Re: NULL scheduler DoS

["Ahmed, Daniele" <ahmeddan@amazon.co.uk>]

Hi, sorry for the late reply.

Here is the full log from the serial console: https://pastebin.com/raw/g9EFtk3y

This still happens on the current unstable as of today.

Thank you
Daniele

On 11/08/2021, 12:09, "Dario Faggioli" <dfaggioli@suse.com> wrote:

    On Mon, 2021-08-09 at 21:38 +0100, Julien Grall wrote:
    > On 09/08/2021 18:35, Julien Grall wrote:
    > > 
    > > This implies that a pCPU may temporarily be assigned to two vCPUs
    > > and we 
    > > expect to be fixed up afterwards. However, a domain may be
    > > destroyed 
    > > before this is happening.
    > > 
    > > So it looks like that unit_deassign() is not able to cope with this
    > > case. From a brief look, I think we may want to check if the pCPU
    > > is in 
    > > the wait list. If it is, then we should bail out.
    > 
    > Actually, I was wrong. It looks like null_unit_remove() is already 
    > checking this condition. Also, the vCPU should be offline (and the
    > unit 
    > as well) because they haven't come online yet:
    > 
    That's what is currently puzzling me.

    I mean, the crash happens in unit_deassign(), called by
    null_unit_remove(), called by sched_destroy_vcpu().

    And I agree that the unit should be offline, but null_unit_remove()
    calls unit_deassign() only if the unit *is* online, so... What's going
    on? :-/

    Regards
    -- 
    Dario Faggioli, Ph.D
    http://about.me/dario.faggioli
    Virtualization Software Engineer
    SUSE Labs, SUSE https://www.suse.com/
    -------------------------------------------------------------------
    <<This happens because _I_ choose it to happen!>> (Raistlin Majere)