +197,7 @@ aldebaran_mode2_perform_reset(struct
> amdgpu_reset_control *reset_ctl,
> 
>  	/* For XGMI wait for all resets to complete before proceed */
>  	if (!r) {
> -		list_for_each_entry(tmp_adev,
> -				     &reset_context->hive->device_list,
> -				     gmc.xgmi.head) {
> +		list_for_each_entry (tmp_adev, &reset_device_list, reset_list) {
>  			if (tmp_adev->gmc.xgmi.num_physical_nodes > 1) {
>  				flush_work(&tmp_adev->reset_cntl-
> >reset_work);
>  				r = tmp_adev->asic_reset_res;
> @@ -186,8 +207,7 @@ aldebaran_mode2_perform_reset(struct
> amdgpu_reset_control *reset_ctl,
>  		}
>  	}
> 
> -	list_for_each_entry(tmp_adev, &reset_context->hive->device_list,
> -			     gmc.xgmi.head) {
> +	list_for_each_entry (tmp_adev, &reset_device_list, reset_list) {
>  		mutex_unlock(&tmp_adev->reset_cntl->reset_lock);
>  		tmp_adev->reset_cntl->active_reset =
> AMD_RESET_METHOD_NONE;
>  	}
> @@ -319,16 +339,30 @@ static int
>  aldebaran_mode2_restore_hwcontext(struct amdgpu_reset_control *reset_ctl,
>  				  struct amdgpu_reset_context *reset_context)
> {
> -	int r;
>  	struct amdgpu_device *tmp_adev = NULL;
> +	struct list_head reset_device_list;
> +	int r;
> 
> -	if (reset_context->hive == NULL) {
> +	if (reset_context->reset_req_dev->ip_versions[MP1_HWIP][0] ==
> +		    IP_VERSION(13, 0, 2) &&
> +	    reset_context->hive == NULL) {
>  		/* Wrong context, return error */
>  		return -EINVAL;
>  	}
> 
> -	list_for_each_entry(tmp_adev, &reset_context->hive->device_list,
> -			     gmc.xgmi.head) {
> +	INIT_LIST_HEAD(&reset_device_list);
> +	if (reset_context->hive) {
> +		list_for_each_entry (tmp_adev,
> +				     &reset_context->hive->device_list,
> +				     gmc.xgmi.head)
> +			list_add_tail(&tmp_adev->reset_list,
> +				      &reset_device_list);
> +	} else {
> +		list_add_tail(&reset_context->reset_req_dev->reset_list,
> +			      &reset_device_list);
> +	}
> +
> +	list_for_each_entry (tmp_adev, &reset_device_list, reset_list) {
>  		dev_info(tmp_adev->dev,
>  			 "GPU reset succeeded, trying to resume\n");
>  		r = aldebaran_mode2_restore_ip(tmp_adev);
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_reset.c
> b/drivers/gpu/drm/amd/amdgpu/amdgpu_reset.c
> index 02afd4115675..e9b9ce80f7d0 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_reset.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_reset.c
> @@ -36,8 +36,8 @@ int amdgpu_reset_init(struct amdgpu_device *adev)  {
>  	int ret = 0;
> 
> -	switch (adev->asic_type) {
> -	case CHIP_ALDEBARAN:
> +	switch (adev->ip_versions[MP1_HWIP][0]) {
> +	case IP_VERSION(13, 0, 2):
>  		ret = aldebaran_reset_init(adev);
>  		break;
>  	default:
> @@ -51,8 +51,8 @@ int amdgpu_reset_fini(struct amdgpu_device *adev)  {
>  	int ret = 0;
> 
> -	switch (adev->asic_type) {
> -	case CHIP_ALDEBARAN:
> +	switch (adev->ip_versions[MP1_HWIP][0]) {
> +	case IP_VERSION(13, 0, 2):
>  		ret = aldebaran_reset_fini(adev);
>  		break;
>  	default:
> --
> 2.25.1


More information about the amd-gfx mailing list