Re: [RFC] [PATCH] cpuset operations causes Badness at mm/slab.c:777 warning

[Paul Jackson <pj@sgi.com>]

> There are no checks necessary. Your function worked fine so far for 
> the case of zero objects with the pointer returned by kmalloc. If the 
> code is correct then it will not dereference the pointer to the zero 
> sized array. If not then we may find a bug and fix it.

I suspect you got lucky.  The check for a full pidarray[] in the routine
pid_array_load() occurs -after- a pid is put in the array.  If a task
showed up in this cpuset at the wrong time, we would fall over and die
in the code:

static int pid_array_load(pid_t *pidarray, int npids, struct cpuset *cs)
{
        int n = 0;
        struct task_struct *g, *p;

        read_lock(&tasklist_lock);

        do_each_thread(g, p) {
                if (p->cpuset == cs) {
                        pidarray[n++] = p->pid;		/* Death if pidarray == NULL */
                        if (unlikely(n == npids))
                                goto array_full;
                }

        } while_each_thread(g, p);

Perhaps if you moved the "if (unlikely(n == npids))" test before the
"pidarray[n++] = p->pid" assignment, it would be safe.

And does the next line of code, the call to sort() after the call of
pid_array_load(), work with pidarray == NULL and npids == 0:

        npids = pid_array_load(pidarray, npids, cs);
        sort(pidarray, npids, sizeof(pid_t), cmppid, NULL);	/* <== ?? */

Off hand, I didn't know.  I guess it must, or you would have already
tripped over it.

-- 
                  I won't rest till it's the best ...
                  Programmer, Linux Scalability
                  Paul Jackson <pj@sgi.com> 1.925.600.0401