At Wed, 26 Nov 2014 15:27:57 +0100, Oliver Neukum wrote: > > On Wed, 2014-11-26 at 15:12 +0100, Takashi Iwai wrote: > > At Wed, 26 Nov 2014 23:05:20 +0900, > > Marcel Holtmann wrote: > > > > > > Hi Oliver, > > > > > > >> In order to paper over this, we may also remember the failing firmware > > > >> and avoid loading it. This might be an easer way than the endless > > > >> fight against UMH race... > > > > > > > > > > > > the full fix would be to implement reset_resume() for btusb. > > > > It seems to me that setup() should be split in two methods, > > > > one to request the firmware from user space and the second > > > > to transfer it to the device. reset_resume() would just need > > > > to repeat the second operation. > > > > > > so when you do hci_register_dev, then hdev->setup is only called once. I > > > really mean only once per lifetime of the hci_dev. So you would need to > > > unregister the hci_dev first before hdev->setup will ever be called > > > again. So I am not sure this is actually the problem here. The problem > > > here is entirely within request_firmware() unless of course we run > > > through the USB probe handlers again. Which I do not see happening here. > > > > > > And we have hdev->setup this way since normally the Bluetooth devices > > > keep their firmware patches and not forget about them and suspend-resume > > > cycles. If the USB device of course jumps of the bus during it then all > > > bets are off anyway. > > > > Usually you can avoid unnecessary rebinding when you provide a proper > > reset_resume callback. I guess that's what Oliver suggested. > > Yes, but even in reset_resume() you would need to redo the setup > part, as the device lost power. The basic problem of requesting > the firmware wouldn't be solved.
Requesting the firmware in the resume path itself is OK. There are many drivers doing so. But the primary problem in btusb is that it's triggered at the wrong timing. (And the second problem is that the firmware loader doesn't cache the non-existing files, so it goes through lengthy code paths for reconfirming that the file doesn't exist.) Takashi -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/