On Tue, 2006-05-02 at 10:16 -0700, Raj, Ashok wrote: > >-----Original Message----- > >From: Bryce Harrington [mailto:bryce@xxxxxxxx] > >Sent: Tuesday, May 02, 2006 10:04 AM > >To: Mary Edie Meredith > >Cc: Raj, Ashok; Hotplug SIG; lhcs-devel@xxxxxxxxxxxxxxxxxxxxx > >Subject: Re: Hotplug CPU issue? > > > >On Tue, May 02, 2006 at 09:59:18AM -0700, Mary Edie Meredith wrote: > >> So that means that an implementation _can have an online file, even > >> though the corresponding cpu is not able to offline. Sounds like > having > >> an online file in sysfs isn't a sure indication of an "off-lineable" > cpu > >> after all. > > > >Well, possibly, unless that situation is considered a bug. > > Not necessarily. Say for e.g there may be platform interrupts that could > be only migrated to certain cpus. Say on a 8way system you have cpus 0-3 > that can take a platform interrupts. And 4-7 cannot handle it, when you > remove each cpu one by one, and you run into the last cpu that can > handle the interrupt and cannot be moved anywhere, the operation will > fail. > > The result of the echo command should tell if it succeded or not. Yes, but from the perspective of a regression test, how do we know that it is OK for the echo command to fail? If the on-line file is there, the script thinks that it should off-line when we echo zero to it. If the echo comes back as failed, without any other information, the scrips would report that as a FAILURE, when it isn't. So how do we know in user-space what the right response should be? > > > > >Mary, have you seen any sort of specification for the cpu hotplug > design > >that indicates when the online file should/shouldn't be present, or > >anything regarding for a given architecture, how to determine what > cpu's > >can be onlined/offlined? > > It started with some power stuff, that depending on what fw is loaded (I > might word it wrong) cpu offlining may not be supported even though > kernel is compiled with support to offline. Then you see the online file > missing. > > We just extended the same for other types as well, for e.g > > If bsp cannot be removed > If CPEI (on ia64) cannot be migrated to another cpu other than what ACPI > told us its associated with etc. > > > > > > >Bryce > -- Mary Edie Meredith Initiative Manager Open Source Development Labs maryedie@xxxxxxxx 503-906-1942
