Re: [PATCH 1/2] kunit: tool: Add x86_64-smp architecture for SMP testing

From: Daniel Latypov <dlatypov@google.com>
To: David Gow <davidgow@google.com>
Cc: Marco Elver <elver@google.com>,
	Brendan Higgins <brendanhiggins@google.com>,
	Shuah Khan <skhan@linuxfoundation.org>,
	Dmitry Vyukov <dvyukov@google.com>,
	KUnit Development <kunit-dev@googlegroups.com>,
	kasan-dev <kasan-dev@googlegroups.com>,
	"open list:KERNEL SELFTEST FRAMEWORK" 
	<linux-kselftest@vger.kernel.org>,
	Linux Kernel Mailing List <linux-kernel@vger.kernel.org>
Subject: Re: [PATCH 1/2] kunit: tool: Add x86_64-smp architecture for SMP testing
Date: Thu, 19 May 2022 10:11:06 -0700	[thread overview]
Message-ID: <CAGS_qxqsF-soqSM7-cO+tRD1Rg5fqrA07TGLRruxPE4i_rLdJw@mail.gmail.com> (raw)
In-Reply-To: <CABVgOS=X51T_=hwTumnzL2yECgcshWBp1RT0F3GiT3+Fe_vang@mail.gmail.com>

On Thu, May 19, 2022 at 6:15 AM David Gow <davidgow@google.com> wrote:
>
> I tend to agree that having both would be nice: I think there are
> enough useful "machine configs" that trying to maintain, e.g, a 1:1
> mapping with kernel architectures is going to leave a bunch of things
> on the table, particularly as we add more tests for, e.g., drivers and
> specific CPU models.

I agree that we don't necessarily need to maintain a 1:1 mapping.
But I feel like we should have a pretty convincing reason for doing
so, e.g. support for a CPU that requires we add in a bunch of
kconfigs.

This particular one feels simple enough to me.
Given we already have to put specific instructions in the
kcsan/.kunitconfig, I don't know if there's much of a difference in
cost between these two commands

$ ./tools/testing/kunit/kunit.py run --kunitconfig=kernel/kcsan
--arch=x86_64-smp
$ ./tools/testing/kunit/kunit.py run --kunitconfig=kernel/kcsan
--arch=x86_64 --kconfig_add CONFIG_SMP=y --qemu_args "-smp 8"

I've generally learned to prefer more explicit commands like the
second, even if they're quite a bit longer.
But I have the following biases
* I use FZF heavily, so I don't re-type long commands much
* I'm the person who proposed --kconfig_add and --qemu_args, so of
course I'd think the longer form is easy to understand.
so I'm not in a position to object to this change.

Changing topics:
Users can overwrite the '-smp 8' here via --qemu_args [1], so I'm much
less worried about hard-coding any specific value in this file
anymore.
And given that, I think a more "natural" value for this file would be "-smp 2".
I think anything that needs more than that should explicitly should --qemu_args.

Thoughts?

[1] tested with --qemu_args='-smp 4' --qemu_args='-smp 8'
and I see the following in the test.log
 smpboot: Allowing 8 CPUs, 0 hotplug CPUs
so QEMU respects the last value passed in, as expected.

>
> The problem, of course, is that the --kconfig_add flags don't allow us
> to override anything explicitly stated in either the kunitconfig or
> qemu_config (and I imagine there could be problems with --qemu_config,
> too).

This patch would fix that.
https://lore.kernel.org/linux-kselftest/20220519164512.3180360-1-dlatypov@google.com

It introduces an overwriting priority of
* --kconfig_add
* kunitconfig / --kunitconfig
* qemu_config