Re: [PATCH 1/3] drm/xe/pf: Use migration-friendly context IDs auto-provisioning

[Michal Wajdeczko <michal.wajdeczko@intel.com>]

On 11/6/2025 3:24 PM, Piotr Piórkowski wrote:
> Michal Wajdeczko <michal.wajdeczko@intel.com> wrote on śro [2025-lis-05 19:32:50 +0100]:
>> Instead of trying very hard to find the largest fair number of GuC
>> context IDs that could be allocated for VFs on the current GT, pick
>> some smaller rounded down to power-of-two value that is more likely
>> to be provisioned in the same manner by the other PF instance:
>>
>>  num VFs | num contexts
>>  --------+-------------
>>   63..32 | 1024
>>   31..16 | 2048
>>   15..8  | 4096
>>    7..4  | 8192
>>    3..2  | 16384
>>       1  | 32768 (regular PF)
>>       1  | 64512 (admin only PF)
>>
>> Add also helper function to determine if the PF is admin-only,
>> and for now use .probe_display flag for that.
>>
>> Signed-off-by: Michal Wajdeczko <michal.wajdeczko@intel.com>
>> ---
>>  drivers/gpu/drm/xe/xe_gt_sriov_pf_config.c | 16 ++++++++++++++++
>>  drivers/gpu/drm/xe/xe_sriov_pf_helpers.h   | 11 +++++++++++
>>  2 files changed, 27 insertions(+)
>>
>> diff --git a/drivers/gpu/drm/xe/xe_gt_sriov_pf_config.c b/drivers/gpu/drm/xe/xe_gt_sriov_pf_config.c
>> index d90261a7ab7c..14feda215d5b 100644
>> --- a/drivers/gpu/drm/xe/xe_gt_sriov_pf_config.c
>> +++ b/drivers/gpu/drm/xe/xe_gt_sriov_pf_config.c
>> @@ -985,6 +985,16 @@ int xe_gt_sriov_pf_config_bulk_set_ctxs(struct xe_gt *gt, unsigned int vfid,
>>  					   "GuC context IDs", no_unit, n, err);
>>  }
>>  
>> +static u32 pf_profile_fair_ctxs(struct xe_gt *gt, unsigned int num_vfs)
>> +{
>> +	bool admin_only_pf = xe_sriov_pf_admin_only(gt_to_xe(gt));
>> +
>> +	if (admin_only_pf && num_vfs == 1)
>> +		return ALIGN_DOWN(GUC_ID_MAX, SZ_1K);
>> +
>> +	return rounddown_pow_of_two(GUC_ID_MAX / num_vfs);
>> +}
>> +
>>  static u32 pf_estimate_fair_ctxs(struct xe_gt *gt, unsigned int num_vfs)
>>  {
>>  	struct xe_guc_id_mgr *idm = &gt->uc.guc.submission_state.idm;
>> @@ -1017,6 +1027,7 @@ static u32 pf_estimate_fair_ctxs(struct xe_gt *gt, unsigned int num_vfs)
>>  int xe_gt_sriov_pf_config_set_fair_ctxs(struct xe_gt *gt, unsigned int vfid,
>>  					unsigned int num_vfs)
>>  {
>> +	u32 profile = pf_profile_fair_ctxs(gt, num_vfs);
>>  	u32 fair;
>>  
>>  	xe_gt_assert(gt, vfid);
>> @@ -1029,6 +1040,11 @@ int xe_gt_sriov_pf_config_set_fair_ctxs(struct xe_gt *gt, unsigned int vfid,
>>  	if (!fair)
>>  		return -ENOSPC;
>>  
>> +	fair = min(fair, profile);
>> +	if (fair < profile)
>> +		xe_gt_sriov_info(gt, "Using non-profile provisioning (%s %u vs %u)\n",
>> +				 "GuC context IDs", fair, profile);
>> +
>>  	return xe_gt_sriov_pf_config_bulk_set_ctxs(gt, vfid, num_vfs, fair);
>>  }
>>  
>> diff --git a/drivers/gpu/drm/xe/xe_sriov_pf_helpers.h b/drivers/gpu/drm/xe/xe_sriov_pf_helpers.h
>> index 4a4340fb633a..3ddeba4451cd 100644
>> --- a/drivers/gpu/drm/xe/xe_sriov_pf_helpers.h
>> +++ b/drivers/gpu/drm/xe/xe_sriov_pf_helpers.h
>> @@ -48,6 +48,17 @@ static inline unsigned int xe_sriov_pf_num_vfs(const struct xe_device *xe)
>>  	return pci_num_vf(to_pci_dev(xe->drm.dev));
>>  }
>>  
>> +/**
>> + * xe_sriov_pf_admin_only() - Check if PF is mainly used for VFs administration.
>> + * @xe: the PF &xe_device
>> + *
>> + * Return: True if PF is mainly used for VFs administration.
>> + */
>> +static inline bool xe_sriov_pf_admin_only(const struct xe_device *xe)
>> +{
>> +	return !xe->info.probe_display;
>> +}
> 
> Honestly, I'm not entirely convinced by this condition for pf_admin_only.
> I wonder if we shouldn't set it up differently, e.g., through a dedicated
> parameter. 

yes, that's exactly the plan (but it's on my TODO list and I needed something
sooner, and logic that uses .probe_display shall be correct in 99% cases)

> But for now, I don't think it bothers us too much:
> Reviewed-by: Piotr Piórkowski <piotr.piorkowski@intel.com>

thanks!

> 
>> +
>>  static inline struct mutex *xe_sriov_pf_master_mutex(struct xe_device *xe)
>>  {
>>  	xe_assert(xe, IS_SRIOV_PF(xe));
>> -- 
>> 2.47.1
>>
>