On Sat, 18 Feb 2012, Seth Forshee wrote: > On Fri, Feb 17, 2012 at 04:37:28PM -0500, Alan Stern wrote: > > This patch (as1521) slightly improves the way usb-storage's > > asynchronous scanning thread interacts with the freezer. Rather than > > making itself freezable before waiting for the device to settle, the > > routine won't make itself freezable until after the settling delay. > > > > This is just as correct as the existing code. If the freezer starts > > up before the delay is over then the scanning thread will be ignored > > until the delay ends, at which point it will immediately be frozen. > > On the other hand, if the freezer starts up after the delay is over > > then it will wait for the scanning thread to finish, as it does now. > > It looks to me like this restores the potential for the deadlock I was > seeing previously. If freezing were to start after khubd starts > processing a disconnection but before the scanning thread is awoken from > its delay loop, can't we still end up usb-stor-scan being frozen while > khubd is waiting on it? This just seems to shift the point where the > freezing occurs as compared with pre-3.2. Ah, this is the point I wasn't clear on. We can't cope with an unbind while the scanning thread is frozen, right? Not unless we can somehow kill the thread while it is frozen. And an unbind can occur at any time, even after the thread has been frozen (unless the thread isn't freezable at all). > What if you move set_freezable to inside the body of the if statement on > US_FLIDX_DONT_SCAN? Then we should only freeze if there isn't someone > waiting in quiesce_and_remove_host. There'd still theoretically be a > race, but the window would be so small that it seems unlikely that it > would ever be a problem, and usb-stor-scan should still freeze if it's > going to proceed with the scan. No, it's better to find a solution that always works. I'll convert the scanning thread to a delayed-work routine. Freezable work queues have the nice property that you can cancel a work item while the queue is frozen. Revised patch to come next week... Alan Stern -- 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
