Am 20.04.2017 um 21:58 schrieb Rabin Vincent: > On Thu, Apr 20, 2017 at 11:27:38AM -0700, Eric Anholt wrote: >> I'm confused by what you're saying here. The driver has already been >> converted to not use dmac_map_area (commit >> cf9caf1929882b66922aee698e99e6c8f357bee5), and uses dma_map_sg instead, >> matching the radeon driver you give as a model as far as I can see. >> That commit is in v4.11-rc6 from Stefan's regression report. > Right. Sorry. I must have had an old tag checked out when I looked at > the driver earlier. The DMA API usage in the driver in v4.11-rc6 and > current master looks fine, except for one thing: > > The flush in flush_dcache_page() (from get_user_pages()) was done with a > v6_flush_kern_dcache_page() which always did a clean+invalidate while > the DMA API only does what is required by the direction, which is only a > invalidate for DMA_FROM_DEVICE. Since the driver calls dma_from_sg() on > the entire page, even if userspace sent in an offset into the page, > unrelated data in userspace may be thrown away. > > Does changing the dma API calls to always use DMA_BIDIRECTIONAL make the > test pass? Unfortunately not (at least not that simple). Do we need special DMA mapping attributes here ? Or a dma_sync_sg_for_* ? _______________________________________________ devel mailing list devel@xxxxxxxxxxxxxxxxxxxxxx http://driverdev.linuxdriverproject.org/mailman/listinfo/driverdev-devel
