Re: [RESEND] [RFC PATCH] xen/arm: domain_build: Ignore empty memory bank

[Stefano Stabellini <sstabellini@xxxxxxxxxx>]

On Mon, 4 Jan 2021, Elliott Mitchell wrote:
> On Mon, Dec 21, 2020 at 06:28:35PM +0000, Julien Grall wrote:
> > On 21/12/2020 17:30, Elliott Mitchell wrote:
> > > I doubt this is the only bug exposed by
> > > 5a37207df52066efefe419c677b089a654d37afc.
> > 
> > Are you saying that with my patch dropped, Xen will boot but with it 
> > will not?
> 
> I thought that was the cause.  Yet after a bunch of builds trying to
> ensure I can cause it to reproduce or not, I wasn't able to.  As such I
> now think this is a misattribution.  :-(
> 
> Other candidate on my radar is this showed up near the time I started
> trying other kernel sources.  I now wonder if this is due to the
> device-trees being produced with recent RPF kernels versus those being
> produced with pure mainline.  Presently I'm using a 5.10 RPF kernel and
> device-tree.
> 
> 
> > So I think we first need to figure out what is the offending node and 
> > why it is dt_device_get_address() is returning an error for it.
> > 
> > That said, I agree that we possibly want a check size == 0 (action TBD) 
> > in map_range_to_domain() as the code would do the wrong thing for 0.
> 
> Already stated "/scb/pcie@7d500000/pci@1,0/usb@1,0".

Can you please post the full node for usb@1,0? I would like to check the
corresponding device tree binding to see the expected format.


> Perhaps the code should be ignoring nodes for which
> which dt_device_get_address() fails, or perhaps this should only be done
> for Domain 0 (where it results in panic).

That seems reasonable