Re: [Qemu-ppc] Freescale qemu kvm and virtio

[Scott Wood]

On Thu, 2013-12-12 at 14:12 -0800, Florian Chazal wrote:
> Hi guys,
> 
> 
> I try to virtualize my system running on freescale p4040 (baremetal)
> on top of a p4080 + kvm :
> Every thing seems to work fine till I do intesive I/O in my VM : 
> qemu-system-ppc -enable-kvm -name kvm1 -serial tcp::4444,server,telnet -m 
> 1024 -M ppce500 -nographic -netdev tap,id=tap0,script=no,vhost=on -device  
> e1000,netdev=tap0 -kernel 
> /home/root/uImage--3.8-r14.3-p4080ds-20131211202743.bin -initrd 
> /home/root/rootfs-goodArch.cpio.gz -append "root=dev/vda1 init=sbin/init 
> console=ttyS0,115200 3" -drive 
> file=/media/sda1/d4-fromscratch-v0.8.4.img,if=virtio
> QEMU waiting for connection on: telnet::0.0.0.04444,server
> QEMU 1.4.0 monitor - type 'help' for more information
> (qemu) status
> unknown command: 'status'
> (qemu) info status
> VM status: running
> (qemu) EXT3-fs (sda1): error in ext3_reserve_inode_write: IO failure
> EXT3-fs error (device sda1): ext3_new_block: block(1542) >= blocks count(0) - 
> block_group = 0, es == eff1a400
> EXT3-fs (sda1): error in ext3_reserve_inode_write: IO failure
> EXT3-fs (sda1): error in ext3_orphan_add: IO failure
> EXT3-fs (sda1): error in ext3_reserve_inode_write: IO failure
> EXT3-fs (sda1): error in ext3_orphan_add: IO failure
> EXT3-fs (sda1): error in ext3_reserve_inode_write: IO failure
> EXT3-fs error (device sda1): ext3_new_block: block(1545) >= blocks count(0) - 
> block_group = 0, es == eff1a400
> EXT3-fs (sda1): error in ext3_reserve_inode_write: IO failure
> EXT3-fs (sda1): error in ext3_orphan_add: IO failure
> 
> 
> My qcow2 image is on a usb key and after this happens, the usb key
> file system is remounted by the host in readonly mode due to ext3
> journal corruption.
> 
> 
> I tried the following thingto use io throttle with virtio bps/iops
> option => limiting  bps to 500000 everything was slow as expected but
> then I got a different qemu error
> 
> (qemu) Unable to handle kernel paging request for data at address 0xffbc8be4
> Faulting instruction address: 0xc048e774

Bare addresses are not useful without the vmlinux they came from -- do
you have a full crash dump with symbol names?  Though if the crash is in
the filesystem code, you'll need to report it to the appropriate
filesystem list. 

As for the above ext3 errors, given the message "IO failure" it may be a
hardware problem.

What kernel are you using?  What U-Boot?

Do you have all relevant errata worked around, especially USB errata?
There's a USB erratum that basically says high speed mode can't be used
except on specially screened chips.
> 
> I found people saying that it can be linked to the virtio driver
> (https://bugzilla.redhat.com/show_bug.cgi?id=531827) 

That bug is about corruption in guest's RFS.  You seem to have
corruption in your host filesystem.  I don't see how that is related to
virtio or KVM -- as far as the host filesystem is concerned, QEMU is
just an ordinary userspace process doing I/O.

> and who advice to go for IDE but unfortunately I can't make ide
> working ("qemu-system-ppc -hda" or -drive ...if=ide), the disk is not
> seen my the kernel.

Right, this is not supported on ppce500.

> The freescale documentation only mentions virtio block disk
> (http://www.freescale.com/infocenter/topic/QORIQSDK/2412421.html#id_2412421).
> Is it because freescale implementation for e500mc does not support
> classic ide ?

Yes.  The closest real-hardware analogue would be either a non-legacy
PCI device, or the CCSR SATA device that can be found on some similar
chips such as p3041.  The latter is not supported in QEMU, and I don't
know what would be involved in the former.

-Scott