Re: [dpdk-dev] [PATCH] app/testpmd: do not enable Rx offloads by default

[Thomas Monjalon <thomas@monjalon.net>]

26/01/2018 08:31, Lu, Wenzhuo:
> From: Stephen Hemminger [mailto:stephen@networkplumber.org]
> > On Thu, 25 Jan 2018 10:04:11 +0100
> > Thomas Monjalon <thomas@monjalon.net> wrote:
> > > 25/01/2018 02:11, Lu, Wenzhuo:
> > > > > --- a/app/test-pmd/testpmd.c
> > > > > +++ b/app/test-pmd/testpmd.c
> > > > > @@ -305,9 +305,7 @@ struct fwd_engine * fwd_engines[] = {
> > > > >   */
> > > > >  struct rte_eth_rxmode rx_mode = {
> > > > >  	.max_rx_pkt_len = ETHER_MAX_LEN, /**< Default maximum frame
> > > > > length. */
> > > > > -	.offloads = (DEV_RX_OFFLOAD_VLAN_FILTER |
> > > > > -		     DEV_RX_OFFLOAD_VLAN_STRIP |
> > > > > -		     DEV_RX_OFFLOAD_CRC_STRIP),
> > > > > +	.offloads = 0,
> > > >
> > > > Change the default behavior may trigger other problems. I think TX
> > offload could be a good reference. Get the capability and check what's
> > supported first, then ignore the not supported functions with printing a
> > warning but not block anything...
> > >
> > > I agree that we should check the capabilities before requesting an offload.
> > > But I disagree on another point: we should not enable an offload if
> > > the user did not request it explicitly. It makes things unclear.
> > > This is a testing tool, it should be close to the ethdev API behavior.
> > >
> > > Why these offload flags are silently enabled?
> > 
> > Also all virtual devices ignore CRC strip.
> Look like it's the case the device ignores the flag if it doesn't have the capability.

It is a wrong behaviour!
If a configuration cannot be applied, it must be an error.