i7core_edac: Properly mark const static vars as such
[GitHub/LineageOS/android_kernel_samsung_universal7580.git] / drivers / edac / i7core_edac.c
CommitLineData
52707f91
MCC
1/* Intel i7 core/Nehalem Memory Controller kernel module
2 *
3 * This driver supports yhe 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.
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:
a0c36a1f
MCC
12 * Mauro Carvalho Chehab <mchehab@redhat.com>
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>
a0c36a1f
MCC
34#include <linux/edac.h>
35#include <linux/mmzone.h>
d5381642 36#include <linux/edac_mce.h>
f4742949 37#include <linux/smp.h>
14d2c083 38#include <asm/processor.h>
a0c36a1f
MCC
39
40#include "edac_core.h"
41
18c29002
MCC
42/* Static vars */
43static LIST_HEAD(i7core_edac_list);
44static DEFINE_MUTEX(i7core_edac_lock);
45static int probed;
46
f4742949
MCC
47/*
48 * This is used for Nehalem-EP and Nehalem-EX devices, where the non-core
49 * registers start at bus 255, and are not reported by BIOS.
50 * We currently find devices with only 2 sockets. In order to support more QPI
51 * Quick Path Interconnect, just increment this number.
52 */
53#define MAX_SOCKET_BUSES 2
54
55
a0c36a1f
MCC
56/*
57 * Alter this version for the module when modifications are made
58 */
59#define I7CORE_REVISION " Ver: 1.0.0 " __DATE__
60#define EDAC_MOD_STR "i7core_edac"
61
a0c36a1f
MCC
62/*
63 * Debug macros
64 */
65#define i7core_printk(level, fmt, arg...) \
66 edac_printk(level, "i7core", fmt, ##arg)
67
68#define i7core_mc_printk(mci, level, fmt, arg...) \
69 edac_mc_chipset_printk(mci, level, "i7core", fmt, ##arg)
70
71/*
72 * i7core Memory Controller Registers
73 */
74
e9bd2e73
MCC
75 /* OFFSETS for Device 0 Function 0 */
76
77#define MC_CFG_CONTROL 0x90
78
a0c36a1f
MCC
79 /* OFFSETS for Device 3 Function 0 */
80
81#define MC_CONTROL 0x48
82#define MC_STATUS 0x4c
83#define MC_MAX_DOD 0x64
84
442305b1
MCC
85/*
86 * OFFSETS for Device 3 Function 4, as inicated on Xeon 5500 datasheet:
87 * http://www.arrownac.com/manufacturers/intel/s/nehalem/5500-datasheet-v2.pdf
88 */
89
90#define MC_TEST_ERR_RCV1 0x60
91 #define DIMM2_COR_ERR(r) ((r) & 0x7fff)
92
93#define MC_TEST_ERR_RCV0 0x64
94 #define DIMM1_COR_ERR(r) (((r) >> 16) & 0x7fff)
95 #define DIMM0_COR_ERR(r) ((r) & 0x7fff)
96
b4e8f0b6
MCC
97/* OFFSETS for Device 3 Function 2, as inicated on Xeon 5500 datasheet */
98#define MC_COR_ECC_CNT_0 0x80
99#define MC_COR_ECC_CNT_1 0x84
100#define MC_COR_ECC_CNT_2 0x88
101#define MC_COR_ECC_CNT_3 0x8c
102#define MC_COR_ECC_CNT_4 0x90
103#define MC_COR_ECC_CNT_5 0x94
104
105#define DIMM_TOP_COR_ERR(r) (((r) >> 16) & 0x7fff)
106#define DIMM_BOT_COR_ERR(r) ((r) & 0x7fff)
107
108
a0c36a1f
MCC
109 /* OFFSETS for Devices 4,5 and 6 Function 0 */
110
0b2b7b7e
MCC
111#define MC_CHANNEL_DIMM_INIT_PARAMS 0x58
112 #define THREE_DIMMS_PRESENT (1 << 24)
113 #define SINGLE_QUAD_RANK_PRESENT (1 << 23)
114 #define QUAD_RANK_PRESENT (1 << 22)
115 #define REGISTERED_DIMM (1 << 15)
116
f122a892
MCC
117#define MC_CHANNEL_MAPPER 0x60
118 #define RDLCH(r, ch) ((((r) >> (3 + (ch * 6))) & 0x07) - 1)
119 #define WRLCH(r, ch) ((((r) >> (ch * 6)) & 0x07) - 1)
120
0b2b7b7e
MCC
121#define MC_CHANNEL_RANK_PRESENT 0x7c
122 #define RANK_PRESENT_MASK 0xffff
123
a0c36a1f 124#define MC_CHANNEL_ADDR_MATCH 0xf0
194a40fe
MCC
125#define MC_CHANNEL_ERROR_MASK 0xf8
126#define MC_CHANNEL_ERROR_INJECT 0xfc
127 #define INJECT_ADDR_PARITY 0x10
128 #define INJECT_ECC 0x08
129 #define MASK_CACHELINE 0x06
130 #define MASK_FULL_CACHELINE 0x06
131 #define MASK_MSB32_CACHELINE 0x04
132 #define MASK_LSB32_CACHELINE 0x02
133 #define NO_MASK_CACHELINE 0x00
134 #define REPEAT_EN 0x01
a0c36a1f 135
0b2b7b7e 136 /* OFFSETS for Devices 4,5 and 6 Function 1 */
b990538a 137
0b2b7b7e
MCC
138#define MC_DOD_CH_DIMM0 0x48
139#define MC_DOD_CH_DIMM1 0x4c
140#define MC_DOD_CH_DIMM2 0x50
141 #define RANKOFFSET_MASK ((1 << 12) | (1 << 11) | (1 << 10))
142 #define RANKOFFSET(x) ((x & RANKOFFSET_MASK) >> 10)
143 #define DIMM_PRESENT_MASK (1 << 9)
144 #define DIMM_PRESENT(x) (((x) & DIMM_PRESENT_MASK) >> 9)
854d3349
MCC
145 #define MC_DOD_NUMBANK_MASK ((1 << 8) | (1 << 7))
146 #define MC_DOD_NUMBANK(x) (((x) & MC_DOD_NUMBANK_MASK) >> 7)
147 #define MC_DOD_NUMRANK_MASK ((1 << 6) | (1 << 5))
148 #define MC_DOD_NUMRANK(x) (((x) & MC_DOD_NUMRANK_MASK) >> 5)
41fcb7fe 149 #define MC_DOD_NUMROW_MASK ((1 << 4) | (1 << 3) | (1 << 2))
5566cb7c 150 #define MC_DOD_NUMROW(x) (((x) & MC_DOD_NUMROW_MASK) >> 2)
854d3349
MCC
151 #define MC_DOD_NUMCOL_MASK 3
152 #define MC_DOD_NUMCOL(x) ((x) & MC_DOD_NUMCOL_MASK)
0b2b7b7e 153
f122a892
MCC
154#define MC_RANK_PRESENT 0x7c
155
0b2b7b7e
MCC
156#define MC_SAG_CH_0 0x80
157#define MC_SAG_CH_1 0x84
158#define MC_SAG_CH_2 0x88
159#define MC_SAG_CH_3 0x8c
160#define MC_SAG_CH_4 0x90
161#define MC_SAG_CH_5 0x94
162#define MC_SAG_CH_6 0x98
163#define MC_SAG_CH_7 0x9c
164
165#define MC_RIR_LIMIT_CH_0 0x40
166#define MC_RIR_LIMIT_CH_1 0x44
167#define MC_RIR_LIMIT_CH_2 0x48
168#define MC_RIR_LIMIT_CH_3 0x4C
169#define MC_RIR_LIMIT_CH_4 0x50
170#define MC_RIR_LIMIT_CH_5 0x54
171#define MC_RIR_LIMIT_CH_6 0x58
172#define MC_RIR_LIMIT_CH_7 0x5C
173#define MC_RIR_LIMIT_MASK ((1 << 10) - 1)
174
175#define MC_RIR_WAY_CH 0x80
176 #define MC_RIR_WAY_OFFSET_MASK (((1 << 14) - 1) & ~0x7)
177 #define MC_RIR_WAY_RANK_MASK 0x7
178
a0c36a1f
MCC
179/*
180 * i7core structs
181 */
182
183#define NUM_CHANS 3
442305b1
MCC
184#define MAX_DIMMS 3 /* Max DIMMS per channel */
185#define MAX_MCR_FUNC 4
186#define MAX_CHAN_FUNC 3
a0c36a1f
MCC
187
188struct i7core_info {
189 u32 mc_control;
190 u32 mc_status;
191 u32 max_dod;
f122a892 192 u32 ch_map;
a0c36a1f
MCC
193};
194
194a40fe
MCC
195
196struct i7core_inject {
197 int enable;
198
199 u32 section;
200 u32 type;
201 u32 eccmask;
202
203 /* Error address mask */
204 int channel, dimm, rank, bank, page, col;
205};
206
0b2b7b7e 207struct i7core_channel {
442305b1
MCC
208 u32 ranks;
209 u32 dimms;
0b2b7b7e
MCC
210};
211
8f331907 212struct pci_id_descr {
66607706
MCC
213 int dev;
214 int func;
215 int dev_id;
de06eeef 216 int optional;
8f331907
MCC
217};
218
bd9e19ca 219struct pci_id_table {
1288c18f
MCC
220 const struct pci_id_descr *descr;
221 int n_devs;
bd9e19ca
VM
222};
223
f4742949
MCC
224struct i7core_dev {
225 struct list_head list;
226 u8 socket;
227 struct pci_dev **pdev;
de06eeef 228 int n_devs;
f4742949
MCC
229 struct mem_ctl_info *mci;
230};
231
a0c36a1f 232struct i7core_pvt {
f4742949
MCC
233 struct pci_dev *pci_noncore;
234 struct pci_dev *pci_mcr[MAX_MCR_FUNC + 1];
235 struct pci_dev *pci_ch[NUM_CHANS][MAX_CHAN_FUNC + 1];
236
237 struct i7core_dev *i7core_dev;
67166af4 238
a0c36a1f 239 struct i7core_info info;
194a40fe 240 struct i7core_inject inject;
f4742949 241 struct i7core_channel channel[NUM_CHANS];
67166af4 242
f4742949 243 int channels; /* Number of active channels */
442305b1 244
f4742949
MCC
245 int ce_count_available;
246 int csrow_map[NUM_CHANS][MAX_DIMMS];
b4e8f0b6
MCC
247
248 /* ECC corrected errors counts per udimm */
f4742949
MCC
249 unsigned long udimm_ce_count[MAX_DIMMS];
250 int udimm_last_ce_count[MAX_DIMMS];
b4e8f0b6 251 /* ECC corrected errors counts per rdimm */
f4742949
MCC
252 unsigned long rdimm_ce_count[NUM_CHANS][MAX_DIMMS];
253 int rdimm_last_ce_count[NUM_CHANS][MAX_DIMMS];
442305b1 254
f4742949 255 unsigned int is_registered;
14d2c083 256
d5381642
MCC
257 /* mcelog glue */
258 struct edac_mce edac_mce;
ca9c90ba
MCC
259
260 /* Fifo double buffers */
d5381642 261 struct mce mce_entry[MCE_LOG_LEN];
ca9c90ba
MCC
262 struct mce mce_outentry[MCE_LOG_LEN];
263
264 /* Fifo in/out counters */
265 unsigned mce_in, mce_out;
266
267 /* Count indicator to show errors not got */
268 unsigned mce_overrun;
939747bd
MCC
269
270 /* Struct to control EDAC polling */
271 struct edac_pci_ctl_info *i7core_pci;
a0c36a1f
MCC
272};
273
8f331907
MCC
274#define PCI_DESCR(device, function, device_id) \
275 .dev = (device), \
276 .func = (function), \
277 .dev_id = (device_id)
278
1288c18f 279static const struct pci_id_descr pci_dev_descr_i7core_nehalem[] = {
8f331907
MCC
280 /* Memory controller */
281 { PCI_DESCR(3, 0, PCI_DEVICE_ID_INTEL_I7_MCR) },
282 { PCI_DESCR(3, 1, PCI_DEVICE_ID_INTEL_I7_MC_TAD) },
de06eeef
MCC
283 /* Exists only for RDIMM */
284 { PCI_DESCR(3, 2, PCI_DEVICE_ID_INTEL_I7_MC_RAS), .optional = 1 },
8f331907
MCC
285 { PCI_DESCR(3, 4, PCI_DEVICE_ID_INTEL_I7_MC_TEST) },
286
287 /* Channel 0 */
288 { PCI_DESCR(4, 0, PCI_DEVICE_ID_INTEL_I7_MC_CH0_CTRL) },
289 { PCI_DESCR(4, 1, PCI_DEVICE_ID_INTEL_I7_MC_CH0_ADDR) },
290 { PCI_DESCR(4, 2, PCI_DEVICE_ID_INTEL_I7_MC_CH0_RANK) },
291 { PCI_DESCR(4, 3, PCI_DEVICE_ID_INTEL_I7_MC_CH0_TC) },
292
293 /* Channel 1 */
294 { PCI_DESCR(5, 0, PCI_DEVICE_ID_INTEL_I7_MC_CH1_CTRL) },
295 { PCI_DESCR(5, 1, PCI_DEVICE_ID_INTEL_I7_MC_CH1_ADDR) },
296 { PCI_DESCR(5, 2, PCI_DEVICE_ID_INTEL_I7_MC_CH1_RANK) },
297 { PCI_DESCR(5, 3, PCI_DEVICE_ID_INTEL_I7_MC_CH1_TC) },
298
299 /* Channel 2 */
300 { PCI_DESCR(6, 0, PCI_DEVICE_ID_INTEL_I7_MC_CH2_CTRL) },
301 { PCI_DESCR(6, 1, PCI_DEVICE_ID_INTEL_I7_MC_CH2_ADDR) },
302 { PCI_DESCR(6, 2, PCI_DEVICE_ID_INTEL_I7_MC_CH2_RANK) },
303 { PCI_DESCR(6, 3, PCI_DEVICE_ID_INTEL_I7_MC_CH2_TC) },
310cbb72
MCC
304
305 /* Generic Non-core registers */
306 /*
307 * This is the PCI device on i7core and on Xeon 35xx (8086:2c41)
308 * On Xeon 55xx, however, it has a different id (8086:2c40). So,
309 * the probing code needs to test for the other address in case of
310 * failure of this one
311 */
fd382654 312 { PCI_DESCR(0, 0, PCI_DEVICE_ID_INTEL_I7_NONCORE) },
310cbb72 313
a0c36a1f 314};
8f331907 315
1288c18f 316static const struct pci_id_descr pci_dev_descr_lynnfield[] = {
52a2e4fc
MCC
317 { PCI_DESCR( 3, 0, PCI_DEVICE_ID_INTEL_LYNNFIELD_MCR) },
318 { PCI_DESCR( 3, 1, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_TAD) },
319 { PCI_DESCR( 3, 4, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_TEST) },
320
321 { PCI_DESCR( 4, 0, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH0_CTRL) },
322 { PCI_DESCR( 4, 1, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH0_ADDR) },
323 { PCI_DESCR( 4, 2, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH0_RANK) },
324 { PCI_DESCR( 4, 3, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH0_TC) },
325
508fa179
MCC
326 { PCI_DESCR( 5, 0, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH1_CTRL) },
327 { PCI_DESCR( 5, 1, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH1_ADDR) },
328 { PCI_DESCR( 5, 2, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH1_RANK) },
329 { PCI_DESCR( 5, 3, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH1_TC) },
52a2e4fc 330
f05da2f7
MCC
331 /*
332 * This is the PCI device has an alternate address on some
333 * processors like Core i7 860
334 */
52a2e4fc
MCC
335 { PCI_DESCR( 0, 0, PCI_DEVICE_ID_INTEL_LYNNFIELD_NONCORE) },
336};
337
1288c18f 338static const struct pci_id_descr pci_dev_descr_i7core_westmere[] = {
bd9e19ca
VM
339 /* Memory controller */
340 { PCI_DESCR(3, 0, PCI_DEVICE_ID_INTEL_LYNNFIELD_MCR_REV2) },
341 { PCI_DESCR(3, 1, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_TAD_REV2) },
342 /* Exists only for RDIMM */
343 { PCI_DESCR(3, 2, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_RAS_REV2), .optional = 1 },
344 { PCI_DESCR(3, 4, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_TEST_REV2) },
345
346 /* Channel 0 */
347 { PCI_DESCR(4, 0, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH0_CTRL_REV2) },
348 { PCI_DESCR(4, 1, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH0_ADDR_REV2) },
349 { PCI_DESCR(4, 2, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH0_RANK_REV2) },
350 { PCI_DESCR(4, 3, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH0_TC_REV2) },
351
352 /* Channel 1 */
353 { PCI_DESCR(5, 0, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH1_CTRL_REV2) },
354 { PCI_DESCR(5, 1, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH1_ADDR_REV2) },
355 { PCI_DESCR(5, 2, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH1_RANK_REV2) },
356 { PCI_DESCR(5, 3, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH1_TC_REV2) },
357
358 /* Channel 2 */
359 { PCI_DESCR(6, 0, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH2_CTRL_REV2) },
360 { PCI_DESCR(6, 1, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH2_ADDR_REV2) },
361 { PCI_DESCR(6, 2, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH2_RANK_REV2) },
362 { PCI_DESCR(6, 3, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH2_TC_REV2) },
363
364 /* Generic Non-core registers */
365 { PCI_DESCR(0, 0, PCI_DEVICE_ID_INTEL_LYNNFIELD_NONCORE_REV2) },
366
367};
368
1288c18f
MCC
369#define PCI_ID_TABLE_ENTRY(A) { .descr=A, .n_devs = ARRAY_SIZE(A) }
370static const struct pci_id_table pci_dev_table[] = {
bd9e19ca
VM
371 PCI_ID_TABLE_ENTRY(pci_dev_descr_i7core_nehalem),
372 PCI_ID_TABLE_ENTRY(pci_dev_descr_lynnfield),
373 PCI_ID_TABLE_ENTRY(pci_dev_descr_i7core_westmere),
374};
375
8f331907
MCC
376/*
377 * pci_device_id table for which devices we are looking for
8f331907
MCC
378 */
379static const struct pci_device_id i7core_pci_tbl[] __devinitdata = {
d1fd4fb6 380 {PCI_DEVICE(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_X58_HUB_MGMT)},
f05da2f7 381 {PCI_DEVICE(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_LYNNFIELD_QPI_LINK0)},
8f331907
MCC
382 {0,} /* 0 terminated list. */
383};
384
a0c36a1f
MCC
385/****************************************************************************
386 Anciliary status routines
387 ****************************************************************************/
388
389 /* MC_CONTROL bits */
ef708b53
MCC
390#define CH_ACTIVE(pvt, ch) ((pvt)->info.mc_control & (1 << (8 + ch)))
391#define ECCx8(pvt) ((pvt)->info.mc_control & (1 << 1))
a0c36a1f
MCC
392
393 /* MC_STATUS bits */
61053fde 394#define ECC_ENABLED(pvt) ((pvt)->info.mc_status & (1 << 4))
ef708b53 395#define CH_DISABLED(pvt, ch) ((pvt)->info.mc_status & (1 << ch))
a0c36a1f
MCC
396
397 /* MC_MAX_DOD read functions */
854d3349 398static inline int numdimms(u32 dimms)
a0c36a1f 399{
854d3349 400 return (dimms & 0x3) + 1;
a0c36a1f
MCC
401}
402
854d3349 403static inline int numrank(u32 rank)
a0c36a1f
MCC
404{
405 static int ranks[4] = { 1, 2, 4, -EINVAL };
406
854d3349 407 return ranks[rank & 0x3];
a0c36a1f
MCC
408}
409
854d3349 410static inline int numbank(u32 bank)
a0c36a1f
MCC
411{
412 static int banks[4] = { 4, 8, 16, -EINVAL };
413
854d3349 414 return banks[bank & 0x3];
a0c36a1f
MCC
415}
416
854d3349 417static inline int numrow(u32 row)
a0c36a1f
MCC
418{
419 static int rows[8] = {
420 1 << 12, 1 << 13, 1 << 14, 1 << 15,
421 1 << 16, -EINVAL, -EINVAL, -EINVAL,
422 };
423
854d3349 424 return rows[row & 0x7];
a0c36a1f
MCC
425}
426
854d3349 427static inline int numcol(u32 col)
a0c36a1f
MCC
428{
429 static int cols[8] = {
430 1 << 10, 1 << 11, 1 << 12, -EINVAL,
431 };
854d3349 432 return cols[col & 0x3];
a0c36a1f
MCC
433}
434
f4742949 435static struct i7core_dev *get_i7core_dev(u8 socket)
66607706
MCC
436{
437 struct i7core_dev *i7core_dev;
438
439 list_for_each_entry(i7core_dev, &i7core_edac_list, list) {
440 if (i7core_dev->socket == socket)
441 return i7core_dev;
442 }
443
444 return NULL;
445}
446
a0c36a1f
MCC
447/****************************************************************************
448 Memory check routines
449 ****************************************************************************/
67166af4
MCC
450static struct pci_dev *get_pdev_slot_func(u8 socket, unsigned slot,
451 unsigned func)
ef708b53 452{
66607706 453 struct i7core_dev *i7core_dev = get_i7core_dev(socket);
ef708b53 454 int i;
ef708b53 455
66607706
MCC
456 if (!i7core_dev)
457 return NULL;
458
de06eeef 459 for (i = 0; i < i7core_dev->n_devs; i++) {
66607706 460 if (!i7core_dev->pdev[i])
ef708b53
MCC
461 continue;
462
66607706
MCC
463 if (PCI_SLOT(i7core_dev->pdev[i]->devfn) == slot &&
464 PCI_FUNC(i7core_dev->pdev[i]->devfn) == func) {
465 return i7core_dev->pdev[i];
ef708b53
MCC
466 }
467 }
468
eb94fc40
MCC
469 return NULL;
470}
471
ec6df24c
MCC
472/**
473 * i7core_get_active_channels() - gets the number of channels and csrows
474 * @socket: Quick Path Interconnect socket
475 * @channels: Number of channels that will be returned
476 * @csrows: Number of csrows found
477 *
478 * Since EDAC core needs to know in advance the number of available channels
479 * and csrows, in order to allocate memory for csrows/channels, it is needed
480 * to run two similar steps. At the first step, implemented on this function,
481 * it checks the number of csrows/channels present at one socket.
482 * this is used in order to properly allocate the size of mci components.
483 *
484 * It should be noticed that none of the current available datasheets explain
485 * or even mention how csrows are seen by the memory controller. So, we need
486 * to add a fake description for csrows.
487 * So, this driver is attributing one DIMM memory for one csrow.
488 */
1288c18f 489static int i7core_get_active_channels(const u8 socket, unsigned *channels,
67166af4 490 unsigned *csrows)
eb94fc40
MCC
491{
492 struct pci_dev *pdev = NULL;
493 int i, j;
494 u32 status, control;
495
496 *channels = 0;
497 *csrows = 0;
498
67166af4 499 pdev = get_pdev_slot_func(socket, 3, 0);
b7c76151 500 if (!pdev) {
67166af4
MCC
501 i7core_printk(KERN_ERR, "Couldn't find socket %d fn 3.0!!!\n",
502 socket);
ef708b53 503 return -ENODEV;
b7c76151 504 }
ef708b53
MCC
505
506 /* Device 3 function 0 reads */
507 pci_read_config_dword(pdev, MC_STATUS, &status);
508 pci_read_config_dword(pdev, MC_CONTROL, &control);
509
510 for (i = 0; i < NUM_CHANS; i++) {
eb94fc40 511 u32 dimm_dod[3];
ef708b53
MCC
512 /* Check if the channel is active */
513 if (!(control & (1 << (8 + i))))
514 continue;
515
516 /* Check if the channel is disabled */
41fcb7fe 517 if (status & (1 << i))
ef708b53 518 continue;
ef708b53 519
67166af4 520 pdev = get_pdev_slot_func(socket, i + 4, 1);
eb94fc40 521 if (!pdev) {
67166af4
MCC
522 i7core_printk(KERN_ERR, "Couldn't find socket %d "
523 "fn %d.%d!!!\n",
524 socket, i + 4, 1);
eb94fc40
MCC
525 return -ENODEV;
526 }
527 /* Devices 4-6 function 1 */
528 pci_read_config_dword(pdev,
529 MC_DOD_CH_DIMM0, &dimm_dod[0]);
530 pci_read_config_dword(pdev,
531 MC_DOD_CH_DIMM1, &dimm_dod[1]);
532 pci_read_config_dword(pdev,
533 MC_DOD_CH_DIMM2, &dimm_dod[2]);
534
ef708b53 535 (*channels)++;
eb94fc40
MCC
536
537 for (j = 0; j < 3; j++) {
538 if (!DIMM_PRESENT(dimm_dod[j]))
539 continue;
540 (*csrows)++;
541 }
ef708b53
MCC
542 }
543
c77720b9 544 debugf0("Number of active channels on socket %d: %d\n",
67166af4 545 socket, *channels);
1c6fed80 546
ef708b53
MCC
547 return 0;
548}
549
1288c18f 550static int get_dimm_config(const struct mem_ctl_info *mci, int *csrow)
a0c36a1f
MCC
551{
552 struct i7core_pvt *pvt = mci->pvt_info;
1c6fed80 553 struct csrow_info *csr;
854d3349 554 struct pci_dev *pdev;
ba6c5c62 555 int i, j;
5566cb7c 556 unsigned long last_page = 0;
1c6fed80 557 enum edac_type mode;
854d3349 558 enum mem_type mtype;
a0c36a1f 559
854d3349 560 /* Get data from the MC register, function 0 */
f4742949 561 pdev = pvt->pci_mcr[0];
7dd6953c 562 if (!pdev)
8f331907
MCC
563 return -ENODEV;
564
f122a892 565 /* Device 3 function 0 reads */
7dd6953c
MCC
566 pci_read_config_dword(pdev, MC_CONTROL, &pvt->info.mc_control);
567 pci_read_config_dword(pdev, MC_STATUS, &pvt->info.mc_status);
568 pci_read_config_dword(pdev, MC_MAX_DOD, &pvt->info.max_dod);
569 pci_read_config_dword(pdev, MC_CHANNEL_MAPPER, &pvt->info.ch_map);
f122a892 570
17cb7b0c 571 debugf0("QPI %d control=0x%08x status=0x%08x dod=0x%08x map=0x%08x\n",
4af91889 572 pvt->i7core_dev->socket, pvt->info.mc_control, pvt->info.mc_status,
f122a892 573 pvt->info.max_dod, pvt->info.ch_map);
a0c36a1f 574
1c6fed80 575 if (ECC_ENABLED(pvt)) {
41fcb7fe 576 debugf0("ECC enabled with x%d SDCC\n", ECCx8(pvt) ? 8 : 4);
1c6fed80
MCC
577 if (ECCx8(pvt))
578 mode = EDAC_S8ECD8ED;
579 else
580 mode = EDAC_S4ECD4ED;
581 } else {
a0c36a1f 582 debugf0("ECC disabled\n");
1c6fed80
MCC
583 mode = EDAC_NONE;
584 }
a0c36a1f
MCC
585
586 /* FIXME: need to handle the error codes */
17cb7b0c
MCC
587 debugf0("DOD Max limits: DIMMS: %d, %d-ranked, %d-banked "
588 "x%x x 0x%x\n",
854d3349
MCC
589 numdimms(pvt->info.max_dod),
590 numrank(pvt->info.max_dod >> 2),
276b824c 591 numbank(pvt->info.max_dod >> 4),
854d3349
MCC
592 numrow(pvt->info.max_dod >> 6),
593 numcol(pvt->info.max_dod >> 9));
a0c36a1f 594
0b2b7b7e 595 for (i = 0; i < NUM_CHANS; i++) {
854d3349 596 u32 data, dimm_dod[3], value[8];
0b2b7b7e 597
52a2e4fc
MCC
598 if (!pvt->pci_ch[i][0])
599 continue;
600
0b2b7b7e
MCC
601 if (!CH_ACTIVE(pvt, i)) {
602 debugf0("Channel %i is not active\n", i);
603 continue;
604 }
605 if (CH_DISABLED(pvt, i)) {
606 debugf0("Channel %i is disabled\n", i);
607 continue;
608 }
609
f122a892 610 /* Devices 4-6 function 0 */
f4742949 611 pci_read_config_dword(pvt->pci_ch[i][0],
0b2b7b7e
MCC
612 MC_CHANNEL_DIMM_INIT_PARAMS, &data);
613
f4742949 614 pvt->channel[i].ranks = (data & QUAD_RANK_PRESENT) ?
67166af4 615 4 : 2;
0b2b7b7e 616
854d3349
MCC
617 if (data & REGISTERED_DIMM)
618 mtype = MEM_RDDR3;
14d2c083 619 else
854d3349
MCC
620 mtype = MEM_DDR3;
621#if 0
0b2b7b7e
MCC
622 if (data & THREE_DIMMS_PRESENT)
623 pvt->channel[i].dimms = 3;
624 else if (data & SINGLE_QUAD_RANK_PRESENT)
625 pvt->channel[i].dimms = 1;
626 else
627 pvt->channel[i].dimms = 2;
854d3349
MCC
628#endif
629
630 /* Devices 4-6 function 1 */
f4742949 631 pci_read_config_dword(pvt->pci_ch[i][1],
854d3349 632 MC_DOD_CH_DIMM0, &dimm_dod[0]);
f4742949 633 pci_read_config_dword(pvt->pci_ch[i][1],
854d3349 634 MC_DOD_CH_DIMM1, &dimm_dod[1]);
f4742949 635 pci_read_config_dword(pvt->pci_ch[i][1],
854d3349 636 MC_DOD_CH_DIMM2, &dimm_dod[2]);
0b2b7b7e 637
1c6fed80 638 debugf0("Ch%d phy rd%d, wr%d (0x%08x): "
854d3349 639 "%d ranks, %cDIMMs\n",
1c6fed80
MCC
640 i,
641 RDLCH(pvt->info.ch_map, i), WRLCH(pvt->info.ch_map, i),
642 data,
f4742949 643 pvt->channel[i].ranks,
41fcb7fe 644 (data & REGISTERED_DIMM) ? 'R' : 'U');
854d3349
MCC
645
646 for (j = 0; j < 3; j++) {
647 u32 banks, ranks, rows, cols;
5566cb7c 648 u32 size, npages;
854d3349
MCC
649
650 if (!DIMM_PRESENT(dimm_dod[j]))
651 continue;
652
653 banks = numbank(MC_DOD_NUMBANK(dimm_dod[j]));
654 ranks = numrank(MC_DOD_NUMRANK(dimm_dod[j]));
655 rows = numrow(MC_DOD_NUMROW(dimm_dod[j]));
656 cols = numcol(MC_DOD_NUMCOL(dimm_dod[j]));
657
5566cb7c
MCC
658 /* DDR3 has 8 I/O banks */
659 size = (rows * cols * banks * ranks) >> (20 - 3);
660
f4742949 661 pvt->channel[i].dimms++;
854d3349 662
17cb7b0c
MCC
663 debugf0("\tdimm %d %d Mb offset: %x, "
664 "bank: %d, rank: %d, row: %#x, col: %#x\n",
665 j, size,
854d3349
MCC
666 RANKOFFSET(dimm_dod[j]),
667 banks, ranks, rows, cols);
668
eb94fc40
MCC
669#if PAGE_SHIFT > 20
670 npages = size >> (PAGE_SHIFT - 20);
671#else
672 npages = size << (20 - PAGE_SHIFT);
673#endif
5566cb7c 674
ba6c5c62 675 csr = &mci->csrows[*csrow];
5566cb7c
MCC
676 csr->first_page = last_page + 1;
677 last_page += npages;
678 csr->last_page = last_page;
679 csr->nr_pages = npages;
680
854d3349 681 csr->page_mask = 0;
eb94fc40 682 csr->grain = 8;
ba6c5c62 683 csr->csrow_idx = *csrow;
eb94fc40
MCC
684 csr->nr_channels = 1;
685
686 csr->channels[0].chan_idx = i;
687 csr->channels[0].ce_count = 0;
854d3349 688
f4742949 689 pvt->csrow_map[i][j] = *csrow;
b4e8f0b6 690
854d3349
MCC
691 switch (banks) {
692 case 4:
693 csr->dtype = DEV_X4;
694 break;
695 case 8:
696 csr->dtype = DEV_X8;
697 break;
698 case 16:
699 csr->dtype = DEV_X16;
700 break;
701 default:
702 csr->dtype = DEV_UNKNOWN;
703 }
704
705 csr->edac_mode = mode;
706 csr->mtype = mtype;
707
ba6c5c62 708 (*csrow)++;
854d3349 709 }
1c6fed80 710
854d3349
MCC
711 pci_read_config_dword(pdev, MC_SAG_CH_0, &value[0]);
712 pci_read_config_dword(pdev, MC_SAG_CH_1, &value[1]);
713 pci_read_config_dword(pdev, MC_SAG_CH_2, &value[2]);
714 pci_read_config_dword(pdev, MC_SAG_CH_3, &value[3]);
715 pci_read_config_dword(pdev, MC_SAG_CH_4, &value[4]);
716 pci_read_config_dword(pdev, MC_SAG_CH_5, &value[5]);
717 pci_read_config_dword(pdev, MC_SAG_CH_6, &value[6]);
718 pci_read_config_dword(pdev, MC_SAG_CH_7, &value[7]);
17cb7b0c 719 debugf1("\t[%i] DIVBY3\tREMOVED\tOFFSET\n", i);
854d3349 720 for (j = 0; j < 8; j++)
17cb7b0c 721 debugf1("\t\t%#x\t%#x\t%#x\n",
854d3349
MCC
722 (value[j] >> 27) & 0x1,
723 (value[j] >> 24) & 0x7,
724 (value[j] && ((1 << 24) - 1)));
0b2b7b7e
MCC
725 }
726
a0c36a1f
MCC
727 return 0;
728}
729
194a40fe
MCC
730/****************************************************************************
731 Error insertion routines
732 ****************************************************************************/
733
734/* The i7core has independent error injection features per channel.
735 However, to have a simpler code, we don't allow enabling error injection
736 on more than one channel.
737 Also, since a change at an inject parameter will be applied only at enable,
738 we're disabling error injection on all write calls to the sysfs nodes that
739 controls the error code injection.
740 */
1288c18f 741static int disable_inject(const struct mem_ctl_info *mci)
194a40fe
MCC
742{
743 struct i7core_pvt *pvt = mci->pvt_info;
744
745 pvt->inject.enable = 0;
746
f4742949 747 if (!pvt->pci_ch[pvt->inject.channel][0])
8f331907
MCC
748 return -ENODEV;
749
f4742949 750 pci_write_config_dword(pvt->pci_ch[pvt->inject.channel][0],
4157d9f5 751 MC_CHANNEL_ERROR_INJECT, 0);
8f331907
MCC
752
753 return 0;
194a40fe
MCC
754}
755
756/*
757 * i7core inject inject.section
758 *
759 * accept and store error injection inject.section value
760 * bit 0 - refers to the lower 32-byte half cacheline
761 * bit 1 - refers to the upper 32-byte half cacheline
762 */
763static ssize_t i7core_inject_section_store(struct mem_ctl_info *mci,
764 const char *data, size_t count)
765{
766 struct i7core_pvt *pvt = mci->pvt_info;
767 unsigned long value;
768 int rc;
769
770 if (pvt->inject.enable)
41fcb7fe 771 disable_inject(mci);
194a40fe
MCC
772
773 rc = strict_strtoul(data, 10, &value);
774 if ((rc < 0) || (value > 3))
2068def5 775 return -EIO;
194a40fe
MCC
776
777 pvt->inject.section = (u32) value;
778 return count;
779}
780
781static ssize_t i7core_inject_section_show(struct mem_ctl_info *mci,
782 char *data)
783{
784 struct i7core_pvt *pvt = mci->pvt_info;
785 return sprintf(data, "0x%08x\n", pvt->inject.section);
786}
787
788/*
789 * i7core inject.type
790 *
791 * accept and store error injection inject.section value
792 * bit 0 - repeat enable - Enable error repetition
793 * bit 1 - inject ECC error
794 * bit 2 - inject parity error
795 */
796static ssize_t i7core_inject_type_store(struct mem_ctl_info *mci,
797 const char *data, size_t count)
798{
799 struct i7core_pvt *pvt = mci->pvt_info;
800 unsigned long value;
801 int rc;
802
803 if (pvt->inject.enable)
41fcb7fe 804 disable_inject(mci);
194a40fe
MCC
805
806 rc = strict_strtoul(data, 10, &value);
807 if ((rc < 0) || (value > 7))
2068def5 808 return -EIO;
194a40fe
MCC
809
810 pvt->inject.type = (u32) value;
811 return count;
812}
813
814static ssize_t i7core_inject_type_show(struct mem_ctl_info *mci,
815 char *data)
816{
817 struct i7core_pvt *pvt = mci->pvt_info;
818 return sprintf(data, "0x%08x\n", pvt->inject.type);
819}
820
821/*
822 * i7core_inject_inject.eccmask_store
823 *
824 * The type of error (UE/CE) will depend on the inject.eccmask value:
825 * Any bits set to a 1 will flip the corresponding ECC bit
826 * Correctable errors can be injected by flipping 1 bit or the bits within
827 * a symbol pair (2 consecutive aligned 8-bit pairs - i.e. 7:0 and 15:8 or
828 * 23:16 and 31:24). Flipping bits in two symbol pairs will cause an
829 * uncorrectable error to be injected.
830 */
831static ssize_t i7core_inject_eccmask_store(struct mem_ctl_info *mci,
832 const char *data, size_t count)
833{
834 struct i7core_pvt *pvt = mci->pvt_info;
835 unsigned long value;
836 int rc;
837
838 if (pvt->inject.enable)
41fcb7fe 839 disable_inject(mci);
194a40fe
MCC
840
841 rc = strict_strtoul(data, 10, &value);
842 if (rc < 0)
2068def5 843 return -EIO;
194a40fe
MCC
844
845 pvt->inject.eccmask = (u32) value;
846 return count;
847}
848
849static ssize_t i7core_inject_eccmask_show(struct mem_ctl_info *mci,
850 char *data)
851{
852 struct i7core_pvt *pvt = mci->pvt_info;
853 return sprintf(data, "0x%08x\n", pvt->inject.eccmask);
854}
855
856/*
857 * i7core_addrmatch
858 *
859 * The type of error (UE/CE) will depend on the inject.eccmask value:
860 * Any bits set to a 1 will flip the corresponding ECC bit
861 * Correctable errors can be injected by flipping 1 bit or the bits within
862 * a symbol pair (2 consecutive aligned 8-bit pairs - i.e. 7:0 and 15:8 or
863 * 23:16 and 31:24). Flipping bits in two symbol pairs will cause an
864 * uncorrectable error to be injected.
865 */
194a40fe 866
a5538e53
MCC
867#define DECLARE_ADDR_MATCH(param, limit) \
868static ssize_t i7core_inject_store_##param( \
869 struct mem_ctl_info *mci, \
870 const char *data, size_t count) \
871{ \
cc301b3a 872 struct i7core_pvt *pvt; \
a5538e53
MCC
873 long value; \
874 int rc; \
875 \
cc301b3a
MCC
876 debugf1("%s()\n", __func__); \
877 pvt = mci->pvt_info; \
878 \
a5538e53
MCC
879 if (pvt->inject.enable) \
880 disable_inject(mci); \
881 \
4f87fad1 882 if (!strcasecmp(data, "any") || !strcasecmp(data, "any\n"))\
a5538e53
MCC
883 value = -1; \
884 else { \
885 rc = strict_strtoul(data, 10, &value); \
886 if ((rc < 0) || (value >= limit)) \
887 return -EIO; \
888 } \
889 \
890 pvt->inject.param = value; \
891 \
892 return count; \
893} \
894 \
895static ssize_t i7core_inject_show_##param( \
896 struct mem_ctl_info *mci, \
897 char *data) \
898{ \
cc301b3a
MCC
899 struct i7core_pvt *pvt; \
900 \
901 pvt = mci->pvt_info; \
902 debugf1("%s() pvt=%p\n", __func__, pvt); \
a5538e53
MCC
903 if (pvt->inject.param < 0) \
904 return sprintf(data, "any\n"); \
905 else \
906 return sprintf(data, "%d\n", pvt->inject.param);\
194a40fe
MCC
907}
908
a5538e53
MCC
909#define ATTR_ADDR_MATCH(param) \
910 { \
911 .attr = { \
912 .name = #param, \
913 .mode = (S_IRUGO | S_IWUSR) \
914 }, \
915 .show = i7core_inject_show_##param, \
916 .store = i7core_inject_store_##param, \
917 }
194a40fe 918
a5538e53
MCC
919DECLARE_ADDR_MATCH(channel, 3);
920DECLARE_ADDR_MATCH(dimm, 3);
921DECLARE_ADDR_MATCH(rank, 4);
922DECLARE_ADDR_MATCH(bank, 32);
923DECLARE_ADDR_MATCH(page, 0x10000);
924DECLARE_ADDR_MATCH(col, 0x4000);
194a40fe 925
1288c18f 926static int write_and_test(struct pci_dev *dev, const int where, const u32 val)
276b824c
MCC
927{
928 u32 read;
929 int count;
930
4157d9f5
MCC
931 debugf0("setting pci %02x:%02x.%x reg=%02x value=%08x\n",
932 dev->bus->number, PCI_SLOT(dev->devfn), PCI_FUNC(dev->devfn),
933 where, val);
934
276b824c
MCC
935 for (count = 0; count < 10; count++) {
936 if (count)
b990538a 937 msleep(100);
276b824c
MCC
938 pci_write_config_dword(dev, where, val);
939 pci_read_config_dword(dev, where, &read);
940
941 if (read == val)
942 return 0;
943 }
944
4157d9f5
MCC
945 i7core_printk(KERN_ERR, "Error during set pci %02x:%02x.%x reg=%02x "
946 "write=%08x. Read=%08x\n",
947 dev->bus->number, PCI_SLOT(dev->devfn), PCI_FUNC(dev->devfn),
948 where, val, read);
276b824c
MCC
949
950 return -EINVAL;
951}
952
194a40fe
MCC
953/*
954 * This routine prepares the Memory Controller for error injection.
955 * The error will be injected when some process tries to write to the
956 * memory that matches the given criteria.
957 * The criteria can be set in terms of a mask where dimm, rank, bank, page
958 * and col can be specified.
959 * A -1 value for any of the mask items will make the MCU to ignore
960 * that matching criteria for error injection.
961 *
962 * It should be noticed that the error will only happen after a write operation
963 * on a memory that matches the condition. if REPEAT_EN is not enabled at
964 * inject mask, then it will produce just one error. Otherwise, it will repeat
965 * until the injectmask would be cleaned.
966 *
967 * FIXME: This routine assumes that MAXNUMDIMMS value of MC_MAX_DOD
968 * is reliable enough to check if the MC is using the
969 * three channels. However, this is not clear at the datasheet.
970 */
971static ssize_t i7core_inject_enable_store(struct mem_ctl_info *mci,
972 const char *data, size_t count)
973{
974 struct i7core_pvt *pvt = mci->pvt_info;
975 u32 injectmask;
976 u64 mask = 0;
977 int rc;
978 long enable;
979
f4742949 980 if (!pvt->pci_ch[pvt->inject.channel][0])
8f331907
MCC
981 return 0;
982
194a40fe
MCC
983 rc = strict_strtoul(data, 10, &enable);
984 if ((rc < 0))
985 return 0;
986
987 if (enable) {
988 pvt->inject.enable = 1;
989 } else {
990 disable_inject(mci);
991 return count;
992 }
993
994 /* Sets pvt->inject.dimm mask */
995 if (pvt->inject.dimm < 0)
486dd09f 996 mask |= 1LL << 41;
194a40fe 997 else {
f4742949 998 if (pvt->channel[pvt->inject.channel].dimms > 2)
486dd09f 999 mask |= (pvt->inject.dimm & 0x3LL) << 35;
194a40fe 1000 else
486dd09f 1001 mask |= (pvt->inject.dimm & 0x1LL) << 36;
194a40fe
MCC
1002 }
1003
1004 /* Sets pvt->inject.rank mask */
1005 if (pvt->inject.rank < 0)
486dd09f 1006 mask |= 1LL << 40;
194a40fe 1007 else {
f4742949 1008 if (pvt->channel[pvt->inject.channel].dimms > 2)
486dd09f 1009 mask |= (pvt->inject.rank & 0x1LL) << 34;
194a40fe 1010 else
486dd09f 1011 mask |= (pvt->inject.rank & 0x3LL) << 34;
194a40fe
MCC
1012 }
1013
1014 /* Sets pvt->inject.bank mask */
1015 if (pvt->inject.bank < 0)
486dd09f 1016 mask |= 1LL << 39;
194a40fe 1017 else
486dd09f 1018 mask |= (pvt->inject.bank & 0x15LL) << 30;
194a40fe
MCC
1019
1020 /* Sets pvt->inject.page mask */
1021 if (pvt->inject.page < 0)
486dd09f 1022 mask |= 1LL << 38;
194a40fe 1023 else
486dd09f 1024 mask |= (pvt->inject.page & 0xffff) << 14;
194a40fe
MCC
1025
1026 /* Sets pvt->inject.column mask */
1027 if (pvt->inject.col < 0)
486dd09f 1028 mask |= 1LL << 37;
194a40fe 1029 else
486dd09f 1030 mask |= (pvt->inject.col & 0x3fff);
194a40fe 1031
276b824c
MCC
1032 /*
1033 * bit 0: REPEAT_EN
1034 * bits 1-2: MASK_HALF_CACHELINE
1035 * bit 3: INJECT_ECC
1036 * bit 4: INJECT_ADDR_PARITY
1037 */
1038
1039 injectmask = (pvt->inject.type & 1) |
1040 (pvt->inject.section & 0x3) << 1 |
1041 (pvt->inject.type & 0x6) << (3 - 1);
1042
1043 /* Unlock writes to registers - this register is write only */
f4742949 1044 pci_write_config_dword(pvt->pci_noncore,
67166af4 1045 MC_CFG_CONTROL, 0x2);
e9bd2e73 1046
f4742949 1047 write_and_test(pvt->pci_ch[pvt->inject.channel][0],
194a40fe 1048 MC_CHANNEL_ADDR_MATCH, mask);
f4742949 1049 write_and_test(pvt->pci_ch[pvt->inject.channel][0],
7b029d03 1050 MC_CHANNEL_ADDR_MATCH + 4, mask >> 32L);
7b029d03 1051
f4742949 1052 write_and_test(pvt->pci_ch[pvt->inject.channel][0],
194a40fe
MCC
1053 MC_CHANNEL_ERROR_MASK, pvt->inject.eccmask);
1054
f4742949 1055 write_and_test(pvt->pci_ch[pvt->inject.channel][0],
4157d9f5 1056 MC_CHANNEL_ERROR_INJECT, injectmask);
276b824c 1057
194a40fe 1058 /*
276b824c
MCC
1059 * This is something undocumented, based on my tests
1060 * Without writing 8 to this register, errors aren't injected. Not sure
1061 * why.
194a40fe 1062 */
f4742949 1063 pci_write_config_dword(pvt->pci_noncore,
276b824c 1064 MC_CFG_CONTROL, 8);
194a40fe 1065
41fcb7fe
MCC
1066 debugf0("Error inject addr match 0x%016llx, ecc 0x%08x,"
1067 " inject 0x%08x\n",
194a40fe
MCC
1068 mask, pvt->inject.eccmask, injectmask);
1069
7b029d03 1070
194a40fe
MCC
1071 return count;
1072}
1073
1074static ssize_t i7core_inject_enable_show(struct mem_ctl_info *mci,
1075 char *data)
1076{
1077 struct i7core_pvt *pvt = mci->pvt_info;
7b029d03
MCC
1078 u32 injectmask;
1079
52a2e4fc
MCC
1080 if (!pvt->pci_ch[pvt->inject.channel][0])
1081 return 0;
1082
f4742949 1083 pci_read_config_dword(pvt->pci_ch[pvt->inject.channel][0],
4157d9f5 1084 MC_CHANNEL_ERROR_INJECT, &injectmask);
7b029d03
MCC
1085
1086 debugf0("Inject error read: 0x%018x\n", injectmask);
1087
1088 if (injectmask & 0x0c)
1089 pvt->inject.enable = 1;
1090
194a40fe
MCC
1091 return sprintf(data, "%d\n", pvt->inject.enable);
1092}
1093
f338d736
MCC
1094#define DECLARE_COUNTER(param) \
1095static ssize_t i7core_show_counter_##param( \
1096 struct mem_ctl_info *mci, \
1097 char *data) \
1098{ \
1099 struct i7core_pvt *pvt = mci->pvt_info; \
1100 \
1101 debugf1("%s() \n", __func__); \
1102 if (!pvt->ce_count_available || (pvt->is_registered)) \
1103 return sprintf(data, "data unavailable\n"); \
1104 return sprintf(data, "%lu\n", \
1105 pvt->udimm_ce_count[param]); \
1106}
442305b1 1107
f338d736
MCC
1108#define ATTR_COUNTER(param) \
1109 { \
1110 .attr = { \
1111 .name = __stringify(udimm##param), \
1112 .mode = (S_IRUGO | S_IWUSR) \
1113 }, \
1114 .show = i7core_show_counter_##param \
d88b8507 1115 }
442305b1 1116
f338d736
MCC
1117DECLARE_COUNTER(0);
1118DECLARE_COUNTER(1);
1119DECLARE_COUNTER(2);
442305b1 1120
194a40fe
MCC
1121/*
1122 * Sysfs struct
1123 */
a5538e53 1124
1288c18f 1125static const struct mcidev_sysfs_attribute i7core_addrmatch_attrs[] = {
a5538e53
MCC
1126 ATTR_ADDR_MATCH(channel),
1127 ATTR_ADDR_MATCH(dimm),
1128 ATTR_ADDR_MATCH(rank),
1129 ATTR_ADDR_MATCH(bank),
1130 ATTR_ADDR_MATCH(page),
1131 ATTR_ADDR_MATCH(col),
1288c18f 1132 { } /* End of list */
a5538e53
MCC
1133};
1134
1288c18f 1135static const struct mcidev_sysfs_group i7core_inject_addrmatch = {
a5538e53
MCC
1136 .name = "inject_addrmatch",
1137 .mcidev_attr = i7core_addrmatch_attrs,
1138};
1139
1288c18f 1140static const struct mcidev_sysfs_attribute i7core_udimm_counters_attrs[] = {
f338d736
MCC
1141 ATTR_COUNTER(0),
1142 ATTR_COUNTER(1),
1143 ATTR_COUNTER(2),
64aab720 1144 { .attr = { .name = NULL } }
f338d736
MCC
1145};
1146
1288c18f 1147static const struct mcidev_sysfs_group i7core_udimm_counters = {
f338d736
MCC
1148 .name = "all_channel_counts",
1149 .mcidev_attr = i7core_udimm_counters_attrs,
1150};
1151
1288c18f 1152static const struct mcidev_sysfs_attribute i7core_sysfs_rdimm_attrs[] = {
194a40fe
MCC
1153 {
1154 .attr = {
1155 .name = "inject_section",
1156 .mode = (S_IRUGO | S_IWUSR)
1157 },
1158 .show = i7core_inject_section_show,
1159 .store = i7core_inject_section_store,
1160 }, {
1161 .attr = {
1162 .name = "inject_type",
1163 .mode = (S_IRUGO | S_IWUSR)
1164 },
1165 .show = i7core_inject_type_show,
1166 .store = i7core_inject_type_store,
1167 }, {
1168 .attr = {
1169 .name = "inject_eccmask",
1170 .mode = (S_IRUGO | S_IWUSR)
1171 },
1172 .show = i7core_inject_eccmask_show,
1173 .store = i7core_inject_eccmask_store,
1174 }, {
a5538e53 1175 .grp = &i7core_inject_addrmatch,
194a40fe
MCC
1176 }, {
1177 .attr = {
1178 .name = "inject_enable",
1179 .mode = (S_IRUGO | S_IWUSR)
1180 },
1181 .show = i7core_inject_enable_show,
1182 .store = i7core_inject_enable_store,
1183 },
1288c18f
MCC
1184 { } /* End of list */
1185};
1186
1187static const struct mcidev_sysfs_attribute i7core_sysfs_udimm_attrs[] = {
1188 {
1189 .attr = {
1190 .name = "inject_section",
1191 .mode = (S_IRUGO | S_IWUSR)
1192 },
1193 .show = i7core_inject_section_show,
1194 .store = i7core_inject_section_store,
1195 }, {
1196 .attr = {
1197 .name = "inject_type",
1198 .mode = (S_IRUGO | S_IWUSR)
1199 },
1200 .show = i7core_inject_type_show,
1201 .store = i7core_inject_type_store,
1202 }, {
1203 .attr = {
1204 .name = "inject_eccmask",
1205 .mode = (S_IRUGO | S_IWUSR)
1206 },
1207 .show = i7core_inject_eccmask_show,
1208 .store = i7core_inject_eccmask_store,
1209 }, {
1210 .grp = &i7core_inject_addrmatch,
1211 }, {
1212 .attr = {
1213 .name = "inject_enable",
1214 .mode = (S_IRUGO | S_IWUSR)
1215 },
1216 .show = i7core_inject_enable_show,
1217 .store = i7core_inject_enable_store,
1218 }, {
1219 .grp = &i7core_udimm_counters,
1220 },
1221 { } /* End of list */
194a40fe
MCC
1222};
1223
a0c36a1f
MCC
1224/****************************************************************************
1225 Device initialization routines: put/get, init/exit
1226 ****************************************************************************/
1227
1228/*
1229 * i7core_put_devices 'put' all the devices that we have
1230 * reserved via 'get'
1231 */
13d6e9b6 1232static void i7core_put_devices(struct i7core_dev *i7core_dev)
a0c36a1f 1233{
13d6e9b6 1234 int i;
a0c36a1f 1235
22e6bcbd 1236 debugf0(__FILE__ ": %s()\n", __func__);
de06eeef 1237 for (i = 0; i < i7core_dev->n_devs; i++) {
22e6bcbd
MCC
1238 struct pci_dev *pdev = i7core_dev->pdev[i];
1239 if (!pdev)
1240 continue;
1241 debugf0("Removing dev %02x:%02x.%d\n",
1242 pdev->bus->number,
1243 PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn));
1244 pci_dev_put(pdev);
1245 }
13d6e9b6 1246 kfree(i7core_dev->pdev);
22e6bcbd 1247 list_del(&i7core_dev->list);
13d6e9b6
MCC
1248 kfree(i7core_dev);
1249}
66607706 1250
13d6e9b6
MCC
1251static void i7core_put_all_devices(void)
1252{
42538680 1253 struct i7core_dev *i7core_dev, *tmp;
13d6e9b6 1254
42538680 1255 list_for_each_entry_safe(i7core_dev, tmp, &i7core_edac_list, list)
13d6e9b6 1256 i7core_put_devices(i7core_dev);
a0c36a1f
MCC
1257}
1258
1288c18f 1259static void __init i7core_xeon_pci_fixup(const struct pci_id_table *table)
bc2d7245
KM
1260{
1261 struct pci_dev *pdev = NULL;
1262 int i;
1263 /*
1264 * On Xeon 55xx, the Intel Quckpath Arch Generic Non-core pci buses
1265 * aren't announced by acpi. So, we need to use a legacy scan probing
1266 * to detect them
1267 */
bd9e19ca
VM
1268 while (table && table->descr) {
1269 pdev = pci_get_device(PCI_VENDOR_ID_INTEL, table->descr[0].dev_id, NULL);
1270 if (unlikely(!pdev)) {
1271 for (i = 0; i < MAX_SOCKET_BUSES; i++)
1272 pcibios_scan_specific_bus(255-i);
1273 }
bda14289 1274 pci_dev_put(pdev);
bd9e19ca 1275 table++;
bc2d7245
KM
1276 }
1277}
1278
bda14289
MCC
1279static unsigned i7core_pci_lastbus(void)
1280{
1281 int last_bus = 0, bus;
1282 struct pci_bus *b = NULL;
1283
1284 while ((b = pci_find_next_bus(b)) != NULL) {
1285 bus = b->number;
1286 debugf0("Found bus %d\n", bus);
1287 if (bus > last_bus)
1288 last_bus = bus;
1289 }
1290
1291 debugf0("Last bus %d\n", last_bus);
1292
1293 return last_bus;
1294}
1295
a0c36a1f
MCC
1296/*
1297 * i7core_get_devices Find and perform 'get' operation on the MCH's
1298 * device/functions we want to reference for this driver
1299 *
1300 * Need to 'get' device 16 func 1 and func 2
1301 */
1288c18f
MCC
1302int i7core_get_onedevice(struct pci_dev **prev, const int devno,
1303 const struct pci_id_descr *dev_descr,
1304 const unsigned n_devs,
1305 const unsigned last_bus)
a0c36a1f 1306{
66607706
MCC
1307 struct i7core_dev *i7core_dev;
1308
8f331907 1309 struct pci_dev *pdev = NULL;
67166af4
MCC
1310 u8 bus = 0;
1311 u8 socket = 0;
a0c36a1f 1312
c77720b9 1313 pdev = pci_get_device(PCI_VENDOR_ID_INTEL,
de06eeef 1314 dev_descr->dev_id, *prev);
c77720b9 1315
c77720b9
MCC
1316 /*
1317 * On Xeon 55xx, the Intel Quckpath Arch Generic Non-core regs
1318 * is at addr 8086:2c40, instead of 8086:2c41. So, we need
1319 * to probe for the alternate address in case of failure
1320 */
de06eeef 1321 if (dev_descr->dev_id == PCI_DEVICE_ID_INTEL_I7_NONCORE && !pdev)
c77720b9 1322 pdev = pci_get_device(PCI_VENDOR_ID_INTEL,
fd382654 1323 PCI_DEVICE_ID_INTEL_I7_NONCORE_ALT, *prev);
d1fd4fb6 1324
bd9e19ca 1325 if (dev_descr->dev_id == PCI_DEVICE_ID_INTEL_LYNNFIELD_NONCORE && !pdev)
f05da2f7
MCC
1326 pdev = pci_get_device(PCI_VENDOR_ID_INTEL,
1327 PCI_DEVICE_ID_INTEL_LYNNFIELD_NONCORE_ALT,
1328 *prev);
1329
c77720b9
MCC
1330 if (!pdev) {
1331 if (*prev) {
1332 *prev = pdev;
1333 return 0;
d1fd4fb6
MCC
1334 }
1335
de06eeef 1336 if (dev_descr->optional)
c77720b9 1337 return 0;
310cbb72 1338
bd9e19ca
VM
1339 if (devno == 0)
1340 return -ENODEV;
1341
ab089374 1342 i7core_printk(KERN_INFO,
c77720b9 1343 "Device not found: dev %02x.%d PCI ID %04x:%04x\n",
de06eeef
MCC
1344 dev_descr->dev, dev_descr->func,
1345 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
67166af4 1346
c77720b9
MCC
1347 /* End of list, leave */
1348 return -ENODEV;
1349 }
1350 bus = pdev->bus->number;
67166af4 1351
bda14289 1352 socket = last_bus - bus;
c77720b9 1353
66607706
MCC
1354 i7core_dev = get_i7core_dev(socket);
1355 if (!i7core_dev) {
1356 i7core_dev = kzalloc(sizeof(*i7core_dev), GFP_KERNEL);
1357 if (!i7core_dev)
1358 return -ENOMEM;
de06eeef 1359 i7core_dev->pdev = kzalloc(sizeof(*i7core_dev->pdev) * n_devs,
66607706 1360 GFP_KERNEL);
2a6fae32
AB
1361 if (!i7core_dev->pdev) {
1362 kfree(i7core_dev);
66607706 1363 return -ENOMEM;
2a6fae32 1364 }
66607706 1365 i7core_dev->socket = socket;
de06eeef 1366 i7core_dev->n_devs = n_devs;
66607706 1367 list_add_tail(&i7core_dev->list, &i7core_edac_list);
c77720b9 1368 }
67166af4 1369
66607706 1370 if (i7core_dev->pdev[devno]) {
c77720b9
MCC
1371 i7core_printk(KERN_ERR,
1372 "Duplicated device for "
1373 "dev %02x:%02x.%d PCI ID %04x:%04x\n",
de06eeef
MCC
1374 bus, dev_descr->dev, dev_descr->func,
1375 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
c77720b9
MCC
1376 pci_dev_put(pdev);
1377 return -ENODEV;
1378 }
67166af4 1379
66607706 1380 i7core_dev->pdev[devno] = pdev;
c77720b9
MCC
1381
1382 /* Sanity check */
de06eeef
MCC
1383 if (unlikely(PCI_SLOT(pdev->devfn) != dev_descr->dev ||
1384 PCI_FUNC(pdev->devfn) != dev_descr->func)) {
c77720b9
MCC
1385 i7core_printk(KERN_ERR,
1386 "Device PCI ID %04x:%04x "
1387 "has dev %02x:%02x.%d instead of dev %02x:%02x.%d\n",
de06eeef 1388 PCI_VENDOR_ID_INTEL, dev_descr->dev_id,
c77720b9 1389 bus, PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn),
de06eeef 1390 bus, dev_descr->dev, dev_descr->func);
c77720b9
MCC
1391 return -ENODEV;
1392 }
ef708b53 1393
c77720b9
MCC
1394 /* Be sure that the device is enabled */
1395 if (unlikely(pci_enable_device(pdev) < 0)) {
1396 i7core_printk(KERN_ERR,
1397 "Couldn't enable "
1398 "dev %02x:%02x.%d PCI ID %04x:%04x\n",
de06eeef
MCC
1399 bus, dev_descr->dev, dev_descr->func,
1400 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
c77720b9
MCC
1401 return -ENODEV;
1402 }
ef708b53 1403
d4c27795 1404 debugf0("Detected socket %d dev %02x:%02x.%d PCI ID %04x:%04x\n",
de06eeef
MCC
1405 socket, bus, dev_descr->dev,
1406 dev_descr->func,
1407 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
8f331907 1408
c77720b9 1409 *prev = pdev;
ef708b53 1410
c77720b9
MCC
1411 return 0;
1412}
a0c36a1f 1413
1288c18f 1414static int i7core_get_devices(const struct pci_id_table *table)
c77720b9 1415{
bda14289 1416 int i, rc, last_bus;
c77720b9 1417 struct pci_dev *pdev = NULL;
1288c18f 1418 const struct pci_id_descr *dev_descr;
bd9e19ca 1419
bda14289
MCC
1420 last_bus = i7core_pci_lastbus();
1421
bd9e19ca
VM
1422 while (table && table->descr) {
1423 dev_descr = table->descr;
1424 for (i = 0; i < table->n_devs; i++) {
1425 pdev = NULL;
1426 do {
bda14289
MCC
1427 rc = i7core_get_onedevice(&pdev, i,
1428 &dev_descr[i],
1429 table->n_devs,
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 }
1441 table++;
c77720b9 1442 }
66607706 1443
ef708b53 1444 return 0;
bd9e19ca 1445 return 0;
ef708b53
MCC
1446}
1447
f4742949
MCC
1448static int mci_bind_devs(struct mem_ctl_info *mci,
1449 struct i7core_dev *i7core_dev)
ef708b53
MCC
1450{
1451 struct i7core_pvt *pvt = mci->pvt_info;
1452 struct pci_dev *pdev;
f4742949 1453 int i, func, slot;
ef708b53 1454
f4742949
MCC
1455 /* Associates i7core_dev and mci for future usage */
1456 pvt->i7core_dev = i7core_dev;
1457 i7core_dev->mci = mci;
66607706 1458
f4742949 1459 pvt->is_registered = 0;
de06eeef 1460 for (i = 0; i < i7core_dev->n_devs; i++) {
f4742949
MCC
1461 pdev = i7core_dev->pdev[i];
1462 if (!pdev)
66607706
MCC
1463 continue;
1464
f4742949
MCC
1465 func = PCI_FUNC(pdev->devfn);
1466 slot = PCI_SLOT(pdev->devfn);
1467 if (slot == 3) {
1468 if (unlikely(func > MAX_MCR_FUNC))
1469 goto error;
1470 pvt->pci_mcr[func] = pdev;
1471 } else if (likely(slot >= 4 && slot < 4 + NUM_CHANS)) {
1472 if (unlikely(func > MAX_CHAN_FUNC))
ef708b53 1473 goto error;
f4742949
MCC
1474 pvt->pci_ch[slot - 4][func] = pdev;
1475 } else if (!slot && !func)
1476 pvt->pci_noncore = pdev;
1477 else
1478 goto error;
ef708b53 1479
f4742949
MCC
1480 debugf0("Associated fn %d.%d, dev = %p, socket %d\n",
1481 PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn),
1482 pdev, i7core_dev->socket);
14d2c083 1483
f4742949
MCC
1484 if (PCI_SLOT(pdev->devfn) == 3 &&
1485 PCI_FUNC(pdev->devfn) == 2)
1486 pvt->is_registered = 1;
a0c36a1f 1487 }
e9bd2e73 1488
a0c36a1f 1489 return 0;
ef708b53
MCC
1490
1491error:
1492 i7core_printk(KERN_ERR, "Device %d, function %d "
1493 "is out of the expected range\n",
1494 slot, func);
1495 return -EINVAL;
a0c36a1f
MCC
1496}
1497
442305b1
MCC
1498/****************************************************************************
1499 Error check routines
1500 ****************************************************************************/
f4742949 1501static void i7core_rdimm_update_csrow(struct mem_ctl_info *mci,
1288c18f
MCC
1502 const int chan,
1503 const int dimm,
1504 const int add)
b4e8f0b6
MCC
1505{
1506 char *msg;
1507 struct i7core_pvt *pvt = mci->pvt_info;
f4742949 1508 int row = pvt->csrow_map[chan][dimm], i;
b4e8f0b6
MCC
1509
1510 for (i = 0; i < add; i++) {
1511 msg = kasprintf(GFP_KERNEL, "Corrected error "
f4742949
MCC
1512 "(Socket=%d channel=%d dimm=%d)",
1513 pvt->i7core_dev->socket, chan, dimm);
b4e8f0b6
MCC
1514
1515 edac_mc_handle_fbd_ce(mci, row, 0, msg);
1516 kfree (msg);
1517 }
1518}
1519
1520static void i7core_rdimm_update_ce_count(struct mem_ctl_info *mci,
1288c18f
MCC
1521 const int chan,
1522 const int new0,
1523 const int new1,
1524 const int new2)
b4e8f0b6
MCC
1525{
1526 struct i7core_pvt *pvt = mci->pvt_info;
1527 int add0 = 0, add1 = 0, add2 = 0;
1528 /* Updates CE counters if it is not the first time here */
f4742949 1529 if (pvt->ce_count_available) {
b4e8f0b6
MCC
1530 /* Updates CE counters */
1531
f4742949
MCC
1532 add2 = new2 - pvt->rdimm_last_ce_count[chan][2];
1533 add1 = new1 - pvt->rdimm_last_ce_count[chan][1];
1534 add0 = new0 - pvt->rdimm_last_ce_count[chan][0];
b4e8f0b6
MCC
1535
1536 if (add2 < 0)
1537 add2 += 0x7fff;
f4742949 1538 pvt->rdimm_ce_count[chan][2] += add2;
b4e8f0b6
MCC
1539
1540 if (add1 < 0)
1541 add1 += 0x7fff;
f4742949 1542 pvt->rdimm_ce_count[chan][1] += add1;
b4e8f0b6
MCC
1543
1544 if (add0 < 0)
1545 add0 += 0x7fff;
f4742949 1546 pvt->rdimm_ce_count[chan][0] += add0;
b4e8f0b6 1547 } else
f4742949 1548 pvt->ce_count_available = 1;
b4e8f0b6
MCC
1549
1550 /* Store the new values */
f4742949
MCC
1551 pvt->rdimm_last_ce_count[chan][2] = new2;
1552 pvt->rdimm_last_ce_count[chan][1] = new1;
1553 pvt->rdimm_last_ce_count[chan][0] = new0;
b4e8f0b6
MCC
1554
1555 /*updated the edac core */
1556 if (add0 != 0)
f4742949 1557 i7core_rdimm_update_csrow(mci, chan, 0, add0);
b4e8f0b6 1558 if (add1 != 0)
f4742949 1559 i7core_rdimm_update_csrow(mci, chan, 1, add1);
b4e8f0b6 1560 if (add2 != 0)
f4742949 1561 i7core_rdimm_update_csrow(mci, chan, 2, add2);
b4e8f0b6
MCC
1562
1563}
1564
f4742949 1565static void i7core_rdimm_check_mc_ecc_err(struct mem_ctl_info *mci)
b4e8f0b6
MCC
1566{
1567 struct i7core_pvt *pvt = mci->pvt_info;
1568 u32 rcv[3][2];
1569 int i, new0, new1, new2;
1570
1571 /*Read DEV 3: FUN 2: MC_COR_ECC_CNT regs directly*/
f4742949 1572 pci_read_config_dword(pvt->pci_mcr[2], MC_COR_ECC_CNT_0,
b4e8f0b6 1573 &rcv[0][0]);
f4742949 1574 pci_read_config_dword(pvt->pci_mcr[2], MC_COR_ECC_CNT_1,
b4e8f0b6 1575 &rcv[0][1]);
f4742949 1576 pci_read_config_dword(pvt->pci_mcr[2], MC_COR_ECC_CNT_2,
b4e8f0b6 1577 &rcv[1][0]);
f4742949 1578 pci_read_config_dword(pvt->pci_mcr[2], MC_COR_ECC_CNT_3,
b4e8f0b6 1579 &rcv[1][1]);
f4742949 1580 pci_read_config_dword(pvt->pci_mcr[2], MC_COR_ECC_CNT_4,
b4e8f0b6 1581 &rcv[2][0]);
f4742949 1582 pci_read_config_dword(pvt->pci_mcr[2], MC_COR_ECC_CNT_5,
b4e8f0b6
MCC
1583 &rcv[2][1]);
1584 for (i = 0 ; i < 3; i++) {
1585 debugf3("MC_COR_ECC_CNT%d = 0x%x; MC_COR_ECC_CNT%d = 0x%x\n",
1586 (i * 2), rcv[i][0], (i * 2) + 1, rcv[i][1]);
1587 /*if the channel has 3 dimms*/
f4742949 1588 if (pvt->channel[i].dimms > 2) {
b4e8f0b6
MCC
1589 new0 = DIMM_BOT_COR_ERR(rcv[i][0]);
1590 new1 = DIMM_TOP_COR_ERR(rcv[i][0]);
1591 new2 = DIMM_BOT_COR_ERR(rcv[i][1]);
1592 } else {
1593 new0 = DIMM_TOP_COR_ERR(rcv[i][0]) +
1594 DIMM_BOT_COR_ERR(rcv[i][0]);
1595 new1 = DIMM_TOP_COR_ERR(rcv[i][1]) +
1596 DIMM_BOT_COR_ERR(rcv[i][1]);
1597 new2 = 0;
1598 }
1599
f4742949 1600 i7core_rdimm_update_ce_count(mci, i, new0, new1, new2);
b4e8f0b6
MCC
1601 }
1602}
442305b1
MCC
1603
1604/* This function is based on the device 3 function 4 registers as described on:
1605 * Intel Xeon Processor 5500 Series Datasheet Volume 2
1606 * http://www.intel.com/Assets/PDF/datasheet/321322.pdf
1607 * also available at:
1608 * http://www.arrownac.com/manufacturers/intel/s/nehalem/5500-datasheet-v2.pdf
1609 */
f4742949 1610static void i7core_udimm_check_mc_ecc_err(struct mem_ctl_info *mci)
442305b1
MCC
1611{
1612 struct i7core_pvt *pvt = mci->pvt_info;
1613 u32 rcv1, rcv0;
1614 int new0, new1, new2;
1615
f4742949 1616 if (!pvt->pci_mcr[4]) {
b990538a 1617 debugf0("%s MCR registers not found\n", __func__);
442305b1
MCC
1618 return;
1619 }
1620
b4e8f0b6 1621 /* Corrected test errors */
f4742949
MCC
1622 pci_read_config_dword(pvt->pci_mcr[4], MC_TEST_ERR_RCV1, &rcv1);
1623 pci_read_config_dword(pvt->pci_mcr[4], MC_TEST_ERR_RCV0, &rcv0);
442305b1
MCC
1624
1625 /* Store the new values */
1626 new2 = DIMM2_COR_ERR(rcv1);
1627 new1 = DIMM1_COR_ERR(rcv0);
1628 new0 = DIMM0_COR_ERR(rcv0);
1629
442305b1 1630 /* Updates CE counters if it is not the first time here */
f4742949 1631 if (pvt->ce_count_available) {
442305b1
MCC
1632 /* Updates CE counters */
1633 int add0, add1, add2;
1634
f4742949
MCC
1635 add2 = new2 - pvt->udimm_last_ce_count[2];
1636 add1 = new1 - pvt->udimm_last_ce_count[1];
1637 add0 = new0 - pvt->udimm_last_ce_count[0];
442305b1
MCC
1638
1639 if (add2 < 0)
1640 add2 += 0x7fff;
f4742949 1641 pvt->udimm_ce_count[2] += add2;
442305b1
MCC
1642
1643 if (add1 < 0)
1644 add1 += 0x7fff;
f4742949 1645 pvt->udimm_ce_count[1] += add1;
442305b1
MCC
1646
1647 if (add0 < 0)
1648 add0 += 0x7fff;
f4742949 1649 pvt->udimm_ce_count[0] += add0;
b4e8f0b6
MCC
1650
1651 if (add0 | add1 | add2)
1652 i7core_printk(KERN_ERR, "New Corrected error(s): "
1653 "dimm0: +%d, dimm1: +%d, dimm2 +%d\n",
1654 add0, add1, add2);
442305b1 1655 } else
f4742949 1656 pvt->ce_count_available = 1;
442305b1
MCC
1657
1658 /* Store the new values */
f4742949
MCC
1659 pvt->udimm_last_ce_count[2] = new2;
1660 pvt->udimm_last_ce_count[1] = new1;
1661 pvt->udimm_last_ce_count[0] = new0;
442305b1
MCC
1662}
1663
8a2f118e
MCC
1664/*
1665 * According with tables E-11 and E-12 of chapter E.3.3 of Intel 64 and IA-32
1666 * Architectures Software Developer’s Manual Volume 3B.
f237fcf2
MCC
1667 * Nehalem are defined as family 0x06, model 0x1a
1668 *
1669 * The MCA registers used here are the following ones:
8a2f118e 1670 * struct mce field MCA Register
f237fcf2
MCC
1671 * m->status MSR_IA32_MC8_STATUS
1672 * m->addr MSR_IA32_MC8_ADDR
1673 * m->misc MSR_IA32_MC8_MISC
8a2f118e
MCC
1674 * In the case of Nehalem, the error information is masked at .status and .misc
1675 * fields
1676 */
d5381642 1677static void i7core_mce_output_error(struct mem_ctl_info *mci,
1288c18f 1678 const struct mce *m)
d5381642 1679{
b4e8f0b6 1680 struct i7core_pvt *pvt = mci->pvt_info;
a639539f 1681 char *type, *optype, *err, *msg;
8a2f118e 1682 unsigned long error = m->status & 0x1ff0000l;
a639539f 1683 u32 optypenum = (m->status >> 4) & 0x07;
8a2f118e
MCC
1684 u32 core_err_cnt = (m->status >> 38) && 0x7fff;
1685 u32 dimm = (m->misc >> 16) & 0x3;
1686 u32 channel = (m->misc >> 18) & 0x3;
1687 u32 syndrome = m->misc >> 32;
1688 u32 errnum = find_first_bit(&error, 32);
b4e8f0b6 1689 int csrow;
8a2f118e 1690
c5d34528
MCC
1691 if (m->mcgstatus & 1)
1692 type = "FATAL";
1693 else
1694 type = "NON_FATAL";
1695
a639539f 1696 switch (optypenum) {
b990538a
MCC
1697 case 0:
1698 optype = "generic undef request";
1699 break;
1700 case 1:
1701 optype = "read error";
1702 break;
1703 case 2:
1704 optype = "write error";
1705 break;
1706 case 3:
1707 optype = "addr/cmd error";
1708 break;
1709 case 4:
1710 optype = "scrubbing error";
1711 break;
1712 default:
1713 optype = "reserved";
1714 break;
a639539f
MCC
1715 }
1716
8a2f118e
MCC
1717 switch (errnum) {
1718 case 16:
1719 err = "read ECC error";
1720 break;
1721 case 17:
1722 err = "RAS ECC error";
1723 break;
1724 case 18:
1725 err = "write parity error";
1726 break;
1727 case 19:
1728 err = "redundacy loss";
1729 break;
1730 case 20:
1731 err = "reserved";
1732 break;
1733 case 21:
1734 err = "memory range error";
1735 break;
1736 case 22:
1737 err = "RTID out of range";
1738 break;
1739 case 23:
1740 err = "address parity error";
1741 break;
1742 case 24:
1743 err = "byte enable parity error";
1744 break;
1745 default:
1746 err = "unknown";
d5381642 1747 }
d5381642 1748
f237fcf2 1749 /* FIXME: should convert addr into bank and rank information */
8a2f118e 1750 msg = kasprintf(GFP_ATOMIC,
f4742949 1751 "%s (addr = 0x%08llx, cpu=%d, Dimm=%d, Channel=%d, "
a639539f 1752 "syndrome=0x%08x, count=%d, Err=%08llx:%08llx (%s: %s))\n",
f4742949 1753 type, (long long) m->addr, m->cpu, dimm, channel,
a639539f
MCC
1754 syndrome, core_err_cnt, (long long)m->status,
1755 (long long)m->misc, optype, err);
8a2f118e
MCC
1756
1757 debugf0("%s", msg);
d5381642 1758
f4742949 1759 csrow = pvt->csrow_map[channel][dimm];
b4e8f0b6 1760
d5381642 1761 /* Call the helper to output message */
b4e8f0b6
MCC
1762 if (m->mcgstatus & 1)
1763 edac_mc_handle_fbd_ue(mci, csrow, 0,
1764 0 /* FIXME: should be channel here */, msg);
f4742949 1765 else if (!pvt->is_registered)
b4e8f0b6
MCC
1766 edac_mc_handle_fbd_ce(mci, csrow,
1767 0 /* FIXME: should be channel here */, msg);
8a2f118e
MCC
1768
1769 kfree(msg);
d5381642
MCC
1770}
1771
87d1d272
MCC
1772/*
1773 * i7core_check_error Retrieve and process errors reported by the
1774 * hardware. Called by the Core module.
1775 */
1776static void i7core_check_error(struct mem_ctl_info *mci)
1777{
d5381642
MCC
1778 struct i7core_pvt *pvt = mci->pvt_info;
1779 int i;
1780 unsigned count = 0;
ca9c90ba 1781 struct mce *m;
d5381642 1782
ca9c90ba
MCC
1783 /*
1784 * MCE first step: Copy all mce errors into a temporary buffer
1785 * We use a double buffering here, to reduce the risk of
1786 * loosing an error.
1787 */
1788 smp_rmb();
321ece4d
MCC
1789 count = (pvt->mce_out + MCE_LOG_LEN - pvt->mce_in)
1790 % MCE_LOG_LEN;
ca9c90ba 1791 if (!count)
8a311e17 1792 goto check_ce_error;
f4742949 1793
ca9c90ba 1794 m = pvt->mce_outentry;
321ece4d
MCC
1795 if (pvt->mce_in + count > MCE_LOG_LEN) {
1796 unsigned l = MCE_LOG_LEN - pvt->mce_in;
f4742949 1797
ca9c90ba
MCC
1798 memcpy(m, &pvt->mce_entry[pvt->mce_in], sizeof(*m) * l);
1799 smp_wmb();
1800 pvt->mce_in = 0;
1801 count -= l;
1802 m += l;
1803 }
1804 memcpy(m, &pvt->mce_entry[pvt->mce_in], sizeof(*m) * count);
1805 smp_wmb();
1806 pvt->mce_in += count;
1807
1808 smp_rmb();
1809 if (pvt->mce_overrun) {
1810 i7core_printk(KERN_ERR, "Lost %d memory errors\n",
1811 pvt->mce_overrun);
1812 smp_wmb();
1813 pvt->mce_overrun = 0;
1814 }
d5381642 1815
ca9c90ba
MCC
1816 /*
1817 * MCE second step: parse errors and display
1818 */
d5381642 1819 for (i = 0; i < count; i++)
ca9c90ba 1820 i7core_mce_output_error(mci, &pvt->mce_outentry[i]);
d5381642 1821
ca9c90ba
MCC
1822 /*
1823 * Now, let's increment CE error counts
1824 */
8a311e17 1825check_ce_error:
f4742949
MCC
1826 if (!pvt->is_registered)
1827 i7core_udimm_check_mc_ecc_err(mci);
1828 else
1829 i7core_rdimm_check_mc_ecc_err(mci);
87d1d272
MCC
1830}
1831
d5381642
MCC
1832/*
1833 * i7core_mce_check_error Replicates mcelog routine to get errors
1834 * This routine simply queues mcelog errors, and
1835 * return. The error itself should be handled later
1836 * by i7core_check_error.
6e103be1
MCC
1837 * WARNING: As this routine should be called at NMI time, extra care should
1838 * be taken to avoid deadlocks, and to be as fast as possible.
d5381642
MCC
1839 */
1840static int i7core_mce_check_error(void *priv, struct mce *mce)
1841{
c5d34528
MCC
1842 struct mem_ctl_info *mci = priv;
1843 struct i7core_pvt *pvt = mci->pvt_info;
d5381642 1844
8a2f118e
MCC
1845 /*
1846 * Just let mcelog handle it if the error is
1847 * outside the memory controller
1848 */
1849 if (((mce->status & 0xffff) >> 7) != 1)
1850 return 0;
1851
f237fcf2
MCC
1852 /* Bank 8 registers are the only ones that we know how to handle */
1853 if (mce->bank != 8)
1854 return 0;
1855
3b918c12 1856#ifdef CONFIG_SMP
f4742949 1857 /* Only handle if it is the right mc controller */
6e103be1 1858 if (cpu_data(mce->cpu).phys_proc_id != pvt->i7core_dev->socket)
f4742949 1859 return 0;
3b918c12 1860#endif
f4742949 1861
ca9c90ba 1862 smp_rmb();
321ece4d 1863 if ((pvt->mce_out + 1) % MCE_LOG_LEN == pvt->mce_in) {
ca9c90ba
MCC
1864 smp_wmb();
1865 pvt->mce_overrun++;
1866 return 0;
d5381642 1867 }
6e103be1
MCC
1868
1869 /* Copy memory error at the ringbuffer */
1870 memcpy(&pvt->mce_entry[pvt->mce_out], mce, sizeof(*mce));
ca9c90ba 1871 smp_wmb();
321ece4d 1872 pvt->mce_out = (pvt->mce_out + 1) % MCE_LOG_LEN;
d5381642 1873
c5d34528
MCC
1874 /* Handle fatal errors immediately */
1875 if (mce->mcgstatus & 1)
1876 i7core_check_error(mci);
1877
d5381642 1878 /* Advice mcelog that the error were handled */
8a2f118e 1879 return 1;
d5381642
MCC
1880}
1881
f4742949 1882static int i7core_register_mci(struct i7core_dev *i7core_dev,
1288c18f 1883 const int num_channels, const int num_csrows)
a0c36a1f
MCC
1884{
1885 struct mem_ctl_info *mci;
1886 struct i7core_pvt *pvt;
ba6c5c62 1887 int csrow = 0;
f4742949 1888 int rc;
a0c36a1f 1889
a0c36a1f 1890 /* allocate a new MC control structure */
d4c27795
MCC
1891 mci = edac_mc_alloc(sizeof(*pvt), num_csrows, num_channels,
1892 i7core_dev->socket);
f4742949
MCC
1893 if (unlikely(!mci))
1894 return -ENOMEM;
a0c36a1f
MCC
1895
1896 debugf0("MC: " __FILE__ ": %s(): mci = %p\n", __func__, mci);
1897
f4742949
MCC
1898 /* record ptr to the generic device */
1899 mci->dev = &i7core_dev->pdev[0]->dev;
1900
a0c36a1f 1901 pvt = mci->pvt_info;
ef708b53 1902 memset(pvt, 0, sizeof(*pvt));
67166af4 1903
41fcb7fe
MCC
1904 /*
1905 * FIXME: how to handle RDDR3 at MCI level? It is possible to have
1906 * Mixed RDDR3/UDDR3 with Nehalem, provided that they are on different
1907 * memory channels
1908 */
1909 mci->mtype_cap = MEM_FLAG_DDR3;
a0c36a1f
MCC
1910 mci->edac_ctl_cap = EDAC_FLAG_NONE;
1911 mci->edac_cap = EDAC_FLAG_NONE;
1912 mci->mod_name = "i7core_edac.c";
1913 mci->mod_ver = I7CORE_REVISION;
f4742949
MCC
1914 mci->ctl_name = kasprintf(GFP_KERNEL, "i7 core #%d",
1915 i7core_dev->socket);
1916 mci->dev_name = pci_name(i7core_dev->pdev[0]);
a0c36a1f 1917 mci->ctl_page_to_phys = NULL;
1288c18f
MCC
1918
1919 if (pvt->is_registered)
1920 mci->mc_driver_sysfs_attributes = i7core_sysfs_rdimm_attrs;
1921 else
1922 mci->mc_driver_sysfs_attributes = i7core_sysfs_udimm_attrs;
1923
87d1d272
MCC
1924 /* Set the function pointer to an actual operation function */
1925 mci->edac_check = i7core_check_error;
8f331907 1926
ef708b53 1927 /* Store pci devices at mci for faster access */
f4742949 1928 rc = mci_bind_devs(mci, i7core_dev);
41fcb7fe 1929 if (unlikely(rc < 0))
f4742949 1930 goto fail;
ef708b53
MCC
1931
1932 /* Get dimm basic config */
f4742949 1933 get_dimm_config(mci, &csrow);
ef708b53 1934
a0c36a1f 1935 /* add this new MC control structure to EDAC's list of MCs */
b7c76151 1936 if (unlikely(edac_mc_add_mc(mci))) {
a0c36a1f
MCC
1937 debugf0("MC: " __FILE__
1938 ": %s(): failed edac_mc_add_mc()\n", __func__);
1939 /* FIXME: perhaps some code should go here that disables error
1940 * reporting if we just enabled it
1941 */
b7c76151
MCC
1942
1943 rc = -EINVAL;
f4742949 1944 goto fail;
a0c36a1f
MCC
1945 }
1946
1947 /* allocating generic PCI control info */
939747bd 1948 pvt->i7core_pci = edac_pci_create_generic_ctl(&i7core_dev->pdev[0]->dev,
f4742949 1949 EDAC_MOD_STR);
939747bd 1950 if (unlikely(!pvt->i7core_pci)) {
a0c36a1f
MCC
1951 printk(KERN_WARNING
1952 "%s(): Unable to create PCI control\n",
1953 __func__);
1954 printk(KERN_WARNING
1955 "%s(): PCI error report via EDAC not setup\n",
1956 __func__);
1957 }
1958
194a40fe 1959 /* Default error mask is any memory */
ef708b53 1960 pvt->inject.channel = 0;
194a40fe
MCC
1961 pvt->inject.dimm = -1;
1962 pvt->inject.rank = -1;
1963 pvt->inject.bank = -1;
1964 pvt->inject.page = -1;
1965 pvt->inject.col = -1;
1966
d5381642 1967 /* Registers on edac_mce in order to receive memory errors */
c5d34528 1968 pvt->edac_mce.priv = mci;
d5381642 1969 pvt->edac_mce.check_error = i7core_mce_check_error;
d5381642
MCC
1970
1971 rc = edac_mce_register(&pvt->edac_mce);
b990538a 1972 if (unlikely(rc < 0)) {
d5381642
MCC
1973 debugf0("MC: " __FILE__
1974 ": %s(): failed edac_mce_register()\n", __func__);
f4742949
MCC
1975 }
1976
1977fail:
d4d1ef45
TL
1978 if (rc < 0)
1979 edac_mc_free(mci);
f4742949
MCC
1980 return rc;
1981}
1982
1983/*
1984 * i7core_probe Probe for ONE instance of device to see if it is
1985 * present.
1986 * return:
1987 * 0 for FOUND a device
1988 * < 0 for error code
1989 */
2d95d815 1990
f4742949
MCC
1991static int __devinit i7core_probe(struct pci_dev *pdev,
1992 const struct pci_device_id *id)
1993{
f4742949
MCC
1994 int rc;
1995 struct i7core_dev *i7core_dev;
1996
2d95d815
MCC
1997 /* get the pci devices we want to reserve for our use */
1998 mutex_lock(&i7core_edac_lock);
1999
f4742949 2000 /*
d4c27795 2001 * All memory controllers are allocated at the first pass.
f4742949 2002 */
2d95d815
MCC
2003 if (unlikely(probed >= 1)) {
2004 mutex_unlock(&i7core_edac_lock);
f4742949 2005 return -EINVAL;
2d95d815
MCC
2006 }
2007 probed++;
de06eeef 2008
bd9e19ca 2009 rc = i7core_get_devices(pci_dev_table);
f4742949
MCC
2010 if (unlikely(rc < 0))
2011 goto fail0;
2012
2013 list_for_each_entry(i7core_dev, &i7core_edac_list, list) {
2014 int channels;
2015 int csrows;
2016
2017 /* Check the number of active and not disabled channels */
2018 rc = i7core_get_active_channels(i7core_dev->socket,
2019 &channels, &csrows);
2020 if (unlikely(rc < 0))
2021 goto fail1;
2022
d4c27795
MCC
2023 rc = i7core_register_mci(i7core_dev, channels, csrows);
2024 if (unlikely(rc < 0))
2025 goto fail1;
d5381642
MCC
2026 }
2027
ef708b53 2028 i7core_printk(KERN_INFO, "Driver loaded.\n");
8f331907 2029
66607706 2030 mutex_unlock(&i7core_edac_lock);
a0c36a1f
MCC
2031 return 0;
2032
66607706 2033fail1:
13d6e9b6 2034 i7core_put_all_devices();
66607706
MCC
2035fail0:
2036 mutex_unlock(&i7core_edac_lock);
b7c76151 2037 return rc;
a0c36a1f
MCC
2038}
2039
2040/*
2041 * i7core_remove destructor for one instance of device
2042 *
2043 */
2044static void __devexit i7core_remove(struct pci_dev *pdev)
2045{
2046 struct mem_ctl_info *mci;
22e6bcbd 2047 struct i7core_dev *i7core_dev, *tmp;
939747bd 2048 struct i7core_pvt *pvt;
a0c36a1f
MCC
2049
2050 debugf0(__FILE__ ": %s()\n", __func__);
2051
22e6bcbd
MCC
2052 /*
2053 * we have a trouble here: pdev value for removal will be wrong, since
2054 * it will point to the X58 register used to detect that the machine
2055 * is a Nehalem or upper design. However, due to the way several PCI
2056 * devices are grouped together to provide MC functionality, we need
2057 * to use a different method for releasing the devices
2058 */
87d1d272 2059
66607706 2060 mutex_lock(&i7core_edac_lock);
22e6bcbd 2061 list_for_each_entry_safe(i7core_dev, tmp, &i7core_edac_list, list) {
939747bd
MCC
2062 mci = find_mci_by_dev(&i7core_dev->pdev[0]->dev);
2063 if (unlikely(!mci || !mci->pvt_info)) {
2064 i7core_printk(KERN_ERR,
2065 "Couldn't find mci hanler\n");
2066 } else {
2067 pvt = mci->pvt_info;
22e6bcbd 2068 i7core_dev = pvt->i7core_dev;
939747bd
MCC
2069
2070 if (likely(pvt->i7core_pci))
2071 edac_pci_release_generic_ctl(pvt->i7core_pci);
2072 else
2073 i7core_printk(KERN_ERR,
2074 "Couldn't find mem_ctl_info for socket %d\n",
2075 i7core_dev->socket);
2076 pvt->i7core_pci = NULL;
2077
2078 edac_mc_del_mc(&i7core_dev->pdev[0]->dev);
2079
22e6bcbd
MCC
2080 edac_mce_unregister(&pvt->edac_mce);
2081 kfree(mci->ctl_name);
2082 edac_mc_free(mci);
2083 i7core_put_devices(i7core_dev);
22e6bcbd
MCC
2084 }
2085 }
2d95d815
MCC
2086 probed--;
2087
66607706 2088 mutex_unlock(&i7core_edac_lock);
a0c36a1f
MCC
2089}
2090
a0c36a1f
MCC
2091MODULE_DEVICE_TABLE(pci, i7core_pci_tbl);
2092
2093/*
2094 * i7core_driver pci_driver structure for this module
2095 *
2096 */
2097static struct pci_driver i7core_driver = {
2098 .name = "i7core_edac",
2099 .probe = i7core_probe,
2100 .remove = __devexit_p(i7core_remove),
2101 .id_table = i7core_pci_tbl,
2102};
2103
2104/*
2105 * i7core_init Module entry function
2106 * Try to initialize this module for its devices
2107 */
2108static int __init i7core_init(void)
2109{
2110 int pci_rc;
2111
2112 debugf2("MC: " __FILE__ ": %s()\n", __func__);
2113
2114 /* Ensure that the OPSTATE is set correctly for POLL or NMI */
2115 opstate_init();
2116
bd9e19ca 2117 i7core_xeon_pci_fixup(pci_dev_table);
bc2d7245 2118
a0c36a1f
MCC
2119 pci_rc = pci_register_driver(&i7core_driver);
2120
3ef288a9
MCC
2121 if (pci_rc >= 0)
2122 return 0;
2123
2124 i7core_printk(KERN_ERR, "Failed to register device with error %d.\n",
2125 pci_rc);
2126
2127 return pci_rc;
a0c36a1f
MCC
2128}
2129
2130/*
2131 * i7core_exit() Module exit function
2132 * Unregister the driver
2133 */
2134static void __exit i7core_exit(void)
2135{
2136 debugf2("MC: " __FILE__ ": %s()\n", __func__);
2137 pci_unregister_driver(&i7core_driver);
2138}
2139
2140module_init(i7core_init);
2141module_exit(i7core_exit);
2142
2143MODULE_LICENSE("GPL");
2144MODULE_AUTHOR("Mauro Carvalho Chehab <mchehab@redhat.com>");
2145MODULE_AUTHOR("Red Hat Inc. (http://www.redhat.com)");
2146MODULE_DESCRIPTION("MC Driver for Intel i7 Core memory controllers - "
2147 I7CORE_REVISION);
2148
2149module_param(edac_op_state, int, 0444);
2150MODULE_PARM_DESC(edac_op_state, "EDAC Error Reporting state: 0=Poll,1=NMI");