Re: [PATCH 3/5] mm/vmalloc.c: optimize code in decay_va_pool_node() a little bit

[Shivank Garg <shivankg@amd.com>]

On 4/15/2025 7:35 PM, Baoquan He wrote:
> On 04/15/25 at 03:59pm, Shivank Garg wrote:
>> On 4/15/2025 8:09 AM, Baoquan He wrote:
>>> When purge lazily freed vmap areas, VA stored in vn->pool[] will also be
>>> taken away into free vmap tree partially or completely accordingly, that
>>> is done in decay_va_pool_node(). When doing that, for each pool of node,
>>> the whole list is detached from the pool for handling. At this time,
>>> that pool is empty. It's not necessary to update the pool size each time
>>> when one VA is removed and addded into free vmap tree.
>>>
>>> Here change code to update the pool size when attaching the pool back.
>>>
>>> Signed-off-by: Baoquan He <bhe@redhat.com>
>>> ---
>>>  mm/vmalloc.c | 23 +++++++++++------------
>>>  1 file changed, 11 insertions(+), 12 deletions(-)
>>>
>>> diff --git a/mm/vmalloc.c b/mm/vmalloc.c
>>> index 488d69b56765..bf735c890878 100644
>>> --- a/mm/vmalloc.c
>>> +++ b/mm/vmalloc.c
>>> @@ -2150,7 +2150,7 @@ decay_va_pool_node(struct vmap_node *vn, bool full_decay)
>>>  	LIST_HEAD(decay_list);
>>>  	struct rb_root decay_root = RB_ROOT;
>>>  	struct vmap_area *va, *nva;
>>> -	unsigned long n_decay;
>>> +	unsigned long n_decay, len;
>>>  	int i;
>>>  
>>>  	for (i = 0; i < MAX_VA_SIZE_PAGES; i++) {
>>> @@ -2164,22 +2164,20 @@ decay_va_pool_node(struct vmap_node *vn, bool full_decay)
>>>  		list_replace_init(&vn->pool[i].head, &tmp_list);
>>>  		spin_unlock(&vn->pool_lock);
>>>  
>>> -		if (full_decay)
>>> -			WRITE_ONCE(vn->pool[i].len, 0);
>>> +		len = n_decay = vn->pool[i].len;
>>> +		WRITE_ONCE(vn->pool[i].len, 0);
>>>  
>>>  		/* Decay a pool by ~25% out of left objects. */
>>> -		n_decay = vn->pool[i].len >> 2;
>>> +		if (!full_decay)
>>> +			n_decay >>= 2;
>>> +		len -= n_decay;
>>>  
>>>  		list_for_each_entry_safe(va, nva, &tmp_list, list) {
>>> +			if (!n_decay)
>>> +				break;
>>>  			list_del_init(&va->list);
>>>  			merge_or_add_vmap_area(va, &decay_root, &decay_list);
>>> -
>>> -			if (!full_decay) {
>>> -				WRITE_ONCE(vn->pool[i].len, vn->pool[i].len - 1);
>>> -
>>> -				if (!--n_decay)
>>> -					break;
>>> -			}
>>> +			n_decay--;
>>>  		}
>>>  
>>>  		/*
>>> @@ -2188,9 +2186,10 @@ decay_va_pool_node(struct vmap_node *vn, bool full_decay)
>>>  		 * can populate the pool therefore a simple list replace
>>>  		 * operation takes place here.
>>>  		 */
>>> -		if (!full_decay && !list_empty(&tmp_list)) {
>>> +		if (!list_empty(&tmp_list)) {
>>>  			spin_lock(&vn->pool_lock);
>>>  			list_replace_init(&tmp_list, &vn->pool[i].head);
>>> +			vn->pool[i].len = len;
>>
>> Current logic uses WRITE_ONCE() to update vn->pool[i].len.
>> Could this lead to consistency issues?
> 
> Seems no necessary to use WRITE_ONCE(). I can change back to use
> WRITE_ONCE() just in case. Currently, it's only updated in
> node_alloc(), decay_va_pool_node(), purge_vmap_node(). And the latter
> two are inside vmap_purge_lock area.
> 

Okay.

Reviewed-by: Shivank Garg <shivankg@amd.com>
Tested-by: Shivank Garg <shivankg@amd.com>

Best,
Shivank

>>
>>>  			spin_unlock(&vn->pool_lock);
>>>  		}
>>>  	}
>>
>