[Intel-xe] [PATCH v4 22/22] drm/xe: Initialize GuC earlier during probe

Welty, Brian brian.welty at intel.com
Wed Nov 29 21:48:50 UTC 2023


On 11/28/2023 5:16 PM, Michał Winiarski wrote:
> SR-IOV VF has limited access to MMIO registers. Fortunately, it is able
> to access a curated subset that is needed to initialize the driver by
> communicating with SR-IOV PF using GuC CT.
> Initialize GuC earlier in order to keep the unified probe ordering
> between VF and PF modes.
> 
> Signed-off-by: Michał Winiarski <michal.winiarski at intel.com>
> ---
> v3 -> v4:
> - Introduce xe_gt_init_hwconfig (Matt Brost)
> - Init PAT before loading GuC
> 
>   drivers/gpu/drm/xe/xe_device.c |  6 ++++
>   drivers/gpu/drm/xe/xe_gt.c     | 51 ++++++++++++++++++++++++----------
>   drivers/gpu/drm/xe/xe_gt.h     |  1 +
>   drivers/gpu/drm/xe/xe_uc.c     | 11 ++++++--
>   4 files changed, 52 insertions(+), 17 deletions(-)
> 
> diff --git a/drivers/gpu/drm/xe/xe_device.c b/drivers/gpu/drm/xe/xe_device.c
> index 65e9aa5e6c31e..a236c36cdae3c 100644
> --- a/drivers/gpu/drm/xe/xe_device.c
> +++ b/drivers/gpu/drm/xe/xe_device.c
> @@ -425,6 +425,12 @@ int xe_device_probe(struct xe_device *xe)
>   			return err;
>   	}
>   
> +	for_each_gt(gt, xe, id) {
> +		err = xe_gt_init_hwconfig(gt);
> +		if (err)
> +			return err;
> +	}
> +
>   	err = drmm_add_action_or_reset(&xe->drm, xe_driver_flr_fini, xe);
>   	if (err)
>   		return err;
> diff --git a/drivers/gpu/drm/xe/xe_gt.c b/drivers/gpu/drm/xe/xe_gt.c
> index 964e76f931a28..d2766c2c58f0f 100644
> --- a/drivers/gpu/drm/xe/xe_gt.c
> +++ b/drivers/gpu/drm/xe/xe_gt.c
> @@ -297,7 +297,6 @@ int xe_gt_init_early(struct xe_gt *gt)
>   		return err;
>   
>   	xe_gt_topology_init(gt);
> -	xe_gt_mcr_init(gt);
>   
>   	err = xe_force_wake_put(gt_to_fw(gt), XE_FW_GT);
>   	if (err)
> @@ -336,22 +335,12 @@ static int gt_fw_domain_init(struct xe_gt *gt)
>   	if (err)
>   		goto err_hw_fence_irq;
>   
> -	xe_pat_init(gt);
> -
>   	if (!xe_gt_is_media_type(gt)) {
>   		err = xe_ggtt_init(gt_to_tile(gt)->mem.ggtt);
>   		if (err)
>   			goto err_force_wake;
>   	}
>   
> -	err = xe_uc_init(&gt->uc);
> -	if (err)
> -		goto err_force_wake;
> -
> -	err = xe_uc_init_hwconfig(&gt->uc);
> -	if (err)
> -		goto err_force_wake;
> -
>   	xe_gt_idle_sysfs_init(&gt->gtidle);
>   
>   	/* XXX: Fake that we pull the engine mask from hwconfig blob */
 >       gt->info.engine_mask = gt->info.__engine_mask;


I've been wondering about the XXX above.  I added the line above that 
comes next.
Should that move into xe_gt_init_hwconfig() as well?
And replace it with code to read it from hwconfig blob?

I'm asking because I was looking at a change that would like to make use 
of a valid engine_mask earlier than xe_gt_init().

-Brian



> @@ -415,10 +404,6 @@ static int all_fw_domain_init(struct xe_gt *gt)
>   	if (err)
>   		goto err_force_wake;
>   
> -	err = xe_uc_init_post_hwconfig(&gt->uc);
> -	if (err)
> -		goto err_force_wake;
> -
>   	if (!xe_gt_is_media_type(gt)) {
>   		/*
>   		 * USM has its only SA pool to non-block behind user operations
> @@ -442,6 +427,10 @@ static int all_fw_domain_init(struct xe_gt *gt)
>   		}
>   	}
>   
> +	err = xe_uc_init_post_hwconfig(&gt->uc);
> +	if (err)
> +		goto err_force_wake;
> +
>   	err = xe_uc_init_hw(&gt->uc);
>   	if (err)
>   		goto err_force_wake;
> @@ -462,6 +451,38 @@ static int all_fw_domain_init(struct xe_gt *gt)
>   	return err;
>   }
>   
> +/*
> + * Initialize enough GT to be able to load GuC in order to obtain hwconfig and enable CTB
> + * communication.
> + */
> +int xe_gt_init_hwconfig(struct xe_gt *gt)
> +{
> +	int err;
> +
> +	xe_device_mem_access_get(gt_to_xe(gt));
> +	err = xe_force_wake_get(gt_to_fw(gt), XE_FW_GT);
> +	if (err)
> +		goto out;
> +
> +	xe_gt_mcr_init(gt);
> +	xe_pat_init(gt);
> +
> +	err = xe_uc_init(&gt->uc);
> +	if (err)
> +		goto out_fw;
> +
> +	err = xe_uc_init_hwconfig(&gt->uc);
> +	if (err)
> +		goto out_fw;
> +
> +out_fw:
> +	xe_force_wake_put(gt_to_fw(gt), XE_FW_GT);
> +out:
> +	xe_device_mem_access_put(gt_to_xe(gt));
> +
> +	return err;
> +}
> +
>   int xe_gt_init(struct xe_gt *gt)
>   {
>   	int err;
> diff --git a/drivers/gpu/drm/xe/xe_gt.h b/drivers/gpu/drm/xe/xe_gt.h
> index caded203a8a03..e9d6aeceb56af 100644
> --- a/drivers/gpu/drm/xe/xe_gt.h
> +++ b/drivers/gpu/drm/xe/xe_gt.h
> @@ -31,6 +31,7 @@ static inline bool xe_fault_inject_gt_reset(void)
>   #endif
>   
>   struct xe_gt *xe_gt_alloc(struct xe_tile *tile);
> +int xe_gt_init_hwconfig(struct xe_gt *gt);
>   int xe_gt_init_early(struct xe_gt *gt);
>   int xe_gt_init(struct xe_gt *gt);
>   int xe_gt_record_default_lrcs(struct xe_gt *gt);
> diff --git a/drivers/gpu/drm/xe/xe_uc.c b/drivers/gpu/drm/xe/xe_uc.c
> index 13e76e6805ca1..123baafc9ec38 100644
> --- a/drivers/gpu/drm/xe/xe_uc.c
> +++ b/drivers/gpu/drm/xe/xe_uc.c
> @@ -29,13 +29,15 @@ uc_to_xe(struct xe_uc *uc)
>   /* Should be called once at driver load only */
>   int xe_uc_init(struct xe_uc *uc)
>   {
> +	struct xe_device *xe = uc_to_xe(uc);
>   	int ret;
>   
> +	xe_device_mem_access_get(xe);
> +
>   	/*
>   	 * We call the GuC/HuC init functions even if GuC submission is off to
>   	 * correctly move our tracking of the FW state to "disabled".
>   	 */
> -
>   	ret = xe_guc_init(&uc->guc);
>   	if (ret)
>   		goto err;
> @@ -45,7 +47,7 @@ int xe_uc_init(struct xe_uc *uc)
>   		goto err;
>   
>   	if (!xe_device_uc_enabled(uc_to_xe(uc)))
> -		return 0;
> +		goto out;
>   
>   	ret = xe_wopcm_init(&uc->wopcm);
>   	if (ret)
> @@ -55,9 +57,14 @@ int xe_uc_init(struct xe_uc *uc)
>   	if (ret)
>   		goto err;
>   
> +out:
> +	xe_device_mem_access_put(xe);
> +
>   	return 0;
>   
>   err:
> +	xe_device_mem_access_put(xe);
> +
>   	return ret;
>   }
>   


More information about the Intel-xe mailing list