cyblafb.c

Linux环境下视频显示卡设备的驱动程序源代码

/*
 * Frame buffer driver for Trident Cyberblade/i1 graphics core
 *
 * Copyright 2005 Knut Petersen <[e-mail]>
 *
 * CREDITS:
 *	tridentfb.c by Jani Monoses
 *	see files above for further credits
 *
 */

#define CYBLAFB_DEBUG 0
#define CYBLAFB_KD_GRAPHICS_QUIRK 1

#define CYBLAFB_PIXMAPSIZE 8192

#include <linux/module.h>
#include <linux/string.h>
#include <linux/fb.h>
#include <linux/init.h>
#include <linux/pci.h>
#include <asm/types.h>
#include <video/cyblafb.h>

#define VERSION "0.62"

struct cyblafb_par {
	u32 pseudo_pal[16];
	struct fb_ops ops;
};

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,
};

static char *mode __devinitdata = NULL;
static int bpp __devinitdata = 8;
static int ref __devinitdata = 75;
static int fp __devinitdata;
static int crt __devinitdata;
static int memsize __devinitdata;

static int basestride;
static int vesafb;
static int nativex;
static int center;
static int stretch;
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);

//=========================================
//
// 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

//=========================================
//
// Port access macros for memory mapped io
//
//=========================================

#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)

//======================================
//
// Hardware access inline functions
//
//======================================

static inline u8 read3X4(u32 reg)
{
	out8(0x3D4, reg);
	return in8(0x3D5);
}

static inline u8 read3C4(u32 reg)
{
	out8(0x3C4, reg);
	return in8(0x3C5);
}

static inline u8 read3CE(u32 reg)
{
	out8(0x3CE, reg);
	return in8(0x3CF);
}

static inline void write3X4(u32 reg, u8 val)
{
	out8(0x3D4, reg);
	out8(0x3D5, val);
}

static inline void write3C4(u32 reg, u8 val)
{
	out8(0x3C4, reg);
	out8(0x3C5, val);
}

static inline void write3CE(u32 reg, u8 val)
{
	out8(0x3CE, reg);
	out8(0x3CF, val);
}

static inline void write3C0(u32 reg, u8 val)
{
	in8(0x3DA);		// read to reset index
	out8(0x3C0, reg);
	out8(0x3C0, val);
}

//=================================================
//
// Enable memory mapped io and unprotect registers
//
//=================================================

static void enable_mmio(void)
{
	u8 tmp;

	outb(0x0B, 0x3C4);
	inb(0x3C5);		// Set NEW mode
	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(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
}

//=================================================
//
// Set pixel clock VCLK1
// - 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;

	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++) {
			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 > 0)
		output("pixclock = %d.%02d MHz, k/m/n %x %x %x\n",
		       freq / 100, freq % 100, (hi & 0xc0) >> 6, hi & 0x3f, lo);
}

//================================================
//
// Cyberblade specific Graphics Engine (GE) setup
//
//================================================

static void cyblafb_setup_GE(int pitch, int bpp)
{
	KD_GRAPHICS_RETURN();

	switch (bpp) {
	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 by toggling
	out32(GE24, 0); 	//   d7 of GE24
	write3X4(CR2D, 0x00);	// GE Timinigs, no delays
	out32(GE6C, 0); 	// Pattern and Style, p 129, ok
}

//=====================================================================
//
// 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)
{
	u32 status, i = 100000;

	KD_GRAPHICS_RETURN(0);

	while (((status = in32(GE20)) & 0xFe800000) && i != 0)
		i--;

	if (i == 0) {
		enable_mmio();
		i = 1000000;
		while (((status = in32(GE20)) & 0xFA800000) && i != 0)
			i--;
		if (i == 0) {
			output("GE Timeout, status: %x\n", status);
			if (status & 0x80000000)
				output("Bresenham Engine : Busy\n");
			if (status & 0x40000000)
				output("Setup Engine     : Busy\n");
			if (status & 0x20000000)
				output("SP / DPE         : Busy\n");
			if (status & 0x10000000)
				output("Memory Interface : Busy\n");
			if (status & 0x08000000)
				output("Com Lst Proc     : Busy\n");
			if (status & 0x04000000)
				output("Block Write      : Busy\n");
			if (status & 0x02000000)
				output("Command Buffer   : Full\n");
			if (status & 0x01000000)
				output("RESERVED         : Busy\n");
			if (status & 0x00800000)
				output("PCI Write Buffer : Busy\n");
			cyblafb_setup_GE(info->var.xres,
					 info->var.bits_per_pixel);
		}
	}

	return 0;
}

//==============================
//
// Cyberblade specific fillrect
//
//==============================

static void cyblafb_fillrect(struct fb_info *info, const struct fb_fillrect *fr)
{
	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;
	}

	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)
{
	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);

	if ((ca->sy > ca->dy) || ((ca->sy == ca->dy) && (ca->sx > ca->dx)))
		direction = 0;
	else
		direction = 2;

	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 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;

	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];
	} else {
		fgcol = image->fg_color;
		bgcol = image->bg_color;
	}

	switch (bpp) {
	case 8:
		fgcol |= fgcol << 8;
		bgcol |= bgcol << 8;
	case 16:
		fgcol |= fgcol << 16;
		bgcol |= bgcol << 16;
	default:
		break;
	}

	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;
		}

		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;
			}
		}

		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++);
			}
		}
	}
}

//==========================================================
//
// Check if video mode is acceptable. We change var->??? if
// video mode is slightly off or return error otherwise.
// info->??? must not be changed!
//
//==========================================================

static int cyblafb_check_var(struct fb_var_screeninfo *var,
			     struct fb_info *info)
{
	int bpp = var->bits_per_pixel;

	//
	// we try to support 8, 16, 24 and 32 bpp modes,
	// default to 8
	//
	// there is a 24 bpp mode, but for now we change requests to 32 bpp
	// (This is what tridentfb does ... will be changed in the future)
	//
	//
	if (bpp % 8 != 0 || bpp < 8 || bpp > 32)
		bpp = 8;
	if (bpp == 24)
		bpp = var->bits_per_pixel = 32;

	//
	// interlaced modes are broken, fail if one is requested
	//
	if (var->vmode & FB_VMODE_INTERLACED)
		return -EINVAL;

	//
	// fail if requested resolution is higher than physical
	// flatpanel resolution
	//
	if ((displaytype == DISPLAY_FP) && nativex && var->xres > nativex)
		return -EINVAL;

	//
	// we do not allow vclk to exceed 230 MHz. If the requested
	// vclk is too high, we default to 200 MHz
	//
	if ((bpp == 32 ? 200000000 : 100000000) / var->pixclock > 23000)
		var->pixclock = (bpp == 32 ? 200000000 : 100000000) / 20000;

	//