Re: [PATCH] perf: make perf.data more self-descriptive (v4)

[Peter Zijlstra <peterz@infradead.org>]

On Fri, 2011-09-16 at 09:33 -0600, David Ahern wrote:
> 
> On 09/16/2011 09:19 AM, Peter Zijlstra wrote:
> > On Fri, 2011-09-16 at 08:50 -0600, David Ahern wrote:
> >>
> >> ppc data file: header->magic 454c494646524550
> >> x86 data file: header->magic 454c494646524550
> >>
> > Ok then we need to fix that magic thing..
> 
> Right, but it breaks backwards/forwards compatibility. e.g.,
> 
> diff --git a/tools/perf/util/header.c b/tools/perf/util/header.c
> index 73fa59e..603a8bb9 100644
> --- a/tools/perf/util/header.c
> +++ b/tools/perf/util/header.c
> @@ -65,9 +65,7 @@ char *perf_header__find_event(u64 id)
>     return NULL;
>  }
> 
> -static const char *__perf_magic = "PERFFILE";
> -
> -#define PERF_MAGIC (*(u64 *)__perf_magic)
> +#define PERF_MAGIC 0x5045524646494c45ULL  // "PERFFILE"
> 
>  struct perf_file_attr {
>     struct perf_event_attr  attr;
> 
> That puts the magic in the file based on endianness of the recording
> system. e.g., on x86:
> 
> $ od -c perf-magic.data | head -1
> 0000000   E   L   I   F   F   R   E   P   h  \0  \0  \0  \0  \0  \0  \0
> 
> Which means older perf binaries can't analyze new files (created with
> new magic order) and newer perf binaries can't analyze older perf files
> -- the new command will incorrectly want to do endian swapping on the
> old file.

crap.. sounds like a nice moment to ditch the current format and go for
the multi-file setup.. :-)