Re: Regression booting Linux kernel 5.16.1 under Xen: Failed to update kernel mapping for mfn=6930582 pfn=524288

[Sander Eikelenboom <linux@xxxxxxxxxxxxxx>]

On 18/01/2022 12:13, Juergen Gross wrote:
> On 18.01.22 11:53, Sander Eikelenboom wrote:
> > L.S.,
> >
> > Both Linux kernel 5.16.0 and 5.16.1 fail to boot as Dom0 under
> > xen-unstable and crash early in boot on my hardware.
> > With Linux 5.15.13 it boots fine. Serial log is below.
> ...
>
> > (XEN) [0000001b1ffedeb8] Command line: dom0_mem=2048M,max:2048M
> > loglvl=all guest_loglvl=all console_timestamps=datems
> > vga=gfx-1280x1024x32 no-cpuidle com1=38400,8n1 console=vga,com1
> > ivrs_ioapic[6]=00:14.0 iommu=on,verbose,debug conring_size=128k
> > ucode=scan sched=credit2 gnttab_max_frames=64 reboot=a
> ...
>
> > mapping kernel into physical memory
> > about to get started...
> > [    0.000000] Linux version 5.16.1-20220118-doflr-mac80211debug+
> > (root@serveerstertje) (gcc (Debian 8.3.0-6) 8.3.0, GNU ld (GNU Binutils
> > for Debian) 2.31.1) #1 SMP PREEMPT Tue Jan 18 10:49:09 CET 2022
> > [    0.000000] Command line: root=/dev/mapper/serveerstertje_ssd-root ro
> > verbose earlyprintk=xen mem=2048M console=hvc0 scsi_mod.use_blk_mq=1
> > console=tty0 acpi_enforce_resources=lax max_loop=30 loop_max_part=10
> > r8169.use_dac=1 loglevel=10 nomodeset net.ifnames=0 biosdevname=0
> > xen-pciback.hide=(00:14.2)(04:00.*)(08:00.0)(09:00.*)(0a:00.0)(0d:00.0)
> ...
>
> > [    0.135670] ------------[ cut here ]------------
> > [    0.135690] Failed to update kernel mapping for mfn=6930582 pfn=524288
> > [    0.135701] WARNING: CPU: 0 PID: 0 at arch/x86/xen/setup.c:312
> > xen_remap_memory+0x191/0x26c
> There seems to be a problem with rearranging the dom0 memory layout
> to match that of the host. Especially it is going sideways when trying
> to remap a memory frame to PFN 524288 (0x80000).
>
> You are limiting dom0 memory to 2 GB via Xen command line, but in
> addition to that you specify "mem=2048M" for the dom0 kernel, too. This
> should cap memory from PFN 0x80000 onwards, hence the failure when dom0
> tries to remap a frame into this area.
>
> Can you please try removing the "mem=2048M" from the dom0 parameters?
>
> I think there has been a kernel commit fixing the correct handling of
> the "mem=" parameter in 5.16, which might be the reason why you are
> seeing the crash only now.
>
>
> Juergen
Ah that sounds plausible, will test tomorrow and report back.
Thanks once again !

--
Sander