On Mon, 20 Apr 2009, Martin Fuzzey wrote: > Hi all, > > I'm working on a HCD for the i.MX21 and am having problems with DMA > alignment [my hardware requires DMA buffers to be aligned on a 4 byte > boundary] > > It seems that the usb core calls dma_map_single() from map_urb_for_dma() > but that this only ensures the memory is is a DMAable region, not that > it is correctly aligned. > > Use case is pwc webcam driver which calls usb_control_message() with a > non aligned stack allocated buffer pointer. That is a no-no. Drivers are never supposed to allocate buffers on the stack, for precisely this reason. They are supposed to use kmalloc or something similar. > Does this mean the HCD has to handle this? (I ask because a brief look > at the existing HCDs didn't turn up anything similar). > > If so there are probably two non aligned cases : > 1) small buffers (<max packet) where the data can be written directly to > hardware "data memory" with no DMA > 2) large buffers where the full buffer will have to be copied to a > suitably aligned buffer and DMA still used. > > Case 2) is problematic because the buffers could be large [though in > practice such buffers should be dynamically allocated and hence aligned]. The HCD is not supposed to handle this. It's a bug in the pwc driver. 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
