Linux Kernel  3.7.1
 All Data Structures Namespaces Files Functions Variables Typedefs Enumerations Enumerator Macros Groups Pages
sb_edac.c
Go to the documentation of this file.
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 <[email protected]>
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: 1.0.0 "
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
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 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 
97 static 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 
111 static 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 
126 static 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 
170 static 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 
200 static 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 
210 static 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 
218 static 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 
229 static 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 
246 static 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 
255 static 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 
276 struct sbridge_info {
278 };
279 
283 };
284 
285 struct pci_id_descr {
286  int dev;
287  int func;
288  int dev_id;
289  int optional;
290 };
291 
292 struct pci_id_table {
293  const struct pci_id_descr *descr;
294  int n_devs;
295 };
296 
297 struct sbridge_dev {
298  struct list_head list;
301  struct pci_dev **pdev;
302  int n_devs;
303  struct mem_ctl_info *mci;
304 };
305 
306 struct sbridge_pvt {
309  struct pci_dev *pci_br;
311 
313 
316 
317  /* Memory type detection */
319 
320  /* Fifo double buffers */
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 */
332 };
333 
334 #define PCI_DESCR(device, function, device_id) \
335  .dev = (device), \
336  .func = (function), \
337  .dev_id = (device_id)
338 
339 static const struct pci_id_descr pci_dev_descr_sbridge[] = {
340  /* Processor Home Agent */
342 
343  /* Memory controller */
351 
352  /* System Address Decoder */
355 
356  /* Broadcast Registers */
358 };
359 
360 #define PCI_ID_TABLE_ENTRY(A) { .descr=A, .n_devs = ARRAY_SIZE(A) }
361 static 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  */
369 static DEFINE_PCI_DEVICE_TABLE(sbridge_pci_tbl) = {
371  {0,} /* 0 terminated list. */
372 };
373 
374 
375 /****************************************************************************
376  Ancillary status routines
377  ****************************************************************************/
378 
379 static inline int numrank(u32 mtr)
380 {
381  int ranks = (1 << RANK_CNT_BITS(mtr));
382 
383  if (ranks > 4) {
384  edac_dbg(0, "Invalid number of ranks: %d (max = 4) raw value = %x (%04x)\n",
385  ranks, (unsigned int)RANK_CNT_BITS(mtr), mtr);
386  return -EINVAL;
387  }
388 
389  return ranks;
390 }
391 
392 static inline int numrow(u32 mtr)
393 {
394  int rows = (RANK_WIDTH_BITS(mtr) + 12);
395 
396  if (rows < 13 || rows > 18) {
397  edac_dbg(0, "Invalid number of rows: %d (should be between 14 and 17) raw value = %x (%04x)\n",
398  rows, (unsigned int)RANK_WIDTH_BITS(mtr), mtr);
399  return -EINVAL;
400  }
401 
402  return 1 << rows;
403 }
404 
405 static inline int numcol(u32 mtr)
406 {
407  int cols = (COL_WIDTH_BITS(mtr) + 10);
408 
409  if (cols > 12) {
410  edac_dbg(0, "Invalid number of cols: %d (max = 4) raw value = %x (%04x)\n",
411  cols, (unsigned int)COL_WIDTH_BITS(mtr), mtr);
412  return -EINVAL;
413  }
414 
415  return 1 << cols;
416 }
417 
418 static 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 
430 static 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 
453 static 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  ****************************************************************************/
463 static 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  edac_dbg(1, "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 
491 static int check_if_ecc_is_active(const u8 bus)
492 {
493  struct pci_dev *pdev = NULL;
494  u32 mcmtr;
495 
496  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  }
509  return 0;
510 }
511 
512 static int get_dimm_config(struct mem_ctl_info *mci)
513 {
514  struct sbridge_pvt *pvt = mci->pvt_info;
515  struct dimm_info *dimm;
516  unsigned i, j, banks, ranks, rows, cols, npages;
517  u64 size;
518  u32 reg;
519  enum edac_type mode;
520  enum mem_type mtype;
521 
522  pci_read_config_dword(pvt->pci_br, SAD_TARGET, &reg);
523  pvt->sbridge_dev->source_id = SOURCE_ID(reg);
524 
525  pci_read_config_dword(pvt->pci_br, SAD_CONTROL, &reg);
526  pvt->sbridge_dev->node_id = NODE_ID(reg);
527  edac_dbg(0, "mc#%d: Node ID: %d, source ID: %d\n",
528  pvt->sbridge_dev->mc,
529  pvt->sbridge_dev->node_id,
530  pvt->sbridge_dev->source_id);
531 
532  pci_read_config_dword(pvt->pci_ras, RASENABLES, &reg);
533  if (IS_MIRROR_ENABLED(reg)) {
534  edac_dbg(0, "Memory mirror is enabled\n");
535  pvt->is_mirrored = true;
536  } else {
537  edac_dbg(0, "Memory mirror is disabled\n");
538  pvt->is_mirrored = false;
539  }
540 
541  pci_read_config_dword(pvt->pci_ta, MCMTR, &pvt->info.mcmtr);
542  if (IS_LOCKSTEP_ENABLED(pvt->info.mcmtr)) {
543  edac_dbg(0, "Lockstep is enabled\n");
544  mode = EDAC_S8ECD8ED;
545  pvt->is_lockstep = true;
546  } else {
547  edac_dbg(0, "Lockstep is disabled\n");
548  mode = EDAC_S4ECD4ED;
549  pvt->is_lockstep = false;
550  }
551  if (IS_CLOSE_PG(pvt->info.mcmtr)) {
552  edac_dbg(0, "address map is on closed page mode\n");
553  pvt->is_close_pg = true;
554  } else {
555  edac_dbg(0, "address map is on open page mode\n");
556  pvt->is_close_pg = false;
557  }
558 
559  pci_read_config_dword(pvt->pci_ddrio, RANK_CFG_A, &reg);
560  if (IS_RDIMM_ENABLED(reg)) {
561  /* FIXME: Can also be LRDIMM */
562  edac_dbg(0, "Memory is registered\n");
563  mtype = MEM_RDDR3;
564  } else {
565  edac_dbg(0, "Memory is unregistered\n");
566  mtype = MEM_DDR3;
567  }
568 
569  /* On all supported DDR3 DIMM types, there are 8 banks available */
570  banks = 8;
571 
572  for (i = 0; i < NUM_CHANNELS; i++) {
573  u32 mtr;
574 
575  for (j = 0; j < ARRAY_SIZE(mtr_regs); j++) {
576  dimm = EDAC_DIMM_PTR(mci->layers, mci->dimms, mci->n_layers,
577  i, j, 0);
578  pci_read_config_dword(pvt->pci_tad[i],
579  mtr_regs[j], &mtr);
580  edac_dbg(4, "Channel #%d MTR%d = %x\n", i, j, mtr);
581  if (IS_DIMM_PRESENT(mtr)) {
582  pvt->channel[i].dimms++;
583 
584  ranks = numrank(mtr);
585  rows = numrow(mtr);
586  cols = numcol(mtr);
587 
588  /* DDR3 has 8 I/O banks */
589  size = ((u64)rows * cols * banks * ranks) >> (20 - 3);
590  npages = MiB_TO_PAGES(size);
591 
592  edac_dbg(0, "mc#%d: channel %d, dimm %d, %Ld Mb (%d pages) bank: %d, rank: %d, row: %#x, col: %#x\n",
593  pvt->sbridge_dev->mc, i, j,
594  size, npages,
595  banks, ranks, rows, cols);
596 
597  dimm->nr_pages = npages;
598  dimm->grain = 32;
599  dimm->dtype = (banks == 8) ? DEV_X8 : DEV_X4;
600  dimm->mtype = mtype;
601  dimm->edac_mode = mode;
602  snprintf(dimm->label, sizeof(dimm->label),
603  "CPU_SrcID#%u_Channel#%u_DIMM#%u",
604  pvt->sbridge_dev->source_id, i, j);
605  }
606  }
607  }
608 
609  return 0;
610 }
611 
612 static void get_memory_layout(const struct mem_ctl_info *mci)
613 {
614  struct sbridge_pvt *pvt = mci->pvt_info;
615  int i, j, k, n_sads, n_tads, sad_interl;
616  u32 reg;
617  u64 limit, prv = 0;
618  u64 tmp_mb;
619  u32 mb, kb;
620  u32 rir_way;
621 
622  /*
623  * Step 1) Get TOLM/TOHM ranges
624  */
625 
626  /* Address range is 32:28 */
627  pci_read_config_dword(pvt->pci_sad1, TOLM,
628  &reg);
629  pvt->tolm = GET_TOLM(reg);
630  tmp_mb = (1 + pvt->tolm) >> 20;
631 
632  mb = div_u64_rem(tmp_mb, 1000, &kb);
633  edac_dbg(0, "TOLM: %u.%03u GB (0x%016Lx)\n", mb, kb, (u64)pvt->tolm);
634 
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 
641  mb = div_u64_rem(tmp_mb, 1000, &kb);
642  edac_dbg(0, "TOHM: %u.%03u GB (0x%016Lx)", mb, kb, (u64)pvt->tohm);
643 
644  /*
645  * Step 2) Get SAD range and SAD Interleave list
646  * TAD registers contain the interleave wayness. However, it
647  * seems simpler to just discover it indirectly, with the
648  * algorithm bellow.
649  */
650  prv = 0;
651  for (n_sads = 0; n_sads < MAX_SAD; n_sads++) {
652  /* SAD_LIMIT Address range is 45:26 */
653  pci_read_config_dword(pvt->pci_sad0, dram_rule[n_sads],
654  &reg);
655  limit = SAD_LIMIT(reg);
656 
657  if (!DRAM_RULE_ENABLE(reg))
658  continue;
659 
660  if (limit <= prv)
661  break;
662 
663  tmp_mb = (limit + 1) >> 20;
664  mb = div_u64_rem(tmp_mb, 1000, &kb);
665  edac_dbg(0, "SAD#%d %s up to %u.%03u GB (0x%016Lx) Interleave: %s reg=0x%08x\n",
666  n_sads,
667  get_dram_attr(reg),
668  mb, kb,
669  ((u64)tmp_mb) << 20L,
670  INTERLEAVE_MODE(reg) ? "8:6" : "[8:6]XOR[18:16]",
671  reg);
672  prv = limit;
673 
674  pci_read_config_dword(pvt->pci_sad0, interleave_list[n_sads],
675  &reg);
676  sad_interl = sad_pkg(reg, 0);
677  for (j = 0; j < 8; j++) {
678  if (j > 0 && sad_interl == sad_pkg(reg, j))
679  break;
680 
681  edac_dbg(0, "SAD#%d, interleave #%d: %d\n",
682  n_sads, j, sad_pkg(reg, j));
683  }
684  }
685 
686  /*
687  * Step 3) Get TAD range
688  */
689  prv = 0;
690  for (n_tads = 0; n_tads < MAX_TAD; n_tads++) {
691  pci_read_config_dword(pvt->pci_ha0, tad_dram_rule[n_tads],
692  &reg);
693  limit = TAD_LIMIT(reg);
694  if (limit <= prv)
695  break;
696  tmp_mb = (limit + 1) >> 20;
697 
698  mb = div_u64_rem(tmp_mb, 1000, &kb);
699  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",
700  n_tads, mb, kb,
701  ((u64)tmp_mb) << 20L,
702  (u32)TAD_SOCK(reg),
703  (u32)TAD_CH(reg),
704  (u32)TAD_TGT0(reg),
705  (u32)TAD_TGT1(reg),
706  (u32)TAD_TGT2(reg),
707  (u32)TAD_TGT3(reg),
708  reg);
709  prv = limit;
710  }
711 
712  /*
713  * Step 4) Get TAD offsets, per each channel
714  */
715  for (i = 0; i < NUM_CHANNELS; i++) {
716  if (!pvt->channel[i].dimms)
717  continue;
718  for (j = 0; j < n_tads; j++) {
719  pci_read_config_dword(pvt->pci_tad[i],
720  tad_ch_nilv_offset[j],
721  &reg);
722  tmp_mb = TAD_OFFSET(reg) >> 20;
723  mb = div_u64_rem(tmp_mb, 1000, &kb);
724  edac_dbg(0, "TAD CH#%d, offset #%d: %u.%03u GB (0x%016Lx), reg=0x%08x\n",
725  i, j,
726  mb, kb,
727  ((u64)tmp_mb) << 20L,
728  reg);
729  }
730  }
731 
732  /*
733  * Step 6) Get RIR Wayness/Limit, per each channel
734  */
735  for (i = 0; i < NUM_CHANNELS; i++) {
736  if (!pvt->channel[i].dimms)
737  continue;
738  for (j = 0; j < MAX_RIR_RANGES; j++) {
739  pci_read_config_dword(pvt->pci_tad[i],
740  rir_way_limit[j],
741  &reg);
742 
743  if (!IS_RIR_VALID(reg))
744  continue;
745 
746  tmp_mb = RIR_LIMIT(reg) >> 20;
747  rir_way = 1 << RIR_WAY(reg);
748  mb = div_u64_rem(tmp_mb, 1000, &kb);
749  edac_dbg(0, "CH#%d RIR#%d, limit: %u.%03u GB (0x%016Lx), way: %d, reg=0x%08x\n",
750  i, j,
751  mb, kb,
752  ((u64)tmp_mb) << 20L,
753  rir_way,
754  reg);
755 
756  for (k = 0; k < rir_way; k++) {
757  pci_read_config_dword(pvt->pci_tad[i],
758  rir_offset[j][k],
759  &reg);
760  tmp_mb = RIR_OFFSET(reg) << 6;
761 
762  mb = div_u64_rem(tmp_mb, 1000, &kb);
763  edac_dbg(0, "CH#%d RIR#%d INTL#%d, offset %u.%03u GB (0x%016Lx), tgt: %d, reg=0x%08x\n",
764  i, j, k,
765  mb, kb,
766  ((u64)tmp_mb) << 20L,
767  (u32)RIR_RNK_TGT(reg),
768  reg);
769  }
770  }
771  }
772 }
773 
775 {
776  struct sbridge_dev *sbridge_dev;
777 
778  list_for_each_entry(sbridge_dev, &sbridge_edac_list, list) {
779  if (sbridge_dev->node_id == node_id)
780  return sbridge_dev->mci;
781  }
782  return NULL;
783 }
784 
785 static int get_memory_error_data(struct mem_ctl_info *mci,
786  u64 addr,
787  u8 *socket,
788  long *channel_mask,
789  u8 *rank,
790  char **area_type, char *msg)
791 {
792  struct mem_ctl_info *new_mci;
793  struct sbridge_pvt *pvt = mci->pvt_info;
794  int n_rir, n_sads, n_tads, sad_way, sck_xch;
795  int sad_interl, idx, base_ch;
796  int interleave_mode;
797  unsigned sad_interleave[MAX_INTERLEAVE];
798  u32 reg;
799  u8 ch_way,sck_way;
800  u32 tad_offset;
801  u32 rir_way;
802  u32 mb, kb;
803  u64 ch_addr, offset, limit, prv = 0;
804 
805 
806  /*
807  * Step 0) Check if the address is at special memory ranges
808  * The check bellow is probably enough to fill all cases where
809  * the error is not inside a memory, except for the legacy
810  * range (e. g. VGA addresses). It is unlikely, however, that the
811  * memory controller would generate an error on that range.
812  */
813  if ((addr > (u64) pvt->tolm) && (addr < (1LL << 32))) {
814  sprintf(msg, "Error at TOLM area, on addr 0x%08Lx", addr);
815  return -EINVAL;
816  }
817  if (addr >= (u64)pvt->tohm) {
818  sprintf(msg, "Error at MMIOH area, on addr 0x%016Lx", addr);
819  return -EINVAL;
820  }
821 
822  /*
823  * Step 1) Get socket
824  */
825  for (n_sads = 0; n_sads < MAX_SAD; n_sads++) {
826  pci_read_config_dword(pvt->pci_sad0, dram_rule[n_sads],
827  &reg);
828 
829  if (!DRAM_RULE_ENABLE(reg))
830  continue;
831 
832  limit = SAD_LIMIT(reg);
833  if (limit <= prv) {
834  sprintf(msg, "Can't discover the memory socket");
835  return -EINVAL;
836  }
837  if (addr <= limit)
838  break;
839  prv = limit;
840  }
841  if (n_sads == MAX_SAD) {
842  sprintf(msg, "Can't discover the memory socket");
843  return -EINVAL;
844  }
845  *area_type = get_dram_attr(reg);
846  interleave_mode = INTERLEAVE_MODE(reg);
847 
848  pci_read_config_dword(pvt->pci_sad0, interleave_list[n_sads],
849  &reg);
850  sad_interl = sad_pkg(reg, 0);
851  for (sad_way = 0; sad_way < 8; sad_way++) {
852  if (sad_way > 0 && sad_interl == sad_pkg(reg, sad_way))
853  break;
854  sad_interleave[sad_way] = sad_pkg(reg, sad_way);
855  edac_dbg(0, "SAD interleave #%d: %d\n",
856  sad_way, sad_interleave[sad_way]);
857  }
858  edac_dbg(0, "mc#%d: Error detected on SAD#%d: address 0x%016Lx < 0x%016Lx, Interleave [%d:6]%s\n",
859  pvt->sbridge_dev->mc,
860  n_sads,
861  addr,
862  limit,
863  sad_way + 7,
864  interleave_mode ? "" : "XOR[18:16]");
865  if (interleave_mode)
866  idx = ((addr >> 6) ^ (addr >> 16)) & 7;
867  else
868  idx = (addr >> 6) & 7;
869  switch (sad_way) {
870  case 1:
871  idx = 0;
872  break;
873  case 2:
874  idx = idx & 1;
875  break;
876  case 4:
877  idx = idx & 3;
878  break;
879  case 8:
880  break;
881  default:
882  sprintf(msg, "Can't discover socket interleave");
883  return -EINVAL;
884  }
885  *socket = sad_interleave[idx];
886  edac_dbg(0, "SAD interleave index: %d (wayness %d) = CPU socket %d\n",
887  idx, sad_way, *socket);
888 
889  /*
890  * Move to the proper node structure, in order to access the
891  * right PCI registers
892  */
893  new_mci = get_mci_for_node_id(*socket);
894  if (!new_mci) {
895  sprintf(msg, "Struct for socket #%u wasn't initialized",
896  *socket);
897  return -EINVAL;
898  }
899  mci = new_mci;
900  pvt = mci->pvt_info;
901 
902  /*
903  * Step 2) Get memory channel
904  */
905  prv = 0;
906  for (n_tads = 0; n_tads < MAX_TAD; n_tads++) {
907  pci_read_config_dword(pvt->pci_ha0, tad_dram_rule[n_tads],
908  &reg);
909  limit = TAD_LIMIT(reg);
910  if (limit <= prv) {
911  sprintf(msg, "Can't discover the memory channel");
912  return -EINVAL;
913  }
914  if (addr <= limit)
915  break;
916  prv = limit;
917  }
918  ch_way = TAD_CH(reg) + 1;
919  sck_way = TAD_SOCK(reg) + 1;
920  /*
921  * FIXME: Is it right to always use channel 0 for offsets?
922  */
923  pci_read_config_dword(pvt->pci_tad[0],
924  tad_ch_nilv_offset[n_tads],
925  &tad_offset);
926 
927  if (ch_way == 3)
928  idx = addr >> 6;
929  else
930  idx = addr >> (6 + sck_way);
931  idx = idx % ch_way;
932 
933  /*
934  * FIXME: Shouldn't we use CHN_IDX_OFFSET() here, when ch_way == 3 ???
935  */
936  switch (idx) {
937  case 0:
938  base_ch = TAD_TGT0(reg);
939  break;
940  case 1:
941  base_ch = TAD_TGT1(reg);
942  break;
943  case 2:
944  base_ch = TAD_TGT2(reg);
945  break;
946  case 3:
947  base_ch = TAD_TGT3(reg);
948  break;
949  default:
950  sprintf(msg, "Can't discover the TAD target");
951  return -EINVAL;
952  }
953  *channel_mask = 1 << base_ch;
954 
955  if (pvt->is_mirrored) {
956  *channel_mask |= 1 << ((base_ch + 2) % 4);
957  switch(ch_way) {
958  case 2:
959  case 4:
960  sck_xch = 1 << sck_way * (ch_way >> 1);
961  break;
962  default:
963  sprintf(msg, "Invalid mirror set. Can't decode addr");
964  return -EINVAL;
965  }
966  } else
967  sck_xch = (1 << sck_way) * ch_way;
968 
969  if (pvt->is_lockstep)
970  *channel_mask |= 1 << ((base_ch + 1) % 4);
971 
972  offset = TAD_OFFSET(tad_offset);
973 
974  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",
975  n_tads,
976  addr,
977  limit,
978  (u32)TAD_SOCK(reg),
979  ch_way,
980  offset,
981  idx,
982  base_ch,
983  *channel_mask);
984 
985  /* Calculate channel address */
986  /* Remove the TAD offset */
987 
988  if (offset > addr) {
989  sprintf(msg, "Can't calculate ch addr: TAD offset 0x%08Lx is too high for addr 0x%08Lx!",
990  offset, addr);
991  return -EINVAL;
992  }
993  addr -= offset;
994  /* Store the low bits [0:6] of the addr */
995  ch_addr = addr & 0x7f;
996  /* Remove socket wayness and remove 6 bits */
997  addr >>= 6;
998  addr = div_u64(addr, sck_xch);
999 #if 0
1000  /* Divide by channel way */
1001  addr = addr / ch_way;
1002 #endif
1003  /* Recover the last 6 bits */
1004  ch_addr |= addr << 6;
1005 
1006  /*
1007  * Step 3) Decode rank
1008  */
1009  for (n_rir = 0; n_rir < MAX_RIR_RANGES; n_rir++) {
1010  pci_read_config_dword(pvt->pci_tad[base_ch],
1011  rir_way_limit[n_rir],
1012  &reg);
1013 
1014  if (!IS_RIR_VALID(reg))
1015  continue;
1016 
1017  limit = RIR_LIMIT(reg);
1018  mb = div_u64_rem(limit >> 20, 1000, &kb);
1019  edac_dbg(0, "RIR#%d, limit: %u.%03u GB (0x%016Lx), way: %d\n",
1020  n_rir,
1021  mb, kb,
1022  limit,
1023  1 << RIR_WAY(reg));
1024  if (ch_addr <= limit)
1025  break;
1026  }
1027  if (n_rir == MAX_RIR_RANGES) {
1028  sprintf(msg, "Can't discover the memory rank for ch addr 0x%08Lx",
1029  ch_addr);
1030  return -EINVAL;
1031  }
1032  rir_way = RIR_WAY(reg);
1033  if (pvt->is_close_pg)
1034  idx = (ch_addr >> 6);
1035  else
1036  idx = (ch_addr >> 13); /* FIXME: Datasheet says to shift by 15 */
1037  idx %= 1 << rir_way;
1038 
1039  pci_read_config_dword(pvt->pci_tad[base_ch],
1040  rir_offset[n_rir][idx],
1041  &reg);
1042  *rank = RIR_RNK_TGT(reg);
1043 
1044  edac_dbg(0, "RIR#%d: channel address 0x%08Lx < 0x%08Lx, RIR interleave %d, index %d\n",
1045  n_rir,
1046  ch_addr,
1047  limit,
1048  rir_way,
1049  idx);
1050 
1051  return 0;
1052 }
1053 
1054 /****************************************************************************
1055  Device initialization routines: put/get, init/exit
1056  ****************************************************************************/
1057 
1058 /*
1059  * sbridge_put_all_devices 'put' all the devices that we have
1060  * reserved via 'get'
1061  */
1062 static void sbridge_put_devices(struct sbridge_dev *sbridge_dev)
1063 {
1064  int i;
1065 
1066  edac_dbg(0, "\n");
1067  for (i = 0; i < sbridge_dev->n_devs; i++) {
1068  struct pci_dev *pdev = sbridge_dev->pdev[i];
1069  if (!pdev)
1070  continue;
1071  edac_dbg(0, "Removing dev %02x:%02x.%d\n",
1072  pdev->bus->number,
1073  PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn));
1074  pci_dev_put(pdev);
1075  }
1076 }
1077 
1078 static void sbridge_put_all_devices(void)
1079 {
1080  struct sbridge_dev *sbridge_dev, *tmp;
1081 
1082  list_for_each_entry_safe(sbridge_dev, tmp, &sbridge_edac_list, list) {
1083  sbridge_put_devices(sbridge_dev);
1084  free_sbridge_dev(sbridge_dev);
1085  }
1086 }
1087 
1088 /*
1089  * sbridge_get_all_devices Find and perform 'get' operation on the MCH's
1090  * device/functions we want to reference for this driver
1091  *
1092  * Need to 'get' device 16 func 1 and func 2
1093  */
1094 static int sbridge_get_onedevice(struct pci_dev **prev,
1095  u8 *num_mc,
1096  const struct pci_id_table *table,
1097  const unsigned devno)
1098 {
1099  struct sbridge_dev *sbridge_dev;
1100  const struct pci_id_descr *dev_descr = &table->descr[devno];
1101 
1102  struct pci_dev *pdev = NULL;
1103  u8 bus = 0;
1104 
1106  "Seeking for: dev %02x.%d PCI ID %04x:%04x\n",
1107  dev_descr->dev, dev_descr->func,
1108  PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
1109 
1111  dev_descr->dev_id, *prev);
1112 
1113  if (!pdev) {
1114  if (*prev) {
1115  *prev = pdev;
1116  return 0;
1117  }
1118 
1119  if (dev_descr->optional)
1120  return 0;
1121 
1122  if (devno == 0)
1123  return -ENODEV;
1124 
1126  "Device not found: dev %02x.%d PCI ID %04x:%04x\n",
1127  dev_descr->dev, dev_descr->func,
1128  PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
1129 
1130  /* End of list, leave */
1131  return -ENODEV;
1132  }
1133  bus = pdev->bus->number;
1134 
1135  sbridge_dev = get_sbridge_dev(bus);
1136  if (!sbridge_dev) {
1137  sbridge_dev = alloc_sbridge_dev(bus, table);
1138  if (!sbridge_dev) {
1139  pci_dev_put(pdev);
1140  return -ENOMEM;
1141  }
1142  (*num_mc)++;
1143  }
1144 
1145  if (sbridge_dev->pdev[devno]) {
1147  "Duplicated device for "
1148  "dev %02x:%d.%d PCI ID %04x:%04x\n",
1149  bus, dev_descr->dev, dev_descr->func,
1150  PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
1151  pci_dev_put(pdev);
1152  return -ENODEV;
1153  }
1154 
1155  sbridge_dev->pdev[devno] = pdev;
1156 
1157  /* Sanity check */
1158  if (unlikely(PCI_SLOT(pdev->devfn) != dev_descr->dev ||
1159  PCI_FUNC(pdev->devfn) != dev_descr->func)) {
1161  "Device PCI ID %04x:%04x "
1162  "has dev %02x:%d.%d instead of dev %02x:%02x.%d\n",
1163  PCI_VENDOR_ID_INTEL, dev_descr->dev_id,
1164  bus, PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn),
1165  bus, dev_descr->dev, dev_descr->func);
1166  return -ENODEV;
1167  }
1168 
1169  /* Be sure that the device is enabled */
1170  if (unlikely(pci_enable_device(pdev) < 0)) {
1172  "Couldn't enable "
1173  "dev %02x:%d.%d PCI ID %04x:%04x\n",
1174  bus, dev_descr->dev, dev_descr->func,
1175  PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
1176  return -ENODEV;
1177  }
1178 
1179  edac_dbg(0, "Detected dev %02x:%d.%d PCI ID %04x:%04x\n",
1180  bus, dev_descr->dev, dev_descr->func,
1181  PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
1182 
1183  /*
1184  * As stated on drivers/pci/search.c, the reference count for
1185  * @from is always decremented if it is not %NULL. So, as we need
1186  * to get all devices up to null, we need to do a get for the device
1187  */
1188  pci_dev_get(pdev);
1189 
1190  *prev = pdev;
1191 
1192  return 0;
1193 }
1194 
1195 static int sbridge_get_all_devices(u8 *num_mc)
1196 {
1197  int i, rc;
1198  struct pci_dev *pdev = NULL;
1199  const struct pci_id_table *table = pci_dev_descr_sbridge_table;
1200 
1201  while (table && table->descr) {
1202  for (i = 0; i < table->n_devs; i++) {
1203  pdev = NULL;
1204  do {
1205  rc = sbridge_get_onedevice(&pdev, num_mc,
1206  table, i);
1207  if (rc < 0) {
1208  if (i == 0) {
1209  i = table->n_devs;
1210  break;
1211  }
1212  sbridge_put_all_devices();
1213  return -ENODEV;
1214  }
1215  } while (pdev);
1216  }
1217  table++;
1218  }
1219 
1220  return 0;
1221 }
1222 
1223 static int mci_bind_devs(struct mem_ctl_info *mci,
1224  struct sbridge_dev *sbridge_dev)
1225 {
1226  struct sbridge_pvt *pvt = mci->pvt_info;
1227  struct pci_dev *pdev;
1228  int i, func, slot;
1229 
1230  for (i = 0; i < sbridge_dev->n_devs; i++) {
1231  pdev = sbridge_dev->pdev[i];
1232  if (!pdev)
1233  continue;
1234  slot = PCI_SLOT(pdev->devfn);
1235  func = PCI_FUNC(pdev->devfn);
1236  switch (slot) {
1237  case 12:
1238  switch (func) {
1239  case 6:
1240  pvt->pci_sad0 = pdev;
1241  break;
1242  case 7:
1243  pvt->pci_sad1 = pdev;
1244  break;
1245  default:
1246  goto error;
1247  }
1248  break;
1249  case 13:
1250  switch (func) {
1251  case 6:
1252  pvt->pci_br = pdev;
1253  break;
1254  default:
1255  goto error;
1256  }
1257  break;
1258  case 14:
1259  switch (func) {
1260  case 0:
1261  pvt->pci_ha0 = pdev;
1262  break;
1263  default:
1264  goto error;
1265  }
1266  break;
1267  case 15:
1268  switch (func) {
1269  case 0:
1270  pvt->pci_ta = pdev;
1271  break;
1272  case 1:
1273  pvt->pci_ras = pdev;
1274  break;
1275  case 2:
1276  case 3:
1277  case 4:
1278  case 5:
1279  pvt->pci_tad[func - 2] = pdev;
1280  break;
1281  default:
1282  goto error;
1283  }
1284  break;
1285  case 17:
1286  switch (func) {
1287  case 0:
1288  pvt->pci_ddrio = pdev;
1289  break;
1290  default:
1291  goto error;
1292  }
1293  break;
1294  default:
1295  goto error;
1296  }
1297 
1298  edac_dbg(0, "Associated PCI %02x.%02d.%d with dev = %p\n",
1299  sbridge_dev->bus,
1300  PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn),
1301  pdev);
1302  }
1303 
1304  /* Check if everything were registered */
1305  if (!pvt->pci_sad0 || !pvt->pci_sad1 || !pvt->pci_ha0 ||
1306  !pvt-> pci_tad || !pvt->pci_ras || !pvt->pci_ta ||
1307  !pvt->pci_ddrio)
1308  goto enodev;
1309 
1310  for (i = 0; i < NUM_CHANNELS; i++) {
1311  if (!pvt->pci_tad[i])
1312  goto enodev;
1313  }
1314  return 0;
1315 
1316 enodev:
1317  sbridge_printk(KERN_ERR, "Some needed devices are missing\n");
1318  return -ENODEV;
1319 
1320 error:
1321  sbridge_printk(KERN_ERR, "Device %d, function %d "
1322  "is out of the expected range\n",
1323  slot, func);
1324  return -EINVAL;
1325 }
1326 
1327 /****************************************************************************
1328  Error check routines
1329  ****************************************************************************/
1330 
1331 /*
1332  * While Sandy Bridge has error count registers, SMI BIOS read values from
1333  * and resets the counters. So, they are not reliable for the OS to read
1334  * from them. So, we have no option but to just trust on whatever MCE is
1335  * telling us about the errors.
1336  */
1337 static void sbridge_mce_output_error(struct mem_ctl_info *mci,
1338  const struct mce *m)
1339 {
1340  struct mem_ctl_info *new_mci;
1341  struct sbridge_pvt *pvt = mci->pvt_info;
1342  enum hw_event_mc_err_type tp_event;
1343  char *type, *optype, msg[256];
1344  bool ripv = GET_BITFIELD(m->mcgstatus, 0, 0);
1345  bool overflow = GET_BITFIELD(m->status, 62, 62);
1346  bool uncorrected_error = GET_BITFIELD(m->status, 61, 61);
1347  bool recoverable = GET_BITFIELD(m->status, 56, 56);
1348  u32 core_err_cnt = GET_BITFIELD(m->status, 38, 52);
1349  u32 mscod = GET_BITFIELD(m->status, 16, 31);
1350  u32 errcode = GET_BITFIELD(m->status, 0, 15);
1351  u32 channel = GET_BITFIELD(m->status, 0, 3);
1352  u32 optypenum = GET_BITFIELD(m->status, 4, 6);
1353  long channel_mask, first_channel;
1354  u8 rank, socket;
1355  int rc, dimm;
1356  char *area_type = NULL;
1357 
1358  if (uncorrected_error) {
1359  if (ripv) {
1360  type = "FATAL";
1361  tp_event = HW_EVENT_ERR_FATAL;
1362  } else {
1363  type = "NON_FATAL";
1364  tp_event = HW_EVENT_ERR_UNCORRECTED;
1365  }
1366  } else {
1367  type = "CORRECTED";
1368  tp_event = HW_EVENT_ERR_CORRECTED;
1369  }
1370 
1371  /*
1372  * According with Table 15-9 of the Intel Architecture spec vol 3A,
1373  * memory errors should fit in this mask:
1374  * 000f 0000 1mmm cccc (binary)
1375  * where:
1376  * f = Correction Report Filtering Bit. If 1, subsequent errors
1377  * won't be shown
1378  * mmm = error type
1379  * cccc = channel
1380  * If the mask doesn't match, report an error to the parsing logic
1381  */
1382  if (! ((errcode & 0xef80) == 0x80)) {
1383  optype = "Can't parse: it is not a mem";
1384  } else {
1385  switch (optypenum) {
1386  case 0:
1387  optype = "generic undef request error";
1388  break;
1389  case 1:
1390  optype = "memory read error";
1391  break;
1392  case 2:
1393  optype = "memory write error";
1394  break;
1395  case 3:
1396  optype = "addr/cmd error";
1397  break;
1398  case 4:
1399  optype = "memory scrubbing error";
1400  break;
1401  default:
1402  optype = "reserved";
1403  break;
1404  }
1405  }
1406 
1407  rc = get_memory_error_data(mci, m->addr, &socket,
1408  &channel_mask, &rank, &area_type, msg);
1409  if (rc < 0)
1410  goto err_parsing;
1411  new_mci = get_mci_for_node_id(socket);
1412  if (!new_mci) {
1413  strcpy(msg, "Error: socket got corrupted!");
1414  goto err_parsing;
1415  }
1416  mci = new_mci;
1417  pvt = mci->pvt_info;
1418 
1419  first_channel = find_first_bit(&channel_mask, NUM_CHANNELS);
1420 
1421  if (rank < 4)
1422  dimm = 0;
1423  else if (rank < 8)
1424  dimm = 1;
1425  else
1426  dimm = 2;
1427 
1428 
1429  /*
1430  * FIXME: On some memory configurations (mirror, lockstep), the
1431  * Memory Controller can't point the error to a single DIMM. The
1432  * EDAC core should be handling the channel mask, in order to point
1433  * to the group of dimm's where the error may be happening.
1434  */
1435  snprintf(msg, sizeof(msg),
1436  "%s%s area:%s err_code:%04x:%04x socket:%d channel_mask:%ld rank:%d",
1437  overflow ? " OVERFLOW" : "",
1438  (uncorrected_error && recoverable) ? " recoverable" : "",
1439  area_type,
1440  mscod, errcode,
1441  socket,
1442  channel_mask,
1443  rank);
1444 
1445  edac_dbg(0, "%s\n", msg);
1446 
1447  /* FIXME: need support for channel mask */
1448 
1449  /* Call the helper to output message */
1450  edac_mc_handle_error(tp_event, mci, core_err_cnt,
1451  m->addr >> PAGE_SHIFT, m->addr & ~PAGE_MASK, 0,
1452  channel, dimm, -1,
1453  optype, msg);
1454  return;
1455 err_parsing:
1456  edac_mc_handle_error(tp_event, mci, core_err_cnt, 0, 0, 0,
1457  -1, -1, -1,
1458  msg, "");
1459 
1460 }
1461 
1462 /*
1463  * sbridge_check_error Retrieve and process errors reported by the
1464  * hardware. Called by the Core module.
1465  */
1466 static void sbridge_check_error(struct mem_ctl_info *mci)
1467 {
1468  struct sbridge_pvt *pvt = mci->pvt_info;
1469  int i;
1470  unsigned count = 0;
1471  struct mce *m;
1472 
1473  /*
1474  * MCE first step: Copy all mce errors into a temporary buffer
1475  * We use a double buffering here, to reduce the risk of
1476  * loosing an error.
1477  */
1478  smp_rmb();
1479  count = (pvt->mce_out + MCE_LOG_LEN - pvt->mce_in)
1480  % MCE_LOG_LEN;
1481  if (!count)
1482  return;
1483 
1484  m = pvt->mce_outentry;
1485  if (pvt->mce_in + count > MCE_LOG_LEN) {
1486  unsigned l = MCE_LOG_LEN - pvt->mce_in;
1487 
1488  memcpy(m, &pvt->mce_entry[pvt->mce_in], sizeof(*m) * l);
1489  smp_wmb();
1490  pvt->mce_in = 0;
1491  count -= l;
1492  m += l;
1493  }
1494  memcpy(m, &pvt->mce_entry[pvt->mce_in], sizeof(*m) * count);
1495  smp_wmb();
1496  pvt->mce_in += count;
1497 
1498  smp_rmb();
1499  if (pvt->mce_overrun) {
1500  sbridge_printk(KERN_ERR, "Lost %d memory errors\n",
1501  pvt->mce_overrun);
1502  smp_wmb();
1503  pvt->mce_overrun = 0;
1504  }
1505 
1506  /*
1507  * MCE second step: parse errors and display
1508  */
1509  for (i = 0; i < count; i++)
1510  sbridge_mce_output_error(mci, &pvt->mce_outentry[i]);
1511 }
1512 
1513 /*
1514  * sbridge_mce_check_error Replicates mcelog routine to get errors
1515  * This routine simply queues mcelog errors, and
1516  * return. The error itself should be handled later
1517  * by sbridge_check_error.
1518  * WARNING: As this routine should be called at NMI time, extra care should
1519  * be taken to avoid deadlocks, and to be as fast as possible.
1520  */
1521 static int sbridge_mce_check_error(struct notifier_block *nb, unsigned long val,
1522  void *data)
1523 {
1524  struct mce *mce = (struct mce *)data;
1525  struct mem_ctl_info *mci;
1526  struct sbridge_pvt *pvt;
1527 
1528  mci = get_mci_for_node_id(mce->socketid);
1529  if (!mci)
1530  return NOTIFY_BAD;
1531  pvt = mci->pvt_info;
1532 
1533  /*
1534  * Just let mcelog handle it if the error is
1535  * outside the memory controller. A memory error
1536  * is indicated by bit 7 = 1 and bits = 8-11,13-15 = 0.
1537  * bit 12 has an special meaning.
1538  */
1539  if ((mce->status & 0xefff) >> 7 != 1)
1540  return NOTIFY_DONE;
1541 
1542  printk("sbridge: HANDLING MCE MEMORY ERROR\n");
1543 
1544  printk("CPU %d: Machine Check Exception: %Lx Bank %d: %016Lx\n",
1545  mce->extcpu, mce->mcgstatus, mce->bank, mce->status);
1546  printk("TSC %llx ", mce->tsc);
1547  printk("ADDR %llx ", mce->addr);
1548  printk("MISC %llx ", mce->misc);
1549 
1550  printk("PROCESSOR %u:%x TIME %llu SOCKET %u APIC %x\n",
1551  mce->cpuvendor, mce->cpuid, mce->time,
1552  mce->socketid, mce->apicid);
1553 
1554  /* Only handle if it is the right mc controller */
1555  if (cpu_data(mce->cpu).phys_proc_id != pvt->sbridge_dev->mc)
1556  return NOTIFY_DONE;
1557 
1558  smp_rmb();
1559  if ((pvt->mce_out + 1) % MCE_LOG_LEN == pvt->mce_in) {
1560  smp_wmb();
1561  pvt->mce_overrun++;
1562  return NOTIFY_DONE;
1563  }
1564 
1565  /* Copy memory error at the ringbuffer */
1566  memcpy(&pvt->mce_entry[pvt->mce_out], mce, sizeof(*mce));
1567  smp_wmb();
1568  pvt->mce_out = (pvt->mce_out + 1) % MCE_LOG_LEN;
1569 
1570  /* Handle fatal errors immediately */
1571  if (mce->mcgstatus & 1)
1572  sbridge_check_error(mci);
1573 
1574  /* Advice mcelog that the error were handled */
1575  return NOTIFY_STOP;
1576 }
1577 
1578 static struct notifier_block sbridge_mce_dec = {
1579  .notifier_call = sbridge_mce_check_error,
1580 };
1581 
1582 /****************************************************************************
1583  EDAC register/unregister logic
1584  ****************************************************************************/
1585 
1586 static void sbridge_unregister_mci(struct sbridge_dev *sbridge_dev)
1587 {
1588  struct mem_ctl_info *mci = sbridge_dev->mci;
1589  struct sbridge_pvt *pvt;
1590 
1591  if (unlikely(!mci || !mci->pvt_info)) {
1592  edac_dbg(0, "MC: dev = %p\n", &sbridge_dev->pdev[0]->dev);
1593 
1594  sbridge_printk(KERN_ERR, "Couldn't find mci handler\n");
1595  return;
1596  }
1597 
1598  pvt = mci->pvt_info;
1599 
1600  edac_dbg(0, "MC: mci = %p, dev = %p\n",
1601  mci, &sbridge_dev->pdev[0]->dev);
1602 
1603  /* Remove MC sysfs nodes */
1604  edac_mc_del_mc(mci->pdev);
1605 
1606  edac_dbg(1, "%s: free mci struct\n", mci->ctl_name);
1607  kfree(mci->ctl_name);
1608  edac_mc_free(mci);
1609  sbridge_dev->mci = NULL;
1610 }
1611 
1612 static int sbridge_register_mci(struct sbridge_dev *sbridge_dev)
1613 {
1614  struct mem_ctl_info *mci;
1615  struct edac_mc_layer layers[2];
1616  struct sbridge_pvt *pvt;
1617  int rc;
1618 
1619  /* Check the number of active and not disabled channels */
1620  rc = check_if_ecc_is_active(sbridge_dev->bus);
1621  if (unlikely(rc < 0))
1622  return rc;
1623 
1624  /* allocate a new MC control structure */
1625  layers[0].type = EDAC_MC_LAYER_CHANNEL;
1626  layers[0].size = NUM_CHANNELS;
1627  layers[0].is_virt_csrow = false;
1628  layers[1].type = EDAC_MC_LAYER_SLOT;
1629  layers[1].size = MAX_DIMMS;
1630  layers[1].is_virt_csrow = true;
1631  mci = edac_mc_alloc(sbridge_dev->mc, ARRAY_SIZE(layers), layers,
1632  sizeof(*pvt));
1633 
1634  if (unlikely(!mci))
1635  return -ENOMEM;
1636 
1637  edac_dbg(0, "MC: mci = %p, dev = %p\n",
1638  mci, &sbridge_dev->pdev[0]->dev);
1639 
1640  pvt = mci->pvt_info;
1641  memset(pvt, 0, sizeof(*pvt));
1642 
1643  /* Associate sbridge_dev and mci for future usage */
1644  pvt->sbridge_dev = sbridge_dev;
1645  sbridge_dev->mci = mci;
1646 
1647  mci->mtype_cap = MEM_FLAG_DDR3;
1649  mci->edac_cap = EDAC_FLAG_NONE;
1650  mci->mod_name = "sbridge_edac.c";
1651  mci->mod_ver = SBRIDGE_REVISION;
1652  mci->ctl_name = kasprintf(GFP_KERNEL, "Sandy Bridge Socket#%d", mci->mc_idx);
1653  mci->dev_name = pci_name(sbridge_dev->pdev[0]);
1654  mci->ctl_page_to_phys = NULL;
1655 
1656  /* Set the function pointer to an actual operation function */
1657  mci->edac_check = sbridge_check_error;
1658 
1659  /* Store pci devices at mci for faster access */
1660  rc = mci_bind_devs(mci, sbridge_dev);
1661  if (unlikely(rc < 0))
1662  goto fail0;
1663 
1664  /* Get dimm basic config and the memory layout */
1665  get_dimm_config(mci);
1666  get_memory_layout(mci);
1667 
1668  /* record ptr to the generic device */
1669  mci->pdev = &sbridge_dev->pdev[0]->dev;
1670 
1671  /* add this new MC control structure to EDAC's list of MCs */
1672  if (unlikely(edac_mc_add_mc(mci))) {
1673  edac_dbg(0, "MC: failed edac_mc_add_mc()\n");
1674  rc = -EINVAL;
1675  goto fail0;
1676  }
1677 
1678  return 0;
1679 
1680 fail0:
1681  kfree(mci->ctl_name);
1682  edac_mc_free(mci);
1683  sbridge_dev->mci = NULL;
1684  return rc;
1685 }
1686 
1687 /*
1688  * sbridge_probe Probe for ONE instance of device to see if it is
1689  * present.
1690  * return:
1691  * 0 for FOUND a device
1692  * < 0 for error code
1693  */
1694 
1695 static int __devinit sbridge_probe(struct pci_dev *pdev,
1696  const struct pci_device_id *id)
1697 {
1698  int rc;
1699  u8 mc, num_mc = 0;
1700  struct sbridge_dev *sbridge_dev;
1701 
1702  /* get the pci devices we want to reserve for our use */
1703  mutex_lock(&sbridge_edac_lock);
1704 
1705  /*
1706  * All memory controllers are allocated at the first pass.
1707  */
1708  if (unlikely(probed >= 1)) {
1709  mutex_unlock(&sbridge_edac_lock);
1710  return -ENODEV;
1711  }
1712  probed++;
1713 
1714  rc = sbridge_get_all_devices(&num_mc);
1715  if (unlikely(rc < 0))
1716  goto fail0;
1717  mc = 0;
1718 
1719  list_for_each_entry(sbridge_dev, &sbridge_edac_list, list) {
1720  edac_dbg(0, "Registering MC#%d (%d of %d)\n",
1721  mc, mc + 1, num_mc);
1722  sbridge_dev->mc = mc++;
1723  rc = sbridge_register_mci(sbridge_dev);
1724  if (unlikely(rc < 0))
1725  goto fail1;
1726  }
1727 
1728  sbridge_printk(KERN_INFO, "Driver loaded.\n");
1729 
1730  mutex_unlock(&sbridge_edac_lock);
1731  return 0;
1732 
1733 fail1:
1734  list_for_each_entry(sbridge_dev, &sbridge_edac_list, list)
1735  sbridge_unregister_mci(sbridge_dev);
1736 
1737  sbridge_put_all_devices();
1738 fail0:
1739  mutex_unlock(&sbridge_edac_lock);
1740  return rc;
1741 }
1742 
1743 /*
1744  * sbridge_remove destructor for one instance of device
1745  *
1746  */
1747 static void __devexit sbridge_remove(struct pci_dev *pdev)
1748 {
1749  struct sbridge_dev *sbridge_dev;
1750 
1751  edac_dbg(0, "\n");
1752 
1753  /*
1754  * we have a trouble here: pdev value for removal will be wrong, since
1755  * it will point to the X58 register used to detect that the machine
1756  * is a Nehalem or upper design. However, due to the way several PCI
1757  * devices are grouped together to provide MC functionality, we need
1758  * to use a different method for releasing the devices
1759  */
1760 
1761  mutex_lock(&sbridge_edac_lock);
1762 
1763  if (unlikely(!probed)) {
1764  mutex_unlock(&sbridge_edac_lock);
1765  return;
1766  }
1767 
1768  list_for_each_entry(sbridge_dev, &sbridge_edac_list, list)
1769  sbridge_unregister_mci(sbridge_dev);
1770 
1771  /* Release PCI resources */
1772  sbridge_put_all_devices();
1773 
1774  probed--;
1775 
1776  mutex_unlock(&sbridge_edac_lock);
1777 }
1778 
1779 MODULE_DEVICE_TABLE(pci, sbridge_pci_tbl);
1780 
1781 /*
1782  * sbridge_driver pci_driver structure for this module
1783  *
1784  */
1785 static struct pci_driver sbridge_driver = {
1786  .name = "sbridge_edac",
1787  .probe = sbridge_probe,
1788  .remove = __devexit_p(sbridge_remove),
1789  .id_table = sbridge_pci_tbl,
1790 };
1791 
1792 /*
1793  * sbridge_init Module entry function
1794  * Try to initialize this module for its devices
1795  */
1796 static int __init sbridge_init(void)
1797 {
1798  int pci_rc;
1799 
1800  edac_dbg(2, "\n");
1801 
1802  /* Ensure that the OPSTATE is set correctly for POLL or NMI */
1803  opstate_init();
1804 
1805  pci_rc = pci_register_driver(&sbridge_driver);
1806 
1807  if (pci_rc >= 0) {
1808  mce_register_decode_chain(&sbridge_mce_dec);
1809  return 0;
1810  }
1811 
1812  sbridge_printk(KERN_ERR, "Failed to register device with error %d.\n",
1813  pci_rc);
1814 
1815  return pci_rc;
1816 }
1817 
1818 /*
1819  * sbridge_exit() Module exit function
1820  * Unregister the driver
1821  */
1822 static void __exit sbridge_exit(void)
1823 {
1824  edac_dbg(2, "\n");
1825  pci_unregister_driver(&sbridge_driver);
1826  mce_unregister_decode_chain(&sbridge_mce_dec);
1827 }
1828 
1829 module_init(sbridge_init);
1830 module_exit(sbridge_exit);
1831 
1832 module_param(edac_op_state, int, 0444);
1833 MODULE_PARM_DESC(edac_op_state, "EDAC Error Reporting state: 0=Poll,1=NMI");
1834 
1835 MODULE_LICENSE("GPL");
1836 MODULE_AUTHOR("Mauro Carvalho Chehab <[email protected]>");
1837 MODULE_AUTHOR("Red Hat Inc. (http://www.redhat.com)");
1838 MODULE_DESCRIPTION("MC Driver for Intel Sandy Bridge memory controllers - "