usb: host: Distinguish Kconfig text for Freescale controllers
[zen-stable.git] / drivers / edac / sb_edac.c
blob1dc118d83cc6a25ddb0be845677574e393d6492b
1 /* Intel Sandy Bridge -EN/-EP/-EX Memory Controller kernel module
3 * This driver supports the memory controllers found on the Intel
4 * processor family Sandy Bridge.
6 * This file may be distributed under the terms of the
7 * GNU General Public License version 2 only.
9 * Copyright (c) 2011 by:
10 * Mauro Carvalho Chehab <mchehab@redhat.com>
13 #include <linux/module.h>
14 #include <linux/init.h>
15 #include <linux/pci.h>
16 #include <linux/pci_ids.h>
17 #include <linux/slab.h>
18 #include <linux/delay.h>
19 #include <linux/edac.h>
20 #include <linux/mmzone.h>
21 #include <linux/smp.h>
22 #include <linux/bitmap.h>
23 #include <asm/processor.h>
24 #include <asm/mce.h>
26 #include "edac_core.h"
28 /* Static vars */
29 static LIST_HEAD(sbridge_edac_list);
30 static DEFINE_MUTEX(sbridge_edac_lock);
31 static int probed;
34 * Alter this version for the module when modifications are made
36 #define SBRIDGE_REVISION " Ver: 1.0.0 "
37 #define EDAC_MOD_STR "sbridge_edac"
40 * Debug macros
42 #define sbridge_printk(level, fmt, arg...) \
43 edac_printk(level, "sbridge", fmt, ##arg)
45 #define sbridge_mc_printk(mci, level, fmt, arg...) \
46 edac_mc_chipset_printk(mci, level, "sbridge", fmt, ##arg)
49 * Get a bit field at register value <v>, from bit <lo> to bit <hi>
51 #define GET_BITFIELD(v, lo, hi) \
52 (((v) & ((1ULL << ((hi) - (lo) + 1)) - 1) << (lo)) >> (lo))
55 * sbridge Memory Controller Registers
59 * FIXME: For now, let's order by device function, as it makes
60 * easier for driver's development proccess. This table should be
61 * moved to pci_id.h when submitted upstream
63 #define PCI_DEVICE_ID_INTEL_SBRIDGE_SAD0 0x3cf4 /* 12.6 */
64 #define PCI_DEVICE_ID_INTEL_SBRIDGE_SAD1 0x3cf6 /* 12.7 */
65 #define PCI_DEVICE_ID_INTEL_SBRIDGE_BR 0x3cf5 /* 13.6 */
66 #define PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_HA0 0x3ca0 /* 14.0 */
67 #define PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TA 0x3ca8 /* 15.0 */
68 #define PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_RAS 0x3c71 /* 15.1 */
69 #define PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD0 0x3caa /* 15.2 */
70 #define PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD1 0x3cab /* 15.3 */
71 #define PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD2 0x3cac /* 15.4 */
72 #define PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD3 0x3cad /* 15.5 */
73 #define PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_DDRIO 0x3cb8 /* 17.0 */
76 * Currently, unused, but will be needed in the future
77 * implementations, as they hold the error counters
79 #define PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_ERR0 0x3c72 /* 16.2 */
80 #define PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_ERR1 0x3c73 /* 16.3 */
81 #define PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_ERR2 0x3c76 /* 16.6 */
82 #define PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_ERR3 0x3c77 /* 16.7 */
84 /* Devices 12 Function 6, Offsets 0x80 to 0xcc */
85 static const u32 dram_rule[] = {
86 0x80, 0x88, 0x90, 0x98, 0xa0,
87 0xa8, 0xb0, 0xb8, 0xc0, 0xc8,
89 #define MAX_SAD ARRAY_SIZE(dram_rule)
91 #define SAD_LIMIT(reg) ((GET_BITFIELD(reg, 6, 25) << 26) | 0x3ffffff)
92 #define DRAM_ATTR(reg) GET_BITFIELD(reg, 2, 3)
93 #define INTERLEAVE_MODE(reg) GET_BITFIELD(reg, 1, 1)
94 #define DRAM_RULE_ENABLE(reg) GET_BITFIELD(reg, 0, 0)
96 static char *get_dram_attr(u32 reg)
98 switch(DRAM_ATTR(reg)) {
99 case 0:
100 return "DRAM";
101 case 1:
102 return "MMCFG";
103 case 2:
104 return "NXM";
105 default:
106 return "unknown";
110 static const u32 interleave_list[] = {
111 0x84, 0x8c, 0x94, 0x9c, 0xa4,
112 0xac, 0xb4, 0xbc, 0xc4, 0xcc,
114 #define MAX_INTERLEAVE ARRAY_SIZE(interleave_list)
116 #define SAD_PKG0(reg) GET_BITFIELD(reg, 0, 2)
117 #define SAD_PKG1(reg) GET_BITFIELD(reg, 3, 5)
118 #define SAD_PKG2(reg) GET_BITFIELD(reg, 8, 10)
119 #define SAD_PKG3(reg) GET_BITFIELD(reg, 11, 13)
120 #define SAD_PKG4(reg) GET_BITFIELD(reg, 16, 18)
121 #define SAD_PKG5(reg) GET_BITFIELD(reg, 19, 21)
122 #define SAD_PKG6(reg) GET_BITFIELD(reg, 24, 26)
123 #define SAD_PKG7(reg) GET_BITFIELD(reg, 27, 29)
125 static inline int sad_pkg(u32 reg, int interleave)
127 switch (interleave) {
128 case 0:
129 return SAD_PKG0(reg);
130 case 1:
131 return SAD_PKG1(reg);
132 case 2:
133 return SAD_PKG2(reg);
134 case 3:
135 return SAD_PKG3(reg);
136 case 4:
137 return SAD_PKG4(reg);
138 case 5:
139 return SAD_PKG5(reg);
140 case 6:
141 return SAD_PKG6(reg);
142 case 7:
143 return SAD_PKG7(reg);
144 default:
145 return -EINVAL;
149 /* Devices 12 Function 7 */
151 #define TOLM 0x80
152 #define TOHM 0x84
154 #define GET_TOLM(reg) ((GET_BITFIELD(reg, 0, 3) << 28) | 0x3ffffff)
155 #define GET_TOHM(reg) ((GET_BITFIELD(reg, 0, 20) << 25) | 0x3ffffff)
157 /* Device 13 Function 6 */
159 #define SAD_TARGET 0xf0
161 #define SOURCE_ID(reg) GET_BITFIELD(reg, 9, 11)
163 #define SAD_CONTROL 0xf4
165 #define NODE_ID(reg) GET_BITFIELD(reg, 0, 2)
167 /* Device 14 function 0 */
169 static const u32 tad_dram_rule[] = {
170 0x40, 0x44, 0x48, 0x4c,
171 0x50, 0x54, 0x58, 0x5c,
172 0x60, 0x64, 0x68, 0x6c,
174 #define MAX_TAD ARRAY_SIZE(tad_dram_rule)
176 #define TAD_LIMIT(reg) ((GET_BITFIELD(reg, 12, 31) << 26) | 0x3ffffff)
177 #define TAD_SOCK(reg) GET_BITFIELD(reg, 10, 11)
178 #define TAD_CH(reg) GET_BITFIELD(reg, 8, 9)
179 #define TAD_TGT3(reg) GET_BITFIELD(reg, 6, 7)
180 #define TAD_TGT2(reg) GET_BITFIELD(reg, 4, 5)
181 #define TAD_TGT1(reg) GET_BITFIELD(reg, 2, 3)
182 #define TAD_TGT0(reg) GET_BITFIELD(reg, 0, 1)
184 /* Device 15, function 0 */
186 #define MCMTR 0x7c
188 #define IS_ECC_ENABLED(mcmtr) GET_BITFIELD(mcmtr, 2, 2)
189 #define IS_LOCKSTEP_ENABLED(mcmtr) GET_BITFIELD(mcmtr, 1, 1)
190 #define IS_CLOSE_PG(mcmtr) GET_BITFIELD(mcmtr, 0, 0)
192 /* Device 15, function 1 */
194 #define RASENABLES 0xac
195 #define IS_MIRROR_ENABLED(reg) GET_BITFIELD(reg, 0, 0)
197 /* Device 15, functions 2-5 */
199 static const int mtr_regs[] = {
200 0x80, 0x84, 0x88,
203 #define RANK_DISABLE(mtr) GET_BITFIELD(mtr, 16, 19)
204 #define IS_DIMM_PRESENT(mtr) GET_BITFIELD(mtr, 14, 14)
205 #define RANK_CNT_BITS(mtr) GET_BITFIELD(mtr, 12, 13)
206 #define RANK_WIDTH_BITS(mtr) GET_BITFIELD(mtr, 2, 4)
207 #define COL_WIDTH_BITS(mtr) GET_BITFIELD(mtr, 0, 1)
209 static const u32 tad_ch_nilv_offset[] = {
210 0x90, 0x94, 0x98, 0x9c,
211 0xa0, 0xa4, 0xa8, 0xac,
212 0xb0, 0xb4, 0xb8, 0xbc,
214 #define CHN_IDX_OFFSET(reg) GET_BITFIELD(reg, 28, 29)
215 #define TAD_OFFSET(reg) (GET_BITFIELD(reg, 6, 25) << 26)
217 static const u32 rir_way_limit[] = {
218 0x108, 0x10c, 0x110, 0x114, 0x118,
220 #define MAX_RIR_RANGES ARRAY_SIZE(rir_way_limit)
222 #define IS_RIR_VALID(reg) GET_BITFIELD(reg, 31, 31)
223 #define RIR_WAY(reg) GET_BITFIELD(reg, 28, 29)
224 #define RIR_LIMIT(reg) ((GET_BITFIELD(reg, 1, 10) << 29)| 0x1fffffff)
226 #define MAX_RIR_WAY 8
228 static const u32 rir_offset[MAX_RIR_RANGES][MAX_RIR_WAY] = {
229 { 0x120, 0x124, 0x128, 0x12c, 0x130, 0x134, 0x138, 0x13c },
230 { 0x140, 0x144, 0x148, 0x14c, 0x150, 0x154, 0x158, 0x15c },
231 { 0x160, 0x164, 0x168, 0x16c, 0x170, 0x174, 0x178, 0x17c },
232 { 0x180, 0x184, 0x188, 0x18c, 0x190, 0x194, 0x198, 0x19c },
233 { 0x1a0, 0x1a4, 0x1a8, 0x1ac, 0x1b0, 0x1b4, 0x1b8, 0x1bc },
236 #define RIR_RNK_TGT(reg) GET_BITFIELD(reg, 16, 19)
237 #define RIR_OFFSET(reg) GET_BITFIELD(reg, 2, 14)
239 /* Device 16, functions 2-7 */
242 * FIXME: Implement the error count reads directly
245 static const u32 correrrcnt[] = {
246 0x104, 0x108, 0x10c, 0x110,
249 #define RANK_ODD_OV(reg) GET_BITFIELD(reg, 31, 31)
250 #define RANK_ODD_ERR_CNT(reg) GET_BITFIELD(reg, 16, 30)
251 #define RANK_EVEN_OV(reg) GET_BITFIELD(reg, 15, 15)
252 #define RANK_EVEN_ERR_CNT(reg) GET_BITFIELD(reg, 0, 14)
254 static const u32 correrrthrsld[] = {
255 0x11c, 0x120, 0x124, 0x128,
258 #define RANK_ODD_ERR_THRSLD(reg) GET_BITFIELD(reg, 16, 30)
259 #define RANK_EVEN_ERR_THRSLD(reg) GET_BITFIELD(reg, 0, 14)
262 /* Device 17, function 0 */
264 #define RANK_CFG_A 0x0328
266 #define IS_RDIMM_ENABLED(reg) GET_BITFIELD(reg, 11, 11)
269 * sbridge structs
272 #define NUM_CHANNELS 4
273 #define MAX_DIMMS 3 /* Max DIMMS per channel */
275 struct sbridge_info {
276 u32 mcmtr;
279 struct sbridge_channel {
280 u32 ranks;
281 u32 dimms;
284 struct pci_id_descr {
285 int dev;
286 int func;
287 int dev_id;
288 int optional;
291 struct pci_id_table {
292 const struct pci_id_descr *descr;
293 int n_devs;
296 struct sbridge_dev {
297 struct list_head list;
298 u8 bus, mc;
299 u8 node_id, source_id;
300 struct pci_dev **pdev;
301 int n_devs;
302 struct mem_ctl_info *mci;
305 struct sbridge_pvt {
306 struct pci_dev *pci_ta, *pci_ddrio, *pci_ras;
307 struct pci_dev *pci_sad0, *pci_sad1, *pci_ha0;
308 struct pci_dev *pci_br;
309 struct pci_dev *pci_tad[NUM_CHANNELS];
311 struct sbridge_dev *sbridge_dev;
313 struct sbridge_info info;
314 struct sbridge_channel channel[NUM_CHANNELS];
316 int csrow_map[NUM_CHANNELS][MAX_DIMMS];
318 /* Memory type detection */
319 bool is_mirrored, is_lockstep, is_close_pg;
321 /* Fifo double buffers */
322 struct mce mce_entry[MCE_LOG_LEN];
323 struct mce mce_outentry[MCE_LOG_LEN];
325 /* Fifo in/out counters */
326 unsigned mce_in, mce_out;
328 /* Count indicator to show errors not got */
329 unsigned mce_overrun;
331 /* Memory description */
332 u64 tolm, tohm;
335 #define PCI_DESCR(device, function, device_id) \
336 .dev = (device), \
337 .func = (function), \
338 .dev_id = (device_id)
340 static const struct pci_id_descr pci_dev_descr_sbridge[] = {
341 /* Processor Home Agent */
342 { PCI_DESCR(14, 0, PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_HA0) },
344 /* Memory controller */
345 { PCI_DESCR(15, 0, PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TA) },
346 { PCI_DESCR(15, 1, PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_RAS) },
347 { PCI_DESCR(15, 2, PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD0) },
348 { PCI_DESCR(15, 3, PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD1) },
349 { PCI_DESCR(15, 4, PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD2) },
350 { PCI_DESCR(15, 5, PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD3) },
351 { PCI_DESCR(17, 0, PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_DDRIO) },
353 /* System Address Decoder */
354 { PCI_DESCR(12, 6, PCI_DEVICE_ID_INTEL_SBRIDGE_SAD0) },
355 { PCI_DESCR(12, 7, PCI_DEVICE_ID_INTEL_SBRIDGE_SAD1) },
357 /* Broadcast Registers */
358 { PCI_DESCR(13, 6, PCI_DEVICE_ID_INTEL_SBRIDGE_BR) },
361 #define PCI_ID_TABLE_ENTRY(A) { .descr=A, .n_devs = ARRAY_SIZE(A) }
362 static const struct pci_id_table pci_dev_descr_sbridge_table[] = {
363 PCI_ID_TABLE_ENTRY(pci_dev_descr_sbridge),
364 {0,} /* 0 terminated list. */
368 * pci_device_id table for which devices we are looking for
370 static const struct pci_device_id sbridge_pci_tbl[] __devinitdata = {
371 {PCI_DEVICE(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TA)},
372 {0,} /* 0 terminated list. */
376 /****************************************************************************
377 Anciliary status routines
378 ****************************************************************************/
380 static inline int numrank(u32 mtr)
382 int ranks = (1 << RANK_CNT_BITS(mtr));
384 if (ranks > 4) {
385 debugf0("Invalid number of ranks: %d (max = 4) raw value = %x (%04x)",
386 ranks, (unsigned int)RANK_CNT_BITS(mtr), mtr);
387 return -EINVAL;
390 return ranks;
393 static inline int numrow(u32 mtr)
395 int rows = (RANK_WIDTH_BITS(mtr) + 12);
397 if (rows < 13 || rows > 18) {
398 debugf0("Invalid number of rows: %d (should be between 14 and 17) raw value = %x (%04x)",
399 rows, (unsigned int)RANK_WIDTH_BITS(mtr), mtr);
400 return -EINVAL;
403 return 1 << rows;
406 static inline int numcol(u32 mtr)
408 int cols = (COL_WIDTH_BITS(mtr) + 10);
410 if (cols > 12) {
411 debugf0("Invalid number of cols: %d (max = 4) raw value = %x (%04x)",
412 cols, (unsigned int)COL_WIDTH_BITS(mtr), mtr);
413 return -EINVAL;
416 return 1 << cols;
419 static struct sbridge_dev *get_sbridge_dev(u8 bus)
421 struct sbridge_dev *sbridge_dev;
423 list_for_each_entry(sbridge_dev, &sbridge_edac_list, list) {
424 if (sbridge_dev->bus == bus)
425 return sbridge_dev;
428 return NULL;
431 static struct sbridge_dev *alloc_sbridge_dev(u8 bus,
432 const struct pci_id_table *table)
434 struct sbridge_dev *sbridge_dev;
436 sbridge_dev = kzalloc(sizeof(*sbridge_dev), GFP_KERNEL);
437 if (!sbridge_dev)
438 return NULL;
440 sbridge_dev->pdev = kzalloc(sizeof(*sbridge_dev->pdev) * table->n_devs,
441 GFP_KERNEL);
442 if (!sbridge_dev->pdev) {
443 kfree(sbridge_dev);
444 return NULL;
447 sbridge_dev->bus = bus;
448 sbridge_dev->n_devs = table->n_devs;
449 list_add_tail(&sbridge_dev->list, &sbridge_edac_list);
451 return sbridge_dev;
454 static void free_sbridge_dev(struct sbridge_dev *sbridge_dev)
456 list_del(&sbridge_dev->list);
457 kfree(sbridge_dev->pdev);
458 kfree(sbridge_dev);
461 /****************************************************************************
462 Memory check routines
463 ****************************************************************************/
464 static struct pci_dev *get_pdev_slot_func(u8 bus, unsigned slot,
465 unsigned func)
467 struct sbridge_dev *sbridge_dev = get_sbridge_dev(bus);
468 int i;
470 if (!sbridge_dev)
471 return NULL;
473 for (i = 0; i < sbridge_dev->n_devs; i++) {
474 if (!sbridge_dev->pdev[i])
475 continue;
477 if (PCI_SLOT(sbridge_dev->pdev[i]->devfn) == slot &&
478 PCI_FUNC(sbridge_dev->pdev[i]->devfn) == func) {
479 debugf1("Associated %02x.%02x.%d with %p\n",
480 bus, slot, func, sbridge_dev->pdev[i]);
481 return sbridge_dev->pdev[i];
485 return NULL;
489 * sbridge_get_active_channels() - gets the number of channels and csrows
490 * bus: Device bus
491 * @channels: Number of channels that will be returned
492 * @csrows: Number of csrows found
494 * Since EDAC core needs to know in advance the number of available channels
495 * and csrows, in order to allocate memory for csrows/channels, it is needed
496 * to run two similar steps. At the first step, implemented on this function,
497 * it checks the number of csrows/channels present at one socket, identified
498 * by the associated PCI bus.
499 * this is used in order to properly allocate the size of mci components.
500 * Note: one csrow is one dimm.
502 static int sbridge_get_active_channels(const u8 bus, unsigned *channels,
503 unsigned *csrows)
505 struct pci_dev *pdev = NULL;
506 int i, j;
507 u32 mcmtr;
509 *channels = 0;
510 *csrows = 0;
512 pdev = get_pdev_slot_func(bus, 15, 0);
513 if (!pdev) {
514 sbridge_printk(KERN_ERR, "Couldn't find PCI device "
515 "%2x.%02d.%d!!!\n",
516 bus, 15, 0);
517 return -ENODEV;
520 pci_read_config_dword(pdev, MCMTR, &mcmtr);
521 if (!IS_ECC_ENABLED(mcmtr)) {
522 sbridge_printk(KERN_ERR, "ECC is disabled. Aborting\n");
523 return -ENODEV;
526 for (i = 0; i < NUM_CHANNELS; i++) {
527 u32 mtr;
529 /* Device 15 functions 2 - 5 */
530 pdev = get_pdev_slot_func(bus, 15, 2 + i);
531 if (!pdev) {
532 sbridge_printk(KERN_ERR, "Couldn't find PCI device "
533 "%2x.%02d.%d!!!\n",
534 bus, 15, 2 + i);
535 return -ENODEV;
537 (*channels)++;
539 for (j = 0; j < ARRAY_SIZE(mtr_regs); j++) {
540 pci_read_config_dword(pdev, mtr_regs[j], &mtr);
541 debugf1("Bus#%02x channel #%d MTR%d = %x\n", bus, i, j, mtr);
542 if (IS_DIMM_PRESENT(mtr))
543 (*csrows)++;
547 debugf0("Number of active channels: %d, number of active dimms: %d\n",
548 *channels, *csrows);
550 return 0;
553 static int get_dimm_config(const struct mem_ctl_info *mci)
555 struct sbridge_pvt *pvt = mci->pvt_info;
556 struct csrow_info *csr;
557 int i, j, banks, ranks, rows, cols, size, npages;
558 int csrow = 0;
559 unsigned long last_page = 0;
560 u32 reg;
561 enum edac_type mode;
562 enum mem_type mtype;
564 pci_read_config_dword(pvt->pci_br, SAD_TARGET, &reg);
565 pvt->sbridge_dev->source_id = SOURCE_ID(reg);
567 pci_read_config_dword(pvt->pci_br, SAD_CONTROL, &reg);
568 pvt->sbridge_dev->node_id = NODE_ID(reg);
569 debugf0("mc#%d: Node ID: %d, source ID: %d\n",
570 pvt->sbridge_dev->mc,
571 pvt->sbridge_dev->node_id,
572 pvt->sbridge_dev->source_id);
574 pci_read_config_dword(pvt->pci_ras, RASENABLES, &reg);
575 if (IS_MIRROR_ENABLED(reg)) {
576 debugf0("Memory mirror is enabled\n");
577 pvt->is_mirrored = true;
578 } else {
579 debugf0("Memory mirror is disabled\n");
580 pvt->is_mirrored = false;
583 pci_read_config_dword(pvt->pci_ta, MCMTR, &pvt->info.mcmtr);
584 if (IS_LOCKSTEP_ENABLED(pvt->info.mcmtr)) {
585 debugf0("Lockstep is enabled\n");
586 mode = EDAC_S8ECD8ED;
587 pvt->is_lockstep = true;
588 } else {
589 debugf0("Lockstep is disabled\n");
590 mode = EDAC_S4ECD4ED;
591 pvt->is_lockstep = false;
593 if (IS_CLOSE_PG(pvt->info.mcmtr)) {
594 debugf0("address map is on closed page mode\n");
595 pvt->is_close_pg = true;
596 } else {
597 debugf0("address map is on open page mode\n");
598 pvt->is_close_pg = false;
601 pci_read_config_dword(pvt->pci_ta, RANK_CFG_A, &reg);
602 if (IS_RDIMM_ENABLED(reg)) {
603 /* FIXME: Can also be LRDIMM */
604 debugf0("Memory is registered\n");
605 mtype = MEM_RDDR3;
606 } else {
607 debugf0("Memory is unregistered\n");
608 mtype = MEM_DDR3;
611 /* On all supported DDR3 DIMM types, there are 8 banks available */
612 banks = 8;
614 for (i = 0; i < NUM_CHANNELS; i++) {
615 u32 mtr;
617 for (j = 0; j < ARRAY_SIZE(mtr_regs); j++) {
618 pci_read_config_dword(pvt->pci_tad[i],
619 mtr_regs[j], &mtr);
620 debugf4("Channel #%d MTR%d = %x\n", i, j, mtr);
621 if (IS_DIMM_PRESENT(mtr)) {
622 pvt->channel[i].dimms++;
624 ranks = numrank(mtr);
625 rows = numrow(mtr);
626 cols = numcol(mtr);
628 /* DDR3 has 8 I/O banks */
629 size = (rows * cols * banks * ranks) >> (20 - 3);
630 npages = MiB_TO_PAGES(size);
632 debugf0("mc#%d: channel %d, dimm %d, %d Mb (%d pages) bank: %d, rank: %d, row: %#x, col: %#x\n",
633 pvt->sbridge_dev->mc, i, j,
634 size, npages,
635 banks, ranks, rows, cols);
636 csr = &mci->csrows[csrow];
638 csr->first_page = last_page;
639 csr->last_page = last_page + npages - 1;
640 csr->page_mask = 0UL; /* Unused */
641 csr->nr_pages = npages;
642 csr->grain = 32;
643 csr->csrow_idx = csrow;
644 csr->dtype = (banks == 8) ? DEV_X8 : DEV_X4;
645 csr->ce_count = 0;
646 csr->ue_count = 0;
647 csr->mtype = mtype;
648 csr->edac_mode = mode;
649 csr->nr_channels = 1;
650 csr->channels[0].chan_idx = i;
651 csr->channels[0].ce_count = 0;
652 pvt->csrow_map[i][j] = csrow;
653 snprintf(csr->channels[0].label,
654 sizeof(csr->channels[0].label),
655 "CPU_SrcID#%u_Channel#%u_DIMM#%u",
656 pvt->sbridge_dev->source_id, i, j);
657 last_page += npages;
658 csrow++;
663 return 0;
666 static void get_memory_layout(const struct mem_ctl_info *mci)
668 struct sbridge_pvt *pvt = mci->pvt_info;
669 int i, j, k, n_sads, n_tads, sad_interl;
670 u32 reg;
671 u64 limit, prv = 0;
672 u64 tmp_mb;
673 u32 rir_way;
676 * Step 1) Get TOLM/TOHM ranges
679 /* Address range is 32:28 */
680 pci_read_config_dword(pvt->pci_sad1, TOLM,
681 &reg);
682 pvt->tolm = GET_TOLM(reg);
683 tmp_mb = (1 + pvt->tolm) >> 20;
685 debugf0("TOLM: %Lu.%03Lu GB (0x%016Lx)\n",
686 tmp_mb / 1000, tmp_mb % 1000, (u64)pvt->tolm);
688 /* Address range is already 45:25 */
689 pci_read_config_dword(pvt->pci_sad1, TOHM,
690 &reg);
691 pvt->tohm = GET_TOHM(reg);
692 tmp_mb = (1 + pvt->tohm) >> 20;
694 debugf0("TOHM: %Lu.%03Lu GB (0x%016Lx)",
695 tmp_mb / 1000, tmp_mb % 1000, (u64)pvt->tohm);
698 * Step 2) Get SAD range and SAD Interleave list
699 * TAD registers contain the interleave wayness. However, it
700 * seems simpler to just discover it indirectly, with the
701 * algorithm bellow.
703 prv = 0;
704 for (n_sads = 0; n_sads < MAX_SAD; n_sads++) {
705 /* SAD_LIMIT Address range is 45:26 */
706 pci_read_config_dword(pvt->pci_sad0, dram_rule[n_sads],
707 &reg);
708 limit = SAD_LIMIT(reg);
710 if (!DRAM_RULE_ENABLE(reg))
711 continue;
713 if (limit <= prv)
714 break;
716 tmp_mb = (limit + 1) >> 20;
717 debugf0("SAD#%d %s up to %Lu.%03Lu GB (0x%016Lx) %s reg=0x%08x\n",
718 n_sads,
719 get_dram_attr(reg),
720 tmp_mb / 1000, tmp_mb % 1000,
721 ((u64)tmp_mb) << 20L,
722 INTERLEAVE_MODE(reg) ? "Interleave: 8:6" : "Interleave: [8:6]XOR[18:16]",
723 reg);
724 prv = limit;
726 pci_read_config_dword(pvt->pci_sad0, interleave_list[n_sads],
727 &reg);
728 sad_interl = sad_pkg(reg, 0);
729 for (j = 0; j < 8; j++) {
730 if (j > 0 && sad_interl == sad_pkg(reg, j))
731 break;
733 debugf0("SAD#%d, interleave #%d: %d\n",
734 n_sads, j, sad_pkg(reg, j));
739 * Step 3) Get TAD range
741 prv = 0;
742 for (n_tads = 0; n_tads < MAX_TAD; n_tads++) {
743 pci_read_config_dword(pvt->pci_ha0, tad_dram_rule[n_tads],
744 &reg);
745 limit = TAD_LIMIT(reg);
746 if (limit <= prv)
747 break;
748 tmp_mb = (limit + 1) >> 20;
750 debugf0("TAD#%d: up to %Lu.%03Lu GB (0x%016Lx), socket interleave %d, memory interleave %d, TGT: %d, %d, %d, %d, reg=0x%08x\n",
751 n_tads, tmp_mb / 1000, tmp_mb % 1000,
752 ((u64)tmp_mb) << 20L,
753 (u32)TAD_SOCK(reg),
754 (u32)TAD_CH(reg),
755 (u32)TAD_TGT0(reg),
756 (u32)TAD_TGT1(reg),
757 (u32)TAD_TGT2(reg),
758 (u32)TAD_TGT3(reg),
759 reg);
760 prv = tmp_mb;
764 * Step 4) Get TAD offsets, per each channel
766 for (i = 0; i < NUM_CHANNELS; i++) {
767 if (!pvt->channel[i].dimms)
768 continue;
769 for (j = 0; j < n_tads; j++) {
770 pci_read_config_dword(pvt->pci_tad[i],
771 tad_ch_nilv_offset[j],
772 &reg);
773 tmp_mb = TAD_OFFSET(reg) >> 20;
774 debugf0("TAD CH#%d, offset #%d: %Lu.%03Lu GB (0x%016Lx), reg=0x%08x\n",
775 i, j,
776 tmp_mb / 1000, tmp_mb % 1000,
777 ((u64)tmp_mb) << 20L,
778 reg);
783 * Step 6) Get RIR Wayness/Limit, per each channel
785 for (i = 0; i < NUM_CHANNELS; i++) {
786 if (!pvt->channel[i].dimms)
787 continue;
788 for (j = 0; j < MAX_RIR_RANGES; j++) {
789 pci_read_config_dword(pvt->pci_tad[i],
790 rir_way_limit[j],
791 &reg);
793 if (!IS_RIR_VALID(reg))
794 continue;
796 tmp_mb = RIR_LIMIT(reg) >> 20;
797 rir_way = 1 << RIR_WAY(reg);
798 debugf0("CH#%d RIR#%d, limit: %Lu.%03Lu GB (0x%016Lx), way: %d, reg=0x%08x\n",
799 i, j,
800 tmp_mb / 1000, tmp_mb % 1000,
801 ((u64)tmp_mb) << 20L,
802 rir_way,
803 reg);
805 for (k = 0; k < rir_way; k++) {
806 pci_read_config_dword(pvt->pci_tad[i],
807 rir_offset[j][k],
808 &reg);
809 tmp_mb = RIR_OFFSET(reg) << 6;
811 debugf0("CH#%d RIR#%d INTL#%d, offset %Lu.%03Lu GB (0x%016Lx), tgt: %d, reg=0x%08x\n",
812 i, j, k,
813 tmp_mb / 1000, tmp_mb % 1000,
814 ((u64)tmp_mb) << 20L,
815 (u32)RIR_RNK_TGT(reg),
816 reg);
822 struct mem_ctl_info *get_mci_for_node_id(u8 node_id)
824 struct sbridge_dev *sbridge_dev;
826 list_for_each_entry(sbridge_dev, &sbridge_edac_list, list) {
827 if (sbridge_dev->node_id == node_id)
828 return sbridge_dev->mci;
830 return NULL;
833 static int get_memory_error_data(struct mem_ctl_info *mci,
834 u64 addr,
835 u8 *socket,
836 long *channel_mask,
837 u8 *rank,
838 char *area_type)
840 struct mem_ctl_info *new_mci;
841 struct sbridge_pvt *pvt = mci->pvt_info;
842 char msg[256];
843 int n_rir, n_sads, n_tads, sad_way, sck_xch;
844 int sad_interl, idx, base_ch;
845 int interleave_mode;
846 unsigned sad_interleave[MAX_INTERLEAVE];
847 u32 reg;
848 u8 ch_way,sck_way;
849 u32 tad_offset;
850 u32 rir_way;
851 u64 ch_addr, offset, limit, prv = 0;
855 * Step 0) Check if the address is at special memory ranges
856 * The check bellow is probably enough to fill all cases where
857 * the error is not inside a memory, except for the legacy
858 * range (e. g. VGA addresses). It is unlikely, however, that the
859 * memory controller would generate an error on that range.
861 if ((addr > (u64) pvt->tolm) && (addr < (1L << 32))) {
862 sprintf(msg, "Error at TOLM area, on addr 0x%08Lx", addr);
863 edac_mc_handle_ce_no_info(mci, msg);
864 return -EINVAL;
866 if (addr >= (u64)pvt->tohm) {
867 sprintf(msg, "Error at MMIOH area, on addr 0x%016Lx", addr);
868 edac_mc_handle_ce_no_info(mci, msg);
869 return -EINVAL;
873 * Step 1) Get socket
875 for (n_sads = 0; n_sads < MAX_SAD; n_sads++) {
876 pci_read_config_dword(pvt->pci_sad0, dram_rule[n_sads],
877 &reg);
879 if (!DRAM_RULE_ENABLE(reg))
880 continue;
882 limit = SAD_LIMIT(reg);
883 if (limit <= prv) {
884 sprintf(msg, "Can't discover the memory socket");
885 edac_mc_handle_ce_no_info(mci, msg);
886 return -EINVAL;
888 if (addr <= limit)
889 break;
890 prv = limit;
892 if (n_sads == MAX_SAD) {
893 sprintf(msg, "Can't discover the memory socket");
894 edac_mc_handle_ce_no_info(mci, msg);
895 return -EINVAL;
897 area_type = get_dram_attr(reg);
898 interleave_mode = INTERLEAVE_MODE(reg);
900 pci_read_config_dword(pvt->pci_sad0, interleave_list[n_sads],
901 &reg);
902 sad_interl = sad_pkg(reg, 0);
903 for (sad_way = 0; sad_way < 8; sad_way++) {
904 if (sad_way > 0 && sad_interl == sad_pkg(reg, sad_way))
905 break;
906 sad_interleave[sad_way] = sad_pkg(reg, sad_way);
907 debugf0("SAD interleave #%d: %d\n",
908 sad_way, sad_interleave[sad_way]);
910 debugf0("mc#%d: Error detected on SAD#%d: address 0x%016Lx < 0x%016Lx, Interleave [%d:6]%s\n",
911 pvt->sbridge_dev->mc,
912 n_sads,
913 addr,
914 limit,
915 sad_way + 7,
916 INTERLEAVE_MODE(reg) ? "" : "XOR[18:16]");
917 if (interleave_mode)
918 idx = ((addr >> 6) ^ (addr >> 16)) & 7;
919 else
920 idx = (addr >> 6) & 7;
921 switch (sad_way) {
922 case 1:
923 idx = 0;
924 break;
925 case 2:
926 idx = idx & 1;
927 break;
928 case 4:
929 idx = idx & 3;
930 break;
931 case 8:
932 break;
933 default:
934 sprintf(msg, "Can't discover socket interleave");
935 edac_mc_handle_ce_no_info(mci, msg);
936 return -EINVAL;
938 *socket = sad_interleave[idx];
939 debugf0("SAD interleave index: %d (wayness %d) = CPU socket %d\n",
940 idx, sad_way, *socket);
943 * Move to the proper node structure, in order to access the
944 * right PCI registers
946 new_mci = get_mci_for_node_id(*socket);
947 if (!new_mci) {
948 sprintf(msg, "Struct for socket #%u wasn't initialized",
949 *socket);
950 edac_mc_handle_ce_no_info(mci, msg);
951 return -EINVAL;
953 mci = new_mci;
954 pvt = mci->pvt_info;
957 * Step 2) Get memory channel
959 prv = 0;
960 for (n_tads = 0; n_tads < MAX_TAD; n_tads++) {
961 pci_read_config_dword(pvt->pci_ha0, tad_dram_rule[n_tads],
962 &reg);
963 limit = TAD_LIMIT(reg);
964 if (limit <= prv) {
965 sprintf(msg, "Can't discover the memory channel");
966 edac_mc_handle_ce_no_info(mci, msg);
967 return -EINVAL;
969 if (addr <= limit)
970 break;
971 prv = limit;
973 ch_way = TAD_CH(reg) + 1;
974 sck_way = TAD_SOCK(reg) + 1;
976 * FIXME: Is it right to always use channel 0 for offsets?
978 pci_read_config_dword(pvt->pci_tad[0],
979 tad_ch_nilv_offset[n_tads],
980 &tad_offset);
982 if (ch_way == 3)
983 idx = addr >> 6;
984 else
985 idx = addr >> (6 + sck_way);
986 idx = idx % ch_way;
989 * FIXME: Shouldn't we use CHN_IDX_OFFSET() here, when ch_way == 3 ???
991 switch (idx) {
992 case 0:
993 base_ch = TAD_TGT0(reg);
994 break;
995 case 1:
996 base_ch = TAD_TGT1(reg);
997 break;
998 case 2:
999 base_ch = TAD_TGT2(reg);
1000 break;
1001 case 3:
1002 base_ch = TAD_TGT3(reg);
1003 break;
1004 default:
1005 sprintf(msg, "Can't discover the TAD target");
1006 edac_mc_handle_ce_no_info(mci, msg);
1007 return -EINVAL;
1009 *channel_mask = 1 << base_ch;
1011 if (pvt->is_mirrored) {
1012 *channel_mask |= 1 << ((base_ch + 2) % 4);
1013 switch(ch_way) {
1014 case 2:
1015 case 4:
1016 sck_xch = 1 << sck_way * (ch_way >> 1);
1017 break;
1018 default:
1019 sprintf(msg, "Invalid mirror set. Can't decode addr");
1020 edac_mc_handle_ce_no_info(mci, msg);
1021 return -EINVAL;
1023 } else
1024 sck_xch = (1 << sck_way) * ch_way;
1026 if (pvt->is_lockstep)
1027 *channel_mask |= 1 << ((base_ch + 1) % 4);
1029 offset = TAD_OFFSET(tad_offset);
1031 debugf0("TAD#%d: address 0x%016Lx < 0x%016Lx, socket interleave %d, channel interleave %d (offset 0x%08Lx), index %d, base ch: %d, ch mask: 0x%02lx\n",
1032 n_tads,
1033 addr,
1034 limit,
1035 (u32)TAD_SOCK(reg),
1036 ch_way,
1037 offset,
1038 idx,
1039 base_ch,
1040 *channel_mask);
1042 /* Calculate channel address */
1043 /* Remove the TAD offset */
1045 if (offset > addr) {
1046 sprintf(msg, "Can't calculate ch addr: TAD offset 0x%08Lx is too high for addr 0x%08Lx!",
1047 offset, addr);
1048 edac_mc_handle_ce_no_info(mci, msg);
1049 return -EINVAL;
1051 addr -= offset;
1052 /* Store the low bits [0:6] of the addr */
1053 ch_addr = addr & 0x7f;
1054 /* Remove socket wayness and remove 6 bits */
1055 addr >>= 6;
1056 addr /= sck_xch;
1057 #if 0
1058 /* Divide by channel way */
1059 addr = addr / ch_way;
1060 #endif
1061 /* Recover the last 6 bits */
1062 ch_addr |= addr << 6;
1065 * Step 3) Decode rank
1067 for (n_rir = 0; n_rir < MAX_RIR_RANGES; n_rir++) {
1068 pci_read_config_dword(pvt->pci_tad[base_ch],
1069 rir_way_limit[n_rir],
1070 &reg);
1072 if (!IS_RIR_VALID(reg))
1073 continue;
1075 limit = RIR_LIMIT(reg);
1077 debugf0("RIR#%d, limit: %Lu.%03Lu GB (0x%016Lx), way: %d\n",
1078 n_rir,
1079 (limit >> 20) / 1000, (limit >> 20) % 1000,
1080 limit,
1081 1 << RIR_WAY(reg));
1082 if (ch_addr <= limit)
1083 break;
1085 if (n_rir == MAX_RIR_RANGES) {
1086 sprintf(msg, "Can't discover the memory rank for ch addr 0x%08Lx",
1087 ch_addr);
1088 edac_mc_handle_ce_no_info(mci, msg);
1089 return -EINVAL;
1091 rir_way = RIR_WAY(reg);
1092 if (pvt->is_close_pg)
1093 idx = (ch_addr >> 6);
1094 else
1095 idx = (ch_addr >> 13); /* FIXME: Datasheet says to shift by 15 */
1096 idx %= 1 << rir_way;
1098 pci_read_config_dword(pvt->pci_tad[base_ch],
1099 rir_offset[n_rir][idx],
1100 &reg);
1101 *rank = RIR_RNK_TGT(reg);
1103 debugf0("RIR#%d: channel address 0x%08Lx < 0x%08Lx, RIR interleave %d, index %d\n",
1104 n_rir,
1105 ch_addr,
1106 limit,
1107 rir_way,
1108 idx);
1110 return 0;
1113 /****************************************************************************
1114 Device initialization routines: put/get, init/exit
1115 ****************************************************************************/
1118 * sbridge_put_all_devices 'put' all the devices that we have
1119 * reserved via 'get'
1121 static void sbridge_put_devices(struct sbridge_dev *sbridge_dev)
1123 int i;
1125 debugf0(__FILE__ ": %s()\n", __func__);
1126 for (i = 0; i < sbridge_dev->n_devs; i++) {
1127 struct pci_dev *pdev = sbridge_dev->pdev[i];
1128 if (!pdev)
1129 continue;
1130 debugf0("Removing dev %02x:%02x.%d\n",
1131 pdev->bus->number,
1132 PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn));
1133 pci_dev_put(pdev);
1137 static void sbridge_put_all_devices(void)
1139 struct sbridge_dev *sbridge_dev, *tmp;
1141 list_for_each_entry_safe(sbridge_dev, tmp, &sbridge_edac_list, list) {
1142 sbridge_put_devices(sbridge_dev);
1143 free_sbridge_dev(sbridge_dev);
1148 * sbridge_get_all_devices Find and perform 'get' operation on the MCH's
1149 * device/functions we want to reference for this driver
1151 * Need to 'get' device 16 func 1 and func 2
1153 static int sbridge_get_onedevice(struct pci_dev **prev,
1154 u8 *num_mc,
1155 const struct pci_id_table *table,
1156 const unsigned devno)
1158 struct sbridge_dev *sbridge_dev;
1159 const struct pci_id_descr *dev_descr = &table->descr[devno];
1161 struct pci_dev *pdev = NULL;
1162 u8 bus = 0;
1164 sbridge_printk(KERN_INFO,
1165 "Seeking for: dev %02x.%d PCI ID %04x:%04x\n",
1166 dev_descr->dev, dev_descr->func,
1167 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
1169 pdev = pci_get_device(PCI_VENDOR_ID_INTEL,
1170 dev_descr->dev_id, *prev);
1172 if (!pdev) {
1173 if (*prev) {
1174 *prev = pdev;
1175 return 0;
1178 if (dev_descr->optional)
1179 return 0;
1181 if (devno == 0)
1182 return -ENODEV;
1184 sbridge_printk(KERN_INFO,
1185 "Device not found: dev %02x.%d PCI ID %04x:%04x\n",
1186 dev_descr->dev, dev_descr->func,
1187 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
1189 /* End of list, leave */
1190 return -ENODEV;
1192 bus = pdev->bus->number;
1194 sbridge_dev = get_sbridge_dev(bus);
1195 if (!sbridge_dev) {
1196 sbridge_dev = alloc_sbridge_dev(bus, table);
1197 if (!sbridge_dev) {
1198 pci_dev_put(pdev);
1199 return -ENOMEM;
1201 (*num_mc)++;
1204 if (sbridge_dev->pdev[devno]) {
1205 sbridge_printk(KERN_ERR,
1206 "Duplicated device for "
1207 "dev %02x:%d.%d PCI ID %04x:%04x\n",
1208 bus, dev_descr->dev, dev_descr->func,
1209 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
1210 pci_dev_put(pdev);
1211 return -ENODEV;
1214 sbridge_dev->pdev[devno] = pdev;
1216 /* Sanity check */
1217 if (unlikely(PCI_SLOT(pdev->devfn) != dev_descr->dev ||
1218 PCI_FUNC(pdev->devfn) != dev_descr->func)) {
1219 sbridge_printk(KERN_ERR,
1220 "Device PCI ID %04x:%04x "
1221 "has dev %02x:%d.%d instead of dev %02x:%02x.%d\n",
1222 PCI_VENDOR_ID_INTEL, dev_descr->dev_id,
1223 bus, PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn),
1224 bus, dev_descr->dev, dev_descr->func);
1225 return -ENODEV;
1228 /* Be sure that the device is enabled */
1229 if (unlikely(pci_enable_device(pdev) < 0)) {
1230 sbridge_printk(KERN_ERR,
1231 "Couldn't enable "
1232 "dev %02x:%d.%d PCI ID %04x:%04x\n",
1233 bus, dev_descr->dev, dev_descr->func,
1234 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
1235 return -ENODEV;
1238 debugf0("Detected dev %02x:%d.%d PCI ID %04x:%04x\n",
1239 bus, dev_descr->dev,
1240 dev_descr->func,
1241 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
1244 * As stated on drivers/pci/search.c, the reference count for
1245 * @from is always decremented if it is not %NULL. So, as we need
1246 * to get all devices up to null, we need to do a get for the device
1248 pci_dev_get(pdev);
1250 *prev = pdev;
1252 return 0;
1255 static int sbridge_get_all_devices(u8 *num_mc)
1257 int i, rc;
1258 struct pci_dev *pdev = NULL;
1259 const struct pci_id_table *table = pci_dev_descr_sbridge_table;
1261 while (table && table->descr) {
1262 for (i = 0; i < table->n_devs; i++) {
1263 pdev = NULL;
1264 do {
1265 rc = sbridge_get_onedevice(&pdev, num_mc,
1266 table, i);
1267 if (rc < 0) {
1268 if (i == 0) {
1269 i = table->n_devs;
1270 break;
1272 sbridge_put_all_devices();
1273 return -ENODEV;
1275 } while (pdev);
1277 table++;
1280 return 0;
1283 static int mci_bind_devs(struct mem_ctl_info *mci,
1284 struct sbridge_dev *sbridge_dev)
1286 struct sbridge_pvt *pvt = mci->pvt_info;
1287 struct pci_dev *pdev;
1288 int i, func, slot;
1290 for (i = 0; i < sbridge_dev->n_devs; i++) {
1291 pdev = sbridge_dev->pdev[i];
1292 if (!pdev)
1293 continue;
1294 slot = PCI_SLOT(pdev->devfn);
1295 func = PCI_FUNC(pdev->devfn);
1296 switch (slot) {
1297 case 12:
1298 switch (func) {
1299 case 6:
1300 pvt->pci_sad0 = pdev;
1301 break;
1302 case 7:
1303 pvt->pci_sad1 = pdev;
1304 break;
1305 default:
1306 goto error;
1308 break;
1309 case 13:
1310 switch (func) {
1311 case 6:
1312 pvt->pci_br = pdev;
1313 break;
1314 default:
1315 goto error;
1317 break;
1318 case 14:
1319 switch (func) {
1320 case 0:
1321 pvt->pci_ha0 = pdev;
1322 break;
1323 default:
1324 goto error;
1326 break;
1327 case 15:
1328 switch (func) {
1329 case 0:
1330 pvt->pci_ta = pdev;
1331 break;
1332 case 1:
1333 pvt->pci_ras = pdev;
1334 break;
1335 case 2:
1336 case 3:
1337 case 4:
1338 case 5:
1339 pvt->pci_tad[func - 2] = pdev;
1340 break;
1341 default:
1342 goto error;
1344 break;
1345 case 17:
1346 switch (func) {
1347 case 0:
1348 pvt->pci_ddrio = pdev;
1349 break;
1350 default:
1351 goto error;
1353 break;
1354 default:
1355 goto error;
1358 debugf0("Associated PCI %02x.%02d.%d with dev = %p\n",
1359 sbridge_dev->bus,
1360 PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn),
1361 pdev);
1364 /* Check if everything were registered */
1365 if (!pvt->pci_sad0 || !pvt->pci_sad1 || !pvt->pci_ha0 ||
1366 !pvt-> pci_tad || !pvt->pci_ras || !pvt->pci_ta ||
1367 !pvt->pci_ddrio)
1368 goto enodev;
1370 for (i = 0; i < NUM_CHANNELS; i++) {
1371 if (!pvt->pci_tad[i])
1372 goto enodev;
1374 return 0;
1376 enodev:
1377 sbridge_printk(KERN_ERR, "Some needed devices are missing\n");
1378 return -ENODEV;
1380 error:
1381 sbridge_printk(KERN_ERR, "Device %d, function %d "
1382 "is out of the expected range\n",
1383 slot, func);
1384 return -EINVAL;
1387 /****************************************************************************
1388 Error check routines
1389 ****************************************************************************/
1392 * While Sandy Bridge has error count registers, SMI BIOS read values from
1393 * and resets the counters. So, they are not reliable for the OS to read
1394 * from them. So, we have no option but to just trust on whatever MCE is
1395 * telling us about the errors.
1397 static void sbridge_mce_output_error(struct mem_ctl_info *mci,
1398 const struct mce *m)
1400 struct mem_ctl_info *new_mci;
1401 struct sbridge_pvt *pvt = mci->pvt_info;
1402 char *type, *optype, *msg, *recoverable_msg;
1403 bool ripv = GET_BITFIELD(m->mcgstatus, 0, 0);
1404 bool overflow = GET_BITFIELD(m->status, 62, 62);
1405 bool uncorrected_error = GET_BITFIELD(m->status, 61, 61);
1406 bool recoverable = GET_BITFIELD(m->status, 56, 56);
1407 u32 core_err_cnt = GET_BITFIELD(m->status, 38, 52);
1408 u32 mscod = GET_BITFIELD(m->status, 16, 31);
1409 u32 errcode = GET_BITFIELD(m->status, 0, 15);
1410 u32 channel = GET_BITFIELD(m->status, 0, 3);
1411 u32 optypenum = GET_BITFIELD(m->status, 4, 6);
1412 long channel_mask, first_channel;
1413 u8 rank, socket;
1414 int csrow, rc, dimm;
1415 char *area_type = "Unknown";
1417 if (ripv)
1418 type = "NON_FATAL";
1419 else
1420 type = "FATAL";
1423 * According with Table 15-9 of the Intel Archictecture spec vol 3A,
1424 * memory errors should fit in this mask:
1425 * 000f 0000 1mmm cccc (binary)
1426 * where:
1427 * f = Correction Report Filtering Bit. If 1, subsequent errors
1428 * won't be shown
1429 * mmm = error type
1430 * cccc = channel
1431 * If the mask doesn't match, report an error to the parsing logic
1433 if (! ((errcode & 0xef80) == 0x80)) {
1434 optype = "Can't parse: it is not a mem";
1435 } else {
1436 switch (optypenum) {
1437 case 0:
1438 optype = "generic undef request";
1439 break;
1440 case 1:
1441 optype = "memory read";
1442 break;
1443 case 2:
1444 optype = "memory write";
1445 break;
1446 case 3:
1447 optype = "addr/cmd";
1448 break;
1449 case 4:
1450 optype = "memory scrubbing";
1451 break;
1452 default:
1453 optype = "reserved";
1454 break;
1458 rc = get_memory_error_data(mci, m->addr, &socket,
1459 &channel_mask, &rank, area_type);
1460 if (rc < 0)
1461 return;
1462 new_mci = get_mci_for_node_id(socket);
1463 if (!new_mci) {
1464 edac_mc_handle_ce_no_info(mci, "Error: socket got corrupted!");
1465 return;
1467 mci = new_mci;
1468 pvt = mci->pvt_info;
1470 first_channel = find_first_bit(&channel_mask, NUM_CHANNELS);
1472 if (rank < 4)
1473 dimm = 0;
1474 else if (rank < 8)
1475 dimm = 1;
1476 else
1477 dimm = 2;
1479 csrow = pvt->csrow_map[first_channel][dimm];
1481 if (uncorrected_error && recoverable)
1482 recoverable_msg = " recoverable";
1483 else
1484 recoverable_msg = "";
1487 * FIXME: What should we do with "channel" information on mcelog?
1488 * Probably, we can just discard it, as the channel information
1489 * comes from the get_memory_error_data() address decoding
1491 msg = kasprintf(GFP_ATOMIC,
1492 "%d %s error(s): %s on %s area %s%s: cpu=%d Err=%04x:%04x (ch=%d), "
1493 "addr = 0x%08llx => socket=%d, Channel=%ld(mask=%ld), rank=%d\n",
1494 core_err_cnt,
1495 area_type,
1496 optype,
1497 type,
1498 recoverable_msg,
1499 overflow ? "OVERFLOW" : "",
1500 m->cpu,
1501 mscod, errcode,
1502 channel, /* 1111b means not specified */
1503 (long long) m->addr,
1504 socket,
1505 first_channel, /* This is the real channel on SB */
1506 channel_mask,
1507 rank);
1509 debugf0("%s", msg);
1511 /* Call the helper to output message */
1512 if (uncorrected_error)
1513 edac_mc_handle_fbd_ue(mci, csrow, 0, 0, msg);
1514 else
1515 edac_mc_handle_fbd_ce(mci, csrow, 0, msg);
1517 kfree(msg);
1521 * sbridge_check_error Retrieve and process errors reported by the
1522 * hardware. Called by the Core module.
1524 static void sbridge_check_error(struct mem_ctl_info *mci)
1526 struct sbridge_pvt *pvt = mci->pvt_info;
1527 int i;
1528 unsigned count = 0;
1529 struct mce *m;
1532 * MCE first step: Copy all mce errors into a temporary buffer
1533 * We use a double buffering here, to reduce the risk of
1534 * loosing an error.
1536 smp_rmb();
1537 count = (pvt->mce_out + MCE_LOG_LEN - pvt->mce_in)
1538 % MCE_LOG_LEN;
1539 if (!count)
1540 return;
1542 m = pvt->mce_outentry;
1543 if (pvt->mce_in + count > MCE_LOG_LEN) {
1544 unsigned l = MCE_LOG_LEN - pvt->mce_in;
1546 memcpy(m, &pvt->mce_entry[pvt->mce_in], sizeof(*m) * l);
1547 smp_wmb();
1548 pvt->mce_in = 0;
1549 count -= l;
1550 m += l;
1552 memcpy(m, &pvt->mce_entry[pvt->mce_in], sizeof(*m) * count);
1553 smp_wmb();
1554 pvt->mce_in += count;
1556 smp_rmb();
1557 if (pvt->mce_overrun) {
1558 sbridge_printk(KERN_ERR, "Lost %d memory errors\n",
1559 pvt->mce_overrun);
1560 smp_wmb();
1561 pvt->mce_overrun = 0;
1565 * MCE second step: parse errors and display
1567 for (i = 0; i < count; i++)
1568 sbridge_mce_output_error(mci, &pvt->mce_outentry[i]);
1572 * sbridge_mce_check_error Replicates mcelog routine to get errors
1573 * This routine simply queues mcelog errors, and
1574 * return. The error itself should be handled later
1575 * by sbridge_check_error.
1576 * WARNING: As this routine should be called at NMI time, extra care should
1577 * be taken to avoid deadlocks, and to be as fast as possible.
1579 static int sbridge_mce_check_error(struct notifier_block *nb, unsigned long val,
1580 void *data)
1582 struct mce *mce = (struct mce *)data;
1583 struct mem_ctl_info *mci;
1584 struct sbridge_pvt *pvt;
1586 mci = get_mci_for_node_id(mce->socketid);
1587 if (!mci)
1588 return NOTIFY_BAD;
1589 pvt = mci->pvt_info;
1592 * Just let mcelog handle it if the error is
1593 * outside the memory controller. A memory error
1594 * is indicated by bit 7 = 1 and bits = 8-11,13-15 = 0.
1595 * bit 12 has an special meaning.
1597 if ((mce->status & 0xefff) >> 7 != 1)
1598 return NOTIFY_DONE;
1600 printk("sbridge: HANDLING MCE MEMORY ERROR\n");
1602 printk("CPU %d: Machine Check Exception: %Lx Bank %d: %016Lx\n",
1603 mce->extcpu, mce->mcgstatus, mce->bank, mce->status);
1604 printk("TSC %llx ", mce->tsc);
1605 printk("ADDR %llx ", mce->addr);
1606 printk("MISC %llx ", mce->misc);
1608 printk("PROCESSOR %u:%x TIME %llu SOCKET %u APIC %x\n",
1609 mce->cpuvendor, mce->cpuid, mce->time,
1610 mce->socketid, mce->apicid);
1612 /* Only handle if it is the right mc controller */
1613 if (cpu_data(mce->cpu).phys_proc_id != pvt->sbridge_dev->mc)
1614 return NOTIFY_DONE;
1616 smp_rmb();
1617 if ((pvt->mce_out + 1) % MCE_LOG_LEN == pvt->mce_in) {
1618 smp_wmb();
1619 pvt->mce_overrun++;
1620 return NOTIFY_DONE;
1623 /* Copy memory error at the ringbuffer */
1624 memcpy(&pvt->mce_entry[pvt->mce_out], mce, sizeof(*mce));
1625 smp_wmb();
1626 pvt->mce_out = (pvt->mce_out + 1) % MCE_LOG_LEN;
1628 /* Handle fatal errors immediately */
1629 if (mce->mcgstatus & 1)
1630 sbridge_check_error(mci);
1632 /* Advice mcelog that the error were handled */
1633 return NOTIFY_STOP;
1636 static struct notifier_block sbridge_mce_dec = {
1637 .notifier_call = sbridge_mce_check_error,
1640 /****************************************************************************
1641 EDAC register/unregister logic
1642 ****************************************************************************/
1644 static void sbridge_unregister_mci(struct sbridge_dev *sbridge_dev)
1646 struct mem_ctl_info *mci = sbridge_dev->mci;
1647 struct sbridge_pvt *pvt;
1649 if (unlikely(!mci || !mci->pvt_info)) {
1650 debugf0("MC: " __FILE__ ": %s(): dev = %p\n",
1651 __func__, &sbridge_dev->pdev[0]->dev);
1653 sbridge_printk(KERN_ERR, "Couldn't find mci handler\n");
1654 return;
1657 pvt = mci->pvt_info;
1659 debugf0("MC: " __FILE__ ": %s(): mci = %p, dev = %p\n",
1660 __func__, mci, &sbridge_dev->pdev[0]->dev);
1662 mce_unregister_decode_chain(&sbridge_mce_dec);
1664 /* Remove MC sysfs nodes */
1665 edac_mc_del_mc(mci->dev);
1667 debugf1("%s: free mci struct\n", mci->ctl_name);
1668 kfree(mci->ctl_name);
1669 edac_mc_free(mci);
1670 sbridge_dev->mci = NULL;
1673 static int sbridge_register_mci(struct sbridge_dev *sbridge_dev)
1675 struct mem_ctl_info *mci;
1676 struct sbridge_pvt *pvt;
1677 int rc, channels, csrows;
1679 /* Check the number of active and not disabled channels */
1680 rc = sbridge_get_active_channels(sbridge_dev->bus, &channels, &csrows);
1681 if (unlikely(rc < 0))
1682 return rc;
1684 /* allocate a new MC control structure */
1685 mci = edac_mc_alloc(sizeof(*pvt), csrows, channels, sbridge_dev->mc);
1686 if (unlikely(!mci))
1687 return -ENOMEM;
1689 debugf0("MC: " __FILE__ ": %s(): mci = %p, dev = %p\n",
1690 __func__, mci, &sbridge_dev->pdev[0]->dev);
1692 pvt = mci->pvt_info;
1693 memset(pvt, 0, sizeof(*pvt));
1695 /* Associate sbridge_dev and mci for future usage */
1696 pvt->sbridge_dev = sbridge_dev;
1697 sbridge_dev->mci = mci;
1699 mci->mtype_cap = MEM_FLAG_DDR3;
1700 mci->edac_ctl_cap = EDAC_FLAG_NONE;
1701 mci->edac_cap = EDAC_FLAG_NONE;
1702 mci->mod_name = "sbridge_edac.c";
1703 mci->mod_ver = SBRIDGE_REVISION;
1704 mci->ctl_name = kasprintf(GFP_KERNEL, "Sandy Bridge Socket#%d", mci->mc_idx);
1705 mci->dev_name = pci_name(sbridge_dev->pdev[0]);
1706 mci->ctl_page_to_phys = NULL;
1708 /* Set the function pointer to an actual operation function */
1709 mci->edac_check = sbridge_check_error;
1711 /* Store pci devices at mci for faster access */
1712 rc = mci_bind_devs(mci, sbridge_dev);
1713 if (unlikely(rc < 0))
1714 goto fail0;
1716 /* Get dimm basic config and the memory layout */
1717 get_dimm_config(mci);
1718 get_memory_layout(mci);
1720 /* record ptr to the generic device */
1721 mci->dev = &sbridge_dev->pdev[0]->dev;
1723 /* add this new MC control structure to EDAC's list of MCs */
1724 if (unlikely(edac_mc_add_mc(mci))) {
1725 debugf0("MC: " __FILE__
1726 ": %s(): failed edac_mc_add_mc()\n", __func__);
1727 rc = -EINVAL;
1728 goto fail0;
1731 mce_register_decode_chain(&sbridge_mce_dec);
1732 return 0;
1734 fail0:
1735 kfree(mci->ctl_name);
1736 edac_mc_free(mci);
1737 sbridge_dev->mci = NULL;
1738 return rc;
1742 * sbridge_probe Probe for ONE instance of device to see if it is
1743 * present.
1744 * return:
1745 * 0 for FOUND a device
1746 * < 0 for error code
1749 static int __devinit sbridge_probe(struct pci_dev *pdev,
1750 const struct pci_device_id *id)
1752 int rc;
1753 u8 mc, num_mc = 0;
1754 struct sbridge_dev *sbridge_dev;
1756 /* get the pci devices we want to reserve for our use */
1757 mutex_lock(&sbridge_edac_lock);
1760 * All memory controllers are allocated at the first pass.
1762 if (unlikely(probed >= 1)) {
1763 mutex_unlock(&sbridge_edac_lock);
1764 return -ENODEV;
1766 probed++;
1768 rc = sbridge_get_all_devices(&num_mc);
1769 if (unlikely(rc < 0))
1770 goto fail0;
1771 mc = 0;
1773 list_for_each_entry(sbridge_dev, &sbridge_edac_list, list) {
1774 debugf0("Registering MC#%d (%d of %d)\n", mc, mc + 1, num_mc);
1775 sbridge_dev->mc = mc++;
1776 rc = sbridge_register_mci(sbridge_dev);
1777 if (unlikely(rc < 0))
1778 goto fail1;
1781 sbridge_printk(KERN_INFO, "Driver loaded.\n");
1783 mutex_unlock(&sbridge_edac_lock);
1784 return 0;
1786 fail1:
1787 list_for_each_entry(sbridge_dev, &sbridge_edac_list, list)
1788 sbridge_unregister_mci(sbridge_dev);
1790 sbridge_put_all_devices();
1791 fail0:
1792 mutex_unlock(&sbridge_edac_lock);
1793 return rc;
1797 * sbridge_remove destructor for one instance of device
1800 static void __devexit sbridge_remove(struct pci_dev *pdev)
1802 struct sbridge_dev *sbridge_dev;
1804 debugf0(__FILE__ ": %s()\n", __func__);
1807 * we have a trouble here: pdev value for removal will be wrong, since
1808 * it will point to the X58 register used to detect that the machine
1809 * is a Nehalem or upper design. However, due to the way several PCI
1810 * devices are grouped together to provide MC functionality, we need
1811 * to use a different method for releasing the devices
1814 mutex_lock(&sbridge_edac_lock);
1816 if (unlikely(!probed)) {
1817 mutex_unlock(&sbridge_edac_lock);
1818 return;
1821 list_for_each_entry(sbridge_dev, &sbridge_edac_list, list)
1822 sbridge_unregister_mci(sbridge_dev);
1824 /* Release PCI resources */
1825 sbridge_put_all_devices();
1827 probed--;
1829 mutex_unlock(&sbridge_edac_lock);
1832 MODULE_DEVICE_TABLE(pci, sbridge_pci_tbl);
1835 * sbridge_driver pci_driver structure for this module
1838 static struct pci_driver sbridge_driver = {
1839 .name = "sbridge_edac",
1840 .probe = sbridge_probe,
1841 .remove = __devexit_p(sbridge_remove),
1842 .id_table = sbridge_pci_tbl,
1846 * sbridge_init Module entry function
1847 * Try to initialize this module for its devices
1849 static int __init sbridge_init(void)
1851 int pci_rc;
1853 debugf2("MC: " __FILE__ ": %s()\n", __func__);
1855 /* Ensure that the OPSTATE is set correctly for POLL or NMI */
1856 opstate_init();
1858 pci_rc = pci_register_driver(&sbridge_driver);
1860 if (pci_rc >= 0)
1861 return 0;
1863 sbridge_printk(KERN_ERR, "Failed to register device with error %d.\n",
1864 pci_rc);
1866 return pci_rc;
1870 * sbridge_exit() Module exit function
1871 * Unregister the driver
1873 static void __exit sbridge_exit(void)
1875 debugf2("MC: " __FILE__ ": %s()\n", __func__);
1876 pci_unregister_driver(&sbridge_driver);
1879 module_init(sbridge_init);
1880 module_exit(sbridge_exit);
1882 module_param(edac_op_state, int, 0444);
1883 MODULE_PARM_DESC(edac_op_state, "EDAC Error Reporting state: 0=Poll,1=NMI");
1885 MODULE_LICENSE("GPL");
1886 MODULE_AUTHOR("Mauro Carvalho Chehab <mchehab@redhat.com>");
1887 MODULE_AUTHOR("Red Hat Inc. (http://www.redhat.com)");
1888 MODULE_DESCRIPTION("MC Driver for Intel Sandy Bridge memory controllers - "
1889 SBRIDGE_REVISION);