Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1754058AbYKQRxr (ORCPT ); Mon, 17 Nov 2008 12:53:47 -0500 Received: (majordomo@vger.kernel.org) by vger.kernel.org id S1751793AbYKQRxi (ORCPT ); Mon, 17 Nov 2008 12:53:38 -0500 Received: from theia.rz.uni-saarland.de ([134.96.7.31]:10327 "EHLO theia.rz.uni-saarland.de" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751331AbYKQRxh (ORCPT ); Mon, 17 Nov 2008 12:53:37 -0500 Date: Mon, 17 Nov 2008 18:52:32 +0100 From: Alexander van Heukelum To: LKML Cc: Andi Kleen , "H. Peter Anvin" , Jan Beulich , Ingo Molnar , Glauber Costa , Thomas Gleixner , Nick Piggin Subject: [RFC,v2] x86_64: save_args out of line Message-ID: <20081117175232.GA13766@mailshack.com> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: <1226845741-12470-2-git-send-email-heukelum@fastmail.fm> User-Agent: Mutt/1.5.9i X-Greylist: Sender IP whitelisted, not delayed by milter-greylist-3.0 (theia.rz.uni-saarland.de [134.96.7.31]); Mon, 17 Nov 2008 18:53:02 +0100 (CET) X-AntiVirus: checked by AntiVir MailGate (version: 2.1.2-14; AVE: 7.9.0.31; VDF: 7.1.0.97; host: AntiVir1) Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org Content-Length: 6777 Lines: 232 The macro "interrupt" in entry_64.S generates a lot of code and it is used more and more often. This patch moves most of its contents into an external function. This saves anywhere between 500 and 2500 bytes of text depending on the configuration. Dwarf2-annotations are most probably wrong or missing at all. v2 moves adjusting the stack to the caller. This avoids the ugly shuffle to handle the position of the return address on the stack. After this patch, a typical handler looks like this: : 68 05 ff ff ff pushq $0xffffffffffffff05 48 83 ec 50 sub $0x50,%rsp e8 72 f4 ff ff callq ffffffff80211260 e8 ec 08 00 00 callq ffffffff802126df e9 16 fd ff ff jmpq ffffffff80211b0e 0f 1f 84 00 00 nopl 0x0(%rax,%rax,1) 00 00 00 I think this approach (v2) is much cleaner than using the same strategy as for the exception handlers, where the address of the C-handler is passed to a common entry point which makes an indirect call to the handler. : ff 15 f2 71 1c callq *0x1c71f2(%rip) # 00 6a 00 pushq $0x0 50 push %rax 48 8d 05 d9 11 lea 0x11d9(%rip),%rax # 00 00 e9 4b 99 0f 00 jmpq ffffffff8030bbb0 66 66 2e 0f 1f nopw %cs:0x0(%rax,%rax,1) 84 00 00 00 00 00 The advantage of _this_ way of doing things is that the stubs can probably be made to fit in 16 bytes, but it comes at the cost of doing an unnecessary indirect call. Signed-off-by: Alexander van Heukelum Cc: Andi Kleen Cc: H. Peter Anvin Cc: Jan Beulich Cc: Thomas Gleixner arch/x86/kernel/entry_64.S | 135 ++++++++++++++++++++++++++------------------ 1 files changed, 81 insertions(+), 54 deletions(-) --- Hi all, I just want to give this one more shot ;). Comments? This patch is on top of tip/x86/cleanups and contains some left-over whitespace changes. Greetings, Alexander diff --git a/arch/x86/kernel/entry_64.S b/arch/x86/kernel/entry_64.S index 5492778..d483e07 100644 --- a/arch/x86/kernel/entry_64.S +++ b/arch/x86/kernel/entry_64.S @@ -242,6 +242,78 @@ ENTRY(native_usergs_sysret64) CFI_REL_OFFSET rsp,RSP /*CFI_REL_OFFSET ss,SS*/ .endm + +/* + * initial frame state for interrupts and exceptions + */ + .macro _frame ref + CFI_STARTPROC simple + CFI_SIGNAL_FRAME + CFI_DEF_CFA rsp,SS+8-\ref + /*CFI_REL_OFFSET ss,SS-\ref*/ + CFI_REL_OFFSET rsp,RSP-\ref + /*CFI_REL_OFFSET rflags,EFLAGS-\ref*/ + /*CFI_REL_OFFSET cs,CS-\ref*/ + CFI_REL_OFFSET rip,RIP-\ref + .endm + +/* + * initial frame state for interrupts (and exceptions without error code) + */ +#define INTR_FRAME _frame RIP +/* + * initial frame state for exceptions with error code (and interrupts + * with vector already pushed) + */ +#define XCPT_FRAME _frame ORIG_RAX + +/* save partial stack frame */ +ENTRY(save_args) + XCPT_FRAME + cld + movq %rdi, 8*8+16(%rsp) + CFI_REL_OFFSET rdi, 8*8+16 + movq %rsi, 7*8+16(%rsp) + CFI_REL_OFFSET rsi, 7*8+16 + movq %rdx, 6*8+16(%rsp) + CFI_REL_OFFSET rdx, 6*8+16 + movq %rcx, 5*8+16(%rsp) + CFI_REL_OFFSET rcx, 5*8+16 + movq %rax, 4*8+16(%rsp) + CFI_REL_OFFSET rax, 4*8+16 + movq %r8, 3*8+16(%rsp) + CFI_REL_OFFSET r8, 3*8+16 + movq %r9, 2*8+16(%rsp) + CFI_REL_OFFSET r9, 2*8+16 + movq %r10, 1*8+16(%rsp) + CFI_REL_OFFSET r10, 1*8+16 + movq %r11, 0*8+16(%rsp) + CFI_REL_OFFSET r11, 0*8+16 + leaq -ARGOFFSET+16(%rsp),%rdi /* arg1 for handler */ + movq %rbp, 8(%rsp) /* push %rbp */ + leaq 8(%rsp), %rbp /* mov %rsp, %ebp */ + testl $3, CS(%rdi) + je 1f + SWAPGS + /* + * irqcount is used to check if a CPU is already on an interrupt stack + * or not. While this is essentially redundant with preempt_count it is + * a little cheaper to use a separate counter in the PDA (short of + * moving irq_enter into assembly, which would be too much work) + */ +1: incl %gs:pda_irqcount + jne 2f + pop %rax /* move return address... */ + mov %gs:pda_irqstackptr,%rsp + push %rax /* ... to the new stack */ + /* + * We entered an interrupt context - irqs are off: + */ +2: TRACE_IRQS_OFF + ret + CFI_ENDPROC +END(save_args) + /* * A newly forked process directly context switches into this. */ @@ -608,65 +680,18 @@ ENTRY(stub_rt_sigreturn) END(stub_rt_sigreturn) /* - * initial frame state for interrupts and exceptions - */ - .macro _frame ref - CFI_STARTPROC simple - CFI_SIGNAL_FRAME - CFI_DEF_CFA rsp,SS+8-\ref - /*CFI_REL_OFFSET ss,SS-\ref*/ - CFI_REL_OFFSET rsp,RSP-\ref - /*CFI_REL_OFFSET rflags,EFLAGS-\ref*/ - /*CFI_REL_OFFSET cs,CS-\ref*/ - CFI_REL_OFFSET rip,RIP-\ref - .endm - -/* initial frame state for interrupts (and exceptions without error code) */ -#define INTR_FRAME _frame RIP -/* initial frame state for exceptions with error code (and interrupts with - vector already pushed) */ -#define XCPT_FRAME _frame ORIG_RAX - -/* * Interrupt entry/exit. * * Interrupt entry points save only callee clobbered registers in fast path. - * - * Entry runs with interrupts off. - */ + * + * Entry runs with interrupts off. + */ -/* 0(%rsp): interrupt number */ +/* 0(%rsp): interrupt number */ .macro interrupt func - cld - SAVE_ARGS - leaq -ARGOFFSET(%rsp),%rdi # arg1 for handler - pushq %rbp - /* - * Save rbp twice: One is for marking the stack frame, as usual, and the - * other, to fill pt_regs properly. This is because bx comes right - * before the last saved register in that structure, and not bp. If the - * base pointer were in the place bx is today, this would not be needed. - */ - movq %rbp, -8(%rsp) - CFI_ADJUST_CFA_OFFSET 8 - CFI_REL_OFFSET rbp, 0 - movq %rsp,%rbp - CFI_DEF_CFA_REGISTER rbp - testl $3,CS(%rdi) - je 1f - SWAPGS - /* irqcount is used to check if a CPU is already on an interrupt - stack or not. While this is essentially redundant with preempt_count - it is a little cheaper to use a separate counter in the PDA - (short of moving irq_enter into assembly, which would be too - much work) */ -1: incl %gs:pda_irqcount - cmoveq %gs:pda_irqstackptr,%rsp - push %rbp # backlink for old unwinder - /* - * We entered an interrupt context - irqs are off: - */ - TRACE_IRQS_OFF + subq $10*8, %rsp + CFI_ADJUST_CFA_OFFSET 10*8 + call save_args call \func .endm @@ -806,6 +831,8 @@ END(common_interrupt) /* * APIC interrupts. */ + .p2align 5 + .macro apicinterrupt num,func INTR_FRAME pushq $~(\num) -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/