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/edac_mce.h>
22 #include <linux/smp.h>
23 #include <linux/bitmap.h>
24 #include <asm/processor.h>
26 #include "edac_core.h"
29 static LIST_HEAD(sbridge_edac_list);
30 static DEFINE_MUTEX(sbridge_edac_lock);
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"
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)) {
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) {
129 return SAD_PKG0(reg);
131 return SAD_PKG1(reg);
133 return SAD_PKG2(reg);
135 return SAD_PKG3(reg);
137 return SAD_PKG4(reg);
139 return SAD_PKG5(reg);
141 return SAD_PKG6(reg);
143 return SAD_PKG7(reg);
149 /* Devices 12 Function 7 */
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 */
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[] = {
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)
272 #define NUM_CHANNELS 4
273 #define MAX_DIMMS 3 /* Max DIMMS per channel */
275 struct sbridge_info {
279 struct sbridge_channel {
284 struct pci_id_descr {
291 struct pci_id_table {
292 const struct pci_id_descr *descr;
297 struct list_head list;
299 u8 node_id, source_id;
300 struct pci_dev **pdev;
302 struct mem_ctl_info *mci;
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;
322 struct edac_mce edac_mce;
324 /* Fifo double buffers */
325 struct mce mce_entry[MCE_LOG_LEN];
326 struct mce mce_outentry[MCE_LOG_LEN];
328 /* Fifo in/out counters */
329 unsigned mce_in, mce_out;
331 /* Count indicator to show errors not got */
332 unsigned mce_overrun;
334 /* Memory description */
338 #define PCI_DESCR(device, function, device_id) \
340 .func = (function), \
341 .dev_id = (device_id)
343 static const struct pci_id_descr pci_dev_descr_sbridge[] = {
344 /* Processor Home Agent */
345 { PCI_DESCR(14, 0, PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_HA0) },
347 /* Memory controller */
348 { PCI_DESCR(15, 0, PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TA) },
349 { PCI_DESCR(15, 1, PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_RAS) },
350 { PCI_DESCR(15, 2, PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD0) },
351 { PCI_DESCR(15, 3, PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD1) },
352 { PCI_DESCR(15, 4, PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD2) },
353 { PCI_DESCR(15, 5, PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD3) },
354 { PCI_DESCR(17, 0, PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_DDRIO) },
356 /* System Address Decoder */
357 { PCI_DESCR(12, 6, PCI_DEVICE_ID_INTEL_SBRIDGE_SAD0) },
358 { PCI_DESCR(12, 7, PCI_DEVICE_ID_INTEL_SBRIDGE_SAD1) },
360 /* Broadcast Registers */
361 { PCI_DESCR(13, 6, PCI_DEVICE_ID_INTEL_SBRIDGE_BR) },
364 #define PCI_ID_TABLE_ENTRY(A) { .descr=A, .n_devs = ARRAY_SIZE(A) }
365 static const struct pci_id_table pci_dev_descr_sbridge_table[] = {
366 PCI_ID_TABLE_ENTRY(pci_dev_descr_sbridge),
367 {0,} /* 0 terminated list. */
371 * pci_device_id table for which devices we are looking for
373 static const struct pci_device_id sbridge_pci_tbl[] __devinitdata = {
374 {PCI_DEVICE(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TA)},
375 {0,} /* 0 terminated list. */
379 /****************************************************************************
380 Anciliary status routines
381 ****************************************************************************/
383 static inline int numrank(u32 mtr)
385 int ranks = (1 << RANK_CNT_BITS(mtr));
388 debugf0("Invalid number of ranks: %d (max = 4) raw value = %x (%04x)",
389 ranks, (unsigned int)RANK_CNT_BITS(mtr), mtr);
396 static inline int numrow(u32 mtr)
398 int rows = (RANK_WIDTH_BITS(mtr) + 12);
400 if (rows < 13 || rows > 18) {
401 debugf0("Invalid number of rows: %d (should be between 14 and 17) raw value = %x (%04x)",
402 rows, (unsigned int)RANK_WIDTH_BITS(mtr), mtr);
409 static inline int numcol(u32 mtr)
411 int cols = (COL_WIDTH_BITS(mtr) + 10);
414 debugf0("Invalid number of cols: %d (max = 4) raw value = %x (%04x)",
415 cols, (unsigned int)COL_WIDTH_BITS(mtr), mtr);
422 static struct sbridge_dev *get_sbridge_dev(u8 bus)
424 struct sbridge_dev *sbridge_dev;
426 list_for_each_entry(sbridge_dev, &sbridge_edac_list, list) {
427 if (sbridge_dev->bus == bus)
434 static struct sbridge_dev *alloc_sbridge_dev(u8 bus,
435 const struct pci_id_table *table)
437 struct sbridge_dev *sbridge_dev;
439 sbridge_dev = kzalloc(sizeof(*sbridge_dev), GFP_KERNEL);
443 sbridge_dev->pdev = kzalloc(sizeof(*sbridge_dev->pdev) * table->n_devs,
445 if (!sbridge_dev->pdev) {
450 sbridge_dev->bus = bus;
451 sbridge_dev->n_devs = table->n_devs;
452 list_add_tail(&sbridge_dev->list, &sbridge_edac_list);
457 static void free_sbridge_dev(struct sbridge_dev *sbridge_dev)
459 list_del(&sbridge_dev->list);
460 kfree(sbridge_dev->pdev);
464 /****************************************************************************
465 Memory check routines
466 ****************************************************************************/
467 static struct pci_dev *get_pdev_slot_func(u8 bus, unsigned slot,
470 struct sbridge_dev *sbridge_dev = get_sbridge_dev(bus);
476 for (i = 0; i < sbridge_dev->n_devs; i++) {
477 if (!sbridge_dev->pdev[i])
480 if (PCI_SLOT(sbridge_dev->pdev[i]->devfn) == slot &&
481 PCI_FUNC(sbridge_dev->pdev[i]->devfn) == func) {
482 debugf1("Associated %02x.%02x.%d with %p\n",
483 bus, slot, func, sbridge_dev->pdev[i]);
484 return sbridge_dev->pdev[i];
492 * sbridge_get_active_channels() - gets the number of channels and csrows
494 * @channels: Number of channels that will be returned
495 * @csrows: Number of csrows found
497 * Since EDAC core needs to know in advance the number of available channels
498 * and csrows, in order to allocate memory for csrows/channels, it is needed
499 * to run two similar steps. At the first step, implemented on this function,
500 * it checks the number of csrows/channels present at one socket, identified
501 * by the associated PCI bus.
502 * this is used in order to properly allocate the size of mci components.
503 * Note: one csrow is one dimm.
505 static int sbridge_get_active_channels(const u8 bus, unsigned *channels,
508 struct pci_dev *pdev = NULL;
515 pdev = get_pdev_slot_func(bus, 15, 0);
517 sbridge_printk(KERN_ERR, "Couldn't find PCI device "
523 pci_read_config_dword(pdev, MCMTR, &mcmtr);
524 if (!IS_ECC_ENABLED(mcmtr)) {
525 sbridge_printk(KERN_ERR, "ECC is disabled. Aborting\n");
529 for (i = 0; i < NUM_CHANNELS; i++) {
532 /* Device 15 functions 2 - 5 */
533 pdev = get_pdev_slot_func(bus, 15, 2 + i);
535 sbridge_printk(KERN_ERR, "Couldn't find PCI device "
542 for (j = 0; j < ARRAY_SIZE(mtr_regs); j++) {
543 pci_read_config_dword(pdev, mtr_regs[j], &mtr);
544 debugf1("Bus#%02x channel #%d MTR%d = %x\n", bus, i, j, mtr);
545 if (IS_DIMM_PRESENT(mtr))
550 debugf0("Number of active channels: %d, number of active dimms: %d\n",
556 static int get_dimm_config(const struct mem_ctl_info *mci)
558 struct sbridge_pvt *pvt = mci->pvt_info;
559 struct csrow_info *csr;
560 int i, j, banks, ranks, rows, cols, size, npages;
562 unsigned long last_page = 0;
566 pci_read_config_dword(pvt->pci_br, SAD_TARGET, ®);
567 pvt->sbridge_dev->source_id = SOURCE_ID(reg);
569 pci_read_config_dword(pvt->pci_br, SAD_CONTROL, ®);
570 pvt->sbridge_dev->node_id = NODE_ID(reg);
571 debugf0("mc#%d: Node ID: %d, source ID: %d\n",
572 pvt->sbridge_dev->mc,
573 pvt->sbridge_dev->node_id,
574 pvt->sbridge_dev->source_id);
576 pci_read_config_dword(pvt->pci_ras, RASENABLES, ®);
577 if (IS_MIRROR_ENABLED(reg)) {
578 debugf0("Memory mirror is enabled\n");
579 pvt->is_mirrored = true;
581 debugf0("Memory mirror is disabled\n");
582 pvt->is_mirrored = false;
585 pci_read_config_dword(pvt->pci_ta, MCMTR, &pvt->info.mcmtr);
586 if (IS_LOCKSTEP_ENABLED(pvt->info.mcmtr)) {
587 debugf0("Lockstep is enabled\n");
588 mode = EDAC_S8ECD8ED;
589 pvt->is_lockstep = true;
591 debugf0("Lockstep is disabled\n");
592 mode = EDAC_S4ECD4ED;
593 pvt->is_lockstep = false;
595 if (IS_CLOSE_PG(pvt->info.mcmtr)) {
596 debugf0("address map is on closed page mode\n");
597 pvt->is_close_pg = true;
599 debugf0("address map is on open page mode\n");
600 pvt->is_close_pg = false;
603 pci_read_config_dword(pvt->pci_ta, RANK_CFG_A, ®);
604 if (IS_RDIMM_ENABLED(reg)) {
605 /* FIXME: Can also be LRDIMM */
606 debugf0("Memory is registered\n");
609 debugf0("Memory is unregistered\n");
613 /* On all supported DDR3 DIMM types, there are 8 banks available */
616 for (i = 0; i < NUM_CHANNELS; i++) {
619 for (j = 0; j < ARRAY_SIZE(mtr_regs); j++) {
620 pci_read_config_dword(pvt->pci_tad[i],
622 debugf4("Channel #%d MTR%d = %x\n", i, j, mtr);
623 if (IS_DIMM_PRESENT(mtr)) {
624 pvt->channel[i].dimms++;
626 ranks = numrank(mtr);
630 /* DDR3 has 8 I/O banks */
631 size = (rows * cols * banks * ranks) >> (20 - 3);
632 npages = MiB_TO_PAGES(size);
634 debugf0("mc#%d: channel %d, dimm %d, %d Mb (%d pages) bank: %d, rank: %d, row: %#x, col: %#x\n",
635 pvt->sbridge_dev->mc, i, j,
637 banks, ranks, rows, cols);
638 csr = &mci->csrows[csrow];
640 csr->first_page = last_page;
641 csr->last_page = last_page + npages - 1;
642 csr->page_mask = 0UL; /* Unused */
643 csr->nr_pages = npages;
645 csr->csrow_idx = csrow;
646 csr->dtype = (banks == 8) ? DEV_X8 : DEV_X4;
650 csr->edac_mode = mode;
651 csr->nr_channels = 1;
652 csr->channels[0].chan_idx = i;
653 csr->channels[0].ce_count = 0;
654 pvt->csrow_map[i][j] = csrow;
655 snprintf(csr->channels[0].label,
656 sizeof(csr->channels[0].label),
657 "CPU_SrcID#%u_Channel#%u_DIMM#%u",
658 pvt->sbridge_dev->source_id, i, j);
668 static void get_memory_layout(const struct mem_ctl_info *mci)
670 struct sbridge_pvt *pvt = mci->pvt_info;
671 int i, j, k, n_sads, n_tads, sad_interl;
678 * Step 1) Get TOLM/TOHM ranges
681 /* Address range is 32:28 */
682 pci_read_config_dword(pvt->pci_sad1, TOLM,
684 pvt->tolm = GET_TOLM(reg);
685 tmp_mb = (1 + pvt->tolm) >> 20;
687 debugf0("TOLM: %Lu.%03Lu GB (0x%016Lx)\n",
688 tmp_mb / 1000, tmp_mb % 1000, (u64)pvt->tolm);
690 /* Address range is already 45:25 */
691 pci_read_config_dword(pvt->pci_sad1, TOHM,
693 pvt->tohm = GET_TOHM(reg);
694 tmp_mb = (1 + pvt->tohm) >> 20;
696 debugf0("TOHM: %Lu.%03Lu GB (0x%016Lx)",
697 tmp_mb / 1000, tmp_mb % 1000, (u64)pvt->tohm);
700 * Step 2) Get SAD range and SAD Interleave list
701 * TAD registers contain the interleave wayness. However, it
702 * seems simpler to just discover it indirectly, with the
706 for (n_sads = 0; n_sads < MAX_SAD; n_sads++) {
707 /* SAD_LIMIT Address range is 45:26 */
708 pci_read_config_dword(pvt->pci_sad0, dram_rule[n_sads],
710 limit = SAD_LIMIT(reg);
712 if (!DRAM_RULE_ENABLE(reg))
718 tmp_mb = (limit + 1) >> 20;
719 debugf0("SAD#%d %s up to %Lu.%03Lu GB (0x%016Lx) %s reg=0x%08x\n",
722 tmp_mb / 1000, tmp_mb % 1000,
723 ((u64)tmp_mb) << 20L,
724 INTERLEAVE_MODE(reg) ? "Interleave: 8:6" : "Interleave: [8:6]XOR[18:16]",
728 pci_read_config_dword(pvt->pci_sad0, interleave_list[n_sads],
730 sad_interl = sad_pkg(reg, 0);
731 for (j = 0; j < 8; j++) {
732 if (j > 0 && sad_interl == sad_pkg(reg, j))
735 debugf0("SAD#%d, interleave #%d: %d\n",
736 n_sads, j, sad_pkg(reg, j));
741 * Step 3) Get TAD range
744 for (n_tads = 0; n_tads < MAX_TAD; n_tads++) {
745 pci_read_config_dword(pvt->pci_ha0, tad_dram_rule[n_tads],
747 limit = TAD_LIMIT(reg);
750 tmp_mb = (limit + 1) >> 20;
752 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",
753 n_tads, tmp_mb / 1000, tmp_mb % 1000,
754 ((u64)tmp_mb) << 20L,
766 * Step 4) Get TAD offsets, per each channel
768 for (i = 0; i < NUM_CHANNELS; i++) {
769 if (!pvt->channel[i].dimms)
771 for (j = 0; j < n_tads; j++) {
772 pci_read_config_dword(pvt->pci_tad[i],
773 tad_ch_nilv_offset[j],
775 tmp_mb = TAD_OFFSET(reg) >> 20;
776 debugf0("TAD CH#%d, offset #%d: %Lu.%03Lu GB (0x%016Lx), reg=0x%08x\n",
778 tmp_mb / 1000, tmp_mb % 1000,
779 ((u64)tmp_mb) << 20L,
785 * Step 6) Get RIR Wayness/Limit, per each channel
787 for (i = 0; i < NUM_CHANNELS; i++) {
788 if (!pvt->channel[i].dimms)
790 for (j = 0; j < MAX_RIR_RANGES; j++) {
791 pci_read_config_dword(pvt->pci_tad[i],
795 if (!IS_RIR_VALID(reg))
798 tmp_mb = RIR_LIMIT(reg) >> 20;
799 rir_way = 1 << RIR_WAY(reg);
800 debugf0("CH#%d RIR#%d, limit: %Lu.%03Lu GB (0x%016Lx), way: %d, reg=0x%08x\n",
802 tmp_mb / 1000, tmp_mb % 1000,
803 ((u64)tmp_mb) << 20L,
807 for (k = 0; k < rir_way; k++) {
808 pci_read_config_dword(pvt->pci_tad[i],
811 tmp_mb = RIR_OFFSET(reg) << 6;
813 debugf0("CH#%d RIR#%d INTL#%d, offset %Lu.%03Lu GB (0x%016Lx), tgt: %d, reg=0x%08x\n",
815 tmp_mb / 1000, tmp_mb % 1000,
816 ((u64)tmp_mb) << 20L,
817 (u32)RIR_RNK_TGT(reg),
824 struct mem_ctl_info *get_mci_for_node_id(u8 node_id)
826 struct sbridge_dev *sbridge_dev;
828 list_for_each_entry(sbridge_dev, &sbridge_edac_list, list) {
829 if (sbridge_dev->node_id == node_id)
830 return sbridge_dev->mci;
835 static int get_memory_error_data(struct mem_ctl_info *mci,
842 struct mem_ctl_info *new_mci;
843 struct sbridge_pvt *pvt = mci->pvt_info;
845 int n_rir, n_sads, n_tads, sad_way, sck_xch;
846 int sad_interl, idx, base_ch;
848 unsigned sad_interleave[MAX_INTERLEAVE];
853 u64 ch_addr, offset, limit, prv = 0;
857 * Step 0) Check if the address is at special memory ranges
858 * The check bellow is probably enough to fill all cases where
859 * the error is not inside a memory, except for the legacy
860 * range (e. g. VGA addresses). It is unlikely, however, that the
861 * memory controller would generate an error on that range.
863 if ((addr > (u64) pvt->tolm) && (addr < (1L << 32))) {
864 sprintf(msg, "Error at TOLM area, on addr 0x%08Lx", addr);
865 edac_mc_handle_ce_no_info(mci, msg);
868 if (addr >= (u64)pvt->tohm) {
869 sprintf(msg, "Error at MMIOH area, on addr 0x%016Lx", addr);
870 edac_mc_handle_ce_no_info(mci, msg);
877 for (n_sads = 0; n_sads < MAX_SAD; n_sads++) {
878 pci_read_config_dword(pvt->pci_sad0, dram_rule[n_sads],
881 if (!DRAM_RULE_ENABLE(reg))
884 limit = SAD_LIMIT(reg);
886 sprintf(msg, "Can't discover the memory socket");
887 edac_mc_handle_ce_no_info(mci, msg);
894 if (n_sads == MAX_SAD) {
895 sprintf(msg, "Can't discover the memory socket");
896 edac_mc_handle_ce_no_info(mci, msg);
899 area_type = get_dram_attr(reg);
900 interleave_mode = INTERLEAVE_MODE(reg);
902 pci_read_config_dword(pvt->pci_sad0, interleave_list[n_sads],
904 sad_interl = sad_pkg(reg, 0);
905 for (sad_way = 0; sad_way < 8; sad_way++) {
906 if (sad_way > 0 && sad_interl == sad_pkg(reg, sad_way))
908 sad_interleave[sad_way] = sad_pkg(reg, sad_way);
909 debugf0("SAD interleave #%d: %d\n",
910 sad_way, sad_interleave[sad_way]);
912 debugf0("mc#%d: Error detected on SAD#%d: address 0x%016Lx < 0x%016Lx, Interleave [%d:6]%s\n",
913 pvt->sbridge_dev->mc,
918 INTERLEAVE_MODE(reg) ? "" : "XOR[18:16]");
920 idx = ((addr >> 6) ^ (addr >> 16)) & 7;
922 idx = (addr >> 6) & 7;
936 sprintf(msg, "Can't discover socket interleave");
937 edac_mc_handle_ce_no_info(mci, msg);
940 *socket = sad_interleave[idx];
941 debugf0("SAD interleave index: %d (wayness %d) = CPU socket %d\n",
942 idx, sad_way, *socket);
945 * Move to the proper node structure, in order to access the
946 * right PCI registers
948 new_mci = get_mci_for_node_id(*socket);
950 sprintf(msg, "Struct for socket #%u wasn't initialized",
952 edac_mc_handle_ce_no_info(mci, msg);
959 * Step 2) Get memory channel
962 for (n_tads = 0; n_tads < MAX_TAD; n_tads++) {
963 pci_read_config_dword(pvt->pci_ha0, tad_dram_rule[n_tads],
965 limit = TAD_LIMIT(reg);
967 sprintf(msg, "Can't discover the memory channel");
968 edac_mc_handle_ce_no_info(mci, msg);
975 ch_way = TAD_CH(reg) + 1;
976 sck_way = TAD_SOCK(reg) + 1;
978 * FIXME: Is it right to always use channel 0 for offsets?
980 pci_read_config_dword(pvt->pci_tad[0],
981 tad_ch_nilv_offset[n_tads],
987 idx = addr >> (6 + sck_way);
991 * FIXME: Shouldn't we use CHN_IDX_OFFSET() here, when ch_way == 3 ???
995 base_ch = TAD_TGT0(reg);
998 base_ch = TAD_TGT1(reg);
1001 base_ch = TAD_TGT2(reg);
1004 base_ch = TAD_TGT3(reg);
1007 sprintf(msg, "Can't discover the TAD target");
1008 edac_mc_handle_ce_no_info(mci, msg);
1011 *channel_mask = 1 << base_ch;
1013 if (pvt->is_mirrored) {
1014 *channel_mask |= 1 << ((base_ch + 2) % 4);
1018 sck_xch = 1 << sck_way * (ch_way >> 1);
1021 sprintf(msg, "Invalid mirror set. Can't decode addr");
1022 edac_mc_handle_ce_no_info(mci, msg);
1026 sck_xch = (1 << sck_way) * ch_way;
1028 if (pvt->is_lockstep)
1029 *channel_mask |= 1 << ((base_ch + 1) % 4);
1031 offset = TAD_OFFSET(tad_offset);
1033 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",
1044 /* Calculate channel address */
1045 /* Remove the TAD offset */
1047 if (offset > addr) {
1048 sprintf(msg, "Can't calculate ch addr: TAD offset 0x%08Lx is too high for addr 0x%08Lx!",
1050 edac_mc_handle_ce_no_info(mci, msg);
1054 /* Store the low bits [0:6] of the addr */
1055 ch_addr = addr & 0x7f;
1056 /* Remove socket wayness and remove 6 bits */
1060 /* Divide by channel way */
1061 addr = addr / ch_way;
1063 /* Recover the last 6 bits */
1064 ch_addr |= addr << 6;
1067 * Step 3) Decode rank
1069 for (n_rir = 0; n_rir < MAX_RIR_RANGES; n_rir++) {
1070 pci_read_config_dword(pvt->pci_tad[base_ch],
1071 rir_way_limit[n_rir],
1074 if (!IS_RIR_VALID(reg))
1077 limit = RIR_LIMIT(reg);
1079 debugf0("RIR#%d, limit: %Lu.%03Lu GB (0x%016Lx), way: %d\n",
1081 (limit >> 20) / 1000, (limit >> 20) % 1000,
1084 if (ch_addr <= limit)
1087 if (n_rir == MAX_RIR_RANGES) {
1088 sprintf(msg, "Can't discover the memory rank for ch addr 0x%08Lx",
1090 edac_mc_handle_ce_no_info(mci, msg);
1093 rir_way = RIR_WAY(reg);
1094 if (pvt->is_close_pg)
1095 idx = (ch_addr >> 6);
1097 idx = (ch_addr >> 13); /* FIXME: Datasheet says to shift by 15 */
1098 idx %= 1 << rir_way;
1100 pci_read_config_dword(pvt->pci_tad[base_ch],
1101 rir_offset[n_rir][idx],
1103 *rank = RIR_RNK_TGT(reg);
1105 debugf0("RIR#%d: channel address 0x%08Lx < 0x%08Lx, RIR interleave %d, index %d\n",
1115 /****************************************************************************
1116 Device initialization routines: put/get, init/exit
1117 ****************************************************************************/
1120 * sbridge_put_all_devices 'put' all the devices that we have
1121 * reserved via 'get'
1123 static void sbridge_put_devices(struct sbridge_dev *sbridge_dev)
1127 debugf0(__FILE__ ": %s()\n", __func__);
1128 for (i = 0; i < sbridge_dev->n_devs; i++) {
1129 struct pci_dev *pdev = sbridge_dev->pdev[i];
1132 debugf0("Removing dev %02x:%02x.%d\n",
1134 PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn));
1139 static void sbridge_put_all_devices(void)
1141 struct sbridge_dev *sbridge_dev, *tmp;
1143 list_for_each_entry_safe(sbridge_dev, tmp, &sbridge_edac_list, list) {
1144 sbridge_put_devices(sbridge_dev);
1145 free_sbridge_dev(sbridge_dev);
1150 * sbridge_get_all_devices Find and perform 'get' operation on the MCH's
1151 * device/functions we want to reference for this driver
1153 * Need to 'get' device 16 func 1 and func 2
1155 static int sbridge_get_onedevice(struct pci_dev **prev,
1157 const struct pci_id_table *table,
1158 const unsigned devno)
1160 struct sbridge_dev *sbridge_dev;
1161 const struct pci_id_descr *dev_descr = &table->descr[devno];
1163 struct pci_dev *pdev = NULL;
1166 sbridge_printk(KERN_INFO,
1167 "Seeking for: dev %02x.%d PCI ID %04x:%04x\n",
1168 dev_descr->dev, dev_descr->func,
1169 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
1171 pdev = pci_get_device(PCI_VENDOR_ID_INTEL,
1172 dev_descr->dev_id, *prev);
1180 if (dev_descr->optional)
1186 sbridge_printk(KERN_INFO,
1187 "Device not found: dev %02x.%d PCI ID %04x:%04x\n",
1188 dev_descr->dev, dev_descr->func,
1189 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
1191 /* End of list, leave */
1194 bus = pdev->bus->number;
1196 sbridge_dev = get_sbridge_dev(bus);
1198 sbridge_dev = alloc_sbridge_dev(bus, table);
1206 if (sbridge_dev->pdev[devno]) {
1207 sbridge_printk(KERN_ERR,
1208 "Duplicated device for "
1209 "dev %02x:%d.%d PCI ID %04x:%04x\n",
1210 bus, dev_descr->dev, dev_descr->func,
1211 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
1216 sbridge_dev->pdev[devno] = pdev;
1219 if (unlikely(PCI_SLOT(pdev->devfn) != dev_descr->dev ||
1220 PCI_FUNC(pdev->devfn) != dev_descr->func)) {
1221 sbridge_printk(KERN_ERR,
1222 "Device PCI ID %04x:%04x "
1223 "has dev %02x:%d.%d instead of dev %02x:%02x.%d\n",
1224 PCI_VENDOR_ID_INTEL, dev_descr->dev_id,
1225 bus, PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn),
1226 bus, dev_descr->dev, dev_descr->func);
1230 /* Be sure that the device is enabled */
1231 if (unlikely(pci_enable_device(pdev) < 0)) {
1232 sbridge_printk(KERN_ERR,
1234 "dev %02x:%d.%d PCI ID %04x:%04x\n",
1235 bus, dev_descr->dev, dev_descr->func,
1236 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
1240 debugf0("Detected dev %02x:%d.%d PCI ID %04x:%04x\n",
1241 bus, dev_descr->dev,
1243 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
1246 * As stated on drivers/pci/search.c, the reference count for
1247 * @from is always decremented if it is not %NULL. So, as we need
1248 * to get all devices up to null, we need to do a get for the device
1257 static int sbridge_get_all_devices(u8 *num_mc)
1260 struct pci_dev *pdev = NULL;
1261 const struct pci_id_table *table = pci_dev_descr_sbridge_table;
1263 while (table && table->descr) {
1264 for (i = 0; i < table->n_devs; i++) {
1267 rc = sbridge_get_onedevice(&pdev, num_mc,
1274 sbridge_put_all_devices();
1285 static int mci_bind_devs(struct mem_ctl_info *mci,
1286 struct sbridge_dev *sbridge_dev)
1288 struct sbridge_pvt *pvt = mci->pvt_info;
1289 struct pci_dev *pdev;
1292 for (i = 0; i < sbridge_dev->n_devs; i++) {
1293 pdev = sbridge_dev->pdev[i];
1296 slot = PCI_SLOT(pdev->devfn);
1297 func = PCI_FUNC(pdev->devfn);
1302 pvt->pci_sad0 = pdev;
1305 pvt->pci_sad1 = pdev;
1323 pvt->pci_ha0 = pdev;
1335 pvt->pci_ras = pdev;
1341 pvt->pci_tad[func - 2] = pdev;
1350 pvt->pci_ddrio = pdev;
1360 debugf0("Associated PCI %02x.%02d.%d with dev = %p\n",
1362 PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn),
1366 /* Check if everything were registered */
1367 if (!pvt->pci_sad0 || !pvt->pci_sad1 || !pvt->pci_ha0 ||
1368 !pvt-> pci_tad || !pvt->pci_ras || !pvt->pci_ta ||
1372 for (i = 0; i < NUM_CHANNELS; i++) {
1373 if (!pvt->pci_tad[i])
1379 sbridge_printk(KERN_ERR, "Some needed devices are missing\n");
1383 sbridge_printk(KERN_ERR, "Device %d, function %d "
1384 "is out of the expected range\n",
1389 /****************************************************************************
1390 Error check routines
1391 ****************************************************************************/
1394 * While Sandy Bridge has error count registers, SMI BIOS read values from
1395 * and resets the counters. So, they are not reliable for the OS to read
1396 * from them. So, we have no option but to just trust on whatever MCE is
1397 * telling us about the errors.
1399 static void sbridge_mce_output_error(struct mem_ctl_info *mci,
1400 const struct mce *m)
1402 struct mem_ctl_info *new_mci;
1403 struct sbridge_pvt *pvt = mci->pvt_info;
1404 char *type, *optype, *msg, *recoverable_msg;
1405 bool ripv = GET_BITFIELD(m->mcgstatus, 0, 0);
1406 bool overflow = GET_BITFIELD(m->status, 62, 62);
1407 bool uncorrected_error = GET_BITFIELD(m->status, 61, 61);
1408 bool recoverable = GET_BITFIELD(m->status, 56, 56);
1409 u32 core_err_cnt = GET_BITFIELD(m->status, 38, 52);
1410 u32 mscod = GET_BITFIELD(m->status, 16, 31);
1411 u32 errcode = GET_BITFIELD(m->status, 0, 15);
1412 u32 channel = GET_BITFIELD(m->status, 0, 3);
1413 u32 optypenum = GET_BITFIELD(m->status, 4, 6);
1414 long channel_mask, first_channel;
1416 int csrow, rc, dimm;
1417 char *area_type = "Unknown";
1425 * According with Table 15-9 of the Intel Archictecture spec vol 3A,
1426 * memory errors should fit in this mask:
1427 * 000f 0000 1mmm cccc (binary)
1429 * f = Correction Report Filtering Bit. If 1, subsequent errors
1433 * If the mask doesn't match, report an error to the parsing logic
1435 if (! ((errcode & 0xef80) == 0x80)) {
1436 optype = "Can't parse: it is not a mem";
1438 switch (optypenum) {
1440 optype = "generic undef request";
1443 optype = "memory read";
1446 optype = "memory write";
1449 optype = "addr/cmd";
1452 optype = "memory scrubbing";
1455 optype = "reserved";
1460 rc = get_memory_error_data(mci, m->addr, &socket,
1461 &channel_mask, &rank, area_type);
1464 new_mci = get_mci_for_node_id(socket);
1466 edac_mc_handle_ce_no_info(mci, "Error: socket got corrupted!");
1470 pvt = mci->pvt_info;
1472 first_channel = find_first_bit(&channel_mask, NUM_CHANNELS);
1481 csrow = pvt->csrow_map[first_channel][dimm];
1483 if (uncorrected_error && recoverable)
1484 recoverable_msg = " recoverable";
1486 recoverable_msg = "";
1489 * FIXME: What should we do with "channel" information on mcelog?
1490 * Probably, we can just discard it, as the channel information
1491 * comes from the get_memory_error_data() address decoding
1493 msg = kasprintf(GFP_ATOMIC,
1494 "%d %s error(s): %s on %s area %s%s: cpu=%d Err=%04x:%04x (ch=%d), "
1495 "addr = 0x%08llx => socket=%d, Channel=%ld(mask=%ld), rank=%d\n",
1501 overflow ? "OVERFLOW" : "",
1504 channel, /* 1111b means not specified */
1505 (long long) m->addr,
1507 first_channel, /* This is the real channel on SB */
1513 /* Call the helper to output message */
1514 if (uncorrected_error)
1515 edac_mc_handle_fbd_ue(mci, csrow, 0, 0, msg);
1517 edac_mc_handle_fbd_ce(mci, csrow, 0, msg);
1523 * sbridge_check_error Retrieve and process errors reported by the
1524 * hardware. Called by the Core module.
1526 static void sbridge_check_error(struct mem_ctl_info *mci)
1528 struct sbridge_pvt *pvt = mci->pvt_info;
1534 * MCE first step: Copy all mce errors into a temporary buffer
1535 * We use a double buffering here, to reduce the risk of
1539 count = (pvt->mce_out + MCE_LOG_LEN - pvt->mce_in)
1544 m = pvt->mce_outentry;
1545 if (pvt->mce_in + count > MCE_LOG_LEN) {
1546 unsigned l = MCE_LOG_LEN - pvt->mce_in;
1548 memcpy(m, &pvt->mce_entry[pvt->mce_in], sizeof(*m) * l);
1554 memcpy(m, &pvt->mce_entry[pvt->mce_in], sizeof(*m) * count);
1556 pvt->mce_in += count;
1559 if (pvt->mce_overrun) {
1560 sbridge_printk(KERN_ERR, "Lost %d memory errors\n",
1563 pvt->mce_overrun = 0;
1567 * MCE second step: parse errors and display
1569 for (i = 0; i < count; i++)
1570 sbridge_mce_output_error(mci, &pvt->mce_outentry[i]);
1574 * sbridge_mce_check_error Replicates mcelog routine to get errors
1575 * This routine simply queues mcelog errors, and
1576 * return. The error itself should be handled later
1577 * by sbridge_check_error.
1578 * WARNING: As this routine should be called at NMI time, extra care should
1579 * be taken to avoid deadlocks, and to be as fast as possible.
1581 static int sbridge_mce_check_error(void *priv, struct mce *mce)
1583 struct mem_ctl_info *mci = priv;
1584 struct sbridge_pvt *pvt = mci->pvt_info;
1587 * Just let mcelog handle it if the error is
1588 * outside the memory controller. A memory error
1589 * is indicated by bit 7 = 1 and bits = 8-11,13-15 = 0.
1590 * bit 12 has an special meaning.
1592 if ((mce->status & 0xefff) >> 7 != 1)
1595 printk("sbridge: HANDLING MCE MEMORY ERROR\n");
1597 printk("CPU %d: Machine Check Exception: %Lx Bank %d: %016Lx\n",
1598 mce->extcpu, mce->mcgstatus, mce->bank, mce->status);
1599 printk("TSC %llx ", mce->tsc);
1600 printk("ADDR %llx ", mce->addr);
1601 printk("MISC %llx ", mce->misc);
1603 printk("PROCESSOR %u:%x TIME %llu SOCKET %u APIC %x\n",
1604 mce->cpuvendor, mce->cpuid, mce->time,
1605 mce->socketid, mce->apicid);
1608 /* Only handle if it is the right mc controller */
1609 if (cpu_data(mce->cpu).phys_proc_id != pvt->sbridge_dev->mc)
1614 if ((pvt->mce_out + 1) % MCE_LOG_LEN == pvt->mce_in) {
1620 /* Copy memory error at the ringbuffer */
1621 memcpy(&pvt->mce_entry[pvt->mce_out], mce, sizeof(*mce));
1623 pvt->mce_out = (pvt->mce_out + 1) % MCE_LOG_LEN;
1625 /* Handle fatal errors immediately */
1626 if (mce->mcgstatus & 1)
1627 sbridge_check_error(mci);
1629 /* Advice mcelog that the error were handled */
1633 /****************************************************************************
1634 EDAC register/unregister logic
1635 ****************************************************************************/
1637 static void sbridge_unregister_mci(struct sbridge_dev *sbridge_dev)
1639 struct mem_ctl_info *mci = sbridge_dev->mci;
1640 struct sbridge_pvt *pvt;
1642 if (unlikely(!mci || !mci->pvt_info)) {
1643 debugf0("MC: " __FILE__ ": %s(): dev = %p\n",
1644 __func__, &sbridge_dev->pdev[0]->dev);
1646 sbridge_printk(KERN_ERR, "Couldn't find mci handler\n");
1650 pvt = mci->pvt_info;
1652 debugf0("MC: " __FILE__ ": %s(): mci = %p, dev = %p\n",
1653 __func__, mci, &sbridge_dev->pdev[0]->dev);
1655 /* Disable MCE NMI handler */
1656 edac_mce_unregister(&pvt->edac_mce);
1658 /* Remove MC sysfs nodes */
1659 edac_mc_del_mc(mci->dev);
1661 debugf1("%s: free mci struct\n", mci->ctl_name);
1662 kfree(mci->ctl_name);
1664 sbridge_dev->mci = NULL;
1667 static int sbridge_register_mci(struct sbridge_dev *sbridge_dev)
1669 struct mem_ctl_info *mci;
1670 struct sbridge_pvt *pvt;
1671 int rc, channels, csrows;
1673 /* Check the number of active and not disabled channels */
1674 rc = sbridge_get_active_channels(sbridge_dev->bus, &channels, &csrows);
1675 if (unlikely(rc < 0))
1678 /* allocate a new MC control structure */
1679 mci = edac_mc_alloc(sizeof(*pvt), csrows, channels, sbridge_dev->mc);
1683 debugf0("MC: " __FILE__ ": %s(): mci = %p, dev = %p\n",
1684 __func__, mci, &sbridge_dev->pdev[0]->dev);
1686 pvt = mci->pvt_info;
1687 memset(pvt, 0, sizeof(*pvt));
1689 /* Associate sbridge_dev and mci for future usage */
1690 pvt->sbridge_dev = sbridge_dev;
1691 sbridge_dev->mci = mci;
1693 mci->mtype_cap = MEM_FLAG_DDR3;
1694 mci->edac_ctl_cap = EDAC_FLAG_NONE;
1695 mci->edac_cap = EDAC_FLAG_NONE;
1696 mci->mod_name = "sbridge_edac.c";
1697 mci->mod_ver = SBRIDGE_REVISION;
1698 mci->ctl_name = kasprintf(GFP_KERNEL, "Sandy Bridge Socket#%d", mci->mc_idx);
1699 mci->dev_name = pci_name(sbridge_dev->pdev[0]);
1700 mci->ctl_page_to_phys = NULL;
1702 /* Set the function pointer to an actual operation function */
1703 mci->edac_check = sbridge_check_error;
1705 /* Store pci devices at mci for faster access */
1706 rc = mci_bind_devs(mci, sbridge_dev);
1707 if (unlikely(rc < 0))
1710 /* Get dimm basic config and the memory layout */
1711 get_dimm_config(mci);
1712 get_memory_layout(mci);
1714 /* record ptr to the generic device */
1715 mci->dev = &sbridge_dev->pdev[0]->dev;
1717 /* add this new MC control structure to EDAC's list of MCs */
1718 if (unlikely(edac_mc_add_mc(mci))) {
1719 debugf0("MC: " __FILE__
1720 ": %s(): failed edac_mc_add_mc()\n", __func__);
1725 /* Registers on edac_mce in order to receive memory errors */
1726 pvt->edac_mce.priv = mci;
1727 pvt->edac_mce.check_error = sbridge_mce_check_error;
1728 rc = edac_mce_register(&pvt->edac_mce);
1729 if (unlikely(rc < 0)) {
1730 debugf0("MC: " __FILE__
1731 ": %s(): failed edac_mce_register()\n", __func__);
1737 edac_mc_del_mc(mci->dev);
1740 kfree(mci->ctl_name);
1742 sbridge_dev->mci = NULL;
1747 * sbridge_probe Probe for ONE instance of device to see if it is
1750 * 0 for FOUND a device
1751 * < 0 for error code
1754 static int __devinit sbridge_probe(struct pci_dev *pdev,
1755 const struct pci_device_id *id)
1759 struct sbridge_dev *sbridge_dev;
1761 /* get the pci devices we want to reserve for our use */
1762 mutex_lock(&sbridge_edac_lock);
1765 * All memory controllers are allocated at the first pass.
1767 if (unlikely(probed >= 1)) {
1768 mutex_unlock(&sbridge_edac_lock);
1773 rc = sbridge_get_all_devices(&num_mc);
1774 if (unlikely(rc < 0))
1778 list_for_each_entry(sbridge_dev, &sbridge_edac_list, list) {
1779 debugf0("Registering MC#%d (%d of %d)\n", mc, mc + 1, num_mc);
1780 sbridge_dev->mc = mc++;
1781 rc = sbridge_register_mci(sbridge_dev);
1782 if (unlikely(rc < 0))
1786 sbridge_printk(KERN_INFO, "Driver loaded.\n");
1788 mutex_unlock(&sbridge_edac_lock);
1792 list_for_each_entry(sbridge_dev, &sbridge_edac_list, list)
1793 sbridge_unregister_mci(sbridge_dev);
1795 sbridge_put_all_devices();
1797 mutex_unlock(&sbridge_edac_lock);
1802 * sbridge_remove destructor for one instance of device
1805 static void __devexit sbridge_remove(struct pci_dev *pdev)
1807 struct sbridge_dev *sbridge_dev;
1809 debugf0(__FILE__ ": %s()\n", __func__);
1812 * we have a trouble here: pdev value for removal will be wrong, since
1813 * it will point to the X58 register used to detect that the machine
1814 * is a Nehalem or upper design. However, due to the way several PCI
1815 * devices are grouped together to provide MC functionality, we need
1816 * to use a different method for releasing the devices
1819 mutex_lock(&sbridge_edac_lock);
1821 if (unlikely(!probed)) {
1822 mutex_unlock(&sbridge_edac_lock);
1826 list_for_each_entry(sbridge_dev, &sbridge_edac_list, list)
1827 sbridge_unregister_mci(sbridge_dev);
1829 /* Release PCI resources */
1830 sbridge_put_all_devices();
1834 mutex_unlock(&sbridge_edac_lock);
1837 MODULE_DEVICE_TABLE(pci, sbridge_pci_tbl);
1840 * sbridge_driver pci_driver structure for this module
1843 static struct pci_driver sbridge_driver = {
1844 .name = "sbridge_edac",
1845 .probe = sbridge_probe,
1846 .remove = __devexit_p(sbridge_remove),
1847 .id_table = sbridge_pci_tbl,
1851 * sbridge_init Module entry function
1852 * Try to initialize this module for its devices
1854 static int __init sbridge_init(void)
1858 debugf2("MC: " __FILE__ ": %s()\n", __func__);
1860 /* Ensure that the OPSTATE is set correctly for POLL or NMI */
1863 pci_rc = pci_register_driver(&sbridge_driver);
1868 sbridge_printk(KERN_ERR, "Failed to register device with error %d.\n",
1875 * sbridge_exit() Module exit function
1876 * Unregister the driver
1878 static void __exit sbridge_exit(void)
1880 debugf2("MC: " __FILE__ ": %s()\n", __func__);
1881 pci_unregister_driver(&sbridge_driver);
1884 module_init(sbridge_init);
1885 module_exit(sbridge_exit);
1887 module_param(edac_op_state, int, 0444);
1888 MODULE_PARM_DESC(edac_op_state, "EDAC Error Reporting state: 0=Poll,1=NMI");
1890 MODULE_LICENSE("GPL");
1891 MODULE_AUTHOR("Mauro Carvalho Chehab <mchehab@redhat.com>");
1892 MODULE_AUTHOR("Red Hat Inc. (http://www.redhat.com)");
1893 MODULE_DESCRIPTION("MC Driver for Intel Sandy Bridge memory controllers - "