switch-coreboot/arch/x86/geodelx/geodelx.c
Carl-Daniel Hailfinger e1138c9164 Typo: word duplication.
Signed-off-by: Carl-Daniel Hailfinger <c-d.hailfinger.devel.2006@gmx.net>
Acked-by: Carl-Daniel Hailfinger <c-d.hailfinger.devel.2006@gmx.net>


git-svn-id: svn://coreboot.org/repository/coreboot-v3@803 f3766cd6-281f-0410-b1cd-43a5c92072e9
2008-08-22 11:33:19 +00:00

483 lines
14 KiB
C

/*
* This file is part of the coreboot project.
*
* Copyright (C) 2006 Indrek Kruusa <indrek.kruusa@artecdesign.ee>
* Copyright (C) 2006 Ronald G. Minnich <rminnich@gmail.com>
* Copyright (C) 2007 Advanced Micro Devices, Inc.
*
* This program is free software; you can redistribute it and/or modify
* it under the terms of the GNU General Public License as published by
* the Free Software Foundation; either version 2 of the License, or
* (at your option) any later version.
*
* 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., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
*/
#include <types.h>
#include <lib.h>
#include <console.h>
#include <msr.h>
#include <io.h>
#include <cpu.h>
#include <amd_geodelx.h>
#include <spd.h>
#include <legacy.h>
/* All these functions used to be in a lot of fiddly little files. To make it
* easier to find functions, we are merging them here. This file is our first
* real CPU-specific support file and should serve as a model for v3
* CPU-specific support.
*
* So, warning, you might think it makes sense to split this file up, but
* we've tried that, and it sucks.
*/
/**
* Starts Timer 1 for port 61 use.
*
* The command 0x56 means write counter 1 lower 8 bits in next I/O, set the
* counter mode to square wave generator (count down to 0 from programmed
* value twice in a row, alternating the output signal) counting in 16-bit
* binary mode.
*
* 0x12 is counter/timer 1 and signals the PIT to do a RAM refresh
* approximately every 15us.
*
* The PIT typically is generating 1.19318 MHz.
*
* Timer 1 was used for RAM refresh on XT/AT and can be read on port 61.
* Port 61 is used by many timing loops for calibration.
*/
static void start_timer1(void)
{
outb(0x56, I82C54_CONTROL_WORD_REGISTER);
outb(0x12, I82C54_COUNTER1);
}
/**
* Very early initialization needed for almost everything else.
* Currently, all we do is start timer1.
*/
void system_preinit(void)
{
start_timer1();
}
/* CPU bug management */
/**
* Bugtool #465 and #609 PCI cache deadlock.
* TODO: URL?
*
* There is also fix code in cache and PCI functions. This bug is very
* pervasive.
*/
static void pci_deadlock(void)
{
struct msr msr;
/* Forces serialization of all load misses. Setting this bit prevents
* the DM pipe from backing up if a read request has to be held up
* waiting for PCI writes to complete.
*/
msr = rdmsr(CPU_DM_CONFIG0);
msr.lo |= DM_CONFIG0_LOWER_MISSER_SET;
wrmsr(CPU_DM_CONFIG0, msr);
/* Write serialize memory hole to PCI. Need to unWS when something is
* shadowed regardless of cachablility.
*/
msr.lo = 0x021212121;
msr.hi = 0x021212121;
wrmsr(CPU_RCONF_A0_BF, msr);
wrmsr(CPU_RCONF_C0_DF, msr);
wrmsr(CPU_RCONF_E0_FF, msr);
}
/**
* PBZ 3659: The MC reordered transactions incorrectly and breaks coherency.
*
* Disable reording and take a potential performance hit. This is safe to do
* here and not in MC init, since there is nothing to maintain coherency with
* and the cache is not enabled yet.
*/
static void disable_memory_reorder(void)
{
struct msr msr;
msr = rdmsr(MC_CF8F_DATA);
msr.hi |= CF8F_UPPER_REORDER_DIS_SET;
wrmsr(MC_CF8F_DATA, msr);
}
/**
* Fix up register settings to manage known CPU bugs.
*
* For CPU version C3. Should be the only released version.
*/
static void cpu_bug(void)
{
pci_deadlock();
disable_memory_reorder();
printk(BIOS_DEBUG, "Done cpubug fixes\n");
}
/**
* Reset the phase locked loop (PLL) hardware.
*
* After power on as part of this operation, we have to set the clock
* hardware and reboot. Thus, we have to know if we have been here before.
*
* To do this, we use the RSTPLL_LOWER_SWFLAGS_SHIFT flag in the
* msrGlcpSysRstpll. Also, the clocks can either be configured via passed-in
* parameters or hardware straps. Once set, we yank the hardware reset line
* and hlt. We should never reach the hlt, but one never knows.
*
* @param manualconf If non-zero, use passed-in parameters to determine how
* to configure PLL -- manual or automagic. If manual, use
* passed-in parameters pll_hi and pll_lo.
* @param pll_hi Value to use for the high 32 bits of the PLL msr.
* @param pll_lo Value to use for the low 32 bits of the PLL msr.
*/
void pll_reset(int manualconf, u32 pll_hi, u32 pll_lo)
{
struct msr msr_glcp_sys_pll; /* GeodeLink PLL control MSR */
printk(BIOS_DEBUG, "pll_reset: read msr %#x\n", GLCP_SYS_RSTPLL);
msr_glcp_sys_pll = rdmsr(GLCP_SYS_RSTPLL);
printk(BIOS_DEBUG, "_MSR GLCP_SYS_RSTPLL (%08x) value is: %08x:%08x\n",
GLCP_SYS_RSTPLL, msr_glcp_sys_pll.hi, msr_glcp_sys_pll.lo);
post_code(POST_PLL_INIT);
if (!(msr_glcp_sys_pll.lo & (1 << RSTPLL_LOWER_SWFLAGS_SHIFT))) {
printk(BIOS_DEBUG, "Configuring PLL\n");
if (manualconf) {
post_code(POST_PLL_MANUAL);
/* CPU and GLIU mult/div (GLMC_CLK = GLIU_CLK / 2) */
msr_glcp_sys_pll.hi = pll_hi;
/* Hold Count - how long we will sit in reset */
msr_glcp_sys_pll.lo = pll_lo;
} else {
/* Automatic configuration (straps) */
post_code(POST_PLL_STRAP);
/* Hold 0xDE * 16 clocks during reset. AMD recomended
* value for PLL reset from silicon validation.
*/
msr_glcp_sys_pll.lo &=
~(0xFF << RSTPPL_LOWER_HOLD_COUNT_SHIFT);
msr_glcp_sys_pll.lo |=
(0xDE << RSTPPL_LOWER_HOLD_COUNT_SHIFT);
msr_glcp_sys_pll.lo &=
~(RSTPPL_LOWER_COREBYPASS_SET |
RSTPPL_LOWER_MBBYPASS_SET);
msr_glcp_sys_pll.lo |=
RSTPPL_LOWER_COREPD_SET | RSTPPL_LOWER_CLPD_SET;
}
/* Use SWFLAGS to remember: "we've already been here". */
msr_glcp_sys_pll.lo |= (1 << RSTPLL_LOWER_SWFLAGS_SHIFT);
printk(BIOS_INFO, "Resetting the processor after PLL "
"configuration for the changes to take effect\n");
/* "Reset the chip" value */
msr_glcp_sys_pll.lo |= RSTPPL_LOWER_CHIP_RESET_SET;
wrmsr(GLCP_SYS_RSTPLL, msr_glcp_sys_pll);
/* You should never get here... the chip has reset. */
printk(BIOS_EMERG, "CONFIGURING PLL FAILURE -- HALT\n");
post_code(POST_PLL_RESET_FAIL);
hlt();
}
printk(BIOS_DEBUG, "Done pll_reset\n");
}
/**
* Return the CPU clock rate from the PLL MSR.
*
* @return CPU speed in MHz.
*/
u32 cpu_speed(void)
{
u32 speed;
struct msr msr;
msr = rdmsr(GLCP_SYS_RSTPLL);
speed = ((((msr.hi >> RSTPLL_UPPER_CPUMULT_SHIFT)
& RSTPLL_UPPER_CPUMULT_MASK) + 1) * 333) / 10;
if ((((((msr.hi >> RSTPLL_UPPER_CPUMULT_SHIFT)
& RSTPLL_UPPER_CPUMULT_MASK) + 1) * 333) % 10) > 5) {
++speed;
}
return speed;
}
/**
* Return the GeodeLink clock rate from the PLL MSR.
*
* @return GeodeLink speed in MHz.
*/
u32 geode_link_speed(void)
{
u32 speed;
struct msr msr;
msr = rdmsr(GLCP_SYS_RSTPLL);
speed = ((((msr.hi >> RSTPLL_UPPER_GLMULT_SHIFT)
& RSTPLL_UPPER_GLMULT_MASK) + 1) * 333) / 10;
if ((((((msr.hi >> RSTPLL_UPPER_GLMULT_SHIFT)
& RSTPLL_UPPER_GLMULT_MASK) + 1) * 333) % 10) > 5) {
++speed;
}
return speed;
}
/**
* Return the PCI bus clock rate from the PLL MSR.
*
* @return PCI speed in MHz.
*/
u32 pci_speed(void)
{
struct msr msr = rdmsr(GLCP_SYS_RSTPLL);
if (msr.hi & (1 << RSTPPL_LOWER_PCISPEED_SHIFT))
return 66;
else
return 33;
}
static const struct msrinit msr_table[] = {
{CPU_BC_MSS_ARRAY_CTL0, {.hi = 0x00000000, .lo = 0x2814D352}},
{CPU_BC_MSS_ARRAY_CTL1, {.hi = 0x00000000, .lo = 0x1068334D}},
{CPU_BC_MSS_ARRAY_CTL2, {.hi = 0x00000106, .lo = 0x83104104}},
};
/**
* Delay Control Settings table from AMD (MCP 0x4C00000F).
*/
static const struct delay_controls {
u8 dimms;
u8 devices;
u32 slow_hi;
u32 slow_low;
u32 fast_hi;
u32 fast_low;
} delay_control_table[] = {
/* DIMMs Devs Slow (<=333MHz) Fast (>334MHz) */
{ 1, 4, 0x0837100FF, 0x056960004, 0x0827100FF, 0x056960004 },
{ 1, 8, 0x0837100AA, 0x056960004, 0x0827100AA, 0x056960004 },
{ 1, 16, 0x0837100AA, 0x056960004, 0x082710055, 0x056960004 },
{ 2, 8, 0x0837100A5, 0x056960004, 0x082710000, 0x056960004 },
{ 2, 16, 0x0937100A5, 0x056960004, 0x0C27100A5, 0x056960004 },
{ 2, 20, 0x0B37100A5, 0x056960004, 0x0B27100A5, 0x056960004 },
{ 2, 24, 0x0B37100A5, 0x056960004, 0x0B27100A5, 0x056960004 },
{ 2, 32, 0x0B37100A5, 0x056960004, 0x0B2710000, 0x056960004 },
};
/*
* Bit 55 (disable SDCLK 1,3,5) should be set if there is a single DIMM
* in slot 0, but it should be clear for all 2 DIMM settings and if a
* single DIMM is in slot 1. Bits 54:52 should always be set to '111'.
*
* Settings for single DIMM and no VTT termination (like DB800 platform)
* 0xF2F100FF 0x56960004
* -------------------------------------
* ADDR/CTL have 22 ohm series R
* DQ/DQM/DQS have 33 ohm series R
*/
/**
* This is Black Magic DRAM timing juju[1].
*
* DRAM delay depends on CPU clock, memory bus clock, memory bus loading,
* memory bus termination, your middle initial (ha! caught you!), GeodeLink
* clock rate, and DRAM timing specifications.
*
* From this the code computes a number which is "known to work". No,
* hardware is not an exact science. And, finally, if an FS2 (JTAG debugger)
* is hooked up, then just don't do anything. This code was written by a master
* of the Dark Arts at AMD and should not be modified in any way.
*
* [1] (http://www.thefreedictionary.com/juju)
*
* @param dimm0 The SMBus address of DIMM 0 (mainboard dependent).
* @param dimm1 The SMBus address of DIMM 1 (mainboard dependent).
* @param terminated The bus is terminated. (mainboard dependent).
*/
static void set_delay_control(u8 dimm0, u8 dimm1, int terminated)
{
u32 glspeed;
u8 spdbyte0, spdbyte1, dimms, i;
struct msr msr;
glspeed = geode_link_speed();
/* Fix delay controls for DM and IM arrays. */
for (i = 0; i < ARRAY_SIZE(msr_table); i++)
wrmsr(msr_table[i].msrnum, msr_table[i].msr);
msr = rdmsr(GLCP_FIFOCTL);
msr.hi = 0x00000005;
wrmsr(GLCP_FIFOCTL, msr);
/* Enable setting. */
msr.hi = 0;
msr.lo = 0x00000001;
wrmsr(CPU_BC_MSS_ARRAY_CTL_ENA, msr);
/* Debug Delay Control setup check.
* Leave it alone if it has been setup. FS2 or something is here.
*/
msr = rdmsr(GLCP_DELAY_CONTROLS);
if (msr.lo & ~(DELAY_LOWER_STATUS_MASK))
return;
/* Delay Controls based on DIMM loading. UGH!
* Number of devices = module width (SPD 6) / device width (SPD 13)
* * physical banks (SPD 5)
*
* Note: We only support a module width of 64.
*/
dimms = 0;
spdbyte0 = spd_read_byte(dimm0, SPD_PRIMARY_SDRAM_WIDTH);
if (spdbyte0 != 0xFF) {
dimms++;
spdbyte0 = (u8)64 / spdbyte0 *
(u8)(spd_read_byte(dimm0, SPD_NUM_DIMM_BANKS));
} else {
spdbyte0 = 0;
}
spdbyte1 = spd_read_byte(dimm1, SPD_PRIMARY_SDRAM_WIDTH);
if (spdbyte1 != 0xFF) {
dimms++;
spdbyte1 = (u8)64 / spdbyte1 *
(u8)(spd_read_byte(dimm1, SPD_NUM_DIMM_BANKS));
} else {
spdbyte1 = 0;
}
/* Zero GLCP_DELAY_CONTROLS MSR */
msr.hi = msr.lo = 0;
/* Save some power, disable clock to second DIMM if it is empty. */
if (spdbyte1 == 0)
msr.hi |= DELAY_UPPER_DISABLE_CLK135;
spdbyte0 += spdbyte1;
if ((dimms == 1) && (terminated == DRAM_TERMINATED)) {
msr.hi = 0xF2F100FF;
msr.lo = 0x56960004;
} else for (i = 0; i < ARRAY_SIZE(delay_control_table); i++) {
if ((dimms == delay_control_table[i].dimms) &&
(spdbyte0 <= delay_control_table[i].devices)) {
if (glspeed < 334) {
msr.hi |= delay_control_table[i].slow_hi;
msr.lo |= delay_control_table[i].slow_low;
} else {
msr.hi |= delay_control_table[i].fast_hi;
msr.lo |= delay_control_table[i].fast_low;
}
break;
}
}
wrmsr(GLCP_DELAY_CONTROLS, msr);
}
/**
* All CPU register settings, here in one place, and done in the proper order.
*
* @param debug_clock_disable Disable the debug clock to save power. Currently
* ignored, but we need to pick this up from a CMOS
* setting in future.
* @param dimm0 SMBus address of DIMM 0 (mainboard dependent).
* @param dimm1 SMBus address of DIMM 1 (mainboard dependent).
* @param terminated The bus is terminated (mainboard dependent).
*/
void cpu_reg_init(int debug_clock_disable, u8 dimm0, u8 dimm1, int terminated)
{
struct msr msr;
/* Castle 2.0 BTM periodic sync period. */
/* [40:37] 1 sync record per 256 bytes. */
msr = rdmsr(CPU_PF_CONF);
msr.hi |= (0x8 << 5);
wrmsr(CPU_PF_CONF, msr);
/* Castle performance setting.
* Enable Quack for fewer re-RAS on the MC.
*/
msr = rdmsr(GLIU0_ARB);
msr.hi &= ~ARB_UPPER_DACK_EN_SET;
msr.hi |= ARB_UPPER_QUACK_EN_SET;
wrmsr(GLIU0_ARB, msr);
msr = rdmsr(GLIU1_ARB);
msr.hi &= ~ARB_UPPER_DACK_EN_SET;
msr.hi |= ARB_UPPER_QUACK_EN_SET;
wrmsr(GLIU1_ARB, msr);
/* GLIU port active enable, limit south pole masters (AES and PCI) to
* one outstanding transaction.
*/
msr = rdmsr(GLIU1_PORT_ACTIVE);
msr.lo &= ~0x880;
wrmsr(GLIU1_PORT_ACTIVE, msr);
/* Set the Delay Control in GLCP. */
set_delay_control(dimm0, dimm1, terminated);
/* Enable RSDC. */
msr = rdmsr(CPU_AC_SMM_CTL);
msr.lo |= SMM_INST_EN_SET;
wrmsr(CPU_AC_SMM_CTL, msr);
/* FPU imprecise exceptions bit. */
msr = rdmsr(CPU_FPU_MSR_MODE);
msr.lo |= FPU_IE_SET;
wrmsr(CPU_FPU_MSR_MODE, msr);
/* Power savers (do after BIST). */
/* Enable Suspend on HLT & PAUSE instructions. */
msr = rdmsr(CPU_XC_CONFIG);
msr.lo |= XC_CONFIG_SUSP_ON_HLT | XC_CONFIG_SUSP_ON_PAUSE;
wrmsr(CPU_XC_CONFIG, msr);
/* Enable SUSP and allow TSC to run in Suspend (keep speed
* detection happy).
*/
msr = rdmsr(CPU_BC_CONF_0);
msr.lo |= TSC_SUSP_SET | SUSP_EN_SET;
msr.lo &= 0x0F0FFFFFF;
msr.lo |= 0x002000000; /* PBZ213: Set PAUSEDLY = 2. */
wrmsr(CPU_BC_CONF_0, msr);
/* Disable the debug clock to save power. */
/* Note: Leave it enabled for FS2 debug. */
if (debug_clock_disable && 0) {
msr.hi = 0;
msr.lo = 0;
wrmsr(GLCP_DBGCLKCTL, msr);
}
/* Setup throttling delays to proper mode if it is ever enabled. */
msr.hi = 0;
msr.lo = 0x00000603C;
wrmsr(GLCP_TH_OD, msr);
/* Fix CPU bugs. */
#warning testing fixing bugs in initram
cpu_bug();
}