[dpdk-dev] [PATCH v2 3/5] net/ixgbe: fix bitmask of supported Tx flags

[Ananyev, Konstantin]

> -----Original Message-----
> From: Wu, Jingjing
> Sent: Tuesday, February 7, 2017 2:30 AM
> To: Ananyev, Konstantin <konstantin.ananyev at intel.com>; dev at dpdk.org
> Subject: RE: [dpdk-dev] [PATCH v2 3/5] net/ixgbe: fix bitmask of supported Tx flags
> 
> 
> 
> > -----Original Message-----
> > From: Ananyev, Konstantin
> > Sent: Monday, February 6, 2017 8:11 PM
> > To: Wu, Jingjing <jingjing.wu at intel.com>; dev at dpdk.org
> > Subject: RE: [dpdk-dev] [PATCH v2 3/5] net/ixgbe: fix bitmask of supported
> > Tx flags
> >
> >
> >
> > > -----Original Message-----
> > > From: Wu, Jingjing
> > > Sent: Monday, February 6, 2017 8:54 AM
> > > To: Ananyev, Konstantin <konstantin.ananyev at intel.com>; dev at dpdk.org
> > > Subject: RE: [dpdk-dev] [PATCH v2 3/5] net/ixgbe: fix bitmask of
> > > supported Tx flags
> > >
> > >
> > >
> > > > -----Original Message-----
> > > > From: Ananyev, Konstantin
> > > > Sent: Sunday, February 5, 2017 8:11 PM
> > > > To: Ananyev, Konstantin <konstantin.ananyev at intel.com>; Wu, Jingjing
> > > > <jingjing.wu at intel.com>; dev at dpdk.org
> > > > Subject: RE: [dpdk-dev] [PATCH v2 3/5] net/ixgbe: fix bitmask of
> > > > supported Tx flags
> > > >
> > > >
> > > >
> > > > > -----Original Message-----
> > > > > From: dev [mailto:dev-bounces at dpdk.org] On Behalf Of Ananyev,
> > > > > Konstantin
> > > > > Sent: Sunday, February 5, 2017 11:59 AM
> > > > > To: Wu, Jingjing <jingjing.wu at intel.com>; dev at dpdk.org
> > > > > Subject: Re: [dpdk-dev] [PATCH v2 3/5] net/ixgbe: fix bitmask of
> > > > > supported Tx flags
> > > > >
> > > > > Hi Jingjing,
> > > > >
> > > > >
> > > > > > -----Original Message-----
> > > > > > From: Wu, Jingjing
> > > > > > Sent: Saturday, February 4, 2017 3:36 AM
> > > > > > To: dev at dpdk.org
> > > > > > Cc: Wu, Jingjing <jingjing.wu at intel.com>; Ananyev, Konstantin
> > > > > > <konstantin.ananyev at intel.com>
> > > > > > Subject: [PATCH v2 3/5] net/ixgbe: fix bitmask of supported Tx
> > > > > > flags
> > > > > >
> > > > > > Add missed flags to bitmask of all supported packet Tx flags.
> > > > > >
> > > > > > CC: konstantin.ananyev at intel.com
> > > > > > Fixes: 7829b8d52be0 ("net/ixgbe: add Tx preparation")
> > > > > > Signed-off-by: Jingjing Wu <jingjing.wu at intel.com>
> > > > > > ---
> > > > > >  drivers/net/ixgbe/ixgbe_rxtx.c | 17 ++++++++++++++++-
> > > > > >  1 file changed, 16 insertions(+), 1 deletion(-)
> > > > > >
> > > > > > diff --git a/drivers/net/ixgbe/ixgbe_rxtx.c
> > > > > > b/drivers/net/ixgbe/ixgbe_rxtx.c index 36f1c02..8454581 100644
> > > > > > --- a/drivers/net/ixgbe/ixgbe_rxtx.c
> > > > > > +++ b/drivers/net/ixgbe/ixgbe_rxtx.c
> > > > > > @@ -81,13 +81,28 @@
> > > > > >  #include "ixgbe_rxtx.h"
> > > > > >
> > > > > >  /* Bit Mask to indicate what bits required for building TX
> > > > > > context */
> > > > > > +#ifdef RTE_LIBRTE_IEEE1588
> > > > > >  #define IXGBE_TX_OFFLOAD_MASK (			 \
> > > > > >  		PKT_TX_VLAN_PKT |		 \
> > > > > >  		PKT_TX_IP_CKSUM |		 \
> > > > > > +		PKT_TX_IPV4 |			 \
> > > > > >  		PKT_TX_L4_MASK |		 \
> > > > > > +		PKT_TX_IEEE1588_TMST |		 \
> > > > > >  		PKT_TX_TCP_SEG |		 \
> > > > > >  		PKT_TX_MACSEC |			 \
> > > > > > -		PKT_TX_OUTER_IP_CKSUM)
> > > > > > +		PKT_TX_OUTER_IP_CKSUM |		 \
> > > > > > +		PKT_TX_OUTER_IPV4)
> > > > > > +#else
> > > > > > +#define IXGBE_TX_OFFLOAD_MASK (			 \
> > > > > > +		PKT_TX_VLAN_PKT |		 \
> > > > > > +		PKT_TX_IP_CKSUM |		 \
> > > > > > +		PKT_TX_IPV4 |			 \
> > > > >
> > > > > Wonder why ixgbe doesn't have PKT_TX_IPV6?
> > > >
> > > > Same question for e1000 and fm10k.
> > > > Also if you decided to go that way, you'll probably need to update
> > > > TX_OFFLOAD_MASK for enic and vmxnet3.
> > > > That's why I still think it would be much less hassle not to include
> > > > these flags
> > > > (PKT_TX_IPV4 and PKT_TX_IPV6)  into TX_OFFLOAD_MASK at all.
> > > > Konstantin
> > > >
> > > >
> > > Thanks for pointing that. PKT_TX_IPV6 is missed.
> > > About whether include these flags (PKT_TX_IPV4 and PKT_TX_IPV6)  into
> > > TX_OFFLOAD_MASK, I think they should be Included. Think about one NIC
> > may support IPV4 L4 checksum offload, but not support IPV6? Even I don't
> > know who it is.
> > >
> >
> > I don't think such combination is possible now anyway.
> > But ok, if your preference is it to do more work and add (PKT_TX_IPV4 |
> > PKT_TX_IPV6) into all required places, I wouldn't argue.
> >
> > BTW, as a side notice, what will be really good is to have a function that
> > would take tx_offload_capabilities as an input and return tx_offload_mask.
> > That would remove necessity to update/support TX_OFFLOAD_MASK for
> > each PMD, and hopefully will allow to avoid confusion for PMD writers.
> > Though that's probably subject of another patch.
> >
> 
> OK. I think what I did is more than necessary. Let me simplify the change. Thanks!
> 
> And about the querying tx offload capabilities, I think it is already been done
> In rte_eth_dev_info_get. But it used another set of flags which is not TX flags
> Defined in mbuf.

I am not talking about querying device offload capabilities.
I am talking about a function that would take these capabilities reported by the
device and convert them into a set of tx offload flags, that are suitable for these capabilities. 
Something like that:
uint64_t
tx_capa_to flags(uint32_t tx_capa)
{
	uint64_t ol_flags;
	
	ol_flags = 0;

	if (tx_capa &  DEV_TX_OFFLOAD_TCP_TSO)
		ol_flags |= PKT_TX_TCP_SEG |  PKT_TX_IPV4 |  PKT_TX_IP_CKSUM  | PKT_TX_IPV6;
	if (tx_capa &  DEV_TX_OFFLOAD_TCP_CKSUM)
		ol_flags |= PKT_TX_TCP_CKSUM;
	...
	return ol_flags;
}   

Then PMD writer wouldn't need to define TX_OFFLOAD_MASK for his PMD manually.
Instead he can just do at startup/configure phase:
txq->tx_offload_mask = tx_capa_to flags(supported_and_configured_capabilties);

Though, as I said, this is probably subject of another patch/discussion.
Konstantin