“failed to add gnttab to physmap” when starting a HVM guest in XEN with memory over 3200 MB

Uncategorized , , , , 0 Comments

I have some HVM guests on XEN but most of them have memory around or less than 2048 MB. However when I tried to give more than 3200 MB of RAM to them I faced the following message:

# xl create /etc/xen/configs/a.conf
Parsing config from /etc/xen/configs/a.conf
xc: error: panic: xc_dom_boot.c:395: xc_dom_gnttab_hvm_seed: failed to add gnttab to physmap [errno=22]
: Internal error
libxl: error: libxl_dm.c:1388:device_model_spawn_outcome: domain 13 device model: spawn failed (rc=-3)
libxl: error: libxl_create.c:1186:domcreate_devmodel_started: device model did not start: -3
libxl: error: libxl_dm.c:1492:kill_device_model: Device Model already exited

And to make it easier, not all the tried XEN hosts produced the above message (6 in 7 produced).

After a lot of google searches (some with no results at all) I could not get around this problem, and after some trial and error guesses I found the problematic part.

It seems that the default shadow_memory = 8 (in MB) parameter was too low. After finding this out I found a post on the Xen-devel maillist mentioning the approximate requirements of the shadow_memory: there should be 2KB of memory for every one MB of RAM assigned to the guest domain, plus a few MB per vcpu (~2MB).

Anyway, doubling this number seems to be solving this problem, my config now looks like the following:

name = "a"
vcpus = 2
memory = 6000
vif = [ "bridge=br0,mac=aa:ab:ac:ad:ae:af" ]
boot = "n"
vnc = 1
vncpasswd = "xxxxxxxx"

builder = "hvm"
acpi = 1
apic = 1
pae = 1
shadow_memory = 16
sdl = 0
vnclisten  = "0.0.0.0"
vncconsole = 1
stdvga = 0
serial = "pty"
usbdevice = "tablet"

on_poweroff = "destroy"
on_reboot = "restart"
on_crash = "restart"

Leave a Reply

Your email address will not be published. Required fields are marked *