On Thu, 13 Jun 2019, Mayuresh Kulkarni wrote: > Hi Alan, > > Thanks for your review and apologies for late response from me. I was on PTO last week and then in a training this week. > > > Aside from the issues Greg raised, it isn't right because it relies on > > the suspend and resume callbacks for individual interfaces, not for the > > whole device. There are a few other things that should be changed as > > well. > > In our use-case, we open the USB device with our VID/PID and then using that fd > we bind to our interface. So this approach probably worked for our use-case. Yes. But it seems more reliable to use suspend/resume information for the whole device, instead of assuming that the userspace program will have claimed an interface. > > Below is my attempt at doing the same thing (not tested, and it doesn't > > answer all of Greg's objections). It is very similar to your patch. > > Does it work for your application? > > > > I am checking your code-changes and will get back to you on this by next week. > > > (Note: I imagine you might run into trouble because devices generally > > do not get put into runtime suspend immediately. So if you call the > > USBDEVFS_SUSPEND ioctl and then the USBDEVFS_WAIT_FOR_RESUME ioctl, the > > wait will return immediately because the device hasn't yet been > > suspended.) > > > > For this point, I am suggesting below - > How about we return "udev->dev.power.usage_count" from suspend ioctl? > count = 0 -> suspend success so good to call wait-for-resume ioctl > count != 0 -> don't call resume since suspend did not happen. > > Will that work? No, it will not. The usage_count value can change at any time, so it will be out-of-date by the time the ioctl returns. Furthermore, even if usage_count is > 0 when the suspend ioctl returns, it may become 0 later on, and the device will be suspended some time after that. In fact, if you use the default settings for USB autosuspend, the device won't be suspended until 2 seconds after the usage_count becomes 0. So even if the suspend ioctl decrements usage_count to 0, the device still won't be suspended right away. If you call the wait-for-resume ioctl immediately, the call will fail. Alan Stern
