]> bbs.cooldavid.org Git - net-next-2.6.git/blame - drivers/edac/i7core_edac.c
i7core_edac: Properly discover the first QPI device
[net-next-2.6.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
f4742949
MCC
42/*
43 * This is used for Nehalem-EP and Nehalem-EX devices, where the non-core
44 * registers start at bus 255, and are not reported by BIOS.
45 * We currently find devices with only 2 sockets. In order to support more QPI
46 * Quick Path Interconnect, just increment this number.
47 */
48#define MAX_SOCKET_BUSES 2
49
50
a0c36a1f
MCC
51/*
52 * Alter this version for the module when modifications are made
53 */
54#define I7CORE_REVISION " Ver: 1.0.0 " __DATE__
55#define EDAC_MOD_STR "i7core_edac"
56
a0c36a1f
MCC
57/*
58 * Debug macros
59 */
60#define i7core_printk(level, fmt, arg...) \
61 edac_printk(level, "i7core", fmt, ##arg)
62
63#define i7core_mc_printk(mci, level, fmt, arg...) \
64 edac_mc_chipset_printk(mci, level, "i7core", fmt, ##arg)
65
66/*
67 * i7core Memory Controller Registers
68 */
69
e9bd2e73
MCC
70 /* OFFSETS for Device 0 Function 0 */
71
72#define MC_CFG_CONTROL 0x90
73
a0c36a1f
MCC
74 /* OFFSETS for Device 3 Function 0 */
75
76#define MC_CONTROL 0x48
77#define MC_STATUS 0x4c
78#define MC_MAX_DOD 0x64
79
442305b1
MCC
80/*
81 * OFFSETS for Device 3 Function 4, as inicated on Xeon 5500 datasheet:
82 * http://www.arrownac.com/manufacturers/intel/s/nehalem/5500-datasheet-v2.pdf
83 */
84
85#define MC_TEST_ERR_RCV1 0x60
86 #define DIMM2_COR_ERR(r) ((r) & 0x7fff)
87
88#define MC_TEST_ERR_RCV0 0x64
89 #define DIMM1_COR_ERR(r) (((r) >> 16) & 0x7fff)
90 #define DIMM0_COR_ERR(r) ((r) & 0x7fff)
91
b4e8f0b6
MCC
92/* OFFSETS for Device 3 Function 2, as inicated on Xeon 5500 datasheet */
93#define MC_COR_ECC_CNT_0 0x80
94#define MC_COR_ECC_CNT_1 0x84
95#define MC_COR_ECC_CNT_2 0x88
96#define MC_COR_ECC_CNT_3 0x8c
97#define MC_COR_ECC_CNT_4 0x90
98#define MC_COR_ECC_CNT_5 0x94
99
100#define DIMM_TOP_COR_ERR(r) (((r) >> 16) & 0x7fff)
101#define DIMM_BOT_COR_ERR(r) ((r) & 0x7fff)
102
103
a0c36a1f
MCC
104 /* OFFSETS for Devices 4,5 and 6 Function 0 */
105
0b2b7b7e
MCC
106#define MC_CHANNEL_DIMM_INIT_PARAMS 0x58
107 #define THREE_DIMMS_PRESENT (1 << 24)
108 #define SINGLE_QUAD_RANK_PRESENT (1 << 23)
109 #define QUAD_RANK_PRESENT (1 << 22)
110 #define REGISTERED_DIMM (1 << 15)
111
f122a892
MCC
112#define MC_CHANNEL_MAPPER 0x60
113 #define RDLCH(r, ch) ((((r) >> (3 + (ch * 6))) & 0x07) - 1)
114 #define WRLCH(r, ch) ((((r) >> (ch * 6)) & 0x07) - 1)
115
0b2b7b7e
MCC
116#define MC_CHANNEL_RANK_PRESENT 0x7c
117 #define RANK_PRESENT_MASK 0xffff
118
a0c36a1f 119#define MC_CHANNEL_ADDR_MATCH 0xf0
194a40fe
MCC
120#define MC_CHANNEL_ERROR_MASK 0xf8
121#define MC_CHANNEL_ERROR_INJECT 0xfc
122 #define INJECT_ADDR_PARITY 0x10
123 #define INJECT_ECC 0x08
124 #define MASK_CACHELINE 0x06
125 #define MASK_FULL_CACHELINE 0x06
126 #define MASK_MSB32_CACHELINE 0x04
127 #define MASK_LSB32_CACHELINE 0x02
128 #define NO_MASK_CACHELINE 0x00
129 #define REPEAT_EN 0x01
a0c36a1f 130
0b2b7b7e 131 /* OFFSETS for Devices 4,5 and 6 Function 1 */
b990538a 132
0b2b7b7e
MCC
133#define MC_DOD_CH_DIMM0 0x48
134#define MC_DOD_CH_DIMM1 0x4c
135#define MC_DOD_CH_DIMM2 0x50
136 #define RANKOFFSET_MASK ((1 << 12) | (1 << 11) | (1 << 10))
137 #define RANKOFFSET(x) ((x & RANKOFFSET_MASK) >> 10)
138 #define DIMM_PRESENT_MASK (1 << 9)
139 #define DIMM_PRESENT(x) (((x) & DIMM_PRESENT_MASK) >> 9)
854d3349
MCC
140 #define MC_DOD_NUMBANK_MASK ((1 << 8) | (1 << 7))
141 #define MC_DOD_NUMBANK(x) (((x) & MC_DOD_NUMBANK_MASK) >> 7)
142 #define MC_DOD_NUMRANK_MASK ((1 << 6) | (1 << 5))
143 #define MC_DOD_NUMRANK(x) (((x) & MC_DOD_NUMRANK_MASK) >> 5)
41fcb7fe 144 #define MC_DOD_NUMROW_MASK ((1 << 4) | (1 << 3) | (1 << 2))
5566cb7c 145 #define MC_DOD_NUMROW(x) (((x) & MC_DOD_NUMROW_MASK) >> 2)
854d3349
MCC
146 #define MC_DOD_NUMCOL_MASK 3
147 #define MC_DOD_NUMCOL(x) ((x) & MC_DOD_NUMCOL_MASK)
0b2b7b7e 148
f122a892
MCC
149#define MC_RANK_PRESENT 0x7c
150
0b2b7b7e
MCC
151#define MC_SAG_CH_0 0x80
152#define MC_SAG_CH_1 0x84
153#define MC_SAG_CH_2 0x88
154#define MC_SAG_CH_3 0x8c
155#define MC_SAG_CH_4 0x90
156#define MC_SAG_CH_5 0x94
157#define MC_SAG_CH_6 0x98
158#define MC_SAG_CH_7 0x9c
159
160#define MC_RIR_LIMIT_CH_0 0x40
161#define MC_RIR_LIMIT_CH_1 0x44
162#define MC_RIR_LIMIT_CH_2 0x48
163#define MC_RIR_LIMIT_CH_3 0x4C
164#define MC_RIR_LIMIT_CH_4 0x50
165#define MC_RIR_LIMIT_CH_5 0x54
166#define MC_RIR_LIMIT_CH_6 0x58
167#define MC_RIR_LIMIT_CH_7 0x5C
168#define MC_RIR_LIMIT_MASK ((1 << 10) - 1)
169
170#define MC_RIR_WAY_CH 0x80
171 #define MC_RIR_WAY_OFFSET_MASK (((1 << 14) - 1) & ~0x7)
172 #define MC_RIR_WAY_RANK_MASK 0x7
173
a0c36a1f
MCC
174/*
175 * i7core structs
176 */
177
178#define NUM_CHANS 3
442305b1
MCC
179#define MAX_DIMMS 3 /* Max DIMMS per channel */
180#define MAX_MCR_FUNC 4
181#define MAX_CHAN_FUNC 3
a0c36a1f
MCC
182
183struct i7core_info {
184 u32 mc_control;
185 u32 mc_status;
186 u32 max_dod;
f122a892 187 u32 ch_map;
a0c36a1f
MCC
188};
189
194a40fe
MCC
190
191struct i7core_inject {
192 int enable;
193
194 u32 section;
195 u32 type;
196 u32 eccmask;
197
198 /* Error address mask */
199 int channel, dimm, rank, bank, page, col;
200};
201
0b2b7b7e 202struct i7core_channel {
442305b1
MCC
203 u32 ranks;
204 u32 dimms;
0b2b7b7e
MCC
205};
206
8f331907 207struct pci_id_descr {
66607706
MCC
208 int dev;
209 int func;
210 int dev_id;
de06eeef 211 int optional;
8f331907
MCC
212};
213
bd9e19ca
VM
214struct pci_id_table {
215 struct pci_id_descr *descr;
216 int n_devs;
217};
218
f4742949
MCC
219struct i7core_dev {
220 struct list_head list;
221 u8 socket;
222 struct pci_dev **pdev;
de06eeef 223 int n_devs;
f4742949
MCC
224 struct mem_ctl_info *mci;
225};
226
a0c36a1f 227struct i7core_pvt {
f4742949
MCC
228 struct pci_dev *pci_noncore;
229 struct pci_dev *pci_mcr[MAX_MCR_FUNC + 1];
230 struct pci_dev *pci_ch[NUM_CHANS][MAX_CHAN_FUNC + 1];
231
232 struct i7core_dev *i7core_dev;
67166af4 233
a0c36a1f 234 struct i7core_info info;
194a40fe 235 struct i7core_inject inject;
f4742949 236 struct i7core_channel channel[NUM_CHANS];
67166af4 237
f4742949 238 int channels; /* Number of active channels */
442305b1 239
f4742949
MCC
240 int ce_count_available;
241 int csrow_map[NUM_CHANS][MAX_DIMMS];
b4e8f0b6
MCC
242
243 /* ECC corrected errors counts per udimm */
f4742949
MCC
244 unsigned long udimm_ce_count[MAX_DIMMS];
245 int udimm_last_ce_count[MAX_DIMMS];
b4e8f0b6 246 /* ECC corrected errors counts per rdimm */
f4742949
MCC
247 unsigned long rdimm_ce_count[NUM_CHANS][MAX_DIMMS];
248 int rdimm_last_ce_count[NUM_CHANS][MAX_DIMMS];
442305b1 249
f4742949 250 unsigned int is_registered;
14d2c083 251
d5381642
MCC
252 /* mcelog glue */
253 struct edac_mce edac_mce;
ca9c90ba
MCC
254
255 /* Fifo double buffers */
d5381642 256 struct mce mce_entry[MCE_LOG_LEN];
ca9c90ba
MCC
257 struct mce mce_outentry[MCE_LOG_LEN];
258
259 /* Fifo in/out counters */
260 unsigned mce_in, mce_out;
261
262 /* Count indicator to show errors not got */
263 unsigned mce_overrun;
a0c36a1f
MCC
264};
265
66607706
MCC
266/* Static vars */
267static LIST_HEAD(i7core_edac_list);
268static DEFINE_MUTEX(i7core_edac_lock);
a0c36a1f 269
8f331907
MCC
270#define PCI_DESCR(device, function, device_id) \
271 .dev = (device), \
272 .func = (function), \
273 .dev_id = (device_id)
274
bd9e19ca 275struct pci_id_descr pci_dev_descr_i7core_nehalem[] = {
8f331907
MCC
276 /* Memory controller */
277 { PCI_DESCR(3, 0, PCI_DEVICE_ID_INTEL_I7_MCR) },
278 { PCI_DESCR(3, 1, PCI_DEVICE_ID_INTEL_I7_MC_TAD) },
de06eeef
MCC
279 /* Exists only for RDIMM */
280 { PCI_DESCR(3, 2, PCI_DEVICE_ID_INTEL_I7_MC_RAS), .optional = 1 },
8f331907
MCC
281 { PCI_DESCR(3, 4, PCI_DEVICE_ID_INTEL_I7_MC_TEST) },
282
283 /* Channel 0 */
284 { PCI_DESCR(4, 0, PCI_DEVICE_ID_INTEL_I7_MC_CH0_CTRL) },
285 { PCI_DESCR(4, 1, PCI_DEVICE_ID_INTEL_I7_MC_CH0_ADDR) },
286 { PCI_DESCR(4, 2, PCI_DEVICE_ID_INTEL_I7_MC_CH0_RANK) },
287 { PCI_DESCR(4, 3, PCI_DEVICE_ID_INTEL_I7_MC_CH0_TC) },
288
289 /* Channel 1 */
290 { PCI_DESCR(5, 0, PCI_DEVICE_ID_INTEL_I7_MC_CH1_CTRL) },
291 { PCI_DESCR(5, 1, PCI_DEVICE_ID_INTEL_I7_MC_CH1_ADDR) },
292 { PCI_DESCR(5, 2, PCI_DEVICE_ID_INTEL_I7_MC_CH1_RANK) },
293 { PCI_DESCR(5, 3, PCI_DEVICE_ID_INTEL_I7_MC_CH1_TC) },
294
295 /* Channel 2 */
296 { PCI_DESCR(6, 0, PCI_DEVICE_ID_INTEL_I7_MC_CH2_CTRL) },
297 { PCI_DESCR(6, 1, PCI_DEVICE_ID_INTEL_I7_MC_CH2_ADDR) },
298 { PCI_DESCR(6, 2, PCI_DEVICE_ID_INTEL_I7_MC_CH2_RANK) },
299 { PCI_DESCR(6, 3, PCI_DEVICE_ID_INTEL_I7_MC_CH2_TC) },
310cbb72
MCC
300
301 /* Generic Non-core registers */
302 /*
303 * This is the PCI device on i7core and on Xeon 35xx (8086:2c41)
304 * On Xeon 55xx, however, it has a different id (8086:2c40). So,
305 * the probing code needs to test for the other address in case of
306 * failure of this one
307 */
fd382654 308 { PCI_DESCR(0, 0, PCI_DEVICE_ID_INTEL_I7_NONCORE) },
310cbb72 309
a0c36a1f 310};
8f331907 311
52a2e4fc
MCC
312struct pci_id_descr pci_dev_descr_lynnfield[] = {
313 { PCI_DESCR( 3, 0, PCI_DEVICE_ID_INTEL_LYNNFIELD_MCR) },
314 { PCI_DESCR( 3, 1, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_TAD) },
315 { PCI_DESCR( 3, 4, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_TEST) },
316
317 { PCI_DESCR( 4, 0, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH0_CTRL) },
318 { PCI_DESCR( 4, 1, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH0_ADDR) },
319 { PCI_DESCR( 4, 2, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH0_RANK) },
320 { PCI_DESCR( 4, 3, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH0_TC) },
321
508fa179
MCC
322 { PCI_DESCR( 5, 0, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH1_CTRL) },
323 { PCI_DESCR( 5, 1, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH1_ADDR) },
324 { PCI_DESCR( 5, 2, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH1_RANK) },
325 { PCI_DESCR( 5, 3, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH1_TC) },
52a2e4fc 326
f05da2f7
MCC
327 /*
328 * This is the PCI device has an alternate address on some
329 * processors like Core i7 860
330 */
52a2e4fc
MCC
331 { PCI_DESCR( 0, 0, PCI_DEVICE_ID_INTEL_LYNNFIELD_NONCORE) },
332};
333
bd9e19ca
VM
334struct pci_id_descr pci_dev_descr_i7core_westmere[] = {
335 /* Memory controller */
336 { PCI_DESCR(3, 0, PCI_DEVICE_ID_INTEL_LYNNFIELD_MCR_REV2) },
337 { PCI_DESCR(3, 1, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_TAD_REV2) },
338 /* Exists only for RDIMM */
339 { PCI_DESCR(3, 2, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_RAS_REV2), .optional = 1 },
340 { PCI_DESCR(3, 4, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_TEST_REV2) },
341
342 /* Channel 0 */
343 { PCI_DESCR(4, 0, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH0_CTRL_REV2) },
344 { PCI_DESCR(4, 1, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH0_ADDR_REV2) },
345 { PCI_DESCR(4, 2, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH0_RANK_REV2) },
346 { PCI_DESCR(4, 3, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH0_TC_REV2) },
347
348 /* Channel 1 */
349 { PCI_DESCR(5, 0, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH1_CTRL_REV2) },
350 { PCI_DESCR(5, 1, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH1_ADDR_REV2) },
351 { PCI_DESCR(5, 2, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH1_RANK_REV2) },
352 { PCI_DESCR(5, 3, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH1_TC_REV2) },
353
354 /* Channel 2 */
355 { PCI_DESCR(6, 0, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH2_CTRL_REV2) },
356 { PCI_DESCR(6, 1, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH2_ADDR_REV2) },
357 { PCI_DESCR(6, 2, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH2_RANK_REV2) },
358 { PCI_DESCR(6, 3, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH2_TC_REV2) },
359
360 /* Generic Non-core registers */
361 { PCI_DESCR(0, 0, PCI_DEVICE_ID_INTEL_LYNNFIELD_NONCORE_REV2) },
362
363};
364
365#define PCI_ID_TABLE_ENTRY(A) { A, ARRAY_SIZE(A) }
366struct pci_id_table pci_dev_table[] = {
367 PCI_ID_TABLE_ENTRY(pci_dev_descr_i7core_nehalem),
368 PCI_ID_TABLE_ENTRY(pci_dev_descr_lynnfield),
369 PCI_ID_TABLE_ENTRY(pci_dev_descr_i7core_westmere),
370};
371
8f331907
MCC
372/*
373 * pci_device_id table for which devices we are looking for
8f331907
MCC
374 */
375static const struct pci_device_id i7core_pci_tbl[] __devinitdata = {
d1fd4fb6 376 {PCI_DEVICE(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_X58_HUB_MGMT)},
f05da2f7 377 {PCI_DEVICE(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_LYNNFIELD_QPI_LINK0)},
8f331907
MCC
378 {0,} /* 0 terminated list. */
379};
380
a0c36a1f
MCC
381static struct edac_pci_ctl_info *i7core_pci;
382
383/****************************************************************************
384 Anciliary status routines
385 ****************************************************************************/
386
387 /* MC_CONTROL bits */
ef708b53
MCC
388#define CH_ACTIVE(pvt, ch) ((pvt)->info.mc_control & (1 << (8 + ch)))
389#define ECCx8(pvt) ((pvt)->info.mc_control & (1 << 1))
a0c36a1f
MCC
390
391 /* MC_STATUS bits */
61053fde 392#define ECC_ENABLED(pvt) ((pvt)->info.mc_status & (1 << 4))
ef708b53 393#define CH_DISABLED(pvt, ch) ((pvt)->info.mc_status & (1 << ch))
a0c36a1f
MCC
394
395 /* MC_MAX_DOD read functions */
854d3349 396static inline int numdimms(u32 dimms)
a0c36a1f 397{
854d3349 398 return (dimms & 0x3) + 1;
a0c36a1f
MCC
399}
400
854d3349 401static inline int numrank(u32 rank)
a0c36a1f
MCC
402{
403 static int ranks[4] = { 1, 2, 4, -EINVAL };
404
854d3349 405 return ranks[rank & 0x3];
a0c36a1f
MCC
406}
407
854d3349 408static inline int numbank(u32 bank)
a0c36a1f
MCC
409{
410 static int banks[4] = { 4, 8, 16, -EINVAL };
411
854d3349 412 return banks[bank & 0x3];
a0c36a1f
MCC
413}
414
854d3349 415static inline int numrow(u32 row)
a0c36a1f
MCC
416{
417 static int rows[8] = {
418 1 << 12, 1 << 13, 1 << 14, 1 << 15,
419 1 << 16, -EINVAL, -EINVAL, -EINVAL,
420 };
421
854d3349 422 return rows[row & 0x7];
a0c36a1f
MCC
423}
424
854d3349 425static inline int numcol(u32 col)
a0c36a1f
MCC
426{
427 static int cols[8] = {
428 1 << 10, 1 << 11, 1 << 12, -EINVAL,
429 };
854d3349 430 return cols[col & 0x3];
a0c36a1f
MCC
431}
432
f4742949 433static struct i7core_dev *get_i7core_dev(u8 socket)
66607706
MCC
434{
435 struct i7core_dev *i7core_dev;
436
437 list_for_each_entry(i7core_dev, &i7core_edac_list, list) {
438 if (i7core_dev->socket == socket)
439 return i7core_dev;
440 }
441
442 return NULL;
443}
444
a0c36a1f
MCC
445/****************************************************************************
446 Memory check routines
447 ****************************************************************************/
67166af4
MCC
448static struct pci_dev *get_pdev_slot_func(u8 socket, unsigned slot,
449 unsigned func)
ef708b53 450{
66607706 451 struct i7core_dev *i7core_dev = get_i7core_dev(socket);
ef708b53 452 int i;
ef708b53 453
66607706
MCC
454 if (!i7core_dev)
455 return NULL;
456
de06eeef 457 for (i = 0; i < i7core_dev->n_devs; i++) {
66607706 458 if (!i7core_dev->pdev[i])
ef708b53
MCC
459 continue;
460
66607706
MCC
461 if (PCI_SLOT(i7core_dev->pdev[i]->devfn) == slot &&
462 PCI_FUNC(i7core_dev->pdev[i]->devfn) == func) {
463 return i7core_dev->pdev[i];
ef708b53
MCC
464 }
465 }
466
eb94fc40
MCC
467 return NULL;
468}
469
ec6df24c
MCC
470/**
471 * i7core_get_active_channels() - gets the number of channels and csrows
472 * @socket: Quick Path Interconnect socket
473 * @channels: Number of channels that will be returned
474 * @csrows: Number of csrows found
475 *
476 * Since EDAC core needs to know in advance the number of available channels
477 * and csrows, in order to allocate memory for csrows/channels, it is needed
478 * to run two similar steps. At the first step, implemented on this function,
479 * it checks the number of csrows/channels present at one socket.
480 * this is used in order to properly allocate the size of mci components.
481 *
482 * It should be noticed that none of the current available datasheets explain
483 * or even mention how csrows are seen by the memory controller. So, we need
484 * to add a fake description for csrows.
485 * So, this driver is attributing one DIMM memory for one csrow.
486 */
67166af4
MCC
487static int i7core_get_active_channels(u8 socket, unsigned *channels,
488 unsigned *csrows)
eb94fc40
MCC
489{
490 struct pci_dev *pdev = NULL;
491 int i, j;
492 u32 status, control;
493
494 *channels = 0;
495 *csrows = 0;
496
67166af4 497 pdev = get_pdev_slot_func(socket, 3, 0);
b7c76151 498 if (!pdev) {
67166af4
MCC
499 i7core_printk(KERN_ERR, "Couldn't find socket %d fn 3.0!!!\n",
500 socket);
ef708b53 501 return -ENODEV;
b7c76151 502 }
ef708b53
MCC
503
504 /* Device 3 function 0 reads */
505 pci_read_config_dword(pdev, MC_STATUS, &status);
506 pci_read_config_dword(pdev, MC_CONTROL, &control);
507
508 for (i = 0; i < NUM_CHANS; i++) {
eb94fc40 509 u32 dimm_dod[3];
ef708b53
MCC
510 /* Check if the channel is active */
511 if (!(control & (1 << (8 + i))))
512 continue;
513
514 /* Check if the channel is disabled */
41fcb7fe 515 if (status & (1 << i))
ef708b53 516 continue;
ef708b53 517
67166af4 518 pdev = get_pdev_slot_func(socket, i + 4, 1);
eb94fc40 519 if (!pdev) {
67166af4
MCC
520 i7core_printk(KERN_ERR, "Couldn't find socket %d "
521 "fn %d.%d!!!\n",
522 socket, i + 4, 1);
eb94fc40
MCC
523 return -ENODEV;
524 }
525 /* Devices 4-6 function 1 */
526 pci_read_config_dword(pdev,
527 MC_DOD_CH_DIMM0, &dimm_dod[0]);
528 pci_read_config_dword(pdev,
529 MC_DOD_CH_DIMM1, &dimm_dod[1]);
530 pci_read_config_dword(pdev,
531 MC_DOD_CH_DIMM2, &dimm_dod[2]);
532
ef708b53 533 (*channels)++;
eb94fc40
MCC
534
535 for (j = 0; j < 3; j++) {
536 if (!DIMM_PRESENT(dimm_dod[j]))
537 continue;
538 (*csrows)++;
539 }
ef708b53
MCC
540 }
541
c77720b9 542 debugf0("Number of active channels on socket %d: %d\n",
67166af4 543 socket, *channels);
1c6fed80 544
ef708b53
MCC
545 return 0;
546}
547
f4742949 548static int get_dimm_config(struct mem_ctl_info *mci, int *csrow)
a0c36a1f
MCC
549{
550 struct i7core_pvt *pvt = mci->pvt_info;
1c6fed80 551 struct csrow_info *csr;
854d3349 552 struct pci_dev *pdev;
ba6c5c62 553 int i, j;
5566cb7c 554 unsigned long last_page = 0;
1c6fed80 555 enum edac_type mode;
854d3349 556 enum mem_type mtype;
a0c36a1f 557
854d3349 558 /* Get data from the MC register, function 0 */
f4742949 559 pdev = pvt->pci_mcr[0];
7dd6953c 560 if (!pdev)
8f331907
MCC
561 return -ENODEV;
562
f122a892 563 /* Device 3 function 0 reads */
7dd6953c
MCC
564 pci_read_config_dword(pdev, MC_CONTROL, &pvt->info.mc_control);
565 pci_read_config_dword(pdev, MC_STATUS, &pvt->info.mc_status);
566 pci_read_config_dword(pdev, MC_MAX_DOD, &pvt->info.max_dod);
567 pci_read_config_dword(pdev, MC_CHANNEL_MAPPER, &pvt->info.ch_map);
f122a892 568
17cb7b0c 569 debugf0("QPI %d control=0x%08x status=0x%08x dod=0x%08x map=0x%08x\n",
4af91889 570 pvt->i7core_dev->socket, pvt->info.mc_control, pvt->info.mc_status,
f122a892 571 pvt->info.max_dod, pvt->info.ch_map);
a0c36a1f 572
1c6fed80 573 if (ECC_ENABLED(pvt)) {
41fcb7fe 574 debugf0("ECC enabled with x%d SDCC\n", ECCx8(pvt) ? 8 : 4);
1c6fed80
MCC
575 if (ECCx8(pvt))
576 mode = EDAC_S8ECD8ED;
577 else
578 mode = EDAC_S4ECD4ED;
579 } else {
a0c36a1f 580 debugf0("ECC disabled\n");
1c6fed80
MCC
581 mode = EDAC_NONE;
582 }
a0c36a1f
MCC
583
584 /* FIXME: need to handle the error codes */
17cb7b0c
MCC
585 debugf0("DOD Max limits: DIMMS: %d, %d-ranked, %d-banked "
586 "x%x x 0x%x\n",
854d3349
MCC
587 numdimms(pvt->info.max_dod),
588 numrank(pvt->info.max_dod >> 2),
276b824c 589 numbank(pvt->info.max_dod >> 4),
854d3349
MCC
590 numrow(pvt->info.max_dod >> 6),
591 numcol(pvt->info.max_dod >> 9));
a0c36a1f 592
0b2b7b7e 593 for (i = 0; i < NUM_CHANS; i++) {
854d3349 594 u32 data, dimm_dod[3], value[8];
0b2b7b7e 595
52a2e4fc
MCC
596 if (!pvt->pci_ch[i][0])
597 continue;
598
0b2b7b7e
MCC
599 if (!CH_ACTIVE(pvt, i)) {
600 debugf0("Channel %i is not active\n", i);
601 continue;
602 }
603 if (CH_DISABLED(pvt, i)) {
604 debugf0("Channel %i is disabled\n", i);
605 continue;
606 }
607
f122a892 608 /* Devices 4-6 function 0 */
f4742949 609 pci_read_config_dword(pvt->pci_ch[i][0],
0b2b7b7e
MCC
610 MC_CHANNEL_DIMM_INIT_PARAMS, &data);
611
f4742949 612 pvt->channel[i].ranks = (data & QUAD_RANK_PRESENT) ?
67166af4 613 4 : 2;
0b2b7b7e 614
854d3349
MCC
615 if (data & REGISTERED_DIMM)
616 mtype = MEM_RDDR3;
14d2c083 617 else
854d3349
MCC
618 mtype = MEM_DDR3;
619#if 0
0b2b7b7e
MCC
620 if (data & THREE_DIMMS_PRESENT)
621 pvt->channel[i].dimms = 3;
622 else if (data & SINGLE_QUAD_RANK_PRESENT)
623 pvt->channel[i].dimms = 1;
624 else
625 pvt->channel[i].dimms = 2;
854d3349
MCC
626#endif
627
628 /* Devices 4-6 function 1 */
f4742949 629 pci_read_config_dword(pvt->pci_ch[i][1],
854d3349 630 MC_DOD_CH_DIMM0, &dimm_dod[0]);
f4742949 631 pci_read_config_dword(pvt->pci_ch[i][1],
854d3349 632 MC_DOD_CH_DIMM1, &dimm_dod[1]);
f4742949 633 pci_read_config_dword(pvt->pci_ch[i][1],
854d3349 634 MC_DOD_CH_DIMM2, &dimm_dod[2]);
0b2b7b7e 635
1c6fed80 636 debugf0("Ch%d phy rd%d, wr%d (0x%08x): "
854d3349 637 "%d ranks, %cDIMMs\n",
1c6fed80
MCC
638 i,
639 RDLCH(pvt->info.ch_map, i), WRLCH(pvt->info.ch_map, i),
640 data,
f4742949 641 pvt->channel[i].ranks,
41fcb7fe 642 (data & REGISTERED_DIMM) ? 'R' : 'U');
854d3349
MCC
643
644 for (j = 0; j < 3; j++) {
645 u32 banks, ranks, rows, cols;
5566cb7c 646 u32 size, npages;
854d3349
MCC
647
648 if (!DIMM_PRESENT(dimm_dod[j]))
649 continue;
650
651 banks = numbank(MC_DOD_NUMBANK(dimm_dod[j]));
652 ranks = numrank(MC_DOD_NUMRANK(dimm_dod[j]));
653 rows = numrow(MC_DOD_NUMROW(dimm_dod[j]));
654 cols = numcol(MC_DOD_NUMCOL(dimm_dod[j]));
655
5566cb7c
MCC
656 /* DDR3 has 8 I/O banks */
657 size = (rows * cols * banks * ranks) >> (20 - 3);
658
f4742949 659 pvt->channel[i].dimms++;
854d3349 660
17cb7b0c
MCC
661 debugf0("\tdimm %d %d Mb offset: %x, "
662 "bank: %d, rank: %d, row: %#x, col: %#x\n",
663 j, size,
854d3349
MCC
664 RANKOFFSET(dimm_dod[j]),
665 banks, ranks, rows, cols);
666
eb94fc40
MCC
667#if PAGE_SHIFT > 20
668 npages = size >> (PAGE_SHIFT - 20);
669#else
670 npages = size << (20 - PAGE_SHIFT);
671#endif
5566cb7c 672
ba6c5c62 673 csr = &mci->csrows[*csrow];
5566cb7c
MCC
674 csr->first_page = last_page + 1;
675 last_page += npages;
676 csr->last_page = last_page;
677 csr->nr_pages = npages;
678
854d3349 679 csr->page_mask = 0;
eb94fc40 680 csr->grain = 8;
ba6c5c62 681 csr->csrow_idx = *csrow;
eb94fc40
MCC
682 csr->nr_channels = 1;
683
684 csr->channels[0].chan_idx = i;
685 csr->channels[0].ce_count = 0;
854d3349 686
f4742949 687 pvt->csrow_map[i][j] = *csrow;
b4e8f0b6 688
854d3349
MCC
689 switch (banks) {
690 case 4:
691 csr->dtype = DEV_X4;
692 break;
693 case 8:
694 csr->dtype = DEV_X8;
695 break;
696 case 16:
697 csr->dtype = DEV_X16;
698 break;
699 default:
700 csr->dtype = DEV_UNKNOWN;
701 }
702
703 csr->edac_mode = mode;
704 csr->mtype = mtype;
705
ba6c5c62 706 (*csrow)++;
854d3349 707 }
1c6fed80 708
854d3349
MCC
709 pci_read_config_dword(pdev, MC_SAG_CH_0, &value[0]);
710 pci_read_config_dword(pdev, MC_SAG_CH_1, &value[1]);
711 pci_read_config_dword(pdev, MC_SAG_CH_2, &value[2]);
712 pci_read_config_dword(pdev, MC_SAG_CH_3, &value[3]);
713 pci_read_config_dword(pdev, MC_SAG_CH_4, &value[4]);
714 pci_read_config_dword(pdev, MC_SAG_CH_5, &value[5]);
715 pci_read_config_dword(pdev, MC_SAG_CH_6, &value[6]);
716 pci_read_config_dword(pdev, MC_SAG_CH_7, &value[7]);
17cb7b0c 717 debugf1("\t[%i] DIVBY3\tREMOVED\tOFFSET\n", i);
854d3349 718 for (j = 0; j < 8; j++)
17cb7b0c 719 debugf1("\t\t%#x\t%#x\t%#x\n",
854d3349
MCC
720 (value[j] >> 27) & 0x1,
721 (value[j] >> 24) & 0x7,
722 (value[j] && ((1 << 24) - 1)));
0b2b7b7e
MCC
723 }
724
a0c36a1f
MCC
725 return 0;
726}
727
194a40fe
MCC
728/****************************************************************************
729 Error insertion routines
730 ****************************************************************************/
731
732/* The i7core has independent error injection features per channel.
733 However, to have a simpler code, we don't allow enabling error injection
734 on more than one channel.
735 Also, since a change at an inject parameter will be applied only at enable,
736 we're disabling error injection on all write calls to the sysfs nodes that
737 controls the error code injection.
738 */
8f331907 739static int disable_inject(struct mem_ctl_info *mci)
194a40fe
MCC
740{
741 struct i7core_pvt *pvt = mci->pvt_info;
742
743 pvt->inject.enable = 0;
744
f4742949 745 if (!pvt->pci_ch[pvt->inject.channel][0])
8f331907
MCC
746 return -ENODEV;
747
f4742949 748 pci_write_config_dword(pvt->pci_ch[pvt->inject.channel][0],
4157d9f5 749 MC_CHANNEL_ERROR_INJECT, 0);
8f331907
MCC
750
751 return 0;
194a40fe
MCC
752}
753
754/*
755 * i7core inject inject.section
756 *
757 * accept and store error injection inject.section value
758 * bit 0 - refers to the lower 32-byte half cacheline
759 * bit 1 - refers to the upper 32-byte half cacheline
760 */
761static ssize_t i7core_inject_section_store(struct mem_ctl_info *mci,
762 const char *data, size_t count)
763{
764 struct i7core_pvt *pvt = mci->pvt_info;
765 unsigned long value;
766 int rc;
767
768 if (pvt->inject.enable)
41fcb7fe 769 disable_inject(mci);
194a40fe
MCC
770
771 rc = strict_strtoul(data, 10, &value);
772 if ((rc < 0) || (value > 3))
2068def5 773 return -EIO;
194a40fe
MCC
774
775 pvt->inject.section = (u32) value;
776 return count;
777}
778
779static ssize_t i7core_inject_section_show(struct mem_ctl_info *mci,
780 char *data)
781{
782 struct i7core_pvt *pvt = mci->pvt_info;
783 return sprintf(data, "0x%08x\n", pvt->inject.section);
784}
785
786/*
787 * i7core inject.type
788 *
789 * accept and store error injection inject.section value
790 * bit 0 - repeat enable - Enable error repetition
791 * bit 1 - inject ECC error
792 * bit 2 - inject parity error
793 */
794static ssize_t i7core_inject_type_store(struct mem_ctl_info *mci,
795 const char *data, size_t count)
796{
797 struct i7core_pvt *pvt = mci->pvt_info;
798 unsigned long value;
799 int rc;
800
801 if (pvt->inject.enable)
41fcb7fe 802 disable_inject(mci);
194a40fe
MCC
803
804 rc = strict_strtoul(data, 10, &value);
805 if ((rc < 0) || (value > 7))
2068def5 806 return -EIO;
194a40fe
MCC
807
808 pvt->inject.type = (u32) value;
809 return count;
810}
811
812static ssize_t i7core_inject_type_show(struct mem_ctl_info *mci,
813 char *data)
814{
815 struct i7core_pvt *pvt = mci->pvt_info;
816 return sprintf(data, "0x%08x\n", pvt->inject.type);
817}
818
819/*
820 * i7core_inject_inject.eccmask_store
821 *
822 * The type of error (UE/CE) will depend on the inject.eccmask value:
823 * Any bits set to a 1 will flip the corresponding ECC bit
824 * Correctable errors can be injected by flipping 1 bit or the bits within
825 * a symbol pair (2 consecutive aligned 8-bit pairs - i.e. 7:0 and 15:8 or
826 * 23:16 and 31:24). Flipping bits in two symbol pairs will cause an
827 * uncorrectable error to be injected.
828 */
829static ssize_t i7core_inject_eccmask_store(struct mem_ctl_info *mci,
830 const char *data, size_t count)
831{
832 struct i7core_pvt *pvt = mci->pvt_info;
833 unsigned long value;
834 int rc;
835
836 if (pvt->inject.enable)
41fcb7fe 837 disable_inject(mci);
194a40fe
MCC
838
839 rc = strict_strtoul(data, 10, &value);
840 if (rc < 0)
2068def5 841 return -EIO;
194a40fe
MCC
842
843 pvt->inject.eccmask = (u32) value;
844 return count;
845}
846
847static ssize_t i7core_inject_eccmask_show(struct mem_ctl_info *mci,
848 char *data)
849{
850 struct i7core_pvt *pvt = mci->pvt_info;
851 return sprintf(data, "0x%08x\n", pvt->inject.eccmask);
852}
853
854/*
855 * i7core_addrmatch
856 *
857 * The type of error (UE/CE) will depend on the inject.eccmask value:
858 * Any bits set to a 1 will flip the corresponding ECC bit
859 * Correctable errors can be injected by flipping 1 bit or the bits within
860 * a symbol pair (2 consecutive aligned 8-bit pairs - i.e. 7:0 and 15:8 or
861 * 23:16 and 31:24). Flipping bits in two symbol pairs will cause an
862 * uncorrectable error to be injected.
863 */
194a40fe 864
a5538e53
MCC
865#define DECLARE_ADDR_MATCH(param, limit) \
866static ssize_t i7core_inject_store_##param( \
867 struct mem_ctl_info *mci, \
868 const char *data, size_t count) \
869{ \
cc301b3a 870 struct i7core_pvt *pvt; \
a5538e53
MCC
871 long value; \
872 int rc; \
873 \
cc301b3a
MCC
874 debugf1("%s()\n", __func__); \
875 pvt = mci->pvt_info; \
876 \
a5538e53
MCC
877 if (pvt->inject.enable) \
878 disable_inject(mci); \
879 \
4f87fad1 880 if (!strcasecmp(data, "any") || !strcasecmp(data, "any\n"))\
a5538e53
MCC
881 value = -1; \
882 else { \
883 rc = strict_strtoul(data, 10, &value); \
884 if ((rc < 0) || (value >= limit)) \
885 return -EIO; \
886 } \
887 \
888 pvt->inject.param = value; \
889 \
890 return count; \
891} \
892 \
893static ssize_t i7core_inject_show_##param( \
894 struct mem_ctl_info *mci, \
895 char *data) \
896{ \
cc301b3a
MCC
897 struct i7core_pvt *pvt; \
898 \
899 pvt = mci->pvt_info; \
900 debugf1("%s() pvt=%p\n", __func__, pvt); \
a5538e53
MCC
901 if (pvt->inject.param < 0) \
902 return sprintf(data, "any\n"); \
903 else \
904 return sprintf(data, "%d\n", pvt->inject.param);\
194a40fe
MCC
905}
906
a5538e53
MCC
907#define ATTR_ADDR_MATCH(param) \
908 { \
909 .attr = { \
910 .name = #param, \
911 .mode = (S_IRUGO | S_IWUSR) \
912 }, \
913 .show = i7core_inject_show_##param, \
914 .store = i7core_inject_store_##param, \
915 }
194a40fe 916
a5538e53
MCC
917DECLARE_ADDR_MATCH(channel, 3);
918DECLARE_ADDR_MATCH(dimm, 3);
919DECLARE_ADDR_MATCH(rank, 4);
920DECLARE_ADDR_MATCH(bank, 32);
921DECLARE_ADDR_MATCH(page, 0x10000);
922DECLARE_ADDR_MATCH(col, 0x4000);
194a40fe 923
276b824c
MCC
924static int write_and_test(struct pci_dev *dev, int where, u32 val)
925{
926 u32 read;
927 int count;
928
4157d9f5
MCC
929 debugf0("setting pci %02x:%02x.%x reg=%02x value=%08x\n",
930 dev->bus->number, PCI_SLOT(dev->devfn), PCI_FUNC(dev->devfn),
931 where, val);
932
276b824c
MCC
933 for (count = 0; count < 10; count++) {
934 if (count)
b990538a 935 msleep(100);
276b824c
MCC
936 pci_write_config_dword(dev, where, val);
937 pci_read_config_dword(dev, where, &read);
938
939 if (read == val)
940 return 0;
941 }
942
4157d9f5
MCC
943 i7core_printk(KERN_ERR, "Error during set pci %02x:%02x.%x reg=%02x "
944 "write=%08x. Read=%08x\n",
945 dev->bus->number, PCI_SLOT(dev->devfn), PCI_FUNC(dev->devfn),
946 where, val, read);
276b824c
MCC
947
948 return -EINVAL;
949}
950
194a40fe
MCC
951/*
952 * This routine prepares the Memory Controller for error injection.
953 * The error will be injected when some process tries to write to the
954 * memory that matches the given criteria.
955 * The criteria can be set in terms of a mask where dimm, rank, bank, page
956 * and col can be specified.
957 * A -1 value for any of the mask items will make the MCU to ignore
958 * that matching criteria for error injection.
959 *
960 * It should be noticed that the error will only happen after a write operation
961 * on a memory that matches the condition. if REPEAT_EN is not enabled at
962 * inject mask, then it will produce just one error. Otherwise, it will repeat
963 * until the injectmask would be cleaned.
964 *
965 * FIXME: This routine assumes that MAXNUMDIMMS value of MC_MAX_DOD
966 * is reliable enough to check if the MC is using the
967 * three channels. However, this is not clear at the datasheet.
968 */
969static ssize_t i7core_inject_enable_store(struct mem_ctl_info *mci,
970 const char *data, size_t count)
971{
972 struct i7core_pvt *pvt = mci->pvt_info;
973 u32 injectmask;
974 u64 mask = 0;
975 int rc;
976 long enable;
977
f4742949 978 if (!pvt->pci_ch[pvt->inject.channel][0])
8f331907
MCC
979 return 0;
980
194a40fe
MCC
981 rc = strict_strtoul(data, 10, &enable);
982 if ((rc < 0))
983 return 0;
984
985 if (enable) {
986 pvt->inject.enable = 1;
987 } else {
988 disable_inject(mci);
989 return count;
990 }
991
992 /* Sets pvt->inject.dimm mask */
993 if (pvt->inject.dimm < 0)
486dd09f 994 mask |= 1LL << 41;
194a40fe 995 else {
f4742949 996 if (pvt->channel[pvt->inject.channel].dimms > 2)
486dd09f 997 mask |= (pvt->inject.dimm & 0x3LL) << 35;
194a40fe 998 else
486dd09f 999 mask |= (pvt->inject.dimm & 0x1LL) << 36;
194a40fe
MCC
1000 }
1001
1002 /* Sets pvt->inject.rank mask */
1003 if (pvt->inject.rank < 0)
486dd09f 1004 mask |= 1LL << 40;
194a40fe 1005 else {
f4742949 1006 if (pvt->channel[pvt->inject.channel].dimms > 2)
486dd09f 1007 mask |= (pvt->inject.rank & 0x1LL) << 34;
194a40fe 1008 else
486dd09f 1009 mask |= (pvt->inject.rank & 0x3LL) << 34;
194a40fe
MCC
1010 }
1011
1012 /* Sets pvt->inject.bank mask */
1013 if (pvt->inject.bank < 0)
486dd09f 1014 mask |= 1LL << 39;
194a40fe 1015 else
486dd09f 1016 mask |= (pvt->inject.bank & 0x15LL) << 30;
194a40fe
MCC
1017
1018 /* Sets pvt->inject.page mask */
1019 if (pvt->inject.page < 0)
486dd09f 1020 mask |= 1LL << 38;
194a40fe 1021 else
486dd09f 1022 mask |= (pvt->inject.page & 0xffff) << 14;
194a40fe
MCC
1023
1024 /* Sets pvt->inject.column mask */
1025 if (pvt->inject.col < 0)
486dd09f 1026 mask |= 1LL << 37;
194a40fe 1027 else
486dd09f 1028 mask |= (pvt->inject.col & 0x3fff);
194a40fe 1029
276b824c
MCC
1030 /*
1031 * bit 0: REPEAT_EN
1032 * bits 1-2: MASK_HALF_CACHELINE
1033 * bit 3: INJECT_ECC
1034 * bit 4: INJECT_ADDR_PARITY
1035 */
1036
1037 injectmask = (pvt->inject.type & 1) |
1038 (pvt->inject.section & 0x3) << 1 |
1039 (pvt->inject.type & 0x6) << (3 - 1);
1040
1041 /* Unlock writes to registers - this register is write only */
f4742949 1042 pci_write_config_dword(pvt->pci_noncore,
67166af4 1043 MC_CFG_CONTROL, 0x2);
e9bd2e73 1044
f4742949 1045 write_and_test(pvt->pci_ch[pvt->inject.channel][0],
194a40fe 1046 MC_CHANNEL_ADDR_MATCH, mask);
f4742949 1047 write_and_test(pvt->pci_ch[pvt->inject.channel][0],
7b029d03 1048 MC_CHANNEL_ADDR_MATCH + 4, mask >> 32L);
7b029d03 1049
f4742949 1050 write_and_test(pvt->pci_ch[pvt->inject.channel][0],
194a40fe
MCC
1051 MC_CHANNEL_ERROR_MASK, pvt->inject.eccmask);
1052
f4742949 1053 write_and_test(pvt->pci_ch[pvt->inject.channel][0],
4157d9f5 1054 MC_CHANNEL_ERROR_INJECT, injectmask);
276b824c 1055
194a40fe 1056 /*
276b824c
MCC
1057 * This is something undocumented, based on my tests
1058 * Without writing 8 to this register, errors aren't injected. Not sure
1059 * why.
194a40fe 1060 */
f4742949 1061 pci_write_config_dword(pvt->pci_noncore,
276b824c 1062 MC_CFG_CONTROL, 8);
194a40fe 1063
41fcb7fe
MCC
1064 debugf0("Error inject addr match 0x%016llx, ecc 0x%08x,"
1065 " inject 0x%08x\n",
194a40fe
MCC
1066 mask, pvt->inject.eccmask, injectmask);
1067
7b029d03 1068
194a40fe
MCC
1069 return count;
1070}
1071
1072static ssize_t i7core_inject_enable_show(struct mem_ctl_info *mci,
1073 char *data)
1074{
1075 struct i7core_pvt *pvt = mci->pvt_info;
7b029d03
MCC
1076 u32 injectmask;
1077
52a2e4fc
MCC
1078 if (!pvt->pci_ch[pvt->inject.channel][0])
1079 return 0;
1080
f4742949 1081 pci_read_config_dword(pvt->pci_ch[pvt->inject.channel][0],
4157d9f5 1082 MC_CHANNEL_ERROR_INJECT, &injectmask);
7b029d03
MCC
1083
1084 debugf0("Inject error read: 0x%018x\n", injectmask);
1085
1086 if (injectmask & 0x0c)
1087 pvt->inject.enable = 1;
1088
194a40fe
MCC
1089 return sprintf(data, "%d\n", pvt->inject.enable);
1090}
1091
f338d736
MCC
1092#define DECLARE_COUNTER(param) \
1093static ssize_t i7core_show_counter_##param( \
1094 struct mem_ctl_info *mci, \
1095 char *data) \
1096{ \
1097 struct i7core_pvt *pvt = mci->pvt_info; \
1098 \
1099 debugf1("%s() \n", __func__); \
1100 if (!pvt->ce_count_available || (pvt->is_registered)) \
1101 return sprintf(data, "data unavailable\n"); \
1102 return sprintf(data, "%lu\n", \
1103 pvt->udimm_ce_count[param]); \
1104}
442305b1 1105
f338d736
MCC
1106#define ATTR_COUNTER(param) \
1107 { \
1108 .attr = { \
1109 .name = __stringify(udimm##param), \
1110 .mode = (S_IRUGO | S_IWUSR) \
1111 }, \
1112 .show = i7core_show_counter_##param \
d88b8507 1113 }
442305b1 1114
f338d736
MCC
1115DECLARE_COUNTER(0);
1116DECLARE_COUNTER(1);
1117DECLARE_COUNTER(2);
442305b1 1118
194a40fe
MCC
1119/*
1120 * Sysfs struct
1121 */
a5538e53
MCC
1122
1123
1124static struct mcidev_sysfs_attribute i7core_addrmatch_attrs[] = {
1125 ATTR_ADDR_MATCH(channel),
1126 ATTR_ADDR_MATCH(dimm),
1127 ATTR_ADDR_MATCH(rank),
1128 ATTR_ADDR_MATCH(bank),
1129 ATTR_ADDR_MATCH(page),
1130 ATTR_ADDR_MATCH(col),
1131 { .attr = { .name = NULL } }
1132};
1133
a5538e53
MCC
1134static struct mcidev_sysfs_group i7core_inject_addrmatch = {
1135 .name = "inject_addrmatch",
1136 .mcidev_attr = i7core_addrmatch_attrs,
1137};
1138
f338d736
MCC
1139static struct mcidev_sysfs_attribute i7core_udimm_counters_attrs[] = {
1140 ATTR_COUNTER(0),
1141 ATTR_COUNTER(1),
1142 ATTR_COUNTER(2),
1143};
1144
1145static struct mcidev_sysfs_group i7core_udimm_counters = {
1146 .name = "all_channel_counts",
1147 .mcidev_attr = i7core_udimm_counters_attrs,
1148};
1149
a5538e53 1150static struct mcidev_sysfs_attribute i7core_sysfs_attrs[] = {
194a40fe
MCC
1151 {
1152 .attr = {
1153 .name = "inject_section",
1154 .mode = (S_IRUGO | S_IWUSR)
1155 },
1156 .show = i7core_inject_section_show,
1157 .store = i7core_inject_section_store,
1158 }, {
1159 .attr = {
1160 .name = "inject_type",
1161 .mode = (S_IRUGO | S_IWUSR)
1162 },
1163 .show = i7core_inject_type_show,
1164 .store = i7core_inject_type_store,
1165 }, {
1166 .attr = {
1167 .name = "inject_eccmask",
1168 .mode = (S_IRUGO | S_IWUSR)
1169 },
1170 .show = i7core_inject_eccmask_show,
1171 .store = i7core_inject_eccmask_store,
1172 }, {
a5538e53 1173 .grp = &i7core_inject_addrmatch,
194a40fe
MCC
1174 }, {
1175 .attr = {
1176 .name = "inject_enable",
1177 .mode = (S_IRUGO | S_IWUSR)
1178 },
1179 .show = i7core_inject_enable_show,
1180 .store = i7core_inject_enable_store,
1181 },
f338d736 1182 { .attr = { .name = NULL } }, /* Reserved for udimm counters */
42538680 1183 { .attr = { .name = NULL } }
194a40fe
MCC
1184};
1185
a0c36a1f
MCC
1186/****************************************************************************
1187 Device initialization routines: put/get, init/exit
1188 ****************************************************************************/
1189
1190/*
1191 * i7core_put_devices 'put' all the devices that we have
1192 * reserved via 'get'
1193 */
13d6e9b6 1194static void i7core_put_devices(struct i7core_dev *i7core_dev)
a0c36a1f 1195{
13d6e9b6 1196 int i;
a0c36a1f 1197
22e6bcbd 1198 debugf0(__FILE__ ": %s()\n", __func__);
de06eeef 1199 for (i = 0; i < i7core_dev->n_devs; i++) {
22e6bcbd
MCC
1200 struct pci_dev *pdev = i7core_dev->pdev[i];
1201 if (!pdev)
1202 continue;
1203 debugf0("Removing dev %02x:%02x.%d\n",
1204 pdev->bus->number,
1205 PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn));
1206 pci_dev_put(pdev);
1207 }
13d6e9b6 1208 kfree(i7core_dev->pdev);
22e6bcbd 1209 list_del(&i7core_dev->list);
13d6e9b6
MCC
1210 kfree(i7core_dev);
1211}
66607706 1212
13d6e9b6
MCC
1213static void i7core_put_all_devices(void)
1214{
42538680 1215 struct i7core_dev *i7core_dev, *tmp;
13d6e9b6 1216
42538680 1217 list_for_each_entry_safe(i7core_dev, tmp, &i7core_edac_list, list)
13d6e9b6 1218 i7core_put_devices(i7core_dev);
a0c36a1f
MCC
1219}
1220
bd9e19ca 1221static void __init i7core_xeon_pci_fixup(struct pci_id_table *table)
bc2d7245
KM
1222{
1223 struct pci_dev *pdev = NULL;
1224 int i;
1225 /*
1226 * On Xeon 55xx, the Intel Quckpath Arch Generic Non-core pci buses
1227 * aren't announced by acpi. So, we need to use a legacy scan probing
1228 * to detect them
1229 */
bd9e19ca
VM
1230 while (table && table->descr) {
1231 pdev = pci_get_device(PCI_VENDOR_ID_INTEL, table->descr[0].dev_id, NULL);
1232 if (unlikely(!pdev)) {
1233 for (i = 0; i < MAX_SOCKET_BUSES; i++)
1234 pcibios_scan_specific_bus(255-i);
1235 }
bda14289 1236 pci_dev_put(pdev);
bd9e19ca 1237 table++;
bc2d7245
KM
1238 }
1239}
1240
bda14289
MCC
1241static unsigned i7core_pci_lastbus(void)
1242{
1243 int last_bus = 0, bus;
1244 struct pci_bus *b = NULL;
1245
1246 while ((b = pci_find_next_bus(b)) != NULL) {
1247 bus = b->number;
1248 debugf0("Found bus %d\n", bus);
1249 if (bus > last_bus)
1250 last_bus = bus;
1251 }
1252
1253 debugf0("Last bus %d\n", last_bus);
1254
1255 return last_bus;
1256}
1257
a0c36a1f
MCC
1258/*
1259 * i7core_get_devices Find and perform 'get' operation on the MCH's
1260 * device/functions we want to reference for this driver
1261 *
1262 * Need to 'get' device 16 func 1 and func 2
1263 */
de06eeef 1264int i7core_get_onedevice(struct pci_dev **prev, int devno,
bda14289
MCC
1265 struct pci_id_descr *dev_descr, unsigned n_devs,
1266 unsigned last_bus)
a0c36a1f 1267{
66607706
MCC
1268 struct i7core_dev *i7core_dev;
1269
8f331907 1270 struct pci_dev *pdev = NULL;
67166af4
MCC
1271 u8 bus = 0;
1272 u8 socket = 0;
a0c36a1f 1273
c77720b9 1274 pdev = pci_get_device(PCI_VENDOR_ID_INTEL,
de06eeef 1275 dev_descr->dev_id, *prev);
c77720b9 1276
c77720b9
MCC
1277 /*
1278 * On Xeon 55xx, the Intel Quckpath Arch Generic Non-core regs
1279 * is at addr 8086:2c40, instead of 8086:2c41. So, we need
1280 * to probe for the alternate address in case of failure
1281 */
de06eeef 1282 if (dev_descr->dev_id == PCI_DEVICE_ID_INTEL_I7_NONCORE && !pdev)
c77720b9 1283 pdev = pci_get_device(PCI_VENDOR_ID_INTEL,
fd382654 1284 PCI_DEVICE_ID_INTEL_I7_NONCORE_ALT, *prev);
d1fd4fb6 1285
bd9e19ca 1286 if (dev_descr->dev_id == PCI_DEVICE_ID_INTEL_LYNNFIELD_NONCORE && !pdev)
f05da2f7
MCC
1287 pdev = pci_get_device(PCI_VENDOR_ID_INTEL,
1288 PCI_DEVICE_ID_INTEL_LYNNFIELD_NONCORE_ALT,
1289 *prev);
1290
c77720b9
MCC
1291 if (!pdev) {
1292 if (*prev) {
1293 *prev = pdev;
1294 return 0;
d1fd4fb6
MCC
1295 }
1296
de06eeef 1297 if (dev_descr->optional)
c77720b9 1298 return 0;
310cbb72 1299
bd9e19ca
VM
1300 if (devno == 0)
1301 return -ENODEV;
1302
c77720b9
MCC
1303 i7core_printk(KERN_ERR,
1304 "Device not found: dev %02x.%d PCI ID %04x:%04x\n",
de06eeef
MCC
1305 dev_descr->dev, dev_descr->func,
1306 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
67166af4 1307
c77720b9
MCC
1308 /* End of list, leave */
1309 return -ENODEV;
1310 }
1311 bus = pdev->bus->number;
67166af4 1312
bda14289 1313 socket = last_bus - bus;
c77720b9 1314
66607706
MCC
1315 i7core_dev = get_i7core_dev(socket);
1316 if (!i7core_dev) {
1317 i7core_dev = kzalloc(sizeof(*i7core_dev), GFP_KERNEL);
1318 if (!i7core_dev)
1319 return -ENOMEM;
de06eeef 1320 i7core_dev->pdev = kzalloc(sizeof(*i7core_dev->pdev) * n_devs,
66607706 1321 GFP_KERNEL);
2a6fae32
AB
1322 if (!i7core_dev->pdev) {
1323 kfree(i7core_dev);
66607706 1324 return -ENOMEM;
2a6fae32 1325 }
66607706 1326 i7core_dev->socket = socket;
de06eeef 1327 i7core_dev->n_devs = n_devs;
66607706 1328 list_add_tail(&i7core_dev->list, &i7core_edac_list);
c77720b9 1329 }
67166af4 1330
66607706 1331 if (i7core_dev->pdev[devno]) {
c77720b9
MCC
1332 i7core_printk(KERN_ERR,
1333 "Duplicated device for "
1334 "dev %02x:%02x.%d PCI ID %04x:%04x\n",
de06eeef
MCC
1335 bus, dev_descr->dev, dev_descr->func,
1336 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
c77720b9
MCC
1337 pci_dev_put(pdev);
1338 return -ENODEV;
1339 }
67166af4 1340
66607706 1341 i7core_dev->pdev[devno] = pdev;
c77720b9
MCC
1342
1343 /* Sanity check */
de06eeef
MCC
1344 if (unlikely(PCI_SLOT(pdev->devfn) != dev_descr->dev ||
1345 PCI_FUNC(pdev->devfn) != dev_descr->func)) {
c77720b9
MCC
1346 i7core_printk(KERN_ERR,
1347 "Device PCI ID %04x:%04x "
1348 "has dev %02x:%02x.%d instead of dev %02x:%02x.%d\n",
de06eeef 1349 PCI_VENDOR_ID_INTEL, dev_descr->dev_id,
c77720b9 1350 bus, PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn),
de06eeef 1351 bus, dev_descr->dev, dev_descr->func);
c77720b9
MCC
1352 return -ENODEV;
1353 }
ef708b53 1354
c77720b9
MCC
1355 /* Be sure that the device is enabled */
1356 if (unlikely(pci_enable_device(pdev) < 0)) {
1357 i7core_printk(KERN_ERR,
1358 "Couldn't enable "
1359 "dev %02x:%02x.%d PCI ID %04x:%04x\n",
de06eeef
MCC
1360 bus, dev_descr->dev, dev_descr->func,
1361 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
c77720b9
MCC
1362 return -ENODEV;
1363 }
ef708b53 1364
d4c27795 1365 debugf0("Detected socket %d dev %02x:%02x.%d PCI ID %04x:%04x\n",
de06eeef
MCC
1366 socket, bus, dev_descr->dev,
1367 dev_descr->func,
1368 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
8f331907 1369
c77720b9 1370 *prev = pdev;
ef708b53 1371
c77720b9
MCC
1372 return 0;
1373}
a0c36a1f 1374
bd9e19ca 1375static int i7core_get_devices(struct pci_id_table *table)
c77720b9 1376{
bda14289 1377 int i, rc, last_bus;
c77720b9 1378 struct pci_dev *pdev = NULL;
bd9e19ca
VM
1379 struct pci_id_descr *dev_descr;
1380
bda14289
MCC
1381 last_bus = i7core_pci_lastbus();
1382
bd9e19ca
VM
1383 while (table && table->descr) {
1384 dev_descr = table->descr;
1385 for (i = 0; i < table->n_devs; i++) {
1386 pdev = NULL;
1387 do {
bda14289
MCC
1388 rc = i7core_get_onedevice(&pdev, i,
1389 &dev_descr[i],
1390 table->n_devs,
1391 last_bus);
bd9e19ca
VM
1392 if (rc < 0) {
1393 if (i == 0) {
1394 i = table->n_devs;
1395 break;
1396 }
1397 i7core_put_all_devices();
1398 return -ENODEV;
1399 }
1400 } while (pdev);
1401 }
1402 table++;
c77720b9 1403 }
66607706 1404
ef708b53 1405 return 0;
bd9e19ca 1406 return 0;
ef708b53
MCC
1407}
1408
f4742949
MCC
1409static int mci_bind_devs(struct mem_ctl_info *mci,
1410 struct i7core_dev *i7core_dev)
ef708b53
MCC
1411{
1412 struct i7core_pvt *pvt = mci->pvt_info;
1413 struct pci_dev *pdev;
f4742949 1414 int i, func, slot;
ef708b53 1415
f4742949
MCC
1416 /* Associates i7core_dev and mci for future usage */
1417 pvt->i7core_dev = i7core_dev;
1418 i7core_dev->mci = mci;
66607706 1419
f4742949 1420 pvt->is_registered = 0;
de06eeef 1421 for (i = 0; i < i7core_dev->n_devs; i++) {
f4742949
MCC
1422 pdev = i7core_dev->pdev[i];
1423 if (!pdev)
66607706
MCC
1424 continue;
1425
f4742949
MCC
1426 func = PCI_FUNC(pdev->devfn);
1427 slot = PCI_SLOT(pdev->devfn);
1428 if (slot == 3) {
1429 if (unlikely(func > MAX_MCR_FUNC))
1430 goto error;
1431 pvt->pci_mcr[func] = pdev;
1432 } else if (likely(slot >= 4 && slot < 4 + NUM_CHANS)) {
1433 if (unlikely(func > MAX_CHAN_FUNC))
ef708b53 1434 goto error;
f4742949
MCC
1435 pvt->pci_ch[slot - 4][func] = pdev;
1436 } else if (!slot && !func)
1437 pvt->pci_noncore = pdev;
1438 else
1439 goto error;
ef708b53 1440
f4742949
MCC
1441 debugf0("Associated fn %d.%d, dev = %p, socket %d\n",
1442 PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn),
1443 pdev, i7core_dev->socket);
14d2c083 1444
f4742949
MCC
1445 if (PCI_SLOT(pdev->devfn) == 3 &&
1446 PCI_FUNC(pdev->devfn) == 2)
1447 pvt->is_registered = 1;
a0c36a1f 1448 }
e9bd2e73 1449
f338d736
MCC
1450 /*
1451 * Add extra nodes to count errors on udimm
1452 * For registered memory, this is not needed, since the counters
1453 * are already displayed at the standard locations
1454 */
1455 if (!pvt->is_registered)
1456 i7core_sysfs_attrs[ARRAY_SIZE(i7core_sysfs_attrs)-2].grp =
1457 &i7core_udimm_counters;
1458
a0c36a1f 1459 return 0;
ef708b53
MCC
1460
1461error:
1462 i7core_printk(KERN_ERR, "Device %d, function %d "
1463 "is out of the expected range\n",
1464 slot, func);
1465 return -EINVAL;
a0c36a1f
MCC
1466}
1467
442305b1
MCC
1468/****************************************************************************
1469 Error check routines
1470 ****************************************************************************/
f4742949 1471static void i7core_rdimm_update_csrow(struct mem_ctl_info *mci,
b4e8f0b6
MCC
1472 int chan, int dimm, int add)
1473{
1474 char *msg;
1475 struct i7core_pvt *pvt = mci->pvt_info;
f4742949 1476 int row = pvt->csrow_map[chan][dimm], i;
b4e8f0b6
MCC
1477
1478 for (i = 0; i < add; i++) {
1479 msg = kasprintf(GFP_KERNEL, "Corrected error "
f4742949
MCC
1480 "(Socket=%d channel=%d dimm=%d)",
1481 pvt->i7core_dev->socket, chan, dimm);
b4e8f0b6
MCC
1482
1483 edac_mc_handle_fbd_ce(mci, row, 0, msg);
1484 kfree (msg);
1485 }
1486}
1487
1488static void i7core_rdimm_update_ce_count(struct mem_ctl_info *mci,
f4742949 1489 int chan, int new0, int new1, int new2)
b4e8f0b6
MCC
1490{
1491 struct i7core_pvt *pvt = mci->pvt_info;
1492 int add0 = 0, add1 = 0, add2 = 0;
1493 /* Updates CE counters if it is not the first time here */
f4742949 1494 if (pvt->ce_count_available) {
b4e8f0b6
MCC
1495 /* Updates CE counters */
1496
f4742949
MCC
1497 add2 = new2 - pvt->rdimm_last_ce_count[chan][2];
1498 add1 = new1 - pvt->rdimm_last_ce_count[chan][1];
1499 add0 = new0 - pvt->rdimm_last_ce_count[chan][0];
b4e8f0b6
MCC
1500
1501 if (add2 < 0)
1502 add2 += 0x7fff;
f4742949 1503 pvt->rdimm_ce_count[chan][2] += add2;
b4e8f0b6
MCC
1504
1505 if (add1 < 0)
1506 add1 += 0x7fff;
f4742949 1507 pvt->rdimm_ce_count[chan][1] += add1;
b4e8f0b6
MCC
1508
1509 if (add0 < 0)
1510 add0 += 0x7fff;
f4742949 1511 pvt->rdimm_ce_count[chan][0] += add0;
b4e8f0b6 1512 } else
f4742949 1513 pvt->ce_count_available = 1;
b4e8f0b6
MCC
1514
1515 /* Store the new values */
f4742949
MCC
1516 pvt->rdimm_last_ce_count[chan][2] = new2;
1517 pvt->rdimm_last_ce_count[chan][1] = new1;
1518 pvt->rdimm_last_ce_count[chan][0] = new0;
b4e8f0b6
MCC
1519
1520 /*updated the edac core */
1521 if (add0 != 0)
f4742949 1522 i7core_rdimm_update_csrow(mci, chan, 0, add0);
b4e8f0b6 1523 if (add1 != 0)
f4742949 1524 i7core_rdimm_update_csrow(mci, chan, 1, add1);
b4e8f0b6 1525 if (add2 != 0)
f4742949 1526 i7core_rdimm_update_csrow(mci, chan, 2, add2);
b4e8f0b6
MCC
1527
1528}
1529
f4742949 1530static void i7core_rdimm_check_mc_ecc_err(struct mem_ctl_info *mci)
b4e8f0b6
MCC
1531{
1532 struct i7core_pvt *pvt = mci->pvt_info;
1533 u32 rcv[3][2];
1534 int i, new0, new1, new2;
1535
1536 /*Read DEV 3: FUN 2: MC_COR_ECC_CNT regs directly*/
f4742949 1537 pci_read_config_dword(pvt->pci_mcr[2], MC_COR_ECC_CNT_0,
b4e8f0b6 1538 &rcv[0][0]);
f4742949 1539 pci_read_config_dword(pvt->pci_mcr[2], MC_COR_ECC_CNT_1,
b4e8f0b6 1540 &rcv[0][1]);
f4742949 1541 pci_read_config_dword(pvt->pci_mcr[2], MC_COR_ECC_CNT_2,
b4e8f0b6 1542 &rcv[1][0]);
f4742949 1543 pci_read_config_dword(pvt->pci_mcr[2], MC_COR_ECC_CNT_3,
b4e8f0b6 1544 &rcv[1][1]);
f4742949 1545 pci_read_config_dword(pvt->pci_mcr[2], MC_COR_ECC_CNT_4,
b4e8f0b6 1546 &rcv[2][0]);
f4742949 1547 pci_read_config_dword(pvt->pci_mcr[2], MC_COR_ECC_CNT_5,
b4e8f0b6
MCC
1548 &rcv[2][1]);
1549 for (i = 0 ; i < 3; i++) {
1550 debugf3("MC_COR_ECC_CNT%d = 0x%x; MC_COR_ECC_CNT%d = 0x%x\n",
1551 (i * 2), rcv[i][0], (i * 2) + 1, rcv[i][1]);
1552 /*if the channel has 3 dimms*/
f4742949 1553 if (pvt->channel[i].dimms > 2) {
b4e8f0b6
MCC
1554 new0 = DIMM_BOT_COR_ERR(rcv[i][0]);
1555 new1 = DIMM_TOP_COR_ERR(rcv[i][0]);
1556 new2 = DIMM_BOT_COR_ERR(rcv[i][1]);
1557 } else {
1558 new0 = DIMM_TOP_COR_ERR(rcv[i][0]) +
1559 DIMM_BOT_COR_ERR(rcv[i][0]);
1560 new1 = DIMM_TOP_COR_ERR(rcv[i][1]) +
1561 DIMM_BOT_COR_ERR(rcv[i][1]);
1562 new2 = 0;
1563 }
1564
f4742949 1565 i7core_rdimm_update_ce_count(mci, i, new0, new1, new2);
b4e8f0b6
MCC
1566 }
1567}
442305b1
MCC
1568
1569/* This function is based on the device 3 function 4 registers as described on:
1570 * Intel Xeon Processor 5500 Series Datasheet Volume 2
1571 * http://www.intel.com/Assets/PDF/datasheet/321322.pdf
1572 * also available at:
1573 * http://www.arrownac.com/manufacturers/intel/s/nehalem/5500-datasheet-v2.pdf
1574 */
f4742949 1575static void i7core_udimm_check_mc_ecc_err(struct mem_ctl_info *mci)
442305b1
MCC
1576{
1577 struct i7core_pvt *pvt = mci->pvt_info;
1578 u32 rcv1, rcv0;
1579 int new0, new1, new2;
1580
f4742949 1581 if (!pvt->pci_mcr[4]) {
b990538a 1582 debugf0("%s MCR registers not found\n", __func__);
442305b1
MCC
1583 return;
1584 }
1585
b4e8f0b6 1586 /* Corrected test errors */
f4742949
MCC
1587 pci_read_config_dword(pvt->pci_mcr[4], MC_TEST_ERR_RCV1, &rcv1);
1588 pci_read_config_dword(pvt->pci_mcr[4], MC_TEST_ERR_RCV0, &rcv0);
442305b1
MCC
1589
1590 /* Store the new values */
1591 new2 = DIMM2_COR_ERR(rcv1);
1592 new1 = DIMM1_COR_ERR(rcv0);
1593 new0 = DIMM0_COR_ERR(rcv0);
1594
442305b1 1595 /* Updates CE counters if it is not the first time here */
f4742949 1596 if (pvt->ce_count_available) {
442305b1
MCC
1597 /* Updates CE counters */
1598 int add0, add1, add2;
1599
f4742949
MCC
1600 add2 = new2 - pvt->udimm_last_ce_count[2];
1601 add1 = new1 - pvt->udimm_last_ce_count[1];
1602 add0 = new0 - pvt->udimm_last_ce_count[0];
442305b1
MCC
1603
1604 if (add2 < 0)
1605 add2 += 0x7fff;
f4742949 1606 pvt->udimm_ce_count[2] += add2;
442305b1
MCC
1607
1608 if (add1 < 0)
1609 add1 += 0x7fff;
f4742949 1610 pvt->udimm_ce_count[1] += add1;
442305b1
MCC
1611
1612 if (add0 < 0)
1613 add0 += 0x7fff;
f4742949 1614 pvt->udimm_ce_count[0] += add0;
b4e8f0b6
MCC
1615
1616 if (add0 | add1 | add2)
1617 i7core_printk(KERN_ERR, "New Corrected error(s): "
1618 "dimm0: +%d, dimm1: +%d, dimm2 +%d\n",
1619 add0, add1, add2);
442305b1 1620 } else
f4742949 1621 pvt->ce_count_available = 1;
442305b1
MCC
1622
1623 /* Store the new values */
f4742949
MCC
1624 pvt->udimm_last_ce_count[2] = new2;
1625 pvt->udimm_last_ce_count[1] = new1;
1626 pvt->udimm_last_ce_count[0] = new0;
442305b1
MCC
1627}
1628
8a2f118e
MCC
1629/*
1630 * According with tables E-11 and E-12 of chapter E.3.3 of Intel 64 and IA-32
1631 * Architectures Software Developer’s Manual Volume 3B.
f237fcf2
MCC
1632 * Nehalem are defined as family 0x06, model 0x1a
1633 *
1634 * The MCA registers used here are the following ones:
8a2f118e 1635 * struct mce field MCA Register
f237fcf2
MCC
1636 * m->status MSR_IA32_MC8_STATUS
1637 * m->addr MSR_IA32_MC8_ADDR
1638 * m->misc MSR_IA32_MC8_MISC
8a2f118e
MCC
1639 * In the case of Nehalem, the error information is masked at .status and .misc
1640 * fields
1641 */
d5381642
MCC
1642static void i7core_mce_output_error(struct mem_ctl_info *mci,
1643 struct mce *m)
1644{
b4e8f0b6 1645 struct i7core_pvt *pvt = mci->pvt_info;
a639539f 1646 char *type, *optype, *err, *msg;
8a2f118e 1647 unsigned long error = m->status & 0x1ff0000l;
a639539f 1648 u32 optypenum = (m->status >> 4) & 0x07;
8a2f118e
MCC
1649 u32 core_err_cnt = (m->status >> 38) && 0x7fff;
1650 u32 dimm = (m->misc >> 16) & 0x3;
1651 u32 channel = (m->misc >> 18) & 0x3;
1652 u32 syndrome = m->misc >> 32;
1653 u32 errnum = find_first_bit(&error, 32);
b4e8f0b6 1654 int csrow;
8a2f118e 1655
c5d34528
MCC
1656 if (m->mcgstatus & 1)
1657 type = "FATAL";
1658 else
1659 type = "NON_FATAL";
1660
a639539f 1661 switch (optypenum) {
b990538a
MCC
1662 case 0:
1663 optype = "generic undef request";
1664 break;
1665 case 1:
1666 optype = "read error";
1667 break;
1668 case 2:
1669 optype = "write error";
1670 break;
1671 case 3:
1672 optype = "addr/cmd error";
1673 break;
1674 case 4:
1675 optype = "scrubbing error";
1676 break;
1677 default:
1678 optype = "reserved";
1679 break;
a639539f
MCC
1680 }
1681
8a2f118e
MCC
1682 switch (errnum) {
1683 case 16:
1684 err = "read ECC error";
1685 break;
1686 case 17:
1687 err = "RAS ECC error";
1688 break;
1689 case 18:
1690 err = "write parity error";
1691 break;
1692 case 19:
1693 err = "redundacy loss";
1694 break;
1695 case 20:
1696 err = "reserved";
1697 break;
1698 case 21:
1699 err = "memory range error";
1700 break;
1701 case 22:
1702 err = "RTID out of range";
1703 break;
1704 case 23:
1705 err = "address parity error";
1706 break;
1707 case 24:
1708 err = "byte enable parity error";
1709 break;
1710 default:
1711 err = "unknown";
d5381642 1712 }
d5381642 1713
f237fcf2 1714 /* FIXME: should convert addr into bank and rank information */
8a2f118e 1715 msg = kasprintf(GFP_ATOMIC,
f4742949 1716 "%s (addr = 0x%08llx, cpu=%d, Dimm=%d, Channel=%d, "
a639539f 1717 "syndrome=0x%08x, count=%d, Err=%08llx:%08llx (%s: %s))\n",
f4742949 1718 type, (long long) m->addr, m->cpu, dimm, channel,
a639539f
MCC
1719 syndrome, core_err_cnt, (long long)m->status,
1720 (long long)m->misc, optype, err);
8a2f118e
MCC
1721
1722 debugf0("%s", msg);
d5381642 1723
f4742949 1724 csrow = pvt->csrow_map[channel][dimm];
b4e8f0b6 1725
d5381642 1726 /* Call the helper to output message */
b4e8f0b6
MCC
1727 if (m->mcgstatus & 1)
1728 edac_mc_handle_fbd_ue(mci, csrow, 0,
1729 0 /* FIXME: should be channel here */, msg);
f4742949 1730 else if (!pvt->is_registered)
b4e8f0b6
MCC
1731 edac_mc_handle_fbd_ce(mci, csrow,
1732 0 /* FIXME: should be channel here */, msg);
8a2f118e
MCC
1733
1734 kfree(msg);
d5381642
MCC
1735}
1736
87d1d272
MCC
1737/*
1738 * i7core_check_error Retrieve and process errors reported by the
1739 * hardware. Called by the Core module.
1740 */
1741static void i7core_check_error(struct mem_ctl_info *mci)
1742{
d5381642
MCC
1743 struct i7core_pvt *pvt = mci->pvt_info;
1744 int i;
1745 unsigned count = 0;
ca9c90ba 1746 struct mce *m;
d5381642 1747
ca9c90ba
MCC
1748 /*
1749 * MCE first step: Copy all mce errors into a temporary buffer
1750 * We use a double buffering here, to reduce the risk of
1751 * loosing an error.
1752 */
1753 smp_rmb();
321ece4d
MCC
1754 count = (pvt->mce_out + MCE_LOG_LEN - pvt->mce_in)
1755 % MCE_LOG_LEN;
ca9c90ba 1756 if (!count)
8a311e17 1757 goto check_ce_error;
f4742949 1758
ca9c90ba 1759 m = pvt->mce_outentry;
321ece4d
MCC
1760 if (pvt->mce_in + count > MCE_LOG_LEN) {
1761 unsigned l = MCE_LOG_LEN - pvt->mce_in;
f4742949 1762
ca9c90ba
MCC
1763 memcpy(m, &pvt->mce_entry[pvt->mce_in], sizeof(*m) * l);
1764 smp_wmb();
1765 pvt->mce_in = 0;
1766 count -= l;
1767 m += l;
1768 }
1769 memcpy(m, &pvt->mce_entry[pvt->mce_in], sizeof(*m) * count);
1770 smp_wmb();
1771 pvt->mce_in += count;
1772
1773 smp_rmb();
1774 if (pvt->mce_overrun) {
1775 i7core_printk(KERN_ERR, "Lost %d memory errors\n",
1776 pvt->mce_overrun);
1777 smp_wmb();
1778 pvt->mce_overrun = 0;
1779 }
d5381642 1780
ca9c90ba
MCC
1781 /*
1782 * MCE second step: parse errors and display
1783 */
d5381642 1784 for (i = 0; i < count; i++)
ca9c90ba 1785 i7core_mce_output_error(mci, &pvt->mce_outentry[i]);
d5381642 1786
ca9c90ba
MCC
1787 /*
1788 * Now, let's increment CE error counts
1789 */
8a311e17 1790check_ce_error:
f4742949
MCC
1791 if (!pvt->is_registered)
1792 i7core_udimm_check_mc_ecc_err(mci);
1793 else
1794 i7core_rdimm_check_mc_ecc_err(mci);
87d1d272
MCC
1795}
1796
d5381642
MCC
1797/*
1798 * i7core_mce_check_error Replicates mcelog routine to get errors
1799 * This routine simply queues mcelog errors, and
1800 * return. The error itself should be handled later
1801 * by i7core_check_error.
6e103be1
MCC
1802 * WARNING: As this routine should be called at NMI time, extra care should
1803 * be taken to avoid deadlocks, and to be as fast as possible.
d5381642
MCC
1804 */
1805static int i7core_mce_check_error(void *priv, struct mce *mce)
1806{
c5d34528
MCC
1807 struct mem_ctl_info *mci = priv;
1808 struct i7core_pvt *pvt = mci->pvt_info;
d5381642 1809
8a2f118e
MCC
1810 /*
1811 * Just let mcelog handle it if the error is
1812 * outside the memory controller
1813 */
1814 if (((mce->status & 0xffff) >> 7) != 1)
1815 return 0;
1816
f237fcf2
MCC
1817 /* Bank 8 registers are the only ones that we know how to handle */
1818 if (mce->bank != 8)
1819 return 0;
1820
3b918c12 1821#ifdef CONFIG_SMP
f4742949 1822 /* Only handle if it is the right mc controller */
6e103be1 1823 if (cpu_data(mce->cpu).phys_proc_id != pvt->i7core_dev->socket)
f4742949 1824 return 0;
3b918c12 1825#endif
f4742949 1826
ca9c90ba 1827 smp_rmb();
321ece4d 1828 if ((pvt->mce_out + 1) % MCE_LOG_LEN == pvt->mce_in) {
ca9c90ba
MCC
1829 smp_wmb();
1830 pvt->mce_overrun++;
1831 return 0;
d5381642 1832 }
6e103be1
MCC
1833
1834 /* Copy memory error at the ringbuffer */
1835 memcpy(&pvt->mce_entry[pvt->mce_out], mce, sizeof(*mce));
ca9c90ba 1836 smp_wmb();
321ece4d 1837 pvt->mce_out = (pvt->mce_out + 1) % MCE_LOG_LEN;
d5381642 1838
c5d34528
MCC
1839 /* Handle fatal errors immediately */
1840 if (mce->mcgstatus & 1)
1841 i7core_check_error(mci);
1842
d5381642 1843 /* Advice mcelog that the error were handled */
8a2f118e 1844 return 1;
d5381642
MCC
1845}
1846
f4742949
MCC
1847static int i7core_register_mci(struct i7core_dev *i7core_dev,
1848 int num_channels, int num_csrows)
a0c36a1f
MCC
1849{
1850 struct mem_ctl_info *mci;
1851 struct i7core_pvt *pvt;
ba6c5c62 1852 int csrow = 0;
f4742949 1853 int rc;
a0c36a1f 1854
a0c36a1f 1855 /* allocate a new MC control structure */
d4c27795
MCC
1856 mci = edac_mc_alloc(sizeof(*pvt), num_csrows, num_channels,
1857 i7core_dev->socket);
f4742949
MCC
1858 if (unlikely(!mci))
1859 return -ENOMEM;
a0c36a1f
MCC
1860
1861 debugf0("MC: " __FILE__ ": %s(): mci = %p\n", __func__, mci);
1862
f4742949
MCC
1863 /* record ptr to the generic device */
1864 mci->dev = &i7core_dev->pdev[0]->dev;
1865
a0c36a1f 1866 pvt = mci->pvt_info;
ef708b53 1867 memset(pvt, 0, sizeof(*pvt));
67166af4 1868
41fcb7fe
MCC
1869 /*
1870 * FIXME: how to handle RDDR3 at MCI level? It is possible to have
1871 * Mixed RDDR3/UDDR3 with Nehalem, provided that they are on different
1872 * memory channels
1873 */
1874 mci->mtype_cap = MEM_FLAG_DDR3;
a0c36a1f
MCC
1875 mci->edac_ctl_cap = EDAC_FLAG_NONE;
1876 mci->edac_cap = EDAC_FLAG_NONE;
1877 mci->mod_name = "i7core_edac.c";
1878 mci->mod_ver = I7CORE_REVISION;
f4742949
MCC
1879 mci->ctl_name = kasprintf(GFP_KERNEL, "i7 core #%d",
1880 i7core_dev->socket);
1881 mci->dev_name = pci_name(i7core_dev->pdev[0]);
a0c36a1f 1882 mci->ctl_page_to_phys = NULL;
a5538e53 1883 mci->mc_driver_sysfs_attributes = i7core_sysfs_attrs;
87d1d272
MCC
1884 /* Set the function pointer to an actual operation function */
1885 mci->edac_check = i7core_check_error;
8f331907 1886
ef708b53 1887 /* Store pci devices at mci for faster access */
f4742949 1888 rc = mci_bind_devs(mci, i7core_dev);
41fcb7fe 1889 if (unlikely(rc < 0))
f4742949 1890 goto fail;
ef708b53
MCC
1891
1892 /* Get dimm basic config */
f4742949 1893 get_dimm_config(mci, &csrow);
ef708b53 1894
a0c36a1f 1895 /* add this new MC control structure to EDAC's list of MCs */
b7c76151 1896 if (unlikely(edac_mc_add_mc(mci))) {
a0c36a1f
MCC
1897 debugf0("MC: " __FILE__
1898 ": %s(): failed edac_mc_add_mc()\n", __func__);
1899 /* FIXME: perhaps some code should go here that disables error
1900 * reporting if we just enabled it
1901 */
b7c76151
MCC
1902
1903 rc = -EINVAL;
f4742949 1904 goto fail;
a0c36a1f
MCC
1905 }
1906
1907 /* allocating generic PCI control info */
f4742949
MCC
1908 i7core_pci = edac_pci_create_generic_ctl(&i7core_dev->pdev[0]->dev,
1909 EDAC_MOD_STR);
41fcb7fe 1910 if (unlikely(!i7core_pci)) {
a0c36a1f
MCC
1911 printk(KERN_WARNING
1912 "%s(): Unable to create PCI control\n",
1913 __func__);
1914 printk(KERN_WARNING
1915 "%s(): PCI error report via EDAC not setup\n",
1916 __func__);
1917 }
1918
194a40fe 1919 /* Default error mask is any memory */
ef708b53 1920 pvt->inject.channel = 0;
194a40fe
MCC
1921 pvt->inject.dimm = -1;
1922 pvt->inject.rank = -1;
1923 pvt->inject.bank = -1;
1924 pvt->inject.page = -1;
1925 pvt->inject.col = -1;
1926
d5381642 1927 /* Registers on edac_mce in order to receive memory errors */
c5d34528 1928 pvt->edac_mce.priv = mci;
d5381642 1929 pvt->edac_mce.check_error = i7core_mce_check_error;
d5381642
MCC
1930
1931 rc = edac_mce_register(&pvt->edac_mce);
b990538a 1932 if (unlikely(rc < 0)) {
d5381642
MCC
1933 debugf0("MC: " __FILE__
1934 ": %s(): failed edac_mce_register()\n", __func__);
f4742949
MCC
1935 }
1936
1937fail:
d4d1ef45
TL
1938 if (rc < 0)
1939 edac_mc_free(mci);
f4742949
MCC
1940 return rc;
1941}
1942
1943/*
1944 * i7core_probe Probe for ONE instance of device to see if it is
1945 * present.
1946 * return:
1947 * 0 for FOUND a device
1948 * < 0 for error code
1949 */
1950static int __devinit i7core_probe(struct pci_dev *pdev,
1951 const struct pci_device_id *id)
1952{
1953 int dev_idx = id->driver_data;
1954 int rc;
1955 struct i7core_dev *i7core_dev;
1956
1957 /*
d4c27795 1958 * All memory controllers are allocated at the first pass.
f4742949
MCC
1959 */
1960 if (unlikely(dev_idx >= 1))
1961 return -EINVAL;
1962
1963 /* get the pci devices we want to reserve for our use */
1964 mutex_lock(&i7core_edac_lock);
de06eeef 1965
bd9e19ca 1966 rc = i7core_get_devices(pci_dev_table);
f4742949
MCC
1967 if (unlikely(rc < 0))
1968 goto fail0;
1969
1970 list_for_each_entry(i7core_dev, &i7core_edac_list, list) {
1971 int channels;
1972 int csrows;
1973
1974 /* Check the number of active and not disabled channels */
1975 rc = i7core_get_active_channels(i7core_dev->socket,
1976 &channels, &csrows);
1977 if (unlikely(rc < 0))
1978 goto fail1;
1979
d4c27795
MCC
1980 rc = i7core_register_mci(i7core_dev, channels, csrows);
1981 if (unlikely(rc < 0))
1982 goto fail1;
d5381642
MCC
1983 }
1984
ef708b53 1985 i7core_printk(KERN_INFO, "Driver loaded.\n");
8f331907 1986
66607706 1987 mutex_unlock(&i7core_edac_lock);
a0c36a1f
MCC
1988 return 0;
1989
66607706 1990fail1:
13d6e9b6 1991 i7core_put_all_devices();
66607706
MCC
1992fail0:
1993 mutex_unlock(&i7core_edac_lock);
b7c76151 1994 return rc;
a0c36a1f
MCC
1995}
1996
1997/*
1998 * i7core_remove destructor for one instance of device
1999 *
2000 */
2001static void __devexit i7core_remove(struct pci_dev *pdev)
2002{
2003 struct mem_ctl_info *mci;
22e6bcbd 2004 struct i7core_dev *i7core_dev, *tmp;
a0c36a1f
MCC
2005
2006 debugf0(__FILE__ ": %s()\n", __func__);
2007
2008 if (i7core_pci)
2009 edac_pci_release_generic_ctl(i7core_pci);
2010
22e6bcbd
MCC
2011 /*
2012 * we have a trouble here: pdev value for removal will be wrong, since
2013 * it will point to the X58 register used to detect that the machine
2014 * is a Nehalem or upper design. However, due to the way several PCI
2015 * devices are grouped together to provide MC functionality, we need
2016 * to use a different method for releasing the devices
2017 */
87d1d272 2018
66607706 2019 mutex_lock(&i7core_edac_lock);
22e6bcbd
MCC
2020 list_for_each_entry_safe(i7core_dev, tmp, &i7core_edac_list, list) {
2021 mci = edac_mc_del_mc(&i7core_dev->pdev[0]->dev);
2022 if (mci) {
2023 struct i7core_pvt *pvt = mci->pvt_info;
2024
2025 i7core_dev = pvt->i7core_dev;
2026 edac_mce_unregister(&pvt->edac_mce);
2027 kfree(mci->ctl_name);
2028 edac_mc_free(mci);
2029 i7core_put_devices(i7core_dev);
2030 } else {
2031 i7core_printk(KERN_ERR,
2032 "Couldn't find mci for socket %d\n",
2033 i7core_dev->socket);
2034 }
2035 }
66607706 2036 mutex_unlock(&i7core_edac_lock);
a0c36a1f
MCC
2037}
2038
a0c36a1f
MCC
2039MODULE_DEVICE_TABLE(pci, i7core_pci_tbl);
2040
2041/*
2042 * i7core_driver pci_driver structure for this module
2043 *
2044 */
2045static struct pci_driver i7core_driver = {
2046 .name = "i7core_edac",
2047 .probe = i7core_probe,
2048 .remove = __devexit_p(i7core_remove),
2049 .id_table = i7core_pci_tbl,
2050};
2051
2052/*
2053 * i7core_init Module entry function
2054 * Try to initialize this module for its devices
2055 */
2056static int __init i7core_init(void)
2057{
2058 int pci_rc;
2059
2060 debugf2("MC: " __FILE__ ": %s()\n", __func__);
2061
2062 /* Ensure that the OPSTATE is set correctly for POLL or NMI */
2063 opstate_init();
2064
bd9e19ca 2065 i7core_xeon_pci_fixup(pci_dev_table);
bc2d7245 2066
a0c36a1f
MCC
2067 pci_rc = pci_register_driver(&i7core_driver);
2068
3ef288a9
MCC
2069 if (pci_rc >= 0)
2070 return 0;
2071
2072 i7core_printk(KERN_ERR, "Failed to register device with error %d.\n",
2073 pci_rc);
2074
2075 return pci_rc;
a0c36a1f
MCC
2076}
2077
2078/*
2079 * i7core_exit() Module exit function
2080 * Unregister the driver
2081 */
2082static void __exit i7core_exit(void)
2083{
2084 debugf2("MC: " __FILE__ ": %s()\n", __func__);
2085 pci_unregister_driver(&i7core_driver);
2086}
2087
2088module_init(i7core_init);
2089module_exit(i7core_exit);
2090
2091MODULE_LICENSE("GPL");
2092MODULE_AUTHOR("Mauro Carvalho Chehab <mchehab@redhat.com>");
2093MODULE_AUTHOR("Red Hat Inc. (http://www.redhat.com)");
2094MODULE_DESCRIPTION("MC Driver for Intel i7 Core memory controllers - "
2095 I7CORE_REVISION);
2096
2097module_param(edac_op_state, int, 0444);
2098MODULE_PARM_DESC(edac_op_state, "EDAC Error Reporting state: 0=Poll,1=NMI");