Re: [PATCH 1/3] xen/ppc: Enable Boot Allocator

[Julien Grall <julien@xxxxxxx>]

Hi Shawn,

On 15/12/2023 02:50, Shawn Anastasio wrote:
> Hi Julien,
>
> Thank you for the feedback. Most of your points will be addressed by
> following your suggestion of moving ARM's bootfdt.c to common, but I
> wanted to respond with a few points of clarification.
>
> On 12/1/23 5:23 PM, Julien Grall wrote:
> > * fdt_get_mem_rsv_paddr(), this is part of the DT is used to reserve
> > memory. This was superseed to /reserved-memory, but I wonder how
> > widespread this is on PPC?
> As far as I can tell, the DT reserve memory map is not used on
> PPC/PowerNV. This information is instead communicated through
> `reserved-memory` nodes in the DT itself, which the existing code
> handles.
On Arm, there is a mix between using /memreserve (found by Xen via 
fdt_get_mem_rsv_paddr() and reserved-memory.
Looking at Linux, it seems that it will try to parse /memreserve in PPC 
and I have found a few DT using it:
42sh> ack memreserve
boot/dts/currituck.dts
13:/memreserve/ 0x01f00000 0x00100000;  // spin table

boot/dts/akebono.dts
14:/memreserve/ 0x01f00000 0x00100000;  // spin table

boot/dts/iss4xx-mpic.dts
17:/memreserve/ 0x01f00000 0x00100000;

boot/dts/mpc836x_mds.dts
10:/memreserve/ 00000000 1000000;

boot/dts/wii.dts
17: * contiguous RAM range and will BUG() if the memreserve is outside
20:/*/memreserve/ 0x10000000 0x0004000;*/       /* DSP RAM */

I guess we are not targetting any of those hardwares. Even if this is 
the case, I don't think we can simply ignore the node. If you don't want 
to handle them, then I would at least suggest to add a check in Xen to 
confirm the /memreserve/ is empty.
Better to catch such DT early rather than debugging a memory corruption 
later on :).
Cheers,

--
Julien Grall