Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1752001AbcL1TRe (ORCPT ); Wed, 28 Dec 2016 14:17:34 -0500 Received: from mail-io0-f194.google.com ([209.85.223.194]:32946 "EHLO mail-io0-f194.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751924AbcL1TRc (ORCPT ); Wed, 28 Dec 2016 14:17:32 -0500 MIME-Version: 1.0 In-Reply-To: <20161228135358.59f47204@roar.ozlabs.ibm.com> References: <20161225030030.23219-1-npiggin@gmail.com> <20161225030030.23219-3-npiggin@gmail.com> <20161226111654.76ab0957@roar.ozlabs.ibm.com> <20161227211946.3770b6ce@roar.ozlabs.ibm.com> <20161228135358.59f47204@roar.ozlabs.ibm.com> From: Linus Torvalds Date: Wed, 28 Dec 2016 11:17:00 -0800 X-Google-Sender-Auth: U0VNBqrKYQ0_Gzb3kC8pcnKDe_Q Message-ID: Subject: Re: [PATCH 2/2] mm: add PageWaiters indicating tasks are waiting for a page bit To: Nicholas Piggin Cc: Dave Hansen , Bob Peterson , Linux Kernel Mailing List , Steven Whitehouse , Andrew Lutomirski , Andreas Gruenbacher , Peter Zijlstra , linux-mm , Mel Gorman Content-Type: text/plain; charset=UTF-8 Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org Content-Length: 2359 Lines: 54 On Tue, Dec 27, 2016 at 7:53 PM, Nicholas Piggin wrote: >> >> Yeah, that patch is disgusting, and doesn't even help x86. > > No, although it would help some cases (but granted the bitops tend to > be problematic in this regard). To be clear I'm not asking to merge it, > just wondered your opinion. (We need something more for unlock_page > anyway because the memory barrier in the way). The thing is, the patch seems pointless anyway. The "add_return()" kind of cases already return the value, so any code that cares can just use that. And the other cases are downright incorrect, like the removal of "volatile" from the bit test ops. >> It also >> depends on the compiler doing the right thing in ways that are not >> obviously true. > > Can you elaborate on this? GCC will do the optimization (modulo a > regression https://gcc.gnu.org/bugzilla/show_bug.cgi?id=77647) So the removal of volatile is just one example of that. You're basically forcing magical side effects. I've never seen this trick _documented_, and quite frankly, the gcc people have had a history of changing their own documentation when it came to their own extensions (ie they've changed how inline functions work etc). But I also worry about interactions with different gcc versions, or with the LLVM people who try to build the kernel with non-gcc compilers. Finally, it fundamentally can't work on x86 anyway, except for the add-return type of operations, which by definitions are pointless (see above). So everything just screams "this is a horrible approach" to me. > Patch seems okay, but it's kind of a horrible primitive. What if you > did clear_bit_unlock_and_test_bit, which does a __builtin_constant_p > test on the bit numbers and if they are < 7 and == 7, then do the > fastpath? So the problem with that is that it makes no sense *except* in the case where the bit is 7. So why add a "generic" function for something that really isn't generic? I agree that it's a hacky interface, but I also happen to believe that being explicit about what you are actually doing causes less pain. It's not magical, and it's not subtle. There's no question about what it does behind your back, and people won't use it by mistake in the wrong context where it doesn't actually work any better than just doing the obvious thing. Linus