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 NUMBANK_MASK ((1 << 8) | (1 << 7))
113 #define NUMBANK(x) (((x) & NUMBANK_MASK) >> 7)
114 #define NUMRANK_MASK ((1 << 6) | (1 << 5))
115 #define NUMRANK(x) (((x) & NUMRANK_MASK) >> 5)
116 #define NUMROW_MASK ((1 << 4) | (1 << 3))
117 #define NUMROW(x) (((x) & NUMROW_MASK) >> 3)
118 #define NUMCOL_MASK 3
119 #define NUMCOL(x) ((x) & 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 maxnumdimms(struct i7core_pvt *pvt)
273 return (pvt->info.max_dod & 0x3) + 1;
276 static inline int maxnumrank(struct i7core_pvt *pvt)
278 static int ranks[4] = { 1, 2, 4, -EINVAL };
280 return ranks[(pvt->info.max_dod >> 2) & 0x3];
283 static inline int maxnumbank(struct i7core_pvt *pvt)
285 static int banks[4] = { 4, 8, 16, -EINVAL };
287 return banks[(pvt->info.max_dod >> 4) & 0x3];
290 static inline int maxnumrow(struct i7core_pvt *pvt)
292 static int rows[8] = {
293 1 << 12, 1 << 13, 1 << 14, 1 << 15,
294 1 << 16, -EINVAL, -EINVAL, -EINVAL,
297 return rows[((pvt->info.max_dod >> 6) & 0x7)];
300 static inline int maxnumcol(struct i7core_pvt *pvt)
302 static int cols[8] = {
303 1 << 10, 1 << 11, 1 << 12, -EINVAL,
305 return cols[((pvt->info.max_dod >> 9) & 0x3) << 12];
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;
334 /* Device 3 function 0 reads */
335 pci_read_config_dword(pdev, MC_STATUS, &status);
336 pci_read_config_dword(pdev, MC_CONTROL, &control);
338 for (i = 0; i < NUM_CHANS; i++) {
339 /* Check if the channel is active */
340 if (!(control & (1 << (8 + i))))
343 /* Check if the channel is disabled */
344 if (status & (1 << i)) {
354 static int get_dimm_config(struct mem_ctl_info *mci)
356 struct i7core_pvt *pvt = mci->pvt_info;
359 if (!pvt->pci_mcr[0])
362 /* Device 3 function 0 reads */
363 pci_read_config_dword(pvt->pci_mcr[0], MC_CONTROL,
364 &pvt->info.mc_control);
365 pci_read_config_dword(pvt->pci_mcr[0], MC_STATUS,
366 &pvt->info.mc_status);
367 pci_read_config_dword(pvt->pci_mcr[0], MC_MAX_DOD,
369 pci_read_config_dword(pvt->pci_mcr[0], MC_CHANNEL_MAPPER,
372 debugf0("MC control=0x%08x status=0x%08x dod=0x%08x map=0x%08x\n",
373 pvt->info.mc_control, pvt->info.mc_status,
374 pvt->info.max_dod, pvt->info.ch_map);
376 if (ECC_ENABLED(pvt))
377 debugf0("ECC enabled with x%d SDCC\n", ECCx8(pvt)?8:4);
379 debugf0("ECC disabled\n");
381 /* FIXME: need to handle the error codes */
382 debugf0("DOD Maximum limits: DIMMS: %d, %d-ranked, %d-banked\n",
383 maxnumdimms(pvt), maxnumrank(pvt), maxnumbank(pvt));
384 debugf0("DOD Maximum rows x colums = 0x%x x 0x%x\n",
385 maxnumrow(pvt), maxnumcol(pvt));
387 debugf0("Memory channel configuration:\n");
389 for (i = 0; i < NUM_CHANS; i++) {
392 if (!CH_ACTIVE(pvt, i)) {
393 debugf0("Channel %i is not active\n", i);
396 if (CH_DISABLED(pvt, i)) {
397 debugf0("Channel %i is disabled\n", i);
401 /* Devices 4-6 function 0 */
402 pci_read_config_dword(pvt->pci_ch[i][0],
403 MC_CHANNEL_DIMM_INIT_PARAMS, &data);
405 pvt->channel[i].ranks = (data & QUAD_RANK_PRESENT)? 4 : 2;
407 if (data & THREE_DIMMS_PRESENT)
408 pvt->channel[i].dimms = 3;
409 else if (data & SINGLE_QUAD_RANK_PRESENT)
410 pvt->channel[i].dimms = 1;
412 pvt->channel[i].dimms = 2;
414 debugf0("Ch%d (0x%08x): rd ch %d, wr ch %d, "
415 "%d ranks, %d %cDIMMs\n",
417 RDLCH(pvt->info.ch_map, i),
418 WRLCH(pvt->info.ch_map, i),
419 pvt->channel[i].ranks, pvt->channel[i].dimms,
420 (data & REGISTERED_DIMM)? 'R' : 'U' );
426 /****************************************************************************
427 Error insertion routines
428 ****************************************************************************/
430 /* The i7core has independent error injection features per channel.
431 However, to have a simpler code, we don't allow enabling error injection
432 on more than one channel.
433 Also, since a change at an inject parameter will be applied only at enable,
434 we're disabling error injection on all write calls to the sysfs nodes that
435 controls the error code injection.
437 static int disable_inject(struct mem_ctl_info *mci)
439 struct i7core_pvt *pvt = mci->pvt_info;
441 pvt->inject.enable = 0;
443 if (!pvt->pci_ch[pvt->inject.channel][0])
446 pci_write_config_dword(pvt->pci_ch[pvt->inject.channel][0],
447 MC_CHANNEL_ERROR_MASK, 0);
453 * i7core inject inject.section
455 * accept and store error injection inject.section value
456 * bit 0 - refers to the lower 32-byte half cacheline
457 * bit 1 - refers to the upper 32-byte half cacheline
459 static ssize_t i7core_inject_section_store(struct mem_ctl_info *mci,
460 const char *data, size_t count)
462 struct i7core_pvt *pvt = mci->pvt_info;
466 if (pvt->inject.enable)
469 rc = strict_strtoul(data, 10, &value);
470 if ((rc < 0) || (value > 3))
473 pvt->inject.section = (u32) value;
477 static ssize_t i7core_inject_section_show(struct mem_ctl_info *mci,
480 struct i7core_pvt *pvt = mci->pvt_info;
481 return sprintf(data, "0x%08x\n", pvt->inject.section);
487 * accept and store error injection inject.section value
488 * bit 0 - repeat enable - Enable error repetition
489 * bit 1 - inject ECC error
490 * bit 2 - inject parity error
492 static ssize_t i7core_inject_type_store(struct mem_ctl_info *mci,
493 const char *data, size_t count)
495 struct i7core_pvt *pvt = mci->pvt_info;
499 if (pvt->inject.enable)
502 rc = strict_strtoul(data, 10, &value);
503 if ((rc < 0) || (value > 7))
506 pvt->inject.type = (u32) value;
510 static ssize_t i7core_inject_type_show(struct mem_ctl_info *mci,
513 struct i7core_pvt *pvt = mci->pvt_info;
514 return sprintf(data, "0x%08x\n", pvt->inject.type);
518 * i7core_inject_inject.eccmask_store
520 * The type of error (UE/CE) will depend on the inject.eccmask value:
521 * Any bits set to a 1 will flip the corresponding ECC bit
522 * Correctable errors can be injected by flipping 1 bit or the bits within
523 * a symbol pair (2 consecutive aligned 8-bit pairs - i.e. 7:0 and 15:8 or
524 * 23:16 and 31:24). Flipping bits in two symbol pairs will cause an
525 * uncorrectable error to be injected.
527 static ssize_t i7core_inject_eccmask_store(struct mem_ctl_info *mci,
528 const char *data, size_t count)
530 struct i7core_pvt *pvt = mci->pvt_info;
534 if (pvt->inject.enable)
537 rc = strict_strtoul(data, 10, &value);
541 pvt->inject.eccmask = (u32) value;
545 static ssize_t i7core_inject_eccmask_show(struct mem_ctl_info *mci,
548 struct i7core_pvt *pvt = mci->pvt_info;
549 return sprintf(data, "0x%08x\n", pvt->inject.eccmask);
555 * The type of error (UE/CE) will depend on the inject.eccmask value:
556 * Any bits set to a 1 will flip the corresponding ECC bit
557 * Correctable errors can be injected by flipping 1 bit or the bits within
558 * a symbol pair (2 consecutive aligned 8-bit pairs - i.e. 7:0 and 15:8 or
559 * 23:16 and 31:24). Flipping bits in two symbol pairs will cause an
560 * uncorrectable error to be injected.
562 static ssize_t i7core_inject_addrmatch_store(struct mem_ctl_info *mci,
563 const char *data, size_t count)
565 struct i7core_pvt *pvt = mci->pvt_info;
570 if (pvt->inject.enable)
574 cmd = strsep((char **) &data, ":");
577 val = strsep((char **) &data, " \n\t");
581 if (!strcasecmp(val,"any"))
584 rc = strict_strtol(val, 10, &value);
585 if ((rc < 0) || (value < 0))
589 if (!strcasecmp(cmd,"channel")) {
591 pvt->inject.channel = value;
594 } else if (!strcasecmp(cmd,"dimm")) {
596 pvt->inject.dimm = value;
599 } else if (!strcasecmp(cmd,"rank")) {
601 pvt->inject.rank = value;
604 } else if (!strcasecmp(cmd,"bank")) {
606 pvt->inject.bank = value;
609 } else if (!strcasecmp(cmd,"page")) {
611 pvt->inject.page = value;
614 } else if (!strcasecmp(cmd,"col") ||
615 !strcasecmp(cmd,"column")) {
617 pvt->inject.col = value;
626 static ssize_t i7core_inject_addrmatch_show(struct mem_ctl_info *mci,
629 struct i7core_pvt *pvt = mci->pvt_info;
630 char channel[4], dimm[4], bank[4], rank[4], page[7], col[7];
632 if (pvt->inject.channel < 0)
633 sprintf(channel, "any");
635 sprintf(channel, "%d", pvt->inject.channel);
636 if (pvt->inject.dimm < 0)
637 sprintf(dimm, "any");
639 sprintf(dimm, "%d", pvt->inject.dimm);
640 if (pvt->inject.bank < 0)
641 sprintf(bank, "any");
643 sprintf(bank, "%d", pvt->inject.bank);
644 if (pvt->inject.rank < 0)
645 sprintf(rank, "any");
647 sprintf(rank, "%d", pvt->inject.rank);
648 if (pvt->inject.page < 0)
649 sprintf(page, "any");
651 sprintf(page, "0x%04x", pvt->inject.page);
652 if (pvt->inject.col < 0)
655 sprintf(col, "0x%04x", pvt->inject.col);
657 return sprintf(data, "channel: %s\ndimm: %s\nbank: %s\n"
658 "rank: %s\npage: %s\ncolumn: %s\n",
659 channel, dimm, bank, rank, page, col);
663 * This routine prepares the Memory Controller for error injection.
664 * The error will be injected when some process tries to write to the
665 * memory that matches the given criteria.
666 * The criteria can be set in terms of a mask where dimm, rank, bank, page
667 * and col can be specified.
668 * A -1 value for any of the mask items will make the MCU to ignore
669 * that matching criteria for error injection.
671 * It should be noticed that the error will only happen after a write operation
672 * on a memory that matches the condition. if REPEAT_EN is not enabled at
673 * inject mask, then it will produce just one error. Otherwise, it will repeat
674 * until the injectmask would be cleaned.
676 * FIXME: This routine assumes that MAXNUMDIMMS value of MC_MAX_DOD
677 * is reliable enough to check if the MC is using the
678 * three channels. However, this is not clear at the datasheet.
680 static ssize_t i7core_inject_enable_store(struct mem_ctl_info *mci,
681 const char *data, size_t count)
683 struct i7core_pvt *pvt = mci->pvt_info;
689 if (!pvt->pci_ch[pvt->inject.channel][0])
692 rc = strict_strtoul(data, 10, &enable);
697 pvt->inject.enable = 1;
703 /* Sets pvt->inject.dimm mask */
704 if (pvt->inject.dimm < 0)
707 if (pvt->channel[pvt->inject.channel].dimms > 2)
708 mask |= (pvt->inject.dimm & 0x3L) << 35;
710 mask |= (pvt->inject.dimm & 0x1L) << 36;
713 /* Sets pvt->inject.rank mask */
714 if (pvt->inject.rank < 0)
717 if (pvt->channel[pvt->inject.channel].dimms > 2)
718 mask |= (pvt->inject.rank & 0x1L) << 34;
720 mask |= (pvt->inject.rank & 0x3L) << 34;
723 /* Sets pvt->inject.bank mask */
724 if (pvt->inject.bank < 0)
727 mask |= (pvt->inject.bank & 0x15L) << 30;
729 /* Sets pvt->inject.page mask */
730 if (pvt->inject.page < 0)
733 mask |= (pvt->inject.page & 0xffffL) << 14;
735 /* Sets pvt->inject.column mask */
736 if (pvt->inject.col < 0)
739 mask |= (pvt->inject.col & 0x3fffL);
742 pci_write_config_qword(pvt->pci_ch[pvt->inject.channel][0],
743 MC_CHANNEL_ADDR_MATCH, mask);
745 pci_write_config_dword(pvt->pci_ch[pvt->inject.channel][0],
746 MC_CHANNEL_ADDR_MATCH, mask);
747 pci_write_config_dword(pvt->pci_ch[pvt->inject.channel][0],
748 MC_CHANNEL_ADDR_MATCH + 4, mask >> 32L);
754 pci_read_config_qword(pvt->pci_ch[pvt->inject.channel][0],
755 MC_CHANNEL_ADDR_MATCH, &rdmask);
756 debugf0("Inject addr match write 0x%016llx, read: 0x%016llx\n",
759 u32 rdmask1, rdmask2;
761 pci_read_config_dword(pvt->pci_ch[pvt->inject.channel][0],
762 MC_CHANNEL_ADDR_MATCH, &rdmask1);
763 pci_read_config_dword(pvt->pci_ch[pvt->inject.channel][0],
764 MC_CHANNEL_ADDR_MATCH + 4, &rdmask2);
766 debugf0("Inject addr match write 0x%016llx, read: 0x%08x%08x\n",
767 mask, rdmask1, rdmask2);
771 pci_write_config_dword(pvt->pci_ch[pvt->inject.channel][0],
772 MC_CHANNEL_ERROR_MASK, pvt->inject.eccmask);
776 * bits 1-2: MASK_HALF_CACHELINE
778 * bit 4: INJECT_ADDR_PARITY
781 injectmask = (pvt->inject.type & 1) |
782 (pvt->inject.section & 0x3) << 1 |
783 (pvt->inject.type & 0x6) << (3 - 1);
785 pci_write_config_dword(pvt->pci_ch[pvt->inject.channel][0],
786 MC_CHANNEL_ERROR_MASK, injectmask);
788 debugf0("Error inject addr match 0x%016llx, ecc 0x%08x, inject 0x%08x\n",
789 mask, pvt->inject.eccmask, injectmask);
796 static ssize_t i7core_inject_enable_show(struct mem_ctl_info *mci,
799 struct i7core_pvt *pvt = mci->pvt_info;
802 pci_read_config_dword(pvt->pci_ch[pvt->inject.channel][0],
803 MC_CHANNEL_ERROR_MASK, &injectmask);
805 debugf0("Inject error read: 0x%018x\n", injectmask);
807 if (injectmask & 0x0c)
808 pvt->inject.enable = 1;
810 return sprintf(data, "%d\n", pvt->inject.enable);
813 static ssize_t i7core_ce_regs_show(struct mem_ctl_info *mci, char *data)
815 struct i7core_pvt *pvt = mci->pvt_info;
817 if (!pvt->ce_count_available)
818 return sprintf(data, "unavailable\n");
820 return sprintf(data, "dimm0: %lu\ndimm1: %lu\ndimm2: %lu\n",
829 static struct mcidev_sysfs_attribute i7core_inj_attrs[] = {
833 .name = "inject_section",
834 .mode = (S_IRUGO | S_IWUSR)
836 .show = i7core_inject_section_show,
837 .store = i7core_inject_section_store,
840 .name = "inject_type",
841 .mode = (S_IRUGO | S_IWUSR)
843 .show = i7core_inject_type_show,
844 .store = i7core_inject_type_store,
847 .name = "inject_eccmask",
848 .mode = (S_IRUGO | S_IWUSR)
850 .show = i7core_inject_eccmask_show,
851 .store = i7core_inject_eccmask_store,
854 .name = "inject_addrmatch",
855 .mode = (S_IRUGO | S_IWUSR)
857 .show = i7core_inject_addrmatch_show,
858 .store = i7core_inject_addrmatch_store,
861 .name = "inject_enable",
862 .mode = (S_IRUGO | S_IWUSR)
864 .show = i7core_inject_enable_show,
865 .store = i7core_inject_enable_store,
868 .name = "corrected_error_counts",
869 .mode = (S_IRUGO | S_IWUSR)
871 .show = i7core_ce_regs_show,
876 /****************************************************************************
877 Device initialization routines: put/get, init/exit
878 ****************************************************************************/
881 * i7core_put_devices 'put' all the devices that we have
884 static void i7core_put_devices(void)
888 for (i = 0; i < N_DEVS; i++)
889 pci_dev_put(pci_devs[i].pdev);
893 * i7core_get_devices Find and perform 'get' operation on the MCH's
894 * device/functions we want to reference for this driver
896 * Need to 'get' device 16 func 1 and func 2
898 static int i7core_get_devices(void)
901 struct pci_dev *pdev = NULL;
903 for (i = 0; i < N_DEVS; i++) {
904 pdev = pci_get_device(PCI_VENDOR_ID_INTEL,
905 pci_devs[i].dev_id, NULL);
907 pci_devs[i].pdev = pdev;
909 i7core_printk(KERN_ERR,
910 "Device not found: PCI ID %04x:%04x "
911 "(dev %d, func %d)\n",
912 PCI_VENDOR_ID_INTEL, pci_devs[i].dev_id,
913 pci_devs[i].dev,pci_devs[i].func);
915 /* Dev 3 function 2 only exists on chips with RDIMMs */
916 if ((pci_devs[i].dev == 3) && (pci_devs[i].func == 2))
919 /* End of list, leave */
925 if (unlikely(PCI_SLOT(pdev->devfn) != pci_devs[i].dev ||
926 PCI_FUNC(pdev->devfn) != pci_devs[i].func)) {
927 i7core_printk(KERN_ERR,
928 "Device PCI ID %04x:%04x "
929 "has fn %d.%d instead of fn %d.%d\n",
930 PCI_VENDOR_ID_INTEL, pci_devs[i].dev_id,
931 PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn),
932 pci_devs[i].dev, pci_devs[i].func);
937 /* Be sure that the device is enabled */
938 rc = pci_enable_device(pdev);
939 if (unlikely(rc < 0)) {
940 i7core_printk(KERN_ERR,
941 "Couldn't enable PCI ID %04x:%04x "
943 PCI_VENDOR_ID_INTEL, pci_devs[i].dev_id,
944 PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn));
948 i7core_printk(KERN_INFO,
949 "Registered device %0x:%0x fn %d.%d\n",
950 PCI_VENDOR_ID_INTEL, pci_devs[i].dev_id,
951 PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn));
957 i7core_put_devices();
961 static int mci_bind_devs(struct mem_ctl_info *mci)
963 struct i7core_pvt *pvt = mci->pvt_info;
964 struct pci_dev *pdev;
967 for (i = 0; i < N_DEVS; i++) {
968 pdev = pci_devs[i].pdev;
972 func = PCI_FUNC(pdev->devfn);
973 slot = PCI_SLOT(pdev->devfn);
975 if (unlikely(func > MAX_MCR_FUNC))
977 pvt->pci_mcr[func] = pdev;
978 } else if (likely(slot >= 4 && slot < 4 + NUM_CHANS)) {
979 if (unlikely(func > MAX_CHAN_FUNC))
981 pvt->pci_ch[slot - 4][func] = pdev;
985 debugf0("Associated fn %d.%d, dev = %p\n",
986 PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn), pdev);
991 i7core_printk(KERN_ERR, "Device %d, function %d "
992 "is out of the expected range\n",
997 /****************************************************************************
999 ****************************************************************************/
1001 /* This function is based on the device 3 function 4 registers as described on:
1002 * Intel Xeon Processor 5500 Series Datasheet Volume 2
1003 * http://www.intel.com/Assets/PDF/datasheet/321322.pdf
1004 * also available at:
1005 * http://www.arrownac.com/manufacturers/intel/s/nehalem/5500-datasheet-v2.pdf
1007 static void check_mc_test_err(struct mem_ctl_info *mci)
1009 struct i7core_pvt *pvt = mci->pvt_info;
1011 int new0, new1, new2;
1013 if (!pvt->pci_mcr[4]) {
1014 debugf0("%s MCR registers not found\n",__func__);
1018 /* Corrected error reads */
1019 pci_read_config_dword(pvt->pci_mcr[4], MC_TEST_ERR_RCV1, &rcv1);
1020 pci_read_config_dword(pvt->pci_mcr[4], MC_TEST_ERR_RCV0, &rcv0);
1022 /* Store the new values */
1023 new2 = DIMM2_COR_ERR(rcv1);
1024 new1 = DIMM1_COR_ERR(rcv0);
1025 new0 = DIMM0_COR_ERR(rcv0);
1027 debugf2("%s CE rcv1=0x%08x rcv0=0x%08x, %d %d %d\n",
1028 (pvt->ce_count_available ? "UPDATE" : "READ"),
1029 rcv1, rcv0, new0, new1, new2);
1031 /* Updates CE counters if it is not the first time here */
1032 if (pvt->ce_count_available) {
1033 /* Updates CE counters */
1034 int add0, add1, add2;
1036 add2 = new2 - pvt->last_ce_count[2];
1037 add1 = new1 - pvt->last_ce_count[1];
1038 add0 = new0 - pvt->last_ce_count[0];
1042 pvt->ce_count[2] += add2;
1046 pvt->ce_count[1] += add1;
1050 pvt->ce_count[0] += add0;
1052 pvt->ce_count_available = 1;
1054 /* Store the new values */
1055 pvt->last_ce_count[2] = new2;
1056 pvt->last_ce_count[1] = new1;
1057 pvt->last_ce_count[0] = new0;
1061 * i7core_check_error Retrieve and process errors reported by the
1062 * hardware. Called by the Core module.
1064 static void i7core_check_error(struct mem_ctl_info *mci)
1066 check_mc_test_err(mci);
1070 * i7core_probe Probe for ONE instance of device to see if it is
1073 * 0 for FOUND a device
1074 * < 0 for error code
1076 static int __devinit i7core_probe(struct pci_dev *pdev,
1077 const struct pci_device_id *id)
1079 struct mem_ctl_info *mci;
1080 struct i7core_pvt *pvt;
1081 int num_channels = 0;
1083 int dev_idx = id->driver_data;
1085 if (unlikely(dev_idx >= ARRAY_SIZE(i7core_devs)))
1088 /* get the pci devices we want to reserve for our use */
1089 if (unlikely(i7core_get_devices() < 0))
1092 /* Check the number of active and not disabled channels */
1093 if (unlikely (i7core_get_active_channels(&num_channels)) < 0)
1096 /* FIXME: we currently don't know the number of csrows */
1097 num_csrows = num_channels;
1099 /* allocate a new MC control structure */
1100 mci = edac_mc_alloc(sizeof(*pvt), num_csrows, num_channels, 0);
1101 if (unlikely (!mci))
1104 debugf0("MC: " __FILE__ ": %s(): mci = %p\n", __func__, mci);
1106 mci->dev = &pdev->dev; /* record ptr to the generic device */
1108 pvt = mci->pvt_info;
1109 memset(pvt, 0, sizeof(*pvt));
1112 mci->mtype_cap = MEM_FLAG_DDR3; /* FIXME: how to handle RDDR3? */
1113 mci->edac_ctl_cap = EDAC_FLAG_NONE;
1114 mci->edac_cap = EDAC_FLAG_NONE;
1115 mci->mod_name = "i7core_edac.c";
1116 mci->mod_ver = I7CORE_REVISION;
1117 mci->ctl_name = i7core_devs[dev_idx].ctl_name;
1118 mci->dev_name = pci_name(pdev);
1119 mci->ctl_page_to_phys = NULL;
1120 mci->mc_driver_sysfs_attributes = i7core_inj_attrs;
1121 /* Set the function pointer to an actual operation function */
1122 mci->edac_check = i7core_check_error;
1124 /* Store pci devices at mci for faster access */
1125 if (unlikely (mci_bind_devs(mci)) < 0)
1128 /* Get dimm basic config */
1129 get_dimm_config(mci);
1131 /* add this new MC control structure to EDAC's list of MCs */
1132 if (unlikely(edac_mc_add_mc(mci)) < 0) {
1133 debugf0("MC: " __FILE__
1134 ": %s(): failed edac_mc_add_mc()\n", __func__);
1135 /* FIXME: perhaps some code should go here that disables error
1136 * reporting if we just enabled it
1141 /* allocating generic PCI control info */
1142 i7core_pci = edac_pci_create_generic_ctl(&pdev->dev, EDAC_MOD_STR);
1143 if (unlikely (!i7core_pci)) {
1145 "%s(): Unable to create PCI control\n",
1148 "%s(): PCI error report via EDAC not setup\n",
1152 /* Default error mask is any memory */
1153 pvt->inject.channel = 0;
1154 pvt->inject.dimm = -1;
1155 pvt->inject.rank = -1;
1156 pvt->inject.bank = -1;
1157 pvt->inject.page = -1;
1158 pvt->inject.col = -1;
1160 i7core_printk(KERN_INFO, "Driver loaded.\n");
1165 i7core_put_devices();
1173 * i7core_remove destructor for one instance of device
1176 static void __devexit i7core_remove(struct pci_dev *pdev)
1178 struct mem_ctl_info *mci;
1180 debugf0(__FILE__ ": %s()\n", __func__);
1183 edac_pci_release_generic_ctl(i7core_pci);
1185 mci = edac_mc_del_mc(&pdev->dev);
1190 /* retrieve references to resources, and free those resources */
1191 i7core_put_devices();
1196 MODULE_DEVICE_TABLE(pci, i7core_pci_tbl);
1199 * i7core_driver pci_driver structure for this module
1202 static struct pci_driver i7core_driver = {
1203 .name = "i7core_edac",
1204 .probe = i7core_probe,
1205 .remove = __devexit_p(i7core_remove),
1206 .id_table = i7core_pci_tbl,
1210 * i7core_init Module entry function
1211 * Try to initialize this module for its devices
1213 static int __init i7core_init(void)
1217 debugf2("MC: " __FILE__ ": %s()\n", __func__);
1219 /* Ensure that the OPSTATE is set correctly for POLL or NMI */
1222 pci_rc = pci_register_driver(&i7core_driver);
1224 return (pci_rc < 0) ? pci_rc : 0;
1228 * i7core_exit() Module exit function
1229 * Unregister the driver
1231 static void __exit i7core_exit(void)
1233 debugf2("MC: " __FILE__ ": %s()\n", __func__);
1234 pci_unregister_driver(&i7core_driver);
1237 module_init(i7core_init);
1238 module_exit(i7core_exit);
1240 MODULE_LICENSE("GPL");
1241 MODULE_AUTHOR("Mauro Carvalho Chehab <mchehab@redhat.com>");
1242 MODULE_AUTHOR("Red Hat Inc. (http://www.redhat.com)");
1243 MODULE_DESCRIPTION("MC Driver for Intel i7 Core memory controllers - "
1246 module_param(edac_op_state, int, 0444);
1247 MODULE_PARM_DESC(edac_op_state, "EDAC Error Reporting state: 0=Poll,1=NMI");