RE: mainline build failure of powerpc allmodconfig for prom_init_check

From: David Laight <David.Laight@ACULAB.COM>
To: 'Michael Ellerman' <mpe@ellerman.id.au>,
	Sudip Mukherjee <sudipm.mukherjee@gmail.com>,
	Benjamin Herrenschmidt <benh@kernel.crashing.org>,
	Paul Mackerras <paulus@samba.org>,
	Kees Cook <keescook@chromium.org>
Cc: "linuxppc-dev@lists.ozlabs.org" <linuxppc-dev@lists.ozlabs.org>,
	linux-kernel <linux-kernel@vger.kernel.org>,
	"linux-hardening@vger.kernel.org"
	<linux-hardening@vger.kernel.org>,
	"Linus Torvalds" <torvalds@linux-foundation.org>
Subject: RE: mainline build failure of powerpc allmodconfig for prom_init_check
Date: Mon, 18 Jul 2022 07:51:45 +0000	[thread overview]
Message-ID: <72a95f0949d149d1b07bd161f3d59cc1@AcuMS.aculab.com> (raw)
In-Reply-To: <87cze3docs.fsf@mpe.ellerman.id.au>

From: Michael Ellerman
> Sent: 18 July 2022 05:41
...
> So we're memsetting all of args to 254, not zero.
> 
> That's happening because allmodconfig with gcc 12 enables
> CONFIG_INIT_STACK_ALL_PATTERN, whereas gcc 11 doesn't.

I can't help feeling it would be better if that generated
a call to a memset64() function.
Saving loads of tests at the top of the function,
and (most of?) the constant expansion to 64bit.

Although and explicit 'stack clear' function would be better
for the kernel - since it would give the option of patching
it away at startup.

I really can't help feeling that initialising on-stack
arrays will kill performance.
While kernel stack frames have to be relatively small,
in userspace very large on-stack arrays can be allocated
(and correctly bound checked) knowing that the cost is
minimal (maybe a TLB miss).

	David

-
Registered Address Lakeside, Bramley Road, Mount Farm, Milton Keynes, MK1 1PT, UK
Registration No: 1397386 (Wales)