Hi, Peter and Paul, Loongson-3's Store Fill Buffer is nearly the same as your "Store Buffer", and it increases the memory ordering weakness. So, smp_cond_load_acquire() only need a __smp_mb() before the loop, not after every READ_ONCE(). In other word, the following code is just OK:
#define smp_cond_load_acquire(ptr, cond_expr) \ ({ \ typeof(ptr) __PTR = (ptr); \ typeof(*ptr) VAL; \ __smp_mb(); \ for (;;) { \ VAL = READ_ONCE(*__PTR); \ if (cond_expr) \ break; \ cpu_relax(); \ } \ __smp_mb(); \ VAL; \ }) the __smp_mb() before loop is used to avoid "reads prioritised over writes", which is caused by SFB's weak ordering and similar to ARM11MPCore (mentioned by Will Deacon). Huacai ------------------ Original ------------------ From: "Peter Zijlstra"<pet...@infradead.org>; Date: Tue, Jun 19, 2018 03:22 PM To: "陈华才"<che...@lemote.com>; Cc: "Paul Burton"<paul.bur...@mips.com>; "Ralf Baechle"<r...@linux-mips.org>; "James Hogan"<james.ho...@mips.com>; "linux-mips"<linux-m...@linux-mips.org>; "Fuxin Zhang"<zhan...@lemote.com>; "wuzhangjin"<wuzhang...@gmail.com>; "Huacai Chen"<chenhua...@gmail.com>; "stable"<sta...@vger.kernel.org>; "Alan Stern"<st...@rowland.harvard.edu>; "AndreaParri"<andrea.pa...@amarulasolutions.com>; "Will Deacon"<will.dea...@arm.com>; "Boqun Feng"<boqun.f...@gmail.com>; "Nicholas Piggin"<npig...@gmail.com>; "David Howells"<dhowe...@redhat.com>; "Jade Alglave"<j.algl...@ucl.ac.uk>; "Luc Maranget"<luc.maran...@inria.fr>; "Paul E. McKenney"<paul...@linux.vnet.ibm.com>; "Akira Yokosawa"<aki...@gmail.com>; "linux-kernel"<linux-kernel@vger.kernel.org>; Subject: Re: [PATCH] MIPS: implement smp_cond_load_acquire() for Loongson-3 On Tue, Jun 19, 2018 at 02:40:14PM +0800, 陈华才 wrote: > Hi, Paul, > > First of all, could you please check why linux-mips reject e-mails > from lemote.com? Of course I can send e-mails by gmail, but my gmail > can't receive e-mails from linux-mips since March, 2018. Could you please learn to use email? No top posting and wrap lines at 78 chars. > I have already read Documentation/memory-barriers.txt, but I don't > think we should define a smp_read_barrier_depends() for Loongson-3. > Because Loongson-3's behavior isn't like Alpha, and in syntax, this is > not a data-dependent issue. Agreed, this is not a data-dependency issue. > There is no document about Loongson-3's SFB. In my opinion, SFB looks > like the L0 cache but sometimes it is out of cache-coherent machanism > (L1 cache's cross-core coherency is maintained by hardware, but not > always true for SFB). smp_mb() is needed for smp_cond_load_acquire(), > but not every READ_ONCE(). Linux does _NOT_ support non-coherent SMP. If your system is not fully coherent, you're out of luck. But please, explain in excruciating detail what exactly you need that smp_mb for. If, like I posited in my previous email, it is to ensure remote store buffer flushes, then your machine is terminally broken.