[ kvm-Bugs-1807560 ] Font corruption (AMD specific)

[ kvm-Bugs-1807560 ] Font corruption (AMD specific)

[SourceForge.net]

Bugs item #1807560, was opened at 2007-10-04 14:32
Message generated for change (Tracker Item Submitted) made by Item Submitter
You can respond by visiting: 
https://sourceforge.net/tracker/?func=detail&atid=893831&aid=1807560&group_id=180599

Please note that this message will contain a full copy of the comment thread,
including the initial issue submission, for this request,
not just the latest update.
Category: None
Group: None
Status: Open
Resolution: None
Priority: 5
Private: No
Submitted By: Matthew Hall (castleinthesky)
Assigned to: Nobody/Anonymous (nobody)
Summary: Font corruption (AMD specific)

Initial Comment:
Hi,

I'm filing this is a separate bug to 1806338, as it still presents with vnc and works with -no-kvm.

Description:
I have a problem, which has been present since KVM 43 involving a font corruption, which is AMD specific (works fine on my Intel box), and only occurs when I have multiple hosts/processes affined to the same cpu.
Example: http://www.castleinthesky.org/kvm2.png

Background:
I initially came across this a while ago, and it was fairly difficult to reproduce at first as it would only corrupt the fonts of particular applications being run on guests. It was a while until I figured out it only occured when two guests had been affined to the same cpu on the host and were vying for resources.

Host: 4-core AMD Opteron 2212 HE, F7, x86_64, kernel 2.6.22.9-91.fc7, kvm 45
Guest: Multiple F7, i386, 2.6.22.5-76.fc7, qcow2 images.

To test where the problem began i've gone back to KVM 34 and worked forward; running two identical i386 f7 images both with 'taskset 1'.
I've yet to bisect to find the specific commit - thought you might want this initial report.

Tests & Results:

Unless explicitly stated, guest boots with 'nmi_watchdog=0 acpi=force'
Unless explicitly stated, host qemu cmdline options were 'taskset 1 /usr/bin/qemu-system-x86_64 -hda /home/fedora-7/devel-$num.img -net nic,macaddr=52:54:00:12:34:4$num,model=rtl8139 -net tap,script=/etc/kvm/qemu-ifup -boot c -m 1024 -smp 1 -no-fd-bootchk -k en-gb -usb -usbdevice tablet -vnc :$num -daemonize' (where $nums are 1 and 2, of 2 identical images).

Notes:
The timing tests involved running 'rdate -s time.demon.co.uk;while true;do date;sleep 1;done' on the host and guests. All timing tests had a positive/good result for KVM <= 36.
In any tests where the guest had acpi disabled (via a lack of 'acpi=force'), the guests lost several seconds a minute. The clock was usually biased towards one guest, ie. one would lose a second every few minutes, the other would lose several seconds every minute.

In the tests for kvm 43-45 with '-no-kvm-irqchip', the font display problem only occured on particular occasions, ie. it occured sometimes on both hosts, sometimes on one and not the other, and sometimes on neither.


---
KVM 34 		- good

KVM 35		- good

KVM 36		- good

KVM 37		- bad (guest sticks on ps/2 enumeration at boot)
KVM 37, without guest cmdline option 'acpi=force'
		- good

KVM 38		- bad (qemu segfault, host oops)

KVM 39		- bad (guest sticks on ps/2 enumeration at boot)
KVM 39, without guest cmdline option 'acpi=force'
		- good

KVM 40		- bad (guest sticks on ps/2 enumeration at boot)
KVM 40, without guest cmdline option 'acpi=force'
		- bad (fonts corrupt)

KVM 41		- bad (guest sticks on ps/2 enumeration at boot)
KVM 41, without guest cmdline option 'acpi=force'
		- good

KVM 42		- bad (guest sticks on ps/2 enumeration at boot)
KVM 42, without guest cmdline option 'acpi=force'
		- good

KVM 43		- bad (fonts corrupt)

KVM 44		- bad (fonts corrupt)
KVM 44, without guest cmdline option 'acpi=force'
		- bad (fonts corrupt)
KVM 44, with host cmdline options '-no-kvm'
		- bad (guest halts on _spin_unlock_irqrestore) 
KVM 44, with host cmdline options '-no-kvm-irqchip'
		- bad (fonts corrupt)
KVM 44, with host cmdline options '-no-kvm-irqchip -no-kvm'
		- bad (guest crash loop on spinlock)

KVM 45 		- bad (fonts corrupt)
		Notes:	Same results as KVM 44 in other tests.
			Timing test shows clock is noticably slower (both hosts lose 30 seconds every minute equally)
KVM 45, with host cmdline options '-no-kvm'
		- good

KVM Snapshot 20071003 - bad (fonts corrupt)
		Notes:	Same results as KVM 45 in other tests.

----------------------------------------------------------------------

You can respond by visiting: 
https://sourceforge.net/tracker/?func=detail&atid=893831&aid=1807560&group_id=180599

-------------------------------------------------------------------------
This SF.net email is sponsored by: Splunk Inc.
Still grepping through log files to find problems?  Stop.
Now Search log events and configuration files using AJAX and a browser.
Download your FREE copy of Splunk now >> http://get.splunk.com/