On Mon, 12 Jun 2023 21:02:52 +0800 Yunsheng Lin wrote: > Currently page_pool_alloc_frag() is not supported in 32-bit > arch with 64-bit DMA, which seems to be quite common, see > [1], which means driver may need to handle it when using > page_pool_alloc_frag() API. > > In order to simplify the driver's work for supporting page > frag, this patch allows page_pool_alloc_frag() to call > page_pool_alloc_pages() to return a big page frag without it returns an entire (potentially compound) page, not a frag. AFAICT > page splitting because of overlap issue between pp_frag_count > and dma_addr_upper in 'struct page' for those arches. These two lines seem to belong in the first paragraph, > As page_pool_create() with PP_FLAG_PAGE_FRAG is supported in "is" -> "will now be" > 32-bit arch with 64-bit DMA now, mlx5 calls page_pool_create() > with PP_FLAG_PAGE_FRAG and manipulate the page->pp_frag_count > directly using the page_pool_defrag_page(), so add a checking > for it to aoivd writing to page->pp_frag_count that may not > exist in some arch. This paragraph needs some proof reading :( > Note that it may aggravate truesize underestimate problem for > skb as there is no page splitting for those pages, if driver > need a accuate truesize, it may calculate that according to accurate > frag size, page order and PAGE_POOL_DMA_USE_PP_FRAG_COUNT > being true or not. And we may provide a helper for that if it > turns out to be helpful. > > 1. https://lore.kernel.org/all/20211117075652.58299-1-linyunsheng@xxxxxxxxxx/ > + /* Return error here to avoid writing to page->pp_frag_count in > + * mlx5e_page_release_fragmented() for page->pp_frag_count is I don't see any direct access to pp_frag_count anywhere outside of page_pool.h in net-next. PAGE_POOL_DMA_USE_PP_FRAG_COUNT sounds like an internal flag, drivers shouldn't be looking at it, IMO.
