[src/trunk]: src/sys/arch/x86/x86 Reorder these loops to reduce the number of...

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

details:   https://anonhg.NetBSD.org/src/rev/a42d0f8b71cb
branches:  trunk
changeset: 824698:a42d0f8b71cb
user:      maxv <maxv%NetBSD.org@localhost>
date:      Thu Jun 15 06:32:52 2017 +0000

description:
Reorder these loops to reduce the number of enter->flush. I figured out
yesterday that this has a clear impact: a system with 16TB of hard-coded
ram has a 4-second black screen when booting. Now we're down to < 0.5s.

It could be optimized more, but verily I don't have a machine with P1GB
right now.

diffstat:

 sys/arch/x86/x86/pmap.c |  45 +++++++++++++++++++++++----------------------
 1 files changed, 23 insertions(+), 22 deletions(-)

diffs (85 lines):

diff -r a94e4a5a94bf -r a42d0f8b71cb sys/arch/x86/x86/pmap.c
--- a/sys/arch/x86/x86/pmap.c   Thu Jun 15 05:42:27 2017 +0000
+++ b/sys/arch/x86/x86/pmap.c   Thu Jun 15 06:32:52 2017 +0000
@@ -1,4 +1,4 @@
-/*     $NetBSD: pmap.c,v 1.246 2017/06/14 14:17:15 maxv Exp $  */
+/*     $NetBSD: pmap.c,v 1.247 2017/06/15 06:32:52 maxv Exp $  */
 
 /*
  * Copyright (c) 2008, 2010, 2016, 2017 The NetBSD Foundation, Inc.
@@ -171,7 +171,7 @@
  */
 
 #include <sys/cdefs.h>
-__KERNEL_RCSID(0, "$NetBSD: pmap.c,v 1.246 2017/06/14 14:17:15 maxv Exp $");
+__KERNEL_RCSID(0, "$NetBSD: pmap.c,v 1.247 2017/06/15 06:32:52 maxv Exp $");
 
 #include "opt_user_ldt.h"
 #include "opt_lockdebug.h"
@@ -1443,8 +1443,8 @@
        pt_entry_t *pte;
        pd_entry_t *pde;
        phys_ram_seg_t *mc;
-       size_t nL4e, nL3e;
-       int i;
+       size_t nL4e, nL3e, pn, npd;
+       int i, n;
 
        const pd_entry_t pteflags = PG_V | PG_KW | pmap_pg_nx;
 
@@ -1503,36 +1503,37 @@
                /* Allocate L2. */
                L2page_pa = pmap_bootstrap_palloc(nL3e);
 
-               /* Zero out the L2 pages. */
+               KASSERT(pmap_largepages != 0);
+
+               /* Large pages are supported. Just create L2. */
                for (i = 0; i < nL3e; i++) {
                        pdp = L2page_pa + i * PAGE_SIZE;
                        *pte = (pdp & PG_FRAME) | pteflags;
                        pmap_update_pg(tmpva);
 
                        memset((void *)tmpva, 0, PAGE_SIZE);
-               }
-
-               KASSERT(pmap_largepages != 0);
-
-               /* Large pages are supported. Just create L2. */
-               for (i = 0; i < NPDPG * nL3e; i++) {
-                       pdp = (paddr_t)&(((pd_entry_t *)L2page_pa)[i]);
-                       *pte = (pdp & PG_FRAME) | pteflags;
-                       pmap_update_pg(tmpva);
-
-                       pde = (pd_entry_t *)(tmpva + (pdp & ~PG_FRAME));
-                       *pde = ((paddr_t)i << L2_SHIFT) | pteflags |
-                           PG_U | PG_PS | PG_G;
+
+                       pde = (pd_entry_t *)tmpva;
+                       for (n = 0; n < NPDPG; n++) {
+                               pn = (i * NPDPG) + n;
+                               pde[n] = ((paddr_t)pn << L2_SHIFT) | pteflags |
+                                       PG_U | PG_PS | PG_G;
+                       }
                }
 
                /* Fill in the L3 entries, linked to L2. */
-               for (i = 0; i < nL3e; i++) {
-                       pdp = (paddr_t)&(((pd_entry_t *)L3page_pa)[i]);
+               for (i = 0; i < nL4e; i++) {
+                       pdp = L3page_pa + i * PAGE_SIZE;
                        *pte = (pdp & PG_FRAME) | pteflags;
                        pmap_update_pg(tmpva);
 
-                       pde = (pd_entry_t *)(tmpva + (pdp & ~PG_FRAME));
-                       *pde = (L2page_pa + (i << PAGE_SHIFT)) | pteflags | PG_U;
+                       pde = (pd_entry_t *)tmpva;
+                       npd = (i == nL4e - 1) ? (nL3e % NPDPG) : NPDPG;
+                       for (n = 0; n < npd; n++) {
+                               pn = (i * NPDPG) + n;
+                               pde[n] = (L2page_pa + (pn << PAGE_SHIFT)) |
+                                   pteflags | PG_U;
+                       }
                }
        }