On Tue, Sep 24, 2019 at 01:54:01PM +0200, Michal Hocko wrote:

On Tue 24-09-19 13:23:49, Peter Zijlstra wrote:

quoted

On Tue, Sep 24, 2019 at 12:56:22PM +0200, Michal Hocko wrote:

[...]

quoted

quoted

To be honest I really fail to see why to object to a simple semantic
that NUMA_NO_NODE imply all usable cpus. Could you explain that please?

Because it feels wrong. The device needs to be _somewhere_. It simply
cannot be node-less.

What if it doesn't have any numa preference for what ever reason? There
is no other way to express that than NUMA_NO_NODE.

Like I said; how does that physically work? The device needs to be
somewhere. It _must_ have a preference.

Anyway, I am not going to argue more about this because it seems more of
a discussion about "HW shouldn't be doing that although the specification
allows that" which cannot really have any outcome except of "feels
correct/wrong".

We can push back and say we don't respect the specification because it
is batshit insane ;-)

If you really feel strongly about this then we should think of a proper
way to prevent this to happen because an out-of-bound access is
certainly not something we really want, right?

I just genuinely don't understand it. And I refuse to duct tape it.

And as shown in that email here:

  https://lkml.kernel.org/r/5a188e2b-6c07-a9db-fbaa-561e9362d3ba@huawei.com

there is a ton of broken...

15.061682] node node0: has invalid NUMA node(-1), default node of 0 now selected. Readjust it by writing to sysfs numa_node or contact your vendor for updates.
...
15.285602] node node3: has invalid NUMA node(-1), default node of 0 now selected. Readjust it by writing to sysfs numa_node or contact your vendor for updates.

15.360241] cpu cpu0: has invalid NUMA node(-1), default node of 0 now selected. Readjust it by writing to sysfs numa_node or contact your vendor for updates.
...
24.768305] cpu cpu127: has invalid NUMA node(-1), default node of 0 now selected. Readjust it by writing to sysfs numa_node or contact your vendor for updates.

39.623339] clockevents clockevent0: has invalid NUMA node(-1), default node of 0 now selected. Readjust it by writing to sysfs numa_node or contact your vendor for updates.
...
48.769530] clockevents clockevent127: has invalid NUMA node(-1), default node of 0 now selected. Readjust it by writing to sysfs numa_node or contact your vendor for updates.

That's all broken for no reason.. those things actually _have_ a trivial
node affinity.

By silently accepting we let this stuff fester.

Now granted; there's a number of virtual devices that really don't have
a node affinity, but then, those are not hurt by forcing them onto a
random node, they really don't do anything. Like:

48.913502] event_source armv8_pmuv3_0: has invalid NUMA node(-1), default node of 0 now selected. Readjust it by writing to sysfs numa_node or contact your vendor for updates.
48.985462] event_source breakpoint: has invalid NUMA node(-1), default node of 0 now selected. Readjust it by writing to sysfs numa_node or contact your vendor for updates.
49.057120] event_source uprobe: has invalid NUMA node(-1), default node of 0 now selected. Readjust it by writing to sysfs numa_node or contact your vendor for updates.
49.128431] event_source kprobe: has invalid NUMA node(-1), default node of 0 now selected. Readjust it by writing to sysfs numa_node or contact your vendor for updates.
49.199742] event_source tracepoint: has invalid NUMA node(-1), default node of 0 now selected. Readjust it by writing to sysfs numa_node or contact your vendor for updates.
49.271399] event_source software: has invalid NUMA node(-1), default node of 0 now selected. Readjust it by writing to sysfs numa_node or contact your vendor for updates.

That's just fake devices to get a sysfs entry.