On Tue, 14 May 2013, Sarah Sharp wrote: > On Wed, May 15, 2013 at 04:56:00AM +0300, Xenia Ragiadakou wrote: > > Hi Sarah, > > Hi Ksenia! > > > In xhci.c line 4665, the function dma_set_mask is used to enable > > 64-bit DMA addressing > > (if i understood correctly). > > > > However, accordingly to the Documentation/DMA-API-HOWTO.txt, there > > is possibility > > that the platform cannot support the DMA addressing of the device. > > So the return value of dma_set_mask needs to be tested if it equals > > zero and if not > > go to error label. Am I right? > > > > Can we assume that 64 bit or 32 bit DMA addressing will always be supported? > > This is the first time i come through these stuff so my intuition is > > limited. > > All xHCI hosts are required to support DMA, by the xHCI specification. > You just can't get USB 3.0 speeds without DMA. > > Or are you concerned that the xHCI host will support 64-bit DMA, but the > system only supports 32-bit DMA? Because that would be a question for > Alan Stern, who has worked on the USB core for far longer than I have. I suppose it is possible. It's also possible that the xHCI hardware might support 32-bit DMA only. > The xHCI PCI driver sets the hc_driver->flag for HCD_MEMORY, which means > the USB host controller requires DMA. Actually it means that the host controller uses memory-mapped I/O for register accesses. HCD_MEMORY has no connection with DMA. Maybe you're thinking of the uses_dma flag in struct usb_bus. > In drivers/usb/core/hcd-pci.c > (which is what the xHCI PCI driver calls into in its probe function), > the USB core will fail the init if the device doesn't support DMA. No, it won't. The uses_dma flag gets set during initialization in usb_create_shared_hcd(), but it isn't checked at all during init. 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
