Re: [PATCH v3] PCI: aardvark: Use LTSSM state to build link training flag

[Marc Zyngier]

On 2019-10-14 15:00, Remi Pommarel wrote:
> On Mon, Oct 14, 2019 at 02:45:34PM +0100, Marc Zyngier wrote:
> > Hi Remi,
> >
> > On 2019-10-14 14:06, Remi Pommarel wrote:
> > > Hi Lorenzo, Marc,

[...]

> > > Sure, I think this could be considered a fix for the following 
> > commit :
> > > Fixes: 8a3ebd8de328 ("PCI: aardvark: Implement emulated root PCI
> > > bridge config space")
> > >
> > > Moreover, Marc, I am also a bit supprised that you did not have to 
> > use
> > > [1] to even be able to boot.
> >
> > No, I don't have that one, and yet the system boots fine (although 
> > PCI
> > doesn't get much use on this box). I guess I'm lucky...
> >
> > > Also if you want to be completely immune to this kind of SError 
> > (that
> > > could theoretically happen if the link goes down for other reasons 
> > than
> > > being retrained) you would have to use mainline ATF along with 
> > [2]. But
> > > the chances to hit that are low (could only happen in case of link
> > > errors).
> >
> > Now you've got me worried. Can you point me to that ATF patch? I'm 
> > quite
> > curious as to how you recover from an SError on a v8.0 CPU given 
> > that it
> > has no syndrome information and may as well signal "CPU on fire!"...
>
> The patch is at [1]. Please note that this is done quite similarly 
> for
> rcar.
>
> [1] 
> https://review.trustedfirmware.org/c/TF-A/trusted-firmware-a/+/1541

That patch, without any other information, looks quite flaky. Unless 
there
is a strong guarantee that ESR_EL3.ISS==2 only when the PCIe controller
goes wrong, it looks like this only papers over the issue...

That's pretty much independent from the patch at hand in this thread, 
but
I certainly wouldn't trust this ATF patch without some more information
about how the fault is reported to the CPU.

        M.
--
Jazz is not dead. It just smells funny...