Re: [PATCH v6] numa: make node_to_cpumask_map() NUMA_NO_NODE aware

[Michal Hocko <mhocko@kernel.org>]

On Mon 23-09-19 17:15:19, Peter Zijlstra wrote:
> On Tue, Sep 17, 2019 at 08:48:54PM +0800, Yunsheng Lin wrote:
> > When passing the return value of dev_to_node() to cpumask_of_node()
> > without checking if the device's node id is NUMA_NO_NODE, there is
> > global-out-of-bounds detected by KASAN.
> > 
> > From the discussion [1], NUMA_NO_NODE really means no node affinity,
> > which also means all cpus should be usable. So the cpumask_of_node()
> > should always return all cpus online when user passes the node id as
> > NUMA_NO_NODE, just like similar semantic that page allocator handles
> > NUMA_NO_NODE.
> > 
> > But we cannot really copy the page allocator logic. Simply because the
> > page allocator doesn't enforce the near node affinity. It just picks it
> > up as a preferred node but then it is free to fallback to any other numa
> > node. This is not the case here and node_to_cpumask_map will only restrict
> > to the particular node's cpus which would have really non deterministic
> > behavior depending on where the code is executed. So in fact we really
> > want to return cpu_online_mask for NUMA_NO_NODE.
> > 
> > Also there is a debugging version of node_to_cpumask_map() for x86 and
> > arm64, which is only used when CONFIG_DEBUG_PER_CPU_MAPS is defined, this
> > patch changes it to handle NUMA_NO_NODE as normal node_to_cpumask_map().
> > 
> > [1] https://lore.kernel.org/patchwork/patch/1125789/
> 
> That is bloody unusable, don't do that. Use:
> 
>   https://lkml.kernel.org/r/$MSGID
> 
> if anything. Then I can find it in my local mbox without having to
> resort to touching a mouse and shitty browser software.
> 
> (also patchwork is absolute crap for reading email threads)
> 
> Anyway, I found it -- I think, I refused to click the link. I replied
> there.
> 
> > Signed-off-by: Yunsheng Lin <linyunsheng@huawei.com>
> > Suggested-by: Michal Hocko <mhocko@kernel.org>
> > Acked-by: Michal Hocko <mhocko@suse.com>
> 
> 
> 
> > diff --git a/arch/x86/mm/numa.c b/arch/x86/mm/numa.c
> > index 4123100e..9859acb 100644
> > --- a/arch/x86/mm/numa.c
> > +++ b/arch/x86/mm/numa.c
> > @@ -861,6 +861,9 @@ void numa_remove_cpu(int cpu)
> >   */
> >  const struct cpumask *cpumask_of_node(int node)
> >  {
> > +	if (node == NUMA_NO_NODE)
> > +		return cpu_online_mask;
> 
> This mandates the caller holds cpus_read_lock() or something, I'm pretty
> sure that if I put:
> 
> 	lockdep_assert_cpus_held();

Is this documented somewhere? Also how does that differ from a normal
case when a proper node is used? The cpumask will always be dynamic in
the cpu hotplug presence, right?

> here, it comes apart real quick. Without holding the cpu hotplug lock,
> the online mask is gibberish.

Can the returned cpu mask go away?
-- 
Michal Hocko
SUSE Labs