switcher: change TrustedStack underflow to exit #328
Conversation
|
If we're handling this path, maybe we should initialise cra in new threads to the switcher's return entry point so that return from an entry point is a graceful thread exit? |
|
Ah interesting, a use case for manually I'll whip something up. |
nwf
force-pushed
the
202410-switcher-exit_again
branch
from
0409f59
to
4385a7a
Compare
|
CI failures (so far; there may be others lurking) look to be due to CHERIoT-SAFE's Ibex submodule not including microsoft/cheriot-ibex@439b27ce ; Kunyan is looking at what I'd thought would be an easy PR, microsoft/cheriot-safe#18 . |
nwf
force-pushed
the
202410-switcher-exit_again
branch
from
4385a7a
to
46db994
Compare
nwf
force-pushed
the
202410-switcher-exit_again
branch
from
3956116
to
71ceb67
Compare
davidchisnall
force-pushed
the
202410-switcher-exit_again
branch
from
71ceb67
to
b57561d
Compare
Rather than infinitely looping such a damaged thread, request that the scheduler stop running it (via .Lset_mcause_and_exit_thread). This should never happen, as .Lpop_trusted_stack_frame tries to exit any thread that runs out of trusted stack frames, but the scheduler could ask us to resume such a thing anyway.
nwf
force-pushed
the
202410-switcher-exit_again
branch
from
b57561d
to
13dbb90
Compare
| #pragma clang diagnostic push | ||
| #pragma clang diagnostic ignored "-Wc99-designator" | ||
| constexpr SealingType Sentries[] = { | ||
| [int(InterruptStatus::Enabled)] = ReturnSentryEnabling, | ||
| [int(InterruptStatus::Disabled)] = ReturnSentryDisabling}; | ||
| #pragma clang diagnostic pop |
There was a problem hiding this comment.
This makes me a bit nervous. I'd normally do this kind of compile-time map with a template so that we get a compile error if something is invalid.
There was a problem hiding this comment.
I was just following seal_entry immediately above. I'd be happy to rework both however desired, tho'.
There was a problem hiding this comment.
Yes, my code above is also bad. I think we want something like:
template<InterruptStatus T>
constexpr SealingType SealingTypeForInterruptStatus;
template<>
constexpr SealingType SealingTypeForInterruptStatus<InterruptStatus::Enabled> = ReturnSentryEnabling;
...Assuming I got the syntax right.
There was a problem hiding this comment.
Ah, no, this needs to be a dynamic thing. Never mind. Leave as is for now.
Add its offset to the image header and change the header magic number.
If a thread returns from its topmost compartment, it should now directly enter the thread exit path without first going through a trap on an untagged cra.
Now that the initial cra for threads is a sentry to a graceful exit and not nullptr, there's no need to detect and notch out exceptions arising from thread exits. Co-authored-by: Robert Norton <[email protected]>
nwf
force-pushed
the
202410-switcher-exit_again
branch
from
13dbb90
to
0dfbced
Compare
Rather than infinitely looping such a damaged thread, request that the scheduler stop running it (via .Lset_mcause_and_exit_thread). This should never happen, as .Lpop_trusted_stack_frame tries to exit any thread that runs out of trusted stack frames, but the scheduler could ask us to resume such a thing anyway.