Re: [PATCH v2 1/3] mm, printk: introduce new format string for flags

[Vlastimil Babka]

On 12/10/2015 05:04 AM, Steven Rostedt wrote:
> On Thu, Dec 10, 2015 at 11:59:44AM +0900, Joonsoo Kim wrote:
> > Ccing, Steven to ask trace-cmd problem.
> >
> > I'd like to use %pgp in tracepoint output. It works well when I do
> > 'cat /sys/kernel/debug/tracing/trace' but not works well when I do
> > './trace-cmd report'. It prints following error log.
> >
> >    [page_ref:page_ref_unfreeze] bad op token &
> >    [page_ref:page_ref_set] bad op token &
> >    [page_ref:page_ref_mod_unless] bad op token &
> >    [page_ref:page_ref_mod_and_test] bad op token &
> >    [page_ref:page_ref_mod_and_return] bad op token &
> >    [page_ref:page_ref_mod] bad op token &
> >    [page_ref:page_ref_freeze] bad op token &
> >
> > Following is the format I used.
> >
> > TP_printk("pfn=0x%lx flags=%pgp count=%d mapcount=%d mapping=%p mt=%d val=%d ret=%d",
> >                  __entry->pfn, &__entry->flags, __entry->count,
> >                  __entry->mapcount, __entry->mapping, __entry->mt,
> >                  __entry->val, __entry->ret)
> >
> > Could it be solved by 'trace-cmd' itself?

You mean that trace-cmd/parse-events.c would interpret the raw value of 
flags by itself? That would mean the flags became fixed ABI, not a good 
idea...

> > Or it's better to pass flags by value?

If it's value (as opposed to a pointer in %pgp), that doesn't change 
much wrt. having to intepret them?

> > Or should I use something like show_gfp_flags()?

Sounds like least pain to me, at least for now. We just need to have the 
translation tables available as #define with __print_flags() in some 
trace/events header, like the existing trace/events/gfpflags.h for gfp 
flags. These tables can still be reused within mm/debug.c or printk code 
without copy/paste, like I did in "[PATCH v2 6/9] mm, debug: introduce 
dump_gfpflag_names() for symbolic printing of gfp_flags" [1]. Maybe it's 
not the most elegant solution, but works without changing parse-events.c 
using the existing format export.

So if you agree, I can do this in the next spin.

[1] https://lkml.org/lkml/2015/11/24/354

> Yes this can be solved in perf and trace-cmd via the parse-events.c file. And
> as soon as that happens, whatever method we decide upon becomes a userspace
> ABI. So don't think you can change it later.
>
> -- Steve

--
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/