[dpdk-dev] [PATCH] Function __mempool_get_bulk() returns wrong count.

Ananyev, Konstantin konstantin.ananyev at intel.com
Mon Sep 29 01:00:22 CEST 2014



> -----Original Message-----
> From: Wiles, Roger Keith [mailto:keith.wiles at windriver.com]
> Sent: Sunday, September 28, 2014 11:57 PM
> To: Ananyev, Konstantin
> Cc: <dev at dpdk.org>
> Subject: Re: [dpdk-dev] [PATCH] Function __mempool_get_bulk() returns wrong count.
> 
> 
> On Sep 28, 2014, at 5:25 PM, Ananyev, Konstantin <konstantin.ananyev at intel.com> wrote:
> 
> >
> >
> >> -----Original Message-----
> >> From: dev [mailto:dev-bounces at dpdk.org] On Behalf Of Wiles, Roger Keith
> >> Sent: Saturday, September 27, 2014 7:42 PM
> >> To: <dev at dpdk.org>
> >> Subject: [dpdk-dev] [PATCH] Function __mempool_get_bulk() returns wrong count.
> >>
> >>
> >> When __mempool_get_bulk() grabs entries from the cache it
> >> returns zero instead of the number of entries obtained. Plus
> >> the stats were increased by the wrong count of objects.
> >>
> >> Signed-off-by: Keith Wiles <keith.wiles at windriver.com>
> >> ---
> >> lib/librte_mempool/rte_mempool.h | 6 +++---
> >> 1 file changed, 3 insertions(+), 3 deletions(-)
> >>
> >> diff --git a/lib/librte_mempool/rte_mempool.h b/lib/librte_mempool/rte_mempool.h
> >> index 299d4d7..6750e78 100644
> >> --- a/lib/librte_mempool/rte_mempool.h
> >> +++ b/lib/librte_mempool/rte_mempool.h
> >> @@ -988,9 +988,9 @@ __mempool_get_bulk(struct rte_mempool *mp, void **obj_table,
> >>
> >>        cache->len -= n;
> >>
> >> -       __MEMPOOL_STAT_ADD(mp, get_success, n_orig);
> >> +       __MEMPOOL_STAT_ADD(mp, get_success, n);
> >
> > As I can see n == n_orig.
> > We can completely remove n_orig, but from other side - I don't see any harm here.
> 
> In the RFC patch I sent I remove n_orig.
> >
> >>
> >> -       return 0;
> >> +       return n;
> >
> > As I can see, __mempool_get_bulk supposed to return 0,
> > if all n objects were allocated from mbuf, or a negative error code otherwise.
> > Check all usages of __mempool_get_bulk(), plus the fact that it does below:
> > ret = rte_ring_mc_dequeue_bulk(mp->ring, obj_table, n);
> > and rte_ring_mc_dequeue_bulk() is just wrapper for __rte_ring_mc_do_dequeue(..., n, RTE_RING_QUEUE_FIXED);
> > I.e. - either allocate all n objects, or return with failure.
> > So, yes we should return 0 here.
> > The only thing that probably needs to be done here: fix the comments.
> > Instead of:
> > - >=0: Success; number of objects supplied.
> > Something like:
> > - 0: Success; n objects supplied.
> >
> >>
> >> ring_dequeue:
> >> #endif /* RTE_MEMPOOL_CACHE_MAX_SIZE > 0 */
> >> @@ -1004,7 +1004,7 @@ ring_dequeue:
> >>        if (ret < 0)
> >>                __MEMPOOL_STAT_ADD(mp, get_fail, n_orig);
> >>        else
> >> -               __MEMPOOL_STAT_ADD(mp, get_success, n_orig);
> >> +               __MEMPOOL_STAT_ADD(mp, get_success, ret);
> >
> > That seems incorrect tom me.
> > ret would be either 0 on success, or negative error value.
> 
> Notice 'if (ret < 0)' above so ret can not be negative in this case only zero or positive.

It can't be positive here. Only zero.
See above why.

> >
> > Konstantin
> >
> >
> >>
> >>        return ret;
> >> }
> >> --
> >> 2.1.0Keith Wiles, Principal Technologist with CTO office, Wind River mobile 972-213-5533
> >
> >
> > As I can see
> 
> Keith Wiles, Principal Technologist with CTO office, Wind River mobile 972-213-5533
> 



More information about the dev mailing list