[dpdk-dev] fm10k: fix wrong VLAN value in RX mbuf

  VLAN value should be copied from RX descriptor to mbuf,
this patch fixes this issue.

Signed-off-by: Shaopeng He <shaopeng.he@intel.com>
---
 drivers/net/fm10k/fm10k_rxtx.c | 2 ++
 1 file changed, 2 insertions(+)
  

Hi, 

Worth to adding comments that vlan_tci is only valid in case RTE_LIBRTE_FM10K_RX_OLFLAGS_ENABLE  is turned on and 
Flag PKT_RX_VLAN_PKT is set.
 
Best Regards,
Mark


> -----Original Message-----
> From: He, Shaopeng
> Sent: Wednesday, November 18, 2015 4:50 PM
> To: dev@dpdk.org
> Cc: Chen, Jing D; Qiu, Michael; He, Shaopeng
> Subject: [PATCH] fm10k: fix wrong VLAN value in RX mbuf
> 
> VLAN value should be copied from RX descriptor to mbuf,
> this patch fixes this issue.
> 
> Signed-off-by: Shaopeng He <shaopeng.he@intel.com>
> ---
>  drivers/net/fm10k/fm10k_rxtx.c | 2 ++
>  1 file changed, 2 insertions(+)
> 
> diff --git a/drivers/net/fm10k/fm10k_rxtx.c
> b/drivers/net/fm10k/fm10k_rxtx.c
> index 1bac28d..eeb635e 100644
> --- a/drivers/net/fm10k/fm10k_rxtx.c
> +++ b/drivers/net/fm10k/fm10k_rxtx.c
> @@ -146,6 +146,7 @@ fm10k_recv_pkts(void *rx_queue, struct rte_mbuf
> **rx_pkts,
>  #endif
> 
>  		mbuf->hash.rss = desc.d.rss;
> +		mbuf->vlan_tci = desc.w.vlan &
> FM10K_RXD_VLAN_ID_MASK;
> 
>  		rx_pkts[count] = mbuf;
>  		if (++next_dd == q->nb_desc) {
> @@ -292,6 +293,7 @@ fm10k_recv_scattered_pkts(void *rx_queue, struct
> rte_mbuf **rx_pkts,
>  		rx_desc_to_ol_flags(first_seg, &desc);
>  #endif
>  		first_seg->hash.rss = desc.d.rss;
> +		first_seg->vlan_tci = desc.w.vlan &
> FM10K_RXD_VLAN_ID_MASK;
> 
>  		/* Prefetch data of first segment, if configured to do so. */
>  		rte_packet_prefetch((char *)first_seg->buf_addr +
> --
> 1.9.3
  

On Wed, 18 Nov 2015 16:50:09 +0800
Shaopeng He <shaopeng.he@intel.com> wrote:

> VLAN value should be copied from RX descriptor to mbuf,
> this patch fixes this issue.
> 
> Signed-off-by: Shaopeng He <shaopeng.he@intel.com>
> ---
>  drivers/net/fm10k/fm10k_rxtx.c | 2 ++
>  1 file changed, 2 insertions(+)
> 
> diff --git a/drivers/net/fm10k/fm10k_rxtx.c b/drivers/net/fm10k/fm10k_rxtx.c
> index 1bac28d..eeb635e 100644
> --- a/drivers/net/fm10k/fm10k_rxtx.c
> +++ b/drivers/net/fm10k/fm10k_rxtx.c
> @@ -146,6 +146,7 @@ fm10k_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts,
>  #endif
>  
>  		mbuf->hash.rss = desc.d.rss;
> +		mbuf->vlan_tci = desc.w.vlan & FM10K_RXD_VLAN_ID_MASK;
>  
>  		rx_pkts[count] = mbuf;
>  		if (++next_dd == q->nb_desc) {
> @@ -292,6 +293,7 @@ fm10k_recv_scattered_pkts(void *rx_queue, struct rte_mbuf **rx_pkts,
>  		rx_desc_to_ol_flags(first_seg, &desc);
>  #endif
>  		first_seg->hash.rss = desc.d.rss;
> +		first_seg->vlan_tci = desc.w.vlan & FM10K_RXD_VLAN_ID_MASK;
>  
>  		/* Prefetch data of first segment, if configured to do so. */
>  		rte_packet_prefetch((char *)first_seg->buf_addr +

Good catch, this bug means VLAN's would not work right now.
But I think it isn't quite right.

By masking the VLAN you are losing the received priority bits.
Other drivers like ixgbe preserve the priority bits.
  

Hi, Mark

> -----Original Message-----
> From: Chen, Jing D
> Sent: Thursday, November 19, 2015 3:05 PM
> To: He, Shaopeng; dev@dpdk.org
> Cc: Qiu, Michael
> Subject: RE: [PATCH] fm10k: fix wrong VLAN value in RX mbuf
> 
> Hi,
> 
> Worth to adding comments that vlan_tci is only valid in case
> RTE_LIBRTE_FM10K_RX_OLFLAGS_ENABLE  is turned on and Flag
> PKT_RX_VLAN_PKT is set.
Good suggestion. But, actually, we have a little different case in fm10k:
fm10k has a switch core inside, which associates a VLAN ID for each packet.
For those packets coming in without a VLAN, the port default VLAN ID will
be used. So for fm10k, vlan_tci is always valid. 
For Flag PKT_RX_VLAN_PKT, currently it will be set when there is inner VLAN
in the packet. This is not correct, I will fix it by setting PKT_RX_VLAN_PKT
always on in next version.

To verify this, different situations have been tested on
current implementation:
1. Ixia sent VLAN 51 packet, DPDK received: vlan_tci=51, PKT_RX_VLAN_PKT=0
2. sent non-VLAN packet, ether port's default VLAN ID is 1
    received: vlan_tci=1, PKT_RX_VLAN_PKT=0
3. sent double-VLAN packet (outer 51, inner 17),
    received: vlan_tci=51, PKT_RX_VLAN_PKT=1

Thanks,
--Shaopeng
> 
> Best Regards,
> Mark
> 
> 
> > -----Original Message-----
> > From: He, Shaopeng
> > Sent: Wednesday, November 18, 2015 4:50 PM
> > To: dev@dpdk.org
> > Cc: Chen, Jing D; Qiu, Michael; He, Shaopeng
> > Subject: [PATCH] fm10k: fix wrong VLAN value in RX mbuf
> >
> > VLAN value should be copied from RX descriptor to mbuf, this patch
> > fixes this issue.
> >
> > Signed-off-by: Shaopeng He <shaopeng.he@intel.com>
> > ---
> >  drivers/net/fm10k/fm10k_rxtx.c | 2 ++
> >  1 file changed, 2 insertions(+)
> >
> > diff --git a/drivers/net/fm10k/fm10k_rxtx.c
> > b/drivers/net/fm10k/fm10k_rxtx.c index 1bac28d..eeb635e 100644
> > --- a/drivers/net/fm10k/fm10k_rxtx.c
> > +++ b/drivers/net/fm10k/fm10k_rxtx.c
> > @@ -146,6 +146,7 @@ fm10k_recv_pkts(void *rx_queue, struct rte_mbuf
> > **rx_pkts,  #endif
> >
> >  		mbuf->hash.rss = desc.d.rss;
> > +		mbuf->vlan_tci = desc.w.vlan &
> > FM10K_RXD_VLAN_ID_MASK;
> >
> >  		rx_pkts[count] = mbuf;
> >  		if (++next_dd == q->nb_desc) {
> > @@ -292,6 +293,7 @@ fm10k_recv_scattered_pkts(void *rx_queue, struct
> > rte_mbuf **rx_pkts,
> >  		rx_desc_to_ol_flags(first_seg, &desc);  #endif
> >  		first_seg->hash.rss = desc.d.rss;
> > +		first_seg->vlan_tci = desc.w.vlan &
> > FM10K_RXD_VLAN_ID_MASK;
> >
> >  		/* Prefetch data of first segment, if configured to do so. */
> >  		rte_packet_prefetch((char *)first_seg->buf_addr +
> > --
> > 1.9.3
  

Hi, Stephen

> -----Original Message-----
> From: Stephen Hemminger [mailto:stephen@networkplumber.org]
> Sent: Friday, November 20, 2015 1:26 AM
> To: He, Shaopeng
> Cc: dev@dpdk.org
> Subject: Re: [dpdk-dev] [PATCH] fm10k: fix wrong VLAN value in RX mbuf
> 
> On Wed, 18 Nov 2015 16:50:09 +0800
> Shaopeng He <shaopeng.he@intel.com> wrote:
> 
> > VLAN value should be copied from RX descriptor to mbuf, this patch
> > fixes this issue.
> >
> > Signed-off-by: Shaopeng He <shaopeng.he@intel.com>
> > ---
> >  drivers/net/fm10k/fm10k_rxtx.c | 2 ++
> >  1 file changed, 2 insertions(+)
> >
> > diff --git a/drivers/net/fm10k/fm10k_rxtx.c
> > b/drivers/net/fm10k/fm10k_rxtx.c index 1bac28d..eeb635e 100644
> > --- a/drivers/net/fm10k/fm10k_rxtx.c
> > +++ b/drivers/net/fm10k/fm10k_rxtx.c
> > @@ -146,6 +146,7 @@ fm10k_recv_pkts(void *rx_queue, struct rte_mbuf
> > **rx_pkts,  #endif
> >
> >  		mbuf->hash.rss = desc.d.rss;
> > +		mbuf->vlan_tci = desc.w.vlan &
> FM10K_RXD_VLAN_ID_MASK;
> >
> >  		rx_pkts[count] = mbuf;
> >  		if (++next_dd == q->nb_desc) {
> > @@ -292,6 +293,7 @@ fm10k_recv_scattered_pkts(void *rx_queue, struct
> rte_mbuf **rx_pkts,
> >  		rx_desc_to_ol_flags(first_seg, &desc);  #endif
> >  		first_seg->hash.rss = desc.d.rss;
> > +		first_seg->vlan_tci = desc.w.vlan &
> FM10K_RXD_VLAN_ID_MASK;
> >
> >  		/* Prefetch data of first segment, if configured to do so. */
> >  		rte_packet_prefetch((char *)first_seg->buf_addr +
> 
> Good catch, this bug means VLAN's would not work right now.
> But I think it isn't quite right.

Actually, VLAN filter works well, only the vlan_tci in mbuf has wrong value.
And because not much places use the vlan_tci, this bug was not
found  before.

> 
> By masking the VLAN you are losing the received priority bits.
> Other drivers like ixgbe preserve the priority bits.

Yes, you are right, in next version, I will preserve the priority bits like others.

Thanks,
--Shaopeng