On Fri, May 12, 2023 at 12:19:09AM +0800, Ruihan Li wrote: > On Thu, May 11, 2023 at 08:32:01AM -0700, Christoph Hellwig wrote: > > On Thu, May 11, 2023 at 09:44:55PM +0800, Ruihan Li wrote: > > > Christoph's patch perfectly fixes _one_ problem: kmalloc'ed memory > > > cannot be mapped to user space. However, as I detailed in the commit > > > message, this series of patches fixes _three_ problems. > > > > FYI, I agree with you. My simple patch was sent before reading > > your new series, and is a strict subset of it. > > Thank you for the clarification. > > > > I have to say that the original code is quite buggy. In the > > > gen_pool_dma_alloc path, there is no guarantee of page alignment. > > > > I also find this whole interface very problematic to start with, > > but that's a separate discussion for later. > > Yes. I don't think hybrid allocation of DMA memory and normal memory in > one function is a good thing, but currently there is no clear way to fix > this. Mixing memory allocation and page allocation is another bad thing, > and at least, my patch can (hopefully) solve the second (and much > easier) issue. Ok, I'll take these through the usb tree if I can get an ack for the mm-specific patches. Or were you going to send a v2 series? thanks, greg k-h
