Re: RFC: arm64: Handling reserved memory nodes

[Leo Yan <leo.yan@xxxxxxxxxx>]

Hi Julien,

On Wed, Sep 20, 2023 at 11:36:24AM +0100, Julien Grall wrote:

[...]

> > Host dt:
> > memory@40000000 {
> >      reg = <0x00 0x40000000 0x01 0x00>;
> >      device_type = "memory";
> > };
> > 
> > reserved-memory {
> >      #size-cells = <0x02>;
> >      #address-cells = <0x02>;
> >      ranges;
> > 
> >      test@50000000 {
> >          reg = <0x00 0x50000000 0x00 0x10000000>;
> >          no-map;
> >      };
> > };
> > 
> > Xen:
> > (XEN) MODULE[0]: 000000004ac00000 - 000000004ad65000 Xen
> > (XEN) MODULE[1]: 000000004ae00000 - 000000004ae03000 Device Tree
> > (XEN) MODULE[2]: 0000000042c00000 - 000000004aa8ea8b Ramdisk
> > (XEN) MODULE[3]: 0000000040400000 - 0000000042b30000 Kernel
> > (XEN)  RESVD[0]: 0000000050000000 - 000000005fffffff
> > ...
> > (XEN) BANK[0] 0x000000c0000000-0x00000100000000 (1024MB)
> > 
> > Linux dom0:
> > [    0.000000] OF: reserved mem: 0x0000000050000000..0x000000005fffffff 
> > (262144 KiB) nomap non-reusable test@50000000
> 
> So Linux should tell whether a region has been reserved. @Leo, can you share
> with us the serial console? Can you confirm the version of Xen you are
> using?

Finally, I located the issue is caused by the reserved memory node
with "disabled" status. In the end, it mismatches for handling the
disabled memory nodes between the Xen hypervisor and the Linux kernel.

I think I should set the status property from the reserved memory
nodes for my debugging platform. But for the case that the memory
nodes (for both normal and reserved nodes) which are disabled, the Xen
hypervisor should ignore them. So I sent a new patch to address the
issue for disabled memory nodes:

https://lore.kernel.org/xen-devel/20230926053333.180811-1-leo.yan@xxxxxxxxxx/T/#u

Hope now we are clear for the issue. If not, please let me know.

Thanks,
Leo