Re: [Xen-devel] [PATCH v2 0/6] x86: fixes/improvements for scratch cpumask

[Jürgen Groß <jgross@xxxxxxxx>]

On 17.02.20 19:43, Roger Pau Monne wrote:
> Hello,
>
> Commit:
>
> 5500d265a2a8fa63d60c08beb549de8ec82ff7a5
> x86/smp: use APIC ALLBUT destination shorthand when possible
>
> Introduced a bogus usage of the scratch cpumask: it was used in a
> function that could be called from interrupt context, and hence using
> the scratch cpumask there is not safe. Patch #5 is a fix for that usage,
> together with also preventing the usage of any per-CPU variables when
> send_IPI_mask is called from #MC or #NMI context. Previous patches are
> preparatory changes.
>
> Patch #6 adds some debug infrastructure to make sure the scratch cpumask
> is used in the right context, and hence should prevent further missuses.
I wonder whether it wouldn't be better to have a common percpu scratch
cpumask handling instead of introducing local ones all over the
hypervisor.

So basically an array of percpu cpumasks allocated when bringing up a
cpu (this spares memory as the masks wouldn't need to cover NR_CPUS
cpus), a percpu counter of the next free index and get_ and put_
functions acting in a lifo manner.

This would help removing all the still existing cpumasks on the stack
and any illegal nesting would be avoided. The only remaining question
would be the size of the array.

Thoughts?


Juergen

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxxx
https://lists.xenproject.org/mailman/listinfo/xen-devel