Message ID | 20230516161841.37138-7-aleksander.lobakin@intel.com (mailing list archive) |
---|---|
State | Superseded |
Delegated to: | Netdev Maintainers |
Headers | show |
Series | net: intel: start The Great Code Dedup + Page Pool for iavf | expand |
So while this looks ok, Eric Dumazet had looked into some optimization for this touching the core code, and promised me to come up with an even better version a while ago. Eric, what's the state of your optimizations for no-op DMA syncs?
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?
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? > > . >
From: Christoph Hellwig <hch@lst.de> Date: Wed, 17 May 2023 10:14:58 +0200 > So while this looks ok, Eric Dumazet had looked into some optimization > for this touching the core code, and promised me to come up with an > even better version a while ago. Eric, what's the state of your > optimizations for no-op DMA syncs? > Hmm, his last proposals were for avoiding indirect calls when IOMMU is on, but in fact we don't need to synchronize stuff -- DMA IOMMU on x86_64 also usually doesn't synchronize anything, but you don't know that prior to doing an indirect call and dma_need_sync() won't help. I was thinking of adding .dma_need_sync() callback to DMA ops, which could also be called once only on page allocation, like in this patch. Also want to hear how it goes for Eric :) Thanks, Olek
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
From: Jakub Kicinski <kuba@kernel.org> Date: Wed, 17 May 2023 21:08:04 -0700 > 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? When recycling is on, not that fast -- new allocations occur relatively rarely and it's allocations anyway, one `if` doesn't change anything there. And seems like I didn't get the sentence regarding "allocate and throw" :s This condition just disables the shortcut if any new page suddenly requires real DMA syncs (and if it does, then the sync a line below will take way more time than 1 more `if` anyway). Thanks, Olek
On Thu, 18 May 2023 15:34:44 +0200 Alexander Lobakin wrote: > And seems like I didn't get the sentence regarding "allocate and throw" > :s This condition just disables the shortcut if any new page suddenly > requires real DMA syncs (and if it does, then the sync a line below will > take way more time than 1 more `if` anyway). Right, I misread.
diff --git a/include/net/page_pool.h b/include/net/page_pool.h index c8ec2f34722b..8435013de06e 100644 --- a/include/net/page_pool.h +++ b/include/net/page_pool.h @@ -46,6 +46,9 @@ * device driver responsibility */ #define PP_FLAG_PAGE_FRAG BIT(2) /* for page frag feature */ +#define PP_FLAG_DMA_MAYBE_SYNC BIT(3) /* Internal, should not be used in the + * drivers + */ #define PP_FLAG_ALL (PP_FLAG_DMA_MAP |\ PP_FLAG_DMA_SYNC_DEV |\ PP_FLAG_PAGE_FRAG) diff --git a/net/core/page_pool.c b/net/core/page_pool.c index e212e9d7edcb..57f323dee6c4 100644 --- a/net/core/page_pool.c +++ b/net/core/page_pool.c @@ -175,6 +175,10 @@ static int page_pool_init(struct page_pool *pool, /* pool->p.offset has to be set according to the address * offset used by the DMA engine to start copying rx data */ + + /* 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; + } + if (pool->p.flags & PP_FLAG_DMA_SYNC_DEV) page_pool_dma_sync_for_device(pool, page, pool->p.max_len);
Turned out page_pool_put{,_full}_page() can burn quite a bunch of cycles even when on DMA-coherent platforms (like x86) with no active IOMMU or swiotlb, just for the call ladder. Indeed, it's page_pool_put_page() page_pool_put_defragged_page() <- external __page_pool_put_page() page_pool_dma_sync_for_device() <- non-inline dma_sync_single_range_for_device() dma_sync_single_for_device() <- external dma_direct_sync_single_for_device() dev_is_dma_coherent() <- exit For the inline functions, no guarantees the compiler won't uninline them (they're clearly not one-liners and sometimes compilers uninline even 2 + 2). The first external call is necessary, but the rest 2+ are done for nothing each time, plus a bunch of checks here and there. Since Page Pool mappings are long-term and for one "device + addr" pair dma_need_sync() will always return the same value (basically, whether it belongs to an swiotlb pool), addresses can be tested once right after they're obtained and the result can be reused until the page is unmapped. Define new PP flag, which will mean "do DMA syncs for device, but only when needed" and turn it on by default when the driver asks to sync pages. When a page is mapped, check whether it needs syncs and if so, replace that "sync when needed" back to "always do syncs" globally for the whole pool (better safe than sorry). As long as a pool has no pages requiring DMA syncs, this cuts off a good piece of calls and checks. On my x86_64, this gives from 2% to 5% performance benefit with no negative impact for cases when IOMMU is on and the shortcut can't be used. Signed-off-by: Alexander Lobakin <aleksander.lobakin@intel.com> --- include/net/page_pool.h | 3 +++ net/core/page_pool.c | 10 ++++++++++ 2 files changed, 13 insertions(+)