Re: [PATCH] mm/numa_balancing: Teach mpol_to_str about the balancing mode

From: Tvrtko Ursulin
Date: Fri Jun 28 2024 - 04:56:55 EST



On 28/06/2024 04:12, Huang, Ying wrote:
Hi, Matthew,

Matthew Wilcox <willy@xxxxxxxxxxxxx> writes:

On Tue, Jun 25, 2024 at 02:26:05PM +0100, Tvrtko Ursulin wrote:
/*
- * Currently, the only defined flags are mutually exclusive
+ * The below two flags are mutually exclusive:
*/
if (flags & MPOL_F_STATIC_NODES)
p += snprintf(p, buffer + maxlen - p, "static");
else if (flags & MPOL_F_RELATIVE_NODES)
p += snprintf(p, buffer + maxlen - p, "relative");
+
+ if (flags & MPOL_F_NUMA_BALANCING)
+ p += snprintf(p, buffer + maxlen - p, "balancing");
}

So if MPOL_F_STATIC_NODES and MPOL_F_NUMA_BALANCING are set, then we
get a string "staticbalancing"? Is that intended?

Or are these three all mutually exclusive and that should have been
as "else if"?

Yes, this is an issue!

Sigh, my apologies. I was sure I tested it as this patch was part of a larger series I have, but then I decided to extract it and send out and the problems obviously go deeper. What I think happened is that I probably only tested the other direction, setting of via mpol_parse_str().

Andrew please dequeue it if you haven't already?

Dig the git history, in commit 2291990ab36b ("mempolicy: clean-up
mpol-to-str() mempolicy formatting"), the support for multiple flags are
removed. I think that we need to restore it.

Done some basic testing. It was found that when MPOL_F_NUMA_BALANCING
is set, /proc/PID/numa_maps always display "default". That is wrong.

This make me think that this patch has never been tested!

The "default" displaying is introduced in commit 8790c71a18e5
("mm/mempolicy.c: fix mempolicy printing in numa_maps"). We need to fix
it firstly for MPOL_F_NUMA_BALANCING with more accurate filtering. The
fix needs to be backported to -stable kernel.

Will you work on this or I can follow up if you want?

Regards,

Tvrtko