Re: [Qemu-devel] incorrect memory size inside vm

[Andrey Korolyov]

On Tue, Jun 23, 2015 at 3:32 PM, Piotr Rybicki
<address@hidden> wrote:
>> Thanks Piotr, the lack of the host-side memory shrinkage after balloon
>> deflation is interesting anyway, hopefully you may share guest` dmesg
>> bits from compiled-in balloon to check visible signs of the issue
>
>
> It is really hard to see dmesg output via vnc console, when kernel is
> sending so many messages looped :(
>
>> (balloon should be initialized *before* DIMMs are being onlined, so
>> the behavior of the balloon itself it not clear at least for me).
>
>
> Why balloon should be initialized before DIMMs onlined?

For compiled-in balloon, its initialization should be done earlier
than the udev, because the kernel is not switched to init execution
yet.
>
> DIMMS are onlined as soon as udev starts (i have udev rule to online mem,
> cpu), so system has to start init process. I don't see any option to online
> DIMMs before booting reaches init process.
>
> With balloon compiled-in, it starts to shrink memmory (for example to shrink
> 8GB, when 1GB 'base' mem is defined, and 10GB in DIMMs), when there is only
> 1GB 'base' memory. This leads to problem
>
> With balloon loaded as a module, shrinking starts after init has begun
> (because system has to at least mount filesystem where balloon module is
> placed to load)
>
> Well, i can't proove this with dmesg, but with balloon as module - system is
> booting and working as expected. With balloon compiled-in - system does not
> boot saying there is no free mem for balloon to shrink it.

Thanks, this is an unexpected behavior, because you are starting with
small amount of total memory exposed to a balloon at the beginning and
there should be no issues except if the inflation races with plugging
in new DIMMs (just a wild guess), dmesg output will be helpful there
as I pointed earlier.

>
> Best regards
> Piotr Rybicki
>