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

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

Hi,

On 28.08.2018 11:50, Jiri Olsa wrote:
> On Mon, Aug 27, 2018 at 09:16:55PM +0300, Alexey Budankov wrote:
>>
>> Trace file offset are linearly calculated by perf_mmap__push() code 
>> for the next possible write operation, but file position is updated by 
>> the kernel only in the second lseek() syscall after the loop. 
>> The first lseek() syscall reads that file position for 
>> the next loop iterations.
>>
>> record__mmap_read_sync implements sort of a barrier between spilling 
>> ready profiling data to disk.
>>
>> Signed-off-by: Alexey Budankov <alexey.budankov@linux.intel.com>
>> ---
>> Changes in v3:
>> - written comments about nanosleep(0.5ms) call prior aio_suspend()
>>   to cope with intrusiveness of its implementation in glibc;
>> - written comments about rationale behind coping profiling data 
>>   into mmap->data buffer;
>> ---
>>  tools/perf/builtin-record.c | 125 +++++++++++++++++++++++++++++++++++++++++---
>>  tools/perf/util/mmap.c      |  36 ++++++++-----
>>  tools/perf/util/mmap.h      |   2 +-
>>  3 files changed, 143 insertions(+), 20 deletions(-)
>>
>> diff --git a/tools/perf/builtin-record.c b/tools/perf/builtin-record.c
>> index 22ebeb92ac51..4ac61399a09a 100644
>> --- a/tools/perf/builtin-record.c
>> +++ b/tools/perf/builtin-record.c
>> @@ -53,6 +53,7 @@
>>  #include <sys/mman.h>
>>  #include <sys/wait.h>
>>  #include <linux/time64.h>
>> +#include <aio.h>
>>  
>>  struct switch_output {
>>  	bool		 enabled;
>> @@ -121,6 +122,23 @@ static int record__write(struct record *rec, void *bf, size_t size)
>>  	return 0;
>>  }
>>  
>> +static int record__aio_write(int trace_fd, struct aiocb *cblock,
>> +		void *buf, size_t size, off_t off)
>> +{
>> +	cblock->aio_fildes = trace_fd;
>> +	cblock->aio_buf    = buf;
>> +	cblock->aio_nbytes = size;
>> +	cblock->aio_offset = off;
>> +	cblock->aio_sigevent.sigev_notify = SIGEV_NONE;
>> +
>> +	if (aio_write(cblock) == -1) {
>> +		pr_err("failed to queue perf data, error: %m\n");
>> +		return -1;
>> +	}
>> +
>> +	return 0;
>> +}
>> +
>>  static int process_synthesized_event(struct perf_tool *tool,
>>  				     union perf_event *event,
>>  				     struct perf_sample *sample __maybe_unused,
>> @@ -130,12 +148,14 @@ static int process_synthesized_event(struct perf_tool *tool,
>>  	return record__write(rec, event, event->header.size);
>>  }
>>  
>> -static int record__pushfn(void *to, void *bf, size_t size)
>> +static int record__pushfn(void *to, void *bf, size_t size, off_t off)
>>  {
>>  	struct record *rec = to;
>> +	struct perf_mmap *map = bf;
> 
> the argument needs to change for record__pushfn,
> now with your changes, it's no longer 'void *bf',
> but 'struct perf_mmap *map'

Ok. Included into [PATCH v4 2/2].

> 
> also I'm little confused why we have '*to' and cast
> it back to 'struct record', but so be it ;-)

Supported. :)

> 
> thanks,
> jirka
>