On Mon, 31 Dec 2018, Mathias Nyman wrote: > On 21.12.2018 04:58, Alan Stern wrote: > > On Thu, 20 Dec 2018, Marc MERLIN wrote: > > > >> On Mon, Dec 17, 2018 at 09:45:44PM -0800, Marc MERLIN wrote: > >>> On Mon, Dec 17, 2018 at 10:52:57AM -0500, Alan Stern wrote: > >>>>> Point taken. I think I was thinking about USB ports that were routed to > >>>>> either a xHCI or an EHCI chip depending on what's plugged into them, but > >>>>> seems that I didn't quite understand how that worked. > >>>> > >>>> It varies. For example, my office computer does exactly what you were > >>>> thinking: It routes SuperSpeed connections to the xHCI controller and > >>>> it routes high speed connections (even on the same port!) to one of the > >>>> EHCI controllers. > >>> > >>> Thanks, so I'm not crazy, I thought I had seen this before. > >>> I had a look at my work server today and thankfully it at least had a > >>> USB3 setting, which when I turned it off, xhci was replaced with ehci as > >>> expected. > >> > >> Again, thanks Alan. > >> I just wrote a web page that hopefully google will index and show in > >> addition or in front of the other page I mentioned and that wasn't clear > >> enough about the 32 device limit being due to the xhci controller being > >> active, not due to USB3 devices being on the bus, or not. > >> > >> http://marc.merlins.org/perso/linux/post_2018-12-20_Getting-Around-USB3-xhci-32-Device-Limit-_Max-number-of-devices-this-xHCI-host-supports-is-32_.html > >> > >> Further reading: > >> https://forums.intel.com/s/question/0D50P00004905stSAA/hardware-limitations-on-usb-endpoints-xhci?language=en_US > >> > >> Thanks again, > > > > Just a small addition, many Intel xHCI controllers now support 64 devices. > > It's possible to get the max device slots xHCI hardware supports from a xHC register. > (bits 7:0 of the HCSPARAMS1 capability register) > > This can be found from debugfs, but it's not very practical: > > mount -t debugfs none /sys/kernel/debug/ > grep HCSPARAMS1 /sys/kernel/debug/usb/xhci/0000\:00\:14.0/reg-cap > HCSPARAMS1 = 0x1a000840 > > We only care about bits 7:0, by ignoring the other bits we get 0x00000040, > which is 64 in decimal. > > So this xHCI supports 64 device slots. The limits that Marc encountered were on both the number of device slots and the number of endpoint slots. In his case, each device has 3 endpoints. Alan Stern
