On Thursday 28 October 2004 14:42, Pavel Machek wrote: > Hi! > > > > I have to agree with some of the things Pavel wrote. If the swap device > > > is a USB flash disk, for instance, you can't > > > write to it without enabling > > > interrupts (and the APIC hardware). Not unless you want to go ahead and > > > write a completely new USB controller driver that uses > > > polling rather than interrupts. > > > > That should work fine as is. EHCI can work without interrupts (and will if you prevent IRQs from arriving). OHCI can't, the driver can't process TDs until the donelist IRQ hands them from HC to HCD. But that's no problem; resuming that flash disk would necessarily resume the HCD, which would necessarily re-enable all that stuff. It should act just fine now, given patches I've posted.(*) So as Patrick said, "fine as-is". > > Remember that we have two lists of devices - regular devices and system > > devices. The stopping and starting (or freezing and thawing) that I'm > > talking about is only for regular devices. System devices are not even > > touched at all before writing the image, so that behavior wouldn't change. > > > > Note, though, that is likely a bug, since we don't appear to be saving any > > system device state.. > > Yes, its a bug, and people tell me that clocks are wrong after resume > due to this. I'll need to fix it sooner on later. STR needs to suspend/resume those timers too, right? If we go the "STD = STR plus image-to-disk" route, we should need just a single patch ... ;) - Dave (*) Notably these two, but there's still some khubd issue. - usbcore and sleep state transitions http://marc.theaimsgroup.com/?l=linux-usb-devel&m=109881500807667&w=2 - EHCI: http://marc.theaimsgroup.com/?l=linux-usb-devel&m=109881507722637&w=2
