LKML Archive on lore.kernel.org
help / color / mirror / Atom feed
From: Jeremy Fitzhardinge <jeremy@goop.org>
To: Andi Kleen <ak@muc.de>
Cc: Andrew Morton <akpm@linux-foundation.org>,
	linux-kernel@vger.kernel.org, virtualization@lists.osdl.org,
	xen-devel@lists.xensource.com, Chris Wright <chrisw@sous-sol.org>,
	Zachary Amsden <zach@vmware.com>
Subject: [patch 05/21] Xen-paravirt: paravirt_ops: hooks to set up initial pagetable
Date: Thu, 15 Feb 2007 18:24:54 -0800	[thread overview]
Message-ID: <20070216022530.856717545@goop.org> (raw)
In-Reply-To: <20070216022449.739760547@goop.org>

[-- Attachment #1: paravirt-memory-init.patch --]
[-- Type: text/plain, Size: 12019 bytes --]

This patch introduces paravirt_ops hooks to control how the kernel's
initial pagetable is set up.

In the case of a native boot, the very early bootstrap code creates a
simple non-PAE pagetable to map the kernel and physical memory.  When
the VM subsystem is initialized, it creates a proper pagetable which
respects the PAE mode, large pages, etc.

When booting under a hypervisor, there are many possibilities for what
paging environment the hypervisor establishes for the guest kernel, so
the constructon of the kernel's pagetable depends on the hypervisor.

In the case of Xen, the hypervisor boots the kernel with a fully
constructed pagetable, which is already using PAE if necessary.  Also,
Xen requires particular care when constructing pagetables to make sure
all pagetables are always mapped read-only.

In order to make this easier, kernel's initial pagetable construction
has been changed to only allocate and initialize a pagetable page if
there's no page already present in the pagetable.  This allows the Xen
paravirt backend to make a copy of the hypervisor-provided pagetable,
allowing the kernel to establish any more mappings it needs while
keeping the existing ones.

A slightly subtle point which is worth highlighting here is that Xen
requires all kernel mappings to share the same pte_t pages between all
pagetables, so that updating a kernel page's mapping in one pagetable
is reflected in all other pagetables.  This makes it possible to
allocate a page and attach it to a pagetable without having to
explicitly enumerate that page's mapping in all pagetables.

Signed-off-by: Jeremy Fitzhardinge <jeremy@xensource.com>

--
 arch/i386/kernel/paravirt.c |   40 ++++++++++++++++++
 arch/i386/mm/init.c         |   89 +++++++++++++++++++++++++-----------------
 include/asm-i386/paravirt.h |   54 +++++++++++++++++++++++++
 include/asm-i386/pgtable.h  |    3 +
 4 files changed, 150 insertions(+), 36 deletions(-)

===================================================================
--- a/arch/i386/kernel/paravirt.c
+++ b/arch/i386/kernel/paravirt.c
@@ -379,6 +379,43 @@ static void native_io_delay(void)
 {
 	asm volatile("outb %al,$0x80");
 }
+
+void native_pagetable_setup_start(pgd_t *base)
+{
+#ifdef CONFIG_X86_PAE
+	int i;
+
+	/*
+	 * Init entries of the first-level page table to the
+	 * zero page, if they haven't already been set up.
+	 *
+	 * In a normal native boot, we'll be running on a
+	 * pagetable rooted in swapper_pg_dir, but not in PAE
+	 * mode, so this will end up clobbering the mappings
+	 * for the lower 24Mbytes of the address space,
+	 * without affecting the kernel address space.
+	 */
+	for (i = 0; i < USER_PTRS_PER_PGD; i++)
+		set_pgd(&base[i],
+			__pgd(__pa(empty_zero_page) | _PAGE_PRESENT));
+	memset(&base[USER_PTRS_PER_PGD], 0, sizeof(pgd_t));
+#endif
+}
+
+void native_pagetable_setup_done(pgd_t *base)
+{
+#ifdef CONFIG_X86_PAE
+	/*
+	 * Add low memory identity-mappings - SMP needs it when
+	 * starting up on an AP from real-mode. In the non-PAE
+	 * case we already have these mappings through head.S.
+	 * All user-space mappings are explicitly cleared after
+	 * SMP startup.
+	 */
+	set_pgd(&base[0], base[USER_PTRS_PER_PGD]);
+#endif
+}
+
 
 static void native_flush_tlb(void)
 {
@@ -627,6 +664,9 @@ struct paravirt_ops paravirt_ops = {
 #endif
 	.set_lazy_mode = (void *)native_nop,
 
+	.pagetable_setup_start = native_pagetable_setup_start,
+	.pagetable_setup_done = native_pagetable_setup_done,
+
 	.flush_tlb_user = native_flush_tlb,
 	.flush_tlb_kernel = native_flush_tlb_global,
 	.flush_tlb_single = native_flush_tlb_single,
===================================================================
--- a/arch/i386/mm/init.c
+++ b/arch/i386/mm/init.c
@@ -42,6 +42,7 @@
 #include <asm/tlb.h>
 #include <asm/tlbflush.h>
 #include <asm/sections.h>
+#include <asm/paravirt.h>
 
 unsigned int __VMALLOC_RESERVE = 128 << 20;
 
@@ -62,6 +63,8 @@ static pmd_t * __init one_md_table_init(
 		
 #ifdef CONFIG_X86_PAE
 	pmd_table = (pmd_t *) alloc_bootmem_low_pages(PAGE_SIZE);
+	memset(pmd_table, 0, PAGE_SIZE);
+
 	paravirt_alloc_pd(__pa(pmd_table) >> PAGE_SHIFT);
 	set_pgd(pgd, __pgd(__pa(pmd_table) | _PAGE_PRESENT));
 	pud = pud_offset(pgd, 0);
@@ -83,12 +86,11 @@ static pte_t * __init one_page_table_ini
 {
 	if (pmd_none(*pmd)) {
 		pte_t *page_table = (pte_t *) alloc_bootmem_low_pages(PAGE_SIZE);
+		memset(page_table, 0, PAGE_SIZE);
+
 		paravirt_alloc_pt(__pa(page_table) >> PAGE_SHIFT);
 		set_pmd(pmd, __pmd(__pa(page_table) | _PAGE_TABLE));
-		if (page_table != pte_offset_kernel(pmd, 0))
-			BUG();	
-
-		return page_table;
+		BUG_ON(page_table != pte_offset_kernel(pmd, 0));
 	}
 	
 	return pte_offset_kernel(pmd, 0);
@@ -119,7 +121,7 @@ static void __init page_table_range_init
 	pgd = pgd_base + pgd_idx;
 
 	for ( ; (pgd_idx < PTRS_PER_PGD) && (vaddr != end); pgd++, pgd_idx++) {
-		if (pgd_none(*pgd)) 
+		if (!(pgd_val(*pgd) & _PAGE_PRESENT)) 
 			one_md_table_init(pgd);
 		pud = pud_offset(pgd, vaddr);
 		pmd = pmd_offset(pud, vaddr);
@@ -158,7 +160,11 @@ static void __init kernel_physical_mappi
 	pfn = 0;
 
 	for (; pgd_idx < PTRS_PER_PGD; pgd++, pgd_idx++) {
-		pmd = one_md_table_init(pgd);
+		if (!(pgd_val(*pgd) & _PAGE_PRESENT))
+			pmd = one_md_table_init(pgd);
+		else
+			pmd = pmd_offset(pud_offset(pgd, PAGE_OFFSET), PAGE_OFFSET);
+
 		if (pfn >= max_low_pfn)
 			continue;
 		for (pmd_idx = 0; pmd_idx < PTRS_PER_PMD && pfn < max_low_pfn; pmd++, pmd_idx++) {
@@ -167,20 +173,26 @@ static void __init kernel_physical_mappi
 			/* Map with big pages if possible, otherwise create normal page tables. */
 			if (cpu_has_pse) {
 				unsigned int address2 = (pfn + PTRS_PER_PTE - 1) * PAGE_SIZE + PAGE_OFFSET + PAGE_SIZE-1;
-
-				if (is_kernel_text(address) || is_kernel_text(address2))
-					set_pmd(pmd, pfn_pmd(pfn, PAGE_KERNEL_LARGE_EXEC));
-				else
-					set_pmd(pmd, pfn_pmd(pfn, PAGE_KERNEL_LARGE));
+				if (!pmd_present(*pmd)) {
+					if (is_kernel_text(address) || is_kernel_text(address2))
+						set_pmd(pmd, pfn_pmd(pfn, PAGE_KERNEL_LARGE_EXEC));
+					else
+						set_pmd(pmd, pfn_pmd(pfn, PAGE_KERNEL_LARGE));
+				}
 				pfn += PTRS_PER_PTE;
 			} else {
 				pte = one_page_table_init(pmd);
 
-				for (pte_ofs = 0; pte_ofs < PTRS_PER_PTE && pfn < max_low_pfn; pte++, pfn++, pte_ofs++) {
-						if (is_kernel_text(address))
-							set_pte(pte, pfn_pte(pfn, PAGE_KERNEL_EXEC));
-						else
-							set_pte(pte, pfn_pte(pfn, PAGE_KERNEL));
+				for (pte_ofs = 0; 
+				     pte_ofs < PTRS_PER_PTE && pfn < max_low_pfn;
+				     pte++, pfn++, pte_ofs++, address += PAGE_SIZE) {
+					if (pte_present(*pte))
+						continue;
+
+					if (is_kernel_text(address))
+						set_pte(pte, pfn_pte(pfn, PAGE_KERNEL_EXEC));
+					else
+						set_pte(pte, pfn_pte(pfn, PAGE_KERNEL));
 				}
 			}
 		}
@@ -337,19 +349,32 @@ extern void __init remap_numa_kva(void);
 #define remap_numa_kva() do {} while (0)
 #endif
 
+/* 
+ * Build a proper pagetable for the kernel mappings.  Up until this
+ * point, we've been running on some set of pagetables constructed by
+ * the boot process.
+ *
+ * If we're booting on native hardware, this will be a pagetable
+ * constructed in arch/i386/kernel/head.S, and not running in PAE mode
+ * (even if we'll end up running in PAE).  The root of the pagetable
+ * will be swapper_pg_dir.
+ *
+ * If we're booting paravirtualized under a hypervisor, then there are
+ * more options: we may already be running PAE, and the pagetable may
+ * or may not be based in swapper_pg_dir.  In any case,
+ * paravirt_pagetable_setup_start() will set up swapper_pg_dir
+ * appropriately for the rest of the initialization to work.
+ *
+ * In general, pagetable_init() assumes that the pagetable may already
+ * be partially populated, and so it avoids stomping on any existing
+ * mappings.
+ */
 static void __init pagetable_init (void)
 {
-	unsigned long vaddr;
+	unsigned long vaddr, end;
 	pgd_t *pgd_base = swapper_pg_dir;
 
-#ifdef CONFIG_X86_PAE
-	int i;
-	/* Init entries of the first-level page table to the zero page */
-	for (i = 0; i < PTRS_PER_PGD; i++)
-		set_pgd(pgd_base + i, __pgd(__pa(empty_zero_page) | _PAGE_PRESENT));
-#else
-	paravirt_alloc_pd(__pa(swapper_pg_dir) >> PAGE_SHIFT);
-#endif
+	paravirt_pagetable_setup_start(pgd_base);
 
 	/* Enable PSE if available */
 	if (cpu_has_pse) {
@@ -371,20 +396,12 @@ static void __init pagetable_init (void)
 	 * created - mappings will be set by set_fixmap():
 	 */
 	vaddr = __fix_to_virt(__end_of_fixed_addresses - 1) & PMD_MASK;
-	page_table_range_init(vaddr, 0, pgd_base);
+	end = (FIXADDR_TOP + PMD_SIZE - 1) & PMD_MASK;
+	page_table_range_init(vaddr, end, pgd_base);
 
 	permanent_kmaps_init(pgd_base);
 
-#ifdef CONFIG_X86_PAE
-	/*
-	 * Add low memory identity-mappings - SMP needs it when
-	 * starting up on an AP from real-mode. In the non-PAE
-	 * case we already have these mappings through head.S.
-	 * All user-space mappings are explicitly cleared after
-	 * SMP startup.
-	 */
-	set_pgd(&pgd_base[0], pgd_base[USER_PTRS_PER_PGD]);
-#endif
+	paravirt_pagetable_setup_done(pgd_base);
 }
 
 #if defined(CONFIG_SOFTWARE_SUSPEND) || defined(CONFIG_ACPI_SLEEP)
===================================================================
--- a/include/asm-i386/paravirt.h
+++ b/include/asm-i386/paravirt.h
@@ -49,6 +49,9 @@ struct paravirt_ops
 	void (*arch_setup)(void);
 	char *(*memory_setup)(void);
 	void (*init_IRQ)(void);
+
+	void (*pagetable_setup_start)(pgd_t *pgd_base);
+	void (*pagetable_setup_done)(pgd_t *pgd_base);
 
 	void (*banner)(void);
 
@@ -185,6 +188,8 @@ struct paravirt_ops
 
 extern struct paravirt_ops paravirt_ops;
 
+void native_pagetable_setup_start(pgd_t *pgd);
+
 #ifdef CONFIG_X86_PAE
 unsigned long long native_pte_val(pte_t);
 unsigned long long native_pmd_val(pmd_t);
@@ -389,6 +394,17 @@ static inline void setup_secondary_clock
 }
 #endif
 
+static inline void paravirt_pagetable_setup_start(pgd_t *base)
+{
+	if (paravirt_ops.pagetable_setup_start)
+		(*paravirt_ops.pagetable_setup_start)(base);
+}
+
+static inline void paravirt_pagetable_setup_done(pgd_t *base)
+{
+	if (paravirt_ops.pagetable_setup_done)
+		(*paravirt_ops.pagetable_setup_done)(base);
+}
 
 void native_set_pte(pte_t *ptep, pte_t pteval);
 void native_set_pte_at(struct mm_struct *mm, u32 addr, pte_t *ptep, pte_t pteval);
@@ -615,5 +631,43 @@ 772:;						\
 	call *paravirt_ops+PARAVIRT_read_cr0
 
 #endif /* __ASSEMBLY__ */
+#else  /* !CONFIG_PARAVIRT */
+#include <asm/pgtable.h>
+
+static inline void paravirt_pagetable_setup_start(pgd_t *base)
+{
+#ifdef CONFIG_X86_PAE
+	int i;
+
+	/*
+	 * Init entries of the first-level page table to the
+	 * zero page, if they haven't already been set up.
+	 *
+	 * In a normal native boot, we'll be running on a
+	 * pagetable rooted in swapper_pg_dir, but not in PAE
+	 * mode, so this will end up clobbering the mappings
+	 * for the lower 24Mbytes of the address space,
+	 * without affecting the kernel address space.
+	 */
+	for (i = 0; i < USER_PTRS_PER_PGD; i++)
+		set_pgd(&base[i],
+			__pgd(__pa(empty_zero_page) | _PAGE_PRESENT));
+	memset(&base[USER_PTRS_PER_PGD], 0, sizeof(pgd_t));
+#endif
+}
+
+static inline void paravirt_pagetable_setup_done(pgd_t *base)
+{
+#ifdef CONFIG_X86_PAE
+	/*
+	 * Add low memory identity-mappings - SMP needs it when
+	 * starting up on an AP from real-mode. In the non-PAE
+	 * case we already have these mappings through head.S.
+	 * All user-space mappings are explicitly cleared after
+	 * SMP startup.
+	 */
+	set_pgd(&base[0], base[USER_PTRS_PER_PGD]);
+#endif
+}
 #endif /* CONFIG_PARAVIRT */
 #endif	/* __ASM_PARAVIRT_H */
===================================================================
--- a/include/asm-i386/pgtable.h
+++ b/include/asm-i386/pgtable.h
@@ -15,7 +15,10 @@
 #include <asm/processor.h>
 #include <asm/fixmap.h>
 #include <linux/threads.h>
+
+#ifdef CONFIG_PARAVIRT		/* guarded to prevent cyclic dependency */
 #include <asm/paravirt.h>
+#endif
 
 #ifndef _I386_BITOPS_H
 #include <asm/bitops.h>

-- 


  parent reply	other threads:[~2007-02-16  4:46 UTC|newest]

Thread overview: 97+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2007-02-16  2:24 [patch 00/21] Xen-paravirt: Xen guest implementation for paravirt_ops interface Jeremy Fitzhardinge
2007-02-16  2:24 ` [patch 01/21] Xen-paravirt: Fix typo in sync_constant_test_bit()s name Jeremy Fitzhardinge
2007-02-16  2:24 ` [patch 02/21] Xen-paravirt: ignore vgacon if hardware not present Jeremy Fitzhardinge
2007-02-16  2:24 ` [patch 03/21] Xen-paravirt: Add pagetable accessors to pack and unpack pagetable entries Jeremy Fitzhardinge
2007-02-16  2:24 ` [patch 04/21] Xen-paravirt: =================================================================== Jeremy Fitzhardinge
2007-02-16  9:12   ` Andi Kleen
2007-02-16 17:32     ` Jeremy Fitzhardinge
2007-02-16  2:24 ` Jeremy Fitzhardinge [this message]
2007-02-16  2:24 ` [patch 06/21] Xen-paravirt: paravirt_ops: allocate a fixmap slot Jeremy Fitzhardinge
2007-02-16  2:24 ` [patch 07/21] Xen-paravirt: remove ctor for pgd cache Jeremy Fitzhardinge
2007-02-16  6:04   ` Zachary Amsden
2007-02-16  8:39   ` Pekka Enberg
2007-02-16  9:19     ` Nick Piggin
2007-02-16 17:19       ` Jeremy Fitzhardinge
2007-02-16 20:33   ` Christoph Lameter
2007-02-16 23:33     ` Jeremy Fitzhardinge
2007-02-16  2:24 ` [patch 08/21] Xen-paravirt: Allow paravirt backend to choose kernel PMD sharing Jeremy Fitzhardinge
2007-02-16  2:24 ` [patch 09/21] Xen-paravirt: Allow paravirt backend to select PGD allocation alignment Jeremy Fitzhardinge
2007-02-16  2:24 ` [patch 10/21] Xen-paravirt: add hooks to intercept mm creation and destruction Jeremy Fitzhardinge
2007-02-16  6:34   ` Andrew Morton
2007-02-16  6:55     ` Jeremy Fitzhardinge
2007-02-22 20:16   ` Pavel Machek
2007-02-22 20:21     ` Jeremy Fitzhardinge
2007-02-22 20:23     ` Jeremy Fitzhardinge
2007-02-16  2:25 ` [patch 11/21] Xen-paravirt: Add apply_to_page_range() which applies a function to a pte range Jeremy Fitzhardinge
2007-02-16  6:37   ` Andrew Morton
2007-02-16  7:06     ` Jeremy Fitzhardinge
2007-02-16  7:19       ` Andrew Morton
2007-02-16  7:23         ` Jeremy Fitzhardinge
2007-02-16  7:31     ` Nick Piggin
2007-02-16  2:25 ` [patch 12/21] Xen-paravirt: Allocate and free vmalloc areas Jeremy Fitzhardinge
2007-02-16  6:43   ` Andrew Morton
2007-02-16  7:08     ` Jeremy Fitzhardinge
2007-02-16  7:24       ` Andrew Morton
2007-02-16  7:30         ` Jeremy Fitzhardinge
2007-02-16  7:49           ` Andrew Morton
2007-02-16  9:18           ` Andi Kleen
2007-02-16 11:10             ` [Xen-devel] " Keir Fraser
2007-02-16 11:34               ` Andi Kleen
2007-02-16 16:46             ` Jeremy Fitzhardinge
2007-02-16 17:10               ` [Xen-devel] " Keir Fraser
2007-02-16 17:12                 ` Keir Fraser
2007-02-16 17:27                   ` Jeremy Fitzhardinge
2007-02-16 19:06                     ` Keir Fraser
2007-02-16 19:19                       ` Keir Fraser
2007-02-16 19:26                         ` Jeremy Fitzhardinge
2007-02-16 23:29                           ` Keir Fraser
2007-02-16 23:41                             ` Jeremy Fitzhardinge
2007-02-16 17:26                 ` Hollis Blanchard
2007-02-16  7:48         ` Nick Piggin
2007-02-16  2:25 ` [patch 13/21] Xen-paravirt: Add nosegneg capability to the vsyscall page notes Jeremy Fitzhardinge
2007-02-16  6:06   ` Zachary Amsden
2007-02-16  2:25 ` [patch 14/21] Xen-paravirt: Add XEN config options and disable unsupported config options Jeremy Fitzhardinge
2007-02-16  6:14   ` Dan Hecht
2007-02-16  7:04     ` Jeremy Fitzhardinge
2007-02-16  7:52       ` Dan Hecht
2007-02-16  8:05         ` Jeremy Fitzhardinge
2007-02-16  8:37           ` Dan Hecht
2007-02-16 10:49             ` Keir Fraser
2007-02-16  7:06     ` Andrew Morton
2007-02-16  7:25       ` Jeremy Fitzhardinge
2007-02-16 10:09         ` Keir Fraser
2007-02-16 10:19           ` [Xen-devel] " Zachary Amsden
2007-02-16 10:44             ` Keir Fraser
2007-02-16 10:55               ` Zachary Amsden
2007-02-17  4:04             ` Rusty Russell
2007-02-16  6:15   ` Zachary Amsden
2007-02-16  7:33     ` Eric W. Biederman
2007-02-16  8:14       ` Ian Campbell
2007-02-16  8:24         ` Eric W. Biederman
2007-02-16  8:31           ` Zachary Amsden
2007-02-16 13:53       ` [Xen-devel] " Gerd Hoffmann
2007-02-18 11:32     ` Avi Kivity
2007-02-16  2:25 ` [patch 16/21] Xen-paravirt: Core Xen implementation Jeremy Fitzhardinge
2007-02-16  2:25 ` [patch 17/21] Xen-paravirt: Add the Xen virtual console driver Jeremy Fitzhardinge
2007-02-16  6:48   ` Andrew Morton
2007-02-16  2:25 ` [patch 18/21] Xen-paravirt: Add Xen grant table support Jeremy Fitzhardinge
2007-02-16  6:52   ` Andrew Morton
2007-02-16  6:58     ` Jeremy Fitzhardinge
2007-02-16  2:25 ` [patch 19/21] Xen-paravirt: Add the Xenbus sysfs and virtual device hotplug driver Jeremy Fitzhardinge
2007-02-16  2:25 ` [patch 20/21] Xen-paravirt: Add Xen virtual block device driver Jeremy Fitzhardinge
2007-02-16  2:25 ` [patch 21/21] Xen-paravirt: Add the Xen virtual network " Jeremy Fitzhardinge
2007-02-16  6:59 ` [patch 00/21] Xen-paravirt: Xen guest implementation for paravirt_ops interface Andrew Morton
2007-02-16  7:20   ` Jeremy Fitzhardinge
2007-02-16 20:49 ` Christoph Lameter
2007-02-16 21:04   ` Zachary Amsden
2007-02-16 21:13     ` Christoph Lameter
2007-02-16 21:48       ` Zachary Amsden
2007-02-16 21:59         ` Christoph Lameter
2007-02-16 22:10           ` Zachary Amsden
2007-02-17 13:51           ` Andi Kleen
2007-02-21 18:37             ` Christoph Lameter
2007-02-21 18:55               ` Zachary Amsden
2007-02-21 20:02               ` Jeremy Fitzhardinge
2007-02-17  5:05         ` Rusty Russell
2007-02-16 23:49   ` Jeremy Fitzhardinge
2007-02-17  4:58   ` Rusty Russell

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=20070216022530.856717545@goop.org \
    --to=jeremy@goop.org \
    --cc=ak@muc.de \
    --cc=akpm@linux-foundation.org \
    --cc=chrisw@sous-sol.org \
    --cc=linux-kernel@vger.kernel.org \
    --cc=virtualization@lists.osdl.org \
    --cc=xen-devel@lists.xensource.com \
    --cc=zach@vmware.com \
    --subject='Re: [patch 05/21] Xen-paravirt: paravirt_ops: hooks to set up initial pagetable' \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).