From: David Miller Date: Wed, 10 Feb 2010 20:12:17 +0000 Subject: Re: [PATCH] sparc: Align clone and signal stacks to 16 bytes. Message-Id: <20100210.121217.83298044.davem@davemloft.net> List-Id: References: <20100209.165006.142022647.davem@davemloft.net> In-Reply-To: <20100209.165006.142022647.davem@davemloft.net> MIME-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 7bit To: sparclinux@vger.kernel.org From: Dennis Gilmore Date: Wed, 10 Feb 2010 13:59:00 -0600 > http://sources.redhat.com/bugzilla/show_bug.cgi?id855 As the patch in this bug does, it should be fixed in glibc. GCC had the same exact issues when creating precompiled headers in c++. > http://sources.redhat.com/bugzilla/show_bug.cgi?id283 Hey, Ulrich Drepper unreasonably flames someone for trying to fix a bug, I've never seen that before! :-) His comment about "all of userland expects page-size alignment to be sufficient" is completely wrong, you only need look at the precompiled header support code in GCC to see but one example. Simple sys_errlist versioning build fixes for Sparc from mid-January still have not been applied, and I supposed STT_GNU_IFUNC support for sparc last week to the glibc lists and that hasn't gotten looked at either. So the lack of any real looking into this issue is no surprise. This is not a kernel bug, shared mappings have stringent, architecture specific requirement wrt. mmap() area alignment in order to handle cache aliasing issues properly. MAP_SHARED mappings are special, you have to be aware of the issues if you're going to use them. You can't just map them wherever you like.