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>
30 #include <linux/edac_mce.h>
31 #include <linux/spinlock.h>
33 #include "edac_core.h"
35 /* To use the new pci_[read/write]_config_qword instead of two dword */
39 * Alter this version for the module when modifications are made
41 #define I7CORE_REVISION " Ver: 1.0.0 " __DATE__
42 #define EDAC_MOD_STR "i7core_edac"
44 /* HACK: temporary, just to enable all logs, for now */
46 #define debugf0(fmt, arg...) edac_printk(KERN_INFO, "i7core", fmt, ##arg)
51 #define i7core_printk(level, fmt, arg...) \
52 edac_printk(level, "i7core", fmt, ##arg)
54 #define i7core_mc_printk(mci, level, fmt, arg...) \
55 edac_mc_chipset_printk(mci, level, "i7core", fmt, ##arg)
58 * i7core Memory Controller Registers
61 /* OFFSETS for Device 0 Function 0 */
63 #define MC_CFG_CONTROL 0x90
65 /* OFFSETS for Device 3 Function 0 */
67 #define MC_CONTROL 0x48
68 #define MC_STATUS 0x4c
69 #define MC_MAX_DOD 0x64
72 * OFFSETS for Device 3 Function 4, as inicated on Xeon 5500 datasheet:
73 * http://www.arrownac.com/manufacturers/intel/s/nehalem/5500-datasheet-v2.pdf
76 #define MC_TEST_ERR_RCV1 0x60
77 #define DIMM2_COR_ERR(r) ((r) & 0x7fff)
79 #define MC_TEST_ERR_RCV0 0x64
80 #define DIMM1_COR_ERR(r) (((r) >> 16) & 0x7fff)
81 #define DIMM0_COR_ERR(r) ((r) & 0x7fff)
83 /* OFFSETS for Devices 4,5 and 6 Function 0 */
85 #define MC_CHANNEL_DIMM_INIT_PARAMS 0x58
86 #define THREE_DIMMS_PRESENT (1 << 24)
87 #define SINGLE_QUAD_RANK_PRESENT (1 << 23)
88 #define QUAD_RANK_PRESENT (1 << 22)
89 #define REGISTERED_DIMM (1 << 15)
91 #define MC_CHANNEL_MAPPER 0x60
92 #define RDLCH(r, ch) ((((r) >> (3 + (ch * 6))) & 0x07) - 1)
93 #define WRLCH(r, ch) ((((r) >> (ch * 6)) & 0x07) - 1)
95 #define MC_CHANNEL_RANK_PRESENT 0x7c
96 #define RANK_PRESENT_MASK 0xffff
98 #define MC_CHANNEL_ADDR_MATCH 0xf0
99 #define MC_CHANNEL_ERROR_MASK 0xf8
100 #define MC_CHANNEL_ERROR_INJECT 0xfc
101 #define INJECT_ADDR_PARITY 0x10
102 #define INJECT_ECC 0x08
103 #define MASK_CACHELINE 0x06
104 #define MASK_FULL_CACHELINE 0x06
105 #define MASK_MSB32_CACHELINE 0x04
106 #define MASK_LSB32_CACHELINE 0x02
107 #define NO_MASK_CACHELINE 0x00
108 #define REPEAT_EN 0x01
110 /* OFFSETS for Devices 4,5 and 6 Function 1 */
111 #define MC_DOD_CH_DIMM0 0x48
112 #define MC_DOD_CH_DIMM1 0x4c
113 #define MC_DOD_CH_DIMM2 0x50
114 #define RANKOFFSET_MASK ((1 << 12) | (1 << 11) | (1 << 10))
115 #define RANKOFFSET(x) ((x & RANKOFFSET_MASK) >> 10)
116 #define DIMM_PRESENT_MASK (1 << 9)
117 #define DIMM_PRESENT(x) (((x) & DIMM_PRESENT_MASK) >> 9)
118 #define MC_DOD_NUMBANK_MASK ((1 << 8) | (1 << 7))
119 #define MC_DOD_NUMBANK(x) (((x) & MC_DOD_NUMBANK_MASK) >> 7)
120 #define MC_DOD_NUMRANK_MASK ((1 << 6) | (1 << 5))
121 #define MC_DOD_NUMRANK(x) (((x) & MC_DOD_NUMRANK_MASK) >> 5)
122 #define MC_DOD_NUMROW_MASK ((1 << 4) | (1 << 3) | (1 << 2))
123 #define MC_DOD_NUMROW(x) (((x) & MC_DOD_NUMROW_MASK) >> 2)
124 #define MC_DOD_NUMCOL_MASK 3
125 #define MC_DOD_NUMCOL(x) ((x) & MC_DOD_NUMCOL_MASK)
127 #define MC_RANK_PRESENT 0x7c
129 #define MC_SAG_CH_0 0x80
130 #define MC_SAG_CH_1 0x84
131 #define MC_SAG_CH_2 0x88
132 #define MC_SAG_CH_3 0x8c
133 #define MC_SAG_CH_4 0x90
134 #define MC_SAG_CH_5 0x94
135 #define MC_SAG_CH_6 0x98
136 #define MC_SAG_CH_7 0x9c
138 #define MC_RIR_LIMIT_CH_0 0x40
139 #define MC_RIR_LIMIT_CH_1 0x44
140 #define MC_RIR_LIMIT_CH_2 0x48
141 #define MC_RIR_LIMIT_CH_3 0x4C
142 #define MC_RIR_LIMIT_CH_4 0x50
143 #define MC_RIR_LIMIT_CH_5 0x54
144 #define MC_RIR_LIMIT_CH_6 0x58
145 #define MC_RIR_LIMIT_CH_7 0x5C
146 #define MC_RIR_LIMIT_MASK ((1 << 10) - 1)
148 #define MC_RIR_WAY_CH 0x80
149 #define MC_RIR_WAY_OFFSET_MASK (((1 << 14) - 1) & ~0x7)
150 #define MC_RIR_WAY_RANK_MASK 0x7
157 #define MAX_DIMMS 3 /* Max DIMMS per channel */
158 #define MAX_MCR_FUNC 4
159 #define MAX_CHAN_FUNC 3
169 struct i7core_inject {
176 /* Error address mask */
177 int channel, dimm, rank, bank, page, col;
180 struct i7core_channel {
185 struct pci_id_descr {
189 struct pci_dev *pdev;
193 struct pci_dev *pci_noncore;
194 struct pci_dev *pci_mcr[MAX_MCR_FUNC + 1];
195 struct pci_dev *pci_ch[NUM_CHANS][MAX_CHAN_FUNC + 1];
196 struct i7core_info info;
197 struct i7core_inject inject;
198 struct i7core_channel channel[NUM_CHANS];
199 int channels; /* Number of active channels */
201 int ce_count_available;
202 unsigned long ce_count[MAX_DIMMS]; /* ECC corrected errors counts per dimm */
203 int last_ce_count[MAX_DIMMS];
206 struct edac_mce edac_mce;
207 struct mce mce_entry[MCE_LOG_LEN];
212 /* Device name and register DID (Device ID) */
213 struct i7core_dev_info {
214 const char *ctl_name; /* name for this device */
215 u16 fsb_mapping_errors; /* DID for the branchmap,control */
218 #define PCI_DESCR(device, function, device_id) \
220 .func = (function), \
221 .dev_id = (device_id)
223 struct pci_id_descr pci_devs[] = {
224 /* Memory controller */
225 { PCI_DESCR(3, 0, PCI_DEVICE_ID_INTEL_I7_MCR) },
226 { PCI_DESCR(3, 1, PCI_DEVICE_ID_INTEL_I7_MC_TAD) },
227 { PCI_DESCR(3, 2, PCI_DEVICE_ID_INTEL_I7_MC_RAS) }, /* if RDIMM is supported */
228 { PCI_DESCR(3, 4, PCI_DEVICE_ID_INTEL_I7_MC_TEST) },
230 /* Generic Non-core registers */
231 { PCI_DESCR(0, 0, PCI_DEVICE_ID_INTEL_I7_NOCORE) },
234 { PCI_DESCR(4, 0, PCI_DEVICE_ID_INTEL_I7_MC_CH0_CTRL) },
235 { PCI_DESCR(4, 1, PCI_DEVICE_ID_INTEL_I7_MC_CH0_ADDR) },
236 { PCI_DESCR(4, 2, PCI_DEVICE_ID_INTEL_I7_MC_CH0_RANK) },
237 { PCI_DESCR(4, 3, PCI_DEVICE_ID_INTEL_I7_MC_CH0_TC) },
240 { PCI_DESCR(5, 0, PCI_DEVICE_ID_INTEL_I7_MC_CH1_CTRL) },
241 { PCI_DESCR(5, 1, PCI_DEVICE_ID_INTEL_I7_MC_CH1_ADDR) },
242 { PCI_DESCR(5, 2, PCI_DEVICE_ID_INTEL_I7_MC_CH1_RANK) },
243 { PCI_DESCR(5, 3, PCI_DEVICE_ID_INTEL_I7_MC_CH1_TC) },
246 { PCI_DESCR(6, 0, PCI_DEVICE_ID_INTEL_I7_MC_CH2_CTRL) },
247 { PCI_DESCR(6, 1, PCI_DEVICE_ID_INTEL_I7_MC_CH2_ADDR) },
248 { PCI_DESCR(6, 2, PCI_DEVICE_ID_INTEL_I7_MC_CH2_RANK) },
249 { PCI_DESCR(6, 3, PCI_DEVICE_ID_INTEL_I7_MC_CH2_TC) },
251 #define N_DEVS ARRAY_SIZE(pci_devs)
254 * pci_device_id table for which devices we are looking for
255 * This should match the first device at pci_devs table
257 static const struct pci_device_id i7core_pci_tbl[] __devinitdata = {
258 {PCI_DEVICE(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_I7_MCR)},
259 {0,} /* 0 terminated list. */
263 /* Table of devices attributes supported by this driver */
264 static const struct i7core_dev_info i7core_devs[] = {
266 .ctl_name = "i7 Core",
267 .fsb_mapping_errors = PCI_DEVICE_ID_INTEL_I7_MCR,
271 static struct edac_pci_ctl_info *i7core_pci;
273 /****************************************************************************
274 Anciliary status routines
275 ****************************************************************************/
277 /* MC_CONTROL bits */
278 #define CH_ACTIVE(pvt, ch) ((pvt)->info.mc_control & (1 << (8 + ch)))
279 #define ECCx8(pvt) ((pvt)->info.mc_control & (1 << 1))
282 #define ECC_ENABLED(pvt) ((pvt)->info.mc_status & (1 << 3))
283 #define CH_DISABLED(pvt, ch) ((pvt)->info.mc_status & (1 << ch))
285 /* MC_MAX_DOD read functions */
286 static inline int numdimms(u32 dimms)
288 return (dimms & 0x3) + 1;
291 static inline int numrank(u32 rank)
293 static int ranks[4] = { 1, 2, 4, -EINVAL };
295 return ranks[rank & 0x3];
298 static inline int numbank(u32 bank)
300 static int banks[4] = { 4, 8, 16, -EINVAL };
302 return banks[bank & 0x3];
305 static inline int numrow(u32 row)
307 static int rows[8] = {
308 1 << 12, 1 << 13, 1 << 14, 1 << 15,
309 1 << 16, -EINVAL, -EINVAL, -EINVAL,
312 return rows[row & 0x7];
315 static inline int numcol(u32 col)
317 static int cols[8] = {
318 1 << 10, 1 << 11, 1 << 12, -EINVAL,
320 return cols[col & 0x3];
324 /****************************************************************************
325 Memory check routines
326 ****************************************************************************/
327 static struct pci_dev *get_pdev_slot_func(int slot, int func)
331 for (i = 0; i < N_DEVS; i++) {
332 if (!pci_devs[i].pdev)
335 if (PCI_SLOT(pci_devs[i].pdev->devfn) == slot &&
336 PCI_FUNC(pci_devs[i].pdev->devfn) == func) {
337 return pci_devs[i].pdev;
344 static int i7core_get_active_channels(int *channels, int *csrows)
346 struct pci_dev *pdev = NULL;
353 pdev = get_pdev_slot_func(3, 0);
355 i7core_printk(KERN_ERR, "Couldn't find fn 3.0!!!\n");
359 /* Device 3 function 0 reads */
360 pci_read_config_dword(pdev, MC_STATUS, &status);
361 pci_read_config_dword(pdev, MC_CONTROL, &control);
363 for (i = 0; i < NUM_CHANS; i++) {
365 /* Check if the channel is active */
366 if (!(control & (1 << (8 + i))))
369 /* Check if the channel is disabled */
370 if (status & (1 << i))
373 pdev = get_pdev_slot_func(i + 4, 1);
375 i7core_printk(KERN_ERR, "Couldn't find fn %d.%d!!!\n",
379 /* Devices 4-6 function 1 */
380 pci_read_config_dword(pdev,
381 MC_DOD_CH_DIMM0, &dimm_dod[0]);
382 pci_read_config_dword(pdev,
383 MC_DOD_CH_DIMM1, &dimm_dod[1]);
384 pci_read_config_dword(pdev,
385 MC_DOD_CH_DIMM2, &dimm_dod[2]);
389 for (j = 0; j < 3; j++) {
390 if (!DIMM_PRESENT(dimm_dod[j]))
396 debugf0("Number of active channels: %d\n", *channels);
401 static int get_dimm_config(struct mem_ctl_info *mci)
403 struct i7core_pvt *pvt = mci->pvt_info;
404 struct csrow_info *csr;
405 struct pci_dev *pdev;
407 unsigned long last_page = 0;
411 /* Get data from the MC register, function 0 */
412 pdev = pvt->pci_mcr[0];
416 /* Device 3 function 0 reads */
417 pci_read_config_dword(pdev, MC_CONTROL, &pvt->info.mc_control);
418 pci_read_config_dword(pdev, MC_STATUS, &pvt->info.mc_status);
419 pci_read_config_dword(pdev, MC_MAX_DOD, &pvt->info.max_dod);
420 pci_read_config_dword(pdev, MC_CHANNEL_MAPPER, &pvt->info.ch_map);
422 debugf0("MC control=0x%08x status=0x%08x dod=0x%08x map=0x%08x\n",
423 pvt->info.mc_control, pvt->info.mc_status,
424 pvt->info.max_dod, pvt->info.ch_map);
426 if (ECC_ENABLED(pvt)) {
427 debugf0("ECC enabled with x%d SDCC\n", ECCx8(pvt) ? 8 : 4);
429 mode = EDAC_S8ECD8ED;
431 mode = EDAC_S4ECD4ED;
433 debugf0("ECC disabled\n");
437 /* FIXME: need to handle the error codes */
438 debugf0("DOD Max limits: DIMMS: %d, %d-ranked, %d-banked\n",
439 numdimms(pvt->info.max_dod),
440 numrank(pvt->info.max_dod >> 2),
441 numbank(pvt->info.max_dod >> 4));
442 debugf0("DOD Max rows x colums = 0x%x x 0x%x\n",
443 numrow(pvt->info.max_dod >> 6),
444 numcol(pvt->info.max_dod >> 9));
446 debugf0("Memory channel configuration:\n");
448 for (i = 0; i < NUM_CHANS; i++) {
449 u32 data, dimm_dod[3], value[8];
451 if (!CH_ACTIVE(pvt, i)) {
452 debugf0("Channel %i is not active\n", i);
455 if (CH_DISABLED(pvt, i)) {
456 debugf0("Channel %i is disabled\n", i);
460 /* Devices 4-6 function 0 */
461 pci_read_config_dword(pvt->pci_ch[i][0],
462 MC_CHANNEL_DIMM_INIT_PARAMS, &data);
464 pvt->channel[i].ranks = (data & QUAD_RANK_PRESENT) ? 4 : 2;
466 if (data & REGISTERED_DIMM)
471 if (data & THREE_DIMMS_PRESENT)
472 pvt->channel[i].dimms = 3;
473 else if (data & SINGLE_QUAD_RANK_PRESENT)
474 pvt->channel[i].dimms = 1;
476 pvt->channel[i].dimms = 2;
479 /* Devices 4-6 function 1 */
480 pci_read_config_dword(pvt->pci_ch[i][1],
481 MC_DOD_CH_DIMM0, &dimm_dod[0]);
482 pci_read_config_dword(pvt->pci_ch[i][1],
483 MC_DOD_CH_DIMM1, &dimm_dod[1]);
484 pci_read_config_dword(pvt->pci_ch[i][1],
485 MC_DOD_CH_DIMM2, &dimm_dod[2]);
487 debugf0("Ch%d phy rd%d, wr%d (0x%08x): "
488 "%d ranks, %cDIMMs\n",
490 RDLCH(pvt->info.ch_map, i), WRLCH(pvt->info.ch_map, i),
492 pvt->channel[i].ranks,
493 (data & REGISTERED_DIMM) ? 'R' : 'U');
495 for (j = 0; j < 3; j++) {
496 u32 banks, ranks, rows, cols;
499 if (!DIMM_PRESENT(dimm_dod[j]))
502 banks = numbank(MC_DOD_NUMBANK(dimm_dod[j]));
503 ranks = numrank(MC_DOD_NUMRANK(dimm_dod[j]));
504 rows = numrow(MC_DOD_NUMROW(dimm_dod[j]));
505 cols = numcol(MC_DOD_NUMCOL(dimm_dod[j]));
507 /* DDR3 has 8 I/O banks */
508 size = (rows * cols * banks * ranks) >> (20 - 3);
510 pvt->channel[i].dimms++;
512 debugf0("\tdimm %d (0x%08x) %d Mb offset: %x, "
514 "numrank: %d, numrow: %#x, numcol: %#x\n",
515 j, dimm_dod[j], size,
516 RANKOFFSET(dimm_dod[j]),
517 banks, ranks, rows, cols);
520 npages = size >> (PAGE_SHIFT - 20);
522 npages = size << (20 - PAGE_SHIFT);
525 csr = &mci->csrows[csrow];
526 csr->first_page = last_page + 1;
528 csr->last_page = last_page;
529 csr->nr_pages = npages;
533 csr->csrow_idx = csrow;
534 csr->nr_channels = 1;
536 csr->channels[0].chan_idx = i;
537 csr->channels[0].ce_count = 0;
547 csr->dtype = DEV_X16;
550 csr->dtype = DEV_UNKNOWN;
553 csr->edac_mode = mode;
559 pci_read_config_dword(pdev, MC_SAG_CH_0, &value[0]);
560 pci_read_config_dword(pdev, MC_SAG_CH_1, &value[1]);
561 pci_read_config_dword(pdev, MC_SAG_CH_2, &value[2]);
562 pci_read_config_dword(pdev, MC_SAG_CH_3, &value[3]);
563 pci_read_config_dword(pdev, MC_SAG_CH_4, &value[4]);
564 pci_read_config_dword(pdev, MC_SAG_CH_5, &value[5]);
565 pci_read_config_dword(pdev, MC_SAG_CH_6, &value[6]);
566 pci_read_config_dword(pdev, MC_SAG_CH_7, &value[7]);
567 debugf0("\t[%i] DIVBY3\tREMOVED\tOFFSET\n", i);
568 for (j = 0; j < 8; j++)
569 debugf0("\t\t%#x\t%#x\t%#x\n",
570 (value[j] >> 27) & 0x1,
571 (value[j] >> 24) & 0x7,
572 (value[j] && ((1 << 24) - 1)));
578 /****************************************************************************
579 Error insertion routines
580 ****************************************************************************/
582 /* The i7core has independent error injection features per channel.
583 However, to have a simpler code, we don't allow enabling error injection
584 on more than one channel.
585 Also, since a change at an inject parameter will be applied only at enable,
586 we're disabling error injection on all write calls to the sysfs nodes that
587 controls the error code injection.
589 static int disable_inject(struct mem_ctl_info *mci)
591 struct i7core_pvt *pvt = mci->pvt_info;
593 pvt->inject.enable = 0;
595 if (!pvt->pci_ch[pvt->inject.channel][0])
598 pci_write_config_dword(pvt->pci_ch[pvt->inject.channel][0],
599 MC_CHANNEL_ERROR_MASK, 0);
605 * i7core inject inject.section
607 * accept and store error injection inject.section value
608 * bit 0 - refers to the lower 32-byte half cacheline
609 * bit 1 - refers to the upper 32-byte half cacheline
611 static ssize_t i7core_inject_section_store(struct mem_ctl_info *mci,
612 const char *data, size_t count)
614 struct i7core_pvt *pvt = mci->pvt_info;
618 if (pvt->inject.enable)
621 rc = strict_strtoul(data, 10, &value);
622 if ((rc < 0) || (value > 3))
625 pvt->inject.section = (u32) value;
629 static ssize_t i7core_inject_section_show(struct mem_ctl_info *mci,
632 struct i7core_pvt *pvt = mci->pvt_info;
633 return sprintf(data, "0x%08x\n", pvt->inject.section);
639 * accept and store error injection inject.section value
640 * bit 0 - repeat enable - Enable error repetition
641 * bit 1 - inject ECC error
642 * bit 2 - inject parity error
644 static ssize_t i7core_inject_type_store(struct mem_ctl_info *mci,
645 const char *data, size_t count)
647 struct i7core_pvt *pvt = mci->pvt_info;
651 if (pvt->inject.enable)
654 rc = strict_strtoul(data, 10, &value);
655 if ((rc < 0) || (value > 7))
658 pvt->inject.type = (u32) value;
662 static ssize_t i7core_inject_type_show(struct mem_ctl_info *mci,
665 struct i7core_pvt *pvt = mci->pvt_info;
666 return sprintf(data, "0x%08x\n", pvt->inject.type);
670 * i7core_inject_inject.eccmask_store
672 * The type of error (UE/CE) will depend on the inject.eccmask value:
673 * Any bits set to a 1 will flip the corresponding ECC bit
674 * Correctable errors can be injected by flipping 1 bit or the bits within
675 * a symbol pair (2 consecutive aligned 8-bit pairs - i.e. 7:0 and 15:8 or
676 * 23:16 and 31:24). Flipping bits in two symbol pairs will cause an
677 * uncorrectable error to be injected.
679 static ssize_t i7core_inject_eccmask_store(struct mem_ctl_info *mci,
680 const char *data, size_t count)
682 struct i7core_pvt *pvt = mci->pvt_info;
686 if (pvt->inject.enable)
689 rc = strict_strtoul(data, 10, &value);
693 pvt->inject.eccmask = (u32) value;
697 static ssize_t i7core_inject_eccmask_show(struct mem_ctl_info *mci,
700 struct i7core_pvt *pvt = mci->pvt_info;
701 return sprintf(data, "0x%08x\n", pvt->inject.eccmask);
707 * The type of error (UE/CE) will depend on the inject.eccmask value:
708 * Any bits set to a 1 will flip the corresponding ECC bit
709 * Correctable errors can be injected by flipping 1 bit or the bits within
710 * a symbol pair (2 consecutive aligned 8-bit pairs - i.e. 7:0 and 15:8 or
711 * 23:16 and 31:24). Flipping bits in two symbol pairs will cause an
712 * uncorrectable error to be injected.
714 static ssize_t i7core_inject_addrmatch_store(struct mem_ctl_info *mci,
715 const char *data, size_t count)
717 struct i7core_pvt *pvt = mci->pvt_info;
722 if (pvt->inject.enable)
726 cmd = strsep((char **) &data, ":");
729 val = strsep((char **) &data, " \n\t");
733 if (!strcasecmp(val, "any"))
736 rc = strict_strtol(val, 10, &value);
737 if ((rc < 0) || (value < 0))
741 if (!strcasecmp(cmd, "channel")) {
743 pvt->inject.channel = value;
746 } else if (!strcasecmp(cmd, "dimm")) {
748 pvt->inject.dimm = value;
751 } else if (!strcasecmp(cmd, "rank")) {
753 pvt->inject.rank = value;
756 } else if (!strcasecmp(cmd, "bank")) {
758 pvt->inject.bank = value;
761 } else if (!strcasecmp(cmd, "page")) {
763 pvt->inject.page = value;
766 } else if (!strcasecmp(cmd, "col") ||
767 !strcasecmp(cmd, "column")) {
769 pvt->inject.col = value;
778 static ssize_t i7core_inject_addrmatch_show(struct mem_ctl_info *mci,
781 struct i7core_pvt *pvt = mci->pvt_info;
782 char channel[4], dimm[4], bank[4], rank[4], page[7], col[7];
784 if (pvt->inject.channel < 0)
785 sprintf(channel, "any");
787 sprintf(channel, "%d", pvt->inject.channel);
788 if (pvt->inject.dimm < 0)
789 sprintf(dimm, "any");
791 sprintf(dimm, "%d", pvt->inject.dimm);
792 if (pvt->inject.bank < 0)
793 sprintf(bank, "any");
795 sprintf(bank, "%d", pvt->inject.bank);
796 if (pvt->inject.rank < 0)
797 sprintf(rank, "any");
799 sprintf(rank, "%d", pvt->inject.rank);
800 if (pvt->inject.page < 0)
801 sprintf(page, "any");
803 sprintf(page, "0x%04x", pvt->inject.page);
804 if (pvt->inject.col < 0)
807 sprintf(col, "0x%04x", pvt->inject.col);
809 return sprintf(data, "channel: %s\ndimm: %s\nbank: %s\n"
810 "rank: %s\npage: %s\ncolumn: %s\n",
811 channel, dimm, bank, rank, page, col);
815 * This routine prepares the Memory Controller for error injection.
816 * The error will be injected when some process tries to write to the
817 * memory that matches the given criteria.
818 * The criteria can be set in terms of a mask where dimm, rank, bank, page
819 * and col can be specified.
820 * A -1 value for any of the mask items will make the MCU to ignore
821 * that matching criteria for error injection.
823 * It should be noticed that the error will only happen after a write operation
824 * on a memory that matches the condition. if REPEAT_EN is not enabled at
825 * inject mask, then it will produce just one error. Otherwise, it will repeat
826 * until the injectmask would be cleaned.
828 * FIXME: This routine assumes that MAXNUMDIMMS value of MC_MAX_DOD
829 * is reliable enough to check if the MC is using the
830 * three channels. However, this is not clear at the datasheet.
832 static ssize_t i7core_inject_enable_store(struct mem_ctl_info *mci,
833 const char *data, size_t count)
835 struct i7core_pvt *pvt = mci->pvt_info;
841 if (!pvt->pci_ch[pvt->inject.channel][0])
844 rc = strict_strtoul(data, 10, &enable);
849 pvt->inject.enable = 1;
855 /* Sets pvt->inject.dimm mask */
856 if (pvt->inject.dimm < 0)
859 if (pvt->channel[pvt->inject.channel].dimms > 2)
860 mask |= (pvt->inject.dimm & 0x3L) << 35;
862 mask |= (pvt->inject.dimm & 0x1L) << 36;
865 /* Sets pvt->inject.rank mask */
866 if (pvt->inject.rank < 0)
869 if (pvt->channel[pvt->inject.channel].dimms > 2)
870 mask |= (pvt->inject.rank & 0x1L) << 34;
872 mask |= (pvt->inject.rank & 0x3L) << 34;
875 /* Sets pvt->inject.bank mask */
876 if (pvt->inject.bank < 0)
879 mask |= (pvt->inject.bank & 0x15L) << 30;
881 /* Sets pvt->inject.page mask */
882 if (pvt->inject.page < 0)
885 mask |= (pvt->inject.page & 0xffffL) << 14;
887 /* Sets pvt->inject.column mask */
888 if (pvt->inject.col < 0)
891 mask |= (pvt->inject.col & 0x3fffL);
893 /* Unlock writes to registers */
894 pci_write_config_dword(pvt->pci_noncore, MC_CFG_CONTROL, 0x2);
897 /* Zeroes error count registers */
898 pci_write_config_dword(pvt->pci_mcr[4], MC_TEST_ERR_RCV1, 0);
899 pci_write_config_dword(pvt->pci_mcr[4], MC_TEST_ERR_RCV0, 0);
900 pvt->ce_count_available = 0;
904 pci_write_config_qword(pvt->pci_ch[pvt->inject.channel][0],
905 MC_CHANNEL_ADDR_MATCH, mask);
907 pci_write_config_dword(pvt->pci_ch[pvt->inject.channel][0],
908 MC_CHANNEL_ADDR_MATCH, mask);
909 pci_write_config_dword(pvt->pci_ch[pvt->inject.channel][0],
910 MC_CHANNEL_ADDR_MATCH + 4, mask >> 32L);
916 pci_read_config_qword(pvt->pci_ch[pvt->inject.channel][0],
917 MC_CHANNEL_ADDR_MATCH, &rdmask);
918 debugf0("Inject addr match write 0x%016llx, read: 0x%016llx\n",
921 u32 rdmask1, rdmask2;
923 pci_read_config_dword(pvt->pci_ch[pvt->inject.channel][0],
924 MC_CHANNEL_ADDR_MATCH, &rdmask1);
925 pci_read_config_dword(pvt->pci_ch[pvt->inject.channel][0],
926 MC_CHANNEL_ADDR_MATCH + 4, &rdmask2);
928 debugf0("Inject addr match write 0x%016llx, read: 0x%08x 0x%08x\n",
929 mask, rdmask1, rdmask2);
933 pci_write_config_dword(pvt->pci_ch[pvt->inject.channel][0],
934 MC_CHANNEL_ERROR_MASK, pvt->inject.eccmask);
938 * bits 1-2: MASK_HALF_CACHELINE
940 * bit 4: INJECT_ADDR_PARITY
943 injectmask = (pvt->inject.type & 1) |
944 (pvt->inject.section & 0x3) << 1 |
945 (pvt->inject.type & 0x6) << (3 - 1);
947 pci_write_config_dword(pvt->pci_ch[pvt->inject.channel][0],
948 MC_CHANNEL_ERROR_MASK, injectmask);
951 /* lock writes to registers */
952 pci_write_config_dword(pvt->pci_noncore, MC_CFG_CONTROL, 0);
954 debugf0("Error inject addr match 0x%016llx, ecc 0x%08x,"
956 mask, pvt->inject.eccmask, injectmask);
962 static ssize_t i7core_inject_enable_show(struct mem_ctl_info *mci,
965 struct i7core_pvt *pvt = mci->pvt_info;
968 pci_read_config_dword(pvt->pci_ch[pvt->inject.channel][0],
969 MC_CHANNEL_ERROR_MASK, &injectmask);
971 debugf0("Inject error read: 0x%018x\n", injectmask);
973 if (injectmask & 0x0c)
974 pvt->inject.enable = 1;
976 return sprintf(data, "%d\n", pvt->inject.enable);
979 static ssize_t i7core_ce_regs_show(struct mem_ctl_info *mci, char *data)
981 struct i7core_pvt *pvt = mci->pvt_info;
983 if (!pvt->ce_count_available)
984 return sprintf(data, "unavailable\n");
986 return sprintf(data, "dimm0: %lu\ndimm1: %lu\ndimm2: %lu\n",
995 static struct mcidev_sysfs_attribute i7core_inj_attrs[] = {
999 .name = "inject_section",
1000 .mode = (S_IRUGO | S_IWUSR)
1002 .show = i7core_inject_section_show,
1003 .store = i7core_inject_section_store,
1006 .name = "inject_type",
1007 .mode = (S_IRUGO | S_IWUSR)
1009 .show = i7core_inject_type_show,
1010 .store = i7core_inject_type_store,
1013 .name = "inject_eccmask",
1014 .mode = (S_IRUGO | S_IWUSR)
1016 .show = i7core_inject_eccmask_show,
1017 .store = i7core_inject_eccmask_store,
1020 .name = "inject_addrmatch",
1021 .mode = (S_IRUGO | S_IWUSR)
1023 .show = i7core_inject_addrmatch_show,
1024 .store = i7core_inject_addrmatch_store,
1027 .name = "inject_enable",
1028 .mode = (S_IRUGO | S_IWUSR)
1030 .show = i7core_inject_enable_show,
1031 .store = i7core_inject_enable_store,
1034 .name = "corrected_error_counts",
1035 .mode = (S_IRUGO | S_IWUSR)
1037 .show = i7core_ce_regs_show,
1042 /****************************************************************************
1043 Device initialization routines: put/get, init/exit
1044 ****************************************************************************/
1047 * i7core_put_devices 'put' all the devices that we have
1048 * reserved via 'get'
1050 static void i7core_put_devices(void)
1054 for (i = 0; i < N_DEVS; i++)
1055 pci_dev_put(pci_devs[i].pdev);
1059 * i7core_get_devices Find and perform 'get' operation on the MCH's
1060 * device/functions we want to reference for this driver
1062 * Need to 'get' device 16 func 1 and func 2
1064 static int i7core_get_devices(void)
1067 struct pci_dev *pdev = NULL;
1069 for (i = 0; i < N_DEVS; i++) {
1070 pdev = pci_get_device(PCI_VENDOR_ID_INTEL,
1071 pci_devs[i].dev_id, NULL);
1073 pci_devs[i].pdev = pdev;
1075 i7core_printk(KERN_ERR,
1076 "Device not found: PCI ID %04x:%04x "
1077 "(dev %d, func %d)\n",
1078 PCI_VENDOR_ID_INTEL, pci_devs[i].dev_id,
1079 pci_devs[i].dev, pci_devs[i].func);
1081 /* Dev 3 function 2 only exists on chips with RDIMMs */
1082 if ((pci_devs[i].dev == 3) && (pci_devs[i].func == 2))
1085 /* End of list, leave */
1091 if (unlikely(PCI_SLOT(pdev->devfn) != pci_devs[i].dev ||
1092 PCI_FUNC(pdev->devfn) != pci_devs[i].func)) {
1093 i7core_printk(KERN_ERR,
1094 "Device PCI ID %04x:%04x "
1095 "has fn %d.%d instead of fn %d.%d\n",
1096 PCI_VENDOR_ID_INTEL, pci_devs[i].dev_id,
1097 PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn),
1098 pci_devs[i].dev, pci_devs[i].func);
1103 /* Be sure that the device is enabled */
1104 rc = pci_enable_device(pdev);
1105 if (unlikely(rc < 0)) {
1106 i7core_printk(KERN_ERR,
1107 "Couldn't enable PCI ID %04x:%04x "
1109 PCI_VENDOR_ID_INTEL, pci_devs[i].dev_id,
1110 PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn));
1114 i7core_printk(KERN_INFO,
1115 "Registered device %0x:%0x fn %d.%d\n",
1116 PCI_VENDOR_ID_INTEL, pci_devs[i].dev_id,
1117 PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn));
1123 i7core_put_devices();
1127 static int mci_bind_devs(struct mem_ctl_info *mci)
1129 struct i7core_pvt *pvt = mci->pvt_info;
1130 struct pci_dev *pdev;
1133 for (i = 0; i < N_DEVS; i++) {
1134 pdev = pci_devs[i].pdev;
1138 func = PCI_FUNC(pdev->devfn);
1139 slot = PCI_SLOT(pdev->devfn);
1141 if (unlikely(func > MAX_MCR_FUNC))
1143 pvt->pci_mcr[func] = pdev;
1144 } else if (likely(slot >= 4 && slot < 4 + NUM_CHANS)) {
1145 if (unlikely(func > MAX_CHAN_FUNC))
1147 pvt->pci_ch[slot - 4][func] = pdev;
1148 } else if (!slot && !func)
1149 pvt->pci_noncore = pdev;
1153 debugf0("Associated fn %d.%d, dev = %p\n",
1154 PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn), pdev);
1160 i7core_printk(KERN_ERR, "Device %d, function %d "
1161 "is out of the expected range\n",
1166 /****************************************************************************
1167 Error check routines
1168 ****************************************************************************/
1170 /* This function is based on the device 3 function 4 registers as described on:
1171 * Intel Xeon Processor 5500 Series Datasheet Volume 2
1172 * http://www.intel.com/Assets/PDF/datasheet/321322.pdf
1173 * also available at:
1174 * http://www.arrownac.com/manufacturers/intel/s/nehalem/5500-datasheet-v2.pdf
1176 static void check_mc_test_err(struct mem_ctl_info *mci)
1178 struct i7core_pvt *pvt = mci->pvt_info;
1180 int new0, new1, new2;
1182 if (!pvt->pci_mcr[4]) {
1183 debugf0("%s MCR registers not found\n",__func__);
1187 /* Corrected error reads */
1188 pci_read_config_dword(pvt->pci_mcr[4], MC_TEST_ERR_RCV1, &rcv1);
1189 pci_read_config_dword(pvt->pci_mcr[4], MC_TEST_ERR_RCV0, &rcv0);
1191 /* Store the new values */
1192 new2 = DIMM2_COR_ERR(rcv1);
1193 new1 = DIMM1_COR_ERR(rcv0);
1194 new0 = DIMM0_COR_ERR(rcv0);
1197 debugf2("%s CE rcv1=0x%08x rcv0=0x%08x, %d %d %d\n",
1198 (pvt->ce_count_available ? "UPDATE" : "READ"),
1199 rcv1, rcv0, new0, new1, new2);
1202 /* Updates CE counters if it is not the first time here */
1203 if (pvt->ce_count_available) {
1204 /* Updates CE counters */
1205 int add0, add1, add2;
1207 add2 = new2 - pvt->last_ce_count[2];
1208 add1 = new1 - pvt->last_ce_count[1];
1209 add0 = new0 - pvt->last_ce_count[0];
1213 pvt->ce_count[2] += add2;
1217 pvt->ce_count[1] += add1;
1221 pvt->ce_count[0] += add0;
1223 pvt->ce_count_available = 1;
1225 /* Store the new values */
1226 pvt->last_ce_count[2] = new2;
1227 pvt->last_ce_count[1] = new1;
1228 pvt->last_ce_count[0] = new0;
1231 static void i7core_mce_output_error(struct mem_ctl_info *mci,
1234 debugf0("CPU %d: Machine Check Exception: %16Lx"
1235 "Bank %d: %016Lx\n",
1236 m->cpu, m->mcgstatus, m->bank, m->status);
1238 debugf0("RIP%s %02x:<%016Lx>\n",
1239 !(m->mcgstatus & MCG_STATUS_EIPV) ? " !INEXACT!" : "",
1242 printk(KERN_EMERG "TSC %llx ", m->tsc);
1244 printk("ADDR %llx ", m->addr);
1246 printk("MISC %llx ", m->misc);
1249 snprintf(msg, sizeof(msg),
1250 "%s (Branch=%d DRAM-Bank=%d Buffer ID = %d RDWR=%s "
1251 "RAS=%d CAS=%d %s Err=0x%lx (%s))",
1252 type, branch >> 1, bank, buf_id, rdwr_str(rdwr), ras, cas,
1253 type, allErrors, error_name[errnum]);
1255 /* Call the helper to output message */
1256 edac_mc_handle_fbd_ue(mci, rank, channel, channel + 1, msg);
1261 * i7core_check_error Retrieve and process errors reported by the
1262 * hardware. Called by the Core module.
1264 static void i7core_check_error(struct mem_ctl_info *mci)
1266 struct i7core_pvt *pvt = mci->pvt_info;
1269 struct mce *m = NULL;
1270 unsigned long flags;
1272 debugf0(__FILE__ ": %s()\n", __func__);
1274 /* Copy all mce errors into a temporary buffer */
1275 spin_lock_irqsave(&pvt->mce_lock, flags);
1276 if (pvt->mce_count) {
1277 m = kmalloc(sizeof(*m) * pvt->mce_count, GFP_ATOMIC);
1279 count = pvt->mce_count;
1280 memcpy(m, &pvt->mce_entry, sizeof(*m) * count);
1284 spin_unlock_irqrestore(&pvt->mce_lock, flags);
1286 /* proccess mcelog errors */
1287 for (i = 0; i < count; i++)
1288 i7core_mce_output_error(mci, &m[i]);
1292 /* check memory count errors */
1293 check_mc_test_err(mci);
1297 * i7core_mce_check_error Replicates mcelog routine to get errors
1298 * This routine simply queues mcelog errors, and
1299 * return. The error itself should be handled later
1300 * by i7core_check_error.
1302 static int i7core_mce_check_error(void *priv, struct mce *mce)
1304 struct i7core_pvt *pvt = priv;
1305 unsigned long flags;
1307 debugf0(__FILE__ ": %s()\n", __func__);
1309 spin_lock_irqsave(&pvt->mce_lock, flags);
1310 if (pvt->mce_count < MCE_LOG_LEN) {
1311 memcpy(&pvt->mce_entry[pvt->mce_count], mce, sizeof(*mce));
1314 spin_unlock_irqrestore(&pvt->mce_lock, flags);
1316 /* Advice mcelog that the error were handled */
1318 return 0; // Let's duplicate the log
1322 * i7core_probe Probe for ONE instance of device to see if it is
1325 * 0 for FOUND a device
1326 * < 0 for error code
1328 static int __devinit i7core_probe(struct pci_dev *pdev,
1329 const struct pci_device_id *id)
1331 struct mem_ctl_info *mci;
1332 struct i7core_pvt *pvt;
1335 int dev_idx = id->driver_data;
1338 if (unlikely(dev_idx >= ARRAY_SIZE(i7core_devs)))
1341 /* get the pci devices we want to reserve for our use */
1342 rc = i7core_get_devices();
1343 if (unlikely(rc < 0))
1346 /* Check the number of active and not disabled channels */
1347 rc = i7core_get_active_channels(&num_channels, &num_csrows);
1348 if (unlikely(rc < 0))
1351 /* allocate a new MC control structure */
1352 mci = edac_mc_alloc(sizeof(*pvt), num_csrows, num_channels, 0);
1353 if (unlikely(!mci)) {
1358 debugf0("MC: " __FILE__ ": %s(): mci = %p\n", __func__, mci);
1360 mci->dev = &pdev->dev; /* record ptr to the generic device */
1362 pvt = mci->pvt_info;
1363 memset(pvt, 0, sizeof(*pvt));
1367 * FIXME: how to handle RDDR3 at MCI level? It is possible to have
1368 * Mixed RDDR3/UDDR3 with Nehalem, provided that they are on different
1371 mci->mtype_cap = MEM_FLAG_DDR3;
1372 mci->edac_ctl_cap = EDAC_FLAG_NONE;
1373 mci->edac_cap = EDAC_FLAG_NONE;
1374 mci->mod_name = "i7core_edac.c";
1375 mci->mod_ver = I7CORE_REVISION;
1376 mci->ctl_name = i7core_devs[dev_idx].ctl_name;
1377 mci->dev_name = pci_name(pdev);
1378 mci->ctl_page_to_phys = NULL;
1379 mci->mc_driver_sysfs_attributes = i7core_inj_attrs;
1380 /* Set the function pointer to an actual operation function */
1381 mci->edac_check = i7core_check_error;
1383 /* Store pci devices at mci for faster access */
1384 rc = mci_bind_devs(mci);
1385 if (unlikely(rc < 0))
1388 /* Get dimm basic config */
1389 get_dimm_config(mci);
1391 /* add this new MC control structure to EDAC's list of MCs */
1392 if (unlikely(edac_mc_add_mc(mci))) {
1393 debugf0("MC: " __FILE__
1394 ": %s(): failed edac_mc_add_mc()\n", __func__);
1395 /* FIXME: perhaps some code should go here that disables error
1396 * reporting if we just enabled it
1403 /* allocating generic PCI control info */
1404 i7core_pci = edac_pci_create_generic_ctl(&pdev->dev, EDAC_MOD_STR);
1405 if (unlikely(!i7core_pci)) {
1407 "%s(): Unable to create PCI control\n",
1410 "%s(): PCI error report via EDAC not setup\n",
1414 /* Default error mask is any memory */
1415 pvt->inject.channel = 0;
1416 pvt->inject.dimm = -1;
1417 pvt->inject.rank = -1;
1418 pvt->inject.bank = -1;
1419 pvt->inject.page = -1;
1420 pvt->inject.col = -1;
1422 /* Registers on edac_mce in order to receive memory errors */
1423 pvt->edac_mce.priv = pvt;
1424 pvt->edac_mce.check_error = i7core_mce_check_error;
1425 spin_lock_init(&pvt->mce_lock);
1427 rc = edac_mce_register(&pvt->edac_mce);
1428 if (unlikely (rc < 0)) {
1429 debugf0("MC: " __FILE__
1430 ": %s(): failed edac_mce_register()\n", __func__);
1434 i7core_printk(KERN_INFO, "Driver loaded.\n");
1442 i7core_put_devices();
1447 * i7core_remove destructor for one instance of device
1450 static void __devexit i7core_remove(struct pci_dev *pdev)
1452 struct mem_ctl_info *mci;
1453 struct i7core_pvt *pvt;
1455 debugf0(__FILE__ ": %s()\n", __func__);
1458 edac_pci_release_generic_ctl(i7core_pci);
1461 mci = edac_mc_del_mc(&pdev->dev);
1465 /* Unregisters on edac_mce in order to receive memory errors */
1466 pvt = mci->pvt_info;
1467 edac_mce_unregister(&pvt->edac_mce);
1469 /* retrieve references to resources, and free those resources */
1470 i7core_put_devices();
1475 MODULE_DEVICE_TABLE(pci, i7core_pci_tbl);
1478 * i7core_driver pci_driver structure for this module
1481 static struct pci_driver i7core_driver = {
1482 .name = "i7core_edac",
1483 .probe = i7core_probe,
1484 .remove = __devexit_p(i7core_remove),
1485 .id_table = i7core_pci_tbl,
1489 * i7core_init Module entry function
1490 * Try to initialize this module for its devices
1492 static int __init i7core_init(void)
1496 debugf2("MC: " __FILE__ ": %s()\n", __func__);
1498 /* Ensure that the OPSTATE is set correctly for POLL or NMI */
1501 pci_rc = pci_register_driver(&i7core_driver);
1503 return (pci_rc < 0) ? pci_rc : 0;
1507 * i7core_exit() Module exit function
1508 * Unregister the driver
1510 static void __exit i7core_exit(void)
1512 debugf2("MC: " __FILE__ ": %s()\n", __func__);
1513 pci_unregister_driver(&i7core_driver);
1516 module_init(i7core_init);
1517 module_exit(i7core_exit);
1519 MODULE_LICENSE("GPL");
1520 MODULE_AUTHOR("Mauro Carvalho Chehab <mchehab@redhat.com>");
1521 MODULE_AUTHOR("Red Hat Inc. (http://www.redhat.com)");
1522 MODULE_DESCRIPTION("MC Driver for Intel i7 Core memory controllers - "
1525 module_param(edac_op_state, int, 0444);
1526 MODULE_PARM_DESC(edac_op_state, "EDAC Error Reporting state: 0=Poll,1=NMI");