Sporadic Invalid_clear exception in init

[Norman Feske]

Hi Pirmin,

On 23.02.21 16:34, Pirmin Duss wrote:
> Eventually init throws 'Genode::Bit_array_base::Invalid_clear' and
> terminates (see the backtrace below).
> 
> Maybe it is noteworthy that packages with sub inits that have components
> that use resource saturation are started.
> 
> Does anybody have an idea on how to proceed with debugging the problem?

I have never seen this exception before, which got me thinking.

The backtrace is already quite insightful. Thanks for posting it.

My interpretation: The problem occurs while closing a session. The
server informs init that it has finished the closing of a session (by
calling Parent::session_response), which prompts init to discard the
local copy of the session capability. The reference count of this
capability reaches zero (which is expected), triggering the deletion of
the capability's local ID from init's capability space. For some reason,
the corresponding slot is already marked as free, which is unexpected.

To investigate, it is important to understand the situation a bit
better. E.g., which session type is the troublemaker? Who is the client,
who is the server? Of course, plain instrumentation with debug messages
would produce too much noise. I would hence try to add conditional
messages that are only printed in "interesting" situations. E.g., the
attached patch prints a message with session information, but only in
the unexpected exception case. The output may help to form a better
mental picture of what is happening. E.g., is the session an environment
session or a higher-level session?

Regarding general debugging hints, it depends on how "sporadic" is this
issue? Can you easily provoke it? If not, I would first try to work on
the reproducibility, e.g., by increasing the frequency of init
reconfigurations. I'd also try to strip down the complexity of the
scenario step by step, e.g., by removing one component after another
while validating that the problem persists.

Best wishes! I'm eager to learn what just happened here.

Norman

-- 
Dr.-Ing. Norman Feske
Genode Labs

https://www.genode-labs.com · https://genode.org

Genode Labs GmbH · Amtsgericht Dresden · HRB 28424 · Sitz Dresden
Geschäftsführer: Dr.-Ing. Norman Feske, Christian Helmuth
-------------- next part --------------
A non-text attachment was scrubbed...
Name: instr_except_revert.patch
Type: text/x-patch
Size: 596 bytes
Desc: not available
URL: <http://lists.genode.org/pipermail/users/attachments/20210224/9f7dc8b6/attachment.bin>