Merge branch 'topic/asoc' into for-linus

[linux-2.6-block.git] / drivers / video / aty / mach64_ct.c

/*
 *  ATI Mach64 CT/VT/GT/LT Support
 */

#include <linux/fb.h>
#include <linux/delay.h>
#include <asm/io.h>
#include <video/mach64.h>
#include "atyfb.h"
#ifdef CONFIG_PPC
#include <asm/machdep.h>
#endif

#undef DEBUG

static int aty_valid_pll_ct (const struct fb_info *info, u32 vclk_per, struct pll_ct *pll);
static int aty_dsp_gt       (const struct fb_info *info, u32 bpp, struct pll_ct *pll);
static int aty_var_to_pll_ct(const struct fb_info *info, u32 vclk_per, u32 bpp, union aty_pll *pll);
static u32 aty_pll_to_var_ct(const struct fb_info *info, const union aty_pll *pll);

u8 aty_ld_pll_ct(int offset, const struct atyfb_par *par)
{
	u8 res;

	/* write addr byte */
	aty_st_8(CLOCK_CNTL_ADDR, (offset << 2) & PLL_ADDR, par);
	/* read the register value */
	res = aty_ld_8(CLOCK_CNTL_DATA, par);
	return res;
}

static void aty_st_pll_ct(int offset, u8 val, const struct atyfb_par *par)
{
	/* write addr byte */
	aty_st_8(CLOCK_CNTL_ADDR, ((offset << 2) & PLL_ADDR) | PLL_WR_EN, par);
	/* write the register value */
	aty_st_8(CLOCK_CNTL_DATA, val & PLL_DATA, par);
	aty_st_8(CLOCK_CNTL_ADDR, ((offset << 2) & PLL_ADDR) & ~PLL_WR_EN, par);
}

/*
 * by Daniel Mantione
 *                                  <daniel.mantione@freepascal.org>
 *
 *
 * ATI Mach64 CT clock synthesis description.
 *
 * All clocks on the Mach64 can be calculated using the same principle:
 *
 *       XTALIN * x * FB_DIV
 * CLK = ----------------------
 *       PLL_REF_DIV * POST_DIV
 *
 * XTALIN is a fixed speed clock. Common speeds are 14.31 MHz and 29.50 MHz.
 * PLL_REF_DIV can be set by the user, but is the same for all clocks.
 * FB_DIV can be set by the user for each clock individually, it should be set
 * between 128 and 255, the chip will generate a bad clock signal for too low
 * values.
 * x depends on the type of clock; usually it is 2, but for the MCLK it can also
 * be set to 4.
 * POST_DIV can be set by the user for each clock individually, Possible values
 * are 1,2,4,8 and for some clocks other values are available too.
 * CLK is of course the clock speed that is generated.
 *
 * The Mach64 has these clocks:
 *
 * MCLK			The clock rate of the chip
 * XCLK			The clock rate of the on-chip memory
 * VCLK0		First pixel clock of first CRT controller
 * VCLK1    Second pixel clock of first CRT controller
 * VCLK2		Third pixel clock of first CRT controller
 * VCLK3    Fourth pixel clock of first CRT controller
 * VCLK			Selected pixel clock, one of VCLK0, VCLK1, VCLK2, VCLK3
 * V2CLK		Pixel clock of the second CRT controller.
 * SCLK			Multi-purpose clock
 *
 * - MCLK and XCLK use the same FB_DIV
 * - VCLK0 .. VCLK3 use the same FB_DIV
 * - V2CLK is needed when the second CRTC is used (can be used for dualhead);
 *   i.e. CRT monitor connected to laptop has different resolution than built
 *   in LCD monitor.
 * - SCLK is not available on all cards; it is know to exist on the Rage LT-PRO,
 *   Rage XL and Rage Mobility. It is know not to exist on the Mach64 VT.
 * - V2CLK is not available on all cards, most likely only the Rage LT-PRO,
 *   the Rage XL and the Rage Mobility
 *
 * SCLK can be used to:
 * - Clock the chip instead of MCLK
 * - Replace XTALIN with a user defined frequency
 * - Generate the pixel clock for the LCD monitor (instead of VCLK)
 */

 /*
  * It can be quite hard to calculate XCLK and MCLK if they don't run at the
  * same frequency. Luckily, until now all cards that need asynchrone clock
  * speeds seem to have SCLK.
  * So this driver uses SCLK to clock the chip and XCLK to clock the memory.
  */

/* ------------------------------------------------------------------------- */

/*
 *  PLL programming (Mach64 CT family)
 *
 *
 * This procedure sets the display fifo. The display fifo is a buffer that
 * contains data read from the video memory that waits to be processed by
 * the CRT controller.
 *
 * On the more modern Mach64 variants, the chip doesn't calculate the
 * interval after which the display fifo has to be reloaded from memory
 * automatically, the driver has to do it instead.
 */

#define Maximum_DSP_PRECISION 7
static u8 postdividers[] = {1,2,4,8,3};

static int aty_dsp_gt(const struct fb_info *info, u32 bpp, struct pll_ct *pll)
{
	u32 dsp_off, dsp_on, dsp_xclks;
	u32 multiplier, divider, ras_multiplier, ras_divider, tmp;
	u8 vshift, xshift;
	s8 dsp_precision;

	multiplier = ((u32)pll->mclk_fb_div) * pll->vclk_post_div_real;
	divider = ((u32)pll->vclk_fb_div) * pll->xclk_ref_div;

	ras_multiplier = pll->xclkmaxrasdelay;
	ras_divider = 1;

	if (bpp>=8)
		divider = divider * (bpp >> 2);

	vshift = (6 - 2) - pll->xclk_post_div;	/* FIFO is 64 bits wide in accelerator mode ... */

	if (bpp == 0)
		vshift--;	/* ... but only 32 bits in VGA mode. */

#ifdef CONFIG_FB_ATY_GENERIC_LCD
	if (pll->xres != 0) {
		struct atyfb_par *par = (struct atyfb_par *) info->par;

		multiplier = multiplier * par->lcd_width;
		divider = divider * pll->xres & ~7;

		ras_multiplier = ras_multiplier * par->lcd_width;
		ras_divider = ras_divider * pll->xres & ~7;
	}
#endif
	/* If we don't do this, 32 bits for multiplier & divider won't be
	enough in certain situations! */
	while (((multiplier | divider) & 1) == 0) {
		multiplier = multiplier >> 1;
		divider = divider >> 1;
	}

	/* Determine DSP precision first */
	tmp = ((multiplier * pll->fifo_size) << vshift) / divider;

	for (dsp_precision = -5;  tmp;  dsp_precision++)
		tmp >>= 1;
	if (dsp_precision < 0)
		dsp_precision = 0;
	else if (dsp_precision > Maximum_DSP_PRECISION)
		dsp_precision = Maximum_DSP_PRECISION;

	xshift = 6 - dsp_precision;
	vshift += xshift;

	/* Move on to dsp_off */
	dsp_off = ((multiplier * (pll->fifo_size - 1)) << vshift) / divider -
		(1 << (vshift - xshift));

/*    if (bpp == 0)
        dsp_on = ((multiplier * 20 << vshift) + divider) / divider;
    else */
	{
		dsp_on = ((multiplier << vshift) + divider) / divider;
		tmp = ((ras_multiplier << xshift) + ras_divider) / ras_divider;
		if (dsp_on < tmp)
		dsp_on = tmp;
		dsp_on = dsp_on + (tmp * 2) + (pll->xclkpagefaultdelay << xshift);
	}

	/* Calculate rounding factor and apply it to dsp_on */
	tmp = ((1 << (Maximum_DSP_PRECISION - dsp_precision)) - 1) >> 1;
	dsp_on = ((dsp_on + tmp) / (tmp + 1)) * (tmp + 1);

	if (dsp_on >= ((dsp_off / (tmp + 1)) * (tmp + 1))) {
		dsp_on = dsp_off - (multiplier << vshift) / divider;
		dsp_on = (dsp_on / (tmp + 1)) * (tmp + 1);
	}

	/* Last but not least:  dsp_xclks */
	dsp_xclks = ((multiplier << (vshift + 5)) + divider) / divider;

	/* Get register values. */
	pll->dsp_on_off = (dsp_on << 16) + dsp_off;
	pll->dsp_config = (dsp_precision << 20) | (pll->dsp_loop_latency << 16) | dsp_xclks;
#ifdef DEBUG
	printk("atyfb(%s): dsp_config 0x%08x, dsp_on_off 0x%08x\n",
		__func__, pll->dsp_config, pll->dsp_on_off);
#endif
	return 0;
}

static int aty_valid_pll_ct(const struct fb_info *info, u32 vclk_per, struct pll_ct *pll)
{
	u32 q;
	struct atyfb_par *par = (struct atyfb_par *) info->par;
	int pllvclk;

	/* FIXME: use the VTB/GTB /{3,6,12} post dividers if they're better suited */
	q = par->ref_clk_per * pll->pll_ref_div * 4 / vclk_per;
	if (q < 16*8 || q > 255*8) {
		printk(KERN_CRIT "atyfb: vclk out of range\n");
		return -EINVAL;
	} else {
		pll->vclk_post_div  = (q < 128*8);
		pll->vclk_post_div += (q <  64*8);
		pll->vclk_post_div += (q <  32*8);
	}
	pll->vclk_post_div_real = postdividers[pll->vclk_post_div];
	//    pll->vclk_post_div <<= 6;
	pll->vclk_fb_div = q * pll->vclk_post_div_real / 8;
	pllvclk = (1000000 * 2 * pll->vclk_fb_div) /
		(par->ref_clk_per * pll->pll_ref_div);