Re: [PATCH net-next 06/11] net: page_pool: avoid calling no-op externals when possible

[Alexander Lobakin <aleksander.lobakin@intel.com>]

From: Yunsheng Lin <linyunsheng@huawei.com>
Date: Thu, 18 May 2023 12:54:37 +0800

> On 2023/5/18 12:08, Jakub Kicinski wrote:
>> On Tue, 16 May 2023 18:18:36 +0200 Alexander Lobakin wrote:
>>> +		/* Try to avoid calling no-op syncs */
>>> +		pool->p.flags |= PP_FLAG_DMA_MAYBE_SYNC;
>>> +		pool->p.flags &= ~PP_FLAG_DMA_SYNC_DEV;
>>>  	}
>>>  
>>>  	if (PAGE_POOL_DMA_USE_PP_FRAG_COUNT &&
>>> @@ -323,6 +327,12 @@ static bool page_pool_dma_map(struct page_pool *pool, struct page *page)
>>>  
>>>  	page_pool_set_dma_addr(page, dma);
>>>  
>>> +	if ((pool->p.flags & PP_FLAG_DMA_MAYBE_SYNC) &&
>>> +	    dma_need_sync(pool->p.dev, dma)) {
>>> +		pool->p.flags |= PP_FLAG_DMA_SYNC_DEV;
>>> +		pool->p.flags &= ~PP_FLAG_DMA_MAYBE_SYNC;
>>> +	}
>>
>> is it just me or does it feel cleaner to allocate a page at init,
>> and throw it into the cache, rather than adding a condition to a
>> fast(ish) path?
> 
> Is dma_need_sync() not reliable until a dma map is called?
> Is there any reason why not just clear PP_FLAG_DMA_SYNC_DEV if
> dma_need_sync() is false without introducing the PP_FLAG_DMA_MAYBE_SYNC
> flag?

We can't just clear the flag, because some drivers don't want PP to
synchronize DMA. Without a new flag, we can't distinguish those two.
Example:

1) Driver doesn't set DMA_SYNC_DEV
2) We check for dma_need_sync() and it returns true
3) As a result, we set DMA_SYNC_DEV, although driver does that on its
   own

Thanks,
Olek