diff options
Diffstat (limited to 'src/cpu/x86/mtrr')
| -rw-r--r-- | src/cpu/x86/mtrr/Config.lb | 1 | ||||
| -rw-r--r-- | src/cpu/x86/mtrr/earlymtrr.c | 123 | ||||
| -rw-r--r-- | src/cpu/x86/mtrr/mtrr.c | 378 |
3 files changed, 502 insertions, 0 deletions
diff --git a/src/cpu/x86/mtrr/Config.lb b/src/cpu/x86/mtrr/Config.lb new file mode 100644 index 0000000000..df5c16f8bd --- /dev/null +++ b/src/cpu/x86/mtrr/Config.lb @@ -0,0 +1 @@ +object mtrr.o
\ No newline at end of file diff --git a/src/cpu/x86/mtrr/earlymtrr.c b/src/cpu/x86/mtrr/earlymtrr.c new file mode 100644 index 0000000000..e8fda994af --- /dev/null +++ b/src/cpu/x86/mtrr/earlymtrr.c @@ -0,0 +1,123 @@ +#ifndef EARLYMTRR_C +#define EARLYMTRR_C +#include <cpu/x86/cache.h> +#include <cpu/x86/mtrr.h> +#include <cpu/x86/msr.h> + +/* Validate XIP_ROM_SIZE and XIP_ROM_BASE */ +#if defined(XIP_ROM_SIZE) && !defined(XIP_ROM_BASE) +#error "XIP_ROM_SIZE without XIP_ROM_BASE" +#endif +#if defined(XIP_ROM_BASE) && !defined(XIP_ROM_SIZE) +#error "XIP_ROM_BASE without XIP_ROM_SIZE" +#endif +#if !defined(CONFIG_LB_MEM_TOPK) +#error "CONFIG_LB_MEM_TOPK not defined" +#endif + +#if defined(XIP_ROM_SIZE) && ((XIP_ROM_SIZE & (XIP_ROM_SIZE -1)) != 0) +#error "XIP_ROM_SIZE is not a power of 2" +#endif +#if defined(XIP_ROM_SIZE) && ((XIP_ROM_BASE % XIP_ROM_SIZE) != 0) +#error "XIP_ROM_BASE is not a multiple of XIP_ROM_SIZE" +#endif + +#if (CONFIG_LB_MEM_TOPK & (CONFIG_LB_MEM_TOPK -1)) != 0 +# error "CONFIG_LB_MEM_TOPK must be a power of 2" +#endif + +static void disable_var_mtrr(unsigned reg) +{ + /* The invalid bit is kept in the mask so we simply + * clear the relevent mask register to disable a + * range. + */ + msr_t zero; + zero.lo = zero.hi = 0; + wrmsr(MTRRphysMask_MSR(reg), zero); +} + +static void set_var_mtrr( + unsigned reg, unsigned base, unsigned size, unsigned type) + +{ + /* Bit Bit 32-35 of MTRRphysMask should be set to 1 */ + msr_t basem, maskm; + basem.lo = base | type; + basem.hi = 0; + wrmsr(MTRRphysBase_MSR(reg), basem); + maskm.lo = ~(size - 1) | 0x800; + maskm.hi = 0x0f; + wrmsr(MTRRphysMask_MSR(reg), maskm); +} + +static void cache_lbmem(int type) +{ + /* Enable caching for 0 - 1MB using variable mtrr */ + disable_cache(); + set_var_mtrr(0, 0x00000000, CONFIG_LB_MEM_TOPK << 10, type); + enable_cache(); +} + + +/* the fixed and variable MTTRs are power-up with random values, + * clear them to MTRR_TYPE_UNCACHEABLE for safty. + */ +static void do_early_mtrr_init(const unsigned long *mtrr_msrs) +{ + /* Precondition: + * The cache is not enabled in cr0 nor in MTRRdefType_MSR + * entry32.inc ensures the cache is not enabled in cr0 + */ + msr_t msr; + const unsigned long *msr_addr; + unsigned long cr0; + + print_spew("Clearing mtrr\r\n"); + + /* Inialize all of the relevant msrs to 0 */ + msr.lo = 0; + msr.hi = 0; + unsigned long msr_nr; + for(msr_addr = mtrr_msrs; (msr_nr = *msr_addr); msr_addr++) { + wrmsr(msr_nr, msr); + } + +#if defined(XIP_ROM_SIZE) + /* enable write through caching so we can do execute in place + * on the flash rom. + */ + set_var_mtrr(1, XIP_ROM_BASE, XIP_ROM_SIZE, MTRR_TYPE_WRBACK); +#endif + + /* Set the default memory type and enable fixed and variable MTRRs + */ + /* Enable Variable MTRRs */ + msr.hi = 0x00000000; + msr.lo = 0x00000800; + wrmsr(MTRRdefType_MSR, msr); + +} + +static void early_mtrr_init(void) +{ + static const unsigned long mtrr_msrs[] = { + /* fixed mtrr */ + 0x250, 0x258, 0x259, + 0x268, 0x269, 0x26A, + 0x26B, 0x26C, 0x26D, + 0x26E, 0x26F, + /* var mtrr */ + 0x200, 0x201, 0x202, 0x203, + 0x204, 0x205, 0x206, 0x207, + 0x208, 0x209, 0x20A, 0x20B, + 0x20C, 0x20D, 0x20E, 0x20F, + /* NULL end of table */ + 0 + }; + disable_cache(); + do_early_mtrr_init(mtrr_msrs); + enable_cache(); +} + +#endif /* EARLYMTRR_C */ diff --git a/src/cpu/x86/mtrr/mtrr.c b/src/cpu/x86/mtrr/mtrr.c new file mode 100644 index 0000000000..8e38f23736 --- /dev/null +++ b/src/cpu/x86/mtrr/mtrr.c @@ -0,0 +1,378 @@ +/* + * intel_mtrr.c: setting MTRR to decent values for cache initialization on P6 + * + * Derived from intel_set_mtrr in intel_subr.c and mtrr.c in linux kernel + * + * Copyright 2000 Silicon Integrated System Corporation + * + * 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., 675 Mass Ave, Cambridge, MA 02139, USA. + * + * + * Reference: Intel Architecture Software Developer's Manual, Volume 3: System Programming + */ + +#include <console/console.h> +#include <device/device.h> +#include <cpu/x86/msr.h> +#include <cpu/x86/mtrr.h> +#include <cpu/x86/cache.h> + +#define arraysize(x) (sizeof(x)/sizeof((x)[0])) + +#warning "FIXME I do not properly handle address more than 36 physical address bits" +#ifdef k8 +# define ADDRESS_BITS 40 +#else +# define ADDRESS_BITS 36 +#endif +#define ADDRESS_BITS_HIGH (ADDRESS_BITS - 32) +#define ADDRESS_MASK_HIGH ((1u << ADDRESS_BITS_HIGH) - 1) + +static unsigned int mtrr_msr[] = { + MTRRfix64K_00000_MSR, MTRRfix16K_80000_MSR, MTRRfix16K_A0000_MSR, + MTRRfix4K_C0000_MSR, MTRRfix4K_C8000_MSR, MTRRfix4K_D0000_MSR, MTRRfix4K_D8000_MSR, + MTRRfix4K_E0000_MSR, MTRRfix4K_E8000_MSR, MTRRfix4K_F0000_MSR, MTRRfix4K_F8000_MSR, +}; + + +static void enable_fixed_mtrr(void) +{ + msr_t msr; + + msr = rdmsr(MTRRdefType_MSR); + msr.lo |= 0xc00; + wrmsr(MTRRdefType_MSR, msr); +} + +static void enable_var_mtrr(void) +{ + msr_t msr; + + msr = rdmsr(MTRRdefType_MSR); + msr.lo |= 0x800; + wrmsr(MTRRdefType_MSR, msr); +} + +/* setting variable mtrr, comes from linux kernel source */ +static void set_var_mtrr(unsigned int reg, unsigned long basek, unsigned long sizek, unsigned char type) +{ + msr_t base, mask; + + base.hi = basek >> 22; + base.lo = basek << 10; + + //printk_debug("ADDRESS_MASK_HIGH=%#x\n", ADDRESS_MASK_HIGH); + + if (sizek < 4*1024*1024) { + mask.hi = ADDRESS_MASK_HIGH; + mask.lo = ~((sizek << 10) -1); + } + else { + mask.hi = ADDRESS_MASK_HIGH & (~((sizek >> 22) -1)); + mask.lo = 0; + } + + if (reg >= 8) + return; + + // it is recommended that we disable and enable cache when we + // do this. + disable_cache(); + if (sizek == 0) { + msr_t zero; + zero.lo = zero.hi = 0; + /* The invalid bit is kept in the mask, so we simply clear the + relevant mask register to disable a range. */ + wrmsr (MTRRphysMask_MSR(reg), zero); + } else { + /* Bit 32-35 of MTRRphysMask should be set to 1 */ + base.lo |= type; + mask.lo |= 0x800; + wrmsr (MTRRphysBase_MSR(reg), base); + wrmsr (MTRRphysMask_MSR(reg), mask); + } + enable_cache(); +} + +/* fms: find most sigificant bit set, stolen from Linux Kernel Source. */ +static inline unsigned int fms(unsigned int x) +{ + int r; + + __asm__("bsrl %1,%0\n\t" + "jnz 1f\n\t" + "movl $0,%0\n" + "1:" : "=r" (r) : "g" (x)); + return r; +} + +/* fms: find least sigificant bit set */ +static inline unsigned int fls(unsigned int x) +{ + int r; + + __asm__("bsfl %1,%0\n\t" + "jnz 1f\n\t" + "movl $32,%0\n" + "1:" : "=r" (r) : "g" (x)); + return r; +} + +/* setting up variable and fixed mtrr + * + * From Intel Vol. III Section 9.12.4, the Range Size and Base Alignment has some kind of requirement: + * 1. The range size must be 2^N byte for N >= 12 (i.e 4KB minimum). + * 2. The base address must be 2^N aligned, where the N here is equal to the N in previous + * requirement. So a 8K range must be 8K aligned not 4K aligned. + * + * These requirement is meet by "decompositing" the ramsize into Sum(Cn * 2^n, n = [0..N], Cn = [0, 1]). + * For Cm = 1, there is a WB range of 2^m size at base address Sum(Cm * 2^m, m = [N..n]). + * A 124MB (128MB - 4MB SMA) example: + * ramsize = 124MB == 64MB (at 0MB) + 32MB (at 64MB) + 16MB (at 96MB ) + 8MB (at 112MB) + 4MB (120MB). + * But this wastes a lot of MTRR registers so we use another more "aggresive" way with Uncacheable Regions. + * + * In the Uncacheable Region scheme, we try to cover the whole ramsize by one WB region as possible, + * If (an only if) this can not be done we will try to decomposite the ramesize, the mathematical formula + * whould be ramsize = Sum(Cn * 2^n, n = [0..N], Cn = [-1, 0, 1]). For Cn = -1, a Uncachable Region is used. + * The same 124MB example: + * ramsize = 124MB == 128MB WB (at 0MB) + 4MB UC (at 124MB) + * or a 156MB (128MB + 32MB - 4MB SMA) example: + * ramsize = 156MB == 128MB WB (at 0MB) + 32MB WB (at 128MB) + 4MB UC (at 156MB) + */ +/* 2 MTRRS are reserved for the operating system */ +#if 0 +#define BIOS_MTRRS 6 +#define OS_MTRRS 2 +#else +#define BIOS_MTRRS 8 +#define OS_MTRRS 0 +#endif +#define MTRRS (BIOS_MTRRS + OS_MTRRS) + + +static void set_fixed_mtrrs(unsigned int first, unsigned int last, unsigned char type) +{ + unsigned int i; + unsigned int fixed_msr = NUM_FIXED_RANGES >> 3; + msr_t msr; + msr.lo = msr.hi = 0; /* Shut up gcc */ + for(i = first; i < last; i++) { + /* When I switch to a new msr read it in */ + if (fixed_msr != i >> 3) { + /* But first write out the old msr */ + if (fixed_msr < (NUM_FIXED_RANGES >> 3)) { + disable_cache(); + wrmsr(mtrr_msr[fixed_msr], msr); + enable_cache(); + } + fixed_msr = i>>3; + msr = rdmsr(mtrr_msr[fixed_msr]); + } + if ((i & 7) < 4) { + msr.lo &= ~(0xff << ((i&3)*8)); + msr.lo |= type << ((i&3)*8); + } else { + msr.hi &= ~(0xff << ((i&3)*8)); + msr.hi |= type << ((i&3)*8); + } + } + /* Write out the final msr */ + if (fixed_msr < (NUM_FIXED_RANGES >> 3)) { + disable_cache(); + wrmsr(mtrr_msr[fixed_msr], msr); + enable_cache(); + } +} + +static unsigned fixed_mtrr_index(unsigned long addrk) +{ + unsigned index; + index = (addrk - 0) >> 6; + if (index >= 8) { + index = ((addrk - 8*64) >> 4) + 8; + } + if (index >= 24) { + index = ((addrk - (8*64 + 16*16)) >> 2) + 24; + } + if (index > NUM_FIXED_RANGES) { + index = NUM_FIXED_RANGES; + } + return index; +} + +static unsigned int range_to_mtrr(unsigned int reg, + unsigned long range_startk, unsigned long range_sizek, + unsigned long next_range_startk) +{ + if (!range_sizek || (reg >= BIOS_MTRRS)) { + return reg; + } + while(range_sizek) { + unsigned long max_align, align; + unsigned long sizek; + /* Compute the maximum size I can make a range */ + max_align = fls(range_startk); + align = fms(range_sizek); + if (align > max_align) { + align = max_align; + } + sizek = 1 << align; + printk_debug("Setting variable MTRR %d, base: %4dMB, range: %4dMB, type WB\n", + reg, range_startk >>10, sizek >> 10); + set_var_mtrr(reg++, range_startk, sizek, MTRR_TYPE_WRBACK); + range_startk += sizek; + range_sizek -= sizek; + if (reg >= BIOS_MTRRS) + break; + } + return reg; +} + +static unsigned long resk(uint64_t value) +{ + unsigned long resultk; + if (value < (1ULL << 42)) { + resultk = value >> 10; + } + else { + resultk = 0xffffffff; + } + return resultk; +} + +void x86_setup_mtrrs(void) +{ + /* Try this the simple way of incrementally adding together + * mtrrs. If this doesn't work out we can get smart again + * and clear out the mtrrs. + */ + struct device *dev; + unsigned long range_startk, range_sizek; + unsigned int reg; + + printk_debug("\n"); + /* Initialized the fixed_mtrrs to uncached */ + printk_debug("Setting fixed MTRRs(%d-%d) type: UC\n", + 0, NUM_FIXED_RANGES); + set_fixed_mtrrs(0, NUM_FIXED_RANGES, MTRR_TYPE_UNCACHEABLE); + + /* Now see which of the fixed mtrrs cover ram. + */ + for(dev = all_devices; dev; dev = dev->next) { + struct resource *res, *last; + last = &dev->resource[dev->resources]; + for(res = &dev->resource[0]; res < last; res++) { + unsigned int start_mtrr; + unsigned int last_mtrr; + if (!(res->flags & IORESOURCE_MEM) || + !(res->flags & IORESOURCE_CACHEABLE)) + { + continue; + } + start_mtrr = fixed_mtrr_index(resk(res->base)); + last_mtrr = fixed_mtrr_index(resk((res->base + res->size))); + if (start_mtrr >= NUM_FIXED_RANGES) { + break; + } + printk_debug("Setting fixed MTRRs(%d-%d) Type: WB\n", + start_mtrr, last_mtrr); + set_fixed_mtrrs(start_mtrr, last_mtrr, MTRR_TYPE_WRBACK); + } + } + printk_debug("DONE fixed MTRRs\n"); + /* Cache as many memory areas as possible */ + /* FIXME is there an algorithm for computing the optimal set of mtrrs? + * In some cases it is definitely possible to do better. + */ + range_startk = 0; + range_sizek = 0; + reg = 0; + for(dev = all_devices; dev; dev = dev->next) { + struct resource *res, *last; + last = &dev->resource[dev->resources]; + for(res = &dev->resource[0]; res < last; res++) { + unsigned long basek, sizek; + if (!(res->flags & IORESOURCE_MEM) || + !(res->flags & IORESOURCE_CACHEABLE)) { + continue; + } + basek = resk(res->base); + sizek = resk(res->size); + /* See if I can merge with the last range + * Either I am below 1M and the fixed mtrrs handle it, or + * the ranges touch. + */ + if ((basek <= 1024) || (range_startk + range_sizek == basek)) { + unsigned long endk = basek + sizek; + range_sizek = endk - range_startk; + continue; + } + /* Write the range mtrrs */ + if (range_sizek != 0) { + reg = range_to_mtrr(reg, range_startk, range_sizek, basek); + range_startk = 0; + range_sizek = 0; + if (reg >= BIOS_MTRRS) + goto last_msr; + } + /* Allocate an msr */ + range_startk = basek; + range_sizek = sizek; + } + } + last_msr: + /* Write the last range */ + reg = range_to_mtrr(reg, range_startk, range_sizek, 0); + printk_debug("DONE variable MTRRs\n"); + printk_debug("Clear out the extra MTRR's\n"); + /* Clear out the extra MTRR's */ + while(reg < MTRRS) { + set_var_mtrr(reg++, 0, 0, 0); + } + /* enable fixed MTRR */ + printk_spew("call enable_fixed_mtrr()\n"); + enable_fixed_mtrr(); + printk_spew("call enable_var_mtrr()\n"); + enable_var_mtrr(); + printk_spew("Leave %s\n", __FUNCTION__); + post_code(0x6A); +} + +int x86_mtrr_check(void) +{ + /* Only Pentium Pro and later have MTRR */ + msr_t msr; + printk_debug("\nMTRR check\n"); + + msr = rdmsr(0x2ff); + msr.lo >>= 10; + + printk_debug("Fixed MTRRs : "); + if (msr.lo & 0x01) + printk_debug("Enabled\n"); + else + printk_debug("Disabled\n"); + + printk_debug("Variable MTRRs: "); + if (msr.lo & 0x02) + printk_debug("Enabled\n"); + else + printk_debug("Disabled\n"); + + printk_debug("\n"); + + post_code(0x93); + return ((int) msr.lo); +} |