[ANNOUNCE] kvm-83 release

[ANNOUNCE] kvm-83 release

[Avi Kivity]

The biggest change this time around is support for the AMD IOMMU for 
device assignment.  As it requires extensive core kernel support, it is 
only available on Linux 2.6.29rc1 or higher hosts.

Changes from kvm-82:
- amd iommu support (with Linux 2.6.29+ hosts) (Joerg Roedel)
- merge qemu-svn
   - qcow2 correctness fixes
   - '-drive serial=...' option
   - virtio-net migration fixes
- fix '--with-patched-kernel' build on Ubuntu (Nolan Leake)
- improved tsc handling (Marcelo Tosatti)
   - should reduce timing errors on older Linux guests
- remove cpuid code duplication due to upstream qemu work (Amit Shah)
- guest S3 sleep (Gleb Natapov)
- load assigned devices option rom (Kechao Liu)
- '-no-kvm-pit-reinjection' option to improve timing on RHEL 3 era guests
   (Marcelo Tosatti)
- remove duplicate firmware config interface probe in bios (Sebastian 
Herbszt)
- backward compatibility for marker_synchronize_register() (Eduardo Habkost)
- fix compile warning in tpr patching (Mark McLoughlin)
- don't install kvm headers to /usr/local
- fix smbios memory size in certain cases (Bill Rieske)
- simplify mmu shadow walking function
- stop using vmap() on guest pages (annoys ksm) (Izik Eidus)
- fix warning on MSR_VM_HSAVE_PA in dmesg
- fix i8254 timer count read (Marcelo Tosatti)
- ppc e500 core support (Hollis Blanchard)
- fix vmx real mode guest state validity checks
- make pop instruction emulation a little more generic
- implement 'ret far' instruction emulation
- properly lock i8354 timer
- remove ppc debug support (Hollis Blanchard)
- fix race freeing assigned device interrupt (Sheng Yang)
- fix pci interrupt in freeing in device assignment (Sheng Yang)
- allow disabling msi for assigned devices (Sheng Yang)
- emulate amd microcode level msr (Alexander Graf)
   - helps esx boot on kvm
- make page unshadowing a little faster
- clear pending pit interrupts when unmasking the timer irq
   - fixes xen-on-kvm
- mmu notifiers release method (Marcelo Tosatti)
   - fixes large pages support
- expose ia64 registers to userspace (Jes Sorensen)
- fix type in has_svm() (Joe Perches)
- drop unnecessary memory clearing when allocating shadow page
   (Marcelo Tosatti)
- rename "metaphysical" page attribute to "direct"
- fix irq routing with -no-kernel-irqchip


Notes:
     If you use the modules bundled with kvm-83, you can use any version
of Linux from 2.6.16 upwards.  You may also use kvm-83 userspace with
the kvm modules provided by Linux 2.6.25 or above.  Some features may
only be available in newer releases.

http://kvm.qumranet.com

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

Re: [ANNOUNCE] kvm-83 release

[Farkas Levente]

Avi Kivity wrote:
> The biggest change this time around is support for the AMD IOMMU for
> device assignment.  As it requires extensive core kernel support, it is
> only available on Linux 2.6.29rc1 or higher hosts.

status report on rhel/centos-5.2 host:
- fedora-9 latest kernel now boot again
- fedora-10 installer stop at "waiting for hardware..." so it's
currently not able to upgrade fedora-9 guest neither from cdrom nor form
 preupgrade (ie hard disk).

and these dozens of warning still comes during compilation on
rhel/centos 5.2:
-----------------------------------------
In file included from
/home/robot/rpm/BUILD/kvm-83/kernel/include/asm/kvm_host.h:65,
                 from
/home/robot/rpm/BUILD/kvm-83/kernel/include/linux/kvm_host.h:67,
                 from /home/robot/rpm/BUILD/kvm-83/kernel/x86/lapic.c:60:
/home/robot/rpm/BUILD/kvm-83/kernel/include-compat/asm/msr-index.h:30:1:
warning: "MSR_IA32_PERFCTR0" redefined
In file included from include/asm/processor.h:16,
                 from
/home/robot/rpm/BUILD/kvm-83/kernel/include/asm/kvm_para.h:89,
                 from
/home/robot/rpm/BUILD/kvm-83/kernel/include/linux/kvm_para.h:63,
                 from
/home/robot/rpm/BUILD/kvm-83/kernel/x86/../external-module-compat-comm.h:14,
                 from
/home/robot/rpm/BUILD/kvm-83/kernel/x86/external-module-compat.h:9,
                 from <command line>:1:
include/asm/msr.h:190:1: warning: this is the location of the previous
definition
In file included from
/home/robot/rpm/BUILD/kvm-83/kernel/include/asm/kvm_host.h:65,
                 from
/home/robot/rpm/BUILD/kvm-83/kernel/include/linux/kvm_host.h:67,
                 from /home/robot/rpm/BUILD/kvm-83/kernel/x86/lapic.c:60:
/home/robot/rpm/BUILD/kvm-83/kernel/include-compat/asm/msr-index.h:31:1:
warning: "MSR_IA32_PERFCTR1" redefined
In file included from include/asm/processor.h:16,
                 from
/home/robot/rpm/BUILD/kvm-83/kernel/include/asm/kvm_para.h:89,
                 from
/home/robot/rpm/BUILD/kvm-83/kernel/include/linux/kvm_para.h:63,
                 from
/home/robot/rpm/BUILD/kvm-83/kernel/x86/../external-module-compat-comm.h:14,
                 from
/home/robot/rpm/BUILD/kvm-83/kernel/x86/external-module-compat.h:9,
                 from <command line>:1:
-----------------------------------------


-- 
  Levente                               "Si vis pacem para bellum!"

Re: [ANNOUNCE] kvm-83 release

[John Wong]

Hi, i installed kvm-83 on xubuntu-amd64/kernel-2.6.7-11.
i notice some problem with -net downscript.
my kvm start up script like this:
qemu-system-x86_64 ... ... -net 
tap,ifname=tap0,script=/path/kvm-ifup,downscript=/path/kvm-ifdown
when the guest shutdown, kvm pass all argument to downscript.

the command like this (kvm-83):
/path/kvm-ifdown "tap0,script=/path/kvm-ifup,downscript=/path/kvm-ifdown"

kvm-82 only pass tap0 to downscript like this:
/path/kvm-ifdown "tap0"

and i notice both kvm82/kvm83 not installed modules to 
/lib/modules/2.6.27-11-generic/kernel/arch/x86/kvm/{kvm-amd,kvm-intel,kvm}.ko
it only rename the old modules to {kvm,kvm-intel,kvm-amd}.ko.org.

please help, thank you.


On 2009年01月13日 21:53, Avi Kivity wrote:
> The biggest change this time around is support for the AMD IOMMU for 
> device assignment.  As it requires extensive core kernel support, it 
> is only available on Linux 2.6.29rc1 or higher hosts.
>
> Changes from kvm-82:
> - amd iommu support (with Linux 2.6.29+ hosts) (Joerg Roedel)
> - merge qemu-svn
>   - qcow2 correctness fixes
>   - '-drive serial=...' option
>   - virtio-net migration fixes
> - fix '--with-patched-kernel' build on Ubuntu (Nolan Leake)
> - improved tsc handling (Marcelo Tosatti)
>   - should reduce timing errors on older Linux guests
> - remove cpuid code duplication due to upstream qemu work (Amit Shah)
> - guest S3 sleep (Gleb Natapov)
> - load assigned devices option rom (Kechao Liu)
> - '-no-kvm-pit-reinjection' option to improve timing on RHEL 3 era guests
>   (Marcelo Tosatti)
> - remove duplicate firmware config interface probe in bios (Sebastian 
> Herbszt)
> - backward compatibility for marker_synchronize_register() (Eduardo 
> Habkost)
> - fix compile warning in tpr patching (Mark McLoughlin)
> - don't install kvm headers to /usr/local
> - fix smbios memory size in certain cases (Bill Rieske)
> - simplify mmu shadow walking function
> - stop using vmap() on guest pages (annoys ksm) (Izik Eidus)
> - fix warning on MSR_VM_HSAVE_PA in dmesg
> - fix i8254 timer count read (Marcelo Tosatti)
> - ppc e500 core support (Hollis Blanchard)
> - fix vmx real mode guest state validity checks
> - make pop instruction emulation a little more generic
> - implement 'ret far' instruction emulation
> - properly lock i8354 timer
> - remove ppc debug support (Hollis Blanchard)
> - fix race freeing assigned device interrupt (Sheng Yang)
> - fix pci interrupt in freeing in device assignment (Sheng Yang)
> - allow disabling msi for assigned devices (Sheng Yang)
> - emulate amd microcode level msr (Alexander Graf)
>   - helps esx boot on kvm
> - make page unshadowing a little faster
> - clear pending pit interrupts when unmasking the timer irq
>   - fixes xen-on-kvm
> - mmu notifiers release method (Marcelo Tosatti)
>   - fixes large pages support
> - expose ia64 registers to userspace (Jes Sorensen)
> - fix type in has_svm() (Joe Perches)
> - drop unnecessary memory clearing when allocating shadow page
>   (Marcelo Tosatti)
> - rename "metaphysical" page attribute to "direct"
> - fix irq routing with -no-kernel-irqchip
>
>
> Notes:
>     If you use the modules bundled with kvm-83, you can use any version
> of Linux from 2.6.16 upwards.  You may also use kvm-83 userspace with
> the kvm modules provided by Linux 2.6.25 or above.  Some features may
> only be available in newer releases.
>
> http://kvm.qumranet.com
>

Re: [ANNOUNCE] kvm-83 release

[Mark McLoughlin]

Hi John,

On Wed, 2009-01-14 at 00:15 +0800, John Wong wrote:
> Hi, i installed kvm-83 on xubuntu-amd64/kernel-2.6.7-11.
> i notice some problem with -net downscript.
> my kvm start up script like this:
> qemu-system-x86_64 ... ... -net 
> tap,ifname=tap0,script=/path/kvm-ifup,downscript=/path/kvm-ifdown
> when the guest shutdown, kvm pass all argument to downscript.
> 
> the command like this (kvm-83):
> /path/kvm-ifdown
> "tap0,script=/path/kvm-ifup,downscript=/path/kvm-ifdown"
> 
> kvm-82 only pass tap0 to downscript like this:
> /path/kvm-ifdown "tap0"

Thanks for the report. Can you try this simple change and see if it
fixes it for you?

--- qemu/net.c
+++ qemu/net.c
@@ -1740,7 +1740,7 @@ void net_cleanup(void)
                 TAPState *s = vc->opaque;
 
                 if (strcmp(vc->model, "tap") == 0 &&
-                    sscanf(vc->info_str, "ifname=%63s ", ifname) == 1 &&
+                    sscanf(vc->info_str, "ifname=%63s,", ifname) == 1 &&
                     s->down_script[0])
                     launch_script(s->down_script, ifname, s->fd);
             }

Thanks,
Mark.

Re: [ANNOUNCE] kvm-83 release

[John Wong]

Hi Mark, i want to try, but i don't know where can download snapshot
source code,
i'm long time no use linux.
Where can i download kvm snapshot source code? kernel.org? or?

On 2009年01月14日 01:27, Mark McLoughlin wrote:
> Hi John,
>
> On Wed, 2009-01-14 at 00:15 +0800, John Wong wrote:
>   
>> Hi, i installed kvm-83 on xubuntu-amd64/kernel-2.6.7-11.
>> i notice some problem with -net downscript.
>> my kvm start up script like this:
>> qemu-system-x86_64 ... ... -net 
>> tap,ifname=tap0,script=/path/kvm-ifup,downscript=/path/kvm-ifdown
>> when the guest shutdown, kvm pass all argument to downscript.
>>
>> the command like this (kvm-83):
>> /path/kvm-ifdown
>> "tap0,script=/path/kvm-ifup,downscript=/path/kvm-ifdown"
>>
>> kvm-82 only pass tap0 to downscript like this:
>> /path/kvm-ifdown "tap0"
>>     
>
> Thanks for the report. Can you try this simple change and see if it
> fixes it for you?
>
> --- qemu/net.c
> +++ qemu/net.c
> @@ -1740,7 +1740,7 @@ void net_cleanup(void)
>                  TAPState *s = vc->opaque;
>  
>                  if (strcmp(vc->model, "tap") == 0 &&
> -                    sscanf(vc->info_str, "ifname=%63s ", ifname) == 1 &&
> +                    sscanf(vc->info_str, "ifname=%63s,", ifname) == 1 &&
>                      s->down_script[0])
>                      launch_script(s->down_script, ifname, s->fd);
>              }
>
> Thanks,
> Mark.
>
> --
> 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-83 release

[Amit Shah]

On Wed, Jan 14, 2009 at 01:40:27AM +0800, John Wong wrote:
> Hi Mark, i want to try, but i don't know where can download snapshot
> source code,
> i'm long time no use linux.
> Where can i download kvm snapshot source code? kernel.org? or?

Hi John,

You apply this patch on top of the kvm-83 release that you've already
downloaded. You don't need anything else.

Amit.

Re: [ANNOUNCE] kvm-83 release

[Michael Tokarev]

Mark McLoughlin wrote:
> Hi John,
> 
> On Wed, 2009-01-14 at 00:15 +0800, John Wong wrote:
>> Hi, i installed kvm-83 on xubuntu-amd64/kernel-2.6.7-11.
>> i notice some problem with -net downscript.
>> my kvm start up script like this:
>> qemu-system-x86_64 ... ... -net 
>> tap,ifname=tap0,script=/path/kvm-ifup,downscript=/path/kvm-ifdown
>> when the guest shutdown, kvm pass all argument to downscript.
>>
>> the command like this (kvm-83):
>> /path/kvm-ifdown
>> "tap0,script=/path/kvm-ifup,downscript=/path/kvm-ifdown"
>>
>> kvm-82 only pass tap0 to downscript like this:
>> /path/kvm-ifdown "tap0"
> 
> Thanks for the report. Can you try this simple change and see if it
> fixes it for you?
> 
> --- qemu/net.c
> +++ qemu/net.c
> @@ -1740,7 +1740,7 @@ void net_cleanup(void)
>                  TAPState *s = vc->opaque;
>  
>                  if (strcmp(vc->model, "tap") == 0 &&
> -                    sscanf(vc->info_str, "ifname=%63s ", ifname) == 1 &&
> +                    sscanf(vc->info_str, "ifname=%63s,", ifname) == 1 &&

Hmm.  What if ifname=foo is the last parameter?

/mjt

Re: [ANNOUNCE] kvm-83 release

[Michael Tokarev]

Michael Tokarev wrote:
> Mark McLoughlin wrote:
[]
>> --- qemu/net.c
>> +++ qemu/net.c
>> @@ -1740,7 +1740,7 @@ void net_cleanup(void)
>>                  TAPState *s = vc->opaque;
>>  
>>                  if (strcmp(vc->model, "tap") == 0 &&
>> -                    sscanf(vc->info_str, "ifname=%63s ", ifname) == 1 &&
>> +                    sscanf(vc->info_str, "ifname=%63s,", ifname) == 1 &&
> 
> Hmm.  What if ifname=foo is the last parameter?

And while we're at it.. Why the parsing isn't done like this:

  for(tok = strtok(arg, ","); tok; tok = strtok(NULL, ",")) {
    if (strncmp(tok, "ifname=", 7) == 0)
      strcpy(ifname, tok+7);
    else if ...
  }

-- here and everywhere else?

/mjt

Re: [ANNOUNCE] kvm-83 release

[Anthony Liguori]

Michael Tokarev wrote:
> Michael Tokarev wrote:
>   
>> Mark McLoughlin wrote:
>>     
> []
>   
>>> --- qemu/net.c
>>> +++ qemu/net.c
>>> @@ -1740,7 +1740,7 @@ void net_cleanup(void)
>>>                  TAPState *s = vc->opaque;
>>>  
>>>                  if (strcmp(vc->model, "tap") == 0 &&
>>> -                    sscanf(vc->info_str, "ifname=%63s ", ifname) == 1 &&
>>> +                    sscanf(vc->info_str, "ifname=%63s,", ifname) == 1 &&
>>>       
>> Hmm.  What if ifname=foo is the last parameter?
>>     
>
> And while we're at it.. Why the parsing isn't done like this:
>
>   for(tok = strtok(arg, ","); tok; tok = strtok(NULL, ",")) {
>     if (strncmp(tok, "ifname=", 7) == 0)
>       strcpy(ifname, tok+7);
>     else if ...
>   }
>   

Because strtok is extremely evil.

Regards,

Anthony Liguori

> -- here and everywhere else?
>
> /mjt
> --
> 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-83 release

[Michael Tokarev]

Anthony Liguori wrote:
>> Michael Tokarev wrote:
...
>>>>        if (strcmp(vc->model, "tap") == 0 &&
>>>> -          sscanf(vc->info_str, "ifname=%63s ", ifname) == 1 &&
>>>> +          sscanf(vc->info_str, "ifname=%63s,", ifname) == 1 &&
>>
>> And while we're at it.. Why the parsing isn't done like this:
>>
>>   for(tok = strtok(arg, ","); tok; tok = strtok(NULL, ",")) {
>>     if (strncmp(tok, "ifname=", 7) == 0)
>>       strcpy(ifname, tok+7);
>>     else if ...
>>   }
> 
> Because strtok is extremely evil.

I didn't mean strtok, but the way -- something LIKE that, i.e,
by splitting the string into args and parsing each in turn.
There's getsubopt(3) too, which does exactly that, and which
is used by, say, mount(8) to parse mount options (-o foo=bar
things).  Those home-grown sscanfs tend to become buggy like
in this example...

And if you dislike strtok especially, there are other alternatives.
And speaking of strtok, there's nothing evil in it provided you
don't run another strtok sequence while this one is running
(it's not reenterant, so to say).  Here, let it modify the
existing string and replace delimiters with zeros to terminate
our tokens, -- nothing wrong with that.  It's not worse than
silently truncating ifname size to 63 chars first and to
IF_NAMESIZE second.  If you don't want to modify original
string, strdup() or astrdup() are our friends...  But that's
all details, details, not related to the original issue and
question.

/mjt

Re: [ANNOUNCE] kvm-83 release

[Mark McLoughlin]

On Wed, 2009-01-14 at 00:01 +0300, Michael Tokarev wrote:
> Anthony Liguori wrote:
> >> Michael Tokarev wrote:
> ...
> >>>>        if (strcmp(vc->model, "tap") == 0 &&
> >>>> -          sscanf(vc->info_str, "ifname=%63s ", ifname) == 1 &&
> >>>> +          sscanf(vc->info_str, "ifname=%63s,", ifname) == 1 &&
> >>
> >> And while we're at it.. Why the parsing isn't done like this:
> >>
> >>   for(tok = strtok(arg, ","); tok; tok = strtok(NULL, ",")) {
> >>     if (strncmp(tok, "ifname=", 7) == 0)
> >>       strcpy(ifname, tok+7);
> >>     else if ...
> >>   }
> > 
> > Because strtok is extremely evil.
> 
> I didn't mean strtok, but the way -- something LIKE that, i.e,
> by splitting the string into args and parsing each in turn.

It shouldn't be parsing anything at all, it's a stupid lame hack.

I fixed it with:

  http://git.kernel.org/?p=virt/qemu/qemu.git;a=commit;h=973cbd37ce

Cheers,
Mark.

Re: [ANNOUNCE] kvm-83 release

[Avi Kivity]

Farkas Levente wrote:
> Avi Kivity wrote:
>   
>> The biggest change this time around is support for the AMD IOMMU for
>> device assignment.  As it requires extensive core kernel support, it is
>> only available on Linux 2.6.29rc1 or higher hosts.
>>     
>
> status report on rhel/centos-5.2 host:
> - fedora-9 latest kernel now boot again
> - fedora-10 installer stop at "waiting for hardware..." so it's
> currently not able to upgrade fedora-9 guest neither from cdrom nor form
>  preupgrade (ie hard disk).
>   

Works here.  What's your host cpu type?  Does the error occur if you pin 
the guest to one host cpu?

> and these dozens of warning still comes during compilation on
> rhel/centos 5.2:
> -----------------------------------------
> In file included from
> /home/robot/rpm/BUILD/kvm-83/kernel/include/asm/kvm_host.h:65,
>                  from
> /home/robot/rpm/BUILD/kvm-83/kernel/include/linux/kvm_host.h:67,
>                  from /home/robot/rpm/BUILD/kvm-83/kernel/x86/lapic.c:60:
> /home/robot/rpm/BUILD/kvm-83/kernel/include-compat/asm/msr-index.h:30:1:
> warning: "MSR_IA32_PERFCTR0" redefined
>   

I've fixed those for kvm-84.


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

Re: [ANNOUNCE] kvm-83 release

[Farkas Levente]

Avi Kivity wrote:
> Farkas Levente wrote:
>> Avi Kivity wrote:
>>  
>>> The biggest change this time around is support for the AMD IOMMU for
>>> device assignment.  As it requires extensive core kernel support, it is
>>> only available on Linux 2.6.29rc1 or higher hosts.
>>>     
>>
>> status report on rhel/centos-5.2 host:
>> - fedora-9 latest kernel now boot again
>> - fedora-10 installer stop at "waiting for hardware..." so it's
>> currently not able to upgrade fedora-9 guest neither from cdrom nor form
>>  preupgrade (ie hard disk).
>>   
> 
> Works here.  What's your host cpu type?  Does the error occur if you pin
> the guest to one host cpu?

both the install/upgrade dvd and the preupgrade failed on this host with
guest-7 (but now all other guest are working):
- host:
  - Intel(R) Core(TM)2 Quad CPU Q6600  @ 2.40GHz
  - Intel S3000AHV
  - 8GB RAM
  - CentOS-5.2
  - kernel-2.6.18-92.1.22.el5 x86_64 64bit
- guest-1:
  - CentOS-5.2 - 4 vcpu
  - kernel-2.6.18-92.1.22.el5 i386 32bit
- guest-2:
  - CentOS-5.2 - 4 vcpu
  - kernel-2.6.18-92.1.22.el5 x86_64 64bit
- guest-3:
  - Mandrake-9 - 1 vcpu
  - kernel-2.4.19.16mdk-1-1mdk 32bit
- guest-4:
  - Mandrake-10 - 1 vcpu
  - kernel-2.6.14.2-p4-smp 32bit
- guest-5:
  - Windows XP Professional 32bit - 2 vcpu
- guest-7:
  - Fedora-9 - 4 vcpu
  - kernel-2.6.25.14-108.fc9.i686


-- 
  Levente                               "Si vis pacem para bellum!"