RE: page allocation/attributes question (i386/x86_64 specific)

[<Stuart_Hayes@Dell.com>]

[-- Attachment #1: Type: text/plain, Size: 5598 bytes --]

> 
> So, if I understand correctly what's going on in x86_64, your fix
> wouldn't be applicable to i386.  In x86_64, every large page has a
> correct "ref_prot" that is the normal setting for that page... but in
> i386, the kernel text area does not--it should ideally be split into
> small pages all the time if there are both kernel code & free pages
> residing in the same 2M area.     
> 
> Stuart

(This isn't a submission--I'm just posting this for comments.)

Right now, any large page that touches anywhere from PAGE_OFFSET to
__init_end is initially set up as a large, executable page... but some
of this area contains data & free pages.  The patch below adds a
"cleanup_nx_in_kerneltext()" function, called at the end of
free_initmem(), which changes these pages--except for the range from
"_text" to "_etext"--to PAGE_KERNEL (i.e., non-executable).

This does result in two large pages being split up into small PTEs
permanently, but all the non-code regions will be non-executable, and
change_page_attr() will work correctly.

What do you think of this?  I have tested this on 2.6.12.

(I've attached the patch as a file, too, since my mail server can't be
convinced to not wrap text.)

Stuart

-----


diff -purN --exclude='*.o' --exclude='*.cmd'
linux-2.6.12grep/arch/i386/mm/init.c linux-2.6.12/arch/i386/mm/init.c
--- linux-2.6.12grep/arch/i386/mm/init.c	2005-07-01
15:09:27.000000000 -0500
+++ linux-2.6.12/arch/i386/mm/init.c	2005-07-01 15:13:06.000000000
-0500
@@ -666,6 +666,30 @@ static int noinline do_test_wp_bit(void)
 	return flag;
 }
 
+extern int change_page_attr_perm(struct page *, int, pgprot_t);
+
+/*
+ * In kernel_physical_mapping_init(), any big pages that contained
kernel text area were
+ * set up as big executable pages.  This function should be called when
the initmem
+ * is freed, to correctly set up the executable & non-executable pages
in this area.
+ */
+static void cleanup_nx_in_kerneltext(void)
+{
+	unsigned long from, to;
+
+	if (!nx_enabled) return;
+
+	from = PAGE_OFFSET;
+	to = (unsigned long)_text & PAGE_MASK;
+	for (; from<to; from += PAGE_SIZE)
+		change_page_attr_perm(virt_to_page(from), 1,
PAGE_KERNEL); 
+	
+	from = ((unsigned long)_etext + PAGE_SIZE - 1) & PAGE_MASK;
+	to = ((unsigned long)__init_end + LARGE_PAGE_SIZE) &
LARGE_PAGE_MASK;
+	for (; from<to; from += PAGE_SIZE)
+		change_page_attr_perm(virt_to_page(from), 1,
PAGE_KERNEL); 
+}
+
 void free_initmem(void)
 {
 	unsigned long addr;
@@ -679,6 +703,8 @@ void free_initmem(void)
 		totalram_pages++;
 	}
 	printk (KERN_INFO "Freeing unused kernel memory: %dk freed\n",
(__init_end - __init_begin) >> 10);
+
+	cleanup_nx_in_kerneltext();
 }
 
 #ifdef CONFIG_BLK_DEV_INITRD
diff -purN --exclude='*.o' --exclude='*.cmd'
linux-2.6.12grep/arch/i386/mm/pageattr.c
linux-2.6.12/arch/i386/mm/pageattr.c
--- linux-2.6.12grep/arch/i386/mm/pageattr.c	2005-07-01
15:09:08.000000000 -0500
+++ linux-2.6.12/arch/i386/mm/pageattr.c	2005-07-01
14:56:06.000000000 -0500
@@ -35,7 +35,7 @@ pte_t *lookup_address(unsigned long addr
         return pte_offset_kernel(pmd, address);
 } 
 
-static struct page *split_large_page(unsigned long address, pgprot_t
prot)
+static struct page *split_large_page(unsigned long address, pgprot_t
prot, pgprot_t ref_prot)
 { 
 	int i; 
 	unsigned long addr;
@@ -53,7 +53,7 @@ static struct page *split_large_page(uns
 	pbase = (pte_t *)page_address(base);
 	for (i = 0; i < PTRS_PER_PTE; i++, addr += PAGE_SIZE) {
 		pbase[i] = pfn_pte(addr >> PAGE_SHIFT, 
-				   addr == address ? prot :
PAGE_KERNEL);
+				   addr == address ? prot : ref_prot);
 	}
 	return base;
 } 
@@ -122,7 +122,7 @@ __change_page_attr(struct page *page, pg
 		if ((pte_val(*kpte) & _PAGE_PSE) == 0) { 
 			set_pte_atomic(kpte, mk_pte(page, prot)); 
 		} else {
-			struct page *split = split_large_page(address,
prot); 
+			struct page *split = split_large_page(address,
prot, PAGE_KERNEL); 
 			if (!split)
 				return -ENOMEM;
 			set_pmd_pte(kpte,address,mk_pte(split,
PAGE_KERNEL));
@@ -152,6 +152,38 @@ __change_page_attr(struct page *page, pg
 	return 0;
 } 
 
+static int __change_page_attr_perm (struct page *page, pgprot_t prot)
+{ 
+	pte_t *kpte; 
+	unsigned long address;
+	struct page *kpte_page;
+
+	BUG_ON(PageHighMem(page));
+	address = (unsigned long)page_address(page);
+
+	kpte = lookup_address(address);
+	if (!kpte)
+		return -EINVAL;
+	kpte_page = virt_to_page(kpte);
+
+	if ((pte_val(*kpte) & _PAGE_PSE) == 0) { 
+		set_pte_atomic(kpte, mk_pte(page, prot)); 
+	} else {
+		pgprot_t ref_prot;
+
+		if ((pte_val(*kpte) & _PAGE_NX))
+			ref_prot = PAGE_KERNEL;
+		else
+			ref_prot = PAGE_KERNEL_EXEC;
+		kpte_page = split_large_page(address, prot, ref_prot);
+		if (!kpte_page)
+			return -ENOMEM;
+		set_pmd_pte(kpte,address,mk_pte(kpte_page, ref_prot));
+	}	
+	SetPageReserved(kpte_page);
+	return 0;
+} 
+
 static inline void flush_map(void)
 {
 	on_each_cpu(flush_kernel_map, NULL, 1, 1);
@@ -186,6 +218,22 @@ int change_page_attr(struct page *page, 
 	return err;
 }
 
+int change_page_attr_perm(struct page *page, int numpages, pgprot_t
prot)
+{
+	int err = 0; 
+	int i; 
+	unsigned long flags;
+
+	spin_lock_irqsave(&cpa_lock, flags);
+	for (i = 0; i < numpages; i++, page++) { 
+		err = __change_page_attr_perm(page, prot);
+		if (err) 
+			break; 
+	} 	
+	spin_unlock_irqrestore(&cpa_lock, flags);
+	return err;
+}
+
 void global_flush_tlb(void)
 { 
 	LIST_HEAD(l);

[-- Attachment #2: pass1.patch --]
[-- Type: application/octet-stream, Size: 4194 bytes --]

diff -purN --exclude='*.o' --exclude='*.cmd' linux-2.6.12grep/arch/i386/mm/init.c linux-2.6.12/arch/i386/mm/init.c
--- linux-2.6.12grep/arch/i386/mm/init.c	2005-07-01 15:09:27.000000000 -0500
+++ linux-2.6.12/arch/i386/mm/init.c	2005-07-01 15:13:06.000000000 -0500
@@ -666,6 +666,30 @@ static int noinline do_test_wp_bit(void)
 	return flag;
 }
 
+extern int change_page_attr_perm(struct page *, int, pgprot_t);
+
+/*
+ * In kernel_physical_mapping_init(), any big pages that contained kernel text area were
+ * set up as big executable pages.  This function should be called when the initmem
+ * is freed, to correctly set up the executable & non-executable pages in this area.
+ */
+static void cleanup_nx_in_kerneltext(void)
+{
+	unsigned long from, to;
+
+	if (!nx_enabled) return;
+
+	from = PAGE_OFFSET;
+	to = (unsigned long)_text & PAGE_MASK;
+	for (; from<to; from += PAGE_SIZE)
+		change_page_attr_perm(virt_to_page(from), 1, PAGE_KERNEL); 
+	
+	from = ((unsigned long)_etext + PAGE_SIZE - 1) & PAGE_MASK;
+	to = ((unsigned long)__init_end + LARGE_PAGE_SIZE) & LARGE_PAGE_MASK;
+	for (; from<to; from += PAGE_SIZE)
+		change_page_attr_perm(virt_to_page(from), 1, PAGE_KERNEL); 
+}
+
 void free_initmem(void)
 {
 	unsigned long addr;
@@ -679,6 +703,8 @@ void free_initmem(void)
 		totalram_pages++;
 	}
 	printk (KERN_INFO "Freeing unused kernel memory: %dk freed\n", (__init_end - __init_begin) >> 10);
+
+	cleanup_nx_in_kerneltext();
 }
 
 #ifdef CONFIG_BLK_DEV_INITRD
diff -purN --exclude='*.o' --exclude='*.cmd' linux-2.6.12grep/arch/i386/mm/pageattr.c linux-2.6.12/arch/i386/mm/pageattr.c
--- linux-2.6.12grep/arch/i386/mm/pageattr.c	2005-07-01 15:09:08.000000000 -0500
+++ linux-2.6.12/arch/i386/mm/pageattr.c	2005-07-01 14:56:06.000000000 -0500
@@ -35,7 +35,7 @@ pte_t *lookup_address(unsigned long addr
         return pte_offset_kernel(pmd, address);
 } 
 
-static struct page *split_large_page(unsigned long address, pgprot_t prot)
+static struct page *split_large_page(unsigned long address, pgprot_t prot, pgprot_t ref_prot)
 { 
 	int i; 
 	unsigned long addr;
@@ -53,7 +53,7 @@ static struct page *split_large_page(uns
 	pbase = (pte_t *)page_address(base);
 	for (i = 0; i < PTRS_PER_PTE; i++, addr += PAGE_SIZE) {
 		pbase[i] = pfn_pte(addr >> PAGE_SHIFT, 
-				   addr == address ? prot : PAGE_KERNEL);
+				   addr == address ? prot : ref_prot);
 	}
 	return base;
 } 
@@ -122,7 +122,7 @@ __change_page_attr(struct page *page, pg
 		if ((pte_val(*kpte) & _PAGE_PSE) == 0) { 
 			set_pte_atomic(kpte, mk_pte(page, prot)); 
 		} else {
-			struct page *split = split_large_page(address, prot); 
+			struct page *split = split_large_page(address, prot, PAGE_KERNEL); 
 			if (!split)
 				return -ENOMEM;
 			set_pmd_pte(kpte,address,mk_pte(split, PAGE_KERNEL));
@@ -152,6 +152,38 @@ __change_page_attr(struct page *page, pg
 	return 0;
 } 
 
+static int __change_page_attr_perm (struct page *page, pgprot_t prot)
+{ 
+	pte_t *kpte; 
+	unsigned long address;
+	struct page *kpte_page;
+
+	BUG_ON(PageHighMem(page));
+	address = (unsigned long)page_address(page);
+
+	kpte = lookup_address(address);
+	if (!kpte)
+		return -EINVAL;
+	kpte_page = virt_to_page(kpte);
+
+	if ((pte_val(*kpte) & _PAGE_PSE) == 0) { 
+		set_pte_atomic(kpte, mk_pte(page, prot)); 
+	} else {
+		pgprot_t ref_prot;
+
+		if ((pte_val(*kpte) & _PAGE_NX))
+			ref_prot = PAGE_KERNEL;
+		else
+			ref_prot = PAGE_KERNEL_EXEC;
+		kpte_page = split_large_page(address, prot, ref_prot);
+		if (!kpte_page)
+			return -ENOMEM;
+		set_pmd_pte(kpte,address,mk_pte(kpte_page, ref_prot));
+	}	
+	SetPageReserved(kpte_page);
+	return 0;
+} 
+
 static inline void flush_map(void)
 {
 	on_each_cpu(flush_kernel_map, NULL, 1, 1);
@@ -186,6 +218,22 @@ int change_page_attr(struct page *page, 
 	return err;
 }
 
+int change_page_attr_perm(struct page *page, int numpages, pgprot_t prot)
+{
+	int err = 0; 
+	int i; 
+	unsigned long flags;
+
+	spin_lock_irqsave(&cpa_lock, flags);
+	for (i = 0; i < numpages; i++, page++) { 
+		err = __change_page_attr_perm(page, prot);
+		if (err) 
+			break; 
+	} 	
+	spin_unlock_irqrestore(&cpa_lock, flags);
+	return err;
+}
+
 void global_flush_tlb(void)
 { 
 	LIST_HEAD(l);