On Sat, 11 Jan 2020, Pete Zaitcev wrote: > On Wed, 8 Jan 2020 16:55:13 +0100 > Greg Kroah-Hartman <gregkh@xxxxxxxxxxxxxxxxxxx> wrote: > > > On Mon, Jan 06, 2020 at 10:37:17AM +0100, Tomasz Moń wrote: > > > Make USB device addresses match while sniffing USB communication > > > with usbmon and hardware USB sniffer (OpenVizsla) at the same time. > > > On xHCI root hubs the address is assigned by hardware and can be > > > different than devnum. > > > > This breaks the userspace abi for matching up the devnum with the number > > that is listed by the kernel to userspace, making it really hard to > > match things up for xhci devices now :( > > > > I understand the need to look at this data, but you can't do it in a way > > that will change things like this. > > This sounds reasonable, although perhaps unfortunate. I acked Tomasz's > patch because I thought that XHCI is new, and thus it's no big deal > if its results are different. > > Of course I rushed to examine the packet structure, but yes, there's > no space in the header, not without some trickery. For example, the > upper 8 bits of the ID are probably the same for all packets, so > it may be stuffed in there. Or, it might be possible to create an > extra header and attach it at the end of ISO descriptors. > > Alan's suggestion of leaving the physical address in /sys appeals > the most to me, honestly. One thing though, libpcap will need to > rifle through /sys and then store that address, so its serialization > has to be changed no matter what. Unfortunately, I'm wholly ignorant > as to what syntax it uses and how extensible it is. Alternatively, libpcap can ignore the issue and just display the device numbers, as it does now. A separate program or the user could convert the number to a physical address, if necessary, using the information in sysfs. Alan Stern
