lkml.org 
[lkml]   [2024]   [May]   [6]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
SubjectRe: [PATCH net-next v4 6/7] page_pool: check for DMA sync shortcut earlier
From
From: Robin Murphy <robin.murphy@arm.com>
Date: Thu, 2 May 2024 16:49:48 +0100

> On 24/04/2024 9:52 am, Alexander Lobakin wrote:
>> From: Alexander Lobakin <aleksander.lobakin@intel.com>
>> Date: Tue, 23 Apr 2024 15:58:31 +0200
>>
>>> We can save a couple more function calls in the Page Pool code if we
>>> check for dma_need_sync() earlier, just when we test pp->p.dma_sync.
>>> Move both these checks into an inline wrapper and call the PP wrapper
>>> over the generic DMA sync function only when both are true.
>>> You can't cache the result of dma_need_sync() in &page_pool, as it may
>>> change anytime if an SWIOTLB buffer is allocated or mapped.
>>>
>>> Signed-off-by: Alexander Lobakin <aleksander.lobakin@intel.com>
>>> ---
>>>   net/core/page_pool.c | 31 +++++++++++++++++--------------
>>>   1 file changed, 17 insertions(+), 14 deletions(-)
>>>
>>> diff --git a/net/core/page_pool.c b/net/core/page_pool.c
>>> index 6cf26a68fa91..87319c6365e0 100644
>>> --- a/net/core/page_pool.c
>>> +++ b/net/core/page_pool.c
>>> @@ -398,16 +398,24 @@ static struct page
>>> *__page_pool_get_cached(struct page_pool *pool)
>>>       return page;
>>>   }
>>>   -static void page_pool_dma_sync_for_device(const struct page_pool
>>> *pool,
>>> -                      const struct page *page,
>>> -                      unsigned int dma_sync_size)
>>> +static void __page_pool_dma_sync_for_device(const struct page_pool
>>> *pool,
>>> +                        const struct page *page,
>>> +                        u32 dma_sync_size)
>>>   {
>>>       dma_addr_t dma_addr = page_pool_get_dma_addr(page);
>>>         dma_sync_size = min(dma_sync_size, pool->p.max_len);
>>> -    dma_sync_single_range_for_device(pool->p.dev, dma_addr,
>>> -                     pool->p.offset, dma_sync_size,
>>> -                     pool->p.dma_dir);
>>> +    __dma_sync_single_for_device(pool->p.dev, dma_addr +
>>> pool->p.offset,
>>> +                     dma_sync_size, pool->p.dma_dir);
>>
>> Breh, this breaks builds on !DMA_NEED_SYNC systems, as this function
>> isn't defined there, and my CI didn't catch it in time... I'll ifdef
>> this in the next spin after some reviews for this one.
>
> Hmm, the way things have ended up, do we even need this change? (Other
> than factoring out the pool->dma_sync check, which is clearly nice)
>
> Since __page_pool_dma_sync_for_device() is never called directly, the
> flow we always get is:
>
> page_pool_dma_sync_for_device()
>     dma_dev_need_sync()
>         __page_pool_dma_sync_for_device()
>             ... // a handful of trivial arithmetic
>             __dma_sync_single_for_device()
>
> i.e. still effectively the same order of
> "if (dma_dev_need_sync()) __dma_sync()" invocations as if we'd just used
> the standard dma_sync(), so referencing the unwrapped internals only
> spreads it out a bit more for no real benefit. As an experiment I tried
> the diff below on top, effectively undoing this problematic part, and it
> doesn't seem to make any appreciable difference in a before-and-after
> comparison of the object code, at least for my arm64 build.
>
> Thanks,
> Robin.
>
> ----->8-----
> diff --git a/net/core/page_pool.c b/net/core/page_pool.c
> index 27f3b6db800e..b8ab7d4ca229 100644
> --- a/net/core/page_pool.c
> +++ b/net/core/page_pool.c
> @@ -398,24 +398,20 @@ static struct page *__page_pool_get_cached(struct
> page_pool *pool)
>      return page;
>  }
>  
> -static void __page_pool_dma_sync_for_device(const struct page_pool *pool,
> -                        const struct page *page,
> -                        u32 dma_sync_size)
> -{
> -    dma_addr_t dma_addr = page_pool_get_dma_addr(page);
> -
> -    dma_sync_size = min(dma_sync_size, pool->p.max_len);
> -    __dma_sync_single_for_device(pool->p.dev, dma_addr + pool->p.offset,
> -                     dma_sync_size, pool->p.dma_dir);
> -}
> -
>  static __always_inline void

So this would unconditionally inline the whole sync code into the call
sites, while I wanted to give a chance for the compilers to uninline
__page_pool_dma_sync_for_device().

>  page_pool_dma_sync_for_device(const struct page_pool *pool,
>                    const struct page *page,
>                    u32 dma_sync_size)
>  {
> -    if (pool->dma_sync && dma_dev_need_sync(pool->p.dev))
> -        __page_pool_dma_sync_for_device(pool, page, dma_sync_size);
> +    dma_addr_t dma_addr = page_pool_get_dma_addr(page);
> +
> +    if (!pool->dma_sync)
> +        return;
> +
> +    dma_sync_size = min(dma_sync_size, pool->p.max_len);
> +    dma_sync_single_range_for_device(pool->p.dev, dma_addr,
> +                     pool->p.offset, dma_sync_size,
> +                     pool->p.dma_dir);
>  }
>  
>  static bool page_pool_dma_map(struct page_pool *pool, struct page *page)

Thanks,
Olek

\
 
 \ /
  Last update: 2024-05-27 18:16    [W:0.063 / U:1.212 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site