PM / sleep: Asynchronous threads for suspend_noirq
[linux/fpc-iii.git] / drivers / edac / i7core_edac.c
blob87533ca7752e0105683210913f83731a9b988aab
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 <mchehab@redhat.com>
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 /* Fifo double buffers */
275 struct mce mce_entry[MCE_LOG_LEN];
276 struct mce mce_outentry[MCE_LOG_LEN];
278 /* Fifo in/out counters */
279 unsigned mce_in, mce_out;
281 /* Count indicator to show errors not got */
282 unsigned mce_overrun;
284 /* DCLK Frequency used for computing scrub rate */
285 int dclk_freq;
287 /* Struct to control EDAC polling */
288 struct edac_pci_ctl_info *i7core_pci;
291 #define PCI_DESCR(device, function, device_id) \
292 .dev = (device), \
293 .func = (function), \
294 .dev_id = (device_id)
296 static const struct pci_id_descr pci_dev_descr_i7core_nehalem[] = {
297 /* Memory controller */
298 { PCI_DESCR(3, 0, PCI_DEVICE_ID_INTEL_I7_MCR) },
299 { PCI_DESCR(3, 1, PCI_DEVICE_ID_INTEL_I7_MC_TAD) },
300 /* Exists only for RDIMM */
301 { PCI_DESCR(3, 2, PCI_DEVICE_ID_INTEL_I7_MC_RAS), .optional = 1 },
302 { PCI_DESCR(3, 4, PCI_DEVICE_ID_INTEL_I7_MC_TEST) },
304 /* Channel 0 */
305 { PCI_DESCR(4, 0, PCI_DEVICE_ID_INTEL_I7_MC_CH0_CTRL) },
306 { PCI_DESCR(4, 1, PCI_DEVICE_ID_INTEL_I7_MC_CH0_ADDR) },
307 { PCI_DESCR(4, 2, PCI_DEVICE_ID_INTEL_I7_MC_CH0_RANK) },
308 { PCI_DESCR(4, 3, PCI_DEVICE_ID_INTEL_I7_MC_CH0_TC) },
310 /* Channel 1 */
311 { PCI_DESCR(5, 0, PCI_DEVICE_ID_INTEL_I7_MC_CH1_CTRL) },
312 { PCI_DESCR(5, 1, PCI_DEVICE_ID_INTEL_I7_MC_CH1_ADDR) },
313 { PCI_DESCR(5, 2, PCI_DEVICE_ID_INTEL_I7_MC_CH1_RANK) },
314 { PCI_DESCR(5, 3, PCI_DEVICE_ID_INTEL_I7_MC_CH1_TC) },
316 /* Channel 2 */
317 { PCI_DESCR(6, 0, PCI_DEVICE_ID_INTEL_I7_MC_CH2_CTRL) },
318 { PCI_DESCR(6, 1, PCI_DEVICE_ID_INTEL_I7_MC_CH2_ADDR) },
319 { PCI_DESCR(6, 2, PCI_DEVICE_ID_INTEL_I7_MC_CH2_RANK) },
320 { PCI_DESCR(6, 3, PCI_DEVICE_ID_INTEL_I7_MC_CH2_TC) },
322 /* Generic Non-core registers */
324 * This is the PCI device on i7core and on Xeon 35xx (8086:2c41)
325 * On Xeon 55xx, however, it has a different id (8086:2c40). So,
326 * the probing code needs to test for the other address in case of
327 * failure of this one
329 { PCI_DESCR(0, 0, PCI_DEVICE_ID_INTEL_I7_NONCORE) },
333 static const struct pci_id_descr pci_dev_descr_lynnfield[] = {
334 { PCI_DESCR( 3, 0, PCI_DEVICE_ID_INTEL_LYNNFIELD_MCR) },
335 { PCI_DESCR( 3, 1, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_TAD) },
336 { PCI_DESCR( 3, 4, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_TEST) },
338 { PCI_DESCR( 4, 0, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH0_CTRL) },
339 { PCI_DESCR( 4, 1, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH0_ADDR) },
340 { PCI_DESCR( 4, 2, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH0_RANK) },
341 { PCI_DESCR( 4, 3, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH0_TC) },
343 { PCI_DESCR( 5, 0, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH1_CTRL) },
344 { PCI_DESCR( 5, 1, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH1_ADDR) },
345 { PCI_DESCR( 5, 2, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH1_RANK) },
346 { PCI_DESCR( 5, 3, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH1_TC) },
349 * This is the PCI device has an alternate address on some
350 * processors like Core i7 860
352 { PCI_DESCR( 0, 0, PCI_DEVICE_ID_INTEL_LYNNFIELD_NONCORE) },
355 static const struct pci_id_descr pci_dev_descr_i7core_westmere[] = {
356 /* Memory controller */
357 { PCI_DESCR(3, 0, PCI_DEVICE_ID_INTEL_LYNNFIELD_MCR_REV2) },
358 { PCI_DESCR(3, 1, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_TAD_REV2) },
359 /* Exists only for RDIMM */
360 { PCI_DESCR(3, 2, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_RAS_REV2), .optional = 1 },
361 { PCI_DESCR(3, 4, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_TEST_REV2) },
363 /* Channel 0 */
364 { PCI_DESCR(4, 0, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH0_CTRL_REV2) },
365 { PCI_DESCR(4, 1, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH0_ADDR_REV2) },
366 { PCI_DESCR(4, 2, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH0_RANK_REV2) },
367 { PCI_DESCR(4, 3, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH0_TC_REV2) },
369 /* Channel 1 */
370 { PCI_DESCR(5, 0, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH1_CTRL_REV2) },
371 { PCI_DESCR(5, 1, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH1_ADDR_REV2) },
372 { PCI_DESCR(5, 2, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH1_RANK_REV2) },
373 { PCI_DESCR(5, 3, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH1_TC_REV2) },
375 /* Channel 2 */
376 { PCI_DESCR(6, 0, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH2_CTRL_REV2) },
377 { PCI_DESCR(6, 1, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH2_ADDR_REV2) },
378 { PCI_DESCR(6, 2, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH2_RANK_REV2) },
379 { PCI_DESCR(6, 3, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH2_TC_REV2) },
381 /* Generic Non-core registers */
382 { PCI_DESCR(0, 0, PCI_DEVICE_ID_INTEL_LYNNFIELD_NONCORE_REV2) },
386 #define PCI_ID_TABLE_ENTRY(A) { .descr=A, .n_devs = ARRAY_SIZE(A) }
387 static const struct pci_id_table pci_dev_table[] = {
388 PCI_ID_TABLE_ENTRY(pci_dev_descr_i7core_nehalem),
389 PCI_ID_TABLE_ENTRY(pci_dev_descr_lynnfield),
390 PCI_ID_TABLE_ENTRY(pci_dev_descr_i7core_westmere),
391 {0,} /* 0 terminated list. */
395 * pci_device_id table for which devices we are looking for
397 static const struct pci_device_id i7core_pci_tbl[] = {
398 {PCI_DEVICE(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_X58_HUB_MGMT)},
399 {PCI_DEVICE(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_LYNNFIELD_QPI_LINK0)},
400 {0,} /* 0 terminated list. */
403 /****************************************************************************
404 Ancillary status routines
405 ****************************************************************************/
407 /* MC_CONTROL bits */
408 #define CH_ACTIVE(pvt, ch) ((pvt)->info.mc_control & (1 << (8 + ch)))
409 #define ECCx8(pvt) ((pvt)->info.mc_control & (1 << 1))
411 /* MC_STATUS bits */
412 #define ECC_ENABLED(pvt) ((pvt)->info.mc_status & (1 << 4))
413 #define CH_DISABLED(pvt, ch) ((pvt)->info.mc_status & (1 << ch))
415 /* MC_MAX_DOD read functions */
416 static inline int numdimms(u32 dimms)
418 return (dimms & 0x3) + 1;
421 static inline int numrank(u32 rank)
423 static const int ranks[] = { 1, 2, 4, -EINVAL };
425 return ranks[rank & 0x3];
428 static inline int numbank(u32 bank)
430 static const int banks[] = { 4, 8, 16, -EINVAL };
432 return banks[bank & 0x3];
435 static inline int numrow(u32 row)
437 static const int rows[] = {
438 1 << 12, 1 << 13, 1 << 14, 1 << 15,
439 1 << 16, -EINVAL, -EINVAL, -EINVAL,
442 return rows[row & 0x7];
445 static inline int numcol(u32 col)
447 static const int cols[] = {
448 1 << 10, 1 << 11, 1 << 12, -EINVAL,
450 return cols[col & 0x3];
453 static struct i7core_dev *get_i7core_dev(u8 socket)
455 struct i7core_dev *i7core_dev;
457 list_for_each_entry(i7core_dev, &i7core_edac_list, list) {
458 if (i7core_dev->socket == socket)
459 return i7core_dev;
462 return NULL;
465 static struct i7core_dev *alloc_i7core_dev(u8 socket,
466 const struct pci_id_table *table)
468 struct i7core_dev *i7core_dev;
470 i7core_dev = kzalloc(sizeof(*i7core_dev), GFP_KERNEL);
471 if (!i7core_dev)
472 return NULL;
474 i7core_dev->pdev = kzalloc(sizeof(*i7core_dev->pdev) * table->n_devs,
475 GFP_KERNEL);
476 if (!i7core_dev->pdev) {
477 kfree(i7core_dev);
478 return NULL;
481 i7core_dev->socket = socket;
482 i7core_dev->n_devs = table->n_devs;
483 list_add_tail(&i7core_dev->list, &i7core_edac_list);
485 return i7core_dev;
488 static void free_i7core_dev(struct i7core_dev *i7core_dev)
490 list_del(&i7core_dev->list);
491 kfree(i7core_dev->pdev);
492 kfree(i7core_dev);
495 /****************************************************************************
496 Memory check routines
497 ****************************************************************************/
499 static int get_dimm_config(struct mem_ctl_info *mci)
501 struct i7core_pvt *pvt = mci->pvt_info;
502 struct pci_dev *pdev;
503 int i, j;
504 enum edac_type mode;
505 enum mem_type mtype;
506 struct dimm_info *dimm;
508 /* Get data from the MC register, function 0 */
509 pdev = pvt->pci_mcr[0];
510 if (!pdev)
511 return -ENODEV;
513 /* Device 3 function 0 reads */
514 pci_read_config_dword(pdev, MC_CONTROL, &pvt->info.mc_control);
515 pci_read_config_dword(pdev, MC_STATUS, &pvt->info.mc_status);
516 pci_read_config_dword(pdev, MC_MAX_DOD, &pvt->info.max_dod);
517 pci_read_config_dword(pdev, MC_CHANNEL_MAPPER, &pvt->info.ch_map);
519 edac_dbg(0, "QPI %d control=0x%08x status=0x%08x dod=0x%08x map=0x%08x\n",
520 pvt->i7core_dev->socket, pvt->info.mc_control,
521 pvt->info.mc_status, pvt->info.max_dod, pvt->info.ch_map);
523 if (ECC_ENABLED(pvt)) {
524 edac_dbg(0, "ECC enabled with x%d SDCC\n", ECCx8(pvt) ? 8 : 4);
525 if (ECCx8(pvt))
526 mode = EDAC_S8ECD8ED;
527 else
528 mode = EDAC_S4ECD4ED;
529 } else {
530 edac_dbg(0, "ECC disabled\n");
531 mode = EDAC_NONE;
534 /* FIXME: need to handle the error codes */
535 edac_dbg(0, "DOD Max limits: DIMMS: %d, %d-ranked, %d-banked x%x x 0x%x\n",
536 numdimms(pvt->info.max_dod),
537 numrank(pvt->info.max_dod >> 2),
538 numbank(pvt->info.max_dod >> 4),
539 numrow(pvt->info.max_dod >> 6),
540 numcol(pvt->info.max_dod >> 9));
542 for (i = 0; i < NUM_CHANS; i++) {
543 u32 data, dimm_dod[3], value[8];
545 if (!pvt->pci_ch[i][0])
546 continue;
548 if (!CH_ACTIVE(pvt, i)) {
549 edac_dbg(0, "Channel %i is not active\n", i);
550 continue;
552 if (CH_DISABLED(pvt, i)) {
553 edac_dbg(0, "Channel %i is disabled\n", i);
554 continue;
557 /* Devices 4-6 function 0 */
558 pci_read_config_dword(pvt->pci_ch[i][0],
559 MC_CHANNEL_DIMM_INIT_PARAMS, &data);
562 if (data & THREE_DIMMS_PRESENT)
563 pvt->channel[i].is_3dimms_present = true;
565 if (data & SINGLE_QUAD_RANK_PRESENT)
566 pvt->channel[i].is_single_4rank = true;
568 if (data & QUAD_RANK_PRESENT)
569 pvt->channel[i].has_4rank = true;
571 if (data & REGISTERED_DIMM)
572 mtype = MEM_RDDR3;
573 else
574 mtype = MEM_DDR3;
576 /* Devices 4-6 function 1 */
577 pci_read_config_dword(pvt->pci_ch[i][1],
578 MC_DOD_CH_DIMM0, &dimm_dod[0]);
579 pci_read_config_dword(pvt->pci_ch[i][1],
580 MC_DOD_CH_DIMM1, &dimm_dod[1]);
581 pci_read_config_dword(pvt->pci_ch[i][1],
582 MC_DOD_CH_DIMM2, &dimm_dod[2]);
584 edac_dbg(0, "Ch%d phy rd%d, wr%d (0x%08x): %s%s%s%cDIMMs\n",
586 RDLCH(pvt->info.ch_map, i), WRLCH(pvt->info.ch_map, i),
587 data,
588 pvt->channel[i].is_3dimms_present ? "3DIMMS " : "",
589 pvt->channel[i].is_3dimms_present ? "SINGLE_4R " : "",
590 pvt->channel[i].has_4rank ? "HAS_4R " : "",
591 (data & REGISTERED_DIMM) ? 'R' : 'U');
593 for (j = 0; j < 3; j++) {
594 u32 banks, ranks, rows, cols;
595 u32 size, npages;
597 if (!DIMM_PRESENT(dimm_dod[j]))
598 continue;
600 dimm = EDAC_DIMM_PTR(mci->layers, mci->dimms, mci->n_layers,
601 i, j, 0);
602 banks = numbank(MC_DOD_NUMBANK(dimm_dod[j]));
603 ranks = numrank(MC_DOD_NUMRANK(dimm_dod[j]));
604 rows = numrow(MC_DOD_NUMROW(dimm_dod[j]));
605 cols = numcol(MC_DOD_NUMCOL(dimm_dod[j]));
607 /* DDR3 has 8 I/O banks */
608 size = (rows * cols * banks * ranks) >> (20 - 3);
610 edac_dbg(0, "\tdimm %d %d Mb offset: %x, bank: %d, rank: %d, row: %#x, col: %#x\n",
611 j, size,
612 RANKOFFSET(dimm_dod[j]),
613 banks, ranks, rows, cols);
615 npages = MiB_TO_PAGES(size);
617 dimm->nr_pages = npages;
619 switch (banks) {
620 case 4:
621 dimm->dtype = DEV_X4;
622 break;
623 case 8:
624 dimm->dtype = DEV_X8;
625 break;
626 case 16:
627 dimm->dtype = DEV_X16;
628 break;
629 default:
630 dimm->dtype = DEV_UNKNOWN;
633 snprintf(dimm->label, sizeof(dimm->label),
634 "CPU#%uChannel#%u_DIMM#%u",
635 pvt->i7core_dev->socket, i, j);
636 dimm->grain = 8;
637 dimm->edac_mode = mode;
638 dimm->mtype = mtype;
641 pci_read_config_dword(pdev, MC_SAG_CH_0, &value[0]);
642 pci_read_config_dword(pdev, MC_SAG_CH_1, &value[1]);
643 pci_read_config_dword(pdev, MC_SAG_CH_2, &value[2]);
644 pci_read_config_dword(pdev, MC_SAG_CH_3, &value[3]);
645 pci_read_config_dword(pdev, MC_SAG_CH_4, &value[4]);
646 pci_read_config_dword(pdev, MC_SAG_CH_5, &value[5]);
647 pci_read_config_dword(pdev, MC_SAG_CH_6, &value[6]);
648 pci_read_config_dword(pdev, MC_SAG_CH_7, &value[7]);
649 edac_dbg(1, "\t[%i] DIVBY3\tREMOVED\tOFFSET\n", i);
650 for (j = 0; j < 8; j++)
651 edac_dbg(1, "\t\t%#x\t%#x\t%#x\n",
652 (value[j] >> 27) & 0x1,
653 (value[j] >> 24) & 0x7,
654 (value[j] & ((1 << 24) - 1)));
657 return 0;
660 /****************************************************************************
661 Error insertion routines
662 ****************************************************************************/
664 #define to_mci(k) container_of(k, struct mem_ctl_info, dev)
666 /* The i7core has independent error injection features per channel.
667 However, to have a simpler code, we don't allow enabling error injection
668 on more than one channel.
669 Also, since a change at an inject parameter will be applied only at enable,
670 we're disabling error injection on all write calls to the sysfs nodes that
671 controls the error code injection.
673 static int disable_inject(const struct mem_ctl_info *mci)
675 struct i7core_pvt *pvt = mci->pvt_info;
677 pvt->inject.enable = 0;
679 if (!pvt->pci_ch[pvt->inject.channel][0])
680 return -ENODEV;
682 pci_write_config_dword(pvt->pci_ch[pvt->inject.channel][0],
683 MC_CHANNEL_ERROR_INJECT, 0);
685 return 0;
689 * i7core inject inject.section
691 * accept and store error injection inject.section value
692 * bit 0 - refers to the lower 32-byte half cacheline
693 * bit 1 - refers to the upper 32-byte half cacheline
695 static ssize_t i7core_inject_section_store(struct device *dev,
696 struct device_attribute *mattr,
697 const char *data, size_t count)
699 struct mem_ctl_info *mci = to_mci(dev);
700 struct i7core_pvt *pvt = mci->pvt_info;
701 unsigned long value;
702 int rc;
704 if (pvt->inject.enable)
705 disable_inject(mci);
707 rc = kstrtoul(data, 10, &value);
708 if ((rc < 0) || (value > 3))
709 return -EIO;
711 pvt->inject.section = (u32) value;
712 return count;
715 static ssize_t i7core_inject_section_show(struct device *dev,
716 struct device_attribute *mattr,
717 char *data)
719 struct mem_ctl_info *mci = to_mci(dev);
720 struct i7core_pvt *pvt = mci->pvt_info;
721 return sprintf(data, "0x%08x\n", pvt->inject.section);
725 * i7core inject.type
727 * accept and store error injection inject.section value
728 * bit 0 - repeat enable - Enable error repetition
729 * bit 1 - inject ECC error
730 * bit 2 - inject parity error
732 static ssize_t i7core_inject_type_store(struct device *dev,
733 struct device_attribute *mattr,
734 const char *data, size_t count)
736 struct mem_ctl_info *mci = to_mci(dev);
737 struct i7core_pvt *pvt = mci->pvt_info;
738 unsigned long value;
739 int rc;
741 if (pvt->inject.enable)
742 disable_inject(mci);
744 rc = kstrtoul(data, 10, &value);
745 if ((rc < 0) || (value > 7))
746 return -EIO;
748 pvt->inject.type = (u32) value;
749 return count;
752 static ssize_t i7core_inject_type_show(struct device *dev,
753 struct device_attribute *mattr,
754 char *data)
756 struct mem_ctl_info *mci = to_mci(dev);
757 struct i7core_pvt *pvt = mci->pvt_info;
759 return sprintf(data, "0x%08x\n", pvt->inject.type);
763 * i7core_inject_inject.eccmask_store
765 * The type of error (UE/CE) will depend on the inject.eccmask value:
766 * Any bits set to a 1 will flip the corresponding ECC bit
767 * Correctable errors can be injected by flipping 1 bit or the bits within
768 * a symbol pair (2 consecutive aligned 8-bit pairs - i.e. 7:0 and 15:8 or
769 * 23:16 and 31:24). Flipping bits in two symbol pairs will cause an
770 * uncorrectable error to be injected.
772 static ssize_t i7core_inject_eccmask_store(struct device *dev,
773 struct device_attribute *mattr,
774 const char *data, size_t count)
776 struct mem_ctl_info *mci = to_mci(dev);
777 struct i7core_pvt *pvt = mci->pvt_info;
778 unsigned long value;
779 int rc;
781 if (pvt->inject.enable)
782 disable_inject(mci);
784 rc = kstrtoul(data, 10, &value);
785 if (rc < 0)
786 return -EIO;
788 pvt->inject.eccmask = (u32) value;
789 return count;
792 static ssize_t i7core_inject_eccmask_show(struct device *dev,
793 struct device_attribute *mattr,
794 char *data)
796 struct mem_ctl_info *mci = to_mci(dev);
797 struct i7core_pvt *pvt = mci->pvt_info;
799 return sprintf(data, "0x%08x\n", pvt->inject.eccmask);
803 * i7core_addrmatch
805 * The type of error (UE/CE) will depend on the inject.eccmask value:
806 * Any bits set to a 1 will flip the corresponding ECC bit
807 * Correctable errors can be injected by flipping 1 bit or the bits within
808 * a symbol pair (2 consecutive aligned 8-bit pairs - i.e. 7:0 and 15:8 or
809 * 23:16 and 31:24). Flipping bits in two symbol pairs will cause an
810 * uncorrectable error to be injected.
813 #define DECLARE_ADDR_MATCH(param, limit) \
814 static ssize_t i7core_inject_store_##param( \
815 struct device *dev, \
816 struct device_attribute *mattr, \
817 const char *data, size_t count) \
819 struct mem_ctl_info *mci = dev_get_drvdata(dev); \
820 struct i7core_pvt *pvt; \
821 long value; \
822 int rc; \
824 edac_dbg(1, "\n"); \
825 pvt = mci->pvt_info; \
827 if (pvt->inject.enable) \
828 disable_inject(mci); \
830 if (!strcasecmp(data, "any") || !strcasecmp(data, "any\n"))\
831 value = -1; \
832 else { \
833 rc = kstrtoul(data, 10, &value); \
834 if ((rc < 0) || (value >= limit)) \
835 return -EIO; \
838 pvt->inject.param = value; \
840 return count; \
843 static ssize_t i7core_inject_show_##param( \
844 struct device *dev, \
845 struct device_attribute *mattr, \
846 char *data) \
848 struct mem_ctl_info *mci = dev_get_drvdata(dev); \
849 struct i7core_pvt *pvt; \
851 pvt = mci->pvt_info; \
852 edac_dbg(1, "pvt=%p\n", pvt); \
853 if (pvt->inject.param < 0) \
854 return sprintf(data, "any\n"); \
855 else \
856 return sprintf(data, "%d\n", pvt->inject.param);\
859 #define ATTR_ADDR_MATCH(param) \
860 static DEVICE_ATTR(param, S_IRUGO | S_IWUSR, \
861 i7core_inject_show_##param, \
862 i7core_inject_store_##param)
864 DECLARE_ADDR_MATCH(channel, 3);
865 DECLARE_ADDR_MATCH(dimm, 3);
866 DECLARE_ADDR_MATCH(rank, 4);
867 DECLARE_ADDR_MATCH(bank, 32);
868 DECLARE_ADDR_MATCH(page, 0x10000);
869 DECLARE_ADDR_MATCH(col, 0x4000);
871 ATTR_ADDR_MATCH(channel);
872 ATTR_ADDR_MATCH(dimm);
873 ATTR_ADDR_MATCH(rank);
874 ATTR_ADDR_MATCH(bank);
875 ATTR_ADDR_MATCH(page);
876 ATTR_ADDR_MATCH(col);
878 static int write_and_test(struct pci_dev *dev, const int where, const u32 val)
880 u32 read;
881 int count;
883 edac_dbg(0, "setting pci %02x:%02x.%x reg=%02x value=%08x\n",
884 dev->bus->number, PCI_SLOT(dev->devfn), PCI_FUNC(dev->devfn),
885 where, val);
887 for (count = 0; count < 10; count++) {
888 if (count)
889 msleep(100);
890 pci_write_config_dword(dev, where, val);
891 pci_read_config_dword(dev, where, &read);
893 if (read == val)
894 return 0;
897 i7core_printk(KERN_ERR, "Error during set pci %02x:%02x.%x reg=%02x "
898 "write=%08x. Read=%08x\n",
899 dev->bus->number, PCI_SLOT(dev->devfn), PCI_FUNC(dev->devfn),
900 where, val, read);
902 return -EINVAL;
906 * This routine prepares the Memory Controller for error injection.
907 * The error will be injected when some process tries to write to the
908 * memory that matches the given criteria.
909 * The criteria can be set in terms of a mask where dimm, rank, bank, page
910 * and col can be specified.
911 * A -1 value for any of the mask items will make the MCU to ignore
912 * that matching criteria for error injection.
914 * It should be noticed that the error will only happen after a write operation
915 * on a memory that matches the condition. if REPEAT_EN is not enabled at
916 * inject mask, then it will produce just one error. Otherwise, it will repeat
917 * until the injectmask would be cleaned.
919 * FIXME: This routine assumes that MAXNUMDIMMS value of MC_MAX_DOD
920 * is reliable enough to check if the MC is using the
921 * three channels. However, this is not clear at the datasheet.
923 static ssize_t i7core_inject_enable_store(struct device *dev,
924 struct device_attribute *mattr,
925 const char *data, size_t count)
927 struct mem_ctl_info *mci = to_mci(dev);
928 struct i7core_pvt *pvt = mci->pvt_info;
929 u32 injectmask;
930 u64 mask = 0;
931 int rc;
932 long enable;
934 if (!pvt->pci_ch[pvt->inject.channel][0])
935 return 0;
937 rc = kstrtoul(data, 10, &enable);
938 if ((rc < 0))
939 return 0;
941 if (enable) {
942 pvt->inject.enable = 1;
943 } else {
944 disable_inject(mci);
945 return count;
948 /* Sets pvt->inject.dimm mask */
949 if (pvt->inject.dimm < 0)
950 mask |= 1LL << 41;
951 else {
952 if (pvt->channel[pvt->inject.channel].dimms > 2)
953 mask |= (pvt->inject.dimm & 0x3LL) << 35;
954 else
955 mask |= (pvt->inject.dimm & 0x1LL) << 36;
958 /* Sets pvt->inject.rank mask */
959 if (pvt->inject.rank < 0)
960 mask |= 1LL << 40;
961 else {
962 if (pvt->channel[pvt->inject.channel].dimms > 2)
963 mask |= (pvt->inject.rank & 0x1LL) << 34;
964 else
965 mask |= (pvt->inject.rank & 0x3LL) << 34;
968 /* Sets pvt->inject.bank mask */
969 if (pvt->inject.bank < 0)
970 mask |= 1LL << 39;
971 else
972 mask |= (pvt->inject.bank & 0x15LL) << 30;
974 /* Sets pvt->inject.page mask */
975 if (pvt->inject.page < 0)
976 mask |= 1LL << 38;
977 else
978 mask |= (pvt->inject.page & 0xffff) << 14;
980 /* Sets pvt->inject.column mask */
981 if (pvt->inject.col < 0)
982 mask |= 1LL << 37;
983 else
984 mask |= (pvt->inject.col & 0x3fff);
987 * bit 0: REPEAT_EN
988 * bits 1-2: MASK_HALF_CACHELINE
989 * bit 3: INJECT_ECC
990 * bit 4: INJECT_ADDR_PARITY
993 injectmask = (pvt->inject.type & 1) |
994 (pvt->inject.section & 0x3) << 1 |
995 (pvt->inject.type & 0x6) << (3 - 1);
997 /* Unlock writes to registers - this register is write only */
998 pci_write_config_dword(pvt->pci_noncore,
999 MC_CFG_CONTROL, 0x2);
1001 write_and_test(pvt->pci_ch[pvt->inject.channel][0],
1002 MC_CHANNEL_ADDR_MATCH, mask);
1003 write_and_test(pvt->pci_ch[pvt->inject.channel][0],
1004 MC_CHANNEL_ADDR_MATCH + 4, mask >> 32L);
1006 write_and_test(pvt->pci_ch[pvt->inject.channel][0],
1007 MC_CHANNEL_ERROR_MASK, pvt->inject.eccmask);
1009 write_and_test(pvt->pci_ch[pvt->inject.channel][0],
1010 MC_CHANNEL_ERROR_INJECT, injectmask);
1013 * This is something undocumented, based on my tests
1014 * Without writing 8 to this register, errors aren't injected. Not sure
1015 * why.
1017 pci_write_config_dword(pvt->pci_noncore,
1018 MC_CFG_CONTROL, 8);
1020 edac_dbg(0, "Error inject addr match 0x%016llx, ecc 0x%08x, inject 0x%08x\n",
1021 mask, pvt->inject.eccmask, injectmask);
1024 return count;
1027 static ssize_t i7core_inject_enable_show(struct device *dev,
1028 struct device_attribute *mattr,
1029 char *data)
1031 struct mem_ctl_info *mci = to_mci(dev);
1032 struct i7core_pvt *pvt = mci->pvt_info;
1033 u32 injectmask;
1035 if (!pvt->pci_ch[pvt->inject.channel][0])
1036 return 0;
1038 pci_read_config_dword(pvt->pci_ch[pvt->inject.channel][0],
1039 MC_CHANNEL_ERROR_INJECT, &injectmask);
1041 edac_dbg(0, "Inject error read: 0x%018x\n", injectmask);
1043 if (injectmask & 0x0c)
1044 pvt->inject.enable = 1;
1046 return sprintf(data, "%d\n", pvt->inject.enable);
1049 #define DECLARE_COUNTER(param) \
1050 static ssize_t i7core_show_counter_##param( \
1051 struct device *dev, \
1052 struct device_attribute *mattr, \
1053 char *data) \
1055 struct mem_ctl_info *mci = dev_get_drvdata(dev); \
1056 struct i7core_pvt *pvt = mci->pvt_info; \
1058 edac_dbg(1, "\n"); \
1059 if (!pvt->ce_count_available || (pvt->is_registered)) \
1060 return sprintf(data, "data unavailable\n"); \
1061 return sprintf(data, "%lu\n", \
1062 pvt->udimm_ce_count[param]); \
1065 #define ATTR_COUNTER(param) \
1066 static DEVICE_ATTR(udimm##param, S_IRUGO | S_IWUSR, \
1067 i7core_show_counter_##param, \
1068 NULL)
1070 DECLARE_COUNTER(0);
1071 DECLARE_COUNTER(1);
1072 DECLARE_COUNTER(2);
1074 ATTR_COUNTER(0);
1075 ATTR_COUNTER(1);
1076 ATTR_COUNTER(2);
1079 * inject_addrmatch device sysfs struct
1082 static struct attribute *i7core_addrmatch_attrs[] = {
1083 &dev_attr_channel.attr,
1084 &dev_attr_dimm.attr,
1085 &dev_attr_rank.attr,
1086 &dev_attr_bank.attr,
1087 &dev_attr_page.attr,
1088 &dev_attr_col.attr,
1089 NULL
1092 static struct attribute_group addrmatch_grp = {
1093 .attrs = i7core_addrmatch_attrs,
1096 static const struct attribute_group *addrmatch_groups[] = {
1097 &addrmatch_grp,
1098 NULL
1101 static void addrmatch_release(struct device *device)
1103 edac_dbg(1, "Releasing device %s\n", dev_name(device));
1104 kfree(device);
1107 static struct device_type addrmatch_type = {
1108 .groups = addrmatch_groups,
1109 .release = addrmatch_release,
1113 * all_channel_counts sysfs struct
1116 static struct attribute *i7core_udimm_counters_attrs[] = {
1117 &dev_attr_udimm0.attr,
1118 &dev_attr_udimm1.attr,
1119 &dev_attr_udimm2.attr,
1120 NULL
1123 static struct attribute_group all_channel_counts_grp = {
1124 .attrs = i7core_udimm_counters_attrs,
1127 static const struct attribute_group *all_channel_counts_groups[] = {
1128 &all_channel_counts_grp,
1129 NULL
1132 static void all_channel_counts_release(struct device *device)
1134 edac_dbg(1, "Releasing device %s\n", dev_name(device));
1135 kfree(device);
1138 static struct device_type all_channel_counts_type = {
1139 .groups = all_channel_counts_groups,
1140 .release = all_channel_counts_release,
1144 * inject sysfs attributes
1147 static DEVICE_ATTR(inject_section, S_IRUGO | S_IWUSR,
1148 i7core_inject_section_show, i7core_inject_section_store);
1150 static DEVICE_ATTR(inject_type, S_IRUGO | S_IWUSR,
1151 i7core_inject_type_show, i7core_inject_type_store);
1154 static DEVICE_ATTR(inject_eccmask, S_IRUGO | S_IWUSR,
1155 i7core_inject_eccmask_show, i7core_inject_eccmask_store);
1157 static DEVICE_ATTR(inject_enable, S_IRUGO | S_IWUSR,
1158 i7core_inject_enable_show, i7core_inject_enable_store);
1160 static int i7core_create_sysfs_devices(struct mem_ctl_info *mci)
1162 struct i7core_pvt *pvt = mci->pvt_info;
1163 int rc;
1165 rc = device_create_file(&mci->dev, &dev_attr_inject_section);
1166 if (rc < 0)
1167 return rc;
1168 rc = device_create_file(&mci->dev, &dev_attr_inject_type);
1169 if (rc < 0)
1170 return rc;
1171 rc = device_create_file(&mci->dev, &dev_attr_inject_eccmask);
1172 if (rc < 0)
1173 return rc;
1174 rc = device_create_file(&mci->dev, &dev_attr_inject_enable);
1175 if (rc < 0)
1176 return rc;
1178 pvt->addrmatch_dev = kzalloc(sizeof(*pvt->addrmatch_dev), GFP_KERNEL);
1179 if (!pvt->addrmatch_dev)
1180 return rc;
1182 pvt->addrmatch_dev->type = &addrmatch_type;
1183 pvt->addrmatch_dev->bus = mci->dev.bus;
1184 device_initialize(pvt->addrmatch_dev);
1185 pvt->addrmatch_dev->parent = &mci->dev;
1186 dev_set_name(pvt->addrmatch_dev, "inject_addrmatch");
1187 dev_set_drvdata(pvt->addrmatch_dev, mci);
1189 edac_dbg(1, "creating %s\n", dev_name(pvt->addrmatch_dev));
1191 rc = device_add(pvt->addrmatch_dev);
1192 if (rc < 0)
1193 return rc;
1195 if (!pvt->is_registered) {
1196 pvt->chancounts_dev = kzalloc(sizeof(*pvt->chancounts_dev),
1197 GFP_KERNEL);
1198 if (!pvt->chancounts_dev) {
1199 put_device(pvt->addrmatch_dev);
1200 device_del(pvt->addrmatch_dev);
1201 return rc;
1204 pvt->chancounts_dev->type = &all_channel_counts_type;
1205 pvt->chancounts_dev->bus = mci->dev.bus;
1206 device_initialize(pvt->chancounts_dev);
1207 pvt->chancounts_dev->parent = &mci->dev;
1208 dev_set_name(pvt->chancounts_dev, "all_channel_counts");
1209 dev_set_drvdata(pvt->chancounts_dev, mci);
1211 edac_dbg(1, "creating %s\n", dev_name(pvt->chancounts_dev));
1213 rc = device_add(pvt->chancounts_dev);
1214 if (rc < 0)
1215 return rc;
1217 return 0;
1220 static void i7core_delete_sysfs_devices(struct mem_ctl_info *mci)
1222 struct i7core_pvt *pvt = mci->pvt_info;
1224 edac_dbg(1, "\n");
1226 device_remove_file(&mci->dev, &dev_attr_inject_section);
1227 device_remove_file(&mci->dev, &dev_attr_inject_type);
1228 device_remove_file(&mci->dev, &dev_attr_inject_eccmask);
1229 device_remove_file(&mci->dev, &dev_attr_inject_enable);
1231 if (!pvt->is_registered) {
1232 put_device(pvt->chancounts_dev);
1233 device_del(pvt->chancounts_dev);
1235 put_device(pvt->addrmatch_dev);
1236 device_del(pvt->addrmatch_dev);
1239 /****************************************************************************
1240 Device initialization routines: put/get, init/exit
1241 ****************************************************************************/
1244 * i7core_put_all_devices 'put' all the devices that we have
1245 * reserved via 'get'
1247 static void i7core_put_devices(struct i7core_dev *i7core_dev)
1249 int i;
1251 edac_dbg(0, "\n");
1252 for (i = 0; i < i7core_dev->n_devs; i++) {
1253 struct pci_dev *pdev = i7core_dev->pdev[i];
1254 if (!pdev)
1255 continue;
1256 edac_dbg(0, "Removing dev %02x:%02x.%d\n",
1257 pdev->bus->number,
1258 PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn));
1259 pci_dev_put(pdev);
1263 static void i7core_put_all_devices(void)
1265 struct i7core_dev *i7core_dev, *tmp;
1267 list_for_each_entry_safe(i7core_dev, tmp, &i7core_edac_list, list) {
1268 i7core_put_devices(i7core_dev);
1269 free_i7core_dev(i7core_dev);
1273 static void __init i7core_xeon_pci_fixup(const struct pci_id_table *table)
1275 struct pci_dev *pdev = NULL;
1276 int i;
1279 * On Xeon 55xx, the Intel Quick Path Arch Generic Non-core pci buses
1280 * aren't announced by acpi. So, we need to use a legacy scan probing
1281 * to detect them
1283 while (table && table->descr) {
1284 pdev = pci_get_device(PCI_VENDOR_ID_INTEL, table->descr[0].dev_id, NULL);
1285 if (unlikely(!pdev)) {
1286 for (i = 0; i < MAX_SOCKET_BUSES; i++)
1287 pcibios_scan_specific_bus(255-i);
1289 pci_dev_put(pdev);
1290 table++;
1294 static unsigned i7core_pci_lastbus(void)
1296 int last_bus = 0, bus;
1297 struct pci_bus *b = NULL;
1299 while ((b = pci_find_next_bus(b)) != NULL) {
1300 bus = b->number;
1301 edac_dbg(0, "Found bus %d\n", bus);
1302 if (bus > last_bus)
1303 last_bus = bus;
1306 edac_dbg(0, "Last bus %d\n", last_bus);
1308 return last_bus;
1312 * i7core_get_all_devices Find and perform 'get' operation on the MCH's
1313 * device/functions we want to reference for this driver
1315 * Need to 'get' device 16 func 1 and func 2
1317 static int i7core_get_onedevice(struct pci_dev **prev,
1318 const struct pci_id_table *table,
1319 const unsigned devno,
1320 const unsigned last_bus)
1322 struct i7core_dev *i7core_dev;
1323 const struct pci_id_descr *dev_descr = &table->descr[devno];
1325 struct pci_dev *pdev = NULL;
1326 u8 bus = 0;
1327 u8 socket = 0;
1329 pdev = pci_get_device(PCI_VENDOR_ID_INTEL,
1330 dev_descr->dev_id, *prev);
1333 * On Xeon 55xx, the Intel QuickPath Arch Generic Non-core regs
1334 * is at addr 8086:2c40, instead of 8086:2c41. So, we need
1335 * to probe for the alternate address in case of failure
1337 if (dev_descr->dev_id == PCI_DEVICE_ID_INTEL_I7_NONCORE && !pdev)
1338 pdev = pci_get_device(PCI_VENDOR_ID_INTEL,
1339 PCI_DEVICE_ID_INTEL_I7_NONCORE_ALT, *prev);
1341 if (dev_descr->dev_id == PCI_DEVICE_ID_INTEL_LYNNFIELD_NONCORE && !pdev)
1342 pdev = pci_get_device(PCI_VENDOR_ID_INTEL,
1343 PCI_DEVICE_ID_INTEL_LYNNFIELD_NONCORE_ALT,
1344 *prev);
1346 if (!pdev) {
1347 if (*prev) {
1348 *prev = pdev;
1349 return 0;
1352 if (dev_descr->optional)
1353 return 0;
1355 if (devno == 0)
1356 return -ENODEV;
1358 i7core_printk(KERN_INFO,
1359 "Device not found: dev %02x.%d PCI ID %04x:%04x\n",
1360 dev_descr->dev, dev_descr->func,
1361 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
1363 /* End of list, leave */
1364 return -ENODEV;
1366 bus = pdev->bus->number;
1368 socket = last_bus - bus;
1370 i7core_dev = get_i7core_dev(socket);
1371 if (!i7core_dev) {
1372 i7core_dev = alloc_i7core_dev(socket, table);
1373 if (!i7core_dev) {
1374 pci_dev_put(pdev);
1375 return -ENOMEM;
1379 if (i7core_dev->pdev[devno]) {
1380 i7core_printk(KERN_ERR,
1381 "Duplicated device for "
1382 "dev %02x:%02x.%d PCI ID %04x:%04x\n",
1383 bus, dev_descr->dev, dev_descr->func,
1384 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
1385 pci_dev_put(pdev);
1386 return -ENODEV;
1389 i7core_dev->pdev[devno] = pdev;
1391 /* Sanity check */
1392 if (unlikely(PCI_SLOT(pdev->devfn) != dev_descr->dev ||
1393 PCI_FUNC(pdev->devfn) != dev_descr->func)) {
1394 i7core_printk(KERN_ERR,
1395 "Device PCI ID %04x:%04x "
1396 "has dev %02x:%02x.%d instead of dev %02x:%02x.%d\n",
1397 PCI_VENDOR_ID_INTEL, dev_descr->dev_id,
1398 bus, PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn),
1399 bus, dev_descr->dev, dev_descr->func);
1400 return -ENODEV;
1403 /* Be sure that the device is enabled */
1404 if (unlikely(pci_enable_device(pdev) < 0)) {
1405 i7core_printk(KERN_ERR,
1406 "Couldn't enable "
1407 "dev %02x:%02x.%d PCI ID %04x:%04x\n",
1408 bus, dev_descr->dev, dev_descr->func,
1409 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
1410 return -ENODEV;
1413 edac_dbg(0, "Detected socket %d dev %02x:%02x.%d PCI ID %04x:%04x\n",
1414 socket, bus, dev_descr->dev,
1415 dev_descr->func,
1416 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
1419 * As stated on drivers/pci/search.c, the reference count for
1420 * @from is always decremented if it is not %NULL. So, as we need
1421 * to get all devices up to null, we need to do a get for the device
1423 pci_dev_get(pdev);
1425 *prev = pdev;
1427 return 0;
1430 static int i7core_get_all_devices(void)
1432 int i, rc, last_bus;
1433 struct pci_dev *pdev = NULL;
1434 const struct pci_id_table *table = pci_dev_table;
1436 last_bus = i7core_pci_lastbus();
1438 while (table && table->descr) {
1439 for (i = 0; i < table->n_devs; i++) {
1440 pdev = NULL;
1441 do {
1442 rc = i7core_get_onedevice(&pdev, table, i,
1443 last_bus);
1444 if (rc < 0) {
1445 if (i == 0) {
1446 i = table->n_devs;
1447 break;
1449 i7core_put_all_devices();
1450 return -ENODEV;
1452 } while (pdev);
1454 table++;
1457 return 0;
1460 static int mci_bind_devs(struct mem_ctl_info *mci,
1461 struct i7core_dev *i7core_dev)
1463 struct i7core_pvt *pvt = mci->pvt_info;
1464 struct pci_dev *pdev;
1465 int i, func, slot;
1466 char *family;
1468 pvt->is_registered = false;
1469 pvt->enable_scrub = false;
1470 for (i = 0; i < i7core_dev->n_devs; i++) {
1471 pdev = i7core_dev->pdev[i];
1472 if (!pdev)
1473 continue;
1475 func = PCI_FUNC(pdev->devfn);
1476 slot = PCI_SLOT(pdev->devfn);
1477 if (slot == 3) {
1478 if (unlikely(func > MAX_MCR_FUNC))
1479 goto error;
1480 pvt->pci_mcr[func] = pdev;
1481 } else if (likely(slot >= 4 && slot < 4 + NUM_CHANS)) {
1482 if (unlikely(func > MAX_CHAN_FUNC))
1483 goto error;
1484 pvt->pci_ch[slot - 4][func] = pdev;
1485 } else if (!slot && !func) {
1486 pvt->pci_noncore = pdev;
1488 /* Detect the processor family */
1489 switch (pdev->device) {
1490 case PCI_DEVICE_ID_INTEL_I7_NONCORE:
1491 family = "Xeon 35xx/ i7core";
1492 pvt->enable_scrub = false;
1493 break;
1494 case PCI_DEVICE_ID_INTEL_LYNNFIELD_NONCORE_ALT:
1495 family = "i7-800/i5-700";
1496 pvt->enable_scrub = false;
1497 break;
1498 case PCI_DEVICE_ID_INTEL_LYNNFIELD_NONCORE:
1499 family = "Xeon 34xx";
1500 pvt->enable_scrub = false;
1501 break;
1502 case PCI_DEVICE_ID_INTEL_I7_NONCORE_ALT:
1503 family = "Xeon 55xx";
1504 pvt->enable_scrub = true;
1505 break;
1506 case PCI_DEVICE_ID_INTEL_LYNNFIELD_NONCORE_REV2:
1507 family = "Xeon 56xx / i7-900";
1508 pvt->enable_scrub = true;
1509 break;
1510 default:
1511 family = "unknown";
1512 pvt->enable_scrub = false;
1514 edac_dbg(0, "Detected a processor type %s\n", family);
1515 } else
1516 goto error;
1518 edac_dbg(0, "Associated fn %d.%d, dev = %p, socket %d\n",
1519 PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn),
1520 pdev, i7core_dev->socket);
1522 if (PCI_SLOT(pdev->devfn) == 3 &&
1523 PCI_FUNC(pdev->devfn) == 2)
1524 pvt->is_registered = true;
1527 return 0;
1529 error:
1530 i7core_printk(KERN_ERR, "Device %d, function %d "
1531 "is out of the expected range\n",
1532 slot, func);
1533 return -EINVAL;
1536 /****************************************************************************
1537 Error check routines
1538 ****************************************************************************/
1540 static void i7core_rdimm_update_ce_count(struct mem_ctl_info *mci,
1541 const int chan,
1542 const int new0,
1543 const int new1,
1544 const int new2)
1546 struct i7core_pvt *pvt = mci->pvt_info;
1547 int add0 = 0, add1 = 0, add2 = 0;
1548 /* Updates CE counters if it is not the first time here */
1549 if (pvt->ce_count_available) {
1550 /* Updates CE counters */
1552 add2 = new2 - pvt->rdimm_last_ce_count[chan][2];
1553 add1 = new1 - pvt->rdimm_last_ce_count[chan][1];
1554 add0 = new0 - pvt->rdimm_last_ce_count[chan][0];
1556 if (add2 < 0)
1557 add2 += 0x7fff;
1558 pvt->rdimm_ce_count[chan][2] += add2;
1560 if (add1 < 0)
1561 add1 += 0x7fff;
1562 pvt->rdimm_ce_count[chan][1] += add1;
1564 if (add0 < 0)
1565 add0 += 0x7fff;
1566 pvt->rdimm_ce_count[chan][0] += add0;
1567 } else
1568 pvt->ce_count_available = 1;
1570 /* Store the new values */
1571 pvt->rdimm_last_ce_count[chan][2] = new2;
1572 pvt->rdimm_last_ce_count[chan][1] = new1;
1573 pvt->rdimm_last_ce_count[chan][0] = new0;
1575 /*updated the edac core */
1576 if (add0 != 0)
1577 edac_mc_handle_error(HW_EVENT_ERR_CORRECTED, mci, add0,
1578 0, 0, 0,
1579 chan, 0, -1, "error", "");
1580 if (add1 != 0)
1581 edac_mc_handle_error(HW_EVENT_ERR_CORRECTED, mci, add1,
1582 0, 0, 0,
1583 chan, 1, -1, "error", "");
1584 if (add2 != 0)
1585 edac_mc_handle_error(HW_EVENT_ERR_CORRECTED, mci, add2,
1586 0, 0, 0,
1587 chan, 2, -1, "error", "");
1590 static void i7core_rdimm_check_mc_ecc_err(struct mem_ctl_info *mci)
1592 struct i7core_pvt *pvt = mci->pvt_info;
1593 u32 rcv[3][2];
1594 int i, new0, new1, new2;
1596 /*Read DEV 3: FUN 2: MC_COR_ECC_CNT regs directly*/
1597 pci_read_config_dword(pvt->pci_mcr[2], MC_COR_ECC_CNT_0,
1598 &rcv[0][0]);
1599 pci_read_config_dword(pvt->pci_mcr[2], MC_COR_ECC_CNT_1,
1600 &rcv[0][1]);
1601 pci_read_config_dword(pvt->pci_mcr[2], MC_COR_ECC_CNT_2,
1602 &rcv[1][0]);
1603 pci_read_config_dword(pvt->pci_mcr[2], MC_COR_ECC_CNT_3,
1604 &rcv[1][1]);
1605 pci_read_config_dword(pvt->pci_mcr[2], MC_COR_ECC_CNT_4,
1606 &rcv[2][0]);
1607 pci_read_config_dword(pvt->pci_mcr[2], MC_COR_ECC_CNT_5,
1608 &rcv[2][1]);
1609 for (i = 0 ; i < 3; i++) {
1610 edac_dbg(3, "MC_COR_ECC_CNT%d = 0x%x; MC_COR_ECC_CNT%d = 0x%x\n",
1611 (i * 2), rcv[i][0], (i * 2) + 1, rcv[i][1]);
1612 /*if the channel has 3 dimms*/
1613 if (pvt->channel[i].dimms > 2) {
1614 new0 = DIMM_BOT_COR_ERR(rcv[i][0]);
1615 new1 = DIMM_TOP_COR_ERR(rcv[i][0]);
1616 new2 = DIMM_BOT_COR_ERR(rcv[i][1]);
1617 } else {
1618 new0 = DIMM_TOP_COR_ERR(rcv[i][0]) +
1619 DIMM_BOT_COR_ERR(rcv[i][0]);
1620 new1 = DIMM_TOP_COR_ERR(rcv[i][1]) +
1621 DIMM_BOT_COR_ERR(rcv[i][1]);
1622 new2 = 0;
1625 i7core_rdimm_update_ce_count(mci, i, new0, new1, new2);
1629 /* This function is based on the device 3 function 4 registers as described on:
1630 * Intel Xeon Processor 5500 Series Datasheet Volume 2
1631 * http://www.intel.com/Assets/PDF/datasheet/321322.pdf
1632 * also available at:
1633 * http://www.arrownac.com/manufacturers/intel/s/nehalem/5500-datasheet-v2.pdf
1635 static void i7core_udimm_check_mc_ecc_err(struct mem_ctl_info *mci)
1637 struct i7core_pvt *pvt = mci->pvt_info;
1638 u32 rcv1, rcv0;
1639 int new0, new1, new2;
1641 if (!pvt->pci_mcr[4]) {
1642 edac_dbg(0, "MCR registers not found\n");
1643 return;
1646 /* Corrected test errors */
1647 pci_read_config_dword(pvt->pci_mcr[4], MC_TEST_ERR_RCV1, &rcv1);
1648 pci_read_config_dword(pvt->pci_mcr[4], MC_TEST_ERR_RCV0, &rcv0);
1650 /* Store the new values */
1651 new2 = DIMM2_COR_ERR(rcv1);
1652 new1 = DIMM1_COR_ERR(rcv0);
1653 new0 = DIMM0_COR_ERR(rcv0);
1655 /* Updates CE counters if it is not the first time here */
1656 if (pvt->ce_count_available) {
1657 /* Updates CE counters */
1658 int add0, add1, add2;
1660 add2 = new2 - pvt->udimm_last_ce_count[2];
1661 add1 = new1 - pvt->udimm_last_ce_count[1];
1662 add0 = new0 - pvt->udimm_last_ce_count[0];
1664 if (add2 < 0)
1665 add2 += 0x7fff;
1666 pvt->udimm_ce_count[2] += add2;
1668 if (add1 < 0)
1669 add1 += 0x7fff;
1670 pvt->udimm_ce_count[1] += add1;
1672 if (add0 < 0)
1673 add0 += 0x7fff;
1674 pvt->udimm_ce_count[0] += add0;
1676 if (add0 | add1 | add2)
1677 i7core_printk(KERN_ERR, "New Corrected error(s): "
1678 "dimm0: +%d, dimm1: +%d, dimm2 +%d\n",
1679 add0, add1, add2);
1680 } else
1681 pvt->ce_count_available = 1;
1683 /* Store the new values */
1684 pvt->udimm_last_ce_count[2] = new2;
1685 pvt->udimm_last_ce_count[1] = new1;
1686 pvt->udimm_last_ce_count[0] = new0;
1690 * According with tables E-11 and E-12 of chapter E.3.3 of Intel 64 and IA-32
1691 * Architectures Software Developer’s Manual Volume 3B.
1692 * Nehalem are defined as family 0x06, model 0x1a
1694 * The MCA registers used here are the following ones:
1695 * struct mce field MCA Register
1696 * m->status MSR_IA32_MC8_STATUS
1697 * m->addr MSR_IA32_MC8_ADDR
1698 * m->misc MSR_IA32_MC8_MISC
1699 * In the case of Nehalem, the error information is masked at .status and .misc
1700 * fields
1702 static void i7core_mce_output_error(struct mem_ctl_info *mci,
1703 const struct mce *m)
1705 struct i7core_pvt *pvt = mci->pvt_info;
1706 char *type, *optype, *err;
1707 enum hw_event_mc_err_type tp_event;
1708 unsigned long error = m->status & 0x1ff0000l;
1709 bool uncorrected_error = m->mcgstatus & 1ll << 61;
1710 bool ripv = m->mcgstatus & 1;
1711 u32 optypenum = (m->status >> 4) & 0x07;
1712 u32 core_err_cnt = (m->status >> 38) & 0x7fff;
1713 u32 dimm = (m->misc >> 16) & 0x3;
1714 u32 channel = (m->misc >> 18) & 0x3;
1715 u32 syndrome = m->misc >> 32;
1716 u32 errnum = find_first_bit(&error, 32);
1718 if (uncorrected_error) {
1719 if (ripv) {
1720 type = "FATAL";
1721 tp_event = HW_EVENT_ERR_FATAL;
1722 } else {
1723 type = "NON_FATAL";
1724 tp_event = HW_EVENT_ERR_UNCORRECTED;
1726 } else {
1727 type = "CORRECTED";
1728 tp_event = HW_EVENT_ERR_CORRECTED;
1731 switch (optypenum) {
1732 case 0:
1733 optype = "generic undef request";
1734 break;
1735 case 1:
1736 optype = "read error";
1737 break;
1738 case 2:
1739 optype = "write error";
1740 break;
1741 case 3:
1742 optype = "addr/cmd error";
1743 break;
1744 case 4:
1745 optype = "scrubbing error";
1746 break;
1747 default:
1748 optype = "reserved";
1749 break;
1752 switch (errnum) {
1753 case 16:
1754 err = "read ECC error";
1755 break;
1756 case 17:
1757 err = "RAS ECC error";
1758 break;
1759 case 18:
1760 err = "write parity error";
1761 break;
1762 case 19:
1763 err = "redundacy loss";
1764 break;
1765 case 20:
1766 err = "reserved";
1767 break;
1768 case 21:
1769 err = "memory range error";
1770 break;
1771 case 22:
1772 err = "RTID out of range";
1773 break;
1774 case 23:
1775 err = "address parity error";
1776 break;
1777 case 24:
1778 err = "byte enable parity error";
1779 break;
1780 default:
1781 err = "unknown";
1785 * Call the helper to output message
1786 * FIXME: what to do if core_err_cnt > 1? Currently, it generates
1787 * only one event
1789 if (uncorrected_error || !pvt->is_registered)
1790 edac_mc_handle_error(tp_event, mci, core_err_cnt,
1791 m->addr >> PAGE_SHIFT,
1792 m->addr & ~PAGE_MASK,
1793 syndrome,
1794 channel, dimm, -1,
1795 err, optype);
1799 * i7core_check_error Retrieve and process errors reported by the
1800 * hardware. Called by the Core module.
1802 static void i7core_check_error(struct mem_ctl_info *mci)
1804 struct i7core_pvt *pvt = mci->pvt_info;
1805 int i;
1806 unsigned count = 0;
1807 struct mce *m;
1810 * MCE first step: Copy all mce errors into a temporary buffer
1811 * We use a double buffering here, to reduce the risk of
1812 * losing an error.
1814 smp_rmb();
1815 count = (pvt->mce_out + MCE_LOG_LEN - pvt->mce_in)
1816 % MCE_LOG_LEN;
1817 if (!count)
1818 goto check_ce_error;
1820 m = pvt->mce_outentry;
1821 if (pvt->mce_in + count > MCE_LOG_LEN) {
1822 unsigned l = MCE_LOG_LEN - pvt->mce_in;
1824 memcpy(m, &pvt->mce_entry[pvt->mce_in], sizeof(*m) * l);
1825 smp_wmb();
1826 pvt->mce_in = 0;
1827 count -= l;
1828 m += l;
1830 memcpy(m, &pvt->mce_entry[pvt->mce_in], sizeof(*m) * count);
1831 smp_wmb();
1832 pvt->mce_in += count;
1834 smp_rmb();
1835 if (pvt->mce_overrun) {
1836 i7core_printk(KERN_ERR, "Lost %d memory errors\n",
1837 pvt->mce_overrun);
1838 smp_wmb();
1839 pvt->mce_overrun = 0;
1843 * MCE second step: parse errors and display
1845 for (i = 0; i < count; i++)
1846 i7core_mce_output_error(mci, &pvt->mce_outentry[i]);
1849 * Now, let's increment CE error counts
1851 check_ce_error:
1852 if (!pvt->is_registered)
1853 i7core_udimm_check_mc_ecc_err(mci);
1854 else
1855 i7core_rdimm_check_mc_ecc_err(mci);
1859 * i7core_mce_check_error Replicates mcelog routine to get errors
1860 * This routine simply queues mcelog errors, and
1861 * return. The error itself should be handled later
1862 * by i7core_check_error.
1863 * WARNING: As this routine should be called at NMI time, extra care should
1864 * be taken to avoid deadlocks, and to be as fast as possible.
1866 static int i7core_mce_check_error(struct notifier_block *nb, unsigned long val,
1867 void *data)
1869 struct mce *mce = (struct mce *)data;
1870 struct i7core_dev *i7_dev;
1871 struct mem_ctl_info *mci;
1872 struct i7core_pvt *pvt;
1874 i7_dev = get_i7core_dev(mce->socketid);
1875 if (!i7_dev)
1876 return NOTIFY_BAD;
1878 mci = i7_dev->mci;
1879 pvt = mci->pvt_info;
1882 * Just let mcelog handle it if the error is
1883 * outside the memory controller
1885 if (((mce->status & 0xffff) >> 7) != 1)
1886 return NOTIFY_DONE;
1888 /* Bank 8 registers are the only ones that we know how to handle */
1889 if (mce->bank != 8)
1890 return NOTIFY_DONE;
1892 smp_rmb();
1893 if ((pvt->mce_out + 1) % MCE_LOG_LEN == pvt->mce_in) {
1894 smp_wmb();
1895 pvt->mce_overrun++;
1896 return NOTIFY_DONE;
1899 /* Copy memory error at the ringbuffer */
1900 memcpy(&pvt->mce_entry[pvt->mce_out], mce, sizeof(*mce));
1901 smp_wmb();
1902 pvt->mce_out = (pvt->mce_out + 1) % MCE_LOG_LEN;
1904 /* Handle fatal errors immediately */
1905 if (mce->mcgstatus & 1)
1906 i7core_check_error(mci);
1908 /* Advise mcelog that the errors were handled */
1909 return NOTIFY_STOP;
1912 static struct notifier_block i7_mce_dec = {
1913 .notifier_call = i7core_mce_check_error,
1916 struct memdev_dmi_entry {
1917 u8 type;
1918 u8 length;
1919 u16 handle;
1920 u16 phys_mem_array_handle;
1921 u16 mem_err_info_handle;
1922 u16 total_width;
1923 u16 data_width;
1924 u16 size;
1925 u8 form;
1926 u8 device_set;
1927 u8 device_locator;
1928 u8 bank_locator;
1929 u8 memory_type;
1930 u16 type_detail;
1931 u16 speed;
1932 u8 manufacturer;
1933 u8 serial_number;
1934 u8 asset_tag;
1935 u8 part_number;
1936 u8 attributes;
1937 u32 extended_size;
1938 u16 conf_mem_clk_speed;
1939 } __attribute__((__packed__));
1943 * Decode the DRAM Clock Frequency, be paranoid, make sure that all
1944 * memory devices show the same speed, and if they don't then consider
1945 * all speeds to be invalid.
1947 static void decode_dclk(const struct dmi_header *dh, void *_dclk_freq)
1949 int *dclk_freq = _dclk_freq;
1950 u16 dmi_mem_clk_speed;
1952 if (*dclk_freq == -1)
1953 return;
1955 if (dh->type == DMI_ENTRY_MEM_DEVICE) {
1956 struct memdev_dmi_entry *memdev_dmi_entry =
1957 (struct memdev_dmi_entry *)dh;
1958 unsigned long conf_mem_clk_speed_offset =
1959 (unsigned long)&memdev_dmi_entry->conf_mem_clk_speed -
1960 (unsigned long)&memdev_dmi_entry->type;
1961 unsigned long speed_offset =
1962 (unsigned long)&memdev_dmi_entry->speed -
1963 (unsigned long)&memdev_dmi_entry->type;
1965 /* Check that a DIMM is present */
1966 if (memdev_dmi_entry->size == 0)
1967 return;
1970 * Pick the configured speed if it's available, otherwise
1971 * pick the DIMM speed, or we don't have a speed.
1973 if (memdev_dmi_entry->length > conf_mem_clk_speed_offset) {
1974 dmi_mem_clk_speed =
1975 memdev_dmi_entry->conf_mem_clk_speed;
1976 } else if (memdev_dmi_entry->length > speed_offset) {
1977 dmi_mem_clk_speed = memdev_dmi_entry->speed;
1978 } else {
1979 *dclk_freq = -1;
1980 return;
1983 if (*dclk_freq == 0) {
1984 /* First pass, speed was 0 */
1985 if (dmi_mem_clk_speed > 0) {
1986 /* Set speed if a valid speed is read */
1987 *dclk_freq = dmi_mem_clk_speed;
1988 } else {
1989 /* Otherwise we don't have a valid speed */
1990 *dclk_freq = -1;
1992 } else if (*dclk_freq > 0 &&
1993 *dclk_freq != dmi_mem_clk_speed) {
1995 * If we have a speed, check that all DIMMS are the same
1996 * speed, otherwise set the speed as invalid.
1998 *dclk_freq = -1;
2004 * The default DCLK frequency is used as a fallback if we
2005 * fail to find anything reliable in the DMI. The value
2006 * is taken straight from the datasheet.
2008 #define DEFAULT_DCLK_FREQ 800
2010 static int get_dclk_freq(void)
2012 int dclk_freq = 0;
2014 dmi_walk(decode_dclk, (void *)&dclk_freq);
2016 if (dclk_freq < 1)
2017 return DEFAULT_DCLK_FREQ;
2019 return dclk_freq;
2023 * set_sdram_scrub_rate This routine sets byte/sec bandwidth scrub rate
2024 * to hardware according to SCRUBINTERVAL formula
2025 * found in datasheet.
2027 static int set_sdram_scrub_rate(struct mem_ctl_info *mci, u32 new_bw)
2029 struct i7core_pvt *pvt = mci->pvt_info;
2030 struct pci_dev *pdev;
2031 u32 dw_scrub;
2032 u32 dw_ssr;
2034 /* Get data from the MC register, function 2 */
2035 pdev = pvt->pci_mcr[2];
2036 if (!pdev)
2037 return -ENODEV;
2039 pci_read_config_dword(pdev, MC_SCRUB_CONTROL, &dw_scrub);
2041 if (new_bw == 0) {
2042 /* Prepare to disable petrol scrub */
2043 dw_scrub &= ~STARTSCRUB;
2044 /* Stop the patrol scrub engine */
2045 write_and_test(pdev, MC_SCRUB_CONTROL,
2046 dw_scrub & ~SCRUBINTERVAL_MASK);
2048 /* Get current status of scrub rate and set bit to disable */
2049 pci_read_config_dword(pdev, MC_SSRCONTROL, &dw_ssr);
2050 dw_ssr &= ~SSR_MODE_MASK;
2051 dw_ssr |= SSR_MODE_DISABLE;
2052 } else {
2053 const int cache_line_size = 64;
2054 const u32 freq_dclk_mhz = pvt->dclk_freq;
2055 unsigned long long scrub_interval;
2057 * Translate the desired scrub rate to a register value and
2058 * program the corresponding register value.
2060 scrub_interval = (unsigned long long)freq_dclk_mhz *
2061 cache_line_size * 1000000;
2062 do_div(scrub_interval, new_bw);
2064 if (!scrub_interval || scrub_interval > SCRUBINTERVAL_MASK)
2065 return -EINVAL;
2067 dw_scrub = SCRUBINTERVAL_MASK & scrub_interval;
2069 /* Start the patrol scrub engine */
2070 pci_write_config_dword(pdev, MC_SCRUB_CONTROL,
2071 STARTSCRUB | dw_scrub);
2073 /* Get current status of scrub rate and set bit to enable */
2074 pci_read_config_dword(pdev, MC_SSRCONTROL, &dw_ssr);
2075 dw_ssr &= ~SSR_MODE_MASK;
2076 dw_ssr |= SSR_MODE_ENABLE;
2078 /* Disable or enable scrubbing */
2079 pci_write_config_dword(pdev, MC_SSRCONTROL, dw_ssr);
2081 return new_bw;
2085 * get_sdram_scrub_rate This routine convert current scrub rate value
2086 * into byte/sec bandwidth according to
2087 * SCRUBINTERVAL formula found in datasheet.
2089 static int get_sdram_scrub_rate(struct mem_ctl_info *mci)
2091 struct i7core_pvt *pvt = mci->pvt_info;
2092 struct pci_dev *pdev;
2093 const u32 cache_line_size = 64;
2094 const u32 freq_dclk_mhz = pvt->dclk_freq;
2095 unsigned long long scrub_rate;
2096 u32 scrubval;
2098 /* Get data from the MC register, function 2 */
2099 pdev = pvt->pci_mcr[2];
2100 if (!pdev)
2101 return -ENODEV;
2103 /* Get current scrub control data */
2104 pci_read_config_dword(pdev, MC_SCRUB_CONTROL, &scrubval);
2106 /* Mask highest 8-bits to 0 */
2107 scrubval &= SCRUBINTERVAL_MASK;
2108 if (!scrubval)
2109 return 0;
2111 /* Calculate scrub rate value into byte/sec bandwidth */
2112 scrub_rate = (unsigned long long)freq_dclk_mhz *
2113 1000000 * cache_line_size;
2114 do_div(scrub_rate, scrubval);
2115 return (int)scrub_rate;
2118 static void enable_sdram_scrub_setting(struct mem_ctl_info *mci)
2120 struct i7core_pvt *pvt = mci->pvt_info;
2121 u32 pci_lock;
2123 /* Unlock writes to pci registers */
2124 pci_read_config_dword(pvt->pci_noncore, MC_CFG_CONTROL, &pci_lock);
2125 pci_lock &= ~0x3;
2126 pci_write_config_dword(pvt->pci_noncore, MC_CFG_CONTROL,
2127 pci_lock | MC_CFG_UNLOCK);
2129 mci->set_sdram_scrub_rate = set_sdram_scrub_rate;
2130 mci->get_sdram_scrub_rate = get_sdram_scrub_rate;
2133 static void disable_sdram_scrub_setting(struct mem_ctl_info *mci)
2135 struct i7core_pvt *pvt = mci->pvt_info;
2136 u32 pci_lock;
2138 /* Lock writes to pci registers */
2139 pci_read_config_dword(pvt->pci_noncore, MC_CFG_CONTROL, &pci_lock);
2140 pci_lock &= ~0x3;
2141 pci_write_config_dword(pvt->pci_noncore, MC_CFG_CONTROL,
2142 pci_lock | MC_CFG_LOCK);
2145 static void i7core_pci_ctl_create(struct i7core_pvt *pvt)
2147 pvt->i7core_pci = edac_pci_create_generic_ctl(
2148 &pvt->i7core_dev->pdev[0]->dev,
2149 EDAC_MOD_STR);
2150 if (unlikely(!pvt->i7core_pci))
2151 i7core_printk(KERN_WARNING,
2152 "Unable to setup PCI error report via EDAC\n");
2155 static void i7core_pci_ctl_release(struct i7core_pvt *pvt)
2157 if (likely(pvt->i7core_pci))
2158 edac_pci_release_generic_ctl(pvt->i7core_pci);
2159 else
2160 i7core_printk(KERN_ERR,
2161 "Couldn't find mem_ctl_info for socket %d\n",
2162 pvt->i7core_dev->socket);
2163 pvt->i7core_pci = NULL;
2166 static void i7core_unregister_mci(struct i7core_dev *i7core_dev)
2168 struct mem_ctl_info *mci = i7core_dev->mci;
2169 struct i7core_pvt *pvt;
2171 if (unlikely(!mci || !mci->pvt_info)) {
2172 edac_dbg(0, "MC: dev = %p\n", &i7core_dev->pdev[0]->dev);
2174 i7core_printk(KERN_ERR, "Couldn't find mci handler\n");
2175 return;
2178 pvt = mci->pvt_info;
2180 edac_dbg(0, "MC: mci = %p, dev = %p\n", mci, &i7core_dev->pdev[0]->dev);
2182 /* Disable scrubrate setting */
2183 if (pvt->enable_scrub)
2184 disable_sdram_scrub_setting(mci);
2186 /* Disable EDAC polling */
2187 i7core_pci_ctl_release(pvt);
2189 /* Remove MC sysfs nodes */
2190 i7core_delete_sysfs_devices(mci);
2191 edac_mc_del_mc(mci->pdev);
2193 edac_dbg(1, "%s: free mci struct\n", mci->ctl_name);
2194 kfree(mci->ctl_name);
2195 edac_mc_free(mci);
2196 i7core_dev->mci = NULL;
2199 static int i7core_register_mci(struct i7core_dev *i7core_dev)
2201 struct mem_ctl_info *mci;
2202 struct i7core_pvt *pvt;
2203 int rc;
2204 struct edac_mc_layer layers[2];
2206 /* allocate a new MC control structure */
2208 layers[0].type = EDAC_MC_LAYER_CHANNEL;
2209 layers[0].size = NUM_CHANS;
2210 layers[0].is_virt_csrow = false;
2211 layers[1].type = EDAC_MC_LAYER_SLOT;
2212 layers[1].size = MAX_DIMMS;
2213 layers[1].is_virt_csrow = true;
2214 mci = edac_mc_alloc(i7core_dev->socket, ARRAY_SIZE(layers), layers,
2215 sizeof(*pvt));
2216 if (unlikely(!mci))
2217 return -ENOMEM;
2219 edac_dbg(0, "MC: mci = %p, dev = %p\n", mci, &i7core_dev->pdev[0]->dev);
2221 pvt = mci->pvt_info;
2222 memset(pvt, 0, sizeof(*pvt));
2224 /* Associates i7core_dev and mci for future usage */
2225 pvt->i7core_dev = i7core_dev;
2226 i7core_dev->mci = mci;
2229 * FIXME: how to handle RDDR3 at MCI level? It is possible to have
2230 * Mixed RDDR3/UDDR3 with Nehalem, provided that they are on different
2231 * memory channels
2233 mci->mtype_cap = MEM_FLAG_DDR3;
2234 mci->edac_ctl_cap = EDAC_FLAG_NONE;
2235 mci->edac_cap = EDAC_FLAG_NONE;
2236 mci->mod_name = "i7core_edac.c";
2237 mci->mod_ver = I7CORE_REVISION;
2238 mci->ctl_name = kasprintf(GFP_KERNEL, "i7 core #%d",
2239 i7core_dev->socket);
2240 mci->dev_name = pci_name(i7core_dev->pdev[0]);
2241 mci->ctl_page_to_phys = NULL;
2243 /* Store pci devices at mci for faster access */
2244 rc = mci_bind_devs(mci, i7core_dev);
2245 if (unlikely(rc < 0))
2246 goto fail0;
2249 /* Get dimm basic config */
2250 get_dimm_config(mci);
2251 /* record ptr to the generic device */
2252 mci->pdev = &i7core_dev->pdev[0]->dev;
2253 /* Set the function pointer to an actual operation function */
2254 mci->edac_check = i7core_check_error;
2256 /* Enable scrubrate setting */
2257 if (pvt->enable_scrub)
2258 enable_sdram_scrub_setting(mci);
2260 /* add this new MC control structure to EDAC's list of MCs */
2261 if (unlikely(edac_mc_add_mc(mci))) {
2262 edac_dbg(0, "MC: failed edac_mc_add_mc()\n");
2263 /* FIXME: perhaps some code should go here that disables error
2264 * reporting if we just enabled it
2267 rc = -EINVAL;
2268 goto fail0;
2270 if (i7core_create_sysfs_devices(mci)) {
2271 edac_dbg(0, "MC: failed to create sysfs nodes\n");
2272 edac_mc_del_mc(mci->pdev);
2273 rc = -EINVAL;
2274 goto fail0;
2277 /* Default error mask is any memory */
2278 pvt->inject.channel = 0;
2279 pvt->inject.dimm = -1;
2280 pvt->inject.rank = -1;
2281 pvt->inject.bank = -1;
2282 pvt->inject.page = -1;
2283 pvt->inject.col = -1;
2285 /* allocating generic PCI control info */
2286 i7core_pci_ctl_create(pvt);
2288 /* DCLK for scrub rate setting */
2289 pvt->dclk_freq = get_dclk_freq();
2291 return 0;
2293 fail0:
2294 kfree(mci->ctl_name);
2295 edac_mc_free(mci);
2296 i7core_dev->mci = NULL;
2297 return rc;
2301 * i7core_probe Probe for ONE instance of device to see if it is
2302 * present.
2303 * return:
2304 * 0 for FOUND a device
2305 * < 0 for error code
2308 static int i7core_probe(struct pci_dev *pdev, const struct pci_device_id *id)
2310 int rc, count = 0;
2311 struct i7core_dev *i7core_dev;
2313 /* get the pci devices we want to reserve for our use */
2314 mutex_lock(&i7core_edac_lock);
2317 * All memory controllers are allocated at the first pass.
2319 if (unlikely(probed >= 1)) {
2320 mutex_unlock(&i7core_edac_lock);
2321 return -ENODEV;
2323 probed++;
2325 rc = i7core_get_all_devices();
2326 if (unlikely(rc < 0))
2327 goto fail0;
2329 list_for_each_entry(i7core_dev, &i7core_edac_list, list) {
2330 count++;
2331 rc = i7core_register_mci(i7core_dev);
2332 if (unlikely(rc < 0))
2333 goto fail1;
2337 * Nehalem-EX uses a different memory controller. However, as the
2338 * memory controller is not visible on some Nehalem/Nehalem-EP, we
2339 * need to indirectly probe via a X58 PCI device. The same devices
2340 * are found on (some) Nehalem-EX. So, on those machines, the
2341 * probe routine needs to return -ENODEV, as the actual Memory
2342 * Controller registers won't be detected.
2344 if (!count) {
2345 rc = -ENODEV;
2346 goto fail1;
2349 i7core_printk(KERN_INFO,
2350 "Driver loaded, %d memory controller(s) found.\n",
2351 count);
2353 mutex_unlock(&i7core_edac_lock);
2354 return 0;
2356 fail1:
2357 list_for_each_entry(i7core_dev, &i7core_edac_list, list)
2358 i7core_unregister_mci(i7core_dev);
2360 i7core_put_all_devices();
2361 fail0:
2362 mutex_unlock(&i7core_edac_lock);
2363 return rc;
2367 * i7core_remove destructor for one instance of device
2370 static void i7core_remove(struct pci_dev *pdev)
2372 struct i7core_dev *i7core_dev;
2374 edac_dbg(0, "\n");
2377 * we have a trouble here: pdev value for removal will be wrong, since
2378 * it will point to the X58 register used to detect that the machine
2379 * is a Nehalem or upper design. However, due to the way several PCI
2380 * devices are grouped together to provide MC functionality, we need
2381 * to use a different method for releasing the devices
2384 mutex_lock(&i7core_edac_lock);
2386 if (unlikely(!probed)) {
2387 mutex_unlock(&i7core_edac_lock);
2388 return;
2391 list_for_each_entry(i7core_dev, &i7core_edac_list, list)
2392 i7core_unregister_mci(i7core_dev);
2394 /* Release PCI resources */
2395 i7core_put_all_devices();
2397 probed--;
2399 mutex_unlock(&i7core_edac_lock);
2402 MODULE_DEVICE_TABLE(pci, i7core_pci_tbl);
2405 * i7core_driver pci_driver structure for this module
2408 static struct pci_driver i7core_driver = {
2409 .name = "i7core_edac",
2410 .probe = i7core_probe,
2411 .remove = i7core_remove,
2412 .id_table = i7core_pci_tbl,
2416 * i7core_init Module entry function
2417 * Try to initialize this module for its devices
2419 static int __init i7core_init(void)
2421 int pci_rc;
2423 edac_dbg(2, "\n");
2425 /* Ensure that the OPSTATE is set correctly for POLL or NMI */
2426 opstate_init();
2428 if (use_pci_fixup)
2429 i7core_xeon_pci_fixup(pci_dev_table);
2431 pci_rc = pci_register_driver(&i7core_driver);
2433 if (pci_rc >= 0) {
2434 mce_register_decode_chain(&i7_mce_dec);
2435 return 0;
2438 i7core_printk(KERN_ERR, "Failed to register device with error %d.\n",
2439 pci_rc);
2441 return pci_rc;
2445 * i7core_exit() Module exit function
2446 * Unregister the driver
2448 static void __exit i7core_exit(void)
2450 edac_dbg(2, "\n");
2451 pci_unregister_driver(&i7core_driver);
2452 mce_unregister_decode_chain(&i7_mce_dec);
2455 module_init(i7core_init);
2456 module_exit(i7core_exit);
2458 MODULE_LICENSE("GPL");
2459 MODULE_AUTHOR("Mauro Carvalho Chehab <mchehab@redhat.com>");
2460 MODULE_AUTHOR("Red Hat Inc. (http://www.redhat.com)");
2461 MODULE_DESCRIPTION("MC Driver for Intel i7 Core memory controllers - "
2462 I7CORE_REVISION);
2464 module_param(edac_op_state, int, 0444);
2465 MODULE_PARM_DESC(edac_op_state, "EDAC Error Reporting state: 0=Poll,1=NMI");