Well, it could be a timing related issue that causes a host controller issue to appear. If the path down the stack is slowed when the buffers in the high mem (above 4GB) get remapped into bounce buffers, then possibly the writes to the endpoint ring take more time. Since isoc is time critical, and the host controller checks the ring every time the service interval comes around, rather than when the endpoint doorbell is rung, it's possible there is some sort of hardware race condition. That could cause the NEC host to give bad DMA pointers in the events on the event ring. But this is all theoretical until we can get some more data about what's actually happening. Sarah Sharp On Fri, Oct 01, 2010 at 11:22:14AM +0200, Sander Eikelenboom wrote: > Hello Andiry, > > Could be it's not isoc specific, but i don't see it on ehci/ohci controllers i passthrough. > And it could be somewhere in cleanup code paths that are not frequently hit. > I will try to change some of the dbg output to warn's, perhaps that sheds a light on where the transfer error come from in the first place .. > > -- > > Sander > > Friday, October 1, 2010, 11:14:03 AM, you wrote: > > > Hi Sander, > > > That's interesting findings. I don't think the isoc patches do anything > > differently above the 4Gb mem boundary. I'll try it on my side. > > > Thanks & Best regards, > > Andiry > > > > >> -----Original Message----- > >> From: Sander Eikelenboom [mailto:linux@xxxxxxxxxxxxxx] > >> Sent: Friday, October 01, 2010 4:28 AM > >> To: Xu, Andiry > >> Cc: Sarah Sharp; linux-usb@xxxxxxxxxxxxxxx > >> Subject: Re: XHCI freezes, most up to date patches / tree ? > >> > >> Hi Andiry, > >> > >> It seems i have narrowed down my problem a bit. > >> My system has 8Gb of ram, when i restrict the hypervisor to use only > > 4G > >> the system doesn't freeze. > >> When using the full 8Gb it does (although the guests never get more > > than > >> 1G of mem) > >> > >> When i compare the logs of the 2 cases i get in the guest with the > > xhci > >> controller passed through to it: > >> > >> When restricting to 4G: > >> Sep 28 17:55:02 security kernel: [81744.078288] xhci_hcd 0000:07:00.0: > >> WARN: transfer error on endpoint > >> Sep 28 17:55:02 security kernel: [81744.092653] xhci_hcd 0000:07:00.0: > >> WARN: transfer error on endpoint > >> Sep 28 17:55:02 security kernel: [81744.093647] xhci_hcd 0000:07:00.0: > >> WARN: transfer error on endpoint > >> Sep 28 17:55:02 security kernel: [81744.093647] xhci_hcd 0000:07:00.0: > >> WARN: transfer error on endpoint > >> Sep 28 17:55:02 security kernel: [81744.093647] xhci_hcd 0000:07:00.0: > >> WARN: transfer error on endpoint > >> > >> Without restrictions, i don't get these before the system freezes. > >> It's a hunch, but perhaps this could be related, because when it > > freezes > >> it hasn't got time to write the warning to the syslog and to disk. > >> > >> Is there anything in the cleanup of this warning related to dma/msi-x > >> interrupts that could write or do something that could react > > differently > >> above the 4Gb mem boundary ? > >> > >> -- > >> > >> Sander > >> > >> > >> Tuesday, September 14, 2010, 10:31:21 AM, you wrote: > >> > >> > Hi Sander, > >> > >> > Paul observed freeze issue, but in his last mail, his xHCI > > controller > >> > has been updated and the symptom is observed much less frequently. > > And I > >> > haven't seen his update log since then. > >> > >> > Thanks & Best regards, > >> > Andiry > >> > > >> > >> >> -----Original Message----- > >> >> From: Sander Eikelenboom [mailto:linux@xxxxxxxxxxxxxx] > >> >> Sent: Tuesday, September 14, 2010 4:01 PM > >> >> To: Xu, Andiry > >> >> Cc: Sarah Sharp; linux-usb@xxxxxxxxxxxxxxx > >> >> Subject: Re: XHCI freezes, most up to date patches / tree ? > >> >> > >> >> Hi Andiry, > >> >> > >> >> Yes, since i was testing on baremetal instead of in a VM, i had to > > use > >> >> another program to do the videograbbing, it turned out I was trying > > to > >> > use > >> >> the old v4l1 interface instead of v4l2. > >> >> That should have worked, but resulted in an Oops, using v4l2 (as > >> > intended) > >> >> did work. > >> >> > >> >> During my vacation, I have had the grabber passed through to the VM > >> > (on > >> >> USB2) and it is running for +20 days now (24/7 grabbing), the > > machine > >> > had > >> >> a lot of disk and network access as well. > >> >> So the setup in essence could work, still having to find out what > >> > exactly > >> >> in passthrough in combination with XHCI makes it freeze completely > >> > without > >> >> any Oops or Trace. > >> >> That's why I was interessted in the thread with Paul Zimmerman, > >> > although i > >> >> didn't see a conclusion. > >> >> > >> >> -- > >> >> Sander > >> >> > >> >> > >> >> Tuesday, September 14, 2010, 9:53:00 AM, you wrote: > >> >> > >> >> > Hi Sander, > >> >> > >> >> > As far as I know, there is no recent update patch for xhci isoc. > >> >> > >> >> > In your last mail you mentioned you also observed em28xx driver > > on > >> > USB2 > >> >> > issue. Is that fixed? > >> >> > >> >> > Please apply the patch I sent to you last time when you catch the > >> > freeze > >> >> > symptom. > >> >> > >> >> > Thanks & Best regards, > >> >> > Andiry > >> >> > > >> >> > >> >> >> -----Original Message----- > >> >> >> From: Sander Eikelenboom [mailto:linux@xxxxxxxxxxxxxx] > >> >> >> Sent: Monday, September 13, 2010 7:32 PM > >> >> >> To: Xu, Andiry > >> >> >> Cc: Sarah Sharp; linux-usb@xxxxxxxxxxxxxxx > >> >> >> Subject: XHCI freezes, most up to date patches / tree ? > >> >> >> > >> >> >> Hi Andiry, > >> >> >> > >> >> >> After my two week holiday, i would like to resume my quest in > >> > finding > >> >> > my > >> >> >> problem where my USB3 passthrough on Xen seems to freeze my > >> > complete > >> >> >> system. > >> >> >> It seems to be DMA related, although after consulting with the > > guy > >> > who > >> >> >> works on the pci passthrough / swiotlb on Xen he couldn't find > > Xen > >> >> > related > >> >> >> problems at first hand. > >> >> >> > >> >> >> Of special interest to me is the thread by "Paul Zimmerman" on > > the > >> >> > usb- > >> >> >> devel mailing list. Also reporting a freeze. > >> >> >> I didn't see any outcome of that discussion. > >> >> >> > >> >> >> Is there somewhere were i can get the most recent set of patches > >> > for > >> >> >> xhci/isoc, or is linus 2.6.36-rc4 tree up to date ? > >> >> >> > >> >> >> -- > >> >> >> Sander > >> >> >> > >> >> > >> >> > >> >> > >> >> > >> >> > >> >> -- > >> >> Best regards, > >> >> Sander mailto:linux@xxxxxxxxxxxxxx > >> >> > >> > >> > >> > >> > >> > >> -- > >> Best regards, > >> Sander mailto:linux@xxxxxxxxxxxxxx > >> > > > > > > -- > Best regards, > Sander mailto:linux@xxxxxxxxxxxxxx > -- 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
