[dpdk-dev] [PATCH v3 1/2] net/ixgbe: fix x550 code to handle unidentified PHY

[Zhang, Qi Z]

> -----Original Message-----
> From: Luca Boccassi [mailto:bluca at debian.org]
> Sent: Wednesday, November 7, 2018 5:55 AM
> To: Zhang, Qi Z <qi.z.zhang at intel.com>; Chas Williams <3chas3 at gmail.com>;
> dev at dpdk.org
> Cc: Lu, Wenzhuo <wenzhuo.lu at intel.com>; Ananyev, Konstantin
> <konstantin.ananyev at intel.com>; stable at dpdk.org
> Subject: Re: [PATCH v3 1/2] net/ixgbe: fix x550 code to handle unidentified
> PHY
> 
> On Tue, 2018-11-06 at 23:31 +0000, Zhang, Qi Z wrote:
> > > -----Original Message-----
> > > From: Chas Williams [mailto:3chas3 at gmail.com]
> > > Sent: Monday, November 5, 2018 11:19 AM
> > > To: Zhang, Qi Z <qi.z.zhang at intel.com>; Luca Boccassi <bluca at debian
> > > .org>; dev at dpdk.org
> > > Cc: Lu, Wenzhuo <wenzhuo.lu at intel.com>; Ananyev, Konstantin
> > > <konstantin.ananyev at intel.com>; stable at dpdk.org
> > > Subject: Re: [PATCH v3 1/2] net/ixgbe: fix x550 code to handle
> > > unidentified PHY
> > >
> > >
> > >
> > > On 11/05/2018 12:41 PM, Zhang, Qi Z wrote:
> > > >
> > > >
> > > > > -----Original Message-----
> > > > > From: Luca Boccassi [mailto:bluca at debian.org]
> > > > > Sent: Friday, November 2, 2018 8:19 AM
> > > > > To: dev at dpdk.org
> > > > > Cc: Lu, Wenzhuo <wenzhuo.lu at intel.com>; Ananyev, Konstantin
> > > > > <konstantin.ananyev at intel.com>; Zhang, Qi Z <qi.z.zhang at intel.c
> > > > > om>;
> > > > > 3chas3 at gmail.com; Luca Boccassi <bluca at debian.org>; stable at dpdk
> > > > > .org
> > > > > Subject: [PATCH v3 1/2] net/ixgbe: fix x550 code to handle
> > > > > unidentified PHY
> > > > >
> > > > > ixgbe_identify_sfp_module_X550em() was missing the code to
> > > > > handle
> > > > > unidentified PHY that has been there in 82599 so it was not
> > > > > able to
> > > > > complete initialization of ixgbe sequence if no sfp plugged in.
> > > > > Port it over to return an appropriate type and complete init
> > > > > sequence
> > > > > properly.
> > > > >
> > > > > Fixes: d2e72774e58c ("ixgbe/base: support X550")
> > > > > Cc: stable at dpdk.org
> > > > >
> > > > > Signed-off-by: Luca Boccassi <bluca at debian.org>
> > > > > ---
> > > > > v2: refresh to remove merge conflict with master
> > > > > v3: coalesce fix into ixgbe_identify_sfp_module_X550em to avoid
> > > > >      code duplication, improve comment
> > > > >
> > > > >   drivers/net/ixgbe/base/ixgbe_x550.c | 6 ++++++
> > > > >   1 file changed, 6 insertions(+)
> > > > >
> > > > > diff --git a/drivers/net/ixgbe/base/ixgbe_x550.c
> > > > > b/drivers/net/ixgbe/base/ixgbe_x550.c
> > > > > index f7b98af52..a88d5c86a 100644
> > > > > --- a/drivers/net/ixgbe/base/ixgbe_x550.c
> > > > > +++ b/drivers/net/ixgbe/base/ixgbe_x550.c
> > > > > @@ -1561,6 +1561,12 @@ s32
> > >
> > > ixgbe_identify_sfp_module_X550em(struct
> > > > > ixgbe_hw *hw)
> > > > >
> > > > >   	status = ixgbe_identify_module_generic(hw);
> > > > >
> > > > > +	/* Set PHY type none if no PHY detected to allow init
> > > > > without SFP */
> > > > > +	if (hw->phy.type == ixgbe_phy_unknown) {
> > > > > +		hw->phy.type = ixgbe_phy_none;
> > > >
> > > > Set PHY type to none for a device that does have PHY looks weird.
> > > > does ixgeb_phy_generic works here?
> > >
> > > Yes, it does seem strange but that's what
> > > ixgbe_identify_sfp_module_generic
> > > seems to do:
> > >
> > > 	err_read_i2c_eeprom:
> > > 		hw->phy.sfp_type = ixgbe_sfp_type_not_present;
> > > 		if (hw->phy.type != ixgbe_phy_nl) {
> > > 			hw->phy.id = 0;
> > > 			hw->phy.type = ixgbe_phy_unknown;
> > > 		}
> > >
> > > The QSFP version a little more forceful:
> > >
> > > 	err_read_i2c_eeprom:
> > > 		hw->phy.sfp_type = ixgbe_sfp_type_not_present;
> > > 		hw->phy.id = 0;
> > > 		hw->phy.type = ixgbe_phy_unknown;
> > >
> > > If we go forward without setting the phy_type to none, we will
> > > eventually run
> > > into issues calling other phy routines.
> > >
> > > So should a lack of SFP, reset the PHY type? It's hazy because the
> > > difference
> > > between PHY and SFP isn't that clear to me here.
> >
> > I'm not sure that's the same case:).
> > Just feel that it's better to handle ixgbe_phy_unknown directly for
> > some device id as a special case than just replace it to
> > ixgbe_phy_none to cheat the check path, since that rely on we never
> > change the way to handle ixgbe_phy_none.
> >
> > So still have the question?
> > What is the failure if you go with ixgbe_phy_unknown?
> > Is that possible to work around this like
> > if (phy_type == ixgbe_phy_unknown && dev_id == xxxx)
> > 	...
> 
> Hi,
> 
> Thanks for having a look at this again. If you could please see the
> other answer, from myself, I've quoted the exact error we see and the
> issue it causes.

Yes, I see it failed at eth_ixgbe_dev_init, it will be better if you can provide more detail for the call stack, so we can figure out if we can work around this by handle ixgbe_phy_unknown directly with some special case.

> 
> --
> Kind regards,
> Luca Boccassi