On 2 March 2017 at 20:55, BALATON Zoltan <bala...@eik.bme.hu> wrote: > On Thu, 2 Mar 2017, Peter Maydell wrote: >> >> On 2 March 2017 at 20:18, BALATON Zoltan <bala...@eik.bme.hu> wrote: >>> >>> On Thu, 2 Mar 2017, Peter Maydell wrote: >>>> >>>> >>>> Don't use qemu_register_reset(). Set the appropriate dc->reset >>>> function pointers instead. >>> >>> >>> >>> Any reason for that? This way I could save two more boilerplate functions >>> because I could define reset function once, otherwise I'd need two >>> versions >>> taking sysbus and pci states just to extract the SM501State function and >>> call this function. Do you still think I should do that instead? >> >> >> qemu_register_reset is a pre-QOM method for doing reset; >> the standard QOM way of saying "my device has some reset >> behaviour" is to set its reset method pointer. >> Code calling qemu_register_reset() is generally either (a) doing >> something kind of weird or (b) old device code that hasn't yet >> been converted to QOM. > > > Hmm, does having a device with both sysbus and pci versions qualify as > weird? Does adding a comment saying that this way we don't need two more > reset functions just to call the registered one justifies using > qemu_register_reset() or this function is deprecated and should not be used > and should go with the separate reset functions instead?
Just use the QOM reset method pointers. Every QOM device should have a reset method, and that should be how it gets reset. (PCI is special anyway, it needs to get reset when the guest does a PCI bus reset, and that results in your device's reset method getting called. Functions registered with qemu_register_reset() only get called for full system reset.) thanks -- PMM
