[src/trunk]: src/sys/arch - On-demand zeroing pages with MOVNTI is crazy. It...

[ad <ad%NetBSD.org@localhost>]

details:   https://anonhg.NetBSD.org/src/rev/101f7fcb973e
branches:  trunk
changeset: 465444:101f7fcb973e
user:      ad <ad%NetBSD.org@localhost>
date:      Fri Nov 22 23:36:25 2019 +0000

description:
- On-demand zeroing pages with MOVNTI is crazy.  It empties L1/L2/L3.
- Disable zeroing in the idle loop.  That needs a cache-friendly strategy.

Result: 3 to 4% reduction in kernel build time on my test system.
Inspired by a discussion with Mateusz Guzik and David Maxwell.

diffstat:

 sys/arch/amd64/amd64/locore.S |  24 ++++++++----------------
 sys/arch/x86/x86/cpu.c        |   6 +++---
 2 files changed, 11 insertions(+), 19 deletions(-)

diffs (68 lines):

diff -r b969e1e10229 -r 101f7fcb973e sys/arch/amd64/amd64/locore.S
--- a/sys/arch/amd64/amd64/locore.S     Fri Nov 22 23:01:49 2019 +0000
+++ b/sys/arch/amd64/amd64/locore.S     Fri Nov 22 23:36:25 2019 +0000
@@ -1,4 +1,4 @@
-/*     $NetBSD: locore.S,v 1.191 2019/11/21 19:27:54 ad Exp $  */
+/*     $NetBSD: locore.S,v 1.192 2019/11/22 23:36:25 ad Exp $  */
 
 /*
  * Copyright-o-rama!
@@ -1519,24 +1519,16 @@
 /*
  * void pagezero(vaddr_t va)
  *
- * Zero a page without polluting the cache.
+ * Zero a page.
  */
 ENTRY(pagezero)
-       movq    $-PAGE_SIZE,%rdx
-       subq    %rdx,%rdi
+       pushq   %rbp
+       movq    %rsp,%rbp
+       movq    $(PAGE_SIZE / 8),%rcx
        xorq    %rax,%rax
-1:
-       movnti  %rax,(%rdi,%rdx)
-       movnti  %rax,8(%rdi,%rdx)
-       movnti  %rax,16(%rdi,%rdx)
-       movnti  %rax,24(%rdi,%rdx)
-       movnti  %rax,32(%rdi,%rdx)
-       movnti  %rax,40(%rdi,%rdx)
-       movnti  %rax,48(%rdi,%rdx)
-       movnti  %rax,56(%rdi,%rdx)
-       addq    $64,%rdx
-       jne     1b
-       sfence
+       rep
+       stosq
+       leave
        ret
 END(pagezero)
 
diff -r b969e1e10229 -r 101f7fcb973e sys/arch/x86/x86/cpu.c
--- a/sys/arch/x86/x86/cpu.c    Fri Nov 22 23:01:49 2019 +0000
+++ b/sys/arch/x86/x86/cpu.c    Fri Nov 22 23:36:25 2019 +0000
@@ -1,4 +1,4 @@
-/*     $NetBSD: cpu.c,v 1.174 2019/11/05 20:19:17 maxv Exp $   */
+/*     $NetBSD: cpu.c,v 1.175 2019/11/22 23:36:25 ad Exp $     */
 
 /*
  * Copyright (c) 2000-2012 NetBSD Foundation, Inc.
@@ -62,7 +62,7 @@
  */
 
 #include <sys/cdefs.h>
-__KERNEL_RCSID(0, "$NetBSD: cpu.c,v 1.174 2019/11/05 20:19:17 maxv Exp $");
+__KERNEL_RCSID(0, "$NetBSD: cpu.c,v 1.175 2019/11/22 23:36:25 ad Exp $");
 
 #include "opt_ddb.h"
 #include "opt_mpbios.h"                /* for MPDEBUG */
@@ -723,7 +723,7 @@
        tsc_tc_init();
 
        /* Enable zeroing of pages in the idle loop if we have SSE2. */
-       vm_page_zero_enable = ((cpu_feature[0] & CPUID_SSE2) != 0);
+       vm_page_zero_enable = false; /* ((cpu_feature[0] & CPUID_SSE2) != 0); */
 }
 #endif