src/northbridge/amd/gx2/northbridgeinit.c

/*
 * This file is part of the coreboot project.
 *
 * Copyright (C) 2007 Advanced Micro Devices, Inc.
 * Copyright (C) 2010 Nils Jacobs
 *
 * This program is free software; you can redistribute it and/or modify
 * it under the terms of the GNU General Public License version 2 as
 * published by the Free Software Foundation.
 *
 * This program is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 * GNU General Public License for more details.
 *
 * You should have received a copy of the GNU General Public License
 * along with this program; if not, write to the Free Software
 * Foundation, Inc.
 */

#include <console/console.h>
#include <arch/io.h>
#include <stdint.h>
#include <device/device.h>
#include <device/pci.h>
#include <device/pci_ids.h>
#include <stdlib.h>
#include <string.h>
#include "northbridge.h"
#include <cpu/amd/gx2def.h>
#include <cpu/x86/msr.h>
#include <cpu/x86/cache.h>

struct gliutable
{
	unsigned long desc_name;
	unsigned short desc_type;
	unsigned long hi, lo;
};

struct gliutable gliu0table[] = {
	{.desc_name=GLIU0_P2D_BM_0,	.desc_type= BM,.hi= MSR_MC + 0x0,.lo= 0x0FFF80},		/* 0-7FFFF to MC */
	{.desc_name=GLIU0_P2D_BM_1,	.desc_type= BM,.hi= MSR_MC + 0x0,.lo=(0x80 << 20) + 0x0FFFE0},	/* 80000-9ffff to Mc */
	{.desc_name=GLIU0_P2D_SC_0,	.desc_type= SC_SHADOW,.hi= MSR_MC + 0x0,.lo= 0x03},		/* C0000-Fffff split to MC and PCI (sub decode) A0000-Bffff handled by SoftVideo */
	{.desc_name=GLIU0_P2D_R_0,	.desc_type= R_SYSMEM,.hi= MSR_MC,.lo= 0x0},			/* Catch and fix dynamicly. */
	{.desc_name=GLIU0_P2D_BMO_0,	.desc_type= BMO_SMM,.hi= MSR_MC,.lo= 0x0},			/* Catch and fix dynamicly. */
	{.desc_name=GLIU0_GLD_MSR_COH,	.desc_type= OTHER,.hi= 0x0,.lo= GL0_CPU},
	{.desc_name=GL_END,		.desc_type= GL_END,.hi= 0x0,.lo= 0x0},
};

struct gliutable gliu1table[] = {
	{.desc_name=GLIU1_P2D_BM_0,	.desc_type= BM,.hi= MSR_GL0 + 0x0,.lo= 0x0FFF80},		/* 0-7FFFF to MC */
	{.desc_name=GLIU1_P2D_BM_1,	.desc_type= BM,.hi= MSR_GL0 + 0x0,.lo= (0x80 << 20) + 0x0FFFE0},/* 80000-9ffff to Mc */
	{.desc_name=GLIU1_P2D_SC_0,	.desc_type= SC_SHADOW,.hi= MSR_GL0 + 0x0,.lo= 0x03},		/* C0000-Fffff split to MC and PCI (sub decode) */
	{.desc_name=GLIU1_P2D_R_0,	.desc_type= R_SYSMEM,.hi= MSR_GL0,.lo= 0x0},			/* Catch and fix dynamicly. */
	{.desc_name=GLIU1_P2D_BM_3,	.desc_type= BM_SMM,.hi= MSR_GL0,.lo= 0x0},			/* Catch and fix dynamicly. */
	{.desc_name=GLIU1_GLD_MSR_COH,	.desc_type= OTHER,.hi= 0x0,.lo= GL1_GLIU0},
	{.desc_name=GLIU1_IOD_SC_0,	.desc_type= SCIO,.hi= (GL1_GLCP << 29) + 0x0,.lo= 0x033000F0},	/* FooGlue FPU 0xF0 */
	{.desc_name=GL_END,		.desc_type= GL_END,.hi= 0x0,.lo= 0x0},
};

struct gliutable *gliutables[] = { gliu0table, gliu1table, 0 };

struct msrinit
{
	unsigned long msrnum;
	msr_t msr;
};

struct msrinit ClockGatingDefault[] = {
	{GLIU0_GLD_MSR_PM,	{.hi=0x00,.lo=0x0005}},
	/* MC must stay off in SDR mode. It is turned on in CPUBug??? lotus #77.142 */
	{MC_GLD_MSR_PM,		{.hi=0x00,.lo=0x0000}},
	{GLIU1_GLD_MSR_PM,	{.hi=0x00,.lo=0x0005}},
	{VG_GLD_MSR_PM,		{.hi=0x00,.lo=0x0000}},	/* lotus #77.163 */
	{GP_GLD_MSR_PM,		{.hi=0x00,.lo=0x0001}},
	{DF_GLD_MSR_PM,		{.hi=0x00,.lo=0x0155}},
	{GLCP_GLD_MSR_PM,	{.hi=0x00,.lo=0x0015}},
	{GLPCI_GLD_MSR_PM,	{.hi=0x00,.lo=0x0015}},
	{FG_GLD_MSR_PM,		{.hi=0x00,.lo=0x0000}},	/* Always on */
	{0xffffffff, 				{0xffffffff, 0xffffffff}},
};

/* All On */
struct msrinit ClockGatingAllOn[] = {
	{GLIU0_GLD_MSR_PM,	{.hi=0x00,.lo=0x0FFFFFFFF}},
	{MC_GLD_MSR_PM,		{.hi=0x00,.lo=0x0FFFFFFFF}},
	{GLIU1_GLD_MSR_PM,	{.hi=0x00,.lo=0x0FFFFFFFF}},
	{VG_GLD_MSR_PM,		{.hi=0x00, .lo=0x00}},
	{GP_GLD_MSR_PM,		{.hi=0x00,.lo=0x000000001}},
	{DF_GLD_MSR_PM,		{.hi=0x00,.lo=0x0FFFFFFFF}},
	{GLCP_GLD_MSR_PM,	{.hi=0x00,.lo=0x0FFFFFFFF}},
	{GLPCI_GLD_MSR_PM,	{.hi=0x00,.lo=0x0FFFFFFFF}},
	{FG_GLD_MSR_PM,		{.hi=0x00,.lo=0x0000}},
 	{0xffffffff, 				{0xffffffff, 0xffffffff}},
};

/* Performance */
struct msrinit ClockGatingPerformance[] = {
	{VG_GLD_MSR_PM,		{.hi=0x00,.lo=0x0000}},	/* lotus #77.163 */
	{GP_GLD_MSR_PM,		{.hi=0x00,.lo=0x0001}},
	{DF_GLD_MSR_PM,		{.hi=0x00,.lo=0x0155}},
	{GLCP_GLD_MSR_PM,	{.hi=0x00,.lo=0x0015}},
	{0xffffffff, 				{0xffffffff, 0xffffffff}},
};

/* SET GeodeLink PRIORITY */
struct msrinit GeodeLinkPriorityTable[] = {
	{CPU_GLD_MSR_CONFIG,		{.hi=0x00,.lo=0x0220}},		/* CPU Priority. */
	{DF_GLD_MSR_MASTER_CONF,	{.hi=0x00,.lo=0x0000}},		/* DF Priority. */
	{VG_GLD_MSR_CONFIG,		{.hi=0x00,.lo=0x0720}},		/* VG Primary and Secondary Priority. */
	{GP_GLD_MSR_CONFIG,		{.hi=0x00,.lo=0x0010}},		/* Graphics Priority. */
	{GLPCI_GLD_MSR_CONFIG,		{.hi=0x00,.lo=0x0027}},		/* GLPCI Priority + PID */
	{GLCP_GLD_MSR_CONF,		{.hi=0x00,.lo=0x0001}},		/* GLCP Priority + PID */
	{FG_GLD_MSR_CONFIG,		{.hi=0x00,.lo=0x0622}},		/* FG PID */
	{0x0FFFFFFFF, 			{0x0FFFFFFFF, 0x0FFFFFFFF}},	/* END */
};

static void writeglmsr(struct gliutable *gl)
{
	msr_t msr;

	msr.lo = gl->lo;
	msr.hi = gl->hi;
	wrmsr(gl->desc_name, msr);	/* MSR - see table above */
	printk(BIOS_DEBUG, "%s: MSR 0x%08lx, val 0x%08x:0x%08x\n", __func__, gl->desc_name, msr.hi, msr.lo);
}

static void ShadowInit(struct gliutable *gl)
{
	msr_t msr;

	msr = rdmsr(gl->desc_name);

	if (msr.lo == 0) {
		writeglmsr(gl);
	}
}

static void SysmemInit(struct gliutable *gl)
{
	msr_t msr;
	int sizembytes, sizebytes;

	/* Figure out how much RAM is in the machine and alocate all to the
	 * system. We will adjust for SMM now and Frame Buffer later.
	 */
	sizembytes = sizeram();
	printk(BIOS_DEBUG, "%s: enable for %dMBytes\n", __func__, sizembytes);
	sizebytes = sizembytes << 20;

	sizebytes -= ((SMM_SIZE * 1024) + 1);

	/* 20 bit address The bottom 12 bits go into bits 20-31 in msr.lo
	   The top 8 bits go into 0-7 of msr.hi. */
	sizebytes --;
	msr.hi = (gl->hi & 0xFFFFFF00) | (sizebytes >> 24);
	sizebytes <<= 8;	/* move bits 23:12 in bits 31:20. */
	sizebytes &= 0xfff00000;
	sizebytes |= 0x100;	/* start at 1MB */
	msr.lo = sizebytes;
	wrmsr(gl->desc_name, msr);	/* MSR - see table above */
	printk(BIOS_DEBUG, "%s: MSR 0x%08lx, val 0x%08x:0x%08x\n", __func__,
				gl->desc_name, msr.hi, msr.lo);
}

static void SMMGL0Init(struct gliutable *gl)
{
	msr_t msr;
	int sizebytes = sizeram() << 20;
	long offset;

	sizebytes -= (SMM_SIZE * 1024);

	printk(BIOS_DEBUG, "%s: %d bytes\n", __func__, sizebytes);

	offset = sizebytes - SMM_OFFSET;
	offset = (offset >> 12) & 0x000fffff;
	printk(BIOS_DEBUG, "%s: offset is 0x%08x\n", __func__, SMM_OFFSET);

	msr.hi = offset << 8 | gl->hi;
	msr.hi |= SMM_OFFSET >> 24;

	msr.lo = SMM_OFFSET << 8;
	msr.lo |= ((~(SMM_SIZE * 1024) + 1) >> 12) & 0xfffff;

	wrmsr(gl->desc_name, msr);	/* MSR - See table above */
	printk(BIOS_DEBUG, "%s: MSR 0x%08lx, val 0x%08x:0x%08x\n", __func__, gl->desc_name, msr.hi, msr.lo);
}

static void SMMGL1Init(struct gliutable *gl)
{
	msr_t msr;
	printk(BIOS_DEBUG, "%s:\n", __func__);

	msr.hi = gl->hi;
	/* I don't think this is needed */
	msr.hi &= 0xffffff00;
	msr.hi |= (SMM_OFFSET >> 24);
	msr.lo = (SMM_OFFSET << 8) & 0xfff00000;
	msr.lo |= ((~(SMM_SIZE * 1024) + 1) >> 12) & 0xfffff;

	wrmsr(gl->desc_name, msr);	/* MSR - See table above */
	printk(BIOS_DEBUG, "%s: MSR 0x%08lx, val 0x%08x:0x%08x\n", __func__, gl->desc_name, msr.hi, msr.lo);
}

static void GLIUInit(struct gliutable *gl)
{
	while (gl->desc_type != GL_END) {
		switch (gl->desc_type) {
		default:
			writeglmsr(gl);
		case SC_SHADOW: /* Check for a Shadow entry */
			ShadowInit(gl);
			break;

		case R_SYSMEM: /* check for a SYSMEM entry */
			SysmemInit(gl);
			break;

		case BMO_SMM: /* check for a SMM entry */
			SMMGL0Init(gl);
			break;

		case BM_SMM: /* check for a SMM entry */
			SMMGL1Init(gl);
			break;
		}
		gl++;
	}
}

/* Set up GLPCI settings for reads/write into memory.
 *
 * R0:  0-640KB,
 * R1:  1MB - Top of System Memory
 * R2: SMM Memory
 * R3: Framebuffer? - not set up yet
 * R4: ??
 */
static void GLPCIInit(void)
{
	struct gliutable *gl = 0;
	int i;
	msr_t msr;
	int msrnum;

	/* R0 - GLPCI settings for Conventional Memory space. */
	msr.hi =  (0x09F000 >> 12) << GLPCI_RC_UPPER_TOP_SHIFT;	/* 640 */
	msr.lo =  0;						/* 0 */
	msr.lo |= GLPCI_RC_LOWER_EN_SET + GLPCI_RC_LOWER_PF_SET + GLPCI_RC_LOWER_WC_SET;
	msrnum = GLPCI_RC0;
	wrmsr(msrnum, msr);

	/* R1 - GLPCI settings for SysMem space. */
	/* Get systop from GLIU0 SYSTOP Descriptor */
	for (i = 0; gliu0table[i].desc_name != GL_END; i++) {
		if (gliu0table[i].desc_type == R_SYSMEM) {
			gl = &gliu0table[i];
			break;
		}
	}
	if (gl) {
		unsigned long pah, pal;
		msrnum = gl->desc_name;
		msr = rdmsr(msrnum);
		/* example R_SYSMEM value: 20:00:00:0f:fb:f0:01:00
		 * translates to a base of 0x00100000 and top of 0xffbf0000
		 * base of 1M and top of around 256M
		 */
		/* we have to create a page-aligned (4KB page) address for base and top
		 * so we need a high page aligned addresss (pah) and low page aligned address (pal)
		 * pah is from msr.hi << 12 | msr.low >> 20. pal is msr.lo << 12
		 */
		pah = ((msr.hi & 0xff) << 12) | ((msr.lo >> 20) & 0xfff);
		/* we have the page address. Now make it a page-aligned address */
		pah <<= 12;

		pal = msr.lo << 12;
		msr.hi = pah;
		msr.lo = pal;
		msr.lo |= GLPCI_RC_LOWER_EN_SET | GLPCI_RC_LOWER_PF_SET | GLPCI_RC_LOWER_WC_SET;
		printk(BIOS_DEBUG, "GLPCI R1: system msr.lo 0x%08x msr.hi 0x%08x\n", msr.lo, msr.hi);
		msrnum = GLPCI_RC1;
		wrmsr(msrnum, msr);
	}

	/* R2 - GLPCI settings for SMM space. */
	msr.hi = ((SMM_OFFSET + (SMM_SIZE * 1024 - 1)) >> 12) << GLPCI_RC_UPPER_TOP_SHIFT;
	msr.lo = (SMM_OFFSET >> 12) << GLPCI_RC_LOWER_BASE_SHIFT;
	msr.lo |= GLPCI_RC_LOWER_EN_SET | GLPCI_RC_LOWER_PF_SET;
	printk(BIOS_DEBUG, "GLPCI R2: system msr.lo 0x%08x msr.hi 0x%08x\n", msr.lo, msr.hi);
	msrnum = GLPCI_RC2;
	wrmsr(msrnum, msr);

	/* this is done elsewhere already, but it does no harm to do it more than once */
	/* write serialize memory hole to PCI. Need to unWS when something is shadowed regardless of cachablility. */
	msr.lo = 0x021212121;	/* cache disabled and write serialized */
	msr.hi = 0x021212121;	/* cache disabled and write serialized */

	msrnum = CPU_RCONF_A0_BF;
	wrmsr(msrnum, msr);

	msrnum = CPU_RCONF_C0_DF;
	wrmsr(msrnum, msr);

	msrnum = CPU_RCONF_E0_FF;
	wrmsr(msrnum, msr);

	/* Set Non-Cacheable Read Only for NorthBound Transactions to Memory. The Enable bit is handled in the Shadow setup. */
	msrnum = GLPCI_A0_BF;
	msr.hi = 0x35353535;
	msr.lo = 0x35353535;
	wrmsr(msrnum, msr);

	msrnum = GLPCI_C0_DF;
	msr.hi = 0x35353535;
	msr.lo = 0x35353535;
	wrmsr(msrnum, msr);

	msrnum = GLPCI_E0_FF;
	msr.hi = 0x35353535;
	msr.lo = 0x35353535;
	wrmsr(msrnum, msr);

	/* Set WSREQ */
	msrnum = CPU_DM_CONFIG0;
	msr = rdmsr(msrnum);
	msr.hi &= ~(7 << DM_CONFIG0_UPPER_WSREQ_SHIFT);
	msr.hi |= 2 << DM_CONFIG0_UPPER_WSREQ_SHIFT	;	/* reduce to 1 for safe mode. */
	wrmsr(msrnum, msr);

	/* we are ignoring the 5530 case for now, and perhaps forever. */

	/* 553X NB Init */

	/* Arbiter setup */
	msrnum = GLPCI_ARB;
	msr = rdmsr(msrnum);
	msr.hi |= GLPCI_ARB_UPPER_PRE0_SET | GLPCI_ARB_UPPER_PRE1_SET;
	msr.lo |= GLPCI_ARB_LOWER_IIE_SET;
	wrmsr(msrnum, msr);

	msrnum = GLPCI_CTRL;
	msr = rdmsr(msrnum);

	msr.lo |= GLPCI_CTRL_LOWER_ME_SET | GLPCI_CTRL_LOWER_OWC_SET | GLPCI_CTRL_LOWER_PCD_SET; 	/* (Out will be disabled in CPUBUG649 for < 2.0 parts .) */
	msr.lo |= GLPCI_CTRL_LOWER_LDE_SET;

	msr.lo &= ~(0x03 << GLPCI_CTRL_LOWER_IRFC_SHIFT);
	msr.lo |= 0x02 << GLPCI_CTRL_LOWER_IRFC_SHIFT;

	msr.lo &= ~(0x07 << GLPCI_CTRL_LOWER_IRFT_SHIFT);
	msr.lo |= 0x06 << GLPCI_CTRL_LOWER_IRFT_SHIFT;

	msr.hi &= ~(0x0f << GLPCI_CTRL_UPPER_FTH_SHIFT);
	msr.hi |= 0x0F << GLPCI_CTRL_UPPER_FTH_SHIFT;

	msr.hi &= ~(0x0f << GLPCI_CTRL_UPPER_RTH_SHIFT);
	msr.hi |= 0x0F << GLPCI_CTRL_UPPER_RTH_SHIFT;

	msr.hi &= ~(0x0f << GLPCI_CTRL_UPPER_SBRTH_SHIFT);
	msr.hi |= 0x0F << GLPCI_CTRL_UPPER_SBRTH_SHIFT;

	msr.hi &= ~(0x03 << GLPCI_CTRL_UPPER_WTO_SHIFT);
	msr.hi |= 0x06 << GLPCI_CTRL_UPPER_WTO_SHIFT;

	msr.hi &= ~(0x03 << GLPCI_CTRL_UPPER_ILTO_SHIFT);
	msr.hi |= 0x00 << GLPCI_CTRL_UPPER_ILTO_SHIFT;
	wrmsr(msrnum, msr);

	/* Set GLPCI Latency Timer. */
	msrnum = GLPCI_CTRL;
	msr = rdmsr(msrnum);
	msr.hi |= 0x1F << GLPCI_CTRL_UPPER_LAT_SHIFT; 	/* Change once 1.x is gone. */
	wrmsr(msrnum, msr);

	/* GLPCI_SPARE */
	msrnum = GLPCI_SPARE;
	msr = rdmsr(msrnum);
	msr.lo &= ~0x7;
	msr.lo |= GLPCI_SPARE_LOWER_AILTO_SET | GLPCI_SPARE_LOWER_PPD_SET | GLPCI_SPARE_LOWER_PPC_SET | GLPCI_SPARE_LOWER_MPC_SET | GLPCI_SPARE_LOWER_NSE_SET | GLPCI_SPARE_LOWER_SUPO_SET;
	wrmsr(msrnum, msr);
}

/* Enable Clock Gating. */
static void ClockGatingInit(void)
{
	msr_t msr;
	struct msrinit *gating = ClockGatingDefault;
	int i;

	for (i = 0; gating->msrnum != 0xffffffff; i++) {
		msr = rdmsr(gating->msrnum);
		msr.hi |= gating->msr.hi;
		msr.lo |= gating->msr.lo;
		wrmsr(gating->msrnum, msr);	/* MSR - See the table above */
		gating += 1;
	}
}

static void GeodeLinkPriority(void)
{
	msr_t msr = { 0, 0 };

	struct msrinit *prio = GeodeLinkPriorityTable;
	int i;

	for (i = 0; prio->msrnum != 0xffffffff; i++) {
		msr = rdmsr(prio->msrnum);
		msr.hi |= prio->msr.hi;
		msr.lo &= ~0xfff;
		msr.lo |= prio->msr.lo;
		wrmsr(prio->msrnum, msr);	/* MSR - See the table above */
		prio += 1;
	}
}

/* Get the GLIU0 shadow register settings.
 *
 * If the setShadow function is used then all shadow descriptors
 * will stay sync'ed.
 */
static uint64_t getShadow(void)
{
	msr_t msr = { 0, 0 };

	msr = rdmsr(GLIU0_P2D_SC_0);
	return ( ( (uint64_t) msr.hi ) << 32 ) | msr.lo;
}

/* Set the cache RConf registers for the memory hole.
 *
 * Keeps all cache shadow descriptors sync'ed.
 * This is part of the PCI lockup solution.
 *
 * Entry: EDX:EAX is the shadow settings.
 */
static void setShadowRCONF(uint32_t shadowHi, uint32_t shadowLo)
{
	/* ok this is whacky bit translation time. */
	int bit;
	uint8_t shadowByte;
	msr_t msr = { 0, 0 };
	shadowByte = (uint8_t) (shadowLo >> 16);

	/* load up D000 settings in edx. */
	for (bit = 8; (bit > 4); bit--) {
		msr.hi <<= 8;
		msr.hi |= 1;			/* cache disable PCI/Shadow memory */
		if (shadowByte && (1 << bit))
			msr.hi |= 0x20;		/* write serialize PCI memory */
	}

	/* load up C000 settings in eax. */
	for (; bit; bit--) {
		msr.lo <<= 8;
		msr.lo |= 1;			/* cache disable PCI/Shadow memory */
		if (shadowByte && (1 << bit))
			msr.lo |= 0x20;		/* write serialize PCI memory */
	}

	wrmsr(CPU_RCONF_C0_DF, msr);

	shadowByte = (uint8_t) (shadowLo >> 24);

	/* load up F000 settings in edx. */
	for (bit = 8; (bit > 4); bit--) {
		msr.hi <<= 8;
		msr.hi |= 1;			/* cache disable PCI/Shadow memory */
		if (shadowByte && (1 << bit))
			msr.hi |= 0x20;		/* write serialize PCI memory */
	}

	/* load up E000 settings in eax. */
	for (; bit; bit--) {
		msr.lo <<= 8;
		msr.lo |= 1;			/* cache disable PCI/Shadow memory */
		if (shadowByte && (1 << bit))
			msr.lo |= 0x20;		/* write serialize PCI memory */
	}

	wrmsr(CPU_RCONF_E0_FF, msr);
}

/* Set the GLPCI registers for the memory hole.
 * Keeps all cache shadow descriptors sync'ed.
 * Entry: EDX:EAX is the shadow settings
 */
static void setShadowGLPCI(uint32_t shadowHi, uint32_t shadowLo)
{
	msr_t msr;

	/* Set the Enable Register. */
	msr = rdmsr(GLPCI_REN);
	msr.lo &= 0xFFFF00FF;
	msr.lo |= ((shadowLo & 0xFFFF0000) >> 8);
	wrmsr(GLPCI_REN, msr);
}

/* Set the GLIU SC register settings. Scans descriptor tables for SC_SHADOW.
 * Keeps all shadow descriptors sync'ed.
 * Entry: EDX:EAX is the shadow settings
 */
static void setShadow(uint64_t shadowSettings)
{
	int i;
	msr_t msr;
	struct gliutable *pTable;
	uint32_t shadowLo, shadowHi;

	shadowLo = (uint32_t) shadowSettings;
	shadowHi = (uint32_t) (shadowSettings >> 32);

	setShadowRCONF(shadowHi, shadowLo);
	setShadowGLPCI(shadowHi, shadowLo);

	for (i = 0; gliutables[i]; i++) {
		for (pTable = gliutables[i]; pTable->desc_type != GL_END; pTable++) {
			if (pTable->desc_type == SC_SHADOW) {

				msr = rdmsr(pTable->desc_name);
				msr.lo = (uint32_t) shadowSettings;
				msr.hi &= 0xFFFF0000;		/* maintain PDID in upper EDX */
				msr.hi |= ((uint32_t) (shadowSettings >> 32)) & 0x0000FFFF;
				wrmsr(pTable->desc_name, msr);	/* MSR - See the table above */

			}
		}
	}
}

static void rom_shadow_settings(void)
{
	uint64_t shadowSettings = getShadow();
	shadowSettings &= (uint64_t) 0xFFFF00000000FFFFULL;	/* Disable read & writes */
	shadowSettings |= (uint64_t) 0x0000FFFFFFFF0000ULL;	/* Enable reads for C0000-FFFFF */
	setShadow(shadowSettings);
}

/* Set up RCONF_DEFAULT and any other RCONF registers needed.
 *
 * DEVRC_RCONF_DEFAULT:
 * ROMRC(63:56) = 04h	 ; write protect ROMBASE
 * ROMBASE(36:55) = 0FFFC0h ; Top of PCI/bottom of rom chipselect area
 * DEVRC(35:28) =  39h	 ; cache disabled in PCI memory + WS bit on + Write Combine + write burst.
 * SYSTOP(27:8) = top of system memory
 * SYSRC(7:0) = 00h 		 ; writeback, can set to 08h to make writethrough
 */
#define SYSMEM_RCONF_WRITETHROUGH 8
#define DEVRC_RCONF_DEFAULT 0x21ul
#define ROMBASE_RCONF_DEFAULT 0xFFFC0000
#define ROMRC_RCONF_DEFAULT 0x25

static void enable_L_cache(void)
{
	struct gliutable *gl = 0;
	int i;
	msr_t msr;
	uint8_t SysMemCacheProp;

	/* Locate SYSMEM entry in GLIU0table */
	for (i = 0; gliu0table[i].desc_name != GL_END; i++) {
		if (gliu0table[i].desc_type == R_SYSMEM) {
			gl = &gliu0table[i];
			break;
		}
	}
	if (gl == 0) {
		post_code(0xCE);	/* POST_RCONFInitError */
		while (1);
	}

/* sysdescfound: */
	msr = rdmsr(gl->desc_name);

	/* 20 bit address -  The bottom 12 bits go into bits 20-31 in eax, the
	 * top 8 bits go into 0-7 of edx.
	 */
	msr.lo = (msr.lo & 0xFFFFFF00) | (msr.hi & 0xFF);
	msr.lo = ((msr.lo << 12) | (msr.lo >> 20)) & 0x000FFFFF;
	msr.lo <<= RCONF_DEFAULT_LOWER_SYSTOP_SHIFT;	/* 8 */

	/* Set Default SYSMEM region properties */
	msr.lo &= ~SYSMEM_RCONF_WRITETHROUGH;	/* NOT writethrough == writeback 8 (or ~8) */

	/* Set PCI space cache properties */
	msr.hi = (DEVRC_RCONF_DEFAULT >> 4);	/* setting is split betwwen hi and lo... */
	msr.lo |= (DEVRC_RCONF_DEFAULT << 28);

	/* Set the ROMBASE. This is usually FFFC0000h */
	msr.hi |= (ROMBASE_RCONF_DEFAULT >> 12) << RCONF_DEFAULT_UPPER_ROMBASE_SHIFT;

	/* Set ROMBASE cache properties. */
	msr.hi |= ((ROMRC_RCONF_DEFAULT >> 8) | (ROMRC_RCONF_DEFAULT << 24));

	/* now program RCONF_DEFAULT */
	wrmsr(CPU_RCONF_DEFAULT, msr);
	printk(BIOS_DEBUG, "CPU_RCONF_DEFAULT (1808): 0x%08X:0x%08X\n", msr.hi, msr.lo);

	/* RCONF_BYPASS: Cache tablewalk properties and SMM header access properties. */
	/* Set to match system memory cache properties. */
	msr = rdmsr(CPU_RCONF_DEFAULT);
	SysMemCacheProp = (uint8_t) (msr.lo & 0xFF);
	msr = rdmsr(CPU_RCONF_BYPASS);
	msr.lo = (msr.lo & 0xFFFF0000) | (SysMemCacheProp << 8) | SysMemCacheProp;
	wrmsr(CPU_RCONF_BYPASS, msr);
	printk(BIOS_DEBUG, "CPU_RCONF_BYPASS (180A): 0x%08x : 0x%08x\n", msr.hi, msr.lo);
}

static void setup_gx2_cache(void)
{
	msr_t msr;

	enable_L_cache();

	/* Make sure all INVD instructions are treated as WBINVD.  We do this
	 * because we've found some programs which require this behavior.
	 */
	msr = rdmsr(CPU_DM_CONFIG0);
	msr.lo |= DM_CONFIG0_LOWER_WBINVD_SET;
	wrmsr(CPU_DM_CONFIG0, msr);

	x86_enable_cache();
	wbinvd();
}

uint32_t get_systop(void)
{
	struct gliutable *gl = 0;
	uint32_t systop;
	msr_t msr;
	int i;

	for (i = 0; gliu0table[i].desc_name != GL_END; i++) {
		if (gliu0table[i].desc_type == R_SYSMEM) {
			gl = &gliu0table[i];
			break;
		}
	}
	if (gl) {
		msr = rdmsr(gl->desc_name);
		systop = ((msr.hi & 0xFF) << 24) | ((msr.lo & 0xFFF00000) >> 8);
		systop += 0x1000;	/* 4K */
	} else {
		systop =
		    ((sizeram() - CONFIG_VIDEO_MB) * 1024) - SMM_SIZE - 1024;
	}
	return systop;
}

/* Core Logic initialization: Host bridge. */
void northbridge_init_early(void)
{
	int i;
	printk(BIOS_DEBUG, "Enter %s\n", __func__);

	for (i = 0; gliutables[i]; i++)
		GLIUInit(gliutables[i]);

	/* Now that the descriptor to memory is set up. */
	/* The memory controller needs one read to synch its lines before it can be used. */
	i = *(volatile int *) 0;

	GeodeLinkPriority();

	setup_gx2_cache();

	rom_shadow_settings();

	GLPCIInit();

	ClockGatingInit();

	__asm__ __volatile__("FINIT\n");
	printk(BIOS_DEBUG, "Exit %s\n", __func__);
}