Re: [dpdk-dev] [PATCH v3 2/2] app/testpmd: only config supported RSS hash types

[Adrien Mazarguil <adrien.mazarguil@6wind.com>]

On Wed, Apr 18, 2018 at 01:54:20PM +0000, Xueming(Steven) Li wrote:
> 
> 
> > -----Original Message-----
> > From: Adrien Mazarguil <adrien.mazarguil@6wind.com>
> > Sent: Wednesday, April 18, 2018 9:26 PM
> > To: Xueming(Steven) Li <xuemingl@mellanox.com>
> > Cc: Shahaf Shuler <shahafs@mellanox.com>; Nelio Laranjeiro <notifications@github.com>; Wenzhuo Lu
> > <wenzhuo.lu@intel.com>; Jingjing Wu <jingjing.wu@intel.com>; Thomas Monjalon <thomas@monjalon.net>;
> > dev@dpdk.org
> > Subject: Re: [dpdk-dev] [PATCH v3 2/2] app/testpmd: only config supported RSS hash types
> > 
> > Hi Xueming,
> > 
> > On Wed, Apr 18, 2018 at 07:06:48PM +0800, Xueming Li wrote:
> > > "port config all rss all" command will fail on PMD that not support
> > > any of hard coding RSS hash types. This patch changed hard coding hash
> > > types to supported types retrieved from device info.
> > >
> > > Signed-off-by: Xueming Li <xuemingl@mellanox.com>
> > 
> > Problem is that this patch removes the ability to request "all" RSS types regardless of PMD support.
> > 
> > Users will expect "all" to behave as documented, however doing so will only result in the limited set
> > of types reported by PMDs. For instance at least
> > mlx4 doesn't update the flow_type_rss_offloads field since it has never implemented configuration
> > support for the legacy RSS API.
> > 
> > You should add an argument with a different name (e.g. "supported" or
> > "default") for that and keep the original meaning for "all".
> > 
> > Testpmd documentation has to be updated accordingly.
> 
> You must want to have a look at first part of this patchset:
> http://www.dpdk.org/ml/archives/dev/2018-April/097849.html

I did, however this is does not address my previous comment.

In this patch you're preventing testpmd users from requesting something. If
they want to enable "all" RSS types and some of them happen to be
unsupported on some ports, it means they want to get errors. Otherwise they
might think their request succeeded while it was in fact interpreted in an
undocumented manner.

The point is if users want to enable only "supported" RSS types, they
wouldn't write "all" on the command line right? A new parameter is needed
for that.

> 
> > 
> > One more nit below.
> > 
> > > ---
> > >  app/test-pmd/cmdline.c | 7 +++++++
> > >  1 file changed, 7 insertions(+)
> > >
> > > diff --git a/app/test-pmd/cmdline.c b/app/test-pmd/cmdline.c index
> > > 512e3b55e..d357de7e6 100644
> > > --- a/app/test-pmd/cmdline.c
> > > +++ b/app/test-pmd/cmdline.c
> > > @@ -1998,6 +1998,7 @@ cmd_config_rss_parsed(void *parsed_result,  {
> > >  	struct cmd_config_rss *res = parsed_result;
> > >  	struct rte_eth_rss_conf rss_conf = { .rss_key_len = 0, };
> > > +	struct rte_eth_dev_info dev_info = {0};
> > 
> > This could be declared in the new block where it's used. Also note that "{0}" is nonstandard syntax,
> > use memset() or initialize a field like rss_key_len above.
> > 
> > >  	int diag;
> > >  	uint8_t i;
> > >
> > > @@ -2034,6 +2035,12 @@ cmd_config_rss_parsed(void *parsed_result,
> > >  	}
> > >  	rss_conf.rss_key = NULL;
> > >  	for (i = 0; i < rte_eth_dev_count(); i++) {
> > > +		if (!strcmp(res->value, "all")) {
> > > +			rte_eth_dev_info_get(i, &dev_info);
> > > +			if (dev_info.flow_type_rss_offloads)
> > > +				rss_conf.rss_hf =
> > > +					dev_info.flow_type_rss_offloads;
> > > +		}
> > >  		diag = rte_eth_dev_rss_hash_update(i, &rss_conf);
> > >  		if (diag < 0)
> > >  			printf("Configuration of RSS hash at ethernet port %d "
> > > --
> > > 2.13.3
> > >
> > 
> > 
> > --
> > Adrien Mazarguil
> > 6WIND

-- 
Adrien Mazarguil
6WIND