On Wed, 1 Sep 2010, David Vrabel wrote: > On 01/09/2010 09:18, Oliver Neukum wrote: > > > > 1. Drivers mapping their own buffers. > > The problem here is that I am not sure we have a valid pointer > > which can support a copy to a bounce buffer. > > I don't see why drivers need to map their own buffers or use coherent > allocations. Should this ability be removed? It doesn't play nice with > usbmon or drivers needing to do PIO either. It should work okay for usbmon and PIO. The buffers returned by usb_alloc_coherent() are always CPU-addressable. No, the problems with usbmon and PIO occur when drivers use the scatter-gather interface. Then there _is_ no buffer pointer, just a page and a DMA address. We use coherent buffers only for the benefit of driver that need to do a lot of rapid I/O repeatedly using the same buffer. Removing them would slow such drivers down a little -- I don't know how much. Alan Stern -- To unsubscribe from this list: send the line "unsubscribe linux-usb" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
