[PATCH v6 2/2] net/ixgbe: Fix SFP detection and linking on hotplug

[Jeff Daly]

> -----Original Message-----
> From: Jeff Daly <jeffd at silicom-usa.com>
> Sent: Thursday, April 14, 2022 6:50 AM
> To: Wang, Haiyue <haiyue.wang at intel.com>; dev at dpdk.org
> Cc: stable at dpdk.org; Yang, Qiming <qiming.yang at intel.com>
> Subject: RE: [PATCH v6 2/2] net/ixgbe: Fix SFP detection and linking on
> hotplug
> 
> Caution: This is an external email. Please take care when clicking links or
> opening attachments.
> 
> 
> > -----Original Message-----
> > From: Wang, Haiyue <haiyue.wang at intel.com>
> > Sent: Tuesday, April 12, 2022 10:47 PM
> > To: Jeff Daly <jeffd at silicom-usa.com>; dev at dpdk.org
> > Cc: stable at dpdk.org; Stephen Douthit <stephend at silicom-usa.com>; Yang,
> > Qiming <qiming.yang at intel.com>
> > Subject: RE: [PATCH v6 2/2] net/ixgbe: Fix SFP detection and linking
> > on hotplug
> >
> > Caution: This is an external email. Please take care when clicking
> > links or opening attachments.
> >
> >
> > > -----Original Message-----
> > > From: Jeff Daly <jeffd at silicom-usa.com>
> > > Sent: Wednesday, April 13, 2022 01:42
> > > To: dev at dpdk.org
> > > Cc: stable at dpdk.org; Stephen Douthit <stephend at silicom-usa.com>;
> > Wang,
> > > Haiyue <haiyue.wang at intel.com>
> > > Subject: [PATCH v6 2/2] net/ixgbe: Fix SFP detection and linking on
> > > hotplug
> > >
> > > Currently the ixgbe driver does not ID any SFP except for the first
> > > one plugged in. This can lead to no-link, or incorrect speed conditions.
> > >
> > > For example:
> > >
> > > * If link is initially established with a 1G SFP, and later a 1G/10G
> > > multispeed part is later installed, then the MAC link setup
> > > functions are never called to change from 1000BASE-X to 10GBASE-R
> > > mode, and the link stays running at the slower rate.
> > >
> > > * If link is initially established with a 1G SFP, and later a 10G
> > > only module is later installed, no link is established, since we are
> > > still trasnsmitting in 1000BASE-X mode to a 10GBASE-R only partner.
> > >
> > > Refactor the SFP ID/setup, and link setup code, to more closely
> > > match the flow of the mainline kernel driver which does not have
> > > these issues.  In that driver a service task runs periodically to
> > > handle these operations based on bit flags that have been set
> > > (usually via interrupt or userspace request), and then get cleared
> > > once the requested subtask has been completed.
> > >
> > > Fixes: af75078fece ("first public release")
> > > Cc: stable at dpdk.org
> > >
> >
> > So BIG change for new platform, DON'T CC to stable!
> >
> > > Signed-off-by: Stephen Douthit <stephend at silicom-usa.com>
> > > Signed-off-by: Jeff Daly <jeffd at silicom-usa.com>
> > > ---
> > >  drivers/net/ixgbe/ixgbe_ethdev.c | 533
> > > +++++++++++++++++++++++--------  drivers/net/ixgbe/ixgbe_ethdev.h |
> > > 14 +-
> > >  2 files changed, 410 insertions(+), 137 deletions(-)
> > >
> > > diff --git a/drivers/net/ixgbe/ixgbe_ethdev.c
> > > b/drivers/net/ixgbe/ixgbe_ethdev.c
> > > index f31bbb7895..9e720eee47 100644
> > > --- a/drivers/net/ixgbe/ixgbe_ethdev.c
> > > +++ b/drivers/net/ixgbe/ixgbe_ethdev.c
> > > @@ -236,9 +236,6 @@ static int ixgbe_dev_interrupt_get_status(struct
> > > rte_eth_dev *dev);  static int ixgbe_dev_interrupt_action(struct
> > > rte_eth_dev *dev);  static void ixgbe_dev_interrupt_handler(void
> > > *param);  static void ixgbe_dev_interrupt_delayed_handler(void
> > > *param); -static void *ixgbe_dev_setup_link_thread_handler(void
> > > *param); -static int ixgbe_dev_wait_setup_link_complete(struct
> > rte_eth_dev *dev,
> > > -                                           uint32_t timeout_ms);
> > >
> >
> >
> > > +     /* TODO - Even for platforms where ixgbe_check_sfp_cage()
> > > + gives a
> > clear
> > > +      * status result, if there's no interrupts, or no interrupt for the SFP
> > > +      * cage present pin, even if other interrupts exist, then we still need
> > > +      * to poll here to set the flag.
> > > +      */
> > > +#ifndef RTE_EXEC_ENV_FREEBSD
> > > +     struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
> > > +     struct rte_intr_handle *intr_handle = pci_dev->intr_handle;
> > > +     if (rte_intr_allow_others(intr_handle)) {
> > > +             /* check if lsc interrupt is enabled */
> > > +             if (dev->data->dev_conf.intr_conf.lsc)
> > > +                     have_int = true;
> >
> >
> > > +     /* Check for loss of SFP */
> > > +     /* TODO - For platforms that don't have this flag, do we need to set
> > > +      *  NEED_SFP_SETUP on LSC if we're a SFP platform?
> > > +      */
> > > +     if (hw->mac.type ==  ixgbe_mac_X550EM_a &&
> > > +         (eicr & IXGBE_EICR_GPI_SDP0_X550EM_a))
> > > +             intr->flags |= IXGBE_FLAG_NEED_SFP_SETUP;
> > > +
> > >       return 0;
> > >  }
> >
> >
> > TODO ?
> >
> 
> These were notes that some further refinements could be possible.  I can
> remove the comments if that makes it more acceptable.  The first TODO was
> a note that (for FreeBSD mostly) w/o interrupts the code below (not shown,
> which checks for 'have_int') will run vs. exiting early and relying on the
> NEED_SFP_SETUP flag being set elsewhere.  The 2nd TODO was a question
> from the original developer to himself where (I believe) he was considering
> other implementations which may not use MOD_ABS the same that other
> platforms do.  I don't think this is an issue (that I know of) so removing this
> commend and just moving on would be the best thing, and if someone else
> knows of a situation where this consideration might need to be made, then
> another patch can be submitted to address it.
> 

Ugh.  Apologies for the dumb Outlook editor not wrapping the lines properly,
if you are reading this on patches.dpdk.org it runs off the page.  Repeating below:

These were notes that some further refinements could be possible.  I can 
remove the comments if that makes it more acceptable.  The first TODO was
a note that (for FreeBSD mostly) w/o interrupts the code below (not shown, 
which checks for 'have_int') will run vs. exiting early and relying on the 
NEED_SFP_SETUP flag being set elsewhere.  The 2nd TODO was a question 
from the original developer to himself where (I believe) he was considering
other implementations which may not use MOD_ABS the same that other
platforms do.  I don't think this is an issue (that I know of) so removing this
comment and just moving on would be the best thing, and if someone else
knows of a situation where this consideration might need to be made, then
another patch can be submitted to address it.

> > > --
> > > 2.25.1