[PATCH v2 1/1] arch/fault: don't print logs for pte marker poison errors

[Axel Rasmussen]

On Wed, May 15, 2024 at 3:54 AM Oscar Salvador <osalvador at suse.de> wrote:
>
> On Wed, May 15, 2024 at 12:41:42PM +0200, Borislav Petkov wrote:
> > On Fri, May 10, 2024 at 11:29:26AM -0700, Axel Rasmussen wrote:
> > > @@ -3938,7 +3938,7 @@ static vm_fault_t handle_pte_marker(struct vm_fault *vmf)
> > >
> > >     /* Higher priority than uffd-wp when data corrupted */
> > >     if (marker & PTE_MARKER_POISONED)
> > > -           return VM_FAULT_HWPOISON;
> > > +           return VM_FAULT_HWPOISON | VM_FAULT_HWPOISON_SILENT;
> >
> > If you know here that this poisoning should be silent, why do you have
> > to make it all complicated and propagate it into arch code, waste
> > a separate VM_FAULT flag just for that instead of simply returning here
> > a VM_FAULT_COMPLETED or some other innocuous value which would stop
> > processing the fault?
>
> AFAIK, He only wants it to be silent wrt. the arch fault handler not screaming,
> but he still wants to be able to trigger force_sig_mceerr().

Right, the goal is to still have the process get a SIGBUS, but to
avoid the "MCE error" log message. The basic issue is, unprivileged
users can set these markers up, and thereby completely spam up the
log.

Also since this is a process-specific thing, and it's not a real
hardware poison event, it's unclear system admins care at all at a
global level (this is why we didn't want to switch to just
printk_ratelimited for example). Better to let the process handle the
SIGBUS however it likes for its use case (logging a message elsewhere,
etc.).

That said, one thing I'm not sure about is whether or not
VM_FAULT_SIGBUS is a viable alternative (returned for a new PTE marker
type specific to simulated poison). The goal of the simulated poison
feature is to "closely simulate" a real hardware poison event. If you
live migrate a VM from a host with real poisoned memory, to a new
host: you'd want to keep the same behavior if the guest accessed those
addresses again, so as not to confuse the guest about why it suddenly
became "un-poisoned". At a basic level I think VM_FAULT_SIGBUS gives
us what we want (send SIGBUS to the process, don't log about MCEs),
but I'm not confident I know all the differences vs. VM_FAULT_HWPOISON
on all the arches.

>
>
> --
> Oscar Salvador
> SUSE Labs