x86: cpa: fix the self-test

Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

arch/x86/Kconfig.debug

@@ -40,7 +40,7 @@ comment "Page alloc debug is incompatible with Software Suspend on i386"
 
 config DEBUG_PAGEALLOC
 	bool "Debug page memory allocations"
-	depends on DEBUG_KERNEL
+	depends on DEBUG_KERNEL && X86_32
 	help
 	  Unmap pages from the kernel linear mapping after free_pages().
 	  This results in a large slowdown, but helps to find certain types

arch/x86/mm/init_32.c

@@ -781,8 +781,6 @@ void mark_rodata_ro(void)
 
 void free_init_pages(char *what, unsigned long begin, unsigned long end)
 {
-	unsigned long addr;
-
 #ifdef CONFIG_DEBUG_PAGEALLOC
 	/*
 	 * If debugging page accesses then do not free this memory but
@@ -793,6 +791,8 @@ void free_init_pages(char *what, unsigned long begin, unsigned long end)
 		begin, PAGE_ALIGN(end));
 	set_memory_np(begin, (end - begin) >> PAGE_SHIFT);
 #else
+	unsigned long addr;
+
 	/*
 	 * We just marked the kernel text read only above, now that
 	 * we are going to free part of that, we need to make that

arch/x86/mm/init_64.c

@@ -569,22 +569,6 @@ void free_init_pages(char *what, unsigned long begin, unsigned long end)
 		free_page(addr);
 		totalram_pages++;
 	}
-#ifdef CONFIG_DEBUG_RODATA
-	/*
-	 * This will make the __init pages not present and
-	 * not executable, so that any attempt to use a
-	 * __init function from now on will fault immediately
-	 * rather than supriously later when memory gets reused.
-	 *
-	 * We only do this for DEBUG_RODATA to not break up the
-	 * 2Mb kernel mapping just for this debug feature.
-	 */
-	if (begin >= __START_KERNEL_map) {
-		set_memory_rw(begin, (end - begin)/PAGE_SIZE);
-		set_memory_np(begin, (end - begin)/PAGE_SIZE);
-		set_memory_nx(begin, (end - begin)/PAGE_SIZE);
-	}
-#endif
 #endif
 }
 

arch/x86/mm/pageattr-test.c

@@ -15,8 +15,7 @@
 #include <asm/kdebug.h>
 
 enum {
-	NTEST			= 400,
-	LOWEST_LEVEL		= PG_LEVEL_4K,
+	NTEST			= 4000,
 #ifdef CONFIG_X86_64
 	LPS			= (1 << PMD_SHIFT),
 #elif defined(CONFIG_X86_PAE)
@@ -59,10 +58,10 @@ static __init int print_split(struct split_state *s)
 			continue;
 		}
 
-		if (level == 2 && sizeof(long) == 8) {
+		if (level == PG_LEVEL_1G && sizeof(long) == 8) {
 			s->gpg++;
 			i += GPS/PAGE_SIZE;
-		} else if (level != LOWEST_LEVEL) {
+		} else if (level == PG_LEVEL_2M) {
 			if (!(pte_val(*pte) & _PAGE_PSE)) {
 				printk(KERN_ERR
 					"%lx level %d but not PSE %Lx\n",
@@ -162,7 +161,7 @@ static __init int exercise_pageattr(void)
 			continue;
 		}
 
-		err = __change_page_attr_clear(addr[i], len[i],
+		err = change_page_attr_clear(addr[i], len[i],
 					       __pgprot(_PAGE_GLOBAL));
 		if (err < 0) {
 			printk(KERN_ERR "CPA %d failed %d\n", i, err);
@@ -175,7 +174,7 @@ static __init int exercise_pageattr(void)
 				pte ? (u64)pte_val(*pte) : 0ULL);
 			failed++;
 		}
-		if (level != LOWEST_LEVEL) {
+		if (level != PG_LEVEL_4K) {
 			printk(KERN_ERR "CPA %lx: unexpected level %d\n",
 				addr[i], level);
 			failed++;
@@ -183,7 +182,6 @@ static __init int exercise_pageattr(void)
 
 	}
 	vfree(bm);
-	cpa_flush_all();
 
 	failed += print_split(&sb);
 
@@ -197,7 +195,7 @@ static __init int exercise_pageattr(void)
 			failed++;
 			continue;
 		}
-		err = __change_page_attr_set(addr[i], len[i],
+		err = change_page_attr_set(addr[i], len[i],
 					     __pgprot(_PAGE_GLOBAL));
 		if (err < 0) {
 			printk(KERN_ERR "CPA reverting failed: %d\n", err);
@@ -211,7 +209,6 @@ static __init int exercise_pageattr(void)
 		}
 
 	}
-	cpa_flush_all();
 
 	failed += print_split(&sc);
 

arch/x86/mm/pageattr.c

@@ -197,10 +197,11 @@ static int split_large_page(pte_t *kpte, unsigned long address)
 	unsigned long addr;
 	pte_t *pbase, *tmp;
 	struct page *base;
-	int i, level;
+	unsigned int i, level;
 
 #ifdef CONFIG_DEBUG_PAGEALLOC
-	gfp_flags = GFP_ATOMIC;
+	gfp_flags = __GFP_HIGH | __GFP_NOFAIL | __GFP_NOWARN;
+	gfp_flags = GFP_ATOMIC | __GFP_NOWARN;
 #endif
 	base = alloc_pages(gfp_flags, 0);
 	if (!base)
@@ -224,6 +225,7 @@ static int split_large_page(pte_t *kpte, unsigned long address)
 	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));
 
@@ -248,7 +250,8 @@ out_unlock:
 }
 
 static int
-__change_page_attr(unsigned long address, unsigned long pfn, pgprot_t prot)
+__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;
@@ -267,15 +270,20 @@ repeat:
 	BUG_ON(PageLRU(kpte_page));
 	BUG_ON(PageCompound(kpte_page));
 
-	prot = static_protections(prot, address);
-
 	if (level == PG_LEVEL_4K) {
-		WARN_ON_ONCE(pgprot_val(prot) & _PAGE_PSE);
-		set_pte_atomic(kpte, pfn_pte(pfn, canon_pgprot(prot)));
-	} else {
-		/* Clear the PSE bit for the 4k level pages ! */
-		pgprot_val(prot) = pgprot_val(prot) & ~_PAGE_PSE;
+		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;
@@ -297,22 +305,26 @@ repeat:
  * Modules and drivers should use the set_memory_* APIs instead.
  */
 
-static int change_page_attr_addr(unsigned long address, pgprot_t prot)
+static int
+change_page_attr_addr(unsigned long address, pgprot_t mask_set,
+							pgprot_t mask_clr)
 {
 	int err = 0, kernel_map = 0;
-	unsigned long pfn = __pa(address) >> PAGE_SHIFT;
+	unsigned long pfn;
 
 #ifdef CONFIG_X86_64
 	if (address >= __START_KERNEL_map &&
 			address < __START_KERNEL_map + KERNEL_TEXT_SIZE) {
 
-		address = (unsigned long)__va(__pa(address));
+		address = (unsigned long)__va(__pa((void *)address));
 		kernel_map = 1;
 	}
 #endif
 
-	if (!kernel_map || pte_present(pfn_pte(0, prot))) {
-		err = __change_page_attr(address, pfn, prot);
+	pfn = __pa(address) >> PAGE_SHIFT;
+
+	if (!kernel_map || 1) {
+		err = __change_page_attr(address, pfn, mask_set, mask_clr);
 		if (err)
 			return err;
 	}
@@ -324,12 +336,15 @@ static int change_page_attr_addr(unsigned long address, pgprot_t prot)
 	 */
 	if (__pa(address) < KERNEL_TEXT_SIZE) {
 		unsigned long addr2;
-		pgprot_t prot2;
 
-		addr2 = __START_KERNEL_map + __pa(address);
+		addr2 = __pa(address) + __START_KERNEL_map - phys_base;
 		/* Make sure the kernel mappings stay executable */
-		prot2 = pte_pgprot(pte_mkexec(pfn_pte(0, prot)));
-		err = __change_page_attr(addr2, pfn, prot2);
+		pgprot_val(mask_clr) |= _PAGE_NX;
+		/*
+		 * Our high aliases are imprecise, so do not propagate
+		 * failures back to users:
+		 */
+		__change_page_attr(addr2, pfn, mask_set, mask_clr);
 	}
 #endif
 
@@ -339,26 +354,13 @@ static int change_page_attr_addr(unsigned long address, pgprot_t prot)
 static int __change_page_attr_set_clr(unsigned long addr, int numpages,
 				      pgprot_t mask_set, pgprot_t mask_clr)
 {
-	pgprot_t new_prot;
-	int level;
-	pte_t *pte;
-	int i, ret;
-
-	for (i = 0; i < numpages ; i++) {
-
-		pte = lookup_address(addr, &level);
-		if (!pte)
-			return -EINVAL;
-
-		new_prot = pte_pgprot(*pte);
-
-		pgprot_val(new_prot) &= ~pgprot_val(mask_clr);
-		pgprot_val(new_prot) |= pgprot_val(mask_set);
+	unsigned int i;
+	int ret;
 
-		ret = change_page_attr_addr(addr, new_prot);
+	for (i = 0; i < numpages ; i++, addr += PAGE_SIZE) {
+		ret = change_page_attr_addr(addr, mask_set, mask_clr);
 		if (ret)
 			return ret;
-		addr += PAGE_SIZE;
 	}
 
 	return 0;

include/asm-x86/pgtable.h

@@ -240,6 +240,7 @@ enum {
 	PG_LEVEL_NONE,
 	PG_LEVEL_4K,
 	PG_LEVEL_2M,
+	PG_LEVEL_1G,
 };
 
 /*