Re: [PATCH] crypto: testmgr - don't generate WARN for missing modules

[Eric Biggers <ebiggers@kernel.org>]

On Sat, Aug 20, 2022 at 12:15:41AM +0000, Elliott, Robert (Servers) wrote:
> Per Stephan, it sounds like this was a hacky way to get some/most of
> the modules loaded.
> 
> It'd be good if there was a way to run all registered tests on all
> available modules, not just the ones that someone remembered to put
> in tcrypt.c.

Most algorithms can be allocated via a userspace program using AF_ALG.  The only
exception is algorithm types that AF_ALG doesn't support.

> I do worry this WARN() isn't really helpful even for real self-test
> failures - it's dumping the call trace to alg_test(), not the
> trace to whatever crypto function alg_test called that is failing. 
> With Linus always expressing concern with too many BUG and WARN
> calls, it might be better as just pr_warn() or pr_err().

It's very helpful because WARN is the standard way for the kernel to report that
a kernel bug has been encountered.  A test failure is a kernel bug.

The stack trace printed by WARN indeed isn't useful here, as it will always be
the same.  But it's just a side effect.  The important things here are that a
WARN is triggered at all, and that some log messages that describe what failed
are printed.

- Eric