Am Montag, den 26.02.2018, 18:09 +0200 schrieb Ran Shalit: > On Mon, Feb 26, 2018 at 5:59 PM, Oliver Neukum <oneukum@xxxxxxxx> wrote: > > > > Am Donnerstag, den 15.02.2018, 11:00 +0200 schrieb Ran Shalit: > > > > > > I actually asked about DMA, because I see that it is possible to send > > > urb using DMA allocated buffers or not (coherent and non-coherent) > > > usb_buffer_alloc(). > > > So, How can we actually know if I should use this API with a device or > > > not ? I mean, is it always possible to use the coherent buffer with > > > the device ? > > > > Hi, > > > > it is always possible to use usb_alloc_coherent() on a host. > > But it is generally not sensible. It is for buffers to be reused > > many times. In general use kmalloc() [once per buffer. That is a rule. > > You must not share them.] > > > > HTH > > Oliver > > > Hi Oliver, > > Is the dma engine which is responsible for the transaction is actually > in the usb device (It's not dma controller in host) ? OK, I think there is a fundamental misunderstanding here. There are two sides: host and gadget Either, both or no side may use DMA. The other side does not learn of that. In either case, however, you write a driver for a logical device or a protocol, if you will. What we call a device driver on the host (the gadget has its own equivalents) does not talk to hardware. Essentially you allocate and manipulate URBs. There is a driver for a host (respectively gadget) controller which talks to hardware. It does DMA, not your driver. > If I understand correctly in both ways (kmalloc or > usb_alloc_coherent), then we are allocating buffers might be used for > dma. Yes. We must. > Even if DMA is not used in the transaction, than is shall still be > functional with the same DMA-allocated buffers. Yes. It must. HTH Oliver -- 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
