On Sun, 13 Sep 2015, Roland Weber wrote: > I compiled a kernel 3.17(.0) with my additional debug lines... > and this one does NOT freeze! The output is: > > (first unbind) > ehci-pci 0000:00:1d.0: remove, state 4 > ehci-pci 0000:00:1d.0: roothub graceful disconnect > usb usb1: USB disconnect, device number 1 > ehci-pci 0000:00:1d.0: shutdown urb ffff880072bf4300 ep1in-intr > usb_remove_hcd: calling stop > ehci-pci 0000:00:1d.0: stop > ehci_silence_controller: entry > ehci_halt: entry > ehci_halt: after spin_lock_irq > ehci_halt: about to readl prematurely > ehci_halt: premature readl returned 10001 Note: 10001 instead of 10000, which is what you saw in the other kernel. That could be highly relevant. > I cannot yet tell whether the difference between freeze and non-freeze > is due to code changes, or because I used a different kernel config. > (It brings down compilation times from 3h to 2h.) > > So next I'll compile 3.17.8 with my config. On Sun, 13 Sep 2015, Roland Weber wrote: > Actually, it made more sense to try other precompiled kernels. > I can freeze the precompiled 3.17.0 with unbind/bind/unbind, > but not the one I compiled myself. I'm currently compiling > 3.17.0 with the exact config of the precompiled kernel, to > rule out differences in the build environment as the cause. > > Assuming that this kernel does freeze, I will start to 'bisect' the > differences in the kernel configurations, based on the assumption > that it might give us a clue about where the problem comes from. > I expect this to take several days... please stop me if you've got > a better lead. At some point along the way, can you try adding some printk statements to ehci_suspend() and ehci_resume() in ehci-hcd.c? I'd like to know if they get called during the bind/unbind procedure and are responsible for that 10001 vs. 10000 value. Alan Stern -- To unsubscribe from this list: send the line "unsubscribe linux-pci" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
