Tian, Kevin
2006-Apr-21 07:28 UTC
RE: [Xen-devel] [PATCH]Check the values of MAX_VIRT_CPUS and NR_CPUSfor SMP
>From: Tristan Gingold [mailto:Tristan.Gingold@bull.net] >Sent: 2006年4月21日 15:26 >> > >> >See how we solve this on x86 near the top of do_boot_cpu() in >> >arch/x86/smpboot.c. Essentially we allocate extra idle domains on >> >demand to give us enough idle vcpus for one per physical cpu. >> > >> > -- Keir >> >> Yes, that''s the same story on ia64 where we also allocate idle >> vcpu at do_boot_cpu(). That''s why I''m not sure the real cause >> for Atsushi''s issue. >No, we don''t create extra domains. Therefore it there are more CPUs >than >VCPUs, something goes wrong... >Yes, there''s only one idle domain now. However I''m not sure why something will go wrong when NR_CPUS > MAX_VIRT_CPUS. As long as you want to wake up an AP, you have to go through do_boot_cpu where idle_vcpu will be created. The only drawback is that some physical cpus will not be waken up by "max_cpus = MAX_VIRT_CPUS" you mentioned in another thread. That needs to be fixed which however shouldn''t crash anything since uninitialized cpus won''t be used. Thanks, Kevin _______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
Tristan Gingold
2006-Apr-21 07:39 UTC
Re: [Xen-devel] [PATCH]Check the values of MAX_VIRT_CPUS and NR_CPUSfor SMP
Le Vendredi 21 Avril 2006 09:28, Tian, Kevin a écrit :> From: Tristan Gingold [mailto:Tristan.Gingold@bull.net] > > >Sent: 2006年4月21日 15:26 > > > >> >See how we solve this on x86 near the top of do_boot_cpu() in > >> >arch/x86/smpboot.c. Essentially we allocate extra idle domains on > >> >demand to give us enough idle vcpus for one per physical cpu. > >> > > >> > -- Keir > >> > >> Yes, that''s the same story on ia64 where we also allocate idle > >> vcpu at do_boot_cpu(). That''s why I''m not sure the real cause > >> for Atsushi''s issue. > > > >No, we don''t create extra domains. Therefore it there are more CPUs > >than > >VCPUs, something goes wrong... > > Yes, there''s only one idle domain now. However I''m not sure > why something will go wrong when NR_CPUS > MAX_VIRT_CPUS.bvt has per VCPU data. if #VCPU > #MAX_VIRT_CPUS, bvt reads/writes beyond the bounds! Tristan. _______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
Seemingly Similar Threads
- RE: [PATCH]Check the values of MAX_VIRT_CPUS and NR_CPUSfor SMP
- RE: [PATCH]Check the values of MAX_VIRT_CPUS and NR_CPUSfor SMP
- scheduler independent forced vcpu selection
- [PATCH] update evtchn for SMP
- do_* declarations (was: Re: [Xen-ia64-devel] [PATCH] added multicall)