[PATCH] xl.cfg: document the maxmem= option

classic Classic list List threaded Threaded
8 messages Options
Reply | Threaded
Open this post in threaded view
|

[PATCH] xl.cfg: document the maxmem= option

Olaf Hering-2
# HG changeset patch
# User Olaf Hering <[hidden email]>
# Date 1336159720 -7200
# Node ID c414728d0d12f1c3e416e40cceefca2f0b00578e
# Parent  8f556a70ae0bef47e242f9e7be0a054769fc8277
xl.cfg: document the maxmem= option

Signed-off-by: Olaf Hering <[hidden email]>

diff -r 8f556a70ae0b -r c414728d0d12 docs/man/xl.cfg.pod.5
--- a/docs/man/xl.cfg.pod.5
+++ b/docs/man/xl.cfg.pod.5
@@ -484,6 +484,14 @@ are not using hardware assisted paging (
 mode) and your guest workload consists of a a very large number of
 similar processes then increasing this value may improve performance.
 
+=item B<maxmem=MBYTES>
+
+Specifies the maximum amount of memory the HVM guest can ever see.
+In combination with the memory= option it will enable the PoD (populate
+on demand) mode for the guest iff the values of memory= and maxmem=
+differ. The guest needs a balloon driver in this case. The value of
+maxmem= must be equal or greater than memory=.
+
 =back
 
 =head3 Processor and Platform Features
@@ -955,10 +963,6 @@ XXX
 
 XXX
 
-=item B<maxmem=NUMBER>
-
-XXX
-
 =item B<nodes=XXX>
 
 XXX

_______________________________________________
Xen-devel mailing list
[hidden email]
http://lists.xen.org/xen-devel
Reply | Threaded
Open this post in threaded view
|

Re: [PATCH] xl.cfg: document the maxmem= option

Konrad Rzeszutek Wilk
On Fri, May 04, 2012 at 09:31:42PM +0200, Olaf Hering wrote:

> # HG changeset patch
> # User Olaf Hering <[hidden email]>
> # Date 1336159720 -7200
> # Node ID c414728d0d12f1c3e416e40cceefca2f0b00578e
> # Parent  8f556a70ae0bef47e242f9e7be0a054769fc8277
> xl.cfg: document the maxmem= option
>
> Signed-off-by: Olaf Hering <[hidden email]>
>
> diff -r 8f556a70ae0b -r c414728d0d12 docs/man/xl.cfg.pod.5
> --- a/docs/man/xl.cfg.pod.5
> +++ b/docs/man/xl.cfg.pod.5
> @@ -484,6 +484,14 @@ are not using hardware assisted paging (
>  mode) and your guest workload consists of a a very large number of
>  similar processes then increasing this value may improve performance.
>  
> +=item B<maxmem=MBYTES>
> +
> +Specifies the maximum amount of memory the HVM guest can ever see.

Isn't it also for PV guests (though without the PoD functionality)?

> +In combination with the memory= option it will enable the PoD (populate
> +on demand) mode for the guest iff the values of memory= and maxmem=
> +differ. The guest needs a balloon driver in this case. The value of
> +maxmem= must be equal or greater than memory=.
> +
>  =back
>  
>  =head3 Processor and Platform Features
> @@ -955,10 +963,6 @@ XXX
>  
>  XXX
>  
> -=item B<maxmem=NUMBER>
> -
> -XXX
> -
>  =item B<nodes=XXX>
>  
>  XXX
>
> _______________________________________________
> Xen-devel mailing list
> [hidden email]
> http://lists.xen.org/xen-devel

_______________________________________________
Xen-devel mailing list
[hidden email]
http://lists.xen.org/xen-devel
Reply | Threaded
Open this post in threaded view
|

Re: [PATCH] xl.cfg: document the maxmem= option

Olaf Hering-2
On Fri, May 04, Konrad Rzeszutek Wilk wrote:

> On Fri, May 04, 2012 at 09:31:42PM +0200, Olaf Hering wrote:
> > # HG changeset patch
> > # User Olaf Hering <[hidden email]>
> > # Date 1336159720 -7200
> > # Node ID c414728d0d12f1c3e416e40cceefca2f0b00578e
> > # Parent  8f556a70ae0bef47e242f9e7be0a054769fc8277
> > xl.cfg: document the maxmem= option
> >
> > Signed-off-by: Olaf Hering <[hidden email]>
> >
> > diff -r 8f556a70ae0b -r c414728d0d12 docs/man/xl.cfg.pod.5
> > --- a/docs/man/xl.cfg.pod.5
> > +++ b/docs/man/xl.cfg.pod.5
> > @@ -484,6 +484,14 @@ are not using hardware assisted paging (
> >  mode) and your guest workload consists of a a very large number of
> >  similar processes then increasing this value may improve performance.
> >  
> > +=item B<maxmem=MBYTES>
> > +
> > +Specifies the maximum amount of memory the HVM guest can ever see.
>
> Isn't it also for PV guests (though without the PoD functionality)?

Now that I look again at libxl__build_pre(), it appearently does also
something for PV guests. I will resend the patch.
Thanks for the review.

Olaf

_______________________________________________
Xen-devel mailing list
[hidden email]
http://lists.xen.org/xen-devel
Reply | Threaded
Open this post in threaded view
|

Re: [PATCH] xl.cfg: document the maxmem= option

Ian Campbell-10
On Fri, 2012-05-04 at 16:18 -0400, Olaf Hering wrote:

> On Fri, May 04, Konrad Rzeszutek Wilk wrote:
>
> > On Fri, May 04, 2012 at 09:31:42PM +0200, Olaf Hering wrote:
> > > # HG changeset patch
> > > # User Olaf Hering <[hidden email]>
> > > # Date 1336159720 -7200
> > > # Node ID c414728d0d12f1c3e416e40cceefca2f0b00578e
> > > # Parent  8f556a70ae0bef47e242f9e7be0a054769fc8277
> > > xl.cfg: document the maxmem= option
> > >
> > > Signed-off-by: Olaf Hering <[hidden email]>
> > >
> > > diff -r 8f556a70ae0b -r c414728d0d12 docs/man/xl.cfg.pod.5
> > > --- a/docs/man/xl.cfg.pod.5
> > > +++ b/docs/man/xl.cfg.pod.5
> > > @@ -484,6 +484,14 @@ are not using hardware assisted paging (
> > >  mode) and your guest workload consists of a a very large number of
> > >  similar processes then increasing this value may improve performance.
> > >  
> > > +=item B<maxmem=MBYTES>
> > > +
> > > +Specifies the maximum amount of memory the HVM guest can ever see.
> >
> > Isn't it also for PV guests (though without the PoD functionality)?
>
> Now that I look again at libxl__build_pre(), it appearently does also
> something for PV guests.

Right. Some people call this booting "pre-ballooned". Essentially n PV
the balloon driver can be initialised early enough that something like
PoD is not required -- the missing pages are just incorporated directly
into the balloon.

(strictly speaking the balloon driver isn't actually initialised early,
but rather the pages are reserved/set-aside for it very early on).

Ian.



_______________________________________________
Xen-devel mailing list
[hidden email]
http://lists.xen.org/xen-devel
Reply | Threaded
Open this post in threaded view
|

Re: [PATCH] xl.cfg: document the maxmem= option

Olaf Hering-2
On Sun, May 06, Ian Campbell wrote:

> On Fri, 2012-05-04 at 16:18 -0400, Olaf Hering wrote:
> > On Fri, May 04, Konrad Rzeszutek Wilk wrote:
> >
> > > On Fri, May 04, 2012 at 09:31:42PM +0200, Olaf Hering wrote:
> > > > # HG changeset patch
> > > > # User Olaf Hering <[hidden email]>
> > > > # Date 1336159720 -7200
> > > > # Node ID c414728d0d12f1c3e416e40cceefca2f0b00578e
> > > > # Parent  8f556a70ae0bef47e242f9e7be0a054769fc8277
> > > > xl.cfg: document the maxmem= option
> > > >
> > > > Signed-off-by: Olaf Hering <[hidden email]>
> > > >
> > > > diff -r 8f556a70ae0b -r c414728d0d12 docs/man/xl.cfg.pod.5
> > > > --- a/docs/man/xl.cfg.pod.5
> > > > +++ b/docs/man/xl.cfg.pod.5
> > > > @@ -484,6 +484,14 @@ are not using hardware assisted paging (
> > > >  mode) and your guest workload consists of a a very large number of
> > > >  similar processes then increasing this value may improve performance.
> > > >  
> > > > +=item B<maxmem=MBYTES>
> > > > +
> > > > +Specifies the maximum amount of memory the HVM guest can ever see.
> > >
> > > Isn't it also for PV guests (though without the PoD functionality)?
> >
> > Now that I look again at libxl__build_pre(), it appearently does also
> > something for PV guests.
>
> Right. Some people call this booting "pre-ballooned". Essentially n PV
> the balloon driver can be initialised early enough that something like
> PoD is not required -- the missing pages are just incorporated directly
> into the balloon.


What about this wording?

Specifies the maximum amount of memory a guest can ever see.
The value of maxmem= must be equal or greater than memory=.
In combination with the memory= option it will start the guest "pre-ballooned".
In a HVM guest it will enable the PoD (populate on demand) mode, iff the values of memory= and maxmem= differ.
The guest needs a balloon driver in this case, without a balloon driver it will crash.


Olaf


_______________________________________________
Xen-devel mailing list
[hidden email]
http://lists.xen.org/xen-devel
Reply | Threaded
Open this post in threaded view
|

Re: [PATCH] xl.cfg: document the maxmem= option

Dan Magenheimer-2
In reply to this post by Ian Campbell-10
> From: Ian Campbell [mailto:[hidden email]]
> Sent: Sunday, May 06, 2012 3:36 AM
> To: Olaf Hering
> Cc: [hidden email]; Konrad Rzeszutek Wilk
> Subject: Re: [Xen-devel] [PATCH] xl.cfg: document the maxmem= option
>
> On Fri, 2012-05-04 at 16:18 -0400, Olaf Hering wrote:
> > On Fri, May 04, Konrad Rzeszutek Wilk wrote:
> >
> > > On Fri, May 04, 2012 at 09:31:42PM +0200, Olaf Hering wrote:
> > > > # HG changeset patch
> > > > # User Olaf Hering <[hidden email]>
> > > > # Date 1336159720 -7200
> > > > # Node ID c414728d0d12f1c3e416e40cceefca2f0b00578e
> > > > # Parent  8f556a70ae0bef47e242f9e7be0a054769fc8277
> > > > xl.cfg: document the maxmem= option
> > > >
> > > > Signed-off-by: Olaf Hering <[hidden email]>
> > > >
> > > > diff -r 8f556a70ae0b -r c414728d0d12 docs/man/xl.cfg.pod.5
> > > > --- a/docs/man/xl.cfg.pod.5
> > > > +++ b/docs/man/xl.cfg.pod.5
> > > > @@ -484,6 +484,14 @@ are not using hardware assisted paging (
> > > >  mode) and your guest workload consists of a a very large number of
> > > >  similar processes then increasing this value may improve performance.
> > > >
> > > > +=item B<maxmem=MBYTES>
> > > > +
> > > > +Specifies the maximum amount of memory the HVM guest can ever see.

can _ever_ see...

What about guest OS's that support hotplug memory?

Also, a subtle point, but is this the maximum physical address that
will contain read/write pages or the maximum amount of RAM that the
guest OS should be prepared to map as read/writeable (e.g. after
subtracting off reserved e820 ranges) or ???

_______________________________________________
Xen-devel mailing list
[hidden email]
http://lists.xen.org/xen-devel
Reply | Threaded
Open this post in threaded view
|

Re: [PATCH] xl.cfg: document the maxmem= option

Ian Campbell-10
On Mon, 2012-05-07 at 17:37 +0100, Dan Magenheimer wrote:

> > From: Ian Campbell [mailto:[hidden email]]
> > Sent: Sunday, May 06, 2012 3:36 AM
> > To: Olaf Hering
> > Cc: [hidden email]; Konrad Rzeszutek Wilk
> > Subject: Re: [Xen-devel] [PATCH] xl.cfg: document the maxmem= option
> >
> > On Fri, 2012-05-04 at 16:18 -0400, Olaf Hering wrote:
> > > On Fri, May 04, Konrad Rzeszutek Wilk wrote:
> > >
> > > > On Fri, May 04, 2012 at 09:31:42PM +0200, Olaf Hering wrote:
> > > > > # HG changeset patch
> > > > > # User Olaf Hering <[hidden email]>
> > > > > # Date 1336159720 -7200
> > > > > # Node ID c414728d0d12f1c3e416e40cceefca2f0b00578e
> > > > > # Parent  8f556a70ae0bef47e242f9e7be0a054769fc8277
> > > > > xl.cfg: document the maxmem= option
> > > > >
> > > > > Signed-off-by: Olaf Hering <[hidden email]>
> > > > >
> > > > > diff -r 8f556a70ae0b -r c414728d0d12 docs/man/xl.cfg.pod.5
> > > > > --- a/docs/man/xl.cfg.pod.5
> > > > > +++ b/docs/man/xl.cfg.pod.5
> > > > > @@ -484,6 +484,14 @@ are not using hardware assisted paging (
> > > > >  mode) and your guest workload consists of a a very large number of
> > > > >  similar processes then increasing this value may improve performance.
> > > > >
> > > > > +=item B<maxmem=MBYTES>
> > > > > +
> > > > > +Specifies the maximum amount of memory the HVM guest can ever see.
>
> can _ever_ see...
>
> What about guest OS's that support hotplug memory?

Does the toolstack (need to) expose the capability to drive that as a
host admin? I don't think xl does... In which case this wording can be
tweaked when that feature is added?

> Also, a subtle point, but is this the maximum physical address that
> will contain read/write pages or the maximum amount of RAM that the
> guest OS should be prepared to map as read/writeable (e.g. after
> subtracting off reserved e820 ranges) or ???

I think it is total number of memory pages, not the maximum PFN. e.g.
the maximum address might be 4.5GB on a 4GB machine with a PCI hole.

I *think*...

Ian.


_______________________________________________
Xen-devel mailing list
[hidden email]
http://lists.xen.org/xen-devel
Reply | Threaded
Open this post in threaded view
|

Re: [PATCH] xl.cfg: document the maxmem= option

Dan Magenheimer-2
> From: Ian Campbell [mailto:[hidden email]]
> Sent: Tuesday, May 08, 2012 6:41 AM
> To: Dan Magenheimer
> Cc: Olaf Hering; [hidden email]; Konrad Wilk
> Subject: RE: [Xen-devel] [PATCH] xl.cfg: document the maxmem= option
>
> On Mon, 2012-05-07 at 17:37 +0100, Dan Magenheimer wrote:
> > > From: Ian Campbell [mailto:[hidden email]]
> > > Sent: Sunday, May 06, 2012 3:36 AM
> > > To: Olaf Hering
> > > Cc: [hidden email]; Konrad Rzeszutek Wilk
> > > Subject: Re: [Xen-devel] [PATCH] xl.cfg: document the maxmem= option
> > >
> > > On Fri, 2012-05-04 at 16:18 -0400, Olaf Hering wrote:
> > > > On Fri, May 04, Konrad Rzeszutek Wilk wrote:
> > > >
> > > > > On Fri, May 04, 2012 at 09:31:42PM +0200, Olaf Hering wrote:
> > > > > > # HG changeset patch
> > > > > > # User Olaf Hering <[hidden email]>
> > > > > > # Date 1336159720 -7200
> > > > > > # Node ID c414728d0d12f1c3e416e40cceefca2f0b00578e
> > > > > > # Parent  8f556a70ae0bef47e242f9e7be0a054769fc8277
> > > > > > xl.cfg: document the maxmem= option
> > > > > >
> > > > > > Signed-off-by: Olaf Hering <[hidden email]>
> > > > > >
> > > > > > diff -r 8f556a70ae0b -r c414728d0d12 docs/man/xl.cfg.pod.5
> > > > > > --- a/docs/man/xl.cfg.pod.5
> > > > > > +++ b/docs/man/xl.cfg.pod.5
> > > > > > @@ -484,6 +484,14 @@ are not using hardware assisted paging (
> > > > > >  mode) and your guest workload consists of a a very large number of
> > > > > >  similar processes then increasing this value may improve performance.
> > > > > >
> > > > > > +=item B<maxmem=MBYTES>
> > > > > > +
> > > > > > +Specifies the maximum amount of memory the HVM guest can ever see.
> >
> > can _ever_ see...
> >
> > What about guest OS's that support hotplug memory?
>
> Does the toolstack (need to) expose the capability to drive that as a
> host admin? I don't think xl does... In which case this wording can be
> tweaked when that feature is added?

I don't know if the matching support ever was implemented in xl,
but hotplug memory support for Xen is in upstream Linux.

http://lists.xen.org/archives/html/xen-devel/2011-05/msg01060.html 

If it _is_ in xl, "can ever see" seems a bit too strong... maybe
something like "memory the HVM guest can see unless changed later
via 'xl mem-max'" or "memory the HVM guest can see when it boots"
would be better.

But this is a nit.  It's impossible to both be precise and brief
in descriptions such as these.

> > Also, a subtle point, but is this the maximum physical address that
> > will contain read/write pages or the maximum amount of RAM that the
> > guest OS should be prepared to map as read/writeable (e.g. after
> > subtracting off reserved e820 ranges) or ???
>
> I think it is total number of memory pages, not the maximum PFN. e.g.
> the maximum address might be 4.5GB on a 4GB machine with a PCI hole.
>
> I *think*...

OK.  If so, then the proposed wording is fine, though maybe adding
one word "maximum amount of _usable_ memory" might help.

_______________________________________________
Xen-devel mailing list
[hidden email]
http://lists.xen.org/xen-devel