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);
47 static int use_pci_fixup;
48 module_param(use_pci_fixup, int, 0444);
49 MODULE_PARM_DESC(use_pci_fixup, "Enable PCI fixup to seek for hidden devices");
51 * This is used for Nehalem-EP and Nehalem-EX devices, where the non-core
52 * registers start at bus 255, and are not reported by BIOS.
53 * We currently find devices with only 2 sockets. In order to support more QPI
54 * Quick Path Interconnect, just increment this number.
56 #define MAX_SOCKET_BUSES 2
60 * Alter this version for the module when modifications are made
62 #define I7CORE_REVISION " Ver: 1.0.0 " __DATE__
63 #define EDAC_MOD_STR "i7core_edac"
68 #define i7core_printk(level, fmt, arg...) \
69 edac_printk(level, "i7core", fmt, ##arg)
71 #define i7core_mc_printk(mci, level, fmt, arg...) \
72 edac_mc_chipset_printk(mci, level, "i7core", fmt, ##arg)
75 * i7core Memory Controller Registers
78 /* OFFSETS for Device 0 Function 0 */
80 #define MC_CFG_CONTROL 0x90
82 /* OFFSETS for Device 3 Function 0 */
84 #define MC_CONTROL 0x48
85 #define MC_STATUS 0x4c
86 #define MC_MAX_DOD 0x64
89 * OFFSETS for Device 3 Function 4, as inicated on Xeon 5500 datasheet:
90 * http://www.arrownac.com/manufacturers/intel/s/nehalem/5500-datasheet-v2.pdf
93 #define MC_TEST_ERR_RCV1 0x60
94 #define DIMM2_COR_ERR(r) ((r) & 0x7fff)
96 #define MC_TEST_ERR_RCV0 0x64
97 #define DIMM1_COR_ERR(r) (((r) >> 16) & 0x7fff)
98 #define DIMM0_COR_ERR(r) ((r) & 0x7fff)
100 /* OFFSETS for Device 3 Function 2, as inicated on Xeon 5500 datasheet */
101 #define MC_COR_ECC_CNT_0 0x80
102 #define MC_COR_ECC_CNT_1 0x84
103 #define MC_COR_ECC_CNT_2 0x88
104 #define MC_COR_ECC_CNT_3 0x8c
105 #define MC_COR_ECC_CNT_4 0x90
106 #define MC_COR_ECC_CNT_5 0x94
108 #define DIMM_TOP_COR_ERR(r) (((r) >> 16) & 0x7fff)
109 #define DIMM_BOT_COR_ERR(r) ((r) & 0x7fff)
112 /* OFFSETS for Devices 4,5 and 6 Function 0 */
114 #define MC_CHANNEL_DIMM_INIT_PARAMS 0x58
115 #define THREE_DIMMS_PRESENT (1 << 24)
116 #define SINGLE_QUAD_RANK_PRESENT (1 << 23)
117 #define QUAD_RANK_PRESENT (1 << 22)
118 #define REGISTERED_DIMM (1 << 15)
120 #define MC_CHANNEL_MAPPER 0x60
121 #define RDLCH(r, ch) ((((r) >> (3 + (ch * 6))) & 0x07) - 1)
122 #define WRLCH(r, ch) ((((r) >> (ch * 6)) & 0x07) - 1)
124 #define MC_CHANNEL_RANK_PRESENT 0x7c
125 #define RANK_PRESENT_MASK 0xffff
127 #define MC_CHANNEL_ADDR_MATCH 0xf0
128 #define MC_CHANNEL_ERROR_MASK 0xf8
129 #define MC_CHANNEL_ERROR_INJECT 0xfc
130 #define INJECT_ADDR_PARITY 0x10
131 #define INJECT_ECC 0x08
132 #define MASK_CACHELINE 0x06
133 #define MASK_FULL_CACHELINE 0x06
134 #define MASK_MSB32_CACHELINE 0x04
135 #define MASK_LSB32_CACHELINE 0x02
136 #define NO_MASK_CACHELINE 0x00
137 #define REPEAT_EN 0x01
139 /* OFFSETS for Devices 4,5 and 6 Function 1 */
141 #define MC_DOD_CH_DIMM0 0x48
142 #define MC_DOD_CH_DIMM1 0x4c
143 #define MC_DOD_CH_DIMM2 0x50
144 #define RANKOFFSET_MASK ((1 << 12) | (1 << 11) | (1 << 10))
145 #define RANKOFFSET(x) ((x & RANKOFFSET_MASK) >> 10)
146 #define DIMM_PRESENT_MASK (1 << 9)
147 #define DIMM_PRESENT(x) (((x) & DIMM_PRESENT_MASK) >> 9)
148 #define MC_DOD_NUMBANK_MASK ((1 << 8) | (1 << 7))
149 #define MC_DOD_NUMBANK(x) (((x) & MC_DOD_NUMBANK_MASK) >> 7)
150 #define MC_DOD_NUMRANK_MASK ((1 << 6) | (1 << 5))
151 #define MC_DOD_NUMRANK(x) (((x) & MC_DOD_NUMRANK_MASK) >> 5)
152 #define MC_DOD_NUMROW_MASK ((1 << 4) | (1 << 3) | (1 << 2))
153 #define MC_DOD_NUMROW(x) (((x) & MC_DOD_NUMROW_MASK) >> 2)
154 #define MC_DOD_NUMCOL_MASK 3
155 #define MC_DOD_NUMCOL(x) ((x) & MC_DOD_NUMCOL_MASK)
157 #define MC_RANK_PRESENT 0x7c
159 #define MC_SAG_CH_0 0x80
160 #define MC_SAG_CH_1 0x84
161 #define MC_SAG_CH_2 0x88
162 #define MC_SAG_CH_3 0x8c
163 #define MC_SAG_CH_4 0x90
164 #define MC_SAG_CH_5 0x94
165 #define MC_SAG_CH_6 0x98
166 #define MC_SAG_CH_7 0x9c
168 #define MC_RIR_LIMIT_CH_0 0x40
169 #define MC_RIR_LIMIT_CH_1 0x44
170 #define MC_RIR_LIMIT_CH_2 0x48
171 #define MC_RIR_LIMIT_CH_3 0x4C
172 #define MC_RIR_LIMIT_CH_4 0x50
173 #define MC_RIR_LIMIT_CH_5 0x54
174 #define MC_RIR_LIMIT_CH_6 0x58
175 #define MC_RIR_LIMIT_CH_7 0x5C
176 #define MC_RIR_LIMIT_MASK ((1 << 10) - 1)
178 #define MC_RIR_WAY_CH 0x80
179 #define MC_RIR_WAY_OFFSET_MASK (((1 << 14) - 1) & ~0x7)
180 #define MC_RIR_WAY_RANK_MASK 0x7
187 #define MAX_DIMMS 3 /* Max DIMMS per channel */
188 #define MAX_MCR_FUNC 4
189 #define MAX_CHAN_FUNC 3
199 struct i7core_inject {
206 /* Error address mask */
207 int channel, dimm, rank, bank, page, col;
210 struct i7core_channel {
215 struct pci_id_descr {
222 struct pci_id_table {
223 const struct pci_id_descr *descr;
228 struct list_head list;
230 struct pci_dev **pdev;
232 struct mem_ctl_info *mci;
236 struct pci_dev *pci_noncore;
237 struct pci_dev *pci_mcr[MAX_MCR_FUNC + 1];
238 struct pci_dev *pci_ch[NUM_CHANS][MAX_CHAN_FUNC + 1];
240 struct i7core_dev *i7core_dev;
242 struct i7core_info info;
243 struct i7core_inject inject;
244 struct i7core_channel channel[NUM_CHANS];
246 int channels; /* Number of active channels */
248 int ce_count_available;
249 int csrow_map[NUM_CHANS][MAX_DIMMS];
251 /* ECC corrected errors counts per udimm */
252 unsigned long udimm_ce_count[MAX_DIMMS];
253 int udimm_last_ce_count[MAX_DIMMS];
254 /* ECC corrected errors counts per rdimm */
255 unsigned long rdimm_ce_count[NUM_CHANS][MAX_DIMMS];
256 int rdimm_last_ce_count[NUM_CHANS][MAX_DIMMS];
258 unsigned int is_registered;
261 struct edac_mce edac_mce;
263 /* Fifo double buffers */
264 struct mce mce_entry[MCE_LOG_LEN];
265 struct mce mce_outentry[MCE_LOG_LEN];
267 /* Fifo in/out counters */
268 unsigned mce_in, mce_out;
270 /* Count indicator to show errors not got */
271 unsigned mce_overrun;
273 /* Struct to control EDAC polling */
274 struct edac_pci_ctl_info *i7core_pci;
277 #define PCI_DESCR(device, function, device_id) \
279 .func = (function), \
280 .dev_id = (device_id)
282 static const struct pci_id_descr pci_dev_descr_i7core_nehalem[] = {
283 /* Memory controller */
284 { PCI_DESCR(3, 0, PCI_DEVICE_ID_INTEL_I7_MCR) },
285 { PCI_DESCR(3, 1, PCI_DEVICE_ID_INTEL_I7_MC_TAD) },
286 /* Exists only for RDIMM */
287 { PCI_DESCR(3, 2, PCI_DEVICE_ID_INTEL_I7_MC_RAS), .optional = 1 },
288 { PCI_DESCR(3, 4, PCI_DEVICE_ID_INTEL_I7_MC_TEST) },
291 { PCI_DESCR(4, 0, PCI_DEVICE_ID_INTEL_I7_MC_CH0_CTRL) },
292 { PCI_DESCR(4, 1, PCI_DEVICE_ID_INTEL_I7_MC_CH0_ADDR) },
293 { PCI_DESCR(4, 2, PCI_DEVICE_ID_INTEL_I7_MC_CH0_RANK) },
294 { PCI_DESCR(4, 3, PCI_DEVICE_ID_INTEL_I7_MC_CH0_TC) },
297 { PCI_DESCR(5, 0, PCI_DEVICE_ID_INTEL_I7_MC_CH1_CTRL) },
298 { PCI_DESCR(5, 1, PCI_DEVICE_ID_INTEL_I7_MC_CH1_ADDR) },
299 { PCI_DESCR(5, 2, PCI_DEVICE_ID_INTEL_I7_MC_CH1_RANK) },
300 { PCI_DESCR(5, 3, PCI_DEVICE_ID_INTEL_I7_MC_CH1_TC) },
303 { PCI_DESCR(6, 0, PCI_DEVICE_ID_INTEL_I7_MC_CH2_CTRL) },
304 { PCI_DESCR(6, 1, PCI_DEVICE_ID_INTEL_I7_MC_CH2_ADDR) },
305 { PCI_DESCR(6, 2, PCI_DEVICE_ID_INTEL_I7_MC_CH2_RANK) },
306 { PCI_DESCR(6, 3, PCI_DEVICE_ID_INTEL_I7_MC_CH2_TC) },
308 /* Generic Non-core registers */
310 * This is the PCI device on i7core and on Xeon 35xx (8086:2c41)
311 * On Xeon 55xx, however, it has a different id (8086:2c40). So,
312 * the probing code needs to test for the other address in case of
313 * failure of this one
315 { PCI_DESCR(0, 0, PCI_DEVICE_ID_INTEL_I7_NONCORE) },
319 static const struct pci_id_descr pci_dev_descr_lynnfield[] = {
320 { PCI_DESCR( 3, 0, PCI_DEVICE_ID_INTEL_LYNNFIELD_MCR) },
321 { PCI_DESCR( 3, 1, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_TAD) },
322 { PCI_DESCR( 3, 4, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_TEST) },
324 { PCI_DESCR( 4, 0, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH0_CTRL) },
325 { PCI_DESCR( 4, 1, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH0_ADDR) },
326 { PCI_DESCR( 4, 2, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH0_RANK) },
327 { PCI_DESCR( 4, 3, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH0_TC) },
329 { PCI_DESCR( 5, 0, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH1_CTRL) },
330 { PCI_DESCR( 5, 1, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH1_ADDR) },
331 { PCI_DESCR( 5, 2, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH1_RANK) },
332 { PCI_DESCR( 5, 3, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH1_TC) },
335 * This is the PCI device has an alternate address on some
336 * processors like Core i7 860
338 { PCI_DESCR( 0, 0, PCI_DEVICE_ID_INTEL_LYNNFIELD_NONCORE) },
341 static const struct pci_id_descr pci_dev_descr_i7core_westmere[] = {
342 /* Memory controller */
343 { PCI_DESCR(3, 0, PCI_DEVICE_ID_INTEL_LYNNFIELD_MCR_REV2) },
344 { PCI_DESCR(3, 1, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_TAD_REV2) },
345 /* Exists only for RDIMM */
346 { PCI_DESCR(3, 2, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_RAS_REV2), .optional = 1 },
347 { PCI_DESCR(3, 4, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_TEST_REV2) },
350 { PCI_DESCR(4, 0, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH0_CTRL_REV2) },
351 { PCI_DESCR(4, 1, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH0_ADDR_REV2) },
352 { PCI_DESCR(4, 2, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH0_RANK_REV2) },
353 { PCI_DESCR(4, 3, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH0_TC_REV2) },
356 { PCI_DESCR(5, 0, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH1_CTRL_REV2) },
357 { PCI_DESCR(5, 1, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH1_ADDR_REV2) },
358 { PCI_DESCR(5, 2, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH1_RANK_REV2) },
359 { PCI_DESCR(5, 3, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH1_TC_REV2) },
362 { PCI_DESCR(6, 0, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH2_CTRL_REV2) },
363 { PCI_DESCR(6, 1, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH2_ADDR_REV2) },
364 { PCI_DESCR(6, 2, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH2_RANK_REV2) },
365 { PCI_DESCR(6, 3, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH2_TC_REV2) },
367 /* Generic Non-core registers */
368 { PCI_DESCR(0, 0, PCI_DEVICE_ID_INTEL_LYNNFIELD_NONCORE_REV2) },
372 #define PCI_ID_TABLE_ENTRY(A) { .descr=A, .n_devs = ARRAY_SIZE(A) }
373 static const struct pci_id_table pci_dev_table[] = {
374 PCI_ID_TABLE_ENTRY(pci_dev_descr_i7core_nehalem),
375 PCI_ID_TABLE_ENTRY(pci_dev_descr_lynnfield),
376 PCI_ID_TABLE_ENTRY(pci_dev_descr_i7core_westmere),
380 * pci_device_id table for which devices we are looking for
382 static const struct pci_device_id i7core_pci_tbl[] __devinitdata = {
383 {PCI_DEVICE(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_X58_HUB_MGMT)},
384 {PCI_DEVICE(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_LYNNFIELD_QPI_LINK0)},
385 {0,} /* 0 terminated list. */
388 /****************************************************************************
389 Anciliary status routines
390 ****************************************************************************/
392 /* MC_CONTROL bits */
393 #define CH_ACTIVE(pvt, ch) ((pvt)->info.mc_control & (1 << (8 + ch)))
394 #define ECCx8(pvt) ((pvt)->info.mc_control & (1 << 1))
397 #define ECC_ENABLED(pvt) ((pvt)->info.mc_status & (1 << 4))
398 #define CH_DISABLED(pvt, ch) ((pvt)->info.mc_status & (1 << ch))
400 /* MC_MAX_DOD read functions */
401 static inline int numdimms(u32 dimms)
403 return (dimms & 0x3) + 1;
406 static inline int numrank(u32 rank)
408 static int ranks[4] = { 1, 2, 4, -EINVAL };
410 return ranks[rank & 0x3];
413 static inline int numbank(u32 bank)
415 static int banks[4] = { 4, 8, 16, -EINVAL };
417 return banks[bank & 0x3];
420 static inline int numrow(u32 row)
422 static int rows[8] = {
423 1 << 12, 1 << 13, 1 << 14, 1 << 15,
424 1 << 16, -EINVAL, -EINVAL, -EINVAL,
427 return rows[row & 0x7];
430 static inline int numcol(u32 col)
432 static int cols[8] = {
433 1 << 10, 1 << 11, 1 << 12, -EINVAL,
435 return cols[col & 0x3];
438 static struct i7core_dev *get_i7core_dev(u8 socket)
440 struct i7core_dev *i7core_dev;
442 list_for_each_entry(i7core_dev, &i7core_edac_list, list) {
443 if (i7core_dev->socket == socket)
450 static struct i7core_dev *alloc_i7core_dev(u8 socket,
451 const struct pci_id_table *table)
453 struct i7core_dev *i7core_dev;
455 i7core_dev = kzalloc(sizeof(*i7core_dev), GFP_KERNEL);
459 i7core_dev->pdev = kzalloc(sizeof(*i7core_dev->pdev) * table->n_devs,
461 if (!i7core_dev->pdev) {
466 i7core_dev->socket = socket;
467 i7core_dev->n_devs = table->n_devs;
468 list_add_tail(&i7core_dev->list, &i7core_edac_list);
473 /****************************************************************************
474 Memory check routines
475 ****************************************************************************/
476 static struct pci_dev *get_pdev_slot_func(u8 socket, unsigned slot,
479 struct i7core_dev *i7core_dev = get_i7core_dev(socket);
485 for (i = 0; i < i7core_dev->n_devs; i++) {
486 if (!i7core_dev->pdev[i])
489 if (PCI_SLOT(i7core_dev->pdev[i]->devfn) == slot &&
490 PCI_FUNC(i7core_dev->pdev[i]->devfn) == func) {
491 return i7core_dev->pdev[i];
499 * i7core_get_active_channels() - gets the number of channels and csrows
500 * @socket: Quick Path Interconnect socket
501 * @channels: Number of channels that will be returned
502 * @csrows: Number of csrows found
504 * Since EDAC core needs to know in advance the number of available channels
505 * and csrows, in order to allocate memory for csrows/channels, it is needed
506 * to run two similar steps. At the first step, implemented on this function,
507 * it checks the number of csrows/channels present at one socket.
508 * this is used in order to properly allocate the size of mci components.
510 * It should be noticed that none of the current available datasheets explain
511 * or even mention how csrows are seen by the memory controller. So, we need
512 * to add a fake description for csrows.
513 * So, this driver is attributing one DIMM memory for one csrow.
515 static int i7core_get_active_channels(const u8 socket, unsigned *channels,
518 struct pci_dev *pdev = NULL;
525 pdev = get_pdev_slot_func(socket, 3, 0);
527 i7core_printk(KERN_ERR, "Couldn't find socket %d fn 3.0!!!\n",
532 /* Device 3 function 0 reads */
533 pci_read_config_dword(pdev, MC_STATUS, &status);
534 pci_read_config_dword(pdev, MC_CONTROL, &control);
536 for (i = 0; i < NUM_CHANS; i++) {
538 /* Check if the channel is active */
539 if (!(control & (1 << (8 + i))))
542 /* Check if the channel is disabled */
543 if (status & (1 << i))
546 pdev = get_pdev_slot_func(socket, i + 4, 1);
548 i7core_printk(KERN_ERR, "Couldn't find socket %d "
553 /* Devices 4-6 function 1 */
554 pci_read_config_dword(pdev,
555 MC_DOD_CH_DIMM0, &dimm_dod[0]);
556 pci_read_config_dword(pdev,
557 MC_DOD_CH_DIMM1, &dimm_dod[1]);
558 pci_read_config_dword(pdev,
559 MC_DOD_CH_DIMM2, &dimm_dod[2]);
563 for (j = 0; j < 3; j++) {
564 if (!DIMM_PRESENT(dimm_dod[j]))
570 debugf0("Number of active channels on socket %d: %d\n",
576 static int get_dimm_config(const struct mem_ctl_info *mci, int *csrow)
578 struct i7core_pvt *pvt = mci->pvt_info;
579 struct csrow_info *csr;
580 struct pci_dev *pdev;
582 unsigned long last_page = 0;
586 /* Get data from the MC register, function 0 */
587 pdev = pvt->pci_mcr[0];
591 /* Device 3 function 0 reads */
592 pci_read_config_dword(pdev, MC_CONTROL, &pvt->info.mc_control);
593 pci_read_config_dword(pdev, MC_STATUS, &pvt->info.mc_status);
594 pci_read_config_dword(pdev, MC_MAX_DOD, &pvt->info.max_dod);
595 pci_read_config_dword(pdev, MC_CHANNEL_MAPPER, &pvt->info.ch_map);
597 debugf0("QPI %d control=0x%08x status=0x%08x dod=0x%08x map=0x%08x\n",
598 pvt->i7core_dev->socket, pvt->info.mc_control, pvt->info.mc_status,
599 pvt->info.max_dod, pvt->info.ch_map);
601 if (ECC_ENABLED(pvt)) {
602 debugf0("ECC enabled with x%d SDCC\n", ECCx8(pvt) ? 8 : 4);
604 mode = EDAC_S8ECD8ED;
606 mode = EDAC_S4ECD4ED;
608 debugf0("ECC disabled\n");
612 /* FIXME: need to handle the error codes */
613 debugf0("DOD Max limits: DIMMS: %d, %d-ranked, %d-banked "
615 numdimms(pvt->info.max_dod),
616 numrank(pvt->info.max_dod >> 2),
617 numbank(pvt->info.max_dod >> 4),
618 numrow(pvt->info.max_dod >> 6),
619 numcol(pvt->info.max_dod >> 9));
621 for (i = 0; i < NUM_CHANS; i++) {
622 u32 data, dimm_dod[3], value[8];
624 if (!pvt->pci_ch[i][0])
627 if (!CH_ACTIVE(pvt, i)) {
628 debugf0("Channel %i is not active\n", i);
631 if (CH_DISABLED(pvt, i)) {
632 debugf0("Channel %i is disabled\n", i);
636 /* Devices 4-6 function 0 */
637 pci_read_config_dword(pvt->pci_ch[i][0],
638 MC_CHANNEL_DIMM_INIT_PARAMS, &data);
640 pvt->channel[i].ranks = (data & QUAD_RANK_PRESENT) ?
643 if (data & REGISTERED_DIMM)
648 if (data & THREE_DIMMS_PRESENT)
649 pvt->channel[i].dimms = 3;
650 else if (data & SINGLE_QUAD_RANK_PRESENT)
651 pvt->channel[i].dimms = 1;
653 pvt->channel[i].dimms = 2;
656 /* Devices 4-6 function 1 */
657 pci_read_config_dword(pvt->pci_ch[i][1],
658 MC_DOD_CH_DIMM0, &dimm_dod[0]);
659 pci_read_config_dword(pvt->pci_ch[i][1],
660 MC_DOD_CH_DIMM1, &dimm_dod[1]);
661 pci_read_config_dword(pvt->pci_ch[i][1],
662 MC_DOD_CH_DIMM2, &dimm_dod[2]);
664 debugf0("Ch%d phy rd%d, wr%d (0x%08x): "
665 "%d ranks, %cDIMMs\n",
667 RDLCH(pvt->info.ch_map, i), WRLCH(pvt->info.ch_map, i),
669 pvt->channel[i].ranks,
670 (data & REGISTERED_DIMM) ? 'R' : 'U');
672 for (j = 0; j < 3; j++) {
673 u32 banks, ranks, rows, cols;
676 if (!DIMM_PRESENT(dimm_dod[j]))
679 banks = numbank(MC_DOD_NUMBANK(dimm_dod[j]));
680 ranks = numrank(MC_DOD_NUMRANK(dimm_dod[j]));
681 rows = numrow(MC_DOD_NUMROW(dimm_dod[j]));
682 cols = numcol(MC_DOD_NUMCOL(dimm_dod[j]));
684 /* DDR3 has 8 I/O banks */
685 size = (rows * cols * banks * ranks) >> (20 - 3);
687 pvt->channel[i].dimms++;
689 debugf0("\tdimm %d %d Mb offset: %x, "
690 "bank: %d, rank: %d, row: %#x, col: %#x\n",
692 RANKOFFSET(dimm_dod[j]),
693 banks, ranks, rows, cols);
695 npages = MiB_TO_PAGES(size);
697 csr = &mci->csrows[*csrow];
698 csr->first_page = last_page + 1;
700 csr->last_page = last_page;
701 csr->nr_pages = npages;
705 csr->csrow_idx = *csrow;
706 csr->nr_channels = 1;
708 csr->channels[0].chan_idx = i;
709 csr->channels[0].ce_count = 0;
711 pvt->csrow_map[i][j] = *csrow;
721 csr->dtype = DEV_X16;
724 csr->dtype = DEV_UNKNOWN;
727 csr->edac_mode = mode;
733 pci_read_config_dword(pdev, MC_SAG_CH_0, &value[0]);
734 pci_read_config_dword(pdev, MC_SAG_CH_1, &value[1]);
735 pci_read_config_dword(pdev, MC_SAG_CH_2, &value[2]);
736 pci_read_config_dword(pdev, MC_SAG_CH_3, &value[3]);
737 pci_read_config_dword(pdev, MC_SAG_CH_4, &value[4]);
738 pci_read_config_dword(pdev, MC_SAG_CH_5, &value[5]);
739 pci_read_config_dword(pdev, MC_SAG_CH_6, &value[6]);
740 pci_read_config_dword(pdev, MC_SAG_CH_7, &value[7]);
741 debugf1("\t[%i] DIVBY3\tREMOVED\tOFFSET\n", i);
742 for (j = 0; j < 8; j++)
743 debugf1("\t\t%#x\t%#x\t%#x\n",
744 (value[j] >> 27) & 0x1,
745 (value[j] >> 24) & 0x7,
746 (value[j] && ((1 << 24) - 1)));
752 /****************************************************************************
753 Error insertion routines
754 ****************************************************************************/
756 /* The i7core has independent error injection features per channel.
757 However, to have a simpler code, we don't allow enabling error injection
758 on more than one channel.
759 Also, since a change at an inject parameter will be applied only at enable,
760 we're disabling error injection on all write calls to the sysfs nodes that
761 controls the error code injection.
763 static int disable_inject(const struct mem_ctl_info *mci)
765 struct i7core_pvt *pvt = mci->pvt_info;
767 pvt->inject.enable = 0;
769 if (!pvt->pci_ch[pvt->inject.channel][0])
772 pci_write_config_dword(pvt->pci_ch[pvt->inject.channel][0],
773 MC_CHANNEL_ERROR_INJECT, 0);
779 * i7core inject inject.section
781 * accept and store error injection inject.section value
782 * bit 0 - refers to the lower 32-byte half cacheline
783 * bit 1 - refers to the upper 32-byte half cacheline
785 static ssize_t i7core_inject_section_store(struct mem_ctl_info *mci,
786 const char *data, size_t count)
788 struct i7core_pvt *pvt = mci->pvt_info;
792 if (pvt->inject.enable)
795 rc = strict_strtoul(data, 10, &value);
796 if ((rc < 0) || (value > 3))
799 pvt->inject.section = (u32) value;
803 static ssize_t i7core_inject_section_show(struct mem_ctl_info *mci,
806 struct i7core_pvt *pvt = mci->pvt_info;
807 return sprintf(data, "0x%08x\n", pvt->inject.section);
813 * accept and store error injection inject.section value
814 * bit 0 - repeat enable - Enable error repetition
815 * bit 1 - inject ECC error
816 * bit 2 - inject parity error
818 static ssize_t i7core_inject_type_store(struct mem_ctl_info *mci,
819 const char *data, size_t count)
821 struct i7core_pvt *pvt = mci->pvt_info;
825 if (pvt->inject.enable)
828 rc = strict_strtoul(data, 10, &value);
829 if ((rc < 0) || (value > 7))
832 pvt->inject.type = (u32) value;
836 static ssize_t i7core_inject_type_show(struct mem_ctl_info *mci,
839 struct i7core_pvt *pvt = mci->pvt_info;
840 return sprintf(data, "0x%08x\n", pvt->inject.type);
844 * i7core_inject_inject.eccmask_store
846 * The type of error (UE/CE) will depend on the inject.eccmask value:
847 * Any bits set to a 1 will flip the corresponding ECC bit
848 * Correctable errors can be injected by flipping 1 bit or the bits within
849 * a symbol pair (2 consecutive aligned 8-bit pairs - i.e. 7:0 and 15:8 or
850 * 23:16 and 31:24). Flipping bits in two symbol pairs will cause an
851 * uncorrectable error to be injected.
853 static ssize_t i7core_inject_eccmask_store(struct mem_ctl_info *mci,
854 const char *data, size_t count)
856 struct i7core_pvt *pvt = mci->pvt_info;
860 if (pvt->inject.enable)
863 rc = strict_strtoul(data, 10, &value);
867 pvt->inject.eccmask = (u32) value;
871 static ssize_t i7core_inject_eccmask_show(struct mem_ctl_info *mci,
874 struct i7core_pvt *pvt = mci->pvt_info;
875 return sprintf(data, "0x%08x\n", pvt->inject.eccmask);
881 * The type of error (UE/CE) will depend on the inject.eccmask value:
882 * Any bits set to a 1 will flip the corresponding ECC bit
883 * Correctable errors can be injected by flipping 1 bit or the bits within
884 * a symbol pair (2 consecutive aligned 8-bit pairs - i.e. 7:0 and 15:8 or
885 * 23:16 and 31:24). Flipping bits in two symbol pairs will cause an
886 * uncorrectable error to be injected.
889 #define DECLARE_ADDR_MATCH(param, limit) \
890 static ssize_t i7core_inject_store_##param( \
891 struct mem_ctl_info *mci, \
892 const char *data, size_t count) \
894 struct i7core_pvt *pvt; \
898 debugf1("%s()\n", __func__); \
899 pvt = mci->pvt_info; \
901 if (pvt->inject.enable) \
902 disable_inject(mci); \
904 if (!strcasecmp(data, "any") || !strcasecmp(data, "any\n"))\
907 rc = strict_strtoul(data, 10, &value); \
908 if ((rc < 0) || (value >= limit)) \
912 pvt->inject.param = value; \
917 static ssize_t i7core_inject_show_##param( \
918 struct mem_ctl_info *mci, \
921 struct i7core_pvt *pvt; \
923 pvt = mci->pvt_info; \
924 debugf1("%s() pvt=%p\n", __func__, pvt); \
925 if (pvt->inject.param < 0) \
926 return sprintf(data, "any\n"); \
928 return sprintf(data, "%d\n", pvt->inject.param);\
931 #define ATTR_ADDR_MATCH(param) \
935 .mode = (S_IRUGO | S_IWUSR) \
937 .show = i7core_inject_show_##param, \
938 .store = i7core_inject_store_##param, \
941 DECLARE_ADDR_MATCH(channel, 3);
942 DECLARE_ADDR_MATCH(dimm, 3);
943 DECLARE_ADDR_MATCH(rank, 4);
944 DECLARE_ADDR_MATCH(bank, 32);
945 DECLARE_ADDR_MATCH(page, 0x10000);
946 DECLARE_ADDR_MATCH(col, 0x4000);
948 static int write_and_test(struct pci_dev *dev, const int where, const u32 val)
953 debugf0("setting pci %02x:%02x.%x reg=%02x value=%08x\n",
954 dev->bus->number, PCI_SLOT(dev->devfn), PCI_FUNC(dev->devfn),
957 for (count = 0; count < 10; count++) {
960 pci_write_config_dword(dev, where, val);
961 pci_read_config_dword(dev, where, &read);
967 i7core_printk(KERN_ERR, "Error during set pci %02x:%02x.%x reg=%02x "
968 "write=%08x. Read=%08x\n",
969 dev->bus->number, PCI_SLOT(dev->devfn), PCI_FUNC(dev->devfn),
976 * This routine prepares the Memory Controller for error injection.
977 * The error will be injected when some process tries to write to the
978 * memory that matches the given criteria.
979 * The criteria can be set in terms of a mask where dimm, rank, bank, page
980 * and col can be specified.
981 * A -1 value for any of the mask items will make the MCU to ignore
982 * that matching criteria for error injection.
984 * It should be noticed that the error will only happen after a write operation
985 * on a memory that matches the condition. if REPEAT_EN is not enabled at
986 * inject mask, then it will produce just one error. Otherwise, it will repeat
987 * until the injectmask would be cleaned.
989 * FIXME: This routine assumes that MAXNUMDIMMS value of MC_MAX_DOD
990 * is reliable enough to check if the MC is using the
991 * three channels. However, this is not clear at the datasheet.
993 static ssize_t i7core_inject_enable_store(struct mem_ctl_info *mci,
994 const char *data, size_t count)
996 struct i7core_pvt *pvt = mci->pvt_info;
1002 if (!pvt->pci_ch[pvt->inject.channel][0])
1005 rc = strict_strtoul(data, 10, &enable);
1010 pvt->inject.enable = 1;
1012 disable_inject(mci);
1016 /* Sets pvt->inject.dimm mask */
1017 if (pvt->inject.dimm < 0)
1020 if (pvt->channel[pvt->inject.channel].dimms > 2)
1021 mask |= (pvt->inject.dimm & 0x3LL) << 35;
1023 mask |= (pvt->inject.dimm & 0x1LL) << 36;
1026 /* Sets pvt->inject.rank mask */
1027 if (pvt->inject.rank < 0)
1030 if (pvt->channel[pvt->inject.channel].dimms > 2)
1031 mask |= (pvt->inject.rank & 0x1LL) << 34;
1033 mask |= (pvt->inject.rank & 0x3LL) << 34;
1036 /* Sets pvt->inject.bank mask */
1037 if (pvt->inject.bank < 0)
1040 mask |= (pvt->inject.bank & 0x15LL) << 30;
1042 /* Sets pvt->inject.page mask */
1043 if (pvt->inject.page < 0)
1046 mask |= (pvt->inject.page & 0xffff) << 14;
1048 /* Sets pvt->inject.column mask */
1049 if (pvt->inject.col < 0)
1052 mask |= (pvt->inject.col & 0x3fff);
1056 * bits 1-2: MASK_HALF_CACHELINE
1058 * bit 4: INJECT_ADDR_PARITY
1061 injectmask = (pvt->inject.type & 1) |
1062 (pvt->inject.section & 0x3) << 1 |
1063 (pvt->inject.type & 0x6) << (3 - 1);
1065 /* Unlock writes to registers - this register is write only */
1066 pci_write_config_dword(pvt->pci_noncore,
1067 MC_CFG_CONTROL, 0x2);
1069 write_and_test(pvt->pci_ch[pvt->inject.channel][0],
1070 MC_CHANNEL_ADDR_MATCH, mask);
1071 write_and_test(pvt->pci_ch[pvt->inject.channel][0],
1072 MC_CHANNEL_ADDR_MATCH + 4, mask >> 32L);
1074 write_and_test(pvt->pci_ch[pvt->inject.channel][0],
1075 MC_CHANNEL_ERROR_MASK, pvt->inject.eccmask);
1077 write_and_test(pvt->pci_ch[pvt->inject.channel][0],
1078 MC_CHANNEL_ERROR_INJECT, injectmask);
1081 * This is something undocumented, based on my tests
1082 * Without writing 8 to this register, errors aren't injected. Not sure
1085 pci_write_config_dword(pvt->pci_noncore,
1088 debugf0("Error inject addr match 0x%016llx, ecc 0x%08x,"
1090 mask, pvt->inject.eccmask, injectmask);
1096 static ssize_t i7core_inject_enable_show(struct mem_ctl_info *mci,
1099 struct i7core_pvt *pvt = mci->pvt_info;
1102 if (!pvt->pci_ch[pvt->inject.channel][0])
1105 pci_read_config_dword(pvt->pci_ch[pvt->inject.channel][0],
1106 MC_CHANNEL_ERROR_INJECT, &injectmask);
1108 debugf0("Inject error read: 0x%018x\n", injectmask);
1110 if (injectmask & 0x0c)
1111 pvt->inject.enable = 1;
1113 return sprintf(data, "%d\n", pvt->inject.enable);
1116 #define DECLARE_COUNTER(param) \
1117 static ssize_t i7core_show_counter_##param( \
1118 struct mem_ctl_info *mci, \
1121 struct i7core_pvt *pvt = mci->pvt_info; \
1123 debugf1("%s() \n", __func__); \
1124 if (!pvt->ce_count_available || (pvt->is_registered)) \
1125 return sprintf(data, "data unavailable\n"); \
1126 return sprintf(data, "%lu\n", \
1127 pvt->udimm_ce_count[param]); \
1130 #define ATTR_COUNTER(param) \
1133 .name = __stringify(udimm##param), \
1134 .mode = (S_IRUGO | S_IWUSR) \
1136 .show = i7core_show_counter_##param \
1147 static const struct mcidev_sysfs_attribute i7core_addrmatch_attrs[] = {
1148 ATTR_ADDR_MATCH(channel),
1149 ATTR_ADDR_MATCH(dimm),
1150 ATTR_ADDR_MATCH(rank),
1151 ATTR_ADDR_MATCH(bank),
1152 ATTR_ADDR_MATCH(page),
1153 ATTR_ADDR_MATCH(col),
1154 { } /* End of list */
1157 static const struct mcidev_sysfs_group i7core_inject_addrmatch = {
1158 .name = "inject_addrmatch",
1159 .mcidev_attr = i7core_addrmatch_attrs,
1162 static const struct mcidev_sysfs_attribute i7core_udimm_counters_attrs[] = {
1166 { .attr = { .name = NULL } }
1169 static const struct mcidev_sysfs_group i7core_udimm_counters = {
1170 .name = "all_channel_counts",
1171 .mcidev_attr = i7core_udimm_counters_attrs,
1174 static const struct mcidev_sysfs_attribute i7core_sysfs_rdimm_attrs[] = {
1177 .name = "inject_section",
1178 .mode = (S_IRUGO | S_IWUSR)
1180 .show = i7core_inject_section_show,
1181 .store = i7core_inject_section_store,
1184 .name = "inject_type",
1185 .mode = (S_IRUGO | S_IWUSR)
1187 .show = i7core_inject_type_show,
1188 .store = i7core_inject_type_store,
1191 .name = "inject_eccmask",
1192 .mode = (S_IRUGO | S_IWUSR)
1194 .show = i7core_inject_eccmask_show,
1195 .store = i7core_inject_eccmask_store,
1197 .grp = &i7core_inject_addrmatch,
1200 .name = "inject_enable",
1201 .mode = (S_IRUGO | S_IWUSR)
1203 .show = i7core_inject_enable_show,
1204 .store = i7core_inject_enable_store,
1206 { } /* End of list */
1209 static const struct mcidev_sysfs_attribute i7core_sysfs_udimm_attrs[] = {
1212 .name = "inject_section",
1213 .mode = (S_IRUGO | S_IWUSR)
1215 .show = i7core_inject_section_show,
1216 .store = i7core_inject_section_store,
1219 .name = "inject_type",
1220 .mode = (S_IRUGO | S_IWUSR)
1222 .show = i7core_inject_type_show,
1223 .store = i7core_inject_type_store,
1226 .name = "inject_eccmask",
1227 .mode = (S_IRUGO | S_IWUSR)
1229 .show = i7core_inject_eccmask_show,
1230 .store = i7core_inject_eccmask_store,
1232 .grp = &i7core_inject_addrmatch,
1235 .name = "inject_enable",
1236 .mode = (S_IRUGO | S_IWUSR)
1238 .show = i7core_inject_enable_show,
1239 .store = i7core_inject_enable_store,
1241 .grp = &i7core_udimm_counters,
1243 { } /* End of list */
1246 /****************************************************************************
1247 Device initialization routines: put/get, init/exit
1248 ****************************************************************************/
1251 * i7core_put_devices 'put' all the devices that we have
1252 * reserved via 'get'
1254 static void i7core_put_devices(struct i7core_dev *i7core_dev)
1258 debugf0(__FILE__ ": %s()\n", __func__);
1259 for (i = 0; i < i7core_dev->n_devs; i++) {
1260 struct pci_dev *pdev = i7core_dev->pdev[i];
1263 debugf0("Removing dev %02x:%02x.%d\n",
1265 PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn));
1268 kfree(i7core_dev->pdev);
1271 static void i7core_put_all_devices(void)
1273 struct i7core_dev *i7core_dev, *tmp;
1275 list_for_each_entry_safe(i7core_dev, tmp, &i7core_edac_list, list) {
1276 i7core_put_devices(i7core_dev);
1277 list_del(&i7core_dev->list);
1282 static void __init i7core_xeon_pci_fixup(const struct pci_id_table *table)
1284 struct pci_dev *pdev = NULL;
1288 * On Xeon 55xx, the Intel Quckpath Arch Generic Non-core pci buses
1289 * aren't announced by acpi. So, we need to use a legacy scan probing
1292 while (table && table->descr) {
1293 pdev = pci_get_device(PCI_VENDOR_ID_INTEL, table->descr[0].dev_id, NULL);
1294 if (unlikely(!pdev)) {
1295 for (i = 0; i < MAX_SOCKET_BUSES; i++)
1296 pcibios_scan_specific_bus(255-i);
1303 static unsigned i7core_pci_lastbus(void)
1305 int last_bus = 0, bus;
1306 struct pci_bus *b = NULL;
1308 while ((b = pci_find_next_bus(b)) != NULL) {
1310 debugf0("Found bus %d\n", bus);
1315 debugf0("Last bus %d\n", last_bus);
1321 * i7core_get_devices Find and perform 'get' operation on the MCH's
1322 * device/functions we want to reference for this driver
1324 * Need to 'get' device 16 func 1 and func 2
1326 static int i7core_get_onedevice(struct pci_dev **prev,
1327 const struct pci_id_table *table,
1328 const unsigned devno,
1329 const unsigned last_bus)
1331 struct i7core_dev *i7core_dev;
1332 const struct pci_id_descr *dev_descr = &table->descr[devno];
1334 struct pci_dev *pdev = NULL;
1338 pdev = pci_get_device(PCI_VENDOR_ID_INTEL,
1339 dev_descr->dev_id, *prev);
1342 * On Xeon 55xx, the Intel Quckpath Arch Generic Non-core regs
1343 * is at addr 8086:2c40, instead of 8086:2c41. So, we need
1344 * to probe for the alternate address in case of failure
1346 if (dev_descr->dev_id == PCI_DEVICE_ID_INTEL_I7_NONCORE && !pdev)
1347 pdev = pci_get_device(PCI_VENDOR_ID_INTEL,
1348 PCI_DEVICE_ID_INTEL_I7_NONCORE_ALT, *prev);
1350 if (dev_descr->dev_id == PCI_DEVICE_ID_INTEL_LYNNFIELD_NONCORE && !pdev)
1351 pdev = pci_get_device(PCI_VENDOR_ID_INTEL,
1352 PCI_DEVICE_ID_INTEL_LYNNFIELD_NONCORE_ALT,
1361 if (dev_descr->optional)
1367 i7core_printk(KERN_INFO,
1368 "Device not found: dev %02x.%d PCI ID %04x:%04x\n",
1369 dev_descr->dev, dev_descr->func,
1370 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
1372 /* End of list, leave */
1375 bus = pdev->bus->number;
1377 socket = last_bus - bus;
1379 i7core_dev = get_i7core_dev(socket);
1381 i7core_dev = alloc_i7core_dev(socket, table);
1386 if (i7core_dev->pdev[devno]) {
1387 i7core_printk(KERN_ERR,
1388 "Duplicated device for "
1389 "dev %02x:%02x.%d PCI ID %04x:%04x\n",
1390 bus, dev_descr->dev, dev_descr->func,
1391 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
1396 i7core_dev->pdev[devno] = pdev;
1399 if (unlikely(PCI_SLOT(pdev->devfn) != dev_descr->dev ||
1400 PCI_FUNC(pdev->devfn) != dev_descr->func)) {
1401 i7core_printk(KERN_ERR,
1402 "Device PCI ID %04x:%04x "
1403 "has dev %02x:%02x.%d instead of dev %02x:%02x.%d\n",
1404 PCI_VENDOR_ID_INTEL, dev_descr->dev_id,
1405 bus, PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn),
1406 bus, dev_descr->dev, dev_descr->func);
1410 /* Be sure that the device is enabled */
1411 if (unlikely(pci_enable_device(pdev) < 0)) {
1412 i7core_printk(KERN_ERR,
1414 "dev %02x:%02x.%d PCI ID %04x:%04x\n",
1415 bus, dev_descr->dev, dev_descr->func,
1416 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
1420 debugf0("Detected socket %d dev %02x:%02x.%d PCI ID %04x:%04x\n",
1421 socket, bus, dev_descr->dev,
1423 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
1430 static int i7core_get_devices(const struct pci_id_table *table)
1432 int i, rc, last_bus;
1433 struct pci_dev *pdev = NULL;
1435 last_bus = i7core_pci_lastbus();
1437 while (table && table->descr) {
1438 for (i = 0; i < table->n_devs; i++) {
1441 rc = i7core_get_onedevice(&pdev, table, i,
1448 i7core_put_all_devices();
1459 static int mci_bind_devs(struct mem_ctl_info *mci,
1460 struct i7core_dev *i7core_dev)
1462 struct i7core_pvt *pvt = mci->pvt_info;
1463 struct pci_dev *pdev;
1466 /* Associates i7core_dev and mci for future usage */
1467 pvt->i7core_dev = i7core_dev;
1468 i7core_dev->mci = mci;
1470 pvt->is_registered = 0;
1471 for (i = 0; i < i7core_dev->n_devs; i++) {
1472 pdev = i7core_dev->pdev[i];
1476 func = PCI_FUNC(pdev->devfn);
1477 slot = PCI_SLOT(pdev->devfn);
1479 if (unlikely(func > MAX_MCR_FUNC))
1481 pvt->pci_mcr[func] = pdev;
1482 } else if (likely(slot >= 4 && slot < 4 + NUM_CHANS)) {
1483 if (unlikely(func > MAX_CHAN_FUNC))
1485 pvt->pci_ch[slot - 4][func] = pdev;
1486 } else if (!slot && !func)
1487 pvt->pci_noncore = pdev;
1491 debugf0("Associated fn %d.%d, dev = %p, socket %d\n",
1492 PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn),
1493 pdev, i7core_dev->socket);
1495 if (PCI_SLOT(pdev->devfn) == 3 &&
1496 PCI_FUNC(pdev->devfn) == 2)
1497 pvt->is_registered = 1;
1503 i7core_printk(KERN_ERR, "Device %d, function %d "
1504 "is out of the expected range\n",
1509 /****************************************************************************
1510 Error check routines
1511 ****************************************************************************/
1512 static void i7core_rdimm_update_csrow(struct mem_ctl_info *mci,
1518 struct i7core_pvt *pvt = mci->pvt_info;
1519 int row = pvt->csrow_map[chan][dimm], i;
1521 for (i = 0; i < add; i++) {
1522 msg = kasprintf(GFP_KERNEL, "Corrected error "
1523 "(Socket=%d channel=%d dimm=%d)",
1524 pvt->i7core_dev->socket, chan, dimm);
1526 edac_mc_handle_fbd_ce(mci, row, 0, msg);
1531 static void i7core_rdimm_update_ce_count(struct mem_ctl_info *mci,
1537 struct i7core_pvt *pvt = mci->pvt_info;
1538 int add0 = 0, add1 = 0, add2 = 0;
1539 /* Updates CE counters if it is not the first time here */
1540 if (pvt->ce_count_available) {
1541 /* Updates CE counters */
1543 add2 = new2 - pvt->rdimm_last_ce_count[chan][2];
1544 add1 = new1 - pvt->rdimm_last_ce_count[chan][1];
1545 add0 = new0 - pvt->rdimm_last_ce_count[chan][0];
1549 pvt->rdimm_ce_count[chan][2] += add2;
1553 pvt->rdimm_ce_count[chan][1] += add1;
1557 pvt->rdimm_ce_count[chan][0] += add0;
1559 pvt->ce_count_available = 1;
1561 /* Store the new values */
1562 pvt->rdimm_last_ce_count[chan][2] = new2;
1563 pvt->rdimm_last_ce_count[chan][1] = new1;
1564 pvt->rdimm_last_ce_count[chan][0] = new0;
1566 /*updated the edac core */
1568 i7core_rdimm_update_csrow(mci, chan, 0, add0);
1570 i7core_rdimm_update_csrow(mci, chan, 1, add1);
1572 i7core_rdimm_update_csrow(mci, chan, 2, add2);
1576 static void i7core_rdimm_check_mc_ecc_err(struct mem_ctl_info *mci)
1578 struct i7core_pvt *pvt = mci->pvt_info;
1580 int i, new0, new1, new2;
1582 /*Read DEV 3: FUN 2: MC_COR_ECC_CNT regs directly*/
1583 pci_read_config_dword(pvt->pci_mcr[2], MC_COR_ECC_CNT_0,
1585 pci_read_config_dword(pvt->pci_mcr[2], MC_COR_ECC_CNT_1,
1587 pci_read_config_dword(pvt->pci_mcr[2], MC_COR_ECC_CNT_2,
1589 pci_read_config_dword(pvt->pci_mcr[2], MC_COR_ECC_CNT_3,
1591 pci_read_config_dword(pvt->pci_mcr[2], MC_COR_ECC_CNT_4,
1593 pci_read_config_dword(pvt->pci_mcr[2], MC_COR_ECC_CNT_5,
1595 for (i = 0 ; i < 3; i++) {
1596 debugf3("MC_COR_ECC_CNT%d = 0x%x; MC_COR_ECC_CNT%d = 0x%x\n",
1597 (i * 2), rcv[i][0], (i * 2) + 1, rcv[i][1]);
1598 /*if the channel has 3 dimms*/
1599 if (pvt->channel[i].dimms > 2) {
1600 new0 = DIMM_BOT_COR_ERR(rcv[i][0]);
1601 new1 = DIMM_TOP_COR_ERR(rcv[i][0]);
1602 new2 = DIMM_BOT_COR_ERR(rcv[i][1]);
1604 new0 = DIMM_TOP_COR_ERR(rcv[i][0]) +
1605 DIMM_BOT_COR_ERR(rcv[i][0]);
1606 new1 = DIMM_TOP_COR_ERR(rcv[i][1]) +
1607 DIMM_BOT_COR_ERR(rcv[i][1]);
1611 i7core_rdimm_update_ce_count(mci, i, new0, new1, new2);
1615 /* This function is based on the device 3 function 4 registers as described on:
1616 * Intel Xeon Processor 5500 Series Datasheet Volume 2
1617 * http://www.intel.com/Assets/PDF/datasheet/321322.pdf
1618 * also available at:
1619 * http://www.arrownac.com/manufacturers/intel/s/nehalem/5500-datasheet-v2.pdf
1621 static void i7core_udimm_check_mc_ecc_err(struct mem_ctl_info *mci)
1623 struct i7core_pvt *pvt = mci->pvt_info;
1625 int new0, new1, new2;
1627 if (!pvt->pci_mcr[4]) {
1628 debugf0("%s MCR registers not found\n", __func__);
1632 /* Corrected test errors */
1633 pci_read_config_dword(pvt->pci_mcr[4], MC_TEST_ERR_RCV1, &rcv1);
1634 pci_read_config_dword(pvt->pci_mcr[4], MC_TEST_ERR_RCV0, &rcv0);
1636 /* Store the new values */
1637 new2 = DIMM2_COR_ERR(rcv1);
1638 new1 = DIMM1_COR_ERR(rcv0);
1639 new0 = DIMM0_COR_ERR(rcv0);
1641 /* Updates CE counters if it is not the first time here */
1642 if (pvt->ce_count_available) {
1643 /* Updates CE counters */
1644 int add0, add1, add2;
1646 add2 = new2 - pvt->udimm_last_ce_count[2];
1647 add1 = new1 - pvt->udimm_last_ce_count[1];
1648 add0 = new0 - pvt->udimm_last_ce_count[0];
1652 pvt->udimm_ce_count[2] += add2;
1656 pvt->udimm_ce_count[1] += add1;
1660 pvt->udimm_ce_count[0] += add0;
1662 if (add0 | add1 | add2)
1663 i7core_printk(KERN_ERR, "New Corrected error(s): "
1664 "dimm0: +%d, dimm1: +%d, dimm2 +%d\n",
1667 pvt->ce_count_available = 1;
1669 /* Store the new values */
1670 pvt->udimm_last_ce_count[2] = new2;
1671 pvt->udimm_last_ce_count[1] = new1;
1672 pvt->udimm_last_ce_count[0] = new0;
1676 * According with tables E-11 and E-12 of chapter E.3.3 of Intel 64 and IA-32
1677 * Architectures Software Developer’s Manual Volume 3B.
1678 * Nehalem are defined as family 0x06, model 0x1a
1680 * The MCA registers used here are the following ones:
1681 * struct mce field MCA Register
1682 * m->status MSR_IA32_MC8_STATUS
1683 * m->addr MSR_IA32_MC8_ADDR
1684 * m->misc MSR_IA32_MC8_MISC
1685 * In the case of Nehalem, the error information is masked at .status and .misc
1688 static void i7core_mce_output_error(struct mem_ctl_info *mci,
1689 const struct mce *m)
1691 struct i7core_pvt *pvt = mci->pvt_info;
1692 char *type, *optype, *err, *msg;
1693 unsigned long error = m->status & 0x1ff0000l;
1694 u32 optypenum = (m->status >> 4) & 0x07;
1695 u32 core_err_cnt = (m->status >> 38) && 0x7fff;
1696 u32 dimm = (m->misc >> 16) & 0x3;
1697 u32 channel = (m->misc >> 18) & 0x3;
1698 u32 syndrome = m->misc >> 32;
1699 u32 errnum = find_first_bit(&error, 32);
1702 if (m->mcgstatus & 1)
1707 switch (optypenum) {
1709 optype = "generic undef request";
1712 optype = "read error";
1715 optype = "write error";
1718 optype = "addr/cmd error";
1721 optype = "scrubbing error";
1724 optype = "reserved";
1730 err = "read ECC error";
1733 err = "RAS ECC error";
1736 err = "write parity error";
1739 err = "redundacy loss";
1745 err = "memory range error";
1748 err = "RTID out of range";
1751 err = "address parity error";
1754 err = "byte enable parity error";
1760 /* FIXME: should convert addr into bank and rank information */
1761 msg = kasprintf(GFP_ATOMIC,
1762 "%s (addr = 0x%08llx, cpu=%d, Dimm=%d, Channel=%d, "
1763 "syndrome=0x%08x, count=%d, Err=%08llx:%08llx (%s: %s))\n",
1764 type, (long long) m->addr, m->cpu, dimm, channel,
1765 syndrome, core_err_cnt, (long long)m->status,
1766 (long long)m->misc, optype, err);
1770 csrow = pvt->csrow_map[channel][dimm];
1772 /* Call the helper to output message */
1773 if (m->mcgstatus & 1)
1774 edac_mc_handle_fbd_ue(mci, csrow, 0,
1775 0 /* FIXME: should be channel here */, msg);
1776 else if (!pvt->is_registered)
1777 edac_mc_handle_fbd_ce(mci, csrow,
1778 0 /* FIXME: should be channel here */, msg);
1784 * i7core_check_error Retrieve and process errors reported by the
1785 * hardware. Called by the Core module.
1787 static void i7core_check_error(struct mem_ctl_info *mci)
1789 struct i7core_pvt *pvt = mci->pvt_info;
1795 * MCE first step: Copy all mce errors into a temporary buffer
1796 * We use a double buffering here, to reduce the risk of
1800 count = (pvt->mce_out + MCE_LOG_LEN - pvt->mce_in)
1803 goto check_ce_error;
1805 m = pvt->mce_outentry;
1806 if (pvt->mce_in + count > MCE_LOG_LEN) {
1807 unsigned l = MCE_LOG_LEN - pvt->mce_in;
1809 memcpy(m, &pvt->mce_entry[pvt->mce_in], sizeof(*m) * l);
1815 memcpy(m, &pvt->mce_entry[pvt->mce_in], sizeof(*m) * count);
1817 pvt->mce_in += count;
1820 if (pvt->mce_overrun) {
1821 i7core_printk(KERN_ERR, "Lost %d memory errors\n",
1824 pvt->mce_overrun = 0;
1828 * MCE second step: parse errors and display
1830 for (i = 0; i < count; i++)
1831 i7core_mce_output_error(mci, &pvt->mce_outentry[i]);
1834 * Now, let's increment CE error counts
1837 if (!pvt->is_registered)
1838 i7core_udimm_check_mc_ecc_err(mci);
1840 i7core_rdimm_check_mc_ecc_err(mci);
1844 * i7core_mce_check_error Replicates mcelog routine to get errors
1845 * This routine simply queues mcelog errors, and
1846 * return. The error itself should be handled later
1847 * by i7core_check_error.
1848 * WARNING: As this routine should be called at NMI time, extra care should
1849 * be taken to avoid deadlocks, and to be as fast as possible.
1851 static int i7core_mce_check_error(void *priv, struct mce *mce)
1853 struct mem_ctl_info *mci = priv;
1854 struct i7core_pvt *pvt = mci->pvt_info;
1857 * Just let mcelog handle it if the error is
1858 * outside the memory controller
1860 if (((mce->status & 0xffff) >> 7) != 1)
1863 /* Bank 8 registers are the only ones that we know how to handle */
1868 /* Only handle if it is the right mc controller */
1869 if (cpu_data(mce->cpu).phys_proc_id != pvt->i7core_dev->socket)
1874 if ((pvt->mce_out + 1) % MCE_LOG_LEN == pvt->mce_in) {
1880 /* Copy memory error at the ringbuffer */
1881 memcpy(&pvt->mce_entry[pvt->mce_out], mce, sizeof(*mce));
1883 pvt->mce_out = (pvt->mce_out + 1) % MCE_LOG_LEN;
1885 /* Handle fatal errors immediately */
1886 if (mce->mcgstatus & 1)
1887 i7core_check_error(mci);
1889 /* Advice mcelog that the error were handled */
1893 static int i7core_register_mci(struct i7core_dev *i7core_dev,
1894 const int num_channels, const int num_csrows)
1896 struct mem_ctl_info *mci;
1897 struct i7core_pvt *pvt;
1901 /* allocate a new MC control structure */
1902 mci = edac_mc_alloc(sizeof(*pvt), num_csrows, num_channels,
1903 i7core_dev->socket);
1907 debugf0("MC: " __FILE__ ": %s(): mci = %p, dev = %p\n",
1908 __func__, mci, &i7core_dev->pdev[0]->dev);
1910 /* record ptr to the generic device */
1911 mci->dev = &i7core_dev->pdev[0]->dev;
1913 pvt = mci->pvt_info;
1914 memset(pvt, 0, sizeof(*pvt));
1917 * FIXME: how to handle RDDR3 at MCI level? It is possible to have
1918 * Mixed RDDR3/UDDR3 with Nehalem, provided that they are on different
1921 mci->mtype_cap = MEM_FLAG_DDR3;
1922 mci->edac_ctl_cap = EDAC_FLAG_NONE;
1923 mci->edac_cap = EDAC_FLAG_NONE;
1924 mci->mod_name = "i7core_edac.c";
1925 mci->mod_ver = I7CORE_REVISION;
1926 mci->ctl_name = kasprintf(GFP_KERNEL, "i7 core #%d",
1927 i7core_dev->socket);
1928 mci->dev_name = pci_name(i7core_dev->pdev[0]);
1929 mci->ctl_page_to_phys = NULL;
1931 if (pvt->is_registered)
1932 mci->mc_driver_sysfs_attributes = i7core_sysfs_rdimm_attrs;
1934 mci->mc_driver_sysfs_attributes = i7core_sysfs_udimm_attrs;
1936 /* Set the function pointer to an actual operation function */
1937 mci->edac_check = i7core_check_error;
1939 /* Store pci devices at mci for faster access */
1940 rc = mci_bind_devs(mci, i7core_dev);
1941 if (unlikely(rc < 0))
1944 /* Get dimm basic config */
1945 get_dimm_config(mci, &csrow);
1947 /* add this new MC control structure to EDAC's list of MCs */
1948 if (unlikely(edac_mc_add_mc(mci))) {
1949 debugf0("MC: " __FILE__
1950 ": %s(): failed edac_mc_add_mc()\n", __func__);
1951 /* FIXME: perhaps some code should go here that disables error
1952 * reporting if we just enabled it
1959 /* Default error mask is any memory */
1960 pvt->inject.channel = 0;
1961 pvt->inject.dimm = -1;
1962 pvt->inject.rank = -1;
1963 pvt->inject.bank = -1;
1964 pvt->inject.page = -1;
1965 pvt->inject.col = -1;
1967 /* Registers on edac_mce in order to receive memory errors */
1968 pvt->edac_mce.priv = mci;
1969 pvt->edac_mce.check_error = i7core_mce_check_error;
1971 /* allocating generic PCI control info */
1972 pvt->i7core_pci = edac_pci_create_generic_ctl(&i7core_dev->pdev[0]->dev,
1974 if (unlikely(!pvt->i7core_pci)) {
1976 "%s(): Unable to create PCI control\n",
1979 "%s(): PCI error report via EDAC not setup\n",
1983 rc = edac_mce_register(&pvt->edac_mce);
1984 if (unlikely(rc < 0)) {
1985 debugf0("MC: " __FILE__
1986 ": %s(): failed edac_mce_register()\n", __func__);
1996 * i7core_probe Probe for ONE instance of device to see if it is
1999 * 0 for FOUND a device
2000 * < 0 for error code
2003 static int __devinit i7core_probe(struct pci_dev *pdev,
2004 const struct pci_device_id *id)
2007 struct i7core_dev *i7core_dev;
2009 /* get the pci devices we want to reserve for our use */
2010 mutex_lock(&i7core_edac_lock);
2013 * All memory controllers are allocated at the first pass.
2015 if (unlikely(probed >= 1)) {
2016 mutex_unlock(&i7core_edac_lock);
2021 rc = i7core_get_devices(pci_dev_table);
2022 if (unlikely(rc < 0))
2025 list_for_each_entry(i7core_dev, &i7core_edac_list, list) {
2029 /* Check the number of active and not disabled channels */
2030 rc = i7core_get_active_channels(i7core_dev->socket,
2031 &channels, &csrows);
2032 if (unlikely(rc < 0))
2035 rc = i7core_register_mci(i7core_dev, channels, csrows);
2036 if (unlikely(rc < 0))
2040 i7core_printk(KERN_INFO, "Driver loaded.\n");
2042 mutex_unlock(&i7core_edac_lock);
2046 i7core_put_all_devices();
2048 mutex_unlock(&i7core_edac_lock);
2053 * i7core_remove destructor for one instance of device
2056 static void __devexit i7core_remove(struct pci_dev *pdev)
2058 struct mem_ctl_info *mci;
2059 struct i7core_dev *i7core_dev, *tmp;
2060 struct i7core_pvt *pvt;
2062 debugf0(__FILE__ ": %s()\n", __func__);
2065 * we have a trouble here: pdev value for removal will be wrong, since
2066 * it will point to the X58 register used to detect that the machine
2067 * is a Nehalem or upper design. However, due to the way several PCI
2068 * devices are grouped together to provide MC functionality, we need
2069 * to use a different method for releasing the devices
2072 mutex_lock(&i7core_edac_lock);
2073 list_for_each_entry_safe(i7core_dev, tmp, &i7core_edac_list, list) {
2074 mci = find_mci_by_dev(&i7core_dev->pdev[0]->dev);
2075 if (unlikely(!mci || !mci->pvt_info)) {
2076 debugf0("MC: " __FILE__ ": %s(): dev = %p\n",
2077 __func__, &i7core_dev->pdev[0]->dev);
2079 i7core_printk(KERN_ERR,
2080 "Couldn't find mci hanler\n");
2082 pvt = mci->pvt_info;
2083 i7core_dev = pvt->i7core_dev;
2085 debugf0("MC: " __FILE__ ": %s(): mci = %p, dev = %p\n",
2086 __func__, mci, &i7core_dev->pdev[0]->dev);
2088 /* Disable MCE NMI handler */
2089 edac_mce_unregister(&pvt->edac_mce);
2091 /* Disable EDAC polling */
2092 if (likely(pvt->i7core_pci))
2093 edac_pci_release_generic_ctl(pvt->i7core_pci);
2095 i7core_printk(KERN_ERR,
2096 "Couldn't find mem_ctl_info for socket %d\n",
2097 i7core_dev->socket);
2098 pvt->i7core_pci = NULL;
2100 /* Remove MC sysfs nodes */
2101 edac_mc_del_mc(&i7core_dev->pdev[0]->dev);
2103 debugf1("%s: free mci struct\n", mci->ctl_name);
2104 kfree(mci->ctl_name);
2107 /* Release PCI resources */
2108 i7core_put_devices(i7core_dev);
2109 list_del(&i7core_dev->list);
2115 mutex_unlock(&i7core_edac_lock);
2118 MODULE_DEVICE_TABLE(pci, i7core_pci_tbl);
2121 * i7core_driver pci_driver structure for this module
2124 static struct pci_driver i7core_driver = {
2125 .name = "i7core_edac",
2126 .probe = i7core_probe,
2127 .remove = __devexit_p(i7core_remove),
2128 .id_table = i7core_pci_tbl,
2132 * i7core_init Module entry function
2133 * Try to initialize this module for its devices
2135 static int __init i7core_init(void)
2139 debugf2("MC: " __FILE__ ": %s()\n", __func__);
2141 /* Ensure that the OPSTATE is set correctly for POLL or NMI */
2145 i7core_xeon_pci_fixup(pci_dev_table);
2147 pci_rc = pci_register_driver(&i7core_driver);
2152 i7core_printk(KERN_ERR, "Failed to register device with error %d.\n",
2159 * i7core_exit() Module exit function
2160 * Unregister the driver
2162 static void __exit i7core_exit(void)
2164 debugf2("MC: " __FILE__ ": %s()\n", __func__);
2165 pci_unregister_driver(&i7core_driver);
2168 module_init(i7core_init);
2169 module_exit(i7core_exit);
2171 MODULE_LICENSE("GPL");
2172 MODULE_AUTHOR("Mauro Carvalho Chehab <mchehab@redhat.com>");
2173 MODULE_AUTHOR("Red Hat Inc. (http://www.redhat.com)");
2174 MODULE_DESCRIPTION("MC Driver for Intel i7 Core memory controllers - "
2177 module_param(edac_op_state, int, 0444);
2178 MODULE_PARM_DESC(edac_op_state, "EDAC Error Reporting state: 0=Poll,1=NMI");