Re: [dpdk-dev] [PATCH v3] mempool: remove non-EAL thread note from header

[Olivier Matz <olivier.matz@6wind.com>]

Hi Gage,

On Wed, 29 Mar 2017 15:30:49 +0000, "Eads, Gage" <gage.eads@intel.com> wrote:
> >  -----Original Message-----
> >  From: Olivier Matz [mailto:olivier.matz@6wind.com]
> >  Sent: Friday, March 24, 2017 9:05 AM
> >  To: Eads, Gage <gage.eads@intel.com>
> >  Cc: dev@dpdk.org
> >  Subject: Re: [PATCH v3] mempool: remove non-EAL thread note from header
> >  
> >  Hi Gage,
> >  
> >  On Thu, 23 Mar 2017 09:20:58 -0500, Gage Eads <gage.eads@intel.com>
> >  wrote:  
> >  > Commit 30e6399892276 ("mempool: support non-EAL thread") added the
> >  > capability for non-EAL threads to use the mempool library. This commit
> >  > removes the note indicating that the mempool library cannot be used
> >  > safely by non-EAL threads.
> >  >
> >  > Also, fix a typo.
> >  >
> >  > Signed-off-by: Gage Eads <gage.eads@intel.com>
> >  > ---
> >  > v2: Changed commit message to referenced commit 30e63998 instead of
> >  > 4b5062755
> >  > v3: Fix checkpatch error
> >  >
> >  >  lib/librte_mempool/rte_mempool.h | 6 +-----
> >  >  1 file changed, 1 insertion(+), 5 deletions(-)
> >  >
> >  > diff --git a/lib/librte_mempool/rte_mempool.h
> >  > b/lib/librte_mempool/rte_mempool.h
> >  > index 991feaa..b1186fd 100644
> >  > --- a/lib/librte_mempool/rte_mempool.h
> >  > +++ b/lib/librte_mempool/rte_mempool.h
> >  > @@ -53,11 +53,7 @@
> >  >   *
> >  >   * Note: the mempool implementation is not preemptable. A lcore must
> >  >   * not be interrupted by another task that uses the same mempool
> >  > - * (because it uses a ring which is not preemptable). Also, mempool
> >  > - * functions must not be used outside the DPDK environment: for
> >  > - * example, in linuxapp environment, a thread that is not created by
> >  > - * the EAL must not use mempools. This is due to the per-lcore cache
> >  > - * that won't work as rte_lcore_id() will not return a correct value.
> >  > + * (because it uses a ring which is not preemptible).
> >  >   */
> >  >
> >  >  #include <stdio.h>  
> >  
> >  I agree the comment is not correct today. But I think we should still highlight
> >  that usual functions [ex: rte_mempool_get(), rte_mempool_put()] won't work
> >  when called on a non-EAL thread.  
> 
> My understanding is that non-EAL threads can call those functions, but their performance will suffer since they go directly to the underlying ring. Is that correct?

Yes, that's more correct than what I wrote :)


> 
> >  
> >  Maybe it could be reworded in that way:
> >  
> >  """
> >  Note: the mempool implementation is not preemptable. A lcore must not be
> >  interrupted by another task that uses the same mempool (because it uses a ring
> >  which is not preemptible). Also, usual mempool functions like
> >  rte_mempool_get() or rte_mempool_put() are designed to be called from an
> >  EAL thread due to the internal per-lcore cache. When using a mempool from a
> >  non-EAL thread, a user cache has to be provided to
> >  rte_mempool_generic_get() or rte_mempool_generic_put().
> >  """
> >  
> >  What do you think?  
> 
> Agreed, this looks better. If indeed non-EAL threads can call rte_mempool_get() and _put(), perhaps the last sentence can be reworked like so?
> 
> Due to the lack of caching, rte_mempool_get() or rte_mempool_put() performance
> will suffer when called by non-EAL threads. Instead, non-EAL threads should call
> rte_mempool_generic_get() or rte_mempool_generic_put() with a user cache
> created with rte_mempool_cache_create().

Looks good, thanks!

Olivier


> 
> >  
> >  Thanks,
> >  Olivier