Re: [PATCH 5/9] perf: Add sysfs format attribute for pmu device

[Jiri Olsa]

On Tue, Jan 24, 2012 at 11:40:28AM -0800, Eric W. Biederman wrote:
> Peter Zijlstra <a.p.zijlstra@xxxxxxxxx> writes:
> 
> > On Mon, 2012-01-23 at 16:33 +0100, Jiri Olsa wrote:
> >>   
> >> > > +static ssize_t perf_default_event_format(struct pmu *pmu, char *page)
> >> > > +{
> >> > > +	ssize_t ret;
> >> > > +
> >> > > +	ret  = sprintf(page,       "config  config:0-63\n");
> >> > > +	ret += sprintf(page + ret, "config1 config1:0-63\n");
> >> > > +	ret += sprintf(page + ret, "config2 config2:0-63\n");
> >> > 
> >> > This looks like abuse of sysfs.  The rule is one value per file.
> >> > When you start adding newlines I can't see how this format attribute can
> >> > be considered one value.
> >> 
> >> ops, missed that rule.. I think we can carry the syntax files
> >> within perf, or have it builtin somehow.. or procfs? :) PeterZ?
> >
> > There's plenty of exceptions to this rule though and touching procfs is
> > even less appreciated.
> 
> There may be exceptions but certainly not plenty of exceptions.
> 
> There is also a practical scaling problem.  You have to be careful
> with how big you make that value before you have problems.
> 
> Perhaps what you want is a directory with a bunch of sysfs attributes?
> 
yep, Peter outlined this possibility in here
https://lkml.org/lkml/2012/1/24/206

> All I know is I see buffer overflows waiting to happen whenever I look
> at the code snippet above.

from this point of view, the above data should always be static/hardcoded
for given pmu.. so we will be fine with the page size buffer,
it could be valifated with some sort of BUILD_BUG_ON(data_size < pagesize)

jirka
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/