Re: [PATCH net-next v7 1/3] net: dsa: microchip: implement KSZ87xx Module 3 low-loss cable errata

[Fidelio LAWSON <lawson.fidelio@gmail.com>]

Hi Nicolai,

Thanks again for the detailed testing and feedback.

On 5/28/26 13:43, Nicolai Buchwitz wrote:
> 
> The cached state lives on struct ksz_device, but ksz8_r_phy() is
> invoked per port. Setting the preset on one port is reported back as
> active on the others:
> 
>    phytool write swp1/1/0x1a 1
>    phytool read  swp1/2/0x1a -> 0x0001
>    phytool read  swp1/3/0x1a -> 0x0001
> 
> The commit message already notes that the erratum applies switch-wide.
> So writes from non-primary ports could either be rejected or at least 
> produce
> a dev_info_once() so userspace can see the setting is shared?
> 

I will add a dev_info_once() to inform userspace that the configuration 
is shared across all ports.
> 
> dev->eq_init starts at zero from the kzalloc, but the header added by
> this patch defines KSZ87XX_DSP_EQ_INIT_FACTORY = 0x0F as the hardware
> default. After a fresh boot on my setup (no writes), the read returns 0:
> 
>    phytool read swp1/1/0x1c -> 0x0000
> 
> Should probe set dev->eq_init = KSZ87XX_DSP_EQ_INIT_FACTORY (and
> dev->lpf_bw = KSZ87XX_PHY_LPF_90MHZ), or read the indirect registers
> once on probe, so the cache reflects what the hardware actually holds?
>
Good point, I will fix this by setting dev->eq_init = 
KSZ87XX_DSP_EQ_INIT_FACTORY and dev->lpf_bw = KSZ87XX_PHY_LPF_90MHZ 
during probe.

Thanks,
Fidelio