From 1f66147ac166fcc21b5a1db1bfb508c9079b1d41 Mon Sep 17 00:00:00 2001 From: riastradh Date: Thu, 21 Apr 2022 12:06:31 +0000 Subject: mips/cavium: Take advantage of Octeon's guaranteed r/rw ordering. --- common/lib/libc/arch/mips/atomic/membar_ops.S | 96 ++++++++++++++++++--------- 1 file changed, 66 insertions(+), 30 deletions(-) (limited to 'common') diff --git a/common/lib/libc/arch/mips/atomic/membar_ops.S b/common/lib/libc/arch/mips/atomic/membar_ops.S index 57694df66cc..d7b444ec120 100644 --- a/common/lib/libc/arch/mips/atomic/membar_ops.S +++ b/common/lib/libc/arch/mips/atomic/membar_ops.S @@ -1,4 +1,4 @@ -/* $NetBSD: membar_ops.S,v 1.12 2022/04/09 23:32:51 riastradh Exp $ */ +/* $NetBSD: membar_ops.S,v 1.13 2022/04/21 12:06:31 riastradh Exp $ */ /*- * Copyright (c) 2006, 2007 The NetBSD Foundation, Inc. @@ -38,44 +38,80 @@ LEAF(_membar_sync) j ra BDSYNC END(_membar_sync) +ATOMIC_OP_ALIAS(membar_sync,_membar_sync) + +STRONG_ALIAS(_membar_enter,_membar_sync) +ATOMIC_OP_ALIAS(membar_enter,_membar_sync) #ifdef __OCTEON__ + +/* + * cnMIPS guarantees load-before-load/store ordering without any + * barriers. So the only barriers we need are store-before-load (sync) + * and store-before-store (syncw, i.e., sync 4). See Table 2-32 + * `Execution Ordering Rules' on p. 104 of Cavium OCTEON III CN78XX + * Hardware Reference Manual, CN78XX-HM-0.99E, September 2014: + * + * First Operation DLD [load instruction to a physical + * address that is L2/DRAM] + * Second Operation Any + * Execution Ordering Comments + * + * The second operation cannot appear to execute before + * the first (DLD) operation, regardless of the presence + * or absence of SYNC* instructions. + * + * Note: I'm not sure if this applies to earlier cnMIPS -- can't find + * it in the Cavium Networks OCTEON Plus CN50XX Hardware Reference + * Manual CN50XX-HM-0.99E, July 2008. Experimentally, on an erlite3 + * (Cavium Octeon CN5020-500), I can easily detect reordering of + * store-before-store and store-before-load, but I haven't been able to + * detect any reordering of load-before-load or load-before-store. + * + * Note: On early cnMIPS (CN3xxx), there is an erratum which sometimes + * requires issuing two syncw's in a row. I don't know the details -- + * don't have documentation -- and in Linux it is only used for I/O + * purposes. + * + * Currently we don't build kernels that work on both Octeon and + * non-Octeon MIPS CPUs, so none of this is done with binary patching. + * For userlands we could use a separate shared library on Octeon with + * ld.so.conf to override the symbols with cheaper definitions, but we + * don't do that now. + */ + +LEAF(_membar_acquire) + j ra + nop +END(_membar_acquire) +ATOMIC_OP_ALIAS(membar_acquire,_membar_acquire) + +STRONG_ALIAS(_membar_consumer,_membar_acquire) +ATOMIC_OP_ALIAS(membar_consumer,_membar_acquire) + LEAF(_membar_release) - /* - * syncw is documented as ordering store-before-store in - * - * Cavium OCTEON III CN78XX Hardware Reference Manual, - * CN78XX-HM-0.99E, September 2014. - * - * It's unclear from the documentation the architecture - * guarantees load-before-store ordering without barriers, but - * this code assumes it does. If that assumption is wrong, we - * can only use syncw for membar_producer -- membar_release has - * to use the full sync. - */ j ra syncw END(_membar_release) -#endif +ATOMIC_OP_ALIAS(membar_release,_membar_release) -ATOMIC_OP_ALIAS(membar_sync,_membar_sync) -ATOMIC_OP_ALIAS(membar_acquire,_membar_sync) -STRONG_ALIAS(_membar_acquire,_membar_sync) -ATOMIC_OP_ALIAS(membar_enter,_membar_sync) -STRONG_ALIAS(_membar_enter,_membar_sync) -#ifdef __OCTEON__ -ATOMIC_OP_ALIAS(membar_exit,_membar_release) STRONG_ALIAS(_membar_exit,_membar_release) -ATOMIC_OP_ALIAS(membar_release,_membar_release) -ATOMIC_OP_ALIAS(membar_producer,_membar_release) +ATOMIC_OP_ALIAS(membar_exit,_membar_release) + STRONG_ALIAS(_membar_producer,_membar_release) -#else -ATOMIC_OP_ALIAS(membar_exit,_membar_sync) -STRONG_ALIAS(_membar_exit,_membar_sync) -ATOMIC_OP_ALIAS(membar_release,_membar_sync) +ATOMIC_OP_ALIAS(membar_producer,_membar_release) + +#else /* !__OCTEON__ */ + +STRONG_ALIAS(_membar_acquire,_membar_sync) +ATOMIC_OP_ALIAS(membar_acquire,_membar_sync) STRONG_ALIAS(_membar_release,_membar_sync) -ATOMIC_OP_ALIAS(membar_producer,_membar_sync) +ATOMIC_OP_ALIAS(membar_release,_membar_sync) +STRONG_ALIAS(_membar_exit,_membar_sync) +ATOMIC_OP_ALIAS(membar_exit,_membar_sync) +STRONG_ALIAS(_membar_consumer,_membar_sync) +ATOMIC_OP_ALIAS(membar_consumer,_membar_sync) STRONG_ALIAS(_membar_producer,_membar_sync) +ATOMIC_OP_ALIAS(membar_producer,_membar_sync) + #endif -ATOMIC_OP_ALIAS(membar_consumer,_membar_sync) -STRONG_ALIAS(_membar_consumer,_membar_sync) -- cgit 14e21d3b'>Add new Makefile knob, USE_FORT, which extends USE_SSP by turning on thetls 2007-02-07 * Remove unused header ctype.hhubertf 2006-05-11clear errno before using strtolliamjfoy 2004-07-13Usage -> usage; add missing -n to usage. From Kouichirou Hiratsuka in PR 26221.wiz 2004-03-17- KNFchristos 2004-03-17PR/24796: Colin Percival: CPU-eating loop in lock(1); apply FreeBSD fix.christos 2003-08-07Move UCB-licensed code from 4-clause to 3-clause licence.agc 2002-11-16error handling on strdup failureitojun 2002-09-26Ispell. Begin new sentences on a new line.wiz 2002-03-22Split the notion of building Hesiod, Kerberos, S/key, and YPthorpej 2001-12-01Whitespace cleanup.wiz 2000-10-18Remove INSTALLFLAGS=-fschg, as per change to usr.bin/ssh/ssh/Makefile.simonb 2000-07-29Just another skey compile fix.martin 2000-04-14Don't declare 'extern opt*' getopt variables.simonb