Merge tag 'io_uring-5.11-2021-01-16' of git://git.kernel.dk/linux-block
[linux/fpc-iii.git] / drivers / edac / skx_common.c
blob81c3e2ec6f568a0b295774cc6bfdb3722e1af775
1 // SPDX-License-Identifier: GPL-2.0
2 /*
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>
21 #include <asm/mce.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 skx_show_retry_log_f skx_show_retry_rd_err_log;
41 static u64 skx_tolm, skx_tohm;
42 static LIST_HEAD(dev_edac_list);
44 int __init skx_adxl_get(void)
46 const char * const *names;
47 int i, j;
49 names = adxl_get_component_names();
50 if (!names) {
51 skx_printk(KERN_NOTICE, "No firmware support for address translation.\n");
52 return -ENODEV;
55 for (i = 0; i < INDEX_MAX; i++) {
56 for (j = 0; names[j]; j++) {
57 if (!strcmp(component_names[i], names[j])) {
58 component_indices[i] = j;
59 break;
63 if (!names[j])
64 goto err;
67 adxl_component_names = names;
68 while (*names++)
69 adxl_component_count++;
71 adxl_values = kcalloc(adxl_component_count, sizeof(*adxl_values),
72 GFP_KERNEL);
73 if (!adxl_values) {
74 adxl_component_count = 0;
75 return -ENOMEM;
78 adxl_msg = kzalloc(MSG_SIZE, GFP_KERNEL);
79 if (!adxl_msg) {
80 adxl_component_count = 0;
81 kfree(adxl_values);
82 return -ENOMEM;
85 return 0;
86 err:
87 skx_printk(KERN_ERR, "'%s' is not matched from DSM parameters: ",
88 component_names[i]);
89 for (j = 0; names[j]; j++)
90 skx_printk(KERN_CONT, "%s ", names[j]);
91 skx_printk(KERN_CONT, "\n");
93 return -ENODEV;
96 void __exit skx_adxl_put(void)
98 kfree(adxl_values);
99 kfree(adxl_msg);
102 static bool skx_adxl_decode(struct decoded_addr *res)
104 struct skx_dev *d;
105 int i, len = 0;
107 if (res->addr >= skx_tohm || (res->addr >= skx_tolm &&
108 res->addr < BIT_ULL(32))) {
109 edac_dbg(0, "Address 0x%llx out of range\n", res->addr);
110 return false;
113 if (adxl_decode(res->addr, adxl_values)) {
114 edac_dbg(0, "Failed to decode 0x%llx\n", res->addr);
115 return false;
118 res->socket = (int)adxl_values[component_indices[INDEX_SOCKET]];
119 res->imc = (int)adxl_values[component_indices[INDEX_MEMCTRL]];
120 res->channel = (int)adxl_values[component_indices[INDEX_CHANNEL]];
121 res->dimm = (int)adxl_values[component_indices[INDEX_DIMM]];
123 if (res->imc > NUM_IMC - 1) {
124 skx_printk(KERN_ERR, "Bad imc %d\n", res->imc);
125 return false;
128 list_for_each_entry(d, &dev_edac_list, list) {
129 if (d->imc[0].src_id == res->socket) {
130 res->dev = d;
131 break;
135 if (!res->dev) {
136 skx_printk(KERN_ERR, "No device for src_id %d imc %d\n",
137 res->socket, res->imc);
138 return false;
141 for (i = 0; i < adxl_component_count; i++) {
142 if (adxl_values[i] == ~0x0ull)
143 continue;
145 len += snprintf(adxl_msg + len, MSG_SIZE - len, " %s:0x%llx",
146 adxl_component_names[i], adxl_values[i]);
147 if (MSG_SIZE - len <= 0)
148 break;
151 return true;
154 void skx_set_decode(skx_decode_f decode, skx_show_retry_log_f show_retry_log)
156 skx_decode = decode;
157 skx_show_retry_rd_err_log = show_retry_log;
160 int skx_get_src_id(struct skx_dev *d, int off, u8 *id)
162 u32 reg;
164 if (pci_read_config_dword(d->util_all, off, &reg)) {
165 skx_printk(KERN_ERR, "Failed to read src id\n");
166 return -ENODEV;
169 *id = GET_BITFIELD(reg, 12, 14);
170 return 0;
173 int skx_get_node_id(struct skx_dev *d, u8 *id)
175 u32 reg;
177 if (pci_read_config_dword(d->util_all, 0xf4, &reg)) {
178 skx_printk(KERN_ERR, "Failed to read node id\n");
179 return -ENODEV;
182 *id = GET_BITFIELD(reg, 0, 2);
183 return 0;
186 static int get_width(u32 mtr)
188 switch (GET_BITFIELD(mtr, 8, 9)) {
189 case 0:
190 return DEV_X4;
191 case 1:
192 return DEV_X8;
193 case 2:
194 return DEV_X16;
196 return DEV_UNKNOWN;
200 * We use the per-socket device @cfg->did to count how many sockets are present,
201 * and to detemine which PCI buses are associated with each socket. Allocate
202 * and build the full list of all the skx_dev structures that we need here.
204 int skx_get_all_bus_mappings(struct res_config *cfg, struct list_head **list)
206 struct pci_dev *pdev, *prev;
207 struct skx_dev *d;
208 u32 reg;
209 int ndev = 0;
211 prev = NULL;
212 for (;;) {
213 pdev = pci_get_device(PCI_VENDOR_ID_INTEL, cfg->decs_did, prev);
214 if (!pdev)
215 break;
216 ndev++;
217 d = kzalloc(sizeof(*d), GFP_KERNEL);
218 if (!d) {
219 pci_dev_put(pdev);
220 return -ENOMEM;
223 if (pci_read_config_dword(pdev, cfg->busno_cfg_offset, &reg)) {
224 kfree(d);
225 pci_dev_put(pdev);
226 skx_printk(KERN_ERR, "Failed to read bus idx\n");
227 return -ENODEV;
230 d->bus[0] = GET_BITFIELD(reg, 0, 7);
231 d->bus[1] = GET_BITFIELD(reg, 8, 15);
232 if (cfg->type == SKX) {
233 d->seg = pci_domain_nr(pdev->bus);
234 d->bus[2] = GET_BITFIELD(reg, 16, 23);
235 d->bus[3] = GET_BITFIELD(reg, 24, 31);
236 } else {
237 d->seg = GET_BITFIELD(reg, 16, 23);
240 edac_dbg(2, "busses: 0x%x, 0x%x, 0x%x, 0x%x\n",
241 d->bus[0], d->bus[1], d->bus[2], d->bus[3]);
242 list_add_tail(&d->list, &dev_edac_list);
243 prev = pdev;
246 if (list)
247 *list = &dev_edac_list;
248 return ndev;
251 int skx_get_hi_lo(unsigned int did, int off[], u64 *tolm, u64 *tohm)
253 struct pci_dev *pdev;
254 u32 reg;
256 pdev = pci_get_device(PCI_VENDOR_ID_INTEL, did, NULL);
257 if (!pdev) {
258 edac_dbg(2, "Can't get tolm/tohm\n");
259 return -ENODEV;
262 if (pci_read_config_dword(pdev, off[0], &reg)) {
263 skx_printk(KERN_ERR, "Failed to read tolm\n");
264 goto fail;
266 skx_tolm = reg;
268 if (pci_read_config_dword(pdev, off[1], &reg)) {
269 skx_printk(KERN_ERR, "Failed to read lower tohm\n");
270 goto fail;
272 skx_tohm = reg;
274 if (pci_read_config_dword(pdev, off[2], &reg)) {
275 skx_printk(KERN_ERR, "Failed to read upper tohm\n");
276 goto fail;
278 skx_tohm |= (u64)reg << 32;
280 pci_dev_put(pdev);
281 *tolm = skx_tolm;
282 *tohm = skx_tohm;
283 edac_dbg(2, "tolm = 0x%llx tohm = 0x%llx\n", skx_tolm, skx_tohm);
284 return 0;
285 fail:
286 pci_dev_put(pdev);
287 return -ENODEV;
290 static int skx_get_dimm_attr(u32 reg, int lobit, int hibit, int add,
291 int minval, int maxval, const char *name)
293 u32 val = GET_BITFIELD(reg, lobit, hibit);
295 if (val < minval || val > maxval) {
296 edac_dbg(2, "bad %s = %d (raw=0x%x)\n", name, val, reg);
297 return -EINVAL;
299 return val + add;
302 #define numrank(reg) skx_get_dimm_attr(reg, 12, 13, 0, 0, 2, "ranks")
303 #define numrow(reg) skx_get_dimm_attr(reg, 2, 4, 12, 1, 6, "rows")
304 #define numcol(reg) skx_get_dimm_attr(reg, 0, 1, 10, 0, 2, "cols")
306 int skx_get_dimm_info(u32 mtr, u32 mcmtr, u32 amap, struct dimm_info *dimm,
307 struct skx_imc *imc, int chan, int dimmno,
308 struct res_config *cfg)
310 int banks, ranks, rows, cols, npages;
311 enum mem_type mtype;
312 u64 size;
314 ranks = numrank(mtr);
315 rows = numrow(mtr);
316 cols = numcol(mtr);
318 if (cfg->support_ddr5 && (amap & 0x8)) {
319 banks = 32;
320 mtype = MEM_DDR5;
321 } else {
322 banks = 16;
323 mtype = MEM_DDR4;
327 * Compute size in 8-byte (2^3) words, then shift to MiB (2^20)
329 size = ((1ull << (rows + cols + ranks)) * banks) >> (20 - 3);
330 npages = MiB_TO_PAGES(size);
332 edac_dbg(0, "mc#%d: channel %d, dimm %d, %lld MiB (%d pages) bank: %d, rank: %d, row: 0x%x, col: 0x%x\n",
333 imc->mc, chan, dimmno, size, npages,
334 banks, 1 << ranks, rows, cols);
336 imc->chan[chan].dimms[dimmno].close_pg = GET_BITFIELD(mcmtr, 0, 0);
337 imc->chan[chan].dimms[dimmno].bank_xor_enable = GET_BITFIELD(mcmtr, 9, 9);
338 imc->chan[chan].dimms[dimmno].fine_grain_bank = GET_BITFIELD(amap, 0, 0);
339 imc->chan[chan].dimms[dimmno].rowbits = rows;
340 imc->chan[chan].dimms[dimmno].colbits = cols;
342 dimm->nr_pages = npages;
343 dimm->grain = 32;
344 dimm->dtype = get_width(mtr);
345 dimm->mtype = mtype;
346 dimm->edac_mode = EDAC_SECDED; /* likely better than this */
347 snprintf(dimm->label, sizeof(dimm->label), "CPU_SrcID#%u_MC#%u_Chan#%u_DIMM#%u",
348 imc->src_id, imc->lmc, chan, dimmno);
350 return 1;
353 int skx_get_nvdimm_info(struct dimm_info *dimm, struct skx_imc *imc,
354 int chan, int dimmno, const char *mod_str)
356 int smbios_handle;
357 u32 dev_handle;
358 u16 flags;
359 u64 size = 0;
361 dev_handle = ACPI_NFIT_BUILD_DEVICE_HANDLE(dimmno, chan, imc->lmc,
362 imc->src_id, 0);
364 smbios_handle = nfit_get_smbios_id(dev_handle, &flags);
365 if (smbios_handle == -EOPNOTSUPP) {
366 pr_warn_once("%s: Can't find size of NVDIMM. Try enabling CONFIG_ACPI_NFIT\n", mod_str);
367 goto unknown_size;
370 if (smbios_handle < 0) {
371 skx_printk(KERN_ERR, "Can't find handle for NVDIMM ADR=0x%x\n", dev_handle);
372 goto unknown_size;
375 if (flags & ACPI_NFIT_MEM_MAP_FAILED) {
376 skx_printk(KERN_ERR, "NVDIMM ADR=0x%x is not mapped\n", dev_handle);
377 goto unknown_size;
380 size = dmi_memdev_size(smbios_handle);
381 if (size == ~0ull)
382 skx_printk(KERN_ERR, "Can't find size for NVDIMM ADR=0x%x/SMBIOS=0x%x\n",
383 dev_handle, smbios_handle);
385 unknown_size:
386 dimm->nr_pages = size >> PAGE_SHIFT;
387 dimm->grain = 32;
388 dimm->dtype = DEV_UNKNOWN;
389 dimm->mtype = MEM_NVDIMM;
390 dimm->edac_mode = EDAC_SECDED; /* likely better than this */
392 edac_dbg(0, "mc#%d: channel %d, dimm %d, %llu MiB (%u pages)\n",
393 imc->mc, chan, dimmno, size >> 20, dimm->nr_pages);
395 snprintf(dimm->label, sizeof(dimm->label), "CPU_SrcID#%u_MC#%u_Chan#%u_DIMM#%u",
396 imc->src_id, imc->lmc, chan, dimmno);
398 return (size == 0 || size == ~0ull) ? 0 : 1;
401 int skx_register_mci(struct skx_imc *imc, struct pci_dev *pdev,
402 const char *ctl_name, const char *mod_str,
403 get_dimm_config_f get_dimm_config,
404 struct res_config *cfg)
406 struct mem_ctl_info *mci;
407 struct edac_mc_layer layers[2];
408 struct skx_pvt *pvt;
409 int rc;
411 /* Allocate a new MC control structure */
412 layers[0].type = EDAC_MC_LAYER_CHANNEL;
413 layers[0].size = NUM_CHANNELS;
414 layers[0].is_virt_csrow = false;
415 layers[1].type = EDAC_MC_LAYER_SLOT;
416 layers[1].size = NUM_DIMMS;
417 layers[1].is_virt_csrow = true;
418 mci = edac_mc_alloc(imc->mc, ARRAY_SIZE(layers), layers,
419 sizeof(struct skx_pvt));
421 if (unlikely(!mci))
422 return -ENOMEM;
424 edac_dbg(0, "MC#%d: mci = %p\n", imc->mc, mci);
426 /* Associate skx_dev and mci for future usage */
427 imc->mci = mci;
428 pvt = mci->pvt_info;
429 pvt->imc = imc;
431 mci->ctl_name = kasprintf(GFP_KERNEL, "%s#%d IMC#%d", ctl_name,
432 imc->node_id, imc->lmc);
433 if (!mci->ctl_name) {
434 rc = -ENOMEM;
435 goto fail0;
438 mci->mtype_cap = MEM_FLAG_DDR4 | MEM_FLAG_NVDIMM;
439 if (cfg->support_ddr5)
440 mci->mtype_cap |= MEM_FLAG_DDR5;
441 mci->edac_ctl_cap = EDAC_FLAG_NONE;
442 mci->edac_cap = EDAC_FLAG_NONE;
443 mci->mod_name = mod_str;
444 mci->dev_name = pci_name(pdev);
445 mci->ctl_page_to_phys = NULL;
447 rc = get_dimm_config(mci, cfg);
448 if (rc < 0)
449 goto fail;
451 /* Record ptr to the generic device */
452 mci->pdev = &pdev->dev;
454 /* Add this new MC control structure to EDAC's list of MCs */
455 if (unlikely(edac_mc_add_mc(mci))) {
456 edac_dbg(0, "MC: failed edac_mc_add_mc()\n");
457 rc = -EINVAL;
458 goto fail;
461 return 0;
463 fail:
464 kfree(mci->ctl_name);
465 fail0:
466 edac_mc_free(mci);
467 imc->mci = NULL;
468 return rc;
471 static void skx_unregister_mci(struct skx_imc *imc)
473 struct mem_ctl_info *mci = imc->mci;
475 if (!mci)
476 return;
478 edac_dbg(0, "MC%d: mci = %p\n", imc->mc, mci);
480 /* Remove MC sysfs nodes */
481 edac_mc_del_mc(mci->pdev);
483 edac_dbg(1, "%s: free mci struct\n", mci->ctl_name);
484 kfree(mci->ctl_name);
485 edac_mc_free(mci);
488 static void skx_mce_output_error(struct mem_ctl_info *mci,
489 const struct mce *m,
490 struct decoded_addr *res)
492 enum hw_event_mc_err_type tp_event;
493 char *optype;
494 bool ripv = GET_BITFIELD(m->mcgstatus, 0, 0);
495 bool overflow = GET_BITFIELD(m->status, 62, 62);
496 bool uncorrected_error = GET_BITFIELD(m->status, 61, 61);
497 bool recoverable;
498 int len;
499 u32 core_err_cnt = GET_BITFIELD(m->status, 38, 52);
500 u32 mscod = GET_BITFIELD(m->status, 16, 31);
501 u32 errcode = GET_BITFIELD(m->status, 0, 15);
502 u32 optypenum = GET_BITFIELD(m->status, 4, 6);
504 recoverable = GET_BITFIELD(m->status, 56, 56);
506 if (uncorrected_error) {
507 core_err_cnt = 1;
508 if (ripv) {
509 tp_event = HW_EVENT_ERR_UNCORRECTED;
510 } else {
511 tp_event = HW_EVENT_ERR_FATAL;
513 } else {
514 tp_event = HW_EVENT_ERR_CORRECTED;
518 * According to Intel Architecture spec vol 3B,
519 * Table 15-10 "IA32_MCi_Status [15:0] Compound Error Code Encoding"
520 * memory errors should fit one of these masks:
521 * 000f 0000 1mmm cccc (binary)
522 * 000f 0010 1mmm cccc (binary) [RAM used as cache]
523 * where:
524 * f = Correction Report Filtering Bit. If 1, subsequent errors
525 * won't be shown
526 * mmm = error type
527 * cccc = channel
528 * If the mask doesn't match, report an error to the parsing logic
530 if (!((errcode & 0xef80) == 0x80 || (errcode & 0xef80) == 0x280)) {
531 optype = "Can't parse: it is not a mem";
532 } else {
533 switch (optypenum) {
534 case 0:
535 optype = "generic undef request error";
536 break;
537 case 1:
538 optype = "memory read error";
539 break;
540 case 2:
541 optype = "memory write error";
542 break;
543 case 3:
544 optype = "addr/cmd error";
545 break;
546 case 4:
547 optype = "memory scrubbing error";
548 break;
549 default:
550 optype = "reserved";
551 break;
554 if (adxl_component_count) {
555 len = snprintf(skx_msg, MSG_SIZE, "%s%s err_code:0x%04x:0x%04x %s",
556 overflow ? " OVERFLOW" : "",
557 (uncorrected_error && recoverable) ? " recoverable" : "",
558 mscod, errcode, adxl_msg);
559 } else {
560 len = snprintf(skx_msg, MSG_SIZE,
561 "%s%s err_code:0x%04x:0x%04x socket:%d imc:%d rank:%d bg:%d ba:%d row:0x%x col:0x%x",
562 overflow ? " OVERFLOW" : "",
563 (uncorrected_error && recoverable) ? " recoverable" : "",
564 mscod, errcode,
565 res->socket, res->imc, res->rank,
566 res->bank_group, res->bank_address, res->row, res->column);
569 if (skx_show_retry_rd_err_log)
570 skx_show_retry_rd_err_log(res, skx_msg + len, MSG_SIZE - len);
572 edac_dbg(0, "%s\n", skx_msg);
574 /* Call the helper to output message */
575 edac_mc_handle_error(tp_event, mci, core_err_cnt,
576 m->addr >> PAGE_SHIFT, m->addr & ~PAGE_MASK, 0,
577 res->channel, res->dimm, -1,
578 optype, skx_msg);
581 int skx_mce_check_error(struct notifier_block *nb, unsigned long val,
582 void *data)
584 struct mce *mce = (struct mce *)data;
585 struct decoded_addr res;
586 struct mem_ctl_info *mci;
587 char *type;
589 if (mce->kflags & MCE_HANDLED_CEC)
590 return NOTIFY_DONE;
592 /* ignore unless this is memory related with an address */
593 if ((mce->status & 0xefff) >> 7 != 1 || !(mce->status & MCI_STATUS_ADDRV))
594 return NOTIFY_DONE;
596 memset(&res, 0, sizeof(res));
597 res.addr = mce->addr;
599 if (adxl_component_count) {
600 if (!skx_adxl_decode(&res))
601 return NOTIFY_DONE;
602 } else if (!skx_decode || !skx_decode(&res)) {
603 return NOTIFY_DONE;
606 mci = res.dev->imc[res.imc].mci;
608 if (!mci)
609 return NOTIFY_DONE;
611 if (mce->mcgstatus & MCG_STATUS_MCIP)
612 type = "Exception";
613 else
614 type = "Event";
616 skx_mc_printk(mci, KERN_DEBUG, "HANDLING MCE MEMORY ERROR\n");
618 skx_mc_printk(mci, KERN_DEBUG, "CPU %d: Machine Check %s: 0x%llx "
619 "Bank %d: 0x%llx\n", mce->extcpu, type,
620 mce->mcgstatus, mce->bank, mce->status);
621 skx_mc_printk(mci, KERN_DEBUG, "TSC 0x%llx ", mce->tsc);
622 skx_mc_printk(mci, KERN_DEBUG, "ADDR 0x%llx ", mce->addr);
623 skx_mc_printk(mci, KERN_DEBUG, "MISC 0x%llx ", mce->misc);
625 skx_mc_printk(mci, KERN_DEBUG, "PROCESSOR %u:0x%x TIME %llu SOCKET "
626 "%u APIC 0x%x\n", mce->cpuvendor, mce->cpuid,
627 mce->time, mce->socketid, mce->apicid);
629 skx_mce_output_error(mci, mce, &res);
631 mce->kflags |= MCE_HANDLED_EDAC;
632 return NOTIFY_DONE;
635 void skx_remove(void)
637 int i, j;
638 struct skx_dev *d, *tmp;
640 edac_dbg(0, "\n");
642 list_for_each_entry_safe(d, tmp, &dev_edac_list, list) {
643 list_del(&d->list);
644 for (i = 0; i < NUM_IMC; i++) {
645 if (d->imc[i].mci)
646 skx_unregister_mci(&d->imc[i]);
648 if (d->imc[i].mdev)
649 pci_dev_put(d->imc[i].mdev);
651 if (d->imc[i].mbase)
652 iounmap(d->imc[i].mbase);
654 for (j = 0; j < NUM_CHANNELS; j++) {
655 if (d->imc[i].chan[j].cdev)
656 pci_dev_put(d->imc[i].chan[j].cdev);
659 if (d->util_all)
660 pci_dev_put(d->util_all);
661 if (d->sad_all)
662 pci_dev_put(d->sad_all);
663 if (d->uracu)
664 pci_dev_put(d->uracu);
666 kfree(d);