> > > But I think Miroslav's suggestion to revert 1d98a69e5cef would be even
> > > better.
> >
> > AFAIK, Miroslav wanted to point out that your opinion was inconsistent.
>
> How is my opinion inconsistent?
>
> Is there something wrong with the original approach, which was reverted
> with
>
> 1d98a69e5cef ("livepatch: Remove reliable stacktrace check in klp_try_switch_task()")
>
> ?
>
> As I mentioned, that has some advantages:
>
> - The generated code is simpler/faster because it uses a build-time
> check.
>
> - The code is more readable IMO. Especially if the check is done higher
> up the call stack by reverting 1d98a69e5cef. That way the arch
> support short-circuiting is done in the logical place, before doing
> any more unnecessary work. It's faster, but also, more importantly,
> it's less surprising.
Correct. I forgot we removed return from klp_enable_patch() if
klp_have_reliable_stack() errors out and we only warn now. So reverting
1d98a69e5cef definitely makes sense.
My...
"We removed it in 1d98a69e5cef ("livepatch: Remove reliable stacktrace
check in klp_try_switch_task()") and I do think it does not belong here.
We can check for the facility right at the beginning in klp_enable_patch()
and it is not necessary to do it every time klp_check_stack() is called."
...from the other email is rubbish then.
Miroslav
