On Wed, 2 Jul 2014, David Laight wrote: > From: Raghavendra > > Thank you for responding Peter, > > > Raghavendra wrote: > > >> I have a query regarding DMA(Direct Memory Access) for the usb devices. > > > USB devices never do DMA. > > > > > >> As far as Linux is concerned, how the DMA action being taking place for > > >> USB devices. > > > It doesn't take place. > > > > > >> As per my understanding, the USB host controller is taking care of > > >> the DMA operations. > > > That's correct. When an URB is submitted by host software the host > > > controller asks the device for data, and when data arrives from the > > > device the host controller writes the data into memory using DMA. > > As I understand, the USB host controllers(on x86) are basically PCI > > devices and PCI devices are capable of bus mastering, thus performing DMA. > > But, if we take the case of a non x86 platform (say ARM), the USB host > > controller may or may not be capable of performing DMA. That's right. > > In that case how > > far the cases are valid if we try to push the data from the USB device > > through DMA > > For example, if we take the case of a simple USB mouse > > driver(usbmouse.c), it uses DMA to obtain the data. DMA gets used if the host controller supports it. > > What if this driver > > is associated with a host controller which doesnt have the support for > > DMA. What will happen in that scenario? Then DMA isn't used. PIO gets used instead. > It is extremely unlikely that a USB host controller will be unable to > do DMA - in the sense that it transfers USB data to/from host memory > buffers. There are plenty of USB host controllers that can't do DMA. The kernel has to take responsibility for transferring data between the controller memory and the host memory, using PIO. > However the host buffer addresses are not made available to the USB > target hardware - so it can't select which buffer is used. > > It is just possible that the memory the host controller uses is resident > on (say) a PCI card. But the normal bandwidth requirements make that > unlikely. Host controllers always have data buffers. > Another exception might be for slow devices (keyboard/mouse) where the > amount of data is very limited - but that won't be a host. 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
