Date: Tue, 04 Nov 2014 10:50:29 -0700 From: Ian Lepore <ian@FreeBSD.org> To: Bruce Evans <brde@optusnet.com.au> Cc: svn-src-head@freebsd.org, svn-src-all@freebsd.org, src-committers@freebsd.org Subject: Re: svn commit: r274088 - head/sys/kern Message-ID: <1415123429.1200.75.camel@revolution.hippie.lan> In-Reply-To: <20141105023323.O1105@besplex.bde.org> References: <201411041129.sA4BTnwX030600@svn.freebsd.org> <20141104114041.GA21297@dft-labs.eu> <20141105023323.O1105@besplex.bde.org>
next in thread | previous in thread | raw e-mail | index | archive | help
On Wed, 2014-11-05 at 03:19 +1100, Bruce Evans wrote:
> [...]
> Another unsuitable alignment is by the MD ALIGNBYTES macro. This is
> (sizeof(register_t) - 1) on all arches except mips 32-bit where it is 7
> sparc64 where it is 15. On arm, it is 3, but arm also has
> STACKALIGNBYTES = 7. The register size is really too small to use for
> malloc() on 32-bit arches. Its technical correctness depends on no
> C objects having more than 32-bit alignment. On i386, int64_t and
> double should have 64-bit alignment, but this is not required unless
> CFLAGS includes -malign-double which breaks the ABI in userland and
> is irrelevant for the kernel.
>
In arm/include/_align.h we have:
/*
* Round p (pointer or byte index) up to a correctly-aligned value
* for all data types (int, long, ...). [more words snipped]
*/
#define _ALIGNBYTES (sizeof(int) - 1)
So that's clearly wrong, because int64_t and double types require 8-byte
alignment. When it comes to fixing it, I could:
* include _types.h and use sizeof(__int64_t)
* use sizeof(long long)
* use sizeof(double)
* just hardcode '7' with a comment that says why
What are the pros and cons of the various options? Is one of them
preferable? Is there something even better I overlooked?
-- Ian
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?1415123429.1200.75.camel>