[PATCH v7] numa: make node_to_cpumask_map() NUMA_NO_NODE aware

[Michal Hocko]

On Wed 30-10-19 11:28:00, Peter Zijlstra wrote:
> On Wed, Oct 30, 2019 at 11:22:29AM +0100, Michal Hocko wrote:
> > On Wed 30-10-19 11:14:49, Peter Zijlstra wrote:
> > > On Wed, Oct 30, 2019 at 05:34:28PM +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://lkml.org/lkml/2019/9/11/66
> > > > Signed-off-by: Yunsheng Lin <linyunsheng at huawei.com>
> > > > Suggested-by: Michal Hocko <mhocko at kernel.org>
> > > > Acked-by: Michal Hocko <mhocko at suse.com>
> > > > Acked-by: Paul Burton <paul.burton at mips.com> # MIPS bits
> > > 
> > > Still:
> > > 
> > > Nacked-by: Peter Zijlstra (Intel) <peterz at infradead.org>
> > 
> > Do you have any other proposal that doesn't make any wild guesses about
> > which node to use instead of the undefined one?
> 
> It only makes 'wild' guesses when the BIOS is shit and it complains
> about that.

I really do not see how this is any better than simply using the online
cpu mask in the same "broken" situation. We are effectivelly talking
about a suboptimal path for suboptimal setups. I haven't heard any
actual technical argument why cpu_online_mask is any worse than adding
some sort of failover guessing which node to use as a replacement.

I completely do you point about complaining loud about broken BIOS/fw.
It seems we just disagree where we should workaround those issues
because as of now we simply do generate semi random behavior because of
an uninitialized memory access.

> Or do you like you BIOS broken?

I do not see anything like that in my response nor in my previous
communication. Moreover a patch to warn about this should be on the way
to get merged AFAIK.

-- 
Michal Hocko
SUSE Labs