On Tue, Jun 09, 2020 at 11:09:14AM +0200, Takashi Iwai wrote: > On Tue, 09 Jun 2020 10:43:05 +0200, > Christoph Hellwig wrote: > > > > On Tue, Jun 09, 2020 at 10:05:26AM +0200, Takashi Iwai wrote: > > > > >From the disassembly it seems like a vmalloc allocation is NULL, which > > > > seems really weird as this patch shouldn't make a difference for them, > > > > and I also only see a single places that allocates the field, and that > > > > checks for an allocation failure. But the sound code is a little > > > > hard to unwind sometimes. > > > > > > It's not clear which sound device being affected, but if it's > > > HD-audio on x86, runtime->dma_area points to a vmapped buffer from > > > SG-pages allocated by dma_alloc_coherent(). > > > > > > OTOH, if it's a USB-audio, runtime->dma_area is a buffer by > > > vmalloc(). > > > > Err, you can't just vmap a buffer returned from dma_alloc_coherent, > > dma_alloc_coherent returns values are opaque and can't be used > > for virt_to_page. Whatever that code did has already been broken > > per the DMA API contract and on many architectures and just happend > > to work on x86 by accident. > > Hmm, that's bad. > > How would be a proper way to get the virtually mapped SG-buffer pages > with coherent memory? (Also allowing user-space mmap, too) dma_mmap_coherent / dma_mmap_attrs for userspace. We don't really have a good way for kernel space mappings.
