[dpdk-dev] [PATCH v2 05/39] examples/l3fwd: move to ethdev offloads API

[Ananyev, Konstantin]

Hi Jerin,

> -----Original Message-----
> From: Jerin Jacob [mailto:jerin.jacob at caviumnetworks.com]
> Sent: Wednesday, December 13, 2017 7:55 AM
> To: Shahaf Shuler <shahafs at mellanox.com>
> Cc: Ananyev, Konstantin <konstantin.ananyev at intel.com>; dev at dpdk.org; Nicolau, Radu <radu.nicolau at intel.com>;
> arybchenko at solarflare.com
> Subject: Re: [dpdk-dev] [PATCH v2 05/39] examples/l3fwd: move to ethdev offloads API
> 
> -----Original Message-----
> > Date: Wed, 13 Dec 2017 07:21:01 +0000
> > From: Shahaf Shuler <shahafs at mellanox.com>
> > To: "Ananyev, Konstantin" <konstantin.ananyev at intel.com>, "dev at dpdk.org"
> >  <dev at dpdk.org>, "Nicolau, Radu" <radu.nicolau at intel.com>,
> >  "arybchenko at solarflare.com" <arybchenko at solarflare.com>
> > Subject: Re: [dpdk-dev] [PATCH v2 05/39] examples/l3fwd: move to ethdev
> >  offloads API
> >
> > Tuesday, December 12, 2017 7:12 PM, Ananyev, Konstantin:
> > > > -----Original Message-----
> > > > From: Shahaf Shuler [mailto:shahafs at mellanox.com]
> > > > Sent: Tuesday, December 12, 2017 12:26 PM
> > > > To: dev at dpdk.org; Ananyev, Konstantin
> > > <konstantin.ananyev at intel.com>;
> > > > Nicolau, Radu <radu.nicolau at intel.com>; arybchenko at solarflare.com
> > > > Subject: [PATCH v2 05/39] examples/l3fwd: move to ethdev offloads API
> > > >
> > > > Ethdev offloads API has changed since:
> > > >
> > > > commit ce17eddefc20 ("ethdev: introduce Rx queue offloads API") commit
> > > > cba7f53b717d ("ethdev: introduce Tx queue offloads API")
> > > >
> > > > This commit support the new API.
> > > >
> > > > Signed-off-by: Shahaf Shuler <shahafs at mellanox.com>
> > > > ---
> > > >  examples/l3fwd/main.c | 40 ++++++++++++++++++++++++++++++--------
> > > --
> > > >  1 file changed, 30 insertions(+), 10 deletions(-)
> > > >
> > > > diff --git a/examples/l3fwd/main.c b/examples/l3fwd/main.c index
> > > > 6229568..3bdf4d5 100644
> > > > --- a/examples/l3fwd/main.c
> > > > +++ b/examples/l3fwd/main.c
> > > > @@ -149,11 +149,9 @@ struct lcore_params {
> > > >  		.mq_mode = ETH_MQ_RX_RSS,
> > > >  		.max_rx_pkt_len = ETHER_MAX_LEN,
> > > >  		.split_hdr_size = 0,
> > > > -		.header_split   = 0, /**< Header Split disabled */
> > > > -		.hw_ip_checksum = 1, /**< IP checksum offload enabled */
> > > > -		.hw_vlan_filter = 0, /**< VLAN filtering disabled */
> > > > -		.jumbo_frame    = 0, /**< Jumbo Frame Support disabled */
> > > > -		.hw_strip_crc   = 1, /**< CRC stripped by hardware */
> > > > +		.ignore_offload_bitfield = 1,
> > > > +		.offloads = (DEV_RX_OFFLOAD_CRC_STRIP |
> > > > +			     DEV_RX_OFFLOAD_CHECKSUM),
> > > >  	},
> > > >  	.rx_adv_conf = {
> > > >  		.rss_conf = {
> > > > @@ -163,6 +161,7 @@ struct lcore_params {
> > > >  	},
> > > >  	.txmode = {
> > > >  		.mq_mode = ETH_MQ_TX_NONE,
> > > > +		.offloads = DEV_TX_OFFLOAD_MBUF_FAST_FREE,
> > >
> > > Hmm, does it mean a new warning for all PMDs (majority) which don't
> > > support DEV_TX_OFFLOAD_MBUF_FAST_FREE?
> >
> > Good point.
> > Unlike other offloads which are must for the application proper run, this one it only for optimizing the performance and should be set only
> if PMD supports.
> > Am continuing to aggregate reasons why the DEV_TX_OFFLOAD_MBUF_FAST_FREE should not be defined as an offload. Anyway we
> passed that...
> >
> > I will fix on v3.
> 
> Removing is not an option as the PMDs rely on that flag to will have the
> impact.
> # I see DEV_TX_OFFLOAD_MBUF_FAST_FREE as hint driver to depict the application requirements
> # All the drivers by default can support DEV_TX_OFFLOAD_MBUF_FAST_FREE(They are using the hint or
> not is a different question)
> 
> So, How about setting DEV_TX_OFFLOAD_MBUF_FAST_FREE in all PMD driver as
> dummy one? I think, currently, it can be moved to old API to new API
> transition function till the drivers change to new offload flag scheme.

I don't think anyone plans to remove it right now.
If you believe your PMD does need it, that's ok by me.
Though I still think it is a very limited usage for it, and I don't think
we have to make that flag supported by all PMDs.
Konstantin

> 
> We are planning to change nicvf driver to new offload scheme for this
> release so with this change, we have the performance impact on l3fwd
> application.
> 
> I think, the other option could be to change usage/meaning of
> DEV_TX_OFFLOAD_MBUF_FAST_FREE flag where when the application needs
> multi-pool and reference count scheme then "it sets" the offload flags.
> If so, we don't need to set by default on the these applications.