fbdev: move udlfb out of staging.

udlfb has undergone a fair bit of cleanup recently and is effectively at
the point where it can be liberated from staging purgatory and promoted
to a real driver.

The outstanding cleanups are all minor, with some of them dependent on
drivers/video headers, so these will be done incrementally from udlfb's
new home.

Requested-by: Bernie Thompson <bernie@plugable.com>
Signed-off-by: Paul Mundt <lethal@linux-sh.org>

1 files changed, 1915 insertions, 0 deletions

diff --git a/drivers/video/udlfb.c b/drivers/video/udlfb.c
new file mode 100644
index 00000000000..0cca4873d49
--- /dev/null
+++ b/drivers/video/udlfb.c
@@ -0,0 +1,1915 @@
+/*
+ * udlfb.c -- Framebuffer driver for DisplayLink USB controller
+ *
+ * Copyright (C) 2009 Roberto De Ioris <roberto@unbit.it>
+ * Copyright (C) 2009 Jaya Kumar <jayakumar.lkml@gmail.com>
+ * Copyright (C) 2009 Bernie Thompson <bernie@plugable.com>
+ *
+ * This file is subject to the terms and conditions of the GNU General Public
+ * License v2. See the file COPYING in the main directory of this archive for
+ * more details.
+ *
+ * Layout is based on skeletonfb by James Simmons and Geert Uytterhoeven,
+ * usb-skeleton by GregKH.
+ *
+ * Device-specific portions based on information from Displaylink, with work
+ * from Florian Echtler, Henrik Bjerregaard Pedersen, and others.
+ */
+
+#include <linux/module.h>
+#include <linux/kernel.h>
+#include <linux/init.h>
+#include <linux/usb.h>
+#include <linux/uaccess.h>
+#include <linux/mm.h>
+#include <linux/fb.h>
+#include <linux/vmalloc.h>
+#include <linux/slab.h>
+#include <linux/delay.h>
+#include <video/udlfb.h>
+
+static struct fb_fix_screeninfo dlfb_fix = {
+	.id =           "udlfb",
+	.type =         FB_TYPE_PACKED_PIXELS,
+	.visual =       FB_VISUAL_TRUECOLOR,
+	.xpanstep =     0,
+	.ypanstep =     0,
+	.ywrapstep =    0,
+	.accel =        FB_ACCEL_NONE,
+};
+
+static const u32 udlfb_info_flags = FBINFO_DEFAULT | FBINFO_READS_FAST |
+#ifdef FBINFO_VIRTFB
+		FBINFO_VIRTFB |
+#endif
+		FBINFO_HWACCEL_IMAGEBLIT | FBINFO_HWACCEL_FILLRECT |
+		FBINFO_HWACCEL_COPYAREA | FBINFO_MISC_ALWAYS_SETPAR;
+
+/*
+ * There are many DisplayLink-based products, all with unique PIDs. We are able
+ * to support all volume ones (circa 2009) with a single driver, so we match
+ * globally on VID. TODO: Probe() needs to detect when we might be running
+ * "future" chips, and bail on those, so a compatible driver can match.
+ */
+static struct usb_device_id id_table[] = {
+	{.idVendor = 0x17e9, .match_flags = USB_DEVICE_ID_MATCH_VENDOR,},
+	{},
+};
+MODULE_DEVICE_TABLE(usb, id_table);
+
+/* module options */
+static int console;   /* Optionally allow fbcon to consume first framebuffer */
+static int fb_defio;  /* Optionally enable experimental fb_defio mmap support */
+
+/* dlfb keeps a list of urbs for efficient bulk transfers */
+static void dlfb_urb_completion(struct urb *urb);
+static struct urb *dlfb_get_urb(struct dlfb_data *dev);
+static int dlfb_submit_urb(struct dlfb_data *dev, struct urb * urb, size_t len);
+static int dlfb_alloc_urb_list(struct dlfb_data *dev, int count, size_t size);
+static void dlfb_free_urb_list(struct dlfb_data *dev);
+
+/*
+ * All DisplayLink bulk operations start with 0xAF, followed by specific code
+ * All operations are written to buffers which then later get sent to device
+ */
+static char *dlfb_set_register(char *buf, u8 reg, u8 val)
+{
+	*buf++ = 0xAF;
+	*buf++ = 0x20;
+	*buf++ = reg;
+	*buf++ = val;
+	return buf;
+}
+
+static char *dlfb_vidreg_lock(char *buf)
+{
+	return dlfb_set_register(buf, 0xFF, 0x00);
+}
+
+static char *dlfb_vidreg_unlock(char *buf)
+{
+	return dlfb_set_register(buf, 0xFF, 0xFF);
+}
+
+/*
+ * On/Off for driving the DisplayLink framebuffer to the display
+ *  0x00 H and V sync on
+ *  0x01 H and V sync off (screen blank but powered)
+ *  0x07 DPMS powerdown (requires modeset to come back)
+ */
+static char *dlfb_enable_hvsync(char *buf, bool enable)
+{
+	if (enable)
+		return dlfb_set_register(buf, 0x1F, 0x00);
+	else
+		return dlfb_set_register(buf, 0x1F, 0x07);
+}
+
+static char *dlfb_set_color_depth(char *buf, u8 selection)
+{
+	return dlfb_set_register(buf, 0x00, selection);
+}
+
+static char *dlfb_set_base16bpp(char *wrptr, u32 base)
+{
+	/* the base pointer is 16 bits wide, 0x20 is hi byte. */
+	wrptr = dlfb_set_register(wrptr, 0x20, base >> 16);
+	wrptr = dlfb_set_register(wrptr, 0x21, base >> 8);
+	return dlfb_set_register(wrptr, 0x22, base);
+}
+
+/*
+ * DisplayLink HW has separate 16bpp and 8bpp framebuffers.
+ * In 24bpp modes, the low 323 RGB bits go in the 8bpp framebuffer
+ */
+static char *dlfb_set_base8bpp(char *wrptr, u32 base)
+{
+	wrptr = dlfb_set_register(wrptr, 0x26, base >> 16);
+	wrptr = dlfb_set_register(wrptr, 0x27, base >> 8);
+	return dlfb_set_register(wrptr, 0x28, base);
+}
+
+static char *dlfb_set_register_16(char *wrptr, u8 reg, u16 value)
+{
+	wrptr = dlfb_set_register(wrptr, reg, value >> 8);
+	return dlfb_set_register(wrptr, reg+1, value);
+}
+
+/*
+ * This is kind of weird because the controller takes some
+ * register values in a different byte order than other registers.
+ */
+static char *dlfb_set_register_16be(char *wrptr, u8 reg, u16 value)
+{
+	wrptr = dlfb_set_register(wrptr, reg, value);
+	return dlfb_set_register(wrptr, reg+1, value >> 8);
+}
+
+/*
+ * LFSR is linear feedback shift register. The reason we have this is
+ * because the display controller needs to minimize the clock depth of
+ * various counters used in the display path. So this code reverses the
+ * provided value into the lfsr16 value by counting backwards to get
+ * the value that needs to be set in the hardware comparator to get the
+ * same actual count. This makes sense once you read above a couple of
+ * times and think about it from a hardware perspective.
+ */
+static u16 dlfb_lfsr16(u16 actual_count)
+{
+	u32 lv = 0xFFFF; /* This is the lfsr value that the hw starts with */
+
+	while (actual_count--) {
+		lv =	 ((lv << 1) |
+			(((lv >> 15) ^ (lv >> 4) ^ (lv >> 2) ^ (lv >> 1)) & 1))
+			& 0xFFFF;
+	}
+
+	return (u16) lv;
+}
+
+/*
+ * This does LFSR conversion on the value that is to be written.
+ * See LFSR explanation above for more detail.
+ */
+static char *dlfb_set_register_lfsr16(char *wrptr, u8 reg, u16 value)
+{
+	return dlfb_set_register_16(wrptr, reg, dlfb_lfsr16(value));
+}
+
+/*
+ * This takes a standard fbdev screeninfo struct and all of its monitor mode
+ * details and converts them into the DisplayLink equivalent register commands.
+ */
+static char *dlfb_set_vid_cmds(char *wrptr, struct fb_var_screeninfo *var)
+{
+	u16 xds, yds;
+	u16 xde, yde;
+	u16 yec;
+
+	/* x display start */
+	xds = var->left_margin + var->hsync_len;
+	wrptr = dlfb_set_register_lfsr16(wrptr, 0x01, xds);
+	/* x display end */
+	xde = xds + var->xres;
+	wrptr = dlfb_set_register_lfsr16(wrptr, 0x03, xde);
+
+	/* y display start */
+	yds = var->upper_margin + var->vsync_len;
+	wrptr = dlfb_set_register_lfsr16(wrptr, 0x05, yds);
+	/* y display end */
+	yde = yds + var->yres;
+	wrptr = dlfb_set_register_lfsr16(wrptr, 0x07, yde);
+
+	/* x end count is active + blanking - 1 */
+	wrptr = dlfb_set_register_lfsr16(wrptr, 0x09,
+			xde + var->right_margin - 1);
+
+	/* libdlo hardcodes hsync start to 1 */
+	wrptr = dlfb_set_register_lfsr16(wrptr, 0x0B, 1);
+
+	/* hsync end is width of sync pulse + 1 */
+	wrptr = dlfb_set_register_lfsr16(wrptr, 0x0D, var->hsync_len + 1);
+
+	/* hpixels is active pixels */
+	wrptr = dlfb_set_register_16(wrptr, 0x0F, var->xres);
+
+	/* yendcount is vertical active + vertical blanking */
+	yec = var->yres + var->upper_margin + var->lower_margin +
+			var->vsync_len;
+	wrptr = dlfb_set_register_lfsr16(wrptr, 0x11, yec);
+
+	/* libdlo hardcodes vsync start to 0 */
+	wrptr = dlfb_set_register_lfsr16(wrptr, 0x13, 0);
+
+	/* vsync end is width of vsync pulse */
+	wrptr = dlfb_set_register_lfsr16(wrptr, 0x15, var->vsync_len);
+
+	/* vpixels is active pixels */
+	wrptr = dlfb_set_register_16(wrptr, 0x17, var->yres);
+
+	/* convert picoseconds to 5kHz multiple for pclk5k = x * 1E12/5k */
+	wrptr = dlfb_set_register_16be(wrptr, 0x1B,
+			200*1000*1000/var->pixclock);
+
+	return wrptr;
+}
+
+/*
+ * This takes a standard fbdev screeninfo struct that was fetched or prepared
+ * and then generates the appropriate command sequence that then drives the
+ * display controller.
+ */
+static int dlfb_set_video_mode(struct dlfb_data *dev,
+				struct fb_var_screeninfo *var)
+{
+	char *buf;
+	char *wrptr;
+	int retval = 0;
+	int writesize;
+	struct urb *urb;
+
+	if (!atomic_read(&dev->usb_active))
+		return -EPERM;
+
+	urb = dlfb_get_urb(dev);
+	if (!urb)
+		return -ENOMEM;
+
+	buf = (char *) urb->transfer_buffer;
+
+	/*
+	* This first section has to do with setting the base address on the
+	* controller * associated with the display. There are 2 base
+	* pointers, currently, we only * use the 16 bpp segment.
+	*/
+	wrptr = dlfb_vidreg_lock(buf);
+	wrptr = dlfb_set_color_depth(wrptr, 0x00);
+	/* set base for 16bpp segment to 0 */
+	wrptr = dlfb_set_base16bpp(wrptr, 0);
+	/* set base for 8bpp segment to end of fb */
+	wrptr = dlfb_set_base8bpp(wrptr, dev->info->fix.smem_len);
+
+	wrptr = dlfb_set_vid_cmds(wrptr, var);
+	wrptr = dlfb_enable_hvsync(wrptr, true);
+	wrptr = dlfb_vidreg_unlock(wrptr);
+
+	writesize = wrptr - buf;
+
+	retval = dlfb_submit_urb(dev, urb, writesize);
+
+	return retval;
+}
+
+static int dlfb_ops_mmap(struct fb_info *info, struct vm_area_struct *vma)
+{
+	unsigned long start = vma->vm_start;
+	unsigned long size = vma->vm_end - vma->vm_start;
+	unsigned long offset = vma->vm_pgoff << PAGE_SHIFT;
+	unsigned long page, pos;
+
+	if (offset + size > info->fix.smem_len)
+		return -EINVAL;
+
+	pos = (unsigned long)info->fix.smem_start + offset;
+
+	dl_notice("mmap() framebuffer addr:%lu size:%lu\n",
+		  pos, size);
+
+	while (size > 0) {
+		page = vmalloc_to_pfn((void *)pos);
+		if (remap_pfn_range(vma, start, page, PAGE_SIZE, PAGE_SHARED))
+			return -EAGAIN;
+
+		start += PAGE_SIZE;
+		pos += PAGE_SIZE;
+		if (size > PAGE_SIZE)
+			size -= PAGE_SIZE;
+		else
+			size = 0;
+	}
+
+	vma->vm_flags |= VM_RESERVED;	/* avoid to swap out this VMA */
+	return 0;
+}
+
+/*
+ * Trims identical data from front and back of line
+ * Sets new front buffer address and width
+ * And returns byte count of identical pixels
+ * Assumes CPU natural alignment (unsigned long)
+ * for back and front buffer ptrs and width
+ */
+static int dlfb_trim_hline(const u8 *bback, const u8 **bfront, int *width_bytes)
+{
+	int j, k;
+	const unsigned long *back = (const unsigned long *) bback;
+	const unsigned long *front = (const unsigned long *) *bfront;
+	const int width = *width_bytes / sizeof(unsigned long);
+	int identical = width;
+	int start = width;
+	int end = width;
+
+	prefetch((void *) front);
+	prefetch((void *) back);
+
+	for (j = 0; j < width; j++) {
+		if (back[j] != front[j]) {
+			start = j;
+			break;
+		}
+	}
+
+	for (k = width - 1; k > j; k--) {
+		if (back[k] != front[k]) {
+			end = k+1;
+			break;
+		}
+	}
+
+	identical = start + (width - end);
+	*bfront = (u8 *) &front[start];
+	*width_bytes = (end - start) * sizeof(unsigned long);
+
+	return identical * sizeof(unsigned long);
+}
+
+/*
+ * Render a command stream for an encoded horizontal line segment of pixels.
+ *
+ * A command buffer holds several commands.
+ * It always begins with a fresh command header
+ * (the protocol doesn't require this, but we enforce it to allow
+ * multiple buffers to be potentially encoded and sent in parallel).
+ * A single command encodes one contiguous horizontal line of pixels
+ *
+ * The function relies on the client to do all allocation, so that
+ * rendering can be done directly to output buffers (e.g. USB URBs).
+ * The function fills the supplied command buffer, providing information
+ * on where it left off, so the client may call in again with additional
+ * buffers if the line will take several buffers to complete.
+ *
+ * A single command can transmit a maximum of 256 pixels,
+ * regardless of the compression ratio (protocol design limit).
+ * To the hardware, 0 for a size byte means 256
+ *
+ * Rather than 256 pixel commands which are either rl or raw encoded,
+ * the rlx command simply assumes alternating raw and rl spans within one cmd.
+ * This has a slightly larger header overhead, but produces more even results.
+ * It also processes all data (read and write) in a single pass.
+ * Performance benchmarks of common cases show it having just slightly better
+ * compression than 256 pixel raw or rle commands, with similar CPU consumpion.
+ * But for very rl friendly data, will compress not quite as well.
+ */
+static void dlfb_compress_hline(
+	const uint16_t **pixel_start_ptr,
+	const uint16_t *const pixel_end,
+	uint32_t *device_address_ptr,
+	uint8_t **command_buffer_ptr,
+	const uint8_t *const cmd_buffer_end)
+{
+	const uint16_t *pixel = *pixel_start_ptr;
+	uint32_t dev_addr  = *device_address_ptr;
+	uint8_t *cmd = *command_buffer_ptr;
+	const int bpp = 2;
+
+	while ((pixel_end > pixel) &&
+	       (cmd_buffer_end - MIN_RLX_CMD_BYTES > cmd)) {
+		uint8_t *raw_pixels_count_byte = 0;
+		uint8_t *cmd_pixels_count_byte = 0;
+		const uint16_t *raw_pixel_start = 0;
+		const uint16_t *cmd_pixel_start, *cmd_pixel_end = 0;
+
+		prefetchw((void *) cmd); /* pull in one cache line at least */
+
+		*cmd++ = 0xAF;
+		*cmd++ = 0x6B;
+		*cmd++ = (uint8_t) ((dev_addr >> 16) & 0xFF);
+		*cmd++ = (uint8_t) ((dev_addr >> 8) & 0xFF);
+		*cmd++ = (uint8_t) ((dev_addr) & 0xFF);
+
+		cmd_pixels_count_byte = cmd++; /*  we'll know this later */
+		cmd_pixel_start = pixel;
+
+		raw_pixels_count_byte = cmd++; /*  we'll know this later */
+		raw_pixel_start = pixel;
+
+		cmd_pixel_end = pixel + min(MAX_CMD_PIXELS + 1,
+			min((int)(pixel_end - pixel),
+			    (int)(cmd_buffer_end - cmd) / bpp));
+
+		prefetch_range((void *) pixel, (cmd_pixel_end - pixel) * bpp);
+
+		while (pixel < cmd_pixel_end) {
+			const uint16_t * const repeating_pixel = pixel;
+
+			*(uint16_t *)cmd = cpu_to_be16p(pixel);
+			cmd += 2;
+			pixel++;
+
+			if (unlikely((pixel < cmd_pixel_end) &&
+				     (*pixel == *repeating_pixel))) {
+				/* go back and fill in raw pixel count */
+				*raw_pixels_count_byte = ((repeating_pixel -
+						raw_pixel_start) + 1) & 0xFF;
+
+				while ((pixel < cmd_pixel_end)
+				       && (*pixel == *repeating_pixel)) {
+					pixel++;
+				}
+
+				/* immediately after raw data is repeat byte */
+				*cmd++ = ((pixel - repeating_pixel) - 1) & 0xFF;
+
+				/* Then start another raw pixel span */
+				raw_pixel_start = pixel;
+				raw_pixels_count_byte = cmd++;
+			}
+		}
+
+		if (pixel > raw_pixel_start) {
+			/* finalize last RAW span */
+			*raw_pixels_count_byte = (pixel-raw_pixel_start) & 0xFF;
+		}
+
+		*cmd_pixels_count_byte = (pixel - cmd_pixel_start) & 0xFF;
+		dev_addr += (pixel - cmd_pixel_start) * bpp;
+	}
+
+	if (cmd_buffer_end <= MIN_RLX_CMD_BYTES + cmd) {
+		/* Fill leftover bytes with no-ops */
+		if (cmd_buffer_end > cmd)
+			memset(cmd, 0xAF, cmd_buffer_end - cmd);
+		cmd = (uint8_t *) cmd_buffer_end;
+	}
+
+	*command_buffer_ptr = cmd;
+	*pixel_start_ptr = pixel;
+	*device_address_ptr = dev_addr;
+
+	return;
+}
+
+/*
+ * There are 3 copies of every pixel: The front buffer that the fbdev
+ * client renders to, the actual framebuffer across the USB bus in hardware
+ * (that we can only write to, slowly, and can never read), and (optionally)
+ * our shadow copy that tracks what's been sent to that hardware buffer.
+ */
+static int dlfb_render_hline(struct dlfb_data *dev, struct urb **urb_ptr,
+			      const char *front, char **urb_buf_ptr,
+			      u32 byte_offset, u32 byte_width,
+			      int *ident_ptr, int *sent_ptr)
+{
+	const u8 *line_start, *line_end, *next_pixel;
+	u32 dev_addr = dev->base16 + byte_offset;
+	struct urb *urb = *urb_ptr;
+	u8 *cmd = *urb_buf_ptr;
+	u8 *cmd_end = (u8 *) urb->transfer_buffer + urb->transfer_buffer_length;
+
+	line_start = (u8 *) (front + byte_offset);
+	next_pixel = line_start;
+	line_end = next_pixel + byte_width;
+
+	if (dev->backing_buffer) {
+		int offset;
+		const u8 *back_start = (u8 *) (dev->backing_buffer
+						+ byte_offset);
+
+		*ident_ptr += dlfb_trim_hline(back_start, &next_pixel,
+			&byte_width);
+
+		offset = next_pixel - line_start;
+		line_end = next_pixel + byte_width;
+		dev_addr += offset;
+		back_start += offset;
+		line_start += offset;
+
+		memcpy((char *)back_start, (char *) line_start,
+		       byte_width);
+	}
+
+	while (next_pixel < line_end) {
+
+		dlfb_compress_hline((const uint16_t **) &next_pixel,
+			     (const uint16_t *) line_end, &dev_addr,
+			(u8 **) &cmd, (u8 *) cmd_end);
+
+		if (cmd >= cmd_end) {
+			int len = cmd - (u8 *) urb->transfer_buffer;
+			if (dlfb_submit_urb(dev, urb, len))
+				return 1; /* lost pixels is set */
+			*sent_ptr += len;
+			urb = dlfb_get_urb(dev);
+			if (!urb)
+				return 1; /* lost_pixels is set */
+			*urb_ptr = urb;
+			cmd = urb->transfer_buffer;
+			cmd_end = &cmd[urb->transfer_buffer_length];
+		}
+	}
+
+	*urb_buf_ptr = cmd;
+
+	return 0;
+}
+
+int dlfb_handle_damage(struct dlfb_data *dev, int x, int y,
+	       int width, int height, char *data)
+{
+	int i, ret;
+	char *cmd;
+	cycles_t start_cycles, end_cycles;
+	int bytes_sent = 0;
+	int bytes_identical = 0;
+	struct urb *urb;
+	int aligned_x;
+
+	start_cycles = get_cycles();
+
+	aligned_x = DL_ALIGN_DOWN(x, sizeof(unsigned long));
+	width = DL_ALIGN_UP(width + (x-aligned_x), sizeof(unsigned long));
+	x = aligned_x;
+
+	if ((width <= 0) ||
+	    (x + width > dev->info->var.xres) ||
+	    (y + height > dev->info->var.yres))
+		return -EINVAL;
+
+	if (!atomic_read(&dev->usb_active))
+		return 0;
+
+	urb = dlfb_get_urb(dev);
+	if (!urb)
+		return 0;
+	cmd = urb->transfer_buffer;
+
+	for (i = y; i < y + height ; i++) {
+		const int line_offset = dev->info->fix.line_length * i;
+		const int byte_offset = line_offset + (x * BPP);
+
+		if (dlfb_render_hline(dev, &urb,
+				      (char *) dev->info->fix.smem_start,
+				      &cmd, byte_offset, width * BPP,
+				      &bytes_identical, &bytes_sent))
+			goto error;
+	}
+
+	if (cmd > (char *) urb->transfer_buffer) {
+		/* Send partial buffer remaining before exiting */
+		int len = cmd - (char *) urb->transfer_buffer;
+		ret = dlfb_submit_urb(dev, urb, len);
+		bytes_sent += len;
+	} else
+		dlfb_urb_completion(urb);
+
+error:
+	atomic_add(bytes_sent, &dev->bytes_sent);
+	atomic_add(bytes_identical, &dev->bytes_identical);
+	atomic_add(width*height*2, &dev->bytes_rendered);
+	end_cycles = get_cycles();
+	atomic_add(((unsigned int) ((end_cycles - start_cycles)
+		    >> 10)), /* Kcycles */
+		   &dev->cpu_kcycles_used);
+
+	return 0;
+}
+
+static ssize_t dlfb_ops_read(struct fb_info *info, char __user *buf,
+			 size_t count, loff_t *ppos)
+{
+	ssize_t result = -ENOSYS;
+
+#if defined CONFIG_FB_SYS_FOPS || defined CONFIG_FB_SYS_FOPS_MODULE
+	result = fb_sys_read(info, buf, count, ppos);
+#endif
+
+	return result;
+}
+
+/*
+ * Path triggered by usermode clients who write to filesystem
+ * e.g. cat filename > /dev/fb1
+ * Not used by X Windows or text-mode console. But useful for testing.
+ * Slow because of extra copy and we must assume all pixels dirty.
+ */
+static ssize_t dlfb_ops_write(struct fb_info *info, const char __user *buf,
+			  size_t count, loff_t *ppos)
+{
+	ssize_t result = -ENOSYS;
+	struct dlfb_data *dev = info->par;
+	u32 offset = (u32) *ppos;
+
+#if defined CONFIG_FB_SYS_FOPS || defined CONFIG_FB_SYS_FOPS_MODULE
+
+	result = fb_sys_write(info, buf, count, ppos);
+
+	if (result > 0) {
+		int start = max((int)(offset / info->fix.line_length) - 1, 0);
+		int lines = min((u32)((result / info->fix.line_length) + 1),
+				(u32)info->var.yres);
+
+		dlfb_handle_damage(dev, 0, start, info->var.xres,
+			lines, info->screen_base);
+	}
+#endif
+
+	return result;
+}
+
+/* hardware has native COPY command (see libdlo), but not worth it for fbcon */
+static void dlfb_ops_copyarea(struct fb_info *info,
+				const struct fb_copyarea *area)
+{
+
+	struct dlfb_data *dev = info->par;
+
+#if defined CONFIG_FB_SYS_COPYAREA || defined CONFIG_FB_SYS_COPYAREA_MODULE
+
+	sys_copyarea(info, area);
+
+	dlfb_handle_damage(dev, area->dx, area->dy,
+			area->width, area->height, info->screen_base);
+#endif
+
+}
+
+static void dlfb_ops_imageblit(struct fb_info *info,
+				const struct fb_image *image)
+{
+	struct dlfb_data *dev = info->par;
+
+#if defined CONFIG_FB_SYS_IMAGEBLIT || defined CONFIG_FB_SYS_IMAGEBLIT_MODULE
+
+	sys_imageblit(info, image);
+
+	dlfb_handle_damage(dev, image->dx, image->dy,
+			image->width, image->height, info->screen_base);
+
+#endif
+
+}
+
+static void dlfb_ops_fillrect(struct fb_info *info,
+			  const struct fb_fillrect *rect)
+{
+	struct dlfb_data *dev = info->par;
+
+#if defined CONFIG_FB_SYS_FILLRECT || defined CONFIG_FB_SYS_FILLRECT_MODULE
+
+	sys_fillrect(info, rect);
+
+	dlfb_handle_damage(dev, rect->dx, rect->dy, rect->width,
+			      rect->height, info->screen_base);
+#endif
+
+}
+
+#ifdef CONFIG_FB_DEFERRED_IO
+/*
+ * NOTE: fb_defio.c is holding info->fbdefio.mutex
+ *   Touching ANY framebuffer memory that triggers a page fault
+ *   in fb_defio will cause a deadlock, when it also tries to
+ *   grab the same mutex.
+ */
+static void dlfb_dpy_deferred_io(struct fb_info *info,
+				struct list_head *pagelist)
+{
+	struct page *cur;
+	struct fb_deferred_io *fbdefio = info->fbdefio;
+	struct dlfb_data *dev = info->par;
+	struct urb *urb;
+	char *cmd;
+	cycles_t start_cycles, end_cycles;
+	int bytes_sent = 0;
+	int bytes_identical = 0;
+	int bytes_rendered = 0;
+
+	if (!fb_defio)
+		return;
+
+	if (!atomic_read(&dev->usb_active))
+		return;
+
+	start_cycles = get_cycles();
+
+	urb = dlfb_get_urb(dev);
+	if (!urb)
+		return;
+
+	cmd = urb->transfer_buffer;
+
+	/* walk the written page list and render each to device */
+	list_for_each_entry(cur, &fbdefio->pagelist, lru) {
+
+		if (dlfb_render_hline(dev, &urb, (char *) info->fix.smem_start,
+				  &cmd, cur->index << PAGE_SHIFT,
+				  PAGE_SIZE, &bytes_identical, &bytes_sent))
+			goto error;
+		bytes_rendered += PAGE_SIZE;
+	}
+
+	if (cmd > (char *) urb->transfer_buffer) {
+		/* Send partial buffer remaining before exiting */
+		int len = cmd - (char *) urb->transfer_buffer;
+		dlfb_submit_urb(dev, urb, len);
+		bytes_sent += len;
+	} else
+		dlfb_urb_completion(urb);
+
+error:
+	atomic_add(bytes_sent, &dev->bytes_sent);
+	atomic_add(bytes_identical, &dev->bytes_identical);
+	atomic_add(bytes_rendered, &dev->bytes_rendered);
+	end_cycles = get_cycles();
+	atomic_add(((unsigned int) ((end_cycles - start_cycles)
+		    >> 10)), /* Kcycles */
+		   &dev->cpu_kcycles_used);
+}
+
+#endif
+
+static int dlfb_get_edid(struct dlfb_data *dev, char *edid, int len)
+{
+	int i;
+	int ret;
+	char *rbuf;
+
+	rbuf = kmalloc(2, GFP_KERNEL);
+	if (!rbuf)
+		return 0;
+
+	for (i = 0; i < len; i++) {
+		ret = usb_control_msg(dev->udev,
+				    usb_rcvctrlpipe(dev->udev, 0), (0x02),
+				    (0x80 | (0x02 << 5)), i << 8, 0xA1, rbuf, 2,
+				    HZ);
+		if (ret < 1) {
+			dl_err("Read EDID byte %d failed err %x\n", i, ret);
+			i--;
+			break;
+		}
+		edid[i] = rbuf[1];
+	}
+
+	kfree(rbuf);
+
+	return i;
+}
+
+static int dlfb_ops_ioctl(struct fb_info *info, unsigned int cmd,
+				unsigned long arg)
+{
+
+	struct dlfb_data *dev = info->par;
+	struct dloarea *area = NULL;
+
+	if (!atomic_read(&dev->usb_active))
+		return 0;
+
+	/* TODO: Update X server to get this from sysfs instead */
+	if (cmd == DLFB_IOCTL_RETURN_EDID) {
+		char *edid = (char *)arg;
+		if (copy_to_user(edid, dev->edid, dev->edid_size))
+			return -EFAULT;
+		return 0;
+	}
+
+	/* TODO: Help propose a standard fb.h ioctl to report mmap damage */
+	if (cmd == DLFB_IOCTL_REPORT_DAMAGE) {
+
+		/*
+		 * If we have a damage-aware client, turn fb_defio "off"
+		 * To avoid perf imact of unecessary page fault handling.
+		 * Done by resetting the delay for this fb_info to a very
+		 * long period. Pages will become writable and stay that way.
+		 * Reset to normal value when all clients have closed this fb.
+		 */
+		if (info->fbdefio)
+			info->fbdefio->delay = DL_DEFIO_WRITE_DISABLE;
+
+		area = (struct dloarea *)arg;
+
+		if (area->x < 0)
+			area->x = 0;
+
+		if (area->x > info->var.xres)
+			area->x = info->var.xres;
+
+		if (area->y < 0)
+			area->y = 0;
+
+		if (area->y > info->var.yres)
+			area->y = info->var.yres;
+
+		dlfb_handle_damage(dev, area->x, area->y, area->w, area->h,
+			   info->screen_base);
+	}
+
+	return 0;
+}
+
+/* taken from vesafb */
+static int
+dlfb_ops_setcolreg(unsigned regno, unsigned red, unsigned green,
+	       unsigned blue, unsigned transp, struct fb_info *info)
+{
+	int err = 0;
+
+	if (regno >= info->cmap.len)
+		return 1;
+
+	if (regno < 16) {
+		if (info->var.red.offset == 10) {
+			/* 1:5:5:5 */
+			((u32 *) (info->pseudo_palette))[regno] =
+			    ((red & 0xf800) >> 1) |
+			    ((green & 0xf800) >> 6) | ((blue & 0xf800) >> 11);
+		} else {
+			/* 0:5:6:5 */
+			((u32 *) (info->pseudo_palette))[regno] =
+			    ((red & 0xf800)) |
+			    ((green & 0xfc00) >> 5) | ((blue & 0xf800) >> 11);
+		}
+	}
+
+	return err;
+}
+
+/*
+ * It's common for several clients to have framebuffer open simultaneously.
+ * e.g. both fbcon and X. Makes things interesting.
+ * Assumes caller is holding info->lock (for open and release at least)
+ */
+static int dlfb_ops_open(struct fb_info *info, int user)
+{
+	struct dlfb_data *dev = info->par;
+
+	/*
+	 * fbcon aggressively connects to first framebuffer it finds,
+	 * preventing other clients (X) from working properly. Usually
+	 * not what the user wants. Fail by default with option to enable.
+	 */
+	if ((user == 0) & (!console))
+		return -EBUSY;
+
+	/* If the USB device is gone, we don't accept new opens */
+	if (dev->virtualized)
+		return -ENODEV;
+
+	dev->fb_count++;
+
+	kref_get(&dev->kref);
+
+#ifdef CONFIG_FB_DEFERRED_IO
+	if (fb_defio && (info->fbdefio == NULL)) {
+		/* enable defio at last moment if not disabled by client */
+
+		struct fb_deferred_io *fbdefio;
+
+		fbdefio = kmalloc(sizeof(struct fb_deferred_io), GFP_KERNEL);
+
+		if (fbdefio) {
+			fbdefio->delay = DL_DEFIO_WRITE_DELAY;
+			fbdefio->deferred_io = dlfb_dpy_deferred_io;
+		}
+
+		info->fbdefio = fbdefio;
+		fb_deferred_io_init(info);
+	}
+#endif
+
+	dl_notice("open /dev/fb%d user=%d fb_info=%p count=%d\n",
+	    info->node, user, info, dev->fb_count);
+
+	return 0;
+}
+
+/*
+ * Called when all client interfaces to start transactions have been disabled,
+ * and all references to our device instance (dlfb_data) are released.
+ * Every transaction must have a reference, so we know are fully spun down
+ */
+static void dlfb_free(struct kref *kref)
+{
+	struct dlfb_data *dev = container_of(kref, struct dlfb_data, kref);
+
+	/* this function will wait for all in-flight urbs to complete */
+	if (dev->urbs.count > 0)
+		dlfb_free_urb_list(dev);
+
+	if (dev->backing_buffer)
+		vfree(dev->backing_buffer);
+
+	kfree(dev->edid);
+
+	dl_warn("freeing dlfb_data %p\n", dev);
+
+	kfree(dev);
+}
+
+static void dlfb_release_urb_work(struct work_struct *work)
+{
+	struct urb_node *unode = container_of(work, struct urb_node,
+					      release_urb_work.work);
+
+	up(&unode->dev->urbs.limit_sem);
+}
+
+static void dlfb_free_framebuffer_work(struct work_struct *work)
+{
+	struct dlfb_data *dev = container_of(work, struct dlfb_data,
+					     free_framebuffer_work.work);
+	struct fb_info *info = dev->info;
+	int node = info->node;
+
+	unregister_framebuffer(info);
+
+	if (info->cmap.len != 0)
+		fb_dealloc_cmap(&info->cmap);
+	if (info->monspecs.modedb)
+		fb_destroy_modedb(info->monspecs.modedb);
+	if (info->screen_base)
+		vfree(info->screen_base);
+
+	fb_destroy_modelist(&info->modelist);
+
+	dev->info = 0;
+
+	/* Assume info structure is freed after this point */
+	framebuffer_release(info);
+
+	dl_warn("fb_info for /dev/fb%d has been freed\n", node);
+
+	/* ref taken in probe() as part of registering framebfufer */
+	kref_put(&dev->kref, dlfb_free);
+}
+
+/*
+ * Assumes caller is holding info->lock mutex (for open and release at least)
+ */
+static int dlfb_ops_release(struct fb_info *info, int user)
+{
+	struct dlfb_data *dev = info->par;
+
+	dev->fb_count--;
+
+	/* We can't free fb_info here - fbmem will touch it when we return */
+	if (dev->virtualized && (dev->fb_count == 0))
+		schedule_delayed_work(&dev->free_framebuffer_work, HZ);
+
+#ifdef CONFIG_FB_DEFERRED_IO
+	if ((dev->fb_count == 0) && (info->fbdefio)) {
+		fb_deferred_io_cleanup(info);
+		kfree(info->fbdefio);
+		info->fbdefio = NULL;
+		info->fbops->fb_mmap = dlfb_ops_mmap;
+	}
+#endif
+
+	dl_warn("released /dev/fb%d user=%d count=%d\n",
+		  info->node, user, dev->fb_count);
+
+	kref_put(&dev->kref, dlfb_free);
+
+	return 0;
+}
+
+/*
+ * Check whether a video mode is supported by the DisplayLink chip
+ * We start from monitor's modes, so don't need to filter that here
+ */
+static int dlfb_is_valid_mode(struct fb_videomode *mode,
+		struct fb_info *info)
+{
+	struct dlfb_data *dev = info->par;
+
+	if (mode->xres * mode->yres > dev->sku_pixel_limit) {
+		dl_warn("%dx%d beyond chip capabilities\n",
+		       mode->xres, mode->yres);
+		return 0;
+	}
+
+	dl_info("%dx%d valid mode\n", mode->xres, mode->yres);
+
+	return 1;
+}
+
+static void dlfb_var_color_format(struct fb_var_screeninfo *var)
+{
+	const struct fb_bitfield red = { 11, 5, 0 };
+	const struct fb_bitfield green = { 5, 6, 0 };
+	const struct fb_bitfield blue = { 0, 5, 0 };
+
+	var->bits_per_pixel = 16;
+	var->red = red;
+	var->green = green;
+	var->blue = blue;
+}
+
+static int dlfb_ops_check_var(struct fb_var_screeninfo *var,
+				struct fb_info *info)
+{
+	struct fb_videomode mode;
+
+	/* TODO: support dynamically changing framebuffer size */
+	if ((var->xres * var->yres * 2) > info->fix.smem_len)
+		return -EINVAL;
+
+	/* set device-specific elements of var unrelated to mode */
+	dlfb_var_color_format(var);
+
+	fb_var_to_videomode(&mode, var);
+
+	if (!dlfb_is_valid_mode(&mode, info))
+		return -EINVAL;
+
+	return 0;
+}
+
+static int dlfb_ops_set_par(struct fb_info *info)
+{
+	struct dlfb_data *dev = info->par;
+	int result;
+	u16 *pix_framebuffer;
+	int i;
+
+	dl_notice("set_par mode %dx%d\n", info->var.xres, info->var.yres);
+
+	result = dlfb_set_video_mode(dev, &info->var);
+
+	if ((result == 0) && (dev->fb_count == 0)) {
+
+		/* paint greenscreen */
+
+		pix_framebuffer = (u16 *) info->screen_base;
+		for (i = 0; i < info->fix.smem_len / 2; i++)
+			pix_framebuffer[i] = 0x37e6;
+
+		dlfb_handle_damage(dev, 0, 0, info->var.xres, info->var.yres,
+				   info->screen_base);
+	}
+
+	return result;
+}
+
+/*
+ * In order to come back from full DPMS off, we need to set the mode again
+ */
+static int dlfb_ops_blank(int blank_mode, struct fb_info *info)
+{
+	struct dlfb_data *dev = info->par;
+
+	if (blank_mode != FB_BLANK_UNBLANK) {
+		char *bufptr;
+		struct urb *urb;
+
+		urb = dlfb_get_urb(dev);
+		if (!urb)
+			return 0;
+
+		bufptr = (char *) urb->transfer_buffer;
+		bufptr = dlfb_vidreg_lock(bufptr);
+		bufptr = dlfb_enable_hvsync(bufptr, false);
+		bufptr = dlfb_vidreg_unlock(bufptr);
+
+		dlfb_submit_urb(dev, urb, bufptr -
+				(char *) urb->transfer_buffer);
+	} else {
+		dlfb_set_video_mode(dev, &info->var);
+	}
+
+	return 0;
+}
+
+static struct fb_ops dlfb_ops = {
+	.owner = THIS_MODULE,
+	.fb_read = dlfb_ops_read,
+	.fb_write = dlfb_ops_write,
+	.fb_setcolreg = dlfb_ops_setcolreg,
+	.fb_fillrect = dlfb_ops_fillrect,
+	.fb_copyarea = dlfb_ops_copyarea,
+	.fb_imageblit = dlfb_ops_imageblit,
+	.fb_mmap = dlfb_ops_mmap,
+	.fb_ioctl = dlfb_ops_ioctl,
+	.fb_open = dlfb_ops_open,
+	.fb_release = dlfb_ops_release,
+	.fb_blank = dlfb_ops_blank,
+	.fb_check_var = dlfb_ops_check_var,
+	.fb_set_par = dlfb_ops_set_par,
+};
+
+
+/*
+ * Assumes &info->lock held by caller
+ * Assumes no active clients have framebuffer open
+ */
+static int dlfb_realloc_framebuffer(struct dlfb_data *dev, struct fb_info *info)
+{
+	int retval = -ENOMEM;
+	int old_len = info->fix.smem_len;
+	int new_len;
+	unsigned char *old_fb = info->screen_base;
+	unsigned char *new_fb;
+	unsigned char *new_back;
+
+	dl_warn("Reallocating framebuffer. Addresses will change!\n");
+
+	new_len = info->fix.line_length * info->var.yres;
+
+	if (PAGE_ALIGN(new_len) > old_len) {
+		/*
+		 * Alloc system memory for virtual framebuffer
+		 */
+		new_fb = vmalloc(new_len);
+		if (!new_fb