diff options
| -rw-r--r-- | drivers/edac/amd64_edac.c | 204 | ||||
| -rw-r--r-- | drivers/edac/amd64_edac.h | 9 |
2 files changed, 117 insertions, 96 deletions
diff --git a/drivers/edac/amd64_edac.c b/drivers/edac/amd64_edac.c index 67541e7..70c7d5f 100644 --- a/drivers/edac/amd64_edac.c +++ b/drivers/edac/amd64_edac.c @@ -2624,6 +2624,109 @@ 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; + struct msr *msrs; + int cpu, nbe, idx = 0; + 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); + + msrs = kzalloc(sizeof(struct msr) * cpumask_weight(mask), GFP_KERNEL); + if (!msrs) { + amd64_printk(KERN_WARNING, "%s: error allocating msrs\n", + __func__); + free_cpumask_var(mask); + 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); + free_cpumask_var(mask); + return ret; +} + +static int amd64_toggle_ecc_err_reporting(struct amd64_pvt *pvt, bool on) +{ + cpumask_var_t cmask; + struct msr *msrs = NULL; + int cpu, idx = 0; + + 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); + + msrs = kzalloc(sizeof(struct msr) * cpumask_weight(cmask), GFP_KERNEL); + if (!msrs) { + amd64_printk(KERN_WARNING, "%s: error allocating msrs\n", + __func__); + return -ENOMEM; + } + + rdmsr_on_cpus(cmask, MSR_IA32_MCG_CTL, msrs); + + for_each_cpu(cpu, cmask) { + + if (on) { + if (msrs[idx].l & K8_MSR_MCGCTL_NBE) + pvt->flags.ecc_report = 1; + + msrs[idx].l |= K8_MSR_MCGCTL_NBE; + } else { + /* + * Turn off ECC reporting only when it was off before + */ + if (!pvt->flags.ecc_report) + msrs[idx].l &= ~K8_MSR_MCGCTL_NBE; + } + idx++; + } + wrmsr_on_cpus(cmask, MSR_IA32_MCG_CTL, msrs); + + kfree(msrs); + free_cpumask_var(cmask); + + return 0; +} + /* * Only if 'ecc_enable_override' is set AND BIOS had ECC disabled, do "we" * enable it. @@ -2631,17 +2734,12 @@ 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 struct cpumask *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; + int err = 0; + 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"); @@ -2657,16 +2755,9 @@ 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) @@ -2707,17 +2798,12 @@ static void amd64_enable_ecc_error_reporting(struct mem_ctl_info *mci) static void amd64_restore_ecc_error_reporting(struct amd64_pvt *pvt) { - const struct cpumask *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; + int err = 0; + 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"); @@ -2727,72 +2813,9 @@ static void amd64_restore_ecc_error_reporting(struct amd64_pvt *pvt) /* 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(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; - struct msr *msrs; - int cpu, nbe, idx = 0; - 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); - - msrs = kzalloc(sizeof(struct msr) * cpumask_weight(mask), GFP_KERNEL); - if (!msrs) { - amd64_printk(KERN_WARNING, "%s: error allocating msrs\n", - __func__); - free_cpumask_var(mask); - 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); - free_cpumask_var(mask); - return ret; + if (amd64_toggle_ecc_err_reporting(pvt, OFF)) + amd64_printk(KERN_WARNING, "Error restoring ECC reporting over " + "MCGCTL!\n"); } /* @@ -2921,7 +2944,6 @@ 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 diff --git a/drivers/edac/amd64_edac.h b/drivers/edac/amd64_edac.h index c6f359a..bba6c94 100644 --- a/drivers/edac/amd64_edac.h +++ b/drivers/edac/amd64_edac.h @@ -147,6 +147,8 @@ #define MAX_CS_COUNT 8 #define DRAM_REG_COUNT 8 +#define ON true +#define OFF false /* * PCI-defined configuration space registers @@ -386,10 +388,7 @@ enum { #define K8_NBCAP_DUAL_NODE BIT(1) #define K8_NBCAP_DCT_DUAL BIT(0) -/* - * MSR Regs - */ -#define K8_MSR_MCGCTL 0x017b +/* MSRs */ #define K8_MSR_MCGCTL_NBE BIT(4) #define K8_MSR_MC4CTL 0x0410 @@ -487,7 +486,6 @@ struct amd64_pvt { /* Save old hw registers' values before we modified them */ u32 nbctl_mcgctl_saved; /* When true, following 2 are valid */ u32 old_nbctl; - unsigned long old_mcgctl; /* per core on this node */ /* MC Type Index value: socket F vs Family 10h */ u32 mc_type_index; @@ -495,6 +493,7 @@ struct amd64_pvt { /* misc settings */ struct flags { unsigned long cf8_extcfg:1; + unsigned long ecc_report:1; } flags; }; |
