Avoid hang when 4GB or more DRAM is installed on AMD RS780 UMA systems.
-- When building for UMA, reduce the limit for DRAM below 4GB from E0000000 to C0000000. This is needed to accomodate the UMA frame buffer. -- Correct problem where msr C0010010 bits 21 and 22 (MtrrTom2En and Tom2ForceMemTypeWB) are not set consistently across cores. -- Enable TOM2 only if DRAM is present above 4GB. -- Use AMD Tom2ForceMemTypeWB feature to avoid the need for variable MTRR ranges above 4GB. -- Split function x86_setup_var_mtrrs into standard and AMD versions. The AMD version of the function relies on Tom2ForceMemTypeWB for making DRAM above 4GB type WB. The AMD version also incorporates a change to avoid an unexpected variable MTRR range when using UMA graphics. -- Improve white space consistency for mtrr.c by using tabs in place of spaces. Tested on kino-780am2-fam10 with 2GB and 4GB.
Signed-off-by: Scott Duplichan scott@notabs.org
Index: src/cpu/amd/mtrr/amd_mtrr.c =================================================================== --- src/cpu/amd/mtrr/amd_mtrr.c (revision 6056) +++ src/cpu/amd/mtrr/amd_mtrr.c (working copy) @@ -107,14 +107,14 @@ unsigned long address_bits; struct mem_state state; unsigned long i; - msr_t msr; + msr_t msr, sys_cfg;
/* Enable the access to AMD RdDram and WrDram extension bits */ disable_cache(); - msr = rdmsr(SYSCFG_MSR); - msr.lo |= SYSCFG_MSR_MtrrFixDramModEn; - wrmsr(SYSCFG_MSR, msr); + sys_cfg = rdmsr(SYSCFG_MSR); + sys_cfg.lo |= SYSCFG_MSR_MtrrFixDramModEn; + wrmsr(SYSCFG_MSR, sys_cfg); enable_cache();
printk(BIOS_DEBUG, "\n"); @@ -148,11 +148,18 @@ msr.lo = state.mmio_basek << 10; wrmsr(TOP_MEM, msr);
+ // assume no DRAM above 4GB + sys_cfg.lo &= ~SYSCFG_MSR_TOM2En; + sys_cfg.lo &= ~SYSCFG_MSR_TOM2WB; if(state.tomk > (4*1024*1024)) { /* Setup TOP_MEM2 */ msr.hi = state.tomk >> 22; msr.lo = state.tomk << 10; wrmsr(TOP_MEM2, msr); + // enable tom2 + sys_cfg.lo |= SYSCFG_MSR_TOM2En; + // Make DRAM above 4GB WB without using any variable MTRR ranges + sys_cfg.lo |= SYSCFG_MSR_TOM2WB; }
/* zero the IORR's before we enable to prevent @@ -167,10 +174,9 @@ * Enable the RdMem and WrMem bits in the fixed mtrrs. * Disable access to the RdMem and WrMem in the fixed mtrr. */ - msr = rdmsr(SYSCFG_MSR); - msr.lo |= SYSCFG_MSR_MtrrVarDramEn | SYSCFG_MSR_MtrrFixDramEn | SYSCFG_MSR_TOM2En; - msr.lo &= ~SYSCFG_MSR_MtrrFixDramModEn; - wrmsr(SYSCFG_MSR, msr); + sys_cfg.lo |= SYSCFG_MSR_MtrrVarDramEn | SYSCFG_MSR_MtrrFixDramEn; + sys_cfg.lo &= ~SYSCFG_MSR_MtrrFixDramModEn; + wrmsr(SYSCFG_MSR, sys_cfg);
enable_fixed_mtrr();
@@ -186,5 +192,5 @@ /* Now that I have mapped what is memory and what is not * Setup the mtrrs so we can cache the memory. */ - x86_setup_var_mtrrs(address_bits); + amd_setup_var_mtrrs(address_bits); } Index: src/cpu/x86/mtrr/mtrr.c =================================================================== --- src/cpu/x86/mtrr/mtrr.c (revision 6056) +++ src/cpu/x86/mtrr/mtrr.c (working copy) @@ -364,31 +364,30 @@
void x86_setup_fixed_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. - */ + /* 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. + */
- printk(BIOS_DEBUG, "\n"); - /* Initialized the fixed_mtrrs to uncached */ - printk(BIOS_DEBUG, "Setting fixed MTRRs(%d-%d) Type: UC\n", - 0, NUM_FIXED_RANGES); - set_fixed_mtrrs(0, NUM_FIXED_RANGES, MTRR_TYPE_UNCACHEABLE); + printk(BIOS_DEBUG, "\n"); + /* Initialized the fixed_mtrrs to uncached */ + printk(BIOS_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. - */ - search_global_resources( + /* Now see which of the fixed mtrrs cover ram. */ + search_global_resources( IORESOURCE_MEM | IORESOURCE_CACHEABLE, IORESOURCE_MEM | IORESOURCE_CACHEABLE, set_fixed_mtrr_resource, NULL); - printk(BIOS_DEBUG, "DONE fixed MTRRs\n"); + printk(BIOS_DEBUG, "DONE fixed MTRRs\n");
- /* enable fixed MTRR */ - printk(BIOS_SPEW, "call enable_fixed_mtrr()\n"); - enable_fixed_mtrr(); + /* enable fixed MTRR */ + printk(BIOS_SPEW, "call enable_fixed_mtrr()\n"); + enable_fixed_mtrr();
}
-void x86_setup_var_mtrrs(unsigned address_bits) +static void x86_setup_var_mtrrs(unsigned address_bits) /* this routine needs to know how many address bits a given processor * supports. CPUs get grumpy when you set too many bits in * their mtrr registers :( I would generically call cpuid here @@ -449,6 +448,64 @@ post_code(0x6A); }
+ +void amd_setup_var_mtrrs(unsigned address_bits) +/* this routine needs to know how many address bits a given processor + * supports. CPUs get grumpy when you set too many bits in + * their mtrr registers :( I would generically call cpuid here + * and find out how many physically supported but some cpus are + * buggy, and report more bits then they actually support. + */ +{ + /* 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 var_mtrr_state var_state; + + /* 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. + */ + + var_state.range_startk = 0; + var_state.range_sizek = 0; + var_state.hole_startk = 0; + var_state.hole_sizek = 0; + var_state.reg = 0; + var_state.address_bits = address_bits; + + search_global_resources( + IORESOURCE_MEM | IORESOURCE_CACHEABLE, IORESOURCE_MEM | IORESOURCE_CACHEABLE, + set_var_mtrr_resource, &var_state); + + if (var_state.range_sizek >= 0x100000000ull / 1024 && var_state.range_startk == 0) + { + // no variable MTRR range is needed for WB from 4GB-TOM2 + // because Tom2ForceMemTypeWB will be used for this purpose. + var_state.range_sizek = var_state.hole_startk; + } + +#if (CONFIG_GFXUMA == 1) /* UMA or SP. */ + // For now we assume the UMA space is at the end of memory below 4GB + // Reduce the dram wb mtrr range so that it does not cover the uma at the end + var_state.range_sizek -= (uma_memory_size >> 10); +#endif + /* Write the last range */ + var_state.reg = range_to_mtrr(var_state.reg, var_state.range_startk, + var_state.range_sizek, 0, MTRR_TYPE_WRBACK, var_state.address_bits); + printk(BIOS_DEBUG, "DONE variable MTRRs\n"); + printk(BIOS_DEBUG, "Clear out the extra MTRR's\n"); + /* Clear out the extra MTRR's */ + while(var_state.reg < MTRRS) { + set_var_mtrr(var_state.reg++, 0, 0, 0, var_state.address_bits); + } + printk(BIOS_SPEW, "call enable_var_mtrr()\n"); + enable_var_mtrr(); + printk(BIOS_SPEW, "Leave %s\n", __func__); + post_code(0x6A); +} + void x86_setup_mtrrs(unsigned address_bits) { x86_setup_fixed_mtrrs(); Index: src/include/cpu/amd/mtrr.h =================================================================== --- src/include/cpu/amd/mtrr.h (revision 6056) +++ src/include/cpu/amd/mtrr.h (working copy) @@ -8,6 +8,7 @@ #define MTRR_WRITE_MEM (1 << 3)
#define SYSCFG_MSR 0xC0010010 +#define SYSCFG_MSR_TOM2WB (1 << 22) #define SYSCFG_MSR_TOM2En (1 << 21) #define SYSCFG_MSR_MtrrVarDramEn (1 << 20) #define SYSCFG_MSR_MtrrFixDramModEn (1 << 19) Index: src/include/cpu/x86/mtrr.h =================================================================== --- src/include/cpu/x86/mtrr.h (revision 6056) +++ src/include/cpu/x86/mtrr.h (working copy) @@ -37,7 +37,7 @@ #if !defined (ASSEMBLY) && !defined(__PRE_RAM__) #include <device/device.h> void enable_fixed_mtrr(void); -void x86_setup_var_mtrrs(unsigned address_bits); +void amd_setup_var_mtrrs(unsigned address_bits); void x86_setup_mtrrs(unsigned address_bits); int x86_mtrr_check(void); void set_var_mtrr_resource(void *gp, struct device *dev, struct resource *res); Index: src/northbridge/amd/amdmct/wrappers/mcti.h =================================================================== --- src/northbridge/amd/amdmct/wrappers/mcti.h (revision 6056) +++ src/northbridge/amd/amdmct/wrappers/mcti.h (working copy) @@ -42,11 +42,12 @@ //#define SYSTEM_TYPE MOBILE #endif
-/*---------------------------------------------------------------------------- -COMMENT OUT ALL BUT 1 -----------------------------------------------------------------------------*/ +/*--------------------------------------------------------------------------*/ +#if (CONFIG_GFXUMA) +#define UMA_SUPPORT 1 /*Supported */ +#else #define UMA_SUPPORT 0 /*Not supported */ -//#define UMA_SUPPORT 1 /*Supported */ +#endif
/*---------------------------------------------------------------------------- UPDATE AS NEEDED Index: src/northbridge/amd/amdmct/wrappers/mcti_d.c =================================================================== --- src/northbridge/amd/amdmct/wrappers/mcti_d.c (revision 6056) +++ src/northbridge/amd/amdmct/wrappers/mcti_d.c (working copy) @@ -132,13 +132,17 @@ //val = 1; /* enable */ break; case NV_BottomIO: +#if (UMA_SUPPORT == 0) val = 0xE0; /* address bits [31:24] */ +#elif (UMA_SUPPORT == 1) + val = 0xC0; /* address bits [31:24] */ +#endif break; case NV_BottomUMA: #if (UMA_SUPPORT == 0) val = 0xE0; /* address bits [31:24] */ #elif (UMA_SUPPORT == 1) - val = 0xB0; /* address bits [31:24] */ + val = 0xC0; /* address bits [31:24] */ #endif break; case NV_ECC: