At Tue, 06 May 2008 21:25:53 +1000, Benjamin Herrenschmidt wrote: > > > On Tue, 2008-05-06 at 13:14 +0200, Takashi Iwai wrote: > > > sg allocations are usually lists of page, so virt_to_page shouldn't > > be a > > > problem in the first place, though we still need some way to get the > > > right prot attributes. > > > > The problem is that ALSA SG buffer is composed from pages allocated > > via dma_alloc_coherent(), and the pages are retrieved via virt_to_page > > for DMA pages, as found in sound/core/sgbuf.c. > > > > I didn't care about SG-buffer cases in my initial patch, and it's a > > bit messy. The current code requires page lists becasue it calls > > vmap() to get a virtually linear buffer. > > Hrm... the problem is that you aren't supposed to make up sglists with > the result of dma_alloc_coherent... It might be a limitation of our core > DMA API, but that's we have to deal with today.. > > If you're going to make up sglists and call vmap, you should allocate > pages with normal GFP. If that is a problem vs. DMA'bility of those > pages, then ... we have a problem :-) > > I don't think we can easily update the DMA API at this stage. What we > could do is provide a way to retrieve the struct page array from the > result of dma_alloc_coherent...
In most cases, it can be obtained via pfn_to_page(), I suppose. But, it's definitely arch-specific thingy, and a generic solution would be really appreciated. Alternatively, we can change the ALSA PCM core code that accesses the virtual linear buffer and handles SG-buffers as they are. Maybe it'll give a bit more useful clean-up in the whole memory-management codes in ALSA in the end. Takashi _______________________________________________ Linuxppc-dev mailing list Linuxppc-dev@ozlabs.org https://ozlabs.org/mailman/listinfo/linuxppc-dev
