Re: [PATCH net-next] octeontx2-pf: report port connector type in ethtool link ksettings

[netdev-bot+sashiko]

Thank you for your contribution! Sashiko AI review found 1 potential
issue(s) to consider:

Critical: 0 · High: 0 · Medium: 1 · Low: 0

- [Medium] otx2_get_link_ksettings()
  (drivers/net/ethernet/marvell/octeontx2/nic/otx2_ethtool.c:1235)…

--

octeontx2-pf: report port connector type in ethtool link ksettings

This adds a single assignment to otx2_get_link_ksettings() so that
cmd->base.port is filled from the CGX firmware data word
rsp->fwdata.port, with the intent of letting ethtool display the
physical connector type.

> diff --git a/drivers/net/ethernet/marvell/octeontx2/nic/otx2_ethtool.c b/drivers/net/ethernet/marvell/octeontx2/nic/otx2_ethtool.c
> index 9bee1b91eeaa6..65630de435397 100644
> --- a/drivers/net/ethernet/marvell/octeontx2/nic/otx2_ethtool.c
> +++ b/drivers/net/ethernet/marvell/octeontx2/nic/otx2_ethtool.c
> @@ -1231,6 +1231,8 @@ static int otx2_get_link_ksettings(struct net_device *netdev,
>  				OTX2_MODE_SUPPORTED, cmd);
>  	otx2_get_fec_info(rsp->fwdata.supported_fec,
>  			  OTX2_MODE_SUPPORTED, cmd);
> +
> +	cmd->base.port = rsp->fwdata.port;

[Severity: Medium]

Should this raw firmware word be translated before it is exported as the
ethtool connector type?

The source field is a 64-bit value in af/mbox.h:

struct cgx_lmac_fwdata_s {
	...
	u32 lmac_type;
	u32 portm_idx;
	u64 mgmt_port:1;
	u64 advertised_an:1;
	u64 port;
	...
};

The destination is the __u8 uAPI field in struct ethtool_link_settings,
documented as "@port: Physical connector type; one of %PORT_*", whose value
space is sparse and exhaustive (PORT_TP 0x00 .. PORT_DA 0x05, PORT_NONE
0xef, PORT_OTHER 0xff).

So the assignment silently narrows to 8 bits, and a firmware value of, say,
0x100 becomes 0x00, which userspace prints as Twisted Pair.  A value in
0x06..0xee is not a legal PORT_* code at all and ethtool prints "Unknown!".
The same byte then becomes the base value for ethnl_update_u8(&lsettings->port, ...)
on the netlink set path.

Would a switch or table mapping the firmware encoding to PORT_*, with
PORT_OTHER for anything unrecognised, be preferable here?

Every other consumer of the same firmware structure in this function does
translate: otx2_get_link_mode_info() maps CGX mode bits to
ETHTOOL_LINK_MODE_* bits, otx2_get_fec_info() maps OTX2_FEC_* to
ETHTOOL_LINK_MODE_FEC_*, and otx2_get_fecparam() bounds-checks against
OTX2_FEC_MAX_INDEX.

Is there anything that tells the driver whether the firmware populated this
word at all?

The field was carved out of the reserved tail of struct cgx_lmac_fwdata_s by
commit 49f02e6877d1be ("Octeontx2-af: Debugfs support for firmware data",
LMAC_FWDATA_RESERVED_MEM 1021 -> 1018) without bumping RVU_FWDATA_VERSION,
which is still 0x0001 in af/rvu.h.  The only check applied to the ATF region
is the outer header check:

static inline bool is_rvu_fwdata_valid(struct rvu *rvu)
{
	return (rvu->fwdata->header_magic == RVU_FWDATA_HEADER_MAGIC) &&
		(rvu->fwdata->version == RVU_FWDATA_VERSION);
}

and rvu_mbox_handler_cgx_get_aux_link_info() copies the region verbatim:

		memcpy(&rsp->fwdata,
		       &rvu->fwdata->cgx_fw_data[cgx_id][lmac_id],
		       sizeof(struct cgx_lmac_fwdata_s));

The struct's own rw_valid indicator is not consulted by the new code.  On
firmware that does not fill this word it reads 0, which is exactly PORT_TP,
so a fibre or DAC CGX/RPM SERDES port would be reported as Twisted Pair
rather than left unspecified.  Could the assignment be gated on
fwdata.rw_valid, or on a bumped firmware-data version, and fall back to
PORT_OTHER or PORT_NONE?

One more question about the commit message:

	Populate cmd->base.port from rsp->fwdata.port so users can see values
	such as PORT_TP or PORT_AUI via ethtool.

This states the firmware supplies PORT_* values, but a grep of
drivers/net/ethernet/marvell/octeontx2/ shows this new line is the only
reference to fwdata.port in the tree, and there is no firmware-encoding to
PORT_* table anywhere.  Can the firmware encoding, and the firmware version
that started populating the field, be documented in the changelog?

-- 
Sashiko AI review · https://netdev-ai.bots.linux.dev/sashiko/#/patchset/20260901045857.2779326-1-rkannoth%40marvell.com