Merge bk://bk.arm.linux.org.uk/linux-2.6-rmk

into ppc970.osdl.org:/home/torvalds/v2.6/linux

arch/arm/Kconfig

@@ -601,12 +601,8 @@ source "arch/arm/oprofile/Kconfig"
 
 source "drivers/video/Kconfig"
 
-if ARCH_ACORN || ARCH_CLPS7500 || ARCH_TBOX || ARCH_SHARK || ARCH_SA1100 || PCI
-
 source "sound/Kconfig"
 
-endif
-
 source "drivers/misc/Kconfig"
 
 source "drivers/usb/Kconfig"

arch/arm/mm/consistent.c

@@ -76,34 +76,27 @@ static struct vm_region consistent_head = {
 	.vm_end		= CONSISTENT_END,
 };
 
-#if 0
-static void vm_region_dump(struct vm_region *head, char *fn)
+static struct vm_region *
+vm_region_alloc(struct vm_region *head, size_t size, int gfp)
 {
-	struct vm_region *c;
+	unsigned long addr = head->vm_start, end = head->vm_end - size;
+	unsigned long flags;
+	struct vm_region *c, *new;
 
-	printk("Consistent Allocation Map (%s):\n", fn);
-	list_for_each_entry(c, &head->vm_list, vm_list) {
-		printk(" %p:  %08lx - %08lx   (0x%08x)\n", c,
-		       c->vm_start, c->vm_end, c->vm_end - c->vm_start);
-	}
-}
-#else
-#define vm_region_dump(head,fn)	do { } while(0)
-#endif
+	new = kmalloc(sizeof(struct vm_region), gfp);
+	if (!new)
+		goto out;
 
-static int vm_region_alloc(struct vm_region *head, struct vm_region *new, size_t size)
-{
-	unsigned long addr = head->vm_start, end = head->vm_end - size;
-	struct vm_region *c;
+	spin_lock_irqsave(&consistent_lock, flags);
 
 	list_for_each_entry(c, &head->vm_list, vm_list) {
 		if ((addr + size) < addr)
-			goto out;
+			goto nospc;
 		if ((addr + size) <= c->vm_start)
 			goto found;
 		addr = c->vm_end;
 		if (addr > end)
-			goto out;
+			goto nospc;
 	}
 
  found:
@@ -114,10 +107,14 @@ static int vm_region_alloc(struct vm_region *head, struct vm_region *new, size_t
 	new->vm_start = addr;
 	new->vm_end = addr + size;
 
-	return 0;
+	spin_unlock_irqrestore(&consistent_lock, flags);
+	return new;
 
+ nospc:
+	spin_unlock_irqrestore(&consistent_lock, flags);
+	kfree(new);
  out:
-	return -ENOMEM;
+	return NULL;
 }
 
 static struct vm_region *vm_region_find(struct vm_region *head, unsigned long addr)
@@ -133,28 +130,46 @@ static struct vm_region *vm_region_find(struct vm_region *head, unsigned long ad
 	return c;
 }
 
-/*
- * This allocates one page of cache-coherent memory space and returns
- * both the virtual and a "dma" address to that space.
- */
-void *consistent_alloc(int gfp, size_t size, dma_addr_t *handle,
-		       unsigned long cache_flags)
+#ifdef CONFIG_HUGETLB_PAGE
+#error ARM Coherent DMA allocator does not (yet) support huge TLB
+#endif
+
+static void *
+__dma_alloc(struct device *dev, size_t size, dma_addr_t *handle, int gfp,
+	    pgprot_t prot)
 {
 	struct page *page;
 	struct vm_region *c;
-	unsigned long order, flags;
-	void *ret = NULL;
-	int res;
+	unsigned long order;
+	u64 mask = 0x00ffffff, limit; /* ISA default */
 
 	if (!consistent_pte) {
-		printk(KERN_ERR "consistent_alloc: not initialised\n");
+		printk(KERN_ERR "%s: not initialised\n", __func__);
 		dump_stack();
 		return NULL;
 	}
 
+	if (dev) {
+		mask = dev->coherent_dma_mask;
+		if (mask == 0) {
+			dev_warn(dev, "coherent DMA mask is unset\n");
+			return NULL;
+		}
+	}
+
 	size = PAGE_ALIGN(size);
+	limit = (mask + 1) & ~mask;
+	if ((limit && size >= limit) || size >= (CONSISTENT_END - CONSISTENT_BASE)) {
+		printk(KERN_WARNING "coherent allocation too big (requested %#x mask %#Lx)\n",
+		       size, mask);
+		return NULL;
+	}
+
 	order = get_order(size);
 
+	if (mask != 0xffffffff)
+		gfp |= GFP_DMA;
+
 	page = alloc_pages(gfp, order);
 	if (!page)
 		goto no_page;
@@ -165,36 +180,18 @@ void *consistent_alloc(int gfp, size_t size, dma_addr_t *handle,
 	 */
 	{
 		unsigned long kaddr = (unsigned long)page_address(page);
-		dmac_inv_range(kaddr, kaddr + size);
+		memset(page_address(page), 0, size);
+		dmac_flush_range(kaddr, kaddr + size);
 	}
 
 	/*
-	 * Our housekeeping doesn't need to come from DMA,
-	 * but it must not come from highmem.
-	 */
-	c = kmalloc(sizeof(struct vm_region),
-		    gfp & ~(__GFP_DMA | __GFP_HIGHMEM));
-	if (!c)
-		goto no_remap;
-
-	/*
-	 * Attempt to allocate a virtual address in the
-	 * consistent mapping region.
+	 * Allocate a virtual address in the consistent mapping region.
 	 */
-	spin_lock_irqsave(&consistent_lock, flags);
-	vm_region_dump(&consistent_head, "before alloc");
-
-	res = vm_region_alloc(&consistent_head, c, size);
-
-	vm_region_dump(&consistent_head, "after alloc");
-	spin_unlock_irqrestore(&consistent_lock, flags);
-
-	if (!res) {
+	c = vm_region_alloc(&consistent_head, size,
+			    gfp & ~(__GFP_DMA | __GFP_HIGHMEM));
+	if (c) {
 		pte_t *pte = consistent_pte + CONSISTENT_OFFSET(c->vm_start);
 		struct page *end = page + (1 << order);
-		pgprot_t prot = __pgprot(L_PTE_PRESENT | L_PTE_YOUNG |
-					 L_PTE_DIRTY | L_PTE_WRITE |
-					 cache_flags);
 
 		/*
 		 * Set the "dma handle"
@@ -220,38 +217,43 @@ void *consistent_alloc(int gfp, size_t size, dma_addr_t *handle,
 			page++;
 		}
 
-		ret = (void *)c->vm_start;
+		return (void *)c->vm_start;
 	}
 
- no_remap:
-	if (ret == NULL) {
-		kfree(c);
+	if (page)
 		__free_pages(page, order);
-	}
  no_page:
-	return ret;
+	return NULL;
 }
-EXPORT_SYMBOL(consistent_alloc);
 
 /*
- * Since we have the DMA mask available to us here, we could try to do
- * a normal allocation, and only fall back to a "DMA" allocation if the
- * resulting bus address does not satisfy the dma_mask requirements.
+ * Allocate DMA-coherent memory space and return both the kernel remapped
+ * virtual and bus address for that space.
  */
 void *
 dma_alloc_coherent(struct device *dev, size_t size, dma_addr_t *handle, int gfp)
 {
-	if (dev == NULL || *dev->dma_mask != 0xffffffff)
-		gfp |= GFP_DMA;
-
-	return consistent_alloc(gfp, size, handle, 0);
+	return __dma_alloc(dev, size, handle, gfp,
+			   pgprot_noncached(pgprot_kernel));
 }
 EXPORT_SYMBOL(dma_alloc_coherent);
 
+/*
+ * Allocate a writecombining region, in much the same way as
+ * dma_alloc_coherent above.
+ */
+void *
+dma_alloc_writecombine(struct device *dev, size_t size, dma_addr_t *handle, int gfp)
+{
+	return __dma_alloc(dev, size, handle, gfp,
+			   pgprot_writecombine(pgprot_kernel));
+}
+EXPORT_SYMBOL(dma_alloc_writecombine);
+
 /*
  * free a page as defined by the above mapping.
  */
-void consistent_free(void *vaddr, size_t size, dma_addr_t handle)
+void dma_free_coherent(struct device *dev, size_t size, void *cpu_addr, dma_addr_t handle)
 {
 	struct vm_region *c;
 	unsigned long flags;
@@ -260,15 +262,14 @@ void consistent_free(void *vaddr, size_t size, dma_addr_t handle)
 	size = PAGE_ALIGN(size);
 
 	spin_lock_irqsave(&consistent_lock, flags);
-	vm_region_dump(&consistent_head, "before free");
 
-	c = vm_region_find(&consistent_head, (unsigned long)vaddr);
+	c = vm_region_find(&consistent_head, (unsigned long)cpu_addr);
 	if (!c)
 		goto no_area;
 
 	if ((c->vm_end - c->vm_start) != size) {
-		printk(KERN_ERR "consistent_free: wrong size (%ld != %d)\n",
-		       c->vm_end - c->vm_start, size);
+		printk(KERN_ERR "%s: freeing wrong coherent size (%ld != %d)\n",
+		       __func__, c->vm_end - c->vm_start, size);
 		dump_stack();
 		size = c->vm_end - c->vm_start;
 	}
@@ -292,15 +293,14 @@ void consistent_free(void *vaddr, size_t size, dma_addr_t handle)
 			}
 		}
 
-		printk(KERN_CRIT "consistent_free: bad page in kernel page "
-		       "table\n");
+		printk(KERN_CRIT "%s: bad page in kernel page table\n",
+		       __func__);
 	} while (size -= PAGE_SIZE);
 
 	flush_tlb_kernel_range(c->vm_start, c->vm_end);
 
 	list_del(&c->vm_list);
 
-	vm_region_dump(&consistent_head, "after free");
 	spin_unlock_irqrestore(&consistent_lock, flags);
 
 	kfree(c);
@@ -308,11 +308,11 @@ void consistent_free(void *vaddr, size_t size, dma_addr_t handle)
 
  no_area:
 	spin_unlock_irqrestore(&consistent_lock, flags);
-	printk(KERN_ERR "consistent_free: trying to free "
-	       "invalid area: %p\n", vaddr);
+	printk(KERN_ERR "%s: trying to free invalid coherent area: %p\n",
+	       __func__, cpu_addr);
 	dump_stack();
 }
-EXPORT_SYMBOL(consistent_free);
+EXPORT_SYMBOL(dma_free_coherent);
 
 /*
  * Initialise the consistent memory allocation.
@@ -330,7 +330,7 @@ static int __init consistent_init(void)
 		pgd = pgd_offset(&init_mm, CONSISTENT_BASE);
 		pmd = pmd_alloc(&init_mm, pgd, CONSISTENT_BASE);
 		if (!pmd) {
-			printk(KERN_ERR "consistent_init: no pmd tables\n");
+			printk(KERN_ERR "%s: no pmd tables\n", __func__);
 			ret = -ENOMEM;
 			break;
 		}
@@ -338,7 +338,7 @@ static int __init consistent_init(void)
 
 		pte = pte_alloc_kernel(&init_mm, pmd, CONSISTENT_BASE);
 		if (!pte) {
-			printk(KERN_ERR "consistent_init: no pte tables\n");
+			printk(KERN_ERR "%s: no pte tables\n", __func__);
 			ret = -ENOMEM;
 			break;
 		}

drivers/acorn/char/i2c.c

@@ -34,9 +34,13 @@ extern int (*set_rtc)(void);
 static struct i2c_client *rtc_client;
 static const unsigned char days_in_mon[] = 
 	{ 0, 31, 28, 31, 30, 31, 30, 31, 31, 30, 31, 30, 31 };
-static unsigned int rtc_epoch = 1900;
 
 #define CMOS_CHECKSUM	(63)
+
+/*
+ * Acorn machines store the year in the static RAM at
+ * location 128.
+ */
 #define CMOS_YEAR	(64 + 128)
 
 static inline int rtc_command(int cmd, void *data)
@@ -49,6 +53,38 @@ static inline int rtc_command(int cmd, void *data)
 	return ret;
 }
 
+/*
+ * Update the century + year bytes in the CMOS RAM, ensuring
+ * that the check byte is correctly adjusted for the change.
+ */
+static int rtc_update_year(unsigned int new_year)
+{
+	unsigned char yr[2], chk;
+	struct mem cmos_year  = { CMOS_YEAR, sizeof(yr), yr };
+	struct mem cmos_check = { CMOS_CHECKSUM, 1, &chk };
+	int ret;
+
+	ret = rtc_command(MEM_READ, &cmos_check);
+	if (ret)
+		goto out;
+	ret = rtc_command(MEM_READ, &cmos_year);
+	if (ret)
+		goto out;
+
+	chk -= yr[1] + yr[0];
+
+	yr[1] = new_year / 100;
+	yr[0] = new_year % 100;
+
+	chk += yr[1] + yr[0];
+
+	ret = rtc_command(MEM_WRITE, &cmos_year);
+	if (ret == 0)
+		ret = rtc_command(MEM_WRITE, &cmos_check);
+ out:
+	return ret;
+}
+
 /*
  * Read the current RTC time and date, and update xtime.
  */
@@ -56,45 +92,51 @@ static void get_rtc_time(struct rtc_tm *rtctm, unsigned int *year)
 {
 	unsigned char ctrl, yr[2];
 	struct mem rtcmem = { CMOS_YEAR, sizeof(yr), yr };
+	int real_year, year_offset;
 
 	/*
 	 * Ensure that the RTC is running.
 	 */
 	rtc_command(RTC_GETCTRL, &ctrl);
 	if (ctrl & 0xc0) {
-		unsigned char new_ctrl;
-
-		new_ctrl = ctrl & ~0xc0;
+		unsigned char new_ctrl = ctrl & ~0xc0;
 
-		printk("RTC: resetting control %02X -> %02X\n",
+		printk(KERN_WARNING "RTC: resetting control %02x -> %02x\n",
 		       ctrl, new_ctrl);
 
 		rtc_command(RTC_SETCTRL, &new_ctrl);
 	}
 
-	/*
-	 * Acorn machines store the year in
-	 * the static RAM at location 192.
-	 */
-	if (rtc_command(MEM_READ, &rtcmem))
+	if (rtc_command(RTC_GETDATETIME, rtctm) ||
+	    rtc_command(MEM_READ, &rtcmem))
 		return;
 
-	if (rtc_command(RTC_GETDATETIME, rtctm))
-		return;
+	real_year = yr[0];
 
-	*year = yr[1] * 100 + yr[0];
+	/*
+	 * The RTC year holds the LSB two bits of the current
+	 * year, which should reflect the LSB two bits of the
+	 * CMOS copy of the year.  Any difference indicates
+	 * that we have to correct the CMOS version.
+	 */
+	year_offset = rtctm->year_off - (real_year & 3);
+	if (year_offset < 0)
+		/*
+		 * RTC year wrapped.  Adjust it appropriately.
+		 */
+		year_offset += 4;
+
+	*year = real_year + year_offset + yr[1] * 100;
 }
 
 static int set_rtc_time(struct rtc_tm *rtctm, unsigned int year)
 {
-	unsigned char yr[2], leap, chk;
-	struct mem cmos_year  = { CMOS_YEAR, sizeof(yr), yr };
-	struct mem cmos_check = { CMOS_CHECKSUM, 1, &chk };
+	unsigned char leap;
 	int ret;
 
 	leap = (!(year % 4) && (year % 100)) || !(year % 400);
 
-	if (rtctm->mon > 12 || rtctm->mday == 0)
+	if (rtctm->mon > 12 || rtctm->mon == 0 || rtctm->mday == 0)
 		return -EINVAL;
 
 	if (rtctm->mday > (days_in_mon[rtctm->mon] + (rtctm->mon == 2 && leap)))
@@ -103,21 +145,16 @@ static int set_rtc_time(struct rtc_tm *rtctm, unsigned int year)
 	if (rtctm->hours >= 24 || rtctm->mins >= 60 || rtctm->secs >= 60)
 		return -EINVAL;
 
-	ret = rtc_command(RTC_SETDATETIME, rtctm);
-	if (ret == 0) {
-		rtc_command(MEM_READ, &cmos_check);
-		rtc_command(MEM_READ, &cmos_year);
-
-		chk -= yr[1] + yr[0];
-
-		yr[1] = year / 100;
-		yr[0] = year % 100;
+	/*
+	 * The RTC's own 2-bit year must reflect the least
+	 * significant two bits of the CMOS year.
+	 */
+	rtctm->year_off = (year % 100) & 3;
 
-		chk += yr[1] + yr[0];
+	ret = rtc_command(RTC_SETDATETIME, rtctm);
+	if (ret == 0)
+		ret = rtc_update_year(year);
 
-		rtc_command(MEM_WRITE, &cmos_year);
-		rtc_command(MEM_WRITE, &cmos_check);
-	}
 	return ret;
 }
 
@@ -189,13 +226,12 @@ static int rtc_ioctl(struct inode *inode, struct file *file,
 		rtc_raw.hours    = rtctm.tm_hour;
 		rtc_raw.mday     = rtctm.tm_mday;
 		rtc_raw.mon      = rtctm.tm_mon + 1;
-		rtc_raw.year_off = 2;
 		year             = rtctm.tm_year + 1900;
 		return set_rtc_time(&rtc_raw, year);
 		break;
 
 	case RTC_EPOCH_READ:
-		return put_user(rtc_epoch, (unsigned long *)arg);
+		return put_user(1900, (unsigned long *)arg);
 
 	}
 	return -EINVAL;

drivers/video/acornfb.c

@@ -29,6 +29,8 @@
 #include <linux/slab.h>
 #include <linux/init.h>
 #include <linux/fb.h>
+#include <linux/device.h>
+#include <linux/dma-mapping.h>
 
 #include <asm/hardware.h>
 #include <asm/io.h>
@@ -1254,6 +1256,11 @@ free_unused_pages(unsigned int virtual_start, unsigned int virtual_end)
 	printk("acornfb: freed %dK memory\n", mb_freed);
 }
 
+static struct device acornfb_device = {
+	.bus_id			= "acornfb",
+	.coherent_dma_mask	= 0xffffffff,
+};
+
 int __init
 acornfb_init(void)
 {
@@ -1263,6 +1270,8 @@ acornfb_init(void)
 
 	acornfb_init_fbinfo();
 
+	current_par.dev = &acornfb_device;
+
 	if (current_par.montype == -1)
 		current_par.montype = acornfb_detect_monitortype();
 
@@ -1323,37 +1332,30 @@ acornfb_init(void)
 
 #if defined(HAS_VIDC20)
 	if (!current_par.using_vram) {
+		dma_addr_t handle;
+		void *base;
+
 		/*
 		 * RiscPC needs to allocate the DRAM memory
 		 * for the framebuffer if we are not using
-		 * VRAM.  Archimedes/A5000 machines use a
-		 * fixed address for their framebuffers.
+		 * VRAM.
 		 */
-		unsigned long page, top, base;
-		int order = get_order(size);
-
-		base = __get_free_pages(GFP_KERNEL, order);
-		if (base == 0) {
+		base = dma_alloc_writecombine(current_par.dev, size, &handle,
+					      GFP_KERNEL);
+		if (base == NULL) {
 			printk(KERN_ERR "acornfb: unable to allocate screen "
 			       "memory\n");
 			return -ENOMEM;
 		}
-		top = base + (PAGE_SIZE << order);
-
-		/* Mark the framebuffer pages as reserved so mmap will work. */
-		for (page = base; page < PAGE_ALIGN(base + size); page += PAGE_SIZE)
-			SetPageReserved(virt_to_page(page));
-		/* Hand back any excess pages that we allocated. */
-		for (page = base + size; page < top; page += PAGE_SIZE)
-			free_page(page);
 
-		fb_info.screen_base = (char *)base;
-		fb_info.fix.smem_start = virt_to_phys(fb_info.screen_base);
+		fb_info.screen_base = base;
+		fb_info.fix.smem_start = handle;
 	}
 #endif
 #if defined(HAS_VIDC)
 	/*
-	 * Free unused pages
+	 * Archimedes/A5000 machines use a fixed address for their
+	 * framebuffers.  Free unused pages
 	 */
 	free_unused_pages(PAGE_OFFSET + size, PAGE_OFFSET + MAX_SIZE);
 #endif

drivers/video/acornfb.h

@@ -47,6 +47,7 @@ union palette {
 };
 
 struct acornfb_par {
+	struct device	*dev;
 	unsigned long	screen_end;
 	unsigned int	dram_size;
 	unsigned int	vram_half_sam;

drivers/video/sa1100fb.c

@@ -1595,12 +1595,18 @@ static int __init sa1100fb_map_video_memory(struct sa1100fb_info *fbi)
 	 * of the framebuffer.
 	 */
 	fbi->map_size = PAGE_ALIGN(fbi->fb.fix.smem_len + PAGE_SIZE);
-	fbi->map_cpu = consistent_alloc(GFP_KERNEL, fbi->map_size,
-					&fbi->map_dma, PTE_BUFFERABLE);
+	fbi->map_cpu = dma_alloc_writecombine(fbi->dev, fbi->map_size,
+					      &fbi->map_dma, GFP_KERNEL);
 
 	if (fbi->map_cpu) {
 		fbi->fb.screen_base = fbi->map_cpu + PAGE_SIZE;
 		fbi->screen_dma = fbi->map_dma + PAGE_SIZE;
+		/*
+		 * FIXME: this is actually the wrong thing to place in
+		 * smem_start.  But fbdev suffers from the problem that
+		 * it needs an API which doesn't exist (in this case,
+		 * dma_writecombine_mmap)
+		 */
 		fbi->fb.fix.smem_start = fbi->screen_dma;
 	}
 
@@ -1613,7 +1619,7 @@ static struct fb_monspecs monspecs __initdata = {
 };
 
 
-static struct sa1100fb_info * __init sa1100fb_init_fbinfo(void)
+static struct sa1100fb_info * __init sa1100fb_init_fbinfo(struct device *dev)
 {
 	struct sa1100fb_mach_info *inf;
 	struct sa1100fb_info *fbi;
@@ -1624,6 +1630,7 @@ static struct sa1100fb_info * __init sa1100fb_init_fbinfo(void)
 		return NULL;
 
 	memset(fbi, 0, sizeof(struct sa1100fb_info));
+	fbi->dev = dev;
 
 	strcpy(fbi->fb.fix.id, SA1100_NAME);
 
@@ -1703,7 +1710,7 @@ static int __init sa1100fb_probe(struct device *dev)
 	if (!request_mem_region(0xb0100000, 0x10000, "LCD"))
 		return -EBUSY;
 
-	fbi = sa1100fb_init_fbinfo();
+	fbi = sa1100fb_init_fbinfo(dev);
 	ret = -ENOMEM;
 	if (!fbi)
 		goto failed;

drivers/video/sa1100fb.h

@@ -63,6 +63,7 @@ struct sa1100fb_lcd_reg {
 
 struct sa1100fb_info {
 	struct fb_info		fb;
+	struct device		*dev;
 	struct sa1100fb_rgb	*rgb[NR_RGB];
 
 	u_int			max_bpp;

include/asm-arm/dma-mapping.h

@@ -14,8 +14,6 @@
  * devices.  This is the "generic" version.  The PCI specific version
  * is in pci.h
  */
-extern void *consistent_alloc(int gfp, size_t size, dma_addr_t *handle, unsigned long flags);
-extern void consistent_free(void *vaddr, size_t size, dma_addr_t handle);
 extern void consistent_sync(void *kaddr, size_t size, int rw);
 
 /*
@@ -99,12 +97,26 @@ dma_alloc_coherent(struct device *dev, size_t size, dma_addr_t *handle, int gfp)
  * References to memory and mappings associated with cpu_addr/handle
  * during and after this call executing are illegal.
  */
-static inline void
+extern void
 dma_free_coherent(struct device *dev, size_t size, void *cpu_addr,
-		  dma_addr_t handle)
-{
-	consistent_free(cpu_addr, size, handle);
-}
+		  dma_addr_t handle);
+
+/**
+ * dma_alloc_writecombine - allocate writecombining memory for DMA
+ * @dev: valid struct device pointer, or NULL for ISA and EISA-like devices
+ * @size: required memory size
+ * @handle: bus-specific DMA address
+ *
+ * Allocate some uncached, buffered memory for a device for
+ * performing DMA.  This function allocates pages, and will
+ * return the CPU-viewed address, and sets @handle to be the
+ * device-viewed address.
+ */
+extern void *
+dma_alloc_writecombine(struct device *dev, size_t size, dma_addr_t *handle, int gfp);
+
+#define dma_free_writecombine(dev,size,cpu_addr,handle) \
+	dma_free_coherent(dev,size,cpu_addr,handle)
 
 /**
  * dma_map_single - map a single buffer for streaming DMA