Re: [PATCH] bpf: selftests: handle sparse CPU allocations

[Martynas <m@lambda.lt>]

Oops, thanks for noticing the issue with strlen in the for loop. Changing to "int len = strlen(buff); for (i = 0; i < len; i++) <..>" should fix the issue. I'm going to re-submit the patch.

Also, I put some testing results: https://gist.github.com/brb/5369b5cfd08babb80cf2c4081dc19762

On Thu, Jan 31, 2019, at 7:25 AM, Y Song wrote:
> [ My reply somehow rejected by netdev, this is to send it again. ]
> 
> On Wed, Jan 30, 2019 at 1:19 AM Martynas Pumputis <m@lambda.lt> wrote:
> >
> > Previously, bpf_num_possible_cpus() had a bug when calculating a
> > number of possible CPUs in the case of sparse CPU allocations, as
> > it was considering only the first range or element of
> > /sys/devices/system/cpu/possible.
> >
> > E.g. in the case of "0,2-3" (CPU 1 is not available), the function
> > returned 1 instead of 3.
> >
> > This patch fixes the function by making it parse all CPU ranges and
> > elements.
> >
> > Signed-off-by: Martynas Pumputis <m@lambda.lt>
> > ---
> >  tools/testing/selftests/bpf/bpf_util.h | 29 +++++++++++++++++---------
> >  1 file changed, 19 insertions(+), 10 deletions(-)
> >
> > diff --git a/tools/testing/selftests/bpf/bpf_util.h b/tools/testing/selftests/bpf/bpf_util.h
> > index 315a44fa32af..8cab50408204 100644
> > --- a/tools/testing/selftests/bpf/bpf_util.h
> > +++ b/tools/testing/selftests/bpf/bpf_util.h
> > @@ -13,7 +13,7 @@ static inline unsigned int bpf_num_possible_cpus(void)
> >         unsigned int start, end, possible_cpus = 0;
> >         char buff[128];
> >         FILE *fp;
> > -       int n;
> > +       int n, i, j = 0;
> >
> >         fp = fopen(fcpu, "r");
> >         if (!fp) {
> > @@ -21,17 +21,26 @@ static inline unsigned int bpf_num_possible_cpus(void)
> >                 exit(1);
> >         }
> >
> > -       while (fgets(buff, sizeof(buff), fp)) {
> > -               n = sscanf(buff, "%u-%u", &start, &end);
> > -               if (n == 0) {
> > -                       printf("Failed to retrieve # possible CPUs!\n");
> > -                       exit(1);
> > -               } else if (n == 1) {
> > -                       end = start;
> > +       if (!fgets(buff, sizeof(buff), fp)) {
> > +               printf("Failed to read %s!\n", fcpu);
> > +               exit(1);
> > +       }
> > +
> > +       for (i = 0; i <= strlen(buff); i++) {
> > +               if (buff[i] == ',' || buff[i] == '\0') {
> > +                       buff[i] = '\0';
> 
> This does not sound right. For example, the cpu list "0,2-3",
> you will change "," to '\0" so buffer becomes "0\02-3".
> The next iteration you will get strlen(buff) = 1.
> The "2-3" will be skipped.
> 
> > +                       n = sscanf(&buff[j], "%u-%u", &start, &end);
> > +                       if (n <= 0) {
> > +                               printf("Failed to retrieve # possible CPUs!\n");
> > +                               exit(1);
> > +                       } else if (n == 1) {
> > +                               end = start;
> > +                       }
> > +                       possible_cpus += end - start + 1;
> > +                       j = i + 1;
> >                 }
> > -               possible_cpus = start == 0 ? end + 1 : 0;
> > -               break;
> >         }
> > +
> >         fclose(fp);
> >
> >         return possible_cpus;
> > --
> > 2.20.1
> >