[deliverable/linux.git] / arch / x86 / mm / pageattr.c

/*
 * Copyright 2002 Andi Kleen, SuSE Labs.
 * Thanks to Ben LaHaise for precious feedback.
 */
#include <linux/highmem.h>
#include <linux/bootmem.h>
#include <linux/module.h>
#include <linux/sched.h>
#include <linux/slab.h>
#include <linux/mm.h>

#include <asm/e820.h>
#include <asm/processor.h>
#include <asm/tlbflush.h>
#include <asm/sections.h>
#include <asm/uaccess.h>
#include <asm/pgalloc.h>

static inline int
within(unsigned long addr, unsigned long start, unsigned long end)
{
	return addr >= start && addr < end;
}

/*
 * Flushing functions
 */

/**
 * clflush_cache_range - flush a cache range with clflush
 * @addr:	virtual start address
 * @size:	number of bytes to flush
 *
 * clflush is an unordered instruction which needs fencing with mfence
 * to avoid ordering issues.
 */
void clflush_cache_range(void *vaddr, unsigned int size)
{
	void *vend = vaddr + size - 1;

	mb();

	for (; vaddr < vend; vaddr += boot_cpu_data.x86_clflush_size)
		clflush(vaddr);
	/*
	 * Flush any possible final partial cacheline:
	 */
	clflush(vend);

	mb();
}

static void __cpa_flush_all(void *arg)
{
	/*
	 * Flush all to work around Errata in early athlons regarding
	 * large page flushing.
	 */
	__flush_tlb_all();

	if (boot_cpu_data.x86_model >= 4)
		wbinvd();
}

static void cpa_flush_all(void)
{
	BUG_ON(irqs_disabled());

	on_each_cpu(__cpa_flush_all, NULL, 1, 1);
}

static void __cpa_flush_range(void *arg)
{
	/*
	 * We could optimize that further and do individual per page
	 * tlb invalidates for a low number of pages. Caveat: we must
	 * flush the high aliases on 64bit as well.
	 */
	__flush_tlb_all();
}

static void cpa_flush_range(unsigned long start, int numpages)
{
	unsigned int i, level;
	unsigned long addr;

	BUG_ON(irqs_disabled());
	WARN_ON(PAGE_ALIGN(start) != start);

	on_each_cpu(__cpa_flush_range, NULL, 1, 1);

	/*
	 * We only need to flush on one CPU,
	 * clflush is a MESI-coherent instruction that
	 * will cause all other CPUs to flush the same
	 * cachelines:
	 */
	for (i = 0, addr = start; i < numpages; i++, addr += PAGE_SIZE) {
		pte_t *pte = lookup_address(addr, &level);

		/*
		 * Only flush present addresses:
		 */
		if (pte && pte_present(*pte))
			clflush_cache_range((void *) addr, PAGE_SIZE);
	}
}

/*
 * Certain areas of memory on x86 require very specific protection flags,
 * for example the BIOS area or kernel text. Callers don't always get this
 * right (again, ioremap() on BIOS memory is not uncommon) so this function
 * checks and fixes these known static required protection bits.
 */
static inline pgprot_t static_protections(pgprot_t prot, unsigned long address)
{
	pgprot_t forbidden = __pgprot(0);

	/*
	 * The BIOS area between 640k and 1Mb needs to be executable for
	 * PCI BIOS based config access (CONFIG_PCI_GOBIOS) support.
	 */
	if (within(__pa(address), BIOS_BEGIN, BIOS_END))
		pgprot_val(forbidden) |= _PAGE_NX;

	/*
	 * The kernel text needs to be executable for obvious reasons
	 * Does not cover __inittext since that is gone later on
	 */
	if (within(address, (unsigned long)_text, (unsigned long)_etext))
		pgprot_val(forbidden) |= _PAGE_NX;

#ifdef CONFIG_DEBUG_RODATA
	/* The .rodata section needs to be read-only */
	if (within(address, (unsigned long)__start_rodata,
				(unsigned long)__end_rodata))
		pgprot_val(forbidden) |= _PAGE_RW;
#endif

	prot = __pgprot(pgprot_val(prot) & ~pgprot_val(forbidden));

	return prot;
}

pte_t *lookup_address(unsigned long address, int *level)
{
	pgd_t *pgd = pgd_offset_k(address);
	pud_t *pud;
	pmd_t *pmd;

	*level = PG_LEVEL_NONE;

	if (pgd_none(*pgd))
		return NULL;
	pud = pud_offset(pgd, address);
	if (pud_none(*pud))
		return NULL;
	pmd = pmd_offset(pud, address);
	if (pmd_none(*pmd))
		return NULL;

	*level = PG_LEVEL_2M;
	if (pmd_large(*pmd))
		return (pte_t *)pmd;

	*level = PG_LEVEL_4K;
	return pte_offset_kernel(pmd, address);
}

static void __set_pmd_pte(pte_t *kpte, unsigned long address, pte_t pte)
{
	/* change init_mm */
	set_pte_atomic(kpte, pte);
#ifdef CONFIG_X86_32
	if (!SHARED_KERNEL_PMD) {
		struct page *page;

		for (page = pgd_list; page; page = (struct page *)page->index) {
			pgd_t *pgd;
			pud_t *pud;
			pmd_t *pmd;

			pgd = (pgd_t *)page_address(page) + pgd_index(address);
			pud = pud_offset(pgd, address);
			pmd = pmd_offset(pud, address);
			set_pte_atomic((pte_t *)pmd, pte);
		}
	}
#endif
}

static int split_large_page(pte_t *kpte, unsigned long address)
{
	pgprot_t ref_prot = pte_pgprot(pte_clrhuge(*kpte));
	gfp_t gfp_flags = GFP_KERNEL;
	unsigned long flags;
	unsigned long addr;
	pte_t *pbase, *tmp;
	struct page *base;
	unsigned int i, level;

#ifdef CONFIG_DEBUG_PAGEALLOC
	gfp_flags = __GFP_HIGH | __GFP_NOFAIL | __GFP_NOWARN;
	gfp_flags = GFP_ATOMIC | __GFP_NOWARN;
#endif
	base = alloc_pages(gfp_flags, 0);
	if (!base)
		return -ENOMEM;

	spin_lock_irqsave(&pgd_lock, flags);
	/*
	 * Check for races, another CPU might have split this page
	 * up for us already:
	 */
	tmp = lookup_address(address, &level);
	if (tmp != kpte) {
		WARN_ON_ONCE(1);
		goto out_unlock;
	}

	address = __pa(address);
	addr = address & LARGE_PAGE_MASK;
	pbase = (pte_t *)page_address(base);
#ifdef CONFIG_X86_32
	paravirt_alloc_pt(&init_mm, page_to_pfn(base));
#endif

	pgprot_val(ref_prot) &= ~_PAGE_NX;
	for (i = 0; i < PTRS_PER_PTE; i++, addr += PAGE_SIZE)
		set_pte(&pbase[i], pfn_pte(addr >> PAGE_SHIFT, ref_prot));

	/*
	 * Install the new, split up pagetable. Important detail here:
	 *
	 * On Intel the NX bit of all levels must be cleared to make a
	 * page executable. See section 4.13.2 of Intel 64 and IA-32
	 * Architectures Software Developer's Manual).
	 */
	ref_prot = pte_pgprot(pte_mkexec(pte_clrhuge(*kpte)));
	__set_pmd_pte(kpte, address, mk_pte(base, ref_prot));
	base = NULL;

out_unlock:
	spin_unlock_irqrestore(&pgd_lock, flags);

	if (base)
		__free_pages(base, 0);

	return 0;
}

static int
__change_page_attr(unsigned long address, unsigned long pfn,
		   pgprot_t mask_set, pgprot_t mask_clr)
{
	struct page *kpte_page;
	int level, err = 0;
	pte_t *kpte;

#ifdef CONFIG_X86_32
	BUG_ON(pfn > max_low_pfn);
#endif

repeat:
	kpte = lookup_address(address, &level);
	if (!kpte)
		return -EINVAL;

	kpte_page = virt_to_page(kpte);
	BUG_ON(PageLRU(kpte_page));
	BUG_ON(PageCompound(kpte_page));

	if (level == PG_LEVEL_4K) {
		pgprot_t new_prot = pte_pgprot(*kpte);
		pte_t new_pte, old_pte = *kpte;

		pgprot_val(new_prot) &= ~pgprot_val(mask_clr);
		pgprot_val(new_prot) |= pgprot_val(mask_set);

		new_prot = static_protections(new_prot, address);

		new_pte = pfn_pte(pfn, canon_pgprot(new_prot));
		BUG_ON(pte_pfn(new_pte) != pte_pfn(old_pte));

		set_pte_atomic(kpte, new_pte);
	} else {
		err = split_large_page(kpte, address);
		if (!err)
			goto repeat;
	}
	return err;
}

/**
 * change_page_attr_addr - Change page table attributes in linear mapping
 * @address: Virtual address in linear mapping.
 * @prot:    New page table attribute (PAGE_*)
 *
 * Change page attributes of a page in the direct mapping. This is a variant
 * of change_page_attr() that also works on memory holes that do not have
 * mem_map entry (pfn_valid() is false).
 *
 * See change_page_attr() documentation for more details.
 *
 * Modules and drivers should use the set_memory_* APIs instead.
 */

#define HIGH_MAP_START	__START_KERNEL_map
#define HIGH_MAP_END	(__START_KERNEL_map + KERNEL_TEXT_SIZE)

static int
change_page_attr_addr(unsigned long address, pgprot_t mask_set,
		      pgprot_t mask_clr)
{
	unsigned long phys_addr = __pa(address);
	unsigned long pfn = phys_addr >> PAGE_SHIFT;
	int err;

#ifdef CONFIG_X86_64
	/*
	 * If we are inside the high mapped kernel range, then we
	 * fixup the low mapping first. __va() returns the virtual
	 * address in the linear mapping:
	 */
	if (within(address, HIGH_MAP_START, HIGH_MAP_END))
		address = (unsigned long) __va(phys_addr);
#endif

	err = __change_page_attr(address, pfn, mask_set, mask_clr);
	if (err)
		return err;

#ifdef CONFIG_X86_64
	/*
	 * If the physical address is inside the kernel map, we need
	 * to touch the high mapped kernel as well:
	 */
	if (within(phys_addr, 0, KERNEL_TEXT_SIZE)) {
		/*
		 * Calc the high mapping address. See __phys_addr()
		 * for the non obvious details.
		 */
		address = phys_addr + HIGH_MAP_START - phys_base;
		/* Make sure the kernel mappings stay executable */
		pgprot_val(mask_clr) |= _PAGE_NX;

		/*
		 * Our high aliases are imprecise, because we check
		 * everything between 0 and KERNEL_TEXT_SIZE, so do
		 * not propagate lookup failures back to users:
		 */
		__change_page_attr(address, pfn, mask_set, mask_clr);
	}
#endif
	return err;
}

static int __change_page_attr_set_clr(unsigned long addr, int numpages,
				      pgprot_t mask_set, pgprot_t mask_clr)
{
	unsigned int i;
	int ret;

	for (i = 0; i < numpages ; i++, addr += PAGE_SIZE) {
		ret = change_page_attr_addr(addr, mask_set, mask_clr);
		if (ret)
			return ret;
	}

	return 0;
}

static int change_page_attr_set_clr(unsigned long addr, int numpages,
				    pgprot_t mask_set, pgprot_t mask_clr)
{
	int ret = __change_page_attr_set_clr(addr, numpages, mask_set,
					     mask_clr);

	/*
	 * On success we use clflush, when the CPU supports it to
	 * avoid the wbindv. If the CPU does not support it and in the
	 * error case we fall back to cpa_flush_all (which uses
	 * wbindv):
	 */
	if (!ret && cpu_has_clflush)
		cpa_flush_range(addr, numpages);
	else
		cpa_flush_all();

	return ret;
}

static inline int change_page_attr_set(unsigned long addr, int numpages,
				       pgprot_t mask)
{
	return change_page_attr_set_clr(addr, numpages, mask, __pgprot(0));
}

static inline int change_page_attr_clear(unsigned long addr, int numpages,
					 pgprot_t mask)
{
	return __change_page_attr_set_clr(addr, numpages, __pgprot(0), mask);

}

int set_memory_uc(unsigned long addr, int numpages)
{
	return change_page_attr_set(addr, numpages,
				    __pgprot(_PAGE_PCD | _PAGE_PWT));
}
EXPORT_SYMBOL(set_memory_uc);

int set_memory_wb(unsigned long addr, int numpages)
{
	return change_page_attr_clear(addr, numpages,
				      __pgprot(_PAGE_PCD | _PAGE_PWT));
}
EXPORT_SYMBOL(set_memory_wb);

int set_memory_x(unsigned long addr, int numpages)
{
	return change_page_attr_clear(addr, numpages, __pgprot(_PAGE_NX));
}
EXPORT_SYMBOL(set_memory_x);

int set_memory_nx(unsigned long addr, int numpages)
{
	return change_page_attr_set(addr, numpages, __pgprot(_PAGE_NX));
}
EXPORT_SYMBOL(set_memory_nx);

int set_memory_ro(unsigned long addr, int numpages)
{
	return change_page_attr_clear(addr, numpages, __pgprot(_PAGE_RW));
}

int set_memory_rw(unsigned long addr, int numpages)
{
	return change_page_attr_set(addr, numpages, __pgprot(_PAGE_RW));
}

int set_memory_np(unsigned long addr, int numpages)
{
	return change_page_attr_clear(addr, numpages, __pgprot(_PAGE_PRESENT));
}

int set_pages_uc(struct page *page, int numpages)
{
	unsigned long addr = (unsigned long)page_address(page);

	return set_memory_uc(addr, numpages);
}
EXPORT_SYMBOL(set_pages_uc);

int set_pages_wb(struct page *page, int numpages)
{
	unsigned long addr = (unsigned long)page_address(page);

	return set_memory_wb(addr, numpages);
}
EXPORT_SYMBOL(set_pages_wb);

int set_pages_x(struct page *page, int numpages)
{
	unsigned long addr = (unsigned long)page_address(page);

	return set_memory_x(addr, numpages);
}
EXPORT_SYMBOL(set_pages_x);

int set_pages_nx(struct page *page, int numpages)
{
	unsigned long addr = (unsigned long)page_address(page);

	return set_memory_nx(addr, numpages);
}
EXPORT_SYMBOL(set_pages_nx);

int set_pages_ro(struct page *page, int numpages)
{
	unsigned long addr = (unsigned long)page_address(page);

	return set_memory_ro(addr, numpages);
}

int set_pages_rw(struct page *page, int numpages)
{
	unsigned long addr = (unsigned long)page_address(page);

	return set_memory_rw(addr, numpages);
}


#if defined(CONFIG_DEBUG_PAGEALLOC) || defined(CONFIG_CPA_DEBUG)
static inline int __change_page_attr_set(unsigned long addr, int numpages,
					 pgprot_t mask)
{
	return __change_page_attr_set_clr(addr, numpages, mask, __pgprot(0));
}

static inline int __change_page_attr_clear(unsigned long addr, int numpages,
					   pgprot_t mask)
{
	return __change_page_attr_set_clr(addr, numpages, __pgprot(0), mask);
}
#endif

#ifdef CONFIG_DEBUG_PAGEALLOC

static int __set_pages_p(struct page *page, int numpages)
{
	unsigned long addr = (unsigned long)page_address(page);

	return __change_page_attr_set(addr, numpages,
				      __pgprot(_PAGE_PRESENT | _PAGE_RW));
}

static int __set_pages_np(struct page *page, int numpages)
{
	unsigned long addr = (unsigned long)page_address(page);

	return __change_page_attr_clear(addr, numpages,
					__pgprot(_PAGE_PRESENT));
}

void kernel_map_pages(struct page *page, int numpages, int enable)
{
	if (PageHighMem(page))
		return;
	if (!enable) {
		debug_check_no_locks_freed(page_address(page),
					   numpages * PAGE_SIZE);
	}

	/*
	 * If page allocator is not up yet then do not call c_p_a():
	 */
	if (!debug_pagealloc_enabled)
		return;

	/*
	 * The return value is ignored - the calls cannot fail,
	 * large pages are disabled at boot time:
	 */
	if (enable)
		__set_pages_p(page, numpages);
	else
		__set_pages_np(page, numpages);

	/*
	 * We should perform an IPI and flush all tlbs,
	 * but that can deadlock->flush only current cpu:
	 */
	__flush_tlb_all();
}
#endif

/*
 * The testcases use internal knowledge of the implementation that shouldn't
 * be exposed to the rest of the kernel. Include these directly here.
 */
#ifdef CONFIG_CPA_DEBUG
#include "pageattr-test.c"
#endif
Commit	Line	Data
9f4c815c IM	1	/*
9f4c815c IM	2	* Copyright 2002 Andi Kleen, SuSE Labs.
1da177e4	3	* Thanks to Ben LaHaise for precious feedback.
9f4c815c	4	*/
1da177e4	5	#include <linux/highmem.h>
8192206d	6	#include <linux/bootmem.h>
1da177e4	7	#include <linux/module.h>
9f4c815c	8	#include <linux/sched.h>
1da177e4	9	#include <linux/slab.h>
9f4c815c IM	10	#include <linux/mm.h>
9f4c815c IM	11
950f9d95	12	#include <asm/e820.h>
1da177e4 LT	13	#include <asm/processor.h>
1da177e4 LT	14	#include <asm/tlbflush.h>
f8af095d	15	#include <asm/sections.h>
9f4c815c IM	16	#include <asm/uaccess.h>
9f4c815c IM	17	#include <asm/pgalloc.h>
1da177e4	18
ed724be6 AV	19	static inline int
ed724be6 AV	20	within(unsigned long addr, unsigned long start, unsigned long end)
687c4825	21	{
ed724be6 AV	22	return addr >= start && addr < end;
	23	}
	24
d7c8f21a TG	25	/*
	26	* Flushing functions
	27	*/
cd8ddf1a	28
cd8ddf1a TG	29	/**
	30	* clflush_cache_range - flush a cache range with clflush
	31	* @addr: virtual start address
	32	* @size: number of bytes to flush
	33	*
	34	* clflush is an unordered instruction which needs fencing with mfence
	35	* to avoid ordering issues.
	36	*/
4c61afcd	37	void clflush_cache_range(void *vaddr, unsigned int size)
d7c8f21a	38	{
4c61afcd	39	void *vend = vaddr + size - 1;
d7c8f21a	40
cd8ddf1a	41	mb();
4c61afcd IM	42
	43	for (; vaddr < vend; vaddr += boot_cpu_data.x86_clflush_size)
	44	clflush(vaddr);
	45	/*
	46	* Flush any possible final partial cacheline:
	47	*/
	48	clflush(vend);
	49
cd8ddf1a	50	mb();
d7c8f21a TG	51	}
d7c8f21a TG	52
af1e6844	53	static void __cpa_flush_all(void *arg)
d7c8f21a TG	54	{
	55	/*
	56	* Flush all to work around Errata in early athlons regarding
	57	* large page flushing.
	58	*/
	59	__flush_tlb_all();
	60
	61	if (boot_cpu_data.x86_model >= 4)
	62	wbinvd();
	63	}
	64
af1e6844	65	static void cpa_flush_all(void)
d7c8f21a TG	66	{
	67	BUG_ON(irqs_disabled());
	68
af1e6844	69	on_each_cpu(__cpa_flush_all, NULL, 1, 1);
d7c8f21a TG	70	}
d7c8f21a TG	71
57a6a46a TG	72	static void __cpa_flush_range(void *arg)
57a6a46a TG	73	{
57a6a46a TG	74	/*
	75	* We could optimize that further and do individual per page
	76	* tlb invalidates for a low number of pages. Caveat: we must
	77	* flush the high aliases on 64bit as well.
	78	*/
	79	__flush_tlb_all();
57a6a46a TG	80	}
57a6a46a TG	81
4c61afcd	82	static void cpa_flush_range(unsigned long start, int numpages)
57a6a46a	83	{
4c61afcd IM	84	unsigned int i, level;
	85	unsigned long addr;
	86
57a6a46a	87	BUG_ON(irqs_disabled());
4c61afcd	88	WARN_ON(PAGE_ALIGN(start) != start);
57a6a46a	89
3b233e52	90	on_each_cpu(__cpa_flush_range, NULL, 1, 1);
57a6a46a	91
3b233e52 TG	92	/*
	93	* We only need to flush on one CPU,
	94	* clflush is a MESI-coherent instruction that
	95	* will cause all other CPUs to flush the same
	96	* cachelines:
	97	*/
4c61afcd IM	98	for (i = 0, addr = start; i < numpages; i++, addr += PAGE_SIZE) {
	99	pte_t *pte = lookup_address(addr, &level);
	100
	101	/*
	102	* Only flush present addresses:
	103	*/
	104	if (pte && pte_present(*pte))
	105	clflush_cache_range((void *) addr, PAGE_SIZE);
	106	}
57a6a46a TG	107	}
57a6a46a TG	108
ed724be6 AV	109	/*
	110	* Certain areas of memory on x86 require very specific protection flags,
	111	* for example the BIOS area or kernel text. Callers don't always get this
	112	* right (again, ioremap() on BIOS memory is not uncommon) so this function
	113	* checks and fixes these known static required protection bits.
	114	*/
	115	static inline pgprot_t static_protections(pgprot_t prot, unsigned long address)
	116	{
	117	pgprot_t forbidden = __pgprot(0);
	118
687c4825	119	/*
ed724be6 AV	120	* The BIOS area between 640k and 1Mb needs to be executable for
ed724be6 AV	121	* PCI BIOS based config access (CONFIG_PCI_GOBIOS) support.
687c4825	122	*/
ed724be6 AV	123	if (within(__pa(address), BIOS_BEGIN, BIOS_END))
	124	pgprot_val(forbidden) \|= _PAGE_NX;
	125
	126	/*
	127	* The kernel text needs to be executable for obvious reasons
	128	* Does not cover __inittext since that is gone later on
	129	*/
	130	if (within(address, (unsigned long)_text, (unsigned long)_etext))
	131	pgprot_val(forbidden) \|= _PAGE_NX;
	132
	133	#ifdef CONFIG_DEBUG_RODATA
	134	/* The .rodata section needs to be read-only */
	135	if (within(address, (unsigned long)__start_rodata,
	136	(unsigned long)__end_rodata))
	137	pgprot_val(forbidden) \|= _PAGE_RW;
	138	#endif
	139
	140	prot = __pgprot(pgprot_val(prot) & ~pgprot_val(forbidden));
687c4825 IM	141
	142	return prot;
	143	}
	144
f0646e43	145	pte_t lookup_address(unsigned long address, int level)
9f4c815c	146	{
1da177e4 LT	147	pgd_t *pgd = pgd_offset_k(address);
	148	pud_t *pud;
	149	pmd_t *pmd;
9f4c815c	150
30551bb3 TG	151	*level = PG_LEVEL_NONE;
30551bb3 TG	152
1da177e4 LT	153	if (pgd_none(*pgd))
	154	return NULL;
	155	pud = pud_offset(pgd, address);
	156	if (pud_none(*pud))
	157	return NULL;
	158	pmd = pmd_offset(pud, address);
	159	if (pmd_none(*pmd))
	160	return NULL;
30551bb3 TG	161
30551bb3 TG	162	*level = PG_LEVEL_2M;
1da177e4 LT	163	if (pmd_large(*pmd))
1da177e4 LT	164	return (pte_t *)pmd;
1da177e4	165
30551bb3	166	*level = PG_LEVEL_4K;
9f4c815c IM	167	return pte_offset_kernel(pmd, address);
	168	}
	169
9a3dc780	170	static void __set_pmd_pte(pte_t *kpte, unsigned long address, pte_t pte)
9f4c815c	171	{
9f4c815c IM	172	/* change init_mm */
9f4c815c IM	173	set_pte_atomic(kpte, pte);
44af6c41	174	#ifdef CONFIG_X86_32
e4b71dcf	175	if (!SHARED_KERNEL_PMD) {
44af6c41 IM	176	struct page *page;
	177
	178	for (page = pgd_list; page; page = (struct page *)page->index) {
	179	pgd_t *pgd;
	180	pud_t *pud;
	181	pmd_t *pmd;
	182
	183	pgd = (pgd_t *)page_address(page) + pgd_index(address);
	184	pud = pud_offset(pgd, address);
	185	pmd = pmd_offset(pud, address);
	186	set_pte_atomic((pte_t *)pmd, pte);
	187	}
1da177e4	188	}
44af6c41	189	#endif
1da177e4 LT	190	}
1da177e4 LT	191
7afe15b9	192	static int split_large_page(pte_t *kpte, unsigned long address)
bb5c2dbd	193	{
7afe15b9	194	pgprot_t ref_prot = pte_pgprot(pte_clrhuge(*kpte));
12d6f21e	195	gfp_t gfp_flags = GFP_KERNEL;
9a3dc780	196	unsigned long flags;
bb5c2dbd IM	197	unsigned long addr;
	198	pte_t pbase, tmp;
	199	struct page *base;
86f03989	200	unsigned int i, level;
bb5c2dbd	201
12d6f21e	202	#ifdef CONFIG_DEBUG_PAGEALLOC
86f03989 IM	203	gfp_flags = __GFP_HIGH \| __GFP_NOFAIL \| __GFP_NOWARN;
86f03989 IM	204	gfp_flags = GFP_ATOMIC \| __GFP_NOWARN;
12d6f21e IM	205	#endif
12d6f21e IM	206	base = alloc_pages(gfp_flags, 0);
bb5c2dbd IM	207	if (!base)
	208	return -ENOMEM;
	209
9a3dc780	210	spin_lock_irqsave(&pgd_lock, flags);
bb5c2dbd IM	211	/*
	212	* Check for races, another CPU might have split this page
	213	* up for us already:
	214	*/
	215	tmp = lookup_address(address, &level);
5508a748 IM	216	if (tmp != kpte) {
5508a748 IM	217	WARN_ON_ONCE(1);
bb5c2dbd	218	goto out_unlock;
5508a748	219	}
bb5c2dbd IM	220
	221	address = __pa(address);
	222	addr = address & LARGE_PAGE_MASK;
	223	pbase = (pte_t *)page_address(base);
44af6c41	224	#ifdef CONFIG_X86_32
bb5c2dbd	225	paravirt_alloc_pt(&init_mm, page_to_pfn(base));
44af6c41	226	#endif
bb5c2dbd	227
86f03989	228	pgprot_val(ref_prot) &= ~_PAGE_NX;
bb5c2dbd IM	229	for (i = 0; i < PTRS_PER_PTE; i++, addr += PAGE_SIZE)
	230	set_pte(&pbase[i], pfn_pte(addr >> PAGE_SHIFT, ref_prot));
	231
	232	/*
4c881ca1 HY	233	* Install the new, split up pagetable. Important detail here:
	234	*
	235	* On Intel the NX bit of all levels must be cleared to make a
	236	* page executable. See section 4.13.2 of Intel 64 and IA-32
	237	* Architectures Software Developer's Manual).
bb5c2dbd	238	*/
4c881ca1	239	ref_prot = pte_pgprot(pte_mkexec(pte_clrhuge(*kpte)));
9a3dc780	240	__set_pmd_pte(kpte, address, mk_pte(base, ref_prot));
bb5c2dbd IM	241	base = NULL;
	242
	243	out_unlock:
9a3dc780	244	spin_unlock_irqrestore(&pgd_lock, flags);
bb5c2dbd IM	245
	246	if (base)
	247	__free_pages(base, 0);
	248
	249	return 0;
	250	}
	251
44af6c41	252	static int
86f03989 IM	253	__change_page_attr(unsigned long address, unsigned long pfn,
86f03989 IM	254	pgprot_t mask_set, pgprot_t mask_clr)
9f4c815c	255	{
1da177e4	256	struct page *kpte_page;
bb5c2dbd	257	int level, err = 0;
9f4c815c	258	pte_t *kpte;
1da177e4	259
8192206d IM	260	#ifdef CONFIG_X86_32
	261	BUG_ON(pfn > max_low_pfn);
	262	#endif
1da177e4	263
97f99fed	264	repeat:
f0646e43	265	kpte = lookup_address(address, &level);
1da177e4 LT	266	if (!kpte)
1da177e4 LT	267	return -EINVAL;
9f4c815c	268
1da177e4	269	kpte_page = virt_to_page(kpte);
65d2f0bc AK	270	BUG_ON(PageLRU(kpte_page));
	271	BUG_ON(PageCompound(kpte_page));
	272
30551bb3	273	if (level == PG_LEVEL_4K) {
86f03989 IM	274	pgprot_t new_prot = pte_pgprot(*kpte);
	275	pte_t new_pte, old_pte = *kpte;
	276
	277	pgprot_val(new_prot) &= ~pgprot_val(mask_clr);
	278	pgprot_val(new_prot) \|= pgprot_val(mask_set);
	279
	280	new_prot = static_protections(new_prot, address);
	281
	282	new_pte = pfn_pte(pfn, canon_pgprot(new_prot));
	283	BUG_ON(pte_pfn(new_pte) != pte_pfn(old_pte));
a72a08a4	284
86f03989 IM	285	set_pte_atomic(kpte, new_pte);
86f03989 IM	286	} else {
7afe15b9	287	err = split_large_page(kpte, address);
bb5c2dbd IM	288	if (!err)
bb5c2dbd IM	289	goto repeat;
1da177e4	290	}
bb5c2dbd	291	return err;
9f4c815c	292	}
1da177e4	293
44af6c41 IM	294	/**
	295	* change_page_attr_addr - Change page table attributes in linear mapping
	296	* @address: Virtual address in linear mapping.
44af6c41	297	* @prot: New page table attribute (PAGE_*)
1da177e4	298	*
44af6c41 IM	299	* Change page attributes of a page in the direct mapping. This is a variant
	300	* of change_page_attr() that also works on memory holes that do not have
	301	* mem_map entry (pfn_valid() is false).
9f4c815c	302	*
44af6c41	303	* See change_page_attr() documentation for more details.
75cbade8 AV	304	*
75cbade8 AV	305	* Modules and drivers should use the set_memory_* APIs instead.
1da177e4	306	*/
44af6c41	307
0879750f TG	308	#define HIGH_MAP_START __START_KERNEL_map
	309	#define HIGH_MAP_END (__START_KERNEL_map + KERNEL_TEXT_SIZE)
	310
86f03989 IM	311	static int
86f03989 IM	312	change_page_attr_addr(unsigned long address, pgprot_t mask_set,
0879750f	313	pgprot_t mask_clr)
1da177e4	314	{
0879750f TG	315	unsigned long phys_addr = __pa(address);
	316	unsigned long pfn = phys_addr >> PAGE_SHIFT;
	317	int err;
44af6c41 IM	318
44af6c41 IM	319	#ifdef CONFIG_X86_64
0879750f TG	320	/*
	321	* If we are inside the high mapped kernel range, then we
	322	* fixup the low mapping first. __va() returns the virtual
	323	* address in the linear mapping:
	324	*/
	325	if (within(address, HIGH_MAP_START, HIGH_MAP_END))
	326	address = (unsigned long) __va(phys_addr);
44af6c41 IM	327	#endif
44af6c41 IM	328
0879750f TG	329	err = __change_page_attr(address, pfn, mask_set, mask_clr);
	330	if (err)
	331	return err;
44af6c41	332
44af6c41	333	#ifdef CONFIG_X86_64
488fd995	334	/*
0879750f TG	335	* If the physical address is inside the kernel map, we need
0879750f TG	336	* to touch the high mapped kernel as well:
488fd995	337	*/
0879750f TG	338	if (within(phys_addr, 0, KERNEL_TEXT_SIZE)) {
	339	/*
	340	* Calc the high mapping address. See __phys_addr()
	341	* for the non obvious details.
	342	*/
	343	address = phys_addr + HIGH_MAP_START - phys_base;
488fd995	344	/* Make sure the kernel mappings stay executable */
86f03989	345	pgprot_val(mask_clr) \|= _PAGE_NX;
0879750f	346
86f03989	347	/*
0879750f TG	348	* Our high aliases are imprecise, because we check
	349	* everything between 0 and KERNEL_TEXT_SIZE, so do
	350	* not propagate lookup failures back to users:
86f03989	351	*/
0879750f	352	__change_page_attr(address, pfn, mask_set, mask_clr);
9f4c815c	353	}
488fd995	354	#endif
1da177e4 LT	355	return err;
	356	}
	357
ff31452b TG	358	static int __change_page_attr_set_clr(unsigned long addr, int numpages,
	359	pgprot_t mask_set, pgprot_t mask_clr)
	360	{
86f03989 IM	361	unsigned int i;
86f03989 IM	362	int ret;
ff31452b	363
86f03989 IM	364	for (i = 0; i < numpages ; i++, addr += PAGE_SIZE) {
86f03989 IM	365	ret = change_page_attr_addr(addr, mask_set, mask_clr);
ff31452b TG	366	if (ret)
ff31452b TG	367	return ret;
ff31452b TG	368	}
	369
	370	return 0;
	371	}
	372
	373	static int change_page_attr_set_clr(unsigned long addr, int numpages,
	374	pgprot_t mask_set, pgprot_t mask_clr)
	375	{
	376	int ret = __change_page_attr_set_clr(addr, numpages, mask_set,
	377	mask_clr);
	378
57a6a46a TG	379	/*
	380	* On success we use clflush, when the CPU supports it to
	381	* avoid the wbindv. If the CPU does not support it and in the
af1e6844	382	* error case we fall back to cpa_flush_all (which uses
57a6a46a TG	383	* wbindv):
	384	*/
	385	if (!ret && cpu_has_clflush)
	386	cpa_flush_range(addr, numpages);
	387	else
af1e6844	388	cpa_flush_all();
ff31452b TG	389
	390	return ret;
	391	}
	392
56744546 TG	393	static inline int change_page_attr_set(unsigned long addr, int numpages,
56744546 TG	394	pgprot_t mask)
75cbade8	395	{
56744546	396	return change_page_attr_set_clr(addr, numpages, mask, __pgprot(0));
75cbade8 AV	397	}
75cbade8 AV	398
56744546 TG	399	static inline int change_page_attr_clear(unsigned long addr, int numpages,
56744546 TG	400	pgprot_t mask)
72932c7a	401	{
56744546	402	return __change_page_attr_set_clr(addr, numpages, __pgprot(0), mask);
72932c7a TG	403
	404	}
	405
	406	int set_memory_uc(unsigned long addr, int numpages)
	407	{
	408	return change_page_attr_set(addr, numpages,
	409	__pgprot(_PAGE_PCD \| _PAGE_PWT));
75cbade8 AV	410	}
	411	EXPORT_SYMBOL(set_memory_uc);
	412
	413	int set_memory_wb(unsigned long addr, int numpages)
	414	{
72932c7a TG	415	return change_page_attr_clear(addr, numpages,
72932c7a TG	416	__pgprot(_PAGE_PCD \| _PAGE_PWT));
75cbade8 AV	417	}
	418	EXPORT_SYMBOL(set_memory_wb);
	419
	420	int set_memory_x(unsigned long addr, int numpages)
	421	{
72932c7a	422	return change_page_attr_clear(addr, numpages, __pgprot(_PAGE_NX));
75cbade8 AV	423	}
	424	EXPORT_SYMBOL(set_memory_x);
	425
	426	int set_memory_nx(unsigned long addr, int numpages)
	427	{
72932c7a	428	return change_page_attr_set(addr, numpages, __pgprot(_PAGE_NX));
75cbade8 AV	429	}
	430	EXPORT_SYMBOL(set_memory_nx);
	431
	432	int set_memory_ro(unsigned long addr, int numpages)
	433	{
72932c7a	434	return change_page_attr_clear(addr, numpages, __pgprot(_PAGE_RW));
75cbade8	435	}
75cbade8 AV	436
	437	int set_memory_rw(unsigned long addr, int numpages)
	438	{
72932c7a	439	return change_page_attr_set(addr, numpages, __pgprot(_PAGE_RW));
75cbade8	440	}
f62d0f00 IM	441
	442	int set_memory_np(unsigned long addr, int numpages)
	443	{
72932c7a	444	return change_page_attr_clear(addr, numpages, __pgprot(_PAGE_PRESENT));
f62d0f00	445	}
75cbade8 AV	446
	447	int set_pages_uc(struct page *page, int numpages)
	448	{
	449	unsigned long addr = (unsigned long)page_address(page);
75cbade8	450
d7c8f21a	451	return set_memory_uc(addr, numpages);
75cbade8 AV	452	}
	453	EXPORT_SYMBOL(set_pages_uc);
	454
	455	int set_pages_wb(struct page *page, int numpages)
	456	{
	457	unsigned long addr = (unsigned long)page_address(page);
75cbade8	458
d7c8f21a	459	return set_memory_wb(addr, numpages);
75cbade8 AV	460	}
	461	EXPORT_SYMBOL(set_pages_wb);
	462
	463	int set_pages_x(struct page *page, int numpages)
	464	{
	465	unsigned long addr = (unsigned long)page_address(page);
75cbade8	466
d7c8f21a	467	return set_memory_x(addr, numpages);
75cbade8 AV	468	}
	469	EXPORT_SYMBOL(set_pages_x);
	470
	471	int set_pages_nx(struct page *page, int numpages)
	472	{
	473	unsigned long addr = (unsigned long)page_address(page);
75cbade8	474
d7c8f21a	475	return set_memory_nx(addr, numpages);
75cbade8 AV	476	}
	477	EXPORT_SYMBOL(set_pages_nx);
	478
	479	int set_pages_ro(struct page *page, int numpages)
	480	{
	481	unsigned long addr = (unsigned long)page_address(page);
75cbade8	482
d7c8f21a	483	return set_memory_ro(addr, numpages);
75cbade8	484	}
75cbade8 AV	485
	486	int set_pages_rw(struct page *page, int numpages)
	487	{
	488	unsigned long addr = (unsigned long)page_address(page);
e81d5dc4	489
d7c8f21a	490	return set_memory_rw(addr, numpages);
78c94aba IM	491	}
78c94aba IM	492
1da177e4	493
56744546 TG	494	#if defined(CONFIG_DEBUG_PAGEALLOC) \|\| defined(CONFIG_CPA_DEBUG)
	495	static inline int __change_page_attr_set(unsigned long addr, int numpages,
	496	pgprot_t mask)
	497	{
	498	return __change_page_attr_set_clr(addr, numpages, mask, __pgprot(0));
	499	}
	500
	501	static inline int __change_page_attr_clear(unsigned long addr, int numpages,
	502	pgprot_t mask)
	503	{
	504	return __change_page_attr_set_clr(addr, numpages, __pgprot(0), mask);
	505	}
	506	#endif
	507
1da177e4	508	#ifdef CONFIG_DEBUG_PAGEALLOC
f62d0f00 IM	509
	510	static int __set_pages_p(struct page *page, int numpages)
	511	{
	512	unsigned long addr = (unsigned long)page_address(page);
72932c7a TG	513
	514	return __change_page_attr_set(addr, numpages,
	515	__pgprot(_PAGE_PRESENT \| _PAGE_RW));
f62d0f00 IM	516	}
	517
	518	static int __set_pages_np(struct page *page, int numpages)
	519	{
	520	unsigned long addr = (unsigned long)page_address(page);
72932c7a TG	521
	522	return __change_page_attr_clear(addr, numpages,
	523	__pgprot(_PAGE_PRESENT));
f62d0f00 IM	524	}
f62d0f00 IM	525
1da177e4 LT	526	void kernel_map_pages(struct page *page, int numpages, int enable)
	527	{
	528	if (PageHighMem(page))
	529	return;
9f4c815c	530	if (!enable) {
f9b8404c IM	531	debug_check_no_locks_freed(page_address(page),
f9b8404c IM	532	numpages * PAGE_SIZE);
9f4c815c	533	}
de5097c2	534
12d6f21e IM	535	/*
	536	* If page allocator is not up yet then do not call c_p_a():
	537	*/
	538	if (!debug_pagealloc_enabled)
	539	return;
	540
9f4c815c	541	/*
e4b71dcf IM	542	* The return value is ignored - the calls cannot fail,
e4b71dcf IM	543	* large pages are disabled at boot time:
1da177e4	544	*/
f62d0f00 IM	545	if (enable)
	546	__set_pages_p(page, numpages);
	547	else
	548	__set_pages_np(page, numpages);
9f4c815c IM	549
9f4c815c IM	550	/*
e4b71dcf IM	551	* We should perform an IPI and flush all tlbs,
e4b71dcf IM	552	* but that can deadlock->flush only current cpu:
1da177e4 LT	553	*/
	554	__flush_tlb_all();
	555	}
	556	#endif
d1028a15 AV	557
	558	/*
	559	* The testcases use internal knowledge of the implementation that shouldn't
	560	* be exposed to the rest of the kernel. Include these directly here.
	561	*/
	562	#ifdef CONFIG_CPA_DEBUG
	563	#include "pageattr-test.c"
	564	#endif