GNU Linux-libre 4.4.289-gnu1
[releases.git] / drivers / edac / ie31200_edac.c
1 /*
2  * Intel E3-1200
3  * Copyright (C) 2014 Jason Baron <jbaron@akamai.com>
4  *
5  * Support for the E3-1200 processor family. Heavily based on previous
6  * Intel EDAC drivers.
7  *
8  * Since the DRAM controller is on the cpu chip, we can use its PCI device
9  * id to identify these processors.
10  *
11  * PCI DRAM controller device ids (Taken from The PCI ID Repository - http://pci-ids.ucw.cz/)
12  *
13  * 0108: Xeon E3-1200 Processor Family DRAM Controller
14  * 010c: Xeon E3-1200/2nd Generation Core Processor Family DRAM Controller
15  * 0150: Xeon E3-1200 v2/3rd Gen Core processor DRAM Controller
16  * 0158: Xeon E3-1200 v2/Ivy Bridge DRAM Controller
17  * 015c: Xeon E3-1200 v2/3rd Gen Core processor DRAM Controller
18  * 0c04: Xeon E3-1200 v3/4th Gen Core Processor DRAM Controller
19  * 0c08: Xeon E3-1200 v3 Processor DRAM Controller
20  *
21  * Based on Intel specification:
22  * http://www.intel.com/content/dam/www/public/us/en/documents/datasheets/xeon-e3-1200v3-vol-2-datasheet.pdf
23  * http://www.intel.com/content/www/us/en/processors/xeon/xeon-e3-1200-family-vol-2-datasheet.html
24  *
25  * According to the above datasheet (p.16):
26  * "
27  * 6. Software must not access B0/D0/F0 32-bit memory-mapped registers with
28  * requests that cross a DW boundary.
29  * "
30  *
31  * Thus, we make use of the explicit: lo_hi_readq(), which breaks the readq into
32  * 2 readl() calls. This restriction may be lifted in subsequent chip releases,
33  * but lo_hi_readq() ensures that we are safe across all e3-1200 processors.
34  */
35
36 #include <linux/module.h>
37 #include <linux/init.h>
38 #include <linux/pci.h>
39 #include <linux/pci_ids.h>
40 #include <linux/edac.h>
41
42 #include <linux/io-64-nonatomic-lo-hi.h>
43 #include "edac_core.h"
44
45 #define IE31200_REVISION "1.0"
46 #define EDAC_MOD_STR "ie31200_edac"
47
48 #define ie31200_printk(level, fmt, arg...) \
49         edac_printk(level, "ie31200", fmt, ##arg)
50
51 #define PCI_DEVICE_ID_INTEL_IE31200_HB_1 0x0108
52 #define PCI_DEVICE_ID_INTEL_IE31200_HB_2 0x010c
53 #define PCI_DEVICE_ID_INTEL_IE31200_HB_3 0x0150
54 #define PCI_DEVICE_ID_INTEL_IE31200_HB_4 0x0158
55 #define PCI_DEVICE_ID_INTEL_IE31200_HB_5 0x015c
56 #define PCI_DEVICE_ID_INTEL_IE31200_HB_6 0x0c04
57 #define PCI_DEVICE_ID_INTEL_IE31200_HB_7 0x0c08
58
59 #define IE31200_DIMMS                   4
60 #define IE31200_RANKS                   8
61 #define IE31200_RANKS_PER_CHANNEL       4
62 #define IE31200_DIMMS_PER_CHANNEL       2
63 #define IE31200_CHANNELS                2
64
65 /* Intel IE31200 register addresses - device 0 function 0 - DRAM Controller */
66 #define IE31200_MCHBAR_LOW              0x48
67 #define IE31200_MCHBAR_HIGH             0x4c
68 #define IE31200_MCHBAR_MASK             GENMASK_ULL(38, 15)
69 #define IE31200_MMR_WINDOW_SIZE         BIT(15)
70
71 /*
72  * Error Status Register (16b)
73  *
74  * 15    reserved
75  * 14    Isochronous TBWRR Run Behind FIFO Full
76  *       (ITCV)
77  * 13    Isochronous TBWRR Run Behind FIFO Put
78  *       (ITSTV)
79  * 12    reserved
80  * 11    MCH Thermal Sensor Event
81  *       for SMI/SCI/SERR (GTSE)
82  * 10    reserved
83  *  9    LOCK to non-DRAM Memory Flag (LCKF)
84  *  8    reserved
85  *  7    DRAM Throttle Flag (DTF)
86  *  6:2  reserved
87  *  1    Multi-bit DRAM ECC Error Flag (DMERR)
88  *  0    Single-bit DRAM ECC Error Flag (DSERR)
89  */
90 #define IE31200_ERRSTS                  0xc8
91 #define IE31200_ERRSTS_UE               BIT(1)
92 #define IE31200_ERRSTS_CE               BIT(0)
93 #define IE31200_ERRSTS_BITS             (IE31200_ERRSTS_UE | IE31200_ERRSTS_CE)
94
95 /*
96  * Channel 0 ECC Error Log (64b)
97  *
98  * 63:48 Error Column Address (ERRCOL)
99  * 47:32 Error Row Address (ERRROW)
100  * 31:29 Error Bank Address (ERRBANK)
101  * 28:27 Error Rank Address (ERRRANK)
102  * 26:24 reserved
103  * 23:16 Error Syndrome (ERRSYND)
104  * 15: 2 reserved
105  *    1  Multiple Bit Error Status (MERRSTS)
106  *    0  Correctable Error Status (CERRSTS)
107  */
108 #define IE31200_C0ECCERRLOG                     0x40c8
109 #define IE31200_C1ECCERRLOG                     0x44c8
110 #define IE31200_ECCERRLOG_CE                    BIT(0)
111 #define IE31200_ECCERRLOG_UE                    BIT(1)
112 #define IE31200_ECCERRLOG_RANK_BITS             GENMASK_ULL(28, 27)
113 #define IE31200_ECCERRLOG_RANK_SHIFT            27
114 #define IE31200_ECCERRLOG_SYNDROME_BITS         GENMASK_ULL(23, 16)
115 #define IE31200_ECCERRLOG_SYNDROME_SHIFT        16
116
117 #define IE31200_ECCERRLOG_SYNDROME(log)            \
118         ((log & IE31200_ECCERRLOG_SYNDROME_BITS) >> \
119          IE31200_ECCERRLOG_SYNDROME_SHIFT)
120
121 #define IE31200_CAPID0                  0xe4
122 #define IE31200_CAPID0_PDCD             BIT(4)
123 #define IE31200_CAPID0_DDPCD            BIT(6)
124 #define IE31200_CAPID0_ECC              BIT(1)
125
126 #define IE31200_MAD_DIMM_0_OFFSET       0x5004
127 #define IE31200_MAD_DIMM_SIZE           GENMASK_ULL(7, 0)
128 #define IE31200_MAD_DIMM_A_RANK         BIT(17)
129 #define IE31200_MAD_DIMM_A_WIDTH        BIT(19)
130
131 #define IE31200_PAGES(n)                (n << (28 - PAGE_SHIFT))
132
133 static int nr_channels;
134 static struct pci_dev *mci_pdev;
135 static int ie31200_registered = 1;
136
137 struct ie31200_priv {
138         void __iomem *window;
139 };
140
141 enum ie31200_chips {
142         IE31200 = 0,
143 };
144
145 struct ie31200_dev_info {
146         const char *ctl_name;
147 };
148
149 struct ie31200_error_info {
150         u16 errsts;
151         u16 errsts2;
152         u64 eccerrlog[IE31200_CHANNELS];
153 };
154
155 static const struct ie31200_dev_info ie31200_devs[] = {
156         [IE31200] = {
157                 .ctl_name = "IE31200"
158         },
159 };
160
161 struct dimm_data {
162         u8 size; /* in 256MB multiples */
163         u8 dual_rank : 1,
164            x16_width : 1; /* 0 means x8 width */
165 };
166
167 static int how_many_channels(struct pci_dev *pdev)
168 {
169         int n_channels;
170         unsigned char capid0_2b; /* 2nd byte of CAPID0 */
171
172         pci_read_config_byte(pdev, IE31200_CAPID0 + 1, &capid0_2b);
173
174         /* check PDCD: Dual Channel Disable */
175         if (capid0_2b & IE31200_CAPID0_PDCD) {
176                 edac_dbg(0, "In single channel mode\n");
177                 n_channels = 1;
178         } else {
179                 edac_dbg(0, "In dual channel mode\n");
180                 n_channels = 2;
181         }
182
183         /* check DDPCD - check if both channels are filled */
184         if (capid0_2b & IE31200_CAPID0_DDPCD)
185                 edac_dbg(0, "2 DIMMS per channel disabled\n");
186         else
187                 edac_dbg(0, "2 DIMMS per channel enabled\n");
188
189         return n_channels;
190 }
191
192 static bool ecc_capable(struct pci_dev *pdev)
193 {
194         unsigned char capid0_4b; /* 4th byte of CAPID0 */
195
196         pci_read_config_byte(pdev, IE31200_CAPID0 + 3, &capid0_4b);
197         if (capid0_4b & IE31200_CAPID0_ECC)
198                 return false;
199         return true;
200 }
201
202 static int eccerrlog_row(int channel, u64 log)
203 {
204         int rank = ((log & IE31200_ECCERRLOG_RANK_BITS) >>
205                 IE31200_ECCERRLOG_RANK_SHIFT);
206         return rank | (channel * IE31200_RANKS_PER_CHANNEL);
207 }
208
209 static void ie31200_clear_error_info(struct mem_ctl_info *mci)
210 {
211         /*
212          * Clear any error bits.
213          * (Yes, we really clear bits by writing 1 to them.)
214          */
215         pci_write_bits16(to_pci_dev(mci->pdev), IE31200_ERRSTS,
216                          IE31200_ERRSTS_BITS, IE31200_ERRSTS_BITS);
217 }
218
219 static void ie31200_get_and_clear_error_info(struct mem_ctl_info *mci,
220                                              struct ie31200_error_info *info)
221 {
222         struct pci_dev *pdev;
223         struct ie31200_priv *priv = mci->pvt_info;
224         void __iomem *window = priv->window;
225
226         pdev = to_pci_dev(mci->pdev);
227
228         /*
229          * This is a mess because there is no atomic way to read all the
230          * registers at once and the registers can transition from CE being
231          * overwritten by UE.
232          */
233         pci_read_config_word(pdev, IE31200_ERRSTS, &info->errsts);
234         if (!(info->errsts & IE31200_ERRSTS_BITS))
235                 return;
236
237         info->eccerrlog[0] = lo_hi_readq(window + IE31200_C0ECCERRLOG);
238         if (nr_channels == 2)
239                 info->eccerrlog[1] = lo_hi_readq(window + IE31200_C1ECCERRLOG);
240
241         pci_read_config_word(pdev, IE31200_ERRSTS, &info->errsts2);
242
243         /*
244          * If the error is the same for both reads then the first set
245          * of reads is valid.  If there is a change then there is a CE
246          * with no info and the second set of reads is valid and
247          * should be UE info.
248          */
249         if ((info->errsts ^ info->errsts2) & IE31200_ERRSTS_BITS) {
250                 info->eccerrlog[0] = lo_hi_readq(window + IE31200_C0ECCERRLOG);
251                 if (nr_channels == 2)
252                         info->eccerrlog[1] =
253                                 lo_hi_readq(window + IE31200_C1ECCERRLOG);
254         }
255
256         ie31200_clear_error_info(mci);
257 }
258
259 static void ie31200_process_error_info(struct mem_ctl_info *mci,
260                                        struct ie31200_error_info *info)
261 {
262         int channel;
263         u64 log;
264
265         if (!(info->errsts & IE31200_ERRSTS_BITS))
266                 return;
267
268         if ((info->errsts ^ info->errsts2) & IE31200_ERRSTS_BITS) {
269                 edac_mc_handle_error(HW_EVENT_ERR_UNCORRECTED, mci, 1, 0, 0, 0,
270                                      -1, -1, -1, "UE overwrote CE", "");
271                 info->errsts = info->errsts2;
272         }
273
274         for (channel = 0; channel < nr_channels; channel++) {
275                 log = info->eccerrlog[channel];
276                 if (log & IE31200_ECCERRLOG_UE) {
277                         edac_mc_handle_error(HW_EVENT_ERR_UNCORRECTED, mci, 1,
278                                              0, 0, 0,
279                                              eccerrlog_row(channel, log),
280                                              channel, -1,
281                                              "ie31200 UE", "");
282                 } else if (log & IE31200_ECCERRLOG_CE) {
283                         edac_mc_handle_error(HW_EVENT_ERR_CORRECTED, mci, 1,
284                                              0, 0,
285                                              IE31200_ECCERRLOG_SYNDROME(log),
286                                              eccerrlog_row(channel, log),
287                                              channel, -1,
288                                              "ie31200 CE", "");
289                 }
290         }
291 }
292
293 static void ie31200_check(struct mem_ctl_info *mci)
294 {
295         struct ie31200_error_info info;
296
297         edac_dbg(1, "MC%d\n", mci->mc_idx);
298         ie31200_get_and_clear_error_info(mci, &info);
299         ie31200_process_error_info(mci, &info);
300 }
301
302 static void __iomem *ie31200_map_mchbar(struct pci_dev *pdev)
303 {
304         union {
305                 u64 mchbar;
306                 struct {
307                         u32 mchbar_low;
308                         u32 mchbar_high;
309                 };
310         } u;
311         void __iomem *window;
312
313         pci_read_config_dword(pdev, IE31200_MCHBAR_LOW, &u.mchbar_low);
314         pci_read_config_dword(pdev, IE31200_MCHBAR_HIGH, &u.mchbar_high);
315         u.mchbar &= IE31200_MCHBAR_MASK;
316
317         if (u.mchbar != (resource_size_t)u.mchbar) {
318                 ie31200_printk(KERN_ERR, "mmio space beyond accessible range (0x%llx)\n",
319                                (unsigned long long)u.mchbar);
320                 return NULL;
321         }
322
323         window = ioremap_nocache(u.mchbar, IE31200_MMR_WINDOW_SIZE);
324         if (!window)
325                 ie31200_printk(KERN_ERR, "Cannot map mmio space at 0x%llx\n",
326                                (unsigned long long)u.mchbar);
327
328         return window;
329 }
330
331 static int ie31200_probe1(struct pci_dev *pdev, int dev_idx)
332 {
333         int i, j, ret;
334         struct mem_ctl_info *mci = NULL;
335         struct edac_mc_layer layers[2];
336         struct dimm_data dimm_info[IE31200_CHANNELS][IE31200_DIMMS_PER_CHANNEL];
337         void __iomem *window;
338         struct ie31200_priv *priv;
339         u32 addr_decode;
340
341         edac_dbg(0, "MC:\n");
342
343         if (!ecc_capable(pdev)) {
344                 ie31200_printk(KERN_INFO, "No ECC support\n");
345                 return -ENODEV;
346         }
347
348         nr_channels = how_many_channels(pdev);
349         layers[0].type = EDAC_MC_LAYER_CHIP_SELECT;
350         layers[0].size = IE31200_DIMMS;
351         layers[0].is_virt_csrow = true;
352         layers[1].type = EDAC_MC_LAYER_CHANNEL;
353         layers[1].size = nr_channels;
354         layers[1].is_virt_csrow = false;
355         mci = edac_mc_alloc(0, ARRAY_SIZE(layers), layers,
356                             sizeof(struct ie31200_priv));
357         if (!mci)
358                 return -ENOMEM;
359
360         window = ie31200_map_mchbar(pdev);
361         if (!window) {
362                 ret = -ENODEV;
363                 goto fail_free;
364         }
365
366         edac_dbg(3, "MC: init mci\n");
367         mci->pdev = &pdev->dev;
368         mci->mtype_cap = MEM_FLAG_DDR3;
369         mci->edac_ctl_cap = EDAC_FLAG_SECDED;
370         mci->edac_cap = EDAC_FLAG_SECDED;
371         mci->mod_name = EDAC_MOD_STR;
372         mci->mod_ver = IE31200_REVISION;
373         mci->ctl_name = ie31200_devs[dev_idx].ctl_name;
374         mci->dev_name = pci_name(pdev);
375         mci->edac_check = ie31200_check;
376         mci->ctl_page_to_phys = NULL;
377         priv = mci->pvt_info;
378         priv->window = window;
379
380         /* populate DIMM info */
381         for (i = 0; i < IE31200_CHANNELS; i++) {
382                 addr_decode = readl(window + IE31200_MAD_DIMM_0_OFFSET +
383                                         (i * 4));
384                 edac_dbg(0, "addr_decode: 0x%x\n", addr_decode);
385                 for (j = 0; j < IE31200_DIMMS_PER_CHANNEL; j++) {
386                         dimm_info[i][j].size = (addr_decode >> (j * 8)) &
387                                                 IE31200_MAD_DIMM_SIZE;
388                         dimm_info[i][j].dual_rank = (addr_decode &
389                                 (IE31200_MAD_DIMM_A_RANK << j)) ? 1 : 0;
390                         dimm_info[i][j].x16_width = (addr_decode &
391                                 (IE31200_MAD_DIMM_A_WIDTH << j)) ? 1 : 0;
392                         edac_dbg(0, "size: 0x%x, rank: %d, width: %d\n",
393                                  dimm_info[i][j].size,
394                                  dimm_info[i][j].dual_rank,
395                                  dimm_info[i][j].x16_width);
396                 }
397         }
398
399         /*
400          * The dram rank boundary (DRB) reg values are boundary addresses
401          * for each DRAM rank with a granularity of 64MB.  DRB regs are
402          * cumulative; the last one will contain the total memory
403          * contained in all ranks.
404          */
405         for (i = 0; i < IE31200_DIMMS_PER_CHANNEL; i++) {
406                 for (j = 0; j < IE31200_CHANNELS; j++) {
407                         struct dimm_info *dimm;
408                         unsigned long nr_pages;
409
410                         nr_pages = IE31200_PAGES(dimm_info[j][i].size);
411                         if (nr_pages == 0)
412                                 continue;
413
414                         if (dimm_info[j][i].dual_rank) {
415                                 nr_pages = nr_pages / 2;
416                                 dimm = EDAC_DIMM_PTR(mci->layers, mci->dimms,
417                                                      mci->n_layers, (i * 2) + 1,
418                                                      j, 0);
419                                 dimm->nr_pages = nr_pages;
420                                 edac_dbg(0, "set nr pages: 0x%lx\n", nr_pages);
421                                 dimm->grain = 8; /* just a guess */
422                                 dimm->mtype = MEM_DDR3;
423                                 dimm->dtype = DEV_UNKNOWN;
424                                 dimm->edac_mode = EDAC_UNKNOWN;
425                         }
426                         dimm = EDAC_DIMM_PTR(mci->layers, mci->dimms,
427                                              mci->n_layers, i * 2, j, 0);
428                         dimm->nr_pages = nr_pages;
429                         edac_dbg(0, "set nr pages: 0x%lx\n", nr_pages);
430                         dimm->grain = 8; /* same guess */
431                         dimm->mtype = MEM_DDR3;
432                         dimm->dtype = DEV_UNKNOWN;
433                         dimm->edac_mode = EDAC_UNKNOWN;
434                 }
435         }
436
437         ie31200_clear_error_info(mci);
438
439         if (edac_mc_add_mc(mci)) {
440                 edac_dbg(3, "MC: failed edac_mc_add_mc()\n");
441                 ret = -ENODEV;
442                 goto fail_unmap;
443         }
444
445         /* get this far and it's successful */
446         edac_dbg(3, "MC: success\n");
447         return 0;
448
449 fail_unmap:
450         iounmap(window);
451
452 fail_free:
453         edac_mc_free(mci);
454
455         return ret;
456 }
457
458 static int ie31200_init_one(struct pci_dev *pdev,
459                             const struct pci_device_id *ent)
460 {
461         int rc;
462
463         edac_dbg(0, "MC:\n");
464         if (pci_enable_device(pdev) < 0)
465                 return -EIO;
466         rc = ie31200_probe1(pdev, ent->driver_data);
467         if (rc == 0 && !mci_pdev)
468                 mci_pdev = pci_dev_get(pdev);
469
470         return rc;
471 }
472
473 static void ie31200_remove_one(struct pci_dev *pdev)
474 {
475         struct mem_ctl_info *mci;
476         struct ie31200_priv *priv;
477
478         edac_dbg(0, "\n");
479         pci_dev_put(mci_pdev);
480         mci_pdev = NULL;
481         mci = edac_mc_del_mc(&pdev->dev);
482         if (!mci)
483                 return;
484         priv = mci->pvt_info;
485         iounmap(priv->window);
486         edac_mc_free(mci);
487 }
488
489 static const struct pci_device_id ie31200_pci_tbl[] = {
490         {
491                 PCI_VEND_DEV(INTEL, IE31200_HB_1), PCI_ANY_ID, PCI_ANY_ID, 0, 0,
492                 IE31200},
493         {
494                 PCI_VEND_DEV(INTEL, IE31200_HB_2), PCI_ANY_ID, PCI_ANY_ID, 0, 0,
495                 IE31200},
496         {
497                 PCI_VEND_DEV(INTEL, IE31200_HB_3), PCI_ANY_ID, PCI_ANY_ID, 0, 0,
498                 IE31200},
499         {
500                 PCI_VEND_DEV(INTEL, IE31200_HB_4), PCI_ANY_ID, PCI_ANY_ID, 0, 0,
501                 IE31200},
502         {
503                 PCI_VEND_DEV(INTEL, IE31200_HB_5), PCI_ANY_ID, PCI_ANY_ID, 0, 0,
504                 IE31200},
505         {
506                 PCI_VEND_DEV(INTEL, IE31200_HB_6), PCI_ANY_ID, PCI_ANY_ID, 0, 0,
507                 IE31200},
508         {
509                 PCI_VEND_DEV(INTEL, IE31200_HB_7), PCI_ANY_ID, PCI_ANY_ID, 0, 0,
510                 IE31200},
511         {
512                 0,
513         }            /* 0 terminated list. */
514 };
515 MODULE_DEVICE_TABLE(pci, ie31200_pci_tbl);
516
517 static struct pci_driver ie31200_driver = {
518         .name = EDAC_MOD_STR,
519         .probe = ie31200_init_one,
520         .remove = ie31200_remove_one,
521         .id_table = ie31200_pci_tbl,
522 };
523
524 static int __init ie31200_init(void)
525 {
526         int pci_rc, i;
527
528         edac_dbg(3, "MC:\n");
529         /* Ensure that the OPSTATE is set correctly for POLL or NMI */
530         opstate_init();
531
532         pci_rc = pci_register_driver(&ie31200_driver);
533         if (pci_rc < 0)
534                 goto fail0;
535
536         if (!mci_pdev) {
537                 ie31200_registered = 0;
538                 for (i = 0; ie31200_pci_tbl[i].vendor != 0; i++) {
539                         mci_pdev = pci_get_device(ie31200_pci_tbl[i].vendor,
540                                                   ie31200_pci_tbl[i].device,
541                                                   NULL);
542                         if (mci_pdev)
543                                 break;
544                 }
545                 if (!mci_pdev) {
546                         edac_dbg(0, "ie31200 pci_get_device fail\n");
547                         pci_rc = -ENODEV;
548                         goto fail1;
549                 }
550                 pci_rc = ie31200_init_one(mci_pdev, &ie31200_pci_tbl[i]);
551                 if (pci_rc < 0) {
552                         edac_dbg(0, "ie31200 init fail\n");
553                         pci_rc = -ENODEV;
554                         goto fail1;
555                 }
556         }
557         return 0;
558
559 fail1:
560         pci_unregister_driver(&ie31200_driver);
561 fail0:
562         pci_dev_put(mci_pdev);
563
564         return pci_rc;
565 }
566
567 static void __exit ie31200_exit(void)
568 {
569         edac_dbg(3, "MC:\n");
570         pci_unregister_driver(&ie31200_driver);
571         if (!ie31200_registered)
572                 ie31200_remove_one(mci_pdev);
573 }
574
575 module_init(ie31200_init);
576 module_exit(ie31200_exit);
577
578 MODULE_LICENSE("GPL");
579 MODULE_AUTHOR("Jason Baron <jbaron@akamai.com>");
580 MODULE_DESCRIPTION("MC support for Intel Processor E31200 memory hub controllers");