Merge tag 'perf-core-for-mingo-20160606' of git://git.kernel.org/pub/scm/linux/kernel...
[cascardo/linux.git] / drivers / edac / i7core_edac.c
1 /* Intel i7 core/Nehalem Memory Controller kernel module
2  *
3  * This driver supports the memory controllers found on the Intel
4  * processor families i7core, i7core 7xx/8xx, i5core, Xeon 35xx,
5  * Xeon 55xx and Xeon 56xx also known as Nehalem, Nehalem-EP, Lynnfield
6  * and Westmere-EP.
7  *
8  * This file may be distributed under the terms of the
9  * GNU General Public License version 2 only.
10  *
11  * Copyright (c) 2009-2010 by:
12  *       Mauro Carvalho Chehab
13  *
14  * Red Hat Inc. http://www.redhat.com
15  *
16  * Forked and adapted from the i5400_edac driver
17  *
18  * Based on the following public Intel datasheets:
19  * Intel Core i7 Processor Extreme Edition and Intel Core i7 Processor
20  * Datasheet, Volume 2:
21  *      http://download.intel.com/design/processor/datashts/320835.pdf
22  * Intel Xeon Processor 5500 Series Datasheet Volume 2
23  *      http://www.intel.com/Assets/PDF/datasheet/321322.pdf
24  * also available at:
25  *      http://www.arrownac.com/manufacturers/intel/s/nehalem/5500-datasheet-v2.pdf
26  */
27
28 #include <linux/module.h>
29 #include <linux/init.h>
30 #include <linux/pci.h>
31 #include <linux/pci_ids.h>
32 #include <linux/slab.h>
33 #include <linux/delay.h>
34 #include <linux/dmi.h>
35 #include <linux/edac.h>
36 #include <linux/mmzone.h>
37 #include <linux/smp.h>
38 #include <asm/mce.h>
39 #include <asm/processor.h>
40 #include <asm/div64.h>
41
42 #include "edac_core.h"
43
44 /* Static vars */
45 static LIST_HEAD(i7core_edac_list);
46 static DEFINE_MUTEX(i7core_edac_lock);
47 static int probed;
48
49 static int use_pci_fixup;
50 module_param(use_pci_fixup, int, 0444);
51 MODULE_PARM_DESC(use_pci_fixup, "Enable PCI fixup to seek for hidden devices");
52 /*
53  * This is used for Nehalem-EP and Nehalem-EX devices, where the non-core
54  * registers start at bus 255, and are not reported by BIOS.
55  * We currently find devices with only 2 sockets. In order to support more QPI
56  * Quick Path Interconnect, just increment this number.
57  */
58 #define MAX_SOCKET_BUSES        2
59
60
61 /*
62  * Alter this version for the module when modifications are made
63  */
64 #define I7CORE_REVISION    " Ver: 1.0.0"
65 #define EDAC_MOD_STR      "i7core_edac"
66
67 /*
68  * Debug macros
69  */
70 #define i7core_printk(level, fmt, arg...)                       \
71         edac_printk(level, "i7core", fmt, ##arg)
72
73 #define i7core_mc_printk(mci, level, fmt, arg...)               \
74         edac_mc_chipset_printk(mci, level, "i7core", fmt, ##arg)
75
76 /*
77  * i7core Memory Controller Registers
78  */
79
80         /* OFFSETS for Device 0 Function 0 */
81
82 #define MC_CFG_CONTROL  0x90
83   #define MC_CFG_UNLOCK         0x02
84   #define MC_CFG_LOCK           0x00
85
86         /* OFFSETS for Device 3 Function 0 */
87
88 #define MC_CONTROL      0x48
89 #define MC_STATUS       0x4c
90 #define MC_MAX_DOD      0x64
91
92 /*
93  * OFFSETS for Device 3 Function 4, as indicated on Xeon 5500 datasheet:
94  * http://www.arrownac.com/manufacturers/intel/s/nehalem/5500-datasheet-v2.pdf
95  */
96
97 #define MC_TEST_ERR_RCV1        0x60
98   #define DIMM2_COR_ERR(r)                      ((r) & 0x7fff)
99
100 #define MC_TEST_ERR_RCV0        0x64
101   #define DIMM1_COR_ERR(r)                      (((r) >> 16) & 0x7fff)
102   #define DIMM0_COR_ERR(r)                      ((r) & 0x7fff)
103
104 /* OFFSETS for Device 3 Function 2, as indicated on Xeon 5500 datasheet */
105 #define MC_SSRCONTROL           0x48
106   #define SSR_MODE_DISABLE      0x00
107   #define SSR_MODE_ENABLE       0x01
108   #define SSR_MODE_MASK         0x03
109
110 #define MC_SCRUB_CONTROL        0x4c
111   #define STARTSCRUB            (1 << 24)
112   #define SCRUBINTERVAL_MASK    0xffffff
113
114 #define MC_COR_ECC_CNT_0        0x80
115 #define MC_COR_ECC_CNT_1        0x84
116 #define MC_COR_ECC_CNT_2        0x88
117 #define MC_COR_ECC_CNT_3        0x8c
118 #define MC_COR_ECC_CNT_4        0x90
119 #define MC_COR_ECC_CNT_5        0x94
120
121 #define DIMM_TOP_COR_ERR(r)                     (((r) >> 16) & 0x7fff)
122 #define DIMM_BOT_COR_ERR(r)                     ((r) & 0x7fff)
123
124
125         /* OFFSETS for Devices 4,5 and 6 Function 0 */
126
127 #define MC_CHANNEL_DIMM_INIT_PARAMS 0x58
128   #define THREE_DIMMS_PRESENT           (1 << 24)
129   #define SINGLE_QUAD_RANK_PRESENT      (1 << 23)
130   #define QUAD_RANK_PRESENT             (1 << 22)
131   #define REGISTERED_DIMM               (1 << 15)
132
133 #define MC_CHANNEL_MAPPER       0x60
134   #define RDLCH(r, ch)          ((((r) >> (3 + (ch * 6))) & 0x07) - 1)
135   #define WRLCH(r, ch)          ((((r) >> (ch * 6)) & 0x07) - 1)
136
137 #define MC_CHANNEL_RANK_PRESENT 0x7c
138   #define RANK_PRESENT_MASK             0xffff
139
140 #define MC_CHANNEL_ADDR_MATCH   0xf0
141 #define MC_CHANNEL_ERROR_MASK   0xf8
142 #define MC_CHANNEL_ERROR_INJECT 0xfc
143   #define INJECT_ADDR_PARITY    0x10
144   #define INJECT_ECC            0x08
145   #define MASK_CACHELINE        0x06
146   #define MASK_FULL_CACHELINE   0x06
147   #define MASK_MSB32_CACHELINE  0x04
148   #define MASK_LSB32_CACHELINE  0x02
149   #define NO_MASK_CACHELINE     0x00
150   #define REPEAT_EN             0x01
151
152         /* OFFSETS for Devices 4,5 and 6 Function 1 */
153
154 #define MC_DOD_CH_DIMM0         0x48
155 #define MC_DOD_CH_DIMM1         0x4c
156 #define MC_DOD_CH_DIMM2         0x50
157   #define RANKOFFSET_MASK       ((1 << 12) | (1 << 11) | (1 << 10))
158   #define RANKOFFSET(x)         ((x & RANKOFFSET_MASK) >> 10)
159   #define DIMM_PRESENT_MASK     (1 << 9)
160   #define DIMM_PRESENT(x)       (((x) & DIMM_PRESENT_MASK) >> 9)
161   #define MC_DOD_NUMBANK_MASK           ((1 << 8) | (1 << 7))
162   #define MC_DOD_NUMBANK(x)             (((x) & MC_DOD_NUMBANK_MASK) >> 7)
163   #define MC_DOD_NUMRANK_MASK           ((1 << 6) | (1 << 5))
164   #define MC_DOD_NUMRANK(x)             (((x) & MC_DOD_NUMRANK_MASK) >> 5)
165   #define MC_DOD_NUMROW_MASK            ((1 << 4) | (1 << 3) | (1 << 2))
166   #define MC_DOD_NUMROW(x)              (((x) & MC_DOD_NUMROW_MASK) >> 2)
167   #define MC_DOD_NUMCOL_MASK            3
168   #define MC_DOD_NUMCOL(x)              ((x) & MC_DOD_NUMCOL_MASK)
169
170 #define MC_RANK_PRESENT         0x7c
171
172 #define MC_SAG_CH_0     0x80
173 #define MC_SAG_CH_1     0x84
174 #define MC_SAG_CH_2     0x88
175 #define MC_SAG_CH_3     0x8c
176 #define MC_SAG_CH_4     0x90
177 #define MC_SAG_CH_5     0x94
178 #define MC_SAG_CH_6     0x98
179 #define MC_SAG_CH_7     0x9c
180
181 #define MC_RIR_LIMIT_CH_0       0x40
182 #define MC_RIR_LIMIT_CH_1       0x44
183 #define MC_RIR_LIMIT_CH_2       0x48
184 #define MC_RIR_LIMIT_CH_3       0x4C
185 #define MC_RIR_LIMIT_CH_4       0x50
186 #define MC_RIR_LIMIT_CH_5       0x54
187 #define MC_RIR_LIMIT_CH_6       0x58
188 #define MC_RIR_LIMIT_CH_7       0x5C
189 #define MC_RIR_LIMIT_MASK       ((1 << 10) - 1)
190
191 #define MC_RIR_WAY_CH           0x80
192   #define MC_RIR_WAY_OFFSET_MASK        (((1 << 14) - 1) & ~0x7)
193   #define MC_RIR_WAY_RANK_MASK          0x7
194
195 /*
196  * i7core structs
197  */
198
199 #define NUM_CHANS 3
200 #define MAX_DIMMS 3             /* Max DIMMS per channel */
201 #define MAX_MCR_FUNC  4
202 #define MAX_CHAN_FUNC 3
203
204 struct i7core_info {
205         u32     mc_control;
206         u32     mc_status;
207         u32     max_dod;
208         u32     ch_map;
209 };
210
211
212 struct i7core_inject {
213         int     enable;
214
215         u32     section;
216         u32     type;
217         u32     eccmask;
218
219         /* Error address mask */
220         int channel, dimm, rank, bank, page, col;
221 };
222
223 struct i7core_channel {
224         bool            is_3dimms_present;
225         bool            is_single_4rank;
226         bool            has_4rank;
227         u32             dimms;
228 };
229
230 struct pci_id_descr {
231         int                     dev;
232         int                     func;
233         int                     dev_id;
234         int                     optional;
235 };
236
237 struct pci_id_table {
238         const struct pci_id_descr       *descr;
239         int                             n_devs;
240 };
241
242 struct i7core_dev {
243         struct list_head        list;
244         u8                      socket;
245         struct pci_dev          **pdev;
246         int                     n_devs;
247         struct mem_ctl_info     *mci;
248 };
249
250 struct i7core_pvt {
251         struct device *addrmatch_dev, *chancounts_dev;
252
253         struct pci_dev  *pci_noncore;
254         struct pci_dev  *pci_mcr[MAX_MCR_FUNC + 1];
255         struct pci_dev  *pci_ch[NUM_CHANS][MAX_CHAN_FUNC + 1];
256
257         struct i7core_dev *i7core_dev;
258
259         struct i7core_info      info;
260         struct i7core_inject    inject;
261         struct i7core_channel   channel[NUM_CHANS];
262
263         int             ce_count_available;
264
265                         /* ECC corrected errors counts per udimm */
266         unsigned long   udimm_ce_count[MAX_DIMMS];
267         int             udimm_last_ce_count[MAX_DIMMS];
268                         /* ECC corrected errors counts per rdimm */
269         unsigned long   rdimm_ce_count[NUM_CHANS][MAX_DIMMS];
270         int             rdimm_last_ce_count[NUM_CHANS][MAX_DIMMS];
271
272         bool            is_registered, enable_scrub;
273
274         /* DCLK Frequency used for computing scrub rate */
275         int                     dclk_freq;
276
277         /* Struct to control EDAC polling */
278         struct edac_pci_ctl_info *i7core_pci;
279 };
280
281 #define PCI_DESCR(device, function, device_id)  \
282         .dev = (device),                        \
283         .func = (function),                     \
284         .dev_id = (device_id)
285
286 static const struct pci_id_descr pci_dev_descr_i7core_nehalem[] = {
287                 /* Memory controller */
288         { PCI_DESCR(3, 0, PCI_DEVICE_ID_INTEL_I7_MCR)     },
289         { PCI_DESCR(3, 1, PCI_DEVICE_ID_INTEL_I7_MC_TAD)  },
290                         /* Exists only for RDIMM */
291         { PCI_DESCR(3, 2, PCI_DEVICE_ID_INTEL_I7_MC_RAS), .optional = 1  },
292         { PCI_DESCR(3, 4, PCI_DEVICE_ID_INTEL_I7_MC_TEST) },
293
294                 /* Channel 0 */
295         { PCI_DESCR(4, 0, PCI_DEVICE_ID_INTEL_I7_MC_CH0_CTRL) },
296         { PCI_DESCR(4, 1, PCI_DEVICE_ID_INTEL_I7_MC_CH0_ADDR) },
297         { PCI_DESCR(4, 2, PCI_DEVICE_ID_INTEL_I7_MC_CH0_RANK) },
298         { PCI_DESCR(4, 3, PCI_DEVICE_ID_INTEL_I7_MC_CH0_TC)   },
299
300                 /* Channel 1 */
301         { PCI_DESCR(5, 0, PCI_DEVICE_ID_INTEL_I7_MC_CH1_CTRL) },
302         { PCI_DESCR(5, 1, PCI_DEVICE_ID_INTEL_I7_MC_CH1_ADDR) },
303         { PCI_DESCR(5, 2, PCI_DEVICE_ID_INTEL_I7_MC_CH1_RANK) },
304         { PCI_DESCR(5, 3, PCI_DEVICE_ID_INTEL_I7_MC_CH1_TC)   },
305
306                 /* Channel 2 */
307         { PCI_DESCR(6, 0, PCI_DEVICE_ID_INTEL_I7_MC_CH2_CTRL) },
308         { PCI_DESCR(6, 1, PCI_DEVICE_ID_INTEL_I7_MC_CH2_ADDR) },
309         { PCI_DESCR(6, 2, PCI_DEVICE_ID_INTEL_I7_MC_CH2_RANK) },
310         { PCI_DESCR(6, 3, PCI_DEVICE_ID_INTEL_I7_MC_CH2_TC)   },
311
312                 /* Generic Non-core registers */
313         /*
314          * This is the PCI device on i7core and on Xeon 35xx (8086:2c41)
315          * On Xeon 55xx, however, it has a different id (8086:2c40). So,
316          * the probing code needs to test for the other address in case of
317          * failure of this one
318          */
319         { PCI_DESCR(0, 0, PCI_DEVICE_ID_INTEL_I7_NONCORE)  },
320
321 };
322
323 static const struct pci_id_descr pci_dev_descr_lynnfield[] = {
324         { PCI_DESCR( 3, 0, PCI_DEVICE_ID_INTEL_LYNNFIELD_MCR)         },
325         { PCI_DESCR( 3, 1, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_TAD)      },
326         { PCI_DESCR( 3, 4, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_TEST)     },
327
328         { PCI_DESCR( 4, 0, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH0_CTRL) },
329         { PCI_DESCR( 4, 1, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH0_ADDR) },
330         { PCI_DESCR( 4, 2, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH0_RANK) },
331         { PCI_DESCR( 4, 3, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH0_TC)   },
332
333         { PCI_DESCR( 5, 0, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH1_CTRL) },
334         { PCI_DESCR( 5, 1, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH1_ADDR) },
335         { PCI_DESCR( 5, 2, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH1_RANK) },
336         { PCI_DESCR( 5, 3, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH1_TC)   },
337
338         /*
339          * This is the PCI device has an alternate address on some
340          * processors like Core i7 860
341          */
342         { PCI_DESCR( 0, 0, PCI_DEVICE_ID_INTEL_LYNNFIELD_NONCORE)     },
343 };
344
345 static const struct pci_id_descr pci_dev_descr_i7core_westmere[] = {
346                 /* Memory controller */
347         { PCI_DESCR(3, 0, PCI_DEVICE_ID_INTEL_LYNNFIELD_MCR_REV2)     },
348         { PCI_DESCR(3, 1, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_TAD_REV2)  },
349                         /* Exists only for RDIMM */
350         { PCI_DESCR(3, 2, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_RAS_REV2), .optional = 1  },
351         { PCI_DESCR(3, 4, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_TEST_REV2) },
352
353                 /* Channel 0 */
354         { PCI_DESCR(4, 0, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH0_CTRL_REV2) },
355         { PCI_DESCR(4, 1, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH0_ADDR_REV2) },
356         { PCI_DESCR(4, 2, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH0_RANK_REV2) },
357         { PCI_DESCR(4, 3, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH0_TC_REV2)   },
358
359                 /* Channel 1 */
360         { PCI_DESCR(5, 0, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH1_CTRL_REV2) },
361         { PCI_DESCR(5, 1, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH1_ADDR_REV2) },
362         { PCI_DESCR(5, 2, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH1_RANK_REV2) },
363         { PCI_DESCR(5, 3, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH1_TC_REV2)   },
364
365                 /* Channel 2 */
366         { PCI_DESCR(6, 0, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH2_CTRL_REV2) },
367         { PCI_DESCR(6, 1, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH2_ADDR_REV2) },
368         { PCI_DESCR(6, 2, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH2_RANK_REV2) },
369         { PCI_DESCR(6, 3, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH2_TC_REV2)   },
370
371                 /* Generic Non-core registers */
372         { PCI_DESCR(0, 0, PCI_DEVICE_ID_INTEL_LYNNFIELD_NONCORE_REV2)  },
373
374 };
375
376 #define PCI_ID_TABLE_ENTRY(A) { .descr=A, .n_devs = ARRAY_SIZE(A) }
377 static const struct pci_id_table pci_dev_table[] = {
378         PCI_ID_TABLE_ENTRY(pci_dev_descr_i7core_nehalem),
379         PCI_ID_TABLE_ENTRY(pci_dev_descr_lynnfield),
380         PCI_ID_TABLE_ENTRY(pci_dev_descr_i7core_westmere),
381         {0,}                    /* 0 terminated list. */
382 };
383
384 /*
385  *      pci_device_id   table for which devices we are looking for
386  */
387 static const struct pci_device_id i7core_pci_tbl[] = {
388         {PCI_DEVICE(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_X58_HUB_MGMT)},
389         {PCI_DEVICE(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_LYNNFIELD_QPI_LINK0)},
390         {0,}                    /* 0 terminated list. */
391 };
392
393 /****************************************************************************
394                         Ancillary status routines
395  ****************************************************************************/
396
397         /* MC_CONTROL bits */
398 #define CH_ACTIVE(pvt, ch)      ((pvt)->info.mc_control & (1 << (8 + ch)))
399 #define ECCx8(pvt)              ((pvt)->info.mc_control & (1 << 1))
400
401         /* MC_STATUS bits */
402 #define ECC_ENABLED(pvt)        ((pvt)->info.mc_status & (1 << 4))
403 #define CH_DISABLED(pvt, ch)    ((pvt)->info.mc_status & (1 << ch))
404
405         /* MC_MAX_DOD read functions */
406 static inline int numdimms(u32 dimms)
407 {
408         return (dimms & 0x3) + 1;
409 }
410
411 static inline int numrank(u32 rank)
412 {
413         static const int ranks[] = { 1, 2, 4, -EINVAL };
414
415         return ranks[rank & 0x3];
416 }
417
418 static inline int numbank(u32 bank)
419 {
420         static const int banks[] = { 4, 8, 16, -EINVAL };
421
422         return banks[bank & 0x3];
423 }
424
425 static inline int numrow(u32 row)
426 {
427         static const int rows[] = {
428                 1 << 12, 1 << 13, 1 << 14, 1 << 15,
429                 1 << 16, -EINVAL, -EINVAL, -EINVAL,
430         };
431
432         return rows[row & 0x7];
433 }
434
435 static inline int numcol(u32 col)
436 {
437         static const int cols[] = {
438                 1 << 10, 1 << 11, 1 << 12, -EINVAL,
439         };
440         return cols[col & 0x3];
441 }
442
443 static struct i7core_dev *get_i7core_dev(u8 socket)
444 {
445         struct i7core_dev *i7core_dev;
446
447         list_for_each_entry(i7core_dev, &i7core_edac_list, list) {
448                 if (i7core_dev->socket == socket)
449                         return i7core_dev;
450         }
451
452         return NULL;
453 }
454
455 static struct i7core_dev *alloc_i7core_dev(u8 socket,
456                                            const struct pci_id_table *table)
457 {
458         struct i7core_dev *i7core_dev;
459
460         i7core_dev = kzalloc(sizeof(*i7core_dev), GFP_KERNEL);
461         if (!i7core_dev)
462                 return NULL;
463
464         i7core_dev->pdev = kzalloc(sizeof(*i7core_dev->pdev) * table->n_devs,
465                                    GFP_KERNEL);
466         if (!i7core_dev->pdev) {
467                 kfree(i7core_dev);
468                 return NULL;
469         }
470
471         i7core_dev->socket = socket;
472         i7core_dev->n_devs = table->n_devs;
473         list_add_tail(&i7core_dev->list, &i7core_edac_list);
474
475         return i7core_dev;
476 }
477
478 static void free_i7core_dev(struct i7core_dev *i7core_dev)
479 {
480         list_del(&i7core_dev->list);
481         kfree(i7core_dev->pdev);
482         kfree(i7core_dev);
483 }
484
485 /****************************************************************************
486                         Memory check routines
487  ****************************************************************************/
488
489 static int get_dimm_config(struct mem_ctl_info *mci)
490 {
491         struct i7core_pvt *pvt = mci->pvt_info;
492         struct pci_dev *pdev;
493         int i, j;
494         enum edac_type mode;
495         enum mem_type mtype;
496         struct dimm_info *dimm;
497
498         /* Get data from the MC register, function 0 */
499         pdev = pvt->pci_mcr[0];
500         if (!pdev)
501                 return -ENODEV;
502
503         /* Device 3 function 0 reads */
504         pci_read_config_dword(pdev, MC_CONTROL, &pvt->info.mc_control);
505         pci_read_config_dword(pdev, MC_STATUS, &pvt->info.mc_status);
506         pci_read_config_dword(pdev, MC_MAX_DOD, &pvt->info.max_dod);
507         pci_read_config_dword(pdev, MC_CHANNEL_MAPPER, &pvt->info.ch_map);
508
509         edac_dbg(0, "QPI %d control=0x%08x status=0x%08x dod=0x%08x map=0x%08x\n",
510                  pvt->i7core_dev->socket, pvt->info.mc_control,
511                  pvt->info.mc_status, pvt->info.max_dod, pvt->info.ch_map);
512
513         if (ECC_ENABLED(pvt)) {
514                 edac_dbg(0, "ECC enabled with x%d SDCC\n", ECCx8(pvt) ? 8 : 4);
515                 if (ECCx8(pvt))
516                         mode = EDAC_S8ECD8ED;
517                 else
518                         mode = EDAC_S4ECD4ED;
519         } else {
520                 edac_dbg(0, "ECC disabled\n");
521                 mode = EDAC_NONE;
522         }
523
524         /* FIXME: need to handle the error codes */
525         edac_dbg(0, "DOD Max limits: DIMMS: %d, %d-ranked, %d-banked x%x x 0x%x\n",
526                  numdimms(pvt->info.max_dod),
527                  numrank(pvt->info.max_dod >> 2),
528                  numbank(pvt->info.max_dod >> 4),
529                  numrow(pvt->info.max_dod >> 6),
530                  numcol(pvt->info.max_dod >> 9));
531
532         for (i = 0; i < NUM_CHANS; i++) {
533                 u32 data, dimm_dod[3], value[8];
534
535                 if (!pvt->pci_ch[i][0])
536                         continue;
537
538                 if (!CH_ACTIVE(pvt, i)) {
539                         edac_dbg(0, "Channel %i is not active\n", i);
540                         continue;
541                 }
542                 if (CH_DISABLED(pvt, i)) {
543                         edac_dbg(0, "Channel %i is disabled\n", i);
544                         continue;
545                 }
546
547                 /* Devices 4-6 function 0 */
548                 pci_read_config_dword(pvt->pci_ch[i][0],
549                                 MC_CHANNEL_DIMM_INIT_PARAMS, &data);
550
551
552                 if (data & THREE_DIMMS_PRESENT)
553                         pvt->channel[i].is_3dimms_present = true;
554
555                 if (data & SINGLE_QUAD_RANK_PRESENT)
556                         pvt->channel[i].is_single_4rank = true;
557
558                 if (data & QUAD_RANK_PRESENT)
559                         pvt->channel[i].has_4rank = true;
560
561                 if (data & REGISTERED_DIMM)
562                         mtype = MEM_RDDR3;
563                 else
564                         mtype = MEM_DDR3;
565
566                 /* Devices 4-6 function 1 */
567                 pci_read_config_dword(pvt->pci_ch[i][1],
568                                 MC_DOD_CH_DIMM0, &dimm_dod[0]);
569                 pci_read_config_dword(pvt->pci_ch[i][1],
570                                 MC_DOD_CH_DIMM1, &dimm_dod[1]);
571                 pci_read_config_dword(pvt->pci_ch[i][1],
572                                 MC_DOD_CH_DIMM2, &dimm_dod[2]);
573
574                 edac_dbg(0, "Ch%d phy rd%d, wr%d (0x%08x): %s%s%s%cDIMMs\n",
575                          i,
576                          RDLCH(pvt->info.ch_map, i), WRLCH(pvt->info.ch_map, i),
577                          data,
578                          pvt->channel[i].is_3dimms_present ? "3DIMMS " : "",
579                          pvt->channel[i].is_3dimms_present ? "SINGLE_4R " : "",
580                          pvt->channel[i].has_4rank ? "HAS_4R " : "",
581                          (data & REGISTERED_DIMM) ? 'R' : 'U');
582
583                 for (j = 0; j < 3; j++) {
584                         u32 banks, ranks, rows, cols;
585                         u32 size, npages;
586
587                         if (!DIMM_PRESENT(dimm_dod[j]))
588                                 continue;
589
590                         dimm = EDAC_DIMM_PTR(mci->layers, mci->dimms, mci->n_layers,
591                                        i, j, 0);
592                         banks = numbank(MC_DOD_NUMBANK(dimm_dod[j]));
593                         ranks = numrank(MC_DOD_NUMRANK(dimm_dod[j]));
594                         rows = numrow(MC_DOD_NUMROW(dimm_dod[j]));
595                         cols = numcol(MC_DOD_NUMCOL(dimm_dod[j]));
596
597                         /* DDR3 has 8 I/O banks */
598                         size = (rows * cols * banks * ranks) >> (20 - 3);
599
600                         edac_dbg(0, "\tdimm %d %d Mb offset: %x, bank: %d, rank: %d, row: %#x, col: %#x\n",
601                                  j, size,
602                                  RANKOFFSET(dimm_dod[j]),
603                                  banks, ranks, rows, cols);
604
605                         npages = MiB_TO_PAGES(size);
606
607                         dimm->nr_pages = npages;
608
609                         switch (banks) {
610                         case 4:
611                                 dimm->dtype = DEV_X4;
612                                 break;
613                         case 8:
614                                 dimm->dtype = DEV_X8;
615                                 break;
616                         case 16:
617                                 dimm->dtype = DEV_X16;
618                                 break;
619                         default:
620                                 dimm->dtype = DEV_UNKNOWN;
621                         }
622
623                         snprintf(dimm->label, sizeof(dimm->label),
624                                  "CPU#%uChannel#%u_DIMM#%u",
625                                  pvt->i7core_dev->socket, i, j);
626                         dimm->grain = 8;
627                         dimm->edac_mode = mode;
628                         dimm->mtype = mtype;
629                 }
630
631                 pci_read_config_dword(pdev, MC_SAG_CH_0, &value[0]);
632                 pci_read_config_dword(pdev, MC_SAG_CH_1, &value[1]);
633                 pci_read_config_dword(pdev, MC_SAG_CH_2, &value[2]);
634                 pci_read_config_dword(pdev, MC_SAG_CH_3, &value[3]);
635                 pci_read_config_dword(pdev, MC_SAG_CH_4, &value[4]);
636                 pci_read_config_dword(pdev, MC_SAG_CH_5, &value[5]);
637                 pci_read_config_dword(pdev, MC_SAG_CH_6, &value[6]);
638                 pci_read_config_dword(pdev, MC_SAG_CH_7, &value[7]);
639                 edac_dbg(1, "\t[%i] DIVBY3\tREMOVED\tOFFSET\n", i);
640                 for (j = 0; j < 8; j++)
641                         edac_dbg(1, "\t\t%#x\t%#x\t%#x\n",
642                                  (value[j] >> 27) & 0x1,
643                                  (value[j] >> 24) & 0x7,
644                                  (value[j] & ((1 << 24) - 1)));
645         }
646
647         return 0;
648 }
649
650 /****************************************************************************
651                         Error insertion routines
652  ****************************************************************************/
653
654 #define to_mci(k) container_of(k, struct mem_ctl_info, dev)
655
656 /* The i7core has independent error injection features per channel.
657    However, to have a simpler code, we don't allow enabling error injection
658    on more than one channel.
659    Also, since a change at an inject parameter will be applied only at enable,
660    we're disabling error injection on all write calls to the sysfs nodes that
661    controls the error code injection.
662  */
663 static int disable_inject(const struct mem_ctl_info *mci)
664 {
665         struct i7core_pvt *pvt = mci->pvt_info;
666
667         pvt->inject.enable = 0;
668
669         if (!pvt->pci_ch[pvt->inject.channel][0])
670                 return -ENODEV;
671
672         pci_write_config_dword(pvt->pci_ch[pvt->inject.channel][0],
673                                 MC_CHANNEL_ERROR_INJECT, 0);
674
675         return 0;
676 }
677
678 /*
679  * i7core inject inject.section
680  *
681  *      accept and store error injection inject.section value
682  *      bit 0 - refers to the lower 32-byte half cacheline
683  *      bit 1 - refers to the upper 32-byte half cacheline
684  */
685 static ssize_t i7core_inject_section_store(struct device *dev,
686                                            struct device_attribute *mattr,
687                                            const char *data, size_t count)
688 {
689         struct mem_ctl_info *mci = to_mci(dev);
690         struct i7core_pvt *pvt = mci->pvt_info;
691         unsigned long value;
692         int rc;
693
694         if (pvt->inject.enable)
695                 disable_inject(mci);
696
697         rc = kstrtoul(data, 10, &value);
698         if ((rc < 0) || (value > 3))
699                 return -EIO;
700
701         pvt->inject.section = (u32) value;
702         return count;
703 }
704
705 static ssize_t i7core_inject_section_show(struct device *dev,
706                                           struct device_attribute *mattr,
707                                           char *data)
708 {
709         struct mem_ctl_info *mci = to_mci(dev);
710         struct i7core_pvt *pvt = mci->pvt_info;
711         return sprintf(data, "0x%08x\n", pvt->inject.section);
712 }
713
714 /*
715  * i7core inject.type
716  *
717  *      accept and store error injection inject.section value
718  *      bit 0 - repeat enable - Enable error repetition
719  *      bit 1 - inject ECC error
720  *      bit 2 - inject parity error
721  */
722 static ssize_t i7core_inject_type_store(struct device *dev,
723                                         struct device_attribute *mattr,
724                                         const char *data, size_t count)
725 {
726         struct mem_ctl_info *mci = to_mci(dev);
727 struct i7core_pvt *pvt = mci->pvt_info;
728         unsigned long value;
729         int rc;
730
731         if (pvt->inject.enable)
732                 disable_inject(mci);
733
734         rc = kstrtoul(data, 10, &value);
735         if ((rc < 0) || (value > 7))
736                 return -EIO;
737
738         pvt->inject.type = (u32) value;
739         return count;
740 }
741
742 static ssize_t i7core_inject_type_show(struct device *dev,
743                                        struct device_attribute *mattr,
744                                        char *data)
745 {
746         struct mem_ctl_info *mci = to_mci(dev);
747         struct i7core_pvt *pvt = mci->pvt_info;
748
749         return sprintf(data, "0x%08x\n", pvt->inject.type);
750 }
751
752 /*
753  * i7core_inject_inject.eccmask_store
754  *
755  * The type of error (UE/CE) will depend on the inject.eccmask value:
756  *   Any bits set to a 1 will flip the corresponding ECC bit
757  *   Correctable errors can be injected by flipping 1 bit or the bits within
758  *   a symbol pair (2 consecutive aligned 8-bit pairs - i.e. 7:0 and 15:8 or
759  *   23:16 and 31:24). Flipping bits in two symbol pairs will cause an
760  *   uncorrectable error to be injected.
761  */
762 static ssize_t i7core_inject_eccmask_store(struct device *dev,
763                                            struct device_attribute *mattr,
764                                            const char *data, size_t count)
765 {
766         struct mem_ctl_info *mci = to_mci(dev);
767         struct i7core_pvt *pvt = mci->pvt_info;
768         unsigned long value;
769         int rc;
770
771         if (pvt->inject.enable)
772                 disable_inject(mci);
773
774         rc = kstrtoul(data, 10, &value);
775         if (rc < 0)
776                 return -EIO;
777
778         pvt->inject.eccmask = (u32) value;
779         return count;
780 }
781
782 static ssize_t i7core_inject_eccmask_show(struct device *dev,
783                                           struct device_attribute *mattr,
784                                           char *data)
785 {
786         struct mem_ctl_info *mci = to_mci(dev);
787         struct i7core_pvt *pvt = mci->pvt_info;
788
789         return sprintf(data, "0x%08x\n", pvt->inject.eccmask);
790 }
791
792 /*
793  * i7core_addrmatch
794  *
795  * The type of error (UE/CE) will depend on the inject.eccmask value:
796  *   Any bits set to a 1 will flip the corresponding ECC bit
797  *   Correctable errors can be injected by flipping 1 bit or the bits within
798  *   a symbol pair (2 consecutive aligned 8-bit pairs - i.e. 7:0 and 15:8 or
799  *   23:16 and 31:24). Flipping bits in two symbol pairs will cause an
800  *   uncorrectable error to be injected.
801  */
802
803 #define DECLARE_ADDR_MATCH(param, limit)                        \
804 static ssize_t i7core_inject_store_##param(                     \
805         struct device *dev,                                     \
806         struct device_attribute *mattr,                         \
807         const char *data, size_t count)                         \
808 {                                                               \
809         struct mem_ctl_info *mci = dev_get_drvdata(dev);        \
810         struct i7core_pvt *pvt;                                 \
811         long value;                                             \
812         int rc;                                                 \
813                                                                 \
814         edac_dbg(1, "\n");                                      \
815         pvt = mci->pvt_info;                                    \
816                                                                 \
817         if (pvt->inject.enable)                                 \
818                 disable_inject(mci);                            \
819                                                                 \
820         if (!strcasecmp(data, "any") || !strcasecmp(data, "any\n"))\
821                 value = -1;                                     \
822         else {                                                  \
823                 rc = kstrtoul(data, 10, &value);                \
824                 if ((rc < 0) || (value >= limit))               \
825                         return -EIO;                            \
826         }                                                       \
827                                                                 \
828         pvt->inject.param = value;                              \
829                                                                 \
830         return count;                                           \
831 }                                                               \
832                                                                 \
833 static ssize_t i7core_inject_show_##param(                      \
834         struct device *dev,                                     \
835         struct device_attribute *mattr,                         \
836         char *data)                                             \
837 {                                                               \
838         struct mem_ctl_info *mci = dev_get_drvdata(dev);        \
839         struct i7core_pvt *pvt;                                 \
840                                                                 \
841         pvt = mci->pvt_info;                                    \
842         edac_dbg(1, "pvt=%p\n", pvt);                           \
843         if (pvt->inject.param < 0)                              \
844                 return sprintf(data, "any\n");                  \
845         else                                                    \
846                 return sprintf(data, "%d\n", pvt->inject.param);\
847 }
848
849 #define ATTR_ADDR_MATCH(param)                                  \
850         static DEVICE_ATTR(param, S_IRUGO | S_IWUSR,            \
851                     i7core_inject_show_##param,                 \
852                     i7core_inject_store_##param)
853
854 DECLARE_ADDR_MATCH(channel, 3);
855 DECLARE_ADDR_MATCH(dimm, 3);
856 DECLARE_ADDR_MATCH(rank, 4);
857 DECLARE_ADDR_MATCH(bank, 32);
858 DECLARE_ADDR_MATCH(page, 0x10000);
859 DECLARE_ADDR_MATCH(col, 0x4000);
860
861 ATTR_ADDR_MATCH(channel);
862 ATTR_ADDR_MATCH(dimm);
863 ATTR_ADDR_MATCH(rank);
864 ATTR_ADDR_MATCH(bank);
865 ATTR_ADDR_MATCH(page);
866 ATTR_ADDR_MATCH(col);
867
868 static int write_and_test(struct pci_dev *dev, const int where, const u32 val)
869 {
870         u32 read;
871         int count;
872
873         edac_dbg(0, "setting pci %02x:%02x.%x reg=%02x value=%08x\n",
874                  dev->bus->number, PCI_SLOT(dev->devfn), PCI_FUNC(dev->devfn),
875                  where, val);
876
877         for (count = 0; count < 10; count++) {
878                 if (count)
879                         msleep(100);
880                 pci_write_config_dword(dev, where, val);
881                 pci_read_config_dword(dev, where, &read);
882
883                 if (read == val)
884                         return 0;
885         }
886
887         i7core_printk(KERN_ERR, "Error during set pci %02x:%02x.%x reg=%02x "
888                 "write=%08x. Read=%08x\n",
889                 dev->bus->number, PCI_SLOT(dev->devfn), PCI_FUNC(dev->devfn),
890                 where, val, read);
891
892         return -EINVAL;
893 }
894
895 /*
896  * This routine prepares the Memory Controller for error injection.
897  * The error will be injected when some process tries to write to the
898  * memory that matches the given criteria.
899  * The criteria can be set in terms of a mask where dimm, rank, bank, page
900  * and col can be specified.
901  * A -1 value for any of the mask items will make the MCU to ignore
902  * that matching criteria for error injection.
903  *
904  * It should be noticed that the error will only happen after a write operation
905  * on a memory that matches the condition. if REPEAT_EN is not enabled at
906  * inject mask, then it will produce just one error. Otherwise, it will repeat
907  * until the injectmask would be cleaned.
908  *
909  * FIXME: This routine assumes that MAXNUMDIMMS value of MC_MAX_DOD
910  *    is reliable enough to check if the MC is using the
911  *    three channels. However, this is not clear at the datasheet.
912  */
913 static ssize_t i7core_inject_enable_store(struct device *dev,
914                                           struct device_attribute *mattr,
915                                           const char *data, size_t count)
916 {
917         struct mem_ctl_info *mci = to_mci(dev);
918         struct i7core_pvt *pvt = mci->pvt_info;
919         u32 injectmask;
920         u64 mask = 0;
921         int  rc;
922         long enable;
923
924         if (!pvt->pci_ch[pvt->inject.channel][0])
925                 return 0;
926
927         rc = kstrtoul(data, 10, &enable);
928         if ((rc < 0))
929                 return 0;
930
931         if (enable) {
932                 pvt->inject.enable = 1;
933         } else {
934                 disable_inject(mci);
935                 return count;
936         }
937
938         /* Sets pvt->inject.dimm mask */
939         if (pvt->inject.dimm < 0)
940                 mask |= 1LL << 41;
941         else {
942                 if (pvt->channel[pvt->inject.channel].dimms > 2)
943                         mask |= (pvt->inject.dimm & 0x3LL) << 35;
944                 else
945                         mask |= (pvt->inject.dimm & 0x1LL) << 36;
946         }
947
948         /* Sets pvt->inject.rank mask */
949         if (pvt->inject.rank < 0)
950                 mask |= 1LL << 40;
951         else {
952                 if (pvt->channel[pvt->inject.channel].dimms > 2)
953                         mask |= (pvt->inject.rank & 0x1LL) << 34;
954                 else
955                         mask |= (pvt->inject.rank & 0x3LL) << 34;
956         }
957
958         /* Sets pvt->inject.bank mask */
959         if (pvt->inject.bank < 0)
960                 mask |= 1LL << 39;
961         else
962                 mask |= (pvt->inject.bank & 0x15LL) << 30;
963
964         /* Sets pvt->inject.page mask */
965         if (pvt->inject.page < 0)
966                 mask |= 1LL << 38;
967         else
968                 mask |= (pvt->inject.page & 0xffff) << 14;
969
970         /* Sets pvt->inject.column mask */
971         if (pvt->inject.col < 0)
972                 mask |= 1LL << 37;
973         else
974                 mask |= (pvt->inject.col & 0x3fff);
975
976         /*
977          * bit    0: REPEAT_EN
978          * bits 1-2: MASK_HALF_CACHELINE
979          * bit    3: INJECT_ECC
980          * bit    4: INJECT_ADDR_PARITY
981          */
982
983         injectmask = (pvt->inject.type & 1) |
984                      (pvt->inject.section & 0x3) << 1 |
985                      (pvt->inject.type & 0x6) << (3 - 1);
986
987         /* Unlock writes to registers - this register is write only */
988         pci_write_config_dword(pvt->pci_noncore,
989                                MC_CFG_CONTROL, 0x2);
990
991         write_and_test(pvt->pci_ch[pvt->inject.channel][0],
992                                MC_CHANNEL_ADDR_MATCH, mask);
993         write_and_test(pvt->pci_ch[pvt->inject.channel][0],
994                                MC_CHANNEL_ADDR_MATCH + 4, mask >> 32L);
995
996         write_and_test(pvt->pci_ch[pvt->inject.channel][0],
997                                MC_CHANNEL_ERROR_MASK, pvt->inject.eccmask);
998
999         write_and_test(pvt->pci_ch[pvt->inject.channel][0],
1000                                MC_CHANNEL_ERROR_INJECT, injectmask);
1001
1002         /*
1003          * This is something undocumented, based on my tests
1004          * Without writing 8 to this register, errors aren't injected. Not sure
1005          * why.
1006          */
1007         pci_write_config_dword(pvt->pci_noncore,
1008                                MC_CFG_CONTROL, 8);
1009
1010         edac_dbg(0, "Error inject addr match 0x%016llx, ecc 0x%08x, inject 0x%08x\n",
1011                  mask, pvt->inject.eccmask, injectmask);
1012
1013
1014         return count;
1015 }
1016
1017 static ssize_t i7core_inject_enable_show(struct device *dev,
1018                                          struct device_attribute *mattr,
1019                                          char *data)
1020 {
1021         struct mem_ctl_info *mci = to_mci(dev);
1022         struct i7core_pvt *pvt = mci->pvt_info;
1023         u32 injectmask;
1024
1025         if (!pvt->pci_ch[pvt->inject.channel][0])
1026                 return 0;
1027
1028         pci_read_config_dword(pvt->pci_ch[pvt->inject.channel][0],
1029                                MC_CHANNEL_ERROR_INJECT, &injectmask);
1030
1031         edac_dbg(0, "Inject error read: 0x%018x\n", injectmask);
1032
1033         if (injectmask & 0x0c)
1034                 pvt->inject.enable = 1;
1035
1036         return sprintf(data, "%d\n", pvt->inject.enable);
1037 }
1038
1039 #define DECLARE_COUNTER(param)                                  \
1040 static ssize_t i7core_show_counter_##param(                     \
1041         struct device *dev,                                     \
1042         struct device_attribute *mattr,                         \
1043         char *data)                                             \
1044 {                                                               \
1045         struct mem_ctl_info *mci = dev_get_drvdata(dev);        \
1046         struct i7core_pvt *pvt = mci->pvt_info;                 \
1047                                                                 \
1048         edac_dbg(1, "\n");                                      \
1049         if (!pvt->ce_count_available || (pvt->is_registered))   \
1050                 return sprintf(data, "data unavailable\n");     \
1051         return sprintf(data, "%lu\n",                           \
1052                         pvt->udimm_ce_count[param]);            \
1053 }
1054
1055 #define ATTR_COUNTER(param)                                     \
1056         static DEVICE_ATTR(udimm##param, S_IRUGO | S_IWUSR,     \
1057                     i7core_show_counter_##param,                \
1058                     NULL)
1059
1060 DECLARE_COUNTER(0);
1061 DECLARE_COUNTER(1);
1062 DECLARE_COUNTER(2);
1063
1064 ATTR_COUNTER(0);
1065 ATTR_COUNTER(1);
1066 ATTR_COUNTER(2);
1067
1068 /*
1069  * inject_addrmatch device sysfs struct
1070  */
1071
1072 static struct attribute *i7core_addrmatch_attrs[] = {
1073         &dev_attr_channel.attr,
1074         &dev_attr_dimm.attr,
1075         &dev_attr_rank.attr,
1076         &dev_attr_bank.attr,
1077         &dev_attr_page.attr,
1078         &dev_attr_col.attr,
1079         NULL
1080 };
1081
1082 static struct attribute_group addrmatch_grp = {
1083         .attrs  = i7core_addrmatch_attrs,
1084 };
1085
1086 static const struct attribute_group *addrmatch_groups[] = {
1087         &addrmatch_grp,
1088         NULL
1089 };
1090
1091 static void addrmatch_release(struct device *device)
1092 {
1093         edac_dbg(1, "Releasing device %s\n", dev_name(device));
1094         kfree(device);
1095 }
1096
1097 static struct device_type addrmatch_type = {
1098         .groups         = addrmatch_groups,
1099         .release        = addrmatch_release,
1100 };
1101
1102 /*
1103  * all_channel_counts sysfs struct
1104  */
1105
1106 static struct attribute *i7core_udimm_counters_attrs[] = {
1107         &dev_attr_udimm0.attr,
1108         &dev_attr_udimm1.attr,
1109         &dev_attr_udimm2.attr,
1110         NULL
1111 };
1112
1113 static struct attribute_group all_channel_counts_grp = {
1114         .attrs  = i7core_udimm_counters_attrs,
1115 };
1116
1117 static const struct attribute_group *all_channel_counts_groups[] = {
1118         &all_channel_counts_grp,
1119         NULL
1120 };
1121
1122 static void all_channel_counts_release(struct device *device)
1123 {
1124         edac_dbg(1, "Releasing device %s\n", dev_name(device));
1125         kfree(device);
1126 }
1127
1128 static struct device_type all_channel_counts_type = {
1129         .groups         = all_channel_counts_groups,
1130         .release        = all_channel_counts_release,
1131 };
1132
1133 /*
1134  * inject sysfs attributes
1135  */
1136
1137 static DEVICE_ATTR(inject_section, S_IRUGO | S_IWUSR,
1138                    i7core_inject_section_show, i7core_inject_section_store);
1139
1140 static DEVICE_ATTR(inject_type, S_IRUGO | S_IWUSR,
1141                    i7core_inject_type_show, i7core_inject_type_store);
1142
1143
1144 static DEVICE_ATTR(inject_eccmask, S_IRUGO | S_IWUSR,
1145                    i7core_inject_eccmask_show, i7core_inject_eccmask_store);
1146
1147 static DEVICE_ATTR(inject_enable, S_IRUGO | S_IWUSR,
1148                    i7core_inject_enable_show, i7core_inject_enable_store);
1149
1150 static struct attribute *i7core_dev_attrs[] = {
1151         &dev_attr_inject_section.attr,
1152         &dev_attr_inject_type.attr,
1153         &dev_attr_inject_eccmask.attr,
1154         &dev_attr_inject_enable.attr,
1155         NULL
1156 };
1157
1158 ATTRIBUTE_GROUPS(i7core_dev);
1159
1160 static int i7core_create_sysfs_devices(struct mem_ctl_info *mci)
1161 {
1162         struct i7core_pvt *pvt = mci->pvt_info;
1163         int rc;
1164
1165         pvt->addrmatch_dev = kzalloc(sizeof(*pvt->addrmatch_dev), GFP_KERNEL);
1166         if (!pvt->addrmatch_dev)
1167                 return -ENOMEM;
1168
1169         pvt->addrmatch_dev->type = &addrmatch_type;
1170         pvt->addrmatch_dev->bus = mci->dev.bus;
1171         device_initialize(pvt->addrmatch_dev);
1172         pvt->addrmatch_dev->parent = &mci->dev;
1173         dev_set_name(pvt->addrmatch_dev, "inject_addrmatch");
1174         dev_set_drvdata(pvt->addrmatch_dev, mci);
1175
1176         edac_dbg(1, "creating %s\n", dev_name(pvt->addrmatch_dev));
1177
1178         rc = device_add(pvt->addrmatch_dev);
1179         if (rc < 0)
1180                 return rc;
1181
1182         if (!pvt->is_registered) {
1183                 pvt->chancounts_dev = kzalloc(sizeof(*pvt->chancounts_dev),
1184                                               GFP_KERNEL);
1185                 if (!pvt->chancounts_dev) {
1186                         put_device(pvt->addrmatch_dev);
1187                         device_del(pvt->addrmatch_dev);
1188                         return -ENOMEM;
1189                 }
1190
1191                 pvt->chancounts_dev->type = &all_channel_counts_type;
1192                 pvt->chancounts_dev->bus = mci->dev.bus;
1193                 device_initialize(pvt->chancounts_dev);
1194                 pvt->chancounts_dev->parent = &mci->dev;
1195                 dev_set_name(pvt->chancounts_dev, "all_channel_counts");
1196                 dev_set_drvdata(pvt->chancounts_dev, mci);
1197
1198                 edac_dbg(1, "creating %s\n", dev_name(pvt->chancounts_dev));
1199
1200                 rc = device_add(pvt->chancounts_dev);
1201                 if (rc < 0)
1202                         return rc;
1203         }
1204         return 0;
1205 }
1206
1207 static void i7core_delete_sysfs_devices(struct mem_ctl_info *mci)
1208 {
1209         struct i7core_pvt *pvt = mci->pvt_info;
1210
1211         edac_dbg(1, "\n");
1212
1213         if (!pvt->is_registered) {
1214                 put_device(pvt->chancounts_dev);
1215                 device_del(pvt->chancounts_dev);
1216         }
1217         put_device(pvt->addrmatch_dev);
1218         device_del(pvt->addrmatch_dev);
1219 }
1220
1221 /****************************************************************************
1222         Device initialization routines: put/get, init/exit
1223  ****************************************************************************/
1224
1225 /*
1226  *      i7core_put_all_devices  'put' all the devices that we have
1227  *                              reserved via 'get'
1228  */
1229 static void i7core_put_devices(struct i7core_dev *i7core_dev)
1230 {
1231         int i;
1232
1233         edac_dbg(0, "\n");
1234         for (i = 0; i < i7core_dev->n_devs; i++) {
1235                 struct pci_dev *pdev = i7core_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 i7core_put_all_devices(void)
1246 {
1247         struct i7core_dev *i7core_dev, *tmp;
1248
1249         list_for_each_entry_safe(i7core_dev, tmp, &i7core_edac_list, list) {
1250                 i7core_put_devices(i7core_dev);
1251                 free_i7core_dev(i7core_dev);
1252         }
1253 }
1254
1255 static void __init i7core_xeon_pci_fixup(const struct pci_id_table *table)
1256 {
1257         struct pci_dev *pdev = NULL;
1258         int i;
1259
1260         /*
1261          * On Xeon 55xx, the Intel Quick Path Arch Generic Non-core pci buses
1262          * aren't announced by acpi. So, we need to use a legacy scan probing
1263          * to detect them
1264          */
1265         while (table && table->descr) {
1266                 pdev = pci_get_device(PCI_VENDOR_ID_INTEL, table->descr[0].dev_id, NULL);
1267                 if (unlikely(!pdev)) {
1268                         for (i = 0; i < MAX_SOCKET_BUSES; i++)
1269                                 pcibios_scan_specific_bus(255-i);
1270                 }
1271                 pci_dev_put(pdev);
1272                 table++;
1273         }
1274 }
1275
1276 static unsigned i7core_pci_lastbus(void)
1277 {
1278         int last_bus = 0, bus;
1279         struct pci_bus *b = NULL;
1280
1281         while ((b = pci_find_next_bus(b)) != NULL) {
1282                 bus = b->number;
1283                 edac_dbg(0, "Found bus %d\n", bus);
1284                 if (bus > last_bus)
1285                         last_bus = bus;
1286         }
1287
1288         edac_dbg(0, "Last bus %d\n", last_bus);
1289
1290         return last_bus;
1291 }
1292
1293 /*
1294  *      i7core_get_all_devices  Find and perform 'get' operation on the MCH's
1295  *                      device/functions we want to reference for this driver
1296  *
1297  *                      Need to 'get' device 16 func 1 and func 2
1298  */
1299 static int i7core_get_onedevice(struct pci_dev **prev,
1300                                 const struct pci_id_table *table,
1301                                 const unsigned devno,
1302                                 const unsigned last_bus)
1303 {
1304         struct i7core_dev *i7core_dev;
1305         const struct pci_id_descr *dev_descr = &table->descr[devno];
1306
1307         struct pci_dev *pdev = NULL;
1308         u8 bus = 0;
1309         u8 socket = 0;
1310
1311         pdev = pci_get_device(PCI_VENDOR_ID_INTEL,
1312                               dev_descr->dev_id, *prev);
1313
1314         /*
1315          * On Xeon 55xx, the Intel QuickPath Arch Generic Non-core regs
1316          * is at addr 8086:2c40, instead of 8086:2c41. So, we need
1317          * to probe for the alternate address in case of failure
1318          */
1319         if (dev_descr->dev_id == PCI_DEVICE_ID_INTEL_I7_NONCORE && !pdev) {
1320                 pci_dev_get(*prev);     /* pci_get_device will put it */
1321                 pdev = pci_get_device(PCI_VENDOR_ID_INTEL,
1322                                       PCI_DEVICE_ID_INTEL_I7_NONCORE_ALT, *prev);
1323         }
1324
1325         if (dev_descr->dev_id == PCI_DEVICE_ID_INTEL_LYNNFIELD_NONCORE &&
1326             !pdev) {
1327                 pci_dev_get(*prev);     /* pci_get_device will put it */
1328                 pdev = pci_get_device(PCI_VENDOR_ID_INTEL,
1329                                       PCI_DEVICE_ID_INTEL_LYNNFIELD_NONCORE_ALT,
1330                                       *prev);
1331         }
1332
1333         if (!pdev) {
1334                 if (*prev) {
1335                         *prev = pdev;
1336                         return 0;
1337                 }
1338
1339                 if (dev_descr->optional)
1340                         return 0;
1341
1342                 if (devno == 0)
1343                         return -ENODEV;
1344
1345                 i7core_printk(KERN_INFO,
1346                         "Device not found: dev %02x.%d PCI ID %04x:%04x\n",
1347                         dev_descr->dev, dev_descr->func,
1348                         PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
1349
1350                 /* End of list, leave */
1351                 return -ENODEV;
1352         }
1353         bus = pdev->bus->number;
1354
1355         socket = last_bus - bus;
1356
1357         i7core_dev = get_i7core_dev(socket);
1358         if (!i7core_dev) {
1359                 i7core_dev = alloc_i7core_dev(socket, table);
1360                 if (!i7core_dev) {
1361                         pci_dev_put(pdev);
1362                         return -ENOMEM;
1363                 }
1364         }
1365
1366         if (i7core_dev->pdev[devno]) {
1367                 i7core_printk(KERN_ERR,
1368                         "Duplicated device for "
1369                         "dev %02x:%02x.%d PCI ID %04x:%04x\n",
1370                         bus, dev_descr->dev, dev_descr->func,
1371                         PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
1372                 pci_dev_put(pdev);
1373                 return -ENODEV;
1374         }
1375
1376         i7core_dev->pdev[devno] = pdev;
1377
1378         /* Sanity check */
1379         if (unlikely(PCI_SLOT(pdev->devfn) != dev_descr->dev ||
1380                         PCI_FUNC(pdev->devfn) != dev_descr->func)) {
1381                 i7core_printk(KERN_ERR,
1382                         "Device PCI ID %04x:%04x "
1383                         "has dev %02x:%02x.%d instead of dev %02x:%02x.%d\n",
1384                         PCI_VENDOR_ID_INTEL, dev_descr->dev_id,
1385                         bus, PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn),
1386                         bus, dev_descr->dev, dev_descr->func);
1387                 return -ENODEV;
1388         }
1389
1390         /* Be sure that the device is enabled */
1391         if (unlikely(pci_enable_device(pdev) < 0)) {
1392                 i7core_printk(KERN_ERR,
1393                         "Couldn't enable "
1394                         "dev %02x:%02x.%d PCI ID %04x:%04x\n",
1395                         bus, dev_descr->dev, dev_descr->func,
1396                         PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
1397                 return -ENODEV;
1398         }
1399
1400         edac_dbg(0, "Detected socket %d dev %02x:%02x.%d PCI ID %04x:%04x\n",
1401                  socket, bus, dev_descr->dev,
1402                  dev_descr->func,
1403                  PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
1404
1405         /*
1406          * As stated on drivers/pci/search.c, the reference count for
1407          * @from is always decremented if it is not %NULL. So, as we need
1408          * to get all devices up to null, we need to do a get for the device
1409          */
1410         pci_dev_get(pdev);
1411
1412         *prev = pdev;
1413
1414         return 0;
1415 }
1416
1417 static int i7core_get_all_devices(void)
1418 {
1419         int i, rc, last_bus;
1420         struct pci_dev *pdev = NULL;
1421         const struct pci_id_table *table = pci_dev_table;
1422
1423         last_bus = i7core_pci_lastbus();
1424
1425         while (table && table->descr) {
1426                 for (i = 0; i < table->n_devs; i++) {
1427                         pdev = NULL;
1428                         do {
1429                                 rc = i7core_get_onedevice(&pdev, table, i,
1430                                                           last_bus);
1431                                 if (rc < 0) {
1432                                         if (i == 0) {
1433                                                 i = table->n_devs;
1434                                                 break;
1435                                         }
1436                                         i7core_put_all_devices();
1437                                         return -ENODEV;
1438                                 }
1439                         } while (pdev);
1440                 }
1441                 table++;
1442         }
1443
1444         return 0;
1445 }
1446
1447 static int mci_bind_devs(struct mem_ctl_info *mci,
1448                          struct i7core_dev *i7core_dev)
1449 {
1450         struct i7core_pvt *pvt = mci->pvt_info;
1451         struct pci_dev *pdev;
1452         int i, func, slot;
1453         char *family;
1454
1455         pvt->is_registered = false;
1456         pvt->enable_scrub  = false;
1457         for (i = 0; i < i7core_dev->n_devs; i++) {
1458                 pdev = i7core_dev->pdev[i];
1459                 if (!pdev)
1460                         continue;
1461
1462                 func = PCI_FUNC(pdev->devfn);
1463                 slot = PCI_SLOT(pdev->devfn);
1464                 if (slot == 3) {
1465                         if (unlikely(func > MAX_MCR_FUNC))
1466                                 goto error;
1467                         pvt->pci_mcr[func] = pdev;
1468                 } else if (likely(slot >= 4 && slot < 4 + NUM_CHANS)) {
1469                         if (unlikely(func > MAX_CHAN_FUNC))
1470                                 goto error;
1471                         pvt->pci_ch[slot - 4][func] = pdev;
1472                 } else if (!slot && !func) {
1473                         pvt->pci_noncore = pdev;
1474
1475                         /* Detect the processor family */
1476                         switch (pdev->device) {
1477                         case PCI_DEVICE_ID_INTEL_I7_NONCORE:
1478                                 family = "Xeon 35xx/ i7core";
1479                                 pvt->enable_scrub = false;
1480                                 break;
1481                         case PCI_DEVICE_ID_INTEL_LYNNFIELD_NONCORE_ALT:
1482                                 family = "i7-800/i5-700";
1483                                 pvt->enable_scrub = false;
1484                                 break;
1485                         case PCI_DEVICE_ID_INTEL_LYNNFIELD_NONCORE:
1486                                 family = "Xeon 34xx";
1487                                 pvt->enable_scrub = false;
1488                                 break;
1489                         case PCI_DEVICE_ID_INTEL_I7_NONCORE_ALT:
1490                                 family = "Xeon 55xx";
1491                                 pvt->enable_scrub = true;
1492                                 break;
1493                         case PCI_DEVICE_ID_INTEL_LYNNFIELD_NONCORE_REV2:
1494                                 family = "Xeon 56xx / i7-900";
1495                                 pvt->enable_scrub = true;
1496                                 break;
1497                         default:
1498                                 family = "unknown";
1499                                 pvt->enable_scrub = false;
1500                         }
1501                         edac_dbg(0, "Detected a processor type %s\n", family);
1502                 } else
1503                         goto error;
1504
1505                 edac_dbg(0, "Associated fn %d.%d, dev = %p, socket %d\n",
1506                          PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn),
1507                          pdev, i7core_dev->socket);
1508
1509                 if (PCI_SLOT(pdev->devfn) == 3 &&
1510                         PCI_FUNC(pdev->devfn) == 2)
1511                         pvt->is_registered = true;
1512         }
1513
1514         return 0;
1515
1516 error:
1517         i7core_printk(KERN_ERR, "Device %d, function %d "
1518                       "is out of the expected range\n",
1519                       slot, func);
1520         return -EINVAL;
1521 }
1522
1523 /****************************************************************************
1524                         Error check routines
1525  ****************************************************************************/
1526
1527 static void i7core_rdimm_update_ce_count(struct mem_ctl_info *mci,
1528                                          const int chan,
1529                                          const int new0,
1530                                          const int new1,
1531                                          const int new2)
1532 {
1533         struct i7core_pvt *pvt = mci->pvt_info;
1534         int add0 = 0, add1 = 0, add2 = 0;
1535         /* Updates CE counters if it is not the first time here */
1536         if (pvt->ce_count_available) {
1537                 /* Updates CE counters */
1538
1539                 add2 = new2 - pvt->rdimm_last_ce_count[chan][2];
1540                 add1 = new1 - pvt->rdimm_last_ce_count[chan][1];
1541                 add0 = new0 - pvt->rdimm_last_ce_count[chan][0];
1542
1543                 if (add2 < 0)
1544                         add2 += 0x7fff;
1545                 pvt->rdimm_ce_count[chan][2] += add2;
1546
1547                 if (add1 < 0)
1548                         add1 += 0x7fff;
1549                 pvt->rdimm_ce_count[chan][1] += add1;
1550
1551                 if (add0 < 0)
1552                         add0 += 0x7fff;
1553                 pvt->rdimm_ce_count[chan][0] += add0;
1554         } else
1555                 pvt->ce_count_available = 1;
1556
1557         /* Store the new values */
1558         pvt->rdimm_last_ce_count[chan][2] = new2;
1559         pvt->rdimm_last_ce_count[chan][1] = new1;
1560         pvt->rdimm_last_ce_count[chan][0] = new0;
1561
1562         /*updated the edac core */
1563         if (add0 != 0)
1564                 edac_mc_handle_error(HW_EVENT_ERR_CORRECTED, mci, add0,
1565                                      0, 0, 0,
1566                                      chan, 0, -1, "error", "");
1567         if (add1 != 0)
1568                 edac_mc_handle_error(HW_EVENT_ERR_CORRECTED, mci, add1,
1569                                      0, 0, 0,
1570                                      chan, 1, -1, "error", "");
1571         if (add2 != 0)
1572                 edac_mc_handle_error(HW_EVENT_ERR_CORRECTED, mci, add2,
1573                                      0, 0, 0,
1574                                      chan, 2, -1, "error", "");
1575 }
1576
1577 static void i7core_rdimm_check_mc_ecc_err(struct mem_ctl_info *mci)
1578 {
1579         struct i7core_pvt *pvt = mci->pvt_info;
1580         u32 rcv[3][2];
1581         int i, new0, new1, new2;
1582
1583         /*Read DEV 3: FUN 2:  MC_COR_ECC_CNT regs directly*/
1584         pci_read_config_dword(pvt->pci_mcr[2], MC_COR_ECC_CNT_0,
1585                                                                 &rcv[0][0]);
1586         pci_read_config_dword(pvt->pci_mcr[2], MC_COR_ECC_CNT_1,
1587                                                                 &rcv[0][1]);
1588         pci_read_config_dword(pvt->pci_mcr[2], MC_COR_ECC_CNT_2,
1589                                                                 &rcv[1][0]);
1590         pci_read_config_dword(pvt->pci_mcr[2], MC_COR_ECC_CNT_3,
1591                                                                 &rcv[1][1]);
1592         pci_read_config_dword(pvt->pci_mcr[2], MC_COR_ECC_CNT_4,
1593                                                                 &rcv[2][0]);
1594         pci_read_config_dword(pvt->pci_mcr[2], MC_COR_ECC_CNT_5,
1595                                                                 &rcv[2][1]);
1596         for (i = 0 ; i < 3; i++) {
1597                 edac_dbg(3, "MC_COR_ECC_CNT%d = 0x%x; MC_COR_ECC_CNT%d = 0x%x\n",
1598                          (i * 2), rcv[i][0], (i * 2) + 1, rcv[i][1]);
1599                 /*if the channel has 3 dimms*/
1600                 if (pvt->channel[i].dimms > 2) {
1601                         new0 = DIMM_BOT_COR_ERR(rcv[i][0]);
1602                         new1 = DIMM_TOP_COR_ERR(rcv[i][0]);
1603                         new2 = DIMM_BOT_COR_ERR(rcv[i][1]);
1604                 } else {
1605                         new0 = DIMM_TOP_COR_ERR(rcv[i][0]) +
1606                                         DIMM_BOT_COR_ERR(rcv[i][0]);
1607                         new1 = DIMM_TOP_COR_ERR(rcv[i][1]) +
1608                                         DIMM_BOT_COR_ERR(rcv[i][1]);
1609                         new2 = 0;
1610                 }
1611
1612                 i7core_rdimm_update_ce_count(mci, i, new0, new1, new2);
1613         }
1614 }
1615
1616 /* This function is based on the device 3 function 4 registers as described on:
1617  * Intel Xeon Processor 5500 Series Datasheet Volume 2
1618  *      http://www.intel.com/Assets/PDF/datasheet/321322.pdf
1619  * also available at:
1620  *      http://www.arrownac.com/manufacturers/intel/s/nehalem/5500-datasheet-v2.pdf
1621  */
1622 static void i7core_udimm_check_mc_ecc_err(struct mem_ctl_info *mci)
1623 {
1624         struct i7core_pvt *pvt = mci->pvt_info;
1625         u32 rcv1, rcv0;
1626         int new0, new1, new2;
1627
1628         if (!pvt->pci_mcr[4]) {
1629                 edac_dbg(0, "MCR registers not found\n");
1630                 return;
1631         }
1632
1633         /* Corrected test errors */
1634         pci_read_config_dword(pvt->pci_mcr[4], MC_TEST_ERR_RCV1, &rcv1);
1635         pci_read_config_dword(pvt->pci_mcr[4], MC_TEST_ERR_RCV0, &rcv0);
1636
1637         /* Store the new values */
1638         new2 = DIMM2_COR_ERR(rcv1);
1639         new1 = DIMM1_COR_ERR(rcv0);
1640         new0 = DIMM0_COR_ERR(rcv0);
1641
1642         /* Updates CE counters if it is not the first time here */
1643         if (pvt->ce_count_available) {
1644                 /* Updates CE counters */
1645                 int add0, add1, add2;
1646
1647                 add2 = new2 - pvt->udimm_last_ce_count[2];
1648                 add1 = new1 - pvt->udimm_last_ce_count[1];
1649                 add0 = new0 - pvt->udimm_last_ce_count[0];
1650
1651                 if (add2 < 0)
1652                         add2 += 0x7fff;
1653                 pvt->udimm_ce_count[2] += add2;
1654
1655                 if (add1 < 0)
1656                         add1 += 0x7fff;
1657                 pvt->udimm_ce_count[1] += add1;
1658
1659                 if (add0 < 0)
1660                         add0 += 0x7fff;
1661                 pvt->udimm_ce_count[0] += add0;
1662
1663                 if (add0 | add1 | add2)
1664                         i7core_printk(KERN_ERR, "New Corrected error(s): "
1665                                       "dimm0: +%d, dimm1: +%d, dimm2 +%d\n",
1666                                       add0, add1, add2);
1667         } else
1668                 pvt->ce_count_available = 1;
1669
1670         /* Store the new values */
1671         pvt->udimm_last_ce_count[2] = new2;
1672         pvt->udimm_last_ce_count[1] = new1;
1673         pvt->udimm_last_ce_count[0] = new0;
1674 }
1675
1676 /*
1677  * According with tables E-11 and E-12 of chapter E.3.3 of Intel 64 and IA-32
1678  * Architectures Software Developer’s Manual Volume 3B.
1679  * Nehalem are defined as family 0x06, model 0x1a
1680  *
1681  * The MCA registers used here are the following ones:
1682  *     struct mce field MCA Register
1683  *     m->status        MSR_IA32_MC8_STATUS
1684  *     m->addr          MSR_IA32_MC8_ADDR
1685  *     m->misc          MSR_IA32_MC8_MISC
1686  * In the case of Nehalem, the error information is masked at .status and .misc
1687  * fields
1688  */
1689 static void i7core_mce_output_error(struct mem_ctl_info *mci,
1690                                     const struct mce *m)
1691 {
1692         struct i7core_pvt *pvt = mci->pvt_info;
1693         char *optype, *err;
1694         enum hw_event_mc_err_type tp_event;
1695         unsigned long error = m->status & 0x1ff0000l;
1696         bool uncorrected_error = m->mcgstatus & 1ll << 61;
1697         bool ripv = m->mcgstatus & 1;
1698         u32 optypenum = (m->status >> 4) & 0x07;
1699         u32 core_err_cnt = (m->status >> 38) & 0x7fff;
1700         u32 dimm = (m->misc >> 16) & 0x3;
1701         u32 channel = (m->misc >> 18) & 0x3;
1702         u32 syndrome = m->misc >> 32;
1703         u32 errnum = find_first_bit(&error, 32);
1704
1705         if (uncorrected_error) {
1706                 if (ripv)
1707                         tp_event = HW_EVENT_ERR_FATAL;
1708                 else
1709                         tp_event = HW_EVENT_ERR_UNCORRECTED;
1710         } else {
1711                 tp_event = HW_EVENT_ERR_CORRECTED;
1712         }
1713
1714         switch (optypenum) {
1715         case 0:
1716                 optype = "generic undef request";
1717                 break;
1718         case 1:
1719                 optype = "read error";
1720                 break;
1721         case 2:
1722                 optype = "write error";
1723                 break;
1724         case 3:
1725                 optype = "addr/cmd error";
1726                 break;
1727         case 4:
1728                 optype = "scrubbing error";
1729                 break;
1730         default:
1731                 optype = "reserved";
1732                 break;
1733         }
1734
1735         switch (errnum) {
1736         case 16:
1737                 err = "read ECC error";
1738                 break;
1739         case 17:
1740                 err = "RAS ECC error";
1741                 break;
1742         case 18:
1743                 err = "write parity error";
1744                 break;
1745         case 19:
1746                 err = "redundacy loss";
1747                 break;
1748         case 20:
1749                 err = "reserved";
1750                 break;
1751         case 21:
1752                 err = "memory range error";
1753                 break;
1754         case 22:
1755                 err = "RTID out of range";
1756                 break;
1757         case 23:
1758                 err = "address parity error";
1759                 break;
1760         case 24:
1761                 err = "byte enable parity error";
1762                 break;
1763         default:
1764                 err = "unknown";
1765         }
1766
1767         /*
1768          * Call the helper to output message
1769          * FIXME: what to do if core_err_cnt > 1? Currently, it generates
1770          * only one event
1771          */
1772         if (uncorrected_error || !pvt->is_registered)
1773                 edac_mc_handle_error(tp_event, mci, core_err_cnt,
1774                                      m->addr >> PAGE_SHIFT,
1775                                      m->addr & ~PAGE_MASK,
1776                                      syndrome,
1777                                      channel, dimm, -1,
1778                                      err, optype);
1779 }
1780
1781 /*
1782  *      i7core_check_error      Retrieve and process errors reported by the
1783  *                              hardware. Called by the Core module.
1784  */
1785 static void i7core_check_error(struct mem_ctl_info *mci, struct mce *m)
1786 {
1787         struct i7core_pvt *pvt = mci->pvt_info;
1788
1789         i7core_mce_output_error(mci, m);
1790
1791         /*
1792          * Now, let's increment CE error counts
1793          */
1794         if (!pvt->is_registered)
1795                 i7core_udimm_check_mc_ecc_err(mci);
1796         else
1797                 i7core_rdimm_check_mc_ecc_err(mci);
1798 }
1799
1800 /*
1801  * Check that logging is enabled and that this is the right type
1802  * of error for us to handle.
1803  */
1804 static int i7core_mce_check_error(struct notifier_block *nb, unsigned long val,
1805                                   void *data)
1806 {
1807         struct mce *mce = (struct mce *)data;
1808         struct i7core_dev *i7_dev;
1809         struct mem_ctl_info *mci;
1810         struct i7core_pvt *pvt;
1811
1812         i7_dev = get_i7core_dev(mce->socketid);
1813         if (!i7_dev)
1814                 return NOTIFY_DONE;
1815
1816         mci = i7_dev->mci;
1817         pvt = mci->pvt_info;
1818
1819         /*
1820          * Just let mcelog handle it if the error is
1821          * outside the memory controller
1822          */
1823         if (((mce->status & 0xffff) >> 7) != 1)
1824                 return NOTIFY_DONE;
1825
1826         /* Bank 8 registers are the only ones that we know how to handle */
1827         if (mce->bank != 8)
1828                 return NOTIFY_DONE;
1829
1830         i7core_check_error(mci, mce);
1831
1832         /* Advise mcelog that the errors were handled */
1833         return NOTIFY_STOP;
1834 }
1835
1836 static struct notifier_block i7_mce_dec = {
1837         .notifier_call  = i7core_mce_check_error,
1838 };
1839
1840 struct memdev_dmi_entry {
1841         u8 type;
1842         u8 length;
1843         u16 handle;
1844         u16 phys_mem_array_handle;
1845         u16 mem_err_info_handle;
1846         u16 total_width;
1847         u16 data_width;
1848         u16 size;
1849         u8 form;
1850         u8 device_set;
1851         u8 device_locator;
1852         u8 bank_locator;
1853         u8 memory_type;
1854         u16 type_detail;
1855         u16 speed;
1856         u8 manufacturer;
1857         u8 serial_number;
1858         u8 asset_tag;
1859         u8 part_number;
1860         u8 attributes;
1861         u32 extended_size;
1862         u16 conf_mem_clk_speed;
1863 } __attribute__((__packed__));
1864
1865
1866 /*
1867  * Decode the DRAM Clock Frequency, be paranoid, make sure that all
1868  * memory devices show the same speed, and if they don't then consider
1869  * all speeds to be invalid.
1870  */
1871 static void decode_dclk(const struct dmi_header *dh, void *_dclk_freq)
1872 {
1873         int *dclk_freq = _dclk_freq;
1874         u16 dmi_mem_clk_speed;
1875
1876         if (*dclk_freq == -1)
1877                 return;
1878
1879         if (dh->type == DMI_ENTRY_MEM_DEVICE) {
1880                 struct memdev_dmi_entry *memdev_dmi_entry =
1881                         (struct memdev_dmi_entry *)dh;
1882                 unsigned long conf_mem_clk_speed_offset =
1883                         (unsigned long)&memdev_dmi_entry->conf_mem_clk_speed -
1884                         (unsigned long)&memdev_dmi_entry->type;
1885                 unsigned long speed_offset =
1886                         (unsigned long)&memdev_dmi_entry->speed -
1887                         (unsigned long)&memdev_dmi_entry->type;
1888
1889                 /* Check that a DIMM is present */
1890                 if (memdev_dmi_entry->size == 0)
1891                         return;
1892
1893                 /*
1894                  * Pick the configured speed if it's available, otherwise
1895                  * pick the DIMM speed, or we don't have a speed.
1896                  */
1897                 if (memdev_dmi_entry->length > conf_mem_clk_speed_offset) {
1898                         dmi_mem_clk_speed =
1899                                 memdev_dmi_entry->conf_mem_clk_speed;
1900                 } else if (memdev_dmi_entry->length > speed_offset) {
1901                         dmi_mem_clk_speed = memdev_dmi_entry->speed;
1902                 } else {
1903                         *dclk_freq = -1;
1904                         return;
1905                 }
1906
1907                 if (*dclk_freq == 0) {
1908                         /* First pass, speed was 0 */
1909                         if (dmi_mem_clk_speed > 0) {
1910                                 /* Set speed if a valid speed is read */
1911                                 *dclk_freq = dmi_mem_clk_speed;
1912                         } else {
1913                                 /* Otherwise we don't have a valid speed */
1914                                 *dclk_freq = -1;
1915                         }
1916                 } else if (*dclk_freq > 0 &&
1917                            *dclk_freq != dmi_mem_clk_speed) {
1918                         /*
1919                          * If we have a speed, check that all DIMMS are the same
1920                          * speed, otherwise set the speed as invalid.
1921                          */
1922                         *dclk_freq = -1;
1923                 }
1924         }
1925 }
1926
1927 /*
1928  * The default DCLK frequency is used as a fallback if we
1929  * fail to find anything reliable in the DMI. The value
1930  * is taken straight from the datasheet.
1931  */
1932 #define DEFAULT_DCLK_FREQ 800
1933
1934 static int get_dclk_freq(void)
1935 {
1936         int dclk_freq = 0;
1937
1938         dmi_walk(decode_dclk, (void *)&dclk_freq);
1939
1940         if (dclk_freq < 1)
1941                 return DEFAULT_DCLK_FREQ;
1942
1943         return dclk_freq;
1944 }
1945
1946 /*
1947  * set_sdram_scrub_rate         This routine sets byte/sec bandwidth scrub rate
1948  *                              to hardware according to SCRUBINTERVAL formula
1949  *                              found in datasheet.
1950  */
1951 static int set_sdram_scrub_rate(struct mem_ctl_info *mci, u32 new_bw)
1952 {
1953         struct i7core_pvt *pvt = mci->pvt_info;
1954         struct pci_dev *pdev;
1955         u32 dw_scrub;
1956         u32 dw_ssr;
1957
1958         /* Get data from the MC register, function 2 */
1959         pdev = pvt->pci_mcr[2];
1960         if (!pdev)
1961                 return -ENODEV;
1962
1963         pci_read_config_dword(pdev, MC_SCRUB_CONTROL, &dw_scrub);
1964
1965         if (new_bw == 0) {
1966                 /* Prepare to disable petrol scrub */
1967                 dw_scrub &= ~STARTSCRUB;
1968                 /* Stop the patrol scrub engine */
1969                 write_and_test(pdev, MC_SCRUB_CONTROL,
1970                                dw_scrub & ~SCRUBINTERVAL_MASK);
1971
1972                 /* Get current status of scrub rate and set bit to disable */
1973                 pci_read_config_dword(pdev, MC_SSRCONTROL, &dw_ssr);
1974                 dw_ssr &= ~SSR_MODE_MASK;
1975                 dw_ssr |= SSR_MODE_DISABLE;
1976         } else {
1977                 const int cache_line_size = 64;
1978                 const u32 freq_dclk_mhz = pvt->dclk_freq;
1979                 unsigned long long scrub_interval;
1980                 /*
1981                  * Translate the desired scrub rate to a register value and
1982                  * program the corresponding register value.
1983                  */
1984                 scrub_interval = (unsigned long long)freq_dclk_mhz *
1985                         cache_line_size * 1000000;
1986                 do_div(scrub_interval, new_bw);
1987
1988                 if (!scrub_interval || scrub_interval > SCRUBINTERVAL_MASK)
1989                         return -EINVAL;
1990
1991                 dw_scrub = SCRUBINTERVAL_MASK & scrub_interval;
1992
1993                 /* Start the patrol scrub engine */
1994                 pci_write_config_dword(pdev, MC_SCRUB_CONTROL,
1995                                        STARTSCRUB | dw_scrub);
1996
1997                 /* Get current status of scrub rate and set bit to enable */
1998                 pci_read_config_dword(pdev, MC_SSRCONTROL, &dw_ssr);
1999                 dw_ssr &= ~SSR_MODE_MASK;
2000                 dw_ssr |= SSR_MODE_ENABLE;
2001         }
2002         /* Disable or enable scrubbing */
2003         pci_write_config_dword(pdev, MC_SSRCONTROL, dw_ssr);
2004
2005         return new_bw;
2006 }
2007
2008 /*
2009  * get_sdram_scrub_rate         This routine convert current scrub rate value
2010  *                              into byte/sec bandwidth according to
2011  *                              SCRUBINTERVAL formula found in datasheet.
2012  */
2013 static int get_sdram_scrub_rate(struct mem_ctl_info *mci)
2014 {
2015         struct i7core_pvt *pvt = mci->pvt_info;
2016         struct pci_dev *pdev;
2017         const u32 cache_line_size = 64;
2018         const u32 freq_dclk_mhz = pvt->dclk_freq;
2019         unsigned long long scrub_rate;
2020         u32 scrubval;
2021
2022         /* Get data from the MC register, function 2 */
2023         pdev = pvt->pci_mcr[2];
2024         if (!pdev)
2025                 return -ENODEV;
2026
2027         /* Get current scrub control data */
2028         pci_read_config_dword(pdev, MC_SCRUB_CONTROL, &scrubval);
2029
2030         /* Mask highest 8-bits to 0 */
2031         scrubval &=  SCRUBINTERVAL_MASK;
2032         if (!scrubval)
2033                 return 0;
2034
2035         /* Calculate scrub rate value into byte/sec bandwidth */
2036         scrub_rate =  (unsigned long long)freq_dclk_mhz *
2037                 1000000 * cache_line_size;
2038         do_div(scrub_rate, scrubval);
2039         return (int)scrub_rate;
2040 }
2041
2042 static void enable_sdram_scrub_setting(struct mem_ctl_info *mci)
2043 {
2044         struct i7core_pvt *pvt = mci->pvt_info;
2045         u32 pci_lock;
2046
2047         /* Unlock writes to pci registers */
2048         pci_read_config_dword(pvt->pci_noncore, MC_CFG_CONTROL, &pci_lock);
2049         pci_lock &= ~0x3;
2050         pci_write_config_dword(pvt->pci_noncore, MC_CFG_CONTROL,
2051                                pci_lock | MC_CFG_UNLOCK);
2052
2053         mci->set_sdram_scrub_rate = set_sdram_scrub_rate;
2054         mci->get_sdram_scrub_rate = get_sdram_scrub_rate;
2055 }
2056
2057 static void disable_sdram_scrub_setting(struct mem_ctl_info *mci)
2058 {
2059         struct i7core_pvt *pvt = mci->pvt_info;
2060         u32 pci_lock;
2061
2062         /* Lock writes to pci registers */
2063         pci_read_config_dword(pvt->pci_noncore, MC_CFG_CONTROL, &pci_lock);
2064         pci_lock &= ~0x3;
2065         pci_write_config_dword(pvt->pci_noncore, MC_CFG_CONTROL,
2066                                pci_lock | MC_CFG_LOCK);
2067 }
2068
2069 static void i7core_pci_ctl_create(struct i7core_pvt *pvt)
2070 {
2071         pvt->i7core_pci = edac_pci_create_generic_ctl(
2072                                                 &pvt->i7core_dev->pdev[0]->dev,
2073                                                 EDAC_MOD_STR);
2074         if (unlikely(!pvt->i7core_pci))
2075                 i7core_printk(KERN_WARNING,
2076                               "Unable to setup PCI error report via EDAC\n");
2077 }
2078
2079 static void i7core_pci_ctl_release(struct i7core_pvt *pvt)
2080 {
2081         if (likely(pvt->i7core_pci))
2082                 edac_pci_release_generic_ctl(pvt->i7core_pci);
2083         else
2084                 i7core_printk(KERN_ERR,
2085                                 "Couldn't find mem_ctl_info for socket %d\n",
2086                                 pvt->i7core_dev->socket);
2087         pvt->i7core_pci = NULL;
2088 }
2089
2090 static void i7core_unregister_mci(struct i7core_dev *i7core_dev)
2091 {
2092         struct mem_ctl_info *mci = i7core_dev->mci;
2093         struct i7core_pvt *pvt;
2094
2095         if (unlikely(!mci || !mci->pvt_info)) {
2096                 edac_dbg(0, "MC: dev = %p\n", &i7core_dev->pdev[0]->dev);
2097
2098                 i7core_printk(KERN_ERR, "Couldn't find mci handler\n");
2099                 return;
2100         }
2101
2102         pvt = mci->pvt_info;
2103
2104         edac_dbg(0, "MC: mci = %p, dev = %p\n", mci, &i7core_dev->pdev[0]->dev);
2105
2106         /* Disable scrubrate setting */
2107         if (pvt->enable_scrub)
2108                 disable_sdram_scrub_setting(mci);
2109
2110         /* Disable EDAC polling */
2111         i7core_pci_ctl_release(pvt);
2112
2113         /* Remove MC sysfs nodes */
2114         i7core_delete_sysfs_devices(mci);
2115         edac_mc_del_mc(mci->pdev);
2116
2117         edac_dbg(1, "%s: free mci struct\n", mci->ctl_name);
2118         kfree(mci->ctl_name);
2119         edac_mc_free(mci);
2120         i7core_dev->mci = NULL;
2121 }
2122
2123 static int i7core_register_mci(struct i7core_dev *i7core_dev)
2124 {
2125         struct mem_ctl_info *mci;
2126         struct i7core_pvt *pvt;
2127         int rc;
2128         struct edac_mc_layer layers[2];
2129
2130         /* allocate a new MC control structure */
2131
2132         layers[0].type = EDAC_MC_LAYER_CHANNEL;
2133         layers[0].size = NUM_CHANS;
2134         layers[0].is_virt_csrow = false;
2135         layers[1].type = EDAC_MC_LAYER_SLOT;
2136         layers[1].size = MAX_DIMMS;
2137         layers[1].is_virt_csrow = true;
2138         mci = edac_mc_alloc(i7core_dev->socket, ARRAY_SIZE(layers), layers,
2139                             sizeof(*pvt));
2140         if (unlikely(!mci))
2141                 return -ENOMEM;
2142
2143         edac_dbg(0, "MC: mci = %p, dev = %p\n", mci, &i7core_dev->pdev[0]->dev);
2144
2145         pvt = mci->pvt_info;
2146         memset(pvt, 0, sizeof(*pvt));
2147
2148         /* Associates i7core_dev and mci for future usage */
2149         pvt->i7core_dev = i7core_dev;
2150         i7core_dev->mci = mci;
2151
2152         /*
2153          * FIXME: how to handle RDDR3 at MCI level? It is possible to have
2154          * Mixed RDDR3/UDDR3 with Nehalem, provided that they are on different
2155          * memory channels
2156          */
2157         mci->mtype_cap = MEM_FLAG_DDR3;
2158         mci->edac_ctl_cap = EDAC_FLAG_NONE;
2159         mci->edac_cap = EDAC_FLAG_NONE;
2160         mci->mod_name = "i7core_edac.c";
2161         mci->mod_ver = I7CORE_REVISION;
2162         mci->ctl_name = kasprintf(GFP_KERNEL, "i7 core #%d",
2163                                   i7core_dev->socket);
2164         mci->dev_name = pci_name(i7core_dev->pdev[0]);
2165         mci->ctl_page_to_phys = NULL;
2166
2167         /* Store pci devices at mci for faster access */
2168         rc = mci_bind_devs(mci, i7core_dev);
2169         if (unlikely(rc < 0))
2170                 goto fail0;
2171
2172
2173         /* Get dimm basic config */
2174         get_dimm_config(mci);
2175         /* record ptr to the generic device */
2176         mci->pdev = &i7core_dev->pdev[0]->dev;
2177
2178         /* Enable scrubrate setting */
2179         if (pvt->enable_scrub)
2180                 enable_sdram_scrub_setting(mci);
2181
2182         /* add this new MC control structure to EDAC's list of MCs */
2183         if (unlikely(edac_mc_add_mc_with_groups(mci, i7core_dev_groups))) {
2184                 edac_dbg(0, "MC: failed edac_mc_add_mc()\n");
2185                 /* FIXME: perhaps some code should go here that disables error
2186                  * reporting if we just enabled it
2187                  */
2188
2189                 rc = -EINVAL;
2190                 goto fail0;
2191         }
2192         if (i7core_create_sysfs_devices(mci)) {
2193                 edac_dbg(0, "MC: failed to create sysfs nodes\n");
2194                 edac_mc_del_mc(mci->pdev);
2195                 rc = -EINVAL;
2196                 goto fail0;
2197         }
2198
2199         /* Default error mask is any memory */
2200         pvt->inject.channel = 0;
2201         pvt->inject.dimm = -1;
2202         pvt->inject.rank = -1;
2203         pvt->inject.bank = -1;
2204         pvt->inject.page = -1;
2205         pvt->inject.col = -1;
2206
2207         /* allocating generic PCI control info */
2208         i7core_pci_ctl_create(pvt);
2209
2210         /* DCLK for scrub rate setting */
2211         pvt->dclk_freq = get_dclk_freq();
2212
2213         return 0;
2214
2215 fail0:
2216         kfree(mci->ctl_name);
2217         edac_mc_free(mci);
2218         i7core_dev->mci = NULL;
2219         return rc;
2220 }
2221
2222 /*
2223  *      i7core_probe    Probe for ONE instance of device to see if it is
2224  *                      present.
2225  *      return:
2226  *              0 for FOUND a device
2227  *              < 0 for error code
2228  */
2229
2230 static int i7core_probe(struct pci_dev *pdev, const struct pci_device_id *id)
2231 {
2232         int rc, count = 0;
2233         struct i7core_dev *i7core_dev;
2234
2235         /* get the pci devices we want to reserve for our use */
2236         mutex_lock(&i7core_edac_lock);
2237
2238         /*
2239          * All memory controllers are allocated at the first pass.
2240          */
2241         if (unlikely(probed >= 1)) {
2242                 mutex_unlock(&i7core_edac_lock);
2243                 return -ENODEV;
2244         }
2245         probed++;
2246
2247         rc = i7core_get_all_devices();
2248         if (unlikely(rc < 0))
2249                 goto fail0;
2250
2251         list_for_each_entry(i7core_dev, &i7core_edac_list, list) {
2252                 count++;
2253                 rc = i7core_register_mci(i7core_dev);
2254                 if (unlikely(rc < 0))
2255                         goto fail1;
2256         }
2257
2258         /*
2259          * Nehalem-EX uses a different memory controller. However, as the
2260          * memory controller is not visible on some Nehalem/Nehalem-EP, we
2261          * need to indirectly probe via a X58 PCI device. The same devices
2262          * are found on (some) Nehalem-EX. So, on those machines, the
2263          * probe routine needs to return -ENODEV, as the actual Memory
2264          * Controller registers won't be detected.
2265          */
2266         if (!count) {
2267                 rc = -ENODEV;
2268                 goto fail1;
2269         }
2270
2271         i7core_printk(KERN_INFO,
2272                       "Driver loaded, %d memory controller(s) found.\n",
2273                       count);
2274
2275         mutex_unlock(&i7core_edac_lock);
2276         return 0;
2277
2278 fail1:
2279         list_for_each_entry(i7core_dev, &i7core_edac_list, list)
2280                 i7core_unregister_mci(i7core_dev);
2281
2282         i7core_put_all_devices();
2283 fail0:
2284         mutex_unlock(&i7core_edac_lock);
2285         return rc;
2286 }
2287
2288 /*
2289  *      i7core_remove   destructor for one instance of device
2290  *
2291  */
2292 static void i7core_remove(struct pci_dev *pdev)
2293 {
2294         struct i7core_dev *i7core_dev;
2295
2296         edac_dbg(0, "\n");
2297
2298         /*
2299          * we have a trouble here: pdev value for removal will be wrong, since
2300          * it will point to the X58 register used to detect that the machine
2301          * is a Nehalem or upper design. However, due to the way several PCI
2302          * devices are grouped together to provide MC functionality, we need
2303          * to use a different method for releasing the devices
2304          */
2305
2306         mutex_lock(&i7core_edac_lock);
2307
2308         if (unlikely(!probed)) {
2309                 mutex_unlock(&i7core_edac_lock);
2310                 return;
2311         }
2312
2313         list_for_each_entry(i7core_dev, &i7core_edac_list, list)
2314                 i7core_unregister_mci(i7core_dev);
2315
2316         /* Release PCI resources */
2317         i7core_put_all_devices();
2318
2319         probed--;
2320
2321         mutex_unlock(&i7core_edac_lock);
2322 }
2323
2324 MODULE_DEVICE_TABLE(pci, i7core_pci_tbl);
2325
2326 /*
2327  *      i7core_driver   pci_driver structure for this module
2328  *
2329  */
2330 static struct pci_driver i7core_driver = {
2331         .name     = "i7core_edac",
2332         .probe    = i7core_probe,
2333         .remove   = i7core_remove,
2334         .id_table = i7core_pci_tbl,
2335 };
2336
2337 /*
2338  *      i7core_init             Module entry function
2339  *                      Try to initialize this module for its devices
2340  */
2341 static int __init i7core_init(void)
2342 {
2343         int pci_rc;
2344
2345         edac_dbg(2, "\n");
2346
2347         /* Ensure that the OPSTATE is set correctly for POLL or NMI */
2348         opstate_init();
2349
2350         if (use_pci_fixup)
2351                 i7core_xeon_pci_fixup(pci_dev_table);
2352
2353         pci_rc = pci_register_driver(&i7core_driver);
2354
2355         if (pci_rc >= 0) {
2356                 mce_register_decode_chain(&i7_mce_dec);
2357                 return 0;
2358         }
2359
2360         i7core_printk(KERN_ERR, "Failed to register device with error %d.\n",
2361                       pci_rc);
2362
2363         return pci_rc;
2364 }
2365
2366 /*
2367  *      i7core_exit()   Module exit function
2368  *                      Unregister the driver
2369  */
2370 static void __exit i7core_exit(void)
2371 {
2372         edac_dbg(2, "\n");
2373         pci_unregister_driver(&i7core_driver);
2374         mce_unregister_decode_chain(&i7_mce_dec);
2375 }
2376
2377 module_init(i7core_init);
2378 module_exit(i7core_exit);
2379
2380 MODULE_LICENSE("GPL");
2381 MODULE_AUTHOR("Mauro Carvalho Chehab");
2382 MODULE_AUTHOR("Red Hat Inc. (http://www.redhat.com)");
2383 MODULE_DESCRIPTION("MC Driver for Intel i7 Core memory controllers - "
2384                    I7CORE_REVISION);
2385
2386 module_param(edac_op_state, int, 0444);
2387 MODULE_PARM_DESC(edac_op_state, "EDAC Error Reporting state: 0=Poll,1=NMI");