[PATCH bpf] cpumap: Avoid warning when CONFIG_DEBUG_PER_CPU_MAPS is enabled

[PATCH bpf] cpumap: Avoid warning when CONFIG_DEBUG_PER_CPU_MAPS is enabled

[Toke Høiland-Jørgensen]

When the kernel is built with CONFIG_DEBUG_PER_CPU_MAPS, the cpumap code
can trigger a spurious warning if CONFIG_CPUMASK_OFFSTACK is also set. This
happens because in this configuration, NR_CPUS can be larger than
nr_cpumask_bits, so the initial check in cpu_map_alloc() is not sufficient
to guard against hitting the warning in cpumask_check().

Fix this by using the nr_cpumask_bits variable in the map creation code
instead of the NR_CPUS constant.

Fixes: 6710e1126934 ("bpf: introduce new bpf cpu map type BPF_MAP_TYPE_CPUMAP")
Cc: Jesper Dangaard Brouer <brouer@redhat.com>
Reported-by: Xiumei Mu <xmu@redhat.com>
Signed-off-by: Toke Høiland-Jørgensen <toke@redhat.com>
---
 kernel/bpf/cpumap.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/kernel/bpf/cpumap.c b/kernel/bpf/cpumap.c
index 70f71b154fa5..23902afb3bba 100644
--- a/kernel/bpf/cpumap.c
+++ b/kernel/bpf/cpumap.c
@@ -99,8 +99,8 @@ static struct bpf_map *cpu_map_alloc(union bpf_attr *attr)
 
 	bpf_map_init_from_attr(&cmap->map, attr);
 
-	/* Pre-limit array size based on NR_CPUS, not final CPU check */
-	if (cmap->map.max_entries > NR_CPUS) {
+	/* Pre-limit array size based on nr_cpumask_bits, not final CPU check */
+	if (cmap->map.max_entries > nr_cpumask_bits) {
 		err = -E2BIG;
 		goto free_cmap;
 	}
-- 
2.26.0

Re: [PATCH bpf] cpumap: Avoid warning when CONFIG_DEBUG_PER_CPU_MAPS is enabled

[Jesper Dangaard Brouer]

On Wed, 15 Apr 2020 16:01:51 +0200
Toke Høiland-Jørgensen <toke@redhat.com> wrote:

> When the kernel is built with CONFIG_DEBUG_PER_CPU_MAPS, the cpumap code
> can trigger a spurious warning if CONFIG_CPUMASK_OFFSTACK is also set. This
> happens because in this configuration, NR_CPUS can be larger than
> nr_cpumask_bits, so the initial check in cpu_map_alloc() is not sufficient
> to guard against hitting the warning in cpumask_check().
> 
> Fix this by using the nr_cpumask_bits variable in the map creation code
> instead of the NR_CPUS constant.

Shouldn't you use 'nr_cpu_ids' instead of 'nr_cpumask_bits' ?

Else this will still fail on systems with CONFIG_CPUMASK_OFFSTACK=n.

> 
> Fixes: 6710e1126934 ("bpf: introduce new bpf cpu map type BPF_MAP_TYPE_CPUMAP")
> Cc: Jesper Dangaard Brouer <brouer@redhat.com>
> Reported-by: Xiumei Mu <xmu@redhat.com>
> Signed-off-by: Toke Høiland-Jørgensen <toke@redhat.com>
> ---
>  kernel/bpf/cpumap.c | 4 ++--
>  1 file changed, 2 insertions(+), 2 deletions(-)
> 
> diff --git a/kernel/bpf/cpumap.c b/kernel/bpf/cpumap.c
> index 70f71b154fa5..23902afb3bba 100644
> --- a/kernel/bpf/cpumap.c
> +++ b/kernel/bpf/cpumap.c
> @@ -99,8 +99,8 @@ static struct bpf_map *cpu_map_alloc(union bpf_attr *attr)
>  
>  	bpf_map_init_from_attr(&cmap->map, attr);
>  
> -	/* Pre-limit array size based on NR_CPUS, not final CPU check */
> -	if (cmap->map.max_entries > NR_CPUS) {
> +	/* Pre-limit array size based on nr_cpumask_bits, not final CPU check */
> +	if (cmap->map.max_entries > nr_cpumask_bits) {

Shouldn't you use 'nr_cpu_ids' instead of 'nr_cpumask_bits' ?

>  		err = -E2BIG;
>  		goto free_cmap;
>  	}



-- 
Best regards,
  Jesper Dangaard Brouer
  MSc.CS, Principal Kernel Engineer at Red Hat
  LinkedIn: http://www.linkedin.com/in/brouer

Re: [PATCH bpf] cpumap: Avoid warning when CONFIG_DEBUG_PER_CPU_MAPS is enabled

[Toke Høiland-Jørgensen]

Jesper Dangaard Brouer <brouer@redhat.com> writes:

> On Wed, 15 Apr 2020 16:01:51 +0200
> Toke Høiland-Jørgensen <toke@redhat.com> wrote:
>
>> When the kernel is built with CONFIG_DEBUG_PER_CPU_MAPS, the cpumap code
>> can trigger a spurious warning if CONFIG_CPUMASK_OFFSTACK is also set. This
>> happens because in this configuration, NR_CPUS can be larger than
>> nr_cpumask_bits, so the initial check in cpu_map_alloc() is not sufficient
>> to guard against hitting the warning in cpumask_check().
>> 
>> Fix this by using the nr_cpumask_bits variable in the map creation code
>> instead of the NR_CPUS constant.
>
> Shouldn't you use 'nr_cpu_ids' instead of 'nr_cpumask_bits' ?
>
> Else this will still fail on systems with CONFIG_CPUMASK_OFFSTACK=n.

Jesper and I had an offlist discussion about this, and decided that it's
actually better to keep the check in cpu_map_alloc as-is, and instead
just check against nr_cpumask_bits in cpu_map_update_elem(). Otherwise,
the max-size of the map can vary between machines which can affect BPF
program portability.

I'll send a v2.

-Toke