Re: [PATCH] PCI: Always lift 2.5GT/s restriction in PCIe failed link retraining

[Ilpo Järvinen]

On Mon, 1 Dec 2025, Maciej W. Rozycki wrote:

> On Mon, 1 Dec 2025, Ilpo Järvinen wrote:
> 
> > > +	pcie_capability_read_dword(dev, PCI_EXP_LNKCAP, &lnkcap);
> > > +	if ((lnkctl2 & PCI_EXP_LNKCTL2_TLS) == PCI_EXP_LNKCTL2_TLS_2_5GT &&
> > > +	    (lnkcap & PCI_EXP_LNKCAP_SLS) != PCI_EXP_LNKCAP_SLS_2_5GB) {
> > 
> > I'm trying to recall, if there was some particular reason why 
> > ->supported_speeds couldn't be used in this function. It would avoid the 
> > need to read LinkCap at all.
> 
>  Thanks for the hint.  There's probably none and it's just me missing some 
> of the zillion bits and pieces.  I'll wait a couple of days for any other 
> people to chime in and respin with this update included if everyone is 
> otherwise happy to proceed with this update.
> 
> > > +		if (ret)
> > > +			goto err;
> > >  	}
> > >  
> > >  	return ret;
> > 
> > return 0;
> 
>  It can still return -ENOTTY if neither of the two latter conditionals 
> matched, meaning the quirk was not applicable after all.  ISTR you had 
> issues with the structure of this code before; I am not sure if it can 
> be made any better in a reasonable way.  It is not a failure per se, so 
> the newly-added common error path does not apply.  This is the case for: 
> "Return an error if retraining was not needed[...]" from the introductory 
> comment.
> 
>  Shall I add a comment above the return statement referring to this?

I think it's fine as is, I just didn't review with enough context to 
notice what it was initialized to (the usual thing when adding a 
rollback path is to forget to change the normal path to return 0, thus 
"auto commenting" it without checking enough, I'm sorry about that).

-- 
 i.