powercap: restrict energy meter to root access
[linux/fpc-iii.git] / drivers / edac / i7core_edac.c
blobf95d5b9c5551d8f1ae958bceaaadc987ec35aad5
1 /* Intel i7 core/Nehalem Memory Controller kernel module
3 * This driver supports the 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.
8 * This file may be distributed under the terms of the
9 * GNU General Public License version 2 only.
11 * Copyright (c) 2009-2010 by:
12 * Mauro Carvalho Chehab
14 * Red Hat Inc. http://www.redhat.com
16 * Forked and adapted from the i5400_edac driver
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
28 #include <linux/module.h>
29 #include <linux/init.h>
30 #include <linux/pci.h>
31 #include <linux/pci_ids.h>
32 #include <linux/slab.h>
33 #include <linux/delay.h>
34 #include <linux/dmi.h>
35 #include <linux/edac.h>
36 #include <linux/mmzone.h>
37 #include <linux/smp.h>
38 #include <asm/mce.h>
39 #include <asm/processor.h>
40 #include <asm/div64.h>
42 #include "edac_core.h"
44 /* Static vars */
45 static LIST_HEAD(i7core_edac_list);
46 static DEFINE_MUTEX(i7core_edac_lock);
47 static int probed;
49 static int use_pci_fixup;
50 module_param(use_pci_fixup, int, 0444);
51 MODULE_PARM_DESC(use_pci_fixup, "Enable PCI fixup to seek for hidden devices");
53 * This is used for Nehalem-EP and Nehalem-EX devices, where the non-core
54 * registers start at bus 255, and are not reported by BIOS.
55 * We currently find devices with only 2 sockets. In order to support more QPI
56 * Quick Path Interconnect, just increment this number.
58 #define MAX_SOCKET_BUSES 2
62 * Alter this version for the module when modifications are made
64 #define I7CORE_REVISION " Ver: 1.0.0"
65 #define EDAC_MOD_STR "i7core_edac"
68 * Debug macros
70 #define i7core_printk(level, fmt, arg...) \
71 edac_printk(level, "i7core", fmt, ##arg)
73 #define i7core_mc_printk(mci, level, fmt, arg...) \
74 edac_mc_chipset_printk(mci, level, "i7core", fmt, ##arg)
77 * i7core Memory Controller Registers
80 /* OFFSETS for Device 0 Function 0 */
82 #define MC_CFG_CONTROL 0x90
83 #define MC_CFG_UNLOCK 0x02
84 #define MC_CFG_LOCK 0x00
86 /* OFFSETS for Device 3 Function 0 */
88 #define MC_CONTROL 0x48
89 #define MC_STATUS 0x4c
90 #define MC_MAX_DOD 0x64
93 * OFFSETS for Device 3 Function 4, as indicated on Xeon 5500 datasheet:
94 * http://www.arrownac.com/manufacturers/intel/s/nehalem/5500-datasheet-v2.pdf
97 #define MC_TEST_ERR_RCV1 0x60
98 #define DIMM2_COR_ERR(r) ((r) & 0x7fff)
100 #define MC_TEST_ERR_RCV0 0x64
101 #define DIMM1_COR_ERR(r) (((r) >> 16) & 0x7fff)
102 #define DIMM0_COR_ERR(r) ((r) & 0x7fff)
104 /* OFFSETS for Device 3 Function 2, as indicated on Xeon 5500 datasheet */
105 #define MC_SSRCONTROL 0x48
106 #define SSR_MODE_DISABLE 0x00
107 #define SSR_MODE_ENABLE 0x01
108 #define SSR_MODE_MASK 0x03
110 #define MC_SCRUB_CONTROL 0x4c
111 #define STARTSCRUB (1 << 24)
112 #define SCRUBINTERVAL_MASK 0xffffff
114 #define MC_COR_ECC_CNT_0 0x80
115 #define MC_COR_ECC_CNT_1 0x84
116 #define MC_COR_ECC_CNT_2 0x88
117 #define MC_COR_ECC_CNT_3 0x8c
118 #define MC_COR_ECC_CNT_4 0x90
119 #define MC_COR_ECC_CNT_5 0x94
121 #define DIMM_TOP_COR_ERR(r) (((r) >> 16) & 0x7fff)
122 #define DIMM_BOT_COR_ERR(r) ((r) & 0x7fff)
125 /* OFFSETS for Devices 4,5 and 6 Function 0 */
127 #define MC_CHANNEL_DIMM_INIT_PARAMS 0x58
128 #define THREE_DIMMS_PRESENT (1 << 24)
129 #define SINGLE_QUAD_RANK_PRESENT (1 << 23)
130 #define QUAD_RANK_PRESENT (1 << 22)
131 #define REGISTERED_DIMM (1 << 15)
133 #define MC_CHANNEL_MAPPER 0x60
134 #define RDLCH(r, ch) ((((r) >> (3 + (ch * 6))) & 0x07) - 1)
135 #define WRLCH(r, ch) ((((r) >> (ch * 6)) & 0x07) - 1)
137 #define MC_CHANNEL_RANK_PRESENT 0x7c
138 #define RANK_PRESENT_MASK 0xffff
140 #define MC_CHANNEL_ADDR_MATCH 0xf0
141 #define MC_CHANNEL_ERROR_MASK 0xf8
142 #define MC_CHANNEL_ERROR_INJECT 0xfc
143 #define INJECT_ADDR_PARITY 0x10
144 #define INJECT_ECC 0x08
145 #define MASK_CACHELINE 0x06
146 #define MASK_FULL_CACHELINE 0x06
147 #define MASK_MSB32_CACHELINE 0x04
148 #define MASK_LSB32_CACHELINE 0x02
149 #define NO_MASK_CACHELINE 0x00
150 #define REPEAT_EN 0x01
152 /* OFFSETS for Devices 4,5 and 6 Function 1 */
154 #define MC_DOD_CH_DIMM0 0x48
155 #define MC_DOD_CH_DIMM1 0x4c
156 #define MC_DOD_CH_DIMM2 0x50
157 #define RANKOFFSET_MASK ((1 << 12) | (1 << 11) | (1 << 10))
158 #define RANKOFFSET(x) ((x & RANKOFFSET_MASK) >> 10)
159 #define DIMM_PRESENT_MASK (1 << 9)
160 #define DIMM_PRESENT(x) (((x) & DIMM_PRESENT_MASK) >> 9)
161 #define MC_DOD_NUMBANK_MASK ((1 << 8) | (1 << 7))
162 #define MC_DOD_NUMBANK(x) (((x) & MC_DOD_NUMBANK_MASK) >> 7)
163 #define MC_DOD_NUMRANK_MASK ((1 << 6) | (1 << 5))
164 #define MC_DOD_NUMRANK(x) (((x) & MC_DOD_NUMRANK_MASK) >> 5)
165 #define MC_DOD_NUMROW_MASK ((1 << 4) | (1 << 3) | (1 << 2))
166 #define MC_DOD_NUMROW(x) (((x) & MC_DOD_NUMROW_MASK) >> 2)
167 #define MC_DOD_NUMCOL_MASK 3
168 #define MC_DOD_NUMCOL(x) ((x) & MC_DOD_NUMCOL_MASK)
170 #define MC_RANK_PRESENT 0x7c
172 #define MC_SAG_CH_0 0x80
173 #define MC_SAG_CH_1 0x84
174 #define MC_SAG_CH_2 0x88
175 #define MC_SAG_CH_3 0x8c
176 #define MC_SAG_CH_4 0x90
177 #define MC_SAG_CH_5 0x94
178 #define MC_SAG_CH_6 0x98
179 #define MC_SAG_CH_7 0x9c
181 #define MC_RIR_LIMIT_CH_0 0x40
182 #define MC_RIR_LIMIT_CH_1 0x44
183 #define MC_RIR_LIMIT_CH_2 0x48
184 #define MC_RIR_LIMIT_CH_3 0x4C
185 #define MC_RIR_LIMIT_CH_4 0x50
186 #define MC_RIR_LIMIT_CH_5 0x54
187 #define MC_RIR_LIMIT_CH_6 0x58
188 #define MC_RIR_LIMIT_CH_7 0x5C
189 #define MC_RIR_LIMIT_MASK ((1 << 10) - 1)
191 #define MC_RIR_WAY_CH 0x80
192 #define MC_RIR_WAY_OFFSET_MASK (((1 << 14) - 1) & ~0x7)
193 #define MC_RIR_WAY_RANK_MASK 0x7
196 * i7core structs
199 #define NUM_CHANS 3
200 #define MAX_DIMMS 3 /* Max DIMMS per channel */
201 #define MAX_MCR_FUNC 4
202 #define MAX_CHAN_FUNC 3
204 struct i7core_info {
205 u32 mc_control;
206 u32 mc_status;
207 u32 max_dod;
208 u32 ch_map;
212 struct i7core_inject {
213 int enable;
215 u32 section;
216 u32 type;
217 u32 eccmask;
219 /* Error address mask */
220 int channel, dimm, rank, bank, page, col;
223 struct i7core_channel {
224 bool is_3dimms_present;
225 bool is_single_4rank;
226 bool has_4rank;
227 u32 dimms;
230 struct pci_id_descr {
231 int dev;
232 int func;
233 int dev_id;
234 int optional;
237 struct pci_id_table {
238 const struct pci_id_descr *descr;
239 int n_devs;
242 struct i7core_dev {
243 struct list_head list;
244 u8 socket;
245 struct pci_dev **pdev;
246 int n_devs;
247 struct mem_ctl_info *mci;
250 struct i7core_pvt {
251 struct device *addrmatch_dev, *chancounts_dev;
253 struct pci_dev *pci_noncore;
254 struct pci_dev *pci_mcr[MAX_MCR_FUNC + 1];
255 struct pci_dev *pci_ch[NUM_CHANS][MAX_CHAN_FUNC + 1];
257 struct i7core_dev *i7core_dev;
259 struct i7core_info info;
260 struct i7core_inject inject;
261 struct i7core_channel channel[NUM_CHANS];
263 int ce_count_available;
265 /* ECC corrected errors counts per udimm */
266 unsigned long udimm_ce_count[MAX_DIMMS];
267 int udimm_last_ce_count[MAX_DIMMS];
268 /* ECC corrected errors counts per rdimm */
269 unsigned long rdimm_ce_count[NUM_CHANS][MAX_DIMMS];
270 int rdimm_last_ce_count[NUM_CHANS][MAX_DIMMS];
272 bool is_registered, enable_scrub;
274 /* DCLK Frequency used for computing scrub rate */
275 int dclk_freq;
277 /* Struct to control EDAC polling */
278 struct edac_pci_ctl_info *i7core_pci;
281 #define PCI_DESCR(device, function, device_id) \
282 .dev = (device), \
283 .func = (function), \
284 .dev_id = (device_id)
286 static const struct pci_id_descr pci_dev_descr_i7core_nehalem[] = {
287 /* Memory controller */
288 { PCI_DESCR(3, 0, PCI_DEVICE_ID_INTEL_I7_MCR) },
289 { PCI_DESCR(3, 1, PCI_DEVICE_ID_INTEL_I7_MC_TAD) },
290 /* Exists only for RDIMM */
291 { PCI_DESCR(3, 2, PCI_DEVICE_ID_INTEL_I7_MC_RAS), .optional = 1 },
292 { PCI_DESCR(3, 4, PCI_DEVICE_ID_INTEL_I7_MC_TEST) },
294 /* Channel 0 */
295 { PCI_DESCR(4, 0, PCI_DEVICE_ID_INTEL_I7_MC_CH0_CTRL) },
296 { PCI_DESCR(4, 1, PCI_DEVICE_ID_INTEL_I7_MC_CH0_ADDR) },
297 { PCI_DESCR(4, 2, PCI_DEVICE_ID_INTEL_I7_MC_CH0_RANK) },
298 { PCI_DESCR(4, 3, PCI_DEVICE_ID_INTEL_I7_MC_CH0_TC) },
300 /* Channel 1 */
301 { PCI_DESCR(5, 0, PCI_DEVICE_ID_INTEL_I7_MC_CH1_CTRL) },
302 { PCI_DESCR(5, 1, PCI_DEVICE_ID_INTEL_I7_MC_CH1_ADDR) },
303 { PCI_DESCR(5, 2, PCI_DEVICE_ID_INTEL_I7_MC_CH1_RANK) },
304 { PCI_DESCR(5, 3, PCI_DEVICE_ID_INTEL_I7_MC_CH1_TC) },
306 /* Channel 2 */
307 { PCI_DESCR(6, 0, PCI_DEVICE_ID_INTEL_I7_MC_CH2_CTRL) },
308 { PCI_DESCR(6, 1, PCI_DEVICE_ID_INTEL_I7_MC_CH2_ADDR) },
309 { PCI_DESCR(6, 2, PCI_DEVICE_ID_INTEL_I7_MC_CH2_RANK) },
310 { PCI_DESCR(6, 3, PCI_DEVICE_ID_INTEL_I7_MC_CH2_TC) },
312 /* Generic Non-core registers */
314 * This is the PCI device on i7core and on Xeon 35xx (8086:2c41)
315 * On Xeon 55xx, however, it has a different id (8086:2c40). So,
316 * the probing code needs to test for the other address in case of
317 * failure of this one
319 { PCI_DESCR(0, 0, PCI_DEVICE_ID_INTEL_I7_NONCORE) },
323 static const struct pci_id_descr pci_dev_descr_lynnfield[] = {
324 { PCI_DESCR( 3, 0, PCI_DEVICE_ID_INTEL_LYNNFIELD_MCR) },
325 { PCI_DESCR( 3, 1, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_TAD) },
326 { PCI_DESCR( 3, 4, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_TEST) },
328 { PCI_DESCR( 4, 0, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH0_CTRL) },
329 { PCI_DESCR( 4, 1, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH0_ADDR) },
330 { PCI_DESCR( 4, 2, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH0_RANK) },
331 { PCI_DESCR( 4, 3, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH0_TC) },
333 { PCI_DESCR( 5, 0, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH1_CTRL) },
334 { PCI_DESCR( 5, 1, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH1_ADDR) },
335 { PCI_DESCR( 5, 2, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH1_RANK) },
336 { PCI_DESCR( 5, 3, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH1_TC) },
339 * This is the PCI device has an alternate address on some
340 * processors like Core i7 860
342 { PCI_DESCR( 0, 0, PCI_DEVICE_ID_INTEL_LYNNFIELD_NONCORE) },
345 static const struct pci_id_descr pci_dev_descr_i7core_westmere[] = {
346 /* Memory controller */
347 { PCI_DESCR(3, 0, PCI_DEVICE_ID_INTEL_LYNNFIELD_MCR_REV2) },
348 { PCI_DESCR(3, 1, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_TAD_REV2) },
349 /* Exists only for RDIMM */
350 { PCI_DESCR(3, 2, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_RAS_REV2), .optional = 1 },
351 { PCI_DESCR(3, 4, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_TEST_REV2) },
353 /* Channel 0 */
354 { PCI_DESCR(4, 0, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH0_CTRL_REV2) },
355 { PCI_DESCR(4, 1, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH0_ADDR_REV2) },
356 { PCI_DESCR(4, 2, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH0_RANK_REV2) },
357 { PCI_DESCR(4, 3, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH0_TC_REV2) },
359 /* Channel 1 */
360 { PCI_DESCR(5, 0, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH1_CTRL_REV2) },
361 { PCI_DESCR(5, 1, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH1_ADDR_REV2) },
362 { PCI_DESCR(5, 2, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH1_RANK_REV2) },
363 { PCI_DESCR(5, 3, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH1_TC_REV2) },
365 /* Channel 2 */
366 { PCI_DESCR(6, 0, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH2_CTRL_REV2) },
367 { PCI_DESCR(6, 1, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH2_ADDR_REV2) },
368 { PCI_DESCR(6, 2, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH2_RANK_REV2) },
369 { PCI_DESCR(6, 3, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH2_TC_REV2) },
371 /* Generic Non-core registers */
372 { PCI_DESCR(0, 0, PCI_DEVICE_ID_INTEL_LYNNFIELD_NONCORE_REV2) },
376 #define PCI_ID_TABLE_ENTRY(A) { .descr=A, .n_devs = ARRAY_SIZE(A) }
377 static const struct pci_id_table pci_dev_table[] = {
378 PCI_ID_TABLE_ENTRY(pci_dev_descr_i7core_nehalem),
379 PCI_ID_TABLE_ENTRY(pci_dev_descr_lynnfield),
380 PCI_ID_TABLE_ENTRY(pci_dev_descr_i7core_westmere),
381 {0,} /* 0 terminated list. */
385 * pci_device_id table for which devices we are looking for
387 static const struct pci_device_id i7core_pci_tbl[] = {
388 {PCI_DEVICE(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_X58_HUB_MGMT)},
389 {PCI_DEVICE(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_LYNNFIELD_QPI_LINK0)},
390 {0,} /* 0 terminated list. */
393 /****************************************************************************
394 Ancillary status routines
395 ****************************************************************************/
397 /* MC_CONTROL bits */
398 #define CH_ACTIVE(pvt, ch) ((pvt)->info.mc_control & (1 << (8 + ch)))
399 #define ECCx8(pvt) ((pvt)->info.mc_control & (1 << 1))
401 /* MC_STATUS bits */
402 #define ECC_ENABLED(pvt) ((pvt)->info.mc_status & (1 << 4))
403 #define CH_DISABLED(pvt, ch) ((pvt)->info.mc_status & (1 << ch))
405 /* MC_MAX_DOD read functions */
406 static inline int numdimms(u32 dimms)
408 return (dimms & 0x3) + 1;
411 static inline int numrank(u32 rank)
413 static const int ranks[] = { 1, 2, 4, -EINVAL };
415 return ranks[rank & 0x3];
418 static inline int numbank(u32 bank)
420 static const int banks[] = { 4, 8, 16, -EINVAL };
422 return banks[bank & 0x3];
425 static inline int numrow(u32 row)
427 static const int rows[] = {
428 1 << 12, 1 << 13, 1 << 14, 1 << 15,
429 1 << 16, -EINVAL, -EINVAL, -EINVAL,
432 return rows[row & 0x7];
435 static inline int numcol(u32 col)
437 static const int cols[] = {
438 1 << 10, 1 << 11, 1 << 12, -EINVAL,
440 return cols[col & 0x3];
443 static struct i7core_dev *get_i7core_dev(u8 socket)
445 struct i7core_dev *i7core_dev;
447 list_for_each_entry(i7core_dev, &i7core_edac_list, list) {
448 if (i7core_dev->socket == socket)
449 return i7core_dev;
452 return NULL;
455 static struct i7core_dev *alloc_i7core_dev(u8 socket,
456 const struct pci_id_table *table)
458 struct i7core_dev *i7core_dev;
460 i7core_dev = kzalloc(sizeof(*i7core_dev), GFP_KERNEL);
461 if (!i7core_dev)
462 return NULL;
464 i7core_dev->pdev = kzalloc(sizeof(*i7core_dev->pdev) * table->n_devs,
465 GFP_KERNEL);
466 if (!i7core_dev->pdev) {
467 kfree(i7core_dev);
468 return NULL;
471 i7core_dev->socket = socket;
472 i7core_dev->n_devs = table->n_devs;
473 list_add_tail(&i7core_dev->list, &i7core_edac_list);
475 return i7core_dev;
478 static void free_i7core_dev(struct i7core_dev *i7core_dev)
480 list_del(&i7core_dev->list);
481 kfree(i7core_dev->pdev);
482 kfree(i7core_dev);
485 /****************************************************************************
486 Memory check routines
487 ****************************************************************************/
489 static int get_dimm_config(struct mem_ctl_info *mci)
491 struct i7core_pvt *pvt = mci->pvt_info;
492 struct pci_dev *pdev;
493 int i, j;
494 enum edac_type mode;
495 enum mem_type mtype;
496 struct dimm_info *dimm;
498 /* Get data from the MC register, function 0 */
499 pdev = pvt->pci_mcr[0];
500 if (!pdev)
501 return -ENODEV;
503 /* Device 3 function 0 reads */
504 pci_read_config_dword(pdev, MC_CONTROL, &pvt->info.mc_control);
505 pci_read_config_dword(pdev, MC_STATUS, &pvt->info.mc_status);
506 pci_read_config_dword(pdev, MC_MAX_DOD, &pvt->info.max_dod);
507 pci_read_config_dword(pdev, MC_CHANNEL_MAPPER, &pvt->info.ch_map);
509 edac_dbg(0, "QPI %d control=0x%08x status=0x%08x dod=0x%08x map=0x%08x\n",
510 pvt->i7core_dev->socket, pvt->info.mc_control,
511 pvt->info.mc_status, pvt->info.max_dod, pvt->info.ch_map);
513 if (ECC_ENABLED(pvt)) {
514 edac_dbg(0, "ECC enabled with x%d SDCC\n", ECCx8(pvt) ? 8 : 4);
515 if (ECCx8(pvt))
516 mode = EDAC_S8ECD8ED;
517 else
518 mode = EDAC_S4ECD4ED;
519 } else {
520 edac_dbg(0, "ECC disabled\n");
521 mode = EDAC_NONE;
524 /* FIXME: need to handle the error codes */
525 edac_dbg(0, "DOD Max limits: DIMMS: %d, %d-ranked, %d-banked x%x x 0x%x\n",
526 numdimms(pvt->info.max_dod),
527 numrank(pvt->info.max_dod >> 2),
528 numbank(pvt->info.max_dod >> 4),
529 numrow(pvt->info.max_dod >> 6),
530 numcol(pvt->info.max_dod >> 9));
532 for (i = 0; i < NUM_CHANS; i++) {
533 u32 data, dimm_dod[3], value[8];
535 if (!pvt->pci_ch[i][0])
536 continue;
538 if (!CH_ACTIVE(pvt, i)) {
539 edac_dbg(0, "Channel %i is not active\n", i);
540 continue;
542 if (CH_DISABLED(pvt, i)) {
543 edac_dbg(0, "Channel %i is disabled\n", i);
544 continue;
547 /* Devices 4-6 function 0 */
548 pci_read_config_dword(pvt->pci_ch[i][0],
549 MC_CHANNEL_DIMM_INIT_PARAMS, &data);
552 if (data & THREE_DIMMS_PRESENT)
553 pvt->channel[i].is_3dimms_present = true;
555 if (data & SINGLE_QUAD_RANK_PRESENT)
556 pvt->channel[i].is_single_4rank = true;
558 if (data & QUAD_RANK_PRESENT)
559 pvt->channel[i].has_4rank = true;
561 if (data & REGISTERED_DIMM)
562 mtype = MEM_RDDR3;
563 else
564 mtype = MEM_DDR3;
566 /* Devices 4-6 function 1 */
567 pci_read_config_dword(pvt->pci_ch[i][1],
568 MC_DOD_CH_DIMM0, &dimm_dod[0]);
569 pci_read_config_dword(pvt->pci_ch[i][1],
570 MC_DOD_CH_DIMM1, &dimm_dod[1]);
571 pci_read_config_dword(pvt->pci_ch[i][1],
572 MC_DOD_CH_DIMM2, &dimm_dod[2]);
574 edac_dbg(0, "Ch%d phy rd%d, wr%d (0x%08x): %s%s%s%cDIMMs\n",
576 RDLCH(pvt->info.ch_map, i), WRLCH(pvt->info.ch_map, i),
577 data,
578 pvt->channel[i].is_3dimms_present ? "3DIMMS " : "",
579 pvt->channel[i].is_3dimms_present ? "SINGLE_4R " : "",
580 pvt->channel[i].has_4rank ? "HAS_4R " : "",
581 (data & REGISTERED_DIMM) ? 'R' : 'U');
583 for (j = 0; j < 3; j++) {
584 u32 banks, ranks, rows, cols;
585 u32 size, npages;
587 if (!DIMM_PRESENT(dimm_dod[j]))
588 continue;
590 dimm = EDAC_DIMM_PTR(mci->layers, mci->dimms, mci->n_layers,
591 i, j, 0);
592 banks = numbank(MC_DOD_NUMBANK(dimm_dod[j]));
593 ranks = numrank(MC_DOD_NUMRANK(dimm_dod[j]));
594 rows = numrow(MC_DOD_NUMROW(dimm_dod[j]));
595 cols = numcol(MC_DOD_NUMCOL(dimm_dod[j]));
597 /* DDR3 has 8 I/O banks */
598 size = (rows * cols * banks * ranks) >> (20 - 3);
600 edac_dbg(0, "\tdimm %d %d Mb offset: %x, bank: %d, rank: %d, row: %#x, col: %#x\n",
601 j, size,
602 RANKOFFSET(dimm_dod[j]),
603 banks, ranks, rows, cols);
605 npages = MiB_TO_PAGES(size);
607 dimm->nr_pages = npages;
609 switch (banks) {
610 case 4:
611 dimm->dtype = DEV_X4;
612 break;
613 case 8:
614 dimm->dtype = DEV_X8;
615 break;
616 case 16:
617 dimm->dtype = DEV_X16;
618 break;
619 default:
620 dimm->dtype = DEV_UNKNOWN;
623 snprintf(dimm->label, sizeof(dimm->label),
624 "CPU#%uChannel#%u_DIMM#%u",
625 pvt->i7core_dev->socket, i, j);
626 dimm->grain = 8;
627 dimm->edac_mode = mode;
628 dimm->mtype = mtype;
631 pci_read_config_dword(pdev, MC_SAG_CH_0, &value[0]);
632 pci_read_config_dword(pdev, MC_SAG_CH_1, &value[1]);
633 pci_read_config_dword(pdev, MC_SAG_CH_2, &value[2]);
634 pci_read_config_dword(pdev, MC_SAG_CH_3, &value[3]);
635 pci_read_config_dword(pdev, MC_SAG_CH_4, &value[4]);
636 pci_read_config_dword(pdev, MC_SAG_CH_5, &value[5]);
637 pci_read_config_dword(pdev, MC_SAG_CH_6, &value[6]);
638 pci_read_config_dword(pdev, MC_SAG_CH_7, &value[7]);
639 edac_dbg(1, "\t[%i] DIVBY3\tREMOVED\tOFFSET\n", i);
640 for (j = 0; j < 8; j++)
641 edac_dbg(1, "\t\t%#x\t%#x\t%#x\n",
642 (value[j] >> 27) & 0x1,
643 (value[j] >> 24) & 0x7,
644 (value[j] & ((1 << 24) - 1)));
647 return 0;
650 /****************************************************************************
651 Error insertion routines
652 ****************************************************************************/
654 #define to_mci(k) container_of(k, struct mem_ctl_info, dev)
656 /* The i7core has independent error injection features per channel.
657 However, to have a simpler code, we don't allow enabling error injection
658 on more than one channel.
659 Also, since a change at an inject parameter will be applied only at enable,
660 we're disabling error injection on all write calls to the sysfs nodes that
661 controls the error code injection.
663 static int disable_inject(const struct mem_ctl_info *mci)
665 struct i7core_pvt *pvt = mci->pvt_info;
667 pvt->inject.enable = 0;
669 if (!pvt->pci_ch[pvt->inject.channel][0])
670 return -ENODEV;
672 pci_write_config_dword(pvt->pci_ch[pvt->inject.channel][0],
673 MC_CHANNEL_ERROR_INJECT, 0);
675 return 0;
679 * i7core inject inject.section
681 * accept and store error injection inject.section value
682 * bit 0 - refers to the lower 32-byte half cacheline
683 * bit 1 - refers to the upper 32-byte half cacheline
685 static ssize_t i7core_inject_section_store(struct device *dev,
686 struct device_attribute *mattr,
687 const char *data, size_t count)
689 struct mem_ctl_info *mci = to_mci(dev);
690 struct i7core_pvt *pvt = mci->pvt_info;
691 unsigned long value;
692 int rc;
694 if (pvt->inject.enable)
695 disable_inject(mci);
697 rc = kstrtoul(data, 10, &value);
698 if ((rc < 0) || (value > 3))
699 return -EIO;
701 pvt->inject.section = (u32) value;
702 return count;
705 static ssize_t i7core_inject_section_show(struct device *dev,
706 struct device_attribute *mattr,
707 char *data)
709 struct mem_ctl_info *mci = to_mci(dev);
710 struct i7core_pvt *pvt = mci->pvt_info;
711 return sprintf(data, "0x%08x\n", pvt->inject.section);
715 * i7core inject.type
717 * accept and store error injection inject.section value
718 * bit 0 - repeat enable - Enable error repetition
719 * bit 1 - inject ECC error
720 * bit 2 - inject parity error
722 static ssize_t i7core_inject_type_store(struct device *dev,
723 struct device_attribute *mattr,
724 const char *data, size_t count)
726 struct mem_ctl_info *mci = to_mci(dev);
727 struct i7core_pvt *pvt = mci->pvt_info;
728 unsigned long value;
729 int rc;
731 if (pvt->inject.enable)
732 disable_inject(mci);
734 rc = kstrtoul(data, 10, &value);
735 if ((rc < 0) || (value > 7))
736 return -EIO;
738 pvt->inject.type = (u32) value;
739 return count;
742 static ssize_t i7core_inject_type_show(struct device *dev,
743 struct device_attribute *mattr,
744 char *data)
746 struct mem_ctl_info *mci = to_mci(dev);
747 struct i7core_pvt *pvt = mci->pvt_info;
749 return sprintf(data, "0x%08x\n", pvt->inject.type);
753 * i7core_inject_inject.eccmask_store
755 * The type of error (UE/CE) will depend on the inject.eccmask value:
756 * Any bits set to a 1 will flip the corresponding ECC bit
757 * Correctable errors can be injected by flipping 1 bit or the bits within
758 * a symbol pair (2 consecutive aligned 8-bit pairs - i.e. 7:0 and 15:8 or
759 * 23:16 and 31:24). Flipping bits in two symbol pairs will cause an
760 * uncorrectable error to be injected.
762 static ssize_t i7core_inject_eccmask_store(struct device *dev,
763 struct device_attribute *mattr,
764 const char *data, size_t count)
766 struct mem_ctl_info *mci = to_mci(dev);
767 struct i7core_pvt *pvt = mci->pvt_info;
768 unsigned long value;
769 int rc;
771 if (pvt->inject.enable)
772 disable_inject(mci);
774 rc = kstrtoul(data, 10, &value);
775 if (rc < 0)
776 return -EIO;
778 pvt->inject.eccmask = (u32) value;
779 return count;
782 static ssize_t i7core_inject_eccmask_show(struct device *dev,
783 struct device_attribute *mattr,
784 char *data)
786 struct mem_ctl_info *mci = to_mci(dev);
787 struct i7core_pvt *pvt = mci->pvt_info;
789 return sprintf(data, "0x%08x\n", pvt->inject.eccmask);
793 * i7core_addrmatch
795 * The type of error (UE/CE) will depend on the inject.eccmask value:
796 * Any bits set to a 1 will flip the corresponding ECC bit
797 * Correctable errors can be injected by flipping 1 bit or the bits within
798 * a symbol pair (2 consecutive aligned 8-bit pairs - i.e. 7:0 and 15:8 or
799 * 23:16 and 31:24). Flipping bits in two symbol pairs will cause an
800 * uncorrectable error to be injected.
803 #define DECLARE_ADDR_MATCH(param, limit) \
804 static ssize_t i7core_inject_store_##param( \
805 struct device *dev, \
806 struct device_attribute *mattr, \
807 const char *data, size_t count) \
809 struct mem_ctl_info *mci = dev_get_drvdata(dev); \
810 struct i7core_pvt *pvt; \
811 long value; \
812 int rc; \
814 edac_dbg(1, "\n"); \
815 pvt = mci->pvt_info; \
817 if (pvt->inject.enable) \
818 disable_inject(mci); \
820 if (!strcasecmp(data, "any") || !strcasecmp(data, "any\n"))\
821 value = -1; \
822 else { \
823 rc = kstrtoul(data, 10, &value); \
824 if ((rc < 0) || (value >= limit)) \
825 return -EIO; \
828 pvt->inject.param = value; \
830 return count; \
833 static ssize_t i7core_inject_show_##param( \
834 struct device *dev, \
835 struct device_attribute *mattr, \
836 char *data) \
838 struct mem_ctl_info *mci = dev_get_drvdata(dev); \
839 struct i7core_pvt *pvt; \
841 pvt = mci->pvt_info; \
842 edac_dbg(1, "pvt=%p\n", pvt); \
843 if (pvt->inject.param < 0) \
844 return sprintf(data, "any\n"); \
845 else \
846 return sprintf(data, "%d\n", pvt->inject.param);\
849 #define ATTR_ADDR_MATCH(param) \
850 static DEVICE_ATTR(param, S_IRUGO | S_IWUSR, \
851 i7core_inject_show_##param, \
852 i7core_inject_store_##param)
854 DECLARE_ADDR_MATCH(channel, 3);
855 DECLARE_ADDR_MATCH(dimm, 3);
856 DECLARE_ADDR_MATCH(rank, 4);
857 DECLARE_ADDR_MATCH(bank, 32);
858 DECLARE_ADDR_MATCH(page, 0x10000);
859 DECLARE_ADDR_MATCH(col, 0x4000);
861 ATTR_ADDR_MATCH(channel);
862 ATTR_ADDR_MATCH(dimm);
863 ATTR_ADDR_MATCH(rank);
864 ATTR_ADDR_MATCH(bank);
865 ATTR_ADDR_MATCH(page);
866 ATTR_ADDR_MATCH(col);
868 static int write_and_test(struct pci_dev *dev, const int where, const u32 val)
870 u32 read;
871 int count;
873 edac_dbg(0, "setting pci %02x:%02x.%x reg=%02x value=%08x\n",
874 dev->bus->number, PCI_SLOT(dev->devfn), PCI_FUNC(dev->devfn),
875 where, val);
877 for (count = 0; count < 10; count++) {
878 if (count)
879 msleep(100);
880 pci_write_config_dword(dev, where, val);
881 pci_read_config_dword(dev, where, &read);
883 if (read == val)
884 return 0;
887 i7core_printk(KERN_ERR, "Error during set pci %02x:%02x.%x reg=%02x "
888 "write=%08x. Read=%08x\n",
889 dev->bus->number, PCI_SLOT(dev->devfn), PCI_FUNC(dev->devfn),
890 where, val, read);
892 return -EINVAL;
896 * This routine prepares the Memory Controller for error injection.
897 * The error will be injected when some process tries to write to the
898 * memory that matches the given criteria.
899 * The criteria can be set in terms of a mask where dimm, rank, bank, page
900 * and col can be specified.
901 * A -1 value for any of the mask items will make the MCU to ignore
902 * that matching criteria for error injection.
904 * It should be noticed that the error will only happen after a write operation
905 * on a memory that matches the condition. if REPEAT_EN is not enabled at
906 * inject mask, then it will produce just one error. Otherwise, it will repeat
907 * until the injectmask would be cleaned.
909 * FIXME: This routine assumes that MAXNUMDIMMS value of MC_MAX_DOD
910 * is reliable enough to check if the MC is using the
911 * three channels. However, this is not clear at the datasheet.
913 static ssize_t i7core_inject_enable_store(struct device *dev,
914 struct device_attribute *mattr,
915 const char *data, size_t count)
917 struct mem_ctl_info *mci = to_mci(dev);
918 struct i7core_pvt *pvt = mci->pvt_info;
919 u32 injectmask;
920 u64 mask = 0;
921 int rc;
922 long enable;
924 if (!pvt->pci_ch[pvt->inject.channel][0])
925 return 0;
927 rc = kstrtoul(data, 10, &enable);
928 if ((rc < 0))
929 return 0;
931 if (enable) {
932 pvt->inject.enable = 1;
933 } else {
934 disable_inject(mci);
935 return count;
938 /* Sets pvt->inject.dimm mask */
939 if (pvt->inject.dimm < 0)
940 mask |= 1LL << 41;
941 else {
942 if (pvt->channel[pvt->inject.channel].dimms > 2)
943 mask |= (pvt->inject.dimm & 0x3LL) << 35;
944 else
945 mask |= (pvt->inject.dimm & 0x1LL) << 36;
948 /* Sets pvt->inject.rank mask */
949 if (pvt->inject.rank < 0)
950 mask |= 1LL << 40;
951 else {
952 if (pvt->channel[pvt->inject.channel].dimms > 2)
953 mask |= (pvt->inject.rank & 0x1LL) << 34;
954 else
955 mask |= (pvt->inject.rank & 0x3LL) << 34;
958 /* Sets pvt->inject.bank mask */
959 if (pvt->inject.bank < 0)
960 mask |= 1LL << 39;
961 else
962 mask |= (pvt->inject.bank & 0x15LL) << 30;
964 /* Sets pvt->inject.page mask */
965 if (pvt->inject.page < 0)
966 mask |= 1LL << 38;
967 else
968 mask |= (pvt->inject.page & 0xffff) << 14;
970 /* Sets pvt->inject.column mask */
971 if (pvt->inject.col < 0)
972 mask |= 1LL << 37;
973 else
974 mask |= (pvt->inject.col & 0x3fff);
977 * bit 0: REPEAT_EN
978 * bits 1-2: MASK_HALF_CACHELINE
979 * bit 3: INJECT_ECC
980 * bit 4: INJECT_ADDR_PARITY
983 injectmask = (pvt->inject.type & 1) |
984 (pvt->inject.section & 0x3) << 1 |
985 (pvt->inject.type & 0x6) << (3 - 1);
987 /* Unlock writes to registers - this register is write only */
988 pci_write_config_dword(pvt->pci_noncore,
989 MC_CFG_CONTROL, 0x2);
991 write_and_test(pvt->pci_ch[pvt->inject.channel][0],
992 MC_CHANNEL_ADDR_MATCH, mask);
993 write_and_test(pvt->pci_ch[pvt->inject.channel][0],
994 MC_CHANNEL_ADDR_MATCH + 4, mask >> 32L);
996 write_and_test(pvt->pci_ch[pvt->inject.channel][0],
997 MC_CHANNEL_ERROR_MASK, pvt->inject.eccmask);
999 write_and_test(pvt->pci_ch[pvt->inject.channel][0],
1000 MC_CHANNEL_ERROR_INJECT, injectmask);
1003 * This is something undocumented, based on my tests
1004 * Without writing 8 to this register, errors aren't injected. Not sure
1005 * why.
1007 pci_write_config_dword(pvt->pci_noncore,
1008 MC_CFG_CONTROL, 8);
1010 edac_dbg(0, "Error inject addr match 0x%016llx, ecc 0x%08x, inject 0x%08x\n",
1011 mask, pvt->inject.eccmask, injectmask);
1014 return count;
1017 static ssize_t i7core_inject_enable_show(struct device *dev,
1018 struct device_attribute *mattr,
1019 char *data)
1021 struct mem_ctl_info *mci = to_mci(dev);
1022 struct i7core_pvt *pvt = mci->pvt_info;
1023 u32 injectmask;
1025 if (!pvt->pci_ch[pvt->inject.channel][0])
1026 return 0;
1028 pci_read_config_dword(pvt->pci_ch[pvt->inject.channel][0],
1029 MC_CHANNEL_ERROR_INJECT, &injectmask);
1031 edac_dbg(0, "Inject error read: 0x%018x\n", injectmask);
1033 if (injectmask & 0x0c)
1034 pvt->inject.enable = 1;
1036 return sprintf(data, "%d\n", pvt->inject.enable);
1039 #define DECLARE_COUNTER(param) \
1040 static ssize_t i7core_show_counter_##param( \
1041 struct device *dev, \
1042 struct device_attribute *mattr, \
1043 char *data) \
1045 struct mem_ctl_info *mci = dev_get_drvdata(dev); \
1046 struct i7core_pvt *pvt = mci->pvt_info; \
1048 edac_dbg(1, "\n"); \
1049 if (!pvt->ce_count_available || (pvt->is_registered)) \
1050 return sprintf(data, "data unavailable\n"); \
1051 return sprintf(data, "%lu\n", \
1052 pvt->udimm_ce_count[param]); \
1055 #define ATTR_COUNTER(param) \
1056 static DEVICE_ATTR(udimm##param, S_IRUGO | S_IWUSR, \
1057 i7core_show_counter_##param, \
1058 NULL)
1060 DECLARE_COUNTER(0);
1061 DECLARE_COUNTER(1);
1062 DECLARE_COUNTER(2);
1064 ATTR_COUNTER(0);
1065 ATTR_COUNTER(1);
1066 ATTR_COUNTER(2);
1069 * inject_addrmatch device sysfs struct
1072 static struct attribute *i7core_addrmatch_attrs[] = {
1073 &dev_attr_channel.attr,
1074 &dev_attr_dimm.attr,
1075 &dev_attr_rank.attr,
1076 &dev_attr_bank.attr,
1077 &dev_attr_page.attr,
1078 &dev_attr_col.attr,
1079 NULL
1082 static struct attribute_group addrmatch_grp = {
1083 .attrs = i7core_addrmatch_attrs,
1086 static const struct attribute_group *addrmatch_groups[] = {
1087 &addrmatch_grp,
1088 NULL
1091 static void addrmatch_release(struct device *device)
1093 edac_dbg(1, "Releasing device %s\n", dev_name(device));
1094 kfree(device);
1097 static struct device_type addrmatch_type = {
1098 .groups = addrmatch_groups,
1099 .release = addrmatch_release,
1103 * all_channel_counts sysfs struct
1106 static struct attribute *i7core_udimm_counters_attrs[] = {
1107 &dev_attr_udimm0.attr,
1108 &dev_attr_udimm1.attr,
1109 &dev_attr_udimm2.attr,
1110 NULL
1113 static struct attribute_group all_channel_counts_grp = {
1114 .attrs = i7core_udimm_counters_attrs,
1117 static const struct attribute_group *all_channel_counts_groups[] = {
1118 &all_channel_counts_grp,
1119 NULL
1122 static void all_channel_counts_release(struct device *device)
1124 edac_dbg(1, "Releasing device %s\n", dev_name(device));
1125 kfree(device);
1128 static struct device_type all_channel_counts_type = {
1129 .groups = all_channel_counts_groups,
1130 .release = all_channel_counts_release,
1134 * inject sysfs attributes
1137 static DEVICE_ATTR(inject_section, S_IRUGO | S_IWUSR,
1138 i7core_inject_section_show, i7core_inject_section_store);
1140 static DEVICE_ATTR(inject_type, S_IRUGO | S_IWUSR,
1141 i7core_inject_type_show, i7core_inject_type_store);
1144 static DEVICE_ATTR(inject_eccmask, S_IRUGO | S_IWUSR,
1145 i7core_inject_eccmask_show, i7core_inject_eccmask_store);
1147 static DEVICE_ATTR(inject_enable, S_IRUGO | S_IWUSR,
1148 i7core_inject_enable_show, i7core_inject_enable_store);
1150 static struct attribute *i7core_dev_attrs[] = {
1151 &dev_attr_inject_section.attr,
1152 &dev_attr_inject_type.attr,
1153 &dev_attr_inject_eccmask.attr,
1154 &dev_attr_inject_enable.attr,
1155 NULL
1158 ATTRIBUTE_GROUPS(i7core_dev);
1160 static int i7core_create_sysfs_devices(struct mem_ctl_info *mci)
1162 struct i7core_pvt *pvt = mci->pvt_info;
1163 int rc;
1165 pvt->addrmatch_dev = kzalloc(sizeof(*pvt->addrmatch_dev), GFP_KERNEL);
1166 if (!pvt->addrmatch_dev)
1167 return -ENOMEM;
1169 pvt->addrmatch_dev->type = &addrmatch_type;
1170 pvt->addrmatch_dev->bus = mci->dev.bus;
1171 device_initialize(pvt->addrmatch_dev);
1172 pvt->addrmatch_dev->parent = &mci->dev;
1173 dev_set_name(pvt->addrmatch_dev, "inject_addrmatch");
1174 dev_set_drvdata(pvt->addrmatch_dev, mci);
1176 edac_dbg(1, "creating %s\n", dev_name(pvt->addrmatch_dev));
1178 rc = device_add(pvt->addrmatch_dev);
1179 if (rc < 0)
1180 goto err_put_addrmatch;
1182 if (!pvt->is_registered) {
1183 pvt->chancounts_dev = kzalloc(sizeof(*pvt->chancounts_dev),
1184 GFP_KERNEL);
1185 if (!pvt->chancounts_dev) {
1186 rc = -ENOMEM;
1187 goto err_del_addrmatch;
1190 pvt->chancounts_dev->type = &all_channel_counts_type;
1191 pvt->chancounts_dev->bus = mci->dev.bus;
1192 device_initialize(pvt->chancounts_dev);
1193 pvt->chancounts_dev->parent = &mci->dev;
1194 dev_set_name(pvt->chancounts_dev, "all_channel_counts");
1195 dev_set_drvdata(pvt->chancounts_dev, mci);
1197 edac_dbg(1, "creating %s\n", dev_name(pvt->chancounts_dev));
1199 rc = device_add(pvt->chancounts_dev);
1200 if (rc < 0)
1201 goto err_put_chancounts;
1203 return 0;
1205 err_put_chancounts:
1206 put_device(pvt->chancounts_dev);
1207 err_del_addrmatch:
1208 device_del(pvt->addrmatch_dev);
1209 err_put_addrmatch:
1210 put_device(pvt->addrmatch_dev);
1212 return rc;
1215 static void i7core_delete_sysfs_devices(struct mem_ctl_info *mci)
1217 struct i7core_pvt *pvt = mci->pvt_info;
1219 edac_dbg(1, "\n");
1221 if (!pvt->is_registered) {
1222 device_del(pvt->chancounts_dev);
1223 put_device(pvt->chancounts_dev);
1225 device_del(pvt->addrmatch_dev);
1226 put_device(pvt->addrmatch_dev);
1229 /****************************************************************************
1230 Device initialization routines: put/get, init/exit
1231 ****************************************************************************/
1234 * i7core_put_all_devices 'put' all the devices that we have
1235 * reserved via 'get'
1237 static void i7core_put_devices(struct i7core_dev *i7core_dev)
1239 int i;
1241 edac_dbg(0, "\n");
1242 for (i = 0; i < i7core_dev->n_devs; i++) {
1243 struct pci_dev *pdev = i7core_dev->pdev[i];
1244 if (!pdev)
1245 continue;
1246 edac_dbg(0, "Removing dev %02x:%02x.%d\n",
1247 pdev->bus->number,
1248 PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn));
1249 pci_dev_put(pdev);
1253 static void i7core_put_all_devices(void)
1255 struct i7core_dev *i7core_dev, *tmp;
1257 list_for_each_entry_safe(i7core_dev, tmp, &i7core_edac_list, list) {
1258 i7core_put_devices(i7core_dev);
1259 free_i7core_dev(i7core_dev);
1263 static void __init i7core_xeon_pci_fixup(const struct pci_id_table *table)
1265 struct pci_dev *pdev = NULL;
1266 int i;
1269 * On Xeon 55xx, the Intel Quick Path Arch Generic Non-core pci buses
1270 * aren't announced by acpi. So, we need to use a legacy scan probing
1271 * to detect them
1273 while (table && table->descr) {
1274 pdev = pci_get_device(PCI_VENDOR_ID_INTEL, table->descr[0].dev_id, NULL);
1275 if (unlikely(!pdev)) {
1276 for (i = 0; i < MAX_SOCKET_BUSES; i++)
1277 pcibios_scan_specific_bus(255-i);
1279 pci_dev_put(pdev);
1280 table++;
1284 static unsigned i7core_pci_lastbus(void)
1286 int last_bus = 0, bus;
1287 struct pci_bus *b = NULL;
1289 while ((b = pci_find_next_bus(b)) != NULL) {
1290 bus = b->number;
1291 edac_dbg(0, "Found bus %d\n", bus);
1292 if (bus > last_bus)
1293 last_bus = bus;
1296 edac_dbg(0, "Last bus %d\n", last_bus);
1298 return last_bus;
1302 * i7core_get_all_devices Find and perform 'get' operation on the MCH's
1303 * device/functions we want to reference for this driver
1305 * Need to 'get' device 16 func 1 and func 2
1307 static int i7core_get_onedevice(struct pci_dev **prev,
1308 const struct pci_id_table *table,
1309 const unsigned devno,
1310 const unsigned last_bus)
1312 struct i7core_dev *i7core_dev;
1313 const struct pci_id_descr *dev_descr = &table->descr[devno];
1315 struct pci_dev *pdev = NULL;
1316 u8 bus = 0;
1317 u8 socket = 0;
1319 pdev = pci_get_device(PCI_VENDOR_ID_INTEL,
1320 dev_descr->dev_id, *prev);
1323 * On Xeon 55xx, the Intel QuickPath Arch Generic Non-core regs
1324 * is at addr 8086:2c40, instead of 8086:2c41. So, we need
1325 * to probe for the alternate address in case of failure
1327 if (dev_descr->dev_id == PCI_DEVICE_ID_INTEL_I7_NONCORE && !pdev) {
1328 pci_dev_get(*prev); /* pci_get_device will put it */
1329 pdev = pci_get_device(PCI_VENDOR_ID_INTEL,
1330 PCI_DEVICE_ID_INTEL_I7_NONCORE_ALT, *prev);
1333 if (dev_descr->dev_id == PCI_DEVICE_ID_INTEL_LYNNFIELD_NONCORE &&
1334 !pdev) {
1335 pci_dev_get(*prev); /* pci_get_device will put it */
1336 pdev = pci_get_device(PCI_VENDOR_ID_INTEL,
1337 PCI_DEVICE_ID_INTEL_LYNNFIELD_NONCORE_ALT,
1338 *prev);
1341 if (!pdev) {
1342 if (*prev) {
1343 *prev = pdev;
1344 return 0;
1347 if (dev_descr->optional)
1348 return 0;
1350 if (devno == 0)
1351 return -ENODEV;
1353 i7core_printk(KERN_INFO,
1354 "Device not found: dev %02x.%d PCI ID %04x:%04x\n",
1355 dev_descr->dev, dev_descr->func,
1356 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
1358 /* End of list, leave */
1359 return -ENODEV;
1361 bus = pdev->bus->number;
1363 socket = last_bus - bus;
1365 i7core_dev = get_i7core_dev(socket);
1366 if (!i7core_dev) {
1367 i7core_dev = alloc_i7core_dev(socket, table);
1368 if (!i7core_dev) {
1369 pci_dev_put(pdev);
1370 return -ENOMEM;
1374 if (i7core_dev->pdev[devno]) {
1375 i7core_printk(KERN_ERR,
1376 "Duplicated device for "
1377 "dev %02x:%02x.%d PCI ID %04x:%04x\n",
1378 bus, dev_descr->dev, dev_descr->func,
1379 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
1380 pci_dev_put(pdev);
1381 return -ENODEV;
1384 i7core_dev->pdev[devno] = pdev;
1386 /* Sanity check */
1387 if (unlikely(PCI_SLOT(pdev->devfn) != dev_descr->dev ||
1388 PCI_FUNC(pdev->devfn) != dev_descr->func)) {
1389 i7core_printk(KERN_ERR,
1390 "Device PCI ID %04x:%04x "
1391 "has dev %02x:%02x.%d instead of dev %02x:%02x.%d\n",
1392 PCI_VENDOR_ID_INTEL, dev_descr->dev_id,
1393 bus, PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn),
1394 bus, dev_descr->dev, dev_descr->func);
1395 return -ENODEV;
1398 /* Be sure that the device is enabled */
1399 if (unlikely(pci_enable_device(pdev) < 0)) {
1400 i7core_printk(KERN_ERR,
1401 "Couldn't enable "
1402 "dev %02x:%02x.%d PCI ID %04x:%04x\n",
1403 bus, dev_descr->dev, dev_descr->func,
1404 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
1405 return -ENODEV;
1408 edac_dbg(0, "Detected socket %d dev %02x:%02x.%d PCI ID %04x:%04x\n",
1409 socket, bus, dev_descr->dev,
1410 dev_descr->func,
1411 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
1414 * As stated on drivers/pci/search.c, the reference count for
1415 * @from is always decremented if it is not %NULL. So, as we need
1416 * to get all devices up to null, we need to do a get for the device
1418 pci_dev_get(pdev);
1420 *prev = pdev;
1422 return 0;
1425 static int i7core_get_all_devices(void)
1427 int i, rc, last_bus;
1428 struct pci_dev *pdev = NULL;
1429 const struct pci_id_table *table = pci_dev_table;
1431 last_bus = i7core_pci_lastbus();
1433 while (table && table->descr) {
1434 for (i = 0; i < table->n_devs; i++) {
1435 pdev = NULL;
1436 do {
1437 rc = i7core_get_onedevice(&pdev, table, i,
1438 last_bus);
1439 if (rc < 0) {
1440 if (i == 0) {
1441 i = table->n_devs;
1442 break;
1444 i7core_put_all_devices();
1445 return -ENODEV;
1447 } while (pdev);
1449 table++;
1452 return 0;
1455 static int mci_bind_devs(struct mem_ctl_info *mci,
1456 struct i7core_dev *i7core_dev)
1458 struct i7core_pvt *pvt = mci->pvt_info;
1459 struct pci_dev *pdev;
1460 int i, func, slot;
1461 char *family;
1463 pvt->is_registered = false;
1464 pvt->enable_scrub = false;
1465 for (i = 0; i < i7core_dev->n_devs; i++) {
1466 pdev = i7core_dev->pdev[i];
1467 if (!pdev)
1468 continue;
1470 func = PCI_FUNC(pdev->devfn);
1471 slot = PCI_SLOT(pdev->devfn);
1472 if (slot == 3) {
1473 if (unlikely(func > MAX_MCR_FUNC))
1474 goto error;
1475 pvt->pci_mcr[func] = pdev;
1476 } else if (likely(slot >= 4 && slot < 4 + NUM_CHANS)) {
1477 if (unlikely(func > MAX_CHAN_FUNC))
1478 goto error;
1479 pvt->pci_ch[slot - 4][func] = pdev;
1480 } else if (!slot && !func) {
1481 pvt->pci_noncore = pdev;
1483 /* Detect the processor family */
1484 switch (pdev->device) {
1485 case PCI_DEVICE_ID_INTEL_I7_NONCORE:
1486 family = "Xeon 35xx/ i7core";
1487 pvt->enable_scrub = false;
1488 break;
1489 case PCI_DEVICE_ID_INTEL_LYNNFIELD_NONCORE_ALT:
1490 family = "i7-800/i5-700";
1491 pvt->enable_scrub = false;
1492 break;
1493 case PCI_DEVICE_ID_INTEL_LYNNFIELD_NONCORE:
1494 family = "Xeon 34xx";
1495 pvt->enable_scrub = false;
1496 break;
1497 case PCI_DEVICE_ID_INTEL_I7_NONCORE_ALT:
1498 family = "Xeon 55xx";
1499 pvt->enable_scrub = true;
1500 break;
1501 case PCI_DEVICE_ID_INTEL_LYNNFIELD_NONCORE_REV2:
1502 family = "Xeon 56xx / i7-900";
1503 pvt->enable_scrub = true;
1504 break;
1505 default:
1506 family = "unknown";
1507 pvt->enable_scrub = false;
1509 edac_dbg(0, "Detected a processor type %s\n", family);
1510 } else
1511 goto error;
1513 edac_dbg(0, "Associated fn %d.%d, dev = %p, socket %d\n",
1514 PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn),
1515 pdev, i7core_dev->socket);
1517 if (PCI_SLOT(pdev->devfn) == 3 &&
1518 PCI_FUNC(pdev->devfn) == 2)
1519 pvt->is_registered = true;
1522 return 0;
1524 error:
1525 i7core_printk(KERN_ERR, "Device %d, function %d "
1526 "is out of the expected range\n",
1527 slot, func);
1528 return -EINVAL;
1531 /****************************************************************************
1532 Error check routines
1533 ****************************************************************************/
1535 static void i7core_rdimm_update_ce_count(struct mem_ctl_info *mci,
1536 const int chan,
1537 const int new0,
1538 const int new1,
1539 const int new2)
1541 struct i7core_pvt *pvt = mci->pvt_info;
1542 int add0 = 0, add1 = 0, add2 = 0;
1543 /* Updates CE counters if it is not the first time here */
1544 if (pvt->ce_count_available) {
1545 /* Updates CE counters */
1547 add2 = new2 - pvt->rdimm_last_ce_count[chan][2];
1548 add1 = new1 - pvt->rdimm_last_ce_count[chan][1];
1549 add0 = new0 - pvt->rdimm_last_ce_count[chan][0];
1551 if (add2 < 0)
1552 add2 += 0x7fff;
1553 pvt->rdimm_ce_count[chan][2] += add2;
1555 if (add1 < 0)
1556 add1 += 0x7fff;
1557 pvt->rdimm_ce_count[chan][1] += add1;
1559 if (add0 < 0)
1560 add0 += 0x7fff;
1561 pvt->rdimm_ce_count[chan][0] += add0;
1562 } else
1563 pvt->ce_count_available = 1;
1565 /* Store the new values */
1566 pvt->rdimm_last_ce_count[chan][2] = new2;
1567 pvt->rdimm_last_ce_count[chan][1] = new1;
1568 pvt->rdimm_last_ce_count[chan][0] = new0;
1570 /*updated the edac core */
1571 if (add0 != 0)
1572 edac_mc_handle_error(HW_EVENT_ERR_CORRECTED, mci, add0,
1573 0, 0, 0,
1574 chan, 0, -1, "error", "");
1575 if (add1 != 0)
1576 edac_mc_handle_error(HW_EVENT_ERR_CORRECTED, mci, add1,
1577 0, 0, 0,
1578 chan, 1, -1, "error", "");
1579 if (add2 != 0)
1580 edac_mc_handle_error(HW_EVENT_ERR_CORRECTED, mci, add2,
1581 0, 0, 0,
1582 chan, 2, -1, "error", "");
1585 static void i7core_rdimm_check_mc_ecc_err(struct mem_ctl_info *mci)
1587 struct i7core_pvt *pvt = mci->pvt_info;
1588 u32 rcv[3][2];
1589 int i, new0, new1, new2;
1591 /*Read DEV 3: FUN 2: MC_COR_ECC_CNT regs directly*/
1592 pci_read_config_dword(pvt->pci_mcr[2], MC_COR_ECC_CNT_0,
1593 &rcv[0][0]);
1594 pci_read_config_dword(pvt->pci_mcr[2], MC_COR_ECC_CNT_1,
1595 &rcv[0][1]);
1596 pci_read_config_dword(pvt->pci_mcr[2], MC_COR_ECC_CNT_2,
1597 &rcv[1][0]);
1598 pci_read_config_dword(pvt->pci_mcr[2], MC_COR_ECC_CNT_3,
1599 &rcv[1][1]);
1600 pci_read_config_dword(pvt->pci_mcr[2], MC_COR_ECC_CNT_4,
1601 &rcv[2][0]);
1602 pci_read_config_dword(pvt->pci_mcr[2], MC_COR_ECC_CNT_5,
1603 &rcv[2][1]);
1604 for (i = 0 ; i < 3; i++) {
1605 edac_dbg(3, "MC_COR_ECC_CNT%d = 0x%x; MC_COR_ECC_CNT%d = 0x%x\n",
1606 (i * 2), rcv[i][0], (i * 2) + 1, rcv[i][1]);
1607 /*if the channel has 3 dimms*/
1608 if (pvt->channel[i].dimms > 2) {
1609 new0 = DIMM_BOT_COR_ERR(rcv[i][0]);
1610 new1 = DIMM_TOP_COR_ERR(rcv[i][0]);
1611 new2 = DIMM_BOT_COR_ERR(rcv[i][1]);
1612 } else {
1613 new0 = DIMM_TOP_COR_ERR(rcv[i][0]) +
1614 DIMM_BOT_COR_ERR(rcv[i][0]);
1615 new1 = DIMM_TOP_COR_ERR(rcv[i][1]) +
1616 DIMM_BOT_COR_ERR(rcv[i][1]);
1617 new2 = 0;
1620 i7core_rdimm_update_ce_count(mci, i, new0, new1, new2);
1624 /* This function is based on the device 3 function 4 registers as described on:
1625 * Intel Xeon Processor 5500 Series Datasheet Volume 2
1626 * http://www.intel.com/Assets/PDF/datasheet/321322.pdf
1627 * also available at:
1628 * http://www.arrownac.com/manufacturers/intel/s/nehalem/5500-datasheet-v2.pdf
1630 static void i7core_udimm_check_mc_ecc_err(struct mem_ctl_info *mci)
1632 struct i7core_pvt *pvt = mci->pvt_info;
1633 u32 rcv1, rcv0;
1634 int new0, new1, new2;
1636 if (!pvt->pci_mcr[4]) {
1637 edac_dbg(0, "MCR registers not found\n");
1638 return;
1641 /* Corrected test errors */
1642 pci_read_config_dword(pvt->pci_mcr[4], MC_TEST_ERR_RCV1, &rcv1);
1643 pci_read_config_dword(pvt->pci_mcr[4], MC_TEST_ERR_RCV0, &rcv0);
1645 /* Store the new values */
1646 new2 = DIMM2_COR_ERR(rcv1);
1647 new1 = DIMM1_COR_ERR(rcv0);
1648 new0 = DIMM0_COR_ERR(rcv0);
1650 /* Updates CE counters if it is not the first time here */
1651 if (pvt->ce_count_available) {
1652 /* Updates CE counters */
1653 int add0, add1, add2;
1655 add2 = new2 - pvt->udimm_last_ce_count[2];
1656 add1 = new1 - pvt->udimm_last_ce_count[1];
1657 add0 = new0 - pvt->udimm_last_ce_count[0];
1659 if (add2 < 0)
1660 add2 += 0x7fff;
1661 pvt->udimm_ce_count[2] += add2;
1663 if (add1 < 0)
1664 add1 += 0x7fff;
1665 pvt->udimm_ce_count[1] += add1;
1667 if (add0 < 0)
1668 add0 += 0x7fff;
1669 pvt->udimm_ce_count[0] += add0;
1671 if (add0 | add1 | add2)
1672 i7core_printk(KERN_ERR, "New Corrected error(s): "
1673 "dimm0: +%d, dimm1: +%d, dimm2 +%d\n",
1674 add0, add1, add2);
1675 } else
1676 pvt->ce_count_available = 1;
1678 /* Store the new values */
1679 pvt->udimm_last_ce_count[2] = new2;
1680 pvt->udimm_last_ce_count[1] = new1;
1681 pvt->udimm_last_ce_count[0] = new0;
1685 * According with tables E-11 and E-12 of chapter E.3.3 of Intel 64 and IA-32
1686 * Architectures Software Developer’s Manual Volume 3B.
1687 * Nehalem are defined as family 0x06, model 0x1a
1689 * The MCA registers used here are the following ones:
1690 * struct mce field MCA Register
1691 * m->status MSR_IA32_MC8_STATUS
1692 * m->addr MSR_IA32_MC8_ADDR
1693 * m->misc MSR_IA32_MC8_MISC
1694 * In the case of Nehalem, the error information is masked at .status and .misc
1695 * fields
1697 static void i7core_mce_output_error(struct mem_ctl_info *mci,
1698 const struct mce *m)
1700 struct i7core_pvt *pvt = mci->pvt_info;
1701 char *optype, *err;
1702 enum hw_event_mc_err_type tp_event;
1703 unsigned long error = m->status & 0x1ff0000l;
1704 bool uncorrected_error = m->mcgstatus & 1ll << 61;
1705 bool ripv = m->mcgstatus & 1;
1706 u32 optypenum = (m->status >> 4) & 0x07;
1707 u32 core_err_cnt = (m->status >> 38) & 0x7fff;
1708 u32 dimm = (m->misc >> 16) & 0x3;
1709 u32 channel = (m->misc >> 18) & 0x3;
1710 u32 syndrome = m->misc >> 32;
1711 u32 errnum = find_first_bit(&error, 32);
1713 if (uncorrected_error) {
1714 core_err_cnt = 1;
1715 if (ripv)
1716 tp_event = HW_EVENT_ERR_FATAL;
1717 else
1718 tp_event = HW_EVENT_ERR_UNCORRECTED;
1719 } else {
1720 tp_event = HW_EVENT_ERR_CORRECTED;
1723 switch (optypenum) {
1724 case 0:
1725 optype = "generic undef request";
1726 break;
1727 case 1:
1728 optype = "read error";
1729 break;
1730 case 2:
1731 optype = "write error";
1732 break;
1733 case 3:
1734 optype = "addr/cmd error";
1735 break;
1736 case 4:
1737 optype = "scrubbing error";
1738 break;
1739 default:
1740 optype = "reserved";
1741 break;
1744 switch (errnum) {
1745 case 16:
1746 err = "read ECC error";
1747 break;
1748 case 17:
1749 err = "RAS ECC error";
1750 break;
1751 case 18:
1752 err = "write parity error";
1753 break;
1754 case 19:
1755 err = "redundacy loss";
1756 break;
1757 case 20:
1758 err = "reserved";
1759 break;
1760 case 21:
1761 err = "memory range error";
1762 break;
1763 case 22:
1764 err = "RTID out of range";
1765 break;
1766 case 23:
1767 err = "address parity error";
1768 break;
1769 case 24:
1770 err = "byte enable parity error";
1771 break;
1772 default:
1773 err = "unknown";
1777 * Call the helper to output message
1778 * FIXME: what to do if core_err_cnt > 1? Currently, it generates
1779 * only one event
1781 if (uncorrected_error || !pvt->is_registered)
1782 edac_mc_handle_error(tp_event, mci, core_err_cnt,
1783 m->addr >> PAGE_SHIFT,
1784 m->addr & ~PAGE_MASK,
1785 syndrome,
1786 channel, dimm, -1,
1787 err, optype);
1791 * i7core_check_error Retrieve and process errors reported by the
1792 * hardware. Called by the Core module.
1794 static void i7core_check_error(struct mem_ctl_info *mci, struct mce *m)
1796 struct i7core_pvt *pvt = mci->pvt_info;
1798 i7core_mce_output_error(mci, m);
1801 * Now, let's increment CE error counts
1803 if (!pvt->is_registered)
1804 i7core_udimm_check_mc_ecc_err(mci);
1805 else
1806 i7core_rdimm_check_mc_ecc_err(mci);
1810 * Check that logging is enabled and that this is the right type
1811 * of error for us to handle.
1813 static int i7core_mce_check_error(struct notifier_block *nb, unsigned long val,
1814 void *data)
1816 struct mce *mce = (struct mce *)data;
1817 struct i7core_dev *i7_dev;
1818 struct mem_ctl_info *mci;
1819 struct i7core_pvt *pvt;
1821 i7_dev = get_i7core_dev(mce->socketid);
1822 if (!i7_dev)
1823 return NOTIFY_DONE;
1825 mci = i7_dev->mci;
1826 pvt = mci->pvt_info;
1829 * Just let mcelog handle it if the error is
1830 * outside the memory controller
1832 if (((mce->status & 0xffff) >> 7) != 1)
1833 return NOTIFY_DONE;
1835 /* Bank 8 registers are the only ones that we know how to handle */
1836 if (mce->bank != 8)
1837 return NOTIFY_DONE;
1839 i7core_check_error(mci, mce);
1841 /* Advise mcelog that the errors were handled */
1842 return NOTIFY_STOP;
1845 static struct notifier_block i7_mce_dec = {
1846 .notifier_call = i7core_mce_check_error,
1849 struct memdev_dmi_entry {
1850 u8 type;
1851 u8 length;
1852 u16 handle;
1853 u16 phys_mem_array_handle;
1854 u16 mem_err_info_handle;
1855 u16 total_width;
1856 u16 data_width;
1857 u16 size;
1858 u8 form;
1859 u8 device_set;
1860 u8 device_locator;
1861 u8 bank_locator;
1862 u8 memory_type;
1863 u16 type_detail;
1864 u16 speed;
1865 u8 manufacturer;
1866 u8 serial_number;
1867 u8 asset_tag;
1868 u8 part_number;
1869 u8 attributes;
1870 u32 extended_size;
1871 u16 conf_mem_clk_speed;
1872 } __attribute__((__packed__));
1876 * Decode the DRAM Clock Frequency, be paranoid, make sure that all
1877 * memory devices show the same speed, and if they don't then consider
1878 * all speeds to be invalid.
1880 static void decode_dclk(const struct dmi_header *dh, void *_dclk_freq)
1882 int *dclk_freq = _dclk_freq;
1883 u16 dmi_mem_clk_speed;
1885 if (*dclk_freq == -1)
1886 return;
1888 if (dh->type == DMI_ENTRY_MEM_DEVICE) {
1889 struct memdev_dmi_entry *memdev_dmi_entry =
1890 (struct memdev_dmi_entry *)dh;
1891 unsigned long conf_mem_clk_speed_offset =
1892 (unsigned long)&memdev_dmi_entry->conf_mem_clk_speed -
1893 (unsigned long)&memdev_dmi_entry->type;
1894 unsigned long speed_offset =
1895 (unsigned long)&memdev_dmi_entry->speed -
1896 (unsigned long)&memdev_dmi_entry->type;
1898 /* Check that a DIMM is present */
1899 if (memdev_dmi_entry->size == 0)
1900 return;
1903 * Pick the configured speed if it's available, otherwise
1904 * pick the DIMM speed, or we don't have a speed.
1906 if (memdev_dmi_entry->length > conf_mem_clk_speed_offset) {
1907 dmi_mem_clk_speed =
1908 memdev_dmi_entry->conf_mem_clk_speed;
1909 } else if (memdev_dmi_entry->length > speed_offset) {
1910 dmi_mem_clk_speed = memdev_dmi_entry->speed;
1911 } else {
1912 *dclk_freq = -1;
1913 return;
1916 if (*dclk_freq == 0) {
1917 /* First pass, speed was 0 */
1918 if (dmi_mem_clk_speed > 0) {
1919 /* Set speed if a valid speed is read */
1920 *dclk_freq = dmi_mem_clk_speed;
1921 } else {
1922 /* Otherwise we don't have a valid speed */
1923 *dclk_freq = -1;
1925 } else if (*dclk_freq > 0 &&
1926 *dclk_freq != dmi_mem_clk_speed) {
1928 * If we have a speed, check that all DIMMS are the same
1929 * speed, otherwise set the speed as invalid.
1931 *dclk_freq = -1;
1937 * The default DCLK frequency is used as a fallback if we
1938 * fail to find anything reliable in the DMI. The value
1939 * is taken straight from the datasheet.
1941 #define DEFAULT_DCLK_FREQ 800
1943 static int get_dclk_freq(void)
1945 int dclk_freq = 0;
1947 dmi_walk(decode_dclk, (void *)&dclk_freq);
1949 if (dclk_freq < 1)
1950 return DEFAULT_DCLK_FREQ;
1952 return dclk_freq;
1956 * set_sdram_scrub_rate This routine sets byte/sec bandwidth scrub rate
1957 * to hardware according to SCRUBINTERVAL formula
1958 * found in datasheet.
1960 static int set_sdram_scrub_rate(struct mem_ctl_info *mci, u32 new_bw)
1962 struct i7core_pvt *pvt = mci->pvt_info;
1963 struct pci_dev *pdev;
1964 u32 dw_scrub;
1965 u32 dw_ssr;
1967 /* Get data from the MC register, function 2 */
1968 pdev = pvt->pci_mcr[2];
1969 if (!pdev)
1970 return -ENODEV;
1972 pci_read_config_dword(pdev, MC_SCRUB_CONTROL, &dw_scrub);
1974 if (new_bw == 0) {
1975 /* Prepare to disable petrol scrub */
1976 dw_scrub &= ~STARTSCRUB;
1977 /* Stop the patrol scrub engine */
1978 write_and_test(pdev, MC_SCRUB_CONTROL,
1979 dw_scrub & ~SCRUBINTERVAL_MASK);
1981 /* Get current status of scrub rate and set bit to disable */
1982 pci_read_config_dword(pdev, MC_SSRCONTROL, &dw_ssr);
1983 dw_ssr &= ~SSR_MODE_MASK;
1984 dw_ssr |= SSR_MODE_DISABLE;
1985 } else {
1986 const int cache_line_size = 64;
1987 const u32 freq_dclk_mhz = pvt->dclk_freq;
1988 unsigned long long scrub_interval;
1990 * Translate the desired scrub rate to a register value and
1991 * program the corresponding register value.
1993 scrub_interval = (unsigned long long)freq_dclk_mhz *
1994 cache_line_size * 1000000;
1995 do_div(scrub_interval, new_bw);
1997 if (!scrub_interval || scrub_interval > SCRUBINTERVAL_MASK)
1998 return -EINVAL;
2000 dw_scrub = SCRUBINTERVAL_MASK & scrub_interval;
2002 /* Start the patrol scrub engine */
2003 pci_write_config_dword(pdev, MC_SCRUB_CONTROL,
2004 STARTSCRUB | dw_scrub);
2006 /* Get current status of scrub rate and set bit to enable */
2007 pci_read_config_dword(pdev, MC_SSRCONTROL, &dw_ssr);
2008 dw_ssr &= ~SSR_MODE_MASK;
2009 dw_ssr |= SSR_MODE_ENABLE;
2011 /* Disable or enable scrubbing */
2012 pci_write_config_dword(pdev, MC_SSRCONTROL, dw_ssr);
2014 return new_bw;
2018 * get_sdram_scrub_rate This routine convert current scrub rate value
2019 * into byte/sec bandwidth according to
2020 * SCRUBINTERVAL formula found in datasheet.
2022 static int get_sdram_scrub_rate(struct mem_ctl_info *mci)
2024 struct i7core_pvt *pvt = mci->pvt_info;
2025 struct pci_dev *pdev;
2026 const u32 cache_line_size = 64;
2027 const u32 freq_dclk_mhz = pvt->dclk_freq;
2028 unsigned long long scrub_rate;
2029 u32 scrubval;
2031 /* Get data from the MC register, function 2 */
2032 pdev = pvt->pci_mcr[2];
2033 if (!pdev)
2034 return -ENODEV;
2036 /* Get current scrub control data */
2037 pci_read_config_dword(pdev, MC_SCRUB_CONTROL, &scrubval);
2039 /* Mask highest 8-bits to 0 */
2040 scrubval &= SCRUBINTERVAL_MASK;
2041 if (!scrubval)
2042 return 0;
2044 /* Calculate scrub rate value into byte/sec bandwidth */
2045 scrub_rate = (unsigned long long)freq_dclk_mhz *
2046 1000000 * cache_line_size;
2047 do_div(scrub_rate, scrubval);
2048 return (int)scrub_rate;
2051 static void enable_sdram_scrub_setting(struct mem_ctl_info *mci)
2053 struct i7core_pvt *pvt = mci->pvt_info;
2054 u32 pci_lock;
2056 /* Unlock writes to pci registers */
2057 pci_read_config_dword(pvt->pci_noncore, MC_CFG_CONTROL, &pci_lock);
2058 pci_lock &= ~0x3;
2059 pci_write_config_dword(pvt->pci_noncore, MC_CFG_CONTROL,
2060 pci_lock | MC_CFG_UNLOCK);
2062 mci->set_sdram_scrub_rate = set_sdram_scrub_rate;
2063 mci->get_sdram_scrub_rate = get_sdram_scrub_rate;
2066 static void disable_sdram_scrub_setting(struct mem_ctl_info *mci)
2068 struct i7core_pvt *pvt = mci->pvt_info;
2069 u32 pci_lock;
2071 /* Lock writes to pci registers */
2072 pci_read_config_dword(pvt->pci_noncore, MC_CFG_CONTROL, &pci_lock);
2073 pci_lock &= ~0x3;
2074 pci_write_config_dword(pvt->pci_noncore, MC_CFG_CONTROL,
2075 pci_lock | MC_CFG_LOCK);
2078 static void i7core_pci_ctl_create(struct i7core_pvt *pvt)
2080 pvt->i7core_pci = edac_pci_create_generic_ctl(
2081 &pvt->i7core_dev->pdev[0]->dev,
2082 EDAC_MOD_STR);
2083 if (unlikely(!pvt->i7core_pci))
2084 i7core_printk(KERN_WARNING,
2085 "Unable to setup PCI error report via EDAC\n");
2088 static void i7core_pci_ctl_release(struct i7core_pvt *pvt)
2090 if (likely(pvt->i7core_pci))
2091 edac_pci_release_generic_ctl(pvt->i7core_pci);
2092 else
2093 i7core_printk(KERN_ERR,
2094 "Couldn't find mem_ctl_info for socket %d\n",
2095 pvt->i7core_dev->socket);
2096 pvt->i7core_pci = NULL;
2099 static void i7core_unregister_mci(struct i7core_dev *i7core_dev)
2101 struct mem_ctl_info *mci = i7core_dev->mci;
2102 struct i7core_pvt *pvt;
2104 if (unlikely(!mci || !mci->pvt_info)) {
2105 edac_dbg(0, "MC: dev = %p\n", &i7core_dev->pdev[0]->dev);
2107 i7core_printk(KERN_ERR, "Couldn't find mci handler\n");
2108 return;
2111 pvt = mci->pvt_info;
2113 edac_dbg(0, "MC: mci = %p, dev = %p\n", mci, &i7core_dev->pdev[0]->dev);
2115 /* Disable scrubrate setting */
2116 if (pvt->enable_scrub)
2117 disable_sdram_scrub_setting(mci);
2119 /* Disable EDAC polling */
2120 i7core_pci_ctl_release(pvt);
2122 /* Remove MC sysfs nodes */
2123 i7core_delete_sysfs_devices(mci);
2124 edac_mc_del_mc(mci->pdev);
2126 edac_dbg(1, "%s: free mci struct\n", mci->ctl_name);
2127 kfree(mci->ctl_name);
2128 edac_mc_free(mci);
2129 i7core_dev->mci = NULL;
2132 static int i7core_register_mci(struct i7core_dev *i7core_dev)
2134 struct mem_ctl_info *mci;
2135 struct i7core_pvt *pvt;
2136 int rc;
2137 struct edac_mc_layer layers[2];
2139 /* allocate a new MC control structure */
2141 layers[0].type = EDAC_MC_LAYER_CHANNEL;
2142 layers[0].size = NUM_CHANS;
2143 layers[0].is_virt_csrow = false;
2144 layers[1].type = EDAC_MC_LAYER_SLOT;
2145 layers[1].size = MAX_DIMMS;
2146 layers[1].is_virt_csrow = true;
2147 mci = edac_mc_alloc(i7core_dev->socket, ARRAY_SIZE(layers), layers,
2148 sizeof(*pvt));
2149 if (unlikely(!mci))
2150 return -ENOMEM;
2152 edac_dbg(0, "MC: mci = %p, dev = %p\n", mci, &i7core_dev->pdev[0]->dev);
2154 pvt = mci->pvt_info;
2155 memset(pvt, 0, sizeof(*pvt));
2157 /* Associates i7core_dev and mci for future usage */
2158 pvt->i7core_dev = i7core_dev;
2159 i7core_dev->mci = mci;
2162 * FIXME: how to handle RDDR3 at MCI level? It is possible to have
2163 * Mixed RDDR3/UDDR3 with Nehalem, provided that they are on different
2164 * memory channels
2166 mci->mtype_cap = MEM_FLAG_DDR3;
2167 mci->edac_ctl_cap = EDAC_FLAG_NONE;
2168 mci->edac_cap = EDAC_FLAG_NONE;
2169 mci->mod_name = "i7core_edac.c";
2170 mci->mod_ver = I7CORE_REVISION;
2171 mci->ctl_name = kasprintf(GFP_KERNEL, "i7 core #%d",
2172 i7core_dev->socket);
2173 mci->dev_name = pci_name(i7core_dev->pdev[0]);
2174 mci->ctl_page_to_phys = NULL;
2176 /* Store pci devices at mci for faster access */
2177 rc = mci_bind_devs(mci, i7core_dev);
2178 if (unlikely(rc < 0))
2179 goto fail0;
2182 /* Get dimm basic config */
2183 get_dimm_config(mci);
2184 /* record ptr to the generic device */
2185 mci->pdev = &i7core_dev->pdev[0]->dev;
2187 /* Enable scrubrate setting */
2188 if (pvt->enable_scrub)
2189 enable_sdram_scrub_setting(mci);
2191 /* add this new MC control structure to EDAC's list of MCs */
2192 if (unlikely(edac_mc_add_mc_with_groups(mci, i7core_dev_groups))) {
2193 edac_dbg(0, "MC: failed edac_mc_add_mc()\n");
2194 /* FIXME: perhaps some code should go here that disables error
2195 * reporting if we just enabled it
2198 rc = -EINVAL;
2199 goto fail0;
2201 if (i7core_create_sysfs_devices(mci)) {
2202 edac_dbg(0, "MC: failed to create sysfs nodes\n");
2203 edac_mc_del_mc(mci->pdev);
2204 rc = -EINVAL;
2205 goto fail0;
2208 /* Default error mask is any memory */
2209 pvt->inject.channel = 0;
2210 pvt->inject.dimm = -1;
2211 pvt->inject.rank = -1;
2212 pvt->inject.bank = -1;
2213 pvt->inject.page = -1;
2214 pvt->inject.col = -1;
2216 /* allocating generic PCI control info */
2217 i7core_pci_ctl_create(pvt);
2219 /* DCLK for scrub rate setting */
2220 pvt->dclk_freq = get_dclk_freq();
2222 return 0;
2224 fail0:
2225 kfree(mci->ctl_name);
2226 edac_mc_free(mci);
2227 i7core_dev->mci = NULL;
2228 return rc;
2232 * i7core_probe Probe for ONE instance of device to see if it is
2233 * present.
2234 * return:
2235 * 0 for FOUND a device
2236 * < 0 for error code
2239 static int i7core_probe(struct pci_dev *pdev, const struct pci_device_id *id)
2241 int rc, count = 0;
2242 struct i7core_dev *i7core_dev;
2244 /* get the pci devices we want to reserve for our use */
2245 mutex_lock(&i7core_edac_lock);
2248 * All memory controllers are allocated at the first pass.
2250 if (unlikely(probed >= 1)) {
2251 mutex_unlock(&i7core_edac_lock);
2252 return -ENODEV;
2254 probed++;
2256 rc = i7core_get_all_devices();
2257 if (unlikely(rc < 0))
2258 goto fail0;
2260 list_for_each_entry(i7core_dev, &i7core_edac_list, list) {
2261 count++;
2262 rc = i7core_register_mci(i7core_dev);
2263 if (unlikely(rc < 0))
2264 goto fail1;
2268 * Nehalem-EX uses a different memory controller. However, as the
2269 * memory controller is not visible on some Nehalem/Nehalem-EP, we
2270 * need to indirectly probe via a X58 PCI device. The same devices
2271 * are found on (some) Nehalem-EX. So, on those machines, the
2272 * probe routine needs to return -ENODEV, as the actual Memory
2273 * Controller registers won't be detected.
2275 if (!count) {
2276 rc = -ENODEV;
2277 goto fail1;
2280 i7core_printk(KERN_INFO,
2281 "Driver loaded, %d memory controller(s) found.\n",
2282 count);
2284 mutex_unlock(&i7core_edac_lock);
2285 return 0;
2287 fail1:
2288 list_for_each_entry(i7core_dev, &i7core_edac_list, list)
2289 i7core_unregister_mci(i7core_dev);
2291 i7core_put_all_devices();
2292 fail0:
2293 mutex_unlock(&i7core_edac_lock);
2294 return rc;
2298 * i7core_remove destructor for one instance of device
2301 static void i7core_remove(struct pci_dev *pdev)
2303 struct i7core_dev *i7core_dev;
2305 edac_dbg(0, "\n");
2308 * we have a trouble here: pdev value for removal will be wrong, since
2309 * it will point to the X58 register used to detect that the machine
2310 * is a Nehalem or upper design. However, due to the way several PCI
2311 * devices are grouped together to provide MC functionality, we need
2312 * to use a different method for releasing the devices
2315 mutex_lock(&i7core_edac_lock);
2317 if (unlikely(!probed)) {
2318 mutex_unlock(&i7core_edac_lock);
2319 return;
2322 list_for_each_entry(i7core_dev, &i7core_edac_list, list)
2323 i7core_unregister_mci(i7core_dev);
2325 /* Release PCI resources */
2326 i7core_put_all_devices();
2328 probed--;
2330 mutex_unlock(&i7core_edac_lock);
2333 MODULE_DEVICE_TABLE(pci, i7core_pci_tbl);
2336 * i7core_driver pci_driver structure for this module
2339 static struct pci_driver i7core_driver = {
2340 .name = "i7core_edac",
2341 .probe = i7core_probe,
2342 .remove = i7core_remove,
2343 .id_table = i7core_pci_tbl,
2347 * i7core_init Module entry function
2348 * Try to initialize this module for its devices
2350 static int __init i7core_init(void)
2352 int pci_rc;
2354 edac_dbg(2, "\n");
2356 /* Ensure that the OPSTATE is set correctly for POLL or NMI */
2357 opstate_init();
2359 if (use_pci_fixup)
2360 i7core_xeon_pci_fixup(pci_dev_table);
2362 pci_rc = pci_register_driver(&i7core_driver);
2364 if (pci_rc >= 0) {
2365 mce_register_decode_chain(&i7_mce_dec);
2366 return 0;
2369 i7core_printk(KERN_ERR, "Failed to register device with error %d.\n",
2370 pci_rc);
2372 return pci_rc;
2376 * i7core_exit() Module exit function
2377 * Unregister the driver
2379 static void __exit i7core_exit(void)
2381 edac_dbg(2, "\n");
2382 pci_unregister_driver(&i7core_driver);
2383 mce_unregister_decode_chain(&i7_mce_dec);
2386 module_init(i7core_init);
2387 module_exit(i7core_exit);
2389 MODULE_LICENSE("GPL");
2390 MODULE_AUTHOR("Mauro Carvalho Chehab");
2391 MODULE_AUTHOR("Red Hat Inc. (http://www.redhat.com)");
2392 MODULE_DESCRIPTION("MC Driver for Intel i7 Core memory controllers - "
2393 I7CORE_REVISION);
2395 module_param(edac_op_state, int, 0444);
2396 MODULE_PARM_DESC(edac_op_state, "EDAC Error Reporting state: 0=Poll,1=NMI");