Issues with Xenpaging in Xen-4.2.1

Issues with Xenpaging in Xen-4.2.1

[sahilsuneja]

Hi,

I know Xenpaging is still at an experimental stage, but I still wanted to
test it out. Sadly, I could not get it to work reliably.

I have an AMD RVI enabled linux 3.2 host, and I am using linux 2.6.35-22 HVM
guests. After booting a few guests normally upto the host RAM limit, I do
the following to enable paging for a few guests:

/>/usr/lib/xen/bin/xenpaging -f /var/lib/xen/xenpaging/pagefile-hvm-guest -d
1 &
>xenstore-write /local/domain/1/memory/target-tot_pages $((1024*512))/

(here '1' is the dom_id and pagefile-hvm-guest is the backing paging file
for the hvm guest)

I am then able to boot more VMs and I can see the corresponding pagefiles
increase in size for the xenpaging enabled guests.

The problem is when I go back to the guests on which I enabled xenpaging, I
see the following errors when I run any basic commands like vim, top, less
/proc/meminfo:

/xc: error: Error populating page 100: Internal error
xc: error: Error loading 100 during page-in (22 = Invalid argument):
Internal error/

This happens even when there is paged-out memory from other guests to be
had, and the system is not out of RAM for paging in guest's memory.

It seems like a page-in error on gfn 100, but that is strange as, from I
what I gathered from the xenpaging code, policy_default.c starts paging only
after half of the guest's gfns : current_gfn = max_pages / 2 (inside
policy_default.c: policy_init())

I was expecting there to be other stability issues rather that this kind of
an error.

Could someone please help me understand this and let me know if I am doing
something incorrectly. 

Also, there do not seem to be any interesting messages in xm dmesg or in
/var/log/xen*.

Thanks,

Sahil Suneja
Ph.D. Student
University of Toronto



--
View this message in context: http://xen.1045712.n5.nabble.com/Issues-with-Xenpaging-in-Xen-4-2-1-tp5714431.html
Sent from the Xen - Dev mailing list archive at Nabble.com.

Re: Issues with Xenpaging in Xen-4.2.1

[Olaf Hering]

On Mon, Feb 25, sahilsuneja wrote:

> I know Xenpaging is still at an experimental stage, but I still wanted to
> test it out. Sadly, I could not get it to work reliably.
> 
> I have an AMD RVI enabled linux 3.2 host, and I am using linux 2.6.35-22 HVM

The pvops kernel may not have all required changes to run xenpaging in
dom0. Maybe 3.8 already has these changes. If in doubt try the
SLES/openSuSE kernels as dom0, their backend drivers can deal with paged
granttable entries.


Olaf

Re: Issues with Xenpaging in Xen-4.2.1

[sahilsuneja]

sahilsuneja wrote
> Hi,
> 
> I know Xenpaging is still at an experimental stage, but I still wanted to
> test it out. Sadly, I could not get it to work reliably.
> 
> I have an AMD RVI enabled linux 3.2 host, and I am using linux 2.6.35-22
> HVM guests. After booting a few guests normally upto the host RAM limit, I
> do the following to enable paging for a few guests:
/
>>/usr/lib/xen/bin/xenpaging -f /var/lib/xen/xenpaging/pagefile-hvm-guest -d
1 &
>>xenstore-write /local/domain/1/memory/target-tot_pages $((1024*512))
/
> 
> (here '1' is the dom_id and pagefile-hvm-guest is the backing paging file
> for the hvm guest)
> 
> I am then able to boot more VMs and I can see the corresponding pagefiles
> increase in size for the xenpaging enabled guests.
> 
> The problem is when I go back to the guests on which I enabled xenpaging,
> I see the following errors when I run any basic commands like vim, top,
> less /proc/meminfo:
/
> xc: error: Error populating page 100: Internal error
> xc: error: Error loading 100 during page-in (22 = Invalid argument):
> Internal error
/
> 
> This happens even when there is paged-out memory from other guests to be
> had, and the system is not out of RAM for paging in guest's memory.
> 
> It seems like a page-in error on gfn 100, but that is strange as, from I
> what I gathered from the xenpaging code, policy_default.c starts paging
> only after half of the guest's gfns : current_gfn = max_pages / 2 (inside
> policy_default.c: policy_init())
> 
> I was expecting there to be other stability issues rather that this kind
> of an error.
> 
> Could someone please help me understand this and let me know if I am doing
> something incorrectly. 
> 
> Also, there do not seem to be any interesting messages in xm dmesg or in
> /var/log/xen*.
> 
> Thanks,
> 
> Sahil Suneja
> Ph.D. Student
> University of Toronto





--
View this message in context: http://xen.1045712.n5.nabble.com/Issues-with-Xenpaging-in-Xen-4-2-1-tp5714431p5714493.html
Sent from the Xen - Dev mailing list archive at Nabble.com.

Re: Issues with Xenpaging in Xen-4.2.1

[Sahil Suneja]

> The pvops kernel may not have all required changes to run xenpaging in
> dom0. Maybe 3.8 already has these changes. If in doubt try the
> SLES/openSuSE kernels as dom0, their backend drivers can deal with
> paged granttable entries.

Thanks a lot Olaf!
Confirming that linux 3.8 as dom0 still has the same xenpaging page-in
error wrt gfn #100, but opensuse 12.3 rc1 kernel (linux 3.7.9-1.4-xen)
has a much more stable xenpaging, and a much more rich xen-config
kernel build option set.
Although, I am still not clear what the error meant to begin with.

Best,
Sahil