Re: [PATCH v3 2/2]: perf record: enable asynchronous trace writing

[Alexey Budankov <alexey.budankov@linux.intel.com>]

Hi,

On 28.08.2018 11:53, Jiri Olsa wrote:
> On Mon, Aug 27, 2018 at 09:16:55PM +0300, Alexey Budankov wrote:
> 
> SNIP
> 
>>  	if ((md->start & md->mask) + size != (md->end & md->mask)) {
>>  		buf = &data[md->start & md->mask];
>> -		size = md->mask + 1 - (md->start & md->mask);
>> -		md->start += size;
>> -
>> -		if (push(to, buf, size) < 0) {
>> -			rc = -1;
>> -			goto out;
>> -		}
>> +		size0 = md->mask + 1 - (md->start & md->mask);
>> +		md->start += size0;
>> +		memcpy(md->data, buf, size0);
>>  	}
>>  
>>  	buf = &data[md->start & md->mask];
>>  	size = md->end - md->start;
>>  	md->start += size;
>> +	memcpy(md->data + size0, buf, size);
>>  
>> -	if (push(to, buf, size) < 0) {
>> -		rc = -1;
>> +	rc = push(to, md, size0 + size, *off) < 0 ? -1 : 1;
>> +	if (rc == -1)
>>  		goto out;
>> -	}
>> +
>> +	perf_mmap__get(md);
>> +	*off += size0 + size;
> 
> this get is for the perf_mmap pointer storage in the mmap_aio array right?

Right. perf_mmap__get() here guards whole perf_mmap object and memory 
referenced thru it from premature deallocation because mmap->base kernel 
buffer can be released earlier than aio requests started on mmap->data 
complete and the both buffers are referenced thru the same perf_mmap object.

> 
> I see it's released in record__mmap_read_sync, which might also return
> without releasing it.. this needs to be fixed and explained in here,
> why we take the reference in the first place

So we increment after successful push() from map->base to map->data 
with following aio_write() and decrement when aio_write() is fully 
complete, because it may require restart if the kernel didn't write 
whole chunk at once.

Probably we are still missing one more perf_mmap__put() after:
	pr_err("failed to write perf data, error: %m\n"); 
prior nullifying the appropriate cblock.

Updated [PATCH v4 2/2]:
- Written comment in perf_mmap__push() just before perf_mmap__get();
- Written comment in record__mmap_read_sync() on possible restarting 
  of aio_write() operation and releasing perf_mmap object after all;
- added perf_mmap__put() for the cases of failed aio_write();

Thanks!

> 
> thanks,
> jirka
>