On Tue, 31 Aug 2010 10:04:43 -0700, Fenghua Yu wrote:
> On Tue, Aug 31, 2010 at 12:07:25AM -0700, Jean Delvare wrote:
> > On Mon, 30 Aug 2010 18:02:52 -0700, Fenghua Yu wrote:
> > > On Mon, Aug 30, 2010 at 05:55:48PM -0700, Jin Dongming wrote:
> > > > diff --git a/arch/x86/kernel/cpu/mcheck/therm_throt.c b/arch/x86/kernel/cpu/mcheck/therm_throt.c
> > > > index d9368ee..79d563a 100644
> > > > --- a/arch/x86/kernel/cpu/mcheck/therm_throt.c
> > > > +++ b/arch/x86/kernel/cpu/mcheck/therm_throt.c
> > > > @@ -216,14 +216,27 @@ static __cpuinit int thermal_throttle_add_dev(struct sys_device *sys_dev,
> > > > err = sysfs_add_file_to_group(&sys_dev->kobj,
> > > > &attr_core_power_limit_count.attr,
> > > > thermal_attr_group.name);
> > > > - if (cpu_has(c, X86_FEATURE_PTS))
> > > > + if (err)
> > > > + goto error;
> > > > +
> > > > + if (cpu_has(c, X86_FEATURE_PTS)) {
> > > > err = sysfs_add_file_to_group(&sys_dev->kobj,
> > > > &attr_package_throttle_count.attr,
> > > > thermal_attr_group.name);
> > > > + if (err)
> > > > + goto error;
> > > > +
> > > > if (cpu_has(c, X86_FEATURE_PLN))
> > > > err = sysfs_add_file_to_group(&sys_dev->kobj,
> > > > &attr_package_power_limit_count.attr,
> > > > thermal_attr_group.name);
> > > > + if (err)
> > > > + goto error;
> > > > + }
> > > > +
> > > > + return 0;
> > > > +error:
> > > > + sysfs_remove_group(&sys_dev->kobj, &thermal_attr_group);
> > > >
> > > > return err;
> > > > }
> > >
> > > This fix is incorrect. In this patch, a previous error prevents from adding any
> > > further devices. There shouldn't be such dependency among the devices.
> >
> > I don't quite follow you. Did you mean to write that a previous error
> > prevents from creating further _attributes_ for the same device? This
> > would be true.
> >
> > Now I don't think this is a problem because 1* such errors should never
> > happen anyway and 2* if they do happen then further attempts to create
> > the other attributes are unlikely to succeed either.
>
> I don't think there is dependency among the count files, i.e. failure to add a
> count file to the group shouldn't impact other files. Other filles can still be
> added to the group. In this case, user application only sees part of count
> numbers. And kernel may just warn on the failure instead of providing nothing
> to user.
>
> In the patch, any failure to add a file will remove the whole group. This is
> too strict. Kernel doesn't provide any count number to user application.
Oh, my bad. I missed the call to sysfs_remove_group() when reviewing
the code. I agree with you that it shouldn't be added.
> Agree with you that such errors should never happen anyway. So original code
> works fine.
The original code works indeed (except for the missing curly braces)
but is confusing for the reader (which is why I raised the point and we
are discussing it now). If you are voluntarily ignoring errors, you
should add a comment saying so. And it is also a good practice to use a
dummy variable to store the error value you'll ignore, so that the
intent is clear.
> If to be picky to the error handling, a patch may just ignore returning errors
> from sysfs_add_file_to_group.
This is an option, yes. Unfortunately this also means that such errors
won't be even logged, while I think this would be desirable.
> Or use err |= sysfs_add_file_to_group to collect
> all errors and return err but without calling sysfs_remove_group.
Please never use |= on non-bitwise values, it can only lead to bugs and
confusion.
--
Jean Delvare
_______________________________________________
lm-sensors mailing list
lm-sensors@xxxxxxxxxxxxxx
http://lists.lm-sensors.org/mailman/listinfo/lm-sensors
