Am 17.09.2014 um 18:47 schrieb Peter Maydell: > On 17 September 2014 09:40, Andreas Färber <afaer...@suse.de> wrote: >> We avoided that by not using DeviceClass::reset but CPUClass::reset. >> It's a question of assuring appropriate reset ordering between CPU and >> devices. PowerPC needed a special reset order via hook in (what is now) >> MachineClass. > >> So while I agree that CPU reset registration is not ideal and needs >> changing, I am not convinced that we can generally make the change and >> hope for the best. I wouldn't mind an incremental transition though, >> with arm taking the first step - still leaves the question of exact >> direction. If you look at x86, you will find that despite my protest >> against this inconsistency, the reset hook registration was moved into >> CPU code but none of the other targets changed alongside. > > I don't object to taking a pragmatic approach in the ARM code > (eg this patch). I just wanted to know if you had a preferred > direction we should be taking instead (which as you say we > kind of have to do in an incremental way). It sounds like you > don't have anything concrete in mind so maybe we should just > apply this patch.
Ack. One other concern I have with this patch is the loop assuming that all following CPUs will be of type ARMCPU, but I suspect there will be other code making the same assumption - in that case Reviewed-by. > In general I suspect there are a lot of unresolved issues in > our handling of reset -- it's a complicated area which we > attempt to address in an over-simplistic way at the moment :-( Yes, having test cases for all machines would help refactor these things... Andreas -- SUSE LINUX Products GmbH, Maxfeldstr. 5, 90409 Nürnberg, Germany GF: Jeff Hawn, Jennifer Guild, Felix Imendörffer; HRB 16746 AG Nürnberg
