From 0f1553b89a226d368f2accd443911a443b8d3a75 Mon Sep 17 00:00:00 2001 From: Timothy Pearson Date: Sun, 2 Aug 2015 21:06:39 -0500 Subject: nb/amd/amdfam10: Add HyperTransport probe filter support All modern Opteron processors support the HT probe filter, which helps to increase coherent fabric performance by reducing the number of HT transactions per cache probe. AMD recommends that the probe filter be enabled on all systems with more than two nodes, and it does not hurt to enable it on systems with 2 nodes. Change-Id: I00a27a828260be8685ae622cfa5a4995add95a8e Signed-off-by: Timothy Pearson Reviewed-on: http://review.coreboot.org/12021 Tested-by: build bot (Jenkins) Tested-by: Raptor Engineering Automated Test Stand Reviewed-by: Alexandru Gagniuc --- src/cpu/amd/family_10h-family_15h/init_cpus.c | 18 ++- .../amd/family_10h-family_15h/model_10xxx_init.c | 22 ++++ src/northbridge/amd/amdfam10/northbridge.c | 144 ++++++++++++++++++++- src/northbridge/amd/amdht/h3ncmn.c | 22 ++++ 4 files changed, 203 insertions(+), 3 deletions(-) diff --git a/src/cpu/amd/family_10h-family_15h/init_cpus.c b/src/cpu/amd/family_10h-family_15h/init_cpus.c index 14d8642083..530ffd2c6f 100644 --- a/src/cpu/amd/family_10h-family_15h/init_cpus.c +++ b/src/cpu/amd/family_10h-family_15h/init_cpus.c @@ -347,9 +347,11 @@ static void STOP_CAR_AND_CPU(uint8_t skip_sharedc_config, uint32_t apicid) static u32 init_cpus(u32 cpu_init_detectedx, struct sys_info *sysinfo) { - u32 bsp_apicid = 0; - u32 apicid; + uint32_t bsp_apicid = 0; + uint32_t apicid; + uint32_t dword; uint8_t set_mtrrs; + uint8_t node_count; struct node_core_id id; /* Please refer to the calculations and explaination in cache_as_ram.inc before modifying these values */ @@ -441,6 +443,18 @@ static u32 init_cpus(u32 cpu_init_detectedx, struct sys_info *sysinfo) cpuSetAMDMSR(id.nodeid); + /* Set up HyperTransport probe filter support */ + if (is_gt_rev_d()) { + dword = pci_read_config32(NODE_PCI(id.nodeid, 0), 0x60); + node_count = ((dword >> 4) & 0x7) + 1; + + if (node_count > 1) { + msr_t msr = rdmsr(BU_CFG2_MSR); + msr.hi |= 1 << (42 - 32); + wrmsr(BU_CFG2_MSR, msr); + } + } + #if CONFIG_SET_FIDVID #if CONFIG_LOGICAL_CPUS && CONFIG_SET_FIDVID_CORE0_ONLY // Run on all AP for proper FID/VID setup. diff --git a/src/cpu/amd/family_10h-family_15h/model_10xxx_init.c b/src/cpu/amd/family_10h-family_15h/model_10xxx_init.c index 3bda14c95a..fe9fb9cd0c 100644 --- a/src/cpu/amd/family_10h-family_15h/model_10xxx_init.c +++ b/src/cpu/amd/family_10h-family_15h/model_10xxx_init.c @@ -50,6 +50,28 @@ static inline uint8_t is_fam15h(void) return fam15h; } +static inline uint8_t is_gt_rev_d(void) +{ + uint8_t fam15h = 0; + uint8_t rev_gte_d = 0; + uint32_t family; + uint32_t model; + + family = model = cpuid_eax(0x80000001); + model = ((model & 0xf0000) >> 12) | ((model & 0xf0) >> 4); + family = ((family & 0xf00000) >> 16) | ((family & 0xf00) >> 8); + + if (family >= 0x6f) + /* Family 15h or later */ + fam15h = 1; + + if ((model >= 0x8) || fam15h) + /* Revision D or later */ + rev_gte_d = 1; + + return rev_gte_d; +} + static volatile uint8_t fam15h_startup_flags[MAX_NODES_SUPPORTED][MAX_CORES_SUPPORTED] = {{ 0 }}; static void model_10xxx_init(device_t dev) diff --git a/src/northbridge/amd/amdfam10/northbridge.c b/src/northbridge/amd/amdfam10/northbridge.c index a922496b10..2684d89449 100644 --- a/src/northbridge/amd/amdfam10/northbridge.c +++ b/src/northbridge/amd/amdfam10/northbridge.c @@ -26,10 +26,13 @@ #include #include #include +#include #include +#include #include #include +#include #include #if CONFIG_LOGICAL_CPUS @@ -1523,7 +1526,7 @@ static void cpu_bus_scan(device_t dev) if(i>=32) { busn--; devn-=32; - pbus = pci_domain->link_list->next); + pbus = pci_domain->link_list->next; } #endif @@ -1643,8 +1646,147 @@ static void cpu_bus_scan(device_t dev) } } +static void detect_and_enable_probe_filter(device_t dev) +{ + uint32_t dword; + + uint8_t fam15h = 0; + uint8_t rev_gte_d = 0; + uint8_t dual_node = 0; + unsigned nb_cfg_54; + uint32_t f3xe8; + uint32_t family; + uint32_t model; + + family = model = cpuid_eax(0x80000001); + model = ((model & 0xf0000) >> 12) | ((model & 0xf0) >> 4); + + if (is_fam15h()) { + /* Family 15h or later */ + fam15h = 1; + nb_cfg_54 = 1; + } + + if ((model >= 0x8) || fam15h) + /* Revision D or later */ + rev_gte_d = 1; + + if (rev_gte_d) + /* Check for dual node capability */ + if (f3xe8 & 0x20000000) + dual_node = 1; + + if (rev_gte_d && (sysconf.nodes > 1)) { + /* Enable the probe filter */ + uint8_t i; + uint8_t pfmode = 0x0; + + uint32_t f3x58[MAX_NODES_SUPPORTED]; + uint32_t f3x5c[MAX_NODES_SUPPORTED]; + + printk(BIOS_DEBUG, "Enabling probe filter\n"); + + /* Disable L3 and DRAM scrubbers and configure system for probe filter support */ + for (i = 0; i < sysconf.nodes; i++) { + device_t f2x_dev = dev_find_slot(0, PCI_DEVFN(0x18 + i, 2)); + device_t f3x_dev = dev_find_slot(0, PCI_DEVFN(0x18 + i, 3)); + + f3x58[i] = pci_read_config32(f3x_dev, 0x58); + f3x5c[i] = pci_read_config32(f3x_dev, 0x5c); + pci_write_config32(f3x_dev, 0x58, f3x58[i] & ~((0x1f << 24) | 0x1f)); + pci_write_config32(f3x_dev, 0x5c, f3x5c[i] & ~0x1); + + dword = pci_read_config32(f2x_dev, 0x1b0); + dword &= ~(0x7 << 8); /* CohPrefPrbLmt = 0x0 */ + pci_write_config32(f2x_dev, 0x1b0, dword); + + msr_t msr = rdmsr_amd(BU_CFG2_MSR); + msr.hi |= 1 << (42 - 32); + wrmsr_amd(BU_CFG2_MSR, msr); + + if (is_fam15h()) { + uint8_t subcache_size = 0x0; + uint8_t pref_so_repl = 0x0; + uint32_t f3x1c4 = pci_read_config32(f3x_dev, 0x1c4); + if ((f3x1c4 & 0xffff) == 0xcccc) { + subcache_size = 0x1; + pref_so_repl = 0x2; + pfmode = 0x3; + } else { + pfmode = 0x2; + } + + dword = pci_read_config32(f3x_dev, 0x1d4); + dword |= 0x1 << 29; /* PFLoIndexHashEn = 0x1 */ + dword &= ~(0x3 << 20); /* PFPreferredSORepl = pref_so_repl */ + dword |= (pref_so_repl & 0x3) << 20; + dword |= 0x1 << 17; /* PFWayHashEn = 0x1 */ + dword |= 0xf << 12; /* PFSubCacheEn = 0xf */ + dword &= ~(0x3 << 10); /* PFSubCacheSize3 = subcache_size */ + dword |= (subcache_size & 0x3) << 10; + dword &= ~(0x3 << 8); /* PFSubCacheSize2 = subcache_size */ + dword |= (subcache_size & 0x3) << 8; + dword &= ~(0x3 << 6); /* PFSubCacheSize1 = subcache_size */ + dword |= (subcache_size & 0x3) << 6; + dword &= ~(0x3 << 4); /* PFSubCacheSize0 = subcache_size */ + dword |= (subcache_size & 0x3) << 4; + dword &= ~(0x3 << 2); /* PFWayNum = 0x2 */ + dword |= 0x2 << 2; + pci_write_config32(f3x_dev, 0x1d4, dword); + } else { + pfmode = 0x2; + + dword = pci_read_config32(f3x_dev, 0x1d4); + dword |= 0x1 << 29; /* PFLoIndexHashEn = 0x1 */ + dword &= ~(0x3 << 20); /* PFPreferredSORepl = 0x2 */ + dword |= 0x2 << 20; + dword |= 0xf << 12; /* PFSubCacheEn = 0xf */ + dword &= ~(0x3 << 10); /* PFSubCacheSize3 = 0x0 */ + dword &= ~(0x3 << 8); /* PFSubCacheSize2 = 0x0 */ + dword &= ~(0x3 << 6); /* PFSubCacheSize1 = 0x0 */ + dword &= ~(0x3 << 4); /* PFSubCacheSize0 = 0x0 */ + dword &= ~(0x3 << 2); /* PFWayNum = 0x2 */ + dword |= 0x2 << 2; + pci_write_config32(f3x_dev, 0x1d4, dword); + } + } + + udelay(40); + + disable_cache(); + wbinvd(); + for (i = 0; i < sysconf.nodes; i++) { + device_t f3x_dev = dev_find_slot(0, PCI_DEVFN(0x18 + i, 3)); + + dword = pci_read_config32(f3x_dev, 0x1c4); + dword |= (0x1 << 31); /* L3TagInit = 1 */ + pci_write_config32(f3x_dev, 0x1c4, dword); + do { + } while (pci_read_config32(f3x_dev, 0x1c4) & (0x1 << 31)); + + dword = pci_read_config32(f3x_dev, 0x1d4); + dword &= ~0x3; /* PFMode = pfmode */ + dword |= pfmode & 0x3; + pci_write_config32(f3x_dev, 0x1d4, dword); + do { + } while (!(pci_read_config32(f3x_dev, 0x1d4) & (0x1 << 19))); + } + enable_cache(); + + /* Reenable L3 and DRAM scrubbers */ + for (i = 0; i < sysconf.nodes; i++) { + device_t f3x_dev = dev_find_slot(0, PCI_DEVFN(0x18 + i, 3)); + + pci_write_config32(f3x_dev, 0x58, f3x58[i]); + pci_write_config32(f3x_dev, 0x5c, f3x5c[i]); + } + + } +} + static void cpu_bus_init(device_t dev) { + detect_and_enable_probe_filter(dev); initialize_cpus(dev->link_list); #if CONFIG_AMD_SB_CIMX sb_After_Pci_Init(); diff --git a/src/northbridge/amd/amdht/h3ncmn.c b/src/northbridge/amd/amdht/h3ncmn.c index 01c0e48d0c..8ee929d12e 100644 --- a/src/northbridge/amd/amdht/h3ncmn.c +++ b/src/northbridge/amd/amdht/h3ncmn.c @@ -103,6 +103,28 @@ static inline uint8_t is_fam15h(void) return fam15h; } +static inline uint8_t is_gt_rev_d(void) +{ + uint8_t fam15h = 0; + uint8_t rev_gte_d = 0; + uint32_t family; + uint32_t model; + + family = model = cpuid_eax(0x80000001); + model = ((model & 0xf0000) >> 12) | ((model & 0xf0) >> 4); + family = ((family & 0xf00000) >> 16) | ((family & 0xf00) >> 8); + + if (family >= 0x6f) + /* Family 15h or later */ + fam15h = 1; + + if ((model >= 0x8) || fam15h) + /* Revision D or later */ + rev_gte_d = 1; + + return rev_gte_d; +} + /***************************************************************************//** * * SBDFO -- cgit v1.2.3