VMs gets not enough memory at startup

[marmarek]

Reported by joanna on 25 Dec 2011 19:14 UTC
... and gui agent doesn't startup:

user# cat /var/log/qubes/gui_agent.log
fork cmd: Cannot allocate memory
Waiting on /var/run/xf86-qubes-socket socket...
user# free -m
total used free shared buffers cached
Mem: 132 93 39 0 5 38
-/+ buffers/cache: 50 82
Swap: 1023 6 1017

Migrated-From: https://wiki.qubes-os.org/ticket/392

[marmarek]

Comment by joanna on 25 Dec 2011 19:17 UTC
The VM wasn't able to read meminfo writer key at boot:

Starting system logger: [ OK ]
Starting system message bus: [ OK ]
Starting HAL daemon: [ OK ]
Starting cups: [ OK ]
Executing Qubes Core scripts for AppVM:xenstore-read: couldn't read path qubes-service/meminfo-writer
[ OK ]
Executing Qubes Core scripts:[ 12.716178] EXT4-fs (xvdb): mounted filesystem with ordered data mode. Opts: (null)
[ OK ]
[ 12.728749] u2mfn_mmap: entering, private= (null)
[ 12.728771] u2mfn_mmap: calling remap return 0
[ 12.728786] u2mfn_release, priv= (null)
Executing Qubes Core scripts NetVM:[ OK ]
Starting Qubes GUI agent:[ OK ]
[ 13.054268] u2mfn_mmap: entering, private= (null)
[ 13.054304] u2mfn_mmap: calling remap return 0
[ 13.054338] u2mfn_release, priv= (null)
[ 13.078260] qubes_gui:816 map pfn expected mapping type write-back for 8b000-8c000, got uncached-minus
xenstore-read: couldn't read path qubes-service/qubes-firewall
xenstore-read: couldn't read path qubes-service/qubes-netwatcher

... but the meminfo writer was started anyway (perhaps too later?)...

user# ps ax | grep meminfo
692 ? S 0:01 /usr/lib/qubes/meminfo-writer 30000 100000
939 hvc0 S+ 0:00 grep --color=auto meminfo
user#

[marmarek]

Comment by joanna on 25 Dec 2011 19:17 UTC
The VM wasn't able to read meminfo writer key at boot:

Starting system logger: [ OK ]
Starting system message bus: [ OK ]
Starting HAL daemon: [ OK ]
Starting cups: [ OK ]
Executing Qubes Core scripts for AppVM:xenstore-read: couldn't read path qubes-service/meminfo-writer
[ OK ]
Executing Qubes Core scripts:[ 12.716178] EXT4-fs (xvdb): mounted filesystem with ordered data mode. Opts: (null)
[ OK ]
[ 12.728749] u2mfn_mmap: entering, private= (null)
[ 12.728771] u2mfn_mmap: calling remap return 0
[ 12.728786] u2mfn_release, priv= (null)
Executing Qubes Core scripts NetVM:[ OK ]
Starting Qubes GUI agent:[ OK ]
[ 13.054268] u2mfn_mmap: entering, private= (null)
[ 13.054304] u2mfn_mmap: calling remap return 0
[ 13.054338] u2mfn_release, priv= (null)
[ 13.078260] qubes_gui:816 map pfn expected mapping type write-back for 8b000-8c000, got uncached-minus
xenstore-read: couldn't read path qubes-service/qubes-firewall
xenstore-read: couldn't read path qubes-service/qubes-netwatcher

... but the meminfo writer was started anyway (perhaps too later?)...

user# ps ax | grep meminfo
692 ? S 0:01 /usr/lib/qubes/meminfo-writer 30000 100000
939 hvc0 S+ 0:00 grep --color=auto meminfo
user#

[marmarek]

Comment by joanna on 25 Dec 2011 19:49 UTC
It seems like this memory allocation problem is also responsible for the following error that sometimes occur when one attempts to start a new VM, while having some other running consuming too much memory:

[xen](root@dom0)# qvm-start --console work
--> Cleaning volatile image: /var/lib/qubes/appvms/work/volatile.img...
--> Loading the VM (type = AppVM)...
xc: error: panic: xc_dom_boot.c:79: launch_vm: SETVCPUCONTEXT failed (rc=-1): Internal error
libxl: error: libxl_dom.c:212:libxl__build_pv xc_dom_boot_image failed: Invalid argument
cannot (re-)build domain: -3
ERROR: Failed to load VM config

I wasn't able to start this work VM until I closed all the other VMs (with variable mem) and then qvm-start succeed immediately.

[marmarek]

Comment by joanna on 25 Dec 2011 19:49 UTC
It seems like this memory allocation problem is also responsible for the following error that sometimes occur when one attempts to start a new VM, while having some other running consuming too much memory:

[xen](root@dom0)# qvm-start --console work
--> Cleaning volatile image: /var/lib/qubes/appvms/work/volatile.img...
--> Loading the VM (type = AppVM)...
xc: error: panic: xc_dom_boot.c:79: launch_vm: SETVCPUCONTEXT failed (rc=-1): Internal error
libxl: error: libxl_dom.c:212:libxl__build_pv xc_dom_boot_image failed: Invalid argument
cannot (re-)build domain: -3
ERROR: Failed to load VM config

I wasn't able to start this work VM until I closed all the other VMs (with variable mem) and then qvm-start succeed immediately.

[marmarek]

Comment by joanna on 26 Dec 2011 09:21 UTC
So, the idea is to use qvm-run to start meminfo-writer, so that the VM doesn't give up memory at the very early start, and only after it starts Xorg.

[marmarek]

Comment by joanna on 26 Dec 2011 09:21 UTC
So, the idea is to use qvm-run to start meminfo-writer, so that the VM doesn't give up memory at the very early start, and only after it starts Xorg.

[marmarek]

Modified by marmarek on 26 Dec 2011 22:40 UTC

[marmarek]

Modified by marmarek on 26 Dec 2011 22:40 UTC

[marmarek]

Comment by marmarek on 27 Dec 2011 16:46 UTC
http://git.qubes-os.org/gitweb/?p=marmarek/core.git;a=commit;h=b8b5cf0a17724e7c89ae43a7555cdc8f226aaaa4

[marmarek]

Comment by marmarek on 27 Dec 2011 16:46 UTC
http://git.qubes-os.org/gitweb/?p=marmarek/core.git;a=commit;h=b8b5cf0a17724e7c89ae43a7555cdc8f226aaaa4