Re: [PATCH] sched/cpuacct: show only present CPUs to userspace

[Peter Zijlstra <peterz@infradead.org>]

On Thu, Oct 17, 2024 at 12:21:38PM +0200, Alexander Mikhalitsyn wrote:
> After commit b0c69e1214bc ("drivers: base: Use present CPUs in GENERIC_CPU_DEVICES")
> changed which CPUs are shown in /sys/devices/system/cpu/ (only "present" ones)
> it also makes sense to change cpuacct cgroupv1 code not to report CPUs
> which are not present in the system as it confuses userspace.
> Let's make it consistent.
> 
> A configuration when #(present CPUs) < #(possible CPUs) is easy to get with:
> qemu-system-x86_64
> 	-smp 3,maxcpus=12 \
> 	...
> 
> Cc: James Morse <james.morse@arm.com>
> Cc: Jonathan Cameron <Jonathan.Cameron@huawei.com>
> Cc: Gavin Shan <gshan@redhat.com>
> Cc: "Russell King (Oracle)" <rmk+kernel@armlinux.org.uk>
> Cc: Thomas Gleixner <tglx@linutronix.de>
> Reported-by: Simon Deziel <simon.deziel@canonical.com>
> Closes: https://github.com/canonical/lxd/issues/13324
> Co-developed-by: Simon Deziel <simon.deziel@canonical.com>
> Signed-off-by: Simon Deziel <simon.deziel@canonical.com>
> Signed-off-by: Alexander Mikhalitsyn <aleksandr.mikhalitsyn@canonical.com>
> ---
>  kernel/sched/cpuacct.c | 4 ++--
>  1 file changed, 2 insertions(+), 2 deletions(-)
> 
> diff --git a/kernel/sched/cpuacct.c b/kernel/sched/cpuacct.c
> index 0de9dda09949..0f07fbfdb20e 100644
> --- a/kernel/sched/cpuacct.c
> +++ b/kernel/sched/cpuacct.c
> @@ -213,7 +213,7 @@ static int __cpuacct_percpu_seq_show(struct seq_file *m,
>  	u64 percpu;
>  	int i;
>  
> -	for_each_possible_cpu(i) {
> +	for_each_present_cpu(i) {
>  		percpu = cpuacct_cpuusage_read(ca, i, index);
>  		seq_printf(m, "%llu ", (unsigned long long) percpu);
>  	}
> @@ -247,7 +247,7 @@ static int cpuacct_all_seq_show(struct seq_file *m, void *V)
>  		seq_printf(m, " %s", cpuacct_stat_desc[index]);
>  	seq_puts(m, "\n");
>  
> -	for_each_possible_cpu(cpu) {
> +	for_each_present_cpu(cpu) {
>  		seq_printf(m, "%d", cpu);
>  		for (index = 0; index < CPUACCT_STAT_NSTATS; index++)
>  			seq_printf(m, " %llu",


Doesn't this create problems for machines that support actual physical
hotplug?

Then all of a sudden, when a CPU with non-zero stats goes from present
to !present, the stats also go away, and any userspace looking at the
sum of stats over CPUs sees an unexplained dis-continuity.