[PATCH] change_page_attr and AGP update

Add change_page_attr to change page attributes for the kernel linear map.

Fix AGP driver to use change_page_attr for the AGP buffer.

Clean up AGP driver a bit (only tested on i386/VIA+AMD)

Change ioremap_nocache to use change_page_attr to avoid mappings with
conflicting caching attributes.

arch/i386/mm/Makefile

@@ -9,6 +9,7 @@
 
 O_TARGET := mm.o
 
-obj-y	 := init.o fault.o ioremap.o extable.o
+obj-y	 := init.o fault.o ioremap.o extable.o pageattr.o
+export-objs := pageattr.o
 
 include $(TOPDIR)/Rules.make

arch/i386/mm/ioremap.c

@@ -10,12 +10,13 @@
 
 #include <linux/vmalloc.h>
 #include <linux/init.h>
+#include <linux/slab.h>
 #include <asm/io.h>
 #include <asm/pgalloc.h>
 #include <asm/fixmap.h>
 #include <asm/cacheflush.h>
 #include <asm/tlbflush.h>
-
+#include <asm/pgtable.h>
 
 static inline void remap_area_pte(pte_t * pte, unsigned long address, unsigned long size,
 	unsigned long phys_addr, unsigned long flags)
@@ -155,6 +156,7 @@ void * __ioremap(unsigned long phys_addr, unsigned long size, unsigned long flag
 	area = get_vm_area(size, VM_IOREMAP);
 	if (!area)
 		return NULL;
+	area->phys_addr = phys_addr;
 	addr = area->addr;
 	if (remap_area_pages(VMALLOC_VMADDR(addr), phys_addr, size, flags)) {
 		vfree(addr);
@@ -163,10 +165,71 @@ void * __ioremap(unsigned long phys_addr, unsigned long size, unsigned long flag
 	return (void *) (offset + (char *)addr);
 }
 
+
+/**
+ * ioremap_nocache     -   map bus memory into CPU space
+ * @offset:    bus address of the memory
+ * @size:      size of the resource to map
+ *
+ * ioremap_nocache performs a platform specific sequence of operations to
+ * make bus memory CPU accessible via the readb/readw/readl/writeb/
+ * writew/writel functions and the other mmio helpers. The returned
+ * address is not guaranteed to be usable directly as a virtual
+ * address. 
+ *
+ * This version of ioremap ensures that the memory is marked uncachable
+ * on the CPU as well as honouring existing caching rules from things like
+ * the PCI bus. Note that there are other caches and buffers on many 
+ * busses. In particular driver authors should read up on PCI writes
+ *
+ * It's useful if some control registers are in such an area and
+ * write combining or read caching is not desirable:
+ * 
+ * Must be freed with iounmap.
+ */
+
+void *ioremap_nocache (unsigned long phys_addr, unsigned long size)
+{
+	void *p = __ioremap(phys_addr, size, _PAGE_PCD);
+	if (!p) 
+		return p; 
+
+	if (phys_addr + size < virt_to_phys(high_memory)) { 
+		struct page *ppage = virt_to_page(__va(phys_addr));		
+		unsigned long npages = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
+
+		BUG_ON(phys_addr+size > (unsigned long)high_memory);
+		BUG_ON(phys_addr + size < phys_addr);
+
+		if (change_page_attr(ppage, npages, PAGE_KERNEL_NOCACHE) < 0) { 
+			iounmap(p); 
+			p = NULL;
+		}
+	} 
+
+	return p;					
+}
+
 void iounmap(void *addr)
 {
-	if (addr > high_memory)
-		return vfree((void *) (PAGE_MASK & (unsigned long) addr));
+	struct vm_struct *p;
+	if (addr < high_memory) 
+		return; 
+	p = remove_kernel_area(addr); 
+	if (!p) { 
+		printk("__iounmap: bad address %p\n", addr);
+		return;
+	} 
+
+	BUG_ON(p->phys_addr == 0);  /* not allocated with ioremap */	
+
+	vmfree_area_pages(VMALLOC_VMADDR(p->addr), p->size);	
+	if (p->flags && p->phys_addr < virt_to_phys(high_memory)) { 
+		change_page_attr(virt_to_page(__va(p->phys_addr)),
+				 p->size >> PAGE_SHIFT,
+				 PAGE_KERNEL); 				 
+	} 
+	kfree(p); 
 }
 
 void __init *bt_ioremap(unsigned long phys_addr, unsigned long size)

arch/i386/mm/pageattr.c

+/* 
+ * Copyright 2002 Andi Kleen, SuSE Labs. 
+ * Thanks to Ben LaHaise for precious feedback.
+ */ 
+
+#include <linux/config.h>
+#include <linux/mm.h>
+#include <linux/sched.h>
+#include <linux/highmem.h>
+#include <linux/module.h>
+#include <linux/slab.h>
+#include <asm/uaccess.h>
+#include <asm/processor.h>
+
+static inline pte_t *lookup_address(unsigned long address) 
+{ 
+	pgd_t *pgd = pgd_offset_k(address); 
+	pmd_t *pmd = pmd_offset(pgd, address); 	       
+	if (pmd_large(*pmd))
+		return (pte_t *)pmd;
+        return pte_offset_kernel(pmd, address);
+} 
+
+static struct page *split_large_page(unsigned long address, pgprot_t prot)
+{ 
+	int i; 
+	unsigned long addr;
+	struct page *base = alloc_pages(GFP_KERNEL, 0);
+	pte_t *pbase;
+	if (!base) 
+		return NULL;
+	address = __pa(address);
+	addr = address & LARGE_PAGE_MASK; 
+	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);
+	}
+	return base;
+} 
+
+static void flush_kernel_map(void *dummy) 
+{ 
+	/* Could use CLFLUSH here if the CPU supports it (Hammer,P4) */
+	if (boot_cpu_data.x86_model >= 4) 
+		asm volatile("wbinvd":::"memory"); 
+	/* Flush all to work around Errata in early athlons regarding 
+	 * large page flushing. 
+	 */
+	__flush_tlb_all(); 	
+}
+
+static void set_pmd_pte(pte_t *kpte, unsigned long address, pte_t pte) 
+{ 
+	set_pte_atomic(kpte, pte); 	/* change init_mm */
+#ifndef CONFIG_X86_PAE
+	{
+		struct list_head *l;
+		spin_lock(&mmlist_lock);
+		list_for_each(l, &init_mm.mmlist) { 
+			struct mm_struct *mm = list_entry(l, struct mm_struct, mmlist);
+			pmd_t *pmd = pmd_offset(pgd_offset(mm, address), address);
+			set_pte_atomic((pte_t *)pmd, pte);
+		} 
+		spin_unlock(&mmlist_lock);
+	}
+#endif
+}
+
+/* 
+ * No more special protections in this 2/4MB area - revert to a
+ * large page again. 
+ */
+static inline void revert_page(struct page *kpte_page, unsigned long address)
+{
+	pte_t *linear = (pte_t *) 
+		pmd_offset(pgd_offset(&init_mm, address), address);
+	set_pmd_pte(linear,  address,
+		    pfn_pte((__pa(address) & LARGE_PAGE_MASK) >> PAGE_SHIFT,
+			    PAGE_KERNEL_LARGE));
+}
+
+static int
+__change_page_attr(struct page *page, pgprot_t prot, struct page **oldpage) 
+{ 
+	pte_t *kpte; 
+	unsigned long address;
+	struct page *kpte_page;
+
+#ifdef CONFIG_HIGHMEM
+	if (page >= highmem_start_page) 
+		BUG(); 
+#endif
+	address = (unsigned long)page_address(page);
+
+	kpte = lookup_address(address);
+	kpte_page = virt_to_page(((unsigned long)kpte) & PAGE_MASK);
+	if (pgprot_val(prot) != pgprot_val(PAGE_KERNEL)) { 
+		if ((pte_val(*kpte) & _PAGE_PSE) == 0) { 
+			pte_t old = *kpte;
+			pte_t standard = mk_pte(page, PAGE_KERNEL); 
+
+			set_pte_atomic(kpte, mk_pte(page, prot)); 
+			if (pte_same(old,standard))
+				atomic_inc(&kpte_page->count);
+		} else {
+			struct page *split = split_large_page(address, prot); 
+			if (!split)
+				return -ENOMEM;
+			set_pmd_pte(kpte,address,mk_pte(split, PAGE_KERNEL));
+		}	
+	} else if ((pte_val(*kpte) & _PAGE_PSE) == 0) { 
+		set_pte_atomic(kpte, mk_pte(page, PAGE_KERNEL));
+		atomic_dec(&kpte_page->count); 
+	}
+
+	if (cpu_has_pse && (atomic_read(&kpte_page->count) == 1)) { 
+		*oldpage = kpte_page;
+		revert_page(kpte_page, address);
+	} 
+	return 0;
+} 
+
+static inline void flush_map(void)
+{	
+#ifdef CONFIG_SMP 
+	smp_call_function(flush_kernel_map, NULL, 1, 1);
+#endif	
+	flush_kernel_map(NULL);
+}
+
+struct deferred_page { 
+	struct deferred_page *next; 
+	struct page *fpage;
+}; 
+static struct deferred_page *df_list; /* protected by init_mm.mmap_sem */
+
+/*
+ * Change the page attributes of an page in the linear mapping.
+ *
+ * This should be used when a page is mapped with a different caching policy
+ * than write-back somewhere - some CPUs do not like it when mappings with
+ * different caching policies exist. This changes the page attributes of the
+ * in kernel linear mapping too.
+ * 
+ * The caller needs to ensure that there are no conflicting mappings elsewhere.
+ * This function only deals with the kernel linear map.
+ * 
+ * Caller must call global_flush_tlb() after this.
+ */
+int change_page_attr(struct page *page, int numpages, pgprot_t prot)
+{
+	int err = 0; 
+	struct page *fpage; 
+	int i; 
+
+	down_write(&init_mm.mmap_sem);
+	for (i = 0; i < numpages; i++, page++) { 
+		fpage = NULL;
+		err = __change_page_attr(page, prot, &fpage); 
+		if (err) 
+			break; 
+		if (fpage) { 
+			struct deferred_page *df;
+			df = kmalloc(sizeof(struct deferred_page), GFP_KERNEL); 
+			if (!df) {
+				flush_map();
+				__free_page(fpage);
+			} else { 
+				df->next = df_list;
+				df->fpage = fpage;				
+				df_list = df;
+			} 			
+		} 
+	} 	
+	up_write(&init_mm.mmap_sem); 
+	return err;
+}
+
+void global_flush_tlb(void)
+{ 
+	struct deferred_page *df, *next_df;
+
+	down_read(&init_mm.mmap_sem);
+	df = xchg(&df_list, NULL);
+	up_read(&init_mm.mmap_sem);
+	flush_map();
+	for (; df; df = next_df) { 
+		next_df = df->next;
+		if (df->fpage) 
+			__free_page(df->fpage);
+		kfree(df);
+	} 
+} 
+
+EXPORT_SYMBOL(change_page_attr);
+EXPORT_SYMBOL(global_flush_tlb);

drivers/char/agp/agp.h

@@ -118,8 +118,8 @@ struct agp_bridge_data {
 	int (*remove_memory) (agp_memory *, off_t, int);
 	agp_memory *(*alloc_by_type) (size_t, int);
 	void (*free_by_type) (agp_memory *);
-	unsigned long (*agp_alloc_page) (void);
-	void (*agp_destroy_page) (unsigned long);
+	void *(*agp_alloc_page) (void);
+	void (*agp_destroy_page) (void *);
 	int (*suspend)(void);
 	void (*resume)(void);
 	

include/asm-alpha/agp.h

+#ifndef AGP_H
+#define AGP_H 1
+
+/* dummy for now */
+
+#define map_page_into_agp(page) 
+#define unmap_page_from_agp(page) 
+#define flush_agp_mappings() 
+#define flush_agp_cache() mb()
+
+#endif

include/asm-i386/agp.h

+#ifndef AGP_H
+#define AGP_H 1
+
+#include <asm/pgtable.h>
+
+/* 
+ * Functions to keep the agpgart mappings coherent with the MMU.
+ * The GART gives the CPU a physical alias of pages in memory. The alias region is
+ * mapped uncacheable. Make sure there are no conflicting mappings
+ * with different cachability attributes for the same page. This avoids
+ * data corruption on some CPUs.
+ */
+
+#define map_page_into_agp(page) change_page_attr(page, 1, PAGE_KERNEL_NOCACHE)
+#define unmap_page_from_agp(page) change_page_attr(page, 1, PAGE_KERNEL)
+#define flush_agp_mappings() global_flush_tlb()
+
+/* Could use CLFLUSH here if the cpu supports it. But then it would
+   need to be called for each cacheline of the whole page so it may not be 
+   worth it. Would need a page for it. */
+#define flush_agp_cache() asm volatile("wbinvd":::"memory")
+
+#endif

include/asm-i386/cacheflush.h

@@ -15,4 +15,7 @@
 #define flush_icache_page(vma,pg)		do { } while (0)
 #define flush_icache_user_range(vma,pg,adr,len)	do { } while (0)
 
+void global_flush_tlb(void); 
+int change_page_attr(struct page *page, int numpages, pgprot_t prot);
+
 #endif /* _I386_CACHEFLUSH_H */

include/asm-i386/io.h

@@ -121,31 +121,7 @@ static inline void * ioremap (unsigned long offset, unsigned long size)
 	return __ioremap(offset, size, 0);
 }
 
-/**
- * ioremap_nocache     -   map bus memory into CPU space
- * @offset:    bus address of the memory
- * @size:      size of the resource to map
- *
- * ioremap_nocache performs a platform specific sequence of operations to
- * make bus memory CPU accessible via the readb/readw/readl/writeb/
- * writew/writel functions and the other mmio helpers. The returned
- * address is not guaranteed to be usable directly as a virtual
- * address. 
- *
- * This version of ioremap ensures that the memory is marked uncachable
- * on the CPU as well as honouring existing caching rules from things like
- * the PCI bus. Note that there are other caches and buffers on many 
- * busses. In paticular driver authors should read up on PCI writes
- *
- * It's useful if some control registers are in such an area and
- * write combining or read caching is not desirable:
- */
-
-static inline void * ioremap_nocache (unsigned long offset, unsigned long size)
-{
-        return __ioremap(offset, size, _PAGE_PCD);
-}
-
+extern void * ioremap_nocache (unsigned long offset, unsigned long size);
 extern void iounmap(void *addr);
 
 /*

include/asm-i386/page.h

@@ -6,6 +6,9 @@
 #define PAGE_SIZE	(1UL << PAGE_SHIFT)
 #define PAGE_MASK	(~(PAGE_SIZE-1))
 
+#define LARGE_PAGE_MASK (~(LARGE_PAGE_SIZE-1))
+#define LARGE_PAGE_SIZE (1UL << PMD_SHIFT)
+
 #ifdef __KERNEL__
 #ifndef __ASSEMBLY__
 

include/asm-i386/pgtable-2level.h

@@ -40,6 +40,7 @@ static inline int pgd_present(pgd_t pgd)	{ return 1; }
  * hook is made available.
  */
 #define set_pte(pteptr, pteval) (*(pteptr) = pteval)
+#define set_pte_atomic(pteptr, pteval) set_pte(pteptr,pteval)
 /*
  * (pmds are folded into pgds so this doesnt get actually called,
  * but the define is needed for a generic inline function.)

include/asm-i386/pgtable-3level.h

@@ -49,6 +49,8 @@ static inline void set_pte(pte_t *ptep, pte_t pte)
 	smp_wmb();
 	ptep->pte_low = pte.pte_low;
 }
+#define set_pte_atomic(pteptr,pteval) \
+		set_64bit((unsigned long long *)(pteptr),pte_val(pteval))
 #define set_pmd(pmdptr,pmdval) \
 		set_64bit((unsigned long long *)(pmdptr),pmd_val(pmdval))
 #define set_pgd(pgdptr,pgdval) \

include/asm-i386/pgtable.h

@@ -237,6 +237,9 @@ static inline pte_t pte_modify(pte_t pte, pgprot_t newprot)
 #define pmd_page(pmd) \
 	(mem_map + (pmd_val(pmd) >> PAGE_SHIFT))
 
+#define pmd_large(pmd) \
+	((pmd_val(pmd) & (_PAGE_PSE|_PAGE_PRESENT)) == (_PAGE_PSE|_PAGE_PRESENT))
+
 /* to find an entry in a page-table-directory. */
 #define pgd_index(address) ((address >> PGDIR_SHIFT) & (PTRS_PER_PGD-1))
 

include/asm-ia64/agp.h

+#ifndef AGP_H
+#define AGP_H 1
+
+/* dummy for now */
+
+#define map_page_into_agp(page) 
+#define unmap_page_from_agp(page) 
+#define flush_agp_mappings() 
+#define flush_agp_cache() mb()
+
+#endif

include/asm-sparc64/agp.h

+#ifndef AGP_H
+#define AGP_H 1
+
+/* dummy for now */
+
+#define map_page_into_agp(page) 
+#define unmap_page_from_agp(page) 
+#define flush_agp_mappings() 
+#define flush_agp_cache() mb()
+
+#endif

include/asm-x86_64/agp.h

+#ifndef AGP_H
+#define AGP_H 1
+
+#include <asm/cacheflush.h>
+
+/* 
+ * Functions to keep the agpgart mappings coherent.
+ * The GART gives the CPU a physical alias of memory. The alias is
+ * mapped uncacheable. Make sure there are no conflicting mappings
+ * with different cachability attributes for the same page.
+ */
+
+#define map_page_into_agp(page) \
+      change_page_attr(page, __pgprot(__PAGE_KERNEL | _PAGE_PCD))
+#define unmap_page_from_agp(page) change_page_attr(page, PAGE_KERNEL)
+#define flush_agp_mappings() global_flush_tlb()
+
+/* Could use CLFLUSH here if the cpu supports it. But then it would
+   need to be called for each cacheline of the whole page so it may not be 
+   worth it. Would need a page for it. */
+#define flush_agp_cache() asm volatile("wbinvd":::"memory")
+
+#endif

include/asm-x86_64/cacheflush.h

@@ -15,4 +15,7 @@
 #define flush_icache_page(vma,pg)		do { } while (0)
 #define flush_icache_user_range(vma,pg,adr,len)	do { } while (0)
 
+void global_flush_tlb(void); 
+int change_page_attr(struct page *page, int numpages, pgprot_t prot);
+
 #endif /* _I386_CACHEFLUSH_H */

include/linux/vmalloc.h

@@ -13,6 +13,7 @@ struct vm_struct {
 	unsigned long flags;
 	void * addr;
 	unsigned long size;
+	unsigned long phys_addr;
 	struct vm_struct * next;
 };
 
@@ -23,6 +24,8 @@ extern long vread(char *buf, char *addr, unsigned long count);
 extern void vmfree_area_pages(unsigned long address, unsigned long size);
 extern int vmalloc_area_pages(unsigned long address, unsigned long size,
                               int gfp_mask, pgprot_t prot);
+extern struct vm_struct *remove_kernel_area(void *addr);
+
 /*
  * Various ways to allocate pages.
  */

mm/vmalloc.c

@@ -195,6 +195,7 @@ struct vm_struct * get_vm_area(unsigned long size, unsigned long flags)
 		if (addr > VMALLOC_END-size)
 			goto out;
 	}
+	area->phys_addr = 0;
 	area->flags = flags;
 	area->addr = (void *)addr;
 	area->size = size;
@@ -209,9 +210,25 @@ struct vm_struct * get_vm_area(unsigned long size, unsigned long flags)
 	return NULL;
 }
 
-void vfree(void * addr)
+struct vm_struct *remove_kernel_area(void *addr) 
 {
 	struct vm_struct **p, *tmp;
+	write_lock(&vmlist_lock);
+	for (p = &vmlist ; (tmp = *p) ; p = &tmp->next) {
+		if (tmp->addr == addr) {
+			*p = tmp->next;
+			write_unlock(&vmlist_lock);
+			return tmp;
+		}
+
+	}
+	write_unlock(&vmlist_lock);
+	return NULL;
+} 
+
+void vfree(void * addr)
+{
+	struct vm_struct *tmp;
 
 	if (!addr)
 		return;
@@ -219,17 +236,12 @@ void vfree(void * addr)
 		printk(KERN_ERR "Trying to vfree() bad address (%p)\n", addr);
 		return;
 	}
-	write_lock(&vmlist_lock);
-	for (p = &vmlist ; (tmp = *p) ; p = &tmp->next) {
-		if (tmp->addr == addr) {
-			*p = tmp->next;
+	tmp = remove_kernel_area(addr); 
+	if (tmp) { 
 			vmfree_area_pages(VMALLOC_VMADDR(tmp->addr), tmp->size);
-			write_unlock(&vmlist_lock);
 			kfree(tmp);
 			return;
 		}
-	}
-	write_unlock(&vmlist_lock);
 	printk(KERN_ERR "Trying to vfree() nonexistent vm area (%p)\n", addr);
 }