[dpdk-dev] [PATCH] net/ixgbe: fix MAT enable for VF in multicast

[Zhang, Qi Z]

> -----Original Message-----
> From: Zhao1, Wei
> Sent: Friday, January 4, 2019 4:35 PM
> To: Zhang, Qi Z <qi.z.zhang at intel.com>; dev at dpdk.org
> Cc: stable at dpdk.org; Wu, Jingjing <jingjing.wu at intel.com>
> Subject: RE: [dpdk-dev] [PATCH] net/ixgbe: fix MAT enable for VF in multicast
> 
> 
> 
> > -----Original Message-----
> > From: Zhang, Qi Z
> > Sent: Thursday, January 3, 2019 9:47 PM
> > To: Zhao1, Wei <wei.zhao1 at intel.com>; dev at dpdk.org
> > Cc: stable at dpdk.org; Wu, Jingjing <jingjing.wu at intel.com>; Zhao1, Wei
> > <wei.zhao1 at intel.com>
> > Subject: RE: [dpdk-dev] [PATCH] net/ixgbe: fix MAT enable for VF in
> > multicast
> >
> > Hi Wei
> >
> > > -----Original Message-----
> > > From: dev [mailto:dev-bounces at dpdk.org] On Behalf Of Wei Zhao
> > > Sent: Wednesday, January 2, 2019 2:33 PM
> > > To: dev at dpdk.org
> > > Cc: stable at dpdk.org; Wu, Jingjing <jingjing.wu at intel.com>; Zhao1,
> > > Wei <wei.zhao1 at intel.com>
> > > Subject: [dpdk-dev] [PATCH] net/ixgbe: fix MAT enable for VF in
> > > multicast
> >
> > What is MAT means ?
> > >
> > > In ixgbe PMD code, all vf ars set with bit IXGBE_VMOLR_ROMPE, which
> > > make vf accept packets that match the MTA table, if some vf update
> > > IXGBE_MTA in function ixgbe_vf_set_multicast, then all vf will
> > > receive packets from these address.
> > > So thhere is need to set VMOLR register bit ROPE onlty after this vf
> > > has been set multicast address. If this bit is when pf host doing
> > > initialization, this vf will receive multicast packets with address
> > > written in MTA table. Align to ixgbe pf kernel 5.3.7 code to fix this bug.
> >
> > There are some typo in you commit log.
> 
> Sorry, v2 will commit.
> 
> >
> > >
> > > Fixes: 00e30184daa0 ("ixgbe: add PF support")
> > >
> > > Signed-off-by: Wei Zhao <wei.zhao1 at intel.com>
> > > ---
> > >  drivers/net/ixgbe/ixgbe_pf.c | 6 +++++-
> > >  1 file changed, 5 insertions(+), 1 deletion(-)
> > >
> > > diff --git a/drivers/net/ixgbe/ixgbe_pf.c
> > > b/drivers/net/ixgbe/ixgbe_pf.c index 4b833ff..0f4b96b 100644
> > > --- a/drivers/net/ixgbe/ixgbe_pf.c
> > > +++ b/drivers/net/ixgbe/ixgbe_pf.c
> > > @@ -351,7 +351,7 @@ ixgbe_vf_reset_event(struct rte_eth_dev *dev,
> > > uint16_t vf)
> > >  	int rar_entry = hw->mac.num_rar_entries - (vf + 1);
> > >  	uint32_t vmolr = IXGBE_READ_REG(hw, IXGBE_VMOLR(vf));
> > >
> > > -	vmolr |= (IXGBE_VMOLR_ROPE | IXGBE_VMOLR_ROMPE |
> > > +	vmolr |= (IXGBE_VMOLR_ROPE |
> > >  			IXGBE_VMOLR_BAM | IXGBE_VMOLR_AUPE);
> > >  	IXGBE_WRITE_REG(hw, IXGBE_VMOLR(vf), vmolr);
> > >
> > > @@ -503,6 +503,7 @@ ixgbe_vf_set_multicast(struct rte_eth_dev *dev,
> > > uint32_t vf, uint32_t *msgbuf)
> > >  	const uint32_t IXGBE_MTA_BIT_MASK = (0x1 <<
> > IXGBE_MTA_BIT_SHIFT) -
> > > 1;
> > >  	uint32_t reg_val;
> > >  	int i;
> > > +	u32 vmolr = IXGBE_READ_REG(hw, IXGBE_VMOLR(vf));
> > >
> > >  	/* Disable multicast promiscuous first */
> > >  	ixgbe_disable_vf_mc_promisc(dev, vf); @@ -525,6 +526,9 @@
> > > ixgbe_vf_set_multicast(struct rte_eth_dev *dev, uint32_t vf,
> > > uint32_t
> > > *msgbuf)
> > >  		IXGBE_WRITE_REG(hw, IXGBE_MTA(mta_idx), reg_val);
> > >  	}
> > >
> > > +	vmolr |= IXGBE_VMOLR_ROMPE;
> > > +	IXGBE_WRITE_REG(hw, IXGBE_VMOLR(vf), vmolr);
> >
> >
> > Please correct me if I'm wrong
> >
> > My understand is MTA table is shared by all VFs (I guess also pf), but
> > what if
> 
> Yes, vf share it but not pf, it is used in vf pool switch
> 
> > two VFs both enable multi-cast but with different address filter?
> > Should we prevent the second one to enable multi-cast if any conflict
> > be detected? Otherwise there still will be unexpected behavior.
> 
> Sorry, I do not known what is the confict.
> Because IXGBE_VMOLR is vf specific, that is to say, each vf control itself for
> enable ROMPE.

The conflict what I mean is for example, vf1 set filter for address A while vf2 set filter for address B, then both VFs will receive A and B which is not expected.
But if the only issue for receive unexpected multi-cast packet is just a performance impact then I agree with most of patch, 
one more comment is we may need to reset IXGBE_VMOLR_ROMPE if nb_entries = 0, which is the case to clear all multi-cast filter.

> 
> 
> >
> > > +
> > >  	return 0;
> > >  }
> > >
> > > --
> > > 2.7.5