m68k: Atari fb revival

Update the atari fb to 2.6 by Michael Schmitz,
Reformatting and rewrite of bit plane functions by Roman Zippel,
A few more fixes by Geert Uytterhoeven.
Signed-off-by: Michael Schmitz <schmitz@debian.org>
Signed-off-by: Roman Zippel <zippel@linux-m68k.org>
Signed-off-by: Geert Uytterhoeven <geert@linux-m68k.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

drivers/video/Kconfig

@@ -389,7 +389,10 @@ config FB_ARC
 
 config FB_ATARI
 	bool "Atari native chipset support"
-	depends on (FB = y) && ATARI && BROKEN
+	depends on (FB = y) && ATARI
+	select FB_CFB_FILLRECT
+	select FB_CFB_COPYAREA
+	select FB_CFB_IMAGEBLIT
 	help
 	  This is the frame buffer device driver for the builtin graphics
 	  chipset found in Ataris.

drivers/video/Makefile

@@ -63,7 +63,8 @@ obj-$(CONFIG_FB_TCX)              += tcx.o sbuslib.o
 obj-$(CONFIG_FB_LEO)              += leo.o sbuslib.o
 obj-$(CONFIG_FB_SGIVW)            += sgivwfb.o
 obj-$(CONFIG_FB_ACORN)            += acornfb.o
-obj-$(CONFIG_FB_ATARI)            += atafb.o
+obj-$(CONFIG_FB_ATARI)            += atafb.o c2p.o atafb_mfb.o \
+                                     atafb_iplan2p2.o atafb_iplan2p4.o atafb_iplan2p8.o
 obj-$(CONFIG_FB_MAC)              += macfb.o
 obj-$(CONFIG_FB_HGA)              += hgafb.o
 obj-$(CONFIG_FB_IGA)              += igafb.o

drivers/video/atafb.h

+#ifndef _VIDEO_ATAFB_H
+#define _VIDEO_ATAFB_H
+
+void atafb_mfb_copyarea(struct fb_info *info, u_long next_line, int sy, int sx, int dy,
+			int dx, int height, int width);
+void atafb_mfb_fillrect(struct fb_info *info, u_long next_line, u32 color,
+			int sy, int sx, int height, int width);
+void atafb_mfb_linefill(struct fb_info *info, u_long next_line,
+			int dy, int dx, u32 width,
+			const u8 *data, u32 bgcolor, u32 fgcolor);
+
+void atafb_iplan2p2_copyarea(struct fb_info *info, u_long next_line, int sy, int sx, int dy,
+			     int dx, int height, int width);
+void atafb_iplan2p2_fillrect(struct fb_info *info, u_long next_line, u32 color,
+			     int sy, int sx, int height, int width);
+void atafb_iplan2p2_linefill(struct fb_info *info, u_long next_line,
+			     int dy, int dx, u32 width,
+			     const u8 *data, u32 bgcolor, u32 fgcolor);
+
+void atafb_iplan2p4_copyarea(struct fb_info *info, u_long next_line, int sy, int sx, int dy,
+			     int dx, int height, int width);
+void atafb_iplan2p4_fillrect(struct fb_info *info, u_long next_line, u32 color,
+			     int sy, int sx, int height, int width);
+void atafb_iplan2p4_linefill(struct fb_info *info, u_long next_line,
+			     int dy, int dx, u32 width,
+			     const u8 *data, u32 bgcolor, u32 fgcolor);
+
+void atafb_iplan2p8_copyarea(struct fb_info *info, u_long next_line, int sy, int sx, int dy,
+			     int dx, int height, int width);
+void atafb_iplan2p8_fillrect(struct fb_info *info, u_long next_line, u32 color,
+			     int sy, int sx, int height, int width);
+void atafb_iplan2p8_linefill(struct fb_info *info, u_long next_line,
+			     int dy, int dx, u32 width,
+			     const u8 *data, u32 bgcolor, u32 fgcolor);
+
+#endif /* _VIDEO_ATAFB_H */

drivers/video/atafb_iplan2p2.c

+/*
+ *  linux/drivers/video/iplan2p2.c -- Low level frame buffer operations for
+ *				      interleaved bitplanes à la Atari (2
+ *				      planes, 2 bytes interleave)
+ *
+ *	Created 5 Apr 1997 by Geert Uytterhoeven
+ *
+ *  This file is subject to the terms and conditions of the GNU General Public
+ *  License.  See the file COPYING in the main directory of this archive for
+ *  more details.
+ */
+
+#include <linux/module.h>
+#include <linux/string.h>
+#include <linux/fb.h>
+
+#include <asm/setup.h>
+
+#include "atafb.h"
+
+#define BPL	2
+#include "atafb_utils.h"
+
+void atafb_iplan2p2_copyarea(struct fb_info *info, u_long next_line,
+			     int sy, int sx, int dy, int dx,
+			     int height, int width)
+{
+	/*  bmove() has to distinguish two major cases: If both, source and
+	 *  destination, start at even addresses or both are at odd
+	 *  addresses, just the first odd and last even column (if present)
+	 *  require special treatment (memmove_col()). The rest between
+	 *  then can be copied by normal operations, because all adjacent
+	 *  bytes are affected and are to be stored in the same order.
+	 *    The pathological case is when the move should go from an odd
+	 *  address to an even or vice versa. Since the bytes in the plane
+	 *  words must be assembled in new order, it seems wisest to make
+	 *  all movements by memmove_col().
+	 */
+
+	u8 *src, *dst;
+	u32 *s, *d;
+	int w, l , i, j;
+	u_int colsize;
+	u_int upwards = (dy < sy) || (dy == sy && dx < sx);
+
+	colsize = height;
+	if (!((sx ^ dx) & 15)) {
+		/* odd->odd or even->even */
+
+		if (upwards) {
+			src = (u8 *)info->screen_base + sy * next_line + (sx & ~15) / (8 / BPL);
+			dst = (u8 *)info->screen_base + dy * next_line + (dx & ~15) / (8 / BPL);
+			if (sx & 15) {
+				memmove32_col(dst, src, 0xff00ff, height, next_line - BPL * 2);
+				src += BPL * 2;
+				dst += BPL * 2;
+				width -= 8;
+			}
+			w = width >> 4;
+			if (w) {
+				s = (u32 *)src;
+				d = (u32 *)dst;
+				w *= BPL / 2;
+				l = next_line - w * 4;
+				for (j = height; j > 0; j--) {
+					for (i = w; i > 0; i--)
+						*d++ = *s++;
+					s = (u32 *)((u8 *)s + l);
+					d = (u32 *)((u8 *)d + l);
+				}
+			}
+			if (width & 15)
+				memmove32_col(dst + width / (8 / BPL), src + width / (8 / BPL),
+					      0xff00ff00, height, next_line - BPL * 2);
+		} else {
+			src = (u8 *)info->screen_base + (sy - 1) * next_line + ((sx + width + 8) & ~15) / (8 / BPL);
+			dst = (u8 *)info->screen_base + (dy - 1) * next_line + ((dx + width + 8) & ~15) / (8 / BPL);
+
+			if ((sx + width) & 15) {
+				src -= BPL * 2;
+				dst -= BPL * 2;
+				memmove32_col(dst, src, 0xff00ff00, colsize, -next_line - BPL * 2);
+				width -= 8;
+			}
+			w = width >> 4;
+			if (w) {
+				s = (u32 *)src;
+				d = (u32 *)dst;
+				w *= BPL / 2;
+				l = next_line - w * 4;
+				for (j = height; j > 0; j--) {
+					for (i = w; i > 0; i--)
+						*--d = *--s;
+					s = (u32 *)((u8 *)s - l);
+					d = (u32 *)((u8 *)d - l);
+				}
+			}
+			if (sx & 15)
+				memmove32_col(dst - (width - 16) / (8 / BPL),
+					      src - (width - 16) / (8 / BPL),
+					      0xff00ff, colsize, -next_line - BPL * 2);
+		}
+	} else {
+		/* odd->even or even->odd */
+		if (upwards) {
+			u32 *src32, *dst32;
+			u32 pval[4], v, v1, mask;
+			int i, j, w, f;
+
+			src = (u8 *)info->screen_base + sy * next_line + (sx & ~15) / (8 / BPL);
+			dst = (u8 *)info->screen_base + dy * next_line + (dx & ~15) / (8 / BPL);
+
+			mask = 0xff00ff00;
+			f = 0;
+			w = width;
+			if (sx & 15) {
+				f = 1;
+				w += 8;
+			}
+			if ((sx + width) & 15)
+				f |= 2;
+			w >>= 4;
+			for (i = height; i; i--) {
+				src32 = (u32 *)src;
+				dst32 = (u32 *)dst;
+
+				if (f & 1) {
+					pval[0] = (*src32++ << 8) & mask;
+				} else {
+					pval[0] = dst32[0] & mask;
+				}
+
+				for (j = w; j > 0; j--) {
+					v = *src32++;
+					v1 = v & mask;
+					*dst32++ = pval[0] | (v1 >> 8);
+					pval[0] = (v ^ v1) << 8;
+				}
+
+				if (f & 2) {
+					dst32[0] = (dst32[0] & mask) | pval[0];
+				}
+
+				src += next_line;
+				dst += next_line;
+			}
+		} else {
+			u32 *src32, *dst32;
+			u32 pval[4], v, v1, mask;
+			int i, j, w, f;
+
+			src = (u8 *)info->screen_base + (sy - 1) * next_line + ((sx + width + 8) & ~15) / (8 / BPL);
+			dst = (u8 *)info->screen_base + (dy - 1) * next_line + ((dx + width + 8) & ~15) / (8 / BPL);
+
+			mask = 0xff00ff;
+			f = 0;
+			w = width;
+			if ((dx + width) & 15)
+				f = 1;
+			if (sx & 15) {
+				f |= 2;
+				w += 8;
+			}
+			w >>= 4;
+			for (i = height; i; i--) {
+				src32 = (u32 *)src;
+				dst32 = (u32 *)dst;
+
+				if (f & 1) {
+					pval[0] = dst32[-1] & mask;
+				} else {
+					pval[0] = (*--src32 >> 8) & mask;
+				}
+
+				for (j = w; j > 0; j--) {
+					v = *--src32;
+					v1 = v & mask;
+					*--dst32 = pval[0] | (v1 << 8);
+					pval[0] = (v ^ v1) >> 8;
+				}
+
+				if (!(f & 2)) {
+					dst32[-1] = (dst32[-1] & mask) | pval[0];
+				}
+
+				src -= next_line;
+				dst -= next_line;
+			}
+		}
+	}
+}
+
+void atafb_iplan2p2_fillrect(struct fb_info *info, u_long next_line, u32 color,
+                             int sy, int sx, int height, int width)
+{
+	u32 *dest;
+	int rows, i;
+	u32 cval[4];
+
+	dest = (u32 *)(info->screen_base + sy * next_line + (sx & ~15) / (8 / BPL));
+	if (sx & 15) {
+		u8 *dest8 = (u8 *)dest + 1;
+
+		expand8_col2mask(color, cval);
+
+		for (i = height; i; i--) {
+			fill8_col(dest8, cval);
+			dest8 += next_line;
+		}
+		dest += BPL / 2;
+		width -= 8;
+	}
+
+	expand16_col2mask(color, cval);
+	rows = width >> 4;
+	if (rows) {
+		u32 *d = dest;
+		u32 off = next_line - rows * BPL * 2;
+		for (i = height; i; i--) {
+			d = fill16_col(d, rows, cval);
+			d = (u32 *)((long)d + off);
+		}
+		dest += rows * BPL / 2;
+		width &= 15;
+	}
+
+	if (width) {
+		u8 *dest8 = (u8 *)dest;
+
+		expand8_col2mask(color, cval);
+
+		for (i = height; i; i--) {
+			fill8_col(dest8, cval);
+			dest8 += next_line;
+		}
+	}
+}
+
+void atafb_iplan2p2_linefill(struct fb_info *info, u_long next_line,
+                             int dy, int dx, u32 width,
+                             const u8 *data, u32 bgcolor, u32 fgcolor)
+{
+	u32 *dest;
+	const u16 *data16;
+	int rows;
+	u32 fgm[4], bgm[4], m;
+
+	dest = (u32 *)(info->screen_base + dy * next_line + (dx & ~15) / (8 / BPL));
+	if (dx & 15) {
+		fill8_2col((u8 *)dest + 1, fgcolor, bgcolor, *data++);
+		dest += BPL / 2;
+		width -= 8;
+	}
+
+	if (width >= 16) {
+		data16 = (const u16 *)data;
+		expand16_2col2mask(fgcolor, bgcolor, fgm, bgm);
+
+		for (rows = width / 16; rows; rows--) {
+			u16 d = *data16++;
+			m = d | ((u32)d << 16);
+			*dest++ = (m & fgm[0]) ^ bgm[0];
+		}
+
+		data = (const u8 *)data16;
+		width &= 15;
+	}
+
+	if (width)
+		fill8_2col((u8 *)dest, fgcolor, bgcolor, *data);
+}
+
+#ifdef MODULE
+MODULE_LICENSE("GPL");
+
+int init_module(void)
+{
+	return 0;
+}
+
+void cleanup_module(void)
+{
+}
+#endif /* MODULE */
+
+
+    /*
+     *  Visible symbols for modules
+     */
+
+EXPORT_SYMBOL(atafb_iplan2p2_copyarea);
+EXPORT_SYMBOL(atafb_iplan2p2_fillrect);
+EXPORT_SYMBOL(atafb_iplan2p2_linefill);

drivers/video/atafb_iplan2p4.c

+/*
+ *  linux/drivers/video/iplan2p4.c -- Low level frame buffer operations for
+ *				      interleaved bitplanes à la Atari (4
+ *				      planes, 2 bytes interleave)
+ *
+ *	Created 5 Apr 1997 by Geert Uytterhoeven
+ *
+ *  This file is subject to the terms and conditions of the GNU General Public
+ *  License.  See the file COPYING in the main directory of this archive for
+ *  more details.
+ */
+
+#include <linux/module.h>
+#include <linux/string.h>
+#include <linux/fb.h>
+
+#include <asm/setup.h>
+
+#include "atafb.h"
+
+#define BPL	4
+#include "atafb_utils.h"
+
+void atafb_iplan2p4_copyarea(struct fb_info *info, u_long next_line,
+			     int sy, int sx, int dy, int dx,
+			     int height, int width)
+{
+	/*  bmove() has to distinguish two major cases: If both, source and
+	 *  destination, start at even addresses or both are at odd
+	 *  addresses, just the first odd and last even column (if present)
+	 *  require special treatment (memmove_col()). The rest between
+	 *  then can be copied by normal operations, because all adjacent
+	 *  bytes are affected and are to be stored in the same order.
+	 *    The pathological case is when the move should go from an odd
+	 *  address to an even or vice versa. Since the bytes in the plane
+	 *  words must be assembled in new order, it seems wisest to make
+	 *  all movements by memmove_col().
+	 */
+
+	u8 *src, *dst;
+	u32 *s, *d;
+	int w, l , i, j;
+	u_int colsize;
+	u_int upwards = (dy < sy) || (dy == sy && dx < sx);
+
+	colsize = height;
+	if (!((sx ^ dx) & 15)) {
+		/* odd->odd or even->even */
+
+		if (upwards) {
+			src = (u8 *)info->screen_base + sy * next_line + (sx & ~15) / (8 / BPL);
+			dst = (u8 *)info->screen_base + dy * next_line + (dx & ~15) / (8 / BPL);
+			if (sx & 15) {
+				memmove32_col(dst, src, 0xff00ff, height, next_line - BPL * 2);
+				src += BPL * 2;
+				dst += BPL * 2;
+				width -= 8;
+			}
+			w = width >> 4;
+			if (w) {
+				s = (u32 *)src;
+				d = (u32 *)dst;
+				w *= BPL / 2;
+				l = next_line - w * 4;
+				for (j = height; j > 0; j--) {
+					for (i = w; i > 0; i--)
+						*d++ = *s++;
+					s = (u32 *)((u8 *)s + l);
+					d = (u32 *)((u8 *)d + l);
+				}
+			}
+			if (width & 15)
+				memmove32_col(dst + width / (8 / BPL), src + width / (8 / BPL),
+					      0xff00ff00, height, next_line - BPL * 2);
+		} else {
+			src = (u8 *)info->screen_base + (sy - 1) * next_line + ((sx + width + 8) & ~15) / (8 / BPL);
+			dst = (u8 *)info->screen_base + (dy - 1) * next_line + ((dx + width + 8) & ~15) / (8 / BPL);
+
+			if ((sx + width) & 15) {
+				src -= BPL * 2;
+				dst -= BPL * 2;
+				memmove32_col(dst, src, 0xff00ff00, colsize, -next_line - BPL * 2);
+				width -= 8;
+			}
+			w = width >> 4;
+			if (w) {
+				s = (u32 *)src;
+				d = (u32 *)dst;
+				w *= BPL / 2;
+				l = next_line - w * 4;
+				for (j = height; j > 0; j--) {
+					for (i = w; i > 0; i--)
+						*--d = *--s;
+					s = (u32 *)((u8 *)s - l);
+					d = (u32 *)((u8 *)d - l);
+				}
+			}
+			if (sx & 15)
+				memmove32_col(dst - (width - 16) / (8 / BPL),
+					      src - (width - 16) / (8 / BPL),
+					      0xff00ff, colsize, -next_line - BPL * 2);
+		}
+	} else {
+		/* odd->even or even->odd */
+		if (upwards) {
+			u32 *src32, *dst32;
+			u32 pval[4], v, v1, mask;
+			int i, j, w, f;
+
+			src = (u8 *)info->screen_base + sy * next_line + (sx & ~15) / (8 / BPL);
+			dst = (u8 *)info->screen_base + dy * next_line + (dx & ~15) / (8 / BPL);
+
+			mask = 0xff00ff00;
+			f = 0;
+			w = width;
+			if (sx & 15) {
+				f = 1;
+				w += 8;
+			}
+			if ((sx + width) & 15)
+				f |= 2;
+			w >>= 4;
+			for (i = height; i; i--) {
+				src32 = (u32 *)src;
+				dst32 = (u32 *)dst;
+
+				if (f & 1) {
+					pval[0] = (*src32++ << 8) & mask;
+					pval[1] = (*src32++ << 8) & mask;
+				} else {
+					pval[0] = dst32[0] & mask;
+					pval[1] = dst32[1] & mask;
+				}
+
+				for (j = w; j > 0; j--) {
+					v = *src32++;
+					v1 = v & mask;
+					*dst32++ = pval[0] | (v1 >> 8);
+					pval[0] = (v ^ v1) << 8;
+					v = *src32++;
+					v1 = v & mask;
+					*dst32++ = pval[1] | (v1 >> 8);
+					pval[1] = (v ^ v1) << 8;
+				}
+
+				if (f & 2) {
+					dst32[0] = (dst32[0] & mask) | pval[0];
+					dst32[1] = (dst32[1] & mask) | pval[1];
+				}
+
+				src += next_line;
+				dst += next_line;
+			}
+		} else {
+			u32 *src32, *dst32;
+			u32 pval[4], v, v1, mask;
+			int i, j, w, f;
+
+			src = (u8 *)info->screen_base + (sy - 1) * next_line + ((sx + width + 8) & ~15) / (8 / BPL);
+			dst = (u8 *)info->screen_base + (dy - 1) * next_line + ((dx + width + 8) & ~15) / (8 / BPL);
+
+			mask = 0xff00ff;
+			f = 0;
+			w = width;
+			if ((dx + width) & 15)
+				f = 1;
+			if (sx & 15) {
+				f |= 2;
+				w += 8;
+			}
+			w >>= 4;
+			for (i = height; i; i--) {
+				src32 = (u32 *)src;
+				dst32 = (u32 *)dst;
+
+				if (f & 1) {
+					pval[0] = dst32[-1] & mask;
+					pval[1] = dst32[-2] & mask;
+				} else {
+					pval[0] = (*--src32 >> 8) & mask;
+					pval[1] = (*--src32 >> 8) & mask;
+				}
+
+				for (j = w; j > 0; j--) {
+					v = *--src32;
+					v1 = v & mask;
+					*--dst32 = pval[0] | (v1 << 8);
+					pval[0] = (v ^ v1) >> 8;
+					v = *--src32;
+					v1 = v & mask;
+					*--dst32 = pval[1] | (v1 << 8);
+					pval[1] = (v ^ v1) >> 8;
+				}
+
+				if (!(f & 2)) {
+					dst32[-1] = (dst32[-1] & mask) | pval[0];
+					dst32[-2] = (dst32[-2] & mask) | pval[1];
+				}
+
+				src -= next_line;
+				dst -= next_line;
+			}
+		}
+	}
+}
+
+void atafb_iplan2p4_fillrect(struct fb_info *info, u_long next_line, u32 color,
+                             int sy, int sx, int height, int width)
+{
+	u32 *dest;
+	int rows, i;
+	u32 cval[4];
+
+	dest = (u32 *)(info->screen_base + sy * next_line + (sx & ~15) / (8 / BPL));
+	if (sx & 15) {
+		u8 *dest8 = (u8 *)dest + 1;
+
+		expand8_col2mask(color, cval);
+
+		for (i = height; i; i--) {
+			fill8_col(dest8, cval);
+			dest8 += next_line;
+		}
+		dest += BPL / 2;
+		width -= 8;
+	}
+
+	expand16_col2mask(color, cval);
+	rows = width >> 4;
+	if (rows) {
+		u32 *d = dest;
+		u32 off = next_line - rows * BPL * 2;
+		for (i = height; i; i--) {
+			d = fill16_col(d, rows, cval);
+			d = (u32 *)((long)d + off);
+		}
+		dest += rows * BPL / 2;
+		width &= 15;
+	}
+
+	if (width) {
+		u8 *dest8 = (u8 *)dest;
+
+		expand8_col2mask(color, cval);
+
+		for (i = height; i; i--) {
+			fill8_col(dest8, cval);
+			dest8 += next_line;
+		}
+	}
+}
+
+void atafb_iplan2p4_linefill(struct fb_info *info, u_long next_line,
+                             int dy, int dx, u32 width,
+                             const u8 *data, u32 bgcolor, u32 fgcolor)
+{
+	u32 *dest;
+	const u16 *data16;
+	int rows;
+	u32 fgm[4], bgm[4], m;
+
+	dest = (u32 *)(info->screen_base + dy * next_line + (dx & ~15) / (8 / BPL));
+	if (dx & 15) {
+		fill8_2col((u8 *)dest + 1, fgcolor, bgcolor, *data++);
+		dest += BPL / 2;
+		width -= 8;
+	}
+
+	if (width >= 16) {
+		data16 = (const u16 *)data;
+		expand16_2col2mask(fgcolor, bgcolor, fgm, bgm);
+
+		for (rows = width / 16; rows; rows--) {
+			u16 d = *data16++;
+			m = d | ((u32)d << 16);
+			*dest++ = (m & fgm[0]) ^ bgm[0];
+			*dest++ = (m & fgm[1]) ^ bgm[1];
+		}
+
+		data = (const u8 *)data16;
+		width &= 15;
+	}
+
+	if (width)
+		fill8_2col((u8 *)dest, fgcolor, bgcolor, *data);
+}
+
+#ifdef MODULE
+MODULE_LICENSE("GPL");
+
+int init_module(void)
+{
+	return 0;
+}
+
+void cleanup_module(void)
+{
+}
+#endif /* MODULE */
+
+
+    /*
+     *  Visible symbols for modules
+     */
+
+EXPORT_SYMBOL(atafb_iplan2p4_copyarea);
+EXPORT_SYMBOL(atafb_iplan2p4_fillrect);
+EXPORT_SYMBOL(atafb_iplan2p4_linefill);

drivers/video/atafb_iplan2p8.c

+/*
+ *  linux/drivers/video/iplan2p8.c -- Low level frame buffer operations for
+ *				      interleaved bitplanes à la Atari (8
+ *				      planes, 2 bytes interleave)
+ *
+ *	Created 5 Apr 1997 by Geert Uytterhoeven
+ *
+ *  This file is subject to the terms and conditions of the GNU General Public
+ *  License.  See the file COPYING in the main directory of this archive for
+ *  more details.
+ */
+
+#include <linux/module.h>
+#include <linux/string.h>
+#include <linux/fb.h>
+
+#include <asm/setup.h>
+
+#include "atafb.h"
+
+#define BPL	8
+#include "atafb_utils.h"
+
+
+/* Copies a 8 plane column from 's', height 'h', to 'd'. */
+
+/* This expands a 8 bit color into two longs for two movepl (8 plane)
+ * operations.
+ */
+
+void atafb_iplan2p8_copyarea(struct fb_info *info, u_long next_line,
+			     int sy, int sx, int dy, int dx,
+			     int height, int width)
+{
+	/*  bmove() has to distinguish two major cases: If both, source and
+	 *  destination, start at even addresses or both are at odd
+	 *  addresses, just the first odd and last even column (if present)
+	 *  require special treatment (memmove_col()). The rest between
+	 *  then can be copied by normal operations, because all adjacent
+	 *  bytes are affected and are to be stored in the same order.
+	 *    The pathological case is when the move should go from an odd
+	 *  address to an even or vice versa. Since the bytes in the plane
+	 *  words must be assembled in new order, it seems wisest to make
+	 *  all movements by memmove_col().
+	 */
+
+	u8 *src, *dst;
+	u32 *s, *d;
+	int w, l , i, j;
+	u_int colsize;
+	u_int upwards = (dy < sy) || (dy == sy && dx < sx);
+
+	colsize = height;
+	if (!((sx ^ dx) & 15)) {
+		/* odd->odd or even->even */
+
+		if (upwards) {
+			src = (u8 *)info->screen_base + sy * next_line + (sx & ~15) / (8 / BPL);
+			dst = (u8 *)info->screen_base + dy * next_line + (dx & ~15) / (8 / BPL);
+			if (sx & 15) {
+				memmove32_col(dst, src, 0xff00ff, height, next_line - BPL * 2);
+				src += BPL * 2;
+				dst += BPL * 2;
+				width -= 8;
+			}
+			w = width >> 4;
+			if (w) {
+				s = (u32 *)src;
+				d = (u32 *)dst;
+				w *= BPL / 2;
+				l = next_line - w * 4;
+				for (j = height; j > 0; j--) {
+					for (i = w; i > 0; i--)
+						*d++ = *s++;
+					s = (u32 *)((u8 *)s + l);
+					d = (u32 *)((u8 *)d + l);
+				}
+			}
+			if (width & 15)
+				memmove32_col(dst + width / (8 / BPL), src + width / (8 / BPL),
+					      0xff00ff00, height, next_line - BPL * 2);
+		} else {
+			src = (u8 *)info->screen_base + (sy - 1) * next_line + ((sx + width + 8) & ~15) / (8 / BPL);
+			dst = (u8 *)info->screen_base + (dy - 1) * next_line + ((dx + width + 8) & ~15) / (8 / BPL);
+
+			if ((sx + width) & 15) {
+				src -= BPL * 2;
+				dst -= BPL * 2;
+				memmove32_col(dst, src, 0xff00ff00, colsize, -next_line - BPL * 2);
+				width -= 8;
+			}
+			w = width >> 4;
+			if (w) {
+				s = (u32 *)src;
+				d = (u32 *)dst;
+				w *= BPL / 2;
+				l = next_line - w * 4;
+				for (j = height; j > 0; j--) {
+					for (i = w; i > 0; i--)
+						*--d = *--s;
+					s = (u32 *)((u8 *)s - l);
+					d = (u32 *)((u8 *)d - l);
+				}
+			}
+			if (sx & 15)
+				memmove32_col(dst - (width - 16) / (8 / BPL),
+					      src - (width - 16) / (8 / BPL),
+					      0xff00ff, colsize, -next_line - BPL * 2);
+		}
+	} else {
+		/* odd->even or even->odd */
+		if (upwards) {
+			u32 *src32, *dst32;
+			u32 pval[4], v, v1, mask;
+			int i, j, w, f;
+
+			src = (u8 *)info->screen_base + sy * next_line + (sx & ~15) / (8 / BPL);
+			dst = (u8 *)info->screen_base + dy * next_line + (dx & ~15) / (8 / BPL);
+
+			mask = 0xff00ff00;
+			f = 0;
+			w = width;
+			if (sx & 15) {
+				f = 1;
+				w += 8;
+			}
+			if ((sx + width) & 15)
+				f |= 2;
+			w >>= 4;
+			for (i = height; i; i--) {
+				src32 = (u32 *)src;
+				dst32 = (u32 *)dst;
+
+				if (f & 1) {
+					pval[0] = (*src32++ << 8) & mask;
+					pval[1] = (*src32++ << 8) & mask;
+					pval[2] = (*src32++ << 8) & mask;
+					pval[3] = (*src32++ << 8) & mask;
+				} else {
+					pval[0] = dst32[0] & mask;
+					pval[1] = dst32[1] & mask;
+					pval[2] = dst32[2] & mask;
+					pval[3] = dst32[3] & mask;
+				}
+
+				for (j = w; j > 0; j--) {
+					v = *src32++;
+					v1 = v & mask;
+					*dst32++ = pval[0] | (v1 >> 8);
+					pval[0] = (v ^ v1) << 8;
+					v = *src32++;
+					v1 = v & mask;
+					*dst32++ = pval[1] | (v1 >> 8);
+					pval[1] = (v ^ v1) << 8;
+					v = *src32++;
+					v1 = v & mask;
+					*dst32++ = pval[2] | (v1 >> 8);
+					pval[2] = (v ^ v1) << 8;
+					v = *src32++;
+					v1 = v & mask;
+					*dst32++ = pval[3] | (v1 >> 8);
+					pval[3] = (v ^ v1) << 8;
+				}
+
+				if (f & 2) {
+					dst32[0] = (dst32[0] & mask) | pval[0];
+					dst32[1] = (dst32[1] & mask) | pval[1];
+					dst32[2] = (dst32[2] & mask) | pval[2];
+					dst32[3] = (dst32[3] & mask) | pval[3];
+				}
+
+				src += next_line;
+				dst += next_line;
+			}
+		} else {
+			u32 *src32, *dst32;
+			u32 pval[4], v, v1, mask;
+			int i, j, w, f;
+
+			src = (u8 *)info->screen_base + (sy - 1) * next_line + ((sx + width + 8) & ~15) / (8 / BPL);
+			dst = (u8 *)info->screen_base + (dy - 1) * next_line + ((dx + width + 8) & ~15) / (8 / BPL);
+
+			mask = 0xff00ff;
+			f = 0;
+			w = width;
+			if ((dx + width) & 15)
+				f = 1;
+			if (sx & 15) {
+				f |= 2;
+				w += 8;
+			}
+			w >>= 4;
+			for (i = height; i; i--) {
+				src32 = (u32 *)src;
+				dst32 = (u32 *)dst;
+
+				if (f & 1) {
+					pval[0] = dst32[-1] & mask;
+					pval[1] = dst32[-2] & mask;
+					pval[2] = dst32[-3] & mask;
+					pval[3] = dst32[-4] & mask;
+				} else {
+					pval[0] = (*--src32 >> 8) & mask;
+					pval[1] = (*--src32 >> 8) & mask;
+					pval[2] = (*--src32 >> 8) & mask;
+					pval[3] = (*--src32 >> 8) & mask;
+				}
+
+				for (j = w; j > 0; j--) {
+					v = *--src32;
+					v1 = v & mask;
+					*--dst32 = pval[0] | (v1 << 8);
+					pval[0] = (v ^ v1) >> 8;
+					v = *--src32;
+					v1 = v & mask;
+					*--dst32 = pval[1] | (v1 << 8);
+					pval[1] = (v ^ v1) >> 8;
+					v = *--src32;
+					v1 = v & mask;
+					*--dst32 = pval[2] | (v1 << 8);
+					pval[2] = (v ^ v1) >> 8;
+					v = *--src32;
+					v1 = v & mask;
+					*--dst32 = pval[3] | (v1 << 8);
+					pval[3] = (v ^ v1) >> 8;
+				}
+
+				if (!(f & 2)) {
+					dst32[-1] = (dst32[-1] & mask) | pval[0];
+					dst32[-2] = (dst32[-2] & mask) | pval[1];
+					dst32[-3] = (dst32[-3] & mask) | pval[2];
+					dst32[-4] = (dst32[-4] & mask) | pval[3];
+				}
+
+				src -= next_line;
+				dst -= next_line;
+			}
+		}
+	}
+}
+
+void atafb_iplan2p8_fillrect(struct fb_info *info, u_long next_line, u32 color,
+                             int sy, int sx, int height, int width)
+{
+	u32 *dest;
+	int rows, i;
+	u32 cval[4];
+
+	dest = (u32 *)(info->screen_base + sy * next_line + (sx & ~15) / (8 / BPL));
+	if (sx & 15) {
+		u8 *dest8 = (u8 *)dest + 1;
+
+		expand8_col2mask(color, cval);
+
+		for (i = height; i; i--) {
+			fill8_col(dest8, cval);
+			dest8 += next_line;
+		}
+		dest += BPL / 2;
+		width -= 8;
+	}
+
+	expand16_col2mask(color, cval);
+	rows = width >> 4;
+	if (rows) {
+		u32 *d = dest;
+		u32 off = next_line - rows * BPL * 2;
+		for (i = height; i; i--) {
+			d = fill16_col(d, rows, cval);
+			d = (u32 *)((long)d + off);
+		}
+		dest += rows * BPL / 2;
+		width &= 15;
+	}
+
+	if (width) {
+		u8 *dest8 = (u8 *)dest;
+
+		expand8_col2mask(color, cval);
+
+		for (i = height; i; i--) {
+			fill8_col(dest8, cval);
+			dest8 += next_line;
+		}
+	}
+}
+
+void atafb_iplan2p8_linefill(struct fb_info *info, u_long next_line,
+			     int dy, int dx, u32 width,
+			     const u8 *data, u32 bgcolor, u32 fgcolor)
+{
+	u32 *dest;
+	const u16 *data16;
+	int rows;
+	u32 fgm[4], bgm[4], m;
+
+	dest = (u32 *)(info->screen_base + dy * next_line + (dx & ~15) / (8 / BPL));
+	if (dx & 15) {
+		fill8_2col((u8 *)dest + 1, fgcolor, bgcolor, *data++);
+		dest += BPL / 2;
+		width -= 8;
+	}
+
+	if (width >= 16) {
+		data16 = (const u16 *)data;
+		expand16_2col2mask(fgcolor, bgcolor, fgm, bgm);
+
+		for (rows = width / 16; rows; rows--) {
+			u16 d = *data16++;
+			m = d | ((u32)d << 16);
+			*dest++ = (m & fgm[0]) ^ bgm[0];
+			*dest++ = (m & fgm[1]) ^ bgm[1];
+			*dest++ = (m & fgm[2]) ^ bgm[2];
+			*dest++ = (m & fgm[3]) ^ bgm[3];
+		}
+
+		data = (const u8 *)data16;
+		width &= 15;
+	}
+
+	if (width)
+		fill8_2col((u8 *)dest, fgcolor, bgcolor, *data);
+}
+
+#ifdef MODULE
+MODULE_LICENSE("GPL");
+
+int init_module(void)
+{
+	return 0;
+}
+
+void cleanup_module(void)
+{
+}
+#endif /* MODULE */
+
+
+    /*
+     *  Visible symbols for modules
+     */
+
+EXPORT_SYMBOL(atafb_iplan2p8_copyarea);
+EXPORT_SYMBOL(atafb_iplan2p8_fillrect);
+EXPORT_SYMBOL(atafb_iplan2p8_linefill);

drivers/video/atafb_mfb.c

+/*
+ *  linux/drivers/video/mfb.c -- Low level frame buffer operations for
+ *				 monochrome
+ *
+ *	Created 5 Apr 1997 by Geert Uytterhoeven
+ *
+ *  This file is subject to the terms and conditions of the GNU General Public
+ *  License.  See the file COPYING in the main directory of this archive for
+ *  more details.
+ */
+
+#include <linux/module.h>
+#include <linux/string.h>
+#include <linux/fb.h>
+
+#include "atafb.h"
+#include "atafb_utils.h"
+
+
+    /*
+     *  Monochrome
+     */
+
+void atafb_mfb_copyarea(struct fb_info *info, u_long next_line,
+			int sy, int sx, int dy, int dx,
+			int height, int width)
+{
+	u8 *src, *dest;
+	u_int rows;
+
+	if (sx == 0 && dx == 0 && width == next_line) {
+		src = (u8 *)info->screen_base + sy * (width >> 3);
+		dest = (u8 *)info->screen_base + dy * (width >> 3);
+		fb_memmove(dest, src, height * (width >> 3));
+	} else if (dy <= sy) {
+		src = (u8 *)info->screen_base + sy * next_line + (sx >> 3);
+		dest = (u8 *)info->screen_base + dy * next_line + (dx >> 3);
+		for (rows = height; rows--;) {
+			fb_memmove(dest, src, width >> 3);
+			src += next_line;
+			dest += next_line;
+		}
+	} else {
+		src = (u8 *)info->screen_base + (sy + height - 1) * next_line + (sx >> 3);
+		dest = (u8 *)info->screen_base + (dy + height - 1) * next_line + (dx >> 3);
+		for (rows = height; rows--;) {
+			fb_memmove(dest, src, width >> 3);
+			src -= next_line;
+			dest -= next_line;
+		}
+	}
+}
+
+void atafb_mfb_fillrect(struct fb_info *info, u_long next_line, u32 color,
+			int sy, int sx, int height, int width)
+{
+	u8 *dest;
+	u_int rows;
+
+	dest = (u8 *)info->screen_base + sy * next_line + (sx >> 3);
+
+	if (sx == 0 && width == next_line) {
+		if (color)
+			fb_memset255(dest, height * (width >> 3));
+		else
+			fb_memclear(dest, height * (width >> 3));
+	} else {
+		for (rows = height; rows--; dest += next_line) {
+			if (color)
+				fb_memset255(dest, width >> 3);
+			else
+				fb_memclear_small(dest, width >> 3);
+		}
+	}
+}
+
+void atafb_mfb_linefill(struct fb_info *info, u_long next_line,
+			int dy, int dx, u32 width,
+			const u8 *data, u32 bgcolor, u32 fgcolor)
+{
+	u8 *dest;
+	u_int rows;
+
+	dest = (u8 *)info->screen_base + dy * next_line + (dx >> 3);
+
+	for (rows = width / 8; rows--; /* check margins */ ) {
+		// use fast_memmove or fb_memmove
+		*dest++ = *data++;
+	}
+}
+
+#ifdef MODULE
+MODULE_LICENSE("GPL");
+
+int init_module(void)
+{
+	return 0;
+}
+
+void cleanup_module(void)
+{
+}
+#endif /* MODULE */
+
+
+    /*
+     *  Visible symbols for modules
+     */
+
+EXPORT_SYMBOL(atafb_mfb_copyarea);
+EXPORT_SYMBOL(atafb_mfb_fillrect);
+EXPORT_SYMBOL(atafb_mfb_linefill);