RE: [PATCH V2 3/3] efi: Use efi_rts_workqueue to invoke EFI Runtime Services

[Prakhya, Sai Praneeth]

> >> > pstore writes could potentially be invoked in interrupt context and
> >> > it uses set_variable<>() and query_variable_info<>() to store logs.
> >> > If we invoke efi_runtime_services() through efi_rts_wq while in
> >> > atomic() kernel issues a warning ("scheduling wile in atomic") and
> >> > prints stack trace. One way to overcome this is to not make the
> >> > caller process wait for the worker thread to finish. This approach
> >> > breaks pstore i.e. the log messages aren't written to efi
> >> > variables. Hence, pstore calls
> >> > efi_runtime_services() without using efi_rts_wq or in other words
> >> > efi_rts_wq will be used unconditionally for all the
> >> > efi_runtime_services() except set_variable<>() and
> >> > query_variable_info<>()
> >>
> >> Can't NMIs still come in during this?
> >>
> >> ... or do we assume that since something has already gone wrong, this
> >> doesn't matter?
> >>
> >
> > I think they can come. AFAIK, pstore (if enabled) runs only when we crashed.
> > Since, we are still executing stuff to log messages and NMI's can't be
> > masked, there is still a possibility for NMI's to occur (please correct me if I am
> wrong).
> > But, as you said earlier, I guess it doesn't matter because anyways we are
> going down.
> 
> The problem is that we are not always in a "already going down"
> condition for typical set_variable and query_variable_info calls.

That's correct.

> So are we actually fixing anything with this patchset?

When we are _not_ in interrupt context (eg: process context)
we still use efi_rts_wq to invoke *all* efi_runtime_services().
This solves *someone trying to access user space* while in EFI runtime services
mapping problem. A instance could be, some user space process requests us to
execute efi_runtime_service(), so, kernel switches to efi_pgd (which doesnât have
user space part of process address space) and while executing efi_runtime_service()
perf NMI comes along to profile user data.

> In other words if the NMI vs EFI
> mapping problem requires the workqueue context then we can't have any EFI
> calls outside of that context.  Am I missing how this scheme addresses that
> problem?

I think so, because, we are not trying to solve NMI vs EFI Runtime Service mappings.
AFAIK, they both work together (for x86_64). The problem is, as stated above,
"someone trying to access user space while executing EFI runtime services".
The problem exists because EFI runtime services mappings donât have user space
part of process address space.

I think the problem still persists when we are already in interrupt context and then
we were requested to execute some efi_runtime_service() and then NMI happens
and that NMI handler touches user space.

Please let me know if my explanation didnât make sense or if I misunderstood
your question.

Ard and Matt, please correct me if I stated something that is incorrect.