1 /* Intel i7 core/Nehalem Memory Controller kernel module
3 * This driver supports yhe memory controllers found on the Intel
4 * processor families i7core, i7core 7xx/8xx, i5core, Xeon 35xx,
5 * Xeon 55xx and Xeon 56xx also known as Nehalem, Nehalem-EP, Lynnfield
8 * This file may be distributed under the terms of the
9 * GNU General Public License version 2 only.
11 * Copyright (c) 2009-2010 by:
12 * Mauro Carvalho Chehab <mchehab@redhat.com>
14 * Red Hat Inc. http://www.redhat.com
16 * Forked and adapted from the i5400_edac driver
18 * Based on the following public Intel datasheets:
19 * Intel Core i7 Processor Extreme Edition and Intel Core i7 Processor
20 * Datasheet, Volume 2:
21 * http://download.intel.com/design/processor/datashts/320835.pdf
22 * Intel Xeon Processor 5500 Series Datasheet Volume 2
23 * http://www.intel.com/Assets/PDF/datasheet/321322.pdf
25 * http://www.arrownac.com/manufacturers/intel/s/nehalem/5500-datasheet-v2.pdf
28 #include <linux/module.h>
29 #include <linux/init.h>
30 #include <linux/pci.h>
31 #include <linux/pci_ids.h>
32 #include <linux/slab.h>
33 #include <linux/delay.h>
34 #include <linux/edac.h>
35 #include <linux/mmzone.h>
36 #include <linux/edac_mce.h>
37 #include <linux/smp.h>
38 #include <asm/processor.h>
40 #include "edac_core.h"
43 static LIST_HEAD(i7core_edac_list);
44 static DEFINE_MUTEX(i7core_edac_lock);
48 * This is used for Nehalem-EP and Nehalem-EX devices, where the non-core
49 * registers start at bus 255, and are not reported by BIOS.
50 * We currently find devices with only 2 sockets. In order to support more QPI
51 * Quick Path Interconnect, just increment this number.
53 #define MAX_SOCKET_BUSES 2
57 * Alter this version for the module when modifications are made
59 #define I7CORE_REVISION " Ver: 1.0.0 " __DATE__
60 #define EDAC_MOD_STR "i7core_edac"
65 #define i7core_printk(level, fmt, arg...) \
66 edac_printk(level, "i7core", fmt, ##arg)
68 #define i7core_mc_printk(mci, level, fmt, arg...) \
69 edac_mc_chipset_printk(mci, level, "i7core", fmt, ##arg)
72 * i7core Memory Controller Registers
75 /* OFFSETS for Device 0 Function 0 */
77 #define MC_CFG_CONTROL 0x90
79 /* OFFSETS for Device 3 Function 0 */
81 #define MC_CONTROL 0x48
82 #define MC_STATUS 0x4c
83 #define MC_MAX_DOD 0x64
86 * OFFSETS for Device 3 Function 4, as inicated on Xeon 5500 datasheet:
87 * http://www.arrownac.com/manufacturers/intel/s/nehalem/5500-datasheet-v2.pdf
90 #define MC_TEST_ERR_RCV1 0x60
91 #define DIMM2_COR_ERR(r) ((r) & 0x7fff)
93 #define MC_TEST_ERR_RCV0 0x64
94 #define DIMM1_COR_ERR(r) (((r) >> 16) & 0x7fff)
95 #define DIMM0_COR_ERR(r) ((r) & 0x7fff)
97 /* OFFSETS for Device 3 Function 2, as inicated on Xeon 5500 datasheet */
98 #define MC_COR_ECC_CNT_0 0x80
99 #define MC_COR_ECC_CNT_1 0x84
100 #define MC_COR_ECC_CNT_2 0x88
101 #define MC_COR_ECC_CNT_3 0x8c
102 #define MC_COR_ECC_CNT_4 0x90
103 #define MC_COR_ECC_CNT_5 0x94
105 #define DIMM_TOP_COR_ERR(r) (((r) >> 16) & 0x7fff)
106 #define DIMM_BOT_COR_ERR(r) ((r) & 0x7fff)
109 /* OFFSETS for Devices 4,5 and 6 Function 0 */
111 #define MC_CHANNEL_DIMM_INIT_PARAMS 0x58
112 #define THREE_DIMMS_PRESENT (1 << 24)
113 #define SINGLE_QUAD_RANK_PRESENT (1 << 23)
114 #define QUAD_RANK_PRESENT (1 << 22)
115 #define REGISTERED_DIMM (1 << 15)
117 #define MC_CHANNEL_MAPPER 0x60
118 #define RDLCH(r, ch) ((((r) >> (3 + (ch * 6))) & 0x07) - 1)
119 #define WRLCH(r, ch) ((((r) >> (ch * 6)) & 0x07) - 1)
121 #define MC_CHANNEL_RANK_PRESENT 0x7c
122 #define RANK_PRESENT_MASK 0xffff
124 #define MC_CHANNEL_ADDR_MATCH 0xf0
125 #define MC_CHANNEL_ERROR_MASK 0xf8
126 #define MC_CHANNEL_ERROR_INJECT 0xfc
127 #define INJECT_ADDR_PARITY 0x10
128 #define INJECT_ECC 0x08
129 #define MASK_CACHELINE 0x06
130 #define MASK_FULL_CACHELINE 0x06
131 #define MASK_MSB32_CACHELINE 0x04
132 #define MASK_LSB32_CACHELINE 0x02
133 #define NO_MASK_CACHELINE 0x00
134 #define REPEAT_EN 0x01
136 /* OFFSETS for Devices 4,5 and 6 Function 1 */
138 #define MC_DOD_CH_DIMM0 0x48
139 #define MC_DOD_CH_DIMM1 0x4c
140 #define MC_DOD_CH_DIMM2 0x50
141 #define RANKOFFSET_MASK ((1 << 12) | (1 << 11) | (1 << 10))
142 #define RANKOFFSET(x) ((x & RANKOFFSET_MASK) >> 10)
143 #define DIMM_PRESENT_MASK (1 << 9)
144 #define DIMM_PRESENT(x) (((x) & DIMM_PRESENT_MASK) >> 9)
145 #define MC_DOD_NUMBANK_MASK ((1 << 8) | (1 << 7))
146 #define MC_DOD_NUMBANK(x) (((x) & MC_DOD_NUMBANK_MASK) >> 7)
147 #define MC_DOD_NUMRANK_MASK ((1 << 6) | (1 << 5))
148 #define MC_DOD_NUMRANK(x) (((x) & MC_DOD_NUMRANK_MASK) >> 5)
149 #define MC_DOD_NUMROW_MASK ((1 << 4) | (1 << 3) | (1 << 2))
150 #define MC_DOD_NUMROW(x) (((x) & MC_DOD_NUMROW_MASK) >> 2)
151 #define MC_DOD_NUMCOL_MASK 3
152 #define MC_DOD_NUMCOL(x) ((x) & MC_DOD_NUMCOL_MASK)
154 #define MC_RANK_PRESENT 0x7c
156 #define MC_SAG_CH_0 0x80
157 #define MC_SAG_CH_1 0x84
158 #define MC_SAG_CH_2 0x88
159 #define MC_SAG_CH_3 0x8c
160 #define MC_SAG_CH_4 0x90
161 #define MC_SAG_CH_5 0x94
162 #define MC_SAG_CH_6 0x98
163 #define MC_SAG_CH_7 0x9c
165 #define MC_RIR_LIMIT_CH_0 0x40
166 #define MC_RIR_LIMIT_CH_1 0x44
167 #define MC_RIR_LIMIT_CH_2 0x48
168 #define MC_RIR_LIMIT_CH_3 0x4C
169 #define MC_RIR_LIMIT_CH_4 0x50
170 #define MC_RIR_LIMIT_CH_5 0x54
171 #define MC_RIR_LIMIT_CH_6 0x58
172 #define MC_RIR_LIMIT_CH_7 0x5C
173 #define MC_RIR_LIMIT_MASK ((1 << 10) - 1)
175 #define MC_RIR_WAY_CH 0x80
176 #define MC_RIR_WAY_OFFSET_MASK (((1 << 14) - 1) & ~0x7)
177 #define MC_RIR_WAY_RANK_MASK 0x7
184 #define MAX_DIMMS 3 /* Max DIMMS per channel */
185 #define MAX_MCR_FUNC 4
186 #define MAX_CHAN_FUNC 3
196 struct i7core_inject {
203 /* Error address mask */
204 int channel, dimm, rank, bank, page, col;
207 struct i7core_channel {
212 struct pci_id_descr {
219 struct pci_id_table {
220 struct pci_id_descr *descr;
225 struct list_head list;
227 struct pci_dev **pdev;
229 struct mem_ctl_info *mci;
233 struct pci_dev *pci_noncore;
234 struct pci_dev *pci_mcr[MAX_MCR_FUNC + 1];
235 struct pci_dev *pci_ch[NUM_CHANS][MAX_CHAN_FUNC + 1];
237 struct i7core_dev *i7core_dev;
239 struct i7core_info info;
240 struct i7core_inject inject;
241 struct i7core_channel channel[NUM_CHANS];
243 int channels; /* Number of active channels */
245 int ce_count_available;
246 int csrow_map[NUM_CHANS][MAX_DIMMS];
248 /* ECC corrected errors counts per udimm */
249 unsigned long udimm_ce_count[MAX_DIMMS];
250 int udimm_last_ce_count[MAX_DIMMS];
251 /* ECC corrected errors counts per rdimm */
252 unsigned long rdimm_ce_count[NUM_CHANS][MAX_DIMMS];
253 int rdimm_last_ce_count[NUM_CHANS][MAX_DIMMS];
255 unsigned int is_registered;
258 struct edac_mce edac_mce;
260 /* Fifo double buffers */
261 struct mce mce_entry[MCE_LOG_LEN];
262 struct mce mce_outentry[MCE_LOG_LEN];
264 /* Fifo in/out counters */
265 unsigned mce_in, mce_out;
267 /* Count indicator to show errors not got */
268 unsigned mce_overrun;
270 /* Struct to control EDAC polling */
271 struct edac_pci_ctl_info *i7core_pci;
274 #define PCI_DESCR(device, function, device_id) \
276 .func = (function), \
277 .dev_id = (device_id)
279 struct pci_id_descr pci_dev_descr_i7core_nehalem[] = {
280 /* Memory controller */
281 { PCI_DESCR(3, 0, PCI_DEVICE_ID_INTEL_I7_MCR) },
282 { PCI_DESCR(3, 1, PCI_DEVICE_ID_INTEL_I7_MC_TAD) },
283 /* Exists only for RDIMM */
284 { PCI_DESCR(3, 2, PCI_DEVICE_ID_INTEL_I7_MC_RAS), .optional = 1 },
285 { PCI_DESCR(3, 4, PCI_DEVICE_ID_INTEL_I7_MC_TEST) },
288 { PCI_DESCR(4, 0, PCI_DEVICE_ID_INTEL_I7_MC_CH0_CTRL) },
289 { PCI_DESCR(4, 1, PCI_DEVICE_ID_INTEL_I7_MC_CH0_ADDR) },
290 { PCI_DESCR(4, 2, PCI_DEVICE_ID_INTEL_I7_MC_CH0_RANK) },
291 { PCI_DESCR(4, 3, PCI_DEVICE_ID_INTEL_I7_MC_CH0_TC) },
294 { PCI_DESCR(5, 0, PCI_DEVICE_ID_INTEL_I7_MC_CH1_CTRL) },
295 { PCI_DESCR(5, 1, PCI_DEVICE_ID_INTEL_I7_MC_CH1_ADDR) },
296 { PCI_DESCR(5, 2, PCI_DEVICE_ID_INTEL_I7_MC_CH1_RANK) },
297 { PCI_DESCR(5, 3, PCI_DEVICE_ID_INTEL_I7_MC_CH1_TC) },
300 { PCI_DESCR(6, 0, PCI_DEVICE_ID_INTEL_I7_MC_CH2_CTRL) },
301 { PCI_DESCR(6, 1, PCI_DEVICE_ID_INTEL_I7_MC_CH2_ADDR) },
302 { PCI_DESCR(6, 2, PCI_DEVICE_ID_INTEL_I7_MC_CH2_RANK) },
303 { PCI_DESCR(6, 3, PCI_DEVICE_ID_INTEL_I7_MC_CH2_TC) },
305 /* Generic Non-core registers */
307 * This is the PCI device on i7core and on Xeon 35xx (8086:2c41)
308 * On Xeon 55xx, however, it has a different id (8086:2c40). So,
309 * the probing code needs to test for the other address in case of
310 * failure of this one
312 { PCI_DESCR(0, 0, PCI_DEVICE_ID_INTEL_I7_NONCORE) },
316 struct pci_id_descr pci_dev_descr_lynnfield[] = {
317 { PCI_DESCR( 3, 0, PCI_DEVICE_ID_INTEL_LYNNFIELD_MCR) },
318 { PCI_DESCR( 3, 1, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_TAD) },
319 { PCI_DESCR( 3, 4, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_TEST) },
321 { PCI_DESCR( 4, 0, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH0_CTRL) },
322 { PCI_DESCR( 4, 1, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH0_ADDR) },
323 { PCI_DESCR( 4, 2, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH0_RANK) },
324 { PCI_DESCR( 4, 3, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH0_TC) },
326 { PCI_DESCR( 5, 0, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH1_CTRL) },
327 { PCI_DESCR( 5, 1, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH1_ADDR) },
328 { PCI_DESCR( 5, 2, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH1_RANK) },
329 { PCI_DESCR( 5, 3, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH1_TC) },
332 * This is the PCI device has an alternate address on some
333 * processors like Core i7 860
335 { PCI_DESCR( 0, 0, PCI_DEVICE_ID_INTEL_LYNNFIELD_NONCORE) },
338 struct pci_id_descr pci_dev_descr_i7core_westmere[] = {
339 /* Memory controller */
340 { PCI_DESCR(3, 0, PCI_DEVICE_ID_INTEL_LYNNFIELD_MCR_REV2) },
341 { PCI_DESCR(3, 1, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_TAD_REV2) },
342 /* Exists only for RDIMM */
343 { PCI_DESCR(3, 2, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_RAS_REV2), .optional = 1 },
344 { PCI_DESCR(3, 4, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_TEST_REV2) },
347 { PCI_DESCR(4, 0, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH0_CTRL_REV2) },
348 { PCI_DESCR(4, 1, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH0_ADDR_REV2) },
349 { PCI_DESCR(4, 2, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH0_RANK_REV2) },
350 { PCI_DESCR(4, 3, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH0_TC_REV2) },
353 { PCI_DESCR(5, 0, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH1_CTRL_REV2) },
354 { PCI_DESCR(5, 1, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH1_ADDR_REV2) },
355 { PCI_DESCR(5, 2, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH1_RANK_REV2) },
356 { PCI_DESCR(5, 3, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH1_TC_REV2) },
359 { PCI_DESCR(6, 0, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH2_CTRL_REV2) },
360 { PCI_DESCR(6, 1, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH2_ADDR_REV2) },
361 { PCI_DESCR(6, 2, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH2_RANK_REV2) },
362 { PCI_DESCR(6, 3, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH2_TC_REV2) },
364 /* Generic Non-core registers */
365 { PCI_DESCR(0, 0, PCI_DEVICE_ID_INTEL_LYNNFIELD_NONCORE_REV2) },
369 #define PCI_ID_TABLE_ENTRY(A) { A, ARRAY_SIZE(A) }
370 struct pci_id_table pci_dev_table[] = {
371 PCI_ID_TABLE_ENTRY(pci_dev_descr_i7core_nehalem),
372 PCI_ID_TABLE_ENTRY(pci_dev_descr_lynnfield),
373 PCI_ID_TABLE_ENTRY(pci_dev_descr_i7core_westmere),
377 * pci_device_id table for which devices we are looking for
379 static const struct pci_device_id i7core_pci_tbl[] __devinitdata = {
380 {PCI_DEVICE(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_X58_HUB_MGMT)},
381 {PCI_DEVICE(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_LYNNFIELD_QPI_LINK0)},
382 {0,} /* 0 terminated list. */
385 /****************************************************************************
386 Anciliary status routines
387 ****************************************************************************/
389 /* MC_CONTROL bits */
390 #define CH_ACTIVE(pvt, ch) ((pvt)->info.mc_control & (1 << (8 + ch)))
391 #define ECCx8(pvt) ((pvt)->info.mc_control & (1 << 1))
394 #define ECC_ENABLED(pvt) ((pvt)->info.mc_status & (1 << 4))
395 #define CH_DISABLED(pvt, ch) ((pvt)->info.mc_status & (1 << ch))
397 /* MC_MAX_DOD read functions */
398 static inline int numdimms(u32 dimms)
400 return (dimms & 0x3) + 1;
403 static inline int numrank(u32 rank)
405 static int ranks[4] = { 1, 2, 4, -EINVAL };
407 return ranks[rank & 0x3];
410 static inline int numbank(u32 bank)
412 static int banks[4] = { 4, 8, 16, -EINVAL };
414 return banks[bank & 0x3];
417 static inline int numrow(u32 row)
419 static int rows[8] = {
420 1 << 12, 1 << 13, 1 << 14, 1 << 15,
421 1 << 16, -EINVAL, -EINVAL, -EINVAL,
424 return rows[row & 0x7];
427 static inline int numcol(u32 col)
429 static int cols[8] = {
430 1 << 10, 1 << 11, 1 << 12, -EINVAL,
432 return cols[col & 0x3];
435 static struct i7core_dev *get_i7core_dev(u8 socket)
437 struct i7core_dev *i7core_dev;
439 list_for_each_entry(i7core_dev, &i7core_edac_list, list) {
440 if (i7core_dev->socket == socket)
447 /****************************************************************************
448 Memory check routines
449 ****************************************************************************/
450 static struct pci_dev *get_pdev_slot_func(u8 socket, unsigned slot,
453 struct i7core_dev *i7core_dev = get_i7core_dev(socket);
459 for (i = 0; i < i7core_dev->n_devs; i++) {
460 if (!i7core_dev->pdev[i])
463 if (PCI_SLOT(i7core_dev->pdev[i]->devfn) == slot &&
464 PCI_FUNC(i7core_dev->pdev[i]->devfn) == func) {
465 return i7core_dev->pdev[i];
473 * i7core_get_active_channels() - gets the number of channels and csrows
474 * @socket: Quick Path Interconnect socket
475 * @channels: Number of channels that will be returned
476 * @csrows: Number of csrows found
478 * Since EDAC core needs to know in advance the number of available channels
479 * and csrows, in order to allocate memory for csrows/channels, it is needed
480 * to run two similar steps. At the first step, implemented on this function,
481 * it checks the number of csrows/channels present at one socket.
482 * this is used in order to properly allocate the size of mci components.
484 * It should be noticed that none of the current available datasheets explain
485 * or even mention how csrows are seen by the memory controller. So, we need
486 * to add a fake description for csrows.
487 * So, this driver is attributing one DIMM memory for one csrow.
489 static int i7core_get_active_channels(u8 socket, unsigned *channels,
492 struct pci_dev *pdev = NULL;
499 pdev = get_pdev_slot_func(socket, 3, 0);
501 i7core_printk(KERN_ERR, "Couldn't find socket %d fn 3.0!!!\n",
506 /* Device 3 function 0 reads */
507 pci_read_config_dword(pdev, MC_STATUS, &status);
508 pci_read_config_dword(pdev, MC_CONTROL, &control);
510 for (i = 0; i < NUM_CHANS; i++) {
512 /* Check if the channel is active */
513 if (!(control & (1 << (8 + i))))
516 /* Check if the channel is disabled */
517 if (status & (1 << i))
520 pdev = get_pdev_slot_func(socket, i + 4, 1);
522 i7core_printk(KERN_ERR, "Couldn't find socket %d "
527 /* Devices 4-6 function 1 */
528 pci_read_config_dword(pdev,
529 MC_DOD_CH_DIMM0, &dimm_dod[0]);
530 pci_read_config_dword(pdev,
531 MC_DOD_CH_DIMM1, &dimm_dod[1]);
532 pci_read_config_dword(pdev,
533 MC_DOD_CH_DIMM2, &dimm_dod[2]);
537 for (j = 0; j < 3; j++) {
538 if (!DIMM_PRESENT(dimm_dod[j]))
544 debugf0("Number of active channels on socket %d: %d\n",
550 static int get_dimm_config(struct mem_ctl_info *mci, int *csrow)
552 struct i7core_pvt *pvt = mci->pvt_info;
553 struct csrow_info *csr;
554 struct pci_dev *pdev;
556 unsigned long last_page = 0;
560 /* Get data from the MC register, function 0 */
561 pdev = pvt->pci_mcr[0];
565 /* Device 3 function 0 reads */
566 pci_read_config_dword(pdev, MC_CONTROL, &pvt->info.mc_control);
567 pci_read_config_dword(pdev, MC_STATUS, &pvt->info.mc_status);
568 pci_read_config_dword(pdev, MC_MAX_DOD, &pvt->info.max_dod);
569 pci_read_config_dword(pdev, MC_CHANNEL_MAPPER, &pvt->info.ch_map);
571 debugf0("QPI %d control=0x%08x status=0x%08x dod=0x%08x map=0x%08x\n",
572 pvt->i7core_dev->socket, pvt->info.mc_control, pvt->info.mc_status,
573 pvt->info.max_dod, pvt->info.ch_map);
575 if (ECC_ENABLED(pvt)) {
576 debugf0("ECC enabled with x%d SDCC\n", ECCx8(pvt) ? 8 : 4);
578 mode = EDAC_S8ECD8ED;
580 mode = EDAC_S4ECD4ED;
582 debugf0("ECC disabled\n");
586 /* FIXME: need to handle the error codes */
587 debugf0("DOD Max limits: DIMMS: %d, %d-ranked, %d-banked "
589 numdimms(pvt->info.max_dod),
590 numrank(pvt->info.max_dod >> 2),
591 numbank(pvt->info.max_dod >> 4),
592 numrow(pvt->info.max_dod >> 6),
593 numcol(pvt->info.max_dod >> 9));
595 for (i = 0; i < NUM_CHANS; i++) {
596 u32 data, dimm_dod[3], value[8];
598 if (!pvt->pci_ch[i][0])
601 if (!CH_ACTIVE(pvt, i)) {
602 debugf0("Channel %i is not active\n", i);
605 if (CH_DISABLED(pvt, i)) {
606 debugf0("Channel %i is disabled\n", i);
610 /* Devices 4-6 function 0 */
611 pci_read_config_dword(pvt->pci_ch[i][0],
612 MC_CHANNEL_DIMM_INIT_PARAMS, &data);
614 pvt->channel[i].ranks = (data & QUAD_RANK_PRESENT) ?
617 if (data & REGISTERED_DIMM)
622 if (data & THREE_DIMMS_PRESENT)
623 pvt->channel[i].dimms = 3;
624 else if (data & SINGLE_QUAD_RANK_PRESENT)
625 pvt->channel[i].dimms = 1;
627 pvt->channel[i].dimms = 2;
630 /* Devices 4-6 function 1 */
631 pci_read_config_dword(pvt->pci_ch[i][1],
632 MC_DOD_CH_DIMM0, &dimm_dod[0]);
633 pci_read_config_dword(pvt->pci_ch[i][1],
634 MC_DOD_CH_DIMM1, &dimm_dod[1]);
635 pci_read_config_dword(pvt->pci_ch[i][1],
636 MC_DOD_CH_DIMM2, &dimm_dod[2]);
638 debugf0("Ch%d phy rd%d, wr%d (0x%08x): "
639 "%d ranks, %cDIMMs\n",
641 RDLCH(pvt->info.ch_map, i), WRLCH(pvt->info.ch_map, i),
643 pvt->channel[i].ranks,
644 (data & REGISTERED_DIMM) ? 'R' : 'U');
646 for (j = 0; j < 3; j++) {
647 u32 banks, ranks, rows, cols;
650 if (!DIMM_PRESENT(dimm_dod[j]))
653 banks = numbank(MC_DOD_NUMBANK(dimm_dod[j]));
654 ranks = numrank(MC_DOD_NUMRANK(dimm_dod[j]));
655 rows = numrow(MC_DOD_NUMROW(dimm_dod[j]));
656 cols = numcol(MC_DOD_NUMCOL(dimm_dod[j]));
658 /* DDR3 has 8 I/O banks */
659 size = (rows * cols * banks * ranks) >> (20 - 3);
661 pvt->channel[i].dimms++;
663 debugf0("\tdimm %d %d Mb offset: %x, "
664 "bank: %d, rank: %d, row: %#x, col: %#x\n",
666 RANKOFFSET(dimm_dod[j]),
667 banks, ranks, rows, cols);
670 npages = size >> (PAGE_SHIFT - 20);
672 npages = size << (20 - PAGE_SHIFT);
675 csr = &mci->csrows[*csrow];
676 csr->first_page = last_page + 1;
678 csr->last_page = last_page;
679 csr->nr_pages = npages;
683 csr->csrow_idx = *csrow;
684 csr->nr_channels = 1;
686 csr->channels[0].chan_idx = i;
687 csr->channels[0].ce_count = 0;
689 pvt->csrow_map[i][j] = *csrow;
699 csr->dtype = DEV_X16;
702 csr->dtype = DEV_UNKNOWN;
705 csr->edac_mode = mode;
711 pci_read_config_dword(pdev, MC_SAG_CH_0, &value[0]);
712 pci_read_config_dword(pdev, MC_SAG_CH_1, &value[1]);
713 pci_read_config_dword(pdev, MC_SAG_CH_2, &value[2]);
714 pci_read_config_dword(pdev, MC_SAG_CH_3, &value[3]);
715 pci_read_config_dword(pdev, MC_SAG_CH_4, &value[4]);
716 pci_read_config_dword(pdev, MC_SAG_CH_5, &value[5]);
717 pci_read_config_dword(pdev, MC_SAG_CH_6, &value[6]);
718 pci_read_config_dword(pdev, MC_SAG_CH_7, &value[7]);
719 debugf1("\t[%i] DIVBY3\tREMOVED\tOFFSET\n", i);
720 for (j = 0; j < 8; j++)
721 debugf1("\t\t%#x\t%#x\t%#x\n",
722 (value[j] >> 27) & 0x1,
723 (value[j] >> 24) & 0x7,
724 (value[j] && ((1 << 24) - 1)));
730 /****************************************************************************
731 Error insertion routines
732 ****************************************************************************/
734 /* The i7core has independent error injection features per channel.
735 However, to have a simpler code, we don't allow enabling error injection
736 on more than one channel.
737 Also, since a change at an inject parameter will be applied only at enable,
738 we're disabling error injection on all write calls to the sysfs nodes that
739 controls the error code injection.
741 static int disable_inject(struct mem_ctl_info *mci)
743 struct i7core_pvt *pvt = mci->pvt_info;
745 pvt->inject.enable = 0;
747 if (!pvt->pci_ch[pvt->inject.channel][0])
750 pci_write_config_dword(pvt->pci_ch[pvt->inject.channel][0],
751 MC_CHANNEL_ERROR_INJECT, 0);
757 * i7core inject inject.section
759 * accept and store error injection inject.section value
760 * bit 0 - refers to the lower 32-byte half cacheline
761 * bit 1 - refers to the upper 32-byte half cacheline
763 static ssize_t i7core_inject_section_store(struct mem_ctl_info *mci,
764 const char *data, size_t count)
766 struct i7core_pvt *pvt = mci->pvt_info;
770 if (pvt->inject.enable)
773 rc = strict_strtoul(data, 10, &value);
774 if ((rc < 0) || (value > 3))
777 pvt->inject.section = (u32) value;
781 static ssize_t i7core_inject_section_show(struct mem_ctl_info *mci,
784 struct i7core_pvt *pvt = mci->pvt_info;
785 return sprintf(data, "0x%08x\n", pvt->inject.section);
791 * accept and store error injection inject.section value
792 * bit 0 - repeat enable - Enable error repetition
793 * bit 1 - inject ECC error
794 * bit 2 - inject parity error
796 static ssize_t i7core_inject_type_store(struct mem_ctl_info *mci,
797 const char *data, size_t count)
799 struct i7core_pvt *pvt = mci->pvt_info;
803 if (pvt->inject.enable)
806 rc = strict_strtoul(data, 10, &value);
807 if ((rc < 0) || (value > 7))
810 pvt->inject.type = (u32) value;
814 static ssize_t i7core_inject_type_show(struct mem_ctl_info *mci,
817 struct i7core_pvt *pvt = mci->pvt_info;
818 return sprintf(data, "0x%08x\n", pvt->inject.type);
822 * i7core_inject_inject.eccmask_store
824 * The type of error (UE/CE) will depend on the inject.eccmask value:
825 * Any bits set to a 1 will flip the corresponding ECC bit
826 * Correctable errors can be injected by flipping 1 bit or the bits within
827 * a symbol pair (2 consecutive aligned 8-bit pairs - i.e. 7:0 and 15:8 or
828 * 23:16 and 31:24). Flipping bits in two symbol pairs will cause an
829 * uncorrectable error to be injected.
831 static ssize_t i7core_inject_eccmask_store(struct mem_ctl_info *mci,
832 const char *data, size_t count)
834 struct i7core_pvt *pvt = mci->pvt_info;
838 if (pvt->inject.enable)
841 rc = strict_strtoul(data, 10, &value);
845 pvt->inject.eccmask = (u32) value;
849 static ssize_t i7core_inject_eccmask_show(struct mem_ctl_info *mci,
852 struct i7core_pvt *pvt = mci->pvt_info;
853 return sprintf(data, "0x%08x\n", pvt->inject.eccmask);
859 * The type of error (UE/CE) will depend on the inject.eccmask value:
860 * Any bits set to a 1 will flip the corresponding ECC bit
861 * Correctable errors can be injected by flipping 1 bit or the bits within
862 * a symbol pair (2 consecutive aligned 8-bit pairs - i.e. 7:0 and 15:8 or
863 * 23:16 and 31:24). Flipping bits in two symbol pairs will cause an
864 * uncorrectable error to be injected.
867 #define DECLARE_ADDR_MATCH(param, limit) \
868 static ssize_t i7core_inject_store_##param( \
869 struct mem_ctl_info *mci, \
870 const char *data, size_t count) \
872 struct i7core_pvt *pvt; \
876 debugf1("%s()\n", __func__); \
877 pvt = mci->pvt_info; \
879 if (pvt->inject.enable) \
880 disable_inject(mci); \
882 if (!strcasecmp(data, "any") || !strcasecmp(data, "any\n"))\
885 rc = strict_strtoul(data, 10, &value); \
886 if ((rc < 0) || (value >= limit)) \
890 pvt->inject.param = value; \
895 static ssize_t i7core_inject_show_##param( \
896 struct mem_ctl_info *mci, \
899 struct i7core_pvt *pvt; \
901 pvt = mci->pvt_info; \
902 debugf1("%s() pvt=%p\n", __func__, pvt); \
903 if (pvt->inject.param < 0) \
904 return sprintf(data, "any\n"); \
906 return sprintf(data, "%d\n", pvt->inject.param);\
909 #define ATTR_ADDR_MATCH(param) \
913 .mode = (S_IRUGO | S_IWUSR) \
915 .show = i7core_inject_show_##param, \
916 .store = i7core_inject_store_##param, \
919 DECLARE_ADDR_MATCH(channel, 3);
920 DECLARE_ADDR_MATCH(dimm, 3);
921 DECLARE_ADDR_MATCH(rank, 4);
922 DECLARE_ADDR_MATCH(bank, 32);
923 DECLARE_ADDR_MATCH(page, 0x10000);
924 DECLARE_ADDR_MATCH(col, 0x4000);
926 static int write_and_test(struct pci_dev *dev, int where, u32 val)
931 debugf0("setting pci %02x:%02x.%x reg=%02x value=%08x\n",
932 dev->bus->number, PCI_SLOT(dev->devfn), PCI_FUNC(dev->devfn),
935 for (count = 0; count < 10; count++) {
938 pci_write_config_dword(dev, where, val);
939 pci_read_config_dword(dev, where, &read);
945 i7core_printk(KERN_ERR, "Error during set pci %02x:%02x.%x reg=%02x "
946 "write=%08x. Read=%08x\n",
947 dev->bus->number, PCI_SLOT(dev->devfn), PCI_FUNC(dev->devfn),
954 * This routine prepares the Memory Controller for error injection.
955 * The error will be injected when some process tries to write to the
956 * memory that matches the given criteria.
957 * The criteria can be set in terms of a mask where dimm, rank, bank, page
958 * and col can be specified.
959 * A -1 value for any of the mask items will make the MCU to ignore
960 * that matching criteria for error injection.
962 * It should be noticed that the error will only happen after a write operation
963 * on a memory that matches the condition. if REPEAT_EN is not enabled at
964 * inject mask, then it will produce just one error. Otherwise, it will repeat
965 * until the injectmask would be cleaned.
967 * FIXME: This routine assumes that MAXNUMDIMMS value of MC_MAX_DOD
968 * is reliable enough to check if the MC is using the
969 * three channels. However, this is not clear at the datasheet.
971 static ssize_t i7core_inject_enable_store(struct mem_ctl_info *mci,
972 const char *data, size_t count)
974 struct i7core_pvt *pvt = mci->pvt_info;
980 if (!pvt->pci_ch[pvt->inject.channel][0])
983 rc = strict_strtoul(data, 10, &enable);
988 pvt->inject.enable = 1;
994 /* Sets pvt->inject.dimm mask */
995 if (pvt->inject.dimm < 0)
998 if (pvt->channel[pvt->inject.channel].dimms > 2)
999 mask |= (pvt->inject.dimm & 0x3LL) << 35;
1001 mask |= (pvt->inject.dimm & 0x1LL) << 36;
1004 /* Sets pvt->inject.rank mask */
1005 if (pvt->inject.rank < 0)
1008 if (pvt->channel[pvt->inject.channel].dimms > 2)
1009 mask |= (pvt->inject.rank & 0x1LL) << 34;
1011 mask |= (pvt->inject.rank & 0x3LL) << 34;
1014 /* Sets pvt->inject.bank mask */
1015 if (pvt->inject.bank < 0)
1018 mask |= (pvt->inject.bank & 0x15LL) << 30;
1020 /* Sets pvt->inject.page mask */
1021 if (pvt->inject.page < 0)
1024 mask |= (pvt->inject.page & 0xffff) << 14;
1026 /* Sets pvt->inject.column mask */
1027 if (pvt->inject.col < 0)
1030 mask |= (pvt->inject.col & 0x3fff);
1034 * bits 1-2: MASK_HALF_CACHELINE
1036 * bit 4: INJECT_ADDR_PARITY
1039 injectmask = (pvt->inject.type & 1) |
1040 (pvt->inject.section & 0x3) << 1 |
1041 (pvt->inject.type & 0x6) << (3 - 1);
1043 /* Unlock writes to registers - this register is write only */
1044 pci_write_config_dword(pvt->pci_noncore,
1045 MC_CFG_CONTROL, 0x2);
1047 write_and_test(pvt->pci_ch[pvt->inject.channel][0],
1048 MC_CHANNEL_ADDR_MATCH, mask);
1049 write_and_test(pvt->pci_ch[pvt->inject.channel][0],
1050 MC_CHANNEL_ADDR_MATCH + 4, mask >> 32L);
1052 write_and_test(pvt->pci_ch[pvt->inject.channel][0],
1053 MC_CHANNEL_ERROR_MASK, pvt->inject.eccmask);
1055 write_and_test(pvt->pci_ch[pvt->inject.channel][0],
1056 MC_CHANNEL_ERROR_INJECT, injectmask);
1059 * This is something undocumented, based on my tests
1060 * Without writing 8 to this register, errors aren't injected. Not sure
1063 pci_write_config_dword(pvt->pci_noncore,
1066 debugf0("Error inject addr match 0x%016llx, ecc 0x%08x,"
1068 mask, pvt->inject.eccmask, injectmask);
1074 static ssize_t i7core_inject_enable_show(struct mem_ctl_info *mci,
1077 struct i7core_pvt *pvt = mci->pvt_info;
1080 if (!pvt->pci_ch[pvt->inject.channel][0])
1083 pci_read_config_dword(pvt->pci_ch[pvt->inject.channel][0],
1084 MC_CHANNEL_ERROR_INJECT, &injectmask);
1086 debugf0("Inject error read: 0x%018x\n", injectmask);
1088 if (injectmask & 0x0c)
1089 pvt->inject.enable = 1;
1091 return sprintf(data, "%d\n", pvt->inject.enable);
1094 #define DECLARE_COUNTER(param) \
1095 static ssize_t i7core_show_counter_##param( \
1096 struct mem_ctl_info *mci, \
1099 struct i7core_pvt *pvt = mci->pvt_info; \
1101 debugf1("%s() \n", __func__); \
1102 if (!pvt->ce_count_available || (pvt->is_registered)) \
1103 return sprintf(data, "data unavailable\n"); \
1104 return sprintf(data, "%lu\n", \
1105 pvt->udimm_ce_count[param]); \
1108 #define ATTR_COUNTER(param) \
1111 .name = __stringify(udimm##param), \
1112 .mode = (S_IRUGO | S_IWUSR) \
1114 .show = i7core_show_counter_##param \
1126 static struct mcidev_sysfs_attribute i7core_addrmatch_attrs[] = {
1127 ATTR_ADDR_MATCH(channel),
1128 ATTR_ADDR_MATCH(dimm),
1129 ATTR_ADDR_MATCH(rank),
1130 ATTR_ADDR_MATCH(bank),
1131 ATTR_ADDR_MATCH(page),
1132 ATTR_ADDR_MATCH(col),
1133 { .attr = { .name = NULL } }
1136 static struct mcidev_sysfs_group i7core_inject_addrmatch = {
1137 .name = "inject_addrmatch",
1138 .mcidev_attr = i7core_addrmatch_attrs,
1141 static struct mcidev_sysfs_attribute i7core_udimm_counters_attrs[] = {
1145 { .attr = { .name = NULL } }
1148 static struct mcidev_sysfs_group i7core_udimm_counters = {
1149 .name = "all_channel_counts",
1150 .mcidev_attr = i7core_udimm_counters_attrs,
1153 static struct mcidev_sysfs_attribute i7core_sysfs_attrs[] = {
1156 .name = "inject_section",
1157 .mode = (S_IRUGO | S_IWUSR)
1159 .show = i7core_inject_section_show,
1160 .store = i7core_inject_section_store,
1163 .name = "inject_type",
1164 .mode = (S_IRUGO | S_IWUSR)
1166 .show = i7core_inject_type_show,
1167 .store = i7core_inject_type_store,
1170 .name = "inject_eccmask",
1171 .mode = (S_IRUGO | S_IWUSR)
1173 .show = i7core_inject_eccmask_show,
1174 .store = i7core_inject_eccmask_store,
1176 .grp = &i7core_inject_addrmatch,
1179 .name = "inject_enable",
1180 .mode = (S_IRUGO | S_IWUSR)
1182 .show = i7core_inject_enable_show,
1183 .store = i7core_inject_enable_store,
1185 { .attr = { .name = NULL } }, /* Reserved for udimm counters */
1186 { .attr = { .name = NULL } }
1189 /****************************************************************************
1190 Device initialization routines: put/get, init/exit
1191 ****************************************************************************/
1194 * i7core_put_devices 'put' all the devices that we have
1195 * reserved via 'get'
1197 static void i7core_put_devices(struct i7core_dev *i7core_dev)
1201 debugf0(__FILE__ ": %s()\n", __func__);
1202 for (i = 0; i < i7core_dev->n_devs; i++) {
1203 struct pci_dev *pdev = i7core_dev->pdev[i];
1206 debugf0("Removing dev %02x:%02x.%d\n",
1208 PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn));
1211 kfree(i7core_dev->pdev);
1212 list_del(&i7core_dev->list);
1216 static void i7core_put_all_devices(void)
1218 struct i7core_dev *i7core_dev, *tmp;
1220 list_for_each_entry_safe(i7core_dev, tmp, &i7core_edac_list, list)
1221 i7core_put_devices(i7core_dev);
1224 static void __init i7core_xeon_pci_fixup(struct pci_id_table *table)
1226 struct pci_dev *pdev = NULL;
1229 * On Xeon 55xx, the Intel Quckpath Arch Generic Non-core pci buses
1230 * aren't announced by acpi. So, we need to use a legacy scan probing
1233 while (table && table->descr) {
1234 pdev = pci_get_device(PCI_VENDOR_ID_INTEL, table->descr[0].dev_id, NULL);
1235 if (unlikely(!pdev)) {
1236 for (i = 0; i < MAX_SOCKET_BUSES; i++)
1237 pcibios_scan_specific_bus(255-i);
1244 static unsigned i7core_pci_lastbus(void)
1246 int last_bus = 0, bus;
1247 struct pci_bus *b = NULL;
1249 while ((b = pci_find_next_bus(b)) != NULL) {
1251 debugf0("Found bus %d\n", bus);
1256 debugf0("Last bus %d\n", last_bus);
1262 * i7core_get_devices Find and perform 'get' operation on the MCH's
1263 * device/functions we want to reference for this driver
1265 * Need to 'get' device 16 func 1 and func 2
1267 int i7core_get_onedevice(struct pci_dev **prev, int devno,
1268 struct pci_id_descr *dev_descr, unsigned n_devs,
1271 struct i7core_dev *i7core_dev;
1273 struct pci_dev *pdev = NULL;
1277 pdev = pci_get_device(PCI_VENDOR_ID_INTEL,
1278 dev_descr->dev_id, *prev);
1281 * On Xeon 55xx, the Intel Quckpath Arch Generic Non-core regs
1282 * is at addr 8086:2c40, instead of 8086:2c41. So, we need
1283 * to probe for the alternate address in case of failure
1285 if (dev_descr->dev_id == PCI_DEVICE_ID_INTEL_I7_NONCORE && !pdev)
1286 pdev = pci_get_device(PCI_VENDOR_ID_INTEL,
1287 PCI_DEVICE_ID_INTEL_I7_NONCORE_ALT, *prev);
1289 if (dev_descr->dev_id == PCI_DEVICE_ID_INTEL_LYNNFIELD_NONCORE && !pdev)
1290 pdev = pci_get_device(PCI_VENDOR_ID_INTEL,
1291 PCI_DEVICE_ID_INTEL_LYNNFIELD_NONCORE_ALT,
1300 if (dev_descr->optional)
1306 i7core_printk(KERN_INFO,
1307 "Device not found: dev %02x.%d PCI ID %04x:%04x\n",
1308 dev_descr->dev, dev_descr->func,
1309 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
1311 /* End of list, leave */
1314 bus = pdev->bus->number;
1316 socket = last_bus - bus;
1318 i7core_dev = get_i7core_dev(socket);
1320 i7core_dev = kzalloc(sizeof(*i7core_dev), GFP_KERNEL);
1323 i7core_dev->pdev = kzalloc(sizeof(*i7core_dev->pdev) * n_devs,
1325 if (!i7core_dev->pdev) {
1329 i7core_dev->socket = socket;
1330 i7core_dev->n_devs = n_devs;
1331 list_add_tail(&i7core_dev->list, &i7core_edac_list);
1334 if (i7core_dev->pdev[devno]) {
1335 i7core_printk(KERN_ERR,
1336 "Duplicated device for "
1337 "dev %02x:%02x.%d PCI ID %04x:%04x\n",
1338 bus, dev_descr->dev, dev_descr->func,
1339 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
1344 i7core_dev->pdev[devno] = pdev;
1347 if (unlikely(PCI_SLOT(pdev->devfn) != dev_descr->dev ||
1348 PCI_FUNC(pdev->devfn) != dev_descr->func)) {
1349 i7core_printk(KERN_ERR,
1350 "Device PCI ID %04x:%04x "
1351 "has dev %02x:%02x.%d instead of dev %02x:%02x.%d\n",
1352 PCI_VENDOR_ID_INTEL, dev_descr->dev_id,
1353 bus, PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn),
1354 bus, dev_descr->dev, dev_descr->func);
1358 /* Be sure that the device is enabled */
1359 if (unlikely(pci_enable_device(pdev) < 0)) {
1360 i7core_printk(KERN_ERR,
1362 "dev %02x:%02x.%d PCI ID %04x:%04x\n",
1363 bus, dev_descr->dev, dev_descr->func,
1364 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
1368 debugf0("Detected socket %d dev %02x:%02x.%d PCI ID %04x:%04x\n",
1369 socket, bus, dev_descr->dev,
1371 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
1378 static int i7core_get_devices(struct pci_id_table *table)
1380 int i, rc, last_bus;
1381 struct pci_dev *pdev = NULL;
1382 struct pci_id_descr *dev_descr;
1384 last_bus = i7core_pci_lastbus();
1386 while (table && table->descr) {
1387 dev_descr = table->descr;
1388 for (i = 0; i < table->n_devs; i++) {
1391 rc = i7core_get_onedevice(&pdev, i,
1400 i7core_put_all_devices();
1412 static int mci_bind_devs(struct mem_ctl_info *mci,
1413 struct i7core_dev *i7core_dev)
1415 struct i7core_pvt *pvt = mci->pvt_info;
1416 struct pci_dev *pdev;
1419 /* Associates i7core_dev and mci for future usage */
1420 pvt->i7core_dev = i7core_dev;
1421 i7core_dev->mci = mci;
1423 pvt->is_registered = 0;
1424 for (i = 0; i < i7core_dev->n_devs; i++) {
1425 pdev = i7core_dev->pdev[i];
1429 func = PCI_FUNC(pdev->devfn);
1430 slot = PCI_SLOT(pdev->devfn);
1432 if (unlikely(func > MAX_MCR_FUNC))
1434 pvt->pci_mcr[func] = pdev;
1435 } else if (likely(slot >= 4 && slot < 4 + NUM_CHANS)) {
1436 if (unlikely(func > MAX_CHAN_FUNC))
1438 pvt->pci_ch[slot - 4][func] = pdev;
1439 } else if (!slot && !func)
1440 pvt->pci_noncore = pdev;
1444 debugf0("Associated fn %d.%d, dev = %p, socket %d\n",
1445 PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn),
1446 pdev, i7core_dev->socket);
1448 if (PCI_SLOT(pdev->devfn) == 3 &&
1449 PCI_FUNC(pdev->devfn) == 2)
1450 pvt->is_registered = 1;
1454 * Add extra nodes to count errors on udimm
1455 * For registered memory, this is not needed, since the counters
1456 * are already displayed at the standard locations
1458 if (!pvt->is_registered)
1459 i7core_sysfs_attrs[ARRAY_SIZE(i7core_sysfs_attrs)-2].grp =
1460 &i7core_udimm_counters;
1465 i7core_printk(KERN_ERR, "Device %d, function %d "
1466 "is out of the expected range\n",
1471 /****************************************************************************
1472 Error check routines
1473 ****************************************************************************/
1474 static void i7core_rdimm_update_csrow(struct mem_ctl_info *mci,
1475 int chan, int dimm, int add)
1478 struct i7core_pvt *pvt = mci->pvt_info;
1479 int row = pvt->csrow_map[chan][dimm], i;
1481 for (i = 0; i < add; i++) {
1482 msg = kasprintf(GFP_KERNEL, "Corrected error "
1483 "(Socket=%d channel=%d dimm=%d)",
1484 pvt->i7core_dev->socket, chan, dimm);
1486 edac_mc_handle_fbd_ce(mci, row, 0, msg);
1491 static void i7core_rdimm_update_ce_count(struct mem_ctl_info *mci,
1492 int chan, int new0, int new1, int new2)
1494 struct i7core_pvt *pvt = mci->pvt_info;
1495 int add0 = 0, add1 = 0, add2 = 0;
1496 /* Updates CE counters if it is not the first time here */
1497 if (pvt->ce_count_available) {
1498 /* Updates CE counters */
1500 add2 = new2 - pvt->rdimm_last_ce_count[chan][2];
1501 add1 = new1 - pvt->rdimm_last_ce_count[chan][1];
1502 add0 = new0 - pvt->rdimm_last_ce_count[chan][0];
1506 pvt->rdimm_ce_count[chan][2] += add2;
1510 pvt->rdimm_ce_count[chan][1] += add1;
1514 pvt->rdimm_ce_count[chan][0] += add0;
1516 pvt->ce_count_available = 1;
1518 /* Store the new values */
1519 pvt->rdimm_last_ce_count[chan][2] = new2;
1520 pvt->rdimm_last_ce_count[chan][1] = new1;
1521 pvt->rdimm_last_ce_count[chan][0] = new0;
1523 /*updated the edac core */
1525 i7core_rdimm_update_csrow(mci, chan, 0, add0);
1527 i7core_rdimm_update_csrow(mci, chan, 1, add1);
1529 i7core_rdimm_update_csrow(mci, chan, 2, add2);
1533 static void i7core_rdimm_check_mc_ecc_err(struct mem_ctl_info *mci)
1535 struct i7core_pvt *pvt = mci->pvt_info;
1537 int i, new0, new1, new2;
1539 /*Read DEV 3: FUN 2: MC_COR_ECC_CNT regs directly*/
1540 pci_read_config_dword(pvt->pci_mcr[2], MC_COR_ECC_CNT_0,
1542 pci_read_config_dword(pvt->pci_mcr[2], MC_COR_ECC_CNT_1,
1544 pci_read_config_dword(pvt->pci_mcr[2], MC_COR_ECC_CNT_2,
1546 pci_read_config_dword(pvt->pci_mcr[2], MC_COR_ECC_CNT_3,
1548 pci_read_config_dword(pvt->pci_mcr[2], MC_COR_ECC_CNT_4,
1550 pci_read_config_dword(pvt->pci_mcr[2], MC_COR_ECC_CNT_5,
1552 for (i = 0 ; i < 3; i++) {
1553 debugf3("MC_COR_ECC_CNT%d = 0x%x; MC_COR_ECC_CNT%d = 0x%x\n",
1554 (i * 2), rcv[i][0], (i * 2) + 1, rcv[i][1]);
1555 /*if the channel has 3 dimms*/
1556 if (pvt->channel[i].dimms > 2) {
1557 new0 = DIMM_BOT_COR_ERR(rcv[i][0]);
1558 new1 = DIMM_TOP_COR_ERR(rcv[i][0]);
1559 new2 = DIMM_BOT_COR_ERR(rcv[i][1]);
1561 new0 = DIMM_TOP_COR_ERR(rcv[i][0]) +
1562 DIMM_BOT_COR_ERR(rcv[i][0]);
1563 new1 = DIMM_TOP_COR_ERR(rcv[i][1]) +
1564 DIMM_BOT_COR_ERR(rcv[i][1]);
1568 i7core_rdimm_update_ce_count(mci, i, new0, new1, new2);
1572 /* This function is based on the device 3 function 4 registers as described on:
1573 * Intel Xeon Processor 5500 Series Datasheet Volume 2
1574 * http://www.intel.com/Assets/PDF/datasheet/321322.pdf
1575 * also available at:
1576 * http://www.arrownac.com/manufacturers/intel/s/nehalem/5500-datasheet-v2.pdf
1578 static void i7core_udimm_check_mc_ecc_err(struct mem_ctl_info *mci)
1580 struct i7core_pvt *pvt = mci->pvt_info;
1582 int new0, new1, new2;
1584 if (!pvt->pci_mcr[4]) {
1585 debugf0("%s MCR registers not found\n", __func__);
1589 /* Corrected test errors */
1590 pci_read_config_dword(pvt->pci_mcr[4], MC_TEST_ERR_RCV1, &rcv1);
1591 pci_read_config_dword(pvt->pci_mcr[4], MC_TEST_ERR_RCV0, &rcv0);
1593 /* Store the new values */
1594 new2 = DIMM2_COR_ERR(rcv1);
1595 new1 = DIMM1_COR_ERR(rcv0);
1596 new0 = DIMM0_COR_ERR(rcv0);
1598 /* Updates CE counters if it is not the first time here */
1599 if (pvt->ce_count_available) {
1600 /* Updates CE counters */
1601 int add0, add1, add2;
1603 add2 = new2 - pvt->udimm_last_ce_count[2];
1604 add1 = new1 - pvt->udimm_last_ce_count[1];
1605 add0 = new0 - pvt->udimm_last_ce_count[0];
1609 pvt->udimm_ce_count[2] += add2;
1613 pvt->udimm_ce_count[1] += add1;
1617 pvt->udimm_ce_count[0] += add0;
1619 if (add0 | add1 | add2)
1620 i7core_printk(KERN_ERR, "New Corrected error(s): "
1621 "dimm0: +%d, dimm1: +%d, dimm2 +%d\n",
1624 pvt->ce_count_available = 1;
1626 /* Store the new values */
1627 pvt->udimm_last_ce_count[2] = new2;
1628 pvt->udimm_last_ce_count[1] = new1;
1629 pvt->udimm_last_ce_count[0] = new0;
1633 * According with tables E-11 and E-12 of chapter E.3.3 of Intel 64 and IA-32
1634 * Architectures Software Developer’s Manual Volume 3B.
1635 * Nehalem are defined as family 0x06, model 0x1a
1637 * The MCA registers used here are the following ones:
1638 * struct mce field MCA Register
1639 * m->status MSR_IA32_MC8_STATUS
1640 * m->addr MSR_IA32_MC8_ADDR
1641 * m->misc MSR_IA32_MC8_MISC
1642 * In the case of Nehalem, the error information is masked at .status and .misc
1645 static void i7core_mce_output_error(struct mem_ctl_info *mci,
1648 struct i7core_pvt *pvt = mci->pvt_info;
1649 char *type, *optype, *err, *msg;
1650 unsigned long error = m->status & 0x1ff0000l;
1651 u32 optypenum = (m->status >> 4) & 0x07;
1652 u32 core_err_cnt = (m->status >> 38) && 0x7fff;
1653 u32 dimm = (m->misc >> 16) & 0x3;
1654 u32 channel = (m->misc >> 18) & 0x3;
1655 u32 syndrome = m->misc >> 32;
1656 u32 errnum = find_first_bit(&error, 32);
1659 if (m->mcgstatus & 1)
1664 switch (optypenum) {
1666 optype = "generic undef request";
1669 optype = "read error";
1672 optype = "write error";
1675 optype = "addr/cmd error";
1678 optype = "scrubbing error";
1681 optype = "reserved";
1687 err = "read ECC error";
1690 err = "RAS ECC error";
1693 err = "write parity error";
1696 err = "redundacy loss";
1702 err = "memory range error";
1705 err = "RTID out of range";
1708 err = "address parity error";
1711 err = "byte enable parity error";
1717 /* FIXME: should convert addr into bank and rank information */
1718 msg = kasprintf(GFP_ATOMIC,
1719 "%s (addr = 0x%08llx, cpu=%d, Dimm=%d, Channel=%d, "
1720 "syndrome=0x%08x, count=%d, Err=%08llx:%08llx (%s: %s))\n",
1721 type, (long long) m->addr, m->cpu, dimm, channel,
1722 syndrome, core_err_cnt, (long long)m->status,
1723 (long long)m->misc, optype, err);
1727 csrow = pvt->csrow_map[channel][dimm];
1729 /* Call the helper to output message */
1730 if (m->mcgstatus & 1)
1731 edac_mc_handle_fbd_ue(mci, csrow, 0,
1732 0 /* FIXME: should be channel here */, msg);
1733 else if (!pvt->is_registered)
1734 edac_mc_handle_fbd_ce(mci, csrow,
1735 0 /* FIXME: should be channel here */, msg);
1741 * i7core_check_error Retrieve and process errors reported by the
1742 * hardware. Called by the Core module.
1744 static void i7core_check_error(struct mem_ctl_info *mci)
1746 struct i7core_pvt *pvt = mci->pvt_info;
1752 * MCE first step: Copy all mce errors into a temporary buffer
1753 * We use a double buffering here, to reduce the risk of
1757 count = (pvt->mce_out + MCE_LOG_LEN - pvt->mce_in)
1760 goto check_ce_error;
1762 m = pvt->mce_outentry;
1763 if (pvt->mce_in + count > MCE_LOG_LEN) {
1764 unsigned l = MCE_LOG_LEN - pvt->mce_in;
1766 memcpy(m, &pvt->mce_entry[pvt->mce_in], sizeof(*m) * l);
1772 memcpy(m, &pvt->mce_entry[pvt->mce_in], sizeof(*m) * count);
1774 pvt->mce_in += count;
1777 if (pvt->mce_overrun) {
1778 i7core_printk(KERN_ERR, "Lost %d memory errors\n",
1781 pvt->mce_overrun = 0;
1785 * MCE second step: parse errors and display
1787 for (i = 0; i < count; i++)
1788 i7core_mce_output_error(mci, &pvt->mce_outentry[i]);
1791 * Now, let's increment CE error counts
1794 if (!pvt->is_registered)
1795 i7core_udimm_check_mc_ecc_err(mci);
1797 i7core_rdimm_check_mc_ecc_err(mci);
1801 * i7core_mce_check_error Replicates mcelog routine to get errors
1802 * This routine simply queues mcelog errors, and
1803 * return. The error itself should be handled later
1804 * by i7core_check_error.
1805 * WARNING: As this routine should be called at NMI time, extra care should
1806 * be taken to avoid deadlocks, and to be as fast as possible.
1808 static int i7core_mce_check_error(void *priv, struct mce *mce)
1810 struct mem_ctl_info *mci = priv;
1811 struct i7core_pvt *pvt = mci->pvt_info;
1814 * Just let mcelog handle it if the error is
1815 * outside the memory controller
1817 if (((mce->status & 0xffff) >> 7) != 1)
1820 /* Bank 8 registers are the only ones that we know how to handle */
1825 /* Only handle if it is the right mc controller */
1826 if (cpu_data(mce->cpu).phys_proc_id != pvt->i7core_dev->socket)
1831 if ((pvt->mce_out + 1) % MCE_LOG_LEN == pvt->mce_in) {
1837 /* Copy memory error at the ringbuffer */
1838 memcpy(&pvt->mce_entry[pvt->mce_out], mce, sizeof(*mce));
1840 pvt->mce_out = (pvt->mce_out + 1) % MCE_LOG_LEN;
1842 /* Handle fatal errors immediately */
1843 if (mce->mcgstatus & 1)
1844 i7core_check_error(mci);
1846 /* Advice mcelog that the error were handled */
1850 static int i7core_register_mci(struct i7core_dev *i7core_dev,
1851 int num_channels, int num_csrows)
1853 struct mem_ctl_info *mci;
1854 struct i7core_pvt *pvt;
1858 /* allocate a new MC control structure */
1859 mci = edac_mc_alloc(sizeof(*pvt), num_csrows, num_channels,
1860 i7core_dev->socket);
1864 debugf0("MC: " __FILE__ ": %s(): mci = %p\n", __func__, mci);
1866 /* record ptr to the generic device */
1867 mci->dev = &i7core_dev->pdev[0]->dev;
1869 pvt = mci->pvt_info;
1870 memset(pvt, 0, sizeof(*pvt));
1873 * FIXME: how to handle RDDR3 at MCI level? It is possible to have
1874 * Mixed RDDR3/UDDR3 with Nehalem, provided that they are on different
1877 mci->mtype_cap = MEM_FLAG_DDR3;
1878 mci->edac_ctl_cap = EDAC_FLAG_NONE;
1879 mci->edac_cap = EDAC_FLAG_NONE;
1880 mci->mod_name = "i7core_edac.c";
1881 mci->mod_ver = I7CORE_REVISION;
1882 mci->ctl_name = kasprintf(GFP_KERNEL, "i7 core #%d",
1883 i7core_dev->socket);
1884 mci->dev_name = pci_name(i7core_dev->pdev[0]);
1885 mci->ctl_page_to_phys = NULL;
1886 mci->mc_driver_sysfs_attributes = i7core_sysfs_attrs;
1887 /* Set the function pointer to an actual operation function */
1888 mci->edac_check = i7core_check_error;
1890 /* Store pci devices at mci for faster access */
1891 rc = mci_bind_devs(mci, i7core_dev);
1892 if (unlikely(rc < 0))
1895 /* Get dimm basic config */
1896 get_dimm_config(mci, &csrow);
1898 /* add this new MC control structure to EDAC's list of MCs */
1899 if (unlikely(edac_mc_add_mc(mci))) {
1900 debugf0("MC: " __FILE__
1901 ": %s(): failed edac_mc_add_mc()\n", __func__);
1902 /* FIXME: perhaps some code should go here that disables error
1903 * reporting if we just enabled it
1910 /* allocating generic PCI control info */
1911 pvt->i7core_pci = edac_pci_create_generic_ctl(&i7core_dev->pdev[0]->dev,
1913 if (unlikely(!pvt->i7core_pci)) {
1915 "%s(): Unable to create PCI control\n",
1918 "%s(): PCI error report via EDAC not setup\n",
1922 /* Default error mask is any memory */
1923 pvt->inject.channel = 0;
1924 pvt->inject.dimm = -1;
1925 pvt->inject.rank = -1;
1926 pvt->inject.bank = -1;
1927 pvt->inject.page = -1;
1928 pvt->inject.col = -1;
1930 /* Registers on edac_mce in order to receive memory errors */
1931 pvt->edac_mce.priv = mci;
1932 pvt->edac_mce.check_error = i7core_mce_check_error;
1934 rc = edac_mce_register(&pvt->edac_mce);
1935 if (unlikely(rc < 0)) {
1936 debugf0("MC: " __FILE__
1937 ": %s(): failed edac_mce_register()\n", __func__);
1947 * i7core_probe Probe for ONE instance of device to see if it is
1950 * 0 for FOUND a device
1951 * < 0 for error code
1954 static int __devinit i7core_probe(struct pci_dev *pdev,
1955 const struct pci_device_id *id)
1958 struct i7core_dev *i7core_dev;
1960 /* get the pci devices we want to reserve for our use */
1961 mutex_lock(&i7core_edac_lock);
1964 * All memory controllers are allocated at the first pass.
1966 if (unlikely(probed >= 1)) {
1967 mutex_unlock(&i7core_edac_lock);
1972 rc = i7core_get_devices(pci_dev_table);
1973 if (unlikely(rc < 0))
1976 list_for_each_entry(i7core_dev, &i7core_edac_list, list) {
1980 /* Check the number of active and not disabled channels */
1981 rc = i7core_get_active_channels(i7core_dev->socket,
1982 &channels, &csrows);
1983 if (unlikely(rc < 0))
1986 rc = i7core_register_mci(i7core_dev, channels, csrows);
1987 if (unlikely(rc < 0))
1991 i7core_printk(KERN_INFO, "Driver loaded.\n");
1993 mutex_unlock(&i7core_edac_lock);
1997 i7core_put_all_devices();
1999 mutex_unlock(&i7core_edac_lock);
2004 * i7core_remove destructor for one instance of device
2007 static void __devexit i7core_remove(struct pci_dev *pdev)
2009 struct mem_ctl_info *mci;
2010 struct i7core_dev *i7core_dev, *tmp;
2011 struct i7core_pvt *pvt;
2013 debugf0(__FILE__ ": %s()\n", __func__);
2016 * we have a trouble here: pdev value for removal will be wrong, since
2017 * it will point to the X58 register used to detect that the machine
2018 * is a Nehalem or upper design. However, due to the way several PCI
2019 * devices are grouped together to provide MC functionality, we need
2020 * to use a different method for releasing the devices
2023 mutex_lock(&i7core_edac_lock);
2024 list_for_each_entry_safe(i7core_dev, tmp, &i7core_edac_list, list) {
2025 mci = find_mci_by_dev(&i7core_dev->pdev[0]->dev);
2026 if (unlikely(!mci || !mci->pvt_info)) {
2027 i7core_printk(KERN_ERR,
2028 "Couldn't find mci hanler\n");
2030 pvt = mci->pvt_info;
2031 i7core_dev = pvt->i7core_dev;
2033 if (likely(pvt->i7core_pci))
2034 edac_pci_release_generic_ctl(pvt->i7core_pci);
2036 i7core_printk(KERN_ERR,
2037 "Couldn't find mem_ctl_info for socket %d\n",
2038 i7core_dev->socket);
2039 pvt->i7core_pci = NULL;
2041 edac_mc_del_mc(&i7core_dev->pdev[0]->dev);
2043 edac_mce_unregister(&pvt->edac_mce);
2044 kfree(mci->ctl_name);
2046 i7core_put_devices(i7core_dev);
2051 mutex_unlock(&i7core_edac_lock);
2054 MODULE_DEVICE_TABLE(pci, i7core_pci_tbl);
2057 * i7core_driver pci_driver structure for this module
2060 static struct pci_driver i7core_driver = {
2061 .name = "i7core_edac",
2062 .probe = i7core_probe,
2063 .remove = __devexit_p(i7core_remove),
2064 .id_table = i7core_pci_tbl,
2068 * i7core_init Module entry function
2069 * Try to initialize this module for its devices
2071 static int __init i7core_init(void)
2075 debugf2("MC: " __FILE__ ": %s()\n", __func__);
2077 /* Ensure that the OPSTATE is set correctly for POLL or NMI */
2080 i7core_xeon_pci_fixup(pci_dev_table);
2082 pci_rc = pci_register_driver(&i7core_driver);
2087 i7core_printk(KERN_ERR, "Failed to register device with error %d.\n",
2094 * i7core_exit() Module exit function
2095 * Unregister the driver
2097 static void __exit i7core_exit(void)
2099 debugf2("MC: " __FILE__ ": %s()\n", __func__);
2100 pci_unregister_driver(&i7core_driver);
2103 module_init(i7core_init);
2104 module_exit(i7core_exit);
2106 MODULE_LICENSE("GPL");
2107 MODULE_AUTHOR("Mauro Carvalho Chehab <mchehab@redhat.com>");
2108 MODULE_AUTHOR("Red Hat Inc. (http://www.redhat.com)");
2109 MODULE_DESCRIPTION("MC Driver for Intel i7 Core memory controllers - "
2112 module_param(edac_op_state, int, 0444);
2113 MODULE_PARM_DESC(edac_op_state, "EDAC Error Reporting state: 0=Poll,1=NMI");