Pkg xen devel - Nov 2014 - Bug#767295: Bug#767295: Bug#767295: xl: apparent memory leak

On Thu, 2014-11-06 at 12:14 +0000, Ian Campbell wrote:
> On Wed, 2014-10-29 at 17:36 -0400, Gedalya wrote:
> > When booting domU's running amd64 jessie, I notice some memory
problems
> > with xl.
> 
> I ran it (actually, upstream staging-4.4) under valgrind and it reported
> after a reboot:
> 
> ==5722== 24 bytes in 1 blocks are definitely lost in loss record 35 of 72
> ==5722==    at 0x4026B2D: malloc (vg_replace_malloc.c:299)
> ==5722==    by 0x41739FF: strdup (strdup.c:43)
> ==5722==    by 0x405F38E: libxl__device_disk_local_initiate_attach
(libxl.c:2681)
> ==5722==    by 0x408D2A6: libxl__bootloader_run (libxl_bootloader.c:385)
> ==5722==    by 0x406B50C: initiate_domain_create (libxl_create.c:807)
> ==5722==    by 0x406B50C: do_domain_create (libxl_create.c:1354)
> ==5722==    by 0x406B6AF: libxl_domain_create_new (libxl_create.c:1377)
> ==5722==    by 0x8056182: create_domain (xl_cmdimpl.c:2254)
> ==5722==    by 0x8059F27: main_create (xl_cmdimpl.c:4407)
> ==5722==    by 0x804E119: main (xl.c:360)
> ==5722== 
> ==5722== 28 bytes in 1 blocks are definitely lost in loss record 39 of 72
> ==5722==    at 0x4026B2D: malloc (vg_replace_malloc.c:299)
> ==5722==    by 0x4279EBB: read_message (xs.c:1143)
> ==5722==    by 0x427A89B: read_thread (xs.c:1219)
> ==5722==    by 0x40E7953: start_thread (pthread_create.c:304)
> ==5722==    by 0x41CF95D: clone (clone.S:130)
I've posted a fix for this upstream:
http://lists.xen.org/archives/html/xen-devel/2014-11/msg00542.html

I've requested that it go into the next 4.4.x release. I also plan to
backport it to the package regardless once it is accepted into the dev
branch.

Ian.

On 11/06/2014 09:12 AM, Ian Campbell wrote:
> I've posted a fix for this upstream:
> http://lists.xen.org/archives/html/xen-devel/2014-11/msg00542.html
>
> I've requested that it go into the next 4.4.x release. I also plan to
> backport it to the package regardless once it is accepted into the dev
> branch.
>
> Ian.
So I've tried building xen 4.4.1-3 with this patch (attached), and I 
replaced the binary packages libxen-4.4 and xen-utils-4.4 with the new ones.
The situation is the same: the xl process for jessie amd64 guests starts 
off at a resident size of 15mb as opposed to ~500kb for wheezy i386 
guests. At guest reboot it jumps to 17mb and then, after a few seconds, 
to ~34mb.
I also rebooted the host to help make sure the test is valid.

Thanks,

Gedalya

-------------- next part --------------
A non-text attachment was scrubbed...
Name: 0039-do-not-leak-diskpath-during-localdisk-attach.patch
Type: text/x-patch
Size: 407 bytes
Desc: not available
URL:
<http://lists.alioth.debian.org/pipermail/pkg-xen-devel/attachments/20141107/856667fe/attachment.bin>

On Fri, 2014-11-07 at 21:32 -0500, Gedalya wrote:
> On 11/06/2014 09:12 AM, Ian Campbell wrote:
> > I've posted a fix for this upstream:
> > http://lists.xen.org/archives/html/xen-devel/2014-11/msg00542.html
> >
> > I've requested that it go into the next 4.4.x release. I also plan
to
> > backport it to the package regardless once it is accepted into the dev
> > branch.
> >
> > Ian.
> 
> So I've tried building xen 4.4.1-3 with this patch (attached), and I 
> replaced the binary packages libxen-4.4 and xen-utils-4.4 with the new
ones.
> The situation is the same: the xl process for jessie amd64 guests starts 
> off at a resident size of 15mb as opposed to ~500kb for wheezy i386 
> guests. At guest reboot it jumps to 17mb and then, after a few seconds, 
> to ~34mb.
> I also rebooted the host to help make sure the test is valid.
Please can you try running xl under valgrind, something similar to what
I described earlier should work.

Could you also post your guest cfg file please.

Thanks,
Ian.