On Sun, 2012-12-02 at 15:36 +0100, Robert Jarzmik wrote:
> kbuild test robot <fengguang.wu@xxxxxxxxx> writes:
>
> > tree: git://git.kernel.org/pub/scm/linux/kernel/git/mhocko/mm.git since-3.6
> > head: 422a0f651b5cefa1b6b3ede2e1c9e540a24a6e01
> > commit: 07b81da5f80b27543ddbe3164170c64e0941a812 [493/499] mm: trace filemap add and del
> >
> >
> > sparse warnings:
> >
> > + include/trace/events/filemap.h:14:1: sparse: incompatible types for operation (<)
> > include/trace/events/filemap.h:14:1: left side has type struct page *<noident>
> > include/trace/events/filemap.h:14:1: right side has type int
> > include/trace/events/filemap.h:45:1: sparse: incompatible types for operation (<)
> > include/trace/events/filemap.h:45:1: left side has type struct page *<noident>
> > include/trace/events/filemap.h:45:1: right side has type int
>
> Hi Steven, Frederic and Ingo,
Sorry for the late reply, It's end of year and I'm getting a lot of
pressure at work to get things done.
>
> I just drop this note to make you aware (as FTRACE maintainers) of the sparse
> warning I received.
>
> This sparse warning will touch all submission to the FTRACE events API AFAIK.
> The node of the problem :
> - in include/linux/ftrace_event.h, we have :
> #define is_signed_type(type) (((type)(-1)) < 0)
> - this is used by kernel/trace/trace_events.c
> #define __common_field(type, item)
> ...
> is_signed_type(type)
> ...
>
> Here, if a trace field is a pointer (for example struct page *), we end up with
> this in my case :
>
> static int __attribute__((no_instrument_function))
> ftrace_define_fields_mm_filemap_delete_from_page_cache(struct ftrace_event_call
> *event_call) { struct ftrace_raw_mm_filemap_delete_from_page_cache field; int
> ret; ret = trace_define_field(event_call, "struct page *", "page",
> __builtin_offsetof(typeof(field),page), sizeof(field.page), (((struct page
> *)(-1)) < 0), FILTER_OTHER); if (ret) return ret; ret =
> trace_define_field(event_call, "unsigned long", "i_ino",
> __builtin_offsetof(typeof(field),i_ino), sizeof(field.i_ino), (((unsigned
> long)(-1)) < 0), FILTER_OTHER); if (ret) return ret; ret =
> trace_define_field(event_call, "unsigned long", "index",
> __builtin_offsetof(typeof(field),index), sizeof(field.index), (((unsigned
> long)(-1)) < 0), FILTER_OTHER); if (ret) return ret; ret =
> trace_define_field(event_call, "dev_t", "s_dev",
> __builtin_offsetof(typeof(field),s_dev), sizeof(field.s_dev), (((dev_t)(-1)) <
> 0), FILTER_OTHER); if (ret) return ret;; return ret; }; ;;
>
> And I think, (((struct page *)(-1)) < 0) gives the warning. I don't know if
> "is_signed_type()" makes sense on a pointer, but I think you'll get other
> reports of that kind for any new event added to trace API.
Hmm, this is mostly automated via the macros. Not sure how we can
differentiate a pointer from other fields. Would this be fixed if we
did:
#define is_signed_type(type) (((type)(-1) < (type)0)
?
-- Steve
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@xxxxxxxxx. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>