Re: [PATCH] perf record: handle death by SIGTERM

[David Ahern <dsahern@gmail.com>]

On 5/6/13 4:40 PM, Stephane Eranian wrote:
> This is a good fix. I have run into this infinite loop in perf report
> many times.

The perf_file_header could use an 'I am sane' bit which is only set when 
the file is closed properly. Perhaps we could overload the magic field 
like this:

diff --git a/tools/perf/util/header.c b/tools/perf/util/header.c
index 326068a..cd9fad6 100644
--- a/tools/perf/util/header.c
+++ b/tools/perf/util/header.c
@@ -2364,7 +2364,6 @@ out_err_write:
     }

     f_header = (struct perf_file_header){
-       .magic     = PERF_MAGIC,
         .size      = sizeof(f_header),
         .attr_size = sizeof(f_attr),
         .attrs = {
@@ -2382,6 +2381,8 @@ out_err_write:
     };

     memcpy(&f_header.adds_features, &header->adds_features, 
sizeof(header->adds_features));
+   if (at_exit)
+       f_header.magic     = PERF_MAGIC,

     lseek(fd, 0, SEEK_SET);
     err = do_write(fd, &f_header, sizeof(f_header));

The perf magic is only written when the file is closed properly. In 
Mike's case you end up with the message (which can be enhanced)

   magic/endian check failed
   incompatible file format (rerun with -v to learn more)

which is better than an infinite loop.

David