diff options
Diffstat (limited to 'drivers/edac')
-rw-r--r-- | drivers/edac/Kconfig | 1 | ||||
-rw-r--r-- | drivers/edac/skx_edac.c | 193 |
2 files changed, 181 insertions, 13 deletions
diff --git a/drivers/edac/Kconfig b/drivers/edac/Kconfig index 57304b2e989f..ffd349c12479 100644 --- a/drivers/edac/Kconfig +++ b/drivers/edac/Kconfig @@ -234,6 +234,7 @@ config EDAC_SKX depends on PCI && X86_64 && X86_MCE_INTEL && PCI_MMCONFIG depends on ACPI_NFIT || !ACPI_NFIT # if ACPI_NFIT=m, EDAC_SKX can't be y select DMI + select ACPI_ADXL help Support for error detection and correction the Intel Skylake server Integrated Memory Controllers. If your diff --git a/drivers/edac/skx_edac.c b/drivers/edac/skx_edac.c index dd209e0dd9ab..a99ea61dad32 100644 --- a/drivers/edac/skx_edac.c +++ b/drivers/edac/skx_edac.c @@ -26,6 +26,7 @@ #include <linux/bitmap.h> #include <linux/math64.h> #include <linux/mod_devicetable.h> +#include <linux/adxl.h> #include <acpi/nfit.h> #include <asm/cpu_device_id.h> #include <asm/intel-family.h> @@ -35,6 +36,7 @@ #include "edac_module.h" #define EDAC_MOD_STR "skx_edac" +#define MSG_SIZE 1024 /* * Debug macros @@ -54,6 +56,29 @@ static LIST_HEAD(skx_edac_list); static u64 skx_tolm, skx_tohm; +static char *skx_msg; +static unsigned int nvdimm_count; + +enum { + INDEX_SOCKET, + INDEX_MEMCTRL, + INDEX_CHANNEL, + INDEX_DIMM, + INDEX_MAX +}; + +static const char * const component_names[] = { + [INDEX_SOCKET] = "ProcessorSocketId", + [INDEX_MEMCTRL] = "MemoryControllerId", + [INDEX_CHANNEL] = "ChannelId", + [INDEX_DIMM] = "DimmSlotId", +}; + +static int component_indices[ARRAY_SIZE(component_names)]; +static int adxl_component_count; +static const char * const *adxl_component_names; +static u64 *adxl_values; +static char *adxl_msg; #define NUM_IMC 2 /* memory controllers per socket */ #define NUM_CHANNELS 3 /* channels per memory controller */ @@ -393,6 +418,8 @@ static int get_nvdimm_info(struct dimm_info *dimm, struct skx_imc *imc, u16 flags; u64 size = 0; + nvdimm_count++; + dev_handle = ACPI_NFIT_BUILD_DEVICE_HANDLE(dimmno, chan, imc->lmc, imc->src_id, 0); @@ -941,12 +968,46 @@ static void teardown_skx_debug(void) } #endif /*CONFIG_EDAC_DEBUG*/ +static bool skx_adxl_decode(struct decoded_addr *res) + +{ + int i, len = 0; + + if (res->addr >= skx_tohm || (res->addr >= skx_tolm && + res->addr < BIT_ULL(32))) { + edac_dbg(0, "Address 0x%llx out of range\n", res->addr); + return false; + } + + if (adxl_decode(res->addr, adxl_values)) { + edac_dbg(0, "Failed to decode 0x%llx\n", res->addr); + return false; + } + + res->socket = (int)adxl_values[component_indices[INDEX_SOCKET]]; + res->imc = (int)adxl_values[component_indices[INDEX_MEMCTRL]]; + res->channel = (int)adxl_values[component_indices[INDEX_CHANNEL]]; + res->dimm = (int)adxl_values[component_indices[INDEX_DIMM]]; + + for (i = 0; i < adxl_component_count; i++) { + if (adxl_values[i] == ~0x0ull) + continue; + + len += snprintf(adxl_msg + len, MSG_SIZE - len, " %s:0x%llx", + adxl_component_names[i], adxl_values[i]); + if (MSG_SIZE - len <= 0) + break; + } + + return true; +} + static void skx_mce_output_error(struct mem_ctl_info *mci, const struct mce *m, struct decoded_addr *res) { enum hw_event_mc_err_type tp_event; - char *type, *optype, msg[256]; + char *type, *optype; bool ripv = GET_BITFIELD(m->mcgstatus, 0, 0); bool overflow = GET_BITFIELD(m->status, 62, 62); bool uncorrected_error = GET_BITFIELD(m->status, 61, 61); @@ -1007,22 +1068,47 @@ static void skx_mce_output_error(struct mem_ctl_info *mci, break; } } + if (adxl_component_count) { + snprintf(skx_msg, MSG_SIZE, "%s%s err_code:%04x:%04x %s", + overflow ? " OVERFLOW" : "", + (uncorrected_error && recoverable) ? " recoverable" : "", + mscod, errcode, adxl_msg); + } else { + snprintf(skx_msg, MSG_SIZE, + "%s%s err_code:%04x:%04x socket:%d imc:%d rank:%d bg:%d ba:%d row:%x col:%x", + overflow ? " OVERFLOW" : "", + (uncorrected_error && recoverable) ? " recoverable" : "", + mscod, errcode, + res->socket, res->imc, res->rank, + res->bank_group, res->bank_address, res->row, res->column); + } - snprintf(msg, sizeof(msg), - "%s%s err_code:%04x:%04x socket:%d imc:%d rank:%d bg:%d ba:%d row:%x col:%x", - overflow ? " OVERFLOW" : "", - (uncorrected_error && recoverable) ? " recoverable" : "", - mscod, errcode, - res->socket, res->imc, res->rank, - res->bank_group, res->bank_address, res->row, res->column); - - edac_dbg(0, "%s\n", msg); + edac_dbg(0, "%s\n", skx_msg); /* Call the helper to output message */ edac_mc_handle_error(tp_event, mci, core_err_cnt, m->addr >> PAGE_SHIFT, m->addr & ~PAGE_MASK, 0, res->channel, res->dimm, -1, - optype, msg); + optype, skx_msg); +} + +static struct mem_ctl_info *get_mci(int src_id, int lmc) +{ + struct skx_dev *d; + + if (lmc > NUM_IMC - 1) { + skx_printk(KERN_ERR, "Bad lmc %d\n", lmc); + return NULL; + } + + list_for_each_entry(d, &skx_edac_list, list) { + if (d->imc[0].src_id == src_id) + return d->imc[lmc].mci; + } + + skx_printk(KERN_ERR, "No mci for src_id %d lmc %d\n", src_id, lmc); + + return NULL; } static int skx_mce_check_error(struct notifier_block *nb, unsigned long val, @@ -1040,10 +1126,23 @@ static int skx_mce_check_error(struct notifier_block *nb, unsigned long val, if ((mce->status & 0xefff) >> 7 != 1 || !(mce->status & MCI_STATUS_ADDRV)) return NOTIFY_DONE; + memset(&res, 0, sizeof(res)); res.addr = mce->addr; - if (!skx_decode(&res)) + + if (adxl_component_count) { + if (!skx_adxl_decode(&res)) + return NOTIFY_DONE; + + mci = get_mci(res.socket, res.imc); + } else { + if (!skx_decode(&res)) + return NOTIFY_DONE; + + mci = res.dev->imc[res.imc].mci; + } + + if (!mci) return NOTIFY_DONE; - mci = res.dev->imc[res.imc].mci; if (mce->mcgstatus & MCG_STATUS_MCIP) type = "Exception"; @@ -1094,6 +1193,62 @@ static void skx_remove(void) } } +static void __init skx_adxl_get(void) +{ + const char * const *names; + int i, j; + + names = adxl_get_component_names(); + if (!names) { + skx_printk(KERN_NOTICE, "No firmware support for address translation."); + skx_printk(KERN_CONT, " Only decoding DDR4 address!\n"); + return; + } + + for (i = 0; i < INDEX_MAX; i++) { + for (j = 0; names[j]; j++) { + if (!strcmp(component_names[i], names[j])) { + component_indices[i] = j; + break; + } + } + + if (!names[j]) + goto err; + } + + adxl_component_names = names; + while (*names++) + adxl_component_count++; + + adxl_values = kcalloc(adxl_component_count, sizeof(*adxl_values), + GFP_KERNEL); + if (!adxl_values) { + adxl_component_count = 0; + return; + } + + adxl_msg = kzalloc(MSG_SIZE, GFP_KERNEL); + if (!adxl_msg) { + adxl_component_count = 0; + kfree(adxl_values); + } + + return; +err: + skx_printk(KERN_ERR, "'%s' is not matched from DSM parameters: ", + component_names[i]); + for (j = 0; names[j]; j++) + skx_printk(KERN_CONT, "%s ", names[j]); + skx_printk(KERN_CONT, "\n"); +} + +static void __exit skx_adxl_put(void) +{ + kfree(adxl_values); + kfree(adxl_msg); +} + /* * skx_init: * make sure we are running on the correct cpu model @@ -1158,6 +1313,15 @@ static int __init skx_init(void) } } + skx_msg = kzalloc(MSG_SIZE, GFP_KERNEL); + if (!skx_msg) { + rc = -ENOMEM; + goto fail; + } + + if (nvdimm_count) + skx_adxl_get(); + /* Ensure that the OPSTATE is set correctly for POLL or NMI */ opstate_init(); @@ -1176,6 +1340,9 @@ static void __exit skx_exit(void) edac_dbg(2, "\n"); mce_unregister_decode_chain(&skx_mce_dec); skx_remove(); + if (nvdimm_count) + skx_adxl_put(); + kfree(skx_msg); teardown_skx_debug(); } |