[ANNOUNCE] kvm-80 release

[ANNOUNCE] kvm-80 release

[Avi Kivity]

We're continuing the merge with kvm support in upstream qemu, and this 
is rocking the boat; please test carefully.

Important note: unlike most kvm releases, userspace in this release is 
required the modules from kvm-80 (or above), and will not work with 
older versions or with the modules provided by the upstream kernel. 
This will be corrected in kvm-81.

Changes from kvm-79:
- merge qemu-svn
   - automatic port allocation for vnc and similar
   - improved cdrom media change handling
   - improved debugging support
   - more scsi improvements
   - e1000 vlan offload
- up to 256 vcpus for ia64 (Jes Sorensen)
- only request io port permission if device assignment is enabled (Jes 
Sorensen)
- fix in-kernel ioapic reset (Jan Kiszka)
   - fixes guest reboot failures
- fix balloon for systems with page size different from 4K (Hollis 
Blanchard)
- fix bios memory reporting with >32G guests (Bill Rieske)
- drop inclusion of compiler.h in external modules (Xiantao Zhang)
- use vcpu pointers instead of vcpu numbers (Jes Sorensen)
- goodbye bool (Jes Sorensen)
- stop using separate structure for cpu state; use qemu's instead
   (Jes Sorensen)
- apic testsuite
- fix -no-kvm apic with smp (Jan Kiszka)
- clean up userspace nmi injection (Jan Kiszka)
- adjust external module for 2.6.28 hosts (Mark McLoughlin)
- much improved large mtu/gso support for virtio-net (Mark McLoughlin)
- simplify x86 emulator stack pop operations
- clean up device assignment code (Mark McLoughlin)
- fix extboot returning error incorrectly (Glauber Costa)
- constrain device assignment mmio to page alignment (Han Weidong)
- ppc speedups (Hollis Blanchard)
- fix s390 problem state handling (Christian Borntraeger)
- fix interrupt loss when injecting an nmi
   - fixes Linux guests with nmi watchdog hanging
- fix ia64 build (Xiantao Zhang)
- fix emergency reboot and kdump on host while running VMs
   (Eduardo Habkost)
- fix syncing of self-referencing ptes (Marcelo Tosatti)
- add printk() for ia64 vmm (Xiantao Zhang)
- add crashed vm handler for ia64 (Xiantao Zhang)
- use correct calling convention into ia64 PAL code (Xiantao Zhang)
- request interrupt window after injecting an interrupt if another is 
pending
- plug ppc memory leak when destroying a vm
- sign kvmclock as a paravirt operation (Glauber Costa)
- MSI support for device assigment (Sheng Yang)
- fix shadow pagetables leak (Marcelo Tosatti)
   - fixes an oops
- optimize setting new pte while syncing a page (Marcelo Tosatti)
- reduce code duplication in x86 emulator
- fix s390 refcounting; allow module unload (Christian Borntraeger)
- fix s390 memory leak (Christian Borntraeger)
- prevent tracing calls into unloaded module (Wu Fengguang)
- fix sparse warnings (Hannes Eder)
- clean up irq ack notifiers (Mark McLoughlin)
- powerpc large page support (Hollis Blanchard)
- faster powerpc shadow tlb handling (Hollis Blanchard)
- reduce powerpc guest tlb miss rate on context switches (Hollis Blanchard)
- powerpc exit timing stats (Hollis Blanchard)
- fix -no-kvm-irqchip not working with smp (Jan Kiszka)
- fix rendering regressions introduced by upstream qemu rendering support

-- 
I have a truly marvellous patch that fixes the bug which this
signature is too narrow to contain.

Re: [ANNOUNCE] kvm-80 release

[Michael Tokarev]

[-- Attachment #1: Type: text/plain, Size: 1467 bytes --]

Avi Kivity wrote:
> We're continuing the merge with kvm support in upstream qemu, and this
> is rocking the boat; please test carefully.

Ok, this is a first "bugreport", so to say, and first time I replying to
an announcement - i.e, it' the first time kvm does not work for me right
out of the box... ;)

Here -- attached - is an example of default-install of WinXP running under
freshly-installed kvm-80.  This shot was taken when I pressed Down and Right
arrow keys already, so some parts of the screen actually appeared.

In short: no screen updates.  Not even mouse cursor - the display is constant
no matter what.  Just sometimes, very rare, it redraws some parts, and
continues.  Just like on the example screen - the hourglass is long gone,
the mouse has been moved many times already (by me trying it), yet it's
still here right on the center of the screen...

This happens on 2.6.27 kernel from kernel.org, 32bits, on an amd athlon
x2-64 CPU, with the default cirrus-logic VGA emulation. With -vga std
windows does not appear at all, just a black screen.

The guest (one of a few) is a cloned default winXP (russian) SP3 install,
worked with several previous kvm releases.

I'm trying other games now, like safe mode or reinstalling 'doze with the
current kvm, but for some reason I think it will not do the trick...  So
here we go.

Thanks!

...
> - fix rendering regressions introduced by upstream qemu rendering support

BTW, maybe this is it? ;)

/mjt


[-- Attachment #2: ws.jpg --]
[-- Type: image/jpeg, Size: 10144 bytes --]

Re: [ANNOUNCE] kvm-80 release

[Avi Kivity]

Michael Tokarev wrote:
> In short: no screen updates.  Not even mouse cursor - the display is constant
> no matter what.  Just sometimes, very rare, it redraws some parts, and
> continues.  Just like on the example screen - the hourglass is long gone,
> the mouse has been moved many times already (by me trying it), yet it's
> still here right on the center of the screen...
>
>   

Have you reloaded the modules?  This is what happens when you use 
modules from the stock kernel.  Will fix this for kvm-81.

Windows XP ought to work well in kvm-80.

-- 
I have a truly marvellous patch that fixes the bug which this
signature is too narrow to contain.

Re: [ANNOUNCE] kvm-80 release

[Michael Tokarev]

Avi Kivity wrote:
> Michael Tokarev wrote:
>> In short: no screen updates.  Not even mouse cursor - the display is
>> constant
>> no matter what.  Just sometimes, very rare, it redraws some parts, and
>> continues.  Just like on the example screen - the hourglass is long gone,
>> the mouse has been moved many times already (by me trying it), yet it's
>> still here right on the center of the screen...
> 
> Have you reloaded the modules?  This is what happens when you use
> modules from the stock kernel.  Will fix this for kvm-81.

Yes I reloaded the modules. I even rebooted with kvm-80 modules
in /lib/modules/$kver/... instead of the ones from 2.6.27 kernel.

> Windows XP ought to work well in kvm-80.

I also tried the fix that went into kvm-81-tobe, as pointed out
by Jan Kiszka.  No change here.  Which is quite expectable, since
the fix is for cirrus, but my prob appears with -vga std too.

I'll keep trying for some more time.  But I don't quite know
what to try next.  Even recompiled kvm/qemu with gcc-3.3 as
it used to use (kvm-80 used default gcc) - no difference.
Installing windows from scratch does not help either - when
it tries to display real graphics resolution instead of some
rudimentary graphics it shows during install, the screen
stops refreshing.

Thanks!

/mjt