[dpdk-dev] [PATCH 2/5] net/mlx5: fix secondary process mempool registration

[Shahaf Shuler]

Wednesday, January 24, 2018 10:14 AM, Nélio Laranjeiro:
> Hi Shahaf,
> 
> On Tue, Jan 23, 2018 at 07:08:20PM +0200, Shahaf Shuler wrote:
> > Secondary process is not allowed to register mempools on the flight.
> >
> > The code will return invalid memory key for such case.
> >
> > Fixes: 87ec44ce1651 ("net/mlx5: add operations for secondary process")
> > Cc: stable at dpdk.org
> >
> > Signed-off-by: Shahaf Shuler <shahafs at mellanox.com>
> > Signed-off-by: Xueming Li <xuemingl at mellanox.com>
> > ---
> >  doc/guides/nics/mlx5.rst     |  7 ++++++-
> >  drivers/net/mlx5/mlx5_rxtx.h | 17 ++++++++++++++---
> >  2 files changed, 20 insertions(+), 4 deletions(-)
> >
> > diff --git a/doc/guides/nics/mlx5.rst b/doc/guides/nics/mlx5.rst index
> > bdc2216c0..2f860402f 100644
> > --- a/doc/guides/nics/mlx5.rst
> > +++ b/doc/guides/nics/mlx5.rst
> > @@ -106,7 +106,12 @@ Limitations
> >
> >  - Inner RSS for VXLAN frames is not supported yet.
> >  - Hardware checksum RX offloads for VXLAN inner header are not
> supported yet.
> > -- Forked secondary process not supported.
> > +- For secondary process:
> > +
> > +  - Forked secondary process not supported.
> > +  - All mempools must be initialized before rte_eth_dev_start().
> > +  - Number of mempools must less than
> > + CONFIG_RTE_LIBRTE_MLX5_TX_MP_CACHE
> 
> Why such limitation?  Registering a new memory is independent of searching
> a new one when the cache is too small.
> 
> >  - Flow pattern without any specific vlan will match for vlan packets as well:
> >
> >    When VLAN spec is not specified in the pattern, the matching rule will be
> created with VLAN as a wild card.
> > diff --git a/drivers/net/mlx5/mlx5_rxtx.h
> > b/drivers/net/mlx5/mlx5_rxtx.h index a63364d79..79cdfc793 100644
> > --- a/drivers/net/mlx5/mlx5_rxtx.h
> > +++ b/drivers/net/mlx5/mlx5_rxtx.h
> > @@ -550,6 +550,7 @@ mlx5_tx_mb2mr(struct mlx5_txq_data *txq, struct
> rte_mbuf *mb)
> >  	uint16_t i = txq->mr_cache_idx;
> >  	uintptr_t addr = rte_pktmbuf_mtod(mb, uintptr_t);
> >  	struct mlx5_mr *mr;
> > +	struct rte_mempool *mp;
> >
> >  	assert(i < RTE_DIM(txq->mp2mr));
> >  	if (likely(txq->mp2mr[i]->start <= addr && txq->mp2mr[i]->end >=
> > addr)) @@ -563,14 +564,24 @@ mlx5_tx_mb2mr(struct mlx5_txq_data
> *txq, struct rte_mbuf *mb)
> >  		if (txq->mp2mr[i]->start <= addr &&
> >  		    txq->mp2mr[i]->end >= addr) {
> >  			assert(txq->mp2mr[i]->lkey != (uint32_t)-1);
> > -			assert(rte_cpu_to_be_32(txq->mp2mr[i]->mr->lkey)
> ==
> > -			       txq->mp2mr[i]->lkey);
> > +			if (rte_eal_process_type() == RTE_PROC_PRIMARY) {
> > +				assert(rte_cpu_to_be_32(txq->mp2mr[i]-
> >mr->lkey)
> > +				       == txq->mp2mr[i]->lkey);
> > +			}
> 
> This code should be inside priv_txq_mp2mr_reg() to let the secondary
> search inside the MR list when the cache is too small.

You probably mean below code and not above. Yes good suggestion.
Regarding the above code - the assert in un-accessible from the secondary process as it tries to access ibv_mr.
I will add another patch to remove this assert completely (in order to save the comparison logic). 

> 
> If it does not find any MR it should fail before calling priv_mr_new().
> 
> >  			txq->mr_cache_idx = i;
> >  			return txq->mp2mr[i]->lkey;
> >  		}
> >  	}
> >  	txq->mr_cache_idx = 0;
> > -	mr = mlx5_txq_mp2mr_reg(txq, mlx5_tx_mb2mp(mb), i);
> > +	mp = mlx5_tx_mb2mp(mb);
> > +	if (rte_eal_process_type() != RTE_PROC_PRIMARY) {
> > +		WARN("Using unregistered mempool 0x%p(%s) in secondary
> process,"
> > +		     " please create mempool before rte_eth_dev_start()",
> > +		     (void *)mp, mp->name);
> > +		assert(rte_eal_process_type() == RTE_PROC_PRIMARY);
> > +		return (uint32_t)-1;
> > +	}
> > +	mr = mlx5_txq_mp2mr_reg(txq, mp, i);
> >  	/*
> >  	 * Request the reference to use in this queue, the original one is
> >  	 * kept by the control plane.
> > --
> > 2.12.0
> 
> --
> Nélio Laranjeiro
> 6WIND