Re: NVMM issues

[Robert Nestor <rnestor%mac.com@localhost>]

On Jun 18, 2021, at 3:43 PM, Chavdar Ivanov <ci4ic4%gmail.com@localhost> wrote:

> On Fri, 18 Jun 2021 at 19:36, Robert Nestor <rnestor%mac.com@localhost> wrote:
>> 
>> Playing with FreeDOS 1.2 and 1.3 under nvmm on a NetBSD 9.1-amd64 system and ran into some issues.  Basically I can do an install from the FreeDOS-1.2 CD and run the system afterwards without an issue, but trying to install FreeDOS-1.3 the same way aborts in nvmm.  If the FreeDOS 1.3 install is done on actual hardware it succeeds and the resulting disk image file will boot and run fine under nvmm.  I’ve also tried running an old copy of Norton Symantec Ghost 2003 under both versions (to recover some old files).  It runs find on real hardware but aborts under nvmm.
>> 
>> Oh, to avoid the system reboot during the installation after FreeDOS partitions and formats the new disk, I do this beforehand using qemu-image create, vndconfig, fdisk, and newfs_msdos.
>> 
>> #
>> # This works for a FreeDOS 1.2 install
>> #
>> qemu-system-x86_64 -accel nvmm -cpu 486 -smp 1 -m 768 -cdrom ./FD12LGCY.iso -netdev tap,id=nd0,ifname=tap0,script=no,downscript=no -device rtl8139,netdev=nd0 -drive file=./FreeDOS-1.2.dsk,media=disk,format=raw
>> 
>> #
>> # This fails for a FreeDOS 1.3 install
>> #
>> qemu-system-x86_64 -accel nvmm -cpu 486 -smp 1 -m 768 -cdrom ./FD13LIVE.iso -netdev tap,id=nd0,ifname=tap0,script=no,downscript=no -device rtl8139,netdev=nd0 -drive file=./FreeDOS-1.3.dsk,media=disk,format=raw
>> #
>> # Error displayed when install fails:
>> #
>> NetBSD Virtual Machine Monitor accelerator is operational
>> qemu-system-x86_64: NVMM: Mem Assist Failed [gpa=0xb018f]
>> qemu-system-x86_64: NVMM: Failed to execute a VCPU.
> 
> I got the same running the newest available qemu-nvmm under today's
> -current (well, I prepared the volume myself after the first boot from
> the cdrom and started the actual installation after the reboot; the
> error takes place after the installation script starts copying data on
> the disk); my backing store was a zvol and the command line was c/p-ed
> from my other vm's, displaying over vnc.

That’s where I saw the error as well - in the file copy which seems like it should be a pretty benign operation.  In my case the volumes and such were all on a local disk, so nothing out of the ordinary.  Since the abort happens during the installation of the copy of the files to the new disk I tried just doing an xcopy of everything on the boot disk to a new disk and that works OK which makes the error seem even stranger.

> I think it is worth a pr. I've looked at the gdb trace, but it doesn't
> tell me a lot, to be honest

OK, I’ll file a PR against NVMM then assuming that were the problem might lie.

> 
>> 
>> #
>> # However, if the FreeDOS 1.3 system is installed using actual hardware
>> #  the resulting disk image file boots and runs successfully under nvmm.
>> #
>> 
>> #
>> # Attemping to run an old copy of Norton Symantec Ghost 2003 in nvmm
>> #  produces the following error under either FreeDOS 1.2 or FreeDOS 1.3:
>> # Note: Ghost 2003 runs fine on real hardware on both versions of FreeDOS
>> #  on systems with 32-bit (i386) or 64-bit (amd64).
>> #
>> qemu-system-x86_64: NVMM: Unexpected VM exit code 0xffffffffffffffff [hw=0x9]
>> qemu-system-x86_64: NVMM: Failed to execute a VCPU.
>> 
>> I suspect this is all caused by some bug in nvmm  (or qemu).  Is this worthy of filing a PR and if so should it be against nvmm, qemu or both?
>> 
>> Thanks
>> 
> 
> 
> -- 
> ----