1 // SPDX-License-Identifier: GPL-2.0
4 * Shared code by both skx_edac and i10nm_edac. Originally split out
5 * from the skx_edac driver.
7 * This file is linked into both skx_edac and i10nm_edac drivers. In
8 * order to avoid link errors, this file must be like a pure library
9 * without including symbols and defines which would otherwise conflict,
10 * when linked once into a module and into a built-in object, at the
11 * same time. For example, __this_module symbol references when that
12 * file is being linked into a built-in object.
14 * Copyright (c) 2018, Intel Corporation.
17 #include <linux/acpi.h>
18 #include <linux/dmi.h>
19 #include <linux/adxl.h>
20 #include <acpi/nfit.h>
22 #include "edac_module.h"
23 #include "skx_common.h"
25 static const char * const component_names[] = {
26 [INDEX_SOCKET] = "ProcessorSocketId",
27 [INDEX_MEMCTRL] = "MemoryControllerId",
28 [INDEX_CHANNEL] = "ChannelId",
29 [INDEX_DIMM] = "DimmSlotId",
32 static int component_indices[ARRAY_SIZE(component_names)];
33 static int adxl_component_count;
34 static const char * const *adxl_component_names;
35 static u64 *adxl_values;
36 static char *adxl_msg;
38 static char skx_msg[MSG_SIZE];
39 static skx_decode_f skx_decode;
40 static u64 skx_tolm, skx_tohm;
41 static LIST_HEAD(dev_edac_list);
43 int __init skx_adxl_get(void)
45 const char * const *names;
48 names = adxl_get_component_names();
50 skx_printk(KERN_NOTICE, "No firmware support for address translation.\n");
54 for (i = 0; i < INDEX_MAX; i++) {
55 for (j = 0; names[j]; j++) {
56 if (!strcmp(component_names[i], names[j])) {
57 component_indices[i] = j;
66 adxl_component_names = names;
68 adxl_component_count++;
70 adxl_values = kcalloc(adxl_component_count, sizeof(*adxl_values),
73 adxl_component_count = 0;
77 adxl_msg = kzalloc(MSG_SIZE, GFP_KERNEL);
79 adxl_component_count = 0;
86 skx_printk(KERN_ERR, "'%s' is not matched from DSM parameters: ",
88 for (j = 0; names[j]; j++)
89 skx_printk(KERN_CONT, "%s ", names[j]);
90 skx_printk(KERN_CONT, "\n");
95 void __exit skx_adxl_put(void)
101 static bool skx_adxl_decode(struct decoded_addr *res)
105 if (res->addr >= skx_tohm || (res->addr >= skx_tolm &&
106 res->addr < BIT_ULL(32))) {
107 edac_dbg(0, "Address 0x%llx out of range\n", res->addr);
111 if (adxl_decode(res->addr, adxl_values)) {
112 edac_dbg(0, "Failed to decode 0x%llx\n", res->addr);
116 res->socket = (int)adxl_values[component_indices[INDEX_SOCKET]];
117 res->imc = (int)adxl_values[component_indices[INDEX_MEMCTRL]];
118 res->channel = (int)adxl_values[component_indices[INDEX_CHANNEL]];
119 res->dimm = (int)adxl_values[component_indices[INDEX_DIMM]];
121 for (i = 0; i < adxl_component_count; i++) {
122 if (adxl_values[i] == ~0x0ull)
125 len += snprintf(adxl_msg + len, MSG_SIZE - len, " %s:0x%llx",
126 adxl_component_names[i], adxl_values[i]);
127 if (MSG_SIZE - len <= 0)
134 void skx_set_decode(skx_decode_f decode)
139 int skx_get_src_id(struct skx_dev *d, int off, u8 *id)
143 if (pci_read_config_dword(d->util_all, off, ®)) {
144 skx_printk(KERN_ERR, "Failed to read src id\n");
148 *id = GET_BITFIELD(reg, 12, 14);
152 int skx_get_node_id(struct skx_dev *d, u8 *id)
156 if (pci_read_config_dword(d->util_all, 0xf4, ®)) {
157 skx_printk(KERN_ERR, "Failed to read node id\n");
161 *id = GET_BITFIELD(reg, 0, 2);
165 static int get_width(u32 mtr)
167 switch (GET_BITFIELD(mtr, 8, 9)) {
179 * We use the per-socket device @did to count how many sockets are present,
180 * and to detemine which PCI buses are associated with each socket. Allocate
181 * and build the full list of all the skx_dev structures that we need here.
183 int skx_get_all_bus_mappings(unsigned int did, int off, enum type type,
184 struct list_head **list)
186 struct pci_dev *pdev, *prev;
193 pdev = pci_get_device(PCI_VENDOR_ID_INTEL, did, prev);
197 d = kzalloc(sizeof(*d), GFP_KERNEL);
203 if (pci_read_config_dword(pdev, off, ®)) {
206 skx_printk(KERN_ERR, "Failed to read bus idx\n");
210 d->bus[0] = GET_BITFIELD(reg, 0, 7);
211 d->bus[1] = GET_BITFIELD(reg, 8, 15);
213 d->seg = pci_domain_nr(pdev->bus);
214 d->bus[2] = GET_BITFIELD(reg, 16, 23);
215 d->bus[3] = GET_BITFIELD(reg, 24, 31);
217 d->seg = GET_BITFIELD(reg, 16, 23);
220 edac_dbg(2, "busses: 0x%x, 0x%x, 0x%x, 0x%x\n",
221 d->bus[0], d->bus[1], d->bus[2], d->bus[3]);
222 list_add_tail(&d->list, &dev_edac_list);
227 *list = &dev_edac_list;
231 int skx_get_hi_lo(unsigned int did, int off[], u64 *tolm, u64 *tohm)
233 struct pci_dev *pdev;
236 pdev = pci_get_device(PCI_VENDOR_ID_INTEL, did, NULL);
238 skx_printk(KERN_ERR, "Can't get tolm/tohm\n");
242 if (pci_read_config_dword(pdev, off[0], ®)) {
243 skx_printk(KERN_ERR, "Failed to read tolm\n");
248 if (pci_read_config_dword(pdev, off[1], ®)) {
249 skx_printk(KERN_ERR, "Failed to read lower tohm\n");
254 if (pci_read_config_dword(pdev, off[2], ®)) {
255 skx_printk(KERN_ERR, "Failed to read upper tohm\n");
258 skx_tohm |= (u64)reg << 32;
263 edac_dbg(2, "tolm = 0x%llx tohm = 0x%llx\n", skx_tolm, skx_tohm);
270 static int skx_get_dimm_attr(u32 reg, int lobit, int hibit, int add,
271 int minval, int maxval, const char *name)
273 u32 val = GET_BITFIELD(reg, lobit, hibit);
275 if (val < minval || val > maxval) {
276 edac_dbg(2, "bad %s = %d (raw=0x%x)\n", name, val, reg);
282 #define numrank(reg) skx_get_dimm_attr(reg, 12, 13, 0, 0, 2, "ranks")
283 #define numrow(reg) skx_get_dimm_attr(reg, 2, 4, 12, 1, 6, "rows")
284 #define numcol(reg) skx_get_dimm_attr(reg, 0, 1, 10, 0, 2, "cols")
286 int skx_get_dimm_info(u32 mtr, u32 amap, struct dimm_info *dimm,
287 struct skx_imc *imc, int chan, int dimmno)
289 int banks = 16, ranks, rows, cols, npages;
292 ranks = numrank(mtr);
297 * Compute size in 8-byte (2^3) words, then shift to MiB (2^20)
299 size = ((1ull << (rows + cols + ranks)) * banks) >> (20 - 3);
300 npages = MiB_TO_PAGES(size);
302 edac_dbg(0, "mc#%d: channel %d, dimm %d, %lld MiB (%d pages) bank: %d, rank: %d, row: 0x%x, col: 0x%x\n",
303 imc->mc, chan, dimmno, size, npages,
304 banks, 1 << ranks, rows, cols);
306 imc->chan[chan].dimms[dimmno].close_pg = GET_BITFIELD(mtr, 0, 0);
307 imc->chan[chan].dimms[dimmno].bank_xor_enable = GET_BITFIELD(mtr, 9, 9);
308 imc->chan[chan].dimms[dimmno].fine_grain_bank = GET_BITFIELD(amap, 0, 0);
309 imc->chan[chan].dimms[dimmno].rowbits = rows;
310 imc->chan[chan].dimms[dimmno].colbits = cols;
312 dimm->nr_pages = npages;
314 dimm->dtype = get_width(mtr);
315 dimm->mtype = MEM_DDR4;
316 dimm->edac_mode = EDAC_SECDED; /* likely better than this */
317 snprintf(dimm->label, sizeof(dimm->label), "CPU_SrcID#%u_MC#%u_Chan#%u_DIMM#%u",
318 imc->src_id, imc->lmc, chan, dimmno);
323 int skx_get_nvdimm_info(struct dimm_info *dimm, struct skx_imc *imc,
324 int chan, int dimmno, const char *mod_str)
331 dev_handle = ACPI_NFIT_BUILD_DEVICE_HANDLE(dimmno, chan, imc->lmc,
334 smbios_handle = nfit_get_smbios_id(dev_handle, &flags);
335 if (smbios_handle == -EOPNOTSUPP) {
336 pr_warn_once("%s: Can't find size of NVDIMM. Try enabling CONFIG_ACPI_NFIT\n", mod_str);
340 if (smbios_handle < 0) {
341 skx_printk(KERN_ERR, "Can't find handle for NVDIMM ADR=0x%x\n", dev_handle);
345 if (flags & ACPI_NFIT_MEM_MAP_FAILED) {
346 skx_printk(KERN_ERR, "NVDIMM ADR=0x%x is not mapped\n", dev_handle);
350 size = dmi_memdev_size(smbios_handle);
352 skx_printk(KERN_ERR, "Can't find size for NVDIMM ADR=0x%x/SMBIOS=0x%x\n",
353 dev_handle, smbios_handle);
356 dimm->nr_pages = size >> PAGE_SHIFT;
358 dimm->dtype = DEV_UNKNOWN;
359 dimm->mtype = MEM_NVDIMM;
360 dimm->edac_mode = EDAC_SECDED; /* likely better than this */
362 edac_dbg(0, "mc#%d: channel %d, dimm %d, %llu MiB (%u pages)\n",
363 imc->mc, chan, dimmno, size >> 20, dimm->nr_pages);
365 snprintf(dimm->label, sizeof(dimm->label), "CPU_SrcID#%u_MC#%u_Chan#%u_DIMM#%u",
366 imc->src_id, imc->lmc, chan, dimmno);
368 return (size == 0 || size == ~0ull) ? 0 : 1;
371 int skx_register_mci(struct skx_imc *imc, struct pci_dev *pdev,
372 const char *ctl_name, const char *mod_str,
373 get_dimm_config_f get_dimm_config)
375 struct mem_ctl_info *mci;
376 struct edac_mc_layer layers[2];
380 /* Allocate a new MC control structure */
381 layers[0].type = EDAC_MC_LAYER_CHANNEL;
382 layers[0].size = NUM_CHANNELS;
383 layers[0].is_virt_csrow = false;
384 layers[1].type = EDAC_MC_LAYER_SLOT;
385 layers[1].size = NUM_DIMMS;
386 layers[1].is_virt_csrow = true;
387 mci = edac_mc_alloc(imc->mc, ARRAY_SIZE(layers), layers,
388 sizeof(struct skx_pvt));
393 edac_dbg(0, "MC#%d: mci = %p\n", imc->mc, mci);
395 /* Associate skx_dev and mci for future usage */
400 mci->ctl_name = kasprintf(GFP_KERNEL, "%s#%d IMC#%d", ctl_name,
401 imc->node_id, imc->lmc);
402 if (!mci->ctl_name) {
407 mci->mtype_cap = MEM_FLAG_DDR4 | MEM_FLAG_NVDIMM;
408 mci->edac_ctl_cap = EDAC_FLAG_NONE;
409 mci->edac_cap = EDAC_FLAG_NONE;
410 mci->mod_name = mod_str;
411 mci->dev_name = pci_name(pdev);
412 mci->ctl_page_to_phys = NULL;
414 rc = get_dimm_config(mci);
418 /* Record ptr to the generic device */
419 mci->pdev = &pdev->dev;
421 /* Add this new MC control structure to EDAC's list of MCs */
422 if (unlikely(edac_mc_add_mc(mci))) {
423 edac_dbg(0, "MC: failed edac_mc_add_mc()\n");
431 kfree(mci->ctl_name);
438 static void skx_unregister_mci(struct skx_imc *imc)
440 struct mem_ctl_info *mci = imc->mci;
445 edac_dbg(0, "MC%d: mci = %p\n", imc->mc, mci);
447 /* Remove MC sysfs nodes */
448 edac_mc_del_mc(mci->pdev);
450 edac_dbg(1, "%s: free mci struct\n", mci->ctl_name);
451 kfree(mci->ctl_name);
455 static struct mem_ctl_info *get_mci(int src_id, int lmc)
459 if (lmc > NUM_IMC - 1) {
460 skx_printk(KERN_ERR, "Bad lmc %d\n", lmc);
464 list_for_each_entry(d, &dev_edac_list, list) {
465 if (d->imc[0].src_id == src_id)
466 return d->imc[lmc].mci;
469 skx_printk(KERN_ERR, "No mci for src_id %d lmc %d\n", src_id, lmc);
473 static void skx_mce_output_error(struct mem_ctl_info *mci,
475 struct decoded_addr *res)
477 enum hw_event_mc_err_type tp_event;
479 bool ripv = GET_BITFIELD(m->mcgstatus, 0, 0);
480 bool overflow = GET_BITFIELD(m->status, 62, 62);
481 bool uncorrected_error = GET_BITFIELD(m->status, 61, 61);
483 u32 core_err_cnt = GET_BITFIELD(m->status, 38, 52);
484 u32 mscod = GET_BITFIELD(m->status, 16, 31);
485 u32 errcode = GET_BITFIELD(m->status, 0, 15);
486 u32 optypenum = GET_BITFIELD(m->status, 4, 6);
488 recoverable = GET_BITFIELD(m->status, 56, 56);
490 if (uncorrected_error) {
494 tp_event = HW_EVENT_ERR_FATAL;
497 tp_event = HW_EVENT_ERR_UNCORRECTED;
501 tp_event = HW_EVENT_ERR_CORRECTED;
505 * According to Intel Architecture spec vol 3B,
506 * Table 15-10 "IA32_MCi_Status [15:0] Compound Error Code Encoding"
507 * memory errors should fit one of these masks:
508 * 000f 0000 1mmm cccc (binary)
509 * 000f 0010 1mmm cccc (binary) [RAM used as cache]
511 * f = Correction Report Filtering Bit. If 1, subsequent errors
515 * If the mask doesn't match, report an error to the parsing logic
517 if (!((errcode & 0xef80) == 0x80 || (errcode & 0xef80) == 0x280)) {
518 optype = "Can't parse: it is not a mem";
522 optype = "generic undef request error";
525 optype = "memory read error";
528 optype = "memory write error";
531 optype = "addr/cmd error";
534 optype = "memory scrubbing error";
541 if (adxl_component_count) {
542 snprintf(skx_msg, MSG_SIZE, "%s%s err_code:0x%04x:0x%04x %s",
543 overflow ? " OVERFLOW" : "",
544 (uncorrected_error && recoverable) ? " recoverable" : "",
545 mscod, errcode, adxl_msg);
547 snprintf(skx_msg, MSG_SIZE,
548 "%s%s err_code:0x%04x:0x%04x socket:%d imc:%d rank:%d bg:%d ba:%d row:0x%x col:0x%x",
549 overflow ? " OVERFLOW" : "",
550 (uncorrected_error && recoverable) ? " recoverable" : "",
552 res->socket, res->imc, res->rank,
553 res->bank_group, res->bank_address, res->row, res->column);
556 edac_dbg(0, "%s\n", skx_msg);
558 /* Call the helper to output message */
559 edac_mc_handle_error(tp_event, mci, core_err_cnt,
560 m->addr >> PAGE_SHIFT, m->addr & ~PAGE_MASK, 0,
561 res->channel, res->dimm, -1,
565 int skx_mce_check_error(struct notifier_block *nb, unsigned long val,
568 struct mce *mce = (struct mce *)data;
569 struct decoded_addr res;
570 struct mem_ctl_info *mci;
573 if (edac_get_report_status() == EDAC_REPORTING_DISABLED)
576 /* ignore unless this is memory related with an address */
577 if ((mce->status & 0xefff) >> 7 != 1 || !(mce->status & MCI_STATUS_ADDRV))
580 memset(&res, 0, sizeof(res));
581 res.addr = mce->addr;
583 if (adxl_component_count) {
584 if (!skx_adxl_decode(&res))
587 mci = get_mci(res.socket, res.imc);
589 if (!skx_decode || !skx_decode(&res))
592 mci = res.dev->imc[res.imc].mci;
598 if (mce->mcgstatus & MCG_STATUS_MCIP)
603 skx_mc_printk(mci, KERN_DEBUG, "HANDLING MCE MEMORY ERROR\n");
605 skx_mc_printk(mci, KERN_DEBUG, "CPU %d: Machine Check %s: 0x%llx "
606 "Bank %d: 0x%llx\n", mce->extcpu, type,
607 mce->mcgstatus, mce->bank, mce->status);
608 skx_mc_printk(mci, KERN_DEBUG, "TSC 0x%llx ", mce->tsc);
609 skx_mc_printk(mci, KERN_DEBUG, "ADDR 0x%llx ", mce->addr);
610 skx_mc_printk(mci, KERN_DEBUG, "MISC 0x%llx ", mce->misc);
612 skx_mc_printk(mci, KERN_DEBUG, "PROCESSOR %u:0x%x TIME %llu SOCKET "
613 "%u APIC 0x%x\n", mce->cpuvendor, mce->cpuid,
614 mce->time, mce->socketid, mce->apicid);
616 skx_mce_output_error(mci, mce, &res);
621 void skx_remove(void)
624 struct skx_dev *d, *tmp;
628 list_for_each_entry_safe(d, tmp, &dev_edac_list, list) {
630 for (i = 0; i < NUM_IMC; i++) {
632 skx_unregister_mci(&d->imc[i]);
635 pci_dev_put(d->imc[i].mdev);
638 iounmap(d->imc[i].mbase);
640 for (j = 0; j < NUM_CHANNELS; j++) {
641 if (d->imc[i].chan[j].cdev)
642 pci_dev_put(d->imc[i].chan[j].cdev);
646 pci_dev_put(d->util_all);
648 pci_dev_put(d->sad_all);
650 pci_dev_put(d->uracu);