Xen devel - Mar 2012 - [PATCH] libxl: use qemu-xen with PV guests by default

qemu-xen offers better disk performances than qemu-xen-traditional
because it supports Linux native AIO.

Signed-off-by: Stefano Stabellini <stefano.stabellini@eu.citrix.com>
---
 tools/libxl/libxl_create.c |   11 ++++++++---
 1 files changed, 8 insertions(+), 3 deletions(-)

diff --git a/tools/libxl/libxl_create.c b/tools/libxl/libxl_create.c
index 8417661..dbff02c 100644
--- a/tools/libxl/libxl_create.c
+++ b/tools/libxl/libxl_create.c
@@ -71,9 +71,14 @@ int libxl__domain_build_info_setdefault(libxl__gc *gc,
         b_info->type != LIBXL_DOMAIN_TYPE_PV)
         return ERROR_INVAL;
 
-    if (!b_info->device_model_version)
-        b_info->device_model_version -           
LIBXL_DEVICE_MODEL_VERSION_QEMU_XEN_TRADITIONAL;
+    if (!b_info->device_model_version) {
+        if (b_info->type == LIBXL_DOMAIN_TYPE_HVM)
+            b_info->device_model_version +               
LIBXL_DEVICE_MODEL_VERSION_QEMU_XEN_TRADITIONAL;
+        else
+            b_info->device_model_version +               
LIBXL_DEVICE_MODEL_VERSION_QEMU_XEN;
+    }
 
     if (b_info->type == LIBXL_DOMAIN_TYPE_HVM) {
         if (!b_info->u.hvm.bios)
-- 
1.7.2.5

On Fri, 2012-03-23 at 14:52 +0000, Stefano Stabellini
wrote:
> qemu-xen offers better disk performances than qemu-xen-traditional
> because it supports Linux native AIO.
The qemu-upstream test flights are failing far more often than not. In
fact I expect the *single* pass (actually a "tolerable fail")
we''ve ever
had was actually a fluke. Is qemu-xen really ready to be used in this
way?

Ian.
> Signed-off-by: Stefano Stabellini <stefano.stabellini@eu.citrix.com>
> ---
>  tools/libxl/libxl_create.c |   11 ++++++++---
>  1 files changed, 8 insertions(+), 3 deletions(-)
> 
> diff --git a/tools/libxl/libxl_create.c b/tools/libxl/libxl_create.c
> index 8417661..dbff02c 100644
> --- a/tools/libxl/libxl_create.c
> +++ b/tools/libxl/libxl_create.c
> @@ -71,9 +71,14 @@ int libxl__domain_build_info_setdefault(libxl__gc *gc,
>          b_info->type != LIBXL_DOMAIN_TYPE_PV)
>          return ERROR_INVAL;
>  
> -    if (!b_info->device_model_version)
> -        b_info->device_model_version > -           
LIBXL_DEVICE_MODEL_VERSION_QEMU_XEN_TRADITIONAL;
> +    if (!b_info->device_model_version) {
> +        if (b_info->type == LIBXL_DOMAIN_TYPE_HVM)
> +            b_info->device_model_version > +               
LIBXL_DEVICE_MODEL_VERSION_QEMU_XEN_TRADITIONAL;
> +        else
> +            b_info->device_model_version > +               
LIBXL_DEVICE_MODEL_VERSION_QEMU_XEN;
> +    }
>  
>      if (b_info->type == LIBXL_DOMAIN_TYPE_HVM) {
>          if (!b_info->u.hvm.bios)

On Mon, 26 Mar 2012, Ian Campbell wrote:
> On Fri, 2012-03-23 at 14:52 +0000, Stefano Stabellini wrote:
> > qemu-xen offers better disk performances than qemu-xen-traditional
> > because it supports Linux native AIO.
> 
> The qemu-upstream test flights are failing far more often than not. In
> fact I expect the *single* pass (actually a "tolerable fail")
we''ve ever
> had was actually a fluke. Is qemu-xen really ready to be used in this
> way?
Actually from the test reports it looks like that''s only true for HVM
guests (especially Windows). This change only affects PV guests.

Stefano Stabellini writes ("[Xen-devel] [PATCH] libxl: use qemu-xen with PV
guests by default"):
> qemu-xen offers better disk performances than qemu-xen-traditional
> because it supports Linux native AIO.
Does this take any account of whether the new qemu-xen dm is
installed ?

Ian.

On Tue, 2012-04-17 at 18:17 +0100, Ian Jackson wrote:
> Stefano Stabellini writes ("[Xen-devel] [PATCH] libxl: use qemu-xen
with PV guests by default"):
> > qemu-xen offers better disk performances than qemu-xen-traditional
> > because it supports Linux native AIO.
> 
> Does this take any account of whether the new qemu-xen dm is
> installed ?
I don''t think it is possible to install xen-unstable without
qemu-upstream-xen getting built and installed? At least not building
from source, perhaps a distro might package them separately but there
really ought to be a depends relationship in that case? Would we want to
support a Xen without qemu-upstream-xen configuration?

Ian.

On Wed, 2012-04-18 at 16:35 +0100, Stefano Stabellini
wrote:
> On Wed, 18 Apr 2012, Ian Campbell wrote:
> > On Tue, 2012-04-17 at 18:17 +0100, Ian Jackson wrote:
> > > Stefano Stabellini writes ("[Xen-devel] [PATCH] libxl: use
qemu-xen with PV guests by default"):
> > > > qemu-xen offers better disk performances than
qemu-xen-traditional
> > > > because it supports Linux native AIO.
> > > 
> > > Does this take any account of whether the new qemu-xen dm is
> > > installed ?
> > 
> > I don''t think it is possible to install xen-unstable without
> > qemu-upstream-xen getting built and installed? At least not building
> > from source, perhaps a distro might package them separately but there
> > really ought to be a depends relationship in that case?
> 
> I would think so.
> 
> > Would we want to
> > support a Xen without qemu-upstream-xen configuration?
> 
> What if we check for the existence of the file in libxl?
> 
> If this is not acceptable I vote for assuming that qemu-xen is installed
> in the system.
I think an access(R_X) check or whatever + log and ERROR_INVAL would be
find.

Ian.

On Wed, 18 Apr 2012, Ian Campbell wrote:
> On Tue, 2012-04-17 at 18:17 +0100, Ian Jackson wrote:
> > Stefano Stabellini writes ("[Xen-devel] [PATCH] libxl: use
qemu-xen with PV guests by default"):
> > > qemu-xen offers better disk performances than
qemu-xen-traditional
> > > because it supports Linux native AIO.
> > 
> > Does this take any account of whether the new qemu-xen dm is
> > installed ?
> 
> I don''t think it is possible to install xen-unstable without
> qemu-upstream-xen getting built and installed? At least not building
> from source, perhaps a distro might package them separately but there
> really ought to be a depends relationship in that case?
I would think so.
> Would we want to
> support a Xen without qemu-upstream-xen configuration?
What if we check for the existence of the file in libxl?

If this is not acceptable I vote for assuming that qemu-xen is installed
in the system.