Re: [PATCH 2/9] drm/xe/pf: Add functions for VRAM provisioning

["Piotr Piórkowski" <piotr.piorkowski@intel.com>]

Piotr Piórkowski <piotr.piorkowski@intel.com> wrote on pon [2026-lut-16 16:02:20 +0100]:
> Michal Wajdeczko <michal.wajdeczko@intel.com> wrote on nie [2026-lut-15 21:33:16 +0100]:
> > We already have functions to configure VF LMEM (aka VRAM) on the
> > tile/GT level, used by the auto-provisioning and debugfs, but we
> > also need functions that will work on the device level that will
> > configure VRAM on all tiles at once.
> > 
> > We will use these new functions in upcoming patch.
> > 
> > Signed-off-by: Michal Wajdeczko <michal.wajdeczko@intel.com>
> > ---
> >  drivers/gpu/drm/xe/xe_sriov_pf_provision.c | 108 +++++++++++++++++++++
> >  drivers/gpu/drm/xe/xe_sriov_pf_provision.h |   4 +
> >  2 files changed, 112 insertions(+)
> > 
> > diff --git a/drivers/gpu/drm/xe/xe_sriov_pf_provision.c b/drivers/gpu/drm/xe/xe_sriov_pf_provision.c
> > index 01470c42e8a7..e7187d03fe1b 100644
> > --- a/drivers/gpu/drm/xe/xe_sriov_pf_provision.c
> > +++ b/drivers/gpu/drm/xe/xe_sriov_pf_provision.c
> > @@ -436,3 +436,111 @@ int xe_sriov_pf_provision_query_vf_priority(struct xe_device *xe, unsigned int v
> >  
> >  	return !count ? -ENODATA : 0;
> >  }
> > +
> > +static u64 vram_alignment(struct xe_device *xe)
> > +{
> > +	/* this might be platform dependent */
> > +	return SZ_2M;
> > +}
> > +
> 
> In xe_gt_sriov_pf_config.c we already have an almost identical function:
> static u64 pf_get_lmem_alignment(struct xe_gt *gt)
> {
> 	/* this might be platform dependent */
> 	return SZ_2M;
> }
> 
> In my opinion, it is not worth duplicating it, but rather creating one common function.
> 
> 
> > +static u64 vram_per_tile(struct xe_tile *tile, u64 total)
> > +{
> > +	struct xe_device *xe = tile->xe;
> > +	unsigned int tcount = xe->info.tile_count;
> > +	u64 alignment = vram_alignment(xe);
> > +
> > +	total = round_up(total, tcount * alignment);
> > +	return div_u64(total, tcount);
> > +}
> > +
> > +/**
> > + * xe_sriov_pf_provision_bulk_apply_vram() - Change VRAM provisioning for all VFs.
> > + * @xe: the PF &xe_device
> > + * @size: the VRAM size in [bytes] to set
> > + *
> > + * Change all VFs VRAM (LMEM) provisioning on all tiles.
> > + *
> > + * This function can only be called on PF.
> > + *
> > + * Return: 0 on success or a negative error code on failure.
> > + */
> > +int xe_sriov_pf_provision_bulk_apply_vram(struct xe_device *xe, u64 size)
> > +{
> > +	unsigned int num_vfs = xe_sriov_pf_get_totalvfs(xe);
> > +	struct xe_tile *tile;
> > +	unsigned int id;
> > +	int result = 0;
> > +	int err;
> > +

Reading the next patch, I realized that we still need to check if(xe_device_has_lmtt)
When defining sysfs, it cannot be done trivially, and sysfs will remain visible anyway due to DGFX.

The same for 
xe_sriov_pf_provision_apply_vf_vram

> > +	guard(mutex)(xe_sriov_pf_master_mutex(xe));
> > +
> > +	for_each_tile(tile, xe, id) {
> > +		err = xe_gt_sriov_pf_config_bulk_set_lmem_locked(tile->primary_gt,
> > +								 VFID(1), num_vfs,
> > +								 vram_per_tile(tile, size));
> > +		result = result ?: err;
> > +	}
> > +
> > +	return result;
> > +}
> > +
> > +/**
> > + * xe_sriov_pf_provision_apply_vf_vram() - Change single VF VRAM allocatio.
> typo
> 
> > + * @xe: the PF &xe_device
> > + * @vfid: the VF identifier (can't be 0 == PFID)
> > + * @size: VRAM size to set
> > + *
> > + * Change VF's VRAM provisioning on all tiles/GTs.
> > + *
> > + * This function can only be called on PF.
> > + *
> > + * Return: 0 on success or a negative error code on failure.
> > + */
> > +int xe_sriov_pf_provision_apply_vf_vram(struct xe_device *xe, unsigned int vfid, u64 size)
> > +{
> > +	struct xe_tile *tile;
> > +	unsigned int id;
> > +	int result = 0;
> > +	int err;
> > +
> > +	xe_assert(xe, vfid);
> > +
> > +	guard(mutex)(xe_sriov_pf_master_mutex(xe));
> > +
> > +	for_each_tile(tile, xe, id) {
> > +		err = xe_gt_sriov_pf_config_set_lmem_locked(tile->primary_gt, vfid,
> > +							    vram_per_tile(tile, size));
> > +		result = result ?: err;
> > +	}
> > +
> > +	return result;
> > +}
> > +
> > +/**
> > + * xe_sriov_pf_provision_query_vf_vram() - Query VF's VRAM allocation.
> > + * @xe: the PF &xe_device
> > + * @vfid: the VF identifier (can't be 0 == PFID)
> > + * @prio: placeholder for the returned VRAM size
> > + *
> > + * Query VF's VRAM provisioning from all tiles/GTs.
> > + *
> > + * This function can only be called on PF.
> > + *
> > + * Return: 0 on success or a negative error code on failure.
> > + */
> > +int xe_sriov_pf_provision_query_vf_vram(struct xe_device *xe, unsigned int vfid, u64 *size)
> > +{
> > +	struct xe_tile *tile;
> > +	unsigned int id;
> > +	u64 total = 0;
> NIT: The variable total is unnecessary
> 
> > +
> > +	xe_assert(xe, vfid);
> > +
> > +	guard(mutex)(xe_sriov_pf_master_mutex(xe));
> > +
> > +	for_each_tile(tile, xe, id)
> > +		total += xe_gt_sriov_pf_config_get_lmem_locked(tile->primary_gt, vfid);
> > +
> > +	*size = total;
> > +	return 0;
> > +}
> > diff --git a/drivers/gpu/drm/xe/xe_sriov_pf_provision.h b/drivers/gpu/drm/xe/xe_sriov_pf_provision.h
> > index bccf23d51396..f26f49539697 100644
> > --- a/drivers/gpu/drm/xe/xe_sriov_pf_provision.h
> > +++ b/drivers/gpu/drm/xe/xe_sriov_pf_provision.h
> > @@ -24,6 +24,10 @@ int xe_sriov_pf_provision_bulk_apply_priority(struct xe_device *xe, u32 prio);
> >  int xe_sriov_pf_provision_apply_vf_priority(struct xe_device *xe, unsigned int vfid, u32 prio);
> >  int xe_sriov_pf_provision_query_vf_priority(struct xe_device *xe, unsigned int vfid, u32 *prio);
> >  
> > +int xe_sriov_pf_provision_bulk_apply_vram(struct xe_device *xe, u64 size);
> > +int xe_sriov_pf_provision_apply_vf_vram(struct xe_device *xe, unsigned int vfid, u64 size);
> > +int xe_sriov_pf_provision_query_vf_vram(struct xe_device *xe, unsigned int vfid, u64 *size);
> > +
> >  int xe_sriov_pf_provision_vfs(struct xe_device *xe, unsigned int num_vfs);
> >  int xe_sriov_pf_unprovision_vfs(struct xe_device *xe, unsigned int num_vfs);
> >  
> Code is functionally OK
> Minor comments, with clarification of vram_alignment:
> Reviewed-by: Piotr Piórkowski <piotr.piorkowski@intel.com>
> > -- 
> > 2.47.1
> > 
> 
> -- 

--