Re: [PATCH] drivers/base: Fix unsigned comparison to -1 in CPUMAP_FILE_MAX_BYTES

[Phil Auld <pauld@redhat.com>]

On Tue, Sep 06, 2022 at 11:09:53AM -0700 Yury Norov wrote:
> On Tue, Sep 06, 2022 at 12:04:30PM -0400, Phil Auld wrote:
> > As PAGE_SIZE is unsigned long, -1 > PAGE_SIZE when NR_CPUS <= 3.
> > This leads to very large file sizes:
> > 
> > topology$ ls -l
> > total 0
> > -r--r--r-- 1 root root 18446744073709551615 Sep  5 11:59 core_cpus
> > -r--r--r-- 1 root root                 4096 Sep  5 11:59 core_cpus_list
> > -r--r--r-- 1 root root                 4096 Sep  5 10:58 core_id
> > -r--r--r-- 1 root root 18446744073709551615 Sep  5 10:10 core_siblings
> > -r--r--r-- 1 root root                 4096 Sep  5 11:59 core_siblings_list
> > -r--r--r-- 1 root root 18446744073709551615 Sep  5 11:59 die_cpus
> > -r--r--r-- 1 root root                 4096 Sep  5 11:59 die_cpus_list
> > -r--r--r-- 1 root root                 4096 Sep  5 11:59 die_id
> > -r--r--r-- 1 root root 18446744073709551615 Sep  5 11:59 package_cpus
> > -r--r--r-- 1 root root                 4096 Sep  5 11:59 package_cpus_list
> > -r--r--r-- 1 root root                 4096 Sep  5 10:58 physical_package_id
> > -r--r--r-- 1 root root 18446744073709551615 Sep  5 10:10 thread_siblings
> > -r--r--r-- 1 root root                 4096 Sep  5 11:59 thread_siblings_list
> > 
> > Adjust the inequality to catch the case when NR_CPUS is configured
> > to a small value.
> > 
> > Fixes: 7ee951acd31a ("drivers/base: fix userspace break from using bin_attributes for cpumap and cpulist")
> > Reported-by: feng xiangjun <fengxj325@gmail.com>
> > Signed-off-by: Phil Auld <pauld@redhat.com>
> > Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
> > Cc: "Rafael J. Wysocki" <rafael@kernel.org>
> > Cc: Yury Norov <yury.norov@gmail.com>
> > Cc: stable@vger.kernel.org
> > Cc: feng xiangjun <fengxj325@gmail.com>
> > ---
> >  include/linux/cpumask.h | 5 +++--
> >  1 file changed, 3 insertions(+), 2 deletions(-)
> > 
> > diff --git a/include/linux/cpumask.h b/include/linux/cpumask.h
> > index bd047864c7ac..7b1349612d6d 100644
> > --- a/include/linux/cpumask.h
> > +++ b/include/linux/cpumask.h
> > @@ -1127,9 +1127,10 @@ cpumap_print_list_to_buf(char *buf, const struct cpumask *mask,
> >   * cover a worst-case of every other cpu being on one of two nodes for a
> >   * very large NR_CPUS.
> >   *
> > - *  Use PAGE_SIZE as a minimum for smaller configurations.
> > + *  Use PAGE_SIZE as a minimum for smaller configurations while avoiding
> > + *  unsigned comparison to -1.
> >   */
> > -#define CPUMAP_FILE_MAX_BYTES  ((((NR_CPUS * 9)/32 - 1) > PAGE_SIZE) \
> > +#define CPUMAP_FILE_MAX_BYTES  ((((NR_CPUS * 9)/32) > PAGE_SIZE + 1) \
> 
> Maybe it would be easier to read with less braces and '>=' instead of '>'?
>   #define CPUMAP_FILE_MAX_BYTES \
>         (NR_CPUS * 9 / 32 >= PAGE_SIZE ? NR_CPUS * 9 / 32 - 1 : PAGE_SIZE)
> 
> Anyways, this is a good catch. If you think it doesn't worth an
> update, I can take it in bitmap-for-next as-is.
>

It would work as
   #define CPUMAP_FILE_MAX_BYTES  (((NR_CPUS * 9)/32 > PAGE_SIZE) \
   	   			     ? (NR_CPUS * 9)/32 - 1 : PAGE_SIZE)

Since 4097 > PAGE_SIZE and then you'd get PAGE_SIZE anyway. That is we could just
leave the 1 out of the inequality completely. 

If I recall at least some of the parens are needed to make the compiler happy.

I can resend with the above if you want.  

Cheers,
Phil

> >  					? (NR_CPUS * 9)/32 - 1 : PAGE_SIZE)
> >  #define CPULIST_FILE_MAX_BYTES  (((NR_CPUS * 7)/2 > PAGE_SIZE) ? (NR_CPUS * 7)/2 : PAGE_SIZE)
> >  
> > -- 
> > 2.31.1
> 

--