Messages in this thread | | | Date | Thu, 8 Mar 2012 14:23:35 -0800 | From | Andrew Morton <> | Subject | Re: [patch] mm, hugetlb: add thread name and pid to SHM_HUGETLB mlock rlimit warning |
| |
On Thu, 8 Mar 2012 14:08:30 -0800 (PST) David Rientjes <rientjes@google.com> wrote:
> On Thu, 8 Mar 2012, Andrew Morton wrote: > > > > We have a get_task_comm() that does the task_lock() > > > internally but requires a TASK_COMM_LEN buffer in the calling code. It's > > > just easier for the calling code to the task_lock() itself for a tiny > > > little printk(). > > > > Well for a tiny little printk we could just omit the locking? The > > printk() won't oops and once in a million years one person will see a > > garbled comm[] string? > > > > Sure, but task_lock() shouldn't be highly contended when the thread isn't > forking or exiting (everything else is attaching/detaching from a cgroup > or testing a mempolicy). I've always added it (like in the oom killer for > the same reason) just because the race exists. Taking it for every thread > on the system for one call to the oom killer has never slowed it down.
I wasn't concerned about the performance side of things - just that it's such a pain over such a silly thing.
btw, if the code had done
printk_once(..., get_task_comm(...), ...)
the task_lock() would have been performed just a single time, rather than every time.
| |