Re: [PATCH 1/2] perf: Fix failure to set cpumask when only one cpu

[He Zhe]

On 8/2/19 9:06 PM, Jiri Olsa wrote:
> On Fri, Aug 02, 2019 at 04:29:51PM +0800, zhe.he@xxxxxxxxxxxxx wrote:
>> From: He Zhe <zhe.he@xxxxxxxxxxxxx>
>>
>> The buffer containing string used to set cpumask is overwritten by end of
>> string later in cpu_map__snprint_mask due to not enough memory space, when
>> there is only one cpu. And thus causes the following failure.
>>
>> $ perf ftrace ls
>> failed to reset ftrace
>>
>> This patch fixes the calculation of cpumask string size.
>>
>> Signed-off-by: He Zhe <zhe.he@xxxxxxxxxxxxx>
>> ---
>>  tools/perf/builtin-ftrace.c | 2 +-
>>  1 file changed, 1 insertion(+), 1 deletion(-)
>>
>> diff --git a/tools/perf/builtin-ftrace.c b/tools/perf/builtin-ftrace.c
>> index 66d5a66..0193128 100644
>> --- a/tools/perf/builtin-ftrace.c
>> +++ b/tools/perf/builtin-ftrace.c
>> @@ -173,7 +173,7 @@ static int set_tracing_cpumask(struct cpu_map *cpumap)
>>  	int last_cpu;
>>  
>>  	last_cpu = cpu_map__cpu(cpumap, cpumap->nr - 1);
>> -	mask_size = (last_cpu + 3) / 4 + 1;
>> +	mask_size = last_cpu / 4 + 2; /* one more byte for EOS */
>>  	mask_size += last_cpu / 32; /* ',' is needed for every 32th cpus */
> ugh..  why do we care about last_cpu value in here at all?
>
> feels like using static buffer would be more reasonable

Thanks, and yes, a static buffer would be easy to handle. A 2KB buffer is enough
to cover 8196 cpus, the maximum numbers of cpus we can run with for now.

Let's see if there is any other concerns.

Zhe

>
> jirka
>
>>  
>>  	cpumask = malloc(mask_size);
>> -- 
>> 2.7.4
>>