[PATCH] perf: fix bug mismatch with -c option definition

[PATCH] perf: fix bug mismatch with -c option definition

[Stephane Eranian]

  The -c option defines the user requested sampling period. It was implemented
  using an unsigned int variable but the type of the option was OPT_LONG. Thus,
  the option parser was overwriting memory belonging to other variables, namely
  the mmap_pages leading to a zero page sampling buffer. The bug was exposed
  only when compiling at -O0, probably because the compiler was padding
  variables at higher optimization levels.

  This patch fixes this problem by declaring user_interval as u64. This also
  avoids wrap-around issues for large period on 32-bit systems.
 
  Signed-off-by: Stephane Eranian <eranian@google.com>
--
 tools/perf/builtin-record.c |    8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)

diff --git a/tools/perf/builtin-record.c b/tools/perf/builtin-record.c
index 0f467cf..78f64cc 100644
--- a/tools/perf/builtin-record.c
+++ b/tools/perf/builtin-record.c
@@ -33,8 +33,8 @@ enum write_mode_t {
 
 static int			*fd[MAX_NR_CPUS][MAX_COUNTERS];
 
-static unsigned int		user_interval 			= UINT_MAX;
-static long			default_interval		=      0;
+static u64			user_interval			= ULLONG_MAX;
+static u64			default_interval		=      0;
 
 static int			nr_cpus				=      0;
 static unsigned int		page_size;
@@ -268,7 +268,7 @@ static void create_counter(int counter, int cpu)
 	 * it a weak assumption overridable by the user.
 	 */
 	if (!attr->sample_period || (user_freq != UINT_MAX &&
-				     user_interval != UINT_MAX)) {
+				     user_interval != ULLONG_MAX)) {
 		if (freq) {
 			attr->sample_type	|= PERF_SAMPLE_PERIOD;
 			attr->freq		= 1;
@@ -901,7 +901,7 @@ int cmd_record(int argc, const char **argv, const char *prefix __used)
 	if (!event_array)
 		return -ENOMEM;
 
-	if (user_interval != UINT_MAX)
+	if (user_interval != ULLONG_MAX)
 		default_interval = user_interval;
 	if (user_freq != UINT_MAX)
 		freq = user_freq;

Re: [PATCH] perf: fix bug mismatch with -c option definition

[Frederic Weisbecker]

On Mon, May 17, 2010 at 12:04:01PM +0200, Stephane Eranian wrote:
>   The -c option defines the user requested sampling period. It was implemented
>   using an unsigned int variable but the type of the option was OPT_LONG. Thus,
>   the option parser was overwriting memory belonging to other variables, namely
>   the mmap_pages leading to a zero page sampling buffer. The bug was exposed
>   only when compiling at -O0, probably because the compiler was padding
>   variables at higher optimization levels.
> 
>   This patch fixes this problem by declaring user_interval as u64. This also
>   avoids wrap-around issues for large period on 32-bit systems.
>  
>   Signed-off-by: Stephane Eranian <eranian@google.com>



Acked-by: Frederic Weisbecker <fweisbec@gmail.com>

Small detail: could you avoid the spaces in the beginning of
your changelog lines?

May be that's because you use git-show to dump your patches?
In which case I suggest you to use git-format-patch instead.

Thanks.

Re: [PATCH] perf: fix bug mismatch with -c option definition

[Arnaldo Carvalho de Melo]

Em Mon, May 17, 2010 at 12:04:01PM +0200, Stephane Eranian escreveu:
>   The -c option defines the user requested sampling period. It was implemented
>   using an unsigned int variable but the type of the option was OPT_LONG. Thus,
>   the option parser was overwriting memory belonging to other variables, namely
>   the mmap_pages leading to a zero page sampling buffer. The bug was exposed
>   only when compiling at -O0, probably because the compiler was padding
>   variables at higher optimization levels.

Well spotted!
 
>   This patch fixes this problem by declaring user_interval as u64. This also
>   avoids wrap-around issues for large period on 32-bit systems.
>  
>   Signed-off-by: Stephane Eranian <eranian@google.com>
> --
>  tools/perf/builtin-record.c |    8 ++++----
>  1 file changed, 4 insertions(+), 4 deletions(-)
> 
> diff --git a/tools/perf/builtin-record.c b/tools/perf/builtin-record.c
> index 0f467cf..78f64cc 100644
> --- a/tools/perf/builtin-record.c
> +++ b/tools/perf/builtin-record.c
> @@ -33,8 +33,8 @@ enum write_mode_t {
>  
>  static int			*fd[MAX_NR_CPUS][MAX_COUNTERS];
>  
> -static unsigned int		user_interval 			= UINT_MAX;
> -static long			default_interval		=      0;
> +static u64			user_interval			= ULLONG_MAX;
> +static u64			default_interval		=      0;

The parsing code uses this for OPT_LONG:

        case OPTION_LONG:
                if (unset) {
                        *(long *)opt->value = 0;
                        return 0;
                }
                if (opt->flags & PARSE_OPT_OPTARG && !p->opt) {
                        *(long *)opt->value = opt->defval;
                        return 0;
                }
                if (get_arg(p, opt, flags, &arg))
                        return -1;
                *(long *)opt->value = strtol(arg, (char **)&s, 10);

So I think we should augment the parsing code to have OPTION_ULONG, and,
for handling u64, OPTION_ULLONG.

I'll add that and then modify your patch to use it.

- Arnaldo

Re: [PATCH] perf: fix bug mismatch with -c option definition

[Stephane Eranian]

On Mon, May 17, 2010 at 4:19 PM, Arnaldo Carvalho de Melo
<acme@infradead.org> wrote:
> Em Mon, May 17, 2010 at 12:04:01PM +0200, Stephane Eranian escreveu:
>>   The -c option defines the user requested sampling period. It was implemented
>>   using an unsigned int variable but the type of the option was OPT_LONG. Thus,
>>   the option parser was overwriting memory belonging to other variables, namely
>>   the mmap_pages leading to a zero page sampling buffer. The bug was exposed
>>   only when compiling at -O0, probably because the compiler was padding
>>   variables at higher optimization levels.
>
> Well spotted!

As you can imagine, it was not so trivial to find ;->

>
>>   This patch fixes this problem by declaring user_interval as u64. This also
>>   avoids wrap-around issues for large period on 32-bit systems.
>>
>>   Signed-off-by: Stephane Eranian <eranian@google.com>
>> --
>>  tools/perf/builtin-record.c |    8 ++++----
>>  1 file changed, 4 insertions(+), 4 deletions(-)
>>
>> diff --git a/tools/perf/builtin-record.c b/tools/perf/builtin-record.c
>> index 0f467cf..78f64cc 100644
>> --- a/tools/perf/builtin-record.c
>> +++ b/tools/perf/builtin-record.c
>> @@ -33,8 +33,8 @@ enum write_mode_t {
>>
>>  static int                   *fd[MAX_NR_CPUS][MAX_COUNTERS];
>>
>> -static unsigned int          user_interval                   = UINT_MAX;
>> -static long                  default_interval                =      0;
>> +static u64                   user_interval                   = ULLONG_MAX;
>> +static u64                   default_interval                =      0;
>
> The parsing code uses this for OPT_LONG:
>
>        case OPTION_LONG:
>                if (unset) {
>                        *(long *)opt->value = 0;
>                        return 0;
>                }
>                if (opt->flags & PARSE_OPT_OPTARG && !p->opt) {
>                        *(long *)opt->value = opt->defval;
>                        return 0;
>                }
>                if (get_arg(p, opt, flags, &arg))
>                        return -1;
>                *(long *)opt->value = strtol(arg, (char **)&s, 10);
>
> So I think we should augment the parsing code to have OPTION_ULONG, and,
> for handling u64, OPTION_ULLONG.
>
> I'll add that and then modify your patch to use it.
>
That's fine too.
Thanks.