From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1753524AbcHWQ2Q (ORCPT ); Tue, 23 Aug 2016 12:28:16 -0400 Received: from mail-lf0-f68.google.com ([209.85.215.68]:34094 "EHLO mail-lf0-f68.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1752570AbcHWQ2P (ORCPT ); Tue, 23 Aug 2016 12:28:15 -0400 Date: Tue, 23 Aug 2016 19:28:21 +0300 From: Alexey Dobriyan To: x86@kernel.org, linux-kernel@vger.kernel.org Subject: [PATCH 3/3] x86: branchless copy_page() Message-ID: <20160823162821.GA2433@p183.telecom.by> References: <20160822205229.GA2638@p183.telecom.by> <20160823161920.GA2286@p183.telecom.by> <20160823162533.GB2286@p183.telecom.by> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: <20160823162533.GB2286@p183.telecom.by> User-Agent: Mutt/1.5.24 (2015-08-30) Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org Apply alternatives at the call site instead of function body. Save branch per copied page. Rename individual function, show immediately in profiles which method is being used. RBX and R12 are saved and restored by unrolled-mov implementation, don't clobber them explicitly. Signed-off-by: Alexey Dobriyan --- arch/x86/include/asm/page_64.h | 15 ++++++++++++++- arch/x86/kernel/x8664_ksyms_64.c | 4 +++- arch/x86/lib/copy_page_64.S | 14 ++++---------- 3 files changed, 21 insertions(+), 12 deletions(-) --- a/arch/x86/include/asm/page_64.h +++ b/arch/x86/include/asm/page_64.h @@ -49,8 +49,21 @@ static __always_inline void clear_page(void *page) : "rax", "rcx", "memory" ); } -void copy_page(void *to, void *from); +void copy_page_mov(void *to, void *from); +void copy_page_rep_movsq(void *to, void *from); +void copy_page_rep_movsb(void *to, void *from); +static __always_inline void copy_page(void *to, void *from) +{ + alternative_call_2( + copy_page_mov, + copy_page_rep_movsq, X86_FEATURE_REP_GOOD, + copy_page_rep_movsb, X86_FEATURE_ERMS, + ASM_OUTPUT2("=D" (to), "=S" (from)), + "0" (to), "1" (from) + : "rax", "rcx", "rdx", "r8", "r9", "r10", "r11", "memory" + ); +} #endif /* !__ASSEMBLY__ */ #ifdef CONFIG_X86_VSYSCALL_EMULATION --- a/arch/x86/kernel/x8664_ksyms_64.c +++ b/arch/x86/kernel/x8664_ksyms_64.c @@ -40,7 +40,9 @@ EXPORT_SYMBOL(_copy_to_user); EXPORT_SYMBOL_GPL(memcpy_mcsafe); -EXPORT_SYMBOL(copy_page); +EXPORT_SYMBOL(copy_page_mov); +EXPORT_SYMBOL(copy_page_rep_movsq); +EXPORT_SYMBOL(copy_page_rep_movsb); EXPORT_SYMBOL(clear_page_mov); EXPORT_SYMBOL(clear_page_rep_stosq); EXPORT_SYMBOL(clear_page_rep_stosb); --- a/arch/x86/lib/copy_page_64.S +++ b/arch/x86/lib/copy_page_64.S @@ -1,8 +1,6 @@ /* Written 2003 by Andi Kleen, based on a kernel by Evandro Menezes */ #include -#include -#include /* * Some CPUs run faster using the string copy instructions (sane microcode). @@ -11,15 +9,11 @@ * prefetch distance based on SMP/UP. */ ALIGN -ENTRY(copy_page) - ALTERNATIVE_2 "jmp copy_page_regs", \ - "", X86_FEATURE_REP_GOOD, \ - "jmp copy_page_rep_movsb", X86_FEATURE_ERMS - +ENTRY(copy_page_rep_movsq) movl $4096/8, %ecx rep movsq ret -ENDPROC(copy_page) +ENDPROC(copy_page_rep_movsq) ENTRY(copy_page_rep_movsb) mov $4096, %ecx @@ -27,7 +21,7 @@ ENTRY(copy_page_rep_movsb) ret ENDPROC(copy_page_rep_movsb) -ENTRY(copy_page_regs) +ENTRY(copy_page_mov) subq $2*8, %rsp movq %rbx, (%rsp) movq %r12, 1*8(%rsp) @@ -92,4 +86,4 @@ ENTRY(copy_page_regs) movq 1*8(%rsp), %r12 addq $2*8, %rsp ret -ENDPROC(copy_page_regs) +ENDPROC(copy_page_mov)