From: Jarkko Sakkinen > Sent: 12 October 2021 18:41 > > On Tue, 2021-10-12 at 12:43 -0300, Jason Gunthorpe wrote: > > On Tue, Oct 12, 2021 at 06:29:58PM +0300, Jarkko Sakkinen wrote: > > > On Mon, 2021-10-11 at 00:01 +0800, Cai Huoqing wrote: > > > > Replacing kmalloc/kfree/get_zeroed_page/free_page/dma_map_single/ > > > ~~~~~~~~~ > > > Replace > > > > > > > dma_unmap_single() with dma_alloc_coherent/dma_free_coherent() > > > > helps to reduce code size, and simplify the code, and coherent > > > > DMA will not clear the cache every time. > > > > > > > > Signed-off-by: Cai Huoqing <caihuoqing@xxxxxxxxx> > > > > > > If this does not do functionally anything useful, there's no > > > reason to apply this. > > > > At least in this case it looks like the ibmvtpm is not using the DMA > > API properly. There is no sync after each data transfer. Replacing > > this wrong usage with the coherent API is reasonable. > > Thank you. As long as this is documented to the commit message, > I'm cool with the change itself. > > E.g. something like this would be perfectly fine replacement for the > current commit message: > > "The current usage pattern for the DMA API is inappropriate, as > data transfers are not synced. Replace the existing DMA code > with the coherent DMA API." Why not also say that the DMA access snoop the cache? (I think that was mentioned earlier in the thread.) David - Registered Address Lakeside, Bramley Road, Mount Farm, Milton Keynes, MK1 1PT, UK Registration No: 1397386 (Wales)
