Re: [RFC PATCH] wifi: rt2x00: use VCO_IC 0x2 for RF3053 on 2.4 GHz

[Stanislaw Gruszka]

On Tue, Oct 13, 2026 at 08:49:44PM +0800, Jie Feng wrote:
> On Tue, 13 Oct 2026 at 18:23, Stanislaw Gruszka <stf_xl@xxxxx> wrote:
> >
> > Hi
> >
> > On Mon, Oct 12, 2026 at 09:37:27PM +0800, Ho Jie Feng wrote:
> > > Using the 0x1, the VCO calibration result is 0x7f on channels 6
> > > and below. Using 0x2 fixes it.
> > >
> > > Hardware: Netgear WNDA4100 (USB 0846:9013, RT3593 rev 0402, RF3053)
> > >
> > > Signed-off-by: Ho Jie Feng <hjf3108@xxxxxxxxx>
> > > ---
> > > RFC for this since I am not sure if it is a generic fix, since the
> > > vendor driver also doesn't use VCO_IC=0x2.

I looked at vendor driver it does not set VCO_IC=0x2 when switch 
a channel. However it does it after wake up from suspend:
https://github.com/ashaffer/rt3573sta/blob/09b80e6290769c53736dc09c5ae9025e15bfbe69/chips/rt3593.c#L194

This function: RT3593ReverseRFSleepModeSetup seams to be partially
the same what NICInitRT3593RFRegisters does. 

Also NICInitRT3593RFRegisters it's different from
rt2800_init_rfcsr_3593() i.e. it set RF_R02 before GPIO_SWITCH.

Could you check if modifying rt2800_init_rfcsr_3593() 
to be more the same with both NICInitRT3593RFRegistersr and
RT3593ReverseRFSleepModeSetup (regarding RF_R06 change)
solve the problem?

> > > Is this also a known issue?
I didn't knew about it.

> > Could you elaborate what is the issue and what you mean by
> > "VCO calibration result is 0x7f" ?'
> I had accidentally left the description out, sorry about that.
No problem.

> The adapter doesn't work on the lower channels in 2.4GHz,
> channels 8 and below. This looked like an issue with
> the PLL or something after debugging.
> 
> By tuning to channels 1-12 and using the debugfs to dump the
> RFCSR registers, RFCSR4 is always 0x7f on the failing channels
> consistently. Since there is code to switch RFCSR6 based on
> channel, I tried the same as the 5GHz code to switch RFCSR6=0x2
> for lower channels and it worked on all the channels surprisingly.
> 
> Due to the behaviour of RFCSR4, I think it is a VCO calibration
> result since it seems to change after enabling the calibration.
> 
> I could not find any mention of this issue elsewhere, and suspect
> either this hardware has very few Linux users or my hardware
> is special in some way, so I can't guarantee that setting
> it to 0x2 will work with another device even if it is the
> same WNDA4100 model and chipset, so some testing would
> be appreciated.
I do not have RT3573/RT3593 device by myself and ordinary users
do not read linux-wireless mailing list, so we have rather little
chance to get more test coverage here.

Anyways, if init change that would make rt2800_init_rfcsr_3593()
more like vendor driver do not fix the problem, it's ok to modify
rt2800_config_channel_rf3053(). 

If we want to be more safe with the change, we can limit it with
rt2x00_rt_rev() using 'rev' of your tested device.

Regards
Stanislaw