kernel26-xen/patches-2.6.25-r1/1065-2.6.25-xen-Trying-to-make-the-pagetable-initialization-code-wor.patch

From 0e9f82c8cfc17b0fb15ca0e0951687cc7c2df281 Mon Sep 17 00:00:00 2001
From: Eduardo Habkost <ehabkost@redhat.com>
Date: Wed, 12 Dec 2007 16:26:56 -0200
Subject: [PATCH] Trying to make the pagetable initialization code work under Xen

To be fixed:

- Lots of #ifdef CONFIG_XEN that will break bare metal
- Make it actually work

Wrt. pse_available() in pgtable_64.h:

- Probably we may use the CPU feature code to handle this
- We need it to be variable (not constant 0) if Xen is supported

Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
---
 arch/x86/mm/init_64.c        |  153 ++++++++++++++++++++++++++++++++++++++++--
 include/asm-x86/paravirt.h   |   10 +++
 include/asm-x86/pgalloc_64.h |    1 +
 include/asm-x86/pgtable.h    |    4 +
 4 files changed, 161 insertions(+), 7 deletions(-)

diff --git a/arch/x86/mm/init_64.c b/arch/x86/mm/init_64.c
index b5f1f6e..69d1056 100644
--- a/arch/x86/mm/init_64.c
+++ b/arch/x86/mm/init_64.c
@@ -47,6 +47,22 @@
 #include <asm/numa.h>
 #include <asm/cacheflush.h>
 
+#ifdef CONFIG_XEN
+/*FIXME: ouch. */
+#include <xen/features.h>
+#include <xen/page.h>
+#include <xen/hvc-console.h>
+#include <asm/xen/hypercall.h>
+#include "../xen/init.h"
+#include "../xen/xen-ops.h"
+#endif
+
+#ifdef CONFIG_XEN
+/*FIXME: ouch again */
+#define PTE_SIZE  PAGE_SIZE
+#else
+#endif
+
 const struct dma_mapping_ops *dma_ops;
 EXPORT_SYMBOL(dma_ops);
 
@@ -102,6 +118,9 @@ void show_mem(void)
 
 int after_bootmem;
 
+static unsigned long __initdata table_start;
+static unsigned long __meminitdata table_end;
+
 static __init void *spp_getpage(void)
 {
 	void *ptr;
@@ -232,8 +251,75 @@ native_set_fixmap(enum fixed_addresses idx, unsigned long phys, pgprot_t prot)
 	__native_set_fixmap(idx, pfn_pte(phys >> PAGE_SHIFT, prot));
 }
 
-static unsigned long __initdata table_start;
-static unsigned long __meminitdata table_end;
+
+#ifdef CONFIG_XEN
+
+/*FIXME: this code should be enabled at runtime, somehow */
+
+unsigned long start_pfn;
+
+static __meminit void *alloc_low_page(unsigned long *phys)
+{
+	unsigned long pfn;
+	void *adr;
+
+	if (after_bootmem) {
+		adr = (void *)get_zeroed_page(GFP_ATOMIC);
+		*phys = __pa(adr);
+		return adr;
+	}
+
+	pfn = start_pfn++;
+	*phys  = pfn << PAGE_SHIFT;
+
+	/* The address returned by __va() is not available yet.
+	 * This function is used by init_memory_mapping(). We will
+	 * use the mapping that was set by Xen for us.
+	 */
+	adr = (void*)((*phys) + __START_KERNEL_map);
+	memset(adr, 0, PAGE_SIZE);
+
+	return adr;
+}
+
+
+static inline void __set_pte(pte_t *dst, pte_t val)
+{
+	*dst = val;
+}
+
+
+int make_readonly(unsigned long paddr)
+{
+	extern char __vsyscall_0;
+	int readonly = 0;
+
+	/* Make new page tables read-only. */
+	if (!xen_feature(XENFEAT_writable_page_tables)
+	    && (paddr >= (table_start << PAGE_SHIFT))
+	    && (paddr < (table_end << PAGE_SHIFT)))
+		readonly = 1;
+	/* Make old page tables read-only. */
+	if (!xen_feature(XENFEAT_writable_page_tables)
+	    && (paddr >= (xen_start_info->pt_base - __START_KERNEL_map))
+	    && (paddr < (start_pfn << PAGE_SHIFT)))
+		readonly = 1;
+
+	/*
+	 * No need for writable mapping of kernel image. This also ensures that
+	 * page and descriptor tables embedded inside don't have writable
+	 * mappings. Exclude the vsyscall area here, allowing alternative
+	 * instruction patching to work.
+	 */
+	if ((paddr >= __pa_symbol(&_text)) && (paddr < __pa_symbol(&_end))
+	    && !(paddr >= __pa_symbol(&__vsyscall_0)
+	         && paddr < __pa_symbol(&__vsyscall_0) + PAGE_SIZE))
+		readonly = 1;
+
+	return readonly;
+}
+
+#else /* !XEN */
 
 static __meminit void *alloc_low_page(unsigned long *phys)
 {
@@ -256,12 +342,18 @@ static __meminit void *alloc_low_page(unsigned long *phys)
 	return adr;
 }
 
+#endif  /* !XEN */
+
+
 static __meminit void unmap_low_page(void *adr)
 {
 	if (after_bootmem)
 		return;
 
+#ifndef CONFIG_XEN
+	/*FIXME: this behaviour should be changed in runtime, also */
 	early_iounmap(adr, PAGE_SIZE);
+#endif
 }
 
 /* Must run before zap_low_mappings */
@@ -284,8 +376,9 @@ __meminit void *early_ioremap(unsigned long addr, unsigned long size)
 		vaddr += addr & ~PMD_MASK;
 		addr &= PMD_MASK;
 
-		for (i = 0; i < pmds; i++, addr += PMD_SIZE)
+		for (i = 0; i < pmds; i++, addr += PMD_SIZE) {
 			set_pmd(pmd+i, __pmd(addr | __PAGE_KERNEL_LARGE_EXEC));
+		} /* for pmds */
 		__flush_tlb_all();
 
 		return (void *)vaddr;
@@ -321,7 +414,7 @@ phys_pmd_init(pmd_t *pmd_page, unsigned long address, unsigned long end)
 {
 	int i = pmd_index(address);
 
-	for (; i < PTRS_PER_PMD; i++, address += PMD_SIZE) {
+	for (; i < PTRS_PER_PMD; i++) {
 		pmd_t *pmd = pmd_page + pmd_index(address);
 
 		if (address >= end) {
@@ -335,8 +428,46 @@ phys_pmd_init(pmd_t *pmd_page, unsigned long address, unsigned long end)
 		if (pmd_val(*pmd))
 			continue;
 
-		set_pte((pte_t *)pmd,
-			pfn_pte(address >> PAGE_SHIFT, PAGE_KERNEL_LARGE));
+		if (pse_available()) {
+			/* Lucky us :) */
+			set_pte((pte_t *)pmd,
+				pfn_pte(address >> PAGE_SHIFT,
+				PAGE_KERNEL_LARGE));
+			address += PMD_SIZE;
+		} else {
+			/* No PSE support. Must make lots of PTEs */
+			unsigned long pte_phys;
+			pte_t *pte, *pte_save;
+			int k;
+
+			pte = alloc_low_page(&pte_phys);
+			pte_save = pte;
+			for (k = 0; k < PTRS_PER_PTE; pte++, k++, address += PTE_SIZE) {
+				pteval_t attrs;
+				if ((address >= end)) {
+					__set_pte(pte, __pte(0));
+					continue;
+				}
+
+				/*FIXME: _PAGE_GLOBAL is skipped here,
+				 * but this is Xen-specific. Use pvops
+				 * to handle this
+				 */
+				attrs = __PAGE_KERNEL;
+
+				/*FIXME: somehow we need to make those
+				 * pages readonly on some other place
+				 */
+				if (make_readonly(address))
+					attrs &= ~_PAGE_RW;
+				__set_pte(pte, __pte(address | attrs));
+			}
+			pte = pte_save;
+
+			/*FIXME: this should be done somewhere else */
+			early_make_page_readonly(pte, XENFEAT_writable_page_tables);
+			set_pmd(pmd, pfn_pmd(pte_phys>>PAGE_SHIFT, PAGE_KERNEL));
+		} /* !pse_available() */
 	}
 }
 
@@ -363,11 +494,14 @@ phys_pud_init(pud_t *pud_page, unsigned long addr, unsigned long end)
 		if (addr >= end)
 			break;
 
+#ifndef CONFIG_XEN
+		/*FIXME: how to make this enabled on runtime */
 		if (!after_bootmem &&
 				!e820_any_mapped(addr, addr+PUD_SIZE, 0)) {
 			set_pud(pud, __pud(0));
 			continue;
 		}
+#endif
 
 		if (pud_val(*pud)) {
 			phys_pmd_update(pud, addr, end);
@@ -375,6 +509,8 @@ phys_pud_init(pud_t *pud_page, unsigned long addr, unsigned long end)
 		}
 
 		pmd = alloc_low_page(&pmd_phys);
+		/*FIXME: this (making page readonly) should be done somewhere else */
+		early_make_page_readonly(pmd, XENFEAT_writable_page_tables);
 
 		spin_lock(&init_mm.page_table_lock);
 		set_pud(pud, __pud(pmd_phys | _KERNPG_TABLE));
@@ -440,8 +576,11 @@ void __init_refok init_memory_mapping(unsigned long start, unsigned long end)
 
 		if (after_bootmem)
 			pud = pud_offset(pgd, start & PGDIR_MASK);
-		else
+		else {
 			pud = alloc_low_page(&pud_phys);
+			/*FIXME: this (making page readonly) should be done somewhere else */
+			early_make_page_readonly(pud, XENFEAT_writable_page_tables);
+		}
 
 		next = start + PGDIR_SIZE;
 		if (next > end)
diff --git a/include/asm-x86/paravirt.h b/include/asm-x86/paravirt.h
index c7c4e34..7edbeda 100644
--- a/include/asm-x86/paravirt.h
+++ b/include/asm-x86/paravirt.h
@@ -153,6 +153,16 @@ struct pv_cpu_ops {
 	struct pv_lazy_ops lazy_mode;
 };
 
+
+#ifdef CONFIG_XEN
+/*FIXME: return 0 only if we are really running under Xen */
+#define pse_available() (0)
+#else
+/* No Xen support, PSE is always available */
+#define pse_available() (1)
+#endif
+
+
 struct pv_irq_ops {
 	void (*init_IRQ)(void);
 
diff --git a/include/asm-x86/pgalloc_64.h b/include/asm-x86/pgalloc_64.h
index 982e1cb..b9835d6 100644
--- a/include/asm-x86/pgalloc_64.h
+++ b/include/asm-x86/pgalloc_64.h
@@ -2,6 +2,7 @@
 #define _X86_64_PGALLOC_H
 
 #include <asm/pda.h>
+#include <asm/vsyscall.h>
 #include <linux/threads.h>
 #include <linux/mm.h>
 
diff --git a/include/asm-x86/pgtable.h b/include/asm-x86/pgtable.h
index 9cf472a..6b3979e 100644
--- a/include/asm-x86/pgtable.h
+++ b/include/asm-x86/pgtable.h
@@ -232,6 +232,10 @@ static inline pte_t pte_modify(pte_t pte, pgprot_t newprot)
 
 #define pte_update(mm, addr, ptep)              do { } while (0)
 #define pte_update_defer(mm, addr, ptep)        do { } while (0)
+
+/*FIXME: this have a better place than here */
+#define pse_available() (1)
+
 #endif	/* CONFIG_PARAVIRT */
 
 #endif	/* __ASSEMBLY__ */
-- 
1.5.4.1

1	From 0e9f82c8cfc17b0fb15ca0e0951687cc7c2df281 Mon Sep 17 00:00:00 2001
2	From: Eduardo Habkost <ehabkost@redhat.com>
3	Date: Wed, 12 Dec 2007 16:26:56 -0200
4	Subject: [PATCH] Trying to make the pagetable initialization code work under Xen
5
6	To be fixed:
7
8	- Lots of #ifdef CONFIG_XEN that will break bare metal
9	- Make it actually work
10
11	Wrt. pse_available() in pgtable_64.h:
12
13	- Probably we may use the CPU feature code to handle this
14	- We need it to be variable (not constant 0) if Xen is supported
15
16	Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
17	---
18	arch/x86/mm/init_64.c \| 153 ++++++++++++++++++++++++++++++++++++++++--
19	include/asm-x86/paravirt.h \| 10 +++
20	include/asm-x86/pgalloc_64.h \| 1 +
21	include/asm-x86/pgtable.h \| 4 +
22	4 files changed, 161 insertions(+), 7 deletions(-)
23
24	diff --git a/arch/x86/mm/init_64.c b/arch/x86/mm/init_64.c
25	index b5f1f6e..69d1056 100644
26	--- a/arch/x86/mm/init_64.c
27	+++ b/arch/x86/mm/init_64.c
28	@@ -47,6 +47,22 @@
29	#include <asm/numa.h>
30	#include <asm/cacheflush.h>
31
32	+#ifdef CONFIG_XEN
33	+/FIXME: ouch. /
34	+#include <xen/features.h>
35	+#include <xen/page.h>
36	+#include <xen/hvc-console.h>
37	+#include <asm/xen/hypercall.h>
38	+#include "../xen/init.h"
39	+#include "../xen/xen-ops.h"
40	+#endif
41	+
42	+#ifdef CONFIG_XEN
43	+/FIXME: ouch again /
44	+#define PTE_SIZE PAGE_SIZE
45	+#else
46	+#endif
47	+
48	const struct dma_mapping_ops *dma_ops;
49	EXPORT_SYMBOL(dma_ops);
50
51	@@ -102,6 +118,9 @@ void show_mem(void)
52
53	int after_bootmem;
54
55	+static unsigned long __initdata table_start;
56	+static unsigned long __meminitdata table_end;
57	+
58	static __init void *spp_getpage(void)
59	{
60	void *ptr;
61	@@ -232,8 +251,75 @@ native_set_fixmap(enum fixed_addresses idx, unsigned long phys, pgprot_t prot)
62	__native_set_fixmap(idx, pfn_pte(phys >> PAGE_SHIFT, prot));
63	}
64
65	-static unsigned long __initdata table_start;
66	-static unsigned long __meminitdata table_end;
67	+
68	+#ifdef CONFIG_XEN
69	+
70	+/FIXME: this code should be enabled at runtime, somehow /
71	+
72	+unsigned long start_pfn;
73	+
74	+static __meminit void alloc_low_page(unsigned long phys)
75	+{
76	+ unsigned long pfn;
77	+ void *adr;
78	+
79	+ if (after_bootmem) {
80	+ adr = (void *)get_zeroed_page(GFP_ATOMIC);
81	+ *phys = __pa(adr);
82	+ return adr;
83	+ }
84	+
85	+ pfn = start_pfn++;
86	+ *phys = pfn << PAGE_SHIFT;
87	+
88	+ /* The address returned by __va() is not available yet.
89	+ * This function is used by init_memory_mapping(). We will
90	+ * use the mapping that was set by Xen for us.
91	+ */
92	+ adr = (void)((phys) + __START_KERNEL_map);
93	+ memset(adr, 0, PAGE_SIZE);
94	+
95	+ return adr;
96	+}
97	+
98	+
99	+static inline void __set_pte(pte_t *dst, pte_t val)
100	+{
101	+ *dst = val;
102	+}
103	+
104	+
105	+int make_readonly(unsigned long paddr)
106	+{
107	+ extern char __vsyscall_0;
108	+ int readonly = 0;
109	+
110	+ /* Make new page tables read-only. */
111	+ if (!xen_feature(XENFEAT_writable_page_tables)
112	+ && (paddr >= (table_start << PAGE_SHIFT))
113	+ && (paddr < (table_end << PAGE_SHIFT)))
114	+ readonly = 1;
115	+ /* Make old page tables read-only. */
116	+ if (!xen_feature(XENFEAT_writable_page_tables)
117	+ && (paddr >= (xen_start_info->pt_base - __START_KERNEL_map))
118	+ && (paddr < (start_pfn << PAGE_SHIFT)))
119	+ readonly = 1;
120	+
121	+ /*
122	+ * No need for writable mapping of kernel image. This also ensures that
123	+ * page and descriptor tables embedded inside don't have writable
124	+ * mappings. Exclude the vsyscall area here, allowing alternative
125	+ * instruction patching to work.
126	+ */
127	+ if ((paddr >= __pa_symbol(&_text)) && (paddr < __pa_symbol(&_end))
128	+ && !(paddr >= __pa_symbol(&__vsyscall_0)
129	+ && paddr < __pa_symbol(&__vsyscall_0) + PAGE_SIZE))
130	+ readonly = 1;
131	+
132	+ return readonly;
133	+}
134	+
135	+#else /* !XEN */
136
137	static __meminit void alloc_low_page(unsigned long phys)
138	{
139	@@ -256,12 +342,18 @@ static __meminit void alloc_low_page(unsigned long phys)
140	return adr;
141	}
142
143	+#endif /* !XEN */
144	+
145	+
146	static __meminit void unmap_low_page(void *adr)
147	{
148	if (after_bootmem)
149	return;
150
151	+#ifndef CONFIG_XEN
152	+ /FIXME: this behaviour should be changed in runtime, also /
153	early_iounmap(adr, PAGE_SIZE);
154	+#endif
155	}
156
157	/* Must run before zap_low_mappings */
158	@@ -284,8 +376,9 @@ __meminit void *early_ioremap(unsigned long addr, unsigned long size)
159	vaddr += addr & ~PMD_MASK;
160	addr &= PMD_MASK;
161
162	- for (i = 0; i < pmds; i++, addr += PMD_SIZE)
163	+ for (i = 0; i < pmds; i++, addr += PMD_SIZE) {
164	set_pmd(pmd+i, __pmd(addr \| __PAGE_KERNEL_LARGE_EXEC));
165	+ } /* for pmds */
166	__flush_tlb_all();
167
168	return (void *)vaddr;
169	@@ -321,7 +414,7 @@ phys_pmd_init(pmd_t *pmd_page, unsigned long address, unsigned long end)
170	{
171	int i = pmd_index(address);
172
173	- for (; i < PTRS_PER_PMD; i++, address += PMD_SIZE) {
174	+ for (; i < PTRS_PER_PMD; i++) {
175	pmd_t *pmd = pmd_page + pmd_index(address);
176
177	if (address >= end) {
178	@@ -335,8 +428,46 @@ phys_pmd_init(pmd_t *pmd_page, unsigned long address, unsigned long end)
179	if (pmd_val(*pmd))
180	continue;
181
182	- set_pte((pte_t *)pmd,
183	- pfn_pte(address >> PAGE_SHIFT, PAGE_KERNEL_LARGE));
184	+ if (pse_available()) {
185	+ /* Lucky us :) */
186	+ set_pte((pte_t *)pmd,
187	+ pfn_pte(address >> PAGE_SHIFT,
188	+ PAGE_KERNEL_LARGE));
189	+ address += PMD_SIZE;
190	+ } else {
191	+ /* No PSE support. Must make lots of PTEs */
192	+ unsigned long pte_phys;
193	+ pte_t pte, pte_save;
194	+ int k;
195	+
196	+ pte = alloc_low_page(&pte_phys);
197	+ pte_save = pte;
198	+ for (k = 0; k < PTRS_PER_PTE; pte++, k++, address += PTE_SIZE) {
199	+ pteval_t attrs;
200	+ if ((address >= end)) {
201	+ __set_pte(pte, __pte(0));
202	+ continue;
203	+ }
204	+
205	+ /*FIXME: _PAGE_GLOBAL is skipped here,
206	+ * but this is Xen-specific. Use pvops
207	+ * to handle this
208	+ */
209	+ attrs = __PAGE_KERNEL;
210	+
211	+ /*FIXME: somehow we need to make those
212	+ * pages readonly on some other place
213	+ */
214	+ if (make_readonly(address))
215	+ attrs &= ~_PAGE_RW;
216	+ __set_pte(pte, __pte(address \| attrs));
217	+ }
218	+ pte = pte_save;
219	+
220	+ /FIXME: this should be done somewhere else /
221	+ early_make_page_readonly(pte, XENFEAT_writable_page_tables);
222	+ set_pmd(pmd, pfn_pmd(pte_phys>>PAGE_SHIFT, PAGE_KERNEL));
223	+ } /* !pse_available() */
224	}
225	}
226
227	@@ -363,11 +494,14 @@ phys_pud_init(pud_t *pud_page, unsigned long addr, unsigned long end)
228	if (addr >= end)
229	break;
230
231	+#ifndef CONFIG_XEN
232	+ /FIXME: how to make this enabled on runtime /
233	if (!after_bootmem &&
234	!e820_any_mapped(addr, addr+PUD_SIZE, 0)) {
235	set_pud(pud, __pud(0));
236	continue;
237	}
238	+#endif
239
240	if (pud_val(*pud)) {
241	phys_pmd_update(pud, addr, end);
242	@@ -375,6 +509,8 @@ phys_pud_init(pud_t *pud_page, unsigned long addr, unsigned long end)
243	}
244
245	pmd = alloc_low_page(&pmd_phys);
246	+ /FIXME: this (making page readonly) should be done somewhere else /
247	+ early_make_page_readonly(pmd, XENFEAT_writable_page_tables);
248
249	spin_lock(&init_mm.page_table_lock);
250	set_pud(pud, __pud(pmd_phys \| _KERNPG_TABLE));
251	@@ -440,8 +576,11 @@ void __init_refok init_memory_mapping(unsigned long start, unsigned long end)
252
253	if (after_bootmem)
254	pud = pud_offset(pgd, start & PGDIR_MASK);
255	- else
256	+ else {
257	pud = alloc_low_page(&pud_phys);
258	+ /FIXME: this (making page readonly) should be done somewhere else /
259	+ early_make_page_readonly(pud, XENFEAT_writable_page_tables);
260	+ }
261
262	next = start + PGDIR_SIZE;
263	if (next > end)
264	diff --git a/include/asm-x86/paravirt.h b/include/asm-x86/paravirt.h
265	index c7c4e34..7edbeda 100644
266	--- a/include/asm-x86/paravirt.h
267	+++ b/include/asm-x86/paravirt.h
268	@@ -153,6 +153,16 @@ struct pv_cpu_ops {
269	struct pv_lazy_ops lazy_mode;
270	};
271
272	+
273	+#ifdef CONFIG_XEN
274	+/FIXME: return 0 only if we are really running under Xen /
275	+#define pse_available() (0)
276	+#else
277	+/* No Xen support, PSE is always available */
278	+#define pse_available() (1)
279	+#endif
280	+
281	+
282	struct pv_irq_ops {
283	void (*init_IRQ)(void);
284
285	diff --git a/include/asm-x86/pgalloc_64.h b/include/asm-x86/pgalloc_64.h
286	index 982e1cb..b9835d6 100644
287	--- a/include/asm-x86/pgalloc_64.h
288	+++ b/include/asm-x86/pgalloc_64.h
289	@@ -2,6 +2,7 @@
290	#define _X86_64_PGALLOC_H
291
292	#include <asm/pda.h>
293	+#include <asm/vsyscall.h>
294	#include <linux/threads.h>
295	#include <linux/mm.h>
296
297	diff --git a/include/asm-x86/pgtable.h b/include/asm-x86/pgtable.h
298	index 9cf472a..6b3979e 100644
299	--- a/include/asm-x86/pgtable.h
300	+++ b/include/asm-x86/pgtable.h
301	@@ -232,6 +232,10 @@ static inline pte_t pte_modify(pte_t pte, pgprot_t newprot)
302
303	#define pte_update(mm, addr, ptep) do { } while (0)
304	#define pte_update_defer(mm, addr, ptep) do { } while (0)
305	+
306	+/FIXME: this have a better place than here /
307	+#define pse_available() (1)
308	+
309	#endif /* CONFIG_PARAVIRT */
310
311	#endif /* __ASSEMBLY__ */
312	--
313	1.5.4.1
314