projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
EDAC, thunderx: Remove suspend/resume support
[linux-2.6-block.git] / drivers / edac / skx_edac.c
CommitLineData
4ec656bd
TL		 1/*
2 * EDAC driver for Intel(R) Xeon(R) Skylake processors
3 * Copyright (c) 2016, Intel Corporation.
4 *
5 * This program is free software; you can redistribute it and/or modify it
6 * under the terms and conditions of the GNU General Public License,
7 * version 2, as published by the Free Software Foundation.
8 *
9 * This program is distributed in the hope it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
12 * more details.
13 */
14
15#include <linux/module.h>
16#include <linux/init.h>
17#include <linux/pci.h>
18#include <linux/pci_ids.h>
19#include <linux/slab.h>
20#include <linux/delay.h>
21#include <linux/edac.h>
22#include <linux/mmzone.h>
23#include <linux/smp.h>
24#include <linux/bitmap.h>
25#include <linux/math64.h>
26#include <linux/mod_devicetable.h>
27#include <asm/cpu_device_id.h>
20f4d692 28#include <asm/intel-family.h>
4ec656bd
TL		 29#include <asm/processor.h>
30#include <asm/mce.h>
31
78d88e8a 32#include "edac_module.h"
4ec656bd 33
301375e7
TK		 34#define EDAC_MOD_STR "skx_edac"
35
4ec656bd
TL		 36/*
37 * Debug macros
38 */
39#define skx_printk(level, fmt, arg...) \
40 edac_printk(level, "skx", fmt, ##arg)
41
42#define skx_mc_printk(mci, level, fmt, arg...) \
43 edac_mc_chipset_printk(mci, level, "skx", fmt, ##arg)
44
45/*
46 * Get a bit field at register value <v>, from bit <lo> to bit <hi>
47 */
48#define GET_BITFIELD(v, lo, hi) \
49 (((v) & GENMASK_ULL((hi), (lo))) >> (lo))
50
51static LIST_HEAD(skx_edac_list);
52
53static u64 skx_tolm, skx_tohm;
54
55#define NUM_IMC 2 /* memory controllers per socket */
56#define NUM_CHANNELS 3 /* channels per memory controller */
57#define NUM_DIMMS 2 /* Max DIMMS per channel */
58
59#define MASK26 0x3FFFFFF /* Mask for 2^26 */
60#define MASK29 0x1FFFFFFF /* Mask for 2^29 */
61
62/*
63 * Each cpu socket contains some pci devices that provide global
64 * information, and also some that are local to each of the two
65 * memory controllers on the die.
66 */
67struct skx_dev {
68 struct list_head list;
69 u8 bus[4];
70 struct pci_dev *sad_all;
71 struct pci_dev *util_all;
72 u32 mcroute;
73 struct skx_imc {
74 struct mem_ctl_info *mci;
75 u8 mc; /* system wide mc# */
76 u8 lmc; /* socket relative mc# */
77 u8 src_id, node_id;
78 struct skx_channel {
79 struct pci_dev *cdev;
80 struct skx_dimm {
81 u8 close_pg;
82 u8 bank_xor_enable;
83 u8 fine_grain_bank;
84 u8 rowbits;
85 u8 colbits;
86 } dimms[NUM_DIMMS];
87 } chan[NUM_CHANNELS];
88 } imc[NUM_IMC];
89};
90static int skx_num_sockets;
91
92struct skx_pvt {
93 struct skx_imc *imc;
94};
95
96struct decoded_addr {
97 struct skx_dev *dev;
98 u64 addr;
99 int socket;
100 int imc;
101 int channel;
102 u64 chan_addr;
103 int sktways;
104 int chanways;
105 int dimm;
106 int rank;
107 int channel_rank;
108 u64 rank_address;
109 int row;
110 int column;
111 int bank_address;
112 int bank_group;
113};
114
115static struct skx_dev *get_skx_dev(u8 bus, u8 idx)
116{
117 struct skx_dev *d;
118
119 list_for_each_entry(d, &skx_edac_list, list) {
120 if (d->bus[idx] == bus)
121 return d;
122 }
123
124 return NULL;
125}
126
127enum munittype {
128 CHAN0, CHAN1, CHAN2, SAD_ALL, UTIL_ALL, SAD
129};
130
131struct munit {
132 u16 did;
133 u16 devfn[NUM_IMC];
134 u8 busidx;
135 u8 per_socket;
136 enum munittype mtype;
137};
138
139/*
140 * List of PCI device ids that we need together with some device
141 * number and function numbers to tell which memory controller the
142 * device belongs to.
143 */
144static const struct munit skx_all_munits[] = {
145 { 0x2054, { }, 1, 1, SAD_ALL },
146 { 0x2055, { }, 1, 1, UTIL_ALL },
147 { 0x2040, { PCI_DEVFN(10, 0), PCI_DEVFN(12, 0) }, 2, 2, CHAN0 },
148 { 0x2044, { PCI_DEVFN(10, 4), PCI_DEVFN(12, 4) }, 2, 2, CHAN1 },
149 { 0x2048, { PCI_DEVFN(11, 0), PCI_DEVFN(13, 0) }, 2, 2, CHAN2 },
150 { 0x208e, { }, 1, 0, SAD },
151 { }
152};
153
154/*
155 * We use the per-socket device 0x2016 to count how many sockets are present,
156 * and to detemine which PCI buses are associated with each socket. Allocate
157 * and build the full list of all the skx_dev structures that we need here.
158 */
159static int get_all_bus_mappings(void)
160{
161 struct pci_dev *pdev, *prev;
162 struct skx_dev *d;
163 u32 reg;
164 int ndev = 0;
165
166 prev = NULL;
167 for (;;) {
168 pdev = pci_get_device(PCI_VENDOR_ID_INTEL, 0x2016, prev);
169 if (!pdev)
170 break;
171 ndev++;
172 d = kzalloc(sizeof(*d), GFP_KERNEL);
173 if (!d) {
174 pci_dev_put(pdev);
175 return -ENOMEM;
176 }
177 pci_read_config_dword(pdev, 0xCC, &reg);
178 d->bus[0] = GET_BITFIELD(reg, 0, 7);
179 d->bus[1] = GET_BITFIELD(reg, 8, 15);
180 d->bus[2] = GET_BITFIELD(reg, 16, 23);
181 d->bus[3] = GET_BITFIELD(reg, 24, 31);
182 edac_dbg(2, "busses: %x, %x, %x, %x\n",
183 d->bus[0], d->bus[1], d->bus[2], d->bus[3]);
184 list_add_tail(&d->list, &skx_edac_list);
185 skx_num_sockets++;
186 prev = pdev;
187 }
188
189 return ndev;
190}
191
192static int get_all_munits(const struct munit *m)
193{
194 struct pci_dev *pdev, *prev;
195 struct skx_dev *d;
196 u32 reg;
197 int i = 0, ndev = 0;
198
199 prev = NULL;
200 for (;;) {
201 pdev = pci_get_device(PCI_VENDOR_ID_INTEL, m->did, prev);
202 if (!pdev)
203 break;
204 ndev++;
205 if (m->per_socket == NUM_IMC) {
206 for (i = 0; i < NUM_IMC; i++)
207 if (m->devfn[i] == pdev->devfn)
208 break;
209 if (i == NUM_IMC)
210 goto fail;
211 }
212 d = get_skx_dev(pdev->bus->number, m->busidx);
213 if (!d)
214 goto fail;
215
216 /* Be sure that the device is enabled */
217 if (unlikely(pci_enable_device(pdev) < 0)) {
218 skx_printk(KERN_ERR,
219 "Couldn't enable %04x:%04x\n", PCI_VENDOR_ID_INTEL, m->did);
220 goto fail;
221 }
222
223 switch (m->mtype) {
224 case CHAN0: case CHAN1: case CHAN2:
225 pci_dev_get(pdev);
226 d->imc[i].chan[m->mtype].cdev = pdev;
227 break;
228 case SAD_ALL:
229 pci_dev_get(pdev);
230 d->sad_all = pdev;
231 break;
232 case UTIL_ALL:
233 pci_dev_get(pdev);
234 d->util_all = pdev;
235 break;
236 case SAD:
237 /*
238 * one of these devices per core, including cores
239 * that don't exist on this SKU. Ignore any that
240 * read a route table of zero, make sure all the
241 * non-zero values match.
242 */
243 pci_read_config_dword(pdev, 0xB4, &reg);
244 if (reg != 0) {
245 if (d->mcroute == 0)
246 d->mcroute = reg;
247 else if (d->mcroute != reg) {
248 skx_printk(KERN_ERR,
249 "mcroute mismatch\n");
250 goto fail;
251 }
252 }
253 ndev--;
254 break;
255 }
256
257 prev = pdev;
258 }
259
260 return ndev;
261fail:
262 pci_dev_put(pdev);
263 return -ENODEV;
264}
265
240ea921 266static const struct x86_cpu_id skx_cpuids[] = {
20f4d692 267 { X86_VENDOR_INTEL, 6, INTEL_FAM6_SKYLAKE_X, 0, 0 },
4ec656bd
TL		 268 { }
269};
270MODULE_DEVICE_TABLE(x86cpu, skx_cpuids);
271
272static u8 get_src_id(struct skx_dev *d)
273{
274 u32 reg;
275
276 pci_read_config_dword(d->util_all, 0xF0, &reg);
277
278 return GET_BITFIELD(reg, 12, 14);
279}
280
281static u8 skx_get_node_id(struct skx_dev *d)
282{
283 u32 reg;
284
285 pci_read_config_dword(d->util_all, 0xF4, &reg);
286
287 return GET_BITFIELD(reg, 0, 2);
288}
289
290static int get_dimm_attr(u32 reg, int lobit, int hibit, int add, int minval,
291 int maxval, char *name)
292{
293 u32 val = GET_BITFIELD(reg, lobit, hibit);
294
295 if (val < minval || val > maxval) {
296 edac_dbg(2, "bad %s = %d (raw=%x)\n", name, val, reg);
297 return -EINVAL;
298 }
299 return val + add;
300}
301
302#define IS_DIMM_PRESENT(mtr) GET_BITFIELD((mtr), 15, 15)
303
a9c0a108 304#define numrank(reg) get_dimm_attr((reg), 12, 13, 0, 0, 2, "ranks")
4ec656bd
TL		 305#define numrow(reg) get_dimm_attr((reg), 2, 4, 12, 1, 6, "rows")
306#define numcol(reg) get_dimm_attr((reg), 0, 1, 10, 0, 2, "cols")
307
308static int get_width(u32 mtr)
309{
310 switch (GET_BITFIELD(mtr, 8, 9)) {
311 case 0:
312 return DEV_X4;
313 case 1:
314 return DEV_X8;
315 case 2:
316 return DEV_X16;
317 }
318 return DEV_UNKNOWN;
319}
320
321static int skx_get_hi_lo(void)
322{
323 struct pci_dev *pdev;
324 u32 reg;
325
326 pdev = pci_get_device(PCI_VENDOR_ID_INTEL, 0x2034, NULL);
327 if (!pdev) {
328 edac_dbg(0, "Can't get tolm/tohm\n");
329 return -ENODEV;
330 }
331
332 pci_read_config_dword(pdev, 0xD0, &reg);
333 skx_tolm = reg;
334 pci_read_config_dword(pdev, 0xD4, &reg);
335 skx_tohm = reg;
336 pci_read_config_dword(pdev, 0xD8, &reg);
337 skx_tohm |= (u64)reg << 32;
338
339 pci_dev_put(pdev);
340 edac_dbg(2, "tolm=%llx tohm=%llx\n", skx_tolm, skx_tohm);
341
342 return 0;
343}
344
345static int get_dimm_info(u32 mtr, u32 amap, struct dimm_info *dimm,
346 struct skx_imc *imc, int chan, int dimmno)
347{
348 int banks = 16, ranks, rows, cols, npages;
349 u64 size;
350
351 if (!IS_DIMM_PRESENT(mtr))
352 return 0;
353 ranks = numrank(mtr);
354 rows = numrow(mtr);
355 cols = numcol(mtr);
356
357 /*
358 * Compute size in 8-byte (2^3) words, then shift to MiB (2^20)
359 */
360 size = ((1ull << (rows + cols + ranks)) * banks) >> (20 - 3);
361 npages = MiB_TO_PAGES(size);
362
363 edac_dbg(0, "mc#%d: channel %d, dimm %d, %lld Mb (%d pages) bank: %d, rank: %d, row: %#x, col: %#x\n",
364 imc->mc, chan, dimmno, size, npages,
a9c0a108 365 banks, 1 << ranks, rows, cols);
4ec656bd
TL		 366
367 imc->chan[chan].dimms[dimmno].close_pg = GET_BITFIELD(mtr, 0, 0);
368 imc->chan[chan].dimms[dimmno].bank_xor_enable = GET_BITFIELD(mtr, 9, 9);
369 imc->chan[chan].dimms[dimmno].fine_grain_bank = GET_BITFIELD(amap, 0, 0);
370 imc->chan[chan].dimms[dimmno].rowbits = rows;
371 imc->chan[chan].dimms[dimmno].colbits = cols;
372
373 dimm->nr_pages = npages;
374 dimm->grain = 32;
375 dimm->dtype = get_width(mtr);
376 dimm->mtype = MEM_DDR4;
377 dimm->edac_mode = EDAC_SECDED; /* likely better than this */
378 snprintf(dimm->label, sizeof(dimm->label), "CPU_SrcID#%u_MC#%u_Chan#%u_DIMM#%u",
379 imc->src_id, imc->lmc, chan, dimmno);
380
381 return 1;
382}
383
384#define SKX_GET_MTMTR(dev, reg) \
385 pci_read_config_dword((dev), 0x87c, &reg)
386
387static bool skx_check_ecc(struct pci_dev *pdev)
388{
389 u32 mtmtr;
390
391 SKX_GET_MTMTR(pdev, mtmtr);
392
393 return !!GET_BITFIELD(mtmtr, 2, 2);
394}
395
396static int skx_get_dimm_config(struct mem_ctl_info *mci)
397{
398 struct skx_pvt *pvt = mci->pvt_info;
399 struct skx_imc *imc = pvt->imc;
400 struct dimm_info *dimm;
401 int i, j;
402 u32 mtr, amap;
403 int ndimms;
404
405 for (i = 0; i < NUM_CHANNELS; i++) {
406 ndimms = 0;
407 pci_read_config_dword(imc->chan[i].cdev, 0x8C, &amap);
408 for (j = 0; j < NUM_DIMMS; j++) {
409 dimm = EDAC_DIMM_PTR(mci->layers, mci->dimms,
410 mci->n_layers, i, j, 0);
411 pci_read_config_dword(imc->chan[i].cdev,
412 0x80 + 4*j, &mtr);
413 ndimms += get_dimm_info(mtr, amap, dimm, imc, i, j);
414 }
415 if (ndimms && !skx_check_ecc(imc->chan[0].cdev)) {
416 skx_printk(KERN_ERR, "ECC is disabled on imc %d\n", imc->mc);
417 return -ENODEV;
418 }
419 }
420
421 return 0;
422}
423
424static void skx_unregister_mci(struct skx_imc *imc)
425{
426 struct mem_ctl_info *mci = imc->mci;
427
428 if (!mci)
429 return;
430
431 edac_dbg(0, "MC%d: mci = %p\n", imc->mc, mci);
432
433 /* Remove MC sysfs nodes */
434 edac_mc_del_mc(mci->pdev);
435
436 edac_dbg(1, "%s: free mci struct\n", mci->ctl_name);
437 kfree(mci->ctl_name);
438 edac_mc_free(mci);
439}
440
441static int skx_register_mci(struct skx_imc *imc)
442{
443 struct mem_ctl_info *mci;
444 struct edac_mc_layer layers[2];
445 struct pci_dev *pdev = imc->chan[0].cdev;
446 struct skx_pvt *pvt;
447 int rc;
448
449 /* allocate a new MC control structure */
450 layers[0].type = EDAC_MC_LAYER_CHANNEL;
451 layers[0].size = NUM_CHANNELS;
452 layers[0].is_virt_csrow = false;
453 layers[1].type = EDAC_MC_LAYER_SLOT;
454 layers[1].size = NUM_DIMMS;
455 layers[1].is_virt_csrow = true;
456 mci = edac_mc_alloc(imc->mc, ARRAY_SIZE(layers), layers,
457 sizeof(struct skx_pvt));
458
459 if (unlikely(!mci))
460 return -ENOMEM;
461
462 edac_dbg(0, "MC#%d: mci = %p\n", imc->mc, mci);
463
464 /* Associate skx_dev and mci for future usage */
465 imc->mci = mci;
466 pvt = mci->pvt_info;
467 pvt->imc = imc;
468
75f029c3
AY		 469 mci->ctl_name = kasprintf(GFP_KERNEL, "Skylake Socket#%d IMC#%d", imc->node_id, imc->lmc);
470 if (!mci->ctl_name) {
471 rc = -ENOMEM;
472 goto fail0;
473 }
474
4ec656bd
TL		 475 mci->mtype_cap = MEM_FLAG_DDR4;
476 mci->edac_ctl_cap = EDAC_FLAG_NONE;
477 mci->edac_cap = EDAC_FLAG_NONE;
301375e7 478 mci->mod_name = EDAC_MOD_STR;
4ec656bd 479 mci->dev_name = pci_name(imc->chan[0].cdev);
4ec656bd
TL		 480 mci->ctl_page_to_phys = NULL;
481
482 rc = skx_get_dimm_config(mci);
483 if (rc < 0)
484 goto fail;
485
486 /* record ptr to the generic device */
487 mci->pdev = &pdev->dev;
488
489 /* add this new MC control structure to EDAC's list of MCs */
490 if (unlikely(edac_mc_add_mc(mci))) {
491 edac_dbg(0, "MC: failed edac_mc_add_mc()\n");
492 rc = -EINVAL;
493 goto fail;
494 }
495
496 return 0;
497
498fail:
499 kfree(mci->ctl_name);
75f029c3 500fail0:
4ec656bd
TL		 501 edac_mc_free(mci);
502 imc->mci = NULL;
503 return rc;
504}
505
506#define SKX_MAX_SAD 24
507
508#define SKX_GET_SAD(d, i, reg) \
509 pci_read_config_dword((d)->sad_all, 0x60 + 8 * (i), &reg)
510#define SKX_GET_ILV(d, i, reg) \
511 pci_read_config_dword((d)->sad_all, 0x64 + 8 * (i), &reg)
512
513#define SKX_SAD_MOD3MODE(sad) GET_BITFIELD((sad), 30, 31)
514#define SKX_SAD_MOD3(sad) GET_BITFIELD((sad), 27, 27)
515#define SKX_SAD_LIMIT(sad) (((u64)GET_BITFIELD((sad), 7, 26) << 26) | MASK26)
516#define SKX_SAD_MOD3ASMOD2(sad) GET_BITFIELD((sad), 5, 6)
517#define SKX_SAD_ATTR(sad) GET_BITFIELD((sad), 3, 4)
518#define SKX_SAD_INTERLEAVE(sad) GET_BITFIELD((sad), 1, 2)
519#define SKX_SAD_ENABLE(sad) GET_BITFIELD((sad), 0, 0)
520
521#define SKX_ILV_REMOTE(tgt) (((tgt) & 8) == 0)
522#define SKX_ILV_TARGET(tgt) ((tgt) & 7)
523
524static bool skx_sad_decode(struct decoded_addr *res)
525{
526 struct skx_dev *d = list_first_entry(&skx_edac_list, typeof(*d), list);
527 u64 addr = res->addr;
528 int i, idx, tgt, lchan, shift;
529 u32 sad, ilv;
530 u64 limit, prev_limit;
531 int remote = 0;
532
533 /* Simple sanity check for I/O space or out of range */
534 if (addr >= skx_tohm || (addr >= skx_tolm && addr < BIT_ULL(32))) {
535 edac_dbg(0, "Address %llx out of range\n", addr);
536 return false;
537 }
538
539restart:
540 prev_limit = 0;
541 for (i = 0; i < SKX_MAX_SAD; i++) {
542 SKX_GET_SAD(d, i, sad);
543 limit = SKX_SAD_LIMIT(sad);
544 if (SKX_SAD_ENABLE(sad)) {
545 if (addr >= prev_limit && addr <= limit)
546 goto sad_found;
547 }
548 prev_limit = limit + 1;
549 }
550 edac_dbg(0, "No SAD entry for %llx\n", addr);
551 return false;
552
553sad_found:
554 SKX_GET_ILV(d, i, ilv);
555
556 switch (SKX_SAD_INTERLEAVE(sad)) {
557 case 0:
558 idx = GET_BITFIELD(addr, 6, 8);
559 break;
560 case 1:
561 idx = GET_BITFIELD(addr, 8, 10);
562 break;
563 case 2:
564 idx = GET_BITFIELD(addr, 12, 14);
565 break;
566 case 3:
567 idx = GET_BITFIELD(addr, 30, 32);
568 break;
569 }
570
571 tgt = GET_BITFIELD(ilv, 4 * idx, 4 * idx + 3);
572
573 /* If point to another node, find it and start over */
574 if (SKX_ILV_REMOTE(tgt)) {
575 if (remote) {
576 edac_dbg(0, "Double remote!\n");
577 return false;
578 }
579 remote = 1;
580 list_for_each_entry(d, &skx_edac_list, list) {
581 if (d->imc[0].src_id == SKX_ILV_TARGET(tgt))
582 goto restart;
583 }
584 edac_dbg(0, "Can't find node %d\n", SKX_ILV_TARGET(tgt));
585 return false;
586 }
587
588 if (SKX_SAD_MOD3(sad) == 0)
589 lchan = SKX_ILV_TARGET(tgt);
590 else {
591 switch (SKX_SAD_MOD3MODE(sad)) {
592 case 0:
593 shift = 6;
594 break;
595 case 1:
596 shift = 8;
597 break;
598 case 2:
599 shift = 12;
600 break;
601 default:
602 edac_dbg(0, "illegal mod3mode\n");
603 return false;
604 }
605 switch (SKX_SAD_MOD3ASMOD2(sad)) {
606 case 0:
607 lchan = (addr >> shift) % 3;
608 break;
609 case 1:
610 lchan = (addr >> shift) % 2;
611 break;
612 case 2:
613 lchan = (addr >> shift) % 2;
614 lchan = (lchan << 1) | ~lchan;
615 break;
616 case 3:
617 lchan = ((addr >> shift) % 2) << 1;
618 break;
619 }
620 lchan = (lchan << 1) | (SKX_ILV_TARGET(tgt) & 1);
621 }
622
623 res->dev = d;
624 res->socket = d->imc[0].src_id;
625 res->imc = GET_BITFIELD(d->mcroute, lchan * 3, lchan * 3 + 2);
626 res->channel = GET_BITFIELD(d->mcroute, lchan * 2 + 18, lchan * 2 + 19);
627
628 edac_dbg(2, "%llx: socket=%d imc=%d channel=%d\n",
629 res->addr, res->socket, res->imc, res->channel);
630 return true;
631}
632
633#define SKX_MAX_TAD 8
634
635#define SKX_GET_TADBASE(d, mc, i, reg) \
636 pci_read_config_dword((d)->imc[mc].chan[0].cdev, 0x850 + 4 * (i), &reg)
637#define SKX_GET_TADWAYNESS(d, mc, i, reg) \
638 pci_read_config_dword((d)->imc[mc].chan[0].cdev, 0x880 + 4 * (i), &reg)
639#define SKX_GET_TADCHNILVOFFSET(d, mc, ch, i, reg) \
640 pci_read_config_dword((d)->imc[mc].chan[ch].cdev, 0x90 + 4 * (i), &reg)
641
642#define SKX_TAD_BASE(b) ((u64)GET_BITFIELD((b), 12, 31) << 26)
643#define SKX_TAD_SKT_GRAN(b) GET_BITFIELD((b), 4, 5)
644#define SKX_TAD_CHN_GRAN(b) GET_BITFIELD((b), 6, 7)
645#define SKX_TAD_LIMIT(b) (((u64)GET_BITFIELD((b), 12, 31) << 26) | MASK26)
646#define SKX_TAD_OFFSET(b) ((u64)GET_BITFIELD((b), 4, 23) << 26)
647#define SKX_TAD_SKTWAYS(b) (1 << GET_BITFIELD((b), 10, 11))
648#define SKX_TAD_CHNWAYS(b) (GET_BITFIELD((b), 8, 9) + 1)
649
650/* which bit used for both socket and channel interleave */
651static int skx_granularity[] = { 6, 8, 12, 30 };
652
653static u64 skx_do_interleave(u64 addr, int shift, int ways, u64 lowbits)
654{
655 addr >>= shift;
656 addr /= ways;
657 addr <<= shift;
658
659 return addr | (lowbits & ((1ull << shift) - 1));
660}
661
662static bool skx_tad_decode(struct decoded_addr *res)
663{
664 int i;
665 u32 base, wayness, chnilvoffset;
666 int skt_interleave_bit, chn_interleave_bit;
667 u64 channel_addr;
668
669 for (i = 0; i < SKX_MAX_TAD; i++) {
670 SKX_GET_TADBASE(res->dev, res->imc, i, base);
671 SKX_GET_TADWAYNESS(res->dev, res->imc, i, wayness);
672 if (SKX_TAD_BASE(base) <= res->addr && res->addr <= SKX_TAD_LIMIT(wayness))
673 goto tad_found;
674 }
675 edac_dbg(0, "No TAD entry for %llx\n", res->addr);
676 return false;
677
678tad_found:
679 res->sktways = SKX_TAD_SKTWAYS(wayness);
680 res->chanways = SKX_TAD_CHNWAYS(wayness);
681 skt_interleave_bit = skx_granularity[SKX_TAD_SKT_GRAN(base)];
682 chn_interleave_bit = skx_granularity[SKX_TAD_CHN_GRAN(base)];
683
684 SKX_GET_TADCHNILVOFFSET(res->dev, res->imc, res->channel, i, chnilvoffset);
685 channel_addr = res->addr - SKX_TAD_OFFSET(chnilvoffset);
686
687 if (res->chanways == 3 && skt_interleave_bit > chn_interleave_bit) {
688 /* Must handle channel first, then socket */
689 channel_addr = skx_do_interleave(channel_addr, chn_interleave_bit,
690 res->chanways, channel_addr);
691 channel_addr = skx_do_interleave(channel_addr, skt_interleave_bit,
692 res->sktways, channel_addr);
693 } else {
694 /* Handle socket then channel. Preserve low bits from original address */
695 channel_addr = skx_do_interleave(channel_addr, skt_interleave_bit,
696 res->sktways, res->addr);
697 channel_addr = skx_do_interleave(channel_addr, chn_interleave_bit,
698 res->chanways, res->addr);
699 }
700
701 res->chan_addr = channel_addr;
702
703 edac_dbg(2, "%llx: chan_addr=%llx sktways=%d chanways=%d\n",
704 res->addr, res->chan_addr, res->sktways, res->chanways);
705 return true;
706}
707
708#define SKX_MAX_RIR 4
709
710#define SKX_GET_RIRWAYNESS(d, mc, ch, i, reg) \
711 pci_read_config_dword((d)->imc[mc].chan[ch].cdev, \
712 0x108 + 4 * (i), &reg)
713#define SKX_GET_RIRILV(d, mc, ch, idx, i, reg) \
714 pci_read_config_dword((d)->imc[mc].chan[ch].cdev, \
715 0x120 + 16 * idx + 4 * (i), &reg)
716
717#define SKX_RIR_VALID(b) GET_BITFIELD((b), 31, 31)
718#define SKX_RIR_LIMIT(b) (((u64)GET_BITFIELD((b), 1, 11) << 29) | MASK29)
719#define SKX_RIR_WAYS(b) (1 << GET_BITFIELD((b), 28, 29))
720#define SKX_RIR_CHAN_RANK(b) GET_BITFIELD((b), 16, 19)
721#define SKX_RIR_OFFSET(b) ((u64)(GET_BITFIELD((b), 2, 15) << 26))
722
723static bool skx_rir_decode(struct decoded_addr *res)
724{
725 int i, idx, chan_rank;
726 int shift;
727 u32 rirway, rirlv;
728 u64 rank_addr, prev_limit = 0, limit;
729
730 if (res->dev->imc[res->imc].chan[res->channel].dimms[0].close_pg)
731 shift = 6;
732 else
733 shift = 13;
734
735 for (i = 0; i < SKX_MAX_RIR; i++) {
736 SKX_GET_RIRWAYNESS(res->dev, res->imc, res->channel, i, rirway);
737 limit = SKX_RIR_LIMIT(rirway);
738 if (SKX_RIR_VALID(rirway)) {
739 if (prev_limit <= res->chan_addr &&
740 res->chan_addr <= limit)
741 goto rir_found;
742 }
743 prev_limit = limit;
744 }
745 edac_dbg(0, "No RIR entry for %llx\n", res->addr);
746 return false;
747
748rir_found:
749 rank_addr = res->chan_addr >> shift;
750 rank_addr /= SKX_RIR_WAYS(rirway);
751 rank_addr <<= shift;
752 rank_addr |= res->chan_addr & GENMASK_ULL(shift - 1, 0);
753
754 res->rank_address = rank_addr;
755 idx = (res->chan_addr >> shift) % SKX_RIR_WAYS(rirway);
756
757 SKX_GET_RIRILV(res->dev, res->imc, res->channel, idx, i, rirlv);
758 res->rank_address = rank_addr - SKX_RIR_OFFSET(rirlv);
759 chan_rank = SKX_RIR_CHAN_RANK(rirlv);
760 res->channel_rank = chan_rank;
761 res->dimm = chan_rank / 4;
762 res->rank = chan_rank % 4;
763
764 edac_dbg(2, "%llx: dimm=%d rank=%d chan_rank=%d rank_addr=%llx\n",
765 res->addr, res->dimm, res->rank,
766 res->channel_rank, res->rank_address);
767 return true;
768}
769
770static u8 skx_close_row[] = {
771 15, 16, 17, 18, 20, 21, 22, 28, 10, 11, 12, 13, 29, 30, 31, 32, 33
772};
773static u8 skx_close_column[] = {
774 3, 4, 5, 14, 19, 23, 24, 25, 26, 27
775};
776static u8 skx_open_row[] = {
777 14, 15, 16, 20, 28, 21, 22, 23, 24, 25, 26, 27, 29, 30, 31, 32, 33
778};
779static u8 skx_open_column[] = {
780 3, 4, 5, 6, 7, 8, 9, 10, 11, 12
781};
782static u8 skx_open_fine_column[] = {
783 3, 4, 5, 7, 8, 9, 10, 11, 12, 13
784};
785
786static int skx_bits(u64 addr, int nbits, u8 *bits)
787{
788 int i, res = 0;
789
790 for (i = 0; i < nbits; i++)
791 res |= ((addr >> bits[i]) & 1) << i;
792 return res;
793}
794
795static int skx_bank_bits(u64 addr, int b0, int b1, int do_xor, int x0, int x1)
796{
797 int ret = GET_BITFIELD(addr, b0, b0) | (GET_BITFIELD(addr, b1, b1) << 1);
798
799 if (do_xor)
800 ret ^= GET_BITFIELD(addr, x0, x0) | (GET_BITFIELD(addr, x1, x1) << 1);
801
802 return ret;
803}
804
805static bool skx_mad_decode(struct decoded_addr *r)
806{
807 struct skx_dimm *dimm = &r->dev->imc[r->imc].chan[r->channel].dimms[r->dimm];
808 int bg0 = dimm->fine_grain_bank ? 6 : 13;
809
810 if (dimm->close_pg) {
811 r->row = skx_bits(r->rank_address, dimm->rowbits, skx_close_row);
812 r->column = skx_bits(r->rank_address, dimm->colbits, skx_close_column);
813 r->column |= 0x400; /* C10 is autoprecharge, always set */
814 r->bank_address = skx_bank_bits(r->rank_address, 8, 9, dimm->bank_xor_enable, 22, 28);
815 r->bank_group = skx_bank_bits(r->rank_address, 6, 7, dimm->bank_xor_enable, 20, 21);
816 } else {
817 r->row = skx_bits(r->rank_address, dimm->rowbits, skx_open_row);
818 if (dimm->fine_grain_bank)
819 r->column = skx_bits(r->rank_address, dimm->colbits, skx_open_fine_column);
820 else
821 r->column = skx_bits(r->rank_address, dimm->colbits, skx_open_column);
822 r->bank_address = skx_bank_bits(r->rank_address, 18, 19, dimm->bank_xor_enable, 22, 23);
823 r->bank_group = skx_bank_bits(r->rank_address, bg0, 17, dimm->bank_xor_enable, 20, 21);
824 }
825 r->row &= (1u << dimm->rowbits) - 1;
826
827 edac_dbg(2, "%llx: row=%x col=%x bank_addr=%d bank_group=%d\n",
828 r->addr, r->row, r->column, r->bank_address,
829 r->bank_group);
830 return true;
831}
832
833static bool skx_decode(struct decoded_addr *res)
834{
835
836 return skx_sad_decode(res) && skx_tad_decode(res) &&
837 skx_rir_decode(res) && skx_mad_decode(res);
838}
839
840#ifdef CONFIG_EDAC_DEBUG
841/*
842 * Debug feature. Make /sys/kernel/debug/skx_edac_test/addr.
843 * Write an address to this file to exercise the address decode
844 * logic in this driver.
845 */
846static struct dentry *skx_test;
847static u64 skx_fake_addr;
848
849static int debugfs_u64_set(void *data, u64 val)
850{
851 struct decoded_addr res;
852
853 res.addr = val;
854 skx_decode(&res);
855
856 return 0;
857}
858
859DEFINE_SIMPLE_ATTRIBUTE(fops_u64_wo, NULL, debugfs_u64_set, "%llu\n");
860
861static struct dentry *mydebugfs_create(const char *name, umode_t mode,
862 struct dentry *parent, u64 *value)
863{
864 return debugfs_create_file(name, mode, parent, value, &fops_u64_wo);
865}
866
867static void setup_skx_debug(void)
868{
869 skx_test = debugfs_create_dir("skx_edac_test", NULL);
870 mydebugfs_create("addr", S_IWUSR, skx_test, &skx_fake_addr);
871}
872
873static void teardown_skx_debug(void)
874{
875 debugfs_remove_recursive(skx_test);
876}
877#else
878static void setup_skx_debug(void)
879{
880}
881
882static void teardown_skx_debug(void)
883{
884}
885#endif /*CONFIG_EDAC_DEBUG*/
886
887static void skx_mce_output_error(struct mem_ctl_info *mci,
888 const struct mce *m,
889 struct decoded_addr *res)
890{
891 enum hw_event_mc_err_type tp_event;
892 char *type, *optype, msg[256];
893 bool ripv = GET_BITFIELD(m->mcgstatus, 0, 0);
894 bool overflow = GET_BITFIELD(m->status, 62, 62);
895 bool uncorrected_error = GET_BITFIELD(m->status, 61, 61);
896 bool recoverable;
897 u32 core_err_cnt = GET_BITFIELD(m->status, 38, 52);
898 u32 mscod = GET_BITFIELD(m->status, 16, 31);
899 u32 errcode = GET_BITFIELD(m->status, 0, 15);
900 u32 optypenum = GET_BITFIELD(m->status, 4, 6);
901
902 recoverable = GET_BITFIELD(m->status, 56, 56);
903
904 if (uncorrected_error) {
905 if (ripv) {
906 type = "FATAL";
907 tp_event = HW_EVENT_ERR_FATAL;
908 } else {
909 type = "NON_FATAL";
910 tp_event = HW_EVENT_ERR_UNCORRECTED;
911 }
912 } else {
913 type = "CORRECTED";
914 tp_event = HW_EVENT_ERR_CORRECTED;
915 }
916
917 /*
918 * According with Table 15-9 of the Intel Architecture spec vol 3A,
919 * memory errors should fit in this mask:
920 * 000f 0000 1mmm cccc (binary)
921 * where:
922 * f = Correction Report Filtering Bit. If 1, subsequent errors
923 * won't be shown
924 * mmm = error type
925 * cccc = channel
926 * If the mask doesn't match, report an error to the parsing logic
927 */
928 if (!((errcode & 0xef80) == 0x80)) {
929 optype = "Can't parse: it is not a mem";
930 } else {
931 switch (optypenum) {
932 case 0:
933 optype = "generic undef request error";
934 break;
935 case 1:
936 optype = "memory read error";
937 break;
938 case 2:
939 optype = "memory write error";
940 break;
941 case 3:
942 optype = "addr/cmd error";
943 break;
944 case 4:
945 optype = "memory scrubbing error";
946 break;
947 default:
948 optype = "reserved";
949 break;
950 }
951 }
952
953 snprintf(msg, sizeof(msg),
954 "%s%s err_code:%04x:%04x socket:%d imc:%d rank:%d bg:%d ba:%d row:%x col:%x",
955 overflow ? " OVERFLOW" : "",
956 (uncorrected_error && recoverable) ? " recoverable" : "",
957 mscod, errcode,
958 res->socket, res->imc, res->rank,
959 res->bank_group, res->bank_address, res->row, res->column);
960
961 edac_dbg(0, "%s\n", msg);
962
963 /* Call the helper to output message */
964 edac_mc_handle_error(tp_event, mci, core_err_cnt,
965 m->addr >> PAGE_SHIFT, m->addr & ~PAGE_MASK, 0,
966 res->channel, res->dimm, -1,
967 optype, msg);
968}
969
970static int skx_mce_check_error(struct notifier_block *nb, unsigned long val,
971 void *data)
972{
973 struct mce *mce = (struct mce *)data;
974 struct decoded_addr res;
975 struct mem_ctl_info *mci;
976 char *type;
977
bffc7dec 978 if (edac_get_report_status() == EDAC_REPORTING_DISABLED)
4ec656bd
TL		 979 return NOTIFY_DONE;
980
981 /* ignore unless this is memory related with an address */
982 if ((mce->status & 0xefff) >> 7 != 1 || !(mce->status & MCI_STATUS_ADDRV))
983 return NOTIFY_DONE;
984
985 res.addr = mce->addr;
986 if (!skx_decode(&res))
987 return NOTIFY_DONE;
988 mci = res.dev->imc[res.imc].mci;
989
990 if (mce->mcgstatus & MCG_STATUS_MCIP)
991 type = "Exception";
992 else
993 type = "Event";
994
995 skx_mc_printk(mci, KERN_DEBUG, "HANDLING MCE MEMORY ERROR\n");
996
997 skx_mc_printk(mci, KERN_DEBUG, "CPU %d: Machine Check %s: %Lx "
998 "Bank %d: %016Lx\n", mce->extcpu, type,
999 mce->mcgstatus, mce->bank, mce->status);
1000 skx_mc_printk(mci, KERN_DEBUG, "TSC %llx ", mce->tsc);
1001 skx_mc_printk(mci, KERN_DEBUG, "ADDR %llx ", mce->addr);
1002 skx_mc_printk(mci, KERN_DEBUG, "MISC %llx ", mce->misc);
1003
1004 skx_mc_printk(mci, KERN_DEBUG, "PROCESSOR %u:%x TIME %llu SOCKET "
1005 "%u APIC %x\n", mce->cpuvendor, mce->cpuid,
1006 mce->time, mce->socketid, mce->apicid);
1007
1008 skx_mce_output_error(mci, mce, &res);
1009
1010 return NOTIFY_DONE;
1011}
1012
1013static struct notifier_block skx_mce_dec = {
9026cc82
BP		 1014 .notifier_call = skx_mce_check_error,
1015 .priority = MCE_PRIO_EDAC,
4ec656bd
TL		 1016};
1017
1018static void skx_remove(void)
1019{
1020 int i, j;
1021 struct skx_dev *d, *tmp;
1022
1023 edac_dbg(0, "\n");
1024
1025 list_for_each_entry_safe(d, tmp, &skx_edac_list, list) {
1026 list_del(&d->list);
1027 for (i = 0; i < NUM_IMC; i++) {
1028 skx_unregister_mci(&d->imc[i]);
1029 for (j = 0; j < NUM_CHANNELS; j++)
1030 pci_dev_put(d->imc[i].chan[j].cdev);
1031 }
1032 pci_dev_put(d->util_all);
1033 pci_dev_put(d->sad_all);
1034
1035 kfree(d);
1036 }
1037}
1038
1039/*
1040 * skx_init:
1041 * make sure we are running on the correct cpu model
1042 * search for all the devices we need
1043 * check which DIMMs are present.
1044 */
240ea921 1045static int __init skx_init(void)
4ec656bd
TL		 1046{
1047 const struct x86_cpu_id *id;
1048 const struct munit *m;
301375e7 1049 const char *owner;
4ec656bd
TL		 1050 int rc = 0, i;
1051 u8 mc = 0, src_id, node_id;
1052 struct skx_dev *d;
1053
1054 edac_dbg(2, "\n");
1055
301375e7
TK		 1056 owner = edac_get_owner();
1057 if (owner && strncmp(owner, EDAC_MOD_STR, sizeof(EDAC_MOD_STR)))
1058 return -EBUSY;
1059
4ec656bd
TL		 1060 id = x86_match_cpu(skx_cpuids);
1061 if (!id)
1062 return -ENODEV;
1063
1064 rc = skx_get_hi_lo();
1065 if (rc)
1066 return rc;
1067
1068 rc = get_all_bus_mappings();
1069 if (rc < 0)
1070 goto fail;
1071 if (rc == 0) {
1072 edac_dbg(2, "No memory controllers found\n");
1073 return -ENODEV;
1074 }
1075
1076 for (m = skx_all_munits; m->did; m++) {
1077 rc = get_all_munits(m);
1078 if (rc < 0)
1079 goto fail;
1080 if (rc != m->per_socket * skx_num_sockets) {
1081 edac_dbg(2, "Expected %d, got %d of %x\n",
1082 m->per_socket * skx_num_sockets, rc, m->did);
1083 rc = -ENODEV;
1084 goto fail;
1085 }
1086 }
1087
1088 list_for_each_entry(d, &skx_edac_list, list) {
1089 src_id = get_src_id(d);
1090 node_id = skx_get_node_id(d);
1091 edac_dbg(2, "src_id=%d node_id=%d\n", src_id, node_id);
1092 for (i = 0; i < NUM_IMC; i++) {
1093 d->imc[i].mc = mc++;
1094 d->imc[i].lmc = i;
1095 d->imc[i].src_id = src_id;
1096 d->imc[i].node_id = node_id;
1097 rc = skx_register_mci(&d->imc[i]);
1098 if (rc < 0)
1099 goto fail;
1100 }
1101 }
1102
1103 /* Ensure that the OPSTATE is set correctly for POLL or NMI */
1104 opstate_init();
1105
1106 setup_skx_debug();
1107
1108 mce_register_decode_chain(&skx_mce_dec);
1109
1110 return 0;
1111fail:
1112 skx_remove();
1113 return rc;
1114}
1115
1116static void __exit skx_exit(void)
1117{
1118 edac_dbg(2, "\n");
1119 mce_unregister_decode_chain(&skx_mce_dec);
1120 skx_remove();
1121 teardown_skx_debug();
1122}
1123
1124module_init(skx_init);
1125module_exit(skx_exit);
1126
1127module_param(edac_op_state, int, 0444);
1128MODULE_PARM_DESC(edac_op_state, "EDAC Error Reporting state: 0=Poll,1=NMI");
1129
1130MODULE_LICENSE("GPL v2");
1131MODULE_AUTHOR("Tony Luck");
1132MODULE_DESCRIPTION("MC Driver for Intel Skylake server processors");
Linux 6.x block layer and io_uring tree(s)