1 /* Intel 7 core Memory Controller kernel module (Nehalem)
3 * This file may be distributed under the terms of the
4 * GNU General Public License version 2 only.
6 * Copyright (c) 2009 by:
7 * Mauro Carvalho Chehab <mchehab@redhat.com>
9 * Red Hat Inc. http://www.redhat.com
11 * Forked and adapted from the i5400_edac driver
13 * Based on the following public Intel datasheets:
14 * Intel Core i7 Processor Extreme Edition and Intel Core i7 Processor
15 * Datasheet, Volume 2:
16 * http://download.intel.com/design/processor/datashts/320835.pdf
17 * Intel Xeon Processor 5500 Series Datasheet Volume 2
18 * http://www.intel.com/Assets/PDF/datasheet/321322.pdf
20 * http://www.arrownac.com/manufacturers/intel/s/nehalem/5500-datasheet-v2.pdf
23 #include <linux/module.h>
24 #include <linux/init.h>
25 #include <linux/pci.h>
26 #include <linux/pci_ids.h>
27 #include <linux/slab.h>
28 #include <linux/edac.h>
29 #include <linux/mmzone.h>
31 #include "edac_core.h"
33 /* To use the new pci_[read/write]_config_qword instead of two dword */
37 * Alter this version for the module when modifications are made
39 #define I7CORE_REVISION " Ver: 1.0.0 " __DATE__
40 #define EDAC_MOD_STR "i7core_edac"
42 /* HACK: temporary, just to enable all logs, for now */
44 #define debugf0(fmt, arg...) edac_printk(KERN_INFO, "i7core", fmt, ##arg)
49 #define i7core_printk(level, fmt, arg...) \
50 edac_printk(level, "i7core", fmt, ##arg)
52 #define i7core_mc_printk(mci, level, fmt, arg...) \
53 edac_mc_chipset_printk(mci, level, "i7core", fmt, ##arg)
56 * i7core Memory Controller Registers
59 /* OFFSETS for Device 3 Function 0 */
61 #define MC_CONTROL 0x48
62 #define MC_STATUS 0x4c
63 #define MC_MAX_DOD 0x64
66 * OFFSETS for Device 3 Function 4, as inicated on Xeon 5500 datasheet:
67 * http://www.arrownac.com/manufacturers/intel/s/nehalem/5500-datasheet-v2.pdf
70 #define MC_TEST_ERR_RCV1 0x60
71 #define DIMM2_COR_ERR(r) ((r) & 0x7fff)
73 #define MC_TEST_ERR_RCV0 0x64
74 #define DIMM1_COR_ERR(r) (((r) >> 16) & 0x7fff)
75 #define DIMM0_COR_ERR(r) ((r) & 0x7fff)
77 /* OFFSETS for Devices 4,5 and 6 Function 0 */
79 #define MC_CHANNEL_DIMM_INIT_PARAMS 0x58
80 #define THREE_DIMMS_PRESENT (1 << 24)
81 #define SINGLE_QUAD_RANK_PRESENT (1 << 23)
82 #define QUAD_RANK_PRESENT (1 << 22)
83 #define REGISTERED_DIMM (1 << 15)
85 #define MC_CHANNEL_MAPPER 0x60
86 #define RDLCH(r, ch) ((((r) >> (3 + (ch * 6))) & 0x07) - 1)
87 #define WRLCH(r, ch) ((((r) >> (ch * 6)) & 0x07) - 1)
89 #define MC_CHANNEL_RANK_PRESENT 0x7c
90 #define RANK_PRESENT_MASK 0xffff
92 #define MC_CHANNEL_ADDR_MATCH 0xf0
93 #define MC_CHANNEL_ERROR_MASK 0xf8
94 #define MC_CHANNEL_ERROR_INJECT 0xfc
95 #define INJECT_ADDR_PARITY 0x10
96 #define INJECT_ECC 0x08
97 #define MASK_CACHELINE 0x06
98 #define MASK_FULL_CACHELINE 0x06
99 #define MASK_MSB32_CACHELINE 0x04
100 #define MASK_LSB32_CACHELINE 0x02
101 #define NO_MASK_CACHELINE 0x00
102 #define REPEAT_EN 0x01
104 /* OFFSETS for Devices 4,5 and 6 Function 1 */
105 #define MC_DOD_CH_DIMM0 0x48
106 #define MC_DOD_CH_DIMM1 0x4c
107 #define MC_DOD_CH_DIMM2 0x50
108 #define RANKOFFSET_MASK ((1 << 12) | (1 << 11) | (1 << 10))
109 #define RANKOFFSET(x) ((x & RANKOFFSET_MASK) >> 10)
110 #define DIMM_PRESENT_MASK (1 << 9)
111 #define DIMM_PRESENT(x) (((x) & DIMM_PRESENT_MASK) >> 9)
112 #define MC_DOD_NUMBANK_MASK ((1 << 8) | (1 << 7))
113 #define MC_DOD_NUMBANK(x) (((x) & MC_DOD_NUMBANK_MASK) >> 7)
114 #define MC_DOD_NUMRANK_MASK ((1 << 6) | (1 << 5))
115 #define MC_DOD_NUMRANK(x) (((x) & MC_DOD_NUMRANK_MASK) >> 5)
116 #define MC_DOD_NUMROW_MASK ((1 << 4) | (1 << 3)| (1 << 2))
117 #define MC_DOD_NUMROW(x) (((x) & MC_DOD_NUMROW_MASK) >> 2)
118 #define MC_DOD_NUMCOL_MASK 3
119 #define MC_DOD_NUMCOL(x) ((x) & MC_DOD_NUMCOL_MASK)
121 #define MC_RANK_PRESENT 0x7c
123 #define MC_SAG_CH_0 0x80
124 #define MC_SAG_CH_1 0x84
125 #define MC_SAG_CH_2 0x88
126 #define MC_SAG_CH_3 0x8c
127 #define MC_SAG_CH_4 0x90
128 #define MC_SAG_CH_5 0x94
129 #define MC_SAG_CH_6 0x98
130 #define MC_SAG_CH_7 0x9c
132 #define MC_RIR_LIMIT_CH_0 0x40
133 #define MC_RIR_LIMIT_CH_1 0x44
134 #define MC_RIR_LIMIT_CH_2 0x48
135 #define MC_RIR_LIMIT_CH_3 0x4C
136 #define MC_RIR_LIMIT_CH_4 0x50
137 #define MC_RIR_LIMIT_CH_5 0x54
138 #define MC_RIR_LIMIT_CH_6 0x58
139 #define MC_RIR_LIMIT_CH_7 0x5C
140 #define MC_RIR_LIMIT_MASK ((1 << 10) - 1)
142 #define MC_RIR_WAY_CH 0x80
143 #define MC_RIR_WAY_OFFSET_MASK (((1 << 14) - 1) & ~0x7)
144 #define MC_RIR_WAY_RANK_MASK 0x7
151 #define MAX_DIMMS 3 /* Max DIMMS per channel */
152 #define MAX_MCR_FUNC 4
153 #define MAX_CHAN_FUNC 3
163 struct i7core_inject {
170 /* Error address mask */
171 int channel, dimm, rank, bank, page, col;
174 struct i7core_channel {
179 struct pci_id_descr {
183 struct pci_dev *pdev;
187 struct pci_dev *pci_mcr[MAX_MCR_FUNC + 1];
188 struct pci_dev *pci_ch[NUM_CHANS][MAX_CHAN_FUNC + 1];
189 struct i7core_info info;
190 struct i7core_inject inject;
191 struct i7core_channel channel[NUM_CHANS];
192 int channels; /* Number of active channels */
194 int ce_count_available;
195 unsigned long ce_count[MAX_DIMMS]; /* ECC corrected errors counts per dimm */
196 int last_ce_count[MAX_DIMMS];
200 /* Device name and register DID (Device ID) */
201 struct i7core_dev_info {
202 const char *ctl_name; /* name for this device */
203 u16 fsb_mapping_errors; /* DID for the branchmap,control */
206 #define PCI_DESCR(device, function, device_id) \
208 .func = (function), \
209 .dev_id = (device_id)
211 struct pci_id_descr pci_devs[] = {
212 /* Memory controller */
213 { PCI_DESCR(3, 0, PCI_DEVICE_ID_INTEL_I7_MCR) },
214 { PCI_DESCR(3, 1, PCI_DEVICE_ID_INTEL_I7_MC_TAD) },
215 { PCI_DESCR(3, 2, PCI_DEVICE_ID_INTEL_I7_MC_RAS) }, /* if RDIMM is supported */
216 { PCI_DESCR(3, 4, PCI_DEVICE_ID_INTEL_I7_MC_TEST) },
219 { PCI_DESCR(4, 0, PCI_DEVICE_ID_INTEL_I7_MC_CH0_CTRL) },
220 { PCI_DESCR(4, 1, PCI_DEVICE_ID_INTEL_I7_MC_CH0_ADDR) },
221 { PCI_DESCR(4, 2, PCI_DEVICE_ID_INTEL_I7_MC_CH0_RANK) },
222 { PCI_DESCR(4, 3, PCI_DEVICE_ID_INTEL_I7_MC_CH0_TC) },
225 { PCI_DESCR(5, 0, PCI_DEVICE_ID_INTEL_I7_MC_CH1_CTRL) },
226 { PCI_DESCR(5, 1, PCI_DEVICE_ID_INTEL_I7_MC_CH1_ADDR) },
227 { PCI_DESCR(5, 2, PCI_DEVICE_ID_INTEL_I7_MC_CH1_RANK) },
228 { PCI_DESCR(5, 3, PCI_DEVICE_ID_INTEL_I7_MC_CH1_TC) },
231 { PCI_DESCR(6, 0, PCI_DEVICE_ID_INTEL_I7_MC_CH2_CTRL) },
232 { PCI_DESCR(6, 1, PCI_DEVICE_ID_INTEL_I7_MC_CH2_ADDR) },
233 { PCI_DESCR(6, 2, PCI_DEVICE_ID_INTEL_I7_MC_CH2_RANK) },
234 { PCI_DESCR(6, 3, PCI_DEVICE_ID_INTEL_I7_MC_CH2_TC) },
236 #define N_DEVS ARRAY_SIZE(pci_devs)
239 * pci_device_id table for which devices we are looking for
240 * This should match the first device at pci_devs table
242 static const struct pci_device_id i7core_pci_tbl[] __devinitdata = {
243 {PCI_DEVICE(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_I7_MCR)},
244 {0,} /* 0 terminated list. */
248 /* Table of devices attributes supported by this driver */
249 static const struct i7core_dev_info i7core_devs[] = {
251 .ctl_name = "i7 Core",
252 .fsb_mapping_errors = PCI_DEVICE_ID_INTEL_I7_MCR,
256 static struct edac_pci_ctl_info *i7core_pci;
258 /****************************************************************************
259 Anciliary status routines
260 ****************************************************************************/
262 /* MC_CONTROL bits */
263 #define CH_ACTIVE(pvt, ch) ((pvt)->info.mc_control & (1 << (8 + ch)))
264 #define ECCx8(pvt) ((pvt)->info.mc_control & (1 << 1))
267 #define ECC_ENABLED(pvt) ((pvt)->info.mc_status & (1 << 3))
268 #define CH_DISABLED(pvt, ch) ((pvt)->info.mc_status & (1 << ch))
270 /* MC_MAX_DOD read functions */
271 static inline int numdimms(u32 dimms)
273 return (dimms & 0x3) + 1;
276 static inline int numrank(u32 rank)
278 static int ranks[4] = { 1, 2, 4, -EINVAL };
280 return ranks[rank & 0x3];
283 static inline int numbank(u32 bank)
285 static int banks[4] = { 4, 8, 16, -EINVAL };
287 return banks[bank & 0x3];
290 static inline int numrow(u32 row)
292 static int rows[8] = {
293 1 << 12, 1 << 13, 1 << 14, 1 << 15,
294 1 << 16, -EINVAL, -EINVAL, -EINVAL,
297 return rows[row & 0x7];
300 static inline int numcol(u32 col)
302 static int cols[8] = {
303 1 << 10, 1 << 11, 1 << 12, -EINVAL,
305 return cols[col & 0x3];
309 /****************************************************************************
310 Memory check routines
311 ****************************************************************************/
312 static int i7core_get_active_channels(int *channels)
314 struct pci_dev *pdev = NULL;
320 for (i = 0; i < N_DEVS; i++) {
321 if (!pci_devs[i].pdev)
324 if (PCI_SLOT(pci_devs[i].pdev->devfn) == 3 &&
325 PCI_FUNC(pci_devs[i].pdev->devfn) == 0) {
326 pdev = pci_devs[i].pdev;
332 i7core_printk(KERN_ERR, "Couldn't find fn 3.0!!!\n");
336 /* Device 3 function 0 reads */
337 pci_read_config_dword(pdev, MC_STATUS, &status);
338 pci_read_config_dword(pdev, MC_CONTROL, &control);
340 for (i = 0; i < NUM_CHANS; i++) {
341 /* Check if the channel is active */
342 if (!(control & (1 << (8 + i))))
345 /* Check if the channel is disabled */
346 if (status & (1 << i)) {
353 debugf0("Number of active channels: %d\n", *channels);
358 static int get_dimm_config(struct mem_ctl_info *mci)
360 struct i7core_pvt *pvt = mci->pvt_info;
361 struct csrow_info *csr;
362 struct pci_dev *pdev;
364 unsigned long last_page = 0;
368 /* Get data from the MC register, function 0 */
369 pdev = pvt->pci_mcr[0];
373 /* Device 3 function 0 reads */
374 pci_read_config_dword(pdev, MC_CONTROL, &pvt->info.mc_control);
375 pci_read_config_dword(pdev, MC_STATUS, &pvt->info.mc_status);
376 pci_read_config_dword(pdev, MC_MAX_DOD, &pvt->info.max_dod);
377 pci_read_config_dword(pdev, MC_CHANNEL_MAPPER, &pvt->info.ch_map);
379 debugf0("MC control=0x%08x status=0x%08x dod=0x%08x map=0x%08x\n",
380 pvt->info.mc_control, pvt->info.mc_status,
381 pvt->info.max_dod, pvt->info.ch_map);
383 if (ECC_ENABLED(pvt)) {
384 debugf0("ECC enabled with x%d SDCC\n", ECCx8(pvt) ?8:4);
386 mode = EDAC_S8ECD8ED;
388 mode = EDAC_S4ECD4ED;
390 debugf0("ECC disabled\n");
394 /* FIXME: need to handle the error codes */
395 debugf0("DOD Max limits: DIMMS: %d, %d-ranked, %d-banked\n",
396 numdimms(pvt->info.max_dod),
397 numrank(pvt->info.max_dod >> 2),
398 numbank(pvt->info.max_dod >> 4));
399 debugf0("DOD Max rows x colums = 0x%x x 0x%x\n",
400 numrow(pvt->info.max_dod >> 6),
401 numcol(pvt->info.max_dod >> 9));
403 debugf0("Memory channel configuration:\n");
405 for (i = 0; i < NUM_CHANS; i++) {
406 u32 data, dimm_dod[3], value[8];
408 if (!CH_ACTIVE(pvt, i)) {
409 debugf0("Channel %i is not active\n", i);
412 if (CH_DISABLED(pvt, i)) {
413 debugf0("Channel %i is disabled\n", i);
417 /* Devices 4-6 function 0 */
418 pci_read_config_dword(pvt->pci_ch[i][0],
419 MC_CHANNEL_DIMM_INIT_PARAMS, &data);
421 pvt->channel[i].ranks = (data & QUAD_RANK_PRESENT)? 4 : 2;
423 if (data & REGISTERED_DIMM)
428 if (data & THREE_DIMMS_PRESENT)
429 pvt->channel[i].dimms = 3;
430 else if (data & SINGLE_QUAD_RANK_PRESENT)
431 pvt->channel[i].dimms = 1;
433 pvt->channel[i].dimms = 2;
436 /* Devices 4-6 function 1 */
437 pci_read_config_dword(pvt->pci_ch[i][1],
438 MC_DOD_CH_DIMM0, &dimm_dod[0]);
439 pci_read_config_dword(pvt->pci_ch[i][1],
440 MC_DOD_CH_DIMM1, &dimm_dod[1]);
441 pci_read_config_dword(pvt->pci_ch[i][1],
442 MC_DOD_CH_DIMM2, &dimm_dod[2]);
444 debugf0("Ch%d phy rd%d, wr%d (0x%08x): "
445 "%d ranks, %cDIMMs\n",
447 RDLCH(pvt->info.ch_map, i), WRLCH(pvt->info.ch_map, i),
449 pvt->channel[i].ranks,
450 (data & REGISTERED_DIMM)? 'R' : 'U');
452 for (j = 0; j < 3; j++) {
453 u32 banks, ranks, rows, cols;
456 if (!DIMM_PRESENT(dimm_dod[j]))
459 banks = numbank(MC_DOD_NUMBANK(dimm_dod[j]));
460 ranks = numrank(MC_DOD_NUMRANK(dimm_dod[j]));
461 rows = numrow(MC_DOD_NUMROW(dimm_dod[j]));
462 cols = numcol(MC_DOD_NUMCOL(dimm_dod[j]));
464 /* DDR3 has 8 I/O banks */
465 size = (rows * cols * banks * ranks) >> (20 - 3);
467 pvt->channel[i].dimms++;
469 debugf0("\tdimm %d (0x%08x) %d Mb offset: %x, "
471 "numrank: %d, numrow: %#x, numcol: %#x\n",
472 j, dimm_dod[j], size,
473 RANKOFFSET(dimm_dod[j]),
474 banks, ranks, rows, cols);
476 npages = cols * rows; /* FIXME */
478 csr = &mci->csrows[csrow];
479 csr->first_page = last_page + 1;
481 csr->last_page = last_page;
482 csr->nr_pages = npages;
486 csr->csrow_idx = csrow;
496 csr->dtype = DEV_X16;
499 csr->dtype = DEV_UNKNOWN;
502 csr->edac_mode = mode;
508 pci_read_config_dword(pdev, MC_SAG_CH_0, &value[0]);
509 pci_read_config_dword(pdev, MC_SAG_CH_1, &value[1]);
510 pci_read_config_dword(pdev, MC_SAG_CH_2, &value[2]);
511 pci_read_config_dword(pdev, MC_SAG_CH_3, &value[3]);
512 pci_read_config_dword(pdev, MC_SAG_CH_4, &value[4]);
513 pci_read_config_dword(pdev, MC_SAG_CH_5, &value[5]);
514 pci_read_config_dword(pdev, MC_SAG_CH_6, &value[6]);
515 pci_read_config_dword(pdev, MC_SAG_CH_7, &value[7]);
516 printk("\t[%i] DIVBY3\tREMOVED\tOFFSET\n", i);
517 for (j = 0; j < 8; j++)
518 printk("\t\t%#x\t%#x\t%#x\n",
519 (value[j] >> 27) & 0x1,
520 (value[j] >> 24) & 0x7,
521 (value[j] && ((1 << 24) - 1)));
527 /****************************************************************************
528 Error insertion routines
529 ****************************************************************************/
531 /* The i7core has independent error injection features per channel.
532 However, to have a simpler code, we don't allow enabling error injection
533 on more than one channel.
534 Also, since a change at an inject parameter will be applied only at enable,
535 we're disabling error injection on all write calls to the sysfs nodes that
536 controls the error code injection.
538 static int disable_inject(struct mem_ctl_info *mci)
540 struct i7core_pvt *pvt = mci->pvt_info;
542 pvt->inject.enable = 0;
544 if (!pvt->pci_ch[pvt->inject.channel][0])
547 pci_write_config_dword(pvt->pci_ch[pvt->inject.channel][0],
548 MC_CHANNEL_ERROR_MASK, 0);
554 * i7core inject inject.section
556 * accept and store error injection inject.section value
557 * bit 0 - refers to the lower 32-byte half cacheline
558 * bit 1 - refers to the upper 32-byte half cacheline
560 static ssize_t i7core_inject_section_store(struct mem_ctl_info *mci,
561 const char *data, size_t count)
563 struct i7core_pvt *pvt = mci->pvt_info;
567 if (pvt->inject.enable)
570 rc = strict_strtoul(data, 10, &value);
571 if ((rc < 0) || (value > 3))
574 pvt->inject.section = (u32) value;
578 static ssize_t i7core_inject_section_show(struct mem_ctl_info *mci,
581 struct i7core_pvt *pvt = mci->pvt_info;
582 return sprintf(data, "0x%08x\n", pvt->inject.section);
588 * accept and store error injection inject.section value
589 * bit 0 - repeat enable - Enable error repetition
590 * bit 1 - inject ECC error
591 * bit 2 - inject parity error
593 static ssize_t i7core_inject_type_store(struct mem_ctl_info *mci,
594 const char *data, size_t count)
596 struct i7core_pvt *pvt = mci->pvt_info;
600 if (pvt->inject.enable)
603 rc = strict_strtoul(data, 10, &value);
604 if ((rc < 0) || (value > 7))
607 pvt->inject.type = (u32) value;
611 static ssize_t i7core_inject_type_show(struct mem_ctl_info *mci,
614 struct i7core_pvt *pvt = mci->pvt_info;
615 return sprintf(data, "0x%08x\n", pvt->inject.type);
619 * i7core_inject_inject.eccmask_store
621 * The type of error (UE/CE) will depend on the inject.eccmask value:
622 * Any bits set to a 1 will flip the corresponding ECC bit
623 * Correctable errors can be injected by flipping 1 bit or the bits within
624 * a symbol pair (2 consecutive aligned 8-bit pairs - i.e. 7:0 and 15:8 or
625 * 23:16 and 31:24). Flipping bits in two symbol pairs will cause an
626 * uncorrectable error to be injected.
628 static ssize_t i7core_inject_eccmask_store(struct mem_ctl_info *mci,
629 const char *data, size_t count)
631 struct i7core_pvt *pvt = mci->pvt_info;
635 if (pvt->inject.enable)
638 rc = strict_strtoul(data, 10, &value);
642 pvt->inject.eccmask = (u32) value;
646 static ssize_t i7core_inject_eccmask_show(struct mem_ctl_info *mci,
649 struct i7core_pvt *pvt = mci->pvt_info;
650 return sprintf(data, "0x%08x\n", pvt->inject.eccmask);
656 * The type of error (UE/CE) will depend on the inject.eccmask value:
657 * Any bits set to a 1 will flip the corresponding ECC bit
658 * Correctable errors can be injected by flipping 1 bit or the bits within
659 * a symbol pair (2 consecutive aligned 8-bit pairs - i.e. 7:0 and 15:8 or
660 * 23:16 and 31:24). Flipping bits in two symbol pairs will cause an
661 * uncorrectable error to be injected.
663 static ssize_t i7core_inject_addrmatch_store(struct mem_ctl_info *mci,
664 const char *data, size_t count)
666 struct i7core_pvt *pvt = mci->pvt_info;
671 if (pvt->inject.enable)
675 cmd = strsep((char **) &data, ":");
678 val = strsep((char **) &data, " \n\t");
682 if (!strcasecmp(val,"any"))
685 rc = strict_strtol(val, 10, &value);
686 if ((rc < 0) || (value < 0))
690 if (!strcasecmp(cmd,"channel")) {
692 pvt->inject.channel = value;
695 } else if (!strcasecmp(cmd,"dimm")) {
697 pvt->inject.dimm = value;
700 } else if (!strcasecmp(cmd,"rank")) {
702 pvt->inject.rank = value;
705 } else if (!strcasecmp(cmd,"bank")) {
707 pvt->inject.bank = value;
710 } else if (!strcasecmp(cmd,"page")) {
712 pvt->inject.page = value;
715 } else if (!strcasecmp(cmd,"col") ||
716 !strcasecmp(cmd,"column")) {
718 pvt->inject.col = value;
727 static ssize_t i7core_inject_addrmatch_show(struct mem_ctl_info *mci,
730 struct i7core_pvt *pvt = mci->pvt_info;
731 char channel[4], dimm[4], bank[4], rank[4], page[7], col[7];
733 if (pvt->inject.channel < 0)
734 sprintf(channel, "any");
736 sprintf(channel, "%d", pvt->inject.channel);
737 if (pvt->inject.dimm < 0)
738 sprintf(dimm, "any");
740 sprintf(dimm, "%d", pvt->inject.dimm);
741 if (pvt->inject.bank < 0)
742 sprintf(bank, "any");
744 sprintf(bank, "%d", pvt->inject.bank);
745 if (pvt->inject.rank < 0)
746 sprintf(rank, "any");
748 sprintf(rank, "%d", pvt->inject.rank);
749 if (pvt->inject.page < 0)
750 sprintf(page, "any");
752 sprintf(page, "0x%04x", pvt->inject.page);
753 if (pvt->inject.col < 0)
756 sprintf(col, "0x%04x", pvt->inject.col);
758 return sprintf(data, "channel: %s\ndimm: %s\nbank: %s\n"
759 "rank: %s\npage: %s\ncolumn: %s\n",
760 channel, dimm, bank, rank, page, col);
764 * This routine prepares the Memory Controller for error injection.
765 * The error will be injected when some process tries to write to the
766 * memory that matches the given criteria.
767 * The criteria can be set in terms of a mask where dimm, rank, bank, page
768 * and col can be specified.
769 * A -1 value for any of the mask items will make the MCU to ignore
770 * that matching criteria for error injection.
772 * It should be noticed that the error will only happen after a write operation
773 * on a memory that matches the condition. if REPEAT_EN is not enabled at
774 * inject mask, then it will produce just one error. Otherwise, it will repeat
775 * until the injectmask would be cleaned.
777 * FIXME: This routine assumes that MAXNUMDIMMS value of MC_MAX_DOD
778 * is reliable enough to check if the MC is using the
779 * three channels. However, this is not clear at the datasheet.
781 static ssize_t i7core_inject_enable_store(struct mem_ctl_info *mci,
782 const char *data, size_t count)
784 struct i7core_pvt *pvt = mci->pvt_info;
790 if (!pvt->pci_ch[pvt->inject.channel][0])
793 rc = strict_strtoul(data, 10, &enable);
798 pvt->inject.enable = 1;
804 /* Sets pvt->inject.dimm mask */
805 if (pvt->inject.dimm < 0)
808 if (pvt->channel[pvt->inject.channel].dimms > 2)
809 mask |= (pvt->inject.dimm & 0x3L) << 35;
811 mask |= (pvt->inject.dimm & 0x1L) << 36;
814 /* Sets pvt->inject.rank mask */
815 if (pvt->inject.rank < 0)
818 if (pvt->channel[pvt->inject.channel].dimms > 2)
819 mask |= (pvt->inject.rank & 0x1L) << 34;
821 mask |= (pvt->inject.rank & 0x3L) << 34;
824 /* Sets pvt->inject.bank mask */
825 if (pvt->inject.bank < 0)
828 mask |= (pvt->inject.bank & 0x15L) << 30;
830 /* Sets pvt->inject.page mask */
831 if (pvt->inject.page < 0)
834 mask |= (pvt->inject.page & 0xffffL) << 14;
836 /* Sets pvt->inject.column mask */
837 if (pvt->inject.col < 0)
840 mask |= (pvt->inject.col & 0x3fffL);
843 pci_write_config_qword(pvt->pci_ch[pvt->inject.channel][0],
844 MC_CHANNEL_ADDR_MATCH, mask);
846 pci_write_config_dword(pvt->pci_ch[pvt->inject.channel][0],
847 MC_CHANNEL_ADDR_MATCH, mask);
848 pci_write_config_dword(pvt->pci_ch[pvt->inject.channel][0],
849 MC_CHANNEL_ADDR_MATCH + 4, mask >> 32L);
855 pci_read_config_qword(pvt->pci_ch[pvt->inject.channel][0],
856 MC_CHANNEL_ADDR_MATCH, &rdmask);
857 debugf0("Inject addr match write 0x%016llx, read: 0x%016llx\n",
860 u32 rdmask1, rdmask2;
862 pci_read_config_dword(pvt->pci_ch[pvt->inject.channel][0],
863 MC_CHANNEL_ADDR_MATCH, &rdmask1);
864 pci_read_config_dword(pvt->pci_ch[pvt->inject.channel][0],
865 MC_CHANNEL_ADDR_MATCH + 4, &rdmask2);
867 debugf0("Inject addr match write 0x%016llx, read: 0x%08x%08x\n",
868 mask, rdmask1, rdmask2);
872 pci_write_config_dword(pvt->pci_ch[pvt->inject.channel][0],
873 MC_CHANNEL_ERROR_MASK, pvt->inject.eccmask);
877 * bits 1-2: MASK_HALF_CACHELINE
879 * bit 4: INJECT_ADDR_PARITY
882 injectmask = (pvt->inject.type & 1) |
883 (pvt->inject.section & 0x3) << 1 |
884 (pvt->inject.type & 0x6) << (3 - 1);
886 pci_write_config_dword(pvt->pci_ch[pvt->inject.channel][0],
887 MC_CHANNEL_ERROR_MASK, injectmask);
889 debugf0("Error inject addr match 0x%016llx, ecc 0x%08x, inject 0x%08x\n",
890 mask, pvt->inject.eccmask, injectmask);
897 static ssize_t i7core_inject_enable_show(struct mem_ctl_info *mci,
900 struct i7core_pvt *pvt = mci->pvt_info;
903 pci_read_config_dword(pvt->pci_ch[pvt->inject.channel][0],
904 MC_CHANNEL_ERROR_MASK, &injectmask);
906 debugf0("Inject error read: 0x%018x\n", injectmask);
908 if (injectmask & 0x0c)
909 pvt->inject.enable = 1;
911 return sprintf(data, "%d\n", pvt->inject.enable);
914 static ssize_t i7core_ce_regs_show(struct mem_ctl_info *mci, char *data)
916 struct i7core_pvt *pvt = mci->pvt_info;
918 if (!pvt->ce_count_available)
919 return sprintf(data, "unavailable\n");
921 return sprintf(data, "dimm0: %lu\ndimm1: %lu\ndimm2: %lu\n",
930 static struct mcidev_sysfs_attribute i7core_inj_attrs[] = {
934 .name = "inject_section",
935 .mode = (S_IRUGO | S_IWUSR)
937 .show = i7core_inject_section_show,
938 .store = i7core_inject_section_store,
941 .name = "inject_type",
942 .mode = (S_IRUGO | S_IWUSR)
944 .show = i7core_inject_type_show,
945 .store = i7core_inject_type_store,
948 .name = "inject_eccmask",
949 .mode = (S_IRUGO | S_IWUSR)
951 .show = i7core_inject_eccmask_show,
952 .store = i7core_inject_eccmask_store,
955 .name = "inject_addrmatch",
956 .mode = (S_IRUGO | S_IWUSR)
958 .show = i7core_inject_addrmatch_show,
959 .store = i7core_inject_addrmatch_store,
962 .name = "inject_enable",
963 .mode = (S_IRUGO | S_IWUSR)
965 .show = i7core_inject_enable_show,
966 .store = i7core_inject_enable_store,
969 .name = "corrected_error_counts",
970 .mode = (S_IRUGO | S_IWUSR)
972 .show = i7core_ce_regs_show,
977 /****************************************************************************
978 Device initialization routines: put/get, init/exit
979 ****************************************************************************/
982 * i7core_put_devices 'put' all the devices that we have
985 static void i7core_put_devices(void)
989 for (i = 0; i < N_DEVS; i++)
990 pci_dev_put(pci_devs[i].pdev);
994 * i7core_get_devices Find and perform 'get' operation on the MCH's
995 * device/functions we want to reference for this driver
997 * Need to 'get' device 16 func 1 and func 2
999 static int i7core_get_devices(void)
1002 struct pci_dev *pdev = NULL;
1004 for (i = 0; i < N_DEVS; i++) {
1005 pdev = pci_get_device(PCI_VENDOR_ID_INTEL,
1006 pci_devs[i].dev_id, NULL);
1008 pci_devs[i].pdev = pdev;
1010 i7core_printk(KERN_ERR,
1011 "Device not found: PCI ID %04x:%04x "
1012 "(dev %d, func %d)\n",
1013 PCI_VENDOR_ID_INTEL, pci_devs[i].dev_id,
1014 pci_devs[i].dev,pci_devs[i].func);
1016 /* Dev 3 function 2 only exists on chips with RDIMMs */
1017 if ((pci_devs[i].dev == 3) && (pci_devs[i].func == 2))
1020 /* End of list, leave */
1026 if (unlikely(PCI_SLOT(pdev->devfn) != pci_devs[i].dev ||
1027 PCI_FUNC(pdev->devfn) != pci_devs[i].func)) {
1028 i7core_printk(KERN_ERR,
1029 "Device PCI ID %04x:%04x "
1030 "has fn %d.%d instead of fn %d.%d\n",
1031 PCI_VENDOR_ID_INTEL, pci_devs[i].dev_id,
1032 PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn),
1033 pci_devs[i].dev, pci_devs[i].func);
1038 /* Be sure that the device is enabled */
1039 rc = pci_enable_device(pdev);
1040 if (unlikely(rc < 0)) {
1041 i7core_printk(KERN_ERR,
1042 "Couldn't enable PCI ID %04x:%04x "
1044 PCI_VENDOR_ID_INTEL, pci_devs[i].dev_id,
1045 PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn));
1049 i7core_printk(KERN_INFO,
1050 "Registered device %0x:%0x fn %d.%d\n",
1051 PCI_VENDOR_ID_INTEL, pci_devs[i].dev_id,
1052 PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn));
1058 i7core_put_devices();
1062 static int mci_bind_devs(struct mem_ctl_info *mci)
1064 struct i7core_pvt *pvt = mci->pvt_info;
1065 struct pci_dev *pdev;
1068 for (i = 0; i < N_DEVS; i++) {
1069 pdev = pci_devs[i].pdev;
1073 func = PCI_FUNC(pdev->devfn);
1074 slot = PCI_SLOT(pdev->devfn);
1076 if (unlikely(func > MAX_MCR_FUNC))
1078 pvt->pci_mcr[func] = pdev;
1079 } else if (likely(slot >= 4 && slot < 4 + NUM_CHANS)) {
1080 if (unlikely(func > MAX_CHAN_FUNC))
1082 pvt->pci_ch[slot - 4][func] = pdev;
1086 debugf0("Associated fn %d.%d, dev = %p\n",
1087 PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn), pdev);
1092 i7core_printk(KERN_ERR, "Device %d, function %d "
1093 "is out of the expected range\n",
1098 /****************************************************************************
1099 Error check routines
1100 ****************************************************************************/
1102 /* This function is based on the device 3 function 4 registers as described on:
1103 * Intel Xeon Processor 5500 Series Datasheet Volume 2
1104 * http://www.intel.com/Assets/PDF/datasheet/321322.pdf
1105 * also available at:
1106 * http://www.arrownac.com/manufacturers/intel/s/nehalem/5500-datasheet-v2.pdf
1108 static void check_mc_test_err(struct mem_ctl_info *mci)
1110 struct i7core_pvt *pvt = mci->pvt_info;
1112 int new0, new1, new2;
1114 if (!pvt->pci_mcr[4]) {
1115 debugf0("%s MCR registers not found\n",__func__);
1119 /* Corrected error reads */
1120 pci_read_config_dword(pvt->pci_mcr[4], MC_TEST_ERR_RCV1, &rcv1);
1121 pci_read_config_dword(pvt->pci_mcr[4], MC_TEST_ERR_RCV0, &rcv0);
1123 /* Store the new values */
1124 new2 = DIMM2_COR_ERR(rcv1);
1125 new1 = DIMM1_COR_ERR(rcv0);
1126 new0 = DIMM0_COR_ERR(rcv0);
1128 debugf2("%s CE rcv1=0x%08x rcv0=0x%08x, %d %d %d\n",
1129 (pvt->ce_count_available ? "UPDATE" : "READ"),
1130 rcv1, rcv0, new0, new1, new2);
1132 /* Updates CE counters if it is not the first time here */
1133 if (pvt->ce_count_available) {
1134 /* Updates CE counters */
1135 int add0, add1, add2;
1137 add2 = new2 - pvt->last_ce_count[2];
1138 add1 = new1 - pvt->last_ce_count[1];
1139 add0 = new0 - pvt->last_ce_count[0];
1143 pvt->ce_count[2] += add2;
1147 pvt->ce_count[1] += add1;
1151 pvt->ce_count[0] += add0;
1153 pvt->ce_count_available = 1;
1155 /* Store the new values */
1156 pvt->last_ce_count[2] = new2;
1157 pvt->last_ce_count[1] = new1;
1158 pvt->last_ce_count[0] = new0;
1162 * i7core_check_error Retrieve and process errors reported by the
1163 * hardware. Called by the Core module.
1165 static void i7core_check_error(struct mem_ctl_info *mci)
1167 check_mc_test_err(mci);
1171 * i7core_probe Probe for ONE instance of device to see if it is
1174 * 0 for FOUND a device
1175 * < 0 for error code
1177 static int __devinit i7core_probe(struct pci_dev *pdev,
1178 const struct pci_device_id *id)
1180 struct mem_ctl_info *mci;
1181 struct i7core_pvt *pvt;
1182 int num_channels = 0;
1184 int dev_idx = id->driver_data;
1187 if (unlikely(dev_idx >= ARRAY_SIZE(i7core_devs)))
1190 /* get the pci devices we want to reserve for our use */
1191 rc = i7core_get_devices();
1192 if (unlikely(rc < 0))
1195 /* Check the number of active and not disabled channels */
1196 rc = i7core_get_active_channels(&num_channels);
1197 if (unlikely (rc < 0))
1200 /* FIXME: we currently don't know the number of csrows */
1201 num_csrows = num_channels;
1203 /* allocate a new MC control structure */
1204 mci = edac_mc_alloc(sizeof(*pvt), num_csrows, num_channels, 0);
1205 if (unlikely (!mci)) {
1210 debugf0("MC: " __FILE__ ": %s(): mci = %p\n", __func__, mci);
1212 mci->dev = &pdev->dev; /* record ptr to the generic device */
1214 pvt = mci->pvt_info;
1215 memset(pvt, 0, sizeof(*pvt));
1218 mci->mtype_cap = MEM_FLAG_DDR3; /* FIXME: how to handle RDDR3? */
1219 mci->edac_ctl_cap = EDAC_FLAG_NONE;
1220 mci->edac_cap = EDAC_FLAG_NONE;
1221 mci->mod_name = "i7core_edac.c";
1222 mci->mod_ver = I7CORE_REVISION;
1223 mci->ctl_name = i7core_devs[dev_idx].ctl_name;
1224 mci->dev_name = pci_name(pdev);
1225 mci->ctl_page_to_phys = NULL;
1226 mci->mc_driver_sysfs_attributes = i7core_inj_attrs;
1227 /* Set the function pointer to an actual operation function */
1228 mci->edac_check = i7core_check_error;
1230 /* Store pci devices at mci for faster access */
1231 rc = mci_bind_devs(mci);
1232 if (unlikely (rc < 0))
1235 /* Get dimm basic config */
1236 get_dimm_config(mci);
1238 /* add this new MC control structure to EDAC's list of MCs */
1239 if (unlikely(edac_mc_add_mc(mci))) {
1240 debugf0("MC: " __FILE__
1241 ": %s(): failed edac_mc_add_mc()\n", __func__);
1242 /* FIXME: perhaps some code should go here that disables error
1243 * reporting if we just enabled it
1250 /* allocating generic PCI control info */
1251 i7core_pci = edac_pci_create_generic_ctl(&pdev->dev, EDAC_MOD_STR);
1252 if (unlikely (!i7core_pci)) {
1254 "%s(): Unable to create PCI control\n",
1257 "%s(): PCI error report via EDAC not setup\n",
1261 /* Default error mask is any memory */
1262 pvt->inject.channel = 0;
1263 pvt->inject.dimm = -1;
1264 pvt->inject.rank = -1;
1265 pvt->inject.bank = -1;
1266 pvt->inject.page = -1;
1267 pvt->inject.col = -1;
1269 i7core_printk(KERN_INFO, "Driver loaded.\n");
1277 i7core_put_devices();
1282 * i7core_remove destructor for one instance of device
1285 static void __devexit i7core_remove(struct pci_dev *pdev)
1287 struct mem_ctl_info *mci;
1289 debugf0(__FILE__ ": %s()\n", __func__);
1292 edac_pci_release_generic_ctl(i7core_pci);
1294 mci = edac_mc_del_mc(&pdev->dev);
1299 /* retrieve references to resources, and free those resources */
1300 i7core_put_devices();
1305 MODULE_DEVICE_TABLE(pci, i7core_pci_tbl);
1308 * i7core_driver pci_driver structure for this module
1311 static struct pci_driver i7core_driver = {
1312 .name = "i7core_edac",
1313 .probe = i7core_probe,
1314 .remove = __devexit_p(i7core_remove),
1315 .id_table = i7core_pci_tbl,
1319 * i7core_init Module entry function
1320 * Try to initialize this module for its devices
1322 static int __init i7core_init(void)
1326 debugf2("MC: " __FILE__ ": %s()\n", __func__);
1328 /* Ensure that the OPSTATE is set correctly for POLL or NMI */
1331 pci_rc = pci_register_driver(&i7core_driver);
1333 return (pci_rc < 0) ? pci_rc : 0;
1337 * i7core_exit() Module exit function
1338 * Unregister the driver
1340 static void __exit i7core_exit(void)
1342 debugf2("MC: " __FILE__ ": %s()\n", __func__);
1343 pci_unregister_driver(&i7core_driver);
1346 module_init(i7core_init);
1347 module_exit(i7core_exit);
1349 MODULE_LICENSE("GPL");
1350 MODULE_AUTHOR("Mauro Carvalho Chehab <mchehab@redhat.com>");
1351 MODULE_AUTHOR("Red Hat Inc. (http://www.redhat.com)");
1352 MODULE_DESCRIPTION("MC Driver for Intel i7 Core memory controllers - "
1355 module_param(edac_op_state, int, 0444);
1356 MODULE_PARM_DESC(edac_op_state, "EDAC Error Reporting state: 0=Poll,1=NMI");