projects / platform / kernel / linux-rpi.git / commitdiff
? search:
summary | shortlog | log | commit | commitdiff | tree
raw | patch | inline | side by side (parent: b690fda)
x86/entry/64: Use 'xorl' for faster register clearing
authorDominik Brodowski <linux@dominikbrodowski.net>
Wed, 14 Feb 2018 17:59:24 +0000 (18:59 +0100)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Thu, 15 Mar 2018 09:54:36 +0000 (10:54 +0100)
commit ced5d0bf603fa0baee8ea889e1d70971fd210894 upstream.

On some x86 CPU microarchitectures using 'xorq' to clear general-purpose
registers is slower than 'xorl'. As 'xorl' is sufficient to clear all
64 bits of these registers due to zero-extension [*], switch the x86
64-bit entry code to use 'xorl'.

No change in functionality and no change in code size.

[*] According to Intel 64 and IA-32 Architecture Software Developer's
    Manual, section 3.4.1.1, the result of 32-bit operands are "zero-
    extended to a 64-bit result in the destination general-purpose
    register." The AMD64 Architecture Programmer’s Manual Volume 3,
    Appendix B.1, describes the same behaviour.

Suggested-by: Denys Vlasenko <dvlasenk@redhat.com>
Signed-off-by: Dominik Brodowski <linux@dominikbrodowski.net>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Arjan van de Ven <arjan@linux.intel.com>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: David Woodhouse <dwmw2@infradead.org>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: Josh Poimboeuf <jpoimboe@redhat.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Link: http://lkml.kernel.org/r/20180214175924.23065-3-linux@dominikbrodowski.net
[ Improved on the changelog a bit. ]
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
arch/x86/entry/calling.h patch | blob | history
arch/x86/entry/entry_64_compat.S patch | blob | history

diff --git a/arch/x86/entry/calling.h b/arch/x86/entry/calling.h
index 196b6103edf60095fd77f9c56a9bb21a80b3fadf..5d10b7a85cadd303a0b47534f40494a99e942bb9 100644 (file)
--- a/arch/x86/entry/calling.h
+++ b/arch/x86/entry/calling.h
@@ -117,25 +117,25 @@ For 32-bit we have the following conventions - kernel is built with
        pushq   %rcx            /* pt_regs->cx */
        pushq   \rax            /* pt_regs->ax */
        pushq   %r8             /* pt_regs->r8 */
-       xorq    %r8, %r8        /* nospec   r8 */
+       xorl    %r8d, %r8d      /* nospec   r8 */
        pushq   %r9             /* pt_regs->r9 */
-       xorq    %r9, %r9        /* nospec   r9 */
+       xorl    %r9d, %r9d      /* nospec   r9 */
        pushq   %r10            /* pt_regs->r10 */
-       xorq    %r10, %r10      /* nospec   r10 */
+       xorl    %r10d, %r10d    /* nospec   r10 */
        pushq   %r11            /* pt_regs->r11 */
-       xorq    %r11, %r11      /* nospec   r11*/
+       xorl    %r11d, %r11d    /* nospec   r11*/
        pushq   %rbx            /* pt_regs->rbx */
        xorl    %ebx, %ebx      /* nospec   rbx*/
        pushq   %rbp            /* pt_regs->rbp */
        xorl    %ebp, %ebp      /* nospec   rbp*/
        pushq   %r12            /* pt_regs->r12 */
-       xorq    %r12, %r12      /* nospec   r12*/
+       xorl    %r12d, %r12d    /* nospec   r12*/
        pushq   %r13            /* pt_regs->r13 */
-       xorq    %r13, %r13      /* nospec   r13*/
+       xorl    %r13d, %r13d    /* nospec   r13*/
        pushq   %r14            /* pt_regs->r14 */
-       xorq    %r14, %r14      /* nospec   r14*/
+       xorl    %r14d, %r14d    /* nospec   r14*/
        pushq   %r15            /* pt_regs->r15 */
-       xorq    %r15, %r15      /* nospec   r15*/
+       xorl    %r15d, %r15d    /* nospec   r15*/
        UNWIND_HINT_REGS
        .if \save_ret
        pushq   %rsi            /* return address on top of stack */
diff --git a/arch/x86/entry/entry_64_compat.S b/arch/x86/entry/entry_64_compat.S
index fd65e016e4133f5634545062d098fc25fb5c0b1d..364ea4a207bed9c8d994263855c578dc8b6e5eb2 100644 (file)
--- a/arch/x86/entry/entry_64_compat.S
+++ b/arch/x86/entry/entry_64_compat.S
@@ -85,25 +85,25 @@ ENTRY(entry_SYSENTER_compat)
        pushq   %rcx                    /* pt_regs->cx */
        pushq   $-ENOSYS                /* pt_regs->ax */
        pushq   $0                      /* pt_regs->r8  = 0 */
-       xorq    %r8, %r8                /* nospec   r8 */
+       xorl    %r8d, %r8d              /* nospec   r8 */
        pushq   $0                      /* pt_regs->r9  = 0 */
-       xorq    %r9, %r9                /* nospec   r9 */
+       xorl    %r9d, %r9d              /* nospec   r9 */
        pushq   $0                      /* pt_regs->r10 = 0 */
-       xorq    %r10, %r10              /* nospec   r10 */
+       xorl    %r10d, %r10d            /* nospec   r10 */
        pushq   $0                      /* pt_regs->r11 = 0 */
-       xorq    %r11, %r11              /* nospec   r11 */
+       xorl    %r11d, %r11d            /* nospec   r11 */
        pushq   %rbx                    /* pt_regs->rbx */
        xorl    %ebx, %ebx              /* nospec   rbx */
        pushq   %rbp                    /* pt_regs->rbp (will be overwritten) */
        xorl    %ebp, %ebp              /* nospec   rbp */
        pushq   $0                      /* pt_regs->r12 = 0 */
-       xorq    %r12, %r12              /* nospec   r12 */
+       xorl    %r12d, %r12d            /* nospec   r12 */
        pushq   $0                      /* pt_regs->r13 = 0 */
-       xorq    %r13, %r13              /* nospec   r13 */
+       xorl    %r13d, %r13d            /* nospec   r13 */
        pushq   $0                      /* pt_regs->r14 = 0 */
-       xorq    %r14, %r14              /* nospec   r14 */
+       xorl    %r14d, %r14d            /* nospec   r14 */
        pushq   $0                      /* pt_regs->r15 = 0 */
-       xorq    %r15, %r15              /* nospec   r15 */
+       xorl    %r15d, %r15d            /* nospec   r15 */
        cld
 
        /*
@@ -224,25 +224,25 @@ GLOBAL(entry_SYSCALL_compat_after_hwframe)
        pushq   %rbp                    /* pt_regs->cx (stashed in bp) */
        pushq   $-ENOSYS                /* pt_regs->ax */
        pushq   $0                      /* pt_regs->r8  = 0 */
-       xorq    %r8, %r8                /* nospec   r8 */
+       xorl    %r8d, %r8d              /* nospec   r8 */
        pushq   $0                      /* pt_regs->r9  = 0 */
-       xorq    %r9, %r9                /* nospec   r9 */
+       xorl    %r9d, %r9d              /* nospec   r9 */
        pushq   $0                      /* pt_regs->r10 = 0 */
-       xorq    %r10, %r10              /* nospec   r10 */
+       xorl    %r10d, %r10d            /* nospec   r10 */
        pushq   $0                      /* pt_regs->r11 = 0 */
-       xorq    %r11, %r11              /* nospec   r11 */
+       xorl    %r11d, %r11d            /* nospec   r11 */
        pushq   %rbx                    /* pt_regs->rbx */
        xorl    %ebx, %ebx              /* nospec   rbx */
        pushq   %rbp                    /* pt_regs->rbp (will be overwritten) */
        xorl    %ebp, %ebp              /* nospec   rbp */
        pushq   $0                      /* pt_regs->r12 = 0 */
-       xorq    %r12, %r12              /* nospec   r12 */
+       xorl    %r12d, %r12d            /* nospec   r12 */
        pushq   $0                      /* pt_regs->r13 = 0 */
-       xorq    %r13, %r13              /* nospec   r13 */
+       xorl    %r13d, %r13d            /* nospec   r13 */
        pushq   $0                      /* pt_regs->r14 = 0 */
-       xorq    %r14, %r14              /* nospec   r14 */
+       xorl    %r14d, %r14d            /* nospec   r14 */
        pushq   $0                      /* pt_regs->r15 = 0 */
-       xorq    %r15, %r15              /* nospec   r15 */
+       xorl    %r15d, %r15d            /* nospec   r15 */
 
        /*
         * User mode is traced as though IRQs are on, and SYSENTER
@@ -298,9 +298,9 @@ sysret32_from_system_call:
         */
        SWITCH_TO_USER_CR3_NOSTACK scratch_reg=%r8 scratch_reg2=%r9
 
-       xorq    %r8, %r8
-       xorq    %r9, %r9
-       xorq    %r10, %r10
+       xorl    %r8d, %r8d
+       xorl    %r9d, %r9d
+       xorl    %r10d, %r10d
        swapgs
        sysretl
 END(entry_SYSCALL_compat)
@@ -358,25 +358,25 @@ ENTRY(entry_INT80_compat)
        pushq   %rcx                    /* pt_regs->cx */
        pushq   $-ENOSYS                /* pt_regs->ax */
        pushq   $0                      /* pt_regs->r8  = 0 */
-       xorq    %r8, %r8                /* nospec   r8 */
+       xorl    %r8d, %r8d              /* nospec   r8 */
        pushq   $0                      /* pt_regs->r9  = 0 */
-       xorq    %r9, %r9                /* nospec   r9 */
+       xorl    %r9d, %r9d              /* nospec   r9 */
        pushq   $0                      /* pt_regs->r10 = 0 */
-       xorq    %r10, %r10              /* nospec   r10 */
+       xorl    %r10d, %r10d            /* nospec   r10 */
        pushq   $0                      /* pt_regs->r11 = 0 */
-       xorq    %r11, %r11              /* nospec   r11 */
+       xorl    %r11d, %r11d            /* nospec   r11 */
        pushq   %rbx                    /* pt_regs->rbx */
        xorl    %ebx, %ebx              /* nospec   rbx */
        pushq   %rbp                    /* pt_regs->rbp */
        xorl    %ebp, %ebp              /* nospec   rbp */
        pushq   %r12                    /* pt_regs->r12 */
-       xorq    %r12, %r12              /* nospec   r12 */
+       xorl    %r12d, %r12d            /* nospec   r12 */
        pushq   %r13                    /* pt_regs->r13 */
-       xorq    %r13, %r13              /* nospec   r13 */
+       xorl    %r13d, %r13d            /* nospec   r13 */
        pushq   %r14                    /* pt_regs->r14 */
-       xorq    %r14, %r14              /* nospec   r14 */
+       xorl    %r14d, %r14d            /* nospec   r14 */
        pushq   %r15                    /* pt_regs->r15 */
-       xorq    %r15, %r15              /* nospec   r15 */
+       xorl    %r15d, %r15d            /* nospec   r15 */
        cld
 
        /*
Domain: System / Kernel;