Re: [PATCH] PM / wakeup: show wakeup sources stats in sysfs

[Greg KH]

On Tue, Jun 25, 2019 at 06:33:08PM -0700, Tri Vo wrote:
> On Tue, Jun 25, 2019 at 6:12 PM Greg KH <gregkh@xxxxxxxxxxxxxxxxxxx> wrote:
> > > +static ssize_t wakeup_source_count_show(struct wakeup_source *ws,
> > > +                                     struct wakeup_source_attribute *attr,
> > > +                                     char *buf)
> > > +{
> > > +     unsigned long flags;
> > > +     unsigned long var;
> > > +
> > > +     spin_lock_irqsave(&ws->lock, flags);
> > > +     if (strcmp(attr->attr.name, "active_count") == 0)
> > > +             var = ws->active_count;
> > > +     else if (strcmp(attr->attr.name, "event_count") == 0)
> > > +             var = ws->event_count;
> > > +     else if (strcmp(attr->attr.name, "wakeup_count") == 0)
> > > +             var = ws->wakeup_count;
> > > +     else
> > > +             var = ws->expire_count;
> > > +     spin_unlock_irqrestore(&ws->lock, flags);
> > > +
> > > +     return sprintf(buf, "%lu\n", var);
> > > +}
> >
> > Why is this lock always needed to be grabbed?  You are just reading a
> > value, who cares if it changes inbetween reading it and returning the
> > buffer string as it can change at that point in time anyway?
> 
> Right, we don't care if the value changes in between us reading and
> printing it. However, IIUC not grabbing this lock results in a data
> race, which is undefined behavior.

A data race where?  Writing to the value?  How can that happen?  All you
are doing is incrementing this variable elsewhere, what is the worst
that can happen?

thanks,

greg k-h