Re: [patch,v2] bdi: add a user-tunable cpu_list for the bdi flusher threads

[Dave Chinner <david@fromorbit.com>]

On Tue, Dec 04, 2012 at 09:42:55AM -0500, Jeff Moyer wrote:
> Dave Chinner <david@fromorbit.com> writes:
> 
> > On Mon, Dec 03, 2012 at 01:53:39PM -0500, Jeff Moyer wrote:
> >> +static ssize_t cpu_list_store(struct device *dev,
> >> +		struct device_attribute *attr, const char *buf, size_t count)
> >> +{
> >> +	struct backing_dev_info *bdi = dev_get_drvdata(dev);
> >> +	struct bdi_writeback *wb = &bdi->wb;
> >> +	cpumask_var_t newmask;
> >> +	ssize_t ret;
> >> +	struct task_struct *task;
> >> +
> >> +	if (!alloc_cpumask_var(&newmask, GFP_KERNEL))
> >> +		return -ENOMEM;
> >> +
> >> +	ret = cpulist_parse(buf, newmask);
> >> +	if (!ret) {
> >> +		spin_lock(&bdi->wb_lock);
> >> +		task = wb->task;
> >> +		if (task)
> >> +			get_task_struct(task);
> >> +		spin_unlock(&bdi->wb_lock);
> >> +		if (task) {
> >> +			ret = set_cpus_allowed_ptr(task, newmask);
> >> +			put_task_struct(task);
> >> +		}
> >
> > Why is this set here outside the bdi->flusher_cpumask_mutex?
> 
> The cpumask mutex protects updates to bdi->flusher_cpumask, it has
> nothing to do with the call to set_cpus_allowed.  We are protected from
> concurrent calls to cpu_list_store by the sysfs mutex that is taken on
> entry.  I understand that this is non-obvious, and it wouldn't be wrong
> to hold the mutex here.  If you'd like me to do that for clarity, that
> would be ok with me.

At minimum it needs a comment like this otherwise someone is going
to come along and ask "why is that safe?" like I just did. I'd
prefer the code to be obviously consistent to avoid the need for
commenting about the special case, especially when the obviously
correct code is simpler ;)

Cheers,

Dave.
-- 
Dave Chinner
david@fromorbit.com