Re: [systemd-devel] [PATCH 1/2] detect-virt: detect in best-heuristic order

[Christian Hesse]

Mantas Mikulėnas  on Wed, 2015/11/04 15:55:
> On Wed, Nov 4, 2015 at 3:52 PM, Lennart Poettering 
> wrote:
> 
> > On Wed, 04.11.15 15:54, Andrei Borzenkov (arvidj...@gmail.com) wrote:
> >  
> > > 04.11.2015 00:04, Andrew Jones пишет:  
> > > >afaict, this will fix a regression caused by commit 75f86906c5.
> > > >Where we used to report "kvm" before that patch, without this patch,
> > > >we would only report "qemu".  
> > >
> > > Are you sure it is regression? QEMU is a program (platform) while KVM is
> > > technology. Modern VirtualBox can use KVM as PV interface but it still
> > > remains VirtualBox. QEMU may use KVM as PV interface but it still
> > > remains QEMU.
> > >
> > > Where does it matter? Is anything broken because of this change?  
> >
> > I am pretty sure that vbox should be reported as vbox even if it uses
> > kvm as backend. qemu-kvm should be reported as kvm, and any other qemu
> > as qemu.
> >  
> 
> As I understand it, VirtualBox doesn't use KVM as *backend*; it only
> exposes a KVM-like paravirt interface to *guests*.

It does use kvm since version 5.0.
-- 
main(a){char*c=/*Schoene Gruesse */"B?IJj;MEH"
"CX:;",b;for(a/*Chris   get my mail address:*/=0;b=c[a++];)
putchar(b-1/(/*   gcc -o sig sig.c && ./sig*/b/42*2-3)*42);}


pgpHcb6_mOMPF.pgp
Description: OpenPGP digital signature
___
systemd-devel mailing list
systemd-devel@lists.freedesktop.org
http://lists.freedesktop.org/mailman/listinfo/systemd-devel

Re: [systemd-devel] [PATCH 1/2] detect-virt: detect in best-heuristic order

[poma]

On 04.11.2015 14:52, Lennart Poettering wrote:
> On Wed, 04.11.15 15:54, Andrei Borzenkov (arvidj...@gmail.com) wrote:
> 
>> 04.11.2015 00:04, Andrew Jones пишет:
>>> afaict, this will fix a regression caused by commit 75f86906c5.
>>> Where we used to report "kvm" before that patch, without this patch,
>>> we would only report "qemu".
>>
>> Are you sure it is regression? QEMU is a program (platform) while KVM is
>> technology. Modern VirtualBox can use KVM as PV interface but it still
>> remains VirtualBox. QEMU may use KVM as PV interface but it still remains
>> QEMU.
>>
>> Where does it matter? Is anything broken because of this change?
> 
> I am pretty sure that vbox should be reported as vbox even if it uses
> kvm as backend. qemu-kvm should be reported as kvm, and any other qemu
> as qemu.
> 
> Yeah, it's not fully symmetric, but I am pretty sure that's what most
> people would expect.
> 
> Lennart
> 


Comparing with 'virt-what' it is not symmetrical, yeah.

= QEMU =

- HOST:
# /usr/bin/qemu-system-x86_64 -m 2048 -smp 2 -drive 
file=/iso/Rawhide-Xfce-Live-1104.iso,format=raw

- GUEST:
# dmesg -t | grep DMI
DMI: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.8.1-20150318_183358- 
04/01/2014

# systemd-detect-virt 
vm-other

# virt-what 
qemu

~

= QEMU-KVM =

- HOST:
# /usr/bin/qemu-system-x86_64 -machine accel=kvm -m 2048 -smp 2 -drive 
file=/iso/Rawhide-Xfce-Live-1104.iso,format=raw

- GUEST:
# dmesg -t | grep DMI
DMI: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.8.1-20150318_183358- 
04/01/2014

# systemd-detect-virt 
kvm

# virt-what 
kvm

~

/proc/cmdline 
BOOT_IMAGE=vmlinuz0 initrd=initrd0.img root=live:CDLABEL=Rawhide-Xfce-Live-1104 
rootfstype=auto ro rd.live.image console=ttyS0 console=tty0 3

SW:
kernel-core-4.3.0-1.fc24.x86_64
virt-what-1.15-2.fc23.x86_64
systemd-227-4.gita6bff4a.fc24.x86_64


Ref.
http://pkgs.fedoraproject.org/cgit/virt-what.git/tree/virt-what.spec#n37
http://people.redhat.com/~rjones/virt-what


___
systemd-devel mailing list
systemd-devel@lists.freedesktop.org
http://lists.freedesktop.org/mailman/listinfo/systemd-devel

Re: [systemd-devel] [PATCH 1/2] detect-virt: detect in best-heuristic order

[Andrei Borzenkov]

04.11.2015 00:04, Andrew Jones пишет:

afaict, this will fix a regression caused by commit 75f86906c5.
Where we used to report "kvm" before that patch, without this patch,
we would only report "qemu".


Are you sure it is regression? QEMU is a program (platform) while KVM is 
technology. Modern VirtualBox can use KVM as PV interface but it still 
remains VirtualBox. QEMU may use KVM as PV interface but it still 
remains QEMU.


Where does it matter? Is anything broken because of this change?


The reason is because cpuid detection
must come before dmi detection. Also, both can safely come before
other xen heuristics.
---
  src/basic/virt.c | 16 +---
  1 file changed, 9 insertions(+), 7 deletions(-)

diff --git a/src/basic/virt.c b/src/basic/virt.c
index fb181e5b55ee5..1e10fc755f201 100644
--- a/src/basic/virt.c
+++ b/src/basic/virt.c
@@ -267,12 +267,7 @@ int detect_vm(void) {
  if (cached_found >= 0)
  return cached_found;

-/* Try xen capabilities file first, if not found try
- * high-level hypervisor sysfs file:
- *
- * https://bugs.freedesktop.org/show_bug.cgi?id=77271 */
-
-r = detect_vm_xen();
+r = detect_vm_cpuid();
  if (r < 0)
  return r;
  if (r != VIRTUALIZATION_NONE)
@@ -284,7 +279,14 @@ int detect_vm(void) {
  if (r != VIRTUALIZATION_NONE)
  goto finish;

-r = detect_vm_cpuid();
+/* x86 xen will most likely be detected by cpuid. If not (most likely
+ * because we're not an x86 guest), then we should try the xen 
capabilities
+ * file next. If that's not found, then we check for the high-level
+ * hypervisor sysfs file:
+ *
+ * https://bugs.freedesktop.org/show_bug.cgi?id=77271 */
+
+r = detect_vm_xen();
  if (r < 0)
  return r;
  if (r != VIRTUALIZATION_NONE)



___
systemd-devel mailing list
systemd-devel@lists.freedesktop.org
http://lists.freedesktop.org/mailman/listinfo/systemd-devel

Re: [systemd-devel] [PATCH 1/2] detect-virt: detect in best-heuristic order

[Andrew Jones]

On Wed, Nov 04, 2015 at 03:54:45PM +0300, Andrei Borzenkov wrote:
> 04.11.2015 00:04, Andrew Jones пишет:
> >afaict, this will fix a regression caused by commit 75f86906c5.
> >Where we used to report "kvm" before that patch, without this patch,
> >we would only report "qemu".
> 
> Are you sure it is regression? QEMU is a program (platform) while KVM is
> technology. Modern VirtualBox can use KVM as PV interface but it still
> remains VirtualBox. QEMU may use KVM as PV interface but it still remains
> QEMU.

Sounds like VirtualBox should override the hypervisor cpuid leaf with
it's own signature, like kvmtool does.

> 
> Where does it matter? Is anything broken because of this change?

It's true that programs shouldn't change their behavior when detecting
"kvm" vs. "qemu", as "kvm" just means accelerated "qemu", but "shouldn't"
isn't something we should count on. Changing a behavior that some users
may now expect is a regression.

Thanks,
drew

> 
> >The reason is because cpuid detection
> >must come before dmi detection. Also, both can safely come before
> >other xen heuristics.
> >---
> >  src/basic/virt.c | 16 +---
> >  1 file changed, 9 insertions(+), 7 deletions(-)
> >
> >diff --git a/src/basic/virt.c b/src/basic/virt.c
> >index fb181e5b55ee5..1e10fc755f201 100644
> >--- a/src/basic/virt.c
> >+++ b/src/basic/virt.c
> >@@ -267,12 +267,7 @@ int detect_vm(void) {
> >  if (cached_found >= 0)
> >  return cached_found;
> >
> >-/* Try xen capabilities file first, if not found try
> >- * high-level hypervisor sysfs file:
> >- *
> >- * https://bugs.freedesktop.org/show_bug.cgi?id=77271 */
> >-
> >-r = detect_vm_xen();
> >+r = detect_vm_cpuid();
> >  if (r < 0)
> >  return r;
> >  if (r != VIRTUALIZATION_NONE)
> >@@ -284,7 +279,14 @@ int detect_vm(void) {
> >  if (r != VIRTUALIZATION_NONE)
> >  goto finish;
> >
> >-r = detect_vm_cpuid();
> >+/* x86 xen will most likely be detected by cpuid. If not (most 
> >likely
> >+ * because we're not an x86 guest), then we should try the xen 
> >capabilities
> >+ * file next. If that's not found, then we check for the high-level
> >+ * hypervisor sysfs file:
> >+ *
> >+ * https://bugs.freedesktop.org/show_bug.cgi?id=77271 */
> >+
> >+r = detect_vm_xen();
> >  if (r < 0)
> >  return r;
> >  if (r != VIRTUALIZATION_NONE)
> >
> 
___
systemd-devel mailing list
systemd-devel@lists.freedesktop.org
http://lists.freedesktop.org/mailman/listinfo/systemd-devel

Re: [systemd-devel] [PATCH 1/2] detect-virt: detect in best-heuristic order

[Lennart Poettering]

On Wed, 04.11.15 15:54, Andrei Borzenkov (arvidj...@gmail.com) wrote:

> 04.11.2015 00:04, Andrew Jones пишет:
> >afaict, this will fix a regression caused by commit 75f86906c5.
> >Where we used to report "kvm" before that patch, without this patch,
> >we would only report "qemu".
> 
> Are you sure it is regression? QEMU is a program (platform) while KVM is
> technology. Modern VirtualBox can use KVM as PV interface but it still
> remains VirtualBox. QEMU may use KVM as PV interface but it still remains
> QEMU.
> 
> Where does it matter? Is anything broken because of this change?

I am pretty sure that vbox should be reported as vbox even if it uses
kvm as backend. qemu-kvm should be reported as kvm, and any other qemu
as qemu.

Yeah, it's not fully symmetric, but I am pretty sure that's what most
people would expect.

Lennart

-- 
Lennart Poettering, Red Hat
___
systemd-devel mailing list
systemd-devel@lists.freedesktop.org
http://lists.freedesktop.org/mailman/listinfo/systemd-devel

Re: [systemd-devel] [PATCH 1/2] detect-virt: detect in best-heuristic order

[Mantas Mikulėnas]

On Wed, Nov 4, 2015 at 3:52 PM, Lennart Poettering 
wrote:

> On Wed, 04.11.15 15:54, Andrei Borzenkov (arvidj...@gmail.com) wrote:
>
> > 04.11.2015 00:04, Andrew Jones пишет:
> > >afaict, this will fix a regression caused by commit 75f86906c5.
> > >Where we used to report "kvm" before that patch, without this patch,
> > >we would only report "qemu".
> >
> > Are you sure it is regression? QEMU is a program (platform) while KVM is
> > technology. Modern VirtualBox can use KVM as PV interface but it still
> > remains VirtualBox. QEMU may use KVM as PV interface but it still remains
> > QEMU.
> >
> > Where does it matter? Is anything broken because of this change?
>
> I am pretty sure that vbox should be reported as vbox even if it uses
> kvm as backend. qemu-kvm should be reported as kvm, and any other qemu
> as qemu.
>

As I understand it, VirtualBox doesn't use KVM as *backend*; it only
exposes a KVM-like paravirt interface to *guests*.

-- 
Mantas Mikulėnas 
___
systemd-devel mailing list
systemd-devel@lists.freedesktop.org
http://lists.freedesktop.org/mailman/listinfo/systemd-devel

[systemd-devel] [PATCH 1/2] detect-virt: detect in best-heuristic order

[Andrew Jones]

afaict, this will fix a regression caused by commit 75f86906c5.
Where we used to report "kvm" before that patch, without this patch,
we would only report "qemu". The reason is because cpuid detection
must come before dmi detection. Also, both can safely come before
other xen heuristics.
---
 src/basic/virt.c | 16 +---
 1 file changed, 9 insertions(+), 7 deletions(-)

diff --git a/src/basic/virt.c b/src/basic/virt.c
index fb181e5b55ee5..1e10fc755f201 100644
--- a/src/basic/virt.c
+++ b/src/basic/virt.c
@@ -267,12 +267,7 @@ int detect_vm(void) {
 if (cached_found >= 0)
 return cached_found;
 
-/* Try xen capabilities file first, if not found try
- * high-level hypervisor sysfs file:
- *
- * https://bugs.freedesktop.org/show_bug.cgi?id=77271 */
-
-r = detect_vm_xen();
+r = detect_vm_cpuid();
 if (r < 0)
 return r;
 if (r != VIRTUALIZATION_NONE)
@@ -284,7 +279,14 @@ int detect_vm(void) {
 if (r != VIRTUALIZATION_NONE)
 goto finish;
 
-r = detect_vm_cpuid();
+/* x86 xen will most likely be detected by cpuid. If not (most likely
+ * because we're not an x86 guest), then we should try the xen 
capabilities
+ * file next. If that's not found, then we check for the high-level
+ * hypervisor sysfs file:
+ *
+ * https://bugs.freedesktop.org/show_bug.cgi?id=77271 */
+
+r = detect_vm_xen();
 if (r < 0)
 return r;
 if (r != VIRTUALIZATION_NONE)
-- 
2.4.3

___
systemd-devel mailing list
systemd-devel@lists.freedesktop.org
http://lists.freedesktop.org/mailman/listinfo/systemd-devel