Re: [RFC Patch v1 0/3] Fix using wrong GPF DVSEC location issue

[Jonathan Cameron]

On Fri, 21 Mar 2025 14:55:42 +0800
Li Ming <ming.li@xxxxxxxxxxxx> wrote:

> On 3/21/2025 11:59 AM, Davidlohr Bueso wrote:
> > On Thu, 20 Mar 2025, Davidlohr Bueso wrote:
> >  
> >> On Wed, 19 Mar 2025, Li Ming wrote:
> >>  
> >>> But I am not sure if all dports under a same port will have same
> >>> configuration space layout, if yes, that will not be a problem. If I am
> >>> wrong, please let me know, thanks.  
> >>
> >> Yes, when caching the dvsec was suggested, it was my assumption that the
> >> config space would be the same.  
> >
> > Ultimately I don't know what the expectation is here, but your updates
> > do allow more flexibility from vendors, I guess(?). It's a bit late
> > in the cycle, unfortunately, so if these are to go in for v6.15, they
> > would be considered a fix imo, otherwise perhaps they are wanted for
> > v6.16 or not at all (patch 3 does look useful regardless)?  
> 
> My understanding is that the expectation of the patchset is to avoid using a wrong GPF DVSEC in case of dports under a same port have different config space layout. And I think the change is more closely to the description of CXL spec.
> 
> If the case(dports under a same port have different config space layout) would not happen, maybe add a comment in cxl_gpf_port_setup() is another option.
> 
> Yes, if patch 1 & 2 are considered to be merged, they are worth a fix tag. And patch 3 is an obvious cleanup change.

I think they can indeed have different layout (in theory).
Seems moderately unlikely to occur in real devices, but you never know.

So I think a fixes tag would be valid.

Jonathan

> 
> >
> > Based on some of the topologies listed in qemu, I did some testing (and
> > this was also why the same dvsec config layout) and see things working as
> > expected.  
> 
> Thanks for testing.
> 
> 
> Ming
> 
> [snip]
>