From mboxrd@z Thu Jan 1 00:00:00 1970 From: will.deacon@arm.com (Will Deacon) Date: Thu, 21 Apr 2011 15:35:31 +0100 Subject: [PATCH] ARM: futex: add SMP futex support when !CPU_USE_DOMAINS In-Reply-To: References: <1302267519-18156-1-git-send-email-will.deacon@arm.com> <1303313347.11314.4.camel@e102144-lin.cambridge.arm.com> Message-ID: <003201cc0031$5dd242c0$1976c840$@deacon@arm.com> To: linux-arm-kernel@lists.infradead.org List-Id: linux-arm-kernel.lists.infradead.org Hi Nicolas, > On Wed, 20 Apr 2011, Will Deacon wrote: > > > On Fri, 2011-04-08 at 13:58 +0100, Will Deacon wrote: > > > This patch uses the load/store exclusive instructions to add SMP futex > > > support for ARM. > > > > > > Since the ARM architecture does not provide instructions for > > > unprivileged exclusive memory accesses, we can only provide SMP futexes > > > when CPU domain support is disabled. > > > > > > Signed-off-by: Will Deacon > > > --- > > > arch/arm/include/asm/futex.h | 137 +++++++++++++++++++++++++++-------------- > > > 1 files changed, 90 insertions(+), 47 deletions(-) > > > > Any comments on this patch? The LTP tests for {get,set}_robust_list seem > > happy enough but any review would be welcome. > > I looked at http://article.gmane.org/gmane.linux.ports.arm.kernel/81847 > back then, and one issue I found with that version was the fact that the > #ifdef condition were wrong (not depending on !CONFIG_CPU_USE_DOMAINS. Yes, I only noticed that patch after I'd written my version. Actually, after looking at the current state of the domains code, I don't understand why we have CONFIG_CPU_USE_DOMAINS at all since the definition of __range_ok checks the address regardless of domain usage. > Your version looks better though, just by looking at the net amount of > lines added. Ha, that's certainly the metric of the moment! I might be able to make the exception table macros cleaner by taking some labels as arguments, but I'm not sure if it's worth it or not. Cheers, Will