Messages in this thread |  | | Date | Thu, 17 Oct 2013 10:30:33 -0400 (EDT) | From | Alan Stern <> | Subject | Re: [PATCH v2] xhci: fix usb3 streams |
| |
On Wed, 16 Oct 2013, Sarah Sharp wrote:
> > > xhci maintains a radix tree for each stream endpoint because it must > > > be able to map a trb address to the stream ring. Each ring segment > > > must be added to the ring for this to work. Currently xhci sticks > > > only the first segment of each stream ring into the radix tree.
> > There may be a simpler approach to this problem. > > > > When using a new ring segment, keep the first TRB entry in reserve. > > Don't put a normal TRB in there, instead leave it as a no-op entry > > containing a pointer to the stream ring. (Make the prior Link TRB > > point to the second entry in the new segment instead of the first.) > > > > Then you won't have to add to or remove anything from the radix tree. > > I don't understand how this would help. Are you advocating a different > way of mapping TRB DMA addresses to stream rings that would allow us to > ditch the radix tree all together? > > Ok, so with your solution, we have a virtual stream ring pointer as the > first TRB of the segment. We get an event with the DMA address of a TRB > in one of many stream rings on an endpoint. From that, I think we can > infer the DMA address of the first TRB on the segment, due to the > alignment requirements and ring size. > > And then what do we do with that? We don't have the virtual address of > that first TRB, so the xHCI driver can't read the ring pointer from it. > I'm confused as to what the next steps would be to solve this.
My mistake; I misunderstood the original description of the problem. I didn't realize that "map a trb address" referred to the TRB's DMA address.
BTW, ohci-hcd faces the same problem (of mapping DMA addresses to kernel addresses). It solves the problem with a hash table rather than a radix tree.
Alan Stern
|  |