On Thu, Feb 21, 2013 at 1:42 PM, Alan Stern <stern@xxxxxxxxxxxxxxxxxxx> wrote: > On Wed, 20 Feb 2013, Ronald wrote: > >> Paul Hartman pointed me at the kernel cmdline option log_buf_len=16M. >> It's kind of confusing that the compile time option is mentioning a >> limit while the kernel cmdline option is limit free. I forgot about >> those differences. Attached is a complete log without wrapping (thanks >> Paul!). I have annotated the events with markers such as '::::::'. >> >> If this message fails to make it to the list, then I will post another >> scribd link. > > The output is all too clear. Both of you are using EHCI controllers > made by VIA, and they both exhibit this same problem -- which is > clearly caused by a silicon bug. > > It's entirely possible that the NVIDIA controller for which I wrote > this commit has a related bug (though not exactly the same one, because > that controller worked okay with it). At any rate, it seemed that the > NVIDIA problem was actually fixed by commit 6e0c3339a6f1 rather than by > the problem-causing commit 55bcdce8a822. > > It does look like the only way to work around this hardware bug is to > revert that commit. Below is a patch that does that, and it also > lowers the timeout period from 20 ms to about 4 ms. This won't affect > non-broken systems, because they update the schedule status bits in > well under 1 ms. And it appears that broken systems _never_ update the > status bits, so using a long timeout is pointless. > > Let me know how this works. Upon patching, recompiling kernel and a reboot everything seems to be working as expected. I now see two instances of: ehci-pci 0000:00:10.3: Waited too long for the async schedule status (0/8000), giving up in dmesg around the 1 second mark, with no perceivable delays, and everything else looks normal to me so far. Rebooted two more times and those were the same, as fast as it used to be. Thanks! -- 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
