On Tue, Jun 28, 2016 at 06:06:47PM -0400, Steven Rostedt wrote:
> On Tue, 28 Jun 2016 21:27:29 +0200
> Rasmus Villemoes <li...@rasmusvillemoes.dk> wrote:
>
> > > I probably should make a trace_printk() that doesn't default to the
> > > binary print, to handle things like this.
> > >
> > > trace_printk_ptr()?
> > >
> > > Or even just see if I can find a way that detects this in the fmt
> > > string. Hmm, that probably can't be done at compile time :-/
> >
> > Well, not with gcc itself, but it wouldn't be too hard to make smatch
> > complain loudly if trace_printk is used on a format string with any %p
> > extension (directing people to use trace_printk_ptr()) - the format
> > parsing (and type checking) is already there.
>
> Well, actually gcc can (see below). Although, the more I think about
> this, the more I'm thinking that the bin_printk() should be default
> just copy the pointer content. As the whole point of bin_printk() is to
> print the content at another time. And since pointers should not be
> dereferenced later, it should be saved at the moment the bprintk() is
> called and not dereferenced later.
>
> -- Steve
>
> diff --git a/include/linux/kernel.h b/include/linux/kernel.h
> index 94aa10ffe156..62693900cc4b 100644
> --- a/include/linux/kernel.h
> +++ b/include/linux/kernel.h
> @@ -631,7 +631,24 @@ do {
> \
> \
> __trace_printk_check_format(fmt, ##args); \
> \
> - if (__builtin_constant_p(fmt)) \
> + if (__builtin_constant_p(fmt) && \
> + !__builtin_strstr(fmt, "%pF") && \
> + !__builtin_strstr(fmt, "%pf") && \
> + !__builtin_strstr(fmt, "%pR") && \
> + !__builtin_strstr(fmt, "%pr") && \
> + !__builtin_strstr(fmt, "%pb") && \
> + !__builtin_strstr(fmt, "%pM") && \
> + !__builtin_strstr(fmt, "%pI") && \
> + !__builtin_strstr(fmt, "%pE") && \
> + !__builtin_strstr(fmt, "%pU") && \
> + !__builtin_strstr(fmt, "%pV") && \
> + !__builtin_strstr(fmt, "%pN") && \
> + !__builtin_strstr(fmt, "%pa") && \
> + !__builtin_strstr(fmt, "%pd") && \
> + !__builtin_strstr(fmt, "%pC") && \
> + !__builtin_strstr(fmt, "%pD") && \
> + !__builtin_strstr(fmt, "%pg") && \
> + !__builtin_strstr(fmt, "%pG")) \
> __trace_bprintk(_THIS_IP_, trace_printk_fmt, ##args); \
> else \
> __trace_printk(_THIS_IP_, fmt, ##args); \
is this one working? it seems better to me than adding new trace_printk_ptr
jirka
