[PATCH v7 1/4] drm/xe/vf: Divide GGTT ballooning into allocation and insertion

Lis, Tomasz tomasz.lis at intel.com
Wed Apr 9 20:58:34 UTC 2025


On 08.04.2025 13:59, Michal Wajdeczko wrote:
> On 03.04.2025 20:40, Tomasz Lis wrote:
>> The balloon nodes, which are used to fill areas of GGTT inaccessible
>> for a specific VF, were allocated and inserted into GGTT within one
>> function. To be able to re-use that insertion code during VF
>> migration recovery, we need to split it.
>>
>> This patch separates allocation (init/fini functs) from the insertion
>> of balloons (balloon/deballoon functs). Locks are also moved to ensure
>> calls from post-migration recovery worker will not cause a deadlock.
>>
>> v2: Moved declarations to proper header
>> v3: Rephrased description, introduced "_locked" versions of some
>>    functs, more lockdep checks, some functions renamed, altered error
>>    handling, added missing kerneldocs.
>>
>> Signed-off-by: Tomasz Lis<tomasz.lis at intel.com>
>> ---
>>   drivers/gpu/drm/xe/xe_ggtt.c        |  11 +--
>>   drivers/gpu/drm/xe/xe_gt_sriov_vf.c | 102 +++++++++++++++++++++-------
>>   drivers/gpu/drm/xe/xe_gt_sriov_vf.h |   2 +
>>   3 files changed, 82 insertions(+), 33 deletions(-)
>>
>> diff --git a/drivers/gpu/drm/xe/xe_ggtt.c b/drivers/gpu/drm/xe/xe_ggtt.c
>> index 5fcb2b4c2c13..769a8dc9be6e 100644
>> --- a/drivers/gpu/drm/xe/xe_ggtt.c
>> +++ b/drivers/gpu/drm/xe/xe_ggtt.c
>> @@ -447,14 +447,13 @@ int xe_ggtt_node_insert_balloon(struct xe_ggtt_node *node, u64 start, u64 end)
>>   	xe_tile_assert(ggtt->tile, IS_ALIGNED(start, XE_PAGE_SIZE));
>>   	xe_tile_assert(ggtt->tile, IS_ALIGNED(end, XE_PAGE_SIZE));
>>   	xe_tile_assert(ggtt->tile, !drm_mm_node_allocated(&node->base));
>> +	lockdep_assert_held(&ggtt->lock);
> since lock is now prerequisite, this function shall be renamed to:
>
> 	xe_ggtt_node_insert_balloon_locked()
>
> likely with update to kerneldoc like:
>
> 	"To be used in cases where ggtt->lock is already taken.

It now has the ggtt->lock dependency, but this is ggtt_node function and 
not ggtt function.

But sure, will rename and comment.

>>   
>>   	node->base.color = 0;
>>   	node->base.start = start;
>>   	node->base.size = end - start;
>>   
>> -	mutex_lock(&ggtt->lock);
>>   	err = drm_mm_reserve_node(&ggtt->mm, &node->base);
>> -	mutex_unlock(&ggtt->lock);
>>   
>>   	if (xe_gt_WARN(ggtt->tile->primary_gt, err,
>>   		       "Failed to balloon GGTT %#llx-%#llx (%pe)\n",
>> @@ -477,16 +476,12 @@ void xe_ggtt_node_remove_balloon(struct xe_ggtt_node *node)
> same as in earlier comment, rename function to:
>
> 	xe_ggtt_node_remove_balloon_locked()
ok
>>   		return;
>>   
>>   	if (!drm_mm_node_allocated(&node->base))
>> -		goto free_node;
>> +		return;
>>   
>> +	lockdep_assert_held(&node->ggtt->lock);
> this should be earlier in the function, as by API SLA lock must be
> always taken, not only when node is allocated

but should the asserts verify the API SLA, or real state inconsistencies?

either way - ok, will move.

>>   	xe_ggtt_dump_node(node->ggtt, &node->base, "remove-balloon");
>>   
>> -	mutex_lock(&node->ggtt->lock);
>>   	drm_mm_remove_node(&node->base);
>> -	mutex_unlock(&node->ggtt->lock);
>> -
>> -free_node:
>> -	xe_ggtt_node_fini(node);
>>   }
>>   
>>   /**
>> diff --git a/drivers/gpu/drm/xe/xe_gt_sriov_vf.c b/drivers/gpu/drm/xe/xe_gt_sriov_vf.c
>> index a439261bf4d7..c3ca33725161 100644
>> --- a/drivers/gpu/drm/xe/xe_gt_sriov_vf.c
>> +++ b/drivers/gpu/drm/xe/xe_gt_sriov_vf.c
>> @@ -560,35 +560,38 @@ u64 xe_gt_sriov_vf_lmem(struct xe_gt *gt)
>>   	return gt->sriov.vf.self_config.lmem_size;
>>   }
>>   
>> -static struct xe_ggtt_node *
>> -vf_balloon_ggtt_node(struct xe_ggtt *ggtt, u64 start, u64 end)
>> +static int vf_init_ggtt_balloons(struct xe_gt *gt)
>>   {
>> -	struct xe_ggtt_node *node;
>> -	int err;
>> +	struct xe_tile *tile = gt_to_tile(gt);
>> +	struct xe_ggtt *ggtt = tile->mem.ggtt;
>>   
>> -	node = xe_ggtt_node_init(ggtt);
>> -	if (IS_ERR(node))
>> -		return node;
>> +	tile->sriov.vf.ggtt_balloon[0] = xe_ggtt_node_init(ggtt);
>> +	if (IS_ERR(tile->sriov.vf.ggtt_balloon[0]))
>> +		return PTR_ERR(tile->sriov.vf.ggtt_balloon[0]);
>>   
>> -	err = xe_ggtt_node_insert_balloon(node, start, end);
>> -	if (err) {
>> -		xe_ggtt_node_fini(node);
>> -		return ERR_PTR(err);
>> -	}
>> +	tile->sriov.vf.ggtt_balloon[1] = xe_ggtt_node_init(ggtt);
>> +	if (IS_ERR(tile->sriov.vf.ggtt_balloon[1]))
>> +		return PTR_ERR(tile->sriov.vf.ggtt_balloon[1]);
> what about ggtt_balloon[0] ? no need to fini() it ?
will add.
>>   
>> -	return node;
>> +	return 0;
>>   }
>>   
>> -static int vf_balloon_ggtt(struct xe_gt *gt)
>> +/**
>> + * xe_gt_sriov_vf_balloon_ggtt_locked - Insert balloon nodes to limit used GGTT address range.
>> + * @gt: the &xe_gt struct instance
>> + * Return: 0 on success or a negative error code on failure.
>> + */
>> +int xe_gt_sriov_vf_balloon_ggtt_locked(struct xe_gt *gt)
>>   {
>>   	struct xe_gt_sriov_vf_selfconfig *config = &gt->sriov.vf.self_config;
>>   	struct xe_tile *tile = gt_to_tile(gt);
>> -	struct xe_ggtt *ggtt = tile->mem.ggtt;
>>   	struct xe_device *xe = gt_to_xe(gt);
>>   	u64 start, end;
>> +	int err;
>>   
>>   	xe_gt_assert(gt, IS_SRIOV_VF(xe));
>>   	xe_gt_assert(gt, !xe_gt_is_media_type(gt));
>> +	lockdep_assert_held(&tile->mem.ggtt->lock);
>>   
>>   	if (!config->ggtt_size)
>>   		return -ENODATA;
>> @@ -611,33 +614,76 @@ static int vf_balloon_ggtt(struct xe_gt *gt)
>>   	start = xe_wopcm_size(xe);
>>   	end = config->ggtt_base;
>>   	if (end != start) {
>> -		tile->sriov.vf.ggtt_balloon[0] = vf_balloon_ggtt_node(ggtt, start, end);
>> -		if (IS_ERR(tile->sriov.vf.ggtt_balloon[0]))
>> -			return PTR_ERR(tile->sriov.vf.ggtt_balloon[0]);
>> +		err = xe_ggtt_node_insert_balloon(tile->sriov.vf.ggtt_balloon[0], start, end);
>> +		if (err)
>> +			return err;
>>   	}
>>   
>>   	start = config->ggtt_base + config->ggtt_size;
>>   	end = GUC_GGTT_TOP;
>>   	if (end != start) {
>> -		tile->sriov.vf.ggtt_balloon[1] = vf_balloon_ggtt_node(ggtt, start, end);
>> -		if (IS_ERR(tile->sriov.vf.ggtt_balloon[1])) {
>> +		err = xe_ggtt_node_insert_balloon(tile->sriov.vf.ggtt_balloon[1], start, end);
>> +		if (err) {
>>   			xe_ggtt_node_remove_balloon(tile->sriov.vf.ggtt_balloon[0]);
>> -			return PTR_ERR(tile->sriov.vf.ggtt_balloon[1]);
>> +			return err;
>>   		}
>>   	}
>>   
>>   	return 0;
>>   }
>>   
>> -static void deballoon_ggtt(struct drm_device *drm, void *arg)
>> +static int vf_balloon_ggtt(struct xe_gt *gt)
>>   {
>> -	struct xe_tile *tile = arg;
>> +	struct xe_ggtt *ggtt = gt_to_tile(gt)->mem.ggtt;
>> +	int err;
>> +
>> +	mutex_lock(&ggtt->lock);
>> +	err = xe_gt_sriov_vf_balloon_ggtt_locked(gt);
>> +	mutex_unlock(&ggtt->lock);
>> +
>> +	return err;
>> +}
>> +
>> +/**
>> + * xe_gt_sriov_vf_deballoon_ggtt_locked - Remove balloon nodes which limited used address renge.
>> + * @gt: the &xe_gt struct instance
>> + */
>> +void xe_gt_sriov_vf_deballoon_ggtt_locked(struct xe_gt *gt)
>> +{
>> +	struct xe_tile *tile = gt_to_tile(gt);
>>   
>>   	xe_tile_assert(tile, IS_SRIOV_VF(tile_to_xe(tile)));
>> +	lockdep_assert_held(&tile->mem.ggtt->lock);
> nit: IMO this is redundant as lock shall be asserted in
> xe_ggtt_node_remove_balloon_locked()
ok, will remove.
>> +
>>   	xe_ggtt_node_remove_balloon(tile->sriov.vf.ggtt_balloon[1]);
>>   	xe_ggtt_node_remove_balloon(tile->sriov.vf.ggtt_balloon[0]);
>>   }
>>   
>> +static void vf_deballoon_ggtt(struct xe_gt *gt)
> hmm, in this patch you don't really need to split (de)balloon logic into
> locked/unlocked parts, so maybe keep it as it was and introduce such
> split when really needed

By this logic, we could squish everything into one patch. I do not "need 
to" separate any parts.

Also, this has no impact - introducing this change does not affect final 
code, and does not improve readability of single patches. It's just 
cutting a pie slightly to the left or to the right.

I don't think such request should be part of a review. It seem to be a 
matter of personal aesthetics rather than engineering.

The current division is no less consistent than the proposed one. 
Actually, for code readability, it makes sense to introduce a locking 
wrapper here even if locked version was unused.

> also it's quite unusual that unlocked part is named in completely
> different fashion than locked, can't it be (later) defined as pair of:
>
> 	xe_gt_sriov_vf_deballoon_ggtt_locked()
> 	xe_gt_sriov_vf_deballoon_ggtt()
>
> and
>
> 	xe_gt_sriov_vf_balloon_ggtt_locked()
> 	xe_gt_sriov_vf_balloon_ggtt()
We have established that static functions have shorter names, and we're 
sticking to it. It is not unusual, it's the matter of project coding rules.
>> +{
>> +	struct xe_tile *tile = gt_to_tile(gt);
>> +
>> +	mutex_lock(&tile->mem.ggtt->lock);
>> +	xe_gt_sriov_vf_deballoon_ggtt_locked(gt);
>> +	mutex_unlock(&tile->mem.ggtt->lock);
>> +}
>> +
>> +static void vf_balloon_fini(struct xe_gt *gt)
>> +{
>> +	struct xe_tile *tile = gt_to_tile(gt);
> missing asserts:
>
>    	xe_gt_assert(gt, IS_SRIOV_VF(xe));
>    	xe_gt_assert(gt, !xe_gt_is_media_type(gt));

will add.

-Tomasz

>> +
>> +	xe_ggtt_node_fini(tile->sriov.vf.ggtt_balloon[1]);
>> +	xe_ggtt_node_fini(tile->sriov.vf.ggtt_balloon[0]);
>> +}
>> +
>> +static void deballoon_and_fini_ggtt(struct drm_device *drm, void *arg)
>> +{
>> +	struct xe_tile *tile = arg;
>> +
>> +	vf_deballoon_ggtt(tile->primary_gt);
>> +	vf_balloon_fini(tile->primary_gt);
>> +}
>> +
>>   /**
>>    * xe_gt_sriov_vf_prepare_ggtt - Prepare a VF's GGTT configuration.
>>    * @gt: the &xe_gt
>> @@ -655,11 +701,17 @@ int xe_gt_sriov_vf_prepare_ggtt(struct xe_gt *gt)
>>   	if (xe_gt_is_media_type(gt))
>>   		return 0;
>>   
>> -	err = vf_balloon_ggtt(gt);
>> +	err = vf_init_ggtt_balloons(gt);
>>   	if (err)
>>   		return err;
>>   
>> -	return drmm_add_action_or_reset(&xe->drm, deballoon_ggtt, tile);
>> +	err = vf_balloon_ggtt(gt);
>> +	if (err) {
>> +		vf_balloon_fini(gt);
>> +		return err;
>> +	}
>> +
>> +	return drmm_add_action_or_reset(&xe->drm, deballoon_and_fini_ggtt, tile);
>>   }
>>   
>>   static int relay_action_handshake(struct xe_gt *gt, u32 *major, u32 *minor)
>> diff --git a/drivers/gpu/drm/xe/xe_gt_sriov_vf.h b/drivers/gpu/drm/xe/xe_gt_sriov_vf.h
>> index ba6c5d74e326..d717deb8af91 100644
>> --- a/drivers/gpu/drm/xe/xe_gt_sriov_vf.h
>> +++ b/drivers/gpu/drm/xe/xe_gt_sriov_vf.h
>> @@ -18,6 +18,8 @@ int xe_gt_sriov_vf_query_config(struct xe_gt *gt);
>>   int xe_gt_sriov_vf_connect(struct xe_gt *gt);
>>   int xe_gt_sriov_vf_query_runtime(struct xe_gt *gt);
>>   int xe_gt_sriov_vf_prepare_ggtt(struct xe_gt *gt);
>> +int xe_gt_sriov_vf_balloon_ggtt_locked(struct xe_gt *gt);
>> +void xe_gt_sriov_vf_deballoon_ggtt_locked(struct xe_gt *gt);
>>   int xe_gt_sriov_vf_notify_resfix_done(struct xe_gt *gt);
>>   void xe_gt_sriov_vf_migrated_event_handler(struct xe_gt *gt);
>>   


More information about the Intel-xe mailing list