?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Staging: unisys: Remove RETINT macro
[linux/fpc-iii.git] / drivers / edac / sb_edac.c
blob54e2abe671f70acef2965ec5f74cd8b877ac7089
1 /* 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>
21 #include <linux/smp.h>
22 #include <linux/bitmap.h>
23 #include <linux/math64.h>
24 #include <asm/processor.h>
25 #include <asm/mce.h>
26
27 #include "edac_core.h"
28
29 /* Static vars */
30 static LIST_HEAD(sbridge_edac_list);
31 static DEFINE_MUTEX(sbridge_edac_lock);
32 static int probed;
33
34 /*
35 * Alter this version for the module when modifications are made
36 */
37 #define SBRIDGE_REVISION " Ver: 1.1.0 "
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) & GENMASK_ULL(hi, 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
61 * easier for driver's development process. This table should be
62 * 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 */
86 static const u32 sbridge_dram_rule[] = {
87 0x80, 0x88, 0x90, 0x98, 0xa0,
88 0xa8, 0xb0, 0xb8, 0xc0, 0xc8,
89 };
90
91 static const u32 ibridge_dram_rule[] = {
92 0x60, 0x68, 0x70, 0x78, 0x80,
93 0x88, 0x90, 0x98, 0xa0, 0xa8,
94 0xb0, 0xb8, 0xc0, 0xc8, 0xd0,
95 0xd8, 0xe0, 0xe8, 0xf0, 0xf8,
96 };
97
98 #define SAD_LIMIT(reg) ((GET_BITFIELD(reg, 6, 25) << 26) | 0x3ffffff)
99 #define DRAM_ATTR(reg) GET_BITFIELD(reg, 2, 3)
100 #define INTERLEAVE_MODE(reg) GET_BITFIELD(reg, 1, 1)
101 #define DRAM_RULE_ENABLE(reg) GET_BITFIELD(reg, 0, 0)
102
103 static char *get_dram_attr(u32 reg)
104 {
105 switch(DRAM_ATTR(reg)) {
106 case 0:
107 return "DRAM";
108 case 1:
109 return "MMCFG";
110 case 2:
111 return "NXM";
112 default:
113 return "unknown";
114 }
115 }
116
117 static const u32 sbridge_interleave_list[] = {
118 0x84, 0x8c, 0x94, 0x9c, 0xa4,
119 0xac, 0xb4, 0xbc, 0xc4, 0xcc,
120 };
121
122 static const u32 ibridge_interleave_list[] = {
123 0x64, 0x6c, 0x74, 0x7c, 0x84,
124 0x8c, 0x94, 0x9c, 0xa4, 0xac,
125 0xb4, 0xbc, 0xc4, 0xcc, 0xd4,
126 0xdc, 0xe4, 0xec, 0xf4, 0xfc,
127 };
128
129 struct interleave_pkg {
130 unsigned char start;
131 unsigned char end;
132 };
133
134 static const struct interleave_pkg sbridge_interleave_pkg[] = {
135 { 0, 2 },
136 { 3, 5 },
137 { 8, 10 },
138 { 11, 13 },
139 { 16, 18 },
140 { 19, 21 },
141 { 24, 26 },
142 { 27, 29 },
143 };
144
145 static const struct interleave_pkg ibridge_interleave_pkg[] = {
146 { 0, 3 },
147 { 4, 7 },
148 { 8, 11 },
149 { 12, 15 },
150 { 16, 19 },
151 { 20, 23 },
152 { 24, 27 },
153 { 28, 31 },
154 };
155
156 static inline int sad_pkg(const struct interleave_pkg *table, u32 reg,
157 int interleave)
158 {
159 return GET_BITFIELD(reg, table[interleave].start,
160 table[interleave].end);
161 }
162
163 /* Devices 12 Function 7 */
164
165 #define TOLM 0x80
166 #define TOHM 0x84
167
168 #define GET_TOLM(reg) ((GET_BITFIELD(reg, 0, 3) << 28) | 0x3ffffff)
169 #define GET_TOHM(reg) ((GET_BITFIELD(reg, 0, 20) << 25) | 0x3ffffff)
170
171 /* Device 13 Function 6 */
172
173 #define SAD_TARGET 0xf0
174
175 #define SOURCE_ID(reg) GET_BITFIELD(reg, 9, 11)
176
177 #define SAD_CONTROL 0xf4
178
179 #define NODE_ID(reg) GET_BITFIELD(reg, 0, 2)
180
181 /* Device 14 function 0 */
182
183 static const u32 tad_dram_rule[] = {
184 0x40, 0x44, 0x48, 0x4c,
185 0x50, 0x54, 0x58, 0x5c,
186 0x60, 0x64, 0x68, 0x6c,
187 };
188 #define MAX_TAD ARRAY_SIZE(tad_dram_rule)
189
190 #define TAD_LIMIT(reg) ((GET_BITFIELD(reg, 12, 31) << 26) | 0x3ffffff)
191 #define TAD_SOCK(reg) GET_BITFIELD(reg, 10, 11)
192 #define TAD_CH(reg) GET_BITFIELD(reg, 8, 9)
193 #define TAD_TGT3(reg) GET_BITFIELD(reg, 6, 7)
194 #define TAD_TGT2(reg) GET_BITFIELD(reg, 4, 5)
195 #define TAD_TGT1(reg) GET_BITFIELD(reg, 2, 3)
196 #define TAD_TGT0(reg) GET_BITFIELD(reg, 0, 1)
197
198 /* Device 15, function 0 */
199
200 #define MCMTR 0x7c
201
202 #define IS_ECC_ENABLED(mcmtr) GET_BITFIELD(mcmtr, 2, 2)
203 #define IS_LOCKSTEP_ENABLED(mcmtr) GET_BITFIELD(mcmtr, 1, 1)
204 #define IS_CLOSE_PG(mcmtr) GET_BITFIELD(mcmtr, 0, 0)
205
206 /* Device 15, function 1 */
207
208 #define RASENABLES 0xac
209 #define IS_MIRROR_ENABLED(reg) GET_BITFIELD(reg, 0, 0)
210
211 /* Device 15, functions 2-5 */
212
213 static const int mtr_regs[] = {
214 0x80, 0x84, 0x88,
215 };
216
217 #define RANK_DISABLE(mtr) GET_BITFIELD(mtr, 16, 19)
218 #define IS_DIMM_PRESENT(mtr) GET_BITFIELD(mtr, 14, 14)
219 #define RANK_CNT_BITS(mtr) GET_BITFIELD(mtr, 12, 13)
220 #define RANK_WIDTH_BITS(mtr) GET_BITFIELD(mtr, 2, 4)
221 #define COL_WIDTH_BITS(mtr) GET_BITFIELD(mtr, 0, 1)
222
223 static const u32 tad_ch_nilv_offset[] = {
224 0x90, 0x94, 0x98, 0x9c,
225 0xa0, 0xa4, 0xa8, 0xac,
226 0xb0, 0xb4, 0xb8, 0xbc,
227 };
228 #define CHN_IDX_OFFSET(reg) GET_BITFIELD(reg, 28, 29)
229 #define TAD_OFFSET(reg) (GET_BITFIELD(reg, 6, 25) << 26)
230
231 static const u32 rir_way_limit[] = {
232 0x108, 0x10c, 0x110, 0x114, 0x118,
233 };
234 #define MAX_RIR_RANGES ARRAY_SIZE(rir_way_limit)
235
236 #define IS_RIR_VALID(reg) GET_BITFIELD(reg, 31, 31)
237 #define RIR_WAY(reg) GET_BITFIELD(reg, 28, 29)
238 #define RIR_LIMIT(reg) ((GET_BITFIELD(reg, 1, 10) << 29)| 0x1fffffff)
239
240 #define MAX_RIR_WAY 8
241
242 static const u32 rir_offset[MAX_RIR_RANGES][MAX_RIR_WAY] = {
243 { 0x120, 0x124, 0x128, 0x12c, 0x130, 0x134, 0x138, 0x13c },
244 { 0x140, 0x144, 0x148, 0x14c, 0x150, 0x154, 0x158, 0x15c },
245 { 0x160, 0x164, 0x168, 0x16c, 0x170, 0x174, 0x178, 0x17c },
246 { 0x180, 0x184, 0x188, 0x18c, 0x190, 0x194, 0x198, 0x19c },
247 { 0x1a0, 0x1a4, 0x1a8, 0x1ac, 0x1b0, 0x1b4, 0x1b8, 0x1bc },
248 };
249
250 #define RIR_RNK_TGT(reg) GET_BITFIELD(reg, 16, 19)
251 #define RIR_OFFSET(reg) GET_BITFIELD(reg, 2, 14)
252
253 /* Device 16, functions 2-7 */
254
255 /*
256 * FIXME: Implement the error count reads directly
257 */
258
259 static const u32 correrrcnt[] = {
260 0x104, 0x108, 0x10c, 0x110,
261 };
262
263 #define RANK_ODD_OV(reg) GET_BITFIELD(reg, 31, 31)
264 #define RANK_ODD_ERR_CNT(reg) GET_BITFIELD(reg, 16, 30)
265 #define RANK_EVEN_OV(reg) GET_BITFIELD(reg, 15, 15)
266 #define RANK_EVEN_ERR_CNT(reg) GET_BITFIELD(reg, 0, 14)
267
268 static const u32 correrrthrsld[] = {
269 0x11c, 0x120, 0x124, 0x128,
270 };
271
272 #define RANK_ODD_ERR_THRSLD(reg) GET_BITFIELD(reg, 16, 30)
273 #define RANK_EVEN_ERR_THRSLD(reg) GET_BITFIELD(reg, 0, 14)
274
275
276 /* Device 17, function 0 */
277
278 #define SB_RANK_CFG_A 0x0328
279
280 #define IB_RANK_CFG_A 0x0320
281
282 #define IS_RDIMM_ENABLED(reg) GET_BITFIELD(reg, 11, 11)
283
284 /*
285 * sbridge structs
286 */
287
288 #define NUM_CHANNELS 4
289 #define MAX_DIMMS 3 /* Max DIMMS per channel */
290
291 enum type {
292 SANDY_BRIDGE,
293 IVY_BRIDGE,
294 };
295
296 struct sbridge_pvt;
297 struct sbridge_info {
298 enum type type;
299 u32 mcmtr;
300 u32 rankcfgr;
301 u64 (*get_tolm)(struct sbridge_pvt *pvt);
302 u64 (*get_tohm)(struct sbridge_pvt *pvt);
303 const u32 *dram_rule;
304 const u32 *interleave_list;
305 const struct interleave_pkg *interleave_pkg;
306 u8 max_sad;
307 u8 max_interleave;
308 };
309
310 struct sbridge_channel {
311 u32 ranks;
312 u32 dimms;
313 };
314
315 struct pci_id_descr {
316 int dev;
317 int func;
318 int dev_id;
319 int optional;
320 };
321
322 struct pci_id_table {
323 const struct pci_id_descr *descr;
324 int n_devs;
325 };
326
327 struct sbridge_dev {
328 struct list_head list;
329 u8 bus, mc;
330 u8 node_id, source_id;
331 struct pci_dev **pdev;
332 int n_devs;
333 struct mem_ctl_info *mci;
334 };
335
336 struct sbridge_pvt {
337 struct pci_dev *pci_ta, *pci_ddrio, *pci_ras;
338 struct pci_dev *pci_sad0, *pci_sad1;
339 struct pci_dev *pci_ha0, *pci_ha1;
340 struct pci_dev *pci_br0, *pci_br1;
341 struct pci_dev *pci_tad[NUM_CHANNELS];
342
343 struct sbridge_dev *sbridge_dev;
344
345 struct sbridge_info info;
346 struct sbridge_channel channel[NUM_CHANNELS];
347
348 /* Memory type detection */
349 bool is_mirrored, is_lockstep, is_close_pg;
350
351 /* Fifo double buffers */
352 struct mce mce_entry[MCE_LOG_LEN];
353 struct mce mce_outentry[MCE_LOG_LEN];
354
355 /* Fifo in/out counters */
356 unsigned mce_in, mce_out;
357
358 /* Count indicator to show errors not got */
359 unsigned mce_overrun;
360
361 /* Memory description */
362 u64 tolm, tohm;
363 };
364
365 #define PCI_DESCR(device, function, device_id, opt) \
366 .dev = (device), \
367 .func = (function), \
368 .dev_id = (device_id), \
369 .optional = opt
370
371 static const struct pci_id_descr pci_dev_descr_sbridge[] = {
372 /* Processor Home Agent */
373 { PCI_DESCR(14, 0, PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_HA0, 0) },
374
375 /* Memory controller */
376 { PCI_DESCR(15, 0, PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TA, 0) },
377 { PCI_DESCR(15, 1, PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_RAS, 0) },
378 { PCI_DESCR(15, 2, PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD0, 0) },
379 { PCI_DESCR(15, 3, PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD1, 0) },
380 { PCI_DESCR(15, 4, PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD2, 0) },
381 { PCI_DESCR(15, 5, PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD3, 0) },
382 { PCI_DESCR(17, 0, PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_DDRIO, 1) },
383
384 /* System Address Decoder */
385 { PCI_DESCR(12, 6, PCI_DEVICE_ID_INTEL_SBRIDGE_SAD0, 0) },
386 { PCI_DESCR(12, 7, PCI_DEVICE_ID_INTEL_SBRIDGE_SAD1, 0) },
387
388 /* Broadcast Registers */
389 { PCI_DESCR(13, 6, PCI_DEVICE_ID_INTEL_SBRIDGE_BR, 0) },
390 };
391
392 #define PCI_ID_TABLE_ENTRY(A) { .descr=A, .n_devs = ARRAY_SIZE(A) }
393 static const struct pci_id_table pci_dev_descr_sbridge_table[] = {
394 PCI_ID_TABLE_ENTRY(pci_dev_descr_sbridge),
395 {0,} /* 0 terminated list. */
396 };
397
398 /* This changes depending if 1HA or 2HA:
399 * 1HA:
400 * 0x0eb8 (17.0) is DDRIO0
401 * 2HA:
402 * 0x0ebc (17.4) is DDRIO0
403 */
404 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_1HA_DDRIO0 0x0eb8
405 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_2HA_DDRIO0 0x0ebc
406
407 /* pci ids */
408 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0 0x0ea0
409 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TA 0x0ea8
410 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_RAS 0x0e71
411 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD0 0x0eaa
412 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD1 0x0eab
413 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD2 0x0eac
414 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD3 0x0ead
415 #define PCI_DEVICE_ID_INTEL_IBRIDGE_SAD 0x0ec8
416 #define PCI_DEVICE_ID_INTEL_IBRIDGE_BR0 0x0ec9
417 #define PCI_DEVICE_ID_INTEL_IBRIDGE_BR1 0x0eca
418 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1 0x0e60
419 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TA 0x0e68
420 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_RAS 0x0e79
421 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD0 0x0e6a
422 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD1 0x0e6b
423
424 static const struct pci_id_descr pci_dev_descr_ibridge[] = {
425 /* Processor Home Agent */
426 { PCI_DESCR(14, 0, PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0, 0) },
427
428 /* Memory controller */
429 { PCI_DESCR(15, 0, PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TA, 0) },
430 { PCI_DESCR(15, 1, PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_RAS, 0) },
431 { PCI_DESCR(15, 2, PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD0, 0) },
432 { PCI_DESCR(15, 3, PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD1, 0) },
433 { PCI_DESCR(15, 4, PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD2, 0) },
434 { PCI_DESCR(15, 5, PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD3, 0) },
435
436 /* System Address Decoder */
437 { PCI_DESCR(22, 0, PCI_DEVICE_ID_INTEL_IBRIDGE_SAD, 0) },
438
439 /* Broadcast Registers */
440 { PCI_DESCR(22, 1, PCI_DEVICE_ID_INTEL_IBRIDGE_BR0, 1) },
441 { PCI_DESCR(22, 2, PCI_DEVICE_ID_INTEL_IBRIDGE_BR1, 0) },
442
443 /* Optional, mode 2HA */
444 { PCI_DESCR(28, 0, PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1, 1) },
445 #if 0
446 { PCI_DESCR(29, 0, PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TA, 1) },
447 { PCI_DESCR(29, 1, PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_RAS, 1) },
448 #endif
449 { PCI_DESCR(29, 2, PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD0, 1) },
450 { PCI_DESCR(29, 3, PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD1, 1) },
451
452 { PCI_DESCR(17, 0, PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_1HA_DDRIO0, 1) },
453 { PCI_DESCR(17, 4, PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_2HA_DDRIO0, 1) },
454 };
455
456 static const struct pci_id_table pci_dev_descr_ibridge_table[] = {
457 PCI_ID_TABLE_ENTRY(pci_dev_descr_ibridge),
458 {0,} /* 0 terminated list. */
459 };
460
461 /*
462 * pci_device_id table for which devices we are looking for
463 */
464 static const struct pci_device_id sbridge_pci_tbl[] = {
465 {PCI_DEVICE(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TA)},
466 {PCI_DEVICE(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TA)},
467 {0,} /* 0 terminated list. */
468 };
469
470
471 /****************************************************************************
472 Ancillary status routines
473 ****************************************************************************/
474
475 static inline int numrank(u32 mtr)
476 {
477 int ranks = (1 << RANK_CNT_BITS(mtr));
478
479 if (ranks > 4) {
480 edac_dbg(0, "Invalid number of ranks: %d (max = 4) raw value = %x (%04x)\n",
481 ranks, (unsigned int)RANK_CNT_BITS(mtr), mtr);
482 return -EINVAL;
483 }
484
485 return ranks;
486 }
487
488 static inline int numrow(u32 mtr)
489 {
490 int rows = (RANK_WIDTH_BITS(mtr) + 12);
491
492 if (rows < 13 || rows > 18) {
493 edac_dbg(0, "Invalid number of rows: %d (should be between 14 and 17) raw value = %x (%04x)\n",
494 rows, (unsigned int)RANK_WIDTH_BITS(mtr), mtr);
495 return -EINVAL;
496 }
497
498 return 1 << rows;
499 }
500
501 static inline int numcol(u32 mtr)
502 {
503 int cols = (COL_WIDTH_BITS(mtr) + 10);
504
505 if (cols > 12) {
506 edac_dbg(0, "Invalid number of cols: %d (max = 4) raw value = %x (%04x)\n",
507 cols, (unsigned int)COL_WIDTH_BITS(mtr), mtr);
508 return -EINVAL;
509 }
510
511 return 1 << cols;
512 }
513
514 static struct sbridge_dev *get_sbridge_dev(u8 bus)
515 {
516 struct sbridge_dev *sbridge_dev;
517
518 list_for_each_entry(sbridge_dev, &sbridge_edac_list, list) {
519 if (sbridge_dev->bus == bus)
520 return sbridge_dev;
521 }
522
523 return NULL;
524 }
525
526 static struct sbridge_dev *alloc_sbridge_dev(u8 bus,
527 const struct pci_id_table *table)
528 {
529 struct sbridge_dev *sbridge_dev;
530
531 sbridge_dev = kzalloc(sizeof(*sbridge_dev), GFP_KERNEL);
532 if (!sbridge_dev)
533 return NULL;
534
535 sbridge_dev->pdev = kzalloc(sizeof(*sbridge_dev->pdev) * table->n_devs,
536 GFP_KERNEL);
537 if (!sbridge_dev->pdev) {
538 kfree(sbridge_dev);
539 return NULL;
540 }
541
542 sbridge_dev->bus = bus;
543 sbridge_dev->n_devs = table->n_devs;
544 list_add_tail(&sbridge_dev->list, &sbridge_edac_list);
545
546 return sbridge_dev;
547 }
548
549 static void free_sbridge_dev(struct sbridge_dev *sbridge_dev)
550 {
551 list_del(&sbridge_dev->list);
552 kfree(sbridge_dev->pdev);
553 kfree(sbridge_dev);
554 }
555
556 static u64 sbridge_get_tolm(struct sbridge_pvt *pvt)
557 {
558 u32 reg;
559
560 /* Address range is 32:28 */
561 pci_read_config_dword(pvt->pci_sad1, TOLM, &reg);
562 return GET_TOLM(reg);
563 }
564
565 static u64 sbridge_get_tohm(struct sbridge_pvt *pvt)
566 {
567 u32 reg;
568
569 pci_read_config_dword(pvt->pci_sad1, TOHM, &reg);
570 return GET_TOHM(reg);
571 }
572
573 static u64 ibridge_get_tolm(struct sbridge_pvt *pvt)
574 {
575 u32 reg;
576
577 pci_read_config_dword(pvt->pci_br1, TOLM, &reg);
578
579 return GET_TOLM(reg);
580 }
581
582 static u64 ibridge_get_tohm(struct sbridge_pvt *pvt)
583 {
584 u32 reg;
585
586 pci_read_config_dword(pvt->pci_br1, TOHM, &reg);
587
588 return GET_TOHM(reg);
589 }
590
591 static inline u8 sad_pkg_socket(u8 pkg)
592 {
593 /* on Ivy Bridge, nodeID is SASS, where A is HA and S is node id */
594 return (pkg >> 3) | (pkg & 0x3);
595 }
596
597 static inline u8 sad_pkg_ha(u8 pkg)
598 {
599 return (pkg >> 2) & 0x1;
600 }
601
602 /****************************************************************************
603 Memory check routines
604 ****************************************************************************/
605 static struct pci_dev *get_pdev_slot_func(u8 bus, unsigned slot,
606 unsigned func)
607 {
608 struct sbridge_dev *sbridge_dev = get_sbridge_dev(bus);
609 int i;
610
611 if (!sbridge_dev)
612 return NULL;
613
614 for (i = 0; i < sbridge_dev->n_devs; i++) {
615 if (!sbridge_dev->pdev[i])
616 continue;
617
618 if (PCI_SLOT(sbridge_dev->pdev[i]->devfn) == slot &&
619 PCI_FUNC(sbridge_dev->pdev[i]->devfn) == func) {
620 edac_dbg(1, "Associated %02x.%02x.%d with %p\n",
621 bus, slot, func, sbridge_dev->pdev[i]);
622 return sbridge_dev->pdev[i];
623 }
624 }
625
626 return NULL;
627 }
628
629 /**
630 * check_if_ecc_is_active() - Checks if ECC is active
631 * bus: Device bus
632 */
633 static int check_if_ecc_is_active(const u8 bus)
634 {
635 struct pci_dev *pdev = NULL;
636 u32 mcmtr;
637
638 pdev = get_pdev_slot_func(bus, 15, 0);
639 if (!pdev) {
640 sbridge_printk(KERN_ERR, "Couldn't find PCI device "
641 "%2x.%02d.%d!!!\n",
642 bus, 15, 0);
643 return -ENODEV;
644 }
645
646 pci_read_config_dword(pdev, MCMTR, &mcmtr);
647 if (!IS_ECC_ENABLED(mcmtr)) {
648 sbridge_printk(KERN_ERR, "ECC is disabled. Aborting\n");
649 return -ENODEV;
650 }
651 return 0;
652 }
653
654 static int get_dimm_config(struct mem_ctl_info *mci)
655 {
656 struct sbridge_pvt *pvt = mci->pvt_info;
657 struct dimm_info *dimm;
658 unsigned i, j, banks, ranks, rows, cols, npages;
659 u64 size;
660 u32 reg;
661 enum edac_type mode;
662 enum mem_type mtype;
663
664 pci_read_config_dword(pvt->pci_br0, SAD_TARGET, &reg);
665 pvt->sbridge_dev->source_id = SOURCE_ID(reg);
666
667 pci_read_config_dword(pvt->pci_br0, SAD_CONTROL, &reg);
668 pvt->sbridge_dev->node_id = NODE_ID(reg);
669 edac_dbg(0, "mc#%d: Node ID: %d, source ID: %d\n",
670 pvt->sbridge_dev->mc,
671 pvt->sbridge_dev->node_id,
672 pvt->sbridge_dev->source_id);
673
674 pci_read_config_dword(pvt->pci_ras, RASENABLES, &reg);
675 if (IS_MIRROR_ENABLED(reg)) {
676 edac_dbg(0, "Memory mirror is enabled\n");
677 pvt->is_mirrored = true;
678 } else {
679 edac_dbg(0, "Memory mirror is disabled\n");
680 pvt->is_mirrored = false;
681 }
682
683 pci_read_config_dword(pvt->pci_ta, MCMTR, &pvt->info.mcmtr);
684 if (IS_LOCKSTEP_ENABLED(pvt->info.mcmtr)) {
685 edac_dbg(0, "Lockstep is enabled\n");
686 mode = EDAC_S8ECD8ED;
687 pvt->is_lockstep = true;
688 } else {
689 edac_dbg(0, "Lockstep is disabled\n");
690 mode = EDAC_S4ECD4ED;
691 pvt->is_lockstep = false;
692 }
693 if (IS_CLOSE_PG(pvt->info.mcmtr)) {
694 edac_dbg(0, "address map is on closed page mode\n");
695 pvt->is_close_pg = true;
696 } else {
697 edac_dbg(0, "address map is on open page mode\n");
698 pvt->is_close_pg = false;
699 }
700
701 if (pvt->pci_ddrio) {
702 pci_read_config_dword(pvt->pci_ddrio, pvt->info.rankcfgr,
703 &reg);
704 if (IS_RDIMM_ENABLED(reg)) {
705 /* FIXME: Can also be LRDIMM */
706 edac_dbg(0, "Memory is registered\n");
707 mtype = MEM_RDDR3;
708 } else {
709 edac_dbg(0, "Memory is unregistered\n");
710 mtype = MEM_DDR3;
711 }
712 } else {
713 edac_dbg(0, "Cannot determine memory type\n");
714 mtype = MEM_UNKNOWN;
715 }
716
717 /* On all supported DDR3 DIMM types, there are 8 banks available */
718 banks = 8;
719
720 for (i = 0; i < NUM_CHANNELS; i++) {
721 u32 mtr;
722
723 for (j = 0; j < ARRAY_SIZE(mtr_regs); j++) {
724 dimm = EDAC_DIMM_PTR(mci->layers, mci->dimms, mci->n_layers,
725 i, j, 0);
726 pci_read_config_dword(pvt->pci_tad[i],
727 mtr_regs[j], &mtr);
728 edac_dbg(4, "Channel #%d MTR%d = %x\n", i, j, mtr);
729 if (IS_DIMM_PRESENT(mtr)) {
730 pvt->channel[i].dimms++;
731
732 ranks = numrank(mtr);
733 rows = numrow(mtr);
734 cols = numcol(mtr);
735
736 /* DDR3 has 8 I/O banks */
737 size = ((u64)rows * cols * banks * ranks) >> (20 - 3);
738 npages = MiB_TO_PAGES(size);
739
740 edac_dbg(0, "mc#%d: channel %d, dimm %d, %Ld Mb (%d pages) bank: %d, rank: %d, row: %#x, col: %#x\n",
741 pvt->sbridge_dev->mc, i, j,
742 size, npages,
743 banks, ranks, rows, cols);
744
745 dimm->nr_pages = npages;
746 dimm->grain = 32;
747 dimm->dtype = (banks == 8) ? DEV_X8 : DEV_X4;
748 dimm->mtype = mtype;
749 dimm->edac_mode = mode;
750 snprintf(dimm->label, sizeof(dimm->label),
751 "CPU_SrcID#%u_Channel#%u_DIMM#%u",
752 pvt->sbridge_dev->source_id, i, j);
753 }
754 }
755 }
756
757 return 0;
758 }
759
760 static void get_memory_layout(const struct mem_ctl_info *mci)
761 {
762 struct sbridge_pvt *pvt = mci->pvt_info;
763 int i, j, k, n_sads, n_tads, sad_interl;
764 u32 reg;
765 u64 limit, prv = 0;
766 u64 tmp_mb;
767 u32 mb, kb;
768 u32 rir_way;
769
770 /*
771 * Step 1) Get TOLM/TOHM ranges
772 */
773
774 pvt->tolm = pvt->info.get_tolm(pvt);
775 tmp_mb = (1 + pvt->tolm) >> 20;
776
777 mb = div_u64_rem(tmp_mb, 1000, &kb);
778 edac_dbg(0, "TOLM: %u.%03u GB (0x%016Lx)\n", mb, kb, (u64)pvt->tolm);
779
780 /* Address range is already 45:25 */
781 pvt->tohm = pvt->info.get_tohm(pvt);
782 tmp_mb = (1 + pvt->tohm) >> 20;
783
784 mb = div_u64_rem(tmp_mb, 1000, &kb);
785 edac_dbg(0, "TOHM: %u.%03u GB (0x%016Lx)\n", mb, kb, (u64)pvt->tohm);
786
787 /*
788 * Step 2) Get SAD range and SAD Interleave list
789 * TAD registers contain the interleave wayness. However, it
790 * seems simpler to just discover it indirectly, with the
791 * algorithm bellow.
792 */
793 prv = 0;
794 for (n_sads = 0; n_sads < pvt->info.max_sad; n_sads++) {
795 /* SAD_LIMIT Address range is 45:26 */
796 pci_read_config_dword(pvt->pci_sad0, pvt->info.dram_rule[n_sads],
797 &reg);
798 limit = SAD_LIMIT(reg);
799
800 if (!DRAM_RULE_ENABLE(reg))
801 continue;
802
803 if (limit <= prv)
804 break;
805
806 tmp_mb = (limit + 1) >> 20;
807 mb = div_u64_rem(tmp_mb, 1000, &kb);
808 edac_dbg(0, "SAD#%d %s up to %u.%03u GB (0x%016Lx) Interleave: %s reg=0x%08x\n",
809 n_sads,
810 get_dram_attr(reg),
811 mb, kb,
812 ((u64)tmp_mb) << 20L,
813 INTERLEAVE_MODE(reg) ? "8:6" : "[8:6]XOR[18:16]",
814 reg);
815 prv = limit;
816
817 pci_read_config_dword(pvt->pci_sad0, pvt->info.interleave_list[n_sads],
818 &reg);
819 sad_interl = sad_pkg(pvt->info.interleave_pkg, reg, 0);
820 for (j = 0; j < 8; j++) {
821 u32 pkg = sad_pkg(pvt->info.interleave_pkg, reg, j);
822 if (j > 0 && sad_interl == pkg)
823 break;
824
825 edac_dbg(0, "SAD#%d, interleave #%d: %d\n",
826 n_sads, j, pkg);
827 }
828 }
829
830 /*
831 * Step 3) Get TAD range
832 */
833 prv = 0;
834 for (n_tads = 0; n_tads < MAX_TAD; n_tads++) {
835 pci_read_config_dword(pvt->pci_ha0, tad_dram_rule[n_tads],
836 &reg);
837 limit = TAD_LIMIT(reg);
838 if (limit <= prv)
839 break;
840 tmp_mb = (limit + 1) >> 20;
841
842 mb = div_u64_rem(tmp_mb, 1000, &kb);
843 edac_dbg(0, "TAD#%d: up to %u.%03u GB (0x%016Lx), socket interleave %d, memory interleave %d, TGT: %d, %d, %d, %d, reg=0x%08x\n",
844 n_tads, mb, kb,
845 ((u64)tmp_mb) << 20L,
846 (u32)TAD_SOCK(reg),
847 (u32)TAD_CH(reg),
848 (u32)TAD_TGT0(reg),
849 (u32)TAD_TGT1(reg),
850 (u32)TAD_TGT2(reg),
851 (u32)TAD_TGT3(reg),
852 reg);
853 prv = limit;
854 }
855
856 /*
857 * Step 4) Get TAD offsets, per each channel
858 */
859 for (i = 0; i < NUM_CHANNELS; i++) {
860 if (!pvt->channel[i].dimms)
861 continue;
862 for (j = 0; j < n_tads; j++) {
863 pci_read_config_dword(pvt->pci_tad[i],
864 tad_ch_nilv_offset[j],
865 &reg);
866 tmp_mb = TAD_OFFSET(reg) >> 20;
867 mb = div_u64_rem(tmp_mb, 1000, &kb);
868 edac_dbg(0, "TAD CH#%d, offset #%d: %u.%03u GB (0x%016Lx), reg=0x%08x\n",
869 i, j,
870 mb, kb,
871 ((u64)tmp_mb) << 20L,
872 reg);
873 }
874 }
875
876 /*
877 * Step 6) Get RIR Wayness/Limit, per each channel
878 */
879 for (i = 0; i < NUM_CHANNELS; i++) {
880 if (!pvt->channel[i].dimms)
881 continue;
882 for (j = 0; j < MAX_RIR_RANGES; j++) {
883 pci_read_config_dword(pvt->pci_tad[i],
884 rir_way_limit[j],
885 &reg);
886
887 if (!IS_RIR_VALID(reg))
888 continue;
889
890 tmp_mb = RIR_LIMIT(reg) >> 20;
891 rir_way = 1 << RIR_WAY(reg);
892 mb = div_u64_rem(tmp_mb, 1000, &kb);
893 edac_dbg(0, "CH#%d RIR#%d, limit: %u.%03u GB (0x%016Lx), way: %d, reg=0x%08x\n",
894 i, j,
895 mb, kb,
896 ((u64)tmp_mb) << 20L,
897 rir_way,
898 reg);
899
900 for (k = 0; k < rir_way; k++) {
901 pci_read_config_dword(pvt->pci_tad[i],
902 rir_offset[j][k],
903 &reg);
904 tmp_mb = RIR_OFFSET(reg) << 6;
905
906 mb = div_u64_rem(tmp_mb, 1000, &kb);
907 edac_dbg(0, "CH#%d RIR#%d INTL#%d, offset %u.%03u GB (0x%016Lx), tgt: %d, reg=0x%08x\n",
908 i, j, k,
909 mb, kb,
910 ((u64)tmp_mb) << 20L,
911 (u32)RIR_RNK_TGT(reg),
912 reg);
913 }
914 }
915 }
916 }
917
918 static struct mem_ctl_info *get_mci_for_node_id(u8 node_id)
919 {
920 struct sbridge_dev *sbridge_dev;
921
922 list_for_each_entry(sbridge_dev, &sbridge_edac_list, list) {
923 if (sbridge_dev->node_id == node_id)
924 return sbridge_dev->mci;
925 }
926 return NULL;
927 }
928
929 static int get_memory_error_data(struct mem_ctl_info *mci,
930 u64 addr,
931 u8 *socket,
932 long *channel_mask,
933 u8 *rank,
934 char **area_type, char *msg)
935 {
936 struct mem_ctl_info *new_mci;
937 struct sbridge_pvt *pvt = mci->pvt_info;
938 struct pci_dev *pci_ha;
939 int n_rir, n_sads, n_tads, sad_way, sck_xch;
940 int sad_interl, idx, base_ch;
941 int interleave_mode;
942 unsigned sad_interleave[pvt->info.max_interleave];
943 u32 reg;
944 u8 ch_way, sck_way, pkg, sad_ha = 0;
945 u32 tad_offset;
946 u32 rir_way;
947 u32 mb, kb;
948 u64 ch_addr, offset, limit = 0, prv = 0;
949
950
951 /*
952 * Step 0) Check if the address is at special memory ranges
953 * The check bellow is probably enough to fill all cases where
954 * the error is not inside a memory, except for the legacy
955 * range (e. g. VGA addresses). It is unlikely, however, that the
956 * memory controller would generate an error on that range.
957 */
958 if ((addr > (u64) pvt->tolm) && (addr < (1LL << 32))) {
959 sprintf(msg, "Error at TOLM area, on addr 0x%08Lx", addr);
960 return -EINVAL;
961 }
962 if (addr >= (u64)pvt->tohm) {
963 sprintf(msg, "Error at MMIOH area, on addr 0x%016Lx", addr);
964 return -EINVAL;
965 }
966
967 /*
968 * Step 1) Get socket
969 */
970 for (n_sads = 0; n_sads < pvt->info.max_sad; n_sads++) {
971 pci_read_config_dword(pvt->pci_sad0, pvt->info.dram_rule[n_sads],
972 &reg);
973
974 if (!DRAM_RULE_ENABLE(reg))
975 continue;
976
977 limit = SAD_LIMIT(reg);
978 if (limit <= prv) {
979 sprintf(msg, "Can't discover the memory socket");
980 return -EINVAL;
981 }
982 if (addr <= limit)
983 break;
984 prv = limit;
985 }
986 if (n_sads == pvt->info.max_sad) {
987 sprintf(msg, "Can't discover the memory socket");
988 return -EINVAL;
989 }
990 *area_type = get_dram_attr(reg);
991 interleave_mode = INTERLEAVE_MODE(reg);
992
993 pci_read_config_dword(pvt->pci_sad0, pvt->info.interleave_list[n_sads],
994 &reg);
995
996 if (pvt->info.type == SANDY_BRIDGE) {
997 sad_interl = sad_pkg(pvt->info.interleave_pkg, reg, 0);
998 for (sad_way = 0; sad_way < 8; sad_way++) {
999 u32 pkg = sad_pkg(pvt->info.interleave_pkg, reg, sad_way);
1000 if (sad_way > 0 && sad_interl == pkg)
1001 break;
1002 sad_interleave[sad_way] = pkg;
1003 edac_dbg(0, "SAD interleave #%d: %d\n",
1004 sad_way, sad_interleave[sad_way]);
1005 }
1006 edac_dbg(0, "mc#%d: Error detected on SAD#%d: address 0x%016Lx < 0x%016Lx, Interleave [%d:6]%s\n",
1007 pvt->sbridge_dev->mc,
1008 n_sads,
1009 addr,
1010 limit,
1011 sad_way + 7,
1012 !interleave_mode ? "" : "XOR[18:16]");
1013 if (interleave_mode)
1014 idx = ((addr >> 6) ^ (addr >> 16)) & 7;
1015 else
1016 idx = (addr >> 6) & 7;
1017 switch (sad_way) {
1018 case 1:
1019 idx = 0;
1020 break;
1021 case 2:
1022 idx = idx & 1;
1023 break;
1024 case 4:
1025 idx = idx & 3;
1026 break;
1027 case 8:
1028 break;
1029 default:
1030 sprintf(msg, "Can't discover socket interleave");
1031 return -EINVAL;
1032 }
1033 *socket = sad_interleave[idx];
1034 edac_dbg(0, "SAD interleave index: %d (wayness %d) = CPU socket %d\n",
1035 idx, sad_way, *socket);
1036 } else {
1037 /* Ivy Bridge's SAD mode doesn't support XOR interleave mode */
1038 idx = (addr >> 6) & 7;
1039 pkg = sad_pkg(pvt->info.interleave_pkg, reg, idx);
1040 *socket = sad_pkg_socket(pkg);
1041 sad_ha = sad_pkg_ha(pkg);
1042 edac_dbg(0, "SAD interleave package: %d = CPU socket %d, HA %d\n",
1043 idx, *socket, sad_ha);
1044 }
1045
1046 /*
1047 * Move to the proper node structure, in order to access the
1048 * right PCI registers
1049 */
1050 new_mci = get_mci_for_node_id(*socket);
1051 if (!new_mci) {
1052 sprintf(msg, "Struct for socket #%u wasn't initialized",
1053 *socket);
1054 return -EINVAL;
1055 }
1056 mci = new_mci;
1057 pvt = mci->pvt_info;
1058
1059 /*
1060 * Step 2) Get memory channel
1061 */
1062 prv = 0;
1063 if (pvt->info.type == SANDY_BRIDGE)
1064 pci_ha = pvt->pci_ha0;
1065 else {
1066 if (sad_ha)
1067 pci_ha = pvt->pci_ha1;
1068 else
1069 pci_ha = pvt->pci_ha0;
1070 }
1071 for (n_tads = 0; n_tads < MAX_TAD; n_tads++) {
1072 pci_read_config_dword(pci_ha, tad_dram_rule[n_tads], &reg);
1073 limit = TAD_LIMIT(reg);
1074 if (limit <= prv) {
1075 sprintf(msg, "Can't discover the memory channel");
1076 return -EINVAL;
1077 }
1078 if (addr <= limit)
1079 break;
1080 prv = limit;
1081 }
1082 if (n_tads == MAX_TAD) {
1083 sprintf(msg, "Can't discover the memory channel");
1084 return -EINVAL;
1085 }
1086
1087 ch_way = TAD_CH(reg) + 1;
1088 sck_way = TAD_SOCK(reg) + 1;
1089
1090 if (ch_way == 3)
1091 idx = addr >> 6;
1092 else
1093 idx = addr >> (6 + sck_way);
1094 idx = idx % ch_way;
1095
1096 /*
1097 * FIXME: Shouldn't we use CHN_IDX_OFFSET() here, when ch_way == 3 ???
1098 */
1099 switch (idx) {
1100 case 0:
1101 base_ch = TAD_TGT0(reg);
1102 break;
1103 case 1:
1104 base_ch = TAD_TGT1(reg);
1105 break;
1106 case 2:
1107 base_ch = TAD_TGT2(reg);
1108 break;
1109 case 3:
1110 base_ch = TAD_TGT3(reg);
1111 break;
1112 default:
1113 sprintf(msg, "Can't discover the TAD target");
1114 return -EINVAL;
1115 }
1116 *channel_mask = 1 << base_ch;
1117
1118 pci_read_config_dword(pvt->pci_tad[base_ch],
1119 tad_ch_nilv_offset[n_tads],
1120 &tad_offset);
1121
1122 if (pvt->is_mirrored) {
1123 *channel_mask |= 1 << ((base_ch + 2) % 4);
1124 switch(ch_way) {
1125 case 2:
1126 case 4:
1127 sck_xch = 1 << sck_way * (ch_way >> 1);
1128 break;
1129 default:
1130 sprintf(msg, "Invalid mirror set. Can't decode addr");
1131 return -EINVAL;
1132 }
1133 } else
1134 sck_xch = (1 << sck_way) * ch_way;
1135
1136 if (pvt->is_lockstep)
1137 *channel_mask |= 1 << ((base_ch + 1) % 4);
1138
1139 offset = TAD_OFFSET(tad_offset);
1140
1141 edac_dbg(0, "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",
1142 n_tads,
1143 addr,
1144 limit,
1145 (u32)TAD_SOCK(reg),
1146 ch_way,
1147 offset,
1148 idx,
1149 base_ch,
1150 *channel_mask);
1151
1152 /* Calculate channel address */
1153 /* Remove the TAD offset */
1154
1155 if (offset > addr) {
1156 sprintf(msg, "Can't calculate ch addr: TAD offset 0x%08Lx is too high for addr 0x%08Lx!",
1157 offset, addr);
1158 return -EINVAL;
1159 }
1160 addr -= offset;
1161 /* Store the low bits [0:6] of the addr */
1162 ch_addr = addr & 0x7f;
1163 /* Remove socket wayness and remove 6 bits */
1164 addr >>= 6;
1165 addr = div_u64(addr, sck_xch);
1166 #if 0
1167 /* Divide by channel way */
1168 addr = addr / ch_way;
1169 #endif
1170 /* Recover the last 6 bits */
1171 ch_addr |= addr << 6;
1172
1173 /*
1174 * Step 3) Decode rank
1175 */
1176 for (n_rir = 0; n_rir < MAX_RIR_RANGES; n_rir++) {
1177 pci_read_config_dword(pvt->pci_tad[base_ch],
1178 rir_way_limit[n_rir],
1179 &reg);
1180
1181 if (!IS_RIR_VALID(reg))
1182 continue;
1183
1184 limit = RIR_LIMIT(reg);
1185 mb = div_u64_rem(limit >> 20, 1000, &kb);
1186 edac_dbg(0, "RIR#%d, limit: %u.%03u GB (0x%016Lx), way: %d\n",
1187 n_rir,
1188 mb, kb,
1189 limit,
1190 1 << RIR_WAY(reg));
1191 if (ch_addr <= limit)
1192 break;
1193 }
1194 if (n_rir == MAX_RIR_RANGES) {
1195 sprintf(msg, "Can't discover the memory rank for ch addr 0x%08Lx",
1196 ch_addr);
1197 return -EINVAL;
1198 }
1199 rir_way = RIR_WAY(reg);
1200 if (pvt->is_close_pg)
1201 idx = (ch_addr >> 6);
1202 else
1203 idx = (ch_addr >> 13); /* FIXME: Datasheet says to shift by 15 */
1204 idx %= 1 << rir_way;
1205
1206 pci_read_config_dword(pvt->pci_tad[base_ch],
1207 rir_offset[n_rir][idx],
1208 &reg);
1209 *rank = RIR_RNK_TGT(reg);
1210
1211 edac_dbg(0, "RIR#%d: channel address 0x%08Lx < 0x%08Lx, RIR interleave %d, index %d\n",
1212 n_rir,
1213 ch_addr,
1214 limit,
1215 rir_way,
1216 idx);
1217
1218 return 0;
1219 }
1220
1221 /****************************************************************************
1222 Device initialization routines: put/get, init/exit
1223 ****************************************************************************/
1224
1225 /*
1226 * sbridge_put_all_devices 'put' all the devices that we have
1227 * reserved via 'get'
1228 */
1229 static void sbridge_put_devices(struct sbridge_dev *sbridge_dev)
1230 {
1231 int i;
1232
1233 edac_dbg(0, "\n");
1234 for (i = 0; i < sbridge_dev->n_devs; i++) {
1235 struct pci_dev *pdev = sbridge_dev->pdev[i];
1236 if (!pdev)
1237 continue;
1238 edac_dbg(0, "Removing dev %02x:%02x.%d\n",
1239 pdev->bus->number,
1240 PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn));
1241 pci_dev_put(pdev);
1242 }
1243 }
1244
1245 static void sbridge_put_all_devices(void)
1246 {
1247 struct sbridge_dev *sbridge_dev, *tmp;
1248
1249 list_for_each_entry_safe(sbridge_dev, tmp, &sbridge_edac_list, list) {
1250 sbridge_put_devices(sbridge_dev);
1251 free_sbridge_dev(sbridge_dev);
1252 }
1253 }
1254
1255 static int sbridge_get_onedevice(struct pci_dev **prev,
1256 u8 *num_mc,
1257 const struct pci_id_table *table,
1258 const unsigned devno)
1259 {
1260 struct sbridge_dev *sbridge_dev;
1261 const struct pci_id_descr *dev_descr = &table->descr[devno];
1262
1263 struct pci_dev *pdev = NULL;
1264 u8 bus = 0;
1265
1266 sbridge_printk(KERN_INFO,
1267 "Seeking for: dev %02x.%d PCI ID %04x:%04x\n",
1268 dev_descr->dev, dev_descr->func,
1269 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
1270
1271 pdev = pci_get_device(PCI_VENDOR_ID_INTEL,
1272 dev_descr->dev_id, *prev);
1273
1274 if (!pdev) {
1275 if (*prev) {
1276 *prev = pdev;
1277 return 0;
1278 }
1279
1280 if (dev_descr->optional)
1281 return 0;
1282
1283 if (devno == 0)
1284 return -ENODEV;
1285
1286 sbridge_printk(KERN_INFO,
1287 "Device not found: dev %02x.%d PCI ID %04x:%04x\n",
1288 dev_descr->dev, dev_descr->func,
1289 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
1290
1291 /* End of list, leave */
1292 return -ENODEV;
1293 }
1294 bus = pdev->bus->number;
1295
1296 sbridge_dev = get_sbridge_dev(bus);
1297 if (!sbridge_dev) {
1298 sbridge_dev = alloc_sbridge_dev(bus, table);
1299 if (!sbridge_dev) {
1300 pci_dev_put(pdev);
1301 return -ENOMEM;
1302 }
1303 (*num_mc)++;
1304 }
1305
1306 if (sbridge_dev->pdev[devno]) {
1307 sbridge_printk(KERN_ERR,
1308 "Duplicated device for "
1309 "dev %02x:%d.%d PCI ID %04x:%04x\n",
1310 bus, dev_descr->dev, dev_descr->func,
1311 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
1312 pci_dev_put(pdev);
1313 return -ENODEV;
1314 }
1315
1316 sbridge_dev->pdev[devno] = pdev;
1317
1318 /* Sanity check */
1319 if (unlikely(PCI_SLOT(pdev->devfn) != dev_descr->dev ||
1320 PCI_FUNC(pdev->devfn) != dev_descr->func)) {
1321 sbridge_printk(KERN_ERR,
1322 "Device PCI ID %04x:%04x "
1323 "has dev %02x:%d.%d instead of dev %02x:%02x.%d\n",
1324 PCI_VENDOR_ID_INTEL, dev_descr->dev_id,
1325 bus, PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn),
1326 bus, dev_descr->dev, dev_descr->func);
1327 return -ENODEV;
1328 }
1329
1330 /* Be sure that the device is enabled */
1331 if (unlikely(pci_enable_device(pdev) < 0)) {
1332 sbridge_printk(KERN_ERR,
1333 "Couldn't enable "
1334 "dev %02x:%d.%d PCI ID %04x:%04x\n",
1335 bus, dev_descr->dev, dev_descr->func,
1336 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
1337 return -ENODEV;
1338 }
1339
1340 edac_dbg(0, "Detected dev %02x:%d.%d PCI ID %04x:%04x\n",
1341 bus, dev_descr->dev, dev_descr->func,
1342 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
1343
1344 /*
1345 * As stated on drivers/pci/search.c, the reference count for
1346 * @from is always decremented if it is not %NULL. So, as we need
1347 * to get all devices up to null, we need to do a get for the device
1348 */
1349 pci_dev_get(pdev);
1350
1351 *prev = pdev;
1352
1353 return 0;
1354 }
1355
1356 /*
1357 * sbridge_get_all_devices - Find and perform 'get' operation on the MCH's
1358 * device/functions we want to reference for this driver.
1359 * Need to 'get' device 16 func 1 and func 2.
1360 * @num_mc: pointer to the memory controllers count, to be incremented in case
1361 * of success.
1362 * @table: model specific table
1363 *
1364 * returns 0 in case of success or error code
1365 */
1366 static int sbridge_get_all_devices(u8 *num_mc,
1367 const struct pci_id_table *table)
1368 {
1369 int i, rc;
1370 struct pci_dev *pdev = NULL;
1371
1372 while (table && table->descr) {
1373 for (i = 0; i < table->n_devs; i++) {
1374 pdev = NULL;
1375 do {
1376 rc = sbridge_get_onedevice(&pdev, num_mc,
1377 table, i);
1378 if (rc < 0) {
1379 if (i == 0) {
1380 i = table->n_devs;
1381 break;
1382 }
1383 sbridge_put_all_devices();
1384 return -ENODEV;
1385 }
1386 } while (pdev);
1387 }
1388 table++;
1389 }
1390
1391 return 0;
1392 }
1393
1394 static int sbridge_mci_bind_devs(struct mem_ctl_info *mci,
1395 struct sbridge_dev *sbridge_dev)
1396 {
1397 struct sbridge_pvt *pvt = mci->pvt_info;
1398 struct pci_dev *pdev;
1399 int i, func, slot;
1400
1401 for (i = 0; i < sbridge_dev->n_devs; i++) {
1402 pdev = sbridge_dev->pdev[i];
1403 if (!pdev)
1404 continue;
1405 slot = PCI_SLOT(pdev->devfn);
1406 func = PCI_FUNC(pdev->devfn);
1407 switch (slot) {
1408 case 12:
1409 switch (func) {
1410 case 6:
1411 pvt->pci_sad0 = pdev;
1412 break;
1413 case 7:
1414 pvt->pci_sad1 = pdev;
1415 break;
1416 default:
1417 goto error;
1418 }
1419 break;
1420 case 13:
1421 switch (func) {
1422 case 6:
1423 pvt->pci_br0 = pdev;
1424 break;
1425 default:
1426 goto error;
1427 }
1428 break;
1429 case 14:
1430 switch (func) {
1431 case 0:
1432 pvt->pci_ha0 = pdev;
1433 break;
1434 default:
1435 goto error;
1436 }
1437 break;
1438 case 15:
1439 switch (func) {
1440 case 0:
1441 pvt->pci_ta = pdev;
1442 break;
1443 case 1:
1444 pvt->pci_ras = pdev;
1445 break;
1446 case 2:
1447 case 3:
1448 case 4:
1449 case 5:
1450 pvt->pci_tad[func - 2] = pdev;
1451 break;
1452 default:
1453 goto error;
1454 }
1455 break;
1456 case 17:
1457 switch (func) {
1458 case 0:
1459 pvt->pci_ddrio = pdev;
1460 break;
1461 default:
1462 goto error;
1463 }
1464 break;
1465 default:
1466 goto error;
1467 }
1468
1469 edac_dbg(0, "Associated PCI %02x.%02d.%d with dev = %p\n",
1470 sbridge_dev->bus,
1471 PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn),
1472 pdev);
1473 }
1474
1475 /* Check if everything were registered */
1476 if (!pvt->pci_sad0 || !pvt->pci_sad1 || !pvt->pci_ha0 ||
1477 !pvt-> pci_tad || !pvt->pci_ras || !pvt->pci_ta)
1478 goto enodev;
1479
1480 for (i = 0; i < NUM_CHANNELS; i++) {
1481 if (!pvt->pci_tad[i])
1482 goto enodev;
1483 }
1484 return 0;
1485
1486 enodev:
1487 sbridge_printk(KERN_ERR, "Some needed devices are missing\n");
1488 return -ENODEV;
1489
1490 error:
1491 sbridge_printk(KERN_ERR, "Device %d, function %d "
1492 "is out of the expected range\n",
1493 slot, func);
1494 return -EINVAL;
1495 }
1496
1497 static int ibridge_mci_bind_devs(struct mem_ctl_info *mci,
1498 struct sbridge_dev *sbridge_dev)
1499 {
1500 struct sbridge_pvt *pvt = mci->pvt_info;
1501 struct pci_dev *pdev, *tmp;
1502 int i, func, slot;
1503 bool mode_2ha = false;
1504
1505 tmp = pci_get_device(PCI_VENDOR_ID_INTEL,
1506 PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1, NULL);
1507 if (tmp) {
1508 mode_2ha = true;
1509 pci_dev_put(tmp);
1510 }
1511
1512 for (i = 0; i < sbridge_dev->n_devs; i++) {
1513 pdev = sbridge_dev->pdev[i];
1514 if (!pdev)
1515 continue;
1516 slot = PCI_SLOT(pdev->devfn);
1517 func = PCI_FUNC(pdev->devfn);
1518
1519 switch (slot) {
1520 case 14:
1521 if (func == 0) {
1522 pvt->pci_ha0 = pdev;
1523 break;
1524 }
1525 goto error;
1526 case 15:
1527 switch (func) {
1528 case 0:
1529 pvt->pci_ta = pdev;
1530 break;
1531 case 1:
1532 pvt->pci_ras = pdev;
1533 break;
1534 case 4:
1535 case 5:
1536 /* if we have 2 HAs active, channels 2 and 3
1537 * are in other device */
1538 if (mode_2ha)
1539 break;
1540 /* fall through */
1541 case 2:
1542 case 3:
1543 pvt->pci_tad[func - 2] = pdev;
1544 break;
1545 default:
1546 goto error;
1547 }
1548 break;
1549 case 17:
1550 if (func == 4) {
1551 pvt->pci_ddrio = pdev;
1552 break;
1553 } else if (func == 0) {
1554 if (!mode_2ha)
1555 pvt->pci_ddrio = pdev;
1556 break;
1557 }
1558 goto error;
1559 case 22:
1560 switch (func) {
1561 case 0:
1562 pvt->pci_sad0 = pdev;
1563 break;
1564 case 1:
1565 pvt->pci_br0 = pdev;
1566 break;
1567 case 2:
1568 pvt->pci_br1 = pdev;
1569 break;
1570 default:
1571 goto error;
1572 }
1573 break;
1574 case 28:
1575 if (func == 0) {
1576 pvt->pci_ha1 = pdev;
1577 break;
1578 }
1579 goto error;
1580 case 29:
1581 /* we shouldn't have this device if we have just one
1582 * HA present */
1583 WARN_ON(!mode_2ha);
1584 if (func == 2 || func == 3) {
1585 pvt->pci_tad[func] = pdev;
1586 break;
1587 }
1588 goto error;
1589 default:
1590 goto error;
1591 }
1592
1593 edac_dbg(0, "Associated PCI %02x.%02d.%d with dev = %p\n",
1594 sbridge_dev->bus,
1595 PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn),
1596 pdev);
1597 }
1598
1599 /* Check if everything were registered */
1600 if (!pvt->pci_sad0 || !pvt->pci_ha0 || !pvt->pci_br0 ||
1601 !pvt->pci_br1 || !pvt->pci_tad || !pvt->pci_ras ||
1602 !pvt->pci_ta)
1603 goto enodev;
1604
1605 for (i = 0; i < NUM_CHANNELS; i++) {
1606 if (!pvt->pci_tad[i])
1607 goto enodev;
1608 }
1609 return 0;
1610
1611 enodev:
1612 sbridge_printk(KERN_ERR, "Some needed devices are missing\n");
1613 return -ENODEV;
1614
1615 error:
1616 sbridge_printk(KERN_ERR,
1617 "Device %d, function %d is out of the expected range\n",
1618 slot, func);
1619 return -EINVAL;
1620 }
1621
1622 /****************************************************************************
1623 Error check routines
1624 ****************************************************************************/
1625
1626 /*
1627 * While Sandy Bridge has error count registers, SMI BIOS read values from
1628 * and resets the counters. So, they are not reliable for the OS to read
1629 * from them. So, we have no option but to just trust on whatever MCE is
1630 * telling us about the errors.
1631 */
1632 static void sbridge_mce_output_error(struct mem_ctl_info *mci,
1633 const struct mce *m)
1634 {
1635 struct mem_ctl_info *new_mci;
1636 struct sbridge_pvt *pvt = mci->pvt_info;
1637 enum hw_event_mc_err_type tp_event;
1638 char *type, *optype, msg[256];
1639 bool ripv = GET_BITFIELD(m->mcgstatus, 0, 0);
1640 bool overflow = GET_BITFIELD(m->status, 62, 62);
1641 bool uncorrected_error = GET_BITFIELD(m->status, 61, 61);
1642 bool recoverable;
1643 u32 core_err_cnt = GET_BITFIELD(m->status, 38, 52);
1644 u32 mscod = GET_BITFIELD(m->status, 16, 31);
1645 u32 errcode = GET_BITFIELD(m->status, 0, 15);
1646 u32 channel = GET_BITFIELD(m->status, 0, 3);
1647 u32 optypenum = GET_BITFIELD(m->status, 4, 6);
1648 long channel_mask, first_channel;
1649 u8 rank, socket;
1650 int rc, dimm;
1651 char *area_type = NULL;
1652
1653 if (pvt->info.type == IVY_BRIDGE)
1654 recoverable = true;
1655 else
1656 recoverable = GET_BITFIELD(m->status, 56, 56);
1657
1658 if (uncorrected_error) {
1659 if (ripv) {
1660 type = "FATAL";
1661 tp_event = HW_EVENT_ERR_FATAL;
1662 } else {
1663 type = "NON_FATAL";
1664 tp_event = HW_EVENT_ERR_UNCORRECTED;
1665 }
1666 } else {
1667 type = "CORRECTED";
1668 tp_event = HW_EVENT_ERR_CORRECTED;
1669 }
1670
1671 /*
1672 * According with Table 15-9 of the Intel Architecture spec vol 3A,
1673 * memory errors should fit in this mask:
1674 * 000f 0000 1mmm cccc (binary)
1675 * where:
1676 * f = Correction Report Filtering Bit. If 1, subsequent errors
1677 * won't be shown
1678 * mmm = error type
1679 * cccc = channel
1680 * If the mask doesn't match, report an error to the parsing logic
1681 */
1682 if (! ((errcode & 0xef80) == 0x80)) {
1683 optype = "Can't parse: it is not a mem";
1684 } else {
1685 switch (optypenum) {
1686 case 0:
1687 optype = "generic undef request error";
1688 break;
1689 case 1:
1690 optype = "memory read error";
1691 break;
1692 case 2:
1693 optype = "memory write error";
1694 break;
1695 case 3:
1696 optype = "addr/cmd error";
1697 break;
1698 case 4:
1699 optype = "memory scrubbing error";
1700 break;
1701 default:
1702 optype = "reserved";
1703 break;
1704 }
1705 }
1706
1707 /* Only decode errors with an valid address (ADDRV) */
1708 if (!GET_BITFIELD(m->status, 58, 58))
1709 return;
1710
1711 rc = get_memory_error_data(mci, m->addr, &socket,
1712 &channel_mask, &rank, &area_type, msg);
1713 if (rc < 0)
1714 goto err_parsing;
1715 new_mci = get_mci_for_node_id(socket);
1716 if (!new_mci) {
1717 strcpy(msg, "Error: socket got corrupted!");
1718 goto err_parsing;
1719 }
1720 mci = new_mci;
1721 pvt = mci->pvt_info;
1722
1723 first_channel = find_first_bit(&channel_mask, NUM_CHANNELS);
1724
1725 if (rank < 4)
1726 dimm = 0;
1727 else if (rank < 8)
1728 dimm = 1;
1729 else
1730 dimm = 2;
1731
1732
1733 /*
1734 * FIXME: On some memory configurations (mirror, lockstep), the
1735 * Memory Controller can't point the error to a single DIMM. The
1736 * EDAC core should be handling the channel mask, in order to point
1737 * to the group of dimm's where the error may be happening.
1738 */
1739 snprintf(msg, sizeof(msg),
1740 "%s%s area:%s err_code:%04x:%04x socket:%d channel_mask:%ld rank:%d",
1741 overflow ? " OVERFLOW" : "",
1742 (uncorrected_error && recoverable) ? " recoverable" : "",
1743 area_type,
1744 mscod, errcode,
1745 socket,
1746 channel_mask,
1747 rank);
1748
1749 edac_dbg(0, "%s\n", msg);
1750
1751 /* FIXME: need support for channel mask */
1752
1753 /* Call the helper to output message */
1754 edac_mc_handle_error(tp_event, mci, core_err_cnt,
1755 m->addr >> PAGE_SHIFT, m->addr & ~PAGE_MASK, 0,
1756 channel, dimm, -1,
1757 optype, msg);
1758 return;
1759 err_parsing:
1760 edac_mc_handle_error(tp_event, mci, core_err_cnt, 0, 0, 0,
1761 -1, -1, -1,
1762 msg, "");
1763
1764 }
1765
1766 /*
1767 * sbridge_check_error Retrieve and process errors reported by the
1768 * hardware. Called by the Core module.
1769 */
1770 static void sbridge_check_error(struct mem_ctl_info *mci)
1771 {
1772 struct sbridge_pvt *pvt = mci->pvt_info;
1773 int i;
1774 unsigned count = 0;
1775 struct mce *m;
1776
1777 /*
1778 * MCE first step: Copy all mce errors into a temporary buffer
1779 * We use a double buffering here, to reduce the risk of
1780 * loosing an error.
1781 */
1782 smp_rmb();
1783 count = (pvt->mce_out + MCE_LOG_LEN - pvt->mce_in)
1784 % MCE_LOG_LEN;
1785 if (!count)
1786 return;
1787
1788 m = pvt->mce_outentry;
1789 if (pvt->mce_in + count > MCE_LOG_LEN) {
1790 unsigned l = MCE_LOG_LEN - pvt->mce_in;
1791
1792 memcpy(m, &pvt->mce_entry[pvt->mce_in], sizeof(*m) * l);
1793 smp_wmb();
1794 pvt->mce_in = 0;
1795 count -= l;
1796 m += l;
1797 }
1798 memcpy(m, &pvt->mce_entry[pvt->mce_in], sizeof(*m) * count);
1799 smp_wmb();
1800 pvt->mce_in += count;
1801
1802 smp_rmb();
1803 if (pvt->mce_overrun) {
1804 sbridge_printk(KERN_ERR, "Lost %d memory errors\n",
1805 pvt->mce_overrun);
1806 smp_wmb();
1807 pvt->mce_overrun = 0;
1808 }
1809
1810 /*
1811 * MCE second step: parse errors and display
1812 */
1813 for (i = 0; i < count; i++)
1814 sbridge_mce_output_error(mci, &pvt->mce_outentry[i]);
1815 }
1816
1817 /*
1818 * sbridge_mce_check_error Replicates mcelog routine to get errors
1819 * This routine simply queues mcelog errors, and
1820 * return. The error itself should be handled later
1821 * by sbridge_check_error.
1822 * WARNING: As this routine should be called at NMI time, extra care should
1823 * be taken to avoid deadlocks, and to be as fast as possible.
1824 */
1825 static int sbridge_mce_check_error(struct notifier_block *nb, unsigned long val,
1826 void *data)
1827 {
1828 struct mce *mce = (struct mce *)data;
1829 struct mem_ctl_info *mci;
1830 struct sbridge_pvt *pvt;
1831
1832 if (get_edac_report_status() == EDAC_REPORTING_DISABLED)
1833 return NOTIFY_DONE;
1834
1835 mci = get_mci_for_node_id(mce->socketid);
1836 if (!mci)
1837 return NOTIFY_BAD;
1838 pvt = mci->pvt_info;
1839
1840 /*
1841 * Just let mcelog handle it if the error is
1842 * outside the memory controller. A memory error
1843 * is indicated by bit 7 = 1 and bits = 8-11,13-15 = 0.
1844 * bit 12 has an special meaning.
1845 */
1846 if ((mce->status & 0xefff) >> 7 != 1)
1847 return NOTIFY_DONE;
1848
1849 printk("sbridge: HANDLING MCE MEMORY ERROR\n");
1850
1851 printk("CPU %d: Machine Check Exception: %Lx Bank %d: %016Lx\n",
1852 mce->extcpu, mce->mcgstatus, mce->bank, mce->status);
1853 printk("TSC %llx ", mce->tsc);
1854 printk("ADDR %llx ", mce->addr);
1855 printk("MISC %llx ", mce->misc);
1856
1857 printk("PROCESSOR %u:%x TIME %llu SOCKET %u APIC %x\n",
1858 mce->cpuvendor, mce->cpuid, mce->time,
1859 mce->socketid, mce->apicid);
1860
1861 /* Only handle if it is the right mc controller */
1862 if (cpu_data(mce->cpu).phys_proc_id != pvt->sbridge_dev->mc)
1863 return NOTIFY_DONE;
1864
1865 smp_rmb();
1866 if ((pvt->mce_out + 1) % MCE_LOG_LEN == pvt->mce_in) {
1867 smp_wmb();
1868 pvt->mce_overrun++;
1869 return NOTIFY_DONE;
1870 }
1871
1872 /* Copy memory error at the ringbuffer */
1873 memcpy(&pvt->mce_entry[pvt->mce_out], mce, sizeof(*mce));
1874 smp_wmb();
1875 pvt->mce_out = (pvt->mce_out + 1) % MCE_LOG_LEN;
1876
1877 /* Handle fatal errors immediately */
1878 if (mce->mcgstatus & 1)
1879 sbridge_check_error(mci);
1880
1881 /* Advice mcelog that the error were handled */
1882 return NOTIFY_STOP;
1883 }
1884
1885 static struct notifier_block sbridge_mce_dec = {
1886 .notifier_call = sbridge_mce_check_error,
1887 };
1888
1889 /****************************************************************************
1890 EDAC register/unregister logic
1891 ****************************************************************************/
1892
1893 static void sbridge_unregister_mci(struct sbridge_dev *sbridge_dev)
1894 {
1895 struct mem_ctl_info *mci = sbridge_dev->mci;
1896 struct sbridge_pvt *pvt;
1897
1898 if (unlikely(!mci || !mci->pvt_info)) {
1899 edac_dbg(0, "MC: dev = %p\n", &sbridge_dev->pdev[0]->dev);
1900
1901 sbridge_printk(KERN_ERR, "Couldn't find mci handler\n");
1902 return;
1903 }
1904
1905 pvt = mci->pvt_info;
1906
1907 edac_dbg(0, "MC: mci = %p, dev = %p\n",
1908 mci, &sbridge_dev->pdev[0]->dev);
1909
1910 /* Remove MC sysfs nodes */
1911 edac_mc_del_mc(mci->pdev);
1912
1913 edac_dbg(1, "%s: free mci struct\n", mci->ctl_name);
1914 kfree(mci->ctl_name);
1915 edac_mc_free(mci);
1916 sbridge_dev->mci = NULL;
1917 }
1918
1919 static int sbridge_register_mci(struct sbridge_dev *sbridge_dev, enum type type)
1920 {
1921 struct mem_ctl_info *mci;
1922 struct edac_mc_layer layers[2];
1923 struct sbridge_pvt *pvt;
1924 struct pci_dev *pdev = sbridge_dev->pdev[0];
1925 int rc;
1926
1927 /* Check the number of active and not disabled channels */
1928 rc = check_if_ecc_is_active(sbridge_dev->bus);
1929 if (unlikely(rc < 0))
1930 return rc;
1931
1932 /* allocate a new MC control structure */
1933 layers[0].type = EDAC_MC_LAYER_CHANNEL;
1934 layers[0].size = NUM_CHANNELS;
1935 layers[0].is_virt_csrow = false;
1936 layers[1].type = EDAC_MC_LAYER_SLOT;
1937 layers[1].size = MAX_DIMMS;
1938 layers[1].is_virt_csrow = true;
1939 mci = edac_mc_alloc(sbridge_dev->mc, ARRAY_SIZE(layers), layers,
1940 sizeof(*pvt));
1941
1942 if (unlikely(!mci))
1943 return -ENOMEM;
1944
1945 edac_dbg(0, "MC: mci = %p, dev = %p\n",
1946 mci, &pdev->dev);
1947
1948 pvt = mci->pvt_info;
1949 memset(pvt, 0, sizeof(*pvt));
1950
1951 /* Associate sbridge_dev and mci for future usage */
1952 pvt->sbridge_dev = sbridge_dev;
1953 sbridge_dev->mci = mci;
1954
1955 mci->mtype_cap = MEM_FLAG_DDR3;
1956 mci->edac_ctl_cap = EDAC_FLAG_NONE;
1957 mci->edac_cap = EDAC_FLAG_NONE;
1958 mci->mod_name = "sbridge_edac.c";
1959 mci->mod_ver = SBRIDGE_REVISION;
1960 mci->dev_name = pci_name(pdev);
1961 mci->ctl_page_to_phys = NULL;
1962
1963 /* Set the function pointer to an actual operation function */
1964 mci->edac_check = sbridge_check_error;
1965
1966 pvt->info.type = type;
1967 if (type == IVY_BRIDGE) {
1968 pvt->info.rankcfgr = IB_RANK_CFG_A;
1969 pvt->info.get_tolm = ibridge_get_tolm;
1970 pvt->info.get_tohm = ibridge_get_tohm;
1971 pvt->info.dram_rule = ibridge_dram_rule;
1972 pvt->info.max_sad = ARRAY_SIZE(ibridge_dram_rule);
1973 pvt->info.interleave_list = ibridge_interleave_list;
1974 pvt->info.max_interleave = ARRAY_SIZE(ibridge_interleave_list);
1975 pvt->info.interleave_pkg = ibridge_interleave_pkg;
1976 mci->ctl_name = kasprintf(GFP_KERNEL, "Ivy Bridge Socket#%d", mci->mc_idx);
1977
1978 /* Store pci devices at mci for faster access */
1979 rc = ibridge_mci_bind_devs(mci, sbridge_dev);
1980 if (unlikely(rc < 0))
1981 goto fail0;
1982 } else {
1983 pvt->info.rankcfgr = SB_RANK_CFG_A;
1984 pvt->info.get_tolm = sbridge_get_tolm;
1985 pvt->info.get_tohm = sbridge_get_tohm;
1986 pvt->info.dram_rule = sbridge_dram_rule;
1987 pvt->info.max_sad = ARRAY_SIZE(sbridge_dram_rule);
1988 pvt->info.interleave_list = sbridge_interleave_list;
1989 pvt->info.max_interleave = ARRAY_SIZE(sbridge_interleave_list);
1990 pvt->info.interleave_pkg = sbridge_interleave_pkg;
1991 mci->ctl_name = kasprintf(GFP_KERNEL, "Sandy Bridge Socket#%d", mci->mc_idx);
1992
1993 /* Store pci devices at mci for faster access */
1994 rc = sbridge_mci_bind_devs(mci, sbridge_dev);
1995 if (unlikely(rc < 0))
1996 goto fail0;
1997 }
1998
1999
2000 /* Get dimm basic config and the memory layout */
2001 get_dimm_config(mci);
2002 get_memory_layout(mci);
2003
2004 /* record ptr to the generic device */
2005 mci->pdev = &pdev->dev;
2006
2007 /* add this new MC control structure to EDAC's list of MCs */
2008 if (unlikely(edac_mc_add_mc(mci))) {
2009 edac_dbg(0, "MC: failed edac_mc_add_mc()\n");
2010 rc = -EINVAL;
2011 goto fail0;
2012 }
2013
2014 return 0;
2015
2016 fail0:
2017 kfree(mci->ctl_name);
2018 edac_mc_free(mci);
2019 sbridge_dev->mci = NULL;
2020 return rc;
2021 }
2022
2023 /*
2024 * sbridge_probe Probe for ONE instance of device to see if it is
2025 * present.
2026 * return:
2027 * 0 for FOUND a device
2028 * < 0 for error code
2029 */
2030
2031 static int sbridge_probe(struct pci_dev *pdev, const struct pci_device_id *id)
2032 {
2033 int rc;
2034 u8 mc, num_mc = 0;
2035 struct sbridge_dev *sbridge_dev;
2036 enum type type;
2037
2038 /* get the pci devices we want to reserve for our use */
2039 mutex_lock(&sbridge_edac_lock);
2040
2041 /*
2042 * All memory controllers are allocated at the first pass.
2043 */
2044 if (unlikely(probed >= 1)) {
2045 mutex_unlock(&sbridge_edac_lock);
2046 return -ENODEV;
2047 }
2048 probed++;
2049
2050 if (pdev->device == PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TA) {
2051 rc = sbridge_get_all_devices(&num_mc, pci_dev_descr_ibridge_table);
2052 type = IVY_BRIDGE;
2053 } else {
2054 rc = sbridge_get_all_devices(&num_mc, pci_dev_descr_sbridge_table);
2055 type = SANDY_BRIDGE;
2056 }
2057 if (unlikely(rc < 0))
2058 goto fail0;
2059 mc = 0;
2060
2061 list_for_each_entry(sbridge_dev, &sbridge_edac_list, list) {
2062 edac_dbg(0, "Registering MC#%d (%d of %d)\n",
2063 mc, mc + 1, num_mc);
2064 sbridge_dev->mc = mc++;
2065 rc = sbridge_register_mci(sbridge_dev, type);
2066 if (unlikely(rc < 0))
2067 goto fail1;
2068 }
2069
2070 sbridge_printk(KERN_INFO, "Driver loaded.\n");
2071
2072 mutex_unlock(&sbridge_edac_lock);
2073 return 0;
2074
2075 fail1:
2076 list_for_each_entry(sbridge_dev, &sbridge_edac_list, list)
2077 sbridge_unregister_mci(sbridge_dev);
2078
2079 sbridge_put_all_devices();
2080 fail0:
2081 mutex_unlock(&sbridge_edac_lock);
2082 return rc;
2083 }
2084
2085 /*
2086 * sbridge_remove destructor for one instance of device
2087 *
2088 */
2089 static void sbridge_remove(struct pci_dev *pdev)
2090 {
2091 struct sbridge_dev *sbridge_dev;
2092
2093 edac_dbg(0, "\n");
2094
2095 /*
2096 * we have a trouble here: pdev value for removal will be wrong, since
2097 * it will point to the X58 register used to detect that the machine
2098 * is a Nehalem or upper design. However, due to the way several PCI
2099 * devices are grouped together to provide MC functionality, we need
2100 * to use a different method for releasing the devices
2101 */
2102
2103 mutex_lock(&sbridge_edac_lock);
2104
2105 if (unlikely(!probed)) {
2106 mutex_unlock(&sbridge_edac_lock);
2107 return;
2108 }
2109
2110 list_for_each_entry(sbridge_dev, &sbridge_edac_list, list)
2111 sbridge_unregister_mci(sbridge_dev);
2112
2113 /* Release PCI resources */
2114 sbridge_put_all_devices();
2115
2116 probed--;
2117
2118 mutex_unlock(&sbridge_edac_lock);
2119 }
2120
2121 MODULE_DEVICE_TABLE(pci, sbridge_pci_tbl);
2122
2123 /*
2124 * sbridge_driver pci_driver structure for this module
2125 *
2126 */
2127 static struct pci_driver sbridge_driver = {
2128 .name = "sbridge_edac",
2129 .probe = sbridge_probe,
2130 .remove = sbridge_remove,
2131 .id_table = sbridge_pci_tbl,
2132 };
2133
2134 /*
2135 * sbridge_init Module entry function
2136 * Try to initialize this module for its devices
2137 */
2138 static int __init sbridge_init(void)
2139 {
2140 int pci_rc;
2141
2142 edac_dbg(2, "\n");
2143
2144 /* Ensure that the OPSTATE is set correctly for POLL or NMI */
2145 opstate_init();
2146
2147 pci_rc = pci_register_driver(&sbridge_driver);
2148 if (pci_rc >= 0) {
2149 mce_register_decode_chain(&sbridge_mce_dec);
2150 if (get_edac_report_status() == EDAC_REPORTING_DISABLED)
2151 sbridge_printk(KERN_WARNING, "Loading driver, error reporting disabled.\n");
2152 return 0;
2153 }
2154
2155 sbridge_printk(KERN_ERR, "Failed to register device with error %d.\n",
2156 pci_rc);
2157
2158 return pci_rc;
2159 }
2160
2161 /*
2162 * sbridge_exit() Module exit function
2163 * Unregister the driver
2164 */
2165 static void __exit sbridge_exit(void)
2166 {
2167 edac_dbg(2, "\n");
2168 pci_unregister_driver(&sbridge_driver);
2169 mce_unregister_decode_chain(&sbridge_mce_dec);
2170 }
2171
2172 module_init(sbridge_init);
2173 module_exit(sbridge_exit);
2174
2175 module_param(edac_op_state, int, 0444);
2176 MODULE_PARM_DESC(edac_op_state, "EDAC Error Reporting state: 0=Poll,1=NMI");
2177
2178 MODULE_LICENSE("GPL");
2179 MODULE_AUTHOR("Mauro Carvalho Chehab <mchehab@redhat.com>");
2180 MODULE_AUTHOR("Red Hat Inc. (http://www.redhat.com)");
2181 MODULE_DESCRIPTION("MC Driver for Intel Sandy Bridge and Ivy Bridge memory controllers - "
2182 SBRIDGE_REVISION);
Linux with added FPC-III support