[CRIU] [PATCH 0/5] cpuinfo: Rework cpuinfo dumping, make it optional

Pavel Emelyanov xemul at parallels.com
Thu Sep 18 09:10:09 PDT 2014


On 09/12/2014 02:15 PM, Cyrill Gorcunov wrote:
> On Fri, Sep 12, 2014 at 02:04:45PM +0400, Pavel Emelyanov wrote:
>> Can you sum up the API change you propose with this?
>>
> 
> Currently (with this series applied)
> ====================================
> 
>  - by default cpu caps are not dumped at all, one need to
>    pass --cpu-cap all to force CRIU write cpu caps bits into
>    the image

Why not just --cpu-cap, w/o any arguments?

>  - on restore we test if image with cpuinfo is present on disk,
>    and if so -- test that image is capable with runtime cpu

I like it. Can you do the following. Add two more actions -- dumpcpu
and checkcpu (or smth like that to be consonant with the dump option).

And make sure this API is available via RPC and library.

> that's all for current status
> 
> What's need to be done
> ======================
> 
>  - criud helpers which would take cpuinfo from the stream (via
>    RPC requiest) and test if the cpu it is spinning on is capable
>    with cpuinfo bits obtained from the stream
> 
> That's basically I thought how would it work,
> 
> Now why I think cpuinfo must be bound into criu dump files (ie not
> as some separate file created with "criu dump-cpuinfo" command) --
> the cpuinfo bits are entangled with instructions saved in program's
> executable memory block.
> 
> Thus base functionality is implemented in this series if we need
> anything else helpers -- we can do that on top.
> 
> 	Cyrill
> .
> 



More information about the CRIU mailing list