[LTP] [PATCH v6] Refactor aiocp using new LTP API

[Andrea Cervesato]

Hi Cyril,

On 8/10/22 12:37, Cyril Hrubis wrote:
> Hi!
> The runtime changes do not unfortunatelly work as expected.
>
> First of all we have to check for the runtime in the function that
> prepares the file with a random data, since that is quite slow as well.
> Since that is done from the test setup it actually runs in a separate
> runtime window so I would do just:
>
> diff --git a/testcases/kernel/io/ltp-aiodio/aiocp.c b/testcases/kernel/io/ltp-aiodio/aiocp.c
> index 816407284..2b34d8bbe 100644
> --- a/testcases/kernel/io/ltp-aiodio/aiocp.c
> +++ b/testcases/kernel/io/ltp-aiodio/aiocp.c
> @@ -66,6 +66,9 @@ static void fill_with_rand_data(int fd, long long size)
>          srand(time(NULL));
>
>          while (i < size) {
> +               if (!tst_remaining_runtime())
> +                       tst_brk(TCONF, "Out of runtime!");
> +
>                  for (j = 0; j < bufsize; j++) {
>                          buf[j] = (rand() % (upper - lower + 1)) + lower;
>                          i++;
>
> Also the rand() function is way to slow, it's almost 10x faster for
> 500MB buffer if you move the buffer intialization out of the loop, i.e.
> initialize the buffer with random data once and then write it over and
> over.
Yes, we can reduce the amount of time to generate file (especially if 
it's really big), but I wouldn't
introduce any repetition because we would like to check that file has 
been copied properly during
callbacks which are asynchronous by definition.
>
> Then again we would have to check in the async_run() otherwise we will
> keep pushing new read request over and over until the queue has
> finished.
>
> Lastly we silently skip the check and report PASS when we are out of
> runtime, that's not nice. I would be inclined to report TCONF if we end
> up out of time before we managed to check that the data was copied
> correctly. That wouldn't provide people with a false security.
>