[dpdk-dev] [PATCH v2 1/2] test/mempool_perf: Free mempool on exit

Olivier Matz olivier.matz at 6wind.com
Fri Apr 7 17:51:02 CEST 2017


Hi Santosh,

On Thu,  6 Apr 2017 12:15:48 +0530, Santosh Shukla <santosh.shukla at caviumnetworks.com> wrote:
> Mempool_perf test not freeing pool memory.
> 
> Cc: stable at dpdk.org
> Signed-off-by: Santosh Shukla <santosh.shukla at caviumnetworks.com>
> Acked-by: Shreyansh Jain <shreyansh.jain at nxp.com>
> ---
> v1 --> v2:
>  * Fixed patch context
> 
>  test/test/test_mempool_perf.c | 31 +++++++++++++++++++------------
>  1 file changed, 19 insertions(+), 12 deletions(-)
> 
> diff --git a/test/test/test_mempool_perf.c b/test/test/test_mempool_perf.c
> index ebf1721ac..3c45971ab 100644
> --- a/test/test/test_mempool_perf.c
> +++ b/test/test/test_mempool_perf.c
> @@ -312,6 +312,8 @@ do_one_mempool_test(unsigned cores)
>  static int
>  test_mempool_perf(void)
>  {
> +	int ret = -1;
> +
>  	rte_atomic32_init(&synchro);
>  
>  	/* create a mempool (without cache) */
> @@ -322,7 +324,7 @@ test_mempool_perf(void)
>  						my_obj_init, NULL,
>  						SOCKET_ID_ANY, 0);
>  	if (mp_nocache == NULL)
> -		return -1;
> +		goto err;
>  
>  	/* create a mempool (with cache) */
>  	if (mp_cache == NULL)

[...]

>  
> -	return 0;
> +	ret = 0;
> +
> +err:
> +	rte_mempool_free(mp_cache);
> +	rte_mempool_free(mp_nocache);
> +	return ret;


Since mp_cache and mp_nocache are global variables, this won't
work properly due to the way mempool are created:

	/* create a mempool (without cache) */
	if (mp_nocache == NULL)
	        mp_nocache = rte_mempool_create("perf_test_nocache", MEMPOOL_SIZE,
	                                        MEMPOOL_ELT_SIZE, 0, 0,
	                                        NULL, NULL,
	                                        my_obj_init, NULL,
	                                        SOCKET_ID_ANY, 0);

The if() should be removed, else we'll have a use after free the next
time.

If you want to do more clean-up, you can try to remove the global variables,
but it's maybe harder.

Thanks,
Olivier


More information about the dev mailing list