Re: [PATCH slab/for-next v4 7/8] mm/slab: introduce kfree_rcu_nolock()

[Vlastimil Babka (SUSE)]

On 7/22/26 10:15, Harry Yoo wrote:
> On 7/21/26 10:09 PM, Vlastimil Babka (SUSE) wrote:
>> On 7/20/26 14:44, Harry Yoo (Oracle) wrote:
>>> Currently, k[v]free_rcu() cannot be called in unknown context since
>>> it could lead to a deadlock when called in the middle of k[v]free_rcu().
>>>
>>> Make users' lives easier by introducing kfree_rcu_nolock() variant,
>>> now that kfree_rcu_sheaf() is available on PREEMPT_RT and
>>> __kfree_rcu_sheaf() handles unknown context.
>>>
>>> kfree_rcu_nolock() falls back to the kvfree_rcu batching when sheaves
>>> path fails.
>> 
>> From the code I would say it falls back to defer_kfree_rcu(), which is an
>> irq_work that does kvfree_rcu batching?
> 
> Indeed that was what I intended to say, but I messed up while
> writing it :-) Will fix that.

Thanks!

>> Otherwise the sentence would contradict the following:
>>
>>> In most cases, the sheaves path is expected to succeed
>>> and it's unnecessary to add complexity to the existing kvfree_rcu
>>> batching by teaching it how to handle unknown context.
>> 
>> This talks about trying to perform the batching immediately without defering.
> 
> You mean the earlier sentence "kfree_rcu_nolock() falls back to the
> kvfree_rcu batching when sheaves path fails", right?

I was trying to point out how it contradicts the wording above. To be clear,
with above fixed, this part is fine.

>>> Since defer_kfree_rcu() can be called on caches without sheaves, move
>>> deferred_work_barrier() and rcu_barrier() outside the branch in
>>> kvfree_rcu_barrier_on_cache().
>>>
>>> Signed-off-by: Harry Yoo (Oracle) <harry@xxxxxxxxxx>
>> 
>> Other than that and a comment correction below, LGTM.
>> 
>> Reviewed-by: Vlastimil Babka (SUSE) <vbabka@xxxxxxxxxx>
> 
> Thanks!
> 
>>> ---
>>>  include/linux/rcupdate.h | 19 +++++++++++++++++++
>>>  mm/slab_common.c         | 28 ++++++++++++++++++++++++++--
>>>  mm/slub.c                | 24 +++++++++++++++++++++++-
>>>  3 files changed, 68 insertions(+), 3 deletions(-)
>>>
>>> diff --git a/include/linux/rcupdate.h b/include/linux/rcupdate.h
>>> index ef5bb6981133..ce0fd401352e 100644
>>> --- a/include/linux/rcupdate.h
>>> +++ b/include/linux/rcupdate.h
>>> @@ -1099,6 +1099,7 @@ static inline void rcu_read_unlock_migrate(void)
>>>   * In mm/slab_common.c, no suitable header to include here.
>>>   */
>>>  void kvfree_call_rcu(struct kvfree_rcu_head *head, void *ptr);
>>> +void kfree_call_rcu_nolock(struct kvfree_rcu_head *head, void *ptr);
>>>  
>>>  /*
>>>   * The BUILD_BUG_ON() makes sure the rcu_head offset can be handled. See the
>>> @@ -1124,6 +1125,24 @@ do {								\
>>>  		kvfree_call_rcu(NULL, (void *) (___p));		\
>>>  } while (0)
>>>  
>>> +/**
>>> + * kfree_rcu_nolock() - a version of kfree_rcu() that can be called in any context.
>>> + * @ptr: pointer to kfree for double-argument invocations.
>>> + * @kvrhf: the name of the struct kvfree_rcu_head within the type of @ptr.
>>> + *
>>> + * Objects that are not allocated via kmalloc_nolock() are not supported.
>> 
>> Is this true? If __kfree_rcu_sheaf() succeeds, rcu_sheaf is either flushed
>> normally (call_rcu with rcu_free_sheaf()) or via the irq_work doing the
>> same. The defer_kfree_rcu() fallback ends up doing a normal
>> kvfree_call_rcu(). I don't see any path that ends up doing kfree_nolock().
>> Am I missing something?
> 
> I added this limitation because I thought it doesn't make sense
> to support this when we don't support large kmalloc and vmalloc in
> unknown context.
> 
> Perhaps too limiting than it should be though.

Hmm right. Maybe we could just state that it handles all small kmalloc() and
kmalloc_nolock() allocations without deferring as much as possible, but
large k[v]malloc() allocations that end up being large_kmalloc() or
vmalloc(), will also work but be always deferred?

>>> + * kfree_rcu_nolock() supports 2-arg variant only.
>>> + */
>>> +#define kfree_rcu_nolock(ptr, kvrhf)						\
>>> +do {										\
>>> +	typeof (ptr) ___p = (ptr);						\
>>> +										\
>>> +	if (___p) {								\
>>> +		BUILD_BUG_ON(offsetof(typeof(*(ptr)), kvrhf) >= 4096);		\
>>> +		kfree_call_rcu_nolock(&((___p)->kvrhf), (void *) (___p));	\
>>> +	}									\
>>> +} while (0)
>>> +
>>>  /*
>>>   * Place this after a lock-acquisition primitive to guarantee that
>>>   * an UNLOCK+LOCK pair acts as a full barrier.  This guarantee applies
>