[deliverable/linux.git] / arch / x86 / kernel / cpu / centaur.c

#include <linux/kernel.h>
#include <linux/init.h>
#include <linux/bitops.h>

#include <asm/processor.h>
#include <asm/msr.h>
#include <asm/e820.h>
#include <asm/mtrr.h>

#include "cpu.h"

#ifdef CONFIG_X86_OOSTORE

static u32 __cpuinit power2(u32 x)
{
	u32 s = 1;

	while (s <= x)
		s <<= 1;

	return s >>= 1;
}


/*
 * Set up an actual MCR
 */
static void __cpuinit centaur_mcr_insert(int reg, u32 base, u32 size, int key)
{
	u32 lo, hi;

	hi = base & ~0xFFF;
	lo = ~(size-1);		/* Size is a power of 2 so this makes a mask */
	lo &= ~0xFFF;		/* Remove the ctrl value bits */
	lo |= key;		/* Attribute we wish to set */
	wrmsr(reg+MSR_IDT_MCR0, lo, hi);
	mtrr_centaur_report_mcr(reg, lo, hi);	/* Tell the mtrr driver */
}

/*
 * Figure what we can cover with MCR's
 *
 * Shortcut: We know you can't put 4Gig of RAM on a winchip
 */
static u32 __cpuinit ramtop(void)
{
	u32 clip = 0xFFFFFFFFUL;
	u32 top = 0;
	int i;

	for (i = 0; i < e820.nr_map; i++) {
		unsigned long start, end;

		if (e820.map[i].addr > 0xFFFFFFFFUL)
			continue;
		/*
		 * Don't MCR over reserved space. Ignore the ISA hole
		 * we frob around that catastrophe already
		 */
		if (e820.map[i].type == E820_RESERVED) {
			if (e820.map[i].addr >= 0x100000UL &&
			    e820.map[i].addr < clip)
				clip = e820.map[i].addr;
			continue;
		}
		start = e820.map[i].addr;
		end = e820.map[i].addr + e820.map[i].size;
		if (start >= end)
			continue;
		if (end > top)
			top = end;
	}
	/*
	 * Everything below 'top' should be RAM except for the ISA hole.
	 * Because of the limited MCR's we want to map NV/ACPI into our
	 * MCR range for gunk in RAM
	 *
	 * Clip might cause us to MCR insufficient RAM but that is an
	 * acceptable failure mode and should only bite obscure boxes with
	 * a VESA hole at 15Mb
	 *
	 * The second case Clip sometimes kicks in is when the EBDA is marked
	 * as reserved. Again we fail safe with reasonable results
	 */
	if (top > clip)
		top = clip;

	return top;
}

/*
 * Compute a set of MCR's to give maximum coverage
 */
static int __cpuinit centaur_mcr_compute(int nr, int key)
{
	u32 mem = ramtop();
	u32 root = power2(mem);
	u32 base = root;
	u32 top = root;
	u32 floor = 0;
	int ct = 0;

	while (ct < nr) {
		u32 fspace = 0;
		u32 high;
		u32 low;

		/*
		 * Find the largest block we will fill going upwards
		 */
		high = power2(mem-top);

		/*
		 * Find the largest block we will fill going downwards
		 */
		low = base/2;

		/*
		 * Don't fill below 1Mb going downwards as there
		 * is an ISA hole in the way.
		 */
		if (base <= 1024*1024)
			low = 0;

		/*
		 * See how much space we could cover by filling below
		 * the ISA hole
		 */

		if (floor == 0)
			fspace = 512*1024;
		else if (floor == 512*1024)
			fspace = 128*1024;

		/* And forget ROM space */

		/*
		 * Now install the largest coverage we get
		 */
		if (fspace > high && fspace > low) {
			centaur_mcr_insert(ct, floor, fspace, key);
			floor += fspace;
		} else if (high > low) {
			centaur_mcr_insert(ct, top, high, key);
			top += high;
		} else if (low > 0) {
			base -= low;
			centaur_mcr_insert(ct, base, low, key);
		} else
			break;
		ct++;
	}
	/*
	 * We loaded ct values. We now need to set the mask. The caller
	 * must do this bit.
	 */
	return ct;
}

static void __cpuinit centaur_create_optimal_mcr(void)
{
	int used;
	int i;

	/*
	 * Allocate up to 6 mcrs to mark as much of ram as possible
	 * as write combining and weak write ordered.
	 *
	 * To experiment with: Linux never uses stack operations for
	 * mmio spaces so we could globally enable stack operation wc
	 *
	 * Load the registers with type 31 - full write combining, all
	 * writes weakly ordered.
	 */
	used = centaur_mcr_compute(6, 31);

	/*
	 * Wipe unused MCRs
	 */
	for (i = used; i < 8; i++)
		wrmsr(MSR_IDT_MCR0+i, 0, 0);
}

static void __cpuinit winchip2_create_optimal_mcr(void)
{
	u32 lo, hi;
	int used;
	int i;

	/*
	 * Allocate up to 6 mcrs to mark as much of ram as possible
	 * as write combining, weak store ordered.
	 *
	 * Load the registers with type 25
	 *	8	-	weak write ordering
	 *	16	-	weak read ordering
	 *	1	-	write combining
	 */
	used = centaur_mcr_compute(6, 25);

	/*
	 * Mark the registers we are using.
	 */
	rdmsr(MSR_IDT_MCR_CTRL, lo, hi);
	for (i = 0; i < used; i++)
		lo |= 1<<(9+i);
	wrmsr(MSR_IDT_MCR_CTRL, lo, hi);

	/*
	 * Wipe unused MCRs
	 */

	for (i = used; i < 8; i++)
		wrmsr(MSR_IDT_MCR0+i, 0, 0);
}

/*
 * Handle the MCR key on the Winchip 2.
 */
static void __cpuinit winchip2_unprotect_mcr(void)
{
	u32 lo, hi;
	u32 key;

	rdmsr(MSR_IDT_MCR_CTRL, lo, hi);
	lo &= ~0x1C0;	/* blank bits 8-6 */
	key = (lo>>17) & 7;
	lo |= key<<6;	/* replace with unlock key */
	wrmsr(MSR_IDT_MCR_CTRL, lo, hi);
}

static void __cpuinit winchip2_protect_mcr(void)
{
	u32 lo, hi;

	rdmsr(MSR_IDT_MCR_CTRL, lo, hi);
	lo &= ~0x1C0;	/* blank bits 8-6 */
	wrmsr(MSR_IDT_MCR_CTRL, lo, hi);
}
#endif /* CONFIG_X86_OOSTORE */

#define ACE_PRESENT	(1 << 6)
#define ACE_ENABLED	(1 << 7)
#define ACE_FCR		(1 << 28)	/* MSR_VIA_FCR */

#define RNG_PRESENT	(1 << 2)
#define RNG_ENABLED	(1 << 3)
#define RNG_ENABLE	(1 << 6)	/* MSR_VIA_RNG */

static void __cpuinit init_c3(struct cpuinfo_x86 *c)
{
	u32  lo, hi;

	/* Test for Centaur Extended Feature Flags presence */
	if (cpuid_eax(0xC0000000) >= 0xC0000001) {
		u32 tmp = cpuid_edx(0xC0000001);

		/* enable ACE unit, if present and disabled */
		if ((tmp & (ACE_PRESENT | ACE_ENABLED)) == ACE_PRESENT) {
			rdmsr(MSR_VIA_FCR, lo, hi);
			lo |= ACE_FCR;		/* enable ACE unit */
			wrmsr(MSR_VIA_FCR, lo, hi);
			printk(KERN_INFO "CPU: Enabled ACE h/w crypto\n");
		}

		/* enable RNG unit, if present and disabled */
		if ((tmp & (RNG_PRESENT | RNG_ENABLED)) == RNG_PRESENT) {
			rdmsr(MSR_VIA_RNG, lo, hi);
			lo |= RNG_ENABLE;	/* enable RNG unit */
			wrmsr(MSR_VIA_RNG, lo, hi);
			printk(KERN_INFO "CPU: Enabled h/w RNG\n");
		}

		/* store Centaur Extended Feature Flags as
		 * word 5 of the CPU capability bit array
		 */
		c->x86_capability[5] = cpuid_edx(0xC0000001);
	}

	/* Cyrix III family needs CX8 & PGE explicitly enabled. */
	if (c->x86_model >= 6 && c->x86_model <= 9) {
		rdmsr(MSR_VIA_FCR, lo, hi);
		lo |= (1<<1 | 1<<7);
		wrmsr(MSR_VIA_FCR, lo, hi);
		set_cpu_cap(c, X86_FEATURE_CX8);
	}

	/* Before Nehemiah, the C3's had 3dNOW! */
	if (c->x86_model >= 6 && c->x86_model < 9)
		set_cpu_cap(c, X86_FEATURE_3DNOW);

	get_model_name(c);
	display_cacheinfo(c);
}

enum {
		ECX8		= 1<<1,
		EIERRINT	= 1<<2,
		DPM		= 1<<3,
		DMCE		= 1<<4,
		DSTPCLK		= 1<<5,
		ELINEAR		= 1<<6,
		DSMC		= 1<<7,
		DTLOCK		= 1<<8,
		EDCTLB		= 1<<8,
		EMMX		= 1<<9,
		DPDC		= 1<<11,
		EBRPRED		= 1<<12,
		DIC		= 1<<13,
		DDC		= 1<<14,
		DNA		= 1<<15,
		ERETSTK		= 1<<16,
		E2MMX		= 1<<19,
		EAMD3D		= 1<<20,
};

static void __cpuinit early_init_centaur(struct cpuinfo_x86 *c)
{
	switch (c->x86) {
	case 5:
		/* Emulate MTRRs using Centaur's MCR. */
		set_cpu_cap(c, X86_FEATURE_CENTAUR_MCR);
		break;
	}
}

static void __cpuinit init_centaur(struct cpuinfo_x86 *c)
{

	char *name;
	u32  fcr_set = 0;
	u32  fcr_clr = 0;
	u32  lo, hi, newlo;
	u32  aa, bb, cc, dd;

	/*
	 * Bit 31 in normal CPUID used for nonstandard 3DNow ID;
	 * 3DNow is IDd by bit 31 in extended CPUID (1*32+31) anyway
	 */
	clear_cpu_cap(c, 0*32+31);

	switch (c->x86) {
	case 5:
		switch (c->x86_model) {
		case 4:
			name = "C6";
			fcr_set = ECX8|DSMC|EDCTLB|EMMX|ERETSTK;
			fcr_clr = DPDC;
			printk(KERN_NOTICE "Disabling bugged TSC.\n");
			clear_cpu_cap(c, X86_FEATURE_TSC);
#ifdef CONFIG_X86_OOSTORE
			centaur_create_optimal_mcr();
			/*
			 * Enable:
			 *	write combining on non-stack, non-string
			 *	write combining on string, all types
			 *	weak write ordering
			 *
			 * The C6 original lacks weak read order
			 *
			 * Note 0x120 is write only on Winchip 1
			 */
			wrmsr(MSR_IDT_MCR_CTRL, 0x01F0001F, 0);
#endif
			break;
		case 8:
			switch (c->x86_mask) {
			default:
			name = "2";
				break;
			case 7 ... 9:
				name = "2A";
				break;
			case 10 ... 15:
				name = "2B";
				break;
			}
			fcr_set = ECX8|DSMC|DTLOCK|EMMX|EBRPRED|ERETSTK|
				  E2MMX|EAMD3D;
			fcr_clr = DPDC;
#ifdef CONFIG_X86_OOSTORE
			winchip2_unprotect_mcr();
			winchip2_create_optimal_mcr();
			rdmsr(MSR_IDT_MCR_CTRL, lo, hi);
			/*
			 * Enable:
			 *	write combining on non-stack, non-string
			 *	write combining on string, all types
			 *	weak write ordering
			 */
			lo |= 31;
			wrmsr(MSR_IDT_MCR_CTRL, lo, hi);
			winchip2_protect_mcr();
#endif
			break;
		case 9:
			name = "3";
			fcr_set = ECX8|DSMC|DTLOCK|EMMX|EBRPRED|ERETSTK|
				  E2MMX|EAMD3D;
			fcr_clr = DPDC;
#ifdef CONFIG_X86_OOSTORE
			winchip2_unprotect_mcr();
			winchip2_create_optimal_mcr();
			rdmsr(MSR_IDT_MCR_CTRL, lo, hi);
			/*
			 * Enable:
			 *	write combining on non-stack, non-string
			 *	write combining on string, all types
			 *	weak write ordering
			 */
			lo |= 31;
			wrmsr(MSR_IDT_MCR_CTRL, lo, hi);
			winchip2_protect_mcr();
#endif
			break;
		default:
			name = "??";
		}

		rdmsr(MSR_IDT_FCR1, lo, hi);
		newlo = (lo|fcr_set) & (~fcr_clr);

		if (newlo != lo) {
			printk(KERN_INFO "Centaur FCR was 0x%X now 0x%X\n",
				lo, newlo);
			wrmsr(MSR_IDT_FCR1, newlo, hi);
		} else {
			printk(KERN_INFO "Centaur FCR is 0x%X\n", lo);
		}
		/* Emulate MTRRs using Centaur's MCR. */
		set_cpu_cap(c, X86_FEATURE_CENTAUR_MCR);
		/* Report CX8 */
		set_cpu_cap(c, X86_FEATURE_CX8);
		/* Set 3DNow! on Winchip 2 and above. */
		if (c->x86_model >= 8)
			set_cpu_cap(c, X86_FEATURE_3DNOW);
		/* See if we can find out some more. */
		if (cpuid_eax(0x80000000) >= 0x80000005) {
			/* Yes, we can. */
			cpuid(0x80000005, &aa, &bb, &cc, &dd);
			/* Add L1 data and code cache sizes. */
			c->x86_cache_size = (cc>>24)+(dd>>24);
		}
		sprintf(c->x86_model_id, "WinChip %s", name);
		break;

	case 6:
		init_c3(c);
		break;
	}
}

static unsigned int __cpuinit
centaur_size_cache(struct cpuinfo_x86 *c, unsigned int size)
{
	/* VIA C3 CPUs (670-68F) need further shifting. */
	if ((c->x86 == 6) && ((c->x86_model == 7) || (c->x86_model == 8)))
		size >>= 8;

	/*
	 * There's also an erratum in Nehemiah stepping 1, which
	 * returns '65KB' instead of '64KB'
	 *  - Note, it seems this may only be in engineering samples.
	 */
	if ((c->x86 == 6) && (c->x86_model == 9) &&
				(c->x86_mask == 1) && (size == 65))
		size -= 1;

	return size;
}

static struct cpu_dev centaur_cpu_dev __cpuinitdata = {
	.c_vendor	= "Centaur",
	.c_ident	= { "CentaurHauls" },
	.c_early_init	= early_init_centaur,
	.c_init		= init_centaur,
	.c_size_cache	= centaur_size_cache,
	.c_x86_vendor	= X86_VENDOR_CENTAUR,
};

cpu_dev_register(centaur_cpu_dev);
Commit	Line	Data
1da177e4 LT	1	#include <linux/kernel.h>
	2	#include <linux/init.h>
	3	#include <linux/bitops.h>
edc05e6d	4
1da177e4 LT	5	#include <asm/processor.h>
	6	#include <asm/msr.h>
	7	#include <asm/e820.h>
52f4a91a	8	#include <asm/mtrr.h>
edc05e6d	9
1da177e4 LT	10	#include "cpu.h"
	11
	12	#ifdef CONFIG_X86_OOSTORE
	13
b4af3f7c	14	static u32 __cpuinit power2(u32 x)
1da177e4	15	{
29a9994b	16	u32 s = 1;
edc05e6d IM	17
edc05e6d IM	18	while (s <= x)
29a9994b	19	s <<= 1;
edc05e6d	20
29a9994b	21	return s >>= 1;
1da177e4 LT	22	}
	23
	24
	25	/*
edc05e6d	26	* Set up an actual MCR
1da177e4	27	*/
b4af3f7c	28	static void __cpuinit centaur_mcr_insert(int reg, u32 base, u32 size, int key)
1da177e4 LT	29	{
1da177e4 LT	30	u32 lo, hi;
29a9994b	31
1da177e4 LT	32	hi = base & ~0xFFF;
	33	lo = ~(size-1); /* Size is a power of 2 so this makes a mask */
	34	lo &= ~0xFFF; /* Remove the ctrl value bits */
	35	lo \|= key; /* Attribute we wish to set */
	36	wrmsr(reg+MSR_IDT_MCR0, lo, hi);
	37	mtrr_centaur_report_mcr(reg, lo, hi); /* Tell the mtrr driver */
	38	}
	39
	40	/*
edc05e6d	41	* Figure what we can cover with MCR's
1da177e4	42	*
edc05e6d	43	* Shortcut: We know you can't put 4Gig of RAM on a winchip
1da177e4	44	*/
edc05e6d	45	static u32 __cpuinit ramtop(void)
1da177e4	46	{
1da177e4	47	u32 clip = 0xFFFFFFFFUL;
edc05e6d IM	48	u32 top = 0;
edc05e6d IM	49	int i;
29a9994b	50
1da177e4 LT	51	for (i = 0; i < e820.nr_map; i++) {
	52	unsigned long start, end;
	53
	54	if (e820.map[i].addr > 0xFFFFFFFFUL)
	55	continue;
	56	/*
edc05e6d IM	57	* Don't MCR over reserved space. Ignore the ISA hole
edc05e6d IM	58	* we frob around that catastrophe already
1da177e4	59	*/
edc05e6d IM	60	if (e820.map[i].type == E820_RESERVED) {
	61	if (e820.map[i].addr >= 0x100000UL &&
	62	e820.map[i].addr < clip)
1da177e4 LT	63	clip = e820.map[i].addr;
	64	continue;
	65	}
	66	start = e820.map[i].addr;
	67	end = e820.map[i].addr + e820.map[i].size;
	68	if (start >= end)
	69	continue;
	70	if (end > top)
	71	top = end;
	72	}
edc05e6d IM	73	/*
	74	* Everything below 'top' should be RAM except for the ISA hole.
	75	* Because of the limited MCR's we want to map NV/ACPI into our
	76	* MCR range for gunk in RAM
	77	*
	78	* Clip might cause us to MCR insufficient RAM but that is an
	79	* acceptable failure mode and should only bite obscure boxes with
	80	* a VESA hole at 15Mb
	81	*
	82	* The second case Clip sometimes kicks in is when the EBDA is marked
	83	* as reserved. Again we fail safe with reasonable results
	84	*/
	85	if (top > clip)
29a9994b PC	86	top = clip;
29a9994b PC	87
1da177e4 LT	88	return top;
	89	}
	90
	91	/*
edc05e6d	92	* Compute a set of MCR's to give maximum coverage
1da177e4	93	*/
b4af3f7c	94	static int __cpuinit centaur_mcr_compute(int nr, int key)
1da177e4 LT	95	{
	96	u32 mem = ramtop();
	97	u32 root = power2(mem);
	98	u32 base = root;
	99	u32 top = root;
	100	u32 floor = 0;
	101	int ct = 0;
29a9994b	102
edc05e6d	103	while (ct < nr) {
1da177e4	104	u32 fspace = 0;
edc05e6d IM	105	u32 high;
edc05e6d IM	106	u32 low;
1da177e4 LT	107
1da177e4 LT	108	/*
edc05e6d	109	* Find the largest block we will fill going upwards
1da177e4	110	*/
edc05e6d	111	high = power2(mem-top);
1da177e4 LT	112
1da177e4 LT	113	/*
edc05e6d	114	* Find the largest block we will fill going downwards
1da177e4	115	*/
edc05e6d	116	low = base/2;
1da177e4 LT	117
1da177e4 LT	118	/*
edc05e6d IM	119	* Don't fill below 1Mb going downwards as there
edc05e6d IM	120	* is an ISA hole in the way.
29a9994b	121	*/
29a9994b	122	if (base <= 1024*1024)
1da177e4	123	low = 0;
29a9994b	124
1da177e4	125	/*
edc05e6d IM	126	* See how much space we could cover by filling below
edc05e6d IM	127	* the ISA hole
1da177e4	128	*/
29a9994b PC	129
29a9994b PC	130	if (floor == 0)
1da177e4	131	fspace = 512*1024;
29a9994b	132	else if (floor == 512*1024)
1da177e4 LT	133	fspace = 128*1024;
	134
	135	/* And forget ROM space */
29a9994b	136
1da177e4	137	/*
edc05e6d	138	* Now install the largest coverage we get
1da177e4	139	*/
edc05e6d	140	if (fspace > high && fspace > low) {
1da177e4 LT	141	centaur_mcr_insert(ct, floor, fspace, key);
1da177e4 LT	142	floor += fspace;
edc05e6d	143	} else if (high > low) {
1da177e4 LT	144	centaur_mcr_insert(ct, top, high, key);
1da177e4 LT	145	top += high;
edc05e6d	146	} else if (low > 0) {
1da177e4 LT	147	base -= low;
1da177e4 LT	148	centaur_mcr_insert(ct, base, low, key);
edc05e6d IM	149	} else
edc05e6d IM	150	break;
1da177e4 LT	151	ct++;
	152	}
	153	/*
edc05e6d IM	154	* We loaded ct values. We now need to set the mask. The caller
edc05e6d IM	155	* must do this bit.
1da177e4	156	*/
1da177e4 LT	157	return ct;
	158	}
	159
b4af3f7c	160	static void __cpuinit centaur_create_optimal_mcr(void)
1da177e4	161	{
edc05e6d	162	int used;
1da177e4	163	int i;
edc05e6d	164
1da177e4	165	/*
edc05e6d IM	166	* Allocate up to 6 mcrs to mark as much of ram as possible
edc05e6d IM	167	* as write combining and weak write ordered.
1da177e4	168	*
edc05e6d IM	169	* To experiment with: Linux never uses stack operations for
edc05e6d IM	170	* mmio spaces so we could globally enable stack operation wc
1da177e4	171	*
edc05e6d IM	172	* Load the registers with type 31 - full write combining, all
edc05e6d IM	173	* writes weakly ordered.
1da177e4	174	*/
edc05e6d	175	used = centaur_mcr_compute(6, 31);
1da177e4 LT	176
1da177e4 LT	177	/*
edc05e6d	178	* Wipe unused MCRs
1da177e4	179	*/
29a9994b	180	for (i = used; i < 8; i++)
1da177e4 LT	181	wrmsr(MSR_IDT_MCR0+i, 0, 0);
	182	}
	183
b4af3f7c	184	static void __cpuinit winchip2_create_optimal_mcr(void)
1da177e4 LT	185	{
1da177e4 LT	186	u32 lo, hi;
edc05e6d	187	int used;
1da177e4 LT	188	int i;
	189
	190	/*
edc05e6d IM	191	* Allocate up to 6 mcrs to mark as much of ram as possible
edc05e6d IM	192	* as write combining, weak store ordered.
1da177e4	193	*
edc05e6d IM	194	* Load the registers with type 25
	195	* 8 - weak write ordering
	196	* 16 - weak read ordering
	197	* 1 - write combining
1da177e4	198	*/
edc05e6d	199	used = centaur_mcr_compute(6, 25);
29a9994b	200
1da177e4	201	/*
edc05e6d	202	* Mark the registers we are using.
1da177e4	203	*/
1da177e4	204	rdmsr(MSR_IDT_MCR_CTRL, lo, hi);
29a9994b PC	205	for (i = 0; i < used; i++)
29a9994b PC	206	lo \|= 1<<(9+i);
1da177e4	207	wrmsr(MSR_IDT_MCR_CTRL, lo, hi);
29a9994b	208
1da177e4	209	/*
edc05e6d	210	* Wipe unused MCRs
1da177e4	211	*/
29a9994b PC	212
29a9994b PC	213	for (i = used; i < 8; i++)
1da177e4 LT	214	wrmsr(MSR_IDT_MCR0+i, 0, 0);
	215	}
	216
	217	/*
edc05e6d	218	* Handle the MCR key on the Winchip 2.
1da177e4	219	*/
b4af3f7c	220	static void __cpuinit winchip2_unprotect_mcr(void)
1da177e4 LT	221	{
	222	u32 lo, hi;
	223	u32 key;
29a9994b	224
1da177e4	225	rdmsr(MSR_IDT_MCR_CTRL, lo, hi);
29a9994b	226	lo &= ~0x1C0; /* blank bits 8-6 */
1da177e4 LT	227	key = (lo>>17) & 7;
	228	lo \|= key<<6; /* replace with unlock key */
	229	wrmsr(MSR_IDT_MCR_CTRL, lo, hi);
	230	}
	231
b4af3f7c	232	static void __cpuinit winchip2_protect_mcr(void)
1da177e4 LT	233	{
1da177e4 LT	234	u32 lo, hi;
29a9994b	235
1da177e4	236	rdmsr(MSR_IDT_MCR_CTRL, lo, hi);
29a9994b	237	lo &= ~0x1C0; /* blank bits 8-6 */
1da177e4 LT	238	wrmsr(MSR_IDT_MCR_CTRL, lo, hi);
	239	}
	240	#endif /* CONFIG_X86_OOSTORE */
	241
	242	#define ACE_PRESENT (1 << 6)
	243	#define ACE_ENABLED (1 << 7)
	244	#define ACE_FCR (1 << 28) /* MSR_VIA_FCR */
	245
	246	#define RNG_PRESENT (1 << 2)
	247	#define RNG_ENABLED (1 << 3)
	248	#define RNG_ENABLE (1 << 6) /* MSR_VIA_RNG */
	249
b4af3f7c	250	static void __cpuinit init_c3(struct cpuinfo_x86 *c)
1da177e4 LT	251	{
	252	u32 lo, hi;
	253
	254	/* Test for Centaur Extended Feature Flags presence */
	255	if (cpuid_eax(0xC0000000) >= 0xC0000001) {
	256	u32 tmp = cpuid_edx(0xC0000001);
	257
	258	/* enable ACE unit, if present and disabled */
	259	if ((tmp & (ACE_PRESENT \| ACE_ENABLED)) == ACE_PRESENT) {
29a9994b	260	rdmsr(MSR_VIA_FCR, lo, hi);
1da177e4	261	lo \|= ACE_FCR; /* enable ACE unit */
29a9994b	262	wrmsr(MSR_VIA_FCR, lo, hi);
1da177e4 LT	263	printk(KERN_INFO "CPU: Enabled ACE h/w crypto\n");
	264	}
	265
	266	/* enable RNG unit, if present and disabled */
	267	if ((tmp & (RNG_PRESENT \| RNG_ENABLED)) == RNG_PRESENT) {
29a9994b	268	rdmsr(MSR_VIA_RNG, lo, hi);
1da177e4	269	lo \|= RNG_ENABLE; /* enable RNG unit */
29a9994b	270	wrmsr(MSR_VIA_RNG, lo, hi);
1da177e4 LT	271	printk(KERN_INFO "CPU: Enabled h/w RNG\n");
	272	}
	273
	274	/* store Centaur Extended Feature Flags as
	275	* word 5 of the CPU capability bit array
	276	*/
	277	c->x86_capability[5] = cpuid_edx(0xC0000001);
	278	}
	279
27b46d76	280	/* Cyrix III family needs CX8 & PGE explicitly enabled. */
29a9994b PC	281	if (c->x86_model >= 6 && c->x86_model <= 9) {
29a9994b PC	282	rdmsr(MSR_VIA_FCR, lo, hi);
1da177e4	283	lo \|= (1<<1 \| 1<<7);
29a9994b	284	wrmsr(MSR_VIA_FCR, lo, hi);
e1a94a97	285	set_cpu_cap(c, X86_FEATURE_CX8);
1da177e4 LT	286	}
	287
	288	/* Before Nehemiah, the C3's had 3dNOW! */
29a9994b	289	if (c->x86_model >= 6 && c->x86_model < 9)
e1a94a97	290	set_cpu_cap(c, X86_FEATURE_3DNOW);
1da177e4 LT	291
	292	get_model_name(c);
	293	display_cacheinfo(c);
	294	}
	295
edc05e6d IM	296	enum {
	297	ECX8 = 1<<1,
	298	EIERRINT = 1<<2,
	299	DPM = 1<<3,
	300	DMCE = 1<<4,
	301	DSTPCLK = 1<<5,
	302	ELINEAR = 1<<6,
	303	DSMC = 1<<7,
	304	DTLOCK = 1<<8,
	305	EDCTLB = 1<<8,
	306	EMMX = 1<<9,
	307	DPDC = 1<<11,
	308	EBRPRED = 1<<12,
	309	DIC = 1<<13,
	310	DDC = 1<<14,
	311	DNA = 1<<15,
	312	ERETSTK = 1<<16,
	313	E2MMX = 1<<19,
	314	EAMD3D = 1<<20,
	315	};
	316
5fef55fd YL	317	static void __cpuinit early_init_centaur(struct cpuinfo_x86 *c)
	318	{
	319	switch (c->x86) {
	320	case 5:
	321	/* Emulate MTRRs using Centaur's MCR. */
	322	set_cpu_cap(c, X86_FEATURE_CENTAUR_MCR);
	323	break;
	324	}
	325	}
	326
b4af3f7c	327	static void __cpuinit init_centaur(struct cpuinfo_x86 *c)
1da177e4	328	{
1da177e4 LT	329
1da177e4 LT	330	char *name;
29a9994b PC	331	u32 fcr_set = 0;
	332	u32 fcr_clr = 0;
	333	u32 lo, hi, newlo;
	334	u32 aa, bb, cc, dd;
1da177e4	335
edc05e6d IM	336	/*
	337	* Bit 31 in normal CPUID used for nonstandard 3DNow ID;
	338	* 3DNow is IDd by bit 31 in extended CPUID (1*32+31) anyway
	339	*/
e1a94a97	340	clear_cpu_cap(c, 0*32+31);
1da177e4 LT	341
1da177e4 LT	342	switch (c->x86) {
29a9994b	343	case 5:
edc05e6d IM	344	switch (c->x86_model) {
	345	case 4:
	346	name = "C6";
	347	fcr_set = ECX8\|DSMC\|EDCTLB\|EMMX\|ERETSTK;
	348	fcr_clr = DPDC;
	349	printk(KERN_NOTICE "Disabling bugged TSC.\n");
e1a94a97	350	clear_cpu_cap(c, X86_FEATURE_TSC);
1da177e4	351	#ifdef CONFIG_X86_OOSTORE
edc05e6d IM	352	centaur_create_optimal_mcr();
	353	/*
	354	* Enable:
	355	* write combining on non-stack, non-string
	356	* write combining on string, all types
	357	* weak write ordering
	358	*
	359	* The C6 original lacks weak read order
	360	*
	361	* Note 0x120 is write only on Winchip 1
	362	*/
	363	wrmsr(MSR_IDT_MCR_CTRL, 0x01F0001F, 0);
29a9994b	364	#endif
edc05e6d IM	365	break;
	366	case 8:
	367	switch (c->x86_mask) {
	368	default:
	369	name = "2";
1da177e4	370	break;
edc05e6d IM	371	case 7 ... 9:
	372	name = "2A";
	373	break;
	374	case 10 ... 15:
	375	name = "2B";
	376	break;
	377	}
	378	fcr_set = ECX8\|DSMC\|DTLOCK\|EMMX\|EBRPRED\|ERETSTK\|
	379	E2MMX\|EAMD3D;
	380	fcr_clr = DPDC;
1da177e4	381	#ifdef CONFIG_X86_OOSTORE
edc05e6d IM	382	winchip2_unprotect_mcr();
	383	winchip2_create_optimal_mcr();
	384	rdmsr(MSR_IDT_MCR_CTRL, lo, hi);
	385	/*
	386	* Enable:
	387	* write combining on non-stack, non-string
	388	* write combining on string, all types
	389	* weak write ordering
	390	*/
	391	lo \|= 31;
	392	wrmsr(MSR_IDT_MCR_CTRL, lo, hi);
	393	winchip2_protect_mcr();
1da177e4	394	#endif
edc05e6d IM	395	break;
	396	case 9:
	397	name = "3";
	398	fcr_set = ECX8\|DSMC\|DTLOCK\|EMMX\|EBRPRED\|ERETSTK\|
	399	E2MMX\|EAMD3D;
	400	fcr_clr = DPDC;
1da177e4	401	#ifdef CONFIG_X86_OOSTORE
edc05e6d IM	402	winchip2_unprotect_mcr();
	403	winchip2_create_optimal_mcr();
	404	rdmsr(MSR_IDT_MCR_CTRL, lo, hi);
	405	/*
	406	* Enable:
	407	* write combining on non-stack, non-string
	408	* write combining on string, all types
	409	* weak write ordering
	410	*/
	411	lo \|= 31;
	412	wrmsr(MSR_IDT_MCR_CTRL, lo, hi);
	413	winchip2_protect_mcr();
1da177e4	414	#endif
edc05e6d IM	415	break;
	416	default:
	417	name = "??";
	418	}
1da177e4	419
edc05e6d IM	420	rdmsr(MSR_IDT_FCR1, lo, hi);
edc05e6d IM	421	newlo = (lo\|fcr_set) & (~fcr_clr);
1da177e4	422
edc05e6d IM	423	if (newlo != lo) {
	424	printk(KERN_INFO "Centaur FCR was 0x%X now 0x%X\n",
	425	lo, newlo);
	426	wrmsr(MSR_IDT_FCR1, newlo, hi);
	427	} else {
	428	printk(KERN_INFO "Centaur FCR is 0x%X\n", lo);
	429	}
	430	/* Emulate MTRRs using Centaur's MCR. */
e1a94a97	431	set_cpu_cap(c, X86_FEATURE_CENTAUR_MCR);
edc05e6d	432	/* Report CX8 */
e1a94a97	433	set_cpu_cap(c, X86_FEATURE_CX8);
edc05e6d IM	434	/* Set 3DNow! on Winchip 2 and above. */
edc05e6d IM	435	if (c->x86_model >= 8)
e1a94a97	436	set_cpu_cap(c, X86_FEATURE_3DNOW);
edc05e6d IM	437	/* See if we can find out some more. */
	438	if (cpuid_eax(0x80000000) >= 0x80000005) {
	439	/* Yes, we can. */
	440	cpuid(0x80000005, &aa, &bb, &cc, &dd);
	441	/* Add L1 data and code cache sizes. */
	442	c->x86_cache_size = (cc>>24)+(dd>>24);
	443	}
	444	sprintf(c->x86_model_id, "WinChip %s", name);
	445	break;
1da177e4	446
29a9994b	447	case 6:
edc05e6d IM	448	init_c3(c);
edc05e6d IM	449	break;
1da177e4 LT	450	}
	451	}
	452
edc05e6d IM	453	static unsigned int __cpuinit
edc05e6d IM	454	centaur_size_cache(struct cpuinfo_x86 *c, unsigned int size)
1da177e4 LT	455	{
	456	/* VIA C3 CPUs (670-68F) need further shifting. */
	457	if ((c->x86 == 6) && ((c->x86_model == 7) \|\| (c->x86_model == 8)))
	458	size >>= 8;
	459
edc05e6d IM	460	/*
	461	* There's also an erratum in Nehemiah stepping 1, which
	462	* returns '65KB' instead of '64KB'
	463	* - Note, it seems this may only be in engineering samples.
	464	*/
	465	if ((c->x86 == 6) && (c->x86_model == 9) &&
	466	(c->x86_mask == 1) && (size == 65))
29a9994b	467	size -= 1;
1da177e4 LT	468
	469	return size;
	470	}
	471
95414930	472	static struct cpu_dev centaur_cpu_dev __cpuinitdata = {
1da177e4 LT	473	.c_vendor = "Centaur",
1da177e4 LT	474	.c_ident = { "CentaurHauls" },
5fef55fd	475	.c_early_init = early_init_centaur,
1da177e4 LT	476	.c_init = init_centaur,
1da177e4 LT	477	.c_size_cache = centaur_size_cache,
10a434fc	478	.c_x86_vendor = X86_VENDOR_CENTAUR,
1da177e4 LT	479	};
1da177e4 LT	480
10a434fc	481	cpu_dev_register(centaur_cpu_dev);