Re: [RFC PATCH net-next 0/8] Convert xpcs to phylink_pcs_ops

[Vladimir Oltean <olteanv@gmail.com>]

Hi VK,

On Mon, May 31, 2021 at 10:30:19AM +0800, Wong Vee Khee wrote:
> On Fri, May 28, 2021 at 12:12:30PM +0300, Vladimir Oltean wrote:
> > Hi VK,
> > 
> > On Fri, May 28, 2021 at 10:15:21AM +0800, Wong Vee Khee wrote:
> > > I got the following kernel panic after applying [1], and followed by
> > > this patch series.
> > > 
> > > [1] https://patchwork.kernel.org/project/netdevbpf/patch/20210527155959.3270478-1-olteanv@gmail.com/
> > > 
> > > [   10.742057] libphy: stmmac: probed
> > > [   10.750396] mdio_bus stmmac-1:01: attached PHY driver [unbound] (mii_bus:phy_addr=stmmac-1:01, irq=POLL)
> > > [   10.818222] intel-eth-pci 0000:00:1e.4 (unnamed net_device) (uninitialized): failed to validate link configuration for in-band status
> > > [   10.830348] intel-eth-pci 0000:00:1e.4 (unnamed net_device) (uninitialized): failed to setup phy (-22)
> > 
> > Thanks a lot for testing. Sadly I can't figure out what is the mistake.
> > Could you please add this debugging patch on top and let me know what it
> > prints?
> > 
> 
> Sorry for the late response. Here the debug log:
> 
> [   11.474302] mdio_bus stmmac-1:01: attached PHY driver [unbound] (mii_bus:phy_addr=stmmac-1:01, irq=POLL)
> [   11.495564] mdio_bus stmmac-1:16: xpcs_create: xpcs_id 7996ced0 matched on entry 0
> [   11.503154] mdio_bus stmmac-1:16: xpcs_create: setting entry->supported bit 13
> [   11.510377] mdio_bus stmmac-1:16: xpcs_create: setting entry->supported bit 14
> [   11.517590] mdio_bus stmmac-1:16: xpcs_create: setting entry->supported bit 6
> [   11.524725] mdio_bus stmmac-1:16: xpcs_create: setting entry->supported bit 17
> [   11.531946] mdio_bus stmmac-1:16: xpcs_create: setting entry->supported bit 18
> [   11.539278] mdio_bus stmmac-1:16: xpcs_create: setting entry->supported bit 19
> [   11.541316] ish-hid {33AECD58-B679-4E54-9BD9-A04D34F0C226}: [hid-ish]: enum_devices_done OK, num_hid_devices=6
> [   11.546487] mdio_bus stmmac-1:16: xpcs_create: setting entry->supported bit 15
> [   11.546489] mdio_bus stmmac-1:16: xpcs_create: xpcs->supported 0000000,00000000,000ee040
> [   11.584687] hid-generic 001F:8087:0AC2.0001: device has no listeners, quitting
> [   11.599461] mdio_bus stmmac-1:16: xpcs_validate: provided interface sgmii does not match supported interface 0 (usxgmii)
> [   11.606538] hid-generic 001F:8087:0AC2.0002: device has no listeners, quitting
> [   11.610306] mdio_bus stmmac-1:16: xpcs_validate: provided interface sgmii does not match any supported interface
> [   11.610309] mdio_bus stmmac-1:16: xpcs_validate: provided interface sgmii does not match supported interface 0 (usxgmii)
> [   11.626259] hid-generic 001F:8087:0AC2.0003: device has no listeners, quitting
> [   11.627675] mdio_bus stmmac-1:16: xpcs_validate: provided interface sgmii does not match any supported interface
> [   11.627677] intel-eth-pci 0000:00:1e.4 (unnamed net_device) (uninitialized): failed to validate link configuration for in-band status
> [   11.641996] hid-generic 001F:8087:0AC2.0004: device has no listeners, quitting
> [   11.645729] intel-eth-pci 0000:00:1e.4 (unnamed net_device) (uninitialized): failed to setup phy (-22)

Ha ha, this works as expected, but I was led into error due to the code
structure.

See, everything in pcs-xpcs.c is laid out as if there are different PHY
IDs for SGMII, USXGMII etc. But if you pay close attention, they are all
equal to 0x7996ced0:

#define SYNOPSYS_XPCS_USXGMII_ID	0x7996ced0
#define SYNOPSYS_XPCS_10GKR_ID		0x7996ced0
#define SYNOPSYS_XPCS_XLGMII_ID		0x7996ced0
#define SYNOPSYS_XPCS_SGMII_ID		0x7996ced0
#define SYNOPSYS_XPCS_MASK		0xffffffff

With the old code, it works because the probing code gets a nudge from
the caller of xpcs_probe by being told what is the expected phy_interface_t.
The xpcs then uses the phy_interface_t _as_part_of_ the PHY ID matching
sequence.

So.. yeah. I got the information I needed. I will come back with a way
for the same PCS PHY ID to support multiple PHY interface types.

Thanks again for testing.