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 <linux/math64.h>
24 #include <asm/processor.h>
27 #include "edac_core.h"
30 static LIST_HEAD(sbridge_edac_list);
31 static DEFINE_MUTEX(sbridge_edac_lock);
35 * Alter this version for the module when modifications are made
37 #define SBRIDGE_REVISION " Ver: 1.0.0 "
38 #define EDAC_MOD_STR "sbridge_edac"
43 #define sbridge_printk(level, fmt, arg...) \
44 edac_printk(level, "sbridge", fmt, ##arg)
46 #define sbridge_mc_printk(mci, level, fmt, arg...) \
47 edac_mc_chipset_printk(mci, level, "sbridge", fmt, ##arg)
50 * Get a bit field at register value <v>, from bit <lo> to bit <hi>
52 #define GET_BITFIELD(v, lo, hi) \
53 (((v) & ((1ULL << ((hi) - (lo) + 1)) - 1) << (lo)) >> (lo))
56 * sbridge Memory Controller Registers
60 * FIXME: For now, let's order by device function, as it makes
61 * easier for driver's development process. This table should be
62 * moved to pci_id.h when submitted upstream
64 #define PCI_DEVICE_ID_INTEL_SBRIDGE_SAD0 0x3cf4 /* 12.6 */
65 #define PCI_DEVICE_ID_INTEL_SBRIDGE_SAD1 0x3cf6 /* 12.7 */
66 #define PCI_DEVICE_ID_INTEL_SBRIDGE_BR 0x3cf5 /* 13.6 */
67 #define PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_HA0 0x3ca0 /* 14.0 */
68 #define PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TA 0x3ca8 /* 15.0 */
69 #define PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_RAS 0x3c71 /* 15.1 */
70 #define PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD0 0x3caa /* 15.2 */
71 #define PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD1 0x3cab /* 15.3 */
72 #define PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD2 0x3cac /* 15.4 */
73 #define PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD3 0x3cad /* 15.5 */
74 #define PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_DDRIO 0x3cb8 /* 17.0 */
77 * Currently, unused, but will be needed in the future
78 * implementations, as they hold the error counters
80 #define PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_ERR0 0x3c72 /* 16.2 */
81 #define PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_ERR1 0x3c73 /* 16.3 */
82 #define PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_ERR2 0x3c76 /* 16.6 */
83 #define PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_ERR3 0x3c77 /* 16.7 */
85 /* Devices 12 Function 6, Offsets 0x80 to 0xcc */
86 static const u32 dram_rule[] = {
87 0x80, 0x88, 0x90, 0x98, 0xa0,
88 0xa8, 0xb0, 0xb8, 0xc0, 0xc8,
90 #define MAX_SAD ARRAY_SIZE(dram_rule)
92 #define SAD_LIMIT(reg) ((GET_BITFIELD(reg, 6, 25) << 26) | 0x3ffffff)
93 #define DRAM_ATTR(reg) GET_BITFIELD(reg, 2, 3)
94 #define INTERLEAVE_MODE(reg) GET_BITFIELD(reg, 1, 1)
95 #define DRAM_RULE_ENABLE(reg) GET_BITFIELD(reg, 0, 0)
97 static char *get_dram_attr(u32 reg)
99 switch(DRAM_ATTR(reg)) {
111 static const u32 interleave_list[] = {
112 0x84, 0x8c, 0x94, 0x9c, 0xa4,
113 0xac, 0xb4, 0xbc, 0xc4, 0xcc,
115 #define MAX_INTERLEAVE ARRAY_SIZE(interleave_list)
117 #define SAD_PKG0(reg) GET_BITFIELD(reg, 0, 2)
118 #define SAD_PKG1(reg) GET_BITFIELD(reg, 3, 5)
119 #define SAD_PKG2(reg) GET_BITFIELD(reg, 8, 10)
120 #define SAD_PKG3(reg) GET_BITFIELD(reg, 11, 13)
121 #define SAD_PKG4(reg) GET_BITFIELD(reg, 16, 18)
122 #define SAD_PKG5(reg) GET_BITFIELD(reg, 19, 21)
123 #define SAD_PKG6(reg) GET_BITFIELD(reg, 24, 26)
124 #define SAD_PKG7(reg) GET_BITFIELD(reg, 27, 29)
126 static inline int sad_pkg(u32 reg, int interleave)
128 switch (interleave) {
130 return SAD_PKG0(reg);
132 return SAD_PKG1(reg);
134 return SAD_PKG2(reg);
136 return SAD_PKG3(reg);
138 return SAD_PKG4(reg);
140 return SAD_PKG5(reg);
142 return SAD_PKG6(reg);
144 return SAD_PKG7(reg);
150 /* Devices 12 Function 7 */
155 #define GET_TOLM(reg) ((GET_BITFIELD(reg, 0, 3) << 28) | 0x3ffffff)
156 #define GET_TOHM(reg) ((GET_BITFIELD(reg, 0, 20) << 25) | 0x3ffffff)
158 /* Device 13 Function 6 */
160 #define SAD_TARGET 0xf0
162 #define SOURCE_ID(reg) GET_BITFIELD(reg, 9, 11)
164 #define SAD_CONTROL 0xf4
166 #define NODE_ID(reg) GET_BITFIELD(reg, 0, 2)
168 /* Device 14 function 0 */
170 static const u32 tad_dram_rule[] = {
171 0x40, 0x44, 0x48, 0x4c,
172 0x50, 0x54, 0x58, 0x5c,
173 0x60, 0x64, 0x68, 0x6c,
175 #define MAX_TAD ARRAY_SIZE(tad_dram_rule)
177 #define TAD_LIMIT(reg) ((GET_BITFIELD(reg, 12, 31) << 26) | 0x3ffffff)
178 #define TAD_SOCK(reg) GET_BITFIELD(reg, 10, 11)
179 #define TAD_CH(reg) GET_BITFIELD(reg, 8, 9)
180 #define TAD_TGT3(reg) GET_BITFIELD(reg, 6, 7)
181 #define TAD_TGT2(reg) GET_BITFIELD(reg, 4, 5)
182 #define TAD_TGT1(reg) GET_BITFIELD(reg, 2, 3)
183 #define TAD_TGT0(reg) GET_BITFIELD(reg, 0, 1)
185 /* Device 15, function 0 */
189 #define IS_ECC_ENABLED(mcmtr) GET_BITFIELD(mcmtr, 2, 2)
190 #define IS_LOCKSTEP_ENABLED(mcmtr) GET_BITFIELD(mcmtr, 1, 1)
191 #define IS_CLOSE_PG(mcmtr) GET_BITFIELD(mcmtr, 0, 0)
193 /* Device 15, function 1 */
195 #define RASENABLES 0xac
196 #define IS_MIRROR_ENABLED(reg) GET_BITFIELD(reg, 0, 0)
198 /* Device 15, functions 2-5 */
200 static const int mtr_regs[] = {
204 #define RANK_DISABLE(mtr) GET_BITFIELD(mtr, 16, 19)
205 #define IS_DIMM_PRESENT(mtr) GET_BITFIELD(mtr, 14, 14)
206 #define RANK_CNT_BITS(mtr) GET_BITFIELD(mtr, 12, 13)
207 #define RANK_WIDTH_BITS(mtr) GET_BITFIELD(mtr, 2, 4)
208 #define COL_WIDTH_BITS(mtr) GET_BITFIELD(mtr, 0, 1)
210 static const u32 tad_ch_nilv_offset[] = {
211 0x90, 0x94, 0x98, 0x9c,
212 0xa0, 0xa4, 0xa8, 0xac,
213 0xb0, 0xb4, 0xb8, 0xbc,
215 #define CHN_IDX_OFFSET(reg) GET_BITFIELD(reg, 28, 29)
216 #define TAD_OFFSET(reg) (GET_BITFIELD(reg, 6, 25) << 26)
218 static const u32 rir_way_limit[] = {
219 0x108, 0x10c, 0x110, 0x114, 0x118,
221 #define MAX_RIR_RANGES ARRAY_SIZE(rir_way_limit)
223 #define IS_RIR_VALID(reg) GET_BITFIELD(reg, 31, 31)
224 #define RIR_WAY(reg) GET_BITFIELD(reg, 28, 29)
225 #define RIR_LIMIT(reg) ((GET_BITFIELD(reg, 1, 10) << 29)| 0x1fffffff)
227 #define MAX_RIR_WAY 8
229 static const u32 rir_offset[MAX_RIR_RANGES][MAX_RIR_WAY] = {
230 { 0x120, 0x124, 0x128, 0x12c, 0x130, 0x134, 0x138, 0x13c },
231 { 0x140, 0x144, 0x148, 0x14c, 0x150, 0x154, 0x158, 0x15c },
232 { 0x160, 0x164, 0x168, 0x16c, 0x170, 0x174, 0x178, 0x17c },
233 { 0x180, 0x184, 0x188, 0x18c, 0x190, 0x194, 0x198, 0x19c },
234 { 0x1a0, 0x1a4, 0x1a8, 0x1ac, 0x1b0, 0x1b4, 0x1b8, 0x1bc },
237 #define RIR_RNK_TGT(reg) GET_BITFIELD(reg, 16, 19)
238 #define RIR_OFFSET(reg) GET_BITFIELD(reg, 2, 14)
240 /* Device 16, functions 2-7 */
243 * FIXME: Implement the error count reads directly
246 static const u32 correrrcnt[] = {
247 0x104, 0x108, 0x10c, 0x110,
250 #define RANK_ODD_OV(reg) GET_BITFIELD(reg, 31, 31)
251 #define RANK_ODD_ERR_CNT(reg) GET_BITFIELD(reg, 16, 30)
252 #define RANK_EVEN_OV(reg) GET_BITFIELD(reg, 15, 15)
253 #define RANK_EVEN_ERR_CNT(reg) GET_BITFIELD(reg, 0, 14)
255 static const u32 correrrthrsld[] = {
256 0x11c, 0x120, 0x124, 0x128,
259 #define RANK_ODD_ERR_THRSLD(reg) GET_BITFIELD(reg, 16, 30)
260 #define RANK_EVEN_ERR_THRSLD(reg) GET_BITFIELD(reg, 0, 14)
263 /* Device 17, function 0 */
265 #define RANK_CFG_A 0x0328
267 #define IS_RDIMM_ENABLED(reg) GET_BITFIELD(reg, 11, 11)
273 #define NUM_CHANNELS 4
274 #define MAX_DIMMS 3 /* Max DIMMS per channel */
275 #define CHANNEL_UNSPECIFIED 0xf /* Intel IA32 SDM 15-14 */
277 struct sbridge_info {
281 struct sbridge_channel {
286 struct pci_id_descr {
293 struct pci_id_table {
294 const struct pci_id_descr *descr;
299 struct list_head list;
301 u8 node_id, source_id;
302 struct pci_dev **pdev;
304 struct mem_ctl_info *mci;
308 struct pci_dev *pci_ta, *pci_ddrio, *pci_ras;
309 struct pci_dev *pci_sad0, *pci_sad1, *pci_ha0;
310 struct pci_dev *pci_br;
311 struct pci_dev *pci_tad[NUM_CHANNELS];
313 struct sbridge_dev *sbridge_dev;
315 struct sbridge_info info;
316 struct sbridge_channel channel[NUM_CHANNELS];
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 */
335 #define PCI_DESCR(device, function, device_id, opt) \
337 .func = (function), \
338 .dev_id = (device_id), \
341 static const struct pci_id_descr pci_dev_descr_sbridge[] = {
342 /* Processor Home Agent */
343 { PCI_DESCR(14, 0, PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_HA0, 0) },
345 /* Memory controller */
346 { PCI_DESCR(15, 0, PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TA, 0) },
347 { PCI_DESCR(15, 1, PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_RAS, 0) },
348 { PCI_DESCR(15, 2, PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD0, 0) },
349 { PCI_DESCR(15, 3, PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD1, 0) },
350 { PCI_DESCR(15, 4, PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD2, 0) },
351 { PCI_DESCR(15, 5, PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD3, 0) },
352 { PCI_DESCR(17, 0, PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_DDRIO, 1) },
354 /* System Address Decoder */
355 { PCI_DESCR(12, 6, PCI_DEVICE_ID_INTEL_SBRIDGE_SAD0, 0) },
356 { PCI_DESCR(12, 7, PCI_DEVICE_ID_INTEL_SBRIDGE_SAD1, 0) },
358 /* Broadcast Registers */
359 { PCI_DESCR(13, 6, PCI_DEVICE_ID_INTEL_SBRIDGE_BR, 0) },
362 #define PCI_ID_TABLE_ENTRY(A) { .descr=A, .n_devs = ARRAY_SIZE(A) }
363 static const struct pci_id_table pci_dev_descr_sbridge_table[] = {
364 PCI_ID_TABLE_ENTRY(pci_dev_descr_sbridge),
365 {0,} /* 0 terminated list. */
369 * pci_device_id table for which devices we are looking for
371 static DEFINE_PCI_DEVICE_TABLE(sbridge_pci_tbl) = {
372 {PCI_DEVICE(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TA)},
373 {0,} /* 0 terminated list. */
377 /****************************************************************************
378 Ancillary status routines
379 ****************************************************************************/
381 static inline int numrank(u32 mtr)
383 int ranks = (1 << RANK_CNT_BITS(mtr));
386 edac_dbg(0, "Invalid number of ranks: %d (max = 4) raw value = %x (%04x)\n",
387 ranks, (unsigned int)RANK_CNT_BITS(mtr), mtr);
394 static inline int numrow(u32 mtr)
396 int rows = (RANK_WIDTH_BITS(mtr) + 12);
398 if (rows < 13 || rows > 18) {
399 edac_dbg(0, "Invalid number of rows: %d (should be between 14 and 17) raw value = %x (%04x)\n",
400 rows, (unsigned int)RANK_WIDTH_BITS(mtr), mtr);
407 static inline int numcol(u32 mtr)
409 int cols = (COL_WIDTH_BITS(mtr) + 10);
412 edac_dbg(0, "Invalid number of cols: %d (max = 4) raw value = %x (%04x)\n",
413 cols, (unsigned int)COL_WIDTH_BITS(mtr), mtr);
420 static struct sbridge_dev *get_sbridge_dev(u8 bus)
422 struct sbridge_dev *sbridge_dev;
424 list_for_each_entry(sbridge_dev, &sbridge_edac_list, list) {
425 if (sbridge_dev->bus == bus)
432 static struct sbridge_dev *alloc_sbridge_dev(u8 bus,
433 const struct pci_id_table *table)
435 struct sbridge_dev *sbridge_dev;
437 sbridge_dev = kzalloc(sizeof(*sbridge_dev), GFP_KERNEL);
441 sbridge_dev->pdev = kzalloc(sizeof(*sbridge_dev->pdev) * table->n_devs,
443 if (!sbridge_dev->pdev) {
448 sbridge_dev->bus = bus;
449 sbridge_dev->n_devs = table->n_devs;
450 list_add_tail(&sbridge_dev->list, &sbridge_edac_list);
455 static void free_sbridge_dev(struct sbridge_dev *sbridge_dev)
457 list_del(&sbridge_dev->list);
458 kfree(sbridge_dev->pdev);
462 /****************************************************************************
463 Memory check routines
464 ****************************************************************************/
465 static struct pci_dev *get_pdev_slot_func(u8 bus, unsigned slot,
468 struct sbridge_dev *sbridge_dev = get_sbridge_dev(bus);
474 for (i = 0; i < sbridge_dev->n_devs; i++) {
475 if (!sbridge_dev->pdev[i])
478 if (PCI_SLOT(sbridge_dev->pdev[i]->devfn) == slot &&
479 PCI_FUNC(sbridge_dev->pdev[i]->devfn) == func) {
480 edac_dbg(1, "Associated %02x.%02x.%d with %p\n",
481 bus, slot, func, sbridge_dev->pdev[i]);
482 return sbridge_dev->pdev[i];
490 * check_if_ecc_is_active() - Checks if ECC is active
493 static int check_if_ecc_is_active(const u8 bus)
495 struct pci_dev *pdev = NULL;
498 pdev = get_pdev_slot_func(bus, 15, 0);
500 sbridge_printk(KERN_ERR, "Couldn't find PCI device "
506 pci_read_config_dword(pdev, MCMTR, &mcmtr);
507 if (!IS_ECC_ENABLED(mcmtr)) {
508 sbridge_printk(KERN_ERR, "ECC is disabled. Aborting\n");
514 static int get_dimm_config(struct mem_ctl_info *mci)
516 struct sbridge_pvt *pvt = mci->pvt_info;
517 struct dimm_info *dimm;
518 unsigned i, j, banks, ranks, rows, cols, npages;
524 pci_read_config_dword(pvt->pci_br, SAD_TARGET, ®);
525 pvt->sbridge_dev->source_id = SOURCE_ID(reg);
527 pci_read_config_dword(pvt->pci_br, SAD_CONTROL, ®);
528 pvt->sbridge_dev->node_id = NODE_ID(reg);
529 edac_dbg(0, "mc#%d: Node ID: %d, source ID: %d\n",
530 pvt->sbridge_dev->mc,
531 pvt->sbridge_dev->node_id,
532 pvt->sbridge_dev->source_id);
534 pci_read_config_dword(pvt->pci_ras, RASENABLES, ®);
535 if (IS_MIRROR_ENABLED(reg)) {
536 edac_dbg(0, "Memory mirror is enabled\n");
537 pvt->is_mirrored = true;
539 edac_dbg(0, "Memory mirror is disabled\n");
540 pvt->is_mirrored = false;
543 pci_read_config_dword(pvt->pci_ta, MCMTR, &pvt->info.mcmtr);
544 if (IS_LOCKSTEP_ENABLED(pvt->info.mcmtr)) {
545 edac_dbg(0, "Lockstep is enabled\n");
546 mode = EDAC_S8ECD8ED;
547 pvt->is_lockstep = true;
549 edac_dbg(0, "Lockstep is disabled\n");
550 mode = EDAC_S4ECD4ED;
551 pvt->is_lockstep = false;
553 if (IS_CLOSE_PG(pvt->info.mcmtr)) {
554 edac_dbg(0, "address map is on closed page mode\n");
555 pvt->is_close_pg = true;
557 edac_dbg(0, "address map is on open page mode\n");
558 pvt->is_close_pg = false;
561 if (pvt->pci_ddrio) {
562 pci_read_config_dword(pvt->pci_ddrio, RANK_CFG_A, ®);
563 if (IS_RDIMM_ENABLED(reg)) {
564 /* FIXME: Can also be LRDIMM */
565 edac_dbg(0, "Memory is registered\n");
568 edac_dbg(0, "Memory is unregistered\n");
572 edac_dbg(0, "Cannot determine memory type\n");
576 /* On all supported DDR3 DIMM types, there are 8 banks available */
579 for (i = 0; i < NUM_CHANNELS; i++) {
582 for (j = 0; j < ARRAY_SIZE(mtr_regs); j++) {
583 dimm = EDAC_DIMM_PTR(mci->layers, mci->dimms, mci->n_layers,
585 pci_read_config_dword(pvt->pci_tad[i],
587 edac_dbg(4, "Channel #%d MTR%d = %x\n", i, j, mtr);
588 if (IS_DIMM_PRESENT(mtr)) {
589 pvt->channel[i].dimms++;
591 ranks = numrank(mtr);
595 /* DDR3 has 8 I/O banks */
596 size = ((u64)rows * cols * banks * ranks) >> (20 - 3);
597 npages = MiB_TO_PAGES(size);
599 edac_dbg(0, "mc#%d: channel %d, dimm %d, %Ld Mb (%d pages) bank: %d, rank: %d, row: %#x, col: %#x\n",
600 pvt->sbridge_dev->mc, i, j,
602 banks, ranks, rows, cols);
604 dimm->nr_pages = npages;
606 dimm->dtype = (banks == 8) ? DEV_X8 : DEV_X4;
608 dimm->edac_mode = mode;
609 snprintf(dimm->label, sizeof(dimm->label),
610 "CPU_SrcID#%u_Channel#%u_DIMM#%u",
611 pvt->sbridge_dev->source_id, i, j);
619 static void get_memory_layout(const struct mem_ctl_info *mci)
621 struct sbridge_pvt *pvt = mci->pvt_info;
622 int i, j, k, n_sads, n_tads, sad_interl;
630 * Step 1) Get TOLM/TOHM ranges
633 /* Address range is 32:28 */
634 pci_read_config_dword(pvt->pci_sad1, TOLM,
636 pvt->tolm = GET_TOLM(reg);
637 tmp_mb = (1 + pvt->tolm) >> 20;
639 mb = div_u64_rem(tmp_mb, 1000, &kb);
640 edac_dbg(0, "TOLM: %u.%03u GB (0x%016Lx)\n", mb, kb, (u64)pvt->tolm);
642 /* Address range is already 45:25 */
643 pci_read_config_dword(pvt->pci_sad1, TOHM,
645 pvt->tohm = GET_TOHM(reg);
646 tmp_mb = (1 + pvt->tohm) >> 20;
648 mb = div_u64_rem(tmp_mb, 1000, &kb);
649 edac_dbg(0, "TOHM: %u.%03u GB (0x%016Lx)\n", mb, kb, (u64)pvt->tohm);
652 * Step 2) Get SAD range and SAD Interleave list
653 * TAD registers contain the interleave wayness. However, it
654 * seems simpler to just discover it indirectly, with the
658 for (n_sads = 0; n_sads < MAX_SAD; n_sads++) {
659 /* SAD_LIMIT Address range is 45:26 */
660 pci_read_config_dword(pvt->pci_sad0, dram_rule[n_sads],
662 limit = SAD_LIMIT(reg);
664 if (!DRAM_RULE_ENABLE(reg))
670 tmp_mb = (limit + 1) >> 20;
671 mb = div_u64_rem(tmp_mb, 1000, &kb);
672 edac_dbg(0, "SAD#%d %s up to %u.%03u GB (0x%016Lx) Interleave: %s reg=0x%08x\n",
676 ((u64)tmp_mb) << 20L,
677 INTERLEAVE_MODE(reg) ? "8:6" : "[8:6]XOR[18:16]",
681 pci_read_config_dword(pvt->pci_sad0, interleave_list[n_sads],
683 sad_interl = sad_pkg(reg, 0);
684 for (j = 0; j < 8; j++) {
685 if (j > 0 && sad_interl == sad_pkg(reg, j))
688 edac_dbg(0, "SAD#%d, interleave #%d: %d\n",
689 n_sads, j, sad_pkg(reg, j));
694 * Step 3) Get TAD range
697 for (n_tads = 0; n_tads < MAX_TAD; n_tads++) {
698 pci_read_config_dword(pvt->pci_ha0, tad_dram_rule[n_tads],
700 limit = TAD_LIMIT(reg);
703 tmp_mb = (limit + 1) >> 20;
705 mb = div_u64_rem(tmp_mb, 1000, &kb);
706 edac_dbg(0, "TAD#%d: up to %u.%03u GB (0x%016Lx), socket interleave %d, memory interleave %d, TGT: %d, %d, %d, %d, reg=0x%08x\n",
708 ((u64)tmp_mb) << 20L,
720 * Step 4) Get TAD offsets, per each channel
722 for (i = 0; i < NUM_CHANNELS; i++) {
723 if (!pvt->channel[i].dimms)
725 for (j = 0; j < n_tads; j++) {
726 pci_read_config_dword(pvt->pci_tad[i],
727 tad_ch_nilv_offset[j],
729 tmp_mb = TAD_OFFSET(reg) >> 20;
730 mb = div_u64_rem(tmp_mb, 1000, &kb);
731 edac_dbg(0, "TAD CH#%d, offset #%d: %u.%03u GB (0x%016Lx), reg=0x%08x\n",
734 ((u64)tmp_mb) << 20L,
740 * Step 6) Get RIR Wayness/Limit, per each channel
742 for (i = 0; i < NUM_CHANNELS; i++) {
743 if (!pvt->channel[i].dimms)
745 for (j = 0; j < MAX_RIR_RANGES; j++) {
746 pci_read_config_dword(pvt->pci_tad[i],
750 if (!IS_RIR_VALID(reg))
753 tmp_mb = RIR_LIMIT(reg) >> 20;
754 rir_way = 1 << RIR_WAY(reg);
755 mb = div_u64_rem(tmp_mb, 1000, &kb);
756 edac_dbg(0, "CH#%d RIR#%d, limit: %u.%03u GB (0x%016Lx), way: %d, reg=0x%08x\n",
759 ((u64)tmp_mb) << 20L,
763 for (k = 0; k < rir_way; k++) {
764 pci_read_config_dword(pvt->pci_tad[i],
767 tmp_mb = RIR_OFFSET(reg) << 6;
769 mb = div_u64_rem(tmp_mb, 1000, &kb);
770 edac_dbg(0, "CH#%d RIR#%d INTL#%d, offset %u.%03u GB (0x%016Lx), tgt: %d, reg=0x%08x\n",
773 ((u64)tmp_mb) << 20L,
774 (u32)RIR_RNK_TGT(reg),
781 struct mem_ctl_info *get_mci_for_node_id(u8 node_id)
783 struct sbridge_dev *sbridge_dev;
785 list_for_each_entry(sbridge_dev, &sbridge_edac_list, list) {
786 if (sbridge_dev->node_id == node_id)
787 return sbridge_dev->mci;
792 static int get_memory_error_data(struct mem_ctl_info *mci,
797 char **area_type, char *msg)
799 struct mem_ctl_info *new_mci;
800 struct sbridge_pvt *pvt = mci->pvt_info;
801 int n_rir, n_sads, n_tads, sad_way, sck_xch;
802 int sad_interl, idx, base_ch;
804 unsigned sad_interleave[MAX_INTERLEAVE];
810 u64 ch_addr, offset, limit, prv = 0;
814 * Step 0) Check if the address is at special memory ranges
815 * The check bellow is probably enough to fill all cases where
816 * the error is not inside a memory, except for the legacy
817 * range (e. g. VGA addresses). It is unlikely, however, that the
818 * memory controller would generate an error on that range.
820 if ((addr > (u64) pvt->tolm) && (addr < (1LL << 32))) {
821 sprintf(msg, "Error at TOLM area, on addr 0x%08Lx", addr);
824 if (addr >= (u64)pvt->tohm) {
825 sprintf(msg, "Error at MMIOH area, on addr 0x%016Lx", addr);
832 for (n_sads = 0; n_sads < MAX_SAD; n_sads++) {
833 pci_read_config_dword(pvt->pci_sad0, dram_rule[n_sads],
836 if (!DRAM_RULE_ENABLE(reg))
839 limit = SAD_LIMIT(reg);
841 sprintf(msg, "Can't discover the memory socket");
848 if (n_sads == MAX_SAD) {
849 sprintf(msg, "Can't discover the memory socket");
852 *area_type = get_dram_attr(reg);
853 interleave_mode = INTERLEAVE_MODE(reg);
855 pci_read_config_dword(pvt->pci_sad0, interleave_list[n_sads],
857 sad_interl = sad_pkg(reg, 0);
858 for (sad_way = 0; sad_way < 8; sad_way++) {
859 if (sad_way > 0 && sad_interl == sad_pkg(reg, sad_way))
861 sad_interleave[sad_way] = sad_pkg(reg, sad_way);
862 edac_dbg(0, "SAD interleave #%d: %d\n",
863 sad_way, sad_interleave[sad_way]);
865 edac_dbg(0, "mc#%d: Error detected on SAD#%d: address 0x%016Lx < 0x%016Lx, Interleave [%d:6]%s\n",
866 pvt->sbridge_dev->mc,
871 interleave_mode ? "" : "XOR[18:16]");
873 idx = ((addr >> 6) ^ (addr >> 16)) & 7;
875 idx = (addr >> 6) & 7;
889 sprintf(msg, "Can't discover socket interleave");
892 *socket = sad_interleave[idx];
893 edac_dbg(0, "SAD interleave index: %d (wayness %d) = CPU socket %d\n",
894 idx, sad_way, *socket);
897 * Move to the proper node structure, in order to access the
898 * right PCI registers
900 new_mci = get_mci_for_node_id(*socket);
902 sprintf(msg, "Struct for socket #%u wasn't initialized",
910 * Step 2) Get memory channel
913 for (n_tads = 0; n_tads < MAX_TAD; n_tads++) {
914 pci_read_config_dword(pvt->pci_ha0, tad_dram_rule[n_tads],
916 limit = TAD_LIMIT(reg);
918 sprintf(msg, "Can't discover the memory channel");
925 ch_way = TAD_CH(reg) + 1;
926 sck_way = TAD_SOCK(reg) + 1;
928 * FIXME: Is it right to always use channel 0 for offsets?
930 pci_read_config_dword(pvt->pci_tad[0],
931 tad_ch_nilv_offset[n_tads],
937 idx = addr >> (6 + sck_way);
941 * FIXME: Shouldn't we use CHN_IDX_OFFSET() here, when ch_way == 3 ???
945 base_ch = TAD_TGT0(reg);
948 base_ch = TAD_TGT1(reg);
951 base_ch = TAD_TGT2(reg);
954 base_ch = TAD_TGT3(reg);
957 sprintf(msg, "Can't discover the TAD target");
960 *channel_mask = 1 << base_ch;
962 if (pvt->is_mirrored) {
963 *channel_mask |= 1 << ((base_ch + 2) % 4);
967 sck_xch = 1 << sck_way * (ch_way >> 1);
970 sprintf(msg, "Invalid mirror set. Can't decode addr");
974 sck_xch = (1 << sck_way) * ch_way;
976 if (pvt->is_lockstep)
977 *channel_mask |= 1 << ((base_ch + 1) % 4);
979 offset = TAD_OFFSET(tad_offset);
981 edac_dbg(0, "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",
992 /* Calculate channel address */
993 /* Remove the TAD offset */
996 sprintf(msg, "Can't calculate ch addr: TAD offset 0x%08Lx is too high for addr 0x%08Lx!",
1001 /* Store the low bits [0:6] of the addr */
1002 ch_addr = addr & 0x7f;
1003 /* Remove socket wayness and remove 6 bits */
1005 addr = div_u64(addr, sck_xch);
1007 /* Divide by channel way */
1008 addr = addr / ch_way;
1010 /* Recover the last 6 bits */
1011 ch_addr |= addr << 6;
1014 * Step 3) Decode rank
1016 for (n_rir = 0; n_rir < MAX_RIR_RANGES; n_rir++) {
1017 pci_read_config_dword(pvt->pci_tad[base_ch],
1018 rir_way_limit[n_rir],
1021 if (!IS_RIR_VALID(reg))
1024 limit = RIR_LIMIT(reg);
1025 mb = div_u64_rem(limit >> 20, 1000, &kb);
1026 edac_dbg(0, "RIR#%d, limit: %u.%03u GB (0x%016Lx), way: %d\n",
1031 if (ch_addr <= limit)
1034 if (n_rir == MAX_RIR_RANGES) {
1035 sprintf(msg, "Can't discover the memory rank for ch addr 0x%08Lx",
1039 rir_way = RIR_WAY(reg);
1040 if (pvt->is_close_pg)
1041 idx = (ch_addr >> 6);
1043 idx = (ch_addr >> 13); /* FIXME: Datasheet says to shift by 15 */
1044 idx %= 1 << rir_way;
1046 pci_read_config_dword(pvt->pci_tad[base_ch],
1047 rir_offset[n_rir][idx],
1049 *rank = RIR_RNK_TGT(reg);
1051 edac_dbg(0, "RIR#%d: channel address 0x%08Lx < 0x%08Lx, RIR interleave %d, index %d\n",
1061 /****************************************************************************
1062 Device initialization routines: put/get, init/exit
1063 ****************************************************************************/
1066 * sbridge_put_all_devices 'put' all the devices that we have
1067 * reserved via 'get'
1069 static void sbridge_put_devices(struct sbridge_dev *sbridge_dev)
1074 for (i = 0; i < sbridge_dev->n_devs; i++) {
1075 struct pci_dev *pdev = sbridge_dev->pdev[i];
1078 edac_dbg(0, "Removing dev %02x:%02x.%d\n",
1080 PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn));
1085 static void sbridge_put_all_devices(void)
1087 struct sbridge_dev *sbridge_dev, *tmp;
1089 list_for_each_entry_safe(sbridge_dev, tmp, &sbridge_edac_list, list) {
1090 sbridge_put_devices(sbridge_dev);
1091 free_sbridge_dev(sbridge_dev);
1096 * sbridge_get_all_devices Find and perform 'get' operation on the MCH's
1097 * device/functions we want to reference for this driver
1099 * Need to 'get' device 16 func 1 and func 2
1101 static int sbridge_get_onedevice(struct pci_dev **prev,
1103 const struct pci_id_table *table,
1104 const unsigned devno)
1106 struct sbridge_dev *sbridge_dev;
1107 const struct pci_id_descr *dev_descr = &table->descr[devno];
1109 struct pci_dev *pdev = NULL;
1112 sbridge_printk(KERN_INFO,
1113 "Seeking for: dev %02x.%d PCI ID %04x:%04x\n",
1114 dev_descr->dev, dev_descr->func,
1115 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
1117 pdev = pci_get_device(PCI_VENDOR_ID_INTEL,
1118 dev_descr->dev_id, *prev);
1126 if (dev_descr->optional)
1132 sbridge_printk(KERN_INFO,
1133 "Device not found: dev %02x.%d PCI ID %04x:%04x\n",
1134 dev_descr->dev, dev_descr->func,
1135 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
1137 /* End of list, leave */
1140 bus = pdev->bus->number;
1142 sbridge_dev = get_sbridge_dev(bus);
1144 sbridge_dev = alloc_sbridge_dev(bus, table);
1152 if (sbridge_dev->pdev[devno]) {
1153 sbridge_printk(KERN_ERR,
1154 "Duplicated device for "
1155 "dev %02x:%d.%d PCI ID %04x:%04x\n",
1156 bus, dev_descr->dev, dev_descr->func,
1157 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
1162 sbridge_dev->pdev[devno] = pdev;
1165 if (unlikely(PCI_SLOT(pdev->devfn) != dev_descr->dev ||
1166 PCI_FUNC(pdev->devfn) != dev_descr->func)) {
1167 sbridge_printk(KERN_ERR,
1168 "Device PCI ID %04x:%04x "
1169 "has dev %02x:%d.%d instead of dev %02x:%02x.%d\n",
1170 PCI_VENDOR_ID_INTEL, dev_descr->dev_id,
1171 bus, PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn),
1172 bus, dev_descr->dev, dev_descr->func);
1176 /* Be sure that the device is enabled */
1177 if (unlikely(pci_enable_device(pdev) < 0)) {
1178 sbridge_printk(KERN_ERR,
1180 "dev %02x:%d.%d PCI ID %04x:%04x\n",
1181 bus, dev_descr->dev, dev_descr->func,
1182 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
1186 edac_dbg(0, "Detected dev %02x:%d.%d PCI ID %04x:%04x\n",
1187 bus, dev_descr->dev, dev_descr->func,
1188 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
1191 * As stated on drivers/pci/search.c, the reference count for
1192 * @from is always decremented if it is not %NULL. So, as we need
1193 * to get all devices up to null, we need to do a get for the device
1202 static int sbridge_get_all_devices(u8 *num_mc)
1205 struct pci_dev *pdev = NULL;
1206 const struct pci_id_table *table = pci_dev_descr_sbridge_table;
1208 while (table && table->descr) {
1209 for (i = 0; i < table->n_devs; i++) {
1212 rc = sbridge_get_onedevice(&pdev, num_mc,
1219 sbridge_put_all_devices();
1230 static int mci_bind_devs(struct mem_ctl_info *mci,
1231 struct sbridge_dev *sbridge_dev)
1233 struct sbridge_pvt *pvt = mci->pvt_info;
1234 struct pci_dev *pdev;
1237 for (i = 0; i < sbridge_dev->n_devs; i++) {
1238 pdev = sbridge_dev->pdev[i];
1241 slot = PCI_SLOT(pdev->devfn);
1242 func = PCI_FUNC(pdev->devfn);
1247 pvt->pci_sad0 = pdev;
1250 pvt->pci_sad1 = pdev;
1268 pvt->pci_ha0 = pdev;
1280 pvt->pci_ras = pdev;
1286 pvt->pci_tad[func - 2] = pdev;
1295 pvt->pci_ddrio = pdev;
1305 edac_dbg(0, "Associated PCI %02x.%02d.%d with dev = %p\n",
1307 PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn),
1311 /* Check if everything were registered */
1312 if (!pvt->pci_sad0 || !pvt->pci_sad1 || !pvt->pci_ha0 ||
1313 !pvt-> pci_tad || !pvt->pci_ras || !pvt->pci_ta)
1316 for (i = 0; i < NUM_CHANNELS; i++) {
1317 if (!pvt->pci_tad[i])
1323 sbridge_printk(KERN_ERR, "Some needed devices are missing\n");
1327 sbridge_printk(KERN_ERR, "Device %d, function %d "
1328 "is out of the expected range\n",
1333 /****************************************************************************
1334 Error check routines
1335 ****************************************************************************/
1338 * While Sandy Bridge has error count registers, SMI BIOS read values from
1339 * and resets the counters. So, they are not reliable for the OS to read
1340 * from them. So, we have no option but to just trust on whatever MCE is
1341 * telling us about the errors.
1343 static void sbridge_mce_output_error(struct mem_ctl_info *mci,
1344 const struct mce *m)
1346 struct mem_ctl_info *new_mci;
1347 struct sbridge_pvt *pvt = mci->pvt_info;
1348 enum hw_event_mc_err_type tp_event;
1349 char *type, *optype, msg[256];
1350 bool ripv = GET_BITFIELD(m->mcgstatus, 0, 0);
1351 bool overflow = GET_BITFIELD(m->status, 62, 62);
1352 bool uncorrected_error = GET_BITFIELD(m->status, 61, 61);
1353 bool recoverable = GET_BITFIELD(m->status, 56, 56);
1354 u32 core_err_cnt = GET_BITFIELD(m->status, 38, 52);
1355 u32 mscod = GET_BITFIELD(m->status, 16, 31);
1356 u32 errcode = GET_BITFIELD(m->status, 0, 15);
1357 u32 channel = GET_BITFIELD(m->status, 0, 3);
1358 u32 optypenum = GET_BITFIELD(m->status, 4, 6);
1359 long channel_mask, first_channel;
1362 char *area_type = NULL;
1364 if (uncorrected_error) {
1367 tp_event = HW_EVENT_ERR_FATAL;
1370 tp_event = HW_EVENT_ERR_UNCORRECTED;
1374 tp_event = HW_EVENT_ERR_CORRECTED;
1378 * According with Table 15-9 of the Intel Architecture spec vol 3A,
1379 * memory errors should fit in this mask:
1380 * 000f 0000 1mmm cccc (binary)
1382 * f = Correction Report Filtering Bit. If 1, subsequent errors
1386 * If the mask doesn't match, report an error to the parsing logic
1388 if (! ((errcode & 0xef80) == 0x80)) {
1389 optype = "Can't parse: it is not a mem";
1391 switch (optypenum) {
1393 optype = "generic undef request error";
1396 optype = "memory read error";
1399 optype = "memory write error";
1402 optype = "addr/cmd error";
1405 optype = "memory scrubbing error";
1408 optype = "reserved";
1413 rc = get_memory_error_data(mci, m->addr, &socket,
1414 &channel_mask, &rank, &area_type, msg);
1417 new_mci = get_mci_for_node_id(socket);
1419 strcpy(msg, "Error: socket got corrupted!");
1423 pvt = mci->pvt_info;
1425 first_channel = find_first_bit(&channel_mask, NUM_CHANNELS);
1436 * FIXME: On some memory configurations (mirror, lockstep), the
1437 * Memory Controller can't point the error to a single DIMM. The
1438 * EDAC core should be handling the channel mask, in order to point
1439 * to the group of dimm's where the error may be happening.
1441 snprintf(msg, sizeof(msg),
1442 "%s%s area:%s err_code:%04x:%04x socket:%d channel_mask:%ld rank:%d",
1443 overflow ? " OVERFLOW" : "",
1444 (uncorrected_error && recoverable) ? " recoverable" : "",
1451 edac_dbg(0, "%s\n", msg);
1453 /* FIXME: need support for channel mask */
1455 if (channel == CHANNEL_UNSPECIFIED)
1458 /* Call the helper to output message */
1459 edac_mc_handle_error(tp_event, mci, core_err_cnt,
1460 m->addr >> PAGE_SHIFT, m->addr & ~PAGE_MASK, 0,
1465 edac_mc_handle_error(tp_event, mci, core_err_cnt, 0, 0, 0,
1472 * sbridge_check_error Retrieve and process errors reported by the
1473 * hardware. Called by the Core module.
1475 static void sbridge_check_error(struct mem_ctl_info *mci)
1477 struct sbridge_pvt *pvt = mci->pvt_info;
1483 * MCE first step: Copy all mce errors into a temporary buffer
1484 * We use a double buffering here, to reduce the risk of
1488 count = (pvt->mce_out + MCE_LOG_LEN - pvt->mce_in)
1493 m = pvt->mce_outentry;
1494 if (pvt->mce_in + count > MCE_LOG_LEN) {
1495 unsigned l = MCE_LOG_LEN - pvt->mce_in;
1497 memcpy(m, &pvt->mce_entry[pvt->mce_in], sizeof(*m) * l);
1503 memcpy(m, &pvt->mce_entry[pvt->mce_in], sizeof(*m) * count);
1505 pvt->mce_in += count;
1508 if (pvt->mce_overrun) {
1509 sbridge_printk(KERN_ERR, "Lost %d memory errors\n",
1512 pvt->mce_overrun = 0;
1516 * MCE second step: parse errors and display
1518 for (i = 0; i < count; i++)
1519 sbridge_mce_output_error(mci, &pvt->mce_outentry[i]);
1523 * sbridge_mce_check_error Replicates mcelog routine to get errors
1524 * This routine simply queues mcelog errors, and
1525 * return. The error itself should be handled later
1526 * by sbridge_check_error.
1527 * WARNING: As this routine should be called at NMI time, extra care should
1528 * be taken to avoid deadlocks, and to be as fast as possible.
1530 static int sbridge_mce_check_error(struct notifier_block *nb, unsigned long val,
1533 struct mce *mce = (struct mce *)data;
1534 struct mem_ctl_info *mci;
1535 struct sbridge_pvt *pvt;
1537 mci = get_mci_for_node_id(mce->socketid);
1540 pvt = mci->pvt_info;
1543 * Just let mcelog handle it if the error is
1544 * outside the memory controller. A memory error
1545 * is indicated by bit 7 = 1 and bits = 8-11,13-15 = 0.
1546 * bit 12 has an special meaning.
1548 if ((mce->status & 0xefff) >> 7 != 1)
1551 printk("sbridge: HANDLING MCE MEMORY ERROR\n");
1553 printk("CPU %d: Machine Check Exception: %Lx Bank %d: %016Lx\n",
1554 mce->extcpu, mce->mcgstatus, mce->bank, mce->status);
1555 printk("TSC %llx ", mce->tsc);
1556 printk("ADDR %llx ", mce->addr);
1557 printk("MISC %llx ", mce->misc);
1559 printk("PROCESSOR %u:%x TIME %llu SOCKET %u APIC %x\n",
1560 mce->cpuvendor, mce->cpuid, mce->time,
1561 mce->socketid, mce->apicid);
1563 /* Only handle if it is the right mc controller */
1564 if (cpu_data(mce->cpu).phys_proc_id != pvt->sbridge_dev->mc)
1568 if ((pvt->mce_out + 1) % MCE_LOG_LEN == pvt->mce_in) {
1574 /* Copy memory error at the ringbuffer */
1575 memcpy(&pvt->mce_entry[pvt->mce_out], mce, sizeof(*mce));
1577 pvt->mce_out = (pvt->mce_out + 1) % MCE_LOG_LEN;
1579 /* Handle fatal errors immediately */
1580 if (mce->mcgstatus & 1)
1581 sbridge_check_error(mci);
1583 /* Advice mcelog that the error were handled */
1587 static struct notifier_block sbridge_mce_dec = {
1588 .notifier_call = sbridge_mce_check_error,
1591 /****************************************************************************
1592 EDAC register/unregister logic
1593 ****************************************************************************/
1595 static void sbridge_unregister_mci(struct sbridge_dev *sbridge_dev)
1597 struct mem_ctl_info *mci = sbridge_dev->mci;
1598 struct sbridge_pvt *pvt;
1600 if (unlikely(!mci || !mci->pvt_info)) {
1601 edac_dbg(0, "MC: dev = %p\n", &sbridge_dev->pdev[0]->dev);
1603 sbridge_printk(KERN_ERR, "Couldn't find mci handler\n");
1607 pvt = mci->pvt_info;
1609 edac_dbg(0, "MC: mci = %p, dev = %p\n",
1610 mci, &sbridge_dev->pdev[0]->dev);
1612 /* Remove MC sysfs nodes */
1613 edac_mc_del_mc(mci->pdev);
1615 edac_dbg(1, "%s: free mci struct\n", mci->ctl_name);
1616 kfree(mci->ctl_name);
1618 sbridge_dev->mci = NULL;
1621 static int sbridge_register_mci(struct sbridge_dev *sbridge_dev)
1623 struct mem_ctl_info *mci;
1624 struct edac_mc_layer layers[2];
1625 struct sbridge_pvt *pvt;
1628 /* Check the number of active and not disabled channels */
1629 rc = check_if_ecc_is_active(sbridge_dev->bus);
1630 if (unlikely(rc < 0))
1633 /* allocate a new MC control structure */
1634 layers[0].type = EDAC_MC_LAYER_CHANNEL;
1635 layers[0].size = NUM_CHANNELS;
1636 layers[0].is_virt_csrow = false;
1637 layers[1].type = EDAC_MC_LAYER_SLOT;
1638 layers[1].size = MAX_DIMMS;
1639 layers[1].is_virt_csrow = true;
1640 mci = edac_mc_alloc(sbridge_dev->mc, ARRAY_SIZE(layers), layers,
1646 edac_dbg(0, "MC: mci = %p, dev = %p\n",
1647 mci, &sbridge_dev->pdev[0]->dev);
1649 pvt = mci->pvt_info;
1650 memset(pvt, 0, sizeof(*pvt));
1652 /* Associate sbridge_dev and mci for future usage */
1653 pvt->sbridge_dev = sbridge_dev;
1654 sbridge_dev->mci = mci;
1656 mci->mtype_cap = MEM_FLAG_DDR3;
1657 mci->edac_ctl_cap = EDAC_FLAG_NONE;
1658 mci->edac_cap = EDAC_FLAG_NONE;
1659 mci->mod_name = "sbridge_edac.c";
1660 mci->mod_ver = SBRIDGE_REVISION;
1661 mci->ctl_name = kasprintf(GFP_KERNEL, "Sandy Bridge Socket#%d", mci->mc_idx);
1662 mci->dev_name = pci_name(sbridge_dev->pdev[0]);
1663 mci->ctl_page_to_phys = NULL;
1665 /* Set the function pointer to an actual operation function */
1666 mci->edac_check = sbridge_check_error;
1668 /* Store pci devices at mci for faster access */
1669 rc = mci_bind_devs(mci, sbridge_dev);
1670 if (unlikely(rc < 0))
1673 /* Get dimm basic config and the memory layout */
1674 get_dimm_config(mci);
1675 get_memory_layout(mci);
1677 /* record ptr to the generic device */
1678 mci->pdev = &sbridge_dev->pdev[0]->dev;
1680 /* add this new MC control structure to EDAC's list of MCs */
1681 if (unlikely(edac_mc_add_mc(mci))) {
1682 edac_dbg(0, "MC: failed edac_mc_add_mc()\n");
1690 kfree(mci->ctl_name);
1692 sbridge_dev->mci = NULL;
1697 * sbridge_probe Probe for ONE instance of device to see if it is
1700 * 0 for FOUND a device
1701 * < 0 for error code
1704 static int sbridge_probe(struct pci_dev *pdev, const struct pci_device_id *id)
1708 struct sbridge_dev *sbridge_dev;
1710 /* get the pci devices we want to reserve for our use */
1711 mutex_lock(&sbridge_edac_lock);
1714 * All memory controllers are allocated at the first pass.
1716 if (unlikely(probed >= 1)) {
1717 mutex_unlock(&sbridge_edac_lock);
1722 rc = sbridge_get_all_devices(&num_mc);
1723 if (unlikely(rc < 0))
1727 list_for_each_entry(sbridge_dev, &sbridge_edac_list, list) {
1728 edac_dbg(0, "Registering MC#%d (%d of %d)\n",
1729 mc, mc + 1, num_mc);
1730 sbridge_dev->mc = mc++;
1731 rc = sbridge_register_mci(sbridge_dev);
1732 if (unlikely(rc < 0))
1736 sbridge_printk(KERN_INFO, "Driver loaded.\n");
1738 mutex_unlock(&sbridge_edac_lock);
1742 list_for_each_entry(sbridge_dev, &sbridge_edac_list, list)
1743 sbridge_unregister_mci(sbridge_dev);
1745 sbridge_put_all_devices();
1747 mutex_unlock(&sbridge_edac_lock);
1752 * sbridge_remove destructor for one instance of device
1755 static void sbridge_remove(struct pci_dev *pdev)
1757 struct sbridge_dev *sbridge_dev;
1762 * we have a trouble here: pdev value for removal will be wrong, since
1763 * it will point to the X58 register used to detect that the machine
1764 * is a Nehalem or upper design. However, due to the way several PCI
1765 * devices are grouped together to provide MC functionality, we need
1766 * to use a different method for releasing the devices
1769 mutex_lock(&sbridge_edac_lock);
1771 if (unlikely(!probed)) {
1772 mutex_unlock(&sbridge_edac_lock);
1776 list_for_each_entry(sbridge_dev, &sbridge_edac_list, list)
1777 sbridge_unregister_mci(sbridge_dev);
1779 /* Release PCI resources */
1780 sbridge_put_all_devices();
1784 mutex_unlock(&sbridge_edac_lock);
1787 MODULE_DEVICE_TABLE(pci, sbridge_pci_tbl);
1790 * sbridge_driver pci_driver structure for this module
1793 static struct pci_driver sbridge_driver = {
1794 .name = "sbridge_edac",
1795 .probe = sbridge_probe,
1796 .remove = sbridge_remove,
1797 .id_table = sbridge_pci_tbl,
1801 * sbridge_init Module entry function
1802 * Try to initialize this module for its devices
1804 static int __init sbridge_init(void)
1810 /* Ensure that the OPSTATE is set correctly for POLL or NMI */
1813 pci_rc = pci_register_driver(&sbridge_driver);
1816 mce_register_decode_chain(&sbridge_mce_dec);
1820 sbridge_printk(KERN_ERR, "Failed to register device with error %d.\n",
1827 * sbridge_exit() Module exit function
1828 * Unregister the driver
1830 static void __exit sbridge_exit(void)
1833 pci_unregister_driver(&sbridge_driver);
1834 mce_unregister_decode_chain(&sbridge_mce_dec);
1837 module_init(sbridge_init);
1838 module_exit(sbridge_exit);
1840 module_param(edac_op_state, int, 0444);
1841 MODULE_PARM_DESC(edac_op_state, "EDAC Error Reporting state: 0=Poll,1=NMI");
1843 MODULE_LICENSE("GPL");
1844 MODULE_AUTHOR("Mauro Carvalho Chehab <mchehab@redhat.com>");
1845 MODULE_AUTHOR("Red Hat Inc. (http://www.redhat.com)");
1846 MODULE_DESCRIPTION("MC Driver for Intel Sandy Bridge memory controllers - "