Merge tag 'nfsd-4.7' of git://linux-nfs.org/~bfields/linux
[linux-2.6-block.git] / drivers / edac / i7core_edac.c
CommitLineData
52707f91
MCC
1/* Intel i7 core/Nehalem Memory Controller kernel module
2 *
e7bf068a 3 * This driver supports the memory controllers found on the Intel
52707f91
MCC
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.
a0c36a1f
MCC
7 *
8 * This file may be distributed under the terms of the
9 * GNU General Public License version 2 only.
10 *
52707f91 11 * Copyright (c) 2009-2010 by:
37e59f87 12 * Mauro Carvalho Chehab
a0c36a1f
MCC
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
a0c36a1f
MCC
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>
3b918c12 33#include <linux/delay.h>
535e9c78 34#include <linux/dmi.h>
a0c36a1f
MCC
35#include <linux/edac.h>
36#include <linux/mmzone.h>
f4742949 37#include <linux/smp.h>
4140c542 38#include <asm/mce.h>
14d2c083 39#include <asm/processor.h>
4fad8098 40#include <asm/div64.h>
a0c36a1f
MCC
41
42#include "edac_core.h"
43
18c29002
MCC
44/* Static vars */
45static LIST_HEAD(i7core_edac_list);
46static DEFINE_MUTEX(i7core_edac_lock);
47static int probed;
48
54a08ab1
MCC
49static int use_pci_fixup;
50module_param(use_pci_fixup, int, 0444);
51MODULE_PARM_DESC(use_pci_fixup, "Enable PCI fixup to seek for hidden devices");
f4742949
MCC
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
a0c36a1f
MCC
61/*
62 * Alter this version for the module when modifications are made
63 */
152ba394 64#define I7CORE_REVISION " Ver: 1.0.0"
a0c36a1f
MCC
65#define EDAC_MOD_STR "i7core_edac"
66
a0c36a1f
MCC
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
e9bd2e73
MCC
80 /* OFFSETS for Device 0 Function 0 */
81
82#define MC_CFG_CONTROL 0x90
e8b6a127
SG
83 #define MC_CFG_UNLOCK 0x02
84 #define MC_CFG_LOCK 0x00
e9bd2e73 85
a0c36a1f
MCC
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
442305b1 92/*
15ed103a 93 * OFFSETS for Device 3 Function 4, as indicated on Xeon 5500 datasheet:
442305b1
MCC
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
15ed103a 104/* OFFSETS for Device 3 Function 2, as indicated on Xeon 5500 datasheet */
e8b6a127
SG
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)
535e9c78 112 #define SCRUBINTERVAL_MASK 0xffffff
e8b6a127 113
b4e8f0b6
MCC
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
a0c36a1f
MCC
125 /* OFFSETS for Devices 4,5 and 6 Function 0 */
126
0b2b7b7e
MCC
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
f122a892
MCC
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
0b2b7b7e
MCC
137#define MC_CHANNEL_RANK_PRESENT 0x7c
138 #define RANK_PRESENT_MASK 0xffff
139
a0c36a1f 140#define MC_CHANNEL_ADDR_MATCH 0xf0
194a40fe
MCC
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
a0c36a1f 151
0b2b7b7e 152 /* OFFSETS for Devices 4,5 and 6 Function 1 */
b990538a 153
0b2b7b7e
MCC
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)
854d3349
MCC
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)
41fcb7fe 165 #define MC_DOD_NUMROW_MASK ((1 << 4) | (1 << 3) | (1 << 2))
5566cb7c 166 #define MC_DOD_NUMROW(x) (((x) & MC_DOD_NUMROW_MASK) >> 2)
854d3349
MCC
167 #define MC_DOD_NUMCOL_MASK 3
168 #define MC_DOD_NUMCOL(x) ((x) & MC_DOD_NUMCOL_MASK)
0b2b7b7e 169
f122a892
MCC
170#define MC_RANK_PRESENT 0x7c
171
0b2b7b7e
MCC
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
a0c36a1f
MCC
195/*
196 * i7core structs
197 */
198
199#define NUM_CHANS 3
442305b1
MCC
200#define MAX_DIMMS 3 /* Max DIMMS per channel */
201#define MAX_MCR_FUNC 4
202#define MAX_CHAN_FUNC 3
a0c36a1f
MCC
203
204struct i7core_info {
205 u32 mc_control;
206 u32 mc_status;
207 u32 max_dod;
f122a892 208 u32 ch_map;
a0c36a1f
MCC
209};
210
194a40fe
MCC
211
212struct 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
0b2b7b7e 223struct i7core_channel {
0bf09e82
MCC
224 bool is_3dimms_present;
225 bool is_single_4rank;
226 bool has_4rank;
442305b1 227 u32 dimms;
0b2b7b7e
MCC
228};
229
8f331907 230struct pci_id_descr {
66607706
MCC
231 int dev;
232 int func;
233 int dev_id;
de06eeef 234 int optional;
8f331907
MCC
235};
236
bd9e19ca 237struct pci_id_table {
1288c18f
MCC
238 const struct pci_id_descr *descr;
239 int n_devs;
bd9e19ca
VM
240};
241
f4742949
MCC
242struct i7core_dev {
243 struct list_head list;
244 u8 socket;
245 struct pci_dev **pdev;
de06eeef 246 int n_devs;
f4742949
MCC
247 struct mem_ctl_info *mci;
248};
249
a0c36a1f 250struct i7core_pvt {
356f0a30 251 struct device *addrmatch_dev, *chancounts_dev;
5c4cdb5a 252
f4742949
MCC
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;
67166af4 258
a0c36a1f 259 struct i7core_info info;
194a40fe 260 struct i7core_inject inject;
f4742949 261 struct i7core_channel channel[NUM_CHANS];
67166af4 262
f4742949 263 int ce_count_available;
b4e8f0b6
MCC
264
265 /* ECC corrected errors counts per udimm */
f4742949
MCC
266 unsigned long udimm_ce_count[MAX_DIMMS];
267 int udimm_last_ce_count[MAX_DIMMS];
b4e8f0b6 268 /* ECC corrected errors counts per rdimm */
f4742949
MCC
269 unsigned long rdimm_ce_count[NUM_CHANS][MAX_DIMMS];
270 int rdimm_last_ce_count[NUM_CHANS][MAX_DIMMS];
442305b1 271
27100db0 272 bool is_registered, enable_scrub;
14d2c083 273
535e9c78
NC
274 /* DCLK Frequency used for computing scrub rate */
275 int dclk_freq;
276
939747bd
MCC
277 /* Struct to control EDAC polling */
278 struct edac_pci_ctl_info *i7core_pci;
a0c36a1f
MCC
279};
280
8f331907
MCC
281#define PCI_DESCR(device, function, device_id) \
282 .dev = (device), \
283 .func = (function), \
284 .dev_id = (device_id)
285
1288c18f 286static const struct pci_id_descr pci_dev_descr_i7core_nehalem[] = {
8f331907
MCC
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) },
224e871f 290 /* Exists only for RDIMM */
de06eeef 291 { PCI_DESCR(3, 2, PCI_DEVICE_ID_INTEL_I7_MC_RAS), .optional = 1 },
8f331907
MCC
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) },
224e871f
MCC
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
a0c36a1f 321};
8f331907 322
1288c18f 323static const struct pci_id_descr pci_dev_descr_lynnfield[] = {
52a2e4fc
MCC
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
508fa179
MCC
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) },
224e871f
MCC
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) },
52a2e4fc
MCC
343};
344
1288c18f 345static const struct pci_id_descr pci_dev_descr_i7core_westmere[] = {
bd9e19ca
VM
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) },
224e871f
MCC
370
371 /* Generic Non-core registers */
372 { PCI_DESCR(0, 0, PCI_DEVICE_ID_INTEL_LYNNFIELD_NONCORE_REV2) },
373
bd9e19ca
VM
374};
375
1288c18f
MCC
376#define PCI_ID_TABLE_ENTRY(A) { .descr=A, .n_devs = ARRAY_SIZE(A) }
377static const struct pci_id_table pci_dev_table[] = {
bd9e19ca
VM
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),
3c52cc57 381 {0,} /* 0 terminated list. */
bd9e19ca
VM
382};
383
8f331907
MCC
384/*
385 * pci_device_id table for which devices we are looking for
8f331907 386 */
ba935f40 387static const struct pci_device_id i7core_pci_tbl[] = {
d1fd4fb6 388 {PCI_DEVICE(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_X58_HUB_MGMT)},
f05da2f7 389 {PCI_DEVICE(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_LYNNFIELD_QPI_LINK0)},
8f331907
MCC
390 {0,} /* 0 terminated list. */
391};
392
a0c36a1f 393/****************************************************************************
15ed103a 394 Ancillary status routines
a0c36a1f
MCC
395 ****************************************************************************/
396
397 /* MC_CONTROL bits */
ef708b53
MCC
398#define CH_ACTIVE(pvt, ch) ((pvt)->info.mc_control & (1 << (8 + ch)))
399#define ECCx8(pvt) ((pvt)->info.mc_control & (1 << 1))
a0c36a1f
MCC
400
401 /* MC_STATUS bits */
61053fde 402#define ECC_ENABLED(pvt) ((pvt)->info.mc_status & (1 << 4))
ef708b53 403#define CH_DISABLED(pvt, ch) ((pvt)->info.mc_status & (1 << ch))
a0c36a1f
MCC
404
405 /* MC_MAX_DOD read functions */
854d3349 406static inline int numdimms(u32 dimms)
a0c36a1f 407{
854d3349 408 return (dimms & 0x3) + 1;
a0c36a1f
MCC
409}
410
854d3349 411static inline int numrank(u32 rank)
a0c36a1f 412{
c31d34fe 413 static const int ranks[] = { 1, 2, 4, -EINVAL };
a0c36a1f 414
854d3349 415 return ranks[rank & 0x3];
a0c36a1f
MCC
416}
417
854d3349 418static inline int numbank(u32 bank)
a0c36a1f 419{
c31d34fe 420 static const int banks[] = { 4, 8, 16, -EINVAL };
a0c36a1f 421
854d3349 422 return banks[bank & 0x3];
a0c36a1f
MCC
423}
424
854d3349 425static inline int numrow(u32 row)
a0c36a1f 426{
c31d34fe 427 static const int rows[] = {
a0c36a1f
MCC
428 1 << 12, 1 << 13, 1 << 14, 1 << 15,
429 1 << 16, -EINVAL, -EINVAL, -EINVAL,
430 };
431
854d3349 432 return rows[row & 0x7];
a0c36a1f
MCC
433}
434
854d3349 435static inline int numcol(u32 col)
a0c36a1f 436{
c31d34fe 437 static const int cols[] = {
a0c36a1f
MCC
438 1 << 10, 1 << 11, 1 << 12, -EINVAL,
439 };
854d3349 440 return cols[col & 0x3];
a0c36a1f
MCC
441}
442
f4742949 443static struct i7core_dev *get_i7core_dev(u8 socket)
66607706
MCC
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
848b2f7e
HS
455static 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
2aa9be44
HS
478static 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
a0c36a1f
MCC
485/****************************************************************************
486 Memory check routines
487 ****************************************************************************/
eb94fc40 488
084a4fcc 489static int get_dimm_config(struct mem_ctl_info *mci)
a0c36a1f
MCC
490{
491 struct i7core_pvt *pvt = mci->pvt_info;
854d3349 492 struct pci_dev *pdev;
ba6c5c62 493 int i, j;
1c6fed80 494 enum edac_type mode;
854d3349 495 enum mem_type mtype;
084a4fcc 496 struct dimm_info *dimm;
a0c36a1f 497
854d3349 498 /* Get data from the MC register, function 0 */
f4742949 499 pdev = pvt->pci_mcr[0];
7dd6953c 500 if (!pdev)
8f331907
MCC
501 return -ENODEV;
502
f122a892 503 /* Device 3 function 0 reads */
7dd6953c
MCC
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);
f122a892 508
956b9ba1
JP
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);
a0c36a1f 512
1c6fed80 513 if (ECC_ENABLED(pvt)) {
956b9ba1 514 edac_dbg(0, "ECC enabled with x%d SDCC\n", ECCx8(pvt) ? 8 : 4);
1c6fed80
MCC
515 if (ECCx8(pvt))
516 mode = EDAC_S8ECD8ED;
517 else
518 mode = EDAC_S4ECD4ED;
519 } else {
956b9ba1 520 edac_dbg(0, "ECC disabled\n");
1c6fed80
MCC
521 mode = EDAC_NONE;
522 }
a0c36a1f
MCC
523
524 /* FIXME: need to handle the error codes */
956b9ba1
JP
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));
a0c36a1f 531
0b2b7b7e 532 for (i = 0; i < NUM_CHANS; i++) {
854d3349 533 u32 data, dimm_dod[3], value[8];
0b2b7b7e 534
52a2e4fc
MCC
535 if (!pvt->pci_ch[i][0])
536 continue;
537
0b2b7b7e 538 if (!CH_ACTIVE(pvt, i)) {
956b9ba1 539 edac_dbg(0, "Channel %i is not active\n", i);
0b2b7b7e
MCC
540 continue;
541 }
542 if (CH_DISABLED(pvt, i)) {
956b9ba1 543 edac_dbg(0, "Channel %i is disabled\n", i);
0b2b7b7e
MCC
544 continue;
545 }
546
f122a892 547 /* Devices 4-6 function 0 */
f4742949 548 pci_read_config_dword(pvt->pci_ch[i][0],
0b2b7b7e
MCC
549 MC_CHANNEL_DIMM_INIT_PARAMS, &data);
550
0bf09e82
MCC
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;
0b2b7b7e 560
854d3349
MCC
561 if (data & REGISTERED_DIMM)
562 mtype = MEM_RDDR3;
14d2c083 563 else
854d3349 564 mtype = MEM_DDR3;
854d3349
MCC
565
566 /* Devices 4-6 function 1 */
f4742949 567 pci_read_config_dword(pvt->pci_ch[i][1],
854d3349 568 MC_DOD_CH_DIMM0, &dimm_dod[0]);
f4742949 569 pci_read_config_dword(pvt->pci_ch[i][1],
854d3349 570 MC_DOD_CH_DIMM1, &dimm_dod[1]);
f4742949 571 pci_read_config_dword(pvt->pci_ch[i][1],
854d3349 572 MC_DOD_CH_DIMM2, &dimm_dod[2]);
0b2b7b7e 573
956b9ba1
JP
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');
854d3349
MCC
582
583 for (j = 0; j < 3; j++) {
584 u32 banks, ranks, rows, cols;
5566cb7c 585 u32 size, npages;
854d3349
MCC
586
587 if (!DIMM_PRESENT(dimm_dod[j]))
588 continue;
589
0975c16f
MCC
590 dimm = EDAC_DIMM_PTR(mci->layers, mci->dimms, mci->n_layers,
591 i, j, 0);
854d3349
MCC
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
5566cb7c
MCC
597 /* DDR3 has 8 I/O banks */
598 size = (rows * cols * banks * ranks) >> (20 - 3);
599
956b9ba1
JP
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);
854d3349 604
e9144601 605 npages = MiB_TO_PAGES(size);
5566cb7c 606
a895bf8b 607 dimm->nr_pages = npages;
b4e8f0b6 608
854d3349
MCC
609 switch (banks) {
610 case 4:
084a4fcc 611 dimm->dtype = DEV_X4;
854d3349
MCC
612 break;
613 case 8:
084a4fcc 614 dimm->dtype = DEV_X8;
854d3349
MCC
615 break;
616 case 16:
084a4fcc 617 dimm->dtype = DEV_X16;
854d3349
MCC
618 break;
619 default:
084a4fcc 620 dimm->dtype = DEV_UNKNOWN;
854d3349
MCC
621 }
622
084a4fcc
MCC
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;
854d3349 629 }
1c6fed80 630
854d3349
MCC
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]);
956b9ba1 639 edac_dbg(1, "\t[%i] DIVBY3\tREMOVED\tOFFSET\n", i);
854d3349 640 for (j = 0; j < 8; j++)
956b9ba1
JP
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)));
0b2b7b7e
MCC
645 }
646
a0c36a1f
MCC
647 return 0;
648}
649
194a40fe
MCC
650/****************************************************************************
651 Error insertion routines
652 ****************************************************************************/
653
5c4cdb5a
MCC
654#define to_mci(k) container_of(k, struct mem_ctl_info, dev)
655
194a40fe
MCC
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 */
1288c18f 663static int disable_inject(const struct mem_ctl_info *mci)
194a40fe
MCC
664{
665 struct i7core_pvt *pvt = mci->pvt_info;
666
667 pvt->inject.enable = 0;
668
f4742949 669 if (!pvt->pci_ch[pvt->inject.channel][0])
8f331907
MCC
670 return -ENODEV;
671
f4742949 672 pci_write_config_dword(pvt->pci_ch[pvt->inject.channel][0],
4157d9f5 673 MC_CHANNEL_ERROR_INJECT, 0);
8f331907
MCC
674
675 return 0;
194a40fe
MCC
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 */
5c4cdb5a
MCC
685static ssize_t i7core_inject_section_store(struct device *dev,
686 struct device_attribute *mattr,
194a40fe
MCC
687 const char *data, size_t count)
688{
5c4cdb5a 689 struct mem_ctl_info *mci = to_mci(dev);
194a40fe
MCC
690 struct i7core_pvt *pvt = mci->pvt_info;
691 unsigned long value;
692 int rc;
693
694 if (pvt->inject.enable)
41fcb7fe 695 disable_inject(mci);
194a40fe 696
c7f62fc8 697 rc = kstrtoul(data, 10, &value);
194a40fe 698 if ((rc < 0) || (value > 3))
2068def5 699 return -EIO;
194a40fe
MCC
700
701 pvt->inject.section = (u32) value;
702 return count;
703}
704
5c4cdb5a
MCC
705static ssize_t i7core_inject_section_show(struct device *dev,
706 struct device_attribute *mattr,
707 char *data)
194a40fe 708{
5c4cdb5a 709 struct mem_ctl_info *mci = to_mci(dev);
194a40fe
MCC
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 */
5c4cdb5a
MCC
722static ssize_t i7core_inject_type_store(struct device *dev,
723 struct device_attribute *mattr,
194a40fe
MCC
724 const char *data, size_t count)
725{
5c4cdb5a
MCC
726 struct mem_ctl_info *mci = to_mci(dev);
727struct i7core_pvt *pvt = mci->pvt_info;
194a40fe
MCC
728 unsigned long value;
729 int rc;
730
731 if (pvt->inject.enable)
41fcb7fe 732 disable_inject(mci);
194a40fe 733
c7f62fc8 734 rc = kstrtoul(data, 10, &value);
194a40fe 735 if ((rc < 0) || (value > 7))
2068def5 736 return -EIO;
194a40fe
MCC
737
738 pvt->inject.type = (u32) value;
739 return count;
740}
741
5c4cdb5a
MCC
742static ssize_t i7core_inject_type_show(struct device *dev,
743 struct device_attribute *mattr,
744 char *data)
194a40fe 745{
5c4cdb5a 746 struct mem_ctl_info *mci = to_mci(dev);
194a40fe 747 struct i7core_pvt *pvt = mci->pvt_info;
5c4cdb5a 748
194a40fe
MCC
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 */
5c4cdb5a
MCC
762static ssize_t i7core_inject_eccmask_store(struct device *dev,
763 struct device_attribute *mattr,
764 const char *data, size_t count)
194a40fe 765{
5c4cdb5a 766 struct mem_ctl_info *mci = to_mci(dev);
194a40fe
MCC
767 struct i7core_pvt *pvt = mci->pvt_info;
768 unsigned long value;
769 int rc;
770
771 if (pvt->inject.enable)
41fcb7fe 772 disable_inject(mci);
194a40fe 773
c7f62fc8 774 rc = kstrtoul(data, 10, &value);
194a40fe 775 if (rc < 0)
2068def5 776 return -EIO;
194a40fe
MCC
777
778 pvt->inject.eccmask = (u32) value;
779 return count;
780}
781
5c4cdb5a
MCC
782static ssize_t i7core_inject_eccmask_show(struct device *dev,
783 struct device_attribute *mattr,
784 char *data)
194a40fe 785{
5c4cdb5a 786 struct mem_ctl_info *mci = to_mci(dev);
194a40fe 787 struct i7core_pvt *pvt = mci->pvt_info;
5c4cdb5a 788
194a40fe
MCC
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 */
194a40fe 802
a5538e53
MCC
803#define DECLARE_ADDR_MATCH(param, limit) \
804static ssize_t i7core_inject_store_##param( \
5c4cdb5a
MCC
805 struct device *dev, \
806 struct device_attribute *mattr, \
807 const char *data, size_t count) \
a5538e53 808{ \
42709efb 809 struct mem_ctl_info *mci = dev_get_drvdata(dev); \
cc301b3a 810 struct i7core_pvt *pvt; \
a5538e53
MCC
811 long value; \
812 int rc; \
813 \
956b9ba1 814 edac_dbg(1, "\n"); \
cc301b3a
MCC
815 pvt = mci->pvt_info; \
816 \
a5538e53
MCC
817 if (pvt->inject.enable) \
818 disable_inject(mci); \
819 \
4f87fad1 820 if (!strcasecmp(data, "any") || !strcasecmp(data, "any\n"))\
a5538e53
MCC
821 value = -1; \
822 else { \
c7f62fc8 823 rc = kstrtoul(data, 10, &value); \
a5538e53
MCC
824 if ((rc < 0) || (value >= limit)) \
825 return -EIO; \
826 } \
827 \
828 pvt->inject.param = value; \
829 \
830 return count; \
831} \
832 \
833static ssize_t i7core_inject_show_##param( \
5c4cdb5a
MCC
834 struct device *dev, \
835 struct device_attribute *mattr, \
836 char *data) \
a5538e53 837{ \
42709efb 838 struct mem_ctl_info *mci = dev_get_drvdata(dev); \
cc301b3a
MCC
839 struct i7core_pvt *pvt; \
840 \
841 pvt = mci->pvt_info; \
956b9ba1 842 edac_dbg(1, "pvt=%p\n", pvt); \
a5538e53
MCC
843 if (pvt->inject.param < 0) \
844 return sprintf(data, "any\n"); \
845 else \
846 return sprintf(data, "%d\n", pvt->inject.param);\
194a40fe
MCC
847}
848
a5538e53 849#define ATTR_ADDR_MATCH(param) \
5c4cdb5a
MCC
850 static DEVICE_ATTR(param, S_IRUGO | S_IWUSR, \
851 i7core_inject_show_##param, \
852 i7core_inject_store_##param)
194a40fe 853
a5538e53
MCC
854DECLARE_ADDR_MATCH(channel, 3);
855DECLARE_ADDR_MATCH(dimm, 3);
856DECLARE_ADDR_MATCH(rank, 4);
857DECLARE_ADDR_MATCH(bank, 32);
858DECLARE_ADDR_MATCH(page, 0x10000);
859DECLARE_ADDR_MATCH(col, 0x4000);
194a40fe 860
5c4cdb5a
MCC
861ATTR_ADDR_MATCH(channel);
862ATTR_ADDR_MATCH(dimm);
863ATTR_ADDR_MATCH(rank);
864ATTR_ADDR_MATCH(bank);
865ATTR_ADDR_MATCH(page);
866ATTR_ADDR_MATCH(col);
867
1288c18f 868static int write_and_test(struct pci_dev *dev, const int where, const u32 val)
276b824c
MCC
869{
870 u32 read;
871 int count;
872
956b9ba1
JP
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);
4157d9f5 876
276b824c
MCC
877 for (count = 0; count < 10; count++) {
878 if (count)
b990538a 879 msleep(100);
276b824c
MCC
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
4157d9f5
MCC
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);
276b824c
MCC
891
892 return -EINVAL;
893}
894
194a40fe
MCC
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 */
5c4cdb5a
MCC
913static ssize_t i7core_inject_enable_store(struct device *dev,
914 struct device_attribute *mattr,
915 const char *data, size_t count)
194a40fe 916{
5c4cdb5a 917 struct mem_ctl_info *mci = to_mci(dev);
194a40fe
MCC
918 struct i7core_pvt *pvt = mci->pvt_info;
919 u32 injectmask;
920 u64 mask = 0;
921 int rc;
922 long enable;
923
f4742949 924 if (!pvt->pci_ch[pvt->inject.channel][0])
8f331907
MCC
925 return 0;
926
c7f62fc8 927 rc = kstrtoul(data, 10, &enable);
194a40fe
MCC
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)
486dd09f 940 mask |= 1LL << 41;
194a40fe 941 else {
f4742949 942 if (pvt->channel[pvt->inject.channel].dimms > 2)
486dd09f 943 mask |= (pvt->inject.dimm & 0x3LL) << 35;
194a40fe 944 else
486dd09f 945 mask |= (pvt->inject.dimm & 0x1LL) << 36;
194a40fe
MCC
946 }
947
948 /* Sets pvt->inject.rank mask */
949 if (pvt->inject.rank < 0)
486dd09f 950 mask |= 1LL << 40;
194a40fe 951 else {
f4742949 952 if (pvt->channel[pvt->inject.channel].dimms > 2)
486dd09f 953 mask |= (pvt->inject.rank & 0x1LL) << 34;
194a40fe 954 else
486dd09f 955 mask |= (pvt->inject.rank & 0x3LL) << 34;
194a40fe
MCC
956 }
957
958 /* Sets pvt->inject.bank mask */
959 if (pvt->inject.bank < 0)
486dd09f 960 mask |= 1LL << 39;
194a40fe 961 else
486dd09f 962 mask |= (pvt->inject.bank & 0x15LL) << 30;
194a40fe
MCC
963
964 /* Sets pvt->inject.page mask */
965 if (pvt->inject.page < 0)
486dd09f 966 mask |= 1LL << 38;
194a40fe 967 else
486dd09f 968 mask |= (pvt->inject.page & 0xffff) << 14;
194a40fe
MCC
969
970 /* Sets pvt->inject.column mask */
971 if (pvt->inject.col < 0)
486dd09f 972 mask |= 1LL << 37;
194a40fe 973 else
486dd09f 974 mask |= (pvt->inject.col & 0x3fff);
194a40fe 975
276b824c
MCC
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 */
f4742949 988 pci_write_config_dword(pvt->pci_noncore,
67166af4 989 MC_CFG_CONTROL, 0x2);
e9bd2e73 990
f4742949 991 write_and_test(pvt->pci_ch[pvt->inject.channel][0],
194a40fe 992 MC_CHANNEL_ADDR_MATCH, mask);
f4742949 993 write_and_test(pvt->pci_ch[pvt->inject.channel][0],
7b029d03 994 MC_CHANNEL_ADDR_MATCH + 4, mask >> 32L);
7b029d03 995
f4742949 996 write_and_test(pvt->pci_ch[pvt->inject.channel][0],
194a40fe
MCC
997 MC_CHANNEL_ERROR_MASK, pvt->inject.eccmask);
998
f4742949 999 write_and_test(pvt->pci_ch[pvt->inject.channel][0],
4157d9f5 1000 MC_CHANNEL_ERROR_INJECT, injectmask);
276b824c 1001
194a40fe 1002 /*
276b824c
MCC
1003 * This is something undocumented, based on my tests
1004 * Without writing 8 to this register, errors aren't injected. Not sure
1005 * why.
194a40fe 1006 */
f4742949 1007 pci_write_config_dword(pvt->pci_noncore,
276b824c 1008 MC_CFG_CONTROL, 8);
194a40fe 1009
956b9ba1
JP
1010 edac_dbg(0, "Error inject addr match 0x%016llx, ecc 0x%08x, inject 0x%08x\n",
1011 mask, pvt->inject.eccmask, injectmask);
194a40fe 1012
7b029d03 1013
194a40fe
MCC
1014 return count;
1015}
1016
5c4cdb5a
MCC
1017static ssize_t i7core_inject_enable_show(struct device *dev,
1018 struct device_attribute *mattr,
1019 char *data)
194a40fe 1020{
5c4cdb5a 1021 struct mem_ctl_info *mci = to_mci(dev);
194a40fe 1022 struct i7core_pvt *pvt = mci->pvt_info;
7b029d03
MCC
1023 u32 injectmask;
1024
52a2e4fc
MCC
1025 if (!pvt->pci_ch[pvt->inject.channel][0])
1026 return 0;
1027
f4742949 1028 pci_read_config_dword(pvt->pci_ch[pvt->inject.channel][0],
4157d9f5 1029 MC_CHANNEL_ERROR_INJECT, &injectmask);
7b029d03 1030
956b9ba1 1031 edac_dbg(0, "Inject error read: 0x%018x\n", injectmask);
7b029d03
MCC
1032
1033 if (injectmask & 0x0c)
1034 pvt->inject.enable = 1;
1035
194a40fe
MCC
1036 return sprintf(data, "%d\n", pvt->inject.enable);
1037}
1038
f338d736
MCC
1039#define DECLARE_COUNTER(param) \
1040static ssize_t i7core_show_counter_##param( \
5c4cdb5a
MCC
1041 struct device *dev, \
1042 struct device_attribute *mattr, \
1043 char *data) \
f338d736 1044{ \
42709efb 1045 struct mem_ctl_info *mci = dev_get_drvdata(dev); \
f338d736
MCC
1046 struct i7core_pvt *pvt = mci->pvt_info; \
1047 \
956b9ba1 1048 edac_dbg(1, "\n"); \
f338d736
MCC
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}
442305b1 1054
f338d736 1055#define ATTR_COUNTER(param) \
5c4cdb5a
MCC
1056 static DEVICE_ATTR(udimm##param, S_IRUGO | S_IWUSR, \
1057 i7core_show_counter_##param, \
1058 NULL)
442305b1 1059
f338d736
MCC
1060DECLARE_COUNTER(0);
1061DECLARE_COUNTER(1);
1062DECLARE_COUNTER(2);
442305b1 1063
5c4cdb5a
MCC
1064ATTR_COUNTER(0);
1065ATTR_COUNTER(1);
1066ATTR_COUNTER(2);
1067
194a40fe 1068/*
5c4cdb5a 1069 * inject_addrmatch device sysfs struct
194a40fe 1070 */
a5538e53 1071
5c4cdb5a
MCC
1072static 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
a5538e53
MCC
1080};
1081
5c4cdb5a
MCC
1082static struct attribute_group addrmatch_grp = {
1083 .attrs = i7core_addrmatch_attrs,
a5538e53
MCC
1084};
1085
5c4cdb5a
MCC
1086static const struct attribute_group *addrmatch_groups[] = {
1087 &addrmatch_grp,
1088 NULL
f338d736
MCC
1089};
1090
5c4cdb5a
MCC
1091static void addrmatch_release(struct device *device)
1092{
956b9ba1 1093 edac_dbg(1, "Releasing device %s\n", dev_name(device));
356f0a30 1094 kfree(device);
5c4cdb5a
MCC
1095}
1096
1097static struct device_type addrmatch_type = {
1098 .groups = addrmatch_groups,
1099 .release = addrmatch_release,
f338d736
MCC
1100};
1101
5c4cdb5a
MCC
1102/*
1103 * all_channel_counts sysfs struct
1104 */
1105
1106static struct attribute *i7core_udimm_counters_attrs[] = {
1107 &dev_attr_udimm0.attr,
1108 &dev_attr_udimm1.attr,
1109 &dev_attr_udimm2.attr,
1110 NULL
1288c18f
MCC
1111};
1112
5c4cdb5a
MCC
1113static struct attribute_group all_channel_counts_grp = {
1114 .attrs = i7core_udimm_counters_attrs,
194a40fe
MCC
1115};
1116
5c4cdb5a
MCC
1117static const struct attribute_group *all_channel_counts_groups[] = {
1118 &all_channel_counts_grp,
1119 NULL
194a40fe
MCC
1120};
1121
5c4cdb5a
MCC
1122static void all_channel_counts_release(struct device *device)
1123{
956b9ba1 1124 edac_dbg(1, "Releasing device %s\n", dev_name(device));
356f0a30 1125 kfree(device);
5c4cdb5a
MCC
1126}
1127
1128static 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
1137static DEVICE_ATTR(inject_section, S_IRUGO | S_IWUSR,
1138 i7core_inject_section_show, i7core_inject_section_store);
1139
1140static DEVICE_ATTR(inject_type, S_IRUGO | S_IWUSR,
1141 i7core_inject_type_show, i7core_inject_type_store);
1142
1143
1144static DEVICE_ATTR(inject_eccmask, S_IRUGO | S_IWUSR,
1145 i7core_inject_eccmask_show, i7core_inject_eccmask_store);
1146
1147static DEVICE_ATTR(inject_enable, S_IRUGO | S_IWUSR,
1148 i7core_inject_enable_show, i7core_inject_enable_store);
1149
2eace188
TI
1150static 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
1158ATTRIBUTE_GROUPS(i7core_dev);
1159
5c4cdb5a
MCC
1160static int i7core_create_sysfs_devices(struct mem_ctl_info *mci)
1161{
1162 struct i7core_pvt *pvt = mci->pvt_info;
1163 int rc;
1164
356f0a30
MCC
1165 pvt->addrmatch_dev = kzalloc(sizeof(*pvt->addrmatch_dev), GFP_KERNEL);
1166 if (!pvt->addrmatch_dev)
e97d7e38 1167 return -ENOMEM;
356f0a30
MCC
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);
5c4cdb5a 1175
956b9ba1 1176 edac_dbg(1, "creating %s\n", dev_name(pvt->addrmatch_dev));
5c4cdb5a 1177
356f0a30 1178 rc = device_add(pvt->addrmatch_dev);
5c4cdb5a
MCC
1179 if (rc < 0)
1180 return rc;
1181
1182 if (!pvt->is_registered) {
356f0a30
MCC
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);
e97d7e38 1188 return -ENOMEM;
356f0a30
MCC
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);
5c4cdb5a 1197
956b9ba1 1198 edac_dbg(1, "creating %s\n", dev_name(pvt->chancounts_dev));
5c4cdb5a 1199
356f0a30 1200 rc = device_add(pvt->chancounts_dev);
5c4cdb5a
MCC
1201 if (rc < 0)
1202 return rc;
1203 }
1204 return 0;
1205}
1206
1207static void i7core_delete_sysfs_devices(struct mem_ctl_info *mci)
1208{
1209 struct i7core_pvt *pvt = mci->pvt_info;
1210
956b9ba1 1211 edac_dbg(1, "\n");
5c4cdb5a 1212
5c4cdb5a 1213 if (!pvt->is_registered) {
356f0a30
MCC
1214 put_device(pvt->chancounts_dev);
1215 device_del(pvt->chancounts_dev);
5c4cdb5a 1216 }
356f0a30
MCC
1217 put_device(pvt->addrmatch_dev);
1218 device_del(pvt->addrmatch_dev);
5c4cdb5a
MCC
1219}
1220
a0c36a1f
MCC
1221/****************************************************************************
1222 Device initialization routines: put/get, init/exit
1223 ****************************************************************************/
1224
1225/*
64c10f6e 1226 * i7core_put_all_devices 'put' all the devices that we have
a0c36a1f
MCC
1227 * reserved via 'get'
1228 */
13d6e9b6 1229static void i7core_put_devices(struct i7core_dev *i7core_dev)
a0c36a1f 1230{
13d6e9b6 1231 int i;
a0c36a1f 1232
956b9ba1 1233 edac_dbg(0, "\n");
de06eeef 1234 for (i = 0; i < i7core_dev->n_devs; i++) {
22e6bcbd
MCC
1235 struct pci_dev *pdev = i7core_dev->pdev[i];
1236 if (!pdev)
1237 continue;
956b9ba1
JP
1238 edac_dbg(0, "Removing dev %02x:%02x.%d\n",
1239 pdev->bus->number,
1240 PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn));
22e6bcbd
MCC
1241 pci_dev_put(pdev);
1242 }
13d6e9b6 1243}
66607706 1244
13d6e9b6
MCC
1245static void i7core_put_all_devices(void)
1246{
42538680 1247 struct i7core_dev *i7core_dev, *tmp;
13d6e9b6 1248
39300e71 1249 list_for_each_entry_safe(i7core_dev, tmp, &i7core_edac_list, list) {
13d6e9b6 1250 i7core_put_devices(i7core_dev);
2aa9be44 1251 free_i7core_dev(i7core_dev);
39300e71 1252 }
a0c36a1f
MCC
1253}
1254
1288c18f 1255static void __init i7core_xeon_pci_fixup(const struct pci_id_table *table)
bc2d7245
KM
1256{
1257 struct pci_dev *pdev = NULL;
1258 int i;
54a08ab1 1259
bc2d7245 1260 /*
e7bf068a 1261 * On Xeon 55xx, the Intel Quick Path Arch Generic Non-core pci buses
bc2d7245
KM
1262 * aren't announced by acpi. So, we need to use a legacy scan probing
1263 * to detect them
1264 */
bd9e19ca
VM
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 }
bda14289 1271 pci_dev_put(pdev);
bd9e19ca 1272 table++;
bc2d7245
KM
1273 }
1274}
1275
bda14289
MCC
1276static 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;
956b9ba1 1283 edac_dbg(0, "Found bus %d\n", bus);
bda14289
MCC
1284 if (bus > last_bus)
1285 last_bus = bus;
1286 }
1287
956b9ba1 1288 edac_dbg(0, "Last bus %d\n", last_bus);
bda14289
MCC
1289
1290 return last_bus;
1291}
1292
a0c36a1f 1293/*
64c10f6e 1294 * i7core_get_all_devices Find and perform 'get' operation on the MCH's
a0c36a1f
MCC
1295 * device/functions we want to reference for this driver
1296 *
1297 * Need to 'get' device 16 func 1 and func 2
1298 */
b197cba0
HS
1299static int i7core_get_onedevice(struct pci_dev **prev,
1300 const struct pci_id_table *table,
1301 const unsigned devno,
1302 const unsigned last_bus)
a0c36a1f 1303{
66607706 1304 struct i7core_dev *i7core_dev;
b197cba0 1305 const struct pci_id_descr *dev_descr = &table->descr[devno];
66607706 1306
8f331907 1307 struct pci_dev *pdev = NULL;
67166af4
MCC
1308 u8 bus = 0;
1309 u8 socket = 0;
a0c36a1f 1310
c77720b9 1311 pdev = pci_get_device(PCI_VENDOR_ID_INTEL,
de06eeef 1312 dev_descr->dev_id, *prev);
c77720b9 1313
224e871f 1314 /*
15ed103a 1315 * On Xeon 55xx, the Intel QuickPath Arch Generic Non-core regs
224e871f
MCC
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 */
c0f5eeed
JD
1319 if (dev_descr->dev_id == PCI_DEVICE_ID_INTEL_I7_NONCORE && !pdev) {
1320 pci_dev_get(*prev); /* pci_get_device will put it */
224e871f
MCC
1321 pdev = pci_get_device(PCI_VENDOR_ID_INTEL,
1322 PCI_DEVICE_ID_INTEL_I7_NONCORE_ALT, *prev);
c0f5eeed 1323 }
224e871f 1324
c0f5eeed
JD
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 */
224e871f
MCC
1328 pdev = pci_get_device(PCI_VENDOR_ID_INTEL,
1329 PCI_DEVICE_ID_INTEL_LYNNFIELD_NONCORE_ALT,
1330 *prev);
c0f5eeed 1331 }
224e871f 1332
c77720b9
MCC
1333 if (!pdev) {
1334 if (*prev) {
1335 *prev = pdev;
1336 return 0;
d1fd4fb6
MCC
1337 }
1338
de06eeef 1339 if (dev_descr->optional)
c77720b9 1340 return 0;
310cbb72 1341
bd9e19ca
VM
1342 if (devno == 0)
1343 return -ENODEV;
1344
ab089374 1345 i7core_printk(KERN_INFO,
c77720b9 1346 "Device not found: dev %02x.%d PCI ID %04x:%04x\n",
de06eeef
MCC
1347 dev_descr->dev, dev_descr->func,
1348 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
67166af4 1349
c77720b9
MCC
1350 /* End of list, leave */
1351 return -ENODEV;
1352 }
1353 bus = pdev->bus->number;
67166af4 1354
bda14289 1355 socket = last_bus - bus;
c77720b9 1356
66607706
MCC
1357 i7core_dev = get_i7core_dev(socket);
1358 if (!i7core_dev) {
848b2f7e 1359 i7core_dev = alloc_i7core_dev(socket, table);
2896637b
HS
1360 if (!i7core_dev) {
1361 pci_dev_put(pdev);
66607706 1362 return -ENOMEM;
2896637b 1363 }
c77720b9 1364 }
67166af4 1365
66607706 1366 if (i7core_dev->pdev[devno]) {
c77720b9
MCC
1367 i7core_printk(KERN_ERR,
1368 "Duplicated device for "
1369 "dev %02x:%02x.%d PCI ID %04x:%04x\n",
de06eeef
MCC
1370 bus, dev_descr->dev, dev_descr->func,
1371 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
c77720b9
MCC
1372 pci_dev_put(pdev);
1373 return -ENODEV;
1374 }
67166af4 1375
66607706 1376 i7core_dev->pdev[devno] = pdev;
c77720b9
MCC
1377
1378 /* Sanity check */
de06eeef
MCC
1379 if (unlikely(PCI_SLOT(pdev->devfn) != dev_descr->dev ||
1380 PCI_FUNC(pdev->devfn) != dev_descr->func)) {
c77720b9
MCC
1381 i7core_printk(KERN_ERR,
1382 "Device PCI ID %04x:%04x "
1383 "has dev %02x:%02x.%d instead of dev %02x:%02x.%d\n",
de06eeef 1384 PCI_VENDOR_ID_INTEL, dev_descr->dev_id,
c77720b9 1385 bus, PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn),
de06eeef 1386 bus, dev_descr->dev, dev_descr->func);
c77720b9
MCC
1387 return -ENODEV;
1388 }
ef708b53 1389
c77720b9
MCC
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",
de06eeef
MCC
1395 bus, dev_descr->dev, dev_descr->func,
1396 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
c77720b9
MCC
1397 return -ENODEV;
1398 }
ef708b53 1399
956b9ba1
JP
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);
8f331907 1404
a3e15416
MCC
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
c77720b9 1412 *prev = pdev;
ef708b53 1413
c77720b9
MCC
1414 return 0;
1415}
a0c36a1f 1416
64c10f6e 1417static int i7core_get_all_devices(void)
c77720b9 1418{
3c52cc57 1419 int i, rc, last_bus;
c77720b9 1420 struct pci_dev *pdev = NULL;
3c52cc57 1421 const struct pci_id_table *table = pci_dev_table;
bd9e19ca 1422
bda14289
MCC
1423 last_bus = i7core_pci_lastbus();
1424
3c52cc57 1425 while (table && table->descr) {
bd9e19ca
VM
1426 for (i = 0; i < table->n_devs; i++) {
1427 pdev = NULL;
1428 do {
b197cba0 1429 rc = i7core_get_onedevice(&pdev, table, i,
bda14289 1430 last_bus);
bd9e19ca
VM
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 }
3c52cc57 1441 table++;
c77720b9 1442 }
66607706 1443
ef708b53 1444 return 0;
ef708b53
MCC
1445}
1446
f4742949
MCC
1447static int mci_bind_devs(struct mem_ctl_info *mci,
1448 struct i7core_dev *i7core_dev)
ef708b53
MCC
1449{
1450 struct i7core_pvt *pvt = mci->pvt_info;
1451 struct pci_dev *pdev;
f4742949 1452 int i, func, slot;
27100db0 1453 char *family;
ef708b53 1454
27100db0
MCC
1455 pvt->is_registered = false;
1456 pvt->enable_scrub = false;
de06eeef 1457 for (i = 0; i < i7core_dev->n_devs; i++) {
f4742949
MCC
1458 pdev = i7core_dev->pdev[i];
1459 if (!pdev)
66607706
MCC
1460 continue;
1461
f4742949
MCC
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))
ef708b53 1470 goto error;
f4742949 1471 pvt->pci_ch[slot - 4][func] = pdev;
27100db0 1472 } else if (!slot && !func) {
f4742949 1473 pvt->pci_noncore = pdev;
27100db0
MCC
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 }
956b9ba1 1501 edac_dbg(0, "Detected a processor type %s\n", family);
27100db0 1502 } else
f4742949 1503 goto error;
ef708b53 1504
956b9ba1
JP
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);
14d2c083 1508
f4742949
MCC
1509 if (PCI_SLOT(pdev->devfn) == 3 &&
1510 PCI_FUNC(pdev->devfn) == 2)
27100db0 1511 pvt->is_registered = true;
a0c36a1f 1512 }
e9bd2e73 1513
a0c36a1f 1514 return 0;
ef708b53
MCC
1515
1516error:
1517 i7core_printk(KERN_ERR, "Device %d, function %d "
1518 "is out of the expected range\n",
1519 slot, func);
1520 return -EINVAL;
a0c36a1f
MCC
1521}
1522
442305b1
MCC
1523/****************************************************************************
1524 Error check routines
1525 ****************************************************************************/
b4e8f0b6
MCC
1526
1527static void i7core_rdimm_update_ce_count(struct mem_ctl_info *mci,
1288c18f
MCC
1528 const int chan,
1529 const int new0,
1530 const int new1,
1531 const int new2)
b4e8f0b6
MCC
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 */
f4742949 1536 if (pvt->ce_count_available) {
b4e8f0b6
MCC
1537 /* Updates CE counters */
1538
f4742949
MCC
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];
b4e8f0b6
MCC
1542
1543 if (add2 < 0)
1544 add2 += 0x7fff;
f4742949 1545 pvt->rdimm_ce_count[chan][2] += add2;
b4e8f0b6
MCC
1546
1547 if (add1 < 0)
1548 add1 += 0x7fff;
f4742949 1549 pvt->rdimm_ce_count[chan][1] += add1;
b4e8f0b6
MCC
1550
1551 if (add0 < 0)
1552 add0 += 0x7fff;
f4742949 1553 pvt->rdimm_ce_count[chan][0] += add0;
b4e8f0b6 1554 } else
f4742949 1555 pvt->ce_count_available = 1;
b4e8f0b6
MCC
1556
1557 /* Store the new values */
f4742949
MCC
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;
b4e8f0b6
MCC
1561
1562 /*updated the edac core */
1563 if (add0 != 0)
00d18339
MCC
1564 edac_mc_handle_error(HW_EVENT_ERR_CORRECTED, mci, add0,
1565 0, 0, 0,
1566 chan, 0, -1, "error", "");
b4e8f0b6 1567 if (add1 != 0)
00d18339
MCC
1568 edac_mc_handle_error(HW_EVENT_ERR_CORRECTED, mci, add1,
1569 0, 0, 0,
1570 chan, 1, -1, "error", "");
b4e8f0b6 1571 if (add2 != 0)
00d18339
MCC
1572 edac_mc_handle_error(HW_EVENT_ERR_CORRECTED, mci, add2,
1573 0, 0, 0,
1574 chan, 2, -1, "error", "");
b4e8f0b6
MCC
1575}
1576
f4742949 1577static void i7core_rdimm_check_mc_ecc_err(struct mem_ctl_info *mci)
b4e8f0b6
MCC
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*/
f4742949 1584 pci_read_config_dword(pvt->pci_mcr[2], MC_COR_ECC_CNT_0,
b4e8f0b6 1585 &rcv[0][0]);
f4742949 1586 pci_read_config_dword(pvt->pci_mcr[2], MC_COR_ECC_CNT_1,
b4e8f0b6 1587 &rcv[0][1]);
f4742949 1588 pci_read_config_dword(pvt->pci_mcr[2], MC_COR_ECC_CNT_2,
b4e8f0b6 1589 &rcv[1][0]);
f4742949 1590 pci_read_config_dword(pvt->pci_mcr[2], MC_COR_ECC_CNT_3,
b4e8f0b6 1591 &rcv[1][1]);
f4742949 1592 pci_read_config_dword(pvt->pci_mcr[2], MC_COR_ECC_CNT_4,
b4e8f0b6 1593 &rcv[2][0]);
f4742949 1594 pci_read_config_dword(pvt->pci_mcr[2], MC_COR_ECC_CNT_5,
b4e8f0b6
MCC
1595 &rcv[2][1]);
1596 for (i = 0 ; i < 3; i++) {
956b9ba1
JP
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]);
b4e8f0b6 1599 /*if the channel has 3 dimms*/
f4742949 1600 if (pvt->channel[i].dimms > 2) {
b4e8f0b6
MCC
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
f4742949 1612 i7core_rdimm_update_ce_count(mci, i, new0, new1, new2);
b4e8f0b6
MCC
1613 }
1614}
442305b1
MCC
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 */
f4742949 1622static void i7core_udimm_check_mc_ecc_err(struct mem_ctl_info *mci)
442305b1
MCC
1623{
1624 struct i7core_pvt *pvt = mci->pvt_info;
1625 u32 rcv1, rcv0;
1626 int new0, new1, new2;
1627
f4742949 1628 if (!pvt->pci_mcr[4]) {
956b9ba1 1629 edac_dbg(0, "MCR registers not found\n");
442305b1
MCC
1630 return;
1631 }
1632
b4e8f0b6 1633 /* Corrected test errors */
f4742949
MCC
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);
442305b1
MCC
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
442305b1 1642 /* Updates CE counters if it is not the first time here */
f4742949 1643 if (pvt->ce_count_available) {
442305b1
MCC
1644 /* Updates CE counters */
1645 int add0, add1, add2;
1646
f4742949
MCC
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];
442305b1
MCC
1650
1651 if (add2 < 0)
1652 add2 += 0x7fff;
f4742949 1653 pvt->udimm_ce_count[2] += add2;
442305b1
MCC
1654
1655 if (add1 < 0)
1656 add1 += 0x7fff;
f4742949 1657 pvt->udimm_ce_count[1] += add1;
442305b1
MCC
1658
1659 if (add0 < 0)
1660 add0 += 0x7fff;
f4742949 1661 pvt->udimm_ce_count[0] += add0;
b4e8f0b6
MCC
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);
442305b1 1667 } else
f4742949 1668 pvt->ce_count_available = 1;
442305b1
MCC
1669
1670 /* Store the new values */
f4742949
MCC
1671 pvt->udimm_last_ce_count[2] = new2;
1672 pvt->udimm_last_ce_count[1] = new1;
1673 pvt->udimm_last_ce_count[0] = new0;
442305b1
MCC
1674}
1675
8a2f118e
MCC
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.
f237fcf2
MCC
1679 * Nehalem are defined as family 0x06, model 0x1a
1680 *
1681 * The MCA registers used here are the following ones:
8a2f118e 1682 * struct mce field MCA Register
f237fcf2
MCC
1683 * m->status MSR_IA32_MC8_STATUS
1684 * m->addr MSR_IA32_MC8_ADDR
1685 * m->misc MSR_IA32_MC8_MISC
8a2f118e
MCC
1686 * In the case of Nehalem, the error information is masked at .status and .misc
1687 * fields
1688 */
d5381642 1689static void i7core_mce_output_error(struct mem_ctl_info *mci,
1288c18f 1690 const struct mce *m)
d5381642 1691{
b4e8f0b6 1692 struct i7core_pvt *pvt = mci->pvt_info;
f118920b 1693 char *optype, *err;
0975c16f 1694 enum hw_event_mc_err_type tp_event;
8a2f118e 1695 unsigned long error = m->status & 0x1ff0000l;
0975c16f
MCC
1696 bool uncorrected_error = m->mcgstatus & 1ll << 61;
1697 bool ripv = m->mcgstatus & 1;
a639539f 1698 u32 optypenum = (m->status >> 4) & 0x07;
8cf2d239 1699 u32 core_err_cnt = (m->status >> 38) & 0x7fff;
8a2f118e
MCC
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
0975c16f 1705 if (uncorrected_error) {
f118920b 1706 if (ripv)
0975c16f 1707 tp_event = HW_EVENT_ERR_FATAL;
f118920b 1708 else
0975c16f 1709 tp_event = HW_EVENT_ERR_UNCORRECTED;
0975c16f 1710 } else {
0975c16f
MCC
1711 tp_event = HW_EVENT_ERR_CORRECTED;
1712 }
c5d34528 1713
a639539f 1714 switch (optypenum) {
b990538a
MCC
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;
a639539f
MCC
1733 }
1734
8a2f118e
MCC
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";
d5381642 1765 }
d5381642 1766
0975c16f
MCC
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)
00d18339 1773 edac_mc_handle_error(tp_event, mci, core_err_cnt,
0975c16f
MCC
1774 m->addr >> PAGE_SHIFT,
1775 m->addr & ~PAGE_MASK,
1776 syndrome,
1777 channel, dimm, -1,
00d18339 1778 err, optype);
d5381642
MCC
1779}
1780
87d1d272
MCC
1781/*
1782 * i7core_check_error Retrieve and process errors reported by the
1783 * hardware. Called by the Core module.
1784 */
53595345 1785static void i7core_check_error(struct mem_ctl_info *mci, struct mce *m)
87d1d272 1786{
d5381642 1787 struct i7core_pvt *pvt = mci->pvt_info;
d5381642 1788
53595345 1789 i7core_mce_output_error(mci, m);
d5381642 1790
ca9c90ba
MCC
1791 /*
1792 * Now, let's increment CE error counts
1793 */
f4742949
MCC
1794 if (!pvt->is_registered)
1795 i7core_udimm_check_mc_ecc_err(mci);
1796 else
1797 i7core_rdimm_check_mc_ecc_err(mci);
87d1d272
MCC
1798}
1799
d5381642 1800/*
53595345
TL
1801 * Check that logging is enabled and that this is the right type
1802 * of error for us to handle.
d5381642 1803 */
4140c542
BP
1804static int i7core_mce_check_error(struct notifier_block *nb, unsigned long val,
1805 void *data)
d5381642 1806{
4140c542
BP
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)
c4fc1956 1814 return NOTIFY_DONE;
4140c542
BP
1815
1816 mci = i7_dev->mci;
1817 pvt = mci->pvt_info;
d5381642 1818
8a2f118e
MCC
1819 /*
1820 * Just let mcelog handle it if the error is
1821 * outside the memory controller
1822 */
1823 if (((mce->status & 0xffff) >> 7) != 1)
4140c542 1824 return NOTIFY_DONE;
8a2f118e 1825
f237fcf2
MCC
1826 /* Bank 8 registers are the only ones that we know how to handle */
1827 if (mce->bank != 8)
4140c542 1828 return NOTIFY_DONE;
f237fcf2 1829
53595345 1830 i7core_check_error(mci, mce);
c5d34528 1831
e7bf068a 1832 /* Advise mcelog that the errors were handled */
4140c542 1833 return NOTIFY_STOP;
d5381642
MCC
1834}
1835
4140c542
BP
1836static struct notifier_block i7_mce_dec = {
1837 .notifier_call = i7core_mce_check_error,
1838};
1839
535e9c78
NC
1840struct 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 */
1871static 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
1934static 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
e8b6a127
SG
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 */
1951static 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;
e8b6a127
SG
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 */
535e9c78
NC
1969 write_and_test(pdev, MC_SCRUB_CONTROL,
1970 dw_scrub & ~SCRUBINTERVAL_MASK);
e8b6a127
SG
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 {
535e9c78
NC
1977 const int cache_line_size = 64;
1978 const u32 freq_dclk_mhz = pvt->dclk_freq;
1979 unsigned long long scrub_interval;
e8b6a127
SG
1980 /*
1981 * Translate the desired scrub rate to a register value and
535e9c78 1982 * program the corresponding register value.
e8b6a127 1983 */
535e9c78 1984 scrub_interval = (unsigned long long)freq_dclk_mhz *
4fad8098
SD
1985 cache_line_size * 1000000;
1986 do_div(scrub_interval, new_bw);
535e9c78
NC
1987
1988 if (!scrub_interval || scrub_interval > SCRUBINTERVAL_MASK)
1989 return -EINVAL;
1990
1991 dw_scrub = SCRUBINTERVAL_MASK & scrub_interval;
e8b6a127
SG
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
15ed103a 2010 * into byte/sec bandwidth according to
e8b6a127
SG
2011 * SCRUBINTERVAL formula found in datasheet.
2012 */
2013static 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;
535e9c78
NC
2018 const u32 freq_dclk_mhz = pvt->dclk_freq;
2019 unsigned long long scrub_rate;
e8b6a127
SG
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 */
535e9c78 2031 scrubval &= SCRUBINTERVAL_MASK;
e8b6a127
SG
2032 if (!scrubval)
2033 return 0;
2034
2035 /* Calculate scrub rate value into byte/sec bandwidth */
535e9c78 2036 scrub_rate = (unsigned long long)freq_dclk_mhz *
4fad8098
SD
2037 1000000 * cache_line_size;
2038 do_div(scrub_rate, scrubval);
535e9c78 2039 return (int)scrub_rate;
e8b6a127
SG
2040}
2041
2042static 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
2057static 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
a3aa0a4a
HS
2069static 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))
f9902f24
MCC
2075 i7core_printk(KERN_WARNING,
2076 "Unable to setup PCI error report via EDAC\n");
a3aa0a4a
HS
2077}
2078
2079static 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
1c6edbbe
HS
2090static 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)) {
956b9ba1 2096 edac_dbg(0, "MC: dev = %p\n", &i7core_dev->pdev[0]->dev);
1c6edbbe
HS
2097
2098 i7core_printk(KERN_ERR, "Couldn't find mci handler\n");
2099 return;
2100 }
2101
2102 pvt = mci->pvt_info;
2103
956b9ba1 2104 edac_dbg(0, "MC: mci = %p, dev = %p\n", mci, &i7core_dev->pdev[0]->dev);
1c6edbbe 2105
e8b6a127 2106 /* Disable scrubrate setting */
27100db0
MCC
2107 if (pvt->enable_scrub)
2108 disable_sdram_scrub_setting(mci);
e8b6a127 2109
1c6edbbe
HS
2110 /* Disable EDAC polling */
2111 i7core_pci_ctl_release(pvt);
2112
2113 /* Remove MC sysfs nodes */
5c4cdb5a 2114 i7core_delete_sysfs_devices(mci);
fd687502 2115 edac_mc_del_mc(mci->pdev);
1c6edbbe 2116
956b9ba1 2117 edac_dbg(1, "%s: free mci struct\n", mci->ctl_name);
1c6edbbe
HS
2118 kfree(mci->ctl_name);
2119 edac_mc_free(mci);
2120 i7core_dev->mci = NULL;
2121}
2122
aace4283 2123static int i7core_register_mci(struct i7core_dev *i7core_dev)
a0c36a1f
MCC
2124{
2125 struct mem_ctl_info *mci;
2126 struct i7core_pvt *pvt;
0975c16f
MCC
2127 int rc;
2128 struct edac_mc_layer layers[2];
a0c36a1f 2129
a0c36a1f 2130 /* allocate a new MC control structure */
0975c16f
MCC
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;
ca0907b9 2138 mci = edac_mc_alloc(i7core_dev->socket, ARRAY_SIZE(layers), layers,
0975c16f 2139 sizeof(*pvt));
f4742949
MCC
2140 if (unlikely(!mci))
2141 return -ENOMEM;
a0c36a1f 2142
956b9ba1 2143 edac_dbg(0, "MC: mci = %p, dev = %p\n", mci, &i7core_dev->pdev[0]->dev);
a0c36a1f 2144
a0c36a1f 2145 pvt = mci->pvt_info;
ef708b53 2146 memset(pvt, 0, sizeof(*pvt));
67166af4 2147
6d37d240
MCC
2148 /* Associates i7core_dev and mci for future usage */
2149 pvt->i7core_dev = i7core_dev;
2150 i7core_dev->mci = mci;
2151
41fcb7fe
MCC
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;
a0c36a1f
MCC
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;
f4742949
MCC
2162 mci->ctl_name = kasprintf(GFP_KERNEL, "i7 core #%d",
2163 i7core_dev->socket);
2164 mci->dev_name = pci_name(i7core_dev->pdev[0]);
a0c36a1f 2165 mci->ctl_page_to_phys = NULL;
1288c18f 2166
ef708b53 2167 /* Store pci devices at mci for faster access */
f4742949 2168 rc = mci_bind_devs(mci, i7core_dev);
41fcb7fe 2169 if (unlikely(rc < 0))
628c5ddf 2170 goto fail0;
ef708b53 2171
5939813b 2172
ef708b53 2173 /* Get dimm basic config */
2e5185f7 2174 get_dimm_config(mci);
5939813b 2175 /* record ptr to the generic device */
fd687502 2176 mci->pdev = &i7core_dev->pdev[0]->dev;
ef708b53 2177
e8b6a127 2178 /* Enable scrubrate setting */
27100db0
MCC
2179 if (pvt->enable_scrub)
2180 enable_sdram_scrub_setting(mci);
e8b6a127 2181
a0c36a1f 2182 /* add this new MC control structure to EDAC's list of MCs */
2eace188 2183 if (unlikely(edac_mc_add_mc_with_groups(mci, i7core_dev_groups))) {
956b9ba1 2184 edac_dbg(0, "MC: failed edac_mc_add_mc()\n");
a0c36a1f
MCC
2185 /* FIXME: perhaps some code should go here that disables error
2186 * reporting if we just enabled it
2187 */
b7c76151
MCC
2188
2189 rc = -EINVAL;
628c5ddf 2190 goto fail0;
a0c36a1f 2191 }
5c4cdb5a 2192 if (i7core_create_sysfs_devices(mci)) {
956b9ba1 2193 edac_dbg(0, "MC: failed to create sysfs nodes\n");
5c4cdb5a
MCC
2194 edac_mc_del_mc(mci->pdev);
2195 rc = -EINVAL;
2196 goto fail0;
2197 }
a0c36a1f 2198
194a40fe 2199 /* Default error mask is any memory */
ef708b53 2200 pvt->inject.channel = 0;
194a40fe
MCC
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
a3aa0a4a
HS
2207 /* allocating generic PCI control info */
2208 i7core_pci_ctl_create(pvt);
2209
535e9c78
NC
2210 /* DCLK for scrub rate setting */
2211 pvt->dclk_freq = get_dclk_freq();
2212
628c5ddf
HS
2213 return 0;
2214
628c5ddf
HS
2215fail0:
2216 kfree(mci->ctl_name);
2217 edac_mc_free(mci);
1c6edbbe 2218 i7core_dev->mci = NULL;
f4742949
MCC
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 */
2d95d815 2229
9b3c6e85 2230static int i7core_probe(struct pci_dev *pdev, const struct pci_device_id *id)
f4742949 2231{
40557591 2232 int rc, count = 0;
f4742949
MCC
2233 struct i7core_dev *i7core_dev;
2234
2d95d815
MCC
2235 /* get the pci devices we want to reserve for our use */
2236 mutex_lock(&i7core_edac_lock);
2237
f4742949 2238 /*
d4c27795 2239 * All memory controllers are allocated at the first pass.
f4742949 2240 */
2d95d815
MCC
2241 if (unlikely(probed >= 1)) {
2242 mutex_unlock(&i7core_edac_lock);
76a7bd81 2243 return -ENODEV;
2d95d815
MCC
2244 }
2245 probed++;
de06eeef 2246
64c10f6e 2247 rc = i7core_get_all_devices();
f4742949
MCC
2248 if (unlikely(rc < 0))
2249 goto fail0;
2250
2251 list_for_each_entry(i7core_dev, &i7core_edac_list, list) {
40557591 2252 count++;
aace4283 2253 rc = i7core_register_mci(i7core_dev);
d4c27795
MCC
2254 if (unlikely(rc < 0))
2255 goto fail1;
d5381642
MCC
2256 }
2257
40557591
MCC
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);
8f331907 2274
66607706 2275 mutex_unlock(&i7core_edac_lock);
a0c36a1f
MCC
2276 return 0;
2277
66607706 2278fail1:
88ef5ea9
MCC
2279 list_for_each_entry(i7core_dev, &i7core_edac_list, list)
2280 i7core_unregister_mci(i7core_dev);
2281
13d6e9b6 2282 i7core_put_all_devices();
66607706
MCC
2283fail0:
2284 mutex_unlock(&i7core_edac_lock);
b7c76151 2285 return rc;
a0c36a1f
MCC
2286}
2287
2288/*
2289 * i7core_remove destructor for one instance of device
2290 *
2291 */
9b3c6e85 2292static void i7core_remove(struct pci_dev *pdev)
a0c36a1f 2293{
64c10f6e 2294 struct i7core_dev *i7core_dev;
a0c36a1f 2295
956b9ba1 2296 edac_dbg(0, "\n");
a0c36a1f 2297
22e6bcbd
MCC
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 */
87d1d272 2305
66607706 2306 mutex_lock(&i7core_edac_lock);
71fe0170
HS
2307
2308 if (unlikely(!probed)) {
2309 mutex_unlock(&i7core_edac_lock);
2310 return;
2311 }
2312
88ef5ea9
MCC
2313 list_for_each_entry(i7core_dev, &i7core_edac_list, list)
2314 i7core_unregister_mci(i7core_dev);
64c10f6e
HS
2315
2316 /* Release PCI resources */
2317 i7core_put_all_devices();
2318
2d95d815
MCC
2319 probed--;
2320
66607706 2321 mutex_unlock(&i7core_edac_lock);
a0c36a1f
MCC
2322}
2323
a0c36a1f
MCC
2324MODULE_DEVICE_TABLE(pci, i7core_pci_tbl);
2325
2326/*
2327 * i7core_driver pci_driver structure for this module
2328 *
2329 */
2330static struct pci_driver i7core_driver = {
2331 .name = "i7core_edac",
2332 .probe = i7core_probe,
9b3c6e85 2333 .remove = i7core_remove,
a0c36a1f
MCC
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 */
2341static int __init i7core_init(void)
2342{
2343 int pci_rc;
2344
956b9ba1 2345 edac_dbg(2, "\n");
a0c36a1f
MCC
2346
2347 /* Ensure that the OPSTATE is set correctly for POLL or NMI */
2348 opstate_init();
2349
54a08ab1
MCC
2350 if (use_pci_fixup)
2351 i7core_xeon_pci_fixup(pci_dev_table);
bc2d7245 2352
a0c36a1f
MCC
2353 pci_rc = pci_register_driver(&i7core_driver);
2354
e35fca47
CG
2355 if (pci_rc >= 0) {
2356 mce_register_decode_chain(&i7_mce_dec);
3ef288a9 2357 return 0;
e35fca47 2358 }
3ef288a9
MCC
2359
2360 i7core_printk(KERN_ERR, "Failed to register device with error %d.\n",
2361 pci_rc);
2362
2363 return pci_rc;
a0c36a1f
MCC
2364}
2365
2366/*
2367 * i7core_exit() Module exit function
2368 * Unregister the driver
2369 */
2370static void __exit i7core_exit(void)
2371{
956b9ba1 2372 edac_dbg(2, "\n");
a0c36a1f 2373 pci_unregister_driver(&i7core_driver);
e35fca47 2374 mce_unregister_decode_chain(&i7_mce_dec);
a0c36a1f
MCC
2375}
2376
2377module_init(i7core_init);
2378module_exit(i7core_exit);
2379
2380MODULE_LICENSE("GPL");
37e59f87 2381MODULE_AUTHOR("Mauro Carvalho Chehab");
a0c36a1f
MCC
2382MODULE_AUTHOR("Red Hat Inc. (http://www.redhat.com)");
2383MODULE_DESCRIPTION("MC Driver for Intel i7 Core memory controllers - "
2384 I7CORE_REVISION);
2385
2386module_param(edac_op_state, int, 0444);
2387MODULE_PARM_DESC(edac_op_state, "EDAC Error Reporting state: 0=Poll,1=NMI");