diff options
Diffstat (limited to 'drivers/edac/amd64_edac.c')
| -rw-r--r-- | drivers/edac/amd64_edac.c | 1282 |
1 files changed, 576 insertions, 706 deletions
diff --git a/drivers/edac/amd64_edac.c b/drivers/edac/amd64_edac.c index a38831c8264..c5facd951dd 100644 --- a/drivers/edac/amd64_edac.c +++ b/drivers/edac/amd64_edac.c @@ -13,32 +13,56 @@ module_param(report_gart_errors, int, 0644); static int ecc_enable_override; module_param(ecc_enable_override, int, 0644); +static struct msr *msrs; + /* Lookup table for all possible MC control instances */ struct amd64_pvt; static struct mem_ctl_info *mci_lookup[EDAC_MAX_NUMNODES]; static struct amd64_pvt *pvt_lookup[EDAC_MAX_NUMNODES]; /* - * See F2x80 for K8 and F2x[1,0]80 for Fam10 and later. The table below is only - * for DDR2 DRAM mapping. + * Address to DRAM bank mapping: see F2x80 for K8 and F2x[1,0]80 for Fam10 and + * later. */ -u32 revf_quad_ddr2_shift[] = { - 0, /* 0000b NULL DIMM (128mb) */ - 28, /* 0001b 256mb */ - 29, /* 0010b 512mb */ - 29, /* 0011b 512mb */ - 29, /* 0100b 512mb */ - 30, /* 0101b 1gb */ - 30, /* 0110b 1gb */ - 31, /* 0111b 2gb */ - 31, /* 1000b 2gb */ - 32, /* 1001b 4gb */ - 32, /* 1010b 4gb */ - 33, /* 1011b 8gb */ - 0, /* 1100b future */ - 0, /* 1101b future */ - 0, /* 1110b future */ - 0 /* 1111b future */ +static int ddr2_dbam_revCG[] = { + [0] = 32, + [1] = 64, + [2] = 128, + [3] = 256, + [4] = 512, + [5] = 1024, + [6] = 2048, +}; + +static int ddr2_dbam_revD[] = { + [0] = 32, + [1] = 64, + [2 ... 3] = 128, + [4] = 256, + [5] = 512, + [6] = 256, + [7] = 512, + [8 ... 9] = 1024, + [10] = 2048, +}; + +static int ddr2_dbam[] = { [0] = 128, + [1] = 256, + [2 ... 4] = 512, + [5 ... 6] = 1024, + [7 ... 8] = 2048, + [9 ... 10] = 4096, + [11] = 8192, +}; + +static int ddr3_dbam[] = { [0] = -1, + [1] = 256, + [2] = 512, + [3 ... 4] = -1, + [5 ... 6] = 1024, + [7 ... 8] = 2048, + [9 ... 10] = 4096, + [11] = 8192, }; /* @@ -164,11 +188,9 @@ static int amd64_get_scrub_rate(struct mem_ctl_info *mci, u32 *bw) { struct amd64_pvt *pvt = mci->pvt_info; u32 scrubval = 0; - int status = -1, i, ret = 0; + int status = -1, i; - ret = pci_read_config_dword(pvt->misc_f3_ctl, K8_SCRCTRL, &scrubval); - if (ret) - debugf0("Reading K8_SCRCTRL failed\n"); + amd64_read_pci_cfg(pvt->misc_f3_ctl, K8_SCRCTRL, &scrubval); scrubval = scrubval & 0x001F; @@ -189,7 +211,7 @@ static int amd64_get_scrub_rate(struct mem_ctl_info *mci, u32 *bw) /* Map from a CSROW entry to the mask entry that operates on it */ static inline u32 amd64_map_to_dcs_mask(struct amd64_pvt *pvt, int csrow) { - if (boot_cpu_data.x86 == 0xf && pvt->ext_model < OPTERON_CPU_REV_F) + if (boot_cpu_data.x86 == 0xf && pvt->ext_model < K8_REV_F) return csrow; else return csrow >> 1; @@ -437,7 +459,7 @@ int amd64_get_dram_hole_info(struct mem_ctl_info *mci, u64 *hole_base, u64 base; /* only revE and later have the DRAM Hole Address Register */ - if (boot_cpu_data.x86 == 0xf && pvt->ext_model < OPTERON_CPU_REV_E) { + if (boot_cpu_data.x86 == 0xf && pvt->ext_model < K8_REV_E) { debugf1(" revision %d for node %d does not support DHAR\n", pvt->ext_model, pvt->mc_node_id); return 1; @@ -743,21 +765,6 @@ static void find_csrow_limits(struct mem_ctl_info *mci, int csrow, *input_addr_max = base | mask | pvt->dcs_mask_notused; } -/* - * Extract error address from MCA NB Address Low (section 3.6.4.5) and MCA NB - * Address High (section 3.6.4.6) register values and return the result. Address - * is located in the info structure (nbeah and nbeal), the encoding is device - * specific. - */ -static u64 extract_error_address(struct mem_ctl_info *mci, - struct err_regs *info) -{ - struct amd64_pvt *pvt = mci->pvt_info; - - return pvt->ops->get_error_address(mci, info); -} - - /* Map the Error address to a PAGE and PAGE OFFSET. */ static inline void error_address_to_page_and_offset(u64 error_address, u32 *page, u32 *offset) @@ -787,7 +794,7 @@ static int sys_addr_to_csrow(struct mem_ctl_info *mci, u64 sys_addr) return csrow; } -static int get_channel_from_ecc_syndrome(unsigned short syndrome); +static int get_channel_from_ecc_syndrome(struct mem_ctl_info *, u16); static void amd64_cpu_display_info(struct amd64_pvt *pvt) { @@ -797,7 +804,7 @@ static void amd64_cpu_display_info(struct amd64_pvt *pvt) edac_printk(KERN_DEBUG, EDAC_MC, "F10h CPU detected\n"); else if (boot_cpu_data.x86 == 0xf) edac_printk(KERN_DEBUG, EDAC_MC, "%s detected\n", - (pvt->ext_model >= OPTERON_CPU_REV_F) ? + (pvt->ext_model >= K8_REV_F) ? "Rev F or later" : "Rev E or earlier"); else /* we'll hardly ever ever get here */ @@ -813,7 +820,7 @@ static enum edac_type amd64_determine_edac_cap(struct amd64_pvt *pvt) int bit; enum dev_type edac_cap = EDAC_FLAG_NONE; - bit = (boot_cpu_data.x86 > 0xf || pvt->ext_model >= OPTERON_CPU_REV_F) + bit = (boot_cpu_data.x86 > 0xf || pvt->ext_model >= K8_REV_F) ? 19 : 17; @@ -824,111 +831,86 @@ static enum edac_type amd64_determine_edac_cap(struct amd64_pvt *pvt) } -static void f10_debug_display_dimm_sizes(int ctrl, struct amd64_pvt *pvt, - int ganged); +static void amd64_debug_display_dimm_sizes(int ctrl, struct amd64_pvt *pvt); + +static void amd64_dump_dramcfg_low(u32 dclr, int chan) +{ + debugf1("F2x%d90 (DRAM Cfg Low): 0x%08x\n", chan, dclr); + + debugf1(" DIMM type: %sbuffered; all DIMMs support ECC: %s\n", + (dclr & BIT(16)) ? "un" : "", + (dclr & BIT(19)) ? "yes" : "no"); + + debugf1(" PAR/ERR parity: %s\n", + (dclr & BIT(8)) ? "enabled" : "disabled"); + + debugf1(" DCT 128bit mode width: %s\n", + (dclr & BIT(11)) ? "128b" : "64b"); + + debugf1(" x4 logical DIMMs present: L0: %s L1: %s L2: %s L3: %s\n", + (dclr & BIT(12)) ? "yes" : "no", + (dclr & BIT(13)) ? "yes" : "no", + (dclr & BIT(14)) ? "yes" : "no", + (dclr & BIT(15)) ? "yes" : "no"); +} /* Display and decode various NB registers for debug purposes. */ static void amd64_dump_misc_regs(struct amd64_pvt *pvt) { int ganged; - debugf1(" nbcap:0x%8.08x DctDualCap=%s DualNode=%s 8-Node=%s\n", - pvt->nbcap, - (pvt->nbcap & K8_NBCAP_DCT_DUAL) ? "True" : "False", - (pvt->nbcap & K8_NBCAP_DUAL_NODE) ? "True" : "False", - (pvt->nbcap & K8_NBCAP_8_NODE) ? "True" : "False"); - debugf1(" ECC Capable=%s ChipKill Capable=%s\n", - (pvt->nbcap & K8_NBCAP_SECDED) ? "True" : "False", - (pvt->nbcap & K8_NBCAP_CHIPKILL) ? "True" : "False"); - debugf1(" DramCfg0-low=0x%08x DIMM-ECC=%s Parity=%s Width=%s\n", - pvt->dclr0, - (pvt->dclr0 & BIT(19)) ? "Enabled" : "Disabled", - (pvt->dclr0 & BIT(8)) ? "Enabled" : "Disabled", - (pvt->dclr0 & BIT(11)) ? "128b" : "64b"); - debugf1(" DIMM x4 Present: L0=%s L1=%s L2=%s L3=%s DIMM Type=%s\n", - (pvt->dclr0 & BIT(12)) ? "Y" : "N", - (pvt->dclr0 & BIT(13)) ? "Y" : "N", - (pvt->dclr0 & BIT(14)) ? "Y" : "N", - (pvt->dclr0 & BIT(15)) ? "Y" : "N", - (pvt->dclr0 & BIT(16)) ? "UN-Buffered" : "Buffered"); - - - debugf1(" online-spare: 0x%8.08x\n", pvt->online_spare); + debugf1("F3xE8 (NB Cap): 0x%08x\n", pvt->nbcap); - if (boot_cpu_data.x86 == 0xf) { - debugf1(" dhar: 0x%8.08x Base=0x%08x Offset=0x%08x\n", - pvt->dhar, dhar_base(pvt->dhar), - k8_dhar_offset(pvt->dhar)); - debugf1(" DramHoleValid=%s\n", - (pvt->dhar & DHAR_VALID) ? "True" : "False"); + debugf1(" NB two channel DRAM capable: %s\n", + (pvt->nbcap & K8_NBCAP_DCT_DUAL) ? "yes" : "no"); - debugf1(" dbam-dkt: 0x%8.08x\n", pvt->dbam0); + debugf1(" ECC capable: %s, ChipKill ECC capable: %s\n", + (pvt->nbcap & K8_NBCAP_SECDED) ? "yes" : "no", + (pvt->nbcap & K8_NBCAP_CHIPKILL) ? "yes" : "no"); - /* everything below this point is Fam10h and above */ - return; + amd64_dump_dramcfg_low(pvt->dclr0, 0); - } else { - debugf1(" dhar: 0x%8.08x Base=0x%08x Offset=0x%08x\n", - pvt->dhar, dhar_base(pvt->dhar), - f10_dhar_offset(pvt->dhar)); - debugf1(" DramMemHoistValid=%s DramHoleValid=%s\n", - (pvt->dhar & F10_DRAM_MEM_HOIST_VALID) ? - "True" : "False", - (pvt->dhar & DHAR_VALID) ? - "True" : "False"); - } + debugf1("F3xB0 (Online Spare): 0x%08x\n", pvt->online_spare); + + debugf1("F1xF0 (DRAM Hole Address): 0x%08x, base: 0x%08x, " + "offset: 0x%08x\n", + pvt->dhar, + dhar_base(pvt->dhar), + (boot_cpu_data.x86 == 0xf) ? k8_dhar_offset(pvt->dhar) + : f10_dhar_offset(pvt->dhar)); - /* Only if NOT ganged does dcl1 have valid info */ - if (!dct_ganging_enabled(pvt)) { - debugf1(" DramCfg1-low=0x%08x DIMM-ECC=%s Parity=%s " - "Width=%s\n", pvt->dclr1, - (pvt->dclr1 & BIT(19)) ? "Enabled" : "Disabled", - (pvt->dclr1 & BIT(8)) ? "Enabled" : "Disabled", - (pvt->dclr1 & BIT(11)) ? "128b" : "64b"); - debugf1(" DIMM x4 Present: L0=%s L1=%s L2=%s L3=%s " - "DIMM Type=%s\n", - (pvt->dclr1 & BIT(12)) ? "Y" : "N", - (pvt->dclr1 & BIT(13)) ? "Y" : "N", - (pvt->dclr1 & BIT(14)) ? "Y" : "N", - (pvt->dclr1 & BIT(15)) ? "Y" : "N", - (pvt->dclr1 & BIT(16)) ? "UN-Buffered" : "Buffered"); + debugf1(" DramHoleValid: %s\n", + (pvt->dhar & DHAR_VALID) ? "yes" : "no"); + + /* everything below this point is Fam10h and above */ + if (boot_cpu_data.x86 == 0xf) { + amd64_debug_display_dimm_sizes(0, pvt); + return; } + /* Only if NOT ganged does dclr1 have valid info */ + if (!dct_ganging_enabled(pvt)) + amd64_dump_dramcfg_low(pvt->dclr1, 1); + /* * Determine if ganged and then dump memory sizes for first controller, * and if NOT ganged dump info for 2nd controller. */ ganged = dct_ganging_enabled(pvt); - f10_debug_display_dimm_sizes(0, pvt, ganged); + amd64_debug_display_dimm_sizes(0, pvt); if (!ganged) - f10_debug_display_dimm_sizes(1, pvt, ganged); + amd64_debug_display_dimm_sizes(1, pvt); } /* Read in both of DBAM registers */ static void amd64_read_dbam_reg(struct amd64_pvt *pvt) { - int err = 0; - unsigned int reg; + amd64_read_pci_cfg(pvt->dram_f2_ctl, DBAM0, &pvt->dbam0); - reg = DBAM0; - err = pci_read_config_dword(pvt->dram_f2_ctl, reg, &pvt->dbam0); - if (err) - goto err_reg; - - if (boot_cpu_data.x86 >= 0x10) { - reg = DBAM1; - err = pci_read_config_dword(pvt->dram_f2_ctl, reg, &pvt->dbam1); - - if (err) - goto err_reg; - } - - return; - -err_reg: - debugf0("Error reading F2x%03x.\n", reg); + if (boot_cpu_data.x86 >= 0x10) + amd64_read_pci_cfg(pvt->dram_f2_ctl, DBAM1, &pvt->dbam1); } /* @@ -963,7 +945,7 @@ err_reg: static void amd64_set_dct_base_and_mask(struct amd64_pvt *pvt) { - if (boot_cpu_data.x86 == 0xf && pvt->ext_model < OPTERON_CPU_REV_F) { + if (boot_cpu_data.x86 == 0xf && pvt->ext_model < K8_REV_F) { pvt->dcsb_base = REV_E_DCSB_BASE_BITS; pvt->dcsm_mask = REV_E_DCSM_MASK_BITS; pvt->dcs_mask_notused = REV_E_DCS_NOTUSED_BITS; @@ -991,28 +973,21 @@ static void amd64_set_dct_base_and_mask(struct amd64_pvt *pvt) */ static void amd64_read_dct_base_mask(struct amd64_pvt *pvt) { - int cs, reg, err = 0; + int cs, reg; amd64_set_dct_base_and_mask(pvt); for (cs = 0; cs < pvt->cs_count; cs++) { reg = K8_DCSB0 + (cs * 4); - err = pci_read_config_dword(pvt->dram_f2_ctl, reg, - &pvt->dcsb0[cs]); - if (unlikely(err)) - debugf0("Reading K8_DCSB0[%d] failed\n", cs); - else + if (!amd64_read_pci_cfg(pvt->dram_f2_ctl, reg, &pvt->dcsb0[cs])) debugf0(" DCSB0[%d]=0x%08x reg: F2x%x\n", cs, pvt->dcsb0[cs], reg); /* If DCT are NOT ganged, then read in DCT1's base */ if (boot_cpu_data.x86 >= 0x10 && !dct_ganging_enabled(pvt)) { reg = F10_DCSB1 + (cs * 4); - err = pci_read_config_dword(pvt->dram_f2_ctl, reg, - &pvt->dcsb1[cs]); - if (unlikely(err)) - debugf0("Reading F10_DCSB1[%d] failed\n", cs); - else + if (!amd64_read_pci_cfg(pvt->dram_f2_ctl, reg, + &pvt->dcsb1[cs])) debugf0(" DCSB1[%d]=0x%08x reg: F2x%x\n", cs, pvt->dcsb1[cs], reg); } else { @@ -1022,26 +997,20 @@ static void amd64_read_dct_base_mask(struct amd64_pvt *pvt) for (cs = 0; cs < pvt->num_dcsm; cs++) { reg = K8_DCSM0 + (cs * 4); - err = pci_read_config_dword(pvt->dram_f2_ctl, reg, - &pvt->dcsm0[cs]); - if (unlikely(err)) - debugf0("Reading K8_DCSM0 failed\n"); - else + if (!amd64_read_pci_cfg(pvt->dram_f2_ctl, reg, &pvt->dcsm0[cs])) debugf0(" DCSM0[%d]=0x%08x reg: F2x%x\n", cs, pvt->dcsm0[cs], reg); /* If DCT are NOT ganged, then read in DCT1's mask */ if (boot_cpu_data.x86 >= 0x10 && !dct_ganging_enabled(pvt)) { reg = F10_DCSM1 + (cs * 4); - err = pci_read_config_dword(pvt->dram_f2_ctl, reg, - &pvt->dcsm1[cs]); - if (unlikely(err)) - debugf0("Reading F10_DCSM1[%d] failed\n", cs); - else + if (!amd64_read_pci_cfg(pvt->dram_f2_ctl, reg, + &pvt->dcsm1[cs])) debugf0(" DCSM1[%d]=0x%08x reg: F2x%x\n", cs, pvt->dcsm1[cs], reg); - } else + } else { pvt->dcsm1[cs] = 0; + } } } @@ -1049,18 +1018,16 @@ static enum mem_type amd64_determine_memory_type(struct amd64_pvt *pvt) { enum mem_type type; - if (boot_cpu_data.x86 >= 0x10 || pvt->ext_model >= OPTERON_CPU_REV_F) { - /* Rev F and later */ - type = (pvt->dclr0 & BIT(16)) ? MEM_DDR2 : MEM_RDDR2; + if (boot_cpu_data.x86 >= 0x10 || pvt->ext_model >= K8_REV_F) { + if (pvt->dchr0 & DDR3_MODE) + type = (pvt->dclr0 & BIT(16)) ? MEM_DDR3 : MEM_RDDR3; + else + type = (pvt->dclr0 & BIT(16)) ? MEM_DDR2 : MEM_RDDR2; } else { - /* Rev E and earlier */ type = (pvt->dclr0 & BIT(18)) ? MEM_DDR : MEM_RDDR; } - debugf1(" Memory type is: %s\n", - (type == MEM_DDR2) ? "MEM_DDR2" : - (type == MEM_RDDR2) ? "MEM_RDDR2" : - (type == MEM_DDR) ? "MEM_DDR" : "MEM_RDDR"); + debugf1(" Memory type is: %s\n", edac_mem_types[type]); return type; } @@ -1078,11 +1045,11 @@ static int k8_early_channel_count(struct amd64_pvt *pvt) { int flag, err = 0; - err = pci_read_config_dword(pvt->dram_f2_ctl, F10_DCLR_0, &pvt->dclr0); + err = amd64_read_pci_cfg(pvt->dram_f2_ctl, F10_DCLR_0, &pvt->dclr0); if (err) return err; - if ((boot_cpu_data.x86_model >> 4) >= OPTERON_CPU_REV_F) { + if ((boot_cpu_data.x86_model >> 4) >= K8_REV_F) { /* RevF (NPT) and later */ flag = pvt->dclr0 & F10_WIDTH_128; } else { @@ -1114,22 +1081,15 @@ static void k8_read_dram_base_limit(struct amd64_pvt *pvt, int dram) { u32 low; u32 off = dram << 3; /* 8 bytes between DRAM entries */ - int err; - err = pci_read_config_dword(pvt->addr_f1_ctl, - K8_DRAM_BASE_LOW + off, &low); - if (err) - debugf0("Reading K8_DRAM_BASE_LOW failed\n"); + amd64_read_pci_cfg(pvt->addr_f1_ctl, K8_DRAM_BASE_LOW + off, &low); /* Extract parts into separate data entries */ pvt->dram_base[dram] = ((u64) low & 0xFFFF0000) << 8; pvt->dram_IntlvEn[dram] = (low >> 8) & 0x7; pvt->dram_rw_en[dram] = (low & 0x3); - err = pci_read_config_dword(pvt->addr_f1_ctl, - K8_DRAM_LIMIT_LOW + off, &low); - if (err) - debugf0("Reading K8_DRAM_LIMIT_LOW failed\n"); + amd64_read_pci_cfg(pvt->addr_f1_ctl, K8_DRAM_LIMIT_LOW + off, &low); /* * Extract parts into separate data entries. Limit is the HIGHEST memory @@ -1142,7 +1102,7 @@ static void k8_read_dram_base_limit(struct amd64_pvt *pvt, int dram) static void k8_map_sysaddr_to_csrow(struct mem_ctl_info *mci, struct err_regs *info, - u64 SystemAddress) + u64 sys_addr) { struct mem_ctl_info *src_mci; unsigned short syndrome; @@ -1155,7 +1115,7 @@ static void k8_map_sysaddr_to_csrow(struct mem_ctl_info *mci, /* CHIPKILL enabled */ if (info->nbcfg & K8_NBCFG_CHIPKILL) { - channel = get_channel_from_ecc_syndrome(syndrome); + channel = get_channel_from_ecc_syndrome(mci, syndrome); if (channel < 0) { /* * Syndrome didn't map, so we don't know which of the @@ -1177,64 +1137,46 @@ static void k8_map_sysaddr_to_csrow(struct mem_ctl_info *mci, * was obtained from email communication with someone at AMD. * (Wish the email was placed in this comment - norsk) */ - channel = ((SystemAddress & BIT(3)) != 0); + channel = ((sys_addr & BIT(3)) != 0); } /* * Find out which node the error address belongs to. This may be * different from the node that detected the error. */ - src_mci = find_mc_by_sys_addr(mci, SystemAddress); + src_mci = find_mc_by_sys_addr(mci, sys_addr); if (!src_mci) { amd64_mc_printk(mci, KERN_ERR, "failed to map error address 0x%lx to a node\n", - (unsigned long)SystemAddress); + (unsigned long)sys_addr); edac_mc_handle_ce_no_info(mci, EDAC_MOD_STR); return; } - /* Now map the SystemAddress to a CSROW */ - csrow = sys_addr_to_csrow(src_mci, SystemAddress); + /* Now map the sys_addr to a CSROW */ + csrow = sys_addr_to_csrow(src_mci, sys_addr); if (csrow < 0) { edac_mc_handle_ce_no_info(src_mci, EDAC_MOD_STR); } else { - error_address_to_page_and_offset(SystemAddress, &page, &offset); + error_address_to_page_and_offset(sys_addr, &page, &offset); edac_mc_handle_ce(src_mci, page, offset, syndrome, csrow, channel, EDAC_MOD_STR); } } -/* - * determrine the number of PAGES in for this DIMM's size based on its DRAM - * Address Mapping. - * - * First step is to calc the number of bits to shift a value of 1 left to - * indicate show many pages. Start with the DBAM value as the starting bits, - * then proceed to adjust those shift bits, based on CPU rev and the table. - * See BKDG on the DBAM - */ -static int k8_dbam_map_to_pages(struct amd64_pvt *pvt, int dram_map) +static int k8_dbam_to_chip_select(struct amd64_pvt *pvt, int cs_mode) { - int nr_pages; - - if (pvt->ext_model >= OPTERON_CPU_REV_F) { - nr_pages = 1 << (revf_quad_ddr2_shift[dram_map] - PAGE_SHIFT); - } else { - /* - * RevE and less section; this line is tricky. It collapses the - * table used by RevD and later to one that matches revisions CG - * and earlier. - */ - dram_map -= (pvt->ext_model >= OPTERON_CPU_REV_D) ? - (dram_map > 8 ? 4 : (dram_map > 5 ? - 3 : (dram_map > 2 ? 1 : 0))) : 0; + int *dbam_map; - /* 25 shift is 32MiB minimum DIMM size in RevE and prior */ - nr_pages = 1 << (dram_map + 25 - PAGE_SHIFT); - } + if (pvt->ext_model >= K8_REV_F) + dbam_map = ddr2_dbam; + else if (pvt->ext_model >= K8_REV_D) + dbam_map = ddr2_dbam_revD; + else + dbam_map = ddr2_dbam_revCG; - return nr_pages; + return dbam_map[cs_mode]; } /* @@ -1248,34 +1190,24 @@ static int k8_dbam_map_to_pages(struct amd64_pvt *pvt, int dram_map) static int f10_early_channel_count(struct amd64_pvt *pvt) { int dbams[] = { DBAM0, DBAM1 }; - int err = 0, channels = 0; - int i, j; + int i, j, channels = 0; u32 dbam; - err = pci_read_config_dword(pvt->dram_f2_ctl, F10_DCLR_0, &pvt->dclr0); - if (err) - goto err_reg; - - err = pci_read_config_dword(pvt->dram_f2_ctl, F10_DCLR_1, &pvt->dclr1); - if (err) - goto err_reg; - /* If we are in 128 bit mode, then we are using 2 channels */ if (pvt->dclr0 & F10_WIDTH_128) { - debugf0("Data WIDTH is 128 bits - 2 channels\n"); channels = 2; return channels; } /* - * Need to check if in UN-ganged mode: In such, there are 2 channels, - * but they are NOT in 128 bit mode and thus the above 'dcl0' status bit - * will be OFF. + * Need to check if in unganged mode: In such, there are 2 channels, + * but they are not in 128 bit mode and thus the above 'dclr0' status + * bit will be OFF. * * Need to check DCT0[0] and DCT1[0] to see if only one of them has * their CSEnable bit on. If so, then SINGLE DIMM case. */ - debugf0("Data WIDTH is NOT 128 bits - need more decoding\n"); + debugf0("Data width is not 128 bits - need more decoding\n"); /* * Check DRAM Bank Address Mapping values for each DIMM to see if there @@ -1283,8 +1215,7 @@ static int f10_early_channel_count(struct amd64_pvt *pvt) * both controllers since DIMMs can be placed in either one. */ for (i = 0; i < ARRAY_SIZE(dbams); i++) { - err = pci_read_config_dword(pvt->dram_f2_ctl, dbams[i], &dbam); - if (err) + if (amd64_read_pci_cfg(pvt->dram_f2_ctl, dbams[i], &dbam)) goto err_reg; for (j = 0; j < 4; j++) { @@ -1295,6 +1226,9 @@ static int f10_early_channel_count(struct amd64_pvt *pvt) } } + if (channels > 2) + channels = 2; + debugf0("MCT channel count: %d\n", channels); return channels; @@ -1304,9 +1238,16 @@ err_reg: } -static int f10_dbam_map_to_pages(struct amd64_pvt *pvt, int dram_map) +static int f10_dbam_to_chip_select(struct amd64_pvt *pvt, int cs_mode) { - return 1 << (revf_quad_ddr2_shift[dram_map] - PAGE_SHIFT); + int *dbam_map; + + if (pvt->dchr0 & DDR3_MODE || pvt->dchr1 & DDR3_MODE) + dbam_map = ddr3_dbam; + else + dbam_map = ddr2_dbam; + + return dbam_map[cs_mode]; } /* Enable extended configuration access via 0xCF8 feature */ @@ -1314,7 +1255,7 @@ static void amd64_setup(struct amd64_pvt *pvt) { u32 reg; - pci_read_config_dword(pvt->misc_f3_ctl, F10_NB_CFG_HIGH, ®); + amd64_read_pci_cfg(pvt->misc_f3_ctl, F10_NB_CFG_HIGH, ®); pvt->flags.cf8_extcfg = !!(reg & F10_NB_CFG_LOW_ENABLE_EXT_CFG); reg |= F10_NB_CFG_LOW_ENABLE_EXT_CFG; @@ -1326,7 +1267,7 @@ static void amd64_teardown(struct amd64_pvt *pvt) { u32 reg; - pci_read_config_dword(pvt->misc_f3_ctl, F10_NB_CFG_HIGH, ®); + amd64_read_pci_cfg(pvt->misc_f3_ctl, F10_NB_CFG_HIGH, ®); reg &= ~F10_NB_CFG_LOW_ENABLE_EXT_CFG; if (pvt->flags.cf8_extcfg) @@ -1355,10 +1296,10 @@ static void f10_read_dram_base_limit(struct amd64_pvt *pvt, int dram) high_offset = F10_DRAM_BASE_HIGH + (dram << 3); /* read the 'raw' DRAM BASE Address register */ - pci_read_config_dword(pvt->addr_f1_ctl, low_offset, &low_base); + amd64_read_pci_cfg(pvt->addr_f1_ctl, low_offset, &low_base); /* Read from the ECS data register */ - pci_read_config_dword(pvt->addr_f1_ctl, high_offset, &high_base); + amd64_read_pci_cfg(pvt->addr_f1_ctl, high_offset, &high_base); /* Extract parts into separate data entries */ pvt->dram_rw_en[dram] = (low_base & 0x3); @@ -1375,13 +1316,10 @@ static void f10_read_dram_base_limit(struct amd64_pvt *pvt, int dram) high_offset = F10_DRAM_LIMIT_HIGH + (dram << 3); /* read the 'raw' LIMIT registers */ - pci_read_config_dword(pvt->addr_f1_ctl, low_offset, &low_limit); + amd64_read_pci_cfg(pvt->addr_f1_ctl, low_offset, &low_limit); /* Read from the ECS data register for the HIGH portion */ - pci_read_config_dword(pvt->addr_f1_ctl, high_offset, &high_limit); - - debugf0(" HW Regs: BASE=0x%08x-%08x LIMIT= 0x%08x-%08x\n", - high_base, low_base, high_limit, low_limit); + amd64_read_pci_cfg(pvt->addr_f1_ctl, high_offset, &high_limit); pvt->dram_DstNode[dram] = (low_limit & 0x7); pvt->dram_IntlvSel[dram] = (low_limit >> 8) & 0x7; @@ -1397,32 +1335,35 @@ static void f10_read_dram_base_limit(struct amd64_pvt *pvt, int dram) static void f10_read_dram_ctl_register(struct amd64_pvt *pvt) { - int err = 0; - err = pci_read_config_dword(pvt->dram_f2_ctl, F10_DCTL_SEL_LOW, - &pvt->dram_ctl_select_low); - if (err) { - debugf0("Reading F10_DCTL_SEL_LOW failed\n"); - } else { - debugf0("DRAM_DCTL_SEL_LOW=0x%x DctSelBaseAddr=0x%x\n", - pvt->dram_ctl_select_low, dct_sel_baseaddr(pvt)); - - debugf0(" DRAM DCTs are=%s DRAM Is=%s DRAM-Ctl-" - "sel-hi-range=%s\n", - (dct_ganging_enabled(pvt) ? "GANGED" : "NOT GANGED"), - (dct_dram_enabled(pvt) ? "Enabled" : "Disabled"), - (dct_high_range_enabled(pvt) ? "Enabled" : "Disabled")); - - debugf0(" DctDatIntLv=%s MemCleared=%s DctSelIntLvAddr=0x%x\n", - (dct_data_intlv_enabled(pvt) ? "Enabled" : "Disabled"), - (dct_memory_cleared(pvt) ? "True " : "False "), + if (!amd64_read_pci_cfg(pvt->dram_f2_ctl, F10_DCTL_SEL_LOW, + &pvt->dram_ctl_select_low)) { + debugf0("F2x110 (DCTL Sel. Low): 0x%08x, " + "High range addresses at: 0x%x\n", + pvt->dram_ctl_select_low, + dct_sel_baseaddr(pvt)); + + debugf0(" DCT mode: %s, All DCTs on: %s\n", + (dct_ganging_enabled(pvt) ? "ganged" : "unganged"), + (dct_dram_enabled(pvt) ? "yes" : "no")); + + if (!dct_ganging_enabled(pvt)) + debugf0(" Address range split per DCT: %s\n", + (dct_high_range_enabled(pvt) ? "yes" : "no")); + + debugf0(" DCT data interleave for ECC: %s, " + "DRAM cleared since last warm reset: %s\n", + (dct_data_intlv_enabled(pvt) ? "enabled" : "disabled"), + (dct_memory_cleared(pvt) ? "yes" : "no")); + + debugf0(" DCT channel interleave: %s, " + "DCT interleave bits selector: 0x%x\n", + (dct_interleave_enabled(pvt) ? "enabled" : "disabled"), dct_sel_interleave_addr(pvt)); } - err = pci_read_config_dword(pvt->dram_f2_ctl, F10_DCTL_SEL_HIGH, - &pvt->dram_ctl_select_high); - if (err) - debugf0("Reading F10_DCTL_SEL_HIGH failed\n"); + amd64_read_pci_cfg(pvt->dram_f2_ctl, F10_DCTL_SEL_HIGH, + &pvt->dram_ctl_select_high); } /* @@ -1706,10 +1647,11 @@ static int f10_translate_sysaddr_to_cs(struct amd64_pvt *pvt, u64 sys_addr, } /* - * This the F10h reference code from AMD to map a @sys_addr to NodeID, - * CSROW, Channel. + * For reference see "2.8.5 Routing DRAM Requests" in F10 BKDG. This code maps + * a @sys_addr to NodeID, DCT (channel) and chip select (CSROW). * - * The @sys_addr is usually an error address received from the hardware. + * The @sys_addr is usually an error address received from the hardware + * (MCX_ADDR). */ static void f10_map_sysaddr_to_csrow(struct mem_ctl_info *mci, struct err_regs *info, @@ -1722,133 +1664,80 @@ static void f10_map_sysaddr_to_csrow(struct mem_ctl_info *mci, csrow = f10_translate_sysaddr_to_cs(pvt, sys_addr, &nid, &chan); - if (csrow >= 0) { - error_address_to_page_and_offset(sys_addr, &page, &offset); - - syndrome = HIGH_SYNDROME(info->nbsl) << 8; - syndrome |= LOW_SYNDROME(info->nbsh); - - /* - * Is CHIPKILL on? If so, then we can attempt to use the - * syndrome to isolate which channel the error was on. - */ - if (pvt->nbcfg & K8_NBCFG_CHIPKILL) - chan = get_channel_from_ecc_syndrome(syndrome); - - if (chan >= 0) { - edac_mc_handle_ce(mci, page, offset, syndrome, - csrow, chan, EDAC_MOD_STR); - } else { - /* - * Channel unknown, report all channels on this - * CSROW as failed. - */ - for (chan = 0; chan < mci->csrows[csrow].nr_channels; - chan++) { - edac_mc_handle_ce(mci, page, offset, - syndrome, - csrow, chan, - EDAC_MOD_STR); - } - } - - } else { + if (csrow < 0) { edac_mc_handle_ce_no_info(mci, EDAC_MOD_STR); + return; } -} -/* - * Input (@index) is the DBAM DIMM value (1 of 4) used as an index into a shift - * table (revf_quad_ddr2_shift) which starts at 128MB DIMM size. Index of 0 - * indicates an empty DIMM slot, as reported by Hardware on empty slots. - * - * Normalize to 128MB by subracting 27 bit shift. - */ -static int map_dbam_to_csrow_size(int index) -{ - int mega_bytes = 0; + error_address_to_page_and_offset(sys_addr, &page, &offset); + + syndrome = HIGH_SYNDROME(info->nbsl) << 8; + syndrome |= LOW_SYNDROME(info->nbsh); - if (index > 0 && index <= DBAM_MAX_VALUE) - mega_bytes = ((128 << (revf_quad_ddr2_shift[index]-27))); + /* + * We need the syndromes for channel detection only when we're + * ganged. Otherwise @chan should already contain the channel at + * this point. + */ + if (dct_ganging_enabled(pvt) && pvt->nbcfg & K8_NBCFG_CHIPKILL) + chan = get_channel_from_ecc_syndrome(mci, syndrome); - return mega_bytes; + if (chan >= 0) + edac_mc_handle_ce(mci, page, offset, syndrome, csrow, chan, + EDAC_MOD_STR); + else + /* + * Channel unknown, report all channels on this CSROW as failed. + */ + for (chan = 0; chan < mci->csrows[csrow].nr_channels; chan++) + edac_mc_handle_ce(mci, page, offset, syndrome, + csrow, chan, EDAC_MOD_STR); } /* - * debug routine to display the memory sizes of a DIMM (ganged or not) and it + * debug routine to display the memory sizes of all logical DIMMs and its * CSROWs as well */ -static void f10_debug_display_dimm_sizes(int ctrl, struct amd64_pvt *pvt, - int ganged) +static void amd64_debug_display_dimm_sizes(int ctrl, struct amd64_pvt *pvt) { - int dimm, size0, size1; + int dimm, size0, size1, factor = 0; u32 dbam; u32 *dcsb; - debugf1(" dbam%d: 0x%8.08x CSROW is %s\n", ctrl, - ctrl ? pvt->dbam1 : pvt->dbam0, - ganged ? "GANGED - dbam1 not used" : "NON-GANGED"); + if (boot_cpu_data.x86 == 0xf) { + if (pvt->dclr0 & F10_WIDTH_128) + factor = 1; + + /* K8 families < revF not supported yet */ + if (pvt->ext_model < K8_REV_F) + return; + else + WARN_ON(ctrl != 0); + } + + debugf1("F2x%d80 (DRAM Bank Address Mapping): 0x%08x\n", + ctrl, ctrl ? pvt->dbam1 : pvt->dbam0); dbam = ctrl ? pvt->dbam1 : pvt->dbam0; dcsb = ctrl ? pvt->dcsb1 : pvt->dcsb0; + edac_printk(KERN_DEBUG, EDAC_MC, "DCT%d chip selects:\n", ctrl); + /* Dump memory sizes for DIMM and its CSROWs */ for (dimm = 0; dimm < 4; dimm++) { size0 = 0; if (dcsb[dimm*2] & K8_DCSB_CS_ENABLE) - size0 = map_dbam_to_csrow_size(DBAM_DIMM(dimm, dbam)); + size0 = pvt->ops->dbam_to_cs(pvt, DBAM_DIMM(dimm, dbam)); size1 = 0; if (dcsb[dimm*2 + 1] & K8_DCSB_CS_ENABLE) - size1 = map_dbam_to_csrow_size(DBAM_DIMM(dimm, dbam)); - - debugf1(" CTRL-%d DIMM-%d=%5dMB CSROW-%d=%5dMB " - "CSROW-%d=%5dMB\n", - ctrl, - dimm, - size0 + size1, - dimm * 2, - size0, - dimm * 2 + 1, - size1); - } -} - -/* - * Very early hardware probe on pci_probe thread to determine if this module - * supports the hardware. - * - * Return: - * 0 for OK - * 1 for error - */ -static int f10_probe_valid_hardware(struct amd64_pvt *pvt) -{ - int ret = 0; - - /* - * If we are on a DDR3 machine, we don't know yet if - * we support that properly at this time - */ - if ((pvt->dchr0 & F10_DCHR_Ddr3Mode) || - (pvt->dchr1 & F10_DCHR_Ddr3Mode)) { - - amd64_printk(KERN_WARNING, - "%s() This machine is running with DDR3 memory. " - "This is not currently supported. " - "DCHR0=0x%x DCHR1=0x%x\n", - __func__, pvt->dchr0, pvt->dchr1); - - amd64_printk(KERN_WARNING, - " Contact '%s' module MAINTAINER to help add" - " support.\n", - EDAC_MOD_STR); - - ret = 1; + size1 = pvt->ops->dbam_to_cs(pvt, DBAM_DIMM(dimm, dbam)); + edac_printk(KERN_DEBUG, EDAC_MC, " %d: %5dMB %d: %5dMB\n", + dimm * 2, size0 << factor, + dimm * 2 + 1, size1 << factor); } - return ret; } /* @@ -1868,11 +1757,11 @@ static struct amd64_family_type amd64_family_types[] = { .addr_f1_ctl = PCI_DEVICE_ID_AMD_K8_NB_ADDRMAP, .misc_f3_ctl = PCI_DEVICE_ID_AMD_K8_NB_MISC, .ops = { - .early_channel_count = k8_early_channel_count, - .get_error_address = k8_get_error_address, - .read_dram_base_limit = k8_read_dram_base_limit, - .map_sysaddr_to_csrow = k8_map_sysaddr_to_csrow, - .dbam_map_to_pages = k8_dbam_map_to_pages, + .early_channel_count = k8_early_channel_count, + .get_error_address = k8_get_error_address, + .read_dram_base_limit = k8_read_dram_base_limit, + .map_sysaddr_to_csrow = k8_map_sysaddr_to_csrow, + .dbam_to_cs = k8_dbam_to_chip_select, } }, [F10_CPUS] = { @@ -1880,13 +1769,12 @@ static struct amd64_family_type amd64_family_types[] = { .addr_f1_ctl = PCI_DEVICE_ID_AMD_10H_NB_MAP, .misc_f3_ctl = PCI_DEVICE_ID_AMD_10H_NB_MISC, .ops = { - .probe_valid_hardware = f10_probe_valid_hardware, - .early_channel_count = f10_early_channel_count, - .get_error_address = f10_get_error_address, - .read_dram_base_limit = f10_read_dram_base_limit, - .read_dram_ctl_register = f10_read_dram_ctl_register, - .map_sysaddr_to_csrow = f10_map_sysaddr_to_csrow, - .dbam_map_to_pages = f10_dbam_map_to_pages, + .early_channel_count = f10_early_channel_count, + .get_error_address = f10_get_error_address, + .read_dram_base_limit = f10_read_dram_base_limit, + .read_dram_ctl_register = f10_read_dram_ctl_register, + .map_sysaddr_to_csrow = f10_map_sysaddr_to_csrow, + .dbam_to_cs = f10_dbam_to_chip_select, } }, [F11_CPUS] = { @@ -1894,13 +1782,12 @@ static struct amd64_family_type amd64_family_types[] = { .addr_f1_ctl = PCI_DEVICE_ID_AMD_11H_NB_MAP, .misc_f3_ctl = PCI_DEVICE_ID_AMD_11H_NB_MISC, .ops = { - .probe_valid_hardware = f10_probe_valid_hardware, - .early_channel_count = f10_early_channel_count, - .get_error_address = f10_get_error_address, - .read_dram_base_limit = f10_read_dram_base_limit, - .read_dram_ctl_register = f10_read_dram_ctl_register, - .map_sysaddr_to_csrow = f10_map_sysaddr_to_csrow, - .dbam_map_to_pages = f10_dbam_map_to_pages, + .early_channel_count = f10_early_channel_count, + .get_error_address = f10_get_error_address, + .read_dram_base_limit = f10_read_dram_base_limit, + .read_dram_ctl_register = f10_read_dram_ctl_register, + .map_sysaddr_to_csrow = f10_map_sysaddr_to_csrow, + .dbam_to_cs = f10_dbam_to_chip_select, } }, }; @@ -1923,142 +1810,170 @@ static struct pci_dev *pci_get_related_function(unsigned int vendor, } /* - * syndrome mapping table for ECC ChipKill devices - * - * The comment in each row is the token (nibble) number that is in error. - * The least significant nibble of the syndrome is the mask for the bits - * that are in error (need to be toggled) for the particular nibble. - * - * Each row contains 16 entries. - * The first entry (0th) is the channel number for that row of syndromes. - * The remaining 15 entries are the syndromes for the respective Error - * bit mask index. - * - * 1st index entry is 0x0001 mask, indicating that the rightmost bit is the - * bit in error. - * The 2nd index entry is 0x0010 that the second bit is damaged. - * The 3rd index entry is 0x0011 indicating that the rightmost 2 bits - * are damaged. - * Thus so on until index 15, 0x1111, whose entry has the syndrome - * indicating that all 4 bits are damaged. - * - * A search is performed on this table looking for a given syndrome. - * - * See the AMD documentation for ECC syndromes. This ECC table is valid - * across all the versions of the AMD64 processors. + * These are tables of eigenvectors (one per line) which can be used for the + * construction of the syndrome tables. The modified syndrome search algorithm + * uses those to find the symbol in error and thus the DIMM. * - * A fast lookup is to use the LAST four bits of the 16-bit syndrome as a - * COLUMN index, then search all ROWS of that column, looking for a match - * with the input syndrome. The ROW value will be the token number. - * - * The 0'th entry on that row, can be returned as the CHANNEL (0 or 1) of this - * error. + * Algorithm courtesy of Ross LaFetra from AMD. */ -#define NUMBER_ECC_ROWS 36 -static const unsigned short ecc_chipkill_syndromes[NUMBER_ECC_ROWS][16] = { - /* Channel 0 syndromes */ - {/*0*/ 0, 0xe821, 0x7c32, 0x9413, 0xbb44, 0x5365, 0xc776, 0x2f57, - 0xdd88, 0x35a9, 0xa1ba, 0x499b, 0x66cc, 0x8eed, 0x1afe, 0xf2df }, - {/*1*/ 0, 0x5d31, 0xa612, 0xfb23, 0x9584, 0xc8b5, 0x3396, 0x6ea7, - 0xeac8, 0xb7f9, 0x4cda, 0x11eb, 0x7f4c, 0x227d, 0xd95e, 0x846f }, - {/*2*/ 0, 0x0001, 0x0002, 0x0003, 0x0004, 0x0005, 0x0006, 0x0007, - 0x0008, 0x0009, 0x000a, 0x000b, 0x000c, 0x000d, 0x000e, 0x000f }, - {/*3*/ 0, 0x2021, 0x3032, 0x1013, 0x4044, 0x6065, 0x7076, 0x5057, - 0x8088, 0xa0a9, 0xb0ba, 0x909b, 0xc0cc, 0xe0ed, 0xf0fe, 0xd0df }, - {/*4*/ 0, 0x5041, 0xa082, 0xf0c3, 0x9054, 0xc015, 0x30d6, 0x6097, - 0xe0a8, 0xb0e9, 0x402a, 0x106b, 0x70fc, 0x20bd, 0xd07e, 0x803f }, - {/*5*/ 0, 0xbe21, 0xd732, 0x6913, 0x2144, 0x9f65, 0xf676, 0x4857, - 0x3288, 0x8ca9, 0xe5ba, 0x5b9b, 0x13cc, 0xaded, 0xc4fe, 0x7adf }, - {/*6*/ 0, 0x4951, 0x8ea2, 0xc7f3, 0x5394, 0x1ac5, 0xdd36, 0x9467, - 0xa1e8, 0xe8b9, 0x2f4a, 0x661b, 0xf27c, 0xbb2d, 0x7cde, 0x358f }, - {/*7*/ 0, 0x74e1, 0x9872, 0xec93, 0xd6b4, 0xa255, 0x4ec6, 0x3a27, - 0x6bd8, 0x1f39, 0xf3aa, 0x874b, 0xbd6c, 0xc98d, 0x251e, 0x51ff }, - {/*8*/ 0, 0x15c1, 0x2a42, 0x3f83, 0xcef4, 0xdb35, 0xe4b6, 0xf177, - 0x4758, 0x5299, 0x6d1a, 0x78db, 0x89ac, 0x9c6d, 0xa3ee, 0xb62f }, - {/*9*/ 0, 0x3d01, 0x1602, 0x2b03, 0x8504, 0xb805, 0x9306, 0xae07, - 0xca08, 0xf709, 0xdc0a, 0xe10b, 0x4f0c, 0x720d, 0x590e, 0x640f }, - {/*a*/ 0, 0x9801, 0xec02, 0x7403, 0x6b04, 0xf305, 0x8706, 0x1f07, - 0xbd08, 0x2509, 0x510a, 0xc90b, 0xd60c, 0x4e0d, 0x3a0e, 0xa20f }, - {/*b*/ 0, 0xd131, 0x6212, 0xb323, 0x3884, 0xe9b5, 0x5a96, 0x8ba7, - 0x1cc8, 0xcdf9, 0x7eda, 0xafeb, 0x244c, 0xf57d, 0x465e, 0x976f }, - {/*c*/ 0, 0xe1d1, 0x7262, 0x93b3, 0xb834, 0x59e5, 0xca56, 0x2b87, - 0xdc18, 0x3dc9, 0xae7a, 0x4fab, 0x542c, 0x85fd, 0x164e, 0xf79f }, - {/*d*/ 0, 0x6051, 0xb0a2, 0xd0f3, 0x1094, 0x70c5, 0xa036, 0xc067, - 0x20e8, 0x40b9, 0x904a, 0x601b, 0x307c, 0x502d, 0x80de, 0xe08f }, - {/*e*/ 0, 0xa4c1, 0xf842, 0x5c83, 0xe6f4, 0x4235, 0x1eb6, 0xba77, - 0x7b58, 0xdf99, 0x831a, 0x27db, 0x9dac, 0x396d, 0x65ee, 0xc12f }, - {/*f*/ 0, 0x11c1, 0x2242, 0x3383, 0xc8f4, 0xd935, 0xeab6, 0xfb77, - 0x4c58, 0x5d99, 0x6e1a, 0x7fdb, 0x84ac, 0x956d, 0xa6ee, 0xb72f }, - - /* Channel 1 syndromes */ - {/*10*/ 1, 0x45d1, 0x8a62, 0xcfb3, 0x5e34, 0x1be5, 0xd456, 0x9187, - 0xa718, 0xe2c9, 0x2d7a, 0x68ab, 0xf92c, 0xbcfd, 0x734e, 0x369f }, - {/*11*/ 1, 0x63e1, 0xb172, 0xd293, 0x14b4, 0x7755, 0xa5c6, 0xc627, - 0x28d8, 0x4b39, 0x99aa, 0xfa4b, 0x3c6c, 0x5f8d, 0x8d1e, 0xeeff }, - {/*12*/ 1, 0xb741, 0xd982, 0x6ec3, 0x2254, 0x9515, 0xfbd6, 0x4c97, - 0x33a8, 0x84e9, 0xea2a, 0x5d6b, 0x11fc, 0xa6bd, 0xc87e, 0x7f3f }, - {/*13*/ 1, 0xdd41, 0x6682, 0xbbc3, 0x3554, 0xe815, 0x53d6, 0xce97, - 0x1aa8, 0xc7e9, 0x7c2a, 0xa1fb, 0x2ffc, 0xf2bd, 0x497e, 0x943f }, - {/*14*/ 1, 0x2bd1, 0x3d62, 0x16b3, 0x4f34, 0x64e5, 0x7256, 0x5987, - 0x8518, 0xaec9, 0xb87a, 0x93ab, 0xca2c, 0xe1fd, 0xf74e, 0xdc9f }, - {/*15*/ 1, 0x83c1, 0xc142, 0x4283, 0xa4f4, 0x2735, 0x65b6, 0xe677, - 0xf858, 0x7b99, 0x391a, 0xbadb, 0x5cac, 0xdf6d, 0x9dee, 0x1e2f }, - {/*16*/ 1, 0x8fd1, 0xc562, 0x4ab3, 0xa934, 0x26e5, 0x6c56, 0xe387, - 0xfe18, 0x71c9, 0x3b7a, 0xb4ab, 0x572c, 0xd8fd, 0x924e, 0x1d9f }, - {/*17*/ 1, 0x4791, 0x89e2, 0xce73, 0x5264, 0x15f5, 0xdb86, 0x9c17, - 0xa3b8, 0xe429, 0x2a5a, 0x6dcb, 0xf1dc, 0xb64d, 0x783e, 0x3faf }, - {/*18*/ 1, 0x5781, 0xa9c2, 0xfe43, 0x92a4, 0xc525, 0x3b66, 0x6ce7, - 0xe3f8, 0xb479, 0x4a3a, 0x1dbb, 0x715c, 0x26dd, 0xd89e, 0x8f1f }, - {/*19*/ 1, 0xbf41, 0xd582, 0x6ac3, 0x2954, 0x9615, 0xfcd6, 0x4397, - 0x3ea8, 0x81e9, 0xeb2a, 0x546b, 0x17fc, 0xa8bd, 0xc27e, 0x7d3f }, - {/*1a*/ 1, 0x9891, 0xe1e2, 0x7273, 0x6464, 0xf7f5, 0x8586, 0x1617, - 0xb8b8, 0x2b29, 0x595a, 0xcacb, 0xdcdc, 0x4f4d, 0x3d3e, 0xaeaf }, - {/*1b*/ 1, 0xcce1, 0x4472, 0x8893, 0xfdb4, 0x3f55, 0xb9c6, 0x7527, - 0x56d8, 0x9a39, 0x12aa, 0xde4b, 0xab6c, 0x678d, 0xef1e, 0x23ff }, - {/*1c*/ 1, 0xa761, 0xf9b2, 0x5ed3, 0xe214, 0x4575, 0x1ba6, 0xbcc7, - 0x7328, 0xd449, 0x8a9a, 0x2dfb, 0x913c, 0x365d, 0x688e, 0xcfef }, - {/*1d*/ 1, 0xff61, 0x55b2, 0xaad3, 0x7914, 0x8675, 0x2ca6, 0xd3c7, - 0x9e28, 0x6149, 0xcb9a, 0x34fb, 0xe73c, 0x185d, 0xb28e, 0x4def }, - {/*1e*/ 1, 0x5451, 0xa8a2, 0xfcf3, 0x9694, 0xc2c5, 0x3e36, 0x6a67, - 0xebe8, 0xbfb9, 0x434a, 0x171b, 0x7d7c, 0x292d, 0xd5de, 0x818f }, - {/*1f*/ 1, 0x6fc1, 0xb542, 0xda83, 0x19f4, 0x7635, 0xacb6, 0xc377, - 0x2e58, 0x4199, 0x9b1a, 0xf4db, 0x37ac, 0x586d, 0x82ee, 0xed2f }, - - /* ECC bits are also in the set of tokens and they too can go bad - * first 2 cover channel 0, while the second 2 cover channel 1 - */ - {/*20*/ 0, 0xbe01, 0xd702, 0x6903, 0x2104, 0x9f05, 0xf606, 0x4807, - 0x3208, 0x8c09, 0xe50a, 0x5b0b, 0x130c, 0xad0d, 0xc40e, 0x7a0f }, - {/*21*/ 0, 0x4101, 0x8202, 0xc303, 0x5804, 0x1905, 0xda06, 0x9b07, - 0xac08, 0xed09, 0x2e0a, 0x6f0b, 0x640c, 0xb50d, 0x760e, 0x370f }, - {/*22*/ 1, 0xc441, 0x4882, 0x8cc3, 0xf654, 0x3215, 0xbed6, 0x7a97, - 0x5ba8, 0x9fe9, 0x132a, 0xd76b, 0xadfc, 0x69bd, 0xe57e, 0x213f }, - {/*23*/ 1, 0x7621, 0x9b32, 0xed13, 0xda44, 0xac65, 0x4176, 0x3757, - 0x6f88, 0x19a9, 0xf4ba, 0x829b, 0xb5cc, 0xc3ed, 0x2efe, 0x58df } +static u16 x4_vectors[] = { + 0x2f57, 0x1afe, 0x66cc, 0xdd88, + 0x11eb, 0x3396, 0x7f4c, 0xeac8, + 0x0001, 0x0002, 0x0004, 0x0008, + 0x1013, 0x3032, 0x4044, 0x8088, + 0x106b, 0x30d6, 0x70fc, 0xe0a8, + 0x4857, 0xc4fe, 0x13cc, 0x3288, + 0x1ac5, 0x2f4a, 0x5394, 0xa1e8, + 0x1f39, 0x251e, 0xbd6c, 0x6bd8, + 0x15c1, 0x2a42, 0x89ac, 0x4758, + 0x2b03, 0x1602, 0x4f0c, 0xca08, + 0x1f07, 0x3a0e, 0x6b04, 0xbd08, + 0x8ba7, 0x465e, 0x244c, 0x1cc8, + 0x2b87, 0x164e, 0x642c, 0xdc18, + 0x40b9, 0x80de, 0x1094, 0x20e8, + 0x27db, 0x1eb6, 0x9dac, 0x7b58, + 0x11c1, 0x2242, 0x84ac, 0x4c58, + 0x1be5, 0x2d7a, 0x5e34, 0xa718, + 0x4b39, 0x8d1e, 0x14b4, 0x28d8, + 0x4c97, 0xc87e, 0x11fc, 0x33a8, + 0x8e97, 0x497e, 0x2ffc, 0x1aa8, + 0x16b3, 0x3d62, 0x4f34, 0x8518, + 0x1e2f, 0x391a, 0x5cac, 0xf858, + 0x1d9f, 0x3b7a, 0x572c, 0xfe18, + 0x15f5, 0x2a5a, 0x5264, 0xa3b8, + 0x1dbb, 0x3b66, 0x715c, 0xe3f8, + 0x4397, 0xc27e, 0x17fc, 0x3ea8, + 0x1617, 0x3d3e, 0x6464, 0xb8b8, + 0x23ff, 0x12aa, 0xab6c, 0x56d8, + 0x2dfb, 0x1ba6, 0x913c, 0x7328, + 0x185d, 0x2ca6, 0x7914, 0x9e28, + 0x171b, 0x3e36, 0x7d7c, 0xebe8, + 0x4199, 0x82ee, 0x19f4, 0x2e58, + 0x4807, 0xc40e, 0x130c, 0x3208, + 0x1905, 0x2e0a, 0x5804, 0xac08, + 0x213f, 0x132a, 0xadfc, 0x5ba8, + 0x19a9, 0x2efe, 0xb5cc, 0x6f88, }; -/* - * Given the syndrome argument, scan each of the channel tables for a syndrome - * match. Depending on which table it is found, return the channel number. - */ -static int get_channel_from_ecc_syndrome(unsigned short syndrome) +static u16 x8_vectors[] = { + 0x0145, 0x028a, 0x2374, 0x43c8, 0xa1f0, 0x0520, 0x0a40, 0x1480, + 0x0211, 0x0422, 0x0844, 0x1088, 0x01b0, 0x44e0, 0x23c0, 0xed80, + 0x1011, 0x0116, 0x022c, 0x0458, 0x08b0, 0x8c60, 0x2740, 0x4e80, + 0x0411, 0x0822, 0x1044, 0x0158, 0x02b0, 0x2360, 0x46c0, 0xab80, + 0x0811, 0x1022, 0x012c, 0x0258, 0x04b0, 0x4660, 0x8cc0, 0x2780, + 0x2071, 0x40e2, 0xa0c4, 0x0108, 0x0210, 0x0420, 0x0840, 0x1080, + 0x4071, 0x80e2, 0x0104, 0x0208, 0x0410, 0x0820, 0x1040, 0x2080, + 0x8071, 0x0102, 0x0204, 0x0408, 0x0810, 0x1020, 0x2040, 0x4080, + 0x019d, 0x03d6, 0x136c, 0x2198, 0x50b0, 0xb2e0, 0x0740, 0x0e80, + 0x0189, 0x03ea, 0x072c, 0x0e58, 0x1cb0, 0x56e0, 0x37c0, 0xf580, + 0x01fd, 0x0376, 0x06ec, 0x0bb8, 0x1110, 0x2220, 0x4440, 0x8880, + 0x0163, 0x02c6, 0x1104, 0x0758, 0x0eb0, 0x2be0, 0x6140, 0xc280, + 0x02fd, 0x01c6, 0x0b5c, 0x1108, 0x07b0, 0x25a0, 0x8840, 0x6180, + 0x0801, 0x012e, 0x025c, 0x04b8, 0x1370, 0x26e0, 0x57c0, 0xb580, + 0x0401, 0x0802, 0x015c, 0x02b8, 0x22b0, 0x13e0, 0x7140, 0xe280, + 0x0201, 0x0402, 0x0804, 0x01b8, 0x11b0, 0x31a0, 0x8040, 0x7180, + 0x0101, 0x0202, 0x0404, 0x0808, 0x1010, 0x2020, 0x4040, 0x8080, + 0x0001, 0x0002, 0x0004, 0x0008, 0x0010, 0x0020, 0x0040, 0x0080, + 0x0100, 0x0200, 0x0400, 0x0800, 0x1000, 0x2000, 0x4000, 0x8000, +}; + +static int decode_syndrome(u16 syndrome, u16 *vectors, int num_vecs, + int v_dim) { - int row; - int column; + unsigned int i, err_sym; + + for (err_sym = 0; err_sym < num_vecs / v_dim; err_sym++) { + u16 s = syndrome; + int v_idx = err_sym * v_dim; + int v_end = (err_sym + 1) * v_dim; + + /* walk over all 16 bits of the syndrome */ + for (i = 1; i < (1U << 16); i <<= 1) { + + /* if bit is set in that eigenvector... */ + if (v_idx < v_end && vectors[v_idx] & i) { + u16 ev_comp = vectors[v_idx++]; - /* Determine column to scan */ - column = syndrome & 0xF; + /* ... and bit set in the modified syndrome, */ + if (s & i) { + /* remove it. */ + s ^= ev_comp; - /* Scan all rows, looking for syndrome, or end of table */ - for (row = 0; row < NUMBER_ECC_ROWS; row++) { - if (ecc_chipkill_syndromes[row][column] == syndrome) - return ecc_chipkill_syndromes[row][0]; + if (!s) + return err_sym; + } + + } else if (s & i) + /* can't get to zero, move to next symbol */ + break; + } } debugf0("syndrome(%x) not found\n", syndrome); return -1; } +static int map_err_sym_to_channel(int err_sym, int sym_size) +{ + if (sym_size == 4) + switch (err_sym) { + case 0x20: + case 0x21: + return 0; + break; + case 0x22: + case 0x23: + return 1; + break; + default: + return err_sym >> 4; + break; + } + /* x8 symbols */ + else + switch (err_sym) { + /* imaginary bits not in a DIMM */ + case 0x10: + WARN(1, KERN_ERR "Invalid error symbol: 0x%x\n", + err_sym); + return -1; + break; + + case 0x11: + return 0; + break; + case 0x12: + return 1; + break; + default: + return err_sym >> 3; + break; + } + return -1; +} + +static int get_channel_from_ecc_syndrome(struct mem_ctl_info *mci, u16 syndrome) +{ + struct amd64_pvt *pvt = mci->pvt_info; + u32 value = 0; + int err_sym = 0; + + amd64_read_pci_cfg(pvt->misc_f3_ctl, 0x180, &value); + + /* F3x180[EccSymbolSize]=1, x8 symbols */ + if (boot_cpu_data.x86 == 0x10 && + boot_cpu_data.x86_model > 7 && + value & BIT(25)) { + err_sym = decode_syndrome(syndrome, x8_vectors, + ARRAY_SIZE(x8_vectors), 8); + return map_err_sym_to_channel(err_sym, 8); + } else { + err_sym = decode_syndrome(syndrome, x4_vectors, + ARRAY_SIZE(x4_vectors), 4); + return map_err_sym_to_channel(err_sym, 4); + } +} + /* * Check for valid error in the NB Status High register. If so, proceed to read * NB Status Low, NB Address Low and NB Address High registers and store data @@ -2073,40 +1988,24 @@ static int amd64_get_error_info_regs(struct mem_ctl_info *mci, { struct amd64_pvt *pvt; struct pci_dev *misc_f3_ctl; - int err = 0; pvt = mci->pvt_info; misc_f3_ctl = pvt->misc_f3_ctl; - err = pci_read_config_dword(misc_f3_ctl, K8_NBSH, ®s->nbsh); - if (err) - goto err_reg; + if (amd64_read_pci_cfg(misc_f3_ctl, K8_NBSH, ®s->nbsh)) + return 0; if (!(regs->nbsh & K8_NBSH_VALID_BIT)) return 0; /* valid error, read remaining error information registers */ - err = pci_read_config_dword(misc_f3_ctl, K8_NBSL, ®s->nbsl); - if (err) - goto err_reg; - - err = pci_read_config_dword(misc_f3_ctl, K8_NBEAL, ®s->nbeal); - if (err) - goto err_reg; - - err = pci_read_config_dword(misc_f3_ctl, K8_NBEAH, ®s->nbeah); - if (err) - goto err_reg; - - err = pci_read_config_dword(misc_f3_ctl, K8_NBCFG, ®s->nbcfg); - if (err) - goto err_reg; + if (amd64_read_pci_cfg(misc_f3_ctl, K8_NBSL, ®s->nbsl) || + amd64_read_pci_cfg(misc_f3_ctl, K8_NBEAL, ®s->nbeal) || + amd64_read_pci_cfg(misc_f3_ctl, K8_NBEAH, ®s->nbeah) || + amd64_read_pci_cfg(misc_f3_ctl, K8_NBCFG, ®s->nbcfg)) + return 0; return 1; - -err_reg: - debugf0("Reading error info register failed\n"); - return 0; } /* @@ -2184,7 +2083,7 @@ static void amd64_handle_ce(struct mem_ctl_info *mci, struct err_regs *info) { struct amd64_pvt *pvt = mci->pvt_info; - u64 SystemAddress; + u64 sys_addr; /* Ensure that the Error Address is VALID */ if ((info->nbsh & K8_NBSH_VALID_ERROR_ADDR) == 0) { @@ -2194,22 +2093,23 @@ static void amd64_handle_ce(struct mem_ctl_info *mci, return; } - SystemAddress = extract_error_address(mci, info); + sys_addr = pvt->ops->get_error_address(mci, info); amd64_mc_printk(mci, KERN_ERR, - "CE ERROR_ADDRESS= 0x%llx\n", SystemAddress); + "CE ERROR_ADDRESS= 0x%llx\n", sys_addr); - pvt->ops->map_sysaddr_to_csrow(mci, info, SystemAddress); + pvt->ops->map_sysaddr_to_csrow(mci, info, sys_addr); } /* Handle any Un-correctable Errors (UEs) */ static void amd64_handle_ue(struct mem_ctl_info *mci, struct err_regs *info) { + struct amd64_pvt *pvt = mci->pvt_info; + struct mem_ctl_info *log_mci, *src_mci = NULL; int csrow; - u64 SystemAddress; + u64 sys_addr; u32 page, offset; - struct mem_ctl_info *log_mci, *src_mci = NULL; log_mci = mci; @@ -2220,31 +2120,31 @@ static void amd64_handle_ue(struct mem_ctl_info *mci, return; } - SystemAddress = extract_error_address(mci, info); + sys_addr = pvt->ops->get_error_address(mci, info); /* * Find out which node the error address belongs to. This may be * different from the node that detected the error. */ - src_mci = find_mc_by_sys_addr(mci, SystemAddress); + src_mci = find_mc_by_sys_addr(mci, sys_addr); if (!src_mci) { amd64_mc_printk(mci, KERN_CRIT, "ERROR ADDRESS (0x%lx) value NOT mapped to a MC\n", - (unsigned long)SystemAddress); + (unsigned long)sys_addr); edac_mc_handle_ue_no_info(log_mci, EDAC_MOD_STR); return; } log_mci = src_mci; - csrow = sys_addr_to_csrow(log_mci, SystemAddress); + csrow = sys_addr_to_csrow(log_mci, sys_addr); if (csrow < 0) { amd64_mc_printk(mci, KERN_CRIT, "ERROR_ADDRESS (0x%lx) value NOT mapped to 'csrow'\n", - (unsigned long)SystemAddress); + (unsigned long)sys_addr); edac_mc_handle_ue_no_info(log_mci, EDAC_MOD_STR); } else { - error_address_to_page_and_offset(SystemAddress, &page, &offset); + error_address_to_page_and_offset(sys_addr, &page, &offset); edac_mc_handle_ue(log_mci, page, offset, csrow, EDAC_MOD_STR); } } @@ -2384,30 +2284,26 @@ static void amd64_free_mc_sibling_devices(struct amd64_pvt *pvt) static void amd64_read_mc_registers(struct amd64_pvt *pvt) { u64 msr_val; - int dram, err = 0; + int dram; /* * Retrieve TOP_MEM and TOP_MEM2; no masking off of reserved bits since * those are Read-As-Zero */ - rdmsrl(MSR_K8_TOP_MEM1, msr_val); - pvt->top_mem = msr_val >> 23; - debugf0(" TOP_MEM=0x%08llx\n", pvt->top_mem); + rdmsrl(MSR_K8_TOP_MEM1, pvt->top_mem); + debugf0(" TOP_MEM: 0x%016llx\n", pvt->top_mem); /* check first whether TOP_MEM2 is enabled */ rdmsrl(MSR_K8_SYSCFG, msr_val); if (msr_val & (1U << 21)) { - rdmsrl(MSR_K8_TOP_MEM2, msr_val); - pvt->top_mem2 = msr_val >> 23; - debugf0(" TOP_MEM2=0x%08llx\n", pvt->top_mem2); + rdmsrl(MSR_K8_TOP_MEM2, pvt->top_mem2); + debugf0(" TOP_MEM2: 0x%016llx\n", pvt->top_mem2); } else debugf0(" TOP_MEM2 disabled.\n"); amd64_cpu_display_info(pvt); - err = pci_read_config_dword(pvt->misc_f3_ctl, K8_NBCAP, &pvt->nbcap); - if (err) - goto err_reg; + amd64_read_pci_cfg(pvt->misc_f3_ctl, K8_NBCAP, &pvt->nbcap); if (pvt->ops->read_dram_ctl_register) pvt->ops->read_dram_ctl_register(pvt); @@ -2425,13 +2321,12 @@ static void amd64_read_mc_registers(struct amd64_pvt *pvt) * debug output block away. */ if (pvt->dram_rw_en[dram] != 0) { - debugf1(" DRAM_BASE[%d]: 0x%8.08x-%8.08x " - "DRAM_LIMIT: 0x%8.08x-%8.08x\n", + debugf1(" DRAM-BASE[%d]: 0x%016llx " + "DRAM-LIMIT: 0x%016llx\n", dram, - (u32)(pvt->dram_base[dram] >> 32), - (u32)(pvt->dram_base[dram] & 0xFFFFFFFF), - (u32)(pvt->dram_limit[dram] >> 32), - (u32)(pvt->dram_limit[dram] & 0xFFFFFFFF)); + pvt->dram_base[dram], + pvt->dram_limit[dram]); + debugf1(" IntlvEn=%s %s %s " "IntlvSel=%d DstNode=%d\n", pvt->dram_IntlvEn[dram] ? @@ -2445,44 +2340,20 @@ static void amd64_read_mc_registers(struct amd64_pvt *pvt) amd64_read_dct_base_mask(pvt); - err = pci_read_config_dword(pvt->addr_f1_ctl, K8_DHAR, &pvt->dhar); - if (err) - goto err_reg; - + amd64_read_pci_cfg(pvt->addr_f1_ctl, K8_DHAR, &pvt->dhar); amd64_read_dbam_reg(pvt); - err = pci_read_config_dword(pvt->misc_f3_ctl, - F10_ONLINE_SPARE, &pvt->online_spare); - if (err) - goto err_reg; + amd64_read_pci_cfg(pvt->misc_f3_ctl, + F10_ONLINE_SPARE, &pvt->online_spare); - err = pci_read_config_dword(pvt->dram_f2_ctl, F10_DCLR_0, &pvt->dclr0); - if (err) - goto err_reg; - - err = pci_read_config_dword(pvt->dram_f2_ctl, F10_DCHR_0, &pvt->dchr0); - if (err) - goto err_reg; + amd64_read_pci_cfg(pvt->dram_f2_ctl, F10_DCLR_0, &pvt->dclr0); + amd64_read_pci_cfg(pvt->dram_f2_ctl, F10_DCHR_0, &pvt->dchr0); - if (!dct_ganging_enabled(pvt)) { - err = pci_read_config_dword(pvt->dram_f2_ctl, F10_DCLR_1, - &pvt->dclr1); - if (err) - goto err_reg; - - err = pci_read_config_dword(pvt->dram_f2_ctl, F10_DCHR_1, - &pvt->dchr1); - if (err) - goto err_reg; + if (!dct_ganging_enabled(pvt) && boot_cpu_data.x86 >= 0x10) { + amd64_read_pci_cfg(pvt->dram_f2_ctl, F10_DCLR_1, &pvt->dclr1); + amd64_read_pci_cfg(pvt->dram_f2_ctl, F10_DCHR_1, &pvt->dchr1); } - amd64_dump_misc_regs(pvt); - - return; - -err_reg: - debugf0("Reading an MC register failed\n"); - } /* @@ -2521,7 +2392,7 @@ err_reg: */ static u32 amd64_csrow_nr_pages(int csrow_nr, struct amd64_pvt *pvt) { - u32 dram_map, nr_pages; + u32 cs_mode, nr_pages; /* * The math on this doesn't look right on the surface because x/2*4 can @@ -2530,9 +2401,9 @@ static u32 amd64_csrow_nr_pages(int csrow_nr, struct amd64_pvt *pvt) * number of bits to shift the DBAM register to extract the proper CSROW * field. */ - dram_map = (pvt->dbam0 >> ((csrow_nr / 2) * 4)) & 0xF; + cs_mode = (pvt->dbam0 >> ((csrow_nr / 2) * 4)) & 0xF; - nr_pages = pvt->ops->dbam_map_to_pages(pvt, dram_map); + nr_pages = pvt->ops->dbam_to_cs(pvt, cs_mode) << (20 - PAGE_SHIFT); /* * If dual channel then double the memory size of single channel. @@ -2540,7 +2411,7 @@ static u32 amd64_csrow_nr_pages(int csrow_nr, struct amd64_pvt *pvt) */ nr_pages <<= (pvt->channel_count - 1); - debugf0(" (csrow=%d) DBAM map index= %d\n", csrow_nr, dram_map); + debugf0(" (csrow=%d) DBAM map index= %d\n", csrow_nr, cs_mode); debugf0(" nr_pages= %u channel-count = %d\n", nr_pages, pvt->channel_count); @@ -2556,13 +2427,11 @@ static int amd64_init_csrows(struct mem_ctl_info *mci) struct csrow_info *csrow; struct amd64_pvt *pvt; u64 input_addr_min, input_addr_max, sys_addr; - int i, err = 0, empty = 1; + int i, empty = 1; pvt = mci->pvt_info; - err = pci_read_config_dword(pvt->misc_f3_ctl, K8_NBCFG, &pvt->nbcfg); - if (err) - debugf0("Reading K8_NBCFG failed\n"); + amd64_read_pci_cfg(pvt->misc_f3_ctl, K8_NBCFG, &pvt->nbcfg); debugf0("NBCFG= 0x%x CHIPKILL= %s DRAM ECC= %s\n", pvt->nbcfg, (pvt->nbcfg & K8_NBCFG_CHIPKILL) ? "Enabled" : "Disabled", @@ -2618,6 +2487,90 @@ static int amd64_init_csrows(struct mem_ctl_info *mci) return empty; } +/* get all cores on this DCT */ +static void get_cpus_on_this_dct_cpumask(struct cpumask *mask, int nid) +{ + int cpu; + + for_each_online_cpu(cpu) + if (amd_get_nb_id(cpu) == nid) + cpumask_set_cpu(cpu, mask); +} + +/* check MCG_CTL on all the cpus on this node */ +static bool amd64_nb_mce_bank_enabled_on_node(int nid) +{ + cpumask_var_t mask; + int cpu, nbe; + bool ret = false; + + if (!zalloc_cpumask_var(&mask, GFP_KERNEL)) { + amd64_printk(KERN_WARNING, "%s: error allocating mask\n", + __func__); + return false; + } + + get_cpus_on_this_dct_cpumask(mask, nid); + + rdmsr_on_cpus(mask, MSR_IA32_MCG_CTL, msrs); + + for_each_cpu(cpu, mask) { + struct msr *reg = per_cpu_ptr(msrs, cpu); + nbe = reg->l & K8_MSR_MCGCTL_NBE; + + debugf0("core: %u, MCG_CTL: 0x%llx, NB MSR is %s\n", + cpu, reg->q, + (nbe ? "enabled" : "disabled")); + + if (!nbe) + goto out; + } + ret = true; + +out: + free_cpumask_var(mask); + return ret; +} + +static int amd64_toggle_ecc_err_reporting(struct amd64_pvt *pvt, bool on) +{ + cpumask_var_t cmask; + int cpu; + + if (!zalloc_cpumask_var(&cmask, GFP_KERNEL)) { + amd64_printk(KERN_WARNING, "%s: error allocating mask\n", + __func__); + return false; + } + + get_cpus_on_this_dct_cpumask(cmask, pvt->mc_node_id); + + rdmsr_on_cpus(cmask, MSR_IA32_MCG_CTL, msrs); + + for_each_cpu(cpu, cmask) { + + struct msr *reg = per_cpu_ptr(msrs, cpu); + + if (on) { + if (reg->l & K8_MSR_MCGCTL_NBE) + pvt->flags.ecc_report = 1; + + reg->l |= K8_MSR_MCGCTL_NBE; + } else { + /* + * Turn off ECC reporting only when it was off before + */ + if (!pvt->flags.ecc_report) + reg->l &= ~K8_MSR_MCGCTL_NBE; + } + } + wrmsr_on_cpus(cmask, MSR_IA32_MCG_CTL, msrs); + + free_cpumask_var(cmask); + + return 0; +} + /* * Only if 'ecc_enable_override' is set AND BIOS had ECC disabled, do "we" * enable it. @@ -2625,24 +2578,16 @@ static int amd64_init_csrows(struct mem_ctl_info *mci) static void amd64_enable_ecc_error_reporting(struct mem_ctl_info *mci) { struct amd64_pvt *pvt = mci->pvt_info; - const cpumask_t *cpumask = cpumask_of_node(pvt->mc_node_id); - int cpu, idx = 0, err = 0; - struct msr msrs[cpumask_weight(cpumask)]; - u32 value; - u32 mask = K8_NBCTL_CECCEn | K8_NBCTL_UECCEn; + u32 value, mask = K8_NBCTL_CECCEn | K8_NBCTL_UECCEn; if (!ecc_enable_override) return; - memset(msrs, 0, sizeof(msrs)); - amd64_printk(KERN_WARNING, "'ecc_enable_override' parameter is active, " "Enabling AMD ECC hardware now: CAUTION\n"); - err = pci_read_config_dword(pvt->misc_f3_ctl, K8_NBCTL, &value); - if (err) - debugf0("Reading K8_NBCTL failed\n"); + amd64_read_pci_cfg(pvt->misc_f3_ctl, K8_NBCTL, &value); /* turn on UECCn and CECCEn bits */ pvt->old_nbctl = value & mask; @@ -2651,20 +2596,11 @@ static void amd64_enable_ecc_error_reporting(struct mem_ctl_info *mci) value |= mask; pci_write_config_dword(pvt->misc_f3_ctl, K8_NBCTL, value); - rdmsr_on_cpus(cpumask, K8_MSR_MCGCTL, msrs); - - for_each_cpu(cpu, cpumask) { - if (msrs[idx].l & K8_MSR_MCGCTL_NBE) - set_bit(idx, &pvt->old_mcgctl); - - msrs[idx].l |= K8_MSR_MCGCTL_NBE; - idx++; - } - wrmsr_on_cpus(cpumask, K8_MSR_MCGCTL, msrs); + if (amd64_toggle_ecc_err_reporting(pvt, ON)) + amd64_printk(KERN_WARNING, "Error enabling ECC reporting over " + "MCGCTL!\n"); - err = pci_read_config_dword(pvt->misc_f3_ctl, K8_NBCFG, &value); - if (err) - debugf0("Reading K8_NBCFG failed\n"); + amd64_read_pci_cfg(pvt->misc_f3_ctl, K8_NBCFG, &value); debugf0("NBCFG(1)= 0x%x CHIPKILL= %s ECC_ENABLE= %s\n", value, (value & K8_NBCFG_CHIPKILL) ? "Enabled" : "Disabled", @@ -2679,9 +2615,7 @@ static void amd64_enable_ecc_error_reporting(struct mem_ctl_info *mci) value |= K8_NBCFG_ECC_ENABLE; pci_write_config_dword(pvt->misc_f3_ctl, K8_NBCFG, value); - err = pci_read_config_dword(pvt->misc_f3_ctl, K8_NBCFG, &value); - if (err) - debugf0("Reading K8_NBCFG failed\n"); + amd64_read_pci_cfg(pvt->misc_f3_ctl, K8_NBCFG, &value); if (!(value & K8_NBCFG_ECC_ENABLE)) { amd64_printk(KERN_WARNING, @@ -2701,86 +2635,21 @@ static void amd64_enable_ecc_error_reporting(struct mem_ctl_info *mci) static void amd64_restore_ecc_error_reporting(struct amd64_pvt *pvt) { - const cpumask_t *cpumask = cpumask_of_node(pvt->mc_node_id); - int cpu, idx = 0, err = 0; - struct msr msrs[cpumask_weight(cpumask)]; - u32 value; - u32 mask = K8_NBCTL_CECCEn | K8_NBCTL_UECCEn; + u32 value, mask = K8_NBCTL_CECCEn | K8_NBCTL_UECCEn; if (!pvt->nbctl_mcgctl_saved) return; - memset(msrs, 0, sizeof(msrs)); - - err = pci_read_config_dword(pvt->misc_f3_ctl, K8_NBCTL, &value); - if (err) - debugf0("Reading K8_NBCTL failed\n"); + amd64_read_pci_cfg(pvt->misc_f3_ctl, K8_NBCTL, &value); value &= ~mask; value |= pvt->old_nbctl; /* restore the NB Enable MCGCTL bit */ pci_write_config_dword(pvt->misc_f3_ctl, K8_NBCTL, value); - rdmsr_on_cpus(cpumask, K8_MSR_MCGCTL, msrs); - - for_each_cpu(cpu, cpumask) { - msrs[idx].l &= ~K8_MSR_MCGCTL_NBE; - msrs[idx].l |= - test_bit(idx, &pvt->old_mcgctl) << K8_MSR_MCGCTL_NBE; - idx++; - } - - wrmsr_on_cpus(cpumask, K8_MSR_MCGCTL, msrs); -} - -/* get all cores on this DCT */ -static void get_cpus_on_this_dct_cpumask(cpumask_t *mask, int nid) -{ - int cpu; - - for_each_online_cpu(cpu) - if (amd_get_nb_id(cpu) == nid) - cpumask_set_cpu(cpu, mask); -} - -/* check MCG_CTL on all the cpus on this node */ -static bool amd64_nb_mce_bank_enabled_on_node(int nid) -{ - cpumask_t mask; - struct msr *msrs; - int cpu, nbe, idx = 0; - bool ret = false; - - cpumask_clear(&mask); - - get_cpus_on_this_dct_cpumask(&mask, nid); - - msrs = kzalloc(sizeof(struct msr) * cpumask_weight(&mask), GFP_KERNEL); - if (!msrs) { - amd64_printk(KERN_WARNING, "%s: error allocating msrs\n", - __func__); - return false; - } - - rdmsr_on_cpus(&mask, MSR_IA32_MCG_CTL, msrs); - - for_each_cpu(cpu, &mask) { - nbe = msrs[idx].l & K8_MSR_MCGCTL_NBE; - - debugf0("core: %u, MCG_CTL: 0x%llx, NB MSR is %s\n", - cpu, msrs[idx].q, - (nbe ? "enabled" : "disabled")); - - if (!nbe) - goto out; - - idx++; - } - ret = true; - -out: - kfree(msrs); - return ret; + if (amd64_toggle_ecc_err_reporting(pvt, OFF)) + amd64_printk(KERN_WARNING, "Error restoring ECC reporting over " + "MCGCTL!\n"); } /* @@ -2797,13 +2666,10 @@ static const char *ecc_warning = static int amd64_check_ecc_enabled(struct amd64_pvt *pvt) { u32 value; - int err = 0; u8 ecc_enabled = 0; bool nb_mce_en = false; - err = pci_read_config_dword(pvt->misc_f3_ctl, K8_NBCFG, &value); - if (err) - debugf0("Reading K8_NBCTL failed\n"); + amd64_read_pci_cfg(pvt->misc_f3_ctl, K8_NBCFG, &value); ecc_enabled = !!(value & K8_NBCFG_ECC_ENABLE); if (!ecc_enabled) @@ -2824,9 +2690,8 @@ static int amd64_check_ecc_enabled(struct amd64_pvt *pvt) amd64_printk(KERN_WARNING, "%s", ecc_warning); return -ENODEV; } - } else - /* CLEAR the override, since BIOS controlled it */ ecc_enable_override = 0; + } return 0; } @@ -2909,7 +2774,6 @@ static int amd64_probe_one_instance(struct pci_dev *dram_f2_ctl, pvt->ext_model = boot_cpu_data.x86_model >> 4; pvt->mc_type_index = mc_type_index; pvt->ops = family_ops(mc_type_index); - pvt->old_mcgctl = 0; /* * We have the dram_f2_ctl device as an argument, now go reserve its @@ -2959,17 +2823,10 @@ static int amd64_init_2nd_stage(struct amd64_pvt *pvt) { int node_id = pvt->mc_node_id; struct mem_ctl_info *mci; - int ret, err = 0; + int ret = -ENODEV; amd64_read_mc_registers(pvt); - ret = -ENODEV; - if (pvt->ops->probe_valid_hardware) { - err = pvt->ops->probe_valid_hardware(pvt); - if (err) - goto err_exit; - } - /* * We need to determine how many memory channels there are. Then use * that information for calculating the size of the dynamic instance @@ -3071,16 +2928,15 @@ static void __devexit amd64_remove_one_instance(struct pci_dev *pdev) amd64_free_mc_sibling_devices(pvt); - kfree(pvt); - mci->pvt_info = NULL; - - mci_lookup[pvt->mc_node_id] = NULL; - /* unregister from EDAC MCE */ amd_report_gart_errors(false); amd_unregister_ecc_decoder(amd64_decode_bus_error); /* Free the EDAC CORE resources */ + mci->pvt_info = NULL; + mci_lookup[pvt->mc_node_id] = NULL; + + kfree(pvt); edac_mc_free(mci); } @@ -3157,23 +3013,29 @@ static void amd64_setup_pci_device(void) static int __init amd64_edac_init(void) { int nb, err = -ENODEV; + bool load_ok = false; edac_printk(KERN_INFO, EDAC_MOD_STR, EDAC_AMD64_VERSION "\n"); opstate_init(); if (cache_k8_northbridges() < 0) - return err; + goto err_ret; + + msrs = msrs_alloc(); + if (!msrs) + goto err_ret; err = pci_register_driver(&amd64_pci_driver); if (err) - return err; + goto err_pci; /* * At this point, the array 'pvt_lookup[]' contains pointers to alloc'd * amd64_pvt structs. These will be used in the 2nd stage init function * to finish initialization of the MC instances. */ + err = -ENODEV; for (nb = 0; nb < num_k8_northbridges; nb++) { if (!pvt_lookup[nb]) continue; @@ -3181,16 +3043,21 @@ static int __init amd64_edac_init(void) err = amd64_init_2nd_stage(pvt_lookup[nb]); if (err) goto err_2nd_stage; - } - amd64_setup_pci_device(); + load_ok = true; + } - return 0; + if (load_ok) { + amd64_setup_pci_device(); + return 0; + } err_2nd_stage: - debugf0("2nd stage failed\n"); pci_unregister_driver(&amd64_pci_driver); - +err_pci: + msrs_free(msrs); + msrs = NULL; +err_ret: return err; } @@ -3200,6 +3067,9 @@ static void __exit amd64_edac_exit(void) edac_pci_release_generic_ctl(amd64_ctl_pci); pci_unregister_driver(&amd64_pci_driver); + + msrs_free(msrs); + msrs = NULL; } module_init(amd64_edac_init); |