blob: 4adaf4b7da993c3d6b1f25d430dbdf1aa3e7a875 [file] [log] [blame]
Mauro Carvalho Chehabeebf11a2011-10-20 19:18:01 -02001/* Intel Sandy Bridge -EN/-EP/-EX Memory Controller kernel module
2 *
3 * This driver supports the memory controllers found on the Intel
4 * processor family Sandy Bridge.
5 *
6 * This file may be distributed under the terms of the
7 * GNU General Public License version 2 only.
8 *
9 * Copyright (c) 2011 by:
10 * Mauro Carvalho Chehab <mchehab@redhat.com>
11 */
12
13#include <linux/module.h>
14#include <linux/init.h>
15#include <linux/pci.h>
16#include <linux/pci_ids.h>
17#include <linux/slab.h>
18#include <linux/delay.h>
19#include <linux/edac.h>
20#include <linux/mmzone.h>
Mauro Carvalho Chehabeebf11a2011-10-20 19:18:01 -020021#include <linux/smp.h>
22#include <linux/bitmap.h>
Mauro Carvalho Chehab5b889e32011-11-07 18:26:53 -030023#include <linux/math64.h>
Mauro Carvalho Chehabeebf11a2011-10-20 19:18:01 -020024#include <asm/processor.h>
Mauro Carvalho Chehab3d78c9a2011-10-20 19:33:46 -020025#include <asm/mce.h>
Mauro Carvalho Chehabeebf11a2011-10-20 19:18:01 -020026
27#include "edac_core.h"
28
29/* Static vars */
30static LIST_HEAD(sbridge_edac_list);
31static DEFINE_MUTEX(sbridge_edac_lock);
32static int probed;
33
34/*
35 * Alter this version for the module when modifications are made
36 */
37#define SBRIDGE_REVISION " Ver: 1.0.0 "
38#define EDAC_MOD_STR "sbridge_edac"
39
40/*
41 * Debug macros
42 */
43#define sbridge_printk(level, fmt, arg...) \
44 edac_printk(level, "sbridge", fmt, ##arg)
45
46#define sbridge_mc_printk(mci, level, fmt, arg...) \
47 edac_mc_chipset_printk(mci, level, "sbridge", fmt, ##arg)
48
49/*
50 * Get a bit field at register value <v>, from bit <lo> to bit <hi>
51 */
52#define GET_BITFIELD(v, lo, hi) \
53 (((v) & ((1ULL << ((hi) - (lo) + 1)) - 1) << (lo)) >> (lo))
54
55/*
56 * sbridge Memory Controller Registers
57 */
58
59/*
60 * FIXME: For now, let's order by device function, as it makes
David Mackey15ed1032012-04-17 11:30:52 -070061 * easier for driver's development process. This table should be
Mauro Carvalho Chehabeebf11a2011-10-20 19:18:01 -020062 * moved to pci_id.h when submitted upstream
63 */
64#define PCI_DEVICE_ID_INTEL_SBRIDGE_SAD0 0x3cf4 /* 12.6 */
65#define PCI_DEVICE_ID_INTEL_SBRIDGE_SAD1 0x3cf6 /* 12.7 */
66#define PCI_DEVICE_ID_INTEL_SBRIDGE_BR 0x3cf5 /* 13.6 */
67#define PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_HA0 0x3ca0 /* 14.0 */
68#define PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TA 0x3ca8 /* 15.0 */
69#define PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_RAS 0x3c71 /* 15.1 */
70#define PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD0 0x3caa /* 15.2 */
71#define PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD1 0x3cab /* 15.3 */
72#define PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD2 0x3cac /* 15.4 */
73#define PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD3 0x3cad /* 15.5 */
74#define PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_DDRIO 0x3cb8 /* 17.0 */
75
76 /*
77 * Currently, unused, but will be needed in the future
78 * implementations, as they hold the error counters
79 */
80#define PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_ERR0 0x3c72 /* 16.2 */
81#define PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_ERR1 0x3c73 /* 16.3 */
82#define PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_ERR2 0x3c76 /* 16.6 */
83#define PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_ERR3 0x3c77 /* 16.7 */
84
85/* Devices 12 Function 6, Offsets 0x80 to 0xcc */
86static const u32 dram_rule[] = {
87 0x80, 0x88, 0x90, 0x98, 0xa0,
88 0xa8, 0xb0, 0xb8, 0xc0, 0xc8,
89};
90#define MAX_SAD ARRAY_SIZE(dram_rule)
91
92#define SAD_LIMIT(reg) ((GET_BITFIELD(reg, 6, 25) << 26) | 0x3ffffff)
93#define DRAM_ATTR(reg) GET_BITFIELD(reg, 2, 3)
94#define INTERLEAVE_MODE(reg) GET_BITFIELD(reg, 1, 1)
95#define DRAM_RULE_ENABLE(reg) GET_BITFIELD(reg, 0, 0)
96
97static char *get_dram_attr(u32 reg)
98{
99 switch(DRAM_ATTR(reg)) {
100 case 0:
101 return "DRAM";
102 case 1:
103 return "MMCFG";
104 case 2:
105 return "NXM";
106 default:
107 return "unknown";
108 }
109}
110
111static const u32 interleave_list[] = {
112 0x84, 0x8c, 0x94, 0x9c, 0xa4,
113 0xac, 0xb4, 0xbc, 0xc4, 0xcc,
114};
115#define MAX_INTERLEAVE ARRAY_SIZE(interleave_list)
116
117#define SAD_PKG0(reg) GET_BITFIELD(reg, 0, 2)
118#define SAD_PKG1(reg) GET_BITFIELD(reg, 3, 5)
119#define SAD_PKG2(reg) GET_BITFIELD(reg, 8, 10)
120#define SAD_PKG3(reg) GET_BITFIELD(reg, 11, 13)
121#define SAD_PKG4(reg) GET_BITFIELD(reg, 16, 18)
122#define SAD_PKG5(reg) GET_BITFIELD(reg, 19, 21)
123#define SAD_PKG6(reg) GET_BITFIELD(reg, 24, 26)
124#define SAD_PKG7(reg) GET_BITFIELD(reg, 27, 29)
125
126static inline int sad_pkg(u32 reg, int interleave)
127{
128 switch (interleave) {
129 case 0:
130 return SAD_PKG0(reg);
131 case 1:
132 return SAD_PKG1(reg);
133 case 2:
134 return SAD_PKG2(reg);
135 case 3:
136 return SAD_PKG3(reg);
137 case 4:
138 return SAD_PKG4(reg);
139 case 5:
140 return SAD_PKG5(reg);
141 case 6:
142 return SAD_PKG6(reg);
143 case 7:
144 return SAD_PKG7(reg);
145 default:
146 return -EINVAL;
147 }
148}
149
150/* Devices 12 Function 7 */
151
152#define TOLM 0x80
153#define TOHM 0x84
154
155#define GET_TOLM(reg) ((GET_BITFIELD(reg, 0, 3) << 28) | 0x3ffffff)
156#define GET_TOHM(reg) ((GET_BITFIELD(reg, 0, 20) << 25) | 0x3ffffff)
157
158/* Device 13 Function 6 */
159
160#define SAD_TARGET 0xf0
161
162#define SOURCE_ID(reg) GET_BITFIELD(reg, 9, 11)
163
164#define SAD_CONTROL 0xf4
165
166#define NODE_ID(reg) GET_BITFIELD(reg, 0, 2)
167
168/* Device 14 function 0 */
169
170static const u32 tad_dram_rule[] = {
171 0x40, 0x44, 0x48, 0x4c,
172 0x50, 0x54, 0x58, 0x5c,
173 0x60, 0x64, 0x68, 0x6c,
174};
175#define MAX_TAD ARRAY_SIZE(tad_dram_rule)
176
177#define TAD_LIMIT(reg) ((GET_BITFIELD(reg, 12, 31) << 26) | 0x3ffffff)
178#define TAD_SOCK(reg) GET_BITFIELD(reg, 10, 11)
179#define TAD_CH(reg) GET_BITFIELD(reg, 8, 9)
180#define TAD_TGT3(reg) GET_BITFIELD(reg, 6, 7)
181#define TAD_TGT2(reg) GET_BITFIELD(reg, 4, 5)
182#define TAD_TGT1(reg) GET_BITFIELD(reg, 2, 3)
183#define TAD_TGT0(reg) GET_BITFIELD(reg, 0, 1)
184
185/* Device 15, function 0 */
186
187#define MCMTR 0x7c
188
189#define IS_ECC_ENABLED(mcmtr) GET_BITFIELD(mcmtr, 2, 2)
190#define IS_LOCKSTEP_ENABLED(mcmtr) GET_BITFIELD(mcmtr, 1, 1)
191#define IS_CLOSE_PG(mcmtr) GET_BITFIELD(mcmtr, 0, 0)
192
193/* Device 15, function 1 */
194
195#define RASENABLES 0xac
196#define IS_MIRROR_ENABLED(reg) GET_BITFIELD(reg, 0, 0)
197
198/* Device 15, functions 2-5 */
199
200static const int mtr_regs[] = {
201 0x80, 0x84, 0x88,
202};
203
204#define RANK_DISABLE(mtr) GET_BITFIELD(mtr, 16, 19)
205#define IS_DIMM_PRESENT(mtr) GET_BITFIELD(mtr, 14, 14)
206#define RANK_CNT_BITS(mtr) GET_BITFIELD(mtr, 12, 13)
207#define RANK_WIDTH_BITS(mtr) GET_BITFIELD(mtr, 2, 4)
208#define COL_WIDTH_BITS(mtr) GET_BITFIELD(mtr, 0, 1)
209
210static const u32 tad_ch_nilv_offset[] = {
211 0x90, 0x94, 0x98, 0x9c,
212 0xa0, 0xa4, 0xa8, 0xac,
213 0xb0, 0xb4, 0xb8, 0xbc,
214};
215#define CHN_IDX_OFFSET(reg) GET_BITFIELD(reg, 28, 29)
216#define TAD_OFFSET(reg) (GET_BITFIELD(reg, 6, 25) << 26)
217
218static const u32 rir_way_limit[] = {
219 0x108, 0x10c, 0x110, 0x114, 0x118,
220};
221#define MAX_RIR_RANGES ARRAY_SIZE(rir_way_limit)
222
223#define IS_RIR_VALID(reg) GET_BITFIELD(reg, 31, 31)
224#define RIR_WAY(reg) GET_BITFIELD(reg, 28, 29)
225#define RIR_LIMIT(reg) ((GET_BITFIELD(reg, 1, 10) << 29)| 0x1fffffff)
226
227#define MAX_RIR_WAY 8
228
229static const u32 rir_offset[MAX_RIR_RANGES][MAX_RIR_WAY] = {
230 { 0x120, 0x124, 0x128, 0x12c, 0x130, 0x134, 0x138, 0x13c },
231 { 0x140, 0x144, 0x148, 0x14c, 0x150, 0x154, 0x158, 0x15c },
232 { 0x160, 0x164, 0x168, 0x16c, 0x170, 0x174, 0x178, 0x17c },
233 { 0x180, 0x184, 0x188, 0x18c, 0x190, 0x194, 0x198, 0x19c },
234 { 0x1a0, 0x1a4, 0x1a8, 0x1ac, 0x1b0, 0x1b4, 0x1b8, 0x1bc },
235};
236
237#define RIR_RNK_TGT(reg) GET_BITFIELD(reg, 16, 19)
238#define RIR_OFFSET(reg) GET_BITFIELD(reg, 2, 14)
239
240/* Device 16, functions 2-7 */
241
242/*
243 * FIXME: Implement the error count reads directly
244 */
245
246static const u32 correrrcnt[] = {
247 0x104, 0x108, 0x10c, 0x110,
248};
249
250#define RANK_ODD_OV(reg) GET_BITFIELD(reg, 31, 31)
251#define RANK_ODD_ERR_CNT(reg) GET_BITFIELD(reg, 16, 30)
252#define RANK_EVEN_OV(reg) GET_BITFIELD(reg, 15, 15)
253#define RANK_EVEN_ERR_CNT(reg) GET_BITFIELD(reg, 0, 14)
254
255static const u32 correrrthrsld[] = {
256 0x11c, 0x120, 0x124, 0x128,
257};
258
259#define RANK_ODD_ERR_THRSLD(reg) GET_BITFIELD(reg, 16, 30)
260#define RANK_EVEN_ERR_THRSLD(reg) GET_BITFIELD(reg, 0, 14)
261
262
263/* Device 17, function 0 */
264
265#define RANK_CFG_A 0x0328
266
267#define IS_RDIMM_ENABLED(reg) GET_BITFIELD(reg, 11, 11)
268
269/*
270 * sbridge structs
271 */
272
273#define NUM_CHANNELS 4
274#define MAX_DIMMS 3 /* Max DIMMS per channel */
275
276struct sbridge_info {
277 u32 mcmtr;
278};
279
280struct sbridge_channel {
281 u32 ranks;
282 u32 dimms;
283};
284
285struct pci_id_descr {
286 int dev;
287 int func;
288 int dev_id;
289 int optional;
290};
291
292struct pci_id_table {
293 const struct pci_id_descr *descr;
294 int n_devs;
295};
296
297struct sbridge_dev {
298 struct list_head list;
299 u8 bus, mc;
300 u8 node_id, source_id;
301 struct pci_dev **pdev;
302 int n_devs;
303 struct mem_ctl_info *mci;
304};
305
306struct sbridge_pvt {
307 struct pci_dev *pci_ta, *pci_ddrio, *pci_ras;
308 struct pci_dev *pci_sad0, *pci_sad1, *pci_ha0;
309 struct pci_dev *pci_br;
310 struct pci_dev *pci_tad[NUM_CHANNELS];
311
312 struct sbridge_dev *sbridge_dev;
313
314 struct sbridge_info info;
315 struct sbridge_channel channel[NUM_CHANNELS];
316
Mauro Carvalho Chehabeebf11a2011-10-20 19:18:01 -0200317 /* Memory type detection */
318 bool is_mirrored, is_lockstep, is_close_pg;
319
Mauro Carvalho Chehabeebf11a2011-10-20 19:18:01 -0200320 /* Fifo double buffers */
321 struct mce mce_entry[MCE_LOG_LEN];
322 struct mce mce_outentry[MCE_LOG_LEN];
323
324 /* Fifo in/out counters */
325 unsigned mce_in, mce_out;
326
327 /* Count indicator to show errors not got */
328 unsigned mce_overrun;
329
330 /* Memory description */
331 u64 tolm, tohm;
332};
333
334#define PCI_DESCR(device, function, device_id) \
335 .dev = (device), \
336 .func = (function), \
337 .dev_id = (device_id)
338
339static const struct pci_id_descr pci_dev_descr_sbridge[] = {
340 /* Processor Home Agent */
341 { PCI_DESCR(14, 0, PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_HA0) },
342
343 /* Memory controller */
344 { PCI_DESCR(15, 0, PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TA) },
345 { PCI_DESCR(15, 1, PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_RAS) },
346 { PCI_DESCR(15, 2, PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD0) },
347 { PCI_DESCR(15, 3, PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD1) },
348 { PCI_DESCR(15, 4, PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD2) },
349 { PCI_DESCR(15, 5, PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD3) },
350 { PCI_DESCR(17, 0, PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_DDRIO) },
351
352 /* System Address Decoder */
353 { PCI_DESCR(12, 6, PCI_DEVICE_ID_INTEL_SBRIDGE_SAD0) },
354 { PCI_DESCR(12, 7, PCI_DEVICE_ID_INTEL_SBRIDGE_SAD1) },
355
356 /* Broadcast Registers */
357 { PCI_DESCR(13, 6, PCI_DEVICE_ID_INTEL_SBRIDGE_BR) },
358};
359
360#define PCI_ID_TABLE_ENTRY(A) { .descr=A, .n_devs = ARRAY_SIZE(A) }
361static const struct pci_id_table pci_dev_descr_sbridge_table[] = {
362 PCI_ID_TABLE_ENTRY(pci_dev_descr_sbridge),
363 {0,} /* 0 terminated list. */
364};
365
366/*
367 * pci_device_id table for which devices we are looking for
368 */
Lionel Debroux36c46f32012-02-27 07:41:47 +0100369static DEFINE_PCI_DEVICE_TABLE(sbridge_pci_tbl) = {
Mauro Carvalho Chehabeebf11a2011-10-20 19:18:01 -0200370 {PCI_DEVICE(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TA)},
371 {0,} /* 0 terminated list. */
372};
373
374
375/****************************************************************************
David Mackey15ed1032012-04-17 11:30:52 -0700376 Ancillary status routines
Mauro Carvalho Chehabeebf11a2011-10-20 19:18:01 -0200377 ****************************************************************************/
378
379static inline int numrank(u32 mtr)
380{
381 int ranks = (1 << RANK_CNT_BITS(mtr));
382
383 if (ranks > 4) {
384 debugf0("Invalid number of ranks: %d (max = 4) raw value = %x (%04x)",
385 ranks, (unsigned int)RANK_CNT_BITS(mtr), mtr);
386 return -EINVAL;
387 }
388
389 return ranks;
390}
391
392static inline int numrow(u32 mtr)
393{
394 int rows = (RANK_WIDTH_BITS(mtr) + 12);
395
396 if (rows < 13 || rows > 18) {
397 debugf0("Invalid number of rows: %d (should be between 14 and 17) raw value = %x (%04x)",
398 rows, (unsigned int)RANK_WIDTH_BITS(mtr), mtr);
399 return -EINVAL;
400 }
401
402 return 1 << rows;
403}
404
405static inline int numcol(u32 mtr)
406{
407 int cols = (COL_WIDTH_BITS(mtr) + 10);
408
409 if (cols > 12) {
410 debugf0("Invalid number of cols: %d (max = 4) raw value = %x (%04x)",
411 cols, (unsigned int)COL_WIDTH_BITS(mtr), mtr);
412 return -EINVAL;
413 }
414
415 return 1 << cols;
416}
417
418static struct sbridge_dev *get_sbridge_dev(u8 bus)
419{
420 struct sbridge_dev *sbridge_dev;
421
422 list_for_each_entry(sbridge_dev, &sbridge_edac_list, list) {
423 if (sbridge_dev->bus == bus)
424 return sbridge_dev;
425 }
426
427 return NULL;
428}
429
430static struct sbridge_dev *alloc_sbridge_dev(u8 bus,
431 const struct pci_id_table *table)
432{
433 struct sbridge_dev *sbridge_dev;
434
435 sbridge_dev = kzalloc(sizeof(*sbridge_dev), GFP_KERNEL);
436 if (!sbridge_dev)
437 return NULL;
438
439 sbridge_dev->pdev = kzalloc(sizeof(*sbridge_dev->pdev) * table->n_devs,
440 GFP_KERNEL);
441 if (!sbridge_dev->pdev) {
442 kfree(sbridge_dev);
443 return NULL;
444 }
445
446 sbridge_dev->bus = bus;
447 sbridge_dev->n_devs = table->n_devs;
448 list_add_tail(&sbridge_dev->list, &sbridge_edac_list);
449
450 return sbridge_dev;
451}
452
453static void free_sbridge_dev(struct sbridge_dev *sbridge_dev)
454{
455 list_del(&sbridge_dev->list);
456 kfree(sbridge_dev->pdev);
457 kfree(sbridge_dev);
458}
459
460/****************************************************************************
461 Memory check routines
462 ****************************************************************************/
463static struct pci_dev *get_pdev_slot_func(u8 bus, unsigned slot,
464 unsigned func)
465{
466 struct sbridge_dev *sbridge_dev = get_sbridge_dev(bus);
467 int i;
468
469 if (!sbridge_dev)
470 return NULL;
471
472 for (i = 0; i < sbridge_dev->n_devs; i++) {
473 if (!sbridge_dev->pdev[i])
474 continue;
475
476 if (PCI_SLOT(sbridge_dev->pdev[i]->devfn) == slot &&
477 PCI_FUNC(sbridge_dev->pdev[i]->devfn) == func) {
478 debugf1("Associated %02x.%02x.%d with %p\n",
479 bus, slot, func, sbridge_dev->pdev[i]);
480 return sbridge_dev->pdev[i];
481 }
482 }
483
484 return NULL;
485}
486
487/**
Mauro Carvalho Chehabc36e3e72012-04-16 15:12:22 -0300488 * check_if_ecc_is_active() - Checks if ECC is active
Mauro Carvalho Chehabeebf11a2011-10-20 19:18:01 -0200489 * bus: Device bus
Mauro Carvalho Chehabeebf11a2011-10-20 19:18:01 -0200490 */
Mauro Carvalho Chehabc36e3e72012-04-16 15:12:22 -0300491static int check_if_ecc_is_active(const u8 bus)
Mauro Carvalho Chehabeebf11a2011-10-20 19:18:01 -0200492{
493 struct pci_dev *pdev = NULL;
Mauro Carvalho Chehabeebf11a2011-10-20 19:18:01 -0200494 u32 mcmtr;
495
Mauro Carvalho Chehabeebf11a2011-10-20 19:18:01 -0200496 pdev = get_pdev_slot_func(bus, 15, 0);
497 if (!pdev) {
498 sbridge_printk(KERN_ERR, "Couldn't find PCI device "
499 "%2x.%02d.%d!!!\n",
500 bus, 15, 0);
501 return -ENODEV;
502 }
503
504 pci_read_config_dword(pdev, MCMTR, &mcmtr);
505 if (!IS_ECC_ENABLED(mcmtr)) {
506 sbridge_printk(KERN_ERR, "ECC is disabled. Aborting\n");
507 return -ENODEV;
508 }
Mauro Carvalho Chehabeebf11a2011-10-20 19:18:01 -0200509 return 0;
510}
511
Mauro Carvalho Chehab084a4fc2012-01-27 18:38:08 -0300512static int get_dimm_config(struct mem_ctl_info *mci)
Mauro Carvalho Chehabeebf11a2011-10-20 19:18:01 -0200513{
514 struct sbridge_pvt *pvt = mci->pvt_info;
Mauro Carvalho Chehabc36e3e72012-04-16 15:12:22 -0300515 struct dimm_info *dimm;
Mauro Carvalho Chehabeebf11a2011-10-20 19:18:01 -0200516 int i, j, banks, ranks, rows, cols, size, npages;
Mauro Carvalho Chehabeebf11a2011-10-20 19:18:01 -0200517 u32 reg;
518 enum edac_type mode;
Mark A. Grondonac6e13b52011-10-18 11:02:58 -0200519 enum mem_type mtype;
Mauro Carvalho Chehabeebf11a2011-10-20 19:18:01 -0200520
521 pci_read_config_dword(pvt->pci_br, SAD_TARGET, &reg);
522 pvt->sbridge_dev->source_id = SOURCE_ID(reg);
523
524 pci_read_config_dword(pvt->pci_br, SAD_CONTROL, &reg);
525 pvt->sbridge_dev->node_id = NODE_ID(reg);
526 debugf0("mc#%d: Node ID: %d, source ID: %d\n",
527 pvt->sbridge_dev->mc,
528 pvt->sbridge_dev->node_id,
529 pvt->sbridge_dev->source_id);
530
531 pci_read_config_dword(pvt->pci_ras, RASENABLES, &reg);
532 if (IS_MIRROR_ENABLED(reg)) {
533 debugf0("Memory mirror is enabled\n");
534 pvt->is_mirrored = true;
535 } else {
536 debugf0("Memory mirror is disabled\n");
537 pvt->is_mirrored = false;
538 }
539
540 pci_read_config_dword(pvt->pci_ta, MCMTR, &pvt->info.mcmtr);
541 if (IS_LOCKSTEP_ENABLED(pvt->info.mcmtr)) {
542 debugf0("Lockstep is enabled\n");
543 mode = EDAC_S8ECD8ED;
544 pvt->is_lockstep = true;
545 } else {
546 debugf0("Lockstep is disabled\n");
547 mode = EDAC_S4ECD4ED;
548 pvt->is_lockstep = false;
549 }
550 if (IS_CLOSE_PG(pvt->info.mcmtr)) {
551 debugf0("address map is on closed page mode\n");
552 pvt->is_close_pg = true;
553 } else {
554 debugf0("address map is on open page mode\n");
555 pvt->is_close_pg = false;
556 }
557
558 pci_read_config_dword(pvt->pci_ta, RANK_CFG_A, &reg);
559 if (IS_RDIMM_ENABLED(reg)) {
560 /* FIXME: Can also be LRDIMM */
561 debugf0("Memory is registered\n");
Mark A. Grondonac6e13b52011-10-18 11:02:58 -0200562 mtype = MEM_RDDR3;
Mauro Carvalho Chehabeebf11a2011-10-20 19:18:01 -0200563 } else {
564 debugf0("Memory is unregistered\n");
Mark A. Grondonac6e13b52011-10-18 11:02:58 -0200565 mtype = MEM_DDR3;
Mauro Carvalho Chehabeebf11a2011-10-20 19:18:01 -0200566 }
567
568 /* On all supported DDR3 DIMM types, there are 8 banks available */
569 banks = 8;
570
571 for (i = 0; i < NUM_CHANNELS; i++) {
572 u32 mtr;
573
574 for (j = 0; j < ARRAY_SIZE(mtr_regs); j++) {
Mauro Carvalho Chehabc36e3e72012-04-16 15:12:22 -0300575 dimm = EDAC_DIMM_PTR(mci->layers, mci->dimms, mci->n_layers,
576 i, j, 0);
Mauro Carvalho Chehabeebf11a2011-10-20 19:18:01 -0200577 pci_read_config_dword(pvt->pci_tad[i],
578 mtr_regs[j], &mtr);
579 debugf4("Channel #%d MTR%d = %x\n", i, j, mtr);
580 if (IS_DIMM_PRESENT(mtr)) {
581 pvt->channel[i].dimms++;
582
583 ranks = numrank(mtr);
584 rows = numrow(mtr);
585 cols = numcol(mtr);
586
587 /* DDR3 has 8 I/O banks */
588 size = (rows * cols * banks * ranks) >> (20 - 3);
589 npages = MiB_TO_PAGES(size);
590
591 debugf0("mc#%d: channel %d, dimm %d, %d Mb (%d pages) bank: %d, rank: %d, row: %#x, col: %#x\n",
592 pvt->sbridge_dev->mc, i, j,
593 size, npages,
594 banks, ranks, rows, cols);
Mauro Carvalho Chehabeebf11a2011-10-20 19:18:01 -0200595
Mauro Carvalho Chehaba895bf82012-01-28 09:09:38 -0300596 dimm->nr_pages = npages;
Mauro Carvalho Chehab084a4fc2012-01-27 18:38:08 -0300597 dimm->grain = 32;
598 dimm->dtype = (banks == 8) ? DEV_X8 : DEV_X4;
599 dimm->mtype = mtype;
600 dimm->edac_mode = mode;
601 snprintf(dimm->label, sizeof(dimm->label),
Mauro Carvalho Chehabeebf11a2011-10-20 19:18:01 -0200602 "CPU_SrcID#%u_Channel#%u_DIMM#%u",
603 pvt->sbridge_dev->source_id, i, j);
Mauro Carvalho Chehabeebf11a2011-10-20 19:18:01 -0200604 }
605 }
606 }
607
608 return 0;
609}
610
611static void get_memory_layout(const struct mem_ctl_info *mci)
612{
613 struct sbridge_pvt *pvt = mci->pvt_info;
614 int i, j, k, n_sads, n_tads, sad_interl;
615 u32 reg;
616 u64 limit, prv = 0;
617 u64 tmp_mb;
Mauro Carvalho Chehab5b889e32011-11-07 18:26:53 -0300618 u32 mb, kb;
Mauro Carvalho Chehabeebf11a2011-10-20 19:18:01 -0200619 u32 rir_way;
620
621 /*
622 * Step 1) Get TOLM/TOHM ranges
623 */
624
625 /* Address range is 32:28 */
626 pci_read_config_dword(pvt->pci_sad1, TOLM,
627 &reg);
628 pvt->tolm = GET_TOLM(reg);
629 tmp_mb = (1 + pvt->tolm) >> 20;
630
Mauro Carvalho Chehab5b889e32011-11-07 18:26:53 -0300631 mb = div_u64_rem(tmp_mb, 1000, &kb);
632 debugf0("TOLM: %u.%03u GB (0x%016Lx)\n",
633 mb, kb, (u64)pvt->tolm);
Mauro Carvalho Chehabeebf11a2011-10-20 19:18:01 -0200634
635 /* Address range is already 45:25 */
636 pci_read_config_dword(pvt->pci_sad1, TOHM,
637 &reg);
638 pvt->tohm = GET_TOHM(reg);
639 tmp_mb = (1 + pvt->tohm) >> 20;
640
Mauro Carvalho Chehab5b889e32011-11-07 18:26:53 -0300641 mb = div_u64_rem(tmp_mb, 1000, &kb);
642 debugf0("TOHM: %u.%03u GB (0x%016Lx)",
643 mb, kb, (u64)pvt->tohm);
Mauro Carvalho Chehabeebf11a2011-10-20 19:18:01 -0200644
645 /*
646 * Step 2) Get SAD range and SAD Interleave list
647 * TAD registers contain the interleave wayness. However, it
648 * seems simpler to just discover it indirectly, with the
649 * algorithm bellow.
650 */
651 prv = 0;
652 for (n_sads = 0; n_sads < MAX_SAD; n_sads++) {
653 /* SAD_LIMIT Address range is 45:26 */
654 pci_read_config_dword(pvt->pci_sad0, dram_rule[n_sads],
655 &reg);
656 limit = SAD_LIMIT(reg);
657
658 if (!DRAM_RULE_ENABLE(reg))
659 continue;
660
661 if (limit <= prv)
662 break;
663
664 tmp_mb = (limit + 1) >> 20;
Mauro Carvalho Chehab5b889e32011-11-07 18:26:53 -0300665 mb = div_u64_rem(tmp_mb, 1000, &kb);
666 debugf0("SAD#%d %s up to %u.%03u GB (0x%016Lx) %s reg=0x%08x\n",
Mauro Carvalho Chehabeebf11a2011-10-20 19:18:01 -0200667 n_sads,
668 get_dram_attr(reg),
Mauro Carvalho Chehab5b889e32011-11-07 18:26:53 -0300669 mb, kb,
Mauro Carvalho Chehabeebf11a2011-10-20 19:18:01 -0200670 ((u64)tmp_mb) << 20L,
671 INTERLEAVE_MODE(reg) ? "Interleave: 8:6" : "Interleave: [8:6]XOR[18:16]",
672 reg);
673 prv = limit;
674
675 pci_read_config_dword(pvt->pci_sad0, interleave_list[n_sads],
676 &reg);
677 sad_interl = sad_pkg(reg, 0);
678 for (j = 0; j < 8; j++) {
679 if (j > 0 && sad_interl == sad_pkg(reg, j))
680 break;
681
682 debugf0("SAD#%d, interleave #%d: %d\n",
683 n_sads, j, sad_pkg(reg, j));
684 }
685 }
686
687 /*
688 * Step 3) Get TAD range
689 */
690 prv = 0;
691 for (n_tads = 0; n_tads < MAX_TAD; n_tads++) {
692 pci_read_config_dword(pvt->pci_ha0, tad_dram_rule[n_tads],
693 &reg);
694 limit = TAD_LIMIT(reg);
695 if (limit <= prv)
696 break;
697 tmp_mb = (limit + 1) >> 20;
698
Mauro Carvalho Chehab5b889e32011-11-07 18:26:53 -0300699 mb = div_u64_rem(tmp_mb, 1000, &kb);
700 debugf0("TAD#%d: up to %u.%03u GB (0x%016Lx), socket interleave %d, memory interleave %d, TGT: %d, %d, %d, %d, reg=0x%08x\n",
701 n_tads, mb, kb,
Mauro Carvalho Chehabeebf11a2011-10-20 19:18:01 -0200702 ((u64)tmp_mb) << 20L,
703 (u32)TAD_SOCK(reg),
704 (u32)TAD_CH(reg),
705 (u32)TAD_TGT0(reg),
706 (u32)TAD_TGT1(reg),
707 (u32)TAD_TGT2(reg),
708 (u32)TAD_TGT3(reg),
709 reg);
Hui Wang7fae0db2012-02-06 04:11:01 -0300710 prv = limit;
Mauro Carvalho Chehabeebf11a2011-10-20 19:18:01 -0200711 }
712
713 /*
714 * Step 4) Get TAD offsets, per each channel
715 */
716 for (i = 0; i < NUM_CHANNELS; i++) {
717 if (!pvt->channel[i].dimms)
718 continue;
719 for (j = 0; j < n_tads; j++) {
720 pci_read_config_dword(pvt->pci_tad[i],
721 tad_ch_nilv_offset[j],
722 &reg);
723 tmp_mb = TAD_OFFSET(reg) >> 20;
Mauro Carvalho Chehab5b889e32011-11-07 18:26:53 -0300724 mb = div_u64_rem(tmp_mb, 1000, &kb);
725 debugf0("TAD CH#%d, offset #%d: %u.%03u GB (0x%016Lx), reg=0x%08x\n",
Mauro Carvalho Chehabeebf11a2011-10-20 19:18:01 -0200726 i, j,
Mauro Carvalho Chehab5b889e32011-11-07 18:26:53 -0300727 mb, kb,
Mauro Carvalho Chehabeebf11a2011-10-20 19:18:01 -0200728 ((u64)tmp_mb) << 20L,
729 reg);
730 }
731 }
732
733 /*
734 * Step 6) Get RIR Wayness/Limit, per each channel
735 */
736 for (i = 0; i < NUM_CHANNELS; i++) {
737 if (!pvt->channel[i].dimms)
738 continue;
739 for (j = 0; j < MAX_RIR_RANGES; j++) {
740 pci_read_config_dword(pvt->pci_tad[i],
741 rir_way_limit[j],
742 &reg);
743
744 if (!IS_RIR_VALID(reg))
745 continue;
746
747 tmp_mb = RIR_LIMIT(reg) >> 20;
748 rir_way = 1 << RIR_WAY(reg);
Mauro Carvalho Chehab5b889e32011-11-07 18:26:53 -0300749 mb = div_u64_rem(tmp_mb, 1000, &kb);
750 debugf0("CH#%d RIR#%d, limit: %u.%03u GB (0x%016Lx), way: %d, reg=0x%08x\n",
Mauro Carvalho Chehabeebf11a2011-10-20 19:18:01 -0200751 i, j,
Mauro Carvalho Chehab5b889e32011-11-07 18:26:53 -0300752 mb, kb,
Mauro Carvalho Chehabeebf11a2011-10-20 19:18:01 -0200753 ((u64)tmp_mb) << 20L,
754 rir_way,
755 reg);
756
757 for (k = 0; k < rir_way; k++) {
758 pci_read_config_dword(pvt->pci_tad[i],
759 rir_offset[j][k],
760 &reg);
761 tmp_mb = RIR_OFFSET(reg) << 6;
762
Mauro Carvalho Chehab5b889e32011-11-07 18:26:53 -0300763 mb = div_u64_rem(tmp_mb, 1000, &kb);
764 debugf0("CH#%d RIR#%d INTL#%d, offset %u.%03u GB (0x%016Lx), tgt: %d, reg=0x%08x\n",
Mauro Carvalho Chehabeebf11a2011-10-20 19:18:01 -0200765 i, j, k,
Mauro Carvalho Chehab5b889e32011-11-07 18:26:53 -0300766 mb, kb,
Mauro Carvalho Chehabeebf11a2011-10-20 19:18:01 -0200767 ((u64)tmp_mb) << 20L,
768 (u32)RIR_RNK_TGT(reg),
769 reg);
770 }
771 }
772 }
773}
774
775struct mem_ctl_info *get_mci_for_node_id(u8 node_id)
776{
777 struct sbridge_dev *sbridge_dev;
778
779 list_for_each_entry(sbridge_dev, &sbridge_edac_list, list) {
780 if (sbridge_dev->node_id == node_id)
781 return sbridge_dev->mci;
782 }
783 return NULL;
784}
785
786static int get_memory_error_data(struct mem_ctl_info *mci,
787 u64 addr,
788 u8 *socket,
789 long *channel_mask,
790 u8 *rank,
Mauro Carvalho Chehabe17a2f42a2012-05-11 11:41:45 -0300791 char **area_type, char *msg)
Mauro Carvalho Chehabeebf11a2011-10-20 19:18:01 -0200792{
793 struct mem_ctl_info *new_mci;
794 struct sbridge_pvt *pvt = mci->pvt_info;
Mauro Carvalho Chehabeebf11a2011-10-20 19:18:01 -0200795 int n_rir, n_sads, n_tads, sad_way, sck_xch;
796 int sad_interl, idx, base_ch;
797 int interleave_mode;
798 unsigned sad_interleave[MAX_INTERLEAVE];
799 u32 reg;
800 u8 ch_way,sck_way;
801 u32 tad_offset;
802 u32 rir_way;
Mauro Carvalho Chehab5b889e32011-11-07 18:26:53 -0300803 u32 mb, kb;
Mauro Carvalho Chehabeebf11a2011-10-20 19:18:01 -0200804 u64 ch_addr, offset, limit, prv = 0;
805
806
807 /*
808 * Step 0) Check if the address is at special memory ranges
809 * The check bellow is probably enough to fill all cases where
810 * the error is not inside a memory, except for the legacy
811 * range (e. g. VGA addresses). It is unlikely, however, that the
812 * memory controller would generate an error on that range.
813 */
Mauro Carvalho Chehab5b889e32011-11-07 18:26:53 -0300814 if ((addr > (u64) pvt->tolm) && (addr < (1LL << 32))) {
Mauro Carvalho Chehabeebf11a2011-10-20 19:18:01 -0200815 sprintf(msg, "Error at TOLM area, on addr 0x%08Lx", addr);
Mauro Carvalho Chehabeebf11a2011-10-20 19:18:01 -0200816 return -EINVAL;
817 }
818 if (addr >= (u64)pvt->tohm) {
819 sprintf(msg, "Error at MMIOH area, on addr 0x%016Lx", addr);
Mauro Carvalho Chehabeebf11a2011-10-20 19:18:01 -0200820 return -EINVAL;
821 }
822
823 /*
824 * Step 1) Get socket
825 */
826 for (n_sads = 0; n_sads < MAX_SAD; n_sads++) {
827 pci_read_config_dword(pvt->pci_sad0, dram_rule[n_sads],
828 &reg);
829
830 if (!DRAM_RULE_ENABLE(reg))
831 continue;
832
833 limit = SAD_LIMIT(reg);
834 if (limit <= prv) {
835 sprintf(msg, "Can't discover the memory socket");
Mauro Carvalho Chehabeebf11a2011-10-20 19:18:01 -0200836 return -EINVAL;
837 }
838 if (addr <= limit)
839 break;
840 prv = limit;
841 }
842 if (n_sads == MAX_SAD) {
843 sprintf(msg, "Can't discover the memory socket");
Mauro Carvalho Chehabeebf11a2011-10-20 19:18:01 -0200844 return -EINVAL;
845 }
Mauro Carvalho Chehabe17a2f42a2012-05-11 11:41:45 -0300846 *area_type = get_dram_attr(reg);
Mauro Carvalho Chehabeebf11a2011-10-20 19:18:01 -0200847 interleave_mode = INTERLEAVE_MODE(reg);
848
849 pci_read_config_dword(pvt->pci_sad0, interleave_list[n_sads],
850 &reg);
851 sad_interl = sad_pkg(reg, 0);
852 for (sad_way = 0; sad_way < 8; sad_way++) {
853 if (sad_way > 0 && sad_interl == sad_pkg(reg, sad_way))
854 break;
855 sad_interleave[sad_way] = sad_pkg(reg, sad_way);
856 debugf0("SAD interleave #%d: %d\n",
857 sad_way, sad_interleave[sad_way]);
858 }
859 debugf0("mc#%d: Error detected on SAD#%d: address 0x%016Lx < 0x%016Lx, Interleave [%d:6]%s\n",
860 pvt->sbridge_dev->mc,
861 n_sads,
862 addr,
863 limit,
864 sad_way + 7,
Hui Wangad9c40b2012-02-06 04:11:00 -0300865 interleave_mode ? "" : "XOR[18:16]");
Mauro Carvalho Chehabeebf11a2011-10-20 19:18:01 -0200866 if (interleave_mode)
867 idx = ((addr >> 6) ^ (addr >> 16)) & 7;
868 else
869 idx = (addr >> 6) & 7;
870 switch (sad_way) {
871 case 1:
872 idx = 0;
873 break;
874 case 2:
875 idx = idx & 1;
876 break;
877 case 4:
878 idx = idx & 3;
879 break;
880 case 8:
881 break;
882 default:
883 sprintf(msg, "Can't discover socket interleave");
Mauro Carvalho Chehabeebf11a2011-10-20 19:18:01 -0200884 return -EINVAL;
885 }
886 *socket = sad_interleave[idx];
887 debugf0("SAD interleave index: %d (wayness %d) = CPU socket %d\n",
888 idx, sad_way, *socket);
889
890 /*
891 * Move to the proper node structure, in order to access the
892 * right PCI registers
893 */
894 new_mci = get_mci_for_node_id(*socket);
895 if (!new_mci) {
896 sprintf(msg, "Struct for socket #%u wasn't initialized",
897 *socket);
Mauro Carvalho Chehabeebf11a2011-10-20 19:18:01 -0200898 return -EINVAL;
899 }
900 mci = new_mci;
901 pvt = mci->pvt_info;
902
903 /*
904 * Step 2) Get memory channel
905 */
906 prv = 0;
907 for (n_tads = 0; n_tads < MAX_TAD; n_tads++) {
908 pci_read_config_dword(pvt->pci_ha0, tad_dram_rule[n_tads],
909 &reg);
910 limit = TAD_LIMIT(reg);
911 if (limit <= prv) {
912 sprintf(msg, "Can't discover the memory channel");
Mauro Carvalho Chehabeebf11a2011-10-20 19:18:01 -0200913 return -EINVAL;
914 }
915 if (addr <= limit)
916 break;
917 prv = limit;
918 }
919 ch_way = TAD_CH(reg) + 1;
920 sck_way = TAD_SOCK(reg) + 1;
921 /*
922 * FIXME: Is it right to always use channel 0 for offsets?
923 */
924 pci_read_config_dword(pvt->pci_tad[0],
925 tad_ch_nilv_offset[n_tads],
926 &tad_offset);
927
928 if (ch_way == 3)
929 idx = addr >> 6;
930 else
931 idx = addr >> (6 + sck_way);
932 idx = idx % ch_way;
933
934 /*
935 * FIXME: Shouldn't we use CHN_IDX_OFFSET() here, when ch_way == 3 ???
936 */
937 switch (idx) {
938 case 0:
939 base_ch = TAD_TGT0(reg);
940 break;
941 case 1:
942 base_ch = TAD_TGT1(reg);
943 break;
944 case 2:
945 base_ch = TAD_TGT2(reg);
946 break;
947 case 3:
948 base_ch = TAD_TGT3(reg);
949 break;
950 default:
951 sprintf(msg, "Can't discover the TAD target");
Mauro Carvalho Chehabeebf11a2011-10-20 19:18:01 -0200952 return -EINVAL;
953 }
954 *channel_mask = 1 << base_ch;
955
956 if (pvt->is_mirrored) {
957 *channel_mask |= 1 << ((base_ch + 2) % 4);
958 switch(ch_way) {
959 case 2:
960 case 4:
961 sck_xch = 1 << sck_way * (ch_way >> 1);
962 break;
963 default:
964 sprintf(msg, "Invalid mirror set. Can't decode addr");
Mauro Carvalho Chehabeebf11a2011-10-20 19:18:01 -0200965 return -EINVAL;
966 }
967 } else
968 sck_xch = (1 << sck_way) * ch_way;
969
970 if (pvt->is_lockstep)
971 *channel_mask |= 1 << ((base_ch + 1) % 4);
972
973 offset = TAD_OFFSET(tad_offset);
974
975 debugf0("TAD#%d: address 0x%016Lx < 0x%016Lx, socket interleave %d, channel interleave %d (offset 0x%08Lx), index %d, base ch: %d, ch mask: 0x%02lx\n",
976 n_tads,
977 addr,
978 limit,
979 (u32)TAD_SOCK(reg),
980 ch_way,
981 offset,
982 idx,
983 base_ch,
984 *channel_mask);
985
986 /* Calculate channel address */
987 /* Remove the TAD offset */
988
989 if (offset > addr) {
990 sprintf(msg, "Can't calculate ch addr: TAD offset 0x%08Lx is too high for addr 0x%08Lx!",
991 offset, addr);
Mauro Carvalho Chehabeebf11a2011-10-20 19:18:01 -0200992 return -EINVAL;
993 }
994 addr -= offset;
995 /* Store the low bits [0:6] of the addr */
996 ch_addr = addr & 0x7f;
997 /* Remove socket wayness and remove 6 bits */
998 addr >>= 6;
Mauro Carvalho Chehab5b889e32011-11-07 18:26:53 -0300999 addr = div_u64(addr, sck_xch);
Mauro Carvalho Chehabeebf11a2011-10-20 19:18:01 -02001000#if 0
1001 /* Divide by channel way */
1002 addr = addr / ch_way;
1003#endif
1004 /* Recover the last 6 bits */
1005 ch_addr |= addr << 6;
1006
1007 /*
1008 * Step 3) Decode rank
1009 */
1010 for (n_rir = 0; n_rir < MAX_RIR_RANGES; n_rir++) {
1011 pci_read_config_dword(pvt->pci_tad[base_ch],
1012 rir_way_limit[n_rir],
1013 &reg);
1014
1015 if (!IS_RIR_VALID(reg))
1016 continue;
1017
1018 limit = RIR_LIMIT(reg);
Mauro Carvalho Chehab5b889e32011-11-07 18:26:53 -03001019 mb = div_u64_rem(limit >> 20, 1000, &kb);
1020 debugf0("RIR#%d, limit: %u.%03u GB (0x%016Lx), way: %d\n",
Mauro Carvalho Chehabeebf11a2011-10-20 19:18:01 -02001021 n_rir,
Mauro Carvalho Chehab5b889e32011-11-07 18:26:53 -03001022 mb, kb,
Mauro Carvalho Chehabeebf11a2011-10-20 19:18:01 -02001023 limit,
1024 1 << RIR_WAY(reg));
1025 if (ch_addr <= limit)
1026 break;
1027 }
1028 if (n_rir == MAX_RIR_RANGES) {
1029 sprintf(msg, "Can't discover the memory rank for ch addr 0x%08Lx",
1030 ch_addr);
Mauro Carvalho Chehabeebf11a2011-10-20 19:18:01 -02001031 return -EINVAL;
1032 }
1033 rir_way = RIR_WAY(reg);
1034 if (pvt->is_close_pg)
1035 idx = (ch_addr >> 6);
1036 else
1037 idx = (ch_addr >> 13); /* FIXME: Datasheet says to shift by 15 */
1038 idx %= 1 << rir_way;
1039
1040 pci_read_config_dword(pvt->pci_tad[base_ch],
1041 rir_offset[n_rir][idx],
1042 &reg);
1043 *rank = RIR_RNK_TGT(reg);
1044
1045 debugf0("RIR#%d: channel address 0x%08Lx < 0x%08Lx, RIR interleave %d, index %d\n",
1046 n_rir,
1047 ch_addr,
1048 limit,
1049 rir_way,
1050 idx);
1051
1052 return 0;
1053}
1054
1055/****************************************************************************
1056 Device initialization routines: put/get, init/exit
1057 ****************************************************************************/
1058
1059/*
1060 * sbridge_put_all_devices 'put' all the devices that we have
1061 * reserved via 'get'
1062 */
1063static void sbridge_put_devices(struct sbridge_dev *sbridge_dev)
1064{
1065 int i;
1066
1067 debugf0(__FILE__ ": %s()\n", __func__);
1068 for (i = 0; i < sbridge_dev->n_devs; i++) {
1069 struct pci_dev *pdev = sbridge_dev->pdev[i];
1070 if (!pdev)
1071 continue;
1072 debugf0("Removing dev %02x:%02x.%d\n",
1073 pdev->bus->number,
1074 PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn));
1075 pci_dev_put(pdev);
1076 }
1077}
1078
1079static void sbridge_put_all_devices(void)
1080{
1081 struct sbridge_dev *sbridge_dev, *tmp;
1082
1083 list_for_each_entry_safe(sbridge_dev, tmp, &sbridge_edac_list, list) {
1084 sbridge_put_devices(sbridge_dev);
1085 free_sbridge_dev(sbridge_dev);
1086 }
1087}
1088
1089/*
1090 * sbridge_get_all_devices Find and perform 'get' operation on the MCH's
1091 * device/functions we want to reference for this driver
1092 *
1093 * Need to 'get' device 16 func 1 and func 2
1094 */
1095static int sbridge_get_onedevice(struct pci_dev **prev,
1096 u8 *num_mc,
1097 const struct pci_id_table *table,
1098 const unsigned devno)
1099{
1100 struct sbridge_dev *sbridge_dev;
1101 const struct pci_id_descr *dev_descr = &table->descr[devno];
1102
1103 struct pci_dev *pdev = NULL;
1104 u8 bus = 0;
1105
1106 sbridge_printk(KERN_INFO,
1107 "Seeking for: dev %02x.%d PCI ID %04x:%04x\n",
1108 dev_descr->dev, dev_descr->func,
1109 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
1110
1111 pdev = pci_get_device(PCI_VENDOR_ID_INTEL,
1112 dev_descr->dev_id, *prev);
1113
1114 if (!pdev) {
1115 if (*prev) {
1116 *prev = pdev;
1117 return 0;
1118 }
1119
1120 if (dev_descr->optional)
1121 return 0;
1122
1123 if (devno == 0)
1124 return -ENODEV;
1125
1126 sbridge_printk(KERN_INFO,
1127 "Device not found: dev %02x.%d PCI ID %04x:%04x\n",
1128 dev_descr->dev, dev_descr->func,
1129 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
1130
1131 /* End of list, leave */
1132 return -ENODEV;
1133 }
1134 bus = pdev->bus->number;
1135
1136 sbridge_dev = get_sbridge_dev(bus);
1137 if (!sbridge_dev) {
1138 sbridge_dev = alloc_sbridge_dev(bus, table);
1139 if (!sbridge_dev) {
1140 pci_dev_put(pdev);
1141 return -ENOMEM;
1142 }
1143 (*num_mc)++;
1144 }
1145
1146 if (sbridge_dev->pdev[devno]) {
1147 sbridge_printk(KERN_ERR,
1148 "Duplicated device for "
1149 "dev %02x:%d.%d PCI ID %04x:%04x\n",
1150 bus, dev_descr->dev, dev_descr->func,
1151 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
1152 pci_dev_put(pdev);
1153 return -ENODEV;
1154 }
1155
1156 sbridge_dev->pdev[devno] = pdev;
1157
1158 /* Sanity check */
1159 if (unlikely(PCI_SLOT(pdev->devfn) != dev_descr->dev ||
1160 PCI_FUNC(pdev->devfn) != dev_descr->func)) {
1161 sbridge_printk(KERN_ERR,
1162 "Device PCI ID %04x:%04x "
1163 "has dev %02x:%d.%d instead of dev %02x:%02x.%d\n",
1164 PCI_VENDOR_ID_INTEL, dev_descr->dev_id,
1165 bus, PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn),
1166 bus, dev_descr->dev, dev_descr->func);
1167 return -ENODEV;
1168 }
1169
1170 /* Be sure that the device is enabled */
1171 if (unlikely(pci_enable_device(pdev) < 0)) {
1172 sbridge_printk(KERN_ERR,
1173 "Couldn't enable "
1174 "dev %02x:%d.%d PCI ID %04x:%04x\n",
1175 bus, dev_descr->dev, dev_descr->func,
1176 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
1177 return -ENODEV;
1178 }
1179
1180 debugf0("Detected dev %02x:%d.%d PCI ID %04x:%04x\n",
1181 bus, dev_descr->dev,
1182 dev_descr->func,
1183 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
1184
1185 /*
1186 * As stated on drivers/pci/search.c, the reference count for
1187 * @from is always decremented if it is not %NULL. So, as we need
1188 * to get all devices up to null, we need to do a get for the device
1189 */
1190 pci_dev_get(pdev);
1191
1192 *prev = pdev;
1193
1194 return 0;
1195}
1196
1197static int sbridge_get_all_devices(u8 *num_mc)
1198{
1199 int i, rc;
1200 struct pci_dev *pdev = NULL;
1201 const struct pci_id_table *table = pci_dev_descr_sbridge_table;
1202
1203 while (table && table->descr) {
1204 for (i = 0; i < table->n_devs; i++) {
1205 pdev = NULL;
1206 do {
1207 rc = sbridge_get_onedevice(&pdev, num_mc,
1208 table, i);
1209 if (rc < 0) {
1210 if (i == 0) {
1211 i = table->n_devs;
1212 break;
1213 }
1214 sbridge_put_all_devices();
1215 return -ENODEV;
1216 }
1217 } while (pdev);
1218 }
1219 table++;
1220 }
1221
1222 return 0;
1223}
1224
1225static int mci_bind_devs(struct mem_ctl_info *mci,
1226 struct sbridge_dev *sbridge_dev)
1227{
1228 struct sbridge_pvt *pvt = mci->pvt_info;
1229 struct pci_dev *pdev;
1230 int i, func, slot;
1231
1232 for (i = 0; i < sbridge_dev->n_devs; i++) {
1233 pdev = sbridge_dev->pdev[i];
1234 if (!pdev)
1235 continue;
1236 slot = PCI_SLOT(pdev->devfn);
1237 func = PCI_FUNC(pdev->devfn);
1238 switch (slot) {
1239 case 12:
1240 switch (func) {
1241 case 6:
1242 pvt->pci_sad0 = pdev;
1243 break;
1244 case 7:
1245 pvt->pci_sad1 = pdev;
1246 break;
1247 default:
1248 goto error;
1249 }
1250 break;
1251 case 13:
1252 switch (func) {
1253 case 6:
1254 pvt->pci_br = pdev;
1255 break;
1256 default:
1257 goto error;
1258 }
1259 break;
1260 case 14:
1261 switch (func) {
1262 case 0:
1263 pvt->pci_ha0 = pdev;
1264 break;
1265 default:
1266 goto error;
1267 }
1268 break;
1269 case 15:
1270 switch (func) {
1271 case 0:
1272 pvt->pci_ta = pdev;
1273 break;
1274 case 1:
1275 pvt->pci_ras = pdev;
1276 break;
1277 case 2:
1278 case 3:
1279 case 4:
1280 case 5:
1281 pvt->pci_tad[func - 2] = pdev;
1282 break;
1283 default:
1284 goto error;
1285 }
1286 break;
1287 case 17:
1288 switch (func) {
1289 case 0:
1290 pvt->pci_ddrio = pdev;
1291 break;
1292 default:
1293 goto error;
1294 }
1295 break;
1296 default:
1297 goto error;
1298 }
1299
1300 debugf0("Associated PCI %02x.%02d.%d with dev = %p\n",
1301 sbridge_dev->bus,
1302 PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn),
1303 pdev);
1304 }
1305
1306 /* Check if everything were registered */
1307 if (!pvt->pci_sad0 || !pvt->pci_sad1 || !pvt->pci_ha0 ||
1308 !pvt-> pci_tad || !pvt->pci_ras || !pvt->pci_ta ||
1309 !pvt->pci_ddrio)
1310 goto enodev;
1311
1312 for (i = 0; i < NUM_CHANNELS; i++) {
1313 if (!pvt->pci_tad[i])
1314 goto enodev;
1315 }
1316 return 0;
1317
1318enodev:
1319 sbridge_printk(KERN_ERR, "Some needed devices are missing\n");
1320 return -ENODEV;
1321
1322error:
1323 sbridge_printk(KERN_ERR, "Device %d, function %d "
1324 "is out of the expected range\n",
1325 slot, func);
1326 return -EINVAL;
1327}
1328
1329/****************************************************************************
1330 Error check routines
1331 ****************************************************************************/
1332
1333/*
1334 * While Sandy Bridge has error count registers, SMI BIOS read values from
1335 * and resets the counters. So, they are not reliable for the OS to read
1336 * from them. So, we have no option but to just trust on whatever MCE is
1337 * telling us about the errors.
1338 */
1339static void sbridge_mce_output_error(struct mem_ctl_info *mci,
1340 const struct mce *m)
1341{
1342 struct mem_ctl_info *new_mci;
1343 struct sbridge_pvt *pvt = mci->pvt_info;
Mauro Carvalho Chehabc36e3e72012-04-16 15:12:22 -03001344 enum hw_event_mc_err_type tp_event;
Mauro Carvalho Chehabe17a2f42a2012-05-11 11:41:45 -03001345 char *type, *optype, msg[256];
Mauro Carvalho Chehabeebf11a2011-10-20 19:18:01 -02001346 bool ripv = GET_BITFIELD(m->mcgstatus, 0, 0);
1347 bool overflow = GET_BITFIELD(m->status, 62, 62);
1348 bool uncorrected_error = GET_BITFIELD(m->status, 61, 61);
1349 bool recoverable = GET_BITFIELD(m->status, 56, 56);
1350 u32 core_err_cnt = GET_BITFIELD(m->status, 38, 52);
1351 u32 mscod = GET_BITFIELD(m->status, 16, 31);
1352 u32 errcode = GET_BITFIELD(m->status, 0, 15);
1353 u32 channel = GET_BITFIELD(m->status, 0, 3);
1354 u32 optypenum = GET_BITFIELD(m->status, 4, 6);
1355 long channel_mask, first_channel;
1356 u8 rank, socket;
Mauro Carvalho Chehabc36e3e72012-04-16 15:12:22 -03001357 int rc, dimm;
Mauro Carvalho Chehabe17a2f42a2012-05-11 11:41:45 -03001358 char *area_type = NULL;
Mauro Carvalho Chehabeebf11a2011-10-20 19:18:01 -02001359
Mauro Carvalho Chehabc36e3e72012-04-16 15:12:22 -03001360 if (uncorrected_error) {
1361 if (ripv) {
1362 type = "FATAL";
1363 tp_event = HW_EVENT_ERR_FATAL;
1364 } else {
1365 type = "NON_FATAL";
1366 tp_event = HW_EVENT_ERR_UNCORRECTED;
1367 }
1368 } else {
1369 type = "CORRECTED";
1370 tp_event = HW_EVENT_ERR_CORRECTED;
1371 }
Mauro Carvalho Chehabeebf11a2011-10-20 19:18:01 -02001372
1373 /*
David Mackey15ed1032012-04-17 11:30:52 -07001374 * According with Table 15-9 of the Intel Architecture spec vol 3A,
Mauro Carvalho Chehabeebf11a2011-10-20 19:18:01 -02001375 * memory errors should fit in this mask:
1376 * 000f 0000 1mmm cccc (binary)
1377 * where:
1378 * f = Correction Report Filtering Bit. If 1, subsequent errors
1379 * won't be shown
1380 * mmm = error type
1381 * cccc = channel
1382 * If the mask doesn't match, report an error to the parsing logic
1383 */
1384 if (! ((errcode & 0xef80) == 0x80)) {
1385 optype = "Can't parse: it is not a mem";
1386 } else {
1387 switch (optypenum) {
1388 case 0:
Mauro Carvalho Chehabc36e3e72012-04-16 15:12:22 -03001389 optype = "generic undef request error";
Mauro Carvalho Chehabeebf11a2011-10-20 19:18:01 -02001390 break;
1391 case 1:
Mauro Carvalho Chehabc36e3e72012-04-16 15:12:22 -03001392 optype = "memory read error";
Mauro Carvalho Chehabeebf11a2011-10-20 19:18:01 -02001393 break;
1394 case 2:
Mauro Carvalho Chehabc36e3e72012-04-16 15:12:22 -03001395 optype = "memory write error";
Mauro Carvalho Chehabeebf11a2011-10-20 19:18:01 -02001396 break;
1397 case 3:
Mauro Carvalho Chehabc36e3e72012-04-16 15:12:22 -03001398 optype = "addr/cmd error";
Mauro Carvalho Chehabeebf11a2011-10-20 19:18:01 -02001399 break;
1400 case 4:
Mauro Carvalho Chehabc36e3e72012-04-16 15:12:22 -03001401 optype = "memory scrubbing error";
Mauro Carvalho Chehabeebf11a2011-10-20 19:18:01 -02001402 break;
1403 default:
1404 optype = "reserved";
1405 break;
1406 }
1407 }
1408
1409 rc = get_memory_error_data(mci, m->addr, &socket,
Mauro Carvalho Chehabe17a2f42a2012-05-11 11:41:45 -03001410 &channel_mask, &rank, &area_type, msg);
Mauro Carvalho Chehabeebf11a2011-10-20 19:18:01 -02001411 if (rc < 0)
Mauro Carvalho Chehabc36e3e72012-04-16 15:12:22 -03001412 goto err_parsing;
Mauro Carvalho Chehabeebf11a2011-10-20 19:18:01 -02001413 new_mci = get_mci_for_node_id(socket);
1414 if (!new_mci) {
Mauro Carvalho Chehabc36e3e72012-04-16 15:12:22 -03001415 strcpy(msg, "Error: socket got corrupted!");
1416 goto err_parsing;
Mauro Carvalho Chehabeebf11a2011-10-20 19:18:01 -02001417 }
1418 mci = new_mci;
1419 pvt = mci->pvt_info;
1420
1421 first_channel = find_first_bit(&channel_mask, NUM_CHANNELS);
1422
1423 if (rank < 4)
1424 dimm = 0;
1425 else if (rank < 8)
1426 dimm = 1;
1427 else
1428 dimm = 2;
1429
Mauro Carvalho Chehabeebf11a2011-10-20 19:18:01 -02001430
1431 /*
Mauro Carvalho Chehabe17a2f42a2012-05-11 11:41:45 -03001432 * FIXME: On some memory configurations (mirror, lockstep), the
1433 * Memory Controller can't point the error to a single DIMM. The
1434 * EDAC core should be handling the channel mask, in order to point
1435 * to the group of dimm's where the error may be happening.
Mauro Carvalho Chehabeebf11a2011-10-20 19:18:01 -02001436 */
Mauro Carvalho Chehabc36e3e72012-04-16 15:12:22 -03001437 snprintf(msg, sizeof(msg),
Mauro Carvalho Chehabe17a2f42a2012-05-11 11:41:45 -03001438 "count:%d%s%s area:%s err_code:%04x:%04x socket:%d channel_mask:%ld rank:%d",
1439 core_err_cnt,
1440 overflow ? " OVERFLOW" : "",
1441 (uncorrected_error && recoverable) ? " recoverable" : "",
1442 area_type,
1443 mscod, errcode,
1444 socket,
1445 channel_mask,
1446 rank);
Mauro Carvalho Chehabeebf11a2011-10-20 19:18:01 -02001447
1448 debugf0("%s", msg);
1449
Mauro Carvalho Chehabc36e3e72012-04-16 15:12:22 -03001450 /* FIXME: need support for channel mask */
Mauro Carvalho Chehabeebf11a2011-10-20 19:18:01 -02001451
Mauro Carvalho Chehabc36e3e72012-04-16 15:12:22 -03001452 /* Call the helper to output message */
1453 edac_mc_handle_error(tp_event, mci,
1454 m->addr >> PAGE_SHIFT, m->addr & ~PAGE_MASK, 0,
1455 channel, dimm, -1,
1456 optype, msg, m);
1457 return;
1458err_parsing:
1459 edac_mc_handle_error(tp_event, mci, 0, 0, 0,
1460 -1, -1, -1,
1461 msg, "", m);
1462
Mauro Carvalho Chehabeebf11a2011-10-20 19:18:01 -02001463}
1464
1465/*
1466 * sbridge_check_error Retrieve and process errors reported by the
1467 * hardware. Called by the Core module.
1468 */
1469static void sbridge_check_error(struct mem_ctl_info *mci)
1470{
1471 struct sbridge_pvt *pvt = mci->pvt_info;
1472 int i;
1473 unsigned count = 0;
1474 struct mce *m;
1475
1476 /*
1477 * MCE first step: Copy all mce errors into a temporary buffer
1478 * We use a double buffering here, to reduce the risk of
1479 * loosing an error.
1480 */
1481 smp_rmb();
1482 count = (pvt->mce_out + MCE_LOG_LEN - pvt->mce_in)
1483 % MCE_LOG_LEN;
1484 if (!count)
1485 return;
1486
1487 m = pvt->mce_outentry;
1488 if (pvt->mce_in + count > MCE_LOG_LEN) {
1489 unsigned l = MCE_LOG_LEN - pvt->mce_in;
1490
1491 memcpy(m, &pvt->mce_entry[pvt->mce_in], sizeof(*m) * l);
1492 smp_wmb();
1493 pvt->mce_in = 0;
1494 count -= l;
1495 m += l;
1496 }
1497 memcpy(m, &pvt->mce_entry[pvt->mce_in], sizeof(*m) * count);
1498 smp_wmb();
1499 pvt->mce_in += count;
1500
1501 smp_rmb();
1502 if (pvt->mce_overrun) {
1503 sbridge_printk(KERN_ERR, "Lost %d memory errors\n",
1504 pvt->mce_overrun);
1505 smp_wmb();
1506 pvt->mce_overrun = 0;
1507 }
1508
1509 /*
1510 * MCE second step: parse errors and display
1511 */
1512 for (i = 0; i < count; i++)
1513 sbridge_mce_output_error(mci, &pvt->mce_outentry[i]);
1514}
1515
1516/*
1517 * sbridge_mce_check_error Replicates mcelog routine to get errors
1518 * This routine simply queues mcelog errors, and
1519 * return. The error itself should be handled later
1520 * by sbridge_check_error.
1521 * WARNING: As this routine should be called at NMI time, extra care should
1522 * be taken to avoid deadlocks, and to be as fast as possible.
1523 */
Mauro Carvalho Chehab3d78c9a2011-10-20 19:33:46 -02001524static int sbridge_mce_check_error(struct notifier_block *nb, unsigned long val,
1525 void *data)
Mauro Carvalho Chehabeebf11a2011-10-20 19:18:01 -02001526{
Mauro Carvalho Chehab3d78c9a2011-10-20 19:33:46 -02001527 struct mce *mce = (struct mce *)data;
1528 struct mem_ctl_info *mci;
1529 struct sbridge_pvt *pvt;
1530
1531 mci = get_mci_for_node_id(mce->socketid);
1532 if (!mci)
1533 return NOTIFY_BAD;
1534 pvt = mci->pvt_info;
Mauro Carvalho Chehabeebf11a2011-10-20 19:18:01 -02001535
1536 /*
1537 * Just let mcelog handle it if the error is
1538 * outside the memory controller. A memory error
1539 * is indicated by bit 7 = 1 and bits = 8-11,13-15 = 0.
1540 * bit 12 has an special meaning.
1541 */
1542 if ((mce->status & 0xefff) >> 7 != 1)
Mauro Carvalho Chehab3d78c9a2011-10-20 19:33:46 -02001543 return NOTIFY_DONE;
Mauro Carvalho Chehabeebf11a2011-10-20 19:18:01 -02001544
1545 printk("sbridge: HANDLING MCE MEMORY ERROR\n");
1546
1547 printk("CPU %d: Machine Check Exception: %Lx Bank %d: %016Lx\n",
1548 mce->extcpu, mce->mcgstatus, mce->bank, mce->status);
1549 printk("TSC %llx ", mce->tsc);
1550 printk("ADDR %llx ", mce->addr);
1551 printk("MISC %llx ", mce->misc);
1552
1553 printk("PROCESSOR %u:%x TIME %llu SOCKET %u APIC %x\n",
1554 mce->cpuvendor, mce->cpuid, mce->time,
1555 mce->socketid, mce->apicid);
1556
Mauro Carvalho Chehabeebf11a2011-10-20 19:18:01 -02001557 /* Only handle if it is the right mc controller */
1558 if (cpu_data(mce->cpu).phys_proc_id != pvt->sbridge_dev->mc)
Mauro Carvalho Chehab3d78c9a2011-10-20 19:33:46 -02001559 return NOTIFY_DONE;
Mauro Carvalho Chehabeebf11a2011-10-20 19:18:01 -02001560
1561 smp_rmb();
1562 if ((pvt->mce_out + 1) % MCE_LOG_LEN == pvt->mce_in) {
1563 smp_wmb();
1564 pvt->mce_overrun++;
Mauro Carvalho Chehab3d78c9a2011-10-20 19:33:46 -02001565 return NOTIFY_DONE;
Mauro Carvalho Chehabeebf11a2011-10-20 19:18:01 -02001566 }
1567
1568 /* Copy memory error at the ringbuffer */
1569 memcpy(&pvt->mce_entry[pvt->mce_out], mce, sizeof(*mce));
1570 smp_wmb();
1571 pvt->mce_out = (pvt->mce_out + 1) % MCE_LOG_LEN;
1572
1573 /* Handle fatal errors immediately */
1574 if (mce->mcgstatus & 1)
1575 sbridge_check_error(mci);
1576
1577 /* Advice mcelog that the error were handled */
Mauro Carvalho Chehab3d78c9a2011-10-20 19:33:46 -02001578 return NOTIFY_STOP;
Mauro Carvalho Chehabeebf11a2011-10-20 19:18:01 -02001579}
1580
Mauro Carvalho Chehab3d78c9a2011-10-20 19:33:46 -02001581static struct notifier_block sbridge_mce_dec = {
1582 .notifier_call = sbridge_mce_check_error,
1583};
1584
Mauro Carvalho Chehabeebf11a2011-10-20 19:18:01 -02001585/****************************************************************************
1586 EDAC register/unregister logic
1587 ****************************************************************************/
1588
1589static void sbridge_unregister_mci(struct sbridge_dev *sbridge_dev)
1590{
1591 struct mem_ctl_info *mci = sbridge_dev->mci;
1592 struct sbridge_pvt *pvt;
1593
1594 if (unlikely(!mci || !mci->pvt_info)) {
1595 debugf0("MC: " __FILE__ ": %s(): dev = %p\n",
1596 __func__, &sbridge_dev->pdev[0]->dev);
1597
1598 sbridge_printk(KERN_ERR, "Couldn't find mci handler\n");
1599 return;
1600 }
1601
1602 pvt = mci->pvt_info;
1603
1604 debugf0("MC: " __FILE__ ": %s(): mci = %p, dev = %p\n",
1605 __func__, mci, &sbridge_dev->pdev[0]->dev);
1606
Borislav Petkov3653ada2011-12-04 15:12:09 +01001607 mce_unregister_decode_chain(&sbridge_mce_dec);
Mauro Carvalho Chehabeebf11a2011-10-20 19:18:01 -02001608
1609 /* Remove MC sysfs nodes */
1610 edac_mc_del_mc(mci->dev);
1611
1612 debugf1("%s: free mci struct\n", mci->ctl_name);
1613 kfree(mci->ctl_name);
1614 edac_mc_free(mci);
1615 sbridge_dev->mci = NULL;
1616}
1617
1618static int sbridge_register_mci(struct sbridge_dev *sbridge_dev)
1619{
1620 struct mem_ctl_info *mci;
Mauro Carvalho Chehabc36e3e72012-04-16 15:12:22 -03001621 struct edac_mc_layer layers[2];
Mauro Carvalho Chehabeebf11a2011-10-20 19:18:01 -02001622 struct sbridge_pvt *pvt;
Mauro Carvalho Chehabc36e3e72012-04-16 15:12:22 -03001623 int rc;
Mauro Carvalho Chehabeebf11a2011-10-20 19:18:01 -02001624
1625 /* Check the number of active and not disabled channels */
Mauro Carvalho Chehabc36e3e72012-04-16 15:12:22 -03001626 rc = check_if_ecc_is_active(sbridge_dev->bus);
Mauro Carvalho Chehabeebf11a2011-10-20 19:18:01 -02001627 if (unlikely(rc < 0))
1628 return rc;
1629
1630 /* allocate a new MC control structure */
Mauro Carvalho Chehabc36e3e72012-04-16 15:12:22 -03001631 layers[0].type = EDAC_MC_LAYER_CHANNEL;
1632 layers[0].size = NUM_CHANNELS;
1633 layers[0].is_virt_csrow = false;
1634 layers[1].type = EDAC_MC_LAYER_SLOT;
1635 layers[1].size = MAX_DIMMS;
1636 layers[1].is_virt_csrow = true;
Mauro Carvalho Chehabca0907b2012-05-02 14:37:00 -03001637 mci = edac_mc_alloc(sbridge_dev->mc, ARRAY_SIZE(layers), layers,
Mauro Carvalho Chehabc36e3e72012-04-16 15:12:22 -03001638 sizeof(*pvt));
1639
Mauro Carvalho Chehabeebf11a2011-10-20 19:18:01 -02001640 if (unlikely(!mci))
1641 return -ENOMEM;
1642
1643 debugf0("MC: " __FILE__ ": %s(): mci = %p, dev = %p\n",
1644 __func__, mci, &sbridge_dev->pdev[0]->dev);
1645
1646 pvt = mci->pvt_info;
1647 memset(pvt, 0, sizeof(*pvt));
1648
1649 /* Associate sbridge_dev and mci for future usage */
1650 pvt->sbridge_dev = sbridge_dev;
1651 sbridge_dev->mci = mci;
1652
1653 mci->mtype_cap = MEM_FLAG_DDR3;
1654 mci->edac_ctl_cap = EDAC_FLAG_NONE;
1655 mci->edac_cap = EDAC_FLAG_NONE;
1656 mci->mod_name = "sbridge_edac.c";
1657 mci->mod_ver = SBRIDGE_REVISION;
1658 mci->ctl_name = kasprintf(GFP_KERNEL, "Sandy Bridge Socket#%d", mci->mc_idx);
1659 mci->dev_name = pci_name(sbridge_dev->pdev[0]);
1660 mci->ctl_page_to_phys = NULL;
1661
1662 /* Set the function pointer to an actual operation function */
1663 mci->edac_check = sbridge_check_error;
1664
1665 /* Store pci devices at mci for faster access */
1666 rc = mci_bind_devs(mci, sbridge_dev);
1667 if (unlikely(rc < 0))
1668 goto fail0;
1669
1670 /* Get dimm basic config and the memory layout */
1671 get_dimm_config(mci);
1672 get_memory_layout(mci);
1673
1674 /* record ptr to the generic device */
1675 mci->dev = &sbridge_dev->pdev[0]->dev;
1676
1677 /* add this new MC control structure to EDAC's list of MCs */
1678 if (unlikely(edac_mc_add_mc(mci))) {
1679 debugf0("MC: " __FILE__
1680 ": %s(): failed edac_mc_add_mc()\n", __func__);
1681 rc = -EINVAL;
1682 goto fail0;
1683 }
1684
Borislav Petkov3653ada2011-12-04 15:12:09 +01001685 mce_register_decode_chain(&sbridge_mce_dec);
Mauro Carvalho Chehabeebf11a2011-10-20 19:18:01 -02001686 return 0;
Mauro Carvalho Chehabeebf11a2011-10-20 19:18:01 -02001687
1688fail0:
1689 kfree(mci->ctl_name);
1690 edac_mc_free(mci);
1691 sbridge_dev->mci = NULL;
1692 return rc;
1693}
1694
1695/*
1696 * sbridge_probe Probe for ONE instance of device to see if it is
1697 * present.
1698 * return:
1699 * 0 for FOUND a device
1700 * < 0 for error code
1701 */
1702
1703static int __devinit sbridge_probe(struct pci_dev *pdev,
1704 const struct pci_device_id *id)
1705{
1706 int rc;
1707 u8 mc, num_mc = 0;
1708 struct sbridge_dev *sbridge_dev;
1709
1710 /* get the pci devices we want to reserve for our use */
1711 mutex_lock(&sbridge_edac_lock);
1712
1713 /*
1714 * All memory controllers are allocated at the first pass.
1715 */
1716 if (unlikely(probed >= 1)) {
1717 mutex_unlock(&sbridge_edac_lock);
1718 return -ENODEV;
1719 }
1720 probed++;
1721
1722 rc = sbridge_get_all_devices(&num_mc);
1723 if (unlikely(rc < 0))
1724 goto fail0;
1725 mc = 0;
1726
1727 list_for_each_entry(sbridge_dev, &sbridge_edac_list, list) {
1728 debugf0("Registering MC#%d (%d of %d)\n", mc, mc + 1, num_mc);
1729 sbridge_dev->mc = mc++;
1730 rc = sbridge_register_mci(sbridge_dev);
1731 if (unlikely(rc < 0))
1732 goto fail1;
1733 }
1734
1735 sbridge_printk(KERN_INFO, "Driver loaded.\n");
1736
1737 mutex_unlock(&sbridge_edac_lock);
1738 return 0;
1739
1740fail1:
1741 list_for_each_entry(sbridge_dev, &sbridge_edac_list, list)
1742 sbridge_unregister_mci(sbridge_dev);
1743
1744 sbridge_put_all_devices();
1745fail0:
1746 mutex_unlock(&sbridge_edac_lock);
1747 return rc;
1748}
1749
1750/*
1751 * sbridge_remove destructor for one instance of device
1752 *
1753 */
1754static void __devexit sbridge_remove(struct pci_dev *pdev)
1755{
1756 struct sbridge_dev *sbridge_dev;
1757
1758 debugf0(__FILE__ ": %s()\n", __func__);
1759
1760 /*
1761 * we have a trouble here: pdev value for removal will be wrong, since
1762 * it will point to the X58 register used to detect that the machine
1763 * is a Nehalem or upper design. However, due to the way several PCI
1764 * devices are grouped together to provide MC functionality, we need
1765 * to use a different method for releasing the devices
1766 */
1767
1768 mutex_lock(&sbridge_edac_lock);
1769
1770 if (unlikely(!probed)) {
1771 mutex_unlock(&sbridge_edac_lock);
1772 return;
1773 }
1774
1775 list_for_each_entry(sbridge_dev, &sbridge_edac_list, list)
1776 sbridge_unregister_mci(sbridge_dev);
1777
1778 /* Release PCI resources */
1779 sbridge_put_all_devices();
1780
1781 probed--;
1782
1783 mutex_unlock(&sbridge_edac_lock);
1784}
1785
1786MODULE_DEVICE_TABLE(pci, sbridge_pci_tbl);
1787
1788/*
1789 * sbridge_driver pci_driver structure for this module
1790 *
1791 */
1792static struct pci_driver sbridge_driver = {
1793 .name = "sbridge_edac",
1794 .probe = sbridge_probe,
1795 .remove = __devexit_p(sbridge_remove),
1796 .id_table = sbridge_pci_tbl,
1797};
1798
1799/*
1800 * sbridge_init Module entry function
1801 * Try to initialize this module for its devices
1802 */
1803static int __init sbridge_init(void)
1804{
1805 int pci_rc;
1806
1807 debugf2("MC: " __FILE__ ": %s()\n", __func__);
1808
1809 /* Ensure that the OPSTATE is set correctly for POLL or NMI */
1810 opstate_init();
1811
1812 pci_rc = pci_register_driver(&sbridge_driver);
1813
1814 if (pci_rc >= 0)
1815 return 0;
1816
1817 sbridge_printk(KERN_ERR, "Failed to register device with error %d.\n",
1818 pci_rc);
1819
1820 return pci_rc;
1821}
1822
1823/*
1824 * sbridge_exit() Module exit function
1825 * Unregister the driver
1826 */
1827static void __exit sbridge_exit(void)
1828{
1829 debugf2("MC: " __FILE__ ": %s()\n", __func__);
1830 pci_unregister_driver(&sbridge_driver);
1831}
1832
1833module_init(sbridge_init);
1834module_exit(sbridge_exit);
1835
1836module_param(edac_op_state, int, 0444);
1837MODULE_PARM_DESC(edac_op_state, "EDAC Error Reporting state: 0=Poll,1=NMI");
1838
1839MODULE_LICENSE("GPL");
1840MODULE_AUTHOR("Mauro Carvalho Chehab <mchehab@redhat.com>");
1841MODULE_AUTHOR("Red Hat Inc. (http://www.redhat.com)");
1842MODULE_DESCRIPTION("MC Driver for Intel Sandy Bridge memory controllers - "
1843 SBRIDGE_REVISION);