bitkeeper revision 1.1159.198.1 (41aa44a7joSgzez98niBGEZAU1bASg)

Get rid of all our uses of virt_to_phys():
 1. Many were buggy, so the code needed fixing or clarifying anyway
 2. It means we can now experiment with kludging virt_to_phys() to be
    the same as virt_to_bus(), to get more buggy drivers working. I haven't
    included the kludge in this patch.

linux-2.4.28-xen-sparse/arch/xen/kernel/ldt.c

@@ -179,7 +179,7 @@ static int write_ldt(void * ptr, unsigned long bytecount, int oldmode)
 {
 	struct mm_struct * mm = current->mm;
 	__u32 entry_1, entry_2, *lp;
-	unsigned long phys_lp;
+	unsigned long mach_lp;
 	int error;
 	struct modify_ldt_ldt_s ldt_info;
 
@@ -208,7 +208,7 @@ static int write_ldt(void * ptr, unsigned long bytecount, int oldmode)
 	}
 
 	lp = (__u32 *) ((ldt_info.entry_number << 3) + (char *) mm->context.ldt);
-	phys_lp = arbitrary_virt_to_phys(lp);
+	mach_lp = arbitrary_virt_to_machine(lp);
 
    	/* Allow LDTs to be cleared by the user. */
    	if (ldt_info.base_addr == 0 && ldt_info.limit == 0) {
@@ -241,7 +241,7 @@ static int write_ldt(void * ptr, unsigned long bytecount, int oldmode)
 
 	/* Install the new entry ...  */
 install:
-	error = HYPERVISOR_update_descriptor(phys_lp, entry_1, entry_2);
+	error = HYPERVISOR_update_descriptor(mach_lp, entry_1, entry_2);
 
 out_unlock:
 	up(&mm->context.sem);

linux-2.4.28-xen-sparse/include/asm-xen/pgtable.h

@@ -350,7 +350,7 @@ static inline void make_pages_writable(void *va, unsigned int nr)
     }
 }
 
-static inline unsigned long arbitrary_virt_to_phys(void *va)
+static inline unsigned long arbitrary_virt_to_machine(void *va)
 {
     pgd_t *pgd = pgd_offset_k((unsigned long)va);
     pmd_t *pmd = pmd_offset(pgd, (unsigned long)va);

linux-2.6.9-xen-sparse/arch/xen/i386/kernel/ldt.c

@@ -193,7 +193,7 @@ static int write_ldt(void __user * ptr, unsigned long bytecount, int oldmode)
 {
 	struct mm_struct * mm = current->mm;
 	__u32 entry_1, entry_2, *lp;
-	unsigned long phys_lp;
+	unsigned long mach_lp;
 	int error;
 	struct user_desc ldt_info;
 
@@ -222,7 +222,7 @@ static int write_ldt(void __user * ptr, unsigned long bytecount, int oldmode)
 	}
 
 	lp = (__u32 *) ((ldt_info.entry_number << 3) + (char *) mm->context.ldt);
-	phys_lp = arbitrary_virt_to_phys(lp);
+	mach_lp = arbitrary_virt_to_machine(lp);
 
    	/* Allow LDTs to be cleared by the user. */
    	if (ldt_info.base_addr == 0 && ldt_info.limit == 0) {
@@ -240,7 +240,7 @@ static int write_ldt(void __user * ptr, unsigned long bytecount, int oldmode)
 
 	/* Install the new entry ...  */
 install:
-	error = HYPERVISOR_update_descriptor(phys_lp, entry_1, entry_2);
+	error = HYPERVISOR_update_descriptor(mach_lp, entry_1, entry_2);
 
 out_unlock:
 	up(&mm->context.sem);

linux-2.6.9-xen-sparse/arch/xen/i386/kernel/setup.c

@@ -1389,10 +1389,11 @@ void __init setup_arch(char **cmdline_p)
 	init_mm.end_data = (unsigned long) _edata;
 	init_mm.brk = (PFN_UP(__pa(xen_start_info.pt_base)) + xen_start_info.nr_pt_frames) << PAGE_SHIFT;
 
-	code_resource.start = virt_to_phys(_text);
-	code_resource.end = virt_to_phys(_etext)-1;
-	data_resource.start = virt_to_phys(_etext);
-	data_resource.end = virt_to_phys(_edata)-1;
+	/* XEN: This is nonsense: kernel may not even be contiguous in RAM. */
+	/*code_resource.start = virt_to_phys(_text);*/
+	/*code_resource.end = virt_to_phys(_etext)-1;*/
+	/*data_resource.start = virt_to_phys(_etext);*/
+	/*data_resource.end = virt_to_phys(_edata)-1;*/
 
 	parse_cmdline_early(cmdline_p);
 

linux-2.6.9-xen-sparse/arch/xen/i386/mm/init.c

@@ -435,17 +435,24 @@ void zap_low_mappings (void)
 void __init zone_sizes_init(void)
 {
 	unsigned long zones_size[MAX_NR_ZONES] = {0, 0, 0};
-	unsigned int max_dma, high, low;
+	unsigned int /*max_dma,*/ high, low;
 	
-	max_dma = virt_to_phys((char *)MAX_DMA_ADDRESS) >> PAGE_SHIFT;
+	/*
+	 * XEN: Our notion of "DMA memory" is fake when running over Xen.
+	 * We simply put all RAM in the DMA zone so that those drivers which
+	 * needlessly specify GFP_DMA do not get starved of RAM unnecessarily.
+	 * Those drivers that *do* require lowmem are screwed anyway when
+	 * running over Xen!
+	 */
+	/*max_dma = virt_to_phys((char *)MAX_DMA_ADDRESS) >> PAGE_SHIFT;*/
 	low = max_low_pfn;
 	high = highend_pfn;
 	
-	if (low < max_dma)
+	/*if (low < max_dma)*/
 		zones_size[ZONE_DMA] = low;
-	else {
-		zones_size[ZONE_DMA] = max_dma;
-		zones_size[ZONE_NORMAL] = low - max_dma;
+	/*else*/ {
+		/*zones_size[ZONE_DMA] = max_dma;*/
+		/*zones_size[ZONE_NORMAL] = low - max_dma;*/
 #ifdef CONFIG_HIGHMEM
 		zones_size[ZONE_HIGHMEM] = high - low;
 #endif

linux-2.6.9-xen-sparse/arch/xen/i386/mm/ioremap.c

@@ -36,6 +36,23 @@ void __init bt_iounmap(void *addr, unsigned long size)
 
 #else
 
+/*
+ * Is @address within a RAM page that is local to this virtual machine (i.e.,
+ * not an I/O page; not a RAM page belonging to another VM). See the comment
+ * that accompanies pte_pfn() in pgtable-2level.h to understand why this works.
+ */
+static inline int is_local_ram(unsigned long address)
+{
+	unsigned long mfn = address >> PAGE_SHIFT;
+	unsigned long pfn = mfn_to_pfn(mfn);
+	if (pfn < max_mapnr) {
+		if (pfn_to_mfn(pfn) == mfn)
+			return 1; /* local ram */
+		printk("is_local_ram: ioremapping foreign ram (a bad idea).\n");
+	}
+	return 0; /* i/o memory or foreign ram */
+}
+
 static inline void remap_area_pte(pte_t * pte, unsigned long address, unsigned long size,
 	unsigned long phys_addr, unsigned long flags)
 {
@@ -140,19 +157,16 @@ void __iomem * __ioremap(unsigned long phys_addr, unsigned long size, unsigned l
 	if (!size || last_addr < phys_addr)
 		return NULL;
 
-        if (phys_addr >= 0x0 && last_addr < 0x100000)
-                return isa_bus_to_virt(phys_addr);
-
 	/*
 	 * Don't remap the low PCI/ISA area, it's always mapped..
 	 */
-	if (phys_addr >= 0xA0000 && last_addr < 0x100000)
-		return (void __iomem *) phys_to_virt(phys_addr);
+	if (phys_addr >= 0x0 && last_addr < 0x100000)
+		return isa_bus_to_virt(phys_addr);
 
 	/*
 	 * Don't allow anybody to remap normal RAM that we're using..
 	 */
-	if (machine_to_phys(phys_addr) < virt_to_phys(high_memory)) {
+	if (is_local_ram(phys_addr)) {
 		char *t_addr, *t_end;
 		struct page *page;
 
@@ -219,7 +233,7 @@ void __iomem *ioremap_nocache (unsigned long phys_addr, unsigned long size)
 	/* Guaranteed to be > phys_addr, as per __ioremap() */
 	last_addr = phys_addr + size - 1;
 
-	if (machine_to_phys(last_addr) < virt_to_phys(high_memory)) { 
+	if (is_local_ram(last_addr)) { 
 		struct page *ppage = virt_to_page(bus_to_virt(phys_addr));
 		unsigned long npages;
 
@@ -256,7 +270,7 @@ void iounmap(volatile void __iomem *addr)
 		return;
 	} 
 
-	if (p->flags && machine_to_phys(p->phys_addr) < virt_to_phys(high_memory)) { 
+	if (p->flags && is_local_ram(p->phys_addr)) { 
 		change_page_attr(virt_to_page(bus_to_virt(p->phys_addr)),
 				 p->size >> PAGE_SHIFT,
 				 PAGE_KERNEL); 				 
@@ -276,14 +290,11 @@ void __init *bt_ioremap(unsigned long phys_addr, unsigned long size)
 	if (!size || last_addr < phys_addr)
 		return NULL;
 
-        if (phys_addr >= 0x0 && last_addr < 0x100000)
-                return isa_bus_to_virt(phys_addr);
-
 	/*
 	 * Don't remap the low PCI/ISA area, it's always mapped..
 	 */
-	if (phys_addr >= 0xA0000 && last_addr < 0x100000)
-		return phys_to_virt(phys_addr);
+	if (phys_addr >= 0x0 && last_addr < 0x100000)
+		return isa_bus_to_virt(phys_addr);
 
 	/*
 	 * Mappings have to be page-aligned

linux-2.6.9-xen-sparse/arch/xen/i386/mm/pgtable.c

@@ -195,7 +195,7 @@ void pte_ctor(void *pte, kmem_cache_t *cache, unsigned long unused)
 
 	clear_page(pte);
 	make_page_readonly(pte);
-	queue_pte_pin(virt_to_phys(pte));
+	queue_pte_pin(__pa(pte));
 	flush_page_update_queue();
 }
 
@@ -204,7 +204,7 @@ void pte_dtor(void *pte, kmem_cache_t *cache, unsigned long unused)
 	struct page *page = virt_to_page(pte);
 	ClearPageForeign(page);
 
-	queue_pte_unpin(virt_to_phys(pte));
+	queue_pte_unpin(__pa(pte));
 	make_page_writable(pte);
 	flush_page_update_queue();
 }

linux-2.6.9-xen-sparse/include/asm-xen/asm-i386/io.h

@@ -146,8 +146,8 @@ extern void bt_iounmap(void *addr, unsigned long size);
  *
  * Allow them on x86 for legacy drivers, though.
  */
-#define virt_to_bus(_x) phys_to_machine(virt_to_phys(_x))
-#define bus_to_virt(_x) phys_to_virt(machine_to_phys(_x))
+#define virt_to_bus(_x) phys_to_machine(__pa(_x))
+#define bus_to_virt(_x) __va(machine_to_phys(_x))
 
 /*
  * readX/writeX() are used to access memory mapped devices. On some

linux-2.6.9-xen-sparse/include/asm-xen/asm-i386/pgtable.h

@@ -472,7 +472,7 @@ void make_page_writable(void *va);
 void make_pages_readonly(void *va, unsigned int nr);
 void make_pages_writable(void *va, unsigned int nr);
 
-static inline unsigned long arbitrary_virt_to_phys(void *va)
+static inline unsigned long arbitrary_virt_to_machine(void *va)
 {
 	pgd_t *pgd = pgd_offset_k((unsigned long)va);
 	pmd_t *pmd = pmd_offset(pgd, (unsigned long)va);