Re: [PATCH v6 06/11] printk: console: Introduce sysfs interface for per-console loglevels

[Petr Mladek]

On Wed 2024-11-20 09:49:08, John Ogness wrote:
> On 2024-11-20, Chris Down <chris@xxxxxxxxxxxxxx> wrote:
> >>> +static ssize_t loglevel_show(struct device *dev, struct device_attribute *attr,
> >>> +			     char *buf)
> >>> +{
> >>> +	struct console *con = dev_get_drvdata(dev);
> >>> +
> >>> +	return sysfs_emit(buf, "%d\n", READ_ONCE(con->level));
> >>
> >>While I admire the use of READ_ONCE() properly, it really doesn't matter
> >>for sysfs as it could change right afterwards and no one cares.  So no
> >>need for that here, right?
> >
> > From my reading of the code it looks like we need this to avoid
> > tearing.
> 
> I cannot imagine that any compiler would perform multiple reads to read
> an aligned field of 4-bytes. Particularly since this function only reads
> this one field.

I believe that the chance is very very small. But are you 100% sure, please?

Honestly, it seems that everyone agrees that the READ_ONCE() makes
some sense. I do not understand why some many people wants to remove
it. I personally prefer to be on the safe side.

> At most it is kind of annotating lockless access to con->level. But
> since it is not using data_race(), it would still trigger KCSAN with a
> warning. I recommend removing it.

I actually suggested to make a wrapper similar to
console_srcu_read_flags() and use the data_race() there, see
https://lore.kernel.org/r/Zy4368zf-sJyyzja@xxxxxxxxxxxxxxx

Best Regards,
Petr