Re: [PATCH v9 0/5] arm64: ptdump: View the second stage page-tables

[Sebastian Ene <sebastianene@google.com>]

On Mon, Sep 02, 2024 at 06:11:04AM +0000, Sebastian Ene wrote:
> On Fri, Aug 30, 2024 at 04:00:11PM +0100, Marc Zyngier wrote:
> > On Fri, 30 Aug 2024 15:44:39 +0100,
> > Marc Zyngier <maz@kernel.org> wrote:

Hello Marc,


I tried with a 16Kb host under VHE on Qemu starting a 4kb guest with
kvmtool:

root@seb-qemu-pkvm-host:~#  zcat /proc/config.gz | grep
"CONFIG_ARM64_[164K]*_PAGES=y"
CONFIG_ARM64_16K_PAGES=y
root@seb-qemu-pkvm-host:~# cat
/sys/kernel/debug/kvm/263-4/stage2*
2
---[ Guest IPA ]---
0x0000000000000000-0x0000000001020000       16512K 3
0x0000000001020000-0x0000000001024000          16K 3   R W X AF    
0x0000000001024000-0x0000000002000000       16240K 3
0x0000000002000000-0x0000000080000000        2016M 2
0x0000000080000000-0x0000000084000000          64M 2   R W   AF BLK
0x0000000084000000-0x000000008e000000         160M 2
0x000000008e000000-0x0000000090000000          32M 2   R W   AF BLK
0x0000000090000000-0x0000000098000000         128M 2
0x0000000098000000-0x000000009a000000          32M 2   R W X AF BLK
0x000000009a000000-0x000000009c000000          32M 2   R W   AF BLK

This looks quite right I guess and I wonder how can I repro what you are seeing ?
What kvm-arm.mode is the host running into ?

Thanks,
Seb

> 
> Hello Marc,
> 
> > > 
> > > Hi Seb,
> > 
> > [...]
> > 
> > > I've been giving this a go on my test systems with 16k pages, and it
> > > doesn't really work as advertised:
> > > 
> > > root@babette:/sys/kernel/debug/kvm# cat 2573-13/stage2_*
> > > 2
> > > ---[ Guest IPA ]---
> > > 0x0000000000000000-0x0000000008000000         128M 
> > > 0x0000000008000000-0x00000000090a0000       17024K 3
> > > 0x00000000090a0000-0x00000000090a4000          16K 3   R W X AF    
> > > 0x00000000090a4000-0x000000000a000000       15728K 3
> > > 
> > > Only 16kB mapped? This is a full Linux guest running the Debian
> > > installer, and just the kernel is about 20MB (the VM has 4GB of RAM,
> > > and is using QEMU as the VMM)
> > > 
> > > So clearly something isn't playing as expected. Also, this '128M'
> > > without a level being displayed makes me wonder. It is probably the
> > > QEMU flash, but then the rest of the addresses don't make much sense
> > > (RAM on QEMU is at 1GB, not at 128MB.
> > > 
> > > On another system with kvmtool, I get something similar:
> > > 
> > > root@duodenum:/home/maz# cat /sys/kernel/debug/kvm/*/stage2_*
> > > 2
> > > ---[ Guest IPA ]---
> > > 0x0000000000000000-0x0000000001020000       16512K 3
> > > 0x0000000001020000-0x0000000001024000          16K 3   R W X AF    
> > > 0x0000000001024000-0x0000000002000000       16240K 3
> > > 
> > > and kvmtool places the RAM at 2GB. Clearly not what we're seeing here.
> > > 
> > > Could you please verify this?
> 
> Ughh, this doesn't look right. I will give it a spin with a different
> granule, thanks for bringing me to attention. I will look first at
> mm/ptdump.c if it works as intended.
>  
> 
> > 
> > For the record, on a 4kB host, I get much more plausible results:
> > 
> > root@big-leg-emma:/home/maz# cat /sys/kernel/debug/kvm/632-12/stage2_*
> > 3
> > ---[ Guest IPA ]---
> > 0x0000000000000000-0x0000000000200000           2M 2   R     AF BLK
> > 0x0000000000200000-0x0000000040000000        1022M 2
> > 0x0000000040000000-0x0000000040200000           2M 2   R W X AF BLK
> > 0x0000000040200000-0x0000000044000000          62M 2
> > 0x0000000044000000-0x0000000044200000           2M 2   R W X AF BLK
> > 0x0000000044200000-0x0000000047600000          52M 2
> > 0x0000000047600000-0x0000000047800000           2M 2   R W   AF BLK
> > 0x0000000047800000-0x0000000047e00000           6M 2   R W X AF BLK
> > 0x0000000047e00000-0x0000000048000000           2M 2   R W   AF BLK
> > 0x0000000048000000-0x00000000b9c00000        1820M 2
> > 0x00000000b9c00000-0x00000000b9e00000           2M 2   R W X AF BLK
> > 0x00000000b9e00000-0x00000000bb800000          26M 2
> > 0x00000000bb800000-0x00000000bba00000           2M 2   R W X AF BLK
> > 0x00000000bba00000-0x00000000bbe00000           4M 2   R W   AF BLK
> > 0x00000000bbe00000-0x00000000bc200000           4M 2   R W X AF BLK
> > 0x00000000bc200000-0x00000000bc800000           6M 2   R W   AF BLK
> > 0x00000000bc800000-0x00000000be400000          28M 2
> > 0x00000000be400000-0x00000000bf800000          20M 2   R W X AF BLK
> > 0x00000000bf800000-0x00000000bfe00000           6M 2   R W   AF BLK
> > 0x00000000bfe00000-0x00000000c0000000           2M 2   R W X AF BLK
> > 
> > So 16kB is the one that needs investigating, and I strongly suspect
> > that 64kB is in the same boat...
> > 
> > Thanks,
> > 
> > 	M. (signing off for the day)
> > 
> 
> Thanks,
> Sebastian
> 
> > -- 
> > Without deviation from the norm, progress is not possible.