Re: [PATCH v5 12/18] x86/sgx: Add EPC OOM path to forcefully reclaim EPC

[Haitao Huang]

On Mon, 16 Oct 2023 20:34:57 -0500, Huang, Kai <kai.huang@xxxxxxxxx> wrote:

> On Mon, 2023-10-16 at 19:10 -0500, Haitao Huang wrote:
> > On Mon, 16 Oct 2023 16:09:52 -0500, Huang, Kai <kai.huang@xxxxxxxxx>  
> > wrote:
> > [...]
> >
> > > still need to fix the bug mentioned above here.
> > >
> > > I really think you should just go this simple way:
> > >
> > > When you want to take EPC back from VM, kill the VM.
> > >
> >
> > My only concern is that this is a compromise due to current limitation  
> > (no
> > other sane way to take EPC from VMs). If we define this behavior and it
> > becomes a contract to user space, then we can't change in future.
>
> Why do we need to "define such behaviour"?
>
> This isn't some kinda of kernel/userspace ABI IMHO, but only kernel  
> internal
> implementation.  Here VM is similar to normal host enclaves.  You limit  
> the
> resource, some host enclaves could be killed.  Similarly, VM could also  
> be
> killed too.
>
> And supporting VMM EPC oversubscription doesn't mean VM won't be  
> killed.  The VM
> can still be a target to kill after VM's all EPC pages have been swapped  
> out.
>
> > On the other hand, my understanding the reason you want this behavior is
> > to enforce EPC limit at runtime.
>
> No I just thought this is a bug/issue needs to be fixed.  If anyone  
> believes
> this is not a bug/issue then it's a separate discussion.

AFAIK, before we introducing max_write() callback in this series, no misc  
controller would possibly enforce the limit when misc.max is reduced. e.g.  
I don't think CVMs be killed when ASID limit is reduced and the cgroup was  
full before limit is reduced.

I think EPC pages to VMs could have the same behavior, once they are given  
to a guest, never taken back by the host. For enclaves on host side, pages  
are reclaimable, that allows us to enforce in a similar way to memcg.

Thanks
Haitao