The question about migration/checkpointing

The question about migration/checkpointing

[Sergei Lyubski]

[-- Attachment #1: Type: text/plain, Size: 1660 bytes --]

  Hi guys,

My name is Sergey. My company uses QEMU for  CPU
performance simulation/evaluaton.
Sorry, I found  your emails in QEMU  relatively recent commits  into ram.c
.  Why ram.c . Because I see some weirdness in the  behavior.

Our goal is :
     a)  To boot  Linux , to launch some  TEST/SPEC, to drop a few
checkpoints.
     b)  To resume  execution from  those checkpoints, to attach a
performance simulator and
           to evaluate performance at the "hot" code  areas.

  How we tried to do it .

     a) run QEMU with -monitor telnet options
     b) connect  to QEMU with telnet
     c)  after  LINUX  booted  in  QEMU monitor (telnet)  executed
            migrate -d file://<FULL  CKPT PATH>
            quit
      d)  run QEMU  with option -incoming file://<FULL CKPT PATH>

  Now !!

  If  I  drop the checkpoint  when running  on linux a long ( endless) test
which  prints on      the console,  everything works (!)  and  the console
is properly restored.

  When I drop  the checkpoint  when the linux console is in shell  prompt,
the restoration does  not  work.
   I see either error message  from  ram_load_precopy()
       error_report("Unknown combination of migration flags: 0x%x", flags);

       or

   the console is locked, QEMU restarts and  runs  (ps, top).   The only
solution is to kill
   QEMU .  After that,  QEMU console is distorted and requires to type
RESET.

    Could you please comment on the above issues.
    If you need more info I can send you my screen shots.
    Please note that  that the similar behavior is observed on  both
     qemu 9.0  ( stable-9.0)  and
     qemu 8.x

Thanks in advance.
Sergey Lyubskiy

[-- Attachment #2: Type: text/html, Size: 2347 bytes --]

Re: The question about migration/checkpointing

[Alex Bennée]

Sergei Lyubski <slyubski@gmail.com> writes:

>   Hi guys,
>
> My name is Sergey. My company uses QEMU for  CPU performance simulation/evaluaton.
> Sorry, I found  your emails in QEMU  relatively recent commits  into ram.c .  Why ram.c . Because I see some weirdness in
> the  behavior.
>   
> Our goal is :
>      a)  To boot  Linux , to launch some  TEST/SPEC, to drop a few checkpoints.
>      b)  To resume  execution from  those checkpoints, to attach a performance simulator and
>            to evaluate performance at the "hot" code  areas.
>
>   How we tried to do it .
>
>      a) run QEMU with -monitor telnet options
>      b) connect  to QEMU with telnet 
>      c)  after  LINUX  booted  in  QEMU monitor (telnet)  executed
>             migrate -d file://<FULL  CKPT PATH> 
>             quit
>       d)  run QEMU  with option -incoming file://<FULL CKPT PATH>
>
>   Now !!
>
>   If  I  drop the checkpoint  when running  on linux a long ( endless) test which  prints on      the console,  everything works (!)
>  and  the console is properly restored.
>    
>   When I drop  the checkpoint  when the linux console is in shell  prompt, the restoration does  not  work.
>    I see either error message  from  ram_load_precopy()
>        error_report("Unknown combination of migration flags: 0x%x",
>   flags);

I suspect seeing the value of flags will help with any diagnosis.

>
>        or
>
>    the console is locked, QEMU restarts and  runs  (ps, top).   The only  solution is to kill 
>    QEMU .  After that,  QEMU console is distorted and requires to type
>    RESET.

Are you able to attach with gdb and get a backtrace?

>
>     Could you please comment on the above issues. 
>     If you need more info I can send you my screen shots.
>     Please note that  that the similar behavior is observed on  both
>      qemu 9.0  ( stable-9.0)  and
>      qemu 8.x
>
> Thanks in advance.
> Sergey Lyubskiy

-- 
Alex Bennée
Virtualisation Tech Lead @ Linaro