blob: 19faeffac9dcfbccb7860e1ab9f3ebe0f15a3a45 [file] [log] [blame]
Mauro Carvalho Chehab52707f92010-05-18 20:43:52 -03001/* 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.
Mauro Carvalho Chehaba0c36a12009-06-22 22:41:15 -03007 *
8 * This file may be distributed under the terms of the
9 * GNU General Public License version 2 only.
10 *
Mauro Carvalho Chehab52707f92010-05-18 20:43:52 -030011 * Copyright (c) 2009-2010 by:
Mauro Carvalho Chehaba0c36a12009-06-22 22:41:15 -030012 * 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
Mauro Carvalho Chehaba0c36a12009-06-22 22:41:15 -030028#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>
Randy Dunlap3b918c12009-11-08 01:36:40 -020033#include <linux/delay.h>
Mauro Carvalho Chehaba0c36a12009-06-22 22:41:15 -030034#include <linux/edac.h>
35#include <linux/mmzone.h>
Mauro Carvalho Chehabd5381642009-07-09 22:06:41 -030036#include <linux/edac_mce.h>
Mauro Carvalho Chehabf4742942009-09-05 02:35:08 -030037#include <linux/smp.h>
Mauro Carvalho Chehab14d2c082009-09-02 23:52:36 -030038#include <asm/processor.h>
Mauro Carvalho Chehaba0c36a12009-06-22 22:41:15 -030039
40#include "edac_core.h"
41
Mauro Carvalho Chehab18c29002010-08-10 18:33:27 -030042/* Static vars */
43static LIST_HEAD(i7core_edac_list);
44static DEFINE_MUTEX(i7core_edac_lock);
45static int probed;
46
Mauro Carvalho Chehab54a08ab2010-08-19 15:51:00 -030047static int use_pci_fixup;
48module_param(use_pci_fixup, int, 0444);
49MODULE_PARM_DESC(use_pci_fixup, "Enable PCI fixup to seek for hidden devices");
Mauro Carvalho Chehaba0c36a12009-06-22 22:41:15 -030050/*
Mauro Carvalho Chehabf4742942009-09-05 02:35:08 -030051 * This is used for Nehalem-EP and Nehalem-EX devices, where the non-core
52 * registers start at bus 255, and are not reported by BIOS.
53 * We currently find devices with only 2 sockets. In order to support more QPI
54 * Quick Path Interconnect, just increment this number.
55 */
56#define MAX_SOCKET_BUSES 2
57
58
59/*
Mauro Carvalho Chehaba0c36a12009-06-22 22:41:15 -030060 * Alter this version for the module when modifications are made
61 */
62#define I7CORE_REVISION " Ver: 1.0.0 " __DATE__
63#define EDAC_MOD_STR "i7core_edac"
64
Mauro Carvalho Chehaba0c36a12009-06-22 22:41:15 -030065/*
66 * Debug macros
67 */
68#define i7core_printk(level, fmt, arg...) \
69 edac_printk(level, "i7core", fmt, ##arg)
70
71#define i7core_mc_printk(mci, level, fmt, arg...) \
72 edac_mc_chipset_printk(mci, level, "i7core", fmt, ##arg)
73
74/*
75 * i7core Memory Controller Registers
76 */
77
Mauro Carvalho Chehabe9bd2e72009-07-09 22:14:35 -030078 /* OFFSETS for Device 0 Function 0 */
79
80#define MC_CFG_CONTROL 0x90
81
Mauro Carvalho Chehaba0c36a12009-06-22 22:41:15 -030082 /* OFFSETS for Device 3 Function 0 */
83
84#define MC_CONTROL 0x48
85#define MC_STATUS 0x4c
86#define MC_MAX_DOD 0x64
87
Mauro Carvalho Chehab442305b2009-06-22 22:48:29 -030088/*
89 * OFFSETS for Device 3 Function 4, as inicated on Xeon 5500 datasheet:
90 * http://www.arrownac.com/manufacturers/intel/s/nehalem/5500-datasheet-v2.pdf
91 */
92
93#define MC_TEST_ERR_RCV1 0x60
94 #define DIMM2_COR_ERR(r) ((r) & 0x7fff)
95
96#define MC_TEST_ERR_RCV0 0x64
97 #define DIMM1_COR_ERR(r) (((r) >> 16) & 0x7fff)
98 #define DIMM0_COR_ERR(r) ((r) & 0x7fff)
99
Mauro Carvalho Chehabb4e8f0b2009-09-02 23:49:59 -0300100/* OFFSETS for Device 3 Function 2, as inicated on Xeon 5500 datasheet */
101#define MC_COR_ECC_CNT_0 0x80
102#define MC_COR_ECC_CNT_1 0x84
103#define MC_COR_ECC_CNT_2 0x88
104#define MC_COR_ECC_CNT_3 0x8c
105#define MC_COR_ECC_CNT_4 0x90
106#define MC_COR_ECC_CNT_5 0x94
107
108#define DIMM_TOP_COR_ERR(r) (((r) >> 16) & 0x7fff)
109#define DIMM_BOT_COR_ERR(r) ((r) & 0x7fff)
110
111
Mauro Carvalho Chehaba0c36a12009-06-22 22:41:15 -0300112 /* OFFSETS for Devices 4,5 and 6 Function 0 */
113
Mauro Carvalho Chehab0b2b7b72009-06-22 22:48:29 -0300114#define MC_CHANNEL_DIMM_INIT_PARAMS 0x58
115 #define THREE_DIMMS_PRESENT (1 << 24)
116 #define SINGLE_QUAD_RANK_PRESENT (1 << 23)
117 #define QUAD_RANK_PRESENT (1 << 22)
118 #define REGISTERED_DIMM (1 << 15)
119
Mauro Carvalho Chehabf122a892009-06-22 22:48:29 -0300120#define MC_CHANNEL_MAPPER 0x60
121 #define RDLCH(r, ch) ((((r) >> (3 + (ch * 6))) & 0x07) - 1)
122 #define WRLCH(r, ch) ((((r) >> (ch * 6)) & 0x07) - 1)
123
Mauro Carvalho Chehab0b2b7b72009-06-22 22:48:29 -0300124#define MC_CHANNEL_RANK_PRESENT 0x7c
125 #define RANK_PRESENT_MASK 0xffff
126
Mauro Carvalho Chehaba0c36a12009-06-22 22:41:15 -0300127#define MC_CHANNEL_ADDR_MATCH 0xf0
Mauro Carvalho Chehab194a40f2009-06-22 22:48:28 -0300128#define MC_CHANNEL_ERROR_MASK 0xf8
129#define MC_CHANNEL_ERROR_INJECT 0xfc
130 #define INJECT_ADDR_PARITY 0x10
131 #define INJECT_ECC 0x08
132 #define MASK_CACHELINE 0x06
133 #define MASK_FULL_CACHELINE 0x06
134 #define MASK_MSB32_CACHELINE 0x04
135 #define MASK_LSB32_CACHELINE 0x02
136 #define NO_MASK_CACHELINE 0x00
137 #define REPEAT_EN 0x01
Mauro Carvalho Chehaba0c36a12009-06-22 22:41:15 -0300138
Mauro Carvalho Chehab0b2b7b72009-06-22 22:48:29 -0300139 /* OFFSETS for Devices 4,5 and 6 Function 1 */
Mauro Carvalho Chehabb9905382009-08-05 21:36:35 -0300140
Mauro Carvalho Chehab0b2b7b72009-06-22 22:48:29 -0300141#define MC_DOD_CH_DIMM0 0x48
142#define MC_DOD_CH_DIMM1 0x4c
143#define MC_DOD_CH_DIMM2 0x50
144 #define RANKOFFSET_MASK ((1 << 12) | (1 << 11) | (1 << 10))
145 #define RANKOFFSET(x) ((x & RANKOFFSET_MASK) >> 10)
146 #define DIMM_PRESENT_MASK (1 << 9)
147 #define DIMM_PRESENT(x) (((x) & DIMM_PRESENT_MASK) >> 9)
Mauro Carvalho Chehab854d3342009-06-22 22:48:30 -0300148 #define MC_DOD_NUMBANK_MASK ((1 << 8) | (1 << 7))
149 #define MC_DOD_NUMBANK(x) (((x) & MC_DOD_NUMBANK_MASK) >> 7)
150 #define MC_DOD_NUMRANK_MASK ((1 << 6) | (1 << 5))
151 #define MC_DOD_NUMRANK(x) (((x) & MC_DOD_NUMRANK_MASK) >> 5)
Mauro Carvalho Chehab41fcb7f2009-06-22 22:48:31 -0300152 #define MC_DOD_NUMROW_MASK ((1 << 4) | (1 << 3) | (1 << 2))
Mauro Carvalho Chehab5566cb72009-06-22 22:48:31 -0300153 #define MC_DOD_NUMROW(x) (((x) & MC_DOD_NUMROW_MASK) >> 2)
Mauro Carvalho Chehab854d3342009-06-22 22:48:30 -0300154 #define MC_DOD_NUMCOL_MASK 3
155 #define MC_DOD_NUMCOL(x) ((x) & MC_DOD_NUMCOL_MASK)
Mauro Carvalho Chehab0b2b7b72009-06-22 22:48:29 -0300156
Mauro Carvalho Chehabf122a892009-06-22 22:48:29 -0300157#define MC_RANK_PRESENT 0x7c
158
Mauro Carvalho Chehab0b2b7b72009-06-22 22:48:29 -0300159#define MC_SAG_CH_0 0x80
160#define MC_SAG_CH_1 0x84
161#define MC_SAG_CH_2 0x88
162#define MC_SAG_CH_3 0x8c
163#define MC_SAG_CH_4 0x90
164#define MC_SAG_CH_5 0x94
165#define MC_SAG_CH_6 0x98
166#define MC_SAG_CH_7 0x9c
167
168#define MC_RIR_LIMIT_CH_0 0x40
169#define MC_RIR_LIMIT_CH_1 0x44
170#define MC_RIR_LIMIT_CH_2 0x48
171#define MC_RIR_LIMIT_CH_3 0x4C
172#define MC_RIR_LIMIT_CH_4 0x50
173#define MC_RIR_LIMIT_CH_5 0x54
174#define MC_RIR_LIMIT_CH_6 0x58
175#define MC_RIR_LIMIT_CH_7 0x5C
176#define MC_RIR_LIMIT_MASK ((1 << 10) - 1)
177
178#define MC_RIR_WAY_CH 0x80
179 #define MC_RIR_WAY_OFFSET_MASK (((1 << 14) - 1) & ~0x7)
180 #define MC_RIR_WAY_RANK_MASK 0x7
181
Mauro Carvalho Chehaba0c36a12009-06-22 22:41:15 -0300182/*
183 * i7core structs
184 */
185
186#define NUM_CHANS 3
Mauro Carvalho Chehab442305b2009-06-22 22:48:29 -0300187#define MAX_DIMMS 3 /* Max DIMMS per channel */
188#define MAX_MCR_FUNC 4
189#define MAX_CHAN_FUNC 3
Mauro Carvalho Chehaba0c36a12009-06-22 22:41:15 -0300190
191struct i7core_info {
192 u32 mc_control;
193 u32 mc_status;
194 u32 max_dod;
Mauro Carvalho Chehabf122a892009-06-22 22:48:29 -0300195 u32 ch_map;
Mauro Carvalho Chehaba0c36a12009-06-22 22:41:15 -0300196};
197
Mauro Carvalho Chehab194a40f2009-06-22 22:48:28 -0300198
199struct i7core_inject {
200 int enable;
201
202 u32 section;
203 u32 type;
204 u32 eccmask;
205
206 /* Error address mask */
207 int channel, dimm, rank, bank, page, col;
208};
209
Mauro Carvalho Chehab0b2b7b72009-06-22 22:48:29 -0300210struct i7core_channel {
Mauro Carvalho Chehab442305b2009-06-22 22:48:29 -0300211 u32 ranks;
212 u32 dimms;
Mauro Carvalho Chehab0b2b7b72009-06-22 22:48:29 -0300213};
214
Mauro Carvalho Chehab8f331902009-06-22 22:48:29 -0300215struct pci_id_descr {
Mauro Carvalho Chehab66607702009-09-05 00:52:11 -0300216 int dev;
217 int func;
218 int dev_id;
Mauro Carvalho Chehabde06eee2009-10-14 08:02:40 -0300219 int optional;
Mauro Carvalho Chehab8f331902009-06-22 22:48:29 -0300220};
221
Vernon Mauerybd9e19c2010-05-18 19:02:50 -0300222struct pci_id_table {
Mauro Carvalho Chehab1288c182010-08-10 18:57:01 -0300223 const struct pci_id_descr *descr;
224 int n_devs;
Vernon Mauerybd9e19c2010-05-18 19:02:50 -0300225};
226
Mauro Carvalho Chehabf4742942009-09-05 02:35:08 -0300227struct i7core_dev {
228 struct list_head list;
229 u8 socket;
230 struct pci_dev **pdev;
Mauro Carvalho Chehabde06eee2009-10-14 08:02:40 -0300231 int n_devs;
Mauro Carvalho Chehabf4742942009-09-05 02:35:08 -0300232 struct mem_ctl_info *mci;
233};
234
Mauro Carvalho Chehaba0c36a12009-06-22 22:41:15 -0300235struct i7core_pvt {
Mauro Carvalho Chehabf4742942009-09-05 02:35:08 -0300236 struct pci_dev *pci_noncore;
237 struct pci_dev *pci_mcr[MAX_MCR_FUNC + 1];
238 struct pci_dev *pci_ch[NUM_CHANS][MAX_CHAN_FUNC + 1];
239
240 struct i7core_dev *i7core_dev;
Mauro Carvalho Chehab67166af2009-07-15 06:56:23 -0300241
Mauro Carvalho Chehaba0c36a12009-06-22 22:41:15 -0300242 struct i7core_info info;
Mauro Carvalho Chehab194a40f2009-06-22 22:48:28 -0300243 struct i7core_inject inject;
Mauro Carvalho Chehabf4742942009-09-05 02:35:08 -0300244 struct i7core_channel channel[NUM_CHANS];
Mauro Carvalho Chehab67166af2009-07-15 06:56:23 -0300245
Mauro Carvalho Chehabf4742942009-09-05 02:35:08 -0300246 int channels; /* Number of active channels */
Mauro Carvalho Chehab442305b2009-06-22 22:48:29 -0300247
Mauro Carvalho Chehabf4742942009-09-05 02:35:08 -0300248 int ce_count_available;
249 int csrow_map[NUM_CHANS][MAX_DIMMS];
Mauro Carvalho Chehabb4e8f0b2009-09-02 23:49:59 -0300250
251 /* ECC corrected errors counts per udimm */
Mauro Carvalho Chehabf4742942009-09-05 02:35:08 -0300252 unsigned long udimm_ce_count[MAX_DIMMS];
253 int udimm_last_ce_count[MAX_DIMMS];
Mauro Carvalho Chehabb4e8f0b2009-09-02 23:49:59 -0300254 /* ECC corrected errors counts per rdimm */
Mauro Carvalho Chehabf4742942009-09-05 02:35:08 -0300255 unsigned long rdimm_ce_count[NUM_CHANS][MAX_DIMMS];
256 int rdimm_last_ce_count[NUM_CHANS][MAX_DIMMS];
Mauro Carvalho Chehab442305b2009-06-22 22:48:29 -0300257
Mauro Carvalho Chehabf4742942009-09-05 02:35:08 -0300258 unsigned int is_registered;
Mauro Carvalho Chehab14d2c082009-09-02 23:52:36 -0300259
Mauro Carvalho Chehabd5381642009-07-09 22:06:41 -0300260 /* mcelog glue */
261 struct edac_mce edac_mce;
Mauro Carvalho Chehabca9c90b2009-10-04 10:15:40 -0300262
263 /* Fifo double buffers */
Mauro Carvalho Chehabd5381642009-07-09 22:06:41 -0300264 struct mce mce_entry[MCE_LOG_LEN];
Mauro Carvalho Chehabca9c90b2009-10-04 10:15:40 -0300265 struct mce mce_outentry[MCE_LOG_LEN];
266
267 /* Fifo in/out counters */
268 unsigned mce_in, mce_out;
269
270 /* Count indicator to show errors not got */
271 unsigned mce_overrun;
Mauro Carvalho Chehab939747bd2010-08-10 11:22:01 -0300272
273 /* Struct to control EDAC polling */
274 struct edac_pci_ctl_info *i7core_pci;
Mauro Carvalho Chehaba0c36a12009-06-22 22:41:15 -0300275};
276
Mauro Carvalho Chehab8f331902009-06-22 22:48:29 -0300277#define PCI_DESCR(device, function, device_id) \
278 .dev = (device), \
279 .func = (function), \
280 .dev_id = (device_id)
281
Mauro Carvalho Chehab1288c182010-08-10 18:57:01 -0300282static const struct pci_id_descr pci_dev_descr_i7core_nehalem[] = {
Mauro Carvalho Chehab8f331902009-06-22 22:48:29 -0300283 /* Memory controller */
284 { PCI_DESCR(3, 0, PCI_DEVICE_ID_INTEL_I7_MCR) },
285 { PCI_DESCR(3, 1, PCI_DEVICE_ID_INTEL_I7_MC_TAD) },
Mauro Carvalho Chehabde06eee2009-10-14 08:02:40 -0300286 /* Exists only for RDIMM */
287 { PCI_DESCR(3, 2, PCI_DEVICE_ID_INTEL_I7_MC_RAS), .optional = 1 },
Mauro Carvalho Chehab8f331902009-06-22 22:48:29 -0300288 { PCI_DESCR(3, 4, PCI_DEVICE_ID_INTEL_I7_MC_TEST) },
289
290 /* Channel 0 */
291 { PCI_DESCR(4, 0, PCI_DEVICE_ID_INTEL_I7_MC_CH0_CTRL) },
292 { PCI_DESCR(4, 1, PCI_DEVICE_ID_INTEL_I7_MC_CH0_ADDR) },
293 { PCI_DESCR(4, 2, PCI_DEVICE_ID_INTEL_I7_MC_CH0_RANK) },
294 { PCI_DESCR(4, 3, PCI_DEVICE_ID_INTEL_I7_MC_CH0_TC) },
295
296 /* Channel 1 */
297 { PCI_DESCR(5, 0, PCI_DEVICE_ID_INTEL_I7_MC_CH1_CTRL) },
298 { PCI_DESCR(5, 1, PCI_DEVICE_ID_INTEL_I7_MC_CH1_ADDR) },
299 { PCI_DESCR(5, 2, PCI_DEVICE_ID_INTEL_I7_MC_CH1_RANK) },
300 { PCI_DESCR(5, 3, PCI_DEVICE_ID_INTEL_I7_MC_CH1_TC) },
301
302 /* Channel 2 */
303 { PCI_DESCR(6, 0, PCI_DEVICE_ID_INTEL_I7_MC_CH2_CTRL) },
304 { PCI_DESCR(6, 1, PCI_DEVICE_ID_INTEL_I7_MC_CH2_ADDR) },
305 { PCI_DESCR(6, 2, PCI_DEVICE_ID_INTEL_I7_MC_CH2_RANK) },
306 { PCI_DESCR(6, 3, PCI_DEVICE_ID_INTEL_I7_MC_CH2_TC) },
Mauro Carvalho Chehab310cbb72009-07-17 00:09:10 -0300307
308 /* Generic Non-core registers */
309 /*
310 * This is the PCI device on i7core and on Xeon 35xx (8086:2c41)
311 * On Xeon 55xx, however, it has a different id (8086:2c40). So,
312 * the probing code needs to test for the other address in case of
313 * failure of this one
314 */
Mauro Carvalho Chehabfd382652009-10-14 06:07:07 -0300315 { PCI_DESCR(0, 0, PCI_DEVICE_ID_INTEL_I7_NONCORE) },
Mauro Carvalho Chehab310cbb72009-07-17 00:09:10 -0300316
Mauro Carvalho Chehaba0c36a12009-06-22 22:41:15 -0300317};
Mauro Carvalho Chehab8f331902009-06-22 22:48:29 -0300318
Mauro Carvalho Chehab1288c182010-08-10 18:57:01 -0300319static const struct pci_id_descr pci_dev_descr_lynnfield[] = {
Mauro Carvalho Chehab52a2e4fc2009-10-14 11:21:58 -0300320 { PCI_DESCR( 3, 0, PCI_DEVICE_ID_INTEL_LYNNFIELD_MCR) },
321 { PCI_DESCR( 3, 1, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_TAD) },
322 { PCI_DESCR( 3, 4, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_TEST) },
323
324 { PCI_DESCR( 4, 0, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH0_CTRL) },
325 { PCI_DESCR( 4, 1, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH0_ADDR) },
326 { PCI_DESCR( 4, 2, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH0_RANK) },
327 { PCI_DESCR( 4, 3, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH0_TC) },
328
Mauro Carvalho Chehab508fa172009-10-14 13:44:37 -0300329 { PCI_DESCR( 5, 0, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH1_CTRL) },
330 { PCI_DESCR( 5, 1, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH1_ADDR) },
331 { PCI_DESCR( 5, 2, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH1_RANK) },
332 { PCI_DESCR( 5, 3, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH1_TC) },
Mauro Carvalho Chehab52a2e4fc2009-10-14 11:21:58 -0300333
Mauro Carvalho Chehabf05da2f2009-10-14 13:31:06 -0300334 /*
335 * This is the PCI device has an alternate address on some
336 * processors like Core i7 860
337 */
Mauro Carvalho Chehab52a2e4fc2009-10-14 11:21:58 -0300338 { PCI_DESCR( 0, 0, PCI_DEVICE_ID_INTEL_LYNNFIELD_NONCORE) },
339};
340
Mauro Carvalho Chehab1288c182010-08-10 18:57:01 -0300341static const struct pci_id_descr pci_dev_descr_i7core_westmere[] = {
Vernon Mauerybd9e19c2010-05-18 19:02:50 -0300342 /* Memory controller */
343 { PCI_DESCR(3, 0, PCI_DEVICE_ID_INTEL_LYNNFIELD_MCR_REV2) },
344 { PCI_DESCR(3, 1, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_TAD_REV2) },
345 /* Exists only for RDIMM */
346 { PCI_DESCR(3, 2, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_RAS_REV2), .optional = 1 },
347 { PCI_DESCR(3, 4, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_TEST_REV2) },
348
349 /* Channel 0 */
350 { PCI_DESCR(4, 0, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH0_CTRL_REV2) },
351 { PCI_DESCR(4, 1, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH0_ADDR_REV2) },
352 { PCI_DESCR(4, 2, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH0_RANK_REV2) },
353 { PCI_DESCR(4, 3, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH0_TC_REV2) },
354
355 /* Channel 1 */
356 { PCI_DESCR(5, 0, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH1_CTRL_REV2) },
357 { PCI_DESCR(5, 1, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH1_ADDR_REV2) },
358 { PCI_DESCR(5, 2, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH1_RANK_REV2) },
359 { PCI_DESCR(5, 3, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH1_TC_REV2) },
360
361 /* Channel 2 */
362 { PCI_DESCR(6, 0, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH2_CTRL_REV2) },
363 { PCI_DESCR(6, 1, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH2_ADDR_REV2) },
364 { PCI_DESCR(6, 2, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH2_RANK_REV2) },
365 { PCI_DESCR(6, 3, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH2_TC_REV2) },
366
367 /* Generic Non-core registers */
368 { PCI_DESCR(0, 0, PCI_DEVICE_ID_INTEL_LYNNFIELD_NONCORE_REV2) },
369
370};
371
Mauro Carvalho Chehab1288c182010-08-10 18:57:01 -0300372#define PCI_ID_TABLE_ENTRY(A) { .descr=A, .n_devs = ARRAY_SIZE(A) }
373static const struct pci_id_table pci_dev_table[] = {
Vernon Mauerybd9e19c2010-05-18 19:02:50 -0300374 PCI_ID_TABLE_ENTRY(pci_dev_descr_i7core_nehalem),
375 PCI_ID_TABLE_ENTRY(pci_dev_descr_lynnfield),
376 PCI_ID_TABLE_ENTRY(pci_dev_descr_i7core_westmere),
377};
378
Mauro Carvalho Chehab8f331902009-06-22 22:48:29 -0300379/*
380 * pci_device_id table for which devices we are looking for
Mauro Carvalho Chehab8f331902009-06-22 22:48:29 -0300381 */
382static const struct pci_device_id i7core_pci_tbl[] __devinitdata = {
Mauro Carvalho Chehabd1fd4fb2009-07-10 18:39:53 -0300383 {PCI_DEVICE(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_X58_HUB_MGMT)},
Mauro Carvalho Chehabf05da2f2009-10-14 13:31:06 -0300384 {PCI_DEVICE(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_LYNNFIELD_QPI_LINK0)},
Mauro Carvalho Chehab8f331902009-06-22 22:48:29 -0300385 {0,} /* 0 terminated list. */
386};
387
Mauro Carvalho Chehaba0c36a12009-06-22 22:41:15 -0300388/****************************************************************************
389 Anciliary status routines
390 ****************************************************************************/
391
392 /* MC_CONTROL bits */
Mauro Carvalho Chehabef708b52009-06-22 22:48:30 -0300393#define CH_ACTIVE(pvt, ch) ((pvt)->info.mc_control & (1 << (8 + ch)))
394#define ECCx8(pvt) ((pvt)->info.mc_control & (1 << 1))
Mauro Carvalho Chehaba0c36a12009-06-22 22:41:15 -0300395
396 /* MC_STATUS bits */
Keith Mannthey61053fd2009-09-02 23:46:59 -0300397#define ECC_ENABLED(pvt) ((pvt)->info.mc_status & (1 << 4))
Mauro Carvalho Chehabef708b52009-06-22 22:48:30 -0300398#define CH_DISABLED(pvt, ch) ((pvt)->info.mc_status & (1 << ch))
Mauro Carvalho Chehaba0c36a12009-06-22 22:41:15 -0300399
400 /* MC_MAX_DOD read functions */
Mauro Carvalho Chehab854d3342009-06-22 22:48:30 -0300401static inline int numdimms(u32 dimms)
Mauro Carvalho Chehaba0c36a12009-06-22 22:41:15 -0300402{
Mauro Carvalho Chehab854d3342009-06-22 22:48:30 -0300403 return (dimms & 0x3) + 1;
Mauro Carvalho Chehaba0c36a12009-06-22 22:41:15 -0300404}
405
Mauro Carvalho Chehab854d3342009-06-22 22:48:30 -0300406static inline int numrank(u32 rank)
Mauro Carvalho Chehaba0c36a12009-06-22 22:41:15 -0300407{
408 static int ranks[4] = { 1, 2, 4, -EINVAL };
409
Mauro Carvalho Chehab854d3342009-06-22 22:48:30 -0300410 return ranks[rank & 0x3];
Mauro Carvalho Chehaba0c36a12009-06-22 22:41:15 -0300411}
412
Mauro Carvalho Chehab854d3342009-06-22 22:48:30 -0300413static inline int numbank(u32 bank)
Mauro Carvalho Chehaba0c36a12009-06-22 22:41:15 -0300414{
415 static int banks[4] = { 4, 8, 16, -EINVAL };
416
Mauro Carvalho Chehab854d3342009-06-22 22:48:30 -0300417 return banks[bank & 0x3];
Mauro Carvalho Chehaba0c36a12009-06-22 22:41:15 -0300418}
419
Mauro Carvalho Chehab854d3342009-06-22 22:48:30 -0300420static inline int numrow(u32 row)
Mauro Carvalho Chehaba0c36a12009-06-22 22:41:15 -0300421{
422 static int rows[8] = {
423 1 << 12, 1 << 13, 1 << 14, 1 << 15,
424 1 << 16, -EINVAL, -EINVAL, -EINVAL,
425 };
426
Mauro Carvalho Chehab854d3342009-06-22 22:48:30 -0300427 return rows[row & 0x7];
Mauro Carvalho Chehaba0c36a12009-06-22 22:41:15 -0300428}
429
Mauro Carvalho Chehab854d3342009-06-22 22:48:30 -0300430static inline int numcol(u32 col)
Mauro Carvalho Chehaba0c36a12009-06-22 22:41:15 -0300431{
432 static int cols[8] = {
433 1 << 10, 1 << 11, 1 << 12, -EINVAL,
434 };
Mauro Carvalho Chehab854d3342009-06-22 22:48:30 -0300435 return cols[col & 0x3];
Mauro Carvalho Chehaba0c36a12009-06-22 22:41:15 -0300436}
437
Mauro Carvalho Chehabf4742942009-09-05 02:35:08 -0300438static struct i7core_dev *get_i7core_dev(u8 socket)
Mauro Carvalho Chehab66607702009-09-05 00:52:11 -0300439{
440 struct i7core_dev *i7core_dev;
441
442 list_for_each_entry(i7core_dev, &i7core_edac_list, list) {
443 if (i7core_dev->socket == socket)
444 return i7core_dev;
445 }
446
447 return NULL;
448}
449
Hidetoshi Seto848b2f72010-08-20 04:24:44 -0300450static struct i7core_dev *alloc_i7core_dev(u8 socket,
451 const struct pci_id_table *table)
452{
453 struct i7core_dev *i7core_dev;
454
455 i7core_dev = kzalloc(sizeof(*i7core_dev), GFP_KERNEL);
456 if (!i7core_dev)
457 return NULL;
458
459 i7core_dev->pdev = kzalloc(sizeof(*i7core_dev->pdev) * table->n_devs,
460 GFP_KERNEL);
461 if (!i7core_dev->pdev) {
462 kfree(i7core_dev);
463 return NULL;
464 }
465
466 i7core_dev->socket = socket;
467 i7core_dev->n_devs = table->n_devs;
468 list_add_tail(&i7core_dev->list, &i7core_edac_list);
469
470 return i7core_dev;
471}
472
Hidetoshi Seto2aa9be42010-08-20 04:25:00 -0300473static void free_i7core_dev(struct i7core_dev *i7core_dev)
474{
475 list_del(&i7core_dev->list);
476 kfree(i7core_dev->pdev);
477 kfree(i7core_dev);
478}
479
Mauro Carvalho Chehaba0c36a12009-06-22 22:41:15 -0300480/****************************************************************************
481 Memory check routines
482 ****************************************************************************/
Mauro Carvalho Chehab67166af2009-07-15 06:56:23 -0300483static struct pci_dev *get_pdev_slot_func(u8 socket, unsigned slot,
484 unsigned func)
Mauro Carvalho Chehabef708b52009-06-22 22:48:30 -0300485{
Mauro Carvalho Chehab66607702009-09-05 00:52:11 -0300486 struct i7core_dev *i7core_dev = get_i7core_dev(socket);
Mauro Carvalho Chehabef708b52009-06-22 22:48:30 -0300487 int i;
Mauro Carvalho Chehabef708b52009-06-22 22:48:30 -0300488
Mauro Carvalho Chehab66607702009-09-05 00:52:11 -0300489 if (!i7core_dev)
490 return NULL;
491
Mauro Carvalho Chehabde06eee2009-10-14 08:02:40 -0300492 for (i = 0; i < i7core_dev->n_devs; i++) {
Mauro Carvalho Chehab66607702009-09-05 00:52:11 -0300493 if (!i7core_dev->pdev[i])
Mauro Carvalho Chehabef708b52009-06-22 22:48:30 -0300494 continue;
495
Mauro Carvalho Chehab66607702009-09-05 00:52:11 -0300496 if (PCI_SLOT(i7core_dev->pdev[i]->devfn) == slot &&
497 PCI_FUNC(i7core_dev->pdev[i]->devfn) == func) {
498 return i7core_dev->pdev[i];
Mauro Carvalho Chehabef708b52009-06-22 22:48:30 -0300499 }
500 }
501
Mauro Carvalho Chehabeb94fc42009-06-22 22:48:31 -0300502 return NULL;
503}
504
Mauro Carvalho Chehabec6df242009-07-18 10:44:30 -0300505/**
506 * i7core_get_active_channels() - gets the number of channels and csrows
507 * @socket: Quick Path Interconnect socket
508 * @channels: Number of channels that will be returned
509 * @csrows: Number of csrows found
510 *
511 * Since EDAC core needs to know in advance the number of available channels
512 * and csrows, in order to allocate memory for csrows/channels, it is needed
513 * to run two similar steps. At the first step, implemented on this function,
514 * it checks the number of csrows/channels present at one socket.
515 * this is used in order to properly allocate the size of mci components.
516 *
517 * It should be noticed that none of the current available datasheets explain
518 * or even mention how csrows are seen by the memory controller. So, we need
519 * to add a fake description for csrows.
520 * So, this driver is attributing one DIMM memory for one csrow.
521 */
Mauro Carvalho Chehab1288c182010-08-10 18:57:01 -0300522static int i7core_get_active_channels(const u8 socket, unsigned *channels,
Mauro Carvalho Chehab67166af2009-07-15 06:56:23 -0300523 unsigned *csrows)
Mauro Carvalho Chehabeb94fc42009-06-22 22:48:31 -0300524{
525 struct pci_dev *pdev = NULL;
526 int i, j;
527 u32 status, control;
528
529 *channels = 0;
530 *csrows = 0;
531
Mauro Carvalho Chehab67166af2009-07-15 06:56:23 -0300532 pdev = get_pdev_slot_func(socket, 3, 0);
Mauro Carvalho Chehabb7c76152009-06-22 22:48:30 -0300533 if (!pdev) {
Mauro Carvalho Chehab67166af2009-07-15 06:56:23 -0300534 i7core_printk(KERN_ERR, "Couldn't find socket %d fn 3.0!!!\n",
535 socket);
Mauro Carvalho Chehabef708b52009-06-22 22:48:30 -0300536 return -ENODEV;
Mauro Carvalho Chehabb7c76152009-06-22 22:48:30 -0300537 }
Mauro Carvalho Chehabef708b52009-06-22 22:48:30 -0300538
539 /* Device 3 function 0 reads */
540 pci_read_config_dword(pdev, MC_STATUS, &status);
541 pci_read_config_dword(pdev, MC_CONTROL, &control);
542
543 for (i = 0; i < NUM_CHANS; i++) {
Mauro Carvalho Chehabeb94fc42009-06-22 22:48:31 -0300544 u32 dimm_dod[3];
Mauro Carvalho Chehabef708b52009-06-22 22:48:30 -0300545 /* Check if the channel is active */
546 if (!(control & (1 << (8 + i))))
547 continue;
548
549 /* Check if the channel is disabled */
Mauro Carvalho Chehab41fcb7f2009-06-22 22:48:31 -0300550 if (status & (1 << i))
Mauro Carvalho Chehabef708b52009-06-22 22:48:30 -0300551 continue;
Mauro Carvalho Chehabef708b52009-06-22 22:48:30 -0300552
Mauro Carvalho Chehab67166af2009-07-15 06:56:23 -0300553 pdev = get_pdev_slot_func(socket, i + 4, 1);
Mauro Carvalho Chehabeb94fc42009-06-22 22:48:31 -0300554 if (!pdev) {
Mauro Carvalho Chehab67166af2009-07-15 06:56:23 -0300555 i7core_printk(KERN_ERR, "Couldn't find socket %d "
556 "fn %d.%d!!!\n",
557 socket, i + 4, 1);
Mauro Carvalho Chehabeb94fc42009-06-22 22:48:31 -0300558 return -ENODEV;
559 }
560 /* Devices 4-6 function 1 */
561 pci_read_config_dword(pdev,
562 MC_DOD_CH_DIMM0, &dimm_dod[0]);
563 pci_read_config_dword(pdev,
564 MC_DOD_CH_DIMM1, &dimm_dod[1]);
565 pci_read_config_dword(pdev,
566 MC_DOD_CH_DIMM2, &dimm_dod[2]);
567
Mauro Carvalho Chehabef708b52009-06-22 22:48:30 -0300568 (*channels)++;
Mauro Carvalho Chehabeb94fc42009-06-22 22:48:31 -0300569
570 for (j = 0; j < 3; j++) {
571 if (!DIMM_PRESENT(dimm_dod[j]))
572 continue;
573 (*csrows)++;
574 }
Mauro Carvalho Chehabef708b52009-06-22 22:48:30 -0300575 }
576
Mauro Carvalho Chehabc77720b2009-07-18 10:43:08 -0300577 debugf0("Number of active channels on socket %d: %d\n",
Mauro Carvalho Chehab67166af2009-07-15 06:56:23 -0300578 socket, *channels);
Mauro Carvalho Chehab1c6fed82009-06-22 22:48:30 -0300579
Mauro Carvalho Chehabef708b52009-06-22 22:48:30 -0300580 return 0;
581}
582
Mauro Carvalho Chehab1288c182010-08-10 18:57:01 -0300583static int get_dimm_config(const struct mem_ctl_info *mci, int *csrow)
Mauro Carvalho Chehaba0c36a12009-06-22 22:41:15 -0300584{
585 struct i7core_pvt *pvt = mci->pvt_info;
Mauro Carvalho Chehab1c6fed82009-06-22 22:48:30 -0300586 struct csrow_info *csr;
Mauro Carvalho Chehab854d3342009-06-22 22:48:30 -0300587 struct pci_dev *pdev;
Mauro Carvalho Chehabba6c5c62009-07-15 09:02:32 -0300588 int i, j;
Mauro Carvalho Chehab5566cb72009-06-22 22:48:31 -0300589 unsigned long last_page = 0;
Mauro Carvalho Chehab1c6fed82009-06-22 22:48:30 -0300590 enum edac_type mode;
Mauro Carvalho Chehab854d3342009-06-22 22:48:30 -0300591 enum mem_type mtype;
Mauro Carvalho Chehaba0c36a12009-06-22 22:41:15 -0300592
Mauro Carvalho Chehab854d3342009-06-22 22:48:30 -0300593 /* Get data from the MC register, function 0 */
Mauro Carvalho Chehabf4742942009-09-05 02:35:08 -0300594 pdev = pvt->pci_mcr[0];
Mauro Carvalho Chehab7dd69532009-06-22 22:48:30 -0300595 if (!pdev)
Mauro Carvalho Chehab8f331902009-06-22 22:48:29 -0300596 return -ENODEV;
597
Mauro Carvalho Chehabf122a892009-06-22 22:48:29 -0300598 /* Device 3 function 0 reads */
Mauro Carvalho Chehab7dd69532009-06-22 22:48:30 -0300599 pci_read_config_dword(pdev, MC_CONTROL, &pvt->info.mc_control);
600 pci_read_config_dword(pdev, MC_STATUS, &pvt->info.mc_status);
601 pci_read_config_dword(pdev, MC_MAX_DOD, &pvt->info.max_dod);
602 pci_read_config_dword(pdev, MC_CHANNEL_MAPPER, &pvt->info.ch_map);
Mauro Carvalho Chehabf122a892009-06-22 22:48:29 -0300603
Mauro Carvalho Chehab17cb7b02009-07-20 18:48:18 -0300604 debugf0("QPI %d control=0x%08x status=0x%08x dod=0x%08x map=0x%08x\n",
Mauro Carvalho Chehab4af91882009-09-24 09:58:26 -0300605 pvt->i7core_dev->socket, pvt->info.mc_control, pvt->info.mc_status,
Mauro Carvalho Chehabf122a892009-06-22 22:48:29 -0300606 pvt->info.max_dod, pvt->info.ch_map);
Mauro Carvalho Chehaba0c36a12009-06-22 22:41:15 -0300607
Mauro Carvalho Chehab1c6fed82009-06-22 22:48:30 -0300608 if (ECC_ENABLED(pvt)) {
Mauro Carvalho Chehab41fcb7f2009-06-22 22:48:31 -0300609 debugf0("ECC enabled with x%d SDCC\n", ECCx8(pvt) ? 8 : 4);
Mauro Carvalho Chehab1c6fed82009-06-22 22:48:30 -0300610 if (ECCx8(pvt))
611 mode = EDAC_S8ECD8ED;
612 else
613 mode = EDAC_S4ECD4ED;
614 } else {
Mauro Carvalho Chehaba0c36a12009-06-22 22:41:15 -0300615 debugf0("ECC disabled\n");
Mauro Carvalho Chehab1c6fed82009-06-22 22:48:30 -0300616 mode = EDAC_NONE;
617 }
Mauro Carvalho Chehaba0c36a12009-06-22 22:41:15 -0300618
619 /* FIXME: need to handle the error codes */
Mauro Carvalho Chehab17cb7b02009-07-20 18:48:18 -0300620 debugf0("DOD Max limits: DIMMS: %d, %d-ranked, %d-banked "
621 "x%x x 0x%x\n",
Mauro Carvalho Chehab854d3342009-06-22 22:48:30 -0300622 numdimms(pvt->info.max_dod),
623 numrank(pvt->info.max_dod >> 2),
Mauro Carvalho Chehab276b8242009-07-22 21:45:50 -0300624 numbank(pvt->info.max_dod >> 4),
Mauro Carvalho Chehab854d3342009-06-22 22:48:30 -0300625 numrow(pvt->info.max_dod >> 6),
626 numcol(pvt->info.max_dod >> 9));
Mauro Carvalho Chehaba0c36a12009-06-22 22:41:15 -0300627
Mauro Carvalho Chehab0b2b7b72009-06-22 22:48:29 -0300628 for (i = 0; i < NUM_CHANS; i++) {
Mauro Carvalho Chehab854d3342009-06-22 22:48:30 -0300629 u32 data, dimm_dod[3], value[8];
Mauro Carvalho Chehab0b2b7b72009-06-22 22:48:29 -0300630
Mauro Carvalho Chehab52a2e4fc2009-10-14 11:21:58 -0300631 if (!pvt->pci_ch[i][0])
632 continue;
633
Mauro Carvalho Chehab0b2b7b72009-06-22 22:48:29 -0300634 if (!CH_ACTIVE(pvt, i)) {
635 debugf0("Channel %i is not active\n", i);
636 continue;
637 }
638 if (CH_DISABLED(pvt, i)) {
639 debugf0("Channel %i is disabled\n", i);
640 continue;
641 }
642
Mauro Carvalho Chehabf122a892009-06-22 22:48:29 -0300643 /* Devices 4-6 function 0 */
Mauro Carvalho Chehabf4742942009-09-05 02:35:08 -0300644 pci_read_config_dword(pvt->pci_ch[i][0],
Mauro Carvalho Chehab0b2b7b72009-06-22 22:48:29 -0300645 MC_CHANNEL_DIMM_INIT_PARAMS, &data);
646
Mauro Carvalho Chehabf4742942009-09-05 02:35:08 -0300647 pvt->channel[i].ranks = (data & QUAD_RANK_PRESENT) ?
Mauro Carvalho Chehab67166af2009-07-15 06:56:23 -0300648 4 : 2;
Mauro Carvalho Chehab0b2b7b72009-06-22 22:48:29 -0300649
Mauro Carvalho Chehab854d3342009-06-22 22:48:30 -0300650 if (data & REGISTERED_DIMM)
651 mtype = MEM_RDDR3;
Mauro Carvalho Chehab14d2c082009-09-02 23:52:36 -0300652 else
Mauro Carvalho Chehab854d3342009-06-22 22:48:30 -0300653 mtype = MEM_DDR3;
654#if 0
Mauro Carvalho Chehab0b2b7b72009-06-22 22:48:29 -0300655 if (data & THREE_DIMMS_PRESENT)
656 pvt->channel[i].dimms = 3;
657 else if (data & SINGLE_QUAD_RANK_PRESENT)
658 pvt->channel[i].dimms = 1;
659 else
660 pvt->channel[i].dimms = 2;
Mauro Carvalho Chehab854d3342009-06-22 22:48:30 -0300661#endif
662
663 /* Devices 4-6 function 1 */
Mauro Carvalho Chehabf4742942009-09-05 02:35:08 -0300664 pci_read_config_dword(pvt->pci_ch[i][1],
Mauro Carvalho Chehab854d3342009-06-22 22:48:30 -0300665 MC_DOD_CH_DIMM0, &dimm_dod[0]);
Mauro Carvalho Chehabf4742942009-09-05 02:35:08 -0300666 pci_read_config_dword(pvt->pci_ch[i][1],
Mauro Carvalho Chehab854d3342009-06-22 22:48:30 -0300667 MC_DOD_CH_DIMM1, &dimm_dod[1]);
Mauro Carvalho Chehabf4742942009-09-05 02:35:08 -0300668 pci_read_config_dword(pvt->pci_ch[i][1],
Mauro Carvalho Chehab854d3342009-06-22 22:48:30 -0300669 MC_DOD_CH_DIMM2, &dimm_dod[2]);
Mauro Carvalho Chehab0b2b7b72009-06-22 22:48:29 -0300670
Mauro Carvalho Chehab1c6fed82009-06-22 22:48:30 -0300671 debugf0("Ch%d phy rd%d, wr%d (0x%08x): "
Mauro Carvalho Chehab854d3342009-06-22 22:48:30 -0300672 "%d ranks, %cDIMMs\n",
Mauro Carvalho Chehab1c6fed82009-06-22 22:48:30 -0300673 i,
674 RDLCH(pvt->info.ch_map, i), WRLCH(pvt->info.ch_map, i),
675 data,
Mauro Carvalho Chehabf4742942009-09-05 02:35:08 -0300676 pvt->channel[i].ranks,
Mauro Carvalho Chehab41fcb7f2009-06-22 22:48:31 -0300677 (data & REGISTERED_DIMM) ? 'R' : 'U');
Mauro Carvalho Chehab7dd69532009-06-22 22:48:30 -0300678
Mauro Carvalho Chehab854d3342009-06-22 22:48:30 -0300679 for (j = 0; j < 3; j++) {
680 u32 banks, ranks, rows, cols;
Mauro Carvalho Chehab5566cb72009-06-22 22:48:31 -0300681 u32 size, npages;
Mauro Carvalho Chehab1c6fed82009-06-22 22:48:30 -0300682
Mauro Carvalho Chehab854d3342009-06-22 22:48:30 -0300683 if (!DIMM_PRESENT(dimm_dod[j]))
684 continue;
Mauro Carvalho Chehab1c6fed82009-06-22 22:48:30 -0300685
Mauro Carvalho Chehab854d3342009-06-22 22:48:30 -0300686 banks = numbank(MC_DOD_NUMBANK(dimm_dod[j]));
687 ranks = numrank(MC_DOD_NUMRANK(dimm_dod[j]));
688 rows = numrow(MC_DOD_NUMROW(dimm_dod[j]));
689 cols = numcol(MC_DOD_NUMCOL(dimm_dod[j]));
Mauro Carvalho Chehab1c6fed82009-06-22 22:48:30 -0300690
Mauro Carvalho Chehab5566cb72009-06-22 22:48:31 -0300691 /* DDR3 has 8 I/O banks */
692 size = (rows * cols * banks * ranks) >> (20 - 3);
693
Mauro Carvalho Chehabf4742942009-09-05 02:35:08 -0300694 pvt->channel[i].dimms++;
Mauro Carvalho Chehab854d3342009-06-22 22:48:30 -0300695
Mauro Carvalho Chehab17cb7b02009-07-20 18:48:18 -0300696 debugf0("\tdimm %d %d Mb offset: %x, "
697 "bank: %d, rank: %d, row: %#x, col: %#x\n",
698 j, size,
Mauro Carvalho Chehab854d3342009-06-22 22:48:30 -0300699 RANKOFFSET(dimm_dod[j]),
700 banks, ranks, rows, cols);
701
Mauro Carvalho Chehabe9144602010-08-10 20:26:35 -0300702 npages = MiB_TO_PAGES(size);
Mauro Carvalho Chehab5566cb72009-06-22 22:48:31 -0300703
Mauro Carvalho Chehabba6c5c62009-07-15 09:02:32 -0300704 csr = &mci->csrows[*csrow];
Mauro Carvalho Chehab5566cb72009-06-22 22:48:31 -0300705 csr->first_page = last_page + 1;
706 last_page += npages;
707 csr->last_page = last_page;
708 csr->nr_pages = npages;
709
Mauro Carvalho Chehab854d3342009-06-22 22:48:30 -0300710 csr->page_mask = 0;
Mauro Carvalho Chehabeb94fc42009-06-22 22:48:31 -0300711 csr->grain = 8;
Mauro Carvalho Chehabba6c5c62009-07-15 09:02:32 -0300712 csr->csrow_idx = *csrow;
Mauro Carvalho Chehabeb94fc42009-06-22 22:48:31 -0300713 csr->nr_channels = 1;
714
715 csr->channels[0].chan_idx = i;
716 csr->channels[0].ce_count = 0;
Mauro Carvalho Chehab854d3342009-06-22 22:48:30 -0300717
Mauro Carvalho Chehabf4742942009-09-05 02:35:08 -0300718 pvt->csrow_map[i][j] = *csrow;
Mauro Carvalho Chehabb4e8f0b2009-09-02 23:49:59 -0300719
Mauro Carvalho Chehab854d3342009-06-22 22:48:30 -0300720 switch (banks) {
721 case 4:
722 csr->dtype = DEV_X4;
723 break;
724 case 8:
725 csr->dtype = DEV_X8;
726 break;
727 case 16:
728 csr->dtype = DEV_X16;
729 break;
730 default:
731 csr->dtype = DEV_UNKNOWN;
732 }
733
734 csr->edac_mode = mode;
735 csr->mtype = mtype;
736
Mauro Carvalho Chehabba6c5c62009-07-15 09:02:32 -0300737 (*csrow)++;
Mauro Carvalho Chehab854d3342009-06-22 22:48:30 -0300738 }
739
740 pci_read_config_dword(pdev, MC_SAG_CH_0, &value[0]);
741 pci_read_config_dword(pdev, MC_SAG_CH_1, &value[1]);
742 pci_read_config_dword(pdev, MC_SAG_CH_2, &value[2]);
743 pci_read_config_dword(pdev, MC_SAG_CH_3, &value[3]);
744 pci_read_config_dword(pdev, MC_SAG_CH_4, &value[4]);
745 pci_read_config_dword(pdev, MC_SAG_CH_5, &value[5]);
746 pci_read_config_dword(pdev, MC_SAG_CH_6, &value[6]);
747 pci_read_config_dword(pdev, MC_SAG_CH_7, &value[7]);
Mauro Carvalho Chehab17cb7b02009-07-20 18:48:18 -0300748 debugf1("\t[%i] DIVBY3\tREMOVED\tOFFSET\n", i);
Mauro Carvalho Chehab854d3342009-06-22 22:48:30 -0300749 for (j = 0; j < 8; j++)
Mauro Carvalho Chehab17cb7b02009-07-20 18:48:18 -0300750 debugf1("\t\t%#x\t%#x\t%#x\n",
Mauro Carvalho Chehab854d3342009-06-22 22:48:30 -0300751 (value[j] >> 27) & 0x1,
752 (value[j] >> 24) & 0x7,
753 (value[j] && ((1 << 24) - 1)));
Mauro Carvalho Chehab0b2b7b72009-06-22 22:48:29 -0300754 }
755
Mauro Carvalho Chehaba0c36a12009-06-22 22:41:15 -0300756 return 0;
757}
758
759/****************************************************************************
Mauro Carvalho Chehab194a40f2009-06-22 22:48:28 -0300760 Error insertion routines
761 ****************************************************************************/
762
763/* The i7core has independent error injection features per channel.
764 However, to have a simpler code, we don't allow enabling error injection
765 on more than one channel.
766 Also, since a change at an inject parameter will be applied only at enable,
767 we're disabling error injection on all write calls to the sysfs nodes that
768 controls the error code injection.
769 */
Mauro Carvalho Chehab1288c182010-08-10 18:57:01 -0300770static int disable_inject(const struct mem_ctl_info *mci)
Mauro Carvalho Chehab194a40f2009-06-22 22:48:28 -0300771{
772 struct i7core_pvt *pvt = mci->pvt_info;
773
774 pvt->inject.enable = 0;
775
Mauro Carvalho Chehabf4742942009-09-05 02:35:08 -0300776 if (!pvt->pci_ch[pvt->inject.channel][0])
Mauro Carvalho Chehab8f331902009-06-22 22:48:29 -0300777 return -ENODEV;
778
Mauro Carvalho Chehabf4742942009-09-05 02:35:08 -0300779 pci_write_config_dword(pvt->pci_ch[pvt->inject.channel][0],
Mauro Carvalho Chehab4157d9f2009-08-05 20:27:15 -0300780 MC_CHANNEL_ERROR_INJECT, 0);
Mauro Carvalho Chehab8f331902009-06-22 22:48:29 -0300781
782 return 0;
Mauro Carvalho Chehab194a40f2009-06-22 22:48:28 -0300783}
784
785/*
786 * i7core inject inject.section
787 *
788 * accept and store error injection inject.section value
789 * bit 0 - refers to the lower 32-byte half cacheline
790 * bit 1 - refers to the upper 32-byte half cacheline
791 */
792static ssize_t i7core_inject_section_store(struct mem_ctl_info *mci,
793 const char *data, size_t count)
794{
795 struct i7core_pvt *pvt = mci->pvt_info;
796 unsigned long value;
797 int rc;
798
799 if (pvt->inject.enable)
Mauro Carvalho Chehab41fcb7f2009-06-22 22:48:31 -0300800 disable_inject(mci);
Mauro Carvalho Chehab194a40f2009-06-22 22:48:28 -0300801
802 rc = strict_strtoul(data, 10, &value);
803 if ((rc < 0) || (value > 3))
Mauro Carvalho Chehab2068def2009-08-05 19:28:27 -0300804 return -EIO;
Mauro Carvalho Chehab194a40f2009-06-22 22:48:28 -0300805
806 pvt->inject.section = (u32) value;
807 return count;
808}
809
810static ssize_t i7core_inject_section_show(struct mem_ctl_info *mci,
811 char *data)
812{
813 struct i7core_pvt *pvt = mci->pvt_info;
814 return sprintf(data, "0x%08x\n", pvt->inject.section);
815}
816
817/*
818 * i7core inject.type
819 *
820 * accept and store error injection inject.section value
821 * bit 0 - repeat enable - Enable error repetition
822 * bit 1 - inject ECC error
823 * bit 2 - inject parity error
824 */
825static ssize_t i7core_inject_type_store(struct mem_ctl_info *mci,
826 const char *data, size_t count)
827{
828 struct i7core_pvt *pvt = mci->pvt_info;
829 unsigned long value;
830 int rc;
831
832 if (pvt->inject.enable)
Mauro Carvalho Chehab41fcb7f2009-06-22 22:48:31 -0300833 disable_inject(mci);
Mauro Carvalho Chehab194a40f2009-06-22 22:48:28 -0300834
835 rc = strict_strtoul(data, 10, &value);
836 if ((rc < 0) || (value > 7))
Mauro Carvalho Chehab2068def2009-08-05 19:28:27 -0300837 return -EIO;
Mauro Carvalho Chehab194a40f2009-06-22 22:48:28 -0300838
839 pvt->inject.type = (u32) value;
840 return count;
841}
842
843static ssize_t i7core_inject_type_show(struct mem_ctl_info *mci,
844 char *data)
845{
846 struct i7core_pvt *pvt = mci->pvt_info;
847 return sprintf(data, "0x%08x\n", pvt->inject.type);
848}
849
850/*
851 * i7core_inject_inject.eccmask_store
852 *
853 * The type of error (UE/CE) will depend on the inject.eccmask value:
854 * Any bits set to a 1 will flip the corresponding ECC bit
855 * Correctable errors can be injected by flipping 1 bit or the bits within
856 * a symbol pair (2 consecutive aligned 8-bit pairs - i.e. 7:0 and 15:8 or
857 * 23:16 and 31:24). Flipping bits in two symbol pairs will cause an
858 * uncorrectable error to be injected.
859 */
860static ssize_t i7core_inject_eccmask_store(struct mem_ctl_info *mci,
861 const char *data, size_t count)
862{
863 struct i7core_pvt *pvt = mci->pvt_info;
864 unsigned long value;
865 int rc;
866
867 if (pvt->inject.enable)
Mauro Carvalho Chehab41fcb7f2009-06-22 22:48:31 -0300868 disable_inject(mci);
Mauro Carvalho Chehab194a40f2009-06-22 22:48:28 -0300869
870 rc = strict_strtoul(data, 10, &value);
871 if (rc < 0)
Mauro Carvalho Chehab2068def2009-08-05 19:28:27 -0300872 return -EIO;
Mauro Carvalho Chehab194a40f2009-06-22 22:48:28 -0300873
874 pvt->inject.eccmask = (u32) value;
875 return count;
876}
877
878static ssize_t i7core_inject_eccmask_show(struct mem_ctl_info *mci,
879 char *data)
880{
881 struct i7core_pvt *pvt = mci->pvt_info;
882 return sprintf(data, "0x%08x\n", pvt->inject.eccmask);
883}
884
885/*
886 * i7core_addrmatch
887 *
888 * The type of error (UE/CE) will depend on the inject.eccmask value:
889 * Any bits set to a 1 will flip the corresponding ECC bit
890 * Correctable errors can be injected by flipping 1 bit or the bits within
891 * a symbol pair (2 consecutive aligned 8-bit pairs - i.e. 7:0 and 15:8 or
892 * 23:16 and 31:24). Flipping bits in two symbol pairs will cause an
893 * uncorrectable error to be injected.
894 */
Mauro Carvalho Chehab194a40f2009-06-22 22:48:28 -0300895
Mauro Carvalho Chehaba5538e52009-09-23 18:56:47 -0300896#define DECLARE_ADDR_MATCH(param, limit) \
897static ssize_t i7core_inject_store_##param( \
898 struct mem_ctl_info *mci, \
899 const char *data, size_t count) \
900{ \
Mauro Carvalho Chehabcc301b32009-09-24 16:23:42 -0300901 struct i7core_pvt *pvt; \
Mauro Carvalho Chehaba5538e52009-09-23 18:56:47 -0300902 long value; \
903 int rc; \
904 \
Mauro Carvalho Chehabcc301b32009-09-24 16:23:42 -0300905 debugf1("%s()\n", __func__); \
906 pvt = mci->pvt_info; \
907 \
Mauro Carvalho Chehaba5538e52009-09-23 18:56:47 -0300908 if (pvt->inject.enable) \
909 disable_inject(mci); \
910 \
Mauro Carvalho Chehab4f87fad2009-10-04 11:54:56 -0300911 if (!strcasecmp(data, "any") || !strcasecmp(data, "any\n"))\
Mauro Carvalho Chehaba5538e52009-09-23 18:56:47 -0300912 value = -1; \
913 else { \
914 rc = strict_strtoul(data, 10, &value); \
915 if ((rc < 0) || (value >= limit)) \
916 return -EIO; \
917 } \
918 \
919 pvt->inject.param = value; \
920 \
921 return count; \
922} \
923 \
924static ssize_t i7core_inject_show_##param( \
925 struct mem_ctl_info *mci, \
926 char *data) \
927{ \
Mauro Carvalho Chehabcc301b32009-09-24 16:23:42 -0300928 struct i7core_pvt *pvt; \
929 \
930 pvt = mci->pvt_info; \
931 debugf1("%s() pvt=%p\n", __func__, pvt); \
Mauro Carvalho Chehaba5538e52009-09-23 18:56:47 -0300932 if (pvt->inject.param < 0) \
933 return sprintf(data, "any\n"); \
934 else \
935 return sprintf(data, "%d\n", pvt->inject.param);\
Mauro Carvalho Chehab194a40f2009-06-22 22:48:28 -0300936}
937
Mauro Carvalho Chehaba5538e52009-09-23 18:56:47 -0300938#define ATTR_ADDR_MATCH(param) \
939 { \
940 .attr = { \
941 .name = #param, \
942 .mode = (S_IRUGO | S_IWUSR) \
943 }, \
944 .show = i7core_inject_show_##param, \
945 .store = i7core_inject_store_##param, \
946 }
Mauro Carvalho Chehab194a40f2009-06-22 22:48:28 -0300947
Mauro Carvalho Chehaba5538e52009-09-23 18:56:47 -0300948DECLARE_ADDR_MATCH(channel, 3);
949DECLARE_ADDR_MATCH(dimm, 3);
950DECLARE_ADDR_MATCH(rank, 4);
951DECLARE_ADDR_MATCH(bank, 32);
952DECLARE_ADDR_MATCH(page, 0x10000);
953DECLARE_ADDR_MATCH(col, 0x4000);
Mauro Carvalho Chehab194a40f2009-06-22 22:48:28 -0300954
Mauro Carvalho Chehab1288c182010-08-10 18:57:01 -0300955static int write_and_test(struct pci_dev *dev, const int where, const u32 val)
Mauro Carvalho Chehab276b8242009-07-22 21:45:50 -0300956{
957 u32 read;
958 int count;
959
Mauro Carvalho Chehab4157d9f2009-08-05 20:27:15 -0300960 debugf0("setting pci %02x:%02x.%x reg=%02x value=%08x\n",
961 dev->bus->number, PCI_SLOT(dev->devfn), PCI_FUNC(dev->devfn),
962 where, val);
963
Mauro Carvalho Chehab276b8242009-07-22 21:45:50 -0300964 for (count = 0; count < 10; count++) {
965 if (count)
Mauro Carvalho Chehabb9905382009-08-05 21:36:35 -0300966 msleep(100);
Mauro Carvalho Chehab276b8242009-07-22 21:45:50 -0300967 pci_write_config_dword(dev, where, val);
968 pci_read_config_dword(dev, where, &read);
969
970 if (read == val)
971 return 0;
972 }
973
Mauro Carvalho Chehab4157d9f2009-08-05 20:27:15 -0300974 i7core_printk(KERN_ERR, "Error during set pci %02x:%02x.%x reg=%02x "
975 "write=%08x. Read=%08x\n",
976 dev->bus->number, PCI_SLOT(dev->devfn), PCI_FUNC(dev->devfn),
977 where, val, read);
Mauro Carvalho Chehab276b8242009-07-22 21:45:50 -0300978
979 return -EINVAL;
980}
981
Mauro Carvalho Chehab194a40f2009-06-22 22:48:28 -0300982/*
983 * This routine prepares the Memory Controller for error injection.
984 * The error will be injected when some process tries to write to the
985 * memory that matches the given criteria.
986 * The criteria can be set in terms of a mask where dimm, rank, bank, page
987 * and col can be specified.
988 * A -1 value for any of the mask items will make the MCU to ignore
989 * that matching criteria for error injection.
990 *
991 * It should be noticed that the error will only happen after a write operation
992 * on a memory that matches the condition. if REPEAT_EN is not enabled at
993 * inject mask, then it will produce just one error. Otherwise, it will repeat
994 * until the injectmask would be cleaned.
995 *
996 * FIXME: This routine assumes that MAXNUMDIMMS value of MC_MAX_DOD
997 * is reliable enough to check if the MC is using the
998 * three channels. However, this is not clear at the datasheet.
999 */
1000static ssize_t i7core_inject_enable_store(struct mem_ctl_info *mci,
1001 const char *data, size_t count)
1002{
1003 struct i7core_pvt *pvt = mci->pvt_info;
1004 u32 injectmask;
1005 u64 mask = 0;
1006 int rc;
1007 long enable;
1008
Mauro Carvalho Chehabf4742942009-09-05 02:35:08 -03001009 if (!pvt->pci_ch[pvt->inject.channel][0])
Mauro Carvalho Chehab8f331902009-06-22 22:48:29 -03001010 return 0;
1011
Mauro Carvalho Chehab194a40f2009-06-22 22:48:28 -03001012 rc = strict_strtoul(data, 10, &enable);
1013 if ((rc < 0))
1014 return 0;
1015
1016 if (enable) {
1017 pvt->inject.enable = 1;
1018 } else {
1019 disable_inject(mci);
1020 return count;
1021 }
1022
1023 /* Sets pvt->inject.dimm mask */
1024 if (pvt->inject.dimm < 0)
Alan Cox486dd092009-11-08 01:34:27 -02001025 mask |= 1LL << 41;
Mauro Carvalho Chehab194a40f2009-06-22 22:48:28 -03001026 else {
Mauro Carvalho Chehabf4742942009-09-05 02:35:08 -03001027 if (pvt->channel[pvt->inject.channel].dimms > 2)
Alan Cox486dd092009-11-08 01:34:27 -02001028 mask |= (pvt->inject.dimm & 0x3LL) << 35;
Mauro Carvalho Chehab194a40f2009-06-22 22:48:28 -03001029 else
Alan Cox486dd092009-11-08 01:34:27 -02001030 mask |= (pvt->inject.dimm & 0x1LL) << 36;
Mauro Carvalho Chehab194a40f2009-06-22 22:48:28 -03001031 }
1032
1033 /* Sets pvt->inject.rank mask */
1034 if (pvt->inject.rank < 0)
Alan Cox486dd092009-11-08 01:34:27 -02001035 mask |= 1LL << 40;
Mauro Carvalho Chehab194a40f2009-06-22 22:48:28 -03001036 else {
Mauro Carvalho Chehabf4742942009-09-05 02:35:08 -03001037 if (pvt->channel[pvt->inject.channel].dimms > 2)
Alan Cox486dd092009-11-08 01:34:27 -02001038 mask |= (pvt->inject.rank & 0x1LL) << 34;
Mauro Carvalho Chehab194a40f2009-06-22 22:48:28 -03001039 else
Alan Cox486dd092009-11-08 01:34:27 -02001040 mask |= (pvt->inject.rank & 0x3LL) << 34;
Mauro Carvalho Chehab194a40f2009-06-22 22:48:28 -03001041 }
1042
1043 /* Sets pvt->inject.bank mask */
1044 if (pvt->inject.bank < 0)
Alan Cox486dd092009-11-08 01:34:27 -02001045 mask |= 1LL << 39;
Mauro Carvalho Chehab194a40f2009-06-22 22:48:28 -03001046 else
Alan Cox486dd092009-11-08 01:34:27 -02001047 mask |= (pvt->inject.bank & 0x15LL) << 30;
Mauro Carvalho Chehab194a40f2009-06-22 22:48:28 -03001048
1049 /* Sets pvt->inject.page mask */
1050 if (pvt->inject.page < 0)
Alan Cox486dd092009-11-08 01:34:27 -02001051 mask |= 1LL << 38;
Mauro Carvalho Chehab194a40f2009-06-22 22:48:28 -03001052 else
Alan Cox486dd092009-11-08 01:34:27 -02001053 mask |= (pvt->inject.page & 0xffff) << 14;
Mauro Carvalho Chehab194a40f2009-06-22 22:48:28 -03001054
1055 /* Sets pvt->inject.column mask */
1056 if (pvt->inject.col < 0)
Alan Cox486dd092009-11-08 01:34:27 -02001057 mask |= 1LL << 37;
Mauro Carvalho Chehab194a40f2009-06-22 22:48:28 -03001058 else
Alan Cox486dd092009-11-08 01:34:27 -02001059 mask |= (pvt->inject.col & 0x3fff);
Mauro Carvalho Chehab194a40f2009-06-22 22:48:28 -03001060
Mauro Carvalho Chehab194a40f2009-06-22 22:48:28 -03001061 /*
1062 * bit 0: REPEAT_EN
1063 * bits 1-2: MASK_HALF_CACHELINE
1064 * bit 3: INJECT_ECC
1065 * bit 4: INJECT_ADDR_PARITY
1066 */
1067
Mauro Carvalho Chehab7b029d02009-06-22 22:48:29 -03001068 injectmask = (pvt->inject.type & 1) |
1069 (pvt->inject.section & 0x3) << 1 |
Mauro Carvalho Chehab194a40f2009-06-22 22:48:28 -03001070 (pvt->inject.type & 0x6) << (3 - 1);
1071
Mauro Carvalho Chehab276b8242009-07-22 21:45:50 -03001072 /* Unlock writes to registers - this register is write only */
Mauro Carvalho Chehabf4742942009-09-05 02:35:08 -03001073 pci_write_config_dword(pvt->pci_noncore,
Mauro Carvalho Chehab276b8242009-07-22 21:45:50 -03001074 MC_CFG_CONTROL, 0x2);
Mauro Carvalho Chehab194a40f2009-06-22 22:48:28 -03001075
Mauro Carvalho Chehabf4742942009-09-05 02:35:08 -03001076 write_and_test(pvt->pci_ch[pvt->inject.channel][0],
Mauro Carvalho Chehab276b8242009-07-22 21:45:50 -03001077 MC_CHANNEL_ADDR_MATCH, mask);
Mauro Carvalho Chehabf4742942009-09-05 02:35:08 -03001078 write_and_test(pvt->pci_ch[pvt->inject.channel][0],
Mauro Carvalho Chehab276b8242009-07-22 21:45:50 -03001079 MC_CHANNEL_ADDR_MATCH + 4, mask >> 32L);
1080
Mauro Carvalho Chehabf4742942009-09-05 02:35:08 -03001081 write_and_test(pvt->pci_ch[pvt->inject.channel][0],
Mauro Carvalho Chehab276b8242009-07-22 21:45:50 -03001082 MC_CHANNEL_ERROR_MASK, pvt->inject.eccmask);
1083
Mauro Carvalho Chehabf4742942009-09-05 02:35:08 -03001084 write_and_test(pvt->pci_ch[pvt->inject.channel][0],
Mauro Carvalho Chehab4157d9f2009-08-05 20:27:15 -03001085 MC_CHANNEL_ERROR_INJECT, injectmask);
Mauro Carvalho Chehab276b8242009-07-22 21:45:50 -03001086
1087 /*
1088 * This is something undocumented, based on my tests
1089 * Without writing 8 to this register, errors aren't injected. Not sure
1090 * why.
1091 */
Mauro Carvalho Chehabf4742942009-09-05 02:35:08 -03001092 pci_write_config_dword(pvt->pci_noncore,
Mauro Carvalho Chehab276b8242009-07-22 21:45:50 -03001093 MC_CFG_CONTROL, 8);
1094
Mauro Carvalho Chehab41fcb7f2009-06-22 22:48:31 -03001095 debugf0("Error inject addr match 0x%016llx, ecc 0x%08x,"
1096 " inject 0x%08x\n",
Mauro Carvalho Chehab194a40f2009-06-22 22:48:28 -03001097 mask, pvt->inject.eccmask, injectmask);
1098
Mauro Carvalho Chehab7b029d02009-06-22 22:48:29 -03001099
Mauro Carvalho Chehab194a40f2009-06-22 22:48:28 -03001100 return count;
1101}
1102
1103static ssize_t i7core_inject_enable_show(struct mem_ctl_info *mci,
1104 char *data)
1105{
1106 struct i7core_pvt *pvt = mci->pvt_info;
Mauro Carvalho Chehab7b029d02009-06-22 22:48:29 -03001107 u32 injectmask;
1108
Mauro Carvalho Chehab52a2e4fc2009-10-14 11:21:58 -03001109 if (!pvt->pci_ch[pvt->inject.channel][0])
1110 return 0;
1111
Mauro Carvalho Chehabf4742942009-09-05 02:35:08 -03001112 pci_read_config_dword(pvt->pci_ch[pvt->inject.channel][0],
Mauro Carvalho Chehab4157d9f2009-08-05 20:27:15 -03001113 MC_CHANNEL_ERROR_INJECT, &injectmask);
Mauro Carvalho Chehab7b029d02009-06-22 22:48:29 -03001114
1115 debugf0("Inject error read: 0x%018x\n", injectmask);
1116
1117 if (injectmask & 0x0c)
1118 pvt->inject.enable = 1;
1119
Mauro Carvalho Chehab194a40f2009-06-22 22:48:28 -03001120 return sprintf(data, "%d\n", pvt->inject.enable);
1121}
1122
Mauro Carvalho Chehabf338d732009-09-24 17:25:43 -03001123#define DECLARE_COUNTER(param) \
1124static ssize_t i7core_show_counter_##param( \
1125 struct mem_ctl_info *mci, \
1126 char *data) \
1127{ \
1128 struct i7core_pvt *pvt = mci->pvt_info; \
1129 \
1130 debugf1("%s() \n", __func__); \
1131 if (!pvt->ce_count_available || (pvt->is_registered)) \
1132 return sprintf(data, "data unavailable\n"); \
1133 return sprintf(data, "%lu\n", \
1134 pvt->udimm_ce_count[param]); \
Mauro Carvalho Chehab442305b2009-06-22 22:48:29 -03001135}
1136
Mauro Carvalho Chehabf338d732009-09-24 17:25:43 -03001137#define ATTR_COUNTER(param) \
1138 { \
1139 .attr = { \
1140 .name = __stringify(udimm##param), \
1141 .mode = (S_IRUGO | S_IWUSR) \
1142 }, \
1143 .show = i7core_show_counter_##param \
1144 }
1145
1146DECLARE_COUNTER(0);
1147DECLARE_COUNTER(1);
1148DECLARE_COUNTER(2);
1149
Mauro Carvalho Chehab194a40f2009-06-22 22:48:28 -03001150/*
1151 * Sysfs struct
1152 */
Mauro Carvalho Chehaba5538e52009-09-23 18:56:47 -03001153
Mauro Carvalho Chehab1288c182010-08-10 18:57:01 -03001154static const struct mcidev_sysfs_attribute i7core_addrmatch_attrs[] = {
Mauro Carvalho Chehaba5538e52009-09-23 18:56:47 -03001155 ATTR_ADDR_MATCH(channel),
1156 ATTR_ADDR_MATCH(dimm),
1157 ATTR_ADDR_MATCH(rank),
1158 ATTR_ADDR_MATCH(bank),
1159 ATTR_ADDR_MATCH(page),
1160 ATTR_ADDR_MATCH(col),
Mauro Carvalho Chehab1288c182010-08-10 18:57:01 -03001161 { } /* End of list */
Mauro Carvalho Chehaba5538e52009-09-23 18:56:47 -03001162};
1163
Mauro Carvalho Chehab1288c182010-08-10 18:57:01 -03001164static const struct mcidev_sysfs_group i7core_inject_addrmatch = {
Mauro Carvalho Chehaba5538e52009-09-23 18:56:47 -03001165 .name = "inject_addrmatch",
1166 .mcidev_attr = i7core_addrmatch_attrs,
1167};
1168
Mauro Carvalho Chehab1288c182010-08-10 18:57:01 -03001169static const struct mcidev_sysfs_attribute i7core_udimm_counters_attrs[] = {
Mauro Carvalho Chehabf338d732009-09-24 17:25:43 -03001170 ATTR_COUNTER(0),
1171 ATTR_COUNTER(1),
1172 ATTR_COUNTER(2),
Marcin Slusarz64aab722010-09-30 15:15:30 -07001173 { .attr = { .name = NULL } }
Mauro Carvalho Chehabf338d732009-09-24 17:25:43 -03001174};
1175
Mauro Carvalho Chehab1288c182010-08-10 18:57:01 -03001176static const struct mcidev_sysfs_group i7core_udimm_counters = {
Mauro Carvalho Chehabf338d732009-09-24 17:25:43 -03001177 .name = "all_channel_counts",
1178 .mcidev_attr = i7core_udimm_counters_attrs,
1179};
1180
Mauro Carvalho Chehab1288c182010-08-10 18:57:01 -03001181static const struct mcidev_sysfs_attribute i7core_sysfs_rdimm_attrs[] = {
Mauro Carvalho Chehab194a40f2009-06-22 22:48:28 -03001182 {
1183 .attr = {
1184 .name = "inject_section",
1185 .mode = (S_IRUGO | S_IWUSR)
1186 },
1187 .show = i7core_inject_section_show,
1188 .store = i7core_inject_section_store,
1189 }, {
1190 .attr = {
1191 .name = "inject_type",
1192 .mode = (S_IRUGO | S_IWUSR)
1193 },
1194 .show = i7core_inject_type_show,
1195 .store = i7core_inject_type_store,
1196 }, {
1197 .attr = {
1198 .name = "inject_eccmask",
1199 .mode = (S_IRUGO | S_IWUSR)
1200 },
1201 .show = i7core_inject_eccmask_show,
1202 .store = i7core_inject_eccmask_store,
1203 }, {
Mauro Carvalho Chehaba5538e52009-09-23 18:56:47 -03001204 .grp = &i7core_inject_addrmatch,
Mauro Carvalho Chehab194a40f2009-06-22 22:48:28 -03001205 }, {
1206 .attr = {
1207 .name = "inject_enable",
1208 .mode = (S_IRUGO | S_IWUSR)
1209 },
1210 .show = i7core_inject_enable_show,
1211 .store = i7core_inject_enable_store,
1212 },
Mauro Carvalho Chehab1288c182010-08-10 18:57:01 -03001213 { } /* End of list */
1214};
1215
1216static const struct mcidev_sysfs_attribute i7core_sysfs_udimm_attrs[] = {
1217 {
1218 .attr = {
1219 .name = "inject_section",
1220 .mode = (S_IRUGO | S_IWUSR)
1221 },
1222 .show = i7core_inject_section_show,
1223 .store = i7core_inject_section_store,
1224 }, {
1225 .attr = {
1226 .name = "inject_type",
1227 .mode = (S_IRUGO | S_IWUSR)
1228 },
1229 .show = i7core_inject_type_show,
1230 .store = i7core_inject_type_store,
1231 }, {
1232 .attr = {
1233 .name = "inject_eccmask",
1234 .mode = (S_IRUGO | S_IWUSR)
1235 },
1236 .show = i7core_inject_eccmask_show,
1237 .store = i7core_inject_eccmask_store,
1238 }, {
1239 .grp = &i7core_inject_addrmatch,
1240 }, {
1241 .attr = {
1242 .name = "inject_enable",
1243 .mode = (S_IRUGO | S_IWUSR)
1244 },
1245 .show = i7core_inject_enable_show,
1246 .store = i7core_inject_enable_store,
1247 }, {
1248 .grp = &i7core_udimm_counters,
1249 },
1250 { } /* End of list */
Mauro Carvalho Chehab194a40f2009-06-22 22:48:28 -03001251};
1252
1253/****************************************************************************
Mauro Carvalho Chehaba0c36a12009-06-22 22:41:15 -03001254 Device initialization routines: put/get, init/exit
1255 ****************************************************************************/
1256
1257/*
Hidetoshi Seto64c10f62010-08-20 04:28:14 -03001258 * i7core_put_all_devices 'put' all the devices that we have
Mauro Carvalho Chehaba0c36a12009-06-22 22:41:15 -03001259 * reserved via 'get'
1260 */
Mauro Carvalho Chehab13d6e9b2009-09-05 12:15:20 -03001261static void i7core_put_devices(struct i7core_dev *i7core_dev)
Mauro Carvalho Chehaba0c36a12009-06-22 22:41:15 -03001262{
Mauro Carvalho Chehab13d6e9b2009-09-05 12:15:20 -03001263 int i;
Mauro Carvalho Chehaba0c36a12009-06-22 22:41:15 -03001264
Mauro Carvalho Chehab22e6bcb2009-09-05 23:06:50 -03001265 debugf0(__FILE__ ": %s()\n", __func__);
Mauro Carvalho Chehabde06eee2009-10-14 08:02:40 -03001266 for (i = 0; i < i7core_dev->n_devs; i++) {
Mauro Carvalho Chehab22e6bcb2009-09-05 23:06:50 -03001267 struct pci_dev *pdev = i7core_dev->pdev[i];
1268 if (!pdev)
1269 continue;
1270 debugf0("Removing dev %02x:%02x.%d\n",
1271 pdev->bus->number,
1272 PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn));
1273 pci_dev_put(pdev);
1274 }
Mauro Carvalho Chehab13d6e9b2009-09-05 12:15:20 -03001275}
Mauro Carvalho Chehab66607702009-09-05 00:52:11 -03001276
Mauro Carvalho Chehab13d6e9b2009-09-05 12:15:20 -03001277static void i7core_put_all_devices(void)
1278{
Mauro Carvalho Chehab42538682009-09-24 09:59:13 -03001279 struct i7core_dev *i7core_dev, *tmp;
Mauro Carvalho Chehab13d6e9b2009-09-05 12:15:20 -03001280
Mauro Carvalho Chehab39300e72010-08-11 23:40:15 -03001281 list_for_each_entry_safe(i7core_dev, tmp, &i7core_edac_list, list) {
Mauro Carvalho Chehab13d6e9b2009-09-05 12:15:20 -03001282 i7core_put_devices(i7core_dev);
Hidetoshi Seto2aa9be42010-08-20 04:25:00 -03001283 free_i7core_dev(i7core_dev);
Mauro Carvalho Chehab39300e72010-08-11 23:40:15 -03001284 }
Mauro Carvalho Chehaba0c36a12009-06-22 22:41:15 -03001285}
1286
Mauro Carvalho Chehab1288c182010-08-10 18:57:01 -03001287static void __init i7core_xeon_pci_fixup(const struct pci_id_table *table)
Keith Manntheybc2d7242009-09-03 00:05:05 -03001288{
1289 struct pci_dev *pdev = NULL;
1290 int i;
Mauro Carvalho Chehab54a08ab2010-08-19 15:51:00 -03001291
Keith Manntheybc2d7242009-09-03 00:05:05 -03001292 /*
1293 * On Xeon 55xx, the Intel Quckpath Arch Generic Non-core pci buses
1294 * aren't announced by acpi. So, we need to use a legacy scan probing
1295 * to detect them
1296 */
Vernon Mauerybd9e19c2010-05-18 19:02:50 -03001297 while (table && table->descr) {
1298 pdev = pci_get_device(PCI_VENDOR_ID_INTEL, table->descr[0].dev_id, NULL);
1299 if (unlikely(!pdev)) {
1300 for (i = 0; i < MAX_SOCKET_BUSES; i++)
1301 pcibios_scan_specific_bus(255-i);
1302 }
Mauro Carvalho Chehabbda14282010-06-30 01:41:35 -03001303 pci_dev_put(pdev);
Vernon Mauerybd9e19c2010-05-18 19:02:50 -03001304 table++;
Keith Manntheybc2d7242009-09-03 00:05:05 -03001305 }
1306}
1307
Mauro Carvalho Chehabbda14282010-06-30 01:41:35 -03001308static unsigned i7core_pci_lastbus(void)
1309{
1310 int last_bus = 0, bus;
1311 struct pci_bus *b = NULL;
1312
1313 while ((b = pci_find_next_bus(b)) != NULL) {
1314 bus = b->number;
1315 debugf0("Found bus %d\n", bus);
1316 if (bus > last_bus)
1317 last_bus = bus;
1318 }
1319
1320 debugf0("Last bus %d\n", last_bus);
1321
1322 return last_bus;
1323}
1324
Mauro Carvalho Chehaba0c36a12009-06-22 22:41:15 -03001325/*
Hidetoshi Seto64c10f62010-08-20 04:28:14 -03001326 * i7core_get_all_devices Find and perform 'get' operation on the MCH's
Mauro Carvalho Chehaba0c36a12009-06-22 22:41:15 -03001327 * device/functions we want to reference for this driver
1328 *
1329 * Need to 'get' device 16 func 1 and func 2
1330 */
Hidetoshi Setob197cba2010-08-20 04:24:31 -03001331static int i7core_get_onedevice(struct pci_dev **prev,
1332 const struct pci_id_table *table,
1333 const unsigned devno,
1334 const unsigned last_bus)
Mauro Carvalho Chehaba0c36a12009-06-22 22:41:15 -03001335{
Mauro Carvalho Chehab66607702009-09-05 00:52:11 -03001336 struct i7core_dev *i7core_dev;
Hidetoshi Setob197cba2010-08-20 04:24:31 -03001337 const struct pci_id_descr *dev_descr = &table->descr[devno];
Mauro Carvalho Chehab66607702009-09-05 00:52:11 -03001338
Mauro Carvalho Chehab8f331902009-06-22 22:48:29 -03001339 struct pci_dev *pdev = NULL;
Mauro Carvalho Chehab67166af2009-07-15 06:56:23 -03001340 u8 bus = 0;
1341 u8 socket = 0;
Mauro Carvalho Chehaba0c36a12009-06-22 22:41:15 -03001342
Mauro Carvalho Chehabc77720b2009-07-18 10:43:08 -03001343 pdev = pci_get_device(PCI_VENDOR_ID_INTEL,
Mauro Carvalho Chehabde06eee2009-10-14 08:02:40 -03001344 dev_descr->dev_id, *prev);
Mauro Carvalho Chehabc77720b2009-07-18 10:43:08 -03001345
1346 /*
Mauro Carvalho Chehabc77720b2009-07-18 10:43:08 -03001347 * On Xeon 55xx, the Intel Quckpath Arch Generic Non-core regs
1348 * is at addr 8086:2c40, instead of 8086:2c41. So, we need
1349 * to probe for the alternate address in case of failure
1350 */
Mauro Carvalho Chehabde06eee2009-10-14 08:02:40 -03001351 if (dev_descr->dev_id == PCI_DEVICE_ID_INTEL_I7_NONCORE && !pdev)
Mauro Carvalho Chehabc77720b2009-07-18 10:43:08 -03001352 pdev = pci_get_device(PCI_VENDOR_ID_INTEL,
Mauro Carvalho Chehabfd382652009-10-14 06:07:07 -03001353 PCI_DEVICE_ID_INTEL_I7_NONCORE_ALT, *prev);
Mauro Carvalho Chehabd1fd4fb2009-07-10 18:39:53 -03001354
Vernon Mauerybd9e19c2010-05-18 19:02:50 -03001355 if (dev_descr->dev_id == PCI_DEVICE_ID_INTEL_LYNNFIELD_NONCORE && !pdev)
Mauro Carvalho Chehabf05da2f2009-10-14 13:31:06 -03001356 pdev = pci_get_device(PCI_VENDOR_ID_INTEL,
1357 PCI_DEVICE_ID_INTEL_LYNNFIELD_NONCORE_ALT,
1358 *prev);
1359
Mauro Carvalho Chehabc77720b2009-07-18 10:43:08 -03001360 if (!pdev) {
1361 if (*prev) {
1362 *prev = pdev;
1363 return 0;
Mauro Carvalho Chehabd1fd4fb2009-07-10 18:39:53 -03001364 }
1365
Mauro Carvalho Chehabde06eee2009-10-14 08:02:40 -03001366 if (dev_descr->optional)
Mauro Carvalho Chehabc77720b2009-07-18 10:43:08 -03001367 return 0;
Mauro Carvalho Chehabef708b52009-06-22 22:48:30 -03001368
Vernon Mauerybd9e19c2010-05-18 19:02:50 -03001369 if (devno == 0)
1370 return -ENODEV;
1371
Daniel J Bluemanab089372010-07-23 23:16:52 +01001372 i7core_printk(KERN_INFO,
Mauro Carvalho Chehabc77720b2009-07-18 10:43:08 -03001373 "Device not found: dev %02x.%d PCI ID %04x:%04x\n",
Mauro Carvalho Chehabde06eee2009-10-14 08:02:40 -03001374 dev_descr->dev, dev_descr->func,
1375 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
Mauro Carvalho Chehabef708b52009-06-22 22:48:30 -03001376
Mauro Carvalho Chehabc77720b2009-07-18 10:43:08 -03001377 /* End of list, leave */
1378 return -ENODEV;
1379 }
1380 bus = pdev->bus->number;
Mauro Carvalho Chehab8f331902009-06-22 22:48:29 -03001381
Mauro Carvalho Chehabbda14282010-06-30 01:41:35 -03001382 socket = last_bus - bus;
Mauro Carvalho Chehabc77720b2009-07-18 10:43:08 -03001383
Mauro Carvalho Chehab66607702009-09-05 00:52:11 -03001384 i7core_dev = get_i7core_dev(socket);
1385 if (!i7core_dev) {
Hidetoshi Seto848b2f72010-08-20 04:24:44 -03001386 i7core_dev = alloc_i7core_dev(socket, table);
Mauro Carvalho Chehab66607702009-09-05 00:52:11 -03001387 if (!i7core_dev)
1388 return -ENOMEM;
Mauro Carvalho Chehaba0c36a12009-06-22 22:41:15 -03001389 }
Mauro Carvalho Chehaba0c36a12009-06-22 22:41:15 -03001390
Mauro Carvalho Chehab66607702009-09-05 00:52:11 -03001391 if (i7core_dev->pdev[devno]) {
Mauro Carvalho Chehabc77720b2009-07-18 10:43:08 -03001392 i7core_printk(KERN_ERR,
1393 "Duplicated device for "
1394 "dev %02x:%02x.%d PCI ID %04x:%04x\n",
Mauro Carvalho Chehabde06eee2009-10-14 08:02:40 -03001395 bus, dev_descr->dev, dev_descr->func,
1396 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
Mauro Carvalho Chehabc77720b2009-07-18 10:43:08 -03001397 pci_dev_put(pdev);
1398 return -ENODEV;
1399 }
Mauro Carvalho Chehabef708b52009-06-22 22:48:30 -03001400
Mauro Carvalho Chehab66607702009-09-05 00:52:11 -03001401 i7core_dev->pdev[devno] = pdev;
Mauro Carvalho Chehabc77720b2009-07-18 10:43:08 -03001402
1403 /* Sanity check */
Mauro Carvalho Chehabde06eee2009-10-14 08:02:40 -03001404 if (unlikely(PCI_SLOT(pdev->devfn) != dev_descr->dev ||
1405 PCI_FUNC(pdev->devfn) != dev_descr->func)) {
Mauro Carvalho Chehabc77720b2009-07-18 10:43:08 -03001406 i7core_printk(KERN_ERR,
1407 "Device PCI ID %04x:%04x "
1408 "has dev %02x:%02x.%d instead of dev %02x:%02x.%d\n",
Mauro Carvalho Chehabde06eee2009-10-14 08:02:40 -03001409 PCI_VENDOR_ID_INTEL, dev_descr->dev_id,
Mauro Carvalho Chehabc77720b2009-07-18 10:43:08 -03001410 bus, PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn),
Mauro Carvalho Chehabde06eee2009-10-14 08:02:40 -03001411 bus, dev_descr->dev, dev_descr->func);
Mauro Carvalho Chehabc77720b2009-07-18 10:43:08 -03001412 return -ENODEV;
1413 }
1414
1415 /* Be sure that the device is enabled */
1416 if (unlikely(pci_enable_device(pdev) < 0)) {
1417 i7core_printk(KERN_ERR,
1418 "Couldn't enable "
1419 "dev %02x:%02x.%d PCI ID %04x:%04x\n",
Mauro Carvalho Chehabde06eee2009-10-14 08:02:40 -03001420 bus, dev_descr->dev, dev_descr->func,
1421 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
Mauro Carvalho Chehabc77720b2009-07-18 10:43:08 -03001422 return -ENODEV;
1423 }
1424
Mauro Carvalho Chehabd4c27792009-09-05 04:12:02 -03001425 debugf0("Detected socket %d dev %02x:%02x.%d PCI ID %04x:%04x\n",
Mauro Carvalho Chehabde06eee2009-10-14 08:02:40 -03001426 socket, bus, dev_descr->dev,
1427 dev_descr->func,
1428 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
Mauro Carvalho Chehabc77720b2009-07-18 10:43:08 -03001429
1430 *prev = pdev;
1431
1432 return 0;
1433}
1434
Hidetoshi Seto64c10f62010-08-20 04:28:14 -03001435static int i7core_get_all_devices(void)
Mauro Carvalho Chehabc77720b2009-07-18 10:43:08 -03001436{
Hidetoshi Seto64c10f62010-08-20 04:28:14 -03001437 int i, j, rc, last_bus;
Mauro Carvalho Chehabc77720b2009-07-18 10:43:08 -03001438 struct pci_dev *pdev = NULL;
Hidetoshi Seto64c10f62010-08-20 04:28:14 -03001439 const struct pci_id_table *table;
Mauro Carvalho Chehabc77720b2009-07-18 10:43:08 -03001440
Mauro Carvalho Chehabbda14282010-06-30 01:41:35 -03001441 last_bus = i7core_pci_lastbus();
1442
Hidetoshi Seto64c10f62010-08-20 04:28:14 -03001443 for (j = 0; j < ARRAY_SIZE(pci_dev_table); j++) {
1444 table = &pci_dev_table[j];
Vernon Mauerybd9e19c2010-05-18 19:02:50 -03001445 for (i = 0; i < table->n_devs; i++) {
1446 pdev = NULL;
1447 do {
Hidetoshi Setob197cba2010-08-20 04:24:31 -03001448 rc = i7core_get_onedevice(&pdev, table, i,
Mauro Carvalho Chehabbda14282010-06-30 01:41:35 -03001449 last_bus);
Vernon Mauerybd9e19c2010-05-18 19:02:50 -03001450 if (rc < 0) {
1451 if (i == 0) {
1452 i = table->n_devs;
1453 break;
1454 }
1455 i7core_put_all_devices();
1456 return -ENODEV;
1457 }
1458 } while (pdev);
1459 }
Mauro Carvalho Chehabc77720b2009-07-18 10:43:08 -03001460 }
Mauro Carvalho Chehab66607702009-09-05 00:52:11 -03001461
Mauro Carvalho Chehabc77720b2009-07-18 10:43:08 -03001462 return 0;
Mauro Carvalho Chehabef708b52009-06-22 22:48:30 -03001463}
1464
Mauro Carvalho Chehabf4742942009-09-05 02:35:08 -03001465static int mci_bind_devs(struct mem_ctl_info *mci,
1466 struct i7core_dev *i7core_dev)
Mauro Carvalho Chehabef708b52009-06-22 22:48:30 -03001467{
1468 struct i7core_pvt *pvt = mci->pvt_info;
1469 struct pci_dev *pdev;
Mauro Carvalho Chehabf4742942009-09-05 02:35:08 -03001470 int i, func, slot;
Mauro Carvalho Chehabef708b52009-06-22 22:48:30 -03001471
Mauro Carvalho Chehabf4742942009-09-05 02:35:08 -03001472 /* Associates i7core_dev and mci for future usage */
1473 pvt->i7core_dev = i7core_dev;
1474 i7core_dev->mci = mci;
Mauro Carvalho Chehab66607702009-09-05 00:52:11 -03001475
Mauro Carvalho Chehabf4742942009-09-05 02:35:08 -03001476 pvt->is_registered = 0;
Mauro Carvalho Chehabde06eee2009-10-14 08:02:40 -03001477 for (i = 0; i < i7core_dev->n_devs; i++) {
Mauro Carvalho Chehabf4742942009-09-05 02:35:08 -03001478 pdev = i7core_dev->pdev[i];
1479 if (!pdev)
Mauro Carvalho Chehab66607702009-09-05 00:52:11 -03001480 continue;
1481
Mauro Carvalho Chehabf4742942009-09-05 02:35:08 -03001482 func = PCI_FUNC(pdev->devfn);
1483 slot = PCI_SLOT(pdev->devfn);
1484 if (slot == 3) {
1485 if (unlikely(func > MAX_MCR_FUNC))
Mauro Carvalho Chehabef708b52009-06-22 22:48:30 -03001486 goto error;
Mauro Carvalho Chehabf4742942009-09-05 02:35:08 -03001487 pvt->pci_mcr[func] = pdev;
1488 } else if (likely(slot >= 4 && slot < 4 + NUM_CHANS)) {
1489 if (unlikely(func > MAX_CHAN_FUNC))
1490 goto error;
1491 pvt->pci_ch[slot - 4][func] = pdev;
1492 } else if (!slot && !func)
1493 pvt->pci_noncore = pdev;
1494 else
1495 goto error;
Mauro Carvalho Chehabef708b52009-06-22 22:48:30 -03001496
Mauro Carvalho Chehabf4742942009-09-05 02:35:08 -03001497 debugf0("Associated fn %d.%d, dev = %p, socket %d\n",
1498 PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn),
1499 pdev, i7core_dev->socket);
Mauro Carvalho Chehab14d2c082009-09-02 23:52:36 -03001500
Mauro Carvalho Chehabf4742942009-09-05 02:35:08 -03001501 if (PCI_SLOT(pdev->devfn) == 3 &&
1502 PCI_FUNC(pdev->devfn) == 2)
1503 pvt->is_registered = 1;
Mauro Carvalho Chehabef708b52009-06-22 22:48:30 -03001504 }
Mauro Carvalho Chehabe9bd2e72009-07-09 22:14:35 -03001505
Mauro Carvalho Chehabef708b52009-06-22 22:48:30 -03001506 return 0;
1507
1508error:
1509 i7core_printk(KERN_ERR, "Device %d, function %d "
1510 "is out of the expected range\n",
1511 slot, func);
1512 return -EINVAL;
Mauro Carvalho Chehaba0c36a12009-06-22 22:41:15 -03001513}
1514
Mauro Carvalho Chehab442305b2009-06-22 22:48:29 -03001515/****************************************************************************
1516 Error check routines
1517 ****************************************************************************/
Mauro Carvalho Chehabf4742942009-09-05 02:35:08 -03001518static void i7core_rdimm_update_csrow(struct mem_ctl_info *mci,
Mauro Carvalho Chehab1288c182010-08-10 18:57:01 -03001519 const int chan,
1520 const int dimm,
1521 const int add)
Mauro Carvalho Chehabb4e8f0b2009-09-02 23:49:59 -03001522{
1523 char *msg;
1524 struct i7core_pvt *pvt = mci->pvt_info;
Mauro Carvalho Chehabf4742942009-09-05 02:35:08 -03001525 int row = pvt->csrow_map[chan][dimm], i;
Mauro Carvalho Chehabb4e8f0b2009-09-02 23:49:59 -03001526
1527 for (i = 0; i < add; i++) {
1528 msg = kasprintf(GFP_KERNEL, "Corrected error "
Mauro Carvalho Chehabf4742942009-09-05 02:35:08 -03001529 "(Socket=%d channel=%d dimm=%d)",
1530 pvt->i7core_dev->socket, chan, dimm);
Mauro Carvalho Chehabb4e8f0b2009-09-02 23:49:59 -03001531
1532 edac_mc_handle_fbd_ce(mci, row, 0, msg);
1533 kfree (msg);
1534 }
1535}
1536
1537static void i7core_rdimm_update_ce_count(struct mem_ctl_info *mci,
Mauro Carvalho Chehab1288c182010-08-10 18:57:01 -03001538 const int chan,
1539 const int new0,
1540 const int new1,
1541 const int new2)
Mauro Carvalho Chehabb4e8f0b2009-09-02 23:49:59 -03001542{
1543 struct i7core_pvt *pvt = mci->pvt_info;
1544 int add0 = 0, add1 = 0, add2 = 0;
1545 /* Updates CE counters if it is not the first time here */
Mauro Carvalho Chehabf4742942009-09-05 02:35:08 -03001546 if (pvt->ce_count_available) {
Mauro Carvalho Chehabb4e8f0b2009-09-02 23:49:59 -03001547 /* Updates CE counters */
1548
Mauro Carvalho Chehabf4742942009-09-05 02:35:08 -03001549 add2 = new2 - pvt->rdimm_last_ce_count[chan][2];
1550 add1 = new1 - pvt->rdimm_last_ce_count[chan][1];
1551 add0 = new0 - pvt->rdimm_last_ce_count[chan][0];
Mauro Carvalho Chehabb4e8f0b2009-09-02 23:49:59 -03001552
1553 if (add2 < 0)
1554 add2 += 0x7fff;
Mauro Carvalho Chehabf4742942009-09-05 02:35:08 -03001555 pvt->rdimm_ce_count[chan][2] += add2;
Mauro Carvalho Chehabb4e8f0b2009-09-02 23:49:59 -03001556
1557 if (add1 < 0)
1558 add1 += 0x7fff;
Mauro Carvalho Chehabf4742942009-09-05 02:35:08 -03001559 pvt->rdimm_ce_count[chan][1] += add1;
Mauro Carvalho Chehabb4e8f0b2009-09-02 23:49:59 -03001560
1561 if (add0 < 0)
1562 add0 += 0x7fff;
Mauro Carvalho Chehabf4742942009-09-05 02:35:08 -03001563 pvt->rdimm_ce_count[chan][0] += add0;
Mauro Carvalho Chehabb4e8f0b2009-09-02 23:49:59 -03001564 } else
Mauro Carvalho Chehabf4742942009-09-05 02:35:08 -03001565 pvt->ce_count_available = 1;
Mauro Carvalho Chehabb4e8f0b2009-09-02 23:49:59 -03001566
1567 /* Store the new values */
Mauro Carvalho Chehabf4742942009-09-05 02:35:08 -03001568 pvt->rdimm_last_ce_count[chan][2] = new2;
1569 pvt->rdimm_last_ce_count[chan][1] = new1;
1570 pvt->rdimm_last_ce_count[chan][0] = new0;
Mauro Carvalho Chehabb4e8f0b2009-09-02 23:49:59 -03001571
1572 /*updated the edac core */
1573 if (add0 != 0)
Mauro Carvalho Chehabf4742942009-09-05 02:35:08 -03001574 i7core_rdimm_update_csrow(mci, chan, 0, add0);
Mauro Carvalho Chehabb4e8f0b2009-09-02 23:49:59 -03001575 if (add1 != 0)
Mauro Carvalho Chehabf4742942009-09-05 02:35:08 -03001576 i7core_rdimm_update_csrow(mci, chan, 1, add1);
Mauro Carvalho Chehabb4e8f0b2009-09-02 23:49:59 -03001577 if (add2 != 0)
Mauro Carvalho Chehabf4742942009-09-05 02:35:08 -03001578 i7core_rdimm_update_csrow(mci, chan, 2, add2);
Mauro Carvalho Chehabb4e8f0b2009-09-02 23:49:59 -03001579
1580}
1581
Mauro Carvalho Chehabf4742942009-09-05 02:35:08 -03001582static void i7core_rdimm_check_mc_ecc_err(struct mem_ctl_info *mci)
Mauro Carvalho Chehabb4e8f0b2009-09-02 23:49:59 -03001583{
1584 struct i7core_pvt *pvt = mci->pvt_info;
1585 u32 rcv[3][2];
1586 int i, new0, new1, new2;
1587
1588 /*Read DEV 3: FUN 2: MC_COR_ECC_CNT regs directly*/
Mauro Carvalho Chehabf4742942009-09-05 02:35:08 -03001589 pci_read_config_dword(pvt->pci_mcr[2], MC_COR_ECC_CNT_0,
Mauro Carvalho Chehabb4e8f0b2009-09-02 23:49:59 -03001590 &rcv[0][0]);
Mauro Carvalho Chehabf4742942009-09-05 02:35:08 -03001591 pci_read_config_dword(pvt->pci_mcr[2], MC_COR_ECC_CNT_1,
Mauro Carvalho Chehabb4e8f0b2009-09-02 23:49:59 -03001592 &rcv[0][1]);
Mauro Carvalho Chehabf4742942009-09-05 02:35:08 -03001593 pci_read_config_dword(pvt->pci_mcr[2], MC_COR_ECC_CNT_2,
Mauro Carvalho Chehabb4e8f0b2009-09-02 23:49:59 -03001594 &rcv[1][0]);
Mauro Carvalho Chehabf4742942009-09-05 02:35:08 -03001595 pci_read_config_dword(pvt->pci_mcr[2], MC_COR_ECC_CNT_3,
Mauro Carvalho Chehabb4e8f0b2009-09-02 23:49:59 -03001596 &rcv[1][1]);
Mauro Carvalho Chehabf4742942009-09-05 02:35:08 -03001597 pci_read_config_dword(pvt->pci_mcr[2], MC_COR_ECC_CNT_4,
Mauro Carvalho Chehabb4e8f0b2009-09-02 23:49:59 -03001598 &rcv[2][0]);
Mauro Carvalho Chehabf4742942009-09-05 02:35:08 -03001599 pci_read_config_dword(pvt->pci_mcr[2], MC_COR_ECC_CNT_5,
Mauro Carvalho Chehabb4e8f0b2009-09-02 23:49:59 -03001600 &rcv[2][1]);
1601 for (i = 0 ; i < 3; i++) {
1602 debugf3("MC_COR_ECC_CNT%d = 0x%x; MC_COR_ECC_CNT%d = 0x%x\n",
1603 (i * 2), rcv[i][0], (i * 2) + 1, rcv[i][1]);
1604 /*if the channel has 3 dimms*/
Mauro Carvalho Chehabf4742942009-09-05 02:35:08 -03001605 if (pvt->channel[i].dimms > 2) {
Mauro Carvalho Chehabb4e8f0b2009-09-02 23:49:59 -03001606 new0 = DIMM_BOT_COR_ERR(rcv[i][0]);
1607 new1 = DIMM_TOP_COR_ERR(rcv[i][0]);
1608 new2 = DIMM_BOT_COR_ERR(rcv[i][1]);
1609 } else {
1610 new0 = DIMM_TOP_COR_ERR(rcv[i][0]) +
1611 DIMM_BOT_COR_ERR(rcv[i][0]);
1612 new1 = DIMM_TOP_COR_ERR(rcv[i][1]) +
1613 DIMM_BOT_COR_ERR(rcv[i][1]);
1614 new2 = 0;
1615 }
1616
Mauro Carvalho Chehabf4742942009-09-05 02:35:08 -03001617 i7core_rdimm_update_ce_count(mci, i, new0, new1, new2);
Mauro Carvalho Chehabb4e8f0b2009-09-02 23:49:59 -03001618 }
1619}
Mauro Carvalho Chehab442305b2009-06-22 22:48:29 -03001620
1621/* This function is based on the device 3 function 4 registers as described on:
1622 * Intel Xeon Processor 5500 Series Datasheet Volume 2
1623 * http://www.intel.com/Assets/PDF/datasheet/321322.pdf
1624 * also available at:
1625 * http://www.arrownac.com/manufacturers/intel/s/nehalem/5500-datasheet-v2.pdf
1626 */
Mauro Carvalho Chehabf4742942009-09-05 02:35:08 -03001627static void i7core_udimm_check_mc_ecc_err(struct mem_ctl_info *mci)
Mauro Carvalho Chehab442305b2009-06-22 22:48:29 -03001628{
1629 struct i7core_pvt *pvt = mci->pvt_info;
1630 u32 rcv1, rcv0;
1631 int new0, new1, new2;
1632
Mauro Carvalho Chehabf4742942009-09-05 02:35:08 -03001633 if (!pvt->pci_mcr[4]) {
Mauro Carvalho Chehabb9905382009-08-05 21:36:35 -03001634 debugf0("%s MCR registers not found\n", __func__);
Mauro Carvalho Chehab442305b2009-06-22 22:48:29 -03001635 return;
1636 }
1637
Mauro Carvalho Chehabb4e8f0b2009-09-02 23:49:59 -03001638 /* Corrected test errors */
Mauro Carvalho Chehabf4742942009-09-05 02:35:08 -03001639 pci_read_config_dword(pvt->pci_mcr[4], MC_TEST_ERR_RCV1, &rcv1);
1640 pci_read_config_dword(pvt->pci_mcr[4], MC_TEST_ERR_RCV0, &rcv0);
Mauro Carvalho Chehab442305b2009-06-22 22:48:29 -03001641
1642 /* Store the new values */
1643 new2 = DIMM2_COR_ERR(rcv1);
1644 new1 = DIMM1_COR_ERR(rcv0);
1645 new0 = DIMM0_COR_ERR(rcv0);
1646
Mauro Carvalho Chehab442305b2009-06-22 22:48:29 -03001647 /* Updates CE counters if it is not the first time here */
Mauro Carvalho Chehabf4742942009-09-05 02:35:08 -03001648 if (pvt->ce_count_available) {
Mauro Carvalho Chehab442305b2009-06-22 22:48:29 -03001649 /* Updates CE counters */
1650 int add0, add1, add2;
1651
Mauro Carvalho Chehabf4742942009-09-05 02:35:08 -03001652 add2 = new2 - pvt->udimm_last_ce_count[2];
1653 add1 = new1 - pvt->udimm_last_ce_count[1];
1654 add0 = new0 - pvt->udimm_last_ce_count[0];
Mauro Carvalho Chehab442305b2009-06-22 22:48:29 -03001655
1656 if (add2 < 0)
1657 add2 += 0x7fff;
Mauro Carvalho Chehabf4742942009-09-05 02:35:08 -03001658 pvt->udimm_ce_count[2] += add2;
Mauro Carvalho Chehab442305b2009-06-22 22:48:29 -03001659
1660 if (add1 < 0)
1661 add1 += 0x7fff;
Mauro Carvalho Chehabf4742942009-09-05 02:35:08 -03001662 pvt->udimm_ce_count[1] += add1;
Mauro Carvalho Chehab442305b2009-06-22 22:48:29 -03001663
1664 if (add0 < 0)
1665 add0 += 0x7fff;
Mauro Carvalho Chehabf4742942009-09-05 02:35:08 -03001666 pvt->udimm_ce_count[0] += add0;
Mauro Carvalho Chehabb4e8f0b2009-09-02 23:49:59 -03001667
1668 if (add0 | add1 | add2)
1669 i7core_printk(KERN_ERR, "New Corrected error(s): "
1670 "dimm0: +%d, dimm1: +%d, dimm2 +%d\n",
1671 add0, add1, add2);
Mauro Carvalho Chehab442305b2009-06-22 22:48:29 -03001672 } else
Mauro Carvalho Chehabf4742942009-09-05 02:35:08 -03001673 pvt->ce_count_available = 1;
Mauro Carvalho Chehab442305b2009-06-22 22:48:29 -03001674
1675 /* Store the new values */
Mauro Carvalho Chehabf4742942009-09-05 02:35:08 -03001676 pvt->udimm_last_ce_count[2] = new2;
1677 pvt->udimm_last_ce_count[1] = new1;
1678 pvt->udimm_last_ce_count[0] = new0;
Mauro Carvalho Chehab442305b2009-06-22 22:48:29 -03001679}
1680
Mauro Carvalho Chehab8a2f1182009-07-15 19:01:08 -03001681/*
1682 * According with tables E-11 and E-12 of chapter E.3.3 of Intel 64 and IA-32
1683 * Architectures Software Developer’s Manual Volume 3B.
Mauro Carvalho Chehabf237fcf2009-07-15 19:53:24 -03001684 * Nehalem are defined as family 0x06, model 0x1a
1685 *
1686 * The MCA registers used here are the following ones:
Mauro Carvalho Chehab8a2f1182009-07-15 19:01:08 -03001687 * struct mce field MCA Register
Mauro Carvalho Chehabf237fcf2009-07-15 19:53:24 -03001688 * m->status MSR_IA32_MC8_STATUS
1689 * m->addr MSR_IA32_MC8_ADDR
1690 * m->misc MSR_IA32_MC8_MISC
Mauro Carvalho Chehab8a2f1182009-07-15 19:01:08 -03001691 * In the case of Nehalem, the error information is masked at .status and .misc
1692 * fields
1693 */
Mauro Carvalho Chehabd5381642009-07-09 22:06:41 -03001694static void i7core_mce_output_error(struct mem_ctl_info *mci,
Mauro Carvalho Chehab1288c182010-08-10 18:57:01 -03001695 const struct mce *m)
Mauro Carvalho Chehabd5381642009-07-09 22:06:41 -03001696{
Mauro Carvalho Chehabb4e8f0b2009-09-02 23:49:59 -03001697 struct i7core_pvt *pvt = mci->pvt_info;
Mauro Carvalho Chehaba6395392009-07-17 10:54:23 -03001698 char *type, *optype, *err, *msg;
Mauro Carvalho Chehab8a2f1182009-07-15 19:01:08 -03001699 unsigned long error = m->status & 0x1ff0000l;
Mauro Carvalho Chehaba6395392009-07-17 10:54:23 -03001700 u32 optypenum = (m->status >> 4) & 0x07;
Mauro Carvalho Chehab8a2f1182009-07-15 19:01:08 -03001701 u32 core_err_cnt = (m->status >> 38) && 0x7fff;
1702 u32 dimm = (m->misc >> 16) & 0x3;
1703 u32 channel = (m->misc >> 18) & 0x3;
1704 u32 syndrome = m->misc >> 32;
1705 u32 errnum = find_first_bit(&error, 32);
Mauro Carvalho Chehabb4e8f0b2009-09-02 23:49:59 -03001706 int csrow;
Mauro Carvalho Chehabd5381642009-07-09 22:06:41 -03001707
Mauro Carvalho Chehabc5d34522009-07-17 10:28:15 -03001708 if (m->mcgstatus & 1)
1709 type = "FATAL";
1710 else
1711 type = "NON_FATAL";
1712
Mauro Carvalho Chehaba6395392009-07-17 10:54:23 -03001713 switch (optypenum) {
Mauro Carvalho Chehabb9905382009-08-05 21:36:35 -03001714 case 0:
1715 optype = "generic undef request";
1716 break;
1717 case 1:
1718 optype = "read error";
1719 break;
1720 case 2:
1721 optype = "write error";
1722 break;
1723 case 3:
1724 optype = "addr/cmd error";
1725 break;
1726 case 4:
1727 optype = "scrubbing error";
1728 break;
1729 default:
1730 optype = "reserved";
1731 break;
Mauro Carvalho Chehaba6395392009-07-17 10:54:23 -03001732 }
1733
Mauro Carvalho Chehab8a2f1182009-07-15 19:01:08 -03001734 switch (errnum) {
1735 case 16:
1736 err = "read ECC error";
1737 break;
1738 case 17:
1739 err = "RAS ECC error";
1740 break;
1741 case 18:
1742 err = "write parity error";
1743 break;
1744 case 19:
1745 err = "redundacy loss";
1746 break;
1747 case 20:
1748 err = "reserved";
1749 break;
1750 case 21:
1751 err = "memory range error";
1752 break;
1753 case 22:
1754 err = "RTID out of range";
1755 break;
1756 case 23:
1757 err = "address parity error";
1758 break;
1759 case 24:
1760 err = "byte enable parity error";
1761 break;
1762 default:
1763 err = "unknown";
1764 }
1765
Mauro Carvalho Chehabf237fcf2009-07-15 19:53:24 -03001766 /* FIXME: should convert addr into bank and rank information */
Mauro Carvalho Chehab8a2f1182009-07-15 19:01:08 -03001767 msg = kasprintf(GFP_ATOMIC,
Mauro Carvalho Chehabf4742942009-09-05 02:35:08 -03001768 "%s (addr = 0x%08llx, cpu=%d, Dimm=%d, Channel=%d, "
Mauro Carvalho Chehaba6395392009-07-17 10:54:23 -03001769 "syndrome=0x%08x, count=%d, Err=%08llx:%08llx (%s: %s))\n",
Mauro Carvalho Chehabf4742942009-09-05 02:35:08 -03001770 type, (long long) m->addr, m->cpu, dimm, channel,
Mauro Carvalho Chehaba6395392009-07-17 10:54:23 -03001771 syndrome, core_err_cnt, (long long)m->status,
1772 (long long)m->misc, optype, err);
Mauro Carvalho Chehab8a2f1182009-07-15 19:01:08 -03001773
1774 debugf0("%s", msg);
Mauro Carvalho Chehabd5381642009-07-09 22:06:41 -03001775
Mauro Carvalho Chehabf4742942009-09-05 02:35:08 -03001776 csrow = pvt->csrow_map[channel][dimm];
Mauro Carvalho Chehabb4e8f0b2009-09-02 23:49:59 -03001777
Mauro Carvalho Chehabd5381642009-07-09 22:06:41 -03001778 /* Call the helper to output message */
Mauro Carvalho Chehabb4e8f0b2009-09-02 23:49:59 -03001779 if (m->mcgstatus & 1)
1780 edac_mc_handle_fbd_ue(mci, csrow, 0,
1781 0 /* FIXME: should be channel here */, msg);
Mauro Carvalho Chehabf4742942009-09-05 02:35:08 -03001782 else if (!pvt->is_registered)
Mauro Carvalho Chehabb4e8f0b2009-09-02 23:49:59 -03001783 edac_mc_handle_fbd_ce(mci, csrow,
1784 0 /* FIXME: should be channel here */, msg);
Mauro Carvalho Chehab8a2f1182009-07-15 19:01:08 -03001785
1786 kfree(msg);
Mauro Carvalho Chehabd5381642009-07-09 22:06:41 -03001787}
1788
Mauro Carvalho Chehaba0c36a12009-06-22 22:41:15 -03001789/*
Mauro Carvalho Chehab87d1d272009-06-22 22:48:29 -03001790 * i7core_check_error Retrieve and process errors reported by the
1791 * hardware. Called by the Core module.
1792 */
1793static void i7core_check_error(struct mem_ctl_info *mci)
1794{
Mauro Carvalho Chehabd5381642009-07-09 22:06:41 -03001795 struct i7core_pvt *pvt = mci->pvt_info;
1796 int i;
1797 unsigned count = 0;
Mauro Carvalho Chehabca9c90b2009-10-04 10:15:40 -03001798 struct mce *m;
Mauro Carvalho Chehabd5381642009-07-09 22:06:41 -03001799
Mauro Carvalho Chehabca9c90b2009-10-04 10:15:40 -03001800 /*
1801 * MCE first step: Copy all mce errors into a temporary buffer
1802 * We use a double buffering here, to reduce the risk of
1803 * loosing an error.
1804 */
1805 smp_rmb();
Mauro Carvalho Chehab321ece42009-10-08 13:11:08 -03001806 count = (pvt->mce_out + MCE_LOG_LEN - pvt->mce_in)
1807 % MCE_LOG_LEN;
Mauro Carvalho Chehabca9c90b2009-10-04 10:15:40 -03001808 if (!count)
Vernon Mauery8a311e12010-04-16 19:40:19 -03001809 goto check_ce_error;
Mauro Carvalho Chehabf4742942009-09-05 02:35:08 -03001810
Mauro Carvalho Chehabca9c90b2009-10-04 10:15:40 -03001811 m = pvt->mce_outentry;
Mauro Carvalho Chehab321ece42009-10-08 13:11:08 -03001812 if (pvt->mce_in + count > MCE_LOG_LEN) {
1813 unsigned l = MCE_LOG_LEN - pvt->mce_in;
Mauro Carvalho Chehabca9c90b2009-10-04 10:15:40 -03001814
1815 memcpy(m, &pvt->mce_entry[pvt->mce_in], sizeof(*m) * l);
1816 smp_wmb();
1817 pvt->mce_in = 0;
1818 count -= l;
1819 m += l;
1820 }
1821 memcpy(m, &pvt->mce_entry[pvt->mce_in], sizeof(*m) * count);
1822 smp_wmb();
1823 pvt->mce_in += count;
1824
1825 smp_rmb();
1826 if (pvt->mce_overrun) {
1827 i7core_printk(KERN_ERR, "Lost %d memory errors\n",
1828 pvt->mce_overrun);
1829 smp_wmb();
1830 pvt->mce_overrun = 0;
Mauro Carvalho Chehabd5381642009-07-09 22:06:41 -03001831 }
Mauro Carvalho Chehabf4742942009-09-05 02:35:08 -03001832
Mauro Carvalho Chehabca9c90b2009-10-04 10:15:40 -03001833 /*
1834 * MCE second step: parse errors and display
1835 */
Mauro Carvalho Chehabd5381642009-07-09 22:06:41 -03001836 for (i = 0; i < count; i++)
Mauro Carvalho Chehabca9c90b2009-10-04 10:15:40 -03001837 i7core_mce_output_error(mci, &pvt->mce_outentry[i]);
Mauro Carvalho Chehabd5381642009-07-09 22:06:41 -03001838
Mauro Carvalho Chehabca9c90b2009-10-04 10:15:40 -03001839 /*
1840 * Now, let's increment CE error counts
1841 */
Vernon Mauery8a311e12010-04-16 19:40:19 -03001842check_ce_error:
Mauro Carvalho Chehabf4742942009-09-05 02:35:08 -03001843 if (!pvt->is_registered)
1844 i7core_udimm_check_mc_ecc_err(mci);
1845 else
1846 i7core_rdimm_check_mc_ecc_err(mci);
Mauro Carvalho Chehab87d1d272009-06-22 22:48:29 -03001847}
1848
1849/*
Mauro Carvalho Chehabd5381642009-07-09 22:06:41 -03001850 * i7core_mce_check_error Replicates mcelog routine to get errors
1851 * This routine simply queues mcelog errors, and
1852 * return. The error itself should be handled later
1853 * by i7core_check_error.
Mauro Carvalho Chehab6e103be2009-10-05 09:40:09 -03001854 * WARNING: As this routine should be called at NMI time, extra care should
1855 * be taken to avoid deadlocks, and to be as fast as possible.
Mauro Carvalho Chehabd5381642009-07-09 22:06:41 -03001856 */
1857static int i7core_mce_check_error(void *priv, struct mce *mce)
1858{
Mauro Carvalho Chehabc5d34522009-07-17 10:28:15 -03001859 struct mem_ctl_info *mci = priv;
1860 struct i7core_pvt *pvt = mci->pvt_info;
Mauro Carvalho Chehabd5381642009-07-09 22:06:41 -03001861
Mauro Carvalho Chehab8a2f1182009-07-15 19:01:08 -03001862 /*
1863 * Just let mcelog handle it if the error is
1864 * outside the memory controller
1865 */
1866 if (((mce->status & 0xffff) >> 7) != 1)
1867 return 0;
1868
Mauro Carvalho Chehabf237fcf2009-07-15 19:53:24 -03001869 /* Bank 8 registers are the only ones that we know how to handle */
1870 if (mce->bank != 8)
1871 return 0;
1872
Randy Dunlap3b918c12009-11-08 01:36:40 -02001873#ifdef CONFIG_SMP
Mauro Carvalho Chehabf4742942009-09-05 02:35:08 -03001874 /* Only handle if it is the right mc controller */
Mauro Carvalho Chehab6e103be2009-10-05 09:40:09 -03001875 if (cpu_data(mce->cpu).phys_proc_id != pvt->i7core_dev->socket)
Mauro Carvalho Chehabf4742942009-09-05 02:35:08 -03001876 return 0;
Randy Dunlap3b918c12009-11-08 01:36:40 -02001877#endif
Mauro Carvalho Chehabf4742942009-09-05 02:35:08 -03001878
Mauro Carvalho Chehabca9c90b2009-10-04 10:15:40 -03001879 smp_rmb();
Mauro Carvalho Chehab321ece42009-10-08 13:11:08 -03001880 if ((pvt->mce_out + 1) % MCE_LOG_LEN == pvt->mce_in) {
Mauro Carvalho Chehabca9c90b2009-10-04 10:15:40 -03001881 smp_wmb();
1882 pvt->mce_overrun++;
1883 return 0;
Mauro Carvalho Chehabd5381642009-07-09 22:06:41 -03001884 }
Mauro Carvalho Chehab6e103be2009-10-05 09:40:09 -03001885
1886 /* Copy memory error at the ringbuffer */
1887 memcpy(&pvt->mce_entry[pvt->mce_out], mce, sizeof(*mce));
Mauro Carvalho Chehabca9c90b2009-10-04 10:15:40 -03001888 smp_wmb();
Mauro Carvalho Chehab321ece42009-10-08 13:11:08 -03001889 pvt->mce_out = (pvt->mce_out + 1) % MCE_LOG_LEN;
Mauro Carvalho Chehabd5381642009-07-09 22:06:41 -03001890
Mauro Carvalho Chehabc5d34522009-07-17 10:28:15 -03001891 /* Handle fatal errors immediately */
1892 if (mce->mcgstatus & 1)
1893 i7core_check_error(mci);
1894
Mauro Carvalho Chehabd5381642009-07-09 22:06:41 -03001895 /* Advice mcelog that the error were handled */
Mauro Carvalho Chehab8a2f1182009-07-15 19:01:08 -03001896 return 1;
Mauro Carvalho Chehabd5381642009-07-09 22:06:41 -03001897}
1898
Hidetoshi Setoa3aa0a42010-08-20 04:25:18 -03001899static void i7core_pci_ctl_create(struct i7core_pvt *pvt)
1900{
1901 pvt->i7core_pci = edac_pci_create_generic_ctl(
1902 &pvt->i7core_dev->pdev[0]->dev,
1903 EDAC_MOD_STR);
1904 if (unlikely(!pvt->i7core_pci))
1905 pr_warn("Unable to setup PCI error report via EDAC\n");
1906}
1907
1908static void i7core_pci_ctl_release(struct i7core_pvt *pvt)
1909{
1910 if (likely(pvt->i7core_pci))
1911 edac_pci_release_generic_ctl(pvt->i7core_pci);
1912 else
1913 i7core_printk(KERN_ERR,
1914 "Couldn't find mem_ctl_info for socket %d\n",
1915 pvt->i7core_dev->socket);
1916 pvt->i7core_pci = NULL;
1917}
1918
Mauro Carvalho Chehabf4742942009-09-05 02:35:08 -03001919static int i7core_register_mci(struct i7core_dev *i7core_dev,
Mauro Carvalho Chehab1288c182010-08-10 18:57:01 -03001920 const int num_channels, const int num_csrows)
Mauro Carvalho Chehaba0c36a12009-06-22 22:41:15 -03001921{
1922 struct mem_ctl_info *mci;
1923 struct i7core_pvt *pvt;
Mauro Carvalho Chehabba6c5c62009-07-15 09:02:32 -03001924 int csrow = 0;
Mauro Carvalho Chehabf4742942009-09-05 02:35:08 -03001925 int rc;
Mauro Carvalho Chehabef708b52009-06-22 22:48:30 -03001926
Mauro Carvalho Chehaba0c36a12009-06-22 22:41:15 -03001927 /* allocate a new MC control structure */
Mauro Carvalho Chehabd4c27792009-09-05 04:12:02 -03001928 mci = edac_mc_alloc(sizeof(*pvt), num_csrows, num_channels,
1929 i7core_dev->socket);
Mauro Carvalho Chehabf4742942009-09-05 02:35:08 -03001930 if (unlikely(!mci))
1931 return -ENOMEM;
Mauro Carvalho Chehaba0c36a12009-06-22 22:41:15 -03001932
Mauro Carvalho Chehab3cfd0142010-08-10 23:23:46 -03001933 debugf0("MC: " __FILE__ ": %s(): mci = %p, dev = %p\n",
1934 __func__, mci, &i7core_dev->pdev[0]->dev);
Mauro Carvalho Chehaba0c36a12009-06-22 22:41:15 -03001935
Mauro Carvalho Chehabf4742942009-09-05 02:35:08 -03001936 /* record ptr to the generic device */
1937 mci->dev = &i7core_dev->pdev[0]->dev;
1938
Mauro Carvalho Chehaba0c36a12009-06-22 22:41:15 -03001939 pvt = mci->pvt_info;
Mauro Carvalho Chehabef708b52009-06-22 22:48:30 -03001940 memset(pvt, 0, sizeof(*pvt));
Mauro Carvalho Chehab67166af2009-07-15 06:56:23 -03001941
Mauro Carvalho Chehab41fcb7f2009-06-22 22:48:31 -03001942 /*
1943 * FIXME: how to handle RDDR3 at MCI level? It is possible to have
1944 * Mixed RDDR3/UDDR3 with Nehalem, provided that they are on different
1945 * memory channels
1946 */
1947 mci->mtype_cap = MEM_FLAG_DDR3;
Mauro Carvalho Chehaba0c36a12009-06-22 22:41:15 -03001948 mci->edac_ctl_cap = EDAC_FLAG_NONE;
1949 mci->edac_cap = EDAC_FLAG_NONE;
1950 mci->mod_name = "i7core_edac.c";
1951 mci->mod_ver = I7CORE_REVISION;
Mauro Carvalho Chehabf4742942009-09-05 02:35:08 -03001952 mci->ctl_name = kasprintf(GFP_KERNEL, "i7 core #%d",
1953 i7core_dev->socket);
1954 mci->dev_name = pci_name(i7core_dev->pdev[0]);
Mauro Carvalho Chehaba0c36a12009-06-22 22:41:15 -03001955 mci->ctl_page_to_phys = NULL;
Mauro Carvalho Chehab1288c182010-08-10 18:57:01 -03001956
1957 if (pvt->is_registered)
1958 mci->mc_driver_sysfs_attributes = i7core_sysfs_rdimm_attrs;
1959 else
1960 mci->mc_driver_sysfs_attributes = i7core_sysfs_udimm_attrs;
1961
Mauro Carvalho Chehab87d1d272009-06-22 22:48:29 -03001962 /* Set the function pointer to an actual operation function */
1963 mci->edac_check = i7core_check_error;
Mauro Carvalho Chehab8f331902009-06-22 22:48:29 -03001964
Mauro Carvalho Chehabef708b52009-06-22 22:48:30 -03001965 /* Store pci devices at mci for faster access */
Mauro Carvalho Chehabf4742942009-09-05 02:35:08 -03001966 rc = mci_bind_devs(mci, i7core_dev);
Mauro Carvalho Chehab41fcb7f2009-06-22 22:48:31 -03001967 if (unlikely(rc < 0))
Mauro Carvalho Chehabf4742942009-09-05 02:35:08 -03001968 goto fail;
Mauro Carvalho Chehabef708b52009-06-22 22:48:30 -03001969
1970 /* Get dimm basic config */
Mauro Carvalho Chehabf4742942009-09-05 02:35:08 -03001971 get_dimm_config(mci, &csrow);
Mauro Carvalho Chehabef708b52009-06-22 22:48:30 -03001972
Mauro Carvalho Chehaba0c36a12009-06-22 22:41:15 -03001973 /* add this new MC control structure to EDAC's list of MCs */
Mauro Carvalho Chehabb7c76152009-06-22 22:48:30 -03001974 if (unlikely(edac_mc_add_mc(mci))) {
Mauro Carvalho Chehaba0c36a12009-06-22 22:41:15 -03001975 debugf0("MC: " __FILE__
1976 ": %s(): failed edac_mc_add_mc()\n", __func__);
1977 /* FIXME: perhaps some code should go here that disables error
1978 * reporting if we just enabled it
1979 */
Mauro Carvalho Chehabb7c76152009-06-22 22:48:30 -03001980
1981 rc = -EINVAL;
Mauro Carvalho Chehabf4742942009-09-05 02:35:08 -03001982 goto fail;
Mauro Carvalho Chehaba0c36a12009-06-22 22:41:15 -03001983 }
1984
Mauro Carvalho Chehab194a40f2009-06-22 22:48:28 -03001985 /* Default error mask is any memory */
Mauro Carvalho Chehabef708b52009-06-22 22:48:30 -03001986 pvt->inject.channel = 0;
Mauro Carvalho Chehab194a40f2009-06-22 22:48:28 -03001987 pvt->inject.dimm = -1;
1988 pvt->inject.rank = -1;
1989 pvt->inject.bank = -1;
1990 pvt->inject.page = -1;
1991 pvt->inject.col = -1;
1992
Hidetoshi Setoa3aa0a42010-08-20 04:25:18 -03001993 /* allocating generic PCI control info */
1994 i7core_pci_ctl_create(pvt);
1995
Mauro Carvalho Chehabd5381642009-07-09 22:06:41 -03001996 /* Registers on edac_mce in order to receive memory errors */
Mauro Carvalho Chehabc5d34522009-07-17 10:28:15 -03001997 pvt->edac_mce.priv = mci;
Mauro Carvalho Chehabd5381642009-07-09 22:06:41 -03001998 pvt->edac_mce.check_error = i7core_mce_check_error;
Mauro Carvalho Chehabd5381642009-07-09 22:06:41 -03001999 rc = edac_mce_register(&pvt->edac_mce);
Mauro Carvalho Chehabb9905382009-08-05 21:36:35 -03002000 if (unlikely(rc < 0)) {
Mauro Carvalho Chehabd5381642009-07-09 22:06:41 -03002001 debugf0("MC: " __FILE__
2002 ": %s(): failed edac_mce_register()\n", __func__);
Mauro Carvalho Chehabf4742942009-09-05 02:35:08 -03002003 }
2004
2005fail:
Tony Luckd4d1ef42010-05-18 10:53:25 -03002006 if (rc < 0)
2007 edac_mc_free(mci);
Mauro Carvalho Chehabf4742942009-09-05 02:35:08 -03002008 return rc;
2009}
2010
2011/*
2012 * i7core_probe Probe for ONE instance of device to see if it is
2013 * present.
2014 * return:
2015 * 0 for FOUND a device
2016 * < 0 for error code
2017 */
Mauro Carvalho Chehab2d95d812010-06-30 01:42:21 -03002018
Mauro Carvalho Chehabf4742942009-09-05 02:35:08 -03002019static int __devinit i7core_probe(struct pci_dev *pdev,
2020 const struct pci_device_id *id)
2021{
Mauro Carvalho Chehabf4742942009-09-05 02:35:08 -03002022 int rc;
2023 struct i7core_dev *i7core_dev;
2024
Mauro Carvalho Chehab2d95d812010-06-30 01:42:21 -03002025 /* get the pci devices we want to reserve for our use */
2026 mutex_lock(&i7core_edac_lock);
2027
Mauro Carvalho Chehabf4742942009-09-05 02:35:08 -03002028 /*
Mauro Carvalho Chehabd4c27792009-09-05 04:12:02 -03002029 * All memory controllers are allocated at the first pass.
Mauro Carvalho Chehabf4742942009-09-05 02:35:08 -03002030 */
Mauro Carvalho Chehab2d95d812010-06-30 01:42:21 -03002031 if (unlikely(probed >= 1)) {
2032 mutex_unlock(&i7core_edac_lock);
Mauro Carvalho Chehabf4742942009-09-05 02:35:08 -03002033 return -EINVAL;
Mauro Carvalho Chehab2d95d812010-06-30 01:42:21 -03002034 }
2035 probed++;
Mauro Carvalho Chehabde06eee2009-10-14 08:02:40 -03002036
Hidetoshi Seto64c10f62010-08-20 04:28:14 -03002037 rc = i7core_get_all_devices();
Mauro Carvalho Chehabf4742942009-09-05 02:35:08 -03002038 if (unlikely(rc < 0))
2039 goto fail0;
2040
2041 list_for_each_entry(i7core_dev, &i7core_edac_list, list) {
2042 int channels;
2043 int csrows;
2044
2045 /* Check the number of active and not disabled channels */
2046 rc = i7core_get_active_channels(i7core_dev->socket,
2047 &channels, &csrows);
2048 if (unlikely(rc < 0))
2049 goto fail1;
2050
Mauro Carvalho Chehabd4c27792009-09-05 04:12:02 -03002051 rc = i7core_register_mci(i7core_dev, channels, csrows);
2052 if (unlikely(rc < 0))
2053 goto fail1;
Mauro Carvalho Chehabd5381642009-07-09 22:06:41 -03002054 }
2055
Mauro Carvalho Chehabef708b52009-06-22 22:48:30 -03002056 i7core_printk(KERN_INFO, "Driver loaded.\n");
Mauro Carvalho Chehab8f331902009-06-22 22:48:29 -03002057
Mauro Carvalho Chehab66607702009-09-05 00:52:11 -03002058 mutex_unlock(&i7core_edac_lock);
Mauro Carvalho Chehaba0c36a12009-06-22 22:41:15 -03002059 return 0;
2060
Mauro Carvalho Chehab66607702009-09-05 00:52:11 -03002061fail1:
Mauro Carvalho Chehab13d6e9b2009-09-05 12:15:20 -03002062 i7core_put_all_devices();
Mauro Carvalho Chehab66607702009-09-05 00:52:11 -03002063fail0:
2064 mutex_unlock(&i7core_edac_lock);
Mauro Carvalho Chehabb7c76152009-06-22 22:48:30 -03002065 return rc;
Mauro Carvalho Chehaba0c36a12009-06-22 22:41:15 -03002066}
2067
2068/*
2069 * i7core_remove destructor for one instance of device
2070 *
2071 */
2072static void __devexit i7core_remove(struct pci_dev *pdev)
2073{
2074 struct mem_ctl_info *mci;
Hidetoshi Seto64c10f62010-08-20 04:28:14 -03002075 struct i7core_dev *i7core_dev;
Mauro Carvalho Chehab939747bd2010-08-10 11:22:01 -03002076 struct i7core_pvt *pvt;
Mauro Carvalho Chehaba0c36a12009-06-22 22:41:15 -03002077
2078 debugf0(__FILE__ ": %s()\n", __func__);
2079
Mauro Carvalho Chehab22e6bcb2009-09-05 23:06:50 -03002080 /*
2081 * we have a trouble here: pdev value for removal will be wrong, since
2082 * it will point to the X58 register used to detect that the machine
2083 * is a Nehalem or upper design. However, due to the way several PCI
2084 * devices are grouped together to provide MC functionality, we need
2085 * to use a different method for releasing the devices
2086 */
Mauro Carvalho Chehab87d1d272009-06-22 22:48:29 -03002087
Mauro Carvalho Chehab66607702009-09-05 00:52:11 -03002088 mutex_lock(&i7core_edac_lock);
Hidetoshi Seto64c10f62010-08-20 04:28:14 -03002089 list_for_each_entry(i7core_dev, &i7core_edac_list, list) {
Mauro Carvalho Chehab939747bd2010-08-10 11:22:01 -03002090 mci = find_mci_by_dev(&i7core_dev->pdev[0]->dev);
2091 if (unlikely(!mci || !mci->pvt_info)) {
Mauro Carvalho Chehab3cfd0142010-08-10 23:23:46 -03002092 debugf0("MC: " __FILE__ ": %s(): dev = %p\n",
2093 __func__, &i7core_dev->pdev[0]->dev);
2094
2095 i7core_printk(KERN_ERR,
Mauro Carvalho Chehab939747bd2010-08-10 11:22:01 -03002096 "Couldn't find mci hanler\n");
2097 } else {
2098 pvt = mci->pvt_info;
Mauro Carvalho Chehab22e6bcb2009-09-05 23:06:50 -03002099 i7core_dev = pvt->i7core_dev;
Mauro Carvalho Chehab939747bd2010-08-10 11:22:01 -03002100
Mauro Carvalho Chehab3cfd0142010-08-10 23:23:46 -03002101 debugf0("MC: " __FILE__ ": %s(): mci = %p, dev = %p\n",
2102 __func__, mci, &i7core_dev->pdev[0]->dev);
2103
Mauro Carvalho Chehab41ba6c12010-08-11 00:58:11 -03002104 /* Disable MCE NMI handler */
2105 edac_mce_unregister(&pvt->edac_mce);
2106
2107 /* Disable EDAC polling */
Hidetoshi Setoa3aa0a42010-08-20 04:25:18 -03002108 i7core_pci_ctl_release(pvt);
Mauro Carvalho Chehab939747bd2010-08-10 11:22:01 -03002109
Mauro Carvalho Chehab41ba6c12010-08-11 00:58:11 -03002110 /* Remove MC sysfs nodes */
Mauro Carvalho Chehab939747bd2010-08-10 11:22:01 -03002111 edac_mc_del_mc(&i7core_dev->pdev[0]->dev);
2112
Mauro Carvalho Chehabaccf74f2010-08-16 18:34:37 -03002113 debugf1("%s: free mci struct\n", mci->ctl_name);
Mauro Carvalho Chehab22e6bcb2009-09-05 23:06:50 -03002114 kfree(mci->ctl_name);
2115 edac_mc_free(mci);
Mauro Carvalho Chehab22e6bcb2009-09-05 23:06:50 -03002116 }
2117 }
Hidetoshi Seto64c10f62010-08-20 04:28:14 -03002118
2119 /* Release PCI resources */
2120 i7core_put_all_devices();
2121
Mauro Carvalho Chehab2d95d812010-06-30 01:42:21 -03002122 probed--;
2123
Mauro Carvalho Chehab22e6bcb2009-09-05 23:06:50 -03002124 mutex_unlock(&i7core_edac_lock);
Mauro Carvalho Chehaba0c36a12009-06-22 22:41:15 -03002125}
2126
Mauro Carvalho Chehaba0c36a12009-06-22 22:41:15 -03002127MODULE_DEVICE_TABLE(pci, i7core_pci_tbl);
2128
2129/*
2130 * i7core_driver pci_driver structure for this module
2131 *
2132 */
2133static struct pci_driver i7core_driver = {
2134 .name = "i7core_edac",
2135 .probe = i7core_probe,
2136 .remove = __devexit_p(i7core_remove),
2137 .id_table = i7core_pci_tbl,
2138};
2139
2140/*
2141 * i7core_init Module entry function
2142 * Try to initialize this module for its devices
2143 */
2144static int __init i7core_init(void)
2145{
2146 int pci_rc;
2147
2148 debugf2("MC: " __FILE__ ": %s()\n", __func__);
2149
2150 /* Ensure that the OPSTATE is set correctly for POLL or NMI */
2151 opstate_init();
2152
Mauro Carvalho Chehab54a08ab2010-08-19 15:51:00 -03002153 if (use_pci_fixup)
2154 i7core_xeon_pci_fixup(pci_dev_table);
Keith Manntheybc2d7242009-09-03 00:05:05 -03002155
Mauro Carvalho Chehaba0c36a12009-06-22 22:41:15 -03002156 pci_rc = pci_register_driver(&i7core_driver);
2157
Mauro Carvalho Chehab3ef288a2009-09-02 23:43:33 -03002158 if (pci_rc >= 0)
2159 return 0;
2160
2161 i7core_printk(KERN_ERR, "Failed to register device with error %d.\n",
2162 pci_rc);
2163
2164 return pci_rc;
Mauro Carvalho Chehaba0c36a12009-06-22 22:41:15 -03002165}
2166
2167/*
2168 * i7core_exit() Module exit function
2169 * Unregister the driver
2170 */
2171static void __exit i7core_exit(void)
2172{
2173 debugf2("MC: " __FILE__ ": %s()\n", __func__);
2174 pci_unregister_driver(&i7core_driver);
2175}
2176
2177module_init(i7core_init);
2178module_exit(i7core_exit);
2179
2180MODULE_LICENSE("GPL");
2181MODULE_AUTHOR("Mauro Carvalho Chehab <mchehab@redhat.com>");
2182MODULE_AUTHOR("Red Hat Inc. (http://www.redhat.com)");
2183MODULE_DESCRIPTION("MC Driver for Intel i7 Core memory controllers - "
2184 I7CORE_REVISION);
2185
2186module_param(edac_op_state, int, 0444);
2187MODULE_PARM_DESC(edac_op_state, "EDAC Error Reporting state: 0=Poll,1=NMI");