On Wed, Jul 22, 2009 at 5:56 PM, Alan Stern<stern@xxxxxxxxxxxxxxxxxxx> wrote: > On Wed, 22 Jul 2009, Manuel Lauss wrote: > >> > This clearly is not working right, but I can't tell exactly what's >> > wrong. Can you repeat the test, this time acquiring a usbmon trace >> > starting before the test and ending after the rmmod? You'll have to >> > use usbmon's 0u file, so try to minimize any other USB activity going >> > on at the same time. >> >> I can't reproduce it if I do a trace after ehci-hcd has been inserted; i.e. the >> busy dma-pool only happens if I don't capture any usbmon output after >> insertion and before removal of ehci-hcd. > > So whatever it is, it must be timing-related. > >> I uploaded a trace here [1], it starts with loading of ehci-hcd, then ~5 seconds >> pause, read some data from ACM device, again some pause then unload >> of ehci-hcd. > > Still not clear, but I'm getting some ideas. Let's try another > approach. For this to work you'll have to enable CONFIG_USB_DEBUG as > well as CONFIG_DEBUG_FS. > > Run your test again, without usbmon. But before removing ehci-hcd, > mount a debugfs filesystem and look in the usb/ehci/au1xxx-ehci.0 > directory. There should be a bunch of files with names like > "registers", "async", and "periodic"; let's see what they contain. Debug output is here: http://mlau.at/au1xxx-ehci-trace-b.txt However I think this is entirely my fault: In an effort to reduce idle wakeups I disabled the EHCI IO watchdog (which reduced idle wakeups by ~10/s). Enabling it again gets rid of the WARN_ON output without any patching. On the other hand, 2.6.30 even with the IO watchdog disabled doesn't show the error. Manuel Lauss -- 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