Re: [ANNOUNCE] kvm-72 release

Re: [ANNOUNCE] kvm-72 release

[Henrik Holst]

There is a boot regression in kvm-72:

scsi and virtio images will not survive a "reboot" from within the
guest. Only drives with if=ide survives a rebooting of the guest.

That is, this will create a guest that will survive any number of reboots:
    qemu-system-x86_64 -M -drive file=a,if=ide,media=disk,boot=on -boot c

While this will only survive the first boot:
    qemu-system-x86_64 -M -drive file=a,if=scsi,media=disk,boot=on -boot c OR
    qemu-system-x86_64 -M -drive file=a,if=virtio,media=disk,boot=on -boot c


Also the extboot work done by H. Peter Anvin finally makes the newer
isolinux menus to work but the older isolinux (for example the
installer to ubuntu 7.10) still does not work. It no longer segfaults
though but instead the guest completely hangs. Much improvement
though!

/Henrik Holst

Re: [ANNOUNCE] kvm-72 release

[Anthony Liguori]

Henrik Holst wrote:
> There is a boot regression in kvm-72:
>
> scsi and virtio images will not survive a "reboot" from within the
> guest. Only drives with if=ide survives a rebooting of the guest.
>
> That is, this will create a guest that will survive any number of reboots:
>     qemu-system-x86_64 -M -drive file=a,if=ide,media=disk,boot=on -boot c
>
> While this will only survive the first boot:
>     qemu-system-x86_64 -M -drive file=a,if=scsi,media=disk,boot=on -boot c OR
>     qemu-system-x86_64 -M -drive file=a,if=virtio,media=disk,boot=on -boot c
>   

What does "survive" mean?

> Also the extboot work done by H. Peter Anvin finally makes the newer
> isolinux menus to work but the older isolinux (for example the
> installer to ubuntu 7.10)

Ubuntu doesn't use isolinux, it uses gfxboot.  Older versions of gfxboot 
are known to be broken with KVM on Intel.

Regards,

Anthony Liguori

>  still does not work. It no longer segfaults
> though but instead the guest completely hangs. Much improvement
> though!
>
> /Henrik Holst
> --
> To unsubscribe from this list: send the line "unsubscribe kvm" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
>   

Re: [ANNOUNCE] kvm-72 release

[Henrik Holst]

2008/7/28 Anthony Liguori <anthony@codemonkey.ws>:
> Henrik Holst wrote:
>>
>> There is a boot regression in kvm-72:
>>
>> scsi and virtio images will not survive a "reboot" from within the
>> guest. Only drives with if=ide survives a rebooting of the guest.
>>
>> That is, this will create a guest that will survive any number of reboots:
>>    qemu-system-x86_64 -M -drive file=a,if=ide,media=disk,boot=on -boot c
>>
>> While this will only survive the first boot:
>>    qemu-system-x86_64 -M -drive file=a,if=scsi,media=disk,boot=on -boot c
>> OR
>>    qemu-system-x86_64 -M -drive file=a,if=virtio,media=disk,boot=on -boot
>> c
>>
>
> What does "survive" mean?

survive means that the drive is available after the guest has
rebooted. If we set if=virtio and start a guest everything is working
just fine. Then we type "reboot" in the guest and the guest reboots.
Now we halt in bios since the bios cannot find any bootable devices
anymore. If we use if=ide the bios will find the bootable devices just
like on kvm-71 and the guest simply starts again like on normal
hardware.

>
>> Also the extboot work done by H. Peter Anvin finally makes the newer
>> isolinux menus to work but the older isolinux (for example the
>> installer to ubuntu 7.10)
>
> Ubuntu doesn't use isolinux, it uses gfxboot.  Older versions of gfxboot are
> known to be broken with KVM on Intel.
>
noted, I might have written wrongly. All I meant was that the older
ubuntu installers (like 7.10) no longer segfaults qemu/kvm but instead
now halts the guest, could be good to know for anyone trying to
improve the situation I thought.

/Henrik Holst

Re: [ANNOUNCE] kvm-72 release

[Henrik Holst]

2008/7/28 Henrik Holst <henrik@witsbits.com>:
> 2008/7/28 Anthony Liguori <anthony@codemonkey.ws>:
>> Henrik Holst wrote:
>>>
>>> There is a boot regression in kvm-72:
>>>
>>> scsi and virtio images will not survive a "reboot" from within the
>>> guest. Only drives with if=ide survives a rebooting of the guest.
>>>
>>> That is, this will create a guest that will survive any number of reboots:
>>>    qemu-system-x86_64 -M -drive file=a,if=ide,media=disk,boot=on -boot c
>>>
>>> While this will only survive the first boot:
>>>    qemu-system-x86_64 -M -drive file=a,if=scsi,media=disk,boot=on -boot c
>>> OR
>>>    qemu-system-x86_64 -M -drive file=a,if=virtio,media=disk,boot=on -boot
>>> c
>>>
>>
>> What does "survive" mean?
>
> survive means that the drive is available after the guest has
> rebooted. If we set if=virtio and start a guest everything is working
> just fine. Then we type "reboot" in the guest and the guest reboots.
> Now we halt in bios since the bios cannot find any bootable devices
> anymore. If we use if=ide the bios will find the bootable devices just
> like on kvm-71 and the guest simply starts again like on normal
> hardware.

Since this probably is due to something in qemu it would be great if
the release-log to kvm could include towards which revision of the
qemu svn repository that was merged. Or am I to understand that the
merge was performed on the same day as the new kvm release was
released?

I'm thinking of this line:
  - merge qemu-svn

That way I could myself begin to hunt down the culprit patch to qemu
much more easily. Also the qemu version in kvm seams to contain some
internal patches? I noted that -drive has an "boot" option (in vl.c)
in kvm which is not present in qemu-svn (atleast not in trunk).

>
>>
>>> Also the extboot work done by H. Peter Anvin finally makes the newer
>>> isolinux menus to work but the older isolinux (for example the
>>> installer to ubuntu 7.10)
>>
>> Ubuntu doesn't use isolinux, it uses gfxboot.  Older versions of gfxboot are
>> known to be broken with KVM on Intel.
>>
> noted, I might have written wrongly. All I meant was that the older
> ubuntu installers (like 7.10) no longer segfaults qemu/kvm but instead
> now halts the guest, could be good to know for anyone trying to
> improve the situation I thought.
>
> /Henrik Holst
>

Re: [ANNOUNCE] kvm-72 release

[Henrik Holst]

> Since this probably is due to something in qemu it would be great if
> the release-log to kvm could include towards which revision of the
> qemu svn repository that was merged. Or am I to understand that the
> merge was performed on the same day as the new kvm release was
> released?
>
> I'm thinking of this line:
>  - merge qemu-svn
>
> That way I could myself begin to hunt down the culprit patch to qemu
> much more easily. Also the qemu version in kvm seams to contain some
> internal patches? I noted that -drive has an "boot" option (in vl.c)
> in kvm which is not present in qemu-svn (atleast not in trunk).
>

I have now tried with qemu-svn and that won't even boot up a scsi drive at all!

This will present a bios which founds no boot devices at all:
  qemu -m 256 -drive file=test.img,if=scsi,media=disk -boot c

While this will boot just fine:
  qemu -m 256 -drive file=test.img,if=ide,media=disk -boot c

So I will have to dig into previous revisions of qemu to see where it
went wrong.

/Henrik Holst

Re: [ANNOUNCE] kvm-72 release

[Henrik Holst]

2008/7/28 Henrik Holst <henrik@witsbits.com>:
>> Since this probably is due to something in qemu it would be great if
>> the release-log to kvm could include towards which revision of the
>> qemu svn repository that was merged. Or am I to understand that the
>> merge was performed on the same day as the new kvm release was
>> released?
>>
>> I'm thinking of this line:
>>  - merge qemu-svn
>>
>> That way I could myself begin to hunt down the culprit patch to qemu
>> much more easily. Also the qemu version in kvm seams to contain some
>> internal patches? I noted that -drive has an "boot" option (in vl.c)
>> in kvm which is not present in qemu-svn (atleast not in trunk).
>>
>
> I have now tried with qemu-svn and that won't even boot up a scsi drive at all!
>
> This will present a bios which founds no boot devices at all:
>  qemu -m 256 -drive file=test.img,if=scsi,media=disk -boot c
>
> While this will boot just fine:
>  qemu -m 256 -drive file=test.img,if=ide,media=disk -boot c
>
> So I will have to dig into previous revisions of qemu to see where it
> went wrong.
>
> /Henrik Holst
>

I have now tracked down the regression to extboot.bin, changing it to
the version from kvm-71 makes it all work just fine again. Will try
and find where in the code the regression is, sounds like it is the
recent extboot patches though.

/Henrik Holst

[ANNOUNCE] kvm-72 release

[Avi Kivity]

The most noticeable change in this release, by far, is the removal of 
the QEMU/KVM annotation on the qemu window in favor of a new 'info kvm' 
monitor command.  If you dig deeper, you will see improved stability for 
Windows guests, especially x64 variants.

Changes from kvm-71:
- fix ia64 build breakage (Xiantao Zhang)
- ppc kvmtrace support (Christian Ehrhardt)
- merge qemu-svn
   - fixes suspend/resume of multiple nics
   - adds Virtual Distributed Ethernet (vde) support
- add 'push mem' to patched tpr instructions
- use host kernel anon_inodes support if available
- robustify signalfd() emulation
- avoid losing the iothread on random signals
- fix test for iothread (Jan Kiszka)
- remove hypercall device (Anthony Liguori)
- remove duplicate monitor command (Anthony Liguori)
- remove unused function declaration (Anthony Liguori)
- remove unnecessary change to TARGET_PAGE_SIZE (Anthony Liguori)
- remove qemu/kvm application decoration (Anthony Liguori)
- add 'info kvm' monitor command  (Anthony Liguori)
- s390 libkvm port (Carsten Otte)
- fix compile warning (Carlo Marcelo Arenas Belon)
- fix duplicate #include (Carlo Marcelo Arenas Belon)
- adjust external module for 2.6.27 hosts
- avoid tpr patching when the stack is not valid (Marcelo Tosatti)
   - fixes some variant of Windows 2003 SMP
- fix extboot stomping on reserved interrupt vectors (H. Peter Anvin)
- remove unused -translation option (Anthony Liguori)
- fix preemption during copy_from_user_inatomic() on older hosts
    (Andrea Arcangeli)
- fix ppc cpu initialization (Hollis Blanchard)
- drop shadow ptes on memslot destruction (Marcelo Tosatti)
- improve invalid shadow root page table handling (Marcelo Tosatti)
- simplify mmu page zapping
- simplify interrupt processing on Intel
- remove redundant check for real mode (Mohammed Gamal)
- add clflush emulation (Glauber Costa)
- fix race setting shadow ptes on i386
- consolidate pic isr clearing into a function
- consolidate exception interrupt vector #defines (Jan Kiszka)
- fix interrupt reinjection in real mode (Jan Kiszka)
- unify guest register save/restore code
- x86 hardware task switching fixes (Marcelo Tosatti)
- optimize vmx by avoiding saving the host stack pointer
- fix interrupts being delayed by instruction emulation
   - improves Vista stability
- fix reloading kvm-intel.ko while changing parameters (Sheng Yang)
- fix reboot exception entry build bug (Ingo Molnar)
- ignore no-op DEBUGCTL msr writes (Alexander Graf)
   - fixes Netware
- fix irq state leak on ia64 (Julia Lawall)
- s390 fixes (Carsten Otte, Christian Borntraeger, Martin Schwidefsky)
- ppc guest breakpoint support (Hollis Blanchard)
- optimize ppc by not saving host tlb state (Hollis Blanchard)
- reduce ppc guest tlb writes (Hollis Blanchard)
- avoid guest tlb flushes on guest privilege switches (Hollis Blanchard)


Notes:
     If you use the modules bundled with kvm-72, you can use any version
of Linux from 2.6.16 upwards.  You may also use kvm-72 userspace with
the kvm modules provided by Linux 2.6.22 or above.  Some features may
only be available in newer releases.

http://kvm.qumranet.com


-- 
error compiling committee.c: too many arguments to function

Re: [ANNOUNCE] kvm-72 release

[Xavier Gnata]

Avi Kivity wrote:
> The most noticeable change in this release, by far, is the removal of 
> the QEMU/KVM annotation on the qemu window in favor of a new 'info 
> kvm' monitor command.  If you dig deeper, you will see improved 
> stability for Windows guests, especially x64 variants.
>
> Changes from kvm-71:
> - fix ia64 build breakage (Xiantao Zhang)
> - ppc kvmtrace support (Christian Ehrhardt)
> - merge qemu-svn
>   - fixes suspend/resume of multiple nics
>   - adds Virtual Distributed Ethernet (vde) support
> - add 'push mem' to patched tpr instructions
> - use host kernel anon_inodes support if available
> - robustify signalfd() emulation
> - avoid losing the iothread on random signals
> - fix test for iothread (Jan Kiszka)
> - remove hypercall device (Anthony Liguori)
> - remove duplicate monitor command (Anthony Liguori)
> - remove unused function declaration (Anthony Liguori)
> - remove unnecessary change to TARGET_PAGE_SIZE (Anthony Liguori)
> - remove qemu/kvm application decoration (Anthony Liguori)
> - add 'info kvm' monitor command  (Anthony Liguori)
> - s390 libkvm port (Carsten Otte)
> - fix compile warning (Carlo Marcelo Arenas Belon)
> - fix duplicate #include (Carlo Marcelo Arenas Belon)
> - adjust external module for 2.6.27 hosts
> - avoid tpr patching when the stack is not valid (Marcelo Tosatti)
>   - fixes some variant of Windows 2003 SMP
> - fix extboot stomping on reserved interrupt vectors (H. Peter Anvin)
> - remove unused -translation option (Anthony Liguori)
> - fix preemption during copy_from_user_inatomic() on older hosts
>    (Andrea Arcangeli)
> - fix ppc cpu initialization (Hollis Blanchard)
> - drop shadow ptes on memslot destruction (Marcelo Tosatti)
> - improve invalid shadow root page table handling (Marcelo Tosatti)
> - simplify mmu page zapping
> - simplify interrupt processing on Intel
> - remove redundant check for real mode (Mohammed Gamal)
> - add clflush emulation (Glauber Costa)
> - fix race setting shadow ptes on i386
> - consolidate pic isr clearing into a function
> - consolidate exception interrupt vector #defines (Jan Kiszka)
> - fix interrupt reinjection in real mode (Jan Kiszka)
> - unify guest register save/restore code
> - x86 hardware task switching fixes (Marcelo Tosatti)
> - optimize vmx by avoiding saving the host stack pointer
> - fix interrupts being delayed by instruction emulation
>   - improves Vista stability
> - fix reloading kvm-intel.ko while changing parameters (Sheng Yang)
> - fix reboot exception entry build bug (Ingo Molnar)
> - ignore no-op DEBUGCTL msr writes (Alexander Graf)
>   - fixes Netware
> - fix irq state leak on ia64 (Julia Lawall)
> - s390 fixes (Carsten Otte, Christian Borntraeger, Martin Schwidefsky)
> - ppc guest breakpoint support (Hollis Blanchard)
> - optimize ppc by not saving host tlb state (Hollis Blanchard)
> - reduce ppc guest tlb writes (Hollis Blanchard)
> - avoid guest tlb flushes on guest privilege switches (Hollis Blanchard)
>
>
> Notes:
>     If you use the modules bundled with kvm-72, you can use any version
> of Linux from 2.6.16 upwards.  You may also use kvm-72 userspace with
> the kvm modules provided by Linux 2.6.22 or above.  Some features may
> only be available in newer releases.
>
> http://kvm.qumranet.com
>
It looks like the audio driver/card selection via ./configure is still 
broken. 
This set of patches does the job with kvm-71 :
"[PATCH 1/2] configure: include audio list options for --help output"
und
"[PATCH 0/2] configure: add support for audio-{drv,card}-list"


Xavier

Re: [ANNOUNCE] kvm-72 release

[Xavier Gnata]

Hi,

kvm-72 does not compile with 2.6.26 kernels ( x86-64bits):

./configure && make gives :

/usr/local/src/kvm-72/kernel/anon_inodes.c: In function 
‘kvm_anon_inode_getfd’:
/usr/local/src/kvm-72/kernel/anon_inodes.c:254: error: too many 
arguments to function ‘anon_inode_getfd’

Were does this "int flag" come from ??

Xavier

> The most noticeable change in this release, by far, is the removal of 
> the QEMU/KVM annotation on the qemu window in favor of a new 'info 
> kvm' monitor command.  If you dig deeper, you will see improved 
> stability for Windows guests, especially x64 variants.
>
> Changes from kvm-71:
> - fix ia64 build breakage (Xiantao Zhang)
> - ppc kvmtrace support (Christian Ehrhardt)
> - merge qemu-svn
>   - fixes suspend/resume of multiple nics
>   - adds Virtual Distributed Ethernet (vde) support
> - add 'push mem' to patched tpr instructions
> - use host kernel anon_inodes support if available
> - robustify signalfd() emulation
> - avoid losing the iothread on random signals
> - fix test for iothread (Jan Kiszka)
> - remove hypercall device (Anthony Liguori)
> - remove duplicate monitor command (Anthony Liguori)
> - remove unused function declaration (Anthony Liguori)
> - remove unnecessary change to TARGET_PAGE_SIZE (Anthony Liguori)
> - remove qemu/kvm application decoration (Anthony Liguori)
> - add 'info kvm' monitor command  (Anthony Liguori)
> - s390 libkvm port (Carsten Otte)
> - fix compile warning (Carlo Marcelo Arenas Belon)
> - fix duplicate #include (Carlo Marcelo Arenas Belon)
> - adjust external module for 2.6.27 hosts
> - avoid tpr patching when the stack is not valid (Marcelo Tosatti)
>   - fixes some variant of Windows 2003 SMP
> - fix extboot stomping on reserved interrupt vectors (H. Peter Anvin)
> - remove unused -translation option (Anthony Liguori)
> - fix preemption during copy_from_user_inatomic() on older hosts
>    (Andrea Arcangeli)
> - fix ppc cpu initialization (Hollis Blanchard)
> - drop shadow ptes on memslot destruction (Marcelo Tosatti)
> - improve invalid shadow root page table handling (Marcelo Tosatti)
> - simplify mmu page zapping
> - simplify interrupt processing on Intel
> - remove redundant check for real mode (Mohammed Gamal)
> - add clflush emulation (Glauber Costa)
> - fix race setting shadow ptes on i386
> - consolidate pic isr clearing into a function
> - consolidate exception interrupt vector #defines (Jan Kiszka)
> - fix interrupt reinjection in real mode (Jan Kiszka)
> - unify guest register save/restore code
> - x86 hardware task switching fixes (Marcelo Tosatti)
> - optimize vmx by avoiding saving the host stack pointer
> - fix interrupts being delayed by instruction emulation
>   - improves Vista stability
> - fix reloading kvm-intel.ko while changing parameters (Sheng Yang)
> - fix reboot exception entry build bug (Ingo Molnar)
> - ignore no-op DEBUGCTL msr writes (Alexander Graf)
>   - fixes Netware
> - fix irq state leak on ia64 (Julia Lawall)
> - s390 fixes (Carsten Otte, Christian Borntraeger, Martin Schwidefsky)
> - ppc guest breakpoint support (Hollis Blanchard)
> - optimize ppc by not saving host tlb state (Hollis Blanchard)
> - reduce ppc guest tlb writes (Hollis Blanchard)
> - avoid guest tlb flushes on guest privilege switches (Hollis Blanchard)
>
>
> Notes:
>     If you use the modules bundled with kvm-72, you can use any version
> of Linux from 2.6.16 upwards.  You may also use kvm-72 userspace with
> the kvm modules provided by Linux 2.6.22 or above.  Some features may
> only be available in newer releases.
>
> http://kvm.qumranet.com
>
>

Re: [ANNOUNCE] kvm-72 release

[Jorge Lucángeli Obes]

On Sun, Jul 27, 2008 at 7:11 PM, Xavier Gnata <xavier.gnata@gmail.com> wrote:
> Hi,
>
> kvm-72 does not compile with 2.6.26 kernels ( x86-64bits):
>
> ./configure && make gives :
>
> /usr/local/src/kvm-72/kernel/anon_inodes.c: In function
> 'kvm_anon_inode_getfd':
> /usr/local/src/kvm-72/kernel/anon_inodes.c:254: error: too many arguments to
> function 'anon_inode_getfd'

I'm having the exact same problem, using a 2.6.26 64-bit kernel as well.

Cheers,
Jorge

RE: [ANNOUNCE] kvm-72 release

[Xu, Jiajun]

On Monday, July 28, 2008 11:11 AM kvm-owner@vger.kernel.org wrote:

> On Sun, Jul 27, 2008 at 7:11 PM, Xavier Gnata
> <xavier.gnata@gmail.com> wrote:
>> Hi,
>> 
>> kvm-72 does not compile with 2.6.26 kernels ( x86-64bits):
>> 
>> ./configure && make gives :
>> 
>> /usr/local/src/kvm-72/kernel/anon_inodes.c: In function
>> 'kvm_anon_inode_getfd':
>> /usr/local/src/kvm-72/kernel/anon_inodes.c:254: error: too many
>> arguments to function 'anon_inode_getfd' 
> 
> I'm having the exact same problem, using a 2.6.26 64-bit
> kernel as well.

The problem also exists with 2.6.26 32-bit kernel.


Best Regards
Jiajun