[src/trunk]: src/sys/arch When the i386 port was designed, the bootstrap code...

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

details:   https://anonhg.NetBSD.org/src/rev/9d3ef82257c3
branches:  trunk
changeset: 349725:9d3ef82257c3
user:      maxv <maxv%NetBSD.org@localhost>
date:      Tue Dec 20 14:03:15 2016 +0000

description:
When the i386 port was designed, the bootstrap code needed little physical
memory, and taking it below the kernel image was fine: we had 160 free
pages, and never allocated more than 20. With amd64 however, we create a
direct map, and for this map we need a number of page table pages that is
mostly proportionate to the number of physical addresses available, which
implies that these 160 free pages may not be enough.

In particular, if the CPU does not support 1GB superpages, each 1GB chunk
of physical memory needs a 4k page in the direct map, which means that if
a machine has 160GB of ram, the bootstrap code allocates more than 160
pages, thereby overwriting the I/O mem area. If we push a little further,
if a machine has 512GB of ram, we allocate ~525 pages, and start
overwriting the kernel text, causing the system to go crazy at boot time.

Fix this moving the physical allocation area from below the kernel to above
it. avail_start is now beyond the kernel, and lowmem_rsvd indicates the
reserved low-memory pages. The area [lowmem_rsvd; IOM_BEGIN[ is
internalized into UVM, so there is no pa loss.

The only limit now is the pa of LAPIC, which is located at ~4GB of memory,
so it is perfectly fine.

This change theoretically adds va support for 512GB of ram; and it is a
prerequisite if we want to support more memory anyway.

diffstat:

 sys/arch/amd64/amd64/machdep.c |  29 +++++++++++++----------------
 sys/arch/i386/i386/machdep.c   |  13 +++++++------
 sys/arch/x86/x86/pmap.c        |   5 +++--
 sys/arch/x86/x86/x86_machdep.c |  14 +++++++-------
 4 files changed, 30 insertions(+), 31 deletions(-)

diffs (220 lines):

diff -r 5ff13fa29c04 -r 9d3ef82257c3 sys/arch/amd64/amd64/machdep.c
--- a/sys/arch/amd64/amd64/machdep.c    Tue Dec 20 12:48:30 2016 +0000
+++ b/sys/arch/amd64/amd64/machdep.c    Tue Dec 20 14:03:15 2016 +0000
@@ -1,4 +1,4 @@
-/*     $NetBSD: machdep.c,v 1.239 2016/12/16 20:16:50 maxv Exp $       */
+/*     $NetBSD: machdep.c,v 1.240 2016/12/20 14:03:15 maxv Exp $       */
 
 /*-
  * Copyright (c) 1996, 1997, 1998, 2000, 2006, 2007, 2008, 2011
@@ -111,7 +111,7 @@
  */
 
 #include <sys/cdefs.h>
-__KERNEL_RCSID(0, "$NetBSD: machdep.c,v 1.239 2016/12/16 20:16:50 maxv Exp $");
+__KERNEL_RCSID(0, "$NetBSD: machdep.c,v 1.240 2016/12/20 14:03:15 maxv Exp $");
 
 /* #define XENDEBUG_LOW  */
 
@@ -276,6 +276,7 @@
 
 struct vm_map *phys_map = NULL;
 
+extern paddr_t lowmem_rsvd;
 extern paddr_t avail_start, avail_end;
 #ifdef XEN
 extern paddr_t pmap_pa_start, pmap_pa_end;
@@ -1584,6 +1585,8 @@
 
        uvmexp.ncolors = 2;
 
+       avail_start = first_avail;
+
 #ifndef XEN
        /*
         * Low memory reservations:
@@ -1596,22 +1599,18 @@
         * Page 6:      Temporary page map level 3
         * Page 7:      Temporary page map level 4
         */
-       avail_start = 8 * PAGE_SIZE;
+       lowmem_rsvd = 8 * PAGE_SIZE;
 
        /* Initialize the memory clusters (needed in pmap_boostrap). */
        init_x86_clusters();
-#else  /* XEN */
+#else
        /* Parse Xen command line (replace bootinfo) */
        xen_parse_cmdline(XEN_PARSE_BOOTFLAGS, NULL);
 
-       /* Determine physical address space */
-       avail_start = first_avail;
        avail_end = ctob(xen_start_info.nr_pages);
        pmap_pa_start = (KERNTEXTOFF - KERNBASE);
        pmap_pa_end = avail_end;
-       __PRINTK(("pmap_pa_start 0x%lx avail_start 0x%lx avail_end 0x%lx\n",
-           pmap_pa_start, avail_start, avail_end));
-#endif /* !XEN */
+#endif
 
        /* End of the virtual space we have created so far. */
        kern_end = (vaddr_t)atdevbase + IOM_SIZE;
@@ -1628,14 +1627,12 @@
 
 #ifndef XEN
        /* Internalize the physical pages into the VM system. */
-       init_x86_vm(first_avail);
-#else  /* XEN */
+       init_x86_vm(avail_start);
+#else
        physmem = xen_start_info.nr_pages;
-
-       uvm_page_physload(atop(avail_start),
-               atop(avail_end), atop(avail_start),
-               atop(avail_end), VM_FREELIST_DEFAULT);
-#endif /* !XEN */
+       uvm_page_physload(atop(avail_start), atop(avail_end),
+           atop(avail_start), atop(avail_end), VM_FREELIST_DEFAULT);
+#endif
 
        init_x86_64_msgbuf();
 
diff -r 5ff13fa29c04 -r 9d3ef82257c3 sys/arch/i386/i386/machdep.c
--- a/sys/arch/i386/i386/machdep.c      Tue Dec 20 12:48:30 2016 +0000
+++ b/sys/arch/i386/i386/machdep.c      Tue Dec 20 14:03:15 2016 +0000
@@ -1,4 +1,4 @@
-/*     $NetBSD: machdep.c,v 1.769 2016/12/17 14:27:53 maxv Exp $       */
+/*     $NetBSD: machdep.c,v 1.770 2016/12/20 14:03:15 maxv Exp $       */
 
 /*-
  * Copyright (c) 1996, 1997, 1998, 2000, 2004, 2006, 2008, 2009
@@ -67,7 +67,7 @@
  */
 
 #include <sys/cdefs.h>
-__KERNEL_RCSID(0, "$NetBSD: machdep.c,v 1.769 2016/12/17 14:27:53 maxv Exp $");
+__KERNEL_RCSID(0, "$NetBSD: machdep.c,v 1.770 2016/12/20 14:03:15 maxv Exp $");
 
 #include "opt_beep.h"
 #include "opt_compat_ibcs2.h"
@@ -254,6 +254,7 @@
 
 struct vm_map *phys_map = NULL;
 
+extern paddr_t lowmem_rsvd;
 extern paddr_t avail_start, avail_end;
 #ifdef XEN
 extern paddr_t pmap_pa_start, pmap_pa_end;
@@ -1199,6 +1200,8 @@
         */
        uvmexp.ncolors = 2;
 
+       avail_start = first_avail;
+
 #ifndef XEN
        /*
         * Low memory reservations:
@@ -1209,7 +1212,7 @@
         * Page 4:      Temporary page table for 0MB-4MB
         * Page 5:      Temporary page directory
         */
-       avail_start = 6 * PAGE_SIZE;
+       lowmem_rsvd = 6 * PAGE_SIZE;
 #else /* !XEN */
        /* Parse Xen command line (replace bootinfo) */
        xen_parse_cmdline(XEN_PARSE_BOOTFLAGS, NULL);
@@ -1219,8 +1222,6 @@
        gdt = (void *)xen_dummy_page;
 
        /* Determine physical address space */
-       first_avail = round_page(first_avail);
-       avail_start = first_avail;
        avail_end = ctob((paddr_t)xen_start_info.nr_pages);
        pmap_pa_start = (KERNTEXTOFF - KERNBASE);
        pmap_pa_end = pmap_pa_start + ctob((paddr_t)xen_start_info.nr_pages);
@@ -1261,7 +1262,7 @@
        init_x86_clusters();
 
        /* Internalize the physical pages into the VM system. */
-       init_x86_vm(first_avail);
+       init_x86_vm(avail_start);
 #else /* !XEN */
        XENPRINTK(("load the memory cluster 0x%" PRIx64 " (%" PRId64 ") - "
            "0x%" PRIx64 " (%" PRId64 ")\n",
diff -r 5ff13fa29c04 -r 9d3ef82257c3 sys/arch/x86/x86/pmap.c
--- a/sys/arch/x86/x86/pmap.c   Tue Dec 20 12:48:30 2016 +0000
+++ b/sys/arch/x86/x86/pmap.c   Tue Dec 20 14:03:15 2016 +0000
@@ -1,4 +1,4 @@
-/*     $NetBSD: pmap.c,v 1.233 2016/12/17 13:43:33 maxv Exp $  */
+/*     $NetBSD: pmap.c,v 1.234 2016/12/20 14:03:15 maxv Exp $  */
 
 /*-
  * Copyright (c) 2008, 2010, 2016 The NetBSD Foundation, Inc.
@@ -171,7 +171,7 @@
  */
 
 #include <sys/cdefs.h>
-__KERNEL_RCSID(0, "$NetBSD: pmap.c,v 1.233 2016/12/17 13:43:33 maxv Exp $");
+__KERNEL_RCSID(0, "$NetBSD: pmap.c,v 1.234 2016/12/20 14:03:15 maxv Exp $");
 
 #include "opt_user_ldt.h"
 #include "opt_lockdebug.h"
@@ -393,6 +393,7 @@
  * (shared with machdep.c) describe the physical address space
  * of this machine.
  */
+paddr_t lowmem_rsvd __read_mostly;
 paddr_t avail_start __read_mostly; /* PA of first available physical page */
 paddr_t avail_end __read_mostly; /* PA of last available physical page */
 
diff -r 5ff13fa29c04 -r 9d3ef82257c3 sys/arch/x86/x86/x86_machdep.c
--- a/sys/arch/x86/x86/x86_machdep.c    Tue Dec 20 12:48:30 2016 +0000
+++ b/sys/arch/x86/x86/x86_machdep.c    Tue Dec 20 14:03:15 2016 +0000
@@ -1,4 +1,4 @@
-/*     $NetBSD: x86_machdep.c,v 1.78 2016/12/20 12:48:30 maxv Exp $    */
+/*     $NetBSD: x86_machdep.c,v 1.79 2016/12/20 14:03:15 maxv Exp $    */
 
 /*-
  * Copyright (c) 2002, 2006, 2007 YAMAMOTO Takashi,
@@ -31,7 +31,7 @@
  */
 
 #include <sys/cdefs.h>
-__KERNEL_RCSID(0, "$NetBSD: x86_machdep.c,v 1.78 2016/12/20 12:48:30 maxv Exp $");
+__KERNEL_RCSID(0, "$NetBSD: x86_machdep.c,v 1.79 2016/12/20 14:03:15 maxv Exp $");
 
 #include "opt_modular.h"
 #include "opt_physmem.h"
@@ -482,7 +482,7 @@
        { VM_FREELIST_FIRST16,  16 * 1024 * 1024 },
 };
 
-extern paddr_t avail_start, avail_end;
+extern paddr_t lowmem_rsvd, avail_end;
 
 int
 x86_select_freelist(uint64_t maxaddr)
@@ -833,7 +833,7 @@
 
 /*
  * init_x86_vm: initialize the VM system on x86. We basically internalize as
- * many physical pages as we can, starting at avail_start, but we don't
+ * many physical pages as we can, starting at lowmem_rsvd, but we don't
  * internalize the kernel physical pages (from pa_kstart to pa_kend).
  */
 int
@@ -865,11 +865,11 @@
                seg_end1 = 0;
 
                /* Skip memory before our available starting point. */
-               if (seg_end <= avail_start)
+               if (seg_end <= lowmem_rsvd)
                        continue;
 
-               if (seg_start <= avail_start && avail_start < seg_end) {
-                       seg_start = avail_start;
+               if (seg_start <= lowmem_rsvd && lowmem_rsvd < seg_end) {
+                       seg_start = lowmem_rsvd;
                        if (seg_start == seg_end)
                                continue;
                }