[Devel] [PATCH RH8] ve/proc/net/nr_cpus: cut lines in /proc/net/softnet_stat to number of vcpus in CT

[Pavel Tikhomirov]

Looks good.

On 4/23/21 7:12 PM, Evgenii Shatokhin wrote:
> From: Pavel Tikhomirov <ptikhomirov at virtuozzo.com>
> 
> Counters are not virtualized, we just remove the leak of host's number
> of cpus. Done similar to /proc/cpuinfo, in case of cpu hotplug race with
> reading proc, can actually show arbitrary number of lines as we do no
> locking like get_online_cpus or cpuhp_lock_acquire_read, it seem we do
> so for the sake of simplicity.
> 
> https://jira.sw.ru/browse/PSBM-73238
> Signed-off-by: Pavel Tikhomirov <ptikhomirov at virtuozzo.com>
> 
> eshatokhin@: updated the conditional to match what is now used for
> /proc/cpuinfo (see __c_start() and cpuinfo_op in arch/x86/kernel/cpu/proc.c).
> No changes in semantics.
> 
> Done in the scope of https://jira.sw.ru/browse/PSBM-127794.

Reviewed-by: Pavel Tikhomirov <ptikhomirov at virtuozzo.com>

> Signed-off-by: Evgenii Shatokhin <eshatokhin at virtuozzo.com>
> ---
>   net/core/net-procfs.c | 3 ++-
>   1 file changed, 2 insertions(+), 1 deletion(-)
> 
> diff --git a/net/core/net-procfs.c b/net/core/net-procfs.c
> index 7083d708db3b..ff0da3fe7597 100644
> --- a/net/core/net-procfs.c
> +++ b/net/core/net-procfs.c
> @@ -122,7 +122,8 @@ static struct softnet_data *softnet_get_online(loff_t *pos)
>   
>   	while (*pos < nr_cpu_ids)
>   		if (cpu_online(*pos)) {
> -			sd = &per_cpu(softnet_data, *pos);
> +			if (bitmap_weight(cpumask_bits(cpu_online_mask), *pos) < num_online_vcpus())
> +				sd = &per_cpu(softnet_data, *pos);
>   			break;
>   		} else
>   			++*pos;
> 

-- 
Best regards, Tikhomirov Pavel
Software Developer, Virtuozzo.