[PATCH] Update cyblafb driver

This is a major update to the cyblafb framebuffer driver. Most
of the stuff has been tested in the mm tree.

Main advantages:
============
  - vxres > xres support
  - ywrap and xpan support
  - much faster for almost all modes (e.g. 1280x1024-16bpp
     draws more than 41 full screens of text instead of about 25
     full screens of text per second on authors Epia 5000)
  - module init/exit code fixed
  - bugs triggered by console rotation fixed
  - lots of minor improvements
  - startup modes suitable for high performance scrolling
     in all directions

This diff  also contains a lot of white space fixes.

No side effects are possible, only one single graphics core is affected.

Signed-off-by: Knut Petersen <Knut_Petersen@t-online.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

2 files changed, 875 insertions, 650 deletions

diff --git a/drivers/video/Kconfig b/drivers/video/Kconfig
index cc8e3bf5001b..3f04427c9026 100644
--- a/drivers/video/Kconfig
+++ b/drivers/video/Kconfig
@@ -1151,7 +1151,7 @@ config FB_VOODOO1
 
 config FB_CYBLA
 	tristate "Cyberblade/i1 support"
-	depends on FB && PCI
+	depends on FB && PCI && X86_32 && !64BIT
 	select FB_CFB_IMAGEBLIT
 	select VIDEO_SELECT
 	---help---
diff --git a/drivers/video/cyblafb.c b/drivers/video/cyblafb.c
index 03fbe83d71a8..e9f5dee67e3c 100644
--- a/drivers/video/cyblafb.c
+++ b/drivers/video/cyblafb.c
@@ -7,11 +7,12 @@
  *	tridentfb.c by Jani Monoses
  *	see files above for further credits
  *
- * TODO:
- *
  */
 
 #define CYBLAFB_DEBUG 0
+#define CYBLAFB_KD_GRAPHICS_QUIRK 1
+
+#define CYBLAFB_PIXMAPSIZE 8192
 
 #include <linux/config.h>
 #include <linux/module.h>
@@ -22,7 +23,7 @@
 #include <asm/types.h>
 #include <video/cyblafb.h>
 
-#define VERSION "0.54"
+#define VERSION "0.62"
 
 struct cyblafb_par {
 	u32 pseudo_pal[16];
@@ -32,7 +33,9 @@ struct cyblafb_par {
 static struct fb_fix_screeninfo cyblafb_fix __devinitdata = {
 	.id = "CyBla",
 	.type = FB_TYPE_PACKED_PIXELS,
+	.xpanstep = 1,
 	.ypanstep = 1,
+	.ywrapstep = 1,
 	.visual = FB_VISUAL_PSEUDOCOLOR,
 	.accel = FB_ACCEL_NONE,
 };
@@ -43,8 +46,9 @@ static int ref __devinitdata = 75;
 static int fp __devinitdata;
 static int crt __devinitdata;
 static int memsize __devinitdata;
-static int vesafb __devinitdata;
 
+static int basestride;
+static int vesafb;
 static int nativex;
 static int center;
 static int stretch;
@@ -52,26 +56,50 @@ static int pciwb = 1;
 static int pcirb = 1;
 static int pciwr = 1;
 static int pcirr = 1;
+static int disabled;
 static int verbosity;
 static int displaytype;
 
-static void __iomem * io_virt; // iospace virtual memory address
-
-module_param(mode,charp,0);
-module_param(bpp,int,0);
-module_param(ref,int,0);
-module_param(fp,int,0);
-module_param(crt,int,0);
-module_param(nativex,int,0);
-module_param(center,int,0);
-module_param(stretch,int,0);
-module_param(pciwb,int,0);
-module_param(pcirb,int,0);
-module_param(pciwr,int,0);
-module_param(pcirr,int,0);
-module_param(memsize,int,0);
-module_param(verbosity,int,0);
-module_param(vesafb,int,0);
+static void __iomem *io_virt;	// iospace virtual memory address
+
+module_param(mode, charp, 0);
+module_param(bpp, int, 0);
+module_param(ref, int, 0);
+module_param(fp, int, 0);
+module_param(crt, int, 0);
+module_param(nativex, int, 0);
+module_param(center, int, 0);
+module_param(stretch, int, 0);
+module_param(pciwb, int, 0);
+module_param(pcirb, int, 0);
+module_param(pciwr, int, 0);
+module_param(pcirr, int, 0);
+module_param(memsize, int, 0);
+module_param(verbosity, int, 0);
+
+//=========================================
+//
+// Well, we have to fix the upper layers.
+// Until this has been done, we work around
+// the bugs.
+//
+//=========================================
+
+#if (CYBLAFB_KD_GRAPHICS_QUIRK && CYBLAFB_DEBUG)
+	if (disabled) { \
+		printk("********\n");\
+		dump_stack();\
+		return val;\
+	}
+
+#elif CYBLAFB_KD_GRAPHICS_QUIRK
+#define KD_GRAPHICS_RETURN(val)\
+	if (disabled) {\
+		return val;\
+	}
+#else
+#define KD_GRAPHICS_RETURN(val)
+#endif
 
 //=========================================
 //
@@ -79,10 +107,10 @@ module_param(vesafb,int,0);
 //
 //=========================================
 
-#define out8(r,v) writeb(v,io_virt+r)
-#define out32(r,v) writel(v,io_virt+r)
-#define in8(r) readb(io_virt+r)
-#define in32(r) readl(io_virt+r)
+#define out8(r, v) writeb(v, io_virt + r)
+#define out32(r, v) writel(v, io_virt + r)
+#define in8(r) readb(io_virt + r)
+#define in32(r) readl(io_virt + r)
 
 //======================================
 //
@@ -90,47 +118,47 @@ module_param(vesafb,int,0);
 //
 //======================================
 
-static inline unsigned char read3X4(int reg)
+static inline u8 read3X4(u32 reg)
 {
-	out8(0x3D4,reg);
+	out8(0x3D4, reg);
 	return in8(0x3D5);
 }
 
-static inline unsigned char read3C4(int reg)
+static inline u8 read3C4(u32 reg)
 {
-	out8(0x3C4,reg);
+	out8(0x3C4, reg);
 	return in8(0x3C5);
 }
 
-static inline unsigned char read3CE(int reg)
+static inline u8 read3CE(u32 reg)
 {
-	out8(0x3CE,reg);
+	out8(0x3CE, reg);
 	return in8(0x3CF);
 }
 
-static inline void write3X4(int reg,unsigned char val)
+static inline void write3X4(u32 reg, u8 val)
 {
-	out8(0x3D4,reg);
-	out8(0x3D5,val);
+	out8(0x3D4, reg);
+	out8(0x3D5, val);
 }
 
-static inline void write3C4(int reg,unsigned char val)
+static inline void write3C4(u32 reg, u8 val)
 {
-	out8(0x3C4,reg);
-	out8(0x3C5,val);
+	out8(0x3C4, reg);
+	out8(0x3C5, val);
 }
 
-static inline void write3CE(int reg,unsigned char val)
+static inline void write3CE(u32 reg, u8 val)
 {
-	out8(0x3CE,reg);
-	out8(0x3CF,val);
+	out8(0x3CE, reg);
+	out8(0x3CF, val);
 }
 
-static inline void write3C0(int reg,unsigned char val)
+static inline void write3C0(u32 reg, u8 val)
 {
-	in8(0x3DA);	// read to reset index
-	out8(0x3C0,reg);
-	out8(0x3C0,val);
+	in8(0x3DA);		// read to reset index
+	out8(0x3C0, reg);
+	out8(0x3C0, val);
 }
 
 //=================================================
@@ -139,58 +167,62 @@ static inline void write3C0(int reg,unsigned char val)
 //
 //=================================================
 
-static inline void enable_mmio(void)
+static void enable_mmio(void)
 {
-	int tmp;
+	u8 tmp;
 
-	outb(0x0B,0x3C4);
+	outb(0x0B, 0x3C4);
 	inb(0x3C5);		// Set NEW mode
-	outb(SR0E,0x3C4);	// write enable a lot of extended ports
-	outb(0x80,0x3C5);
+	outb(SR0E, 0x3C4);	// write enable a lot of extended ports
+	outb(0x80, 0x3C5);
 
-	outb(SR11,0x3C4);	// write enable those extended ports that
-	outb(0x87,0x3C5);	// are not affected by SR0E_New
+	outb(SR11, 0x3C4);	// write enable those extended ports that
+	outb(0x87, 0x3C5);	// are not affected by SR0E_New
 
-	outb(CR1E,0x3d4);	// clear write protect bit for port 0x3c2
-	tmp=inb(0x3d5) & 0xBF;
-	outb(CR1E,0x3d4);
-	outb(tmp,0x3d5);
+	outb(CR1E, 0x3d4);	// clear write protect bit for port 0x3c2
+	tmp = inb(0x3d5) & 0xBF;
+	outb(CR1E, 0x3d4);
+	outb(tmp, 0x3d5);
 
-	outb(CR39,0x3D4);
-	outb(inb(0x3D5)|0x01,0x3D5); // Enable mmio, everything else untouched
+	outb(CR39, 0x3D4);
+	outb(inb(0x3D5) | 0x01, 0x3D5); // Enable mmio
 }
 
 //=================================================
 //
 // Set pixel clock VCLK1
-//   - multipliers set elswhere
-//   - freq in units of 0.01 MHz
+// - multipliers set elswhere
+// - freq in units of 0.01 MHz
+//
+// Hardware bug: SR18 >= 250 is broken for the
+//		 cyberblade/i1
 //
 //=================================================
 
 static void set_vclk(struct cyblafb_par *par, int freq)
 {
-	u32 m,n,k;
-	int f,fi,d,di;
-	u8 lo=0,hi=0;
+	u32 m, n, k;
+	int f, fi, d, di;
+	u8 lo = 0, hi = 0;
 
 	d = 2000;
 	k = freq >= 10000 ? 0 : freq >= 5000 ? 1 : freq >= 2500 ? 2 : 3;
-	for(m = 0;m<64;m++)
-	for(n = 0;n<250;n++) { // max 249 is a hardware limit for cybla/i1 !
-		fi = (int)(((5864727*(n+8))/((m+2)*(1<<k)))>>12);
-		if ((di = abs(fi - freq)) < d) {
-			d = di;
-			f = fi;
-			lo = (u8) n;
-			hi = (u8) ((k<<6) | m);
+	for (m = 0; m < 64; m++)
+		for (n = 0; n < 250; n++) {
+			fi = (int)(((5864727 * (n + 8)) /
+				    ((m + 2) * (1 << k))) >> 12);
+			if ((di = abs(fi - freq)) < d) {
+				d = di;
+				f = fi;
+				lo = (u8) n;
+				hi = (u8) ((k << 6) | m);
+			}
 		}
-	}
-	write3C4(SR19,hi);
-	write3C4(SR18,lo);
-	if(verbosity > 1)
+	write3C4(SR19, hi);
+	write3C4(SR18, lo);
+	if (verbosity > 0)
 		output("pixclock = %d.%02d MHz, k/m/n %x %x %x\n",
-		freq/100,freq%100,(hi&0xc0)>>6,hi&0x3f,lo);
+		       freq / 100, freq % 100, (hi & 0xc0) >> 6, hi & 0x3f, lo);
 }
 
 //================================================
@@ -199,83 +231,83 @@ static void set_vclk(struct cyblafb_par *par, int freq)
 //
 //================================================
 
-static void cyblafb_setup_GE(int pitch,int bpp)
+static void cyblafb_setup_GE(int pitch, int bpp)
 {
-	int base = (pitch>>3)<<20;
+	KD_GRAPHICS_RETURN();
 
 	switch (bpp) {
-		case  8: base |= (0<<29); break;
-		case 15: base |= (5<<29); break;
-		case 16: base |= (1<<29); break;
-		case 24:
-		case 32: base |= (2<<29); break;
+	case 8:
+		basestride = ((pitch >> 3) << 20) | (0 << 29);
+		break;
+	case 15:
+		basestride = ((pitch >> 3) << 20) | (5 << 29);
+		break;
+	case 16:
+		basestride = ((pitch >> 3) << 20) | (1 << 29);
+		break;
+	case 24:
+	case 32:
+		basestride = ((pitch >> 3) << 20) | (2 << 29);
+		break;
 	}
 
-	write3X4(CR36,0x90);	// reset GE
-	write3X4(CR36,0x80);	// enable GE
-
-	out32(GE24,1<<7);	// reset all GE pointers
-	out32(GE24,0);
-
-	write3X4(CR2D,0x00);	// GE Timinigs, no delays
-
-	out32(GEB8,base); // Destination Stride / Buffer Base 0, p 133
-	out32(GEBC,base); // Destination Stride / Buffer Base 1, p 133
-	out32(GEC0,base); // Destination Stride / Buffer Base 2, p 133
-	out32(GEC4,base); // Destination Stride / Buffer Base 3, p 133
-	out32(GEC8,base); // Source Stride / Buffer Base 0, p 133
-	out32(GECC,base); // Source Stride / Buffer Base 1, p 133
-	out32(GED0,base); // Source Stride / Buffer Base 2, p 133
-	out32(GED4,base); // Source Stride / Buffer Base 3, p 133
-	out32(GE6C,0);	  // Pattern and Style, p 129, ok
+	write3X4(CR36, 0x90);	// reset GE
+	write3X4(CR36, 0x80);	// enable GE
+	out32(GE24, 1 << 7);	// reset all GE pointers by toggling
+	out32(GE24, 0); 	//   d7 of GE24
+	write3X4(CR2D, 0x00);	// GE Timinigs, no delays
+	out32(GE6C, 0); 	// Pattern and Style, p 129, ok
 }
 
 //=====================================================================
 //
-// Although this is a .fb_sync function that could be enabled in
-// cyblafb_ops, we do not include it there. We sync immediately before
-// new GE operations to improve performance.
+// Cyberblade specific syncing
+//
+//   A timeout might be caused by disabled mmio.
+//   Cause:
+//     - bit CR39 & 1 == 0 upon return, X trident driver bug
+//     - kdm bug (KD_GRAPHICS not set on first switch)
+//     - kernel design flaw (it believes in the correctness
+//	 of kdm/X
+//   First we try to sync ignoring that problem, as most of the
+//   time that will succeed immediately and the enable_mmio()
+//   would only degrade performance.
 //
 //=====================================================================
 
 static int cyblafb_sync(struct fb_info *info)
 {
-	int status, i=100000;
-	while( ((status=in32(GE20)) & 0xFA800000) && i != 0)
+	u32 status, i = 100000;
+
+	KD_GRAPHICS_RETURN(0);
+
+	while (((status = in32(GE20)) & 0xFe800000) && i != 0)
 		i--;
 
 	if (i == 0) {
-		// The timeout might be caused by disabled mmio.
-		// Cause:
-		//   - bit CR39 & 1 == 0 upon return, X trident driver bug
-		//   - kdm bug (KD_GRAPHICS not set on first switch)
-		//   - kernel design flaw (it believes in the correctness
-		//     of kdm/X
-		// So we make sure that mmio is enabled first ...
 		enable_mmio();
-//		show_trace(NULL,&status);
-		i=1000000;
-		while( ((status=in32(GE20)) & 0xFA800000) && i != 0)
+		i = 1000000;
+		while (((status = in32(GE20)) & 0xFA800000) && i != 0)
 			i--;
 		if (i == 0) {
-			output("GE Timeout, status: %x\n",status);
-			if(status & 0x80000000)
+			output("GE Timeout, status: %x\n", status);
+			if (status & 0x80000000)
 				output("Bresenham Engine : Busy\n");
-			if(status & 0x40000000)
+			if (status & 0x40000000)
 				output("Setup Engine     : Busy\n");
-			if(status & 0x20000000)
+			if (status & 0x20000000)
 				output("SP / DPE         : Busy\n");
-			if(status & 0x10000000)
+			if (status & 0x10000000)
 				output("Memory Interface : Busy\n");
-			if(status & 0x08000000)
+			if (status & 0x08000000)
 				output("Com Lst Proc     : Busy\n");
-			if(status & 0x04000000)
+			if (status & 0x04000000)
 				output("Block Write      : Busy\n");
-			if(status & 0x02000000)
+			if (status & 0x02000000)
 				output("Command Buffer   : Full\n");
-			if(status & 0x01000000)
+			if (status & 0x01000000)
 				output("RESERVED         : Busy\n");
-			if(status & 0x00800000)
+			if (status & 0x00800000)
 				output("PCI Write Buffer : Busy\n");
 			cyblafb_setup_GE(info->var.xres,
 					 info->var.bits_per_pixel);
@@ -291,142 +323,193 @@ static int cyblafb_sync(struct fb_info *info)
 //
 //==============================
 
-static void cyblafb_fillrect(struct fb_info * info,
-			     const struct fb_fillrect *fr)
+static void cyblafb_fillrect(struct fb_info *info, const struct fb_fillrect *fr)
 {
-	int bpp = info->var.bits_per_pixel;
-	int col;
+	u32 bpp = info->var.bits_per_pixel, col, desty, height;
+
+	KD_GRAPHICS_RETURN();
 
 	switch (bpp) {
-		default:
-		case 8: col = fr->color;
-			col |= col <<8;
-			col |= col <<16;
-			break;
-		case 16: col = ((u32 *)(info->pseudo_palette))[fr->color];
-			 col |= col <<16;
-			 break;
-		case 32: col = ((u32 *)(info->pseudo_palette))[fr->color];
-			 break;
+	default:
+	case 8:
+		col = fr->color;
+		col |= col << 8;
+		col |= col << 16;
+		break;
+	case 16:
+		col = ((u32 *) (info->pseudo_palette))[fr->color];
+		col |= col << 16;
+		break;
+	case 32:
+		col = ((u32 *) (info->pseudo_palette))[fr->color];
+		break;
 	}
 
-	cyblafb_sync(info);
-
-	out32(GE60,col);
-	out32(GE48,fr->rop ? 0x66:ROP_S);
-	out32(GE44,0x20000000|1<<19|1<<4|2<<2);
-	out32(GE08,point(fr->dx,fr->dy));
-	out32(GE0C,point(fr->dx+fr->width-1,fr->dy+fr->height-1));
-
+	desty = fr->dy;
+	height = fr->height;
+	while (height) {
+		out32(GEB8, basestride | ((desty * info->var.xres_virtual *
+					   bpp) >> 6));
+		out32(GE60, col);
+		out32(GE48, fr->rop ? 0x66 : ROP_S);
+		out32(GE44, 0x20000000 | 1 << 19 | 1 << 4 | 2 << 2);
+		out32(GE08, point(fr->dx, 0));
+		out32(GE0C, point(fr->dx + fr->width - 1,
+				  height > 4096 ? 4095 : height - 1));
+		if (likely(height <= 4096))
+			return;
+		desty += 4096;
+		height -= 4096;
+	}
 }
 
-//==============================
+//================================================
 //
 // Cyberblade specific copyarea
 //
-//==============================
+// This function silently assumes that it never
+// will be called with width or height exceeding
+// 4096.
+//
+//================================================
 
-static void cyblafb_copyarea(struct fb_info *info,
-			     const struct fb_copyarea *ca)
+static void cyblafb_copyarea(struct fb_info *info, const struct fb_copyarea *ca)
 {
-	__u32 s1,s2,d1,d2;
-	int direction;
+	u32 s1, s2, d1, d2, direction;
+
+	KD_GRAPHICS_RETURN();
+
+	s1 = point(ca->sx, 0);
+	s2 = point(ca->sx + ca->width - 1, ca->height - 1);
+	d1 = point(ca->dx, 0);
+	d2 = point(ca->dx + ca->width - 1, ca->height - 1);
 
-	s1 = point(ca->sx,ca->sy);
-	s2 = point(ca->sx+ca->width-1,ca->sy+ca->height-1);
-	d1 = point(ca->dx,ca->dy);
-	d2 = point(ca->dx+ca->width-1,ca->dy+ca->height-1);
 	if ((ca->sy > ca->dy) || ((ca->sy == ca->dy) && (ca->sx > ca->dx)))
 		direction = 0;
 	else
 		direction = 2;
 
-	cyblafb_sync(info);
-
-	out32(GE44,0xa0000000|1<<19|1<<2|direction);
-	out32(GE00,direction?s2:s1);
-	out32(GE04,direction?s1:s2);
-	out32(GE08,direction?d2:d1);
-	out32(GE0C,direction?d1:d2);
-
+	out32(GEB8, basestride | ((ca->dy * info->var.xres_virtual *
+				   info->var.bits_per_pixel) >> 6));
+	out32(GEC8, basestride | ((ca->sy * info->var.xres_virtual *
+				   info->var.bits_per_pixel) >> 6));
+	out32(GE44, 0xa0000000 | 1 << 19 | 1 << 2 | direction);
+	out32(GE00, direction ? s2 : s1);
+	out32(GE04, direction ? s1 : s2);
+	out32(GE08, direction ? d2 : d1);
+	out32(GE0C, direction ? d1 : d2);
 }
 
 //=======================================================================
 //
 // Cyberblade specific imageblit
 //
-// Accelerated for the most usual case, blitting 1-bit deep character
-// character images. Everything else is passed to the generic imageblit.
+// Accelerated for the most usual case, blitting 1 - bit deep
+// character images. Everything else is passed to the generic imageblit
+// unless it is so insane that it is better to printk an alert.
+//
+// Hardware bug: _Never_ blit across pixel column 2048, that will lock
+// the system. We split those blit requests into three blitting
+// operations.
 //
 //=======================================================================
 
 static void cyblafb_imageblit(struct fb_info *info,
 			      const struct fb_image *image)
 {
-
 	u32 fgcol, bgcol;
+	u32 *pd = (u32 *) image->data;
+	u32 bpp = info->var.bits_per_pixel;
 
-	int i;
-	int bpp = info->var.bits_per_pixel;
-	int index = 0;
-	int index_end=image->height * image->width / 8;
-	int width_dds=image->width / 32;
-	int width_dbs=image->width % 32;
-
-	if (image->depth != 1 || bpp < 8 || bpp > 32 || bpp % 8 != 0 ||
-	    image->width % 8 != 0 || image->width == 0 || image->height == 0) {
-		cfb_imageblit(info,image);
+	KD_GRAPHICS_RETURN();
+
+	// Used only for drawing the penguine (image->depth > 1)
+	if (image->depth != 1) {
+		cfb_imageblit(info, image);
+		return;
+	}
+	// That should never happen, but it would be fatal
+	if (image->width == 0 || image->height == 0) {
+		output("imageblit: width/height 0 detected\n");
 		return;
 	}
 
 	if (info->fix.visual == FB_VISUAL_TRUECOLOR ||
 	    info->fix.visual == FB_VISUAL_DIRECTCOLOR) {
-		fgcol = ((u32*)(info->pseudo_palette))[image->fg_color];
-		bgcol = ((u32*)(info->pseudo_palette))[image->bg_color];
+		fgcol = ((u32 *) (info->pseudo_palette))[image->fg_color];
+		bgcol = ((u32 *) (info->pseudo_palette))[image->bg_color];
 	} else {
 		fgcol = image->fg_color;
 		bgcol = image->bg_color;
 	}
 
 	switch (bpp) {
-		case 8:
-			fgcol |= fgcol <<8; fgcol |= fgcol <<16;
-			bgcol |= bgcol <<8; bgcol |= bgcol <<16;
-			break;
-		case 16:
-			fgcol |= fgcol <<16;
-			bgcol |= bgcol <<16;
-			break;
-		default:
-			 break;
+	case 8:
+		fgcol |= fgcol << 8;
+		bgcol |= bgcol << 8;
+	case 16:
+		fgcol |= fgcol << 16;
+		bgcol |= bgcol << 16;
+	default:
+		break;
 	}
 
-	cyblafb_sync(info);
-
-	out32(GE60,fgcol);
-	out32(GE64,bgcol);
-	out32(GE44,0xa0000000 | 1<<20 | 1<<19);
-	out32(GE08,point(image->dx,image->dy));
-	out32(GE0C,point(image->dx+image->width-1,image->dy+image->height-1));
+	out32(GEB8, basestride | ((image->dy * info->var.xres_virtual *
+				   bpp) >> 6));
+	out32(GE60, fgcol);
+	out32(GE64, bgcol);
+
+	if (!(image->dx < 2048 && (image->dx + image->width - 1) >= 2048)) {
+		u32 dds = ((image->width + 31) >> 5) * image->height;
+		out32(GE44, 0xa0000000 | 1 << 20 | 1 << 19);
+		out32(GE08, point(image->dx, 0));
+		out32(GE0C, point(image->dx + image->width - 1,
+				  image->height - 1));
+		while (dds--)
+			out32(GE9C, *pd++);
+	} else {
+		int i, j;
+		u32 ddstotal = (image->width + 31) >> 5;
+		u32 ddsleft = (2048 - image->dx + 31) >> 5;
+		u32 skipleft = ddstotal - ddsleft;
+
+		out32(GE44, 0xa0000000 | 1 << 20 | 1 << 19);
+		out32(GE08, point(image->dx, 0));
+		out32(GE0C, point(2048 - 1, image->height - 1));
+		for (i = 0; i < image->height; i++) {
+			for (j = 0; j < ddsleft; j++)
+				out32(GE9C, *pd++);
+			pd += skipleft;
+		}
 
-	while(index < index_end) {
-		const char *p = image->data + index;
-		for(i=0;i<width_dds;i++) {
-			out32(GE9C,*(u32*)p);
-			p+=4;
-			index+=4;
+		if (image->dx % 32) {
+			out32(GE44, 0xa0000000 | 1 << 20 | 1 << 19);
+			out32(GE08, point(2048, 0));
+			if (image->width > ddsleft << 5)
+				out32(GE0C, point(image->dx + (ddsleft << 5) -
+						  1, image->height - 1));
+			else
+				out32(GE0C, point(image->dx + image->width - 1,
+						  image->height - 1));
+			pd = ((u32 *) image->data) + ddstotal - skipleft - 1;
+			for (i = 0; i < image->height; i++) {
+				out32(GE9C, swab32(swab32(*pd) << ((32 -
+					    (image->dx & 31)) & 31)));
+				pd += ddstotal;
+			}
 		}
-		switch(width_dbs) {
-		case 0: break;
-		case 8:	out32(GE9C,*(u8*)p);
-			index+=1;
-			break;
-		case 16: out32(GE9C,*(u16*)p);
-			index+=2;
-			break;
-		case 24: out32(GE9C,*(u16*)p | *(u8*)(p+2)<<16);
-			index+=3;
-			break;
+
+		if (skipleft) {
+			out32(GE44, 0xa0000000 | 1 << 20 | 1 << 19);
+			out32(GE08, point(image->dx + (ddsleft << 5), 0));
+			out32(GE0C, point(image->dx + image->width - 1,
+					  image->height - 1));
+			pd = (u32 *) image->data;
+			for (i = 0; i < image->height; i++) {
+				pd += ddsleft;
+				for (j = 0; j < skipleft; j++)
+					out32(GE9C, *pd++);
+			}
 		}
 	}
 }
@@ -443,7 +526,6 @@ static int cyblafb_check_var(struct fb_var_screeninfo *var,
 			     struct fb_info *info)
 {
 	int bpp = var->bits_per_pixel;
-	int s,t,maxvyres;
 
 	//
 	// we try to support 8, 16, 24 and 32 bpp modes,
@@ -453,9 +535,9 @@ static int cyblafb_check_var(struct fb_var_screeninfo *var,
 	// (This is what tridentfb does ... will be changed in the future)
 	//
 	//
-	if ( bpp % 8 != 0 || bpp < 8 || bpp >32)
+	if (bpp % 8 != 0 || bpp < 8 || bpp > 32)
 		bpp = 8;
-	if (bpp == 24 )
+	if (bpp == 24)
 		bpp = var->bits_per_pixel = 32;
 
 	//
@@ -472,65 +554,93 @@ static int cyblafb_check_var(struct fb_var_screeninfo *var,
 		return -EINVAL;
 
 	//
-	// xres != xres_virtual is broken, fail if such an
-	// unusual mode is requested
+	// we do not allow vclk to exceed 230 MHz. If the requested
+	// vclk is too high, we default to 200 MHz
 	//
-	if (var->xres != var->xres_virtual)
-		return -EINVAL;
+	if ((bpp == 32 ? 200000000 : 100000000) / var->pixclock > 23000)
+		var->pixclock = (bpp == 32 ? 200000000 : 100000000) / 20000;
 
 	//
-	// we do not allow vclk to exceed 230 MHz
+	// enforce (h|v)sync_len limits
 	//
-	if ((bpp==32 ? 200000000 : 100000000) / var->pixclock > 23000)
-		return -EINVAL;
+	var->hsync_len &= ~7;
+	if(var->hsync_len > 248)
+		var->hsync_len = 248;
+
+	var->vsync_len &= 15;
 
 	//
-	// calc max yres_virtual that would fit in memory
-	// and max yres_virtual that could be used for scrolling
-	// and use minimum of the results as maxvyres
-	//
-	// adjust vyres_virtual to maxvyres if necessary
-	// fail if requested yres is bigger than maxvyres
+	// Enforce horizontal and vertical hardware limits.
+	// 1600x1200 is mentioned as a maximum, but higher resolutions could
+	// work with slow refresh, small margins and short sync.
 	//
-	s = (0x1fffff / (var->xres * bpp/8)) + var->yres;
-	t = info->fix.smem_len / (var->xres * bpp/8);
-	maxvyres = t < s ? t : s;
-	if (maxvyres < var->yres_virtual)
-		var->yres_virtual=maxvyres;
-	if (maxvyres < var->yres)
+	var->xres &= ~7;
+
+	if (((var->xres + var->left_margin + var->right_margin +
+			var->hsync_len) > (bpp == 32 ? 2040 : 4088)) ||
+			((var->yres + var->upper_margin + var->lower_margin +
+			var->vsync_len) > 2047))
 		return -EINVAL;
 
-	switch (bpp) {
-		case 8:
-			var->red.offset = 0;
-			var->green.offset = 0;
-			var->blue.offset = 0;
-			var->red.length = 6;
-			var->green.length = 6;
-			var->blue.length = 6;
-			break;
-		case 16:
-			var->red.offset = 11;
-			var->green.offset = 5;
-			var->blue.offset = 0;
-			var->red.length = 5;
-			var->green.length = 6;
-			var->blue.length = 5;
-			break;
-		case 32:
-			var->red.offset = 16;
-			var->green.offset = 8;
-			var->blue.offset = 0;
-			var->red.length = 8;
-			var->green.length = 8;
-			var->blue.length = 8;
-			break;
-		default:
+	if ((var->xres > 1600) || (var->yres > 1200))
+		output("Mode %dx%d exceeds documented limits.\n",
+					   var->xres, var->yres);
+	//
+	// try to be smart about (x|y)res_virtual problems.
+	//
+	if (var->xres > var->xres_virtual)
+		var->xres_virtual = var->xres;
+	if (var->yres > var->yres_virtual)
+		var->yres_virtual = var->yres;
+
+	if (bpp == 8 || bpp == 16) {
+		if (var->xres_virtual > 4088)
+			var->xres_virtual = 4088;
+	} else {
+		if (var->xres_virtual > 2040)
+			var->xres_virtual = 2040;
+	}
+	var->xres_virtual &= ~7;
+	while (var->xres_virtual * var->yres_virtual * bpp / 8 >
+	       info->fix.smem_len) {
+		if (var->yres_virtual > var->yres)
+			var->yres_virtual--;
+		else if (var->xres_virtual > var->xres)
+			var->xres_virtual -= 8;
+		else
 			return -EINVAL;
 	}
 
-	return 0;
+	switch (bpp) {
+	case 8:
+		var->red.offset = 0;
+		var->green.offset = 0;
+		var->blue.offset = 0;
+		var->red.length = 6;
+		var->green.length = 6;
+		var->blue.length = 6;
+		break;
+	case 16:
+		var->red.offset = 11;
+		var->green.offset = 5;
+		var->blue.offset = 0;
+		var->red.length = 5;
+		var->green.length = 6;
+		var->blue.length = 5;
+		break;
+	case 32:
+		var->red.offset = 16;
+		var->green.offset = 8;
+		var->blue.offset = 0;
+		var->red.length = 8;
+		var->green.length = 8;
+		var->blue.length = 8;
+		break;
+	default:
+		return -EINVAL;
+	}
 
+	return 0;
 }
 
 //=====================================================================
@@ -543,23 +653,25 @@ static int cyblafb_check_var(struct fb_var_screeninfo *var,
 // it, so it is also safe to be used here. BTW: datasheet CR0E on page
 // 90 really is CR1E, the real CRE is documented on page 72.
 //
+// BUT:
+//
+// As of internal version 0.60 we do not use vga panning any longer.
+// Vga panning did not allow us the use of all available video memory
+// and thus prevented ywrap scrolling. We do use the "right view"
+// register now.
+//
+//
 //=====================================================================
 
 static int cyblafb_pan_display(struct fb_var_screeninfo *var,
 			       struct fb_info *info)
 {
-	unsigned int offset;
+	KD_GRAPHICS_RETURN(0);
 
-	offset=(var->xoffset+(var->yoffset*var->xres))*var->bits_per_pixel/32;
 	info->var.xoffset = var->xoffset;
 	info->var.yoffset = var->yoffset;
-
-	write3X4(CR0D,offset & 0xFF);
-	write3X4(CR0C,(offset & 0xFF00) >> 8);
-	write3X4(CR1E,(read3X4(CR1E) & 0xDF) | ((offset & 0x10000) >> 11));
-	write3X4(CR27,(read3X4(CR27) & 0xF8) | ((offset & 0xE0000) >> 17));
-	write3X4(CR2B,(read3X4(CR2B) & 0xDF) | ((offset & 0x100000) >> 15));
-
+	out32(GE10, 0x80000000 | ((var->xoffset + (var->yoffset *
+		    var->xres_virtual)) * var->bits_per_pixel / 32));
 	return 0;
 }
 
@@ -578,56 +690,96 @@ static void regdump(struct cyblafb_par *par)
 		return;
 
 	printk("\n");
-	for(i=0; i<=0xff; i++) {
-		outb(i,0x3d4);
-		printk("CR%02x=%02x ",i,inb(0x3d5));
-		if (i%16==15)
+	for (i = 0; i <= 0xff; i++) {
+		outb(i, 0x3d4);
+		printk("CR%02x=%02x ", i, inb(0x3d5));
+		if (i % 16 == 15)
 			printk("\n");
 	}
 
-	outb(0x30,0x3ce);
-	outb(inb(0x3cf) | 0x40,0x3cf);
-	for(i=0; i<=0x1f; i++) {
-		if (i==0 || (i>2 && i<8) || i==0x10 || i==0x11 || i==0x16) {
-			outb(i,0x3d4);
-			printk("CR%02x=%02x ",i,inb(0x3d5));
+	outb(0x30, 0x3ce);
+	outb(inb(0x3cf) | 0x40, 0x3cf);
+	for (i = 0; i <= 0x1f; i++) {
+		if (i == 0 || (i > 2 && i < 8) || i == 0x10 || i == 0x11
+		    || i == 0x16) {
+			outb(i, 0x3d4);
+			printk("CR%02x=%02x ", i, inb(0x3d5));
 		} else
 			printk("------- ");
-		if (i%16==15)
+		if (i % 16 == 15)
 			printk("\n");
 	}
-	outb(0x30,0x3ce);
-	outb(inb(0x3cf) & 0xbf,0x3cf);
+	outb(0x30, 0x3ce);
+	outb(inb(0x3cf) & 0xbf, 0x3cf);
 
 	printk("\n");
-	for(i=0; i<=0x7f; i++) {
-		outb(i,0x3ce);
-		printk("GR%02x=%02x ",i,inb(0x3cf));
-		if (i%16==15)
+	for (i = 0; i <= 0x7f; i++) {
+		outb(i, 0x3ce);
+		printk("GR%02x=%02x ", i, inb(0x3cf));
+		if (i % 16 == 15)
 			printk("\n");
 	}
 
 	printk("\n");
-	for(i=0; i<=0xff; i++) {
-		outb(i,0x3c4);
-		printk("SR%02x=%02x ",i,inb(0x3c5));
-		if (i%16==15)
+	for (i = 0; i <= 0xff; i++) {
+		outb(i, 0x3c4);
+		printk("SR%02x=%02x ", i, inb(0x3c5));
+		if (i % 16 == 15)
 			printk("\n");
 	}
 
 	printk("\n");
-	for(i=0; i <= 0x1F; i++) {
-		inb(0x3da); // next access is index!
-		outb(i,0x3c0);
-		printk("AR%02x=%02x ",i,inb(0x3c1));
-		if (i%16==15)
+	for (i = 0; i <= 0x1F; i++) {
+		inb(0x3da);	// next access is index!
+		outb(i, 0x3c0);
+		printk("AR%02x=%02x ", i, inb(0x3c1));
+		if (i % 16 == 15)
 			printk("\n");
 	}
 	printk("\n");
 
-	inb(0x3DA);			// reset internal flag to 3c0 index
-	outb(0x20,0x3C0);		// enable attr
+	inb(0x3DA);		// reset internal flag to 3c0 index
+	outb(0x20, 0x3C0);	// enable attr
+
+	return;
+}
+
+//=======================================================================
+//
+// Save State
+//
+// This function is called while a switch to KD_TEXT is in progress,
+// before any of the other functions are called.
+//
+//=======================================================================
 
+static void cyblafb_save_state(struct fb_info *info)
+{
+	struct cyblafb_par *par = info->par;
+	if (verbosity > 0)
+		output("Switching to KD_TEXT\n");
+	disabled = 0;
+	regdump(par);
+	enable_mmio();
+	return;
+}
+
+//=======================================================================
+//
+// Restore State
+//
+// This function is called while a switch to KD_GRAPHICS is in progress,
+// We have to turn on vga style panning registers again because the
+// trident driver of X does not know about GE10.
+//
+//=======================================================================
+
+static void cyblafb_restore_state(struct fb_info *info)
+{
+	if (verbosity > 0)
+		output("Switching to KD_GRAPHICS\n");
+	out32(GE10, 0);
+	disabled = 1;
 	return;
 }
 
@@ -640,32 +792,34 @@ static void regdump(struct cyblafb_par *par)
 static int cyblafb_set_par(struct fb_info *info)
 {
 	struct cyblafb_par *par = info->par;
-	u32
-	htotal,hdispend,hsyncstart,hsyncend,hblankstart,hblankend,preendfetch,
-		vtotal,vdispend,vsyncstart,vsyncend,vblankstart,vblankend;
+	u32 htotal, hdispend, hsyncstart, hsyncend, hblankstart,
+	    hblankend, preendfetch, vtotal, vdispend, vsyncstart,
+	    vsyncend, vblankstart, vblankend;
 	struct fb_var_screeninfo *var = &info->var;
 	int bpp = var->bits_per_pixel;
 	int i;
 
+	KD_GRAPHICS_RETURN(0);
+
 	if (verbosity > 0)
 		output("Switching to new mode: "
 		       "fbset -g %d %d %d %d %d -t %d %d %d %d %d %d %d\n",
-			var->xres,var->yres,var->xres_virtual,
-			var->yres_virtual,var->bits_per_pixel,var->pixclock,
-			var->left_margin,var->right_margin,var->upper_margin,
-			var->lower_margin,var->hsync_len,var->vsync_len);
+		       var->xres, var->yres, var->xres_virtual,
+		       var->yres_virtual, var->bits_per_pixel, var->pixclock,
+		       var->left_margin, var->right_margin, var->upper_margin,
+		       var->lower_margin, var->hsync_len, var->vsync_len);
 
 	htotal = (var->xres + var->left_margin + var->right_margin +
-						 var->hsync_len) / 8 - 5;
-	hdispend = var->xres/8 - 1;
-	hsyncstart = (var->xres + var->right_margin)/8;
-	hsyncend = var->hsync_len/8;
+		  var->hsync_len) / 8 - 5;
+	hdispend = var->xres / 8 - 1;
+	hsyncstart = (var->xres + var->right_margin) / 8;
+	hsyncend = var->hsync_len / 8;
 	hblankstart = hdispend + 1;
 	hblankend = htotal + 3; // should be htotal + 5, bios does it this way
-	preendfetch = ((var->xres >> 3) + 1) * ((bpp+1) >> 3);
+	preendfetch = ((var->xres >> 3) + 1) * ((bpp + 1) >> 3);
 
 	vtotal = var->yres + var->upper_margin + var->lower_margin +
-						 var->vsync_len - 2;
+							var->vsync_len - 2;
 	vdispend = var->yres - 1;
 	vsyncstart = var->yres + var->lower_margin;
 	vblankstart = var->yres;
@@ -674,101 +828,99 @@ static int cyblafb_set_par(struct fb_info *info)
 
 	enable_mmio();		// necessary! ... check X ...
 
-	write3X4(CR11,read3X4(CR11) & 0x7F); // unlock cr00 .. cr07
+	write3X4(CR11, read3X4(CR11) & 0x7F);	// unlock cr00 .. cr07
 
-	write3CE(GR30,8);
+	write3CE(GR30, 8);
 
 	if ((displaytype == DISPLAY_FP) && var->xres < nativex) {
 
 		// stretch or center ?
 
-		out8(0x3C2,0xEB);
+		out8(0x3C2, 0xEB);
 
-		write3CE(GR30,read3CE(GR30) | 0x81); // shadow mode on
+		write3CE(GR30, read3CE(GR30) | 0x81);	// shadow mode on
 
 		if (center) {
-			write3CE(GR52,(read3CE(GR52) & 0x7C) | 0x80);
-			write3CE(GR53,(read3CE(GR53) & 0x7C) | 0x80);
-		}
-		else if (s