Re: [patch 24/25] debugobjects: Refill per CPU pool more agressively

[Leizhen (ThunderTown)]

On 2024/10/8 0:50, Thomas Gleixner wrote:
> Right now the per CPU pools are only refilled when they become
> empty. That's suboptimal especially when there are still non-freed objects
> in the to free list.
> 
> Check whether an allocation from the per CPU pool emptied a batch and try
> to allocate from the free pool if that still has objects available.
> 
>    	    kmem_cache_alloc()	kmem_cache_free()
> Baseline:   295k		245k
> Refill:	    225k		173k

Reviewed-by: Zhen Lei <thunder.leizhen@xxxxxxxxxx>

> 
> Signed-off-by: Thomas Gleixner <tglx@xxxxxxxxxxxxx>
> ---
>  lib/debugobjects.c |   18 ++++++++++++++++++
>  1 file changed, 18 insertions(+)
> 
> --- a/lib/debugobjects.c
> +++ b/lib/debugobjects.c
> @@ -255,6 +255,24 @@ static struct debug_obj *pcpu_alloc(void
>  
>  		if (likely(obj)) {
>  			pcp->cnt--;
> +			/*
> +			 * If this emptied a batch try to refill from the
> +			 * free pool. Don't do that if this was the top-most
> +			 * batch as pcpu_free() expects the per CPU pool
> +			 * to be less than ODEBUG_POOL_PERCPU_SIZE.
> +			 */
> +			if (unlikely(pcp->cnt < (ODEBUG_POOL_PERCPU_SIZE - ODEBUG_BATCH_SIZE) &&
> +				     !(pcp->cnt % ODEBUG_BATCH_SIZE))) {

It seems better to swap the contents on both sides of &&. Because the probability
of the current right side is lower.

> +				/*
> +				 * Don't try to allocate from the regular pool here
> +				 * to not exhaust it prematurely.
> +				 */
> +				if (pool_count(&pool_to_free)) {
> +					guard(raw_spinlock)(&pool_lock);
> +					pool_move_batch(pcp, &pool_to_free);
> +					pcpu_refill_stats();
> +				}
> +			}
>  			return obj;
>  		}
>  
> 
> .
> 

-- 
Regards,
  Zhen Lei