i7core_edac: move static vars to the beginning of the file
[GitHub/mt8127/android_kernel_alcatel_ttab.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
VM
219struct pci_id_table {
220 struct pci_id_descr *descr;
221 int n_devs;
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
bd9e19ca 279struct 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
52a2e4fc
MCC
316struct pci_id_descr pci_dev_descr_lynnfield[] = {
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
bd9e19ca
VM
338struct pci_id_descr pci_dev_descr_i7core_westmere[] = {
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
369#define PCI_ID_TABLE_ENTRY(A) { A, ARRAY_SIZE(A) }
370struct pci_id_table pci_dev_table[] = {
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 */
67166af4
MCC
489static int i7core_get_active_channels(u8 socket, unsigned *channels,
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
f4742949 550static int get_dimm_config(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 */
8f331907 741static int disable_inject(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
276b824c
MCC
926static int write_and_test(struct pci_dev *dev, int where, u32 val)
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
MCC
1124
1125
1126static struct mcidev_sysfs_attribute i7core_addrmatch_attrs[] = {
1127 ATTR_ADDR_MATCH(channel),
1128 ATTR_ADDR_MATCH(dimm),
1129 ATTR_ADDR_MATCH(rank),
1130 ATTR_ADDR_MATCH(bank),
1131 ATTR_ADDR_MATCH(page),
1132 ATTR_ADDR_MATCH(col),
1133 { .attr = { .name = NULL } }
1134};
1135
a5538e53
MCC
1136static struct mcidev_sysfs_group i7core_inject_addrmatch = {
1137 .name = "inject_addrmatch",
1138 .mcidev_attr = i7core_addrmatch_attrs,
1139};
1140
f338d736
MCC
1141static struct mcidev_sysfs_attribute i7core_udimm_counters_attrs[] = {
1142 ATTR_COUNTER(0),
1143 ATTR_COUNTER(1),
1144 ATTR_COUNTER(2),
64aab720 1145 { .attr = { .name = NULL } }
f338d736
MCC
1146};
1147
1148static struct mcidev_sysfs_group i7core_udimm_counters = {
1149 .name = "all_channel_counts",
1150 .mcidev_attr = i7core_udimm_counters_attrs,
1151};
1152
a5538e53 1153static struct mcidev_sysfs_attribute i7core_sysfs_attrs[] = {
194a40fe
MCC
1154 {
1155 .attr = {
1156 .name = "inject_section",
1157 .mode = (S_IRUGO | S_IWUSR)
1158 },
1159 .show = i7core_inject_section_show,
1160 .store = i7core_inject_section_store,
1161 }, {
1162 .attr = {
1163 .name = "inject_type",
1164 .mode = (S_IRUGO | S_IWUSR)
1165 },
1166 .show = i7core_inject_type_show,
1167 .store = i7core_inject_type_store,
1168 }, {
1169 .attr = {
1170 .name = "inject_eccmask",
1171 .mode = (S_IRUGO | S_IWUSR)
1172 },
1173 .show = i7core_inject_eccmask_show,
1174 .store = i7core_inject_eccmask_store,
1175 }, {
a5538e53 1176 .grp = &i7core_inject_addrmatch,
194a40fe
MCC
1177 }, {
1178 .attr = {
1179 .name = "inject_enable",
1180 .mode = (S_IRUGO | S_IWUSR)
1181 },
1182 .show = i7core_inject_enable_show,
1183 .store = i7core_inject_enable_store,
1184 },
f338d736 1185 { .attr = { .name = NULL } }, /* Reserved for udimm counters */
42538680 1186 { .attr = { .name = NULL } }
194a40fe
MCC
1187};
1188
a0c36a1f
MCC
1189/****************************************************************************
1190 Device initialization routines: put/get, init/exit
1191 ****************************************************************************/
1192
1193/*
1194 * i7core_put_devices 'put' all the devices that we have
1195 * reserved via 'get'
1196 */
13d6e9b6 1197static void i7core_put_devices(struct i7core_dev *i7core_dev)
a0c36a1f 1198{
13d6e9b6 1199 int i;
a0c36a1f 1200
22e6bcbd 1201 debugf0(__FILE__ ": %s()\n", __func__);
de06eeef 1202 for (i = 0; i < i7core_dev->n_devs; i++) {
22e6bcbd
MCC
1203 struct pci_dev *pdev = i7core_dev->pdev[i];
1204 if (!pdev)
1205 continue;
1206 debugf0("Removing dev %02x:%02x.%d\n",
1207 pdev->bus->number,
1208 PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn));
1209 pci_dev_put(pdev);
1210 }
13d6e9b6 1211 kfree(i7core_dev->pdev);
22e6bcbd 1212 list_del(&i7core_dev->list);
13d6e9b6
MCC
1213 kfree(i7core_dev);
1214}
66607706 1215
13d6e9b6
MCC
1216static void i7core_put_all_devices(void)
1217{
42538680 1218 struct i7core_dev *i7core_dev, *tmp;
13d6e9b6 1219
42538680 1220 list_for_each_entry_safe(i7core_dev, tmp, &i7core_edac_list, list)
13d6e9b6 1221 i7core_put_devices(i7core_dev);
a0c36a1f
MCC
1222}
1223
bd9e19ca 1224static void __init i7core_xeon_pci_fixup(struct pci_id_table *table)
bc2d7245
KM
1225{
1226 struct pci_dev *pdev = NULL;
1227 int i;
1228 /*
1229 * On Xeon 55xx, the Intel Quckpath Arch Generic Non-core pci buses
1230 * aren't announced by acpi. So, we need to use a legacy scan probing
1231 * to detect them
1232 */
bd9e19ca
VM
1233 while (table && table->descr) {
1234 pdev = pci_get_device(PCI_VENDOR_ID_INTEL, table->descr[0].dev_id, NULL);
1235 if (unlikely(!pdev)) {
1236 for (i = 0; i < MAX_SOCKET_BUSES; i++)
1237 pcibios_scan_specific_bus(255-i);
1238 }
bda14289 1239 pci_dev_put(pdev);
bd9e19ca 1240 table++;
bc2d7245
KM
1241 }
1242}
1243
bda14289
MCC
1244static unsigned i7core_pci_lastbus(void)
1245{
1246 int last_bus = 0, bus;
1247 struct pci_bus *b = NULL;
1248
1249 while ((b = pci_find_next_bus(b)) != NULL) {
1250 bus = b->number;
1251 debugf0("Found bus %d\n", bus);
1252 if (bus > last_bus)
1253 last_bus = bus;
1254 }
1255
1256 debugf0("Last bus %d\n", last_bus);
1257
1258 return last_bus;
1259}
1260
a0c36a1f
MCC
1261/*
1262 * i7core_get_devices Find and perform 'get' operation on the MCH's
1263 * device/functions we want to reference for this driver
1264 *
1265 * Need to 'get' device 16 func 1 and func 2
1266 */
de06eeef 1267int i7core_get_onedevice(struct pci_dev **prev, int devno,
bda14289
MCC
1268 struct pci_id_descr *dev_descr, unsigned n_devs,
1269 unsigned last_bus)
a0c36a1f 1270{
66607706
MCC
1271 struct i7core_dev *i7core_dev;
1272
8f331907 1273 struct pci_dev *pdev = NULL;
67166af4
MCC
1274 u8 bus = 0;
1275 u8 socket = 0;
a0c36a1f 1276
c77720b9 1277 pdev = pci_get_device(PCI_VENDOR_ID_INTEL,
de06eeef 1278 dev_descr->dev_id, *prev);
c77720b9 1279
c77720b9
MCC
1280 /*
1281 * On Xeon 55xx, the Intel Quckpath Arch Generic Non-core regs
1282 * is at addr 8086:2c40, instead of 8086:2c41. So, we need
1283 * to probe for the alternate address in case of failure
1284 */
de06eeef 1285 if (dev_descr->dev_id == PCI_DEVICE_ID_INTEL_I7_NONCORE && !pdev)
c77720b9 1286 pdev = pci_get_device(PCI_VENDOR_ID_INTEL,
fd382654 1287 PCI_DEVICE_ID_INTEL_I7_NONCORE_ALT, *prev);
d1fd4fb6 1288
bd9e19ca 1289 if (dev_descr->dev_id == PCI_DEVICE_ID_INTEL_LYNNFIELD_NONCORE && !pdev)
f05da2f7
MCC
1290 pdev = pci_get_device(PCI_VENDOR_ID_INTEL,
1291 PCI_DEVICE_ID_INTEL_LYNNFIELD_NONCORE_ALT,
1292 *prev);
1293
c77720b9
MCC
1294 if (!pdev) {
1295 if (*prev) {
1296 *prev = pdev;
1297 return 0;
d1fd4fb6
MCC
1298 }
1299
de06eeef 1300 if (dev_descr->optional)
c77720b9 1301 return 0;
310cbb72 1302
bd9e19ca
VM
1303 if (devno == 0)
1304 return -ENODEV;
1305
ab089374 1306 i7core_printk(KERN_INFO,
c77720b9 1307 "Device not found: dev %02x.%d PCI ID %04x:%04x\n",
de06eeef
MCC
1308 dev_descr->dev, dev_descr->func,
1309 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
67166af4 1310
c77720b9
MCC
1311 /* End of list, leave */
1312 return -ENODEV;
1313 }
1314 bus = pdev->bus->number;
67166af4 1315
bda14289 1316 socket = last_bus - bus;
c77720b9 1317
66607706
MCC
1318 i7core_dev = get_i7core_dev(socket);
1319 if (!i7core_dev) {
1320 i7core_dev = kzalloc(sizeof(*i7core_dev), GFP_KERNEL);
1321 if (!i7core_dev)
1322 return -ENOMEM;
de06eeef 1323 i7core_dev->pdev = kzalloc(sizeof(*i7core_dev->pdev) * n_devs,
66607706 1324 GFP_KERNEL);
2a6fae32
AB
1325 if (!i7core_dev->pdev) {
1326 kfree(i7core_dev);
66607706 1327 return -ENOMEM;
2a6fae32 1328 }
66607706 1329 i7core_dev->socket = socket;
de06eeef 1330 i7core_dev->n_devs = n_devs;
66607706 1331 list_add_tail(&i7core_dev->list, &i7core_edac_list);
c77720b9 1332 }
67166af4 1333
66607706 1334 if (i7core_dev->pdev[devno]) {
c77720b9
MCC
1335 i7core_printk(KERN_ERR,
1336 "Duplicated device for "
1337 "dev %02x:%02x.%d PCI ID %04x:%04x\n",
de06eeef
MCC
1338 bus, dev_descr->dev, dev_descr->func,
1339 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
c77720b9
MCC
1340 pci_dev_put(pdev);
1341 return -ENODEV;
1342 }
67166af4 1343
66607706 1344 i7core_dev->pdev[devno] = pdev;
c77720b9
MCC
1345
1346 /* Sanity check */
de06eeef
MCC
1347 if (unlikely(PCI_SLOT(pdev->devfn) != dev_descr->dev ||
1348 PCI_FUNC(pdev->devfn) != dev_descr->func)) {
c77720b9
MCC
1349 i7core_printk(KERN_ERR,
1350 "Device PCI ID %04x:%04x "
1351 "has dev %02x:%02x.%d instead of dev %02x:%02x.%d\n",
de06eeef 1352 PCI_VENDOR_ID_INTEL, dev_descr->dev_id,
c77720b9 1353 bus, PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn),
de06eeef 1354 bus, dev_descr->dev, dev_descr->func);
c77720b9
MCC
1355 return -ENODEV;
1356 }
ef708b53 1357
c77720b9
MCC
1358 /* Be sure that the device is enabled */
1359 if (unlikely(pci_enable_device(pdev) < 0)) {
1360 i7core_printk(KERN_ERR,
1361 "Couldn't enable "
1362 "dev %02x:%02x.%d PCI ID %04x:%04x\n",
de06eeef
MCC
1363 bus, dev_descr->dev, dev_descr->func,
1364 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
c77720b9
MCC
1365 return -ENODEV;
1366 }
ef708b53 1367
d4c27795 1368 debugf0("Detected socket %d dev %02x:%02x.%d PCI ID %04x:%04x\n",
de06eeef
MCC
1369 socket, bus, dev_descr->dev,
1370 dev_descr->func,
1371 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
8f331907 1372
c77720b9 1373 *prev = pdev;
ef708b53 1374
c77720b9
MCC
1375 return 0;
1376}
a0c36a1f 1377
bd9e19ca 1378static int i7core_get_devices(struct pci_id_table *table)
c77720b9 1379{
bda14289 1380 int i, rc, last_bus;
c77720b9 1381 struct pci_dev *pdev = NULL;
bd9e19ca
VM
1382 struct pci_id_descr *dev_descr;
1383
bda14289
MCC
1384 last_bus = i7core_pci_lastbus();
1385
bd9e19ca
VM
1386 while (table && table->descr) {
1387 dev_descr = table->descr;
1388 for (i = 0; i < table->n_devs; i++) {
1389 pdev = NULL;
1390 do {
bda14289
MCC
1391 rc = i7core_get_onedevice(&pdev, i,
1392 &dev_descr[i],
1393 table->n_devs,
1394 last_bus);
bd9e19ca
VM
1395 if (rc < 0) {
1396 if (i == 0) {
1397 i = table->n_devs;
1398 break;
1399 }
1400 i7core_put_all_devices();
1401 return -ENODEV;
1402 }
1403 } while (pdev);
1404 }
1405 table++;
c77720b9 1406 }
66607706 1407
ef708b53 1408 return 0;
bd9e19ca 1409 return 0;
ef708b53
MCC
1410}
1411
f4742949
MCC
1412static int mci_bind_devs(struct mem_ctl_info *mci,
1413 struct i7core_dev *i7core_dev)
ef708b53
MCC
1414{
1415 struct i7core_pvt *pvt = mci->pvt_info;
1416 struct pci_dev *pdev;
f4742949 1417 int i, func, slot;
ef708b53 1418
f4742949
MCC
1419 /* Associates i7core_dev and mci for future usage */
1420 pvt->i7core_dev = i7core_dev;
1421 i7core_dev->mci = mci;
66607706 1422
f4742949 1423 pvt->is_registered = 0;
de06eeef 1424 for (i = 0; i < i7core_dev->n_devs; i++) {
f4742949
MCC
1425 pdev = i7core_dev->pdev[i];
1426 if (!pdev)
66607706
MCC
1427 continue;
1428
f4742949
MCC
1429 func = PCI_FUNC(pdev->devfn);
1430 slot = PCI_SLOT(pdev->devfn);
1431 if (slot == 3) {
1432 if (unlikely(func > MAX_MCR_FUNC))
1433 goto error;
1434 pvt->pci_mcr[func] = pdev;
1435 } else if (likely(slot >= 4 && slot < 4 + NUM_CHANS)) {
1436 if (unlikely(func > MAX_CHAN_FUNC))
ef708b53 1437 goto error;
f4742949
MCC
1438 pvt->pci_ch[slot - 4][func] = pdev;
1439 } else if (!slot && !func)
1440 pvt->pci_noncore = pdev;
1441 else
1442 goto error;
ef708b53 1443
f4742949
MCC
1444 debugf0("Associated fn %d.%d, dev = %p, socket %d\n",
1445 PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn),
1446 pdev, i7core_dev->socket);
14d2c083 1447
f4742949
MCC
1448 if (PCI_SLOT(pdev->devfn) == 3 &&
1449 PCI_FUNC(pdev->devfn) == 2)
1450 pvt->is_registered = 1;
a0c36a1f 1451 }
e9bd2e73 1452
f338d736
MCC
1453 /*
1454 * Add extra nodes to count errors on udimm
1455 * For registered memory, this is not needed, since the counters
1456 * are already displayed at the standard locations
1457 */
1458 if (!pvt->is_registered)
1459 i7core_sysfs_attrs[ARRAY_SIZE(i7core_sysfs_attrs)-2].grp =
1460 &i7core_udimm_counters;
1461
a0c36a1f 1462 return 0;
ef708b53
MCC
1463
1464error:
1465 i7core_printk(KERN_ERR, "Device %d, function %d "
1466 "is out of the expected range\n",
1467 slot, func);
1468 return -EINVAL;
a0c36a1f
MCC
1469}
1470
442305b1
MCC
1471/****************************************************************************
1472 Error check routines
1473 ****************************************************************************/
f4742949 1474static void i7core_rdimm_update_csrow(struct mem_ctl_info *mci,
b4e8f0b6
MCC
1475 int chan, int dimm, int add)
1476{
1477 char *msg;
1478 struct i7core_pvt *pvt = mci->pvt_info;
f4742949 1479 int row = pvt->csrow_map[chan][dimm], i;
b4e8f0b6
MCC
1480
1481 for (i = 0; i < add; i++) {
1482 msg = kasprintf(GFP_KERNEL, "Corrected error "
f4742949
MCC
1483 "(Socket=%d channel=%d dimm=%d)",
1484 pvt->i7core_dev->socket, chan, dimm);
b4e8f0b6
MCC
1485
1486 edac_mc_handle_fbd_ce(mci, row, 0, msg);
1487 kfree (msg);
1488 }
1489}
1490
1491static void i7core_rdimm_update_ce_count(struct mem_ctl_info *mci,
f4742949 1492 int chan, int new0, int new1, int new2)
b4e8f0b6
MCC
1493{
1494 struct i7core_pvt *pvt = mci->pvt_info;
1495 int add0 = 0, add1 = 0, add2 = 0;
1496 /* Updates CE counters if it is not the first time here */
f4742949 1497 if (pvt->ce_count_available) {
b4e8f0b6
MCC
1498 /* Updates CE counters */
1499
f4742949
MCC
1500 add2 = new2 - pvt->rdimm_last_ce_count[chan][2];
1501 add1 = new1 - pvt->rdimm_last_ce_count[chan][1];
1502 add0 = new0 - pvt->rdimm_last_ce_count[chan][0];
b4e8f0b6
MCC
1503
1504 if (add2 < 0)
1505 add2 += 0x7fff;
f4742949 1506 pvt->rdimm_ce_count[chan][2] += add2;
b4e8f0b6
MCC
1507
1508 if (add1 < 0)
1509 add1 += 0x7fff;
f4742949 1510 pvt->rdimm_ce_count[chan][1] += add1;
b4e8f0b6
MCC
1511
1512 if (add0 < 0)
1513 add0 += 0x7fff;
f4742949 1514 pvt->rdimm_ce_count[chan][0] += add0;
b4e8f0b6 1515 } else
f4742949 1516 pvt->ce_count_available = 1;
b4e8f0b6
MCC
1517
1518 /* Store the new values */
f4742949
MCC
1519 pvt->rdimm_last_ce_count[chan][2] = new2;
1520 pvt->rdimm_last_ce_count[chan][1] = new1;
1521 pvt->rdimm_last_ce_count[chan][0] = new0;
b4e8f0b6
MCC
1522
1523 /*updated the edac core */
1524 if (add0 != 0)
f4742949 1525 i7core_rdimm_update_csrow(mci, chan, 0, add0);
b4e8f0b6 1526 if (add1 != 0)
f4742949 1527 i7core_rdimm_update_csrow(mci, chan, 1, add1);
b4e8f0b6 1528 if (add2 != 0)
f4742949 1529 i7core_rdimm_update_csrow(mci, chan, 2, add2);
b4e8f0b6
MCC
1530
1531}
1532
f4742949 1533static void i7core_rdimm_check_mc_ecc_err(struct mem_ctl_info *mci)
b4e8f0b6
MCC
1534{
1535 struct i7core_pvt *pvt = mci->pvt_info;
1536 u32 rcv[3][2];
1537 int i, new0, new1, new2;
1538
1539 /*Read DEV 3: FUN 2: MC_COR_ECC_CNT regs directly*/
f4742949 1540 pci_read_config_dword(pvt->pci_mcr[2], MC_COR_ECC_CNT_0,
b4e8f0b6 1541 &rcv[0][0]);
f4742949 1542 pci_read_config_dword(pvt->pci_mcr[2], MC_COR_ECC_CNT_1,
b4e8f0b6 1543 &rcv[0][1]);
f4742949 1544 pci_read_config_dword(pvt->pci_mcr[2], MC_COR_ECC_CNT_2,
b4e8f0b6 1545 &rcv[1][0]);
f4742949 1546 pci_read_config_dword(pvt->pci_mcr[2], MC_COR_ECC_CNT_3,
b4e8f0b6 1547 &rcv[1][1]);
f4742949 1548 pci_read_config_dword(pvt->pci_mcr[2], MC_COR_ECC_CNT_4,
b4e8f0b6 1549 &rcv[2][0]);
f4742949 1550 pci_read_config_dword(pvt->pci_mcr[2], MC_COR_ECC_CNT_5,
b4e8f0b6
MCC
1551 &rcv[2][1]);
1552 for (i = 0 ; i < 3; i++) {
1553 debugf3("MC_COR_ECC_CNT%d = 0x%x; MC_COR_ECC_CNT%d = 0x%x\n",
1554 (i * 2), rcv[i][0], (i * 2) + 1, rcv[i][1]);
1555 /*if the channel has 3 dimms*/
f4742949 1556 if (pvt->channel[i].dimms > 2) {
b4e8f0b6
MCC
1557 new0 = DIMM_BOT_COR_ERR(rcv[i][0]);
1558 new1 = DIMM_TOP_COR_ERR(rcv[i][0]);
1559 new2 = DIMM_BOT_COR_ERR(rcv[i][1]);
1560 } else {
1561 new0 = DIMM_TOP_COR_ERR(rcv[i][0]) +
1562 DIMM_BOT_COR_ERR(rcv[i][0]);
1563 new1 = DIMM_TOP_COR_ERR(rcv[i][1]) +
1564 DIMM_BOT_COR_ERR(rcv[i][1]);
1565 new2 = 0;
1566 }
1567
f4742949 1568 i7core_rdimm_update_ce_count(mci, i, new0, new1, new2);
b4e8f0b6
MCC
1569 }
1570}
442305b1
MCC
1571
1572/* This function is based on the device 3 function 4 registers as described on:
1573 * Intel Xeon Processor 5500 Series Datasheet Volume 2
1574 * http://www.intel.com/Assets/PDF/datasheet/321322.pdf
1575 * also available at:
1576 * http://www.arrownac.com/manufacturers/intel/s/nehalem/5500-datasheet-v2.pdf
1577 */
f4742949 1578static void i7core_udimm_check_mc_ecc_err(struct mem_ctl_info *mci)
442305b1
MCC
1579{
1580 struct i7core_pvt *pvt = mci->pvt_info;
1581 u32 rcv1, rcv0;
1582 int new0, new1, new2;
1583
f4742949 1584 if (!pvt->pci_mcr[4]) {
b990538a 1585 debugf0("%s MCR registers not found\n", __func__);
442305b1
MCC
1586 return;
1587 }
1588
b4e8f0b6 1589 /* Corrected test errors */
f4742949
MCC
1590 pci_read_config_dword(pvt->pci_mcr[4], MC_TEST_ERR_RCV1, &rcv1);
1591 pci_read_config_dword(pvt->pci_mcr[4], MC_TEST_ERR_RCV0, &rcv0);
442305b1
MCC
1592
1593 /* Store the new values */
1594 new2 = DIMM2_COR_ERR(rcv1);
1595 new1 = DIMM1_COR_ERR(rcv0);
1596 new0 = DIMM0_COR_ERR(rcv0);
1597
442305b1 1598 /* Updates CE counters if it is not the first time here */
f4742949 1599 if (pvt->ce_count_available) {
442305b1
MCC
1600 /* Updates CE counters */
1601 int add0, add1, add2;
1602
f4742949
MCC
1603 add2 = new2 - pvt->udimm_last_ce_count[2];
1604 add1 = new1 - pvt->udimm_last_ce_count[1];
1605 add0 = new0 - pvt->udimm_last_ce_count[0];
442305b1
MCC
1606
1607 if (add2 < 0)
1608 add2 += 0x7fff;
f4742949 1609 pvt->udimm_ce_count[2] += add2;
442305b1
MCC
1610
1611 if (add1 < 0)
1612 add1 += 0x7fff;
f4742949 1613 pvt->udimm_ce_count[1] += add1;
442305b1
MCC
1614
1615 if (add0 < 0)
1616 add0 += 0x7fff;
f4742949 1617 pvt->udimm_ce_count[0] += add0;
b4e8f0b6
MCC
1618
1619 if (add0 | add1 | add2)
1620 i7core_printk(KERN_ERR, "New Corrected error(s): "
1621 "dimm0: +%d, dimm1: +%d, dimm2 +%d\n",
1622 add0, add1, add2);
442305b1 1623 } else
f4742949 1624 pvt->ce_count_available = 1;
442305b1
MCC
1625
1626 /* Store the new values */
f4742949
MCC
1627 pvt->udimm_last_ce_count[2] = new2;
1628 pvt->udimm_last_ce_count[1] = new1;
1629 pvt->udimm_last_ce_count[0] = new0;
442305b1
MCC
1630}
1631
8a2f118e
MCC
1632/*
1633 * According with tables E-11 and E-12 of chapter E.3.3 of Intel 64 and IA-32
1634 * Architectures Software Developer’s Manual Volume 3B.
f237fcf2
MCC
1635 * Nehalem are defined as family 0x06, model 0x1a
1636 *
1637 * The MCA registers used here are the following ones:
8a2f118e 1638 * struct mce field MCA Register
f237fcf2
MCC
1639 * m->status MSR_IA32_MC8_STATUS
1640 * m->addr MSR_IA32_MC8_ADDR
1641 * m->misc MSR_IA32_MC8_MISC
8a2f118e
MCC
1642 * In the case of Nehalem, the error information is masked at .status and .misc
1643 * fields
1644 */
d5381642
MCC
1645static void i7core_mce_output_error(struct mem_ctl_info *mci,
1646 struct mce *m)
1647{
b4e8f0b6 1648 struct i7core_pvt *pvt = mci->pvt_info;
a639539f 1649 char *type, *optype, *err, *msg;
8a2f118e 1650 unsigned long error = m->status & 0x1ff0000l;
a639539f 1651 u32 optypenum = (m->status >> 4) & 0x07;
8a2f118e
MCC
1652 u32 core_err_cnt = (m->status >> 38) && 0x7fff;
1653 u32 dimm = (m->misc >> 16) & 0x3;
1654 u32 channel = (m->misc >> 18) & 0x3;
1655 u32 syndrome = m->misc >> 32;
1656 u32 errnum = find_first_bit(&error, 32);
b4e8f0b6 1657 int csrow;
8a2f118e 1658
c5d34528
MCC
1659 if (m->mcgstatus & 1)
1660 type = "FATAL";
1661 else
1662 type = "NON_FATAL";
1663
a639539f 1664 switch (optypenum) {
b990538a
MCC
1665 case 0:
1666 optype = "generic undef request";
1667 break;
1668 case 1:
1669 optype = "read error";
1670 break;
1671 case 2:
1672 optype = "write error";
1673 break;
1674 case 3:
1675 optype = "addr/cmd error";
1676 break;
1677 case 4:
1678 optype = "scrubbing error";
1679 break;
1680 default:
1681 optype = "reserved";
1682 break;
a639539f
MCC
1683 }
1684
8a2f118e
MCC
1685 switch (errnum) {
1686 case 16:
1687 err = "read ECC error";
1688 break;
1689 case 17:
1690 err = "RAS ECC error";
1691 break;
1692 case 18:
1693 err = "write parity error";
1694 break;
1695 case 19:
1696 err = "redundacy loss";
1697 break;
1698 case 20:
1699 err = "reserved";
1700 break;
1701 case 21:
1702 err = "memory range error";
1703 break;
1704 case 22:
1705 err = "RTID out of range";
1706 break;
1707 case 23:
1708 err = "address parity error";
1709 break;
1710 case 24:
1711 err = "byte enable parity error";
1712 break;
1713 default:
1714 err = "unknown";
d5381642 1715 }
d5381642 1716
f237fcf2 1717 /* FIXME: should convert addr into bank and rank information */
8a2f118e 1718 msg = kasprintf(GFP_ATOMIC,
f4742949 1719 "%s (addr = 0x%08llx, cpu=%d, Dimm=%d, Channel=%d, "
a639539f 1720 "syndrome=0x%08x, count=%d, Err=%08llx:%08llx (%s: %s))\n",
f4742949 1721 type, (long long) m->addr, m->cpu, dimm, channel,
a639539f
MCC
1722 syndrome, core_err_cnt, (long long)m->status,
1723 (long long)m->misc, optype, err);
8a2f118e
MCC
1724
1725 debugf0("%s", msg);
d5381642 1726
f4742949 1727 csrow = pvt->csrow_map[channel][dimm];
b4e8f0b6 1728
d5381642 1729 /* Call the helper to output message */
b4e8f0b6
MCC
1730 if (m->mcgstatus & 1)
1731 edac_mc_handle_fbd_ue(mci, csrow, 0,
1732 0 /* FIXME: should be channel here */, msg);
f4742949 1733 else if (!pvt->is_registered)
b4e8f0b6
MCC
1734 edac_mc_handle_fbd_ce(mci, csrow,
1735 0 /* FIXME: should be channel here */, msg);
8a2f118e
MCC
1736
1737 kfree(msg);
d5381642
MCC
1738}
1739
87d1d272
MCC
1740/*
1741 * i7core_check_error Retrieve and process errors reported by the
1742 * hardware. Called by the Core module.
1743 */
1744static void i7core_check_error(struct mem_ctl_info *mci)
1745{
d5381642
MCC
1746 struct i7core_pvt *pvt = mci->pvt_info;
1747 int i;
1748 unsigned count = 0;
ca9c90ba 1749 struct mce *m;
d5381642 1750
ca9c90ba
MCC
1751 /*
1752 * MCE first step: Copy all mce errors into a temporary buffer
1753 * We use a double buffering here, to reduce the risk of
1754 * loosing an error.
1755 */
1756 smp_rmb();
321ece4d
MCC
1757 count = (pvt->mce_out + MCE_LOG_LEN - pvt->mce_in)
1758 % MCE_LOG_LEN;
ca9c90ba 1759 if (!count)
8a311e17 1760 goto check_ce_error;
f4742949 1761
ca9c90ba 1762 m = pvt->mce_outentry;
321ece4d
MCC
1763 if (pvt->mce_in + count > MCE_LOG_LEN) {
1764 unsigned l = MCE_LOG_LEN - pvt->mce_in;
f4742949 1765
ca9c90ba
MCC
1766 memcpy(m, &pvt->mce_entry[pvt->mce_in], sizeof(*m) * l);
1767 smp_wmb();
1768 pvt->mce_in = 0;
1769 count -= l;
1770 m += l;
1771 }
1772 memcpy(m, &pvt->mce_entry[pvt->mce_in], sizeof(*m) * count);
1773 smp_wmb();
1774 pvt->mce_in += count;
1775
1776 smp_rmb();
1777 if (pvt->mce_overrun) {
1778 i7core_printk(KERN_ERR, "Lost %d memory errors\n",
1779 pvt->mce_overrun);
1780 smp_wmb();
1781 pvt->mce_overrun = 0;
1782 }
d5381642 1783
ca9c90ba
MCC
1784 /*
1785 * MCE second step: parse errors and display
1786 */
d5381642 1787 for (i = 0; i < count; i++)
ca9c90ba 1788 i7core_mce_output_error(mci, &pvt->mce_outentry[i]);
d5381642 1789
ca9c90ba
MCC
1790 /*
1791 * Now, let's increment CE error counts
1792 */
8a311e17 1793check_ce_error:
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
MCC
1800/*
1801 * i7core_mce_check_error Replicates mcelog routine to get errors
1802 * This routine simply queues mcelog errors, and
1803 * return. The error itself should be handled later
1804 * by i7core_check_error.
6e103be1
MCC
1805 * WARNING: As this routine should be called at NMI time, extra care should
1806 * be taken to avoid deadlocks, and to be as fast as possible.
d5381642
MCC
1807 */
1808static int i7core_mce_check_error(void *priv, struct mce *mce)
1809{
c5d34528
MCC
1810 struct mem_ctl_info *mci = priv;
1811 struct i7core_pvt *pvt = mci->pvt_info;
d5381642 1812
8a2f118e
MCC
1813 /*
1814 * Just let mcelog handle it if the error is
1815 * outside the memory controller
1816 */
1817 if (((mce->status & 0xffff) >> 7) != 1)
1818 return 0;
1819
f237fcf2
MCC
1820 /* Bank 8 registers are the only ones that we know how to handle */
1821 if (mce->bank != 8)
1822 return 0;
1823
3b918c12 1824#ifdef CONFIG_SMP
f4742949 1825 /* Only handle if it is the right mc controller */
6e103be1 1826 if (cpu_data(mce->cpu).phys_proc_id != pvt->i7core_dev->socket)
f4742949 1827 return 0;
3b918c12 1828#endif
f4742949 1829
ca9c90ba 1830 smp_rmb();
321ece4d 1831 if ((pvt->mce_out + 1) % MCE_LOG_LEN == pvt->mce_in) {
ca9c90ba
MCC
1832 smp_wmb();
1833 pvt->mce_overrun++;
1834 return 0;
d5381642 1835 }
6e103be1
MCC
1836
1837 /* Copy memory error at the ringbuffer */
1838 memcpy(&pvt->mce_entry[pvt->mce_out], mce, sizeof(*mce));
ca9c90ba 1839 smp_wmb();
321ece4d 1840 pvt->mce_out = (pvt->mce_out + 1) % MCE_LOG_LEN;
d5381642 1841
c5d34528
MCC
1842 /* Handle fatal errors immediately */
1843 if (mce->mcgstatus & 1)
1844 i7core_check_error(mci);
1845
d5381642 1846 /* Advice mcelog that the error were handled */
8a2f118e 1847 return 1;
d5381642
MCC
1848}
1849
f4742949
MCC
1850static int i7core_register_mci(struct i7core_dev *i7core_dev,
1851 int num_channels, int num_csrows)
a0c36a1f
MCC
1852{
1853 struct mem_ctl_info *mci;
1854 struct i7core_pvt *pvt;
ba6c5c62 1855 int csrow = 0;
f4742949 1856 int rc;
a0c36a1f 1857
a0c36a1f 1858 /* allocate a new MC control structure */
d4c27795
MCC
1859 mci = edac_mc_alloc(sizeof(*pvt), num_csrows, num_channels,
1860 i7core_dev->socket);
f4742949
MCC
1861 if (unlikely(!mci))
1862 return -ENOMEM;
a0c36a1f
MCC
1863
1864 debugf0("MC: " __FILE__ ": %s(): mci = %p\n", __func__, mci);
1865
f4742949
MCC
1866 /* record ptr to the generic device */
1867 mci->dev = &i7core_dev->pdev[0]->dev;
1868
a0c36a1f 1869 pvt = mci->pvt_info;
ef708b53 1870 memset(pvt, 0, sizeof(*pvt));
67166af4 1871
41fcb7fe
MCC
1872 /*
1873 * FIXME: how to handle RDDR3 at MCI level? It is possible to have
1874 * Mixed RDDR3/UDDR3 with Nehalem, provided that they are on different
1875 * memory channels
1876 */
1877 mci->mtype_cap = MEM_FLAG_DDR3;
a0c36a1f
MCC
1878 mci->edac_ctl_cap = EDAC_FLAG_NONE;
1879 mci->edac_cap = EDAC_FLAG_NONE;
1880 mci->mod_name = "i7core_edac.c";
1881 mci->mod_ver = I7CORE_REVISION;
f4742949
MCC
1882 mci->ctl_name = kasprintf(GFP_KERNEL, "i7 core #%d",
1883 i7core_dev->socket);
1884 mci->dev_name = pci_name(i7core_dev->pdev[0]);
a0c36a1f 1885 mci->ctl_page_to_phys = NULL;
a5538e53 1886 mci->mc_driver_sysfs_attributes = i7core_sysfs_attrs;
87d1d272
MCC
1887 /* Set the function pointer to an actual operation function */
1888 mci->edac_check = i7core_check_error;
8f331907 1889
ef708b53 1890 /* Store pci devices at mci for faster access */
f4742949 1891 rc = mci_bind_devs(mci, i7core_dev);
41fcb7fe 1892 if (unlikely(rc < 0))
f4742949 1893 goto fail;
ef708b53
MCC
1894
1895 /* Get dimm basic config */
f4742949 1896 get_dimm_config(mci, &csrow);
ef708b53 1897
a0c36a1f 1898 /* add this new MC control structure to EDAC's list of MCs */
b7c76151 1899 if (unlikely(edac_mc_add_mc(mci))) {
a0c36a1f
MCC
1900 debugf0("MC: " __FILE__
1901 ": %s(): failed edac_mc_add_mc()\n", __func__);
1902 /* FIXME: perhaps some code should go here that disables error
1903 * reporting if we just enabled it
1904 */
b7c76151
MCC
1905
1906 rc = -EINVAL;
f4742949 1907 goto fail;
a0c36a1f
MCC
1908 }
1909
1910 /* allocating generic PCI control info */
939747bd 1911 pvt->i7core_pci = edac_pci_create_generic_ctl(&i7core_dev->pdev[0]->dev,
f4742949 1912 EDAC_MOD_STR);
939747bd 1913 if (unlikely(!pvt->i7core_pci)) {
a0c36a1f
MCC
1914 printk(KERN_WARNING
1915 "%s(): Unable to create PCI control\n",
1916 __func__);
1917 printk(KERN_WARNING
1918 "%s(): PCI error report via EDAC not setup\n",
1919 __func__);
1920 }
1921
194a40fe 1922 /* Default error mask is any memory */
ef708b53 1923 pvt->inject.channel = 0;
194a40fe
MCC
1924 pvt->inject.dimm = -1;
1925 pvt->inject.rank = -1;
1926 pvt->inject.bank = -1;
1927 pvt->inject.page = -1;
1928 pvt->inject.col = -1;
1929
d5381642 1930 /* Registers on edac_mce in order to receive memory errors */
c5d34528 1931 pvt->edac_mce.priv = mci;
d5381642 1932 pvt->edac_mce.check_error = i7core_mce_check_error;
d5381642
MCC
1933
1934 rc = edac_mce_register(&pvt->edac_mce);
b990538a 1935 if (unlikely(rc < 0)) {
d5381642
MCC
1936 debugf0("MC: " __FILE__
1937 ": %s(): failed edac_mce_register()\n", __func__);
f4742949
MCC
1938 }
1939
1940fail:
d4d1ef45
TL
1941 if (rc < 0)
1942 edac_mc_free(mci);
f4742949
MCC
1943 return rc;
1944}
1945
1946/*
1947 * i7core_probe Probe for ONE instance of device to see if it is
1948 * present.
1949 * return:
1950 * 0 for FOUND a device
1951 * < 0 for error code
1952 */
2d95d815 1953
f4742949
MCC
1954static int __devinit i7core_probe(struct pci_dev *pdev,
1955 const struct pci_device_id *id)
1956{
f4742949
MCC
1957 int rc;
1958 struct i7core_dev *i7core_dev;
1959
2d95d815
MCC
1960 /* get the pci devices we want to reserve for our use */
1961 mutex_lock(&i7core_edac_lock);
1962
f4742949 1963 /*
d4c27795 1964 * All memory controllers are allocated at the first pass.
f4742949 1965 */
2d95d815
MCC
1966 if (unlikely(probed >= 1)) {
1967 mutex_unlock(&i7core_edac_lock);
f4742949 1968 return -EINVAL;
2d95d815
MCC
1969 }
1970 probed++;
de06eeef 1971
bd9e19ca 1972 rc = i7core_get_devices(pci_dev_table);
f4742949
MCC
1973 if (unlikely(rc < 0))
1974 goto fail0;
1975
1976 list_for_each_entry(i7core_dev, &i7core_edac_list, list) {
1977 int channels;
1978 int csrows;
1979
1980 /* Check the number of active and not disabled channels */
1981 rc = i7core_get_active_channels(i7core_dev->socket,
1982 &channels, &csrows);
1983 if (unlikely(rc < 0))
1984 goto fail1;
1985
d4c27795
MCC
1986 rc = i7core_register_mci(i7core_dev, channels, csrows);
1987 if (unlikely(rc < 0))
1988 goto fail1;
d5381642
MCC
1989 }
1990
ef708b53 1991 i7core_printk(KERN_INFO, "Driver loaded.\n");
8f331907 1992
66607706 1993 mutex_unlock(&i7core_edac_lock);
a0c36a1f
MCC
1994 return 0;
1995
66607706 1996fail1:
13d6e9b6 1997 i7core_put_all_devices();
66607706
MCC
1998fail0:
1999 mutex_unlock(&i7core_edac_lock);
b7c76151 2000 return rc;
a0c36a1f
MCC
2001}
2002
2003/*
2004 * i7core_remove destructor for one instance of device
2005 *
2006 */
2007static void __devexit i7core_remove(struct pci_dev *pdev)
2008{
2009 struct mem_ctl_info *mci;
22e6bcbd 2010 struct i7core_dev *i7core_dev, *tmp;
939747bd 2011 struct i7core_pvt *pvt;
a0c36a1f
MCC
2012
2013 debugf0(__FILE__ ": %s()\n", __func__);
2014
22e6bcbd
MCC
2015 /*
2016 * we have a trouble here: pdev value for removal will be wrong, since
2017 * it will point to the X58 register used to detect that the machine
2018 * is a Nehalem or upper design. However, due to the way several PCI
2019 * devices are grouped together to provide MC functionality, we need
2020 * to use a different method for releasing the devices
2021 */
87d1d272 2022
66607706 2023 mutex_lock(&i7core_edac_lock);
22e6bcbd 2024 list_for_each_entry_safe(i7core_dev, tmp, &i7core_edac_list, list) {
939747bd
MCC
2025 mci = find_mci_by_dev(&i7core_dev->pdev[0]->dev);
2026 if (unlikely(!mci || !mci->pvt_info)) {
2027 i7core_printk(KERN_ERR,
2028 "Couldn't find mci hanler\n");
2029 } else {
2030 pvt = mci->pvt_info;
22e6bcbd 2031 i7core_dev = pvt->i7core_dev;
939747bd
MCC
2032
2033 if (likely(pvt->i7core_pci))
2034 edac_pci_release_generic_ctl(pvt->i7core_pci);
2035 else
2036 i7core_printk(KERN_ERR,
2037 "Couldn't find mem_ctl_info for socket %d\n",
2038 i7core_dev->socket);
2039 pvt->i7core_pci = NULL;
2040
2041 edac_mc_del_mc(&i7core_dev->pdev[0]->dev);
2042
22e6bcbd
MCC
2043 edac_mce_unregister(&pvt->edac_mce);
2044 kfree(mci->ctl_name);
2045 edac_mc_free(mci);
2046 i7core_put_devices(i7core_dev);
22e6bcbd
MCC
2047 }
2048 }
2d95d815
MCC
2049 probed--;
2050
66607706 2051 mutex_unlock(&i7core_edac_lock);
a0c36a1f
MCC
2052}
2053
a0c36a1f
MCC
2054MODULE_DEVICE_TABLE(pci, i7core_pci_tbl);
2055
2056/*
2057 * i7core_driver pci_driver structure for this module
2058 *
2059 */
2060static struct pci_driver i7core_driver = {
2061 .name = "i7core_edac",
2062 .probe = i7core_probe,
2063 .remove = __devexit_p(i7core_remove),
2064 .id_table = i7core_pci_tbl,
2065};
2066
2067/*
2068 * i7core_init Module entry function
2069 * Try to initialize this module for its devices
2070 */
2071static int __init i7core_init(void)
2072{
2073 int pci_rc;
2074
2075 debugf2("MC: " __FILE__ ": %s()\n", __func__);
2076
2077 /* Ensure that the OPSTATE is set correctly for POLL or NMI */
2078 opstate_init();
2079
bd9e19ca 2080 i7core_xeon_pci_fixup(pci_dev_table);
bc2d7245 2081
a0c36a1f
MCC
2082 pci_rc = pci_register_driver(&i7core_driver);
2083
3ef288a9
MCC
2084 if (pci_rc >= 0)
2085 return 0;
2086
2087 i7core_printk(KERN_ERR, "Failed to register device with error %d.\n",
2088 pci_rc);
2089
2090 return pci_rc;
a0c36a1f
MCC
2091}
2092
2093/*
2094 * i7core_exit() Module exit function
2095 * Unregister the driver
2096 */
2097static void __exit i7core_exit(void)
2098{
2099 debugf2("MC: " __FILE__ ": %s()\n", __func__);
2100 pci_unregister_driver(&i7core_driver);
2101}
2102
2103module_init(i7core_init);
2104module_exit(i7core_exit);
2105
2106MODULE_LICENSE("GPL");
2107MODULE_AUTHOR("Mauro Carvalho Chehab <mchehab@redhat.com>");
2108MODULE_AUTHOR("Red Hat Inc. (http://www.redhat.com)");
2109MODULE_DESCRIPTION("MC Driver for Intel i7 Core memory controllers - "
2110 I7CORE_REVISION);
2111
2112module_param(edac_op_state, int, 0444);
2113MODULE_PARM_DESC(edac_op_state, "EDAC Error Reporting state: 0=Poll,1=NMI");