Re: [PATCH net-next v2 2/3] bnxt_en: add .get_module_eeprom_by_page() support

[Ido Schimmel <idosch@idosch.org>]

On Sun, Oct 02, 2022 at 09:51:10PM +0530, Vikas Gupta wrote:
> On Sun, Oct 2, 2022 at 9:04 PM Ido Schimmel <idosch@idosch.org> wrote:
> >
> > On Sat, Oct 01, 2022 at 02:27:10PM -0400, Michael Chan wrote:
> > > +static int bnxt_get_module_eeprom_by_page(struct net_device *dev,
> > > +                                       const struct ethtool_module_eeprom *page_data,
> > > +                                       struct netlink_ext_ack *extack)
> > > +{
> > > +     struct bnxt *bp = netdev_priv(dev);
> > > +     int rc;
> > > +
> > > +     if (bp->link_info.module_status >
> > > +         PORT_PHY_QCFG_RESP_MODULE_STATUS_WARNINGMSG) {
> > > +             NL_SET_ERR_MSG_MOD(extack, "Phy status unknown");
> >
> > Can you make this more helpful to users? The comment above this check in
> > bnxt_get_module_info() suggests that it is possible:
> 
> Do you mean that we should elaborate something like
> NL_SET_ERR_MSG_MOD(extack, "Module may be not inserted or powered down
> or 10G Base-T");

Yes, but even then the exact error is unknown and you would need
something like drgn / kprobes to retrieve the specific module_state for
debug. You can do something like the following (in a separate function):

if (bp->link_info.module_status <=
    PORT_PHY_QCFG_RESP_MODULE_STATUS_WARNINGMSG)
        return 0;

switch (bp->link_info.module_status) {
case PORT_PHY_QCFG_RESP_MODULE_STATUS_PWRDOWN:
	NL_SET_ERR_MSG_MOD(extack, "Transceiver module is powering down");
	break;
case PORT_PHY_QCFG_RESP_MODULE_STATUS_NOTINSERTED:
	NL_SET_ERR_MSG_MOD(extack, "Transceiver module not inserted");
	break;
case PORT_PHY_QCFG_RESP_MODULE_STATUS_CURRENTFAULT:
	NL_SET_ERR_MSG_MOD(extack, "... something that explains what this means ...");
	break;
default:
	NL_SET_ERR_MSG_MOD(extack, "Unknown error");
	break;
}

return -EINVAL;

> 
> >
> > /* No point in going further if phy status indicates
> >  * module is not inserted or if it is powered down or
> >  * if it is of type 10GBase-T
> >  */
> > if (bp->link_info.module_status >
> >         PORT_PHY_QCFG_RESP_MODULE_STATUS_WARNINGMSG)
> >
> > > +             return -EIO;
> > > +     }
> > > +
> > > +     if (bp->hwrm_spec_code < 0x10202) {
> > > +             NL_SET_ERR_MSG_MOD(extack, "Unsupported hwrm spec");
> >
> > Likewise. As a user I do not know what "hwrm spec" means... Maybe:
> >
> > NL_SET_ERR_MSG_MOD(extack, "Firmware version too old");
> >
> >
> > > +             return -EOPNOTSUPP;
> > > +     }
> > > +
> > > +     if (page_data->bank && !(bp->phy_flags & BNXT_PHY_FL_BANK_SEL)) {
> > > +             NL_SET_ERR_MSG_MOD(extack, "Firmware not capable for bank selection");
> > > +             return -EOPNOTSUPP;
> >
> > What happens if you have an old firmware that does not support this
> > functionality and user space tries to dump page 10h from bank 1 of a
> > CMIS module that supports multiple banks?
> >
> > I wanted to say that you would see the wrong information (from bank 0)
> > because the legacy operations do not support banks and bank 0 is
> > assumed. However, because only pages 10h-ffh are banked, user space will
> > get an error from the following check in fallback_set_params():
> >
> > if (request->page)
> >         offset = request->page * ETH_MODULE_EEPROM_PAGE_LEN + offset;
> >
> > [...]
> >
> > if (offset >= modinfo->eeprom_len)
> >         return -EINVAL;
> >
> > I believe it makes sense to be more explicit about it and return an
> > error in fallback_set_params() in case bank is not 0. Please follow up
> > if the above analysis is correct.
> 
> So older firmware do not understand bank > 0 and hence it returns to
> EOPNOTSUPP, which goes to fallback_set_params() and fails for
> if (offset >= modinfo->eeprom_len)
>         return -EINVAL
> As we are not setting modinfo->eeprom_len for CMIS modules in get_module_eeprom.
> With the above said userspace gets EINVAL.
> Let me know if my understanding is correct?

Yes. Basically there is no point for ethtool to even try to invoke the
legacy operations when bank is not zero:

diff --git a/net/ethtool/eeprom.c b/net/ethtool/eeprom.c
index 1c94bb8ea03f..1d6a35c8b6f0 100644
--- a/net/ethtool/eeprom.c
+++ b/net/ethtool/eeprom.c
@@ -60,6 +60,9 @@ static int eeprom_fallback(struct eeprom_req_info *request,
 	u8 *data;
 	int err;
 
+	if (request->bank)
+		return -EINVAL;
+
 	modinfo.cmd = ETHTOOL_GMODULEINFO;
 	err = ethtool_get_module_info_call(dev, &modinfo);
 	if (err < 0)

Not sure how many will actually hit it. I expect drivers supporting
modules with banked pages to implement the new ethtool operation.