Adding Felipe. Xenia, Felipe is the USB "gadget" maintainer, and also added code for non-PCI xHCI hosts. Felipe, Xenia is my summer intern through the FOSS Outreach Program for Women: http://kernelnewbies.org/OPWIntro Felipe, we're discussing the fact that the xHCI host currently does not set up 64-bit coherent DMA, so we get 64-bit addresses for USB buffers, but not xHCI data structures. Xenia has a patch to fix it, which can be found here: http://marc.info/?l=linux-usb&m=137052855612729&w=2 Andy thinks the patch has issues with non-PCI xHCI hosts. On Fri, Jun 07, 2013 at 03:00:47PM +0300, Mathias Nyman wrote: > Andy explained his concern in more detail to me irl. > > The beef is that the dma_mask is a pointer while dma_coherent_mask > is a u64 value in struct device, and dma_set_mask() fails if > dev->dma_mask doesn't point anywhere. > > This is not a problem for PCI enumerated devices because PCI probe > sets a default 32bit mask and makes the dma_mask of all new PCI > devices point to this value. > > For non-pci enumerated devices the device dma_mask may be NULL and > dma_set_mask() may fail even if dma is supported on the host > machine. > > for example ehci-platform.c solves this by adding: > > if (!dev->dev.dma_mask) > dev->dev.dma_mask = &dev->dev.coherent_dma_mask; > if (!dev->dev.coherent_dma_mask) > dev->dev.coherent_dma_mask = DMA_BIT_MASK(32); So it sounds like xhci_plat_setup() in xhci-plat.c needs to do something similar before it calls xhci_gen_setup(). Felipe, do you agree? It's funny that the code worked before. Perhaps setting the dma_mask pointer should be a separate patch for stable? Sarah Sharp -- 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
