Re: [next] x86_64 defconfig failed with clang-10

From: Hyeonggon Yoo <42.hyeyoo@gmail.com>
To: Nick Desaulniers <ndesaulniers@google.com>
Cc: Nathan Chancellor <nathan@kernel.org>,
	Naresh Kamboju <naresh.kamboju@linaro.org>,
	clang-built-linux <clang-built-linux@googlegroups.com>,
	Linux-Next Mailing List <linux-next@vger.kernel.org>,
	lkft-triage@lists.linaro.org,
	Stephen Rothwell <sfr@canb.auug.org.au>,
	Arnd Bergmann <arnd@arndb.de>, Vlastimil Babka <vbabka@suse.cz>,
	Andrew Morton <akpm@linux-foundation.org>
Subject: Re: [next] x86_64 defconfig failed with clang-10
Date: Tue, 18 May 2021 10:04:07 +0900	[thread overview]
Message-ID: <20210518010407.GA83641@hyeyoo> (raw)
In-Reply-To: <CAKwvOd=4wux6NG_6tF6C_xjU0ps+Fh5hfW5a_0U+xcPgL+XJYg@mail.gmail.com>

On Mon, May 17, 2021 at 01:39:46PM -0700, Nick Desaulniers wrote:
> If we can make the compile time validation done instead at runtime for
> clang-10, that would be preferable. Then when we drop clang-10
> support, we can revert that.  Looking at the diff, I think we could
> reinstate parts like so:

> +#if CLANG_VERSION <= 100000
> +       if (size <=  64 * 1024 * 1024) return 26;
> +#endif

> +#if CLANG_VERSION <= 10000
> +       ,INIT_KMALLOC_INFO(67108864, 64M)
> +#endif

I see no reason to add those lines. I removed them because kmalloc
isn't written to support 64MB size, and it doesn't depend on clang
version.

am I misunderstanding your patch?

the problem here is clang-10 is mis-understanding constant argument as
non-constant. and the patch that Nathan linked makes compiler do run-time
evaluation when clang version is less than 11.

Thanks,
Hyeonggon