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
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 <linux/mod_devicetable.h>
25 #include <asm/cpu_device_id.h>
26 #include <asm/processor.h>
29 #include "edac_core.h"
32 static LIST_HEAD(sbridge_edac_list);
35 * Alter this version for the module when modifications are made
37 #define SBRIDGE_REVISION " Ver: 1.1.1 "
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) & GENMASK_ULL(hi, lo)) >> (lo))
55 /* Devices 12 Function 6, Offsets 0x80 to 0xcc */
56 static const u32 sbridge_dram_rule[] = {
57 0x80, 0x88, 0x90, 0x98, 0xa0,
58 0xa8, 0xb0, 0xb8, 0xc0, 0xc8,
61 static const u32 ibridge_dram_rule[] = {
62 0x60, 0x68, 0x70, 0x78, 0x80,
63 0x88, 0x90, 0x98, 0xa0, 0xa8,
64 0xb0, 0xb8, 0xc0, 0xc8, 0xd0,
65 0xd8, 0xe0, 0xe8, 0xf0, 0xf8,
68 static const u32 knl_dram_rule[] = {
69 0x60, 0x68, 0x70, 0x78, 0x80, /* 0-4 */
70 0x88, 0x90, 0x98, 0xa0, 0xa8, /* 5-9 */
71 0xb0, 0xb8, 0xc0, 0xc8, 0xd0, /* 10-14 */
72 0xd8, 0xe0, 0xe8, 0xf0, 0xf8, /* 15-19 */
73 0x100, 0x108, 0x110, 0x118, /* 20-23 */
76 #define DRAM_RULE_ENABLE(reg) GET_BITFIELD(reg, 0, 0)
77 #define A7MODE(reg) GET_BITFIELD(reg, 26, 26)
79 static char *show_dram_attr(u32 attr)
93 static const u32 sbridge_interleave_list[] = {
94 0x84, 0x8c, 0x94, 0x9c, 0xa4,
95 0xac, 0xb4, 0xbc, 0xc4, 0xcc,
98 static const u32 ibridge_interleave_list[] = {
99 0x64, 0x6c, 0x74, 0x7c, 0x84,
100 0x8c, 0x94, 0x9c, 0xa4, 0xac,
101 0xb4, 0xbc, 0xc4, 0xcc, 0xd4,
102 0xdc, 0xe4, 0xec, 0xf4, 0xfc,
105 static const u32 knl_interleave_list[] = {
106 0x64, 0x6c, 0x74, 0x7c, 0x84, /* 0-4 */
107 0x8c, 0x94, 0x9c, 0xa4, 0xac, /* 5-9 */
108 0xb4, 0xbc, 0xc4, 0xcc, 0xd4, /* 10-14 */
109 0xdc, 0xe4, 0xec, 0xf4, 0xfc, /* 15-19 */
110 0x104, 0x10c, 0x114, 0x11c, /* 20-23 */
113 struct interleave_pkg {
118 static const struct interleave_pkg sbridge_interleave_pkg[] = {
129 static const struct interleave_pkg ibridge_interleave_pkg[] = {
140 static inline int sad_pkg(const struct interleave_pkg *table, u32 reg,
143 return GET_BITFIELD(reg, table[interleave].start,
144 table[interleave].end);
147 /* Devices 12 Function 7 */
151 #define HASWELL_TOLM 0xd0
152 #define HASWELL_TOHM_0 0xd4
153 #define HASWELL_TOHM_1 0xd8
154 #define KNL_TOLM 0xd0
155 #define KNL_TOHM_0 0xd4
156 #define KNL_TOHM_1 0xd8
158 #define GET_TOLM(reg) ((GET_BITFIELD(reg, 0, 3) << 28) | 0x3ffffff)
159 #define GET_TOHM(reg) ((GET_BITFIELD(reg, 0, 20) << 25) | 0x3ffffff)
161 /* Device 13 Function 6 */
163 #define SAD_TARGET 0xf0
165 #define SOURCE_ID(reg) GET_BITFIELD(reg, 9, 11)
167 #define SOURCE_ID_KNL(reg) GET_BITFIELD(reg, 12, 14)
169 #define SAD_CONTROL 0xf4
171 /* Device 14 function 0 */
173 static const u32 tad_dram_rule[] = {
174 0x40, 0x44, 0x48, 0x4c,
175 0x50, 0x54, 0x58, 0x5c,
176 0x60, 0x64, 0x68, 0x6c,
178 #define MAX_TAD ARRAY_SIZE(tad_dram_rule)
180 #define TAD_LIMIT(reg) ((GET_BITFIELD(reg, 12, 31) << 26) | 0x3ffffff)
181 #define TAD_SOCK(reg) GET_BITFIELD(reg, 10, 11)
182 #define TAD_CH(reg) GET_BITFIELD(reg, 8, 9)
183 #define TAD_TGT3(reg) GET_BITFIELD(reg, 6, 7)
184 #define TAD_TGT2(reg) GET_BITFIELD(reg, 4, 5)
185 #define TAD_TGT1(reg) GET_BITFIELD(reg, 2, 3)
186 #define TAD_TGT0(reg) GET_BITFIELD(reg, 0, 1)
188 /* Device 15, function 0 */
191 #define KNL_MCMTR 0x624
193 #define IS_ECC_ENABLED(mcmtr) GET_BITFIELD(mcmtr, 2, 2)
194 #define IS_LOCKSTEP_ENABLED(mcmtr) GET_BITFIELD(mcmtr, 1, 1)
195 #define IS_CLOSE_PG(mcmtr) GET_BITFIELD(mcmtr, 0, 0)
197 /* Device 15, function 1 */
199 #define RASENABLES 0xac
200 #define IS_MIRROR_ENABLED(reg) GET_BITFIELD(reg, 0, 0)
202 /* Device 15, functions 2-5 */
204 static const int mtr_regs[] = {
208 static const int knl_mtr_reg = 0xb60;
210 #define RANK_DISABLE(mtr) GET_BITFIELD(mtr, 16, 19)
211 #define IS_DIMM_PRESENT(mtr) GET_BITFIELD(mtr, 14, 14)
212 #define RANK_CNT_BITS(mtr) GET_BITFIELD(mtr, 12, 13)
213 #define RANK_WIDTH_BITS(mtr) GET_BITFIELD(mtr, 2, 4)
214 #define COL_WIDTH_BITS(mtr) GET_BITFIELD(mtr, 0, 1)
216 static const u32 tad_ch_nilv_offset[] = {
217 0x90, 0x94, 0x98, 0x9c,
218 0xa0, 0xa4, 0xa8, 0xac,
219 0xb0, 0xb4, 0xb8, 0xbc,
221 #define CHN_IDX_OFFSET(reg) GET_BITFIELD(reg, 28, 29)
222 #define TAD_OFFSET(reg) (GET_BITFIELD(reg, 6, 25) << 26)
224 static const u32 rir_way_limit[] = {
225 0x108, 0x10c, 0x110, 0x114, 0x118,
227 #define MAX_RIR_RANGES ARRAY_SIZE(rir_way_limit)
229 #define IS_RIR_VALID(reg) GET_BITFIELD(reg, 31, 31)
230 #define RIR_WAY(reg) GET_BITFIELD(reg, 28, 29)
232 #define MAX_RIR_WAY 8
234 static const u32 rir_offset[MAX_RIR_RANGES][MAX_RIR_WAY] = {
235 { 0x120, 0x124, 0x128, 0x12c, 0x130, 0x134, 0x138, 0x13c },
236 { 0x140, 0x144, 0x148, 0x14c, 0x150, 0x154, 0x158, 0x15c },
237 { 0x160, 0x164, 0x168, 0x16c, 0x170, 0x174, 0x178, 0x17c },
238 { 0x180, 0x184, 0x188, 0x18c, 0x190, 0x194, 0x198, 0x19c },
239 { 0x1a0, 0x1a4, 0x1a8, 0x1ac, 0x1b0, 0x1b4, 0x1b8, 0x1bc },
242 #define RIR_RNK_TGT(reg) GET_BITFIELD(reg, 16, 19)
243 #define RIR_OFFSET(reg) GET_BITFIELD(reg, 2, 14)
245 /* Device 16, functions 2-7 */
248 * FIXME: Implement the error count reads directly
251 static const u32 correrrcnt[] = {
252 0x104, 0x108, 0x10c, 0x110,
255 #define RANK_ODD_OV(reg) GET_BITFIELD(reg, 31, 31)
256 #define RANK_ODD_ERR_CNT(reg) GET_BITFIELD(reg, 16, 30)
257 #define RANK_EVEN_OV(reg) GET_BITFIELD(reg, 15, 15)
258 #define RANK_EVEN_ERR_CNT(reg) GET_BITFIELD(reg, 0, 14)
260 static const u32 correrrthrsld[] = {
261 0x11c, 0x120, 0x124, 0x128,
264 #define RANK_ODD_ERR_THRSLD(reg) GET_BITFIELD(reg, 16, 30)
265 #define RANK_EVEN_ERR_THRSLD(reg) GET_BITFIELD(reg, 0, 14)
268 /* Device 17, function 0 */
270 #define SB_RANK_CFG_A 0x0328
272 #define IB_RANK_CFG_A 0x0320
278 #define NUM_CHANNELS 8 /* 2MC per socket, four chan per MC */
279 #define MAX_DIMMS 3 /* Max DIMMS per channel */
280 #define KNL_MAX_CHAS 38 /* KNL max num. of Cache Home Agents */
281 #define KNL_MAX_CHANNELS 6 /* KNL max num. of PCI channels */
282 #define KNL_MAX_EDCS 8 /* Embedded DRAM controllers */
283 #define CHANNEL_UNSPECIFIED 0xf /* Intel IA32 SDM 15-14 */
294 struct sbridge_info {
298 u64 (*get_tolm)(struct sbridge_pvt *pvt);
299 u64 (*get_tohm)(struct sbridge_pvt *pvt);
300 u64 (*rir_limit)(u32 reg);
301 u64 (*sad_limit)(u32 reg);
302 u32 (*interleave_mode)(u32 reg);
303 char* (*show_interleave_mode)(u32 reg);
304 u32 (*dram_attr)(u32 reg);
305 const u32 *dram_rule;
306 const u32 *interleave_list;
307 const struct interleave_pkg *interleave_pkg;
310 u8 (*get_node_id)(struct sbridge_pvt *pvt);
311 enum mem_type (*get_memory_type)(struct sbridge_pvt *pvt);
312 enum dev_type (*get_width)(struct sbridge_pvt *pvt, u32 mtr);
313 struct pci_dev *pci_vtd;
316 struct sbridge_channel {
321 struct pci_id_descr {
326 struct pci_id_table {
327 const struct pci_id_descr *descr;
332 struct list_head list;
334 u8 node_id, source_id;
335 struct pci_dev **pdev;
337 struct mem_ctl_info *mci;
341 struct pci_dev *pci_cha[KNL_MAX_CHAS];
342 struct pci_dev *pci_channel[KNL_MAX_CHANNELS];
343 struct pci_dev *pci_mc0;
344 struct pci_dev *pci_mc1;
345 struct pci_dev *pci_mc0_misc;
346 struct pci_dev *pci_mc1_misc;
347 struct pci_dev *pci_mc_info; /* tolm, tohm */
351 struct pci_dev *pci_ta, *pci_ddrio, *pci_ras;
352 struct pci_dev *pci_sad0, *pci_sad1;
353 struct pci_dev *pci_ha0, *pci_ha1;
354 struct pci_dev *pci_br0, *pci_br1;
355 struct pci_dev *pci_ha1_ta;
356 struct pci_dev *pci_tad[NUM_CHANNELS];
358 struct sbridge_dev *sbridge_dev;
360 struct sbridge_info info;
361 struct sbridge_channel channel[NUM_CHANNELS];
363 /* Memory type detection */
364 bool is_mirrored, is_lockstep, is_close_pg;
367 /* Memory description */
372 #define PCI_DESCR(device_id, opt) \
373 .dev_id = (device_id), \
376 static const struct pci_id_descr pci_dev_descr_sbridge[] = {
377 /* Processor Home Agent */
378 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_HA0, 0) },
380 /* Memory controller */
381 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TA, 0) },
382 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_RAS, 0) },
383 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD0, 0) },
384 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD1, 0) },
385 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD2, 0) },
386 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD3, 0) },
387 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_DDRIO, 1) },
389 /* System Address Decoder */
390 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_SAD0, 0) },
391 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_SAD1, 0) },
393 /* Broadcast Registers */
394 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_BR, 0) },
397 #define PCI_ID_TABLE_ENTRY(A) { .descr=A, .n_devs = ARRAY_SIZE(A) }
398 static const struct pci_id_table pci_dev_descr_sbridge_table[] = {
399 PCI_ID_TABLE_ENTRY(pci_dev_descr_sbridge),
400 {0,} /* 0 terminated list. */
403 /* This changes depending if 1HA or 2HA:
405 * 0x0eb8 (17.0) is DDRIO0
407 * 0x0ebc (17.4) is DDRIO0
409 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_1HA_DDRIO0 0x0eb8
410 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_2HA_DDRIO0 0x0ebc
413 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0 0x0ea0
414 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TA 0x0ea8
415 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_RAS 0x0e71
416 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD0 0x0eaa
417 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD1 0x0eab
418 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD2 0x0eac
419 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD3 0x0ead
420 #define PCI_DEVICE_ID_INTEL_IBRIDGE_SAD 0x0ec8
421 #define PCI_DEVICE_ID_INTEL_IBRIDGE_BR0 0x0ec9
422 #define PCI_DEVICE_ID_INTEL_IBRIDGE_BR1 0x0eca
423 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1 0x0e60
424 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TA 0x0e68
425 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_RAS 0x0e79
426 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD0 0x0e6a
427 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD1 0x0e6b
428 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD2 0x0e6c
429 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD3 0x0e6d
431 static const struct pci_id_descr pci_dev_descr_ibridge[] = {
432 /* Processor Home Agent */
433 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0, 0) },
435 /* Memory controller */
436 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TA, 0) },
437 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_RAS, 0) },
438 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD0, 0) },
439 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD1, 0) },
440 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD2, 0) },
441 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD3, 0) },
443 /* System Address Decoder */
444 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_SAD, 0) },
446 /* Broadcast Registers */
447 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_BR0, 1) },
448 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_BR1, 0) },
450 /* Optional, mode 2HA */
451 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1, 1) },
453 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TA, 1) },
454 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_RAS, 1) },
456 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD0, 1) },
457 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD1, 1) },
458 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD2, 1) },
459 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD3, 1) },
461 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_1HA_DDRIO0, 1) },
462 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_2HA_DDRIO0, 1) },
465 static const struct pci_id_table pci_dev_descr_ibridge_table[] = {
466 PCI_ID_TABLE_ENTRY(pci_dev_descr_ibridge),
467 {0,} /* 0 terminated list. */
470 /* Haswell support */
473 * - 3 DDR3 channels, 2 DPC per channel
476 * - 4 DDR4 channels, 3 DPC per channel
479 * - 4 DDR4 channels, 3 DPC per channel
482 * - each IMC interfaces with a SMI 2 channel
483 * - each SMI channel interfaces with a scalable memory buffer
484 * - each scalable memory buffer supports 4 DDR3/DDR4 channels, 3 DPC
486 #define HASWELL_DDRCRCLKCONTROLS 0xa10 /* Ditto on Broadwell */
487 #define HASWELL_HASYSDEFEATURE2 0x84
488 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_VTD_MISC 0x2f28
489 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0 0x2fa0
490 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1 0x2f60
491 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TA 0x2fa8
492 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_THERMAL 0x2f71
493 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TA 0x2f68
494 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_THERMAL 0x2f79
495 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_CBO_SAD0 0x2ffc
496 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_CBO_SAD1 0x2ffd
497 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD0 0x2faa
498 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD1 0x2fab
499 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD2 0x2fac
500 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD3 0x2fad
501 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD0 0x2f6a
502 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD1 0x2f6b
503 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD2 0x2f6c
504 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD3 0x2f6d
505 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO0 0x2fbd
506 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO1 0x2fbf
507 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO2 0x2fb9
508 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO3 0x2fbb
509 static const struct pci_id_descr pci_dev_descr_haswell[] = {
510 /* first item must be the HA */
511 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0, 0) },
513 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_CBO_SAD0, 0) },
514 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_CBO_SAD1, 0) },
516 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1, 1) },
518 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TA, 0) },
519 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_THERMAL, 0) },
520 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD0, 0) },
521 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD1, 0) },
522 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD2, 1) },
523 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD3, 1) },
525 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO0, 1) },
526 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO1, 1) },
527 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO2, 1) },
528 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO3, 1) },
530 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TA, 1) },
531 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_THERMAL, 1) },
532 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD0, 1) },
533 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD1, 1) },
534 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD2, 1) },
535 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD3, 1) },
538 static const struct pci_id_table pci_dev_descr_haswell_table[] = {
539 PCI_ID_TABLE_ENTRY(pci_dev_descr_haswell),
540 {0,} /* 0 terminated list. */
543 /* Knight's Landing Support */
545 * KNL's memory channels are swizzled between memory controllers.
546 * MC0 is mapped to CH3,5,6 and MC1 is mapped to CH0,1,2
548 #define knl_channel_remap(channel) ((channel + 3) % 6)
550 /* Memory controller, TAD tables, error injection - 2-8-0, 2-9-0 (2 of these) */
551 #define PCI_DEVICE_ID_INTEL_KNL_IMC_MC 0x7840
552 /* DRAM channel stuff; bank addrs, dimmmtr, etc.. 2-8-2 - 2-9-4 (6 of these) */
553 #define PCI_DEVICE_ID_INTEL_KNL_IMC_CHANNEL 0x7843
554 /* kdrwdbu TAD limits/offsets, MCMTR - 2-10-1, 2-11-1 (2 of these) */
555 #define PCI_DEVICE_ID_INTEL_KNL_IMC_TA 0x7844
556 /* CHA broadcast registers, dram rules - 1-29-0 (1 of these) */
557 #define PCI_DEVICE_ID_INTEL_KNL_IMC_SAD0 0x782a
558 /* SAD target - 1-29-1 (1 of these) */
559 #define PCI_DEVICE_ID_INTEL_KNL_IMC_SAD1 0x782b
560 /* Caching / Home Agent */
561 #define PCI_DEVICE_ID_INTEL_KNL_IMC_CHA 0x782c
562 /* Device with TOLM and TOHM, 0-5-0 (1 of these) */
563 #define PCI_DEVICE_ID_INTEL_KNL_IMC_TOLHM 0x7810
566 * KNL differs from SB, IB, and Haswell in that it has multiple
567 * instances of the same device with the same device ID, so we handle that
568 * by creating as many copies in the table as we expect to find.
569 * (Like device ID must be grouped together.)
572 static const struct pci_id_descr pci_dev_descr_knl[] = {
573 [0] = { PCI_DESCR(PCI_DEVICE_ID_INTEL_KNL_IMC_SAD0, 0) },
574 [1] = { PCI_DESCR(PCI_DEVICE_ID_INTEL_KNL_IMC_SAD1, 0) },
575 [2 ... 3] = { PCI_DESCR(PCI_DEVICE_ID_INTEL_KNL_IMC_MC, 0)},
576 [4 ... 41] = { PCI_DESCR(PCI_DEVICE_ID_INTEL_KNL_IMC_CHA, 0) },
577 [42 ... 47] = { PCI_DESCR(PCI_DEVICE_ID_INTEL_KNL_IMC_CHANNEL, 0) },
578 [48] = { PCI_DESCR(PCI_DEVICE_ID_INTEL_KNL_IMC_TA, 0) },
579 [49] = { PCI_DESCR(PCI_DEVICE_ID_INTEL_KNL_IMC_TOLHM, 0) },
582 static const struct pci_id_table pci_dev_descr_knl_table[] = {
583 PCI_ID_TABLE_ENTRY(pci_dev_descr_knl),
592 * - 2 DDR3 channels, 2 DPC per channel
595 * - 4 DDR4 channels, 3 DPC per channel
598 * - 4 DDR4 channels, 3 DPC per channel
601 * - each IMC interfaces with a SMI 2 channel
602 * - each SMI channel interfaces with a scalable memory buffer
603 * - each scalable memory buffer supports 4 DDR3/DDR4 channels, 3 DPC
605 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_VTD_MISC 0x6f28
606 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0 0x6fa0
607 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1 0x6f60
608 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TA 0x6fa8
609 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_THERMAL 0x6f71
610 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TA 0x6f68
611 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_THERMAL 0x6f79
612 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_CBO_SAD0 0x6ffc
613 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_CBO_SAD1 0x6ffd
614 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD0 0x6faa
615 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD1 0x6fab
616 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD2 0x6fac
617 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD3 0x6fad
618 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD0 0x6f6a
619 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD1 0x6f6b
620 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD2 0x6f6c
621 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD3 0x6f6d
622 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_DDRIO0 0x6faf
624 static const struct pci_id_descr pci_dev_descr_broadwell[] = {
625 /* first item must be the HA */
626 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0, 0) },
628 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_CBO_SAD0, 0) },
629 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_CBO_SAD1, 0) },
631 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1, 1) },
633 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TA, 0) },
634 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_THERMAL, 0) },
635 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD0, 0) },
636 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD1, 0) },
637 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD2, 1) },
638 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD3, 1) },
640 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_DDRIO0, 1) },
642 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TA, 1) },
643 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_THERMAL, 1) },
644 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD0, 1) },
645 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD1, 1) },
646 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD2, 1) },
647 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD3, 1) },
650 static const struct pci_id_table pci_dev_descr_broadwell_table[] = {
651 PCI_ID_TABLE_ENTRY(pci_dev_descr_broadwell),
652 {0,} /* 0 terminated list. */
656 /****************************************************************************
657 Ancillary status routines
658 ****************************************************************************/
660 static inline int numrank(enum type type, u32 mtr)
662 int ranks = (1 << RANK_CNT_BITS(mtr));
665 if (type == HASWELL || type == BROADWELL || type == KNIGHTS_LANDING)
669 edac_dbg(0, "Invalid number of ranks: %d (max = %i) raw value = %x (%04x)\n",
670 ranks, max, (unsigned int)RANK_CNT_BITS(mtr), mtr);
677 static inline int numrow(u32 mtr)
679 int rows = (RANK_WIDTH_BITS(mtr) + 12);
681 if (rows < 13 || rows > 18) {
682 edac_dbg(0, "Invalid number of rows: %d (should be between 14 and 17) raw value = %x (%04x)\n",
683 rows, (unsigned int)RANK_WIDTH_BITS(mtr), mtr);
690 static inline int numcol(u32 mtr)
692 int cols = (COL_WIDTH_BITS(mtr) + 10);
695 edac_dbg(0, "Invalid number of cols: %d (max = 4) raw value = %x (%04x)\n",
696 cols, (unsigned int)COL_WIDTH_BITS(mtr), mtr);
703 static struct sbridge_dev *get_sbridge_dev(u8 bus, int multi_bus)
705 struct sbridge_dev *sbridge_dev;
708 * If we have devices scattered across several busses that pertain
709 * to the same memory controller, we'll lump them all together.
712 return list_first_entry_or_null(&sbridge_edac_list,
713 struct sbridge_dev, list);
716 list_for_each_entry(sbridge_dev, &sbridge_edac_list, list) {
717 if (sbridge_dev->bus == bus)
724 static struct sbridge_dev *alloc_sbridge_dev(u8 bus,
725 const struct pci_id_table *table)
727 struct sbridge_dev *sbridge_dev;
729 sbridge_dev = kzalloc(sizeof(*sbridge_dev), GFP_KERNEL);
733 sbridge_dev->pdev = kzalloc(sizeof(*sbridge_dev->pdev) * table->n_devs,
735 if (!sbridge_dev->pdev) {
740 sbridge_dev->bus = bus;
741 sbridge_dev->n_devs = table->n_devs;
742 list_add_tail(&sbridge_dev->list, &sbridge_edac_list);
747 static void free_sbridge_dev(struct sbridge_dev *sbridge_dev)
749 list_del(&sbridge_dev->list);
750 kfree(sbridge_dev->pdev);
754 static u64 sbridge_get_tolm(struct sbridge_pvt *pvt)
758 /* Address range is 32:28 */
759 pci_read_config_dword(pvt->pci_sad1, TOLM, ®);
760 return GET_TOLM(reg);
763 static u64 sbridge_get_tohm(struct sbridge_pvt *pvt)
767 pci_read_config_dword(pvt->pci_sad1, TOHM, ®);
768 return GET_TOHM(reg);
771 static u64 ibridge_get_tolm(struct sbridge_pvt *pvt)
775 pci_read_config_dword(pvt->pci_br1, TOLM, ®);
777 return GET_TOLM(reg);
780 static u64 ibridge_get_tohm(struct sbridge_pvt *pvt)
784 pci_read_config_dword(pvt->pci_br1, TOHM, ®);
786 return GET_TOHM(reg);
789 static u64 rir_limit(u32 reg)
791 return ((u64)GET_BITFIELD(reg, 1, 10) << 29) | 0x1fffffff;
794 static u64 sad_limit(u32 reg)
796 return (GET_BITFIELD(reg, 6, 25) << 26) | 0x3ffffff;
799 static u32 interleave_mode(u32 reg)
801 return GET_BITFIELD(reg, 1, 1);
804 char *show_interleave_mode(u32 reg)
806 return interleave_mode(reg) ? "8:6" : "[8:6]XOR[18:16]";
809 static u32 dram_attr(u32 reg)
811 return GET_BITFIELD(reg, 2, 3);
814 static u64 knl_sad_limit(u32 reg)
816 return (GET_BITFIELD(reg, 7, 26) << 26) | 0x3ffffff;
819 static u32 knl_interleave_mode(u32 reg)
821 return GET_BITFIELD(reg, 1, 2);
824 static char *knl_show_interleave_mode(u32 reg)
828 switch (knl_interleave_mode(reg)) {
830 s = "use address bits [8:6]";
833 s = "use address bits [10:8]";
836 s = "use address bits [14:12]";
839 s = "use address bits [32:30]";
849 static u32 dram_attr_knl(u32 reg)
851 return GET_BITFIELD(reg, 3, 4);
855 static enum mem_type get_memory_type(struct sbridge_pvt *pvt)
860 if (pvt->pci_ddrio) {
861 pci_read_config_dword(pvt->pci_ddrio, pvt->info.rankcfgr,
863 if (GET_BITFIELD(reg, 11, 11))
864 /* FIXME: Can also be LRDIMM */
874 static enum mem_type haswell_get_memory_type(struct sbridge_pvt *pvt)
877 bool registered = false;
878 enum mem_type mtype = MEM_UNKNOWN;
883 pci_read_config_dword(pvt->pci_ddrio,
884 HASWELL_DDRCRCLKCONTROLS, ®);
886 if (GET_BITFIELD(reg, 16, 16))
889 pci_read_config_dword(pvt->pci_ta, MCMTR, ®);
890 if (GET_BITFIELD(reg, 14, 14)) {
906 static enum dev_type knl_get_width(struct sbridge_pvt *pvt, u32 mtr)
908 /* for KNL value is fixed */
912 static enum dev_type sbridge_get_width(struct sbridge_pvt *pvt, u32 mtr)
914 /* there's no way to figure out */
918 static enum dev_type __ibridge_get_width(u32 mtr)
940 static enum dev_type ibridge_get_width(struct sbridge_pvt *pvt, u32 mtr)
943 * ddr3_width on the documentation but also valid for DDR4 on
946 return __ibridge_get_width(GET_BITFIELD(mtr, 7, 8));
949 static enum dev_type broadwell_get_width(struct sbridge_pvt *pvt, u32 mtr)
951 /* ddr3_width on the documentation but also valid for DDR4 */
952 return __ibridge_get_width(GET_BITFIELD(mtr, 8, 9));
955 static enum mem_type knl_get_memory_type(struct sbridge_pvt *pvt)
957 /* DDR4 RDIMMS and LRDIMMS are supported */
961 static u8 get_node_id(struct sbridge_pvt *pvt)
964 pci_read_config_dword(pvt->pci_br0, SAD_CONTROL, ®);
965 return GET_BITFIELD(reg, 0, 2);
968 static u8 haswell_get_node_id(struct sbridge_pvt *pvt)
972 pci_read_config_dword(pvt->pci_sad1, SAD_CONTROL, ®);
973 return GET_BITFIELD(reg, 0, 3);
976 static u8 knl_get_node_id(struct sbridge_pvt *pvt)
980 pci_read_config_dword(pvt->pci_sad1, SAD_CONTROL, ®);
981 return GET_BITFIELD(reg, 0, 2);
985 static u64 haswell_get_tolm(struct sbridge_pvt *pvt)
989 pci_read_config_dword(pvt->info.pci_vtd, HASWELL_TOLM, ®);
990 return (GET_BITFIELD(reg, 26, 31) << 26) | 0x3ffffff;
993 static u64 haswell_get_tohm(struct sbridge_pvt *pvt)
998 pci_read_config_dword(pvt->info.pci_vtd, HASWELL_TOHM_0, ®);
999 rc = GET_BITFIELD(reg, 26, 31);
1000 pci_read_config_dword(pvt->info.pci_vtd, HASWELL_TOHM_1, ®);
1001 rc = ((reg << 6) | rc) << 26;
1003 return rc | 0x1ffffff;
1006 static u64 knl_get_tolm(struct sbridge_pvt *pvt)
1010 pci_read_config_dword(pvt->knl.pci_mc_info, KNL_TOLM, ®);
1011 return (GET_BITFIELD(reg, 26, 31) << 26) | 0x3ffffff;
1014 static u64 knl_get_tohm(struct sbridge_pvt *pvt)
1019 pci_read_config_dword(pvt->knl.pci_mc_info, KNL_TOHM_0, ®_lo);
1020 pci_read_config_dword(pvt->knl.pci_mc_info, KNL_TOHM_1, ®_hi);
1021 rc = ((u64)reg_hi << 32) | reg_lo;
1022 return rc | 0x3ffffff;
1026 static u64 haswell_rir_limit(u32 reg)
1028 return (((u64)GET_BITFIELD(reg, 1, 11) + 1) << 29) - 1;
1031 static inline u8 sad_pkg_socket(u8 pkg)
1033 /* on Ivy Bridge, nodeID is SASS, where A is HA and S is node id */
1034 return ((pkg >> 3) << 2) | (pkg & 0x3);
1037 static inline u8 sad_pkg_ha(u8 pkg)
1039 return (pkg >> 2) & 0x1;
1042 static int haswell_chan_hash(int idx, u64 addr)
1047 * XOR even bits from 12:26 to bit0 of idx,
1048 * odd bits from 13:27 to bit1
1050 for (i = 12; i < 28; i += 2)
1051 idx ^= (addr >> i) & 3;
1056 /****************************************************************************
1057 Memory check routines
1058 ****************************************************************************/
1059 static struct pci_dev *get_pdev_same_bus(u8 bus, u32 id)
1061 struct pci_dev *pdev = NULL;
1064 pdev = pci_get_device(PCI_VENDOR_ID_INTEL, id, pdev);
1065 if (pdev && pdev->bus->number == bus)
1073 * check_if_ecc_is_active() - Checks if ECC is active
1075 * @type: Memory controller type
1076 * returns: 0 in case ECC is active, -ENODEV if it can't be determined or
1079 static int check_if_ecc_is_active(const u8 bus, enum type type)
1081 struct pci_dev *pdev = NULL;
1086 id = PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TA;
1089 id = PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TA;
1092 id = PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TA;
1095 id = PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TA;
1097 case KNIGHTS_LANDING:
1099 * KNL doesn't group things by bus the same way
1100 * SB/IB/Haswell does.
1102 id = PCI_DEVICE_ID_INTEL_KNL_IMC_TA;
1108 if (type != KNIGHTS_LANDING)
1109 pdev = get_pdev_same_bus(bus, id);
1111 pdev = pci_get_device(PCI_VENDOR_ID_INTEL, id, 0);
1114 sbridge_printk(KERN_ERR, "Couldn't find PCI device "
1115 "%04x:%04x! on bus %02d\n",
1116 PCI_VENDOR_ID_INTEL, id, bus);
1120 pci_read_config_dword(pdev,
1121 type == KNIGHTS_LANDING ? KNL_MCMTR : MCMTR, &mcmtr);
1122 if (!IS_ECC_ENABLED(mcmtr)) {
1123 sbridge_printk(KERN_ERR, "ECC is disabled. Aborting\n");
1129 /* Low bits of TAD limit, and some metadata. */
1130 static const u32 knl_tad_dram_limit_lo[] = {
1131 0x400, 0x500, 0x600, 0x700,
1132 0x800, 0x900, 0xa00, 0xb00,
1135 /* Low bits of TAD offset. */
1136 static const u32 knl_tad_dram_offset_lo[] = {
1137 0x404, 0x504, 0x604, 0x704,
1138 0x804, 0x904, 0xa04, 0xb04,
1141 /* High 16 bits of TAD limit and offset. */
1142 static const u32 knl_tad_dram_hi[] = {
1143 0x408, 0x508, 0x608, 0x708,
1144 0x808, 0x908, 0xa08, 0xb08,
1147 /* Number of ways a tad entry is interleaved. */
1148 static const u32 knl_tad_ways[] = {
1153 * Retrieve the n'th Target Address Decode table entry
1154 * from the memory controller's TAD table.
1156 * @pvt: driver private data
1157 * @entry: which entry you want to retrieve
1158 * @mc: which memory controller (0 or 1)
1159 * @offset: output tad range offset
1160 * @limit: output address of first byte above tad range
1161 * @ways: output number of interleave ways
1163 * The offset value has curious semantics. It's a sort of running total
1164 * of the sizes of all the memory regions that aren't mapped in this
1167 static int knl_get_tad(const struct sbridge_pvt *pvt,
1174 u32 reg_limit_lo, reg_offset_lo, reg_hi;
1175 struct pci_dev *pci_mc;
1180 pci_mc = pvt->knl.pci_mc0;
1183 pci_mc = pvt->knl.pci_mc1;
1190 pci_read_config_dword(pci_mc,
1191 knl_tad_dram_limit_lo[entry], ®_limit_lo);
1192 pci_read_config_dword(pci_mc,
1193 knl_tad_dram_offset_lo[entry], ®_offset_lo);
1194 pci_read_config_dword(pci_mc,
1195 knl_tad_dram_hi[entry], ®_hi);
1197 /* Is this TAD entry enabled? */
1198 if (!GET_BITFIELD(reg_limit_lo, 0, 0))
1201 way_id = GET_BITFIELD(reg_limit_lo, 3, 5);
1203 if (way_id < ARRAY_SIZE(knl_tad_ways)) {
1204 *ways = knl_tad_ways[way_id];
1207 sbridge_printk(KERN_ERR,
1208 "Unexpected value %d in mc_tad_limit_lo wayness field\n",
1214 * The least significant 6 bits of base and limit are truncated.
1215 * For limit, we fill the missing bits with 1s.
1217 *offset = ((u64) GET_BITFIELD(reg_offset_lo, 6, 31) << 6) |
1218 ((u64) GET_BITFIELD(reg_hi, 0, 15) << 32);
1219 *limit = ((u64) GET_BITFIELD(reg_limit_lo, 6, 31) << 6) | 63 |
1220 ((u64) GET_BITFIELD(reg_hi, 16, 31) << 32);
1225 /* Determine which memory controller is responsible for a given channel. */
1226 static int knl_channel_mc(int channel)
1228 WARN_ON(channel < 0 || channel >= 6);
1230 return channel < 3 ? 1 : 0;
1234 * Get the Nth entry from EDC_ROUTE_TABLE register.
1235 * (This is the per-tile mapping of logical interleave targets to
1236 * physical EDC modules.)
1248 static u32 knl_get_edc_route(int entry, u32 reg)
1250 WARN_ON(entry >= KNL_MAX_EDCS);
1251 return GET_BITFIELD(reg, entry*3, (entry*3)+2);
1255 * Get the Nth entry from MC_ROUTE_TABLE register.
1256 * (This is the per-tile mapping of logical interleave targets to
1257 * physical DRAM channels modules.)
1259 * entry 0: mc 0:2 channel 18:19
1260 * 1: mc 3:5 channel 20:21
1261 * 2: mc 6:8 channel 22:23
1262 * 3: mc 9:11 channel 24:25
1263 * 4: mc 12:14 channel 26:27
1264 * 5: mc 15:17 channel 28:29
1267 * Though we have 3 bits to identify the MC, we should only see
1268 * the values 0 or 1.
1271 static u32 knl_get_mc_route(int entry, u32 reg)
1275 WARN_ON(entry >= KNL_MAX_CHANNELS);
1277 mc = GET_BITFIELD(reg, entry*3, (entry*3)+2);
1278 chan = GET_BITFIELD(reg, (entry*2) + 18, (entry*2) + 18 + 1);
1280 return knl_channel_remap(mc*3 + chan);
1284 * Render the EDC_ROUTE register in human-readable form.
1285 * Output string s should be at least KNL_MAX_EDCS*2 bytes.
1287 static void knl_show_edc_route(u32 reg, char *s)
1291 for (i = 0; i < KNL_MAX_EDCS; i++) {
1292 s[i*2] = knl_get_edc_route(i, reg) + '0';
1296 s[KNL_MAX_EDCS*2 - 1] = '\0';
1300 * Render the MC_ROUTE register in human-readable form.
1301 * Output string s should be at least KNL_MAX_CHANNELS*2 bytes.
1303 static void knl_show_mc_route(u32 reg, char *s)
1307 for (i = 0; i < KNL_MAX_CHANNELS; i++) {
1308 s[i*2] = knl_get_mc_route(i, reg) + '0';
1312 s[KNL_MAX_CHANNELS*2 - 1] = '\0';
1315 #define KNL_EDC_ROUTE 0xb8
1316 #define KNL_MC_ROUTE 0xb4
1318 /* Is this dram rule backed by regular DRAM in flat mode? */
1319 #define KNL_EDRAM(reg) GET_BITFIELD(reg, 29, 29)
1321 /* Is this dram rule cached? */
1322 #define KNL_CACHEABLE(reg) GET_BITFIELD(reg, 28, 28)
1324 /* Is this rule backed by edc ? */
1325 #define KNL_EDRAM_ONLY(reg) GET_BITFIELD(reg, 29, 29)
1327 /* Is this rule backed by DRAM, cacheable in EDRAM? */
1328 #define KNL_CACHEABLE(reg) GET_BITFIELD(reg, 28, 28)
1330 /* Is this rule mod3? */
1331 #define KNL_MOD3(reg) GET_BITFIELD(reg, 27, 27)
1334 * Figure out how big our RAM modules are.
1336 * The DIMMMTR register in KNL doesn't tell us the size of the DIMMs, so we
1337 * have to figure this out from the SAD rules, interleave lists, route tables,
1340 * SAD rules can have holes in them (e.g. the 3G-4G hole), so we have to
1341 * inspect the TAD rules to figure out how large the SAD regions really are.
1343 * When we know the real size of a SAD region and how many ways it's
1344 * interleaved, we know the individual contribution of each channel to
1347 * Finally, we have to check whether each channel participates in each SAD
1350 * Fortunately, KNL only supports one DIMM per channel, so once we know how
1351 * much memory the channel uses, we know the DIMM is at least that large.
1352 * (The BIOS might possibly choose not to map all available memory, in which
1353 * case we will underreport the size of the DIMM.)
1355 * In theory, we could try to determine the EDC sizes as well, but that would
1356 * only work in flat mode, not in cache mode.
1358 * @mc_sizes: Output sizes of channels (must have space for KNL_MAX_CHANNELS
1361 static int knl_get_dimm_capacity(struct sbridge_pvt *pvt, u64 *mc_sizes)
1363 u64 sad_base, sad_size, sad_limit = 0;
1364 u64 tad_base, tad_size, tad_limit, tad_deadspace, tad_livespace;
1367 int intrlv_ways, tad_ways;
1370 u64 sad_actual_size[2]; /* sad size accounting for holes, per mc */
1371 u32 dram_rule, interleave_reg;
1372 u32 mc_route_reg[KNL_MAX_CHAS];
1373 u32 edc_route_reg[KNL_MAX_CHAS];
1375 char edc_route_string[KNL_MAX_EDCS*2];
1376 char mc_route_string[KNL_MAX_CHANNELS*2];
1381 int participants[KNL_MAX_CHANNELS];
1382 int participant_count = 0;
1384 for (i = 0; i < KNL_MAX_CHANNELS; i++)
1387 /* Read the EDC route table in each CHA. */
1389 for (i = 0; i < KNL_MAX_CHAS; i++) {
1390 pci_read_config_dword(pvt->knl.pci_cha[i],
1391 KNL_EDC_ROUTE, &edc_route_reg[i]);
1393 if (i > 0 && edc_route_reg[i] != edc_route_reg[i-1]) {
1394 knl_show_edc_route(edc_route_reg[i-1],
1396 if (cur_reg_start == i-1)
1397 edac_dbg(0, "edc route table for CHA %d: %s\n",
1398 cur_reg_start, edc_route_string);
1400 edac_dbg(0, "edc route table for CHA %d-%d: %s\n",
1401 cur_reg_start, i-1, edc_route_string);
1405 knl_show_edc_route(edc_route_reg[i-1], edc_route_string);
1406 if (cur_reg_start == i-1)
1407 edac_dbg(0, "edc route table for CHA %d: %s\n",
1408 cur_reg_start, edc_route_string);
1410 edac_dbg(0, "edc route table for CHA %d-%d: %s\n",
1411 cur_reg_start, i-1, edc_route_string);
1413 /* Read the MC route table in each CHA. */
1415 for (i = 0; i < KNL_MAX_CHAS; i++) {
1416 pci_read_config_dword(pvt->knl.pci_cha[i],
1417 KNL_MC_ROUTE, &mc_route_reg[i]);
1419 if (i > 0 && mc_route_reg[i] != mc_route_reg[i-1]) {
1420 knl_show_mc_route(mc_route_reg[i-1], mc_route_string);
1421 if (cur_reg_start == i-1)
1422 edac_dbg(0, "mc route table for CHA %d: %s\n",
1423 cur_reg_start, mc_route_string);
1425 edac_dbg(0, "mc route table for CHA %d-%d: %s\n",
1426 cur_reg_start, i-1, mc_route_string);
1430 knl_show_mc_route(mc_route_reg[i-1], mc_route_string);
1431 if (cur_reg_start == i-1)
1432 edac_dbg(0, "mc route table for CHA %d: %s\n",
1433 cur_reg_start, mc_route_string);
1435 edac_dbg(0, "mc route table for CHA %d-%d: %s\n",
1436 cur_reg_start, i-1, mc_route_string);
1438 /* Process DRAM rules */
1439 for (sad_rule = 0; sad_rule < pvt->info.max_sad; sad_rule++) {
1440 /* previous limit becomes the new base */
1441 sad_base = sad_limit;
1443 pci_read_config_dword(pvt->pci_sad0,
1444 pvt->info.dram_rule[sad_rule], &dram_rule);
1446 if (!DRAM_RULE_ENABLE(dram_rule))
1449 edram_only = KNL_EDRAM_ONLY(dram_rule);
1451 sad_limit = pvt->info.sad_limit(dram_rule)+1;
1452 sad_size = sad_limit - sad_base;
1454 pci_read_config_dword(pvt->pci_sad0,
1455 pvt->info.interleave_list[sad_rule], &interleave_reg);
1458 * Find out how many ways this dram rule is interleaved.
1459 * We stop when we see the first channel again.
1461 first_pkg = sad_pkg(pvt->info.interleave_pkg,
1463 for (intrlv_ways = 1; intrlv_ways < 8; intrlv_ways++) {
1464 pkg = sad_pkg(pvt->info.interleave_pkg,
1465 interleave_reg, intrlv_ways);
1467 if ((pkg & 0x8) == 0) {
1469 * 0 bit means memory is non-local,
1470 * which KNL doesn't support
1472 edac_dbg(0, "Unexpected interleave target %d\n",
1477 if (pkg == first_pkg)
1480 if (KNL_MOD3(dram_rule))
1483 edac_dbg(3, "dram rule %d (base 0x%llx, limit 0x%llx), %d way interleave%s\n",
1488 edram_only ? ", EDRAM" : "");
1491 * Find out how big the SAD region really is by iterating
1492 * over TAD tables (SAD regions may contain holes).
1493 * Each memory controller might have a different TAD table, so
1494 * we have to look at both.
1496 * Livespace is the memory that's mapped in this TAD table,
1497 * deadspace is the holes (this could be the MMIO hole, or it
1498 * could be memory that's mapped by the other TAD table but
1501 for (mc = 0; mc < 2; mc++) {
1502 sad_actual_size[mc] = 0;
1505 tad_rule < ARRAY_SIZE(
1506 knl_tad_dram_limit_lo);
1508 if (knl_get_tad(pvt,
1516 tad_size = (tad_limit+1) -
1517 (tad_livespace + tad_deadspace);
1518 tad_livespace += tad_size;
1519 tad_base = (tad_limit+1) - tad_size;
1521 if (tad_base < sad_base) {
1522 if (tad_limit > sad_base)
1523 edac_dbg(0, "TAD region overlaps lower SAD boundary -- TAD tables may be configured incorrectly.\n");
1524 } else if (tad_base < sad_limit) {
1525 if (tad_limit+1 > sad_limit) {
1526 edac_dbg(0, "TAD region overlaps upper SAD boundary -- TAD tables may be configured incorrectly.\n");
1528 /* TAD region is completely inside SAD region */
1529 edac_dbg(3, "TAD region %d 0x%llx - 0x%llx (%lld bytes) table%d\n",
1531 tad_limit, tad_size,
1533 sad_actual_size[mc] += tad_size;
1536 tad_base = tad_limit+1;
1540 for (mc = 0; mc < 2; mc++) {
1541 edac_dbg(3, " total TAD DRAM footprint in table%d : 0x%llx (%lld bytes)\n",
1542 mc, sad_actual_size[mc], sad_actual_size[mc]);
1545 /* Ignore EDRAM rule */
1549 /* Figure out which channels participate in interleave. */
1550 for (channel = 0; channel < KNL_MAX_CHANNELS; channel++)
1551 participants[channel] = 0;
1553 /* For each channel, does at least one CHA have
1554 * this channel mapped to the given target?
1556 for (channel = 0; channel < KNL_MAX_CHANNELS; channel++) {
1557 for (way = 0; way < intrlv_ways; way++) {
1561 if (KNL_MOD3(dram_rule))
1564 target = 0x7 & sad_pkg(
1565 pvt->info.interleave_pkg, interleave_reg, way);
1567 for (cha = 0; cha < KNL_MAX_CHAS; cha++) {
1568 if (knl_get_mc_route(target,
1569 mc_route_reg[cha]) == channel
1570 && !participants[channel]) {
1571 participant_count++;
1572 participants[channel] = 1;
1579 if (participant_count != intrlv_ways)
1580 edac_dbg(0, "participant_count (%d) != interleave_ways (%d): DIMM size may be incorrect\n",
1581 participant_count, intrlv_ways);
1583 for (channel = 0; channel < KNL_MAX_CHANNELS; channel++) {
1584 mc = knl_channel_mc(channel);
1585 if (participants[channel]) {
1586 edac_dbg(4, "mc channel %d contributes %lld bytes via sad entry %d\n",
1588 sad_actual_size[mc]/intrlv_ways,
1590 mc_sizes[channel] +=
1591 sad_actual_size[mc]/intrlv_ways;
1599 static int get_dimm_config(struct mem_ctl_info *mci)
1601 struct sbridge_pvt *pvt = mci->pvt_info;
1602 struct dimm_info *dimm;
1603 unsigned i, j, banks, ranks, rows, cols, npages;
1606 enum edac_type mode;
1607 enum mem_type mtype;
1608 int channels = pvt->info.type == KNIGHTS_LANDING ?
1609 KNL_MAX_CHANNELS : NUM_CHANNELS;
1610 u64 knl_mc_sizes[KNL_MAX_CHANNELS];
1612 if (pvt->info.type == HASWELL || pvt->info.type == BROADWELL) {
1613 pci_read_config_dword(pvt->pci_ha0, HASWELL_HASYSDEFEATURE2, ®);
1614 pvt->is_chan_hash = GET_BITFIELD(reg, 21, 21);
1616 if (pvt->info.type == HASWELL || pvt->info.type == BROADWELL ||
1617 pvt->info.type == KNIGHTS_LANDING)
1618 pci_read_config_dword(pvt->pci_sad1, SAD_TARGET, ®);
1620 pci_read_config_dword(pvt->pci_br0, SAD_TARGET, ®);
1622 if (pvt->info.type == KNIGHTS_LANDING)
1623 pvt->sbridge_dev->source_id = SOURCE_ID_KNL(reg);
1625 pvt->sbridge_dev->source_id = SOURCE_ID(reg);
1627 pvt->sbridge_dev->node_id = pvt->info.get_node_id(pvt);
1628 edac_dbg(0, "mc#%d: Node ID: %d, source ID: %d\n",
1629 pvt->sbridge_dev->mc,
1630 pvt->sbridge_dev->node_id,
1631 pvt->sbridge_dev->source_id);
1633 /* KNL doesn't support mirroring or lockstep,
1634 * and is always closed page
1636 if (pvt->info.type == KNIGHTS_LANDING) {
1637 mode = EDAC_S4ECD4ED;
1638 pvt->is_mirrored = false;
1640 if (knl_get_dimm_capacity(pvt, knl_mc_sizes) != 0)
1643 pci_read_config_dword(pvt->pci_ras, RASENABLES, ®);
1644 if (IS_MIRROR_ENABLED(reg)) {
1645 edac_dbg(0, "Memory mirror is enabled\n");
1646 pvt->is_mirrored = true;
1648 edac_dbg(0, "Memory mirror is disabled\n");
1649 pvt->is_mirrored = false;
1652 pci_read_config_dword(pvt->pci_ta, MCMTR, &pvt->info.mcmtr);
1653 if (IS_LOCKSTEP_ENABLED(pvt->info.mcmtr)) {
1654 edac_dbg(0, "Lockstep is enabled\n");
1655 mode = EDAC_S8ECD8ED;
1656 pvt->is_lockstep = true;
1658 edac_dbg(0, "Lockstep is disabled\n");
1659 mode = EDAC_S4ECD4ED;
1660 pvt->is_lockstep = false;
1662 if (IS_CLOSE_PG(pvt->info.mcmtr)) {
1663 edac_dbg(0, "address map is on closed page mode\n");
1664 pvt->is_close_pg = true;
1666 edac_dbg(0, "address map is on open page mode\n");
1667 pvt->is_close_pg = false;
1671 mtype = pvt->info.get_memory_type(pvt);
1672 if (mtype == MEM_RDDR3 || mtype == MEM_RDDR4)
1673 edac_dbg(0, "Memory is registered\n");
1674 else if (mtype == MEM_UNKNOWN)
1675 edac_dbg(0, "Cannot determine memory type\n");
1677 edac_dbg(0, "Memory is unregistered\n");
1679 if (mtype == MEM_DDR4 || mtype == MEM_RDDR4)
1684 for (i = 0; i < channels; i++) {
1687 int max_dimms_per_channel;
1689 if (pvt->info.type == KNIGHTS_LANDING) {
1690 max_dimms_per_channel = 1;
1691 if (!pvt->knl.pci_channel[i])
1694 max_dimms_per_channel = ARRAY_SIZE(mtr_regs);
1695 if (!pvt->pci_tad[i])
1699 for (j = 0; j < max_dimms_per_channel; j++) {
1700 dimm = EDAC_DIMM_PTR(mci->layers, mci->dimms, mci->n_layers,
1702 if (pvt->info.type == KNIGHTS_LANDING) {
1703 pci_read_config_dword(pvt->knl.pci_channel[i],
1706 pci_read_config_dword(pvt->pci_tad[i],
1709 edac_dbg(4, "Channel #%d MTR%d = %x\n", i, j, mtr);
1710 if (IS_DIMM_PRESENT(mtr)) {
1711 pvt->channel[i].dimms++;
1713 ranks = numrank(pvt->info.type, mtr);
1715 if (pvt->info.type == KNIGHTS_LANDING) {
1716 /* For DDR4, this is fixed. */
1718 rows = knl_mc_sizes[i] /
1719 ((u64) cols * ranks * banks * 8);
1725 size = ((u64)rows * cols * banks * ranks) >> (20 - 3);
1726 npages = MiB_TO_PAGES(size);
1728 edac_dbg(0, "mc#%d: ha %d channel %d, dimm %d, %lld Mb (%d pages) bank: %d, rank: %d, row: %#x, col: %#x\n",
1729 pvt->sbridge_dev->mc, i/4, i%4, j,
1731 banks, ranks, rows, cols);
1733 dimm->nr_pages = npages;
1735 dimm->dtype = pvt->info.get_width(pvt, mtr);
1736 dimm->mtype = mtype;
1737 dimm->edac_mode = mode;
1738 snprintf(dimm->label, sizeof(dimm->label),
1739 "CPU_SrcID#%u_Ha#%u_Chan#%u_DIMM#%u",
1740 pvt->sbridge_dev->source_id, i/4, i%4, j);
1748 static void get_memory_layout(const struct mem_ctl_info *mci)
1750 struct sbridge_pvt *pvt = mci->pvt_info;
1751 int i, j, k, n_sads, n_tads, sad_interl;
1759 * Step 1) Get TOLM/TOHM ranges
1762 pvt->tolm = pvt->info.get_tolm(pvt);
1763 tmp_mb = (1 + pvt->tolm) >> 20;
1765 gb = div_u64_rem(tmp_mb, 1024, &mb);
1766 edac_dbg(0, "TOLM: %u.%03u GB (0x%016Lx)\n",
1767 gb, (mb*1000)/1024, (u64)pvt->tolm);
1769 /* Address range is already 45:25 */
1770 pvt->tohm = pvt->info.get_tohm(pvt);
1771 tmp_mb = (1 + pvt->tohm) >> 20;
1773 gb = div_u64_rem(tmp_mb, 1024, &mb);
1774 edac_dbg(0, "TOHM: %u.%03u GB (0x%016Lx)\n",
1775 gb, (mb*1000)/1024, (u64)pvt->tohm);
1778 * Step 2) Get SAD range and SAD Interleave list
1779 * TAD registers contain the interleave wayness. However, it
1780 * seems simpler to just discover it indirectly, with the
1784 for (n_sads = 0; n_sads < pvt->info.max_sad; n_sads++) {
1785 /* SAD_LIMIT Address range is 45:26 */
1786 pci_read_config_dword(pvt->pci_sad0, pvt->info.dram_rule[n_sads],
1788 limit = pvt->info.sad_limit(reg);
1790 if (!DRAM_RULE_ENABLE(reg))
1796 tmp_mb = (limit + 1) >> 20;
1797 gb = div_u64_rem(tmp_mb, 1024, &mb);
1798 edac_dbg(0, "SAD#%d %s up to %u.%03u GB (0x%016Lx) Interleave: %s reg=0x%08x\n",
1800 show_dram_attr(pvt->info.dram_attr(reg)),
1802 ((u64)tmp_mb) << 20L,
1803 pvt->info.show_interleave_mode(reg),
1807 pci_read_config_dword(pvt->pci_sad0, pvt->info.interleave_list[n_sads],
1809 sad_interl = sad_pkg(pvt->info.interleave_pkg, reg, 0);
1810 for (j = 0; j < 8; j++) {
1811 u32 pkg = sad_pkg(pvt->info.interleave_pkg, reg, j);
1812 if (j > 0 && sad_interl == pkg)
1815 edac_dbg(0, "SAD#%d, interleave #%d: %d\n",
1820 if (pvt->info.type == KNIGHTS_LANDING)
1824 * Step 3) Get TAD range
1827 for (n_tads = 0; n_tads < MAX_TAD; n_tads++) {
1828 pci_read_config_dword(pvt->pci_ha0, tad_dram_rule[n_tads],
1830 limit = TAD_LIMIT(reg);
1833 tmp_mb = (limit + 1) >> 20;
1835 gb = div_u64_rem(tmp_mb, 1024, &mb);
1836 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",
1837 n_tads, gb, (mb*1000)/1024,
1838 ((u64)tmp_mb) << 20L,
1839 (u32)(1 << TAD_SOCK(reg)),
1840 (u32)TAD_CH(reg) + 1,
1850 * Step 4) Get TAD offsets, per each channel
1852 for (i = 0; i < NUM_CHANNELS; i++) {
1853 if (!pvt->channel[i].dimms)
1855 for (j = 0; j < n_tads; j++) {
1856 pci_read_config_dword(pvt->pci_tad[i],
1857 tad_ch_nilv_offset[j],
1859 tmp_mb = TAD_OFFSET(reg) >> 20;
1860 gb = div_u64_rem(tmp_mb, 1024, &mb);
1861 edac_dbg(0, "TAD CH#%d, offset #%d: %u.%03u GB (0x%016Lx), reg=0x%08x\n",
1864 ((u64)tmp_mb) << 20L,
1870 * Step 6) Get RIR Wayness/Limit, per each channel
1872 for (i = 0; i < NUM_CHANNELS; i++) {
1873 if (!pvt->channel[i].dimms)
1875 for (j = 0; j < MAX_RIR_RANGES; j++) {
1876 pci_read_config_dword(pvt->pci_tad[i],
1880 if (!IS_RIR_VALID(reg))
1883 tmp_mb = pvt->info.rir_limit(reg) >> 20;
1884 rir_way = 1 << RIR_WAY(reg);
1885 gb = div_u64_rem(tmp_mb, 1024, &mb);
1886 edac_dbg(0, "CH#%d RIR#%d, limit: %u.%03u GB (0x%016Lx), way: %d, reg=0x%08x\n",
1889 ((u64)tmp_mb) << 20L,
1893 for (k = 0; k < rir_way; k++) {
1894 pci_read_config_dword(pvt->pci_tad[i],
1897 tmp_mb = RIR_OFFSET(reg) << 6;
1899 gb = div_u64_rem(tmp_mb, 1024, &mb);
1900 edac_dbg(0, "CH#%d RIR#%d INTL#%d, offset %u.%03u GB (0x%016Lx), tgt: %d, reg=0x%08x\n",
1903 ((u64)tmp_mb) << 20L,
1904 (u32)RIR_RNK_TGT(reg),
1911 static struct mem_ctl_info *get_mci_for_node_id(u8 node_id)
1913 struct sbridge_dev *sbridge_dev;
1915 list_for_each_entry(sbridge_dev, &sbridge_edac_list, list) {
1916 if (sbridge_dev->node_id == node_id)
1917 return sbridge_dev->mci;
1922 static int get_memory_error_data(struct mem_ctl_info *mci,
1927 char **area_type, char *msg)
1929 struct mem_ctl_info *new_mci;
1930 struct sbridge_pvt *pvt = mci->pvt_info;
1931 struct pci_dev *pci_ha;
1932 int n_rir, n_sads, n_tads, sad_way, sck_xch;
1933 int sad_interl, idx, base_ch;
1934 int interleave_mode, shiftup = 0;
1935 unsigned sad_interleave[pvt->info.max_interleave];
1937 u8 ch_way, sck_way, pkg, sad_ha = 0, ch_add = 0;
1941 u64 ch_addr, offset, limit = 0, prv = 0;
1945 * Step 0) Check if the address is at special memory ranges
1946 * The check bellow is probably enough to fill all cases where
1947 * the error is not inside a memory, except for the legacy
1948 * range (e. g. VGA addresses). It is unlikely, however, that the
1949 * memory controller would generate an error on that range.
1951 if ((addr > (u64) pvt->tolm) && (addr < (1LL << 32))) {
1952 sprintf(msg, "Error at TOLM area, on addr 0x%08Lx", addr);
1955 if (addr >= (u64)pvt->tohm) {
1956 sprintf(msg, "Error at MMIOH area, on addr 0x%016Lx", addr);
1961 * Step 1) Get socket
1963 for (n_sads = 0; n_sads < pvt->info.max_sad; n_sads++) {
1964 pci_read_config_dword(pvt->pci_sad0, pvt->info.dram_rule[n_sads],
1967 if (!DRAM_RULE_ENABLE(reg))
1970 limit = pvt->info.sad_limit(reg);
1972 sprintf(msg, "Can't discover the memory socket");
1979 if (n_sads == pvt->info.max_sad) {
1980 sprintf(msg, "Can't discover the memory socket");
1984 *area_type = show_dram_attr(pvt->info.dram_attr(dram_rule));
1985 interleave_mode = pvt->info.interleave_mode(dram_rule);
1987 pci_read_config_dword(pvt->pci_sad0, pvt->info.interleave_list[n_sads],
1990 if (pvt->info.type == SANDY_BRIDGE) {
1991 sad_interl = sad_pkg(pvt->info.interleave_pkg, reg, 0);
1992 for (sad_way = 0; sad_way < 8; sad_way++) {
1993 u32 pkg = sad_pkg(pvt->info.interleave_pkg, reg, sad_way);
1994 if (sad_way > 0 && sad_interl == pkg)
1996 sad_interleave[sad_way] = pkg;
1997 edac_dbg(0, "SAD interleave #%d: %d\n",
1998 sad_way, sad_interleave[sad_way]);
2000 edac_dbg(0, "mc#%d: Error detected on SAD#%d: address 0x%016Lx < 0x%016Lx, Interleave [%d:6]%s\n",
2001 pvt->sbridge_dev->mc,
2006 !interleave_mode ? "" : "XOR[18:16]");
2007 if (interleave_mode)
2008 idx = ((addr >> 6) ^ (addr >> 16)) & 7;
2010 idx = (addr >> 6) & 7;
2024 sprintf(msg, "Can't discover socket interleave");
2027 *socket = sad_interleave[idx];
2028 edac_dbg(0, "SAD interleave index: %d (wayness %d) = CPU socket %d\n",
2029 idx, sad_way, *socket);
2030 } else if (pvt->info.type == HASWELL || pvt->info.type == BROADWELL) {
2031 int bits, a7mode = A7MODE(dram_rule);
2034 /* A7 mode swaps P9 with P6 */
2035 bits = GET_BITFIELD(addr, 7, 8) << 1;
2036 bits |= GET_BITFIELD(addr, 9, 9);
2038 bits = GET_BITFIELD(addr, 6, 8);
2040 if (interleave_mode == 0) {
2041 /* interleave mode will XOR {8,7,6} with {18,17,16} */
2042 idx = GET_BITFIELD(addr, 16, 18);
2047 pkg = sad_pkg(pvt->info.interleave_pkg, reg, idx);
2048 *socket = sad_pkg_socket(pkg);
2049 sad_ha = sad_pkg_ha(pkg);
2054 /* MCChanShiftUpEnable */
2055 pci_read_config_dword(pvt->pci_ha0,
2056 HASWELL_HASYSDEFEATURE2, ®);
2057 shiftup = GET_BITFIELD(reg, 22, 22);
2060 edac_dbg(0, "SAD interleave package: %d = CPU socket %d, HA %i, shiftup: %i\n",
2061 idx, *socket, sad_ha, shiftup);
2063 /* Ivy Bridge's SAD mode doesn't support XOR interleave mode */
2064 idx = (addr >> 6) & 7;
2065 pkg = sad_pkg(pvt->info.interleave_pkg, reg, idx);
2066 *socket = sad_pkg_socket(pkg);
2067 sad_ha = sad_pkg_ha(pkg);
2070 edac_dbg(0, "SAD interleave package: %d = CPU socket %d, HA %d\n",
2071 idx, *socket, sad_ha);
2077 * Move to the proper node structure, in order to access the
2078 * right PCI registers
2080 new_mci = get_mci_for_node_id(*socket);
2082 sprintf(msg, "Struct for socket #%u wasn't initialized",
2087 pvt = mci->pvt_info;
2090 * Step 2) Get memory channel
2093 if (pvt->info.type == SANDY_BRIDGE)
2094 pci_ha = pvt->pci_ha0;
2097 pci_ha = pvt->pci_ha1;
2099 pci_ha = pvt->pci_ha0;
2101 for (n_tads = 0; n_tads < MAX_TAD; n_tads++) {
2102 pci_read_config_dword(pci_ha, tad_dram_rule[n_tads], ®);
2103 limit = TAD_LIMIT(reg);
2105 sprintf(msg, "Can't discover the memory channel");
2112 if (n_tads == MAX_TAD) {
2113 sprintf(msg, "Can't discover the memory channel");
2117 ch_way = TAD_CH(reg) + 1;
2118 sck_way = TAD_SOCK(reg);
2123 idx = (addr >> (6 + sck_way + shiftup)) & 0x3;
2124 if (pvt->is_chan_hash)
2125 idx = haswell_chan_hash(idx, addr);
2130 * FIXME: Shouldn't we use CHN_IDX_OFFSET() here, when ch_way == 3 ???
2134 base_ch = TAD_TGT0(reg);
2137 base_ch = TAD_TGT1(reg);
2140 base_ch = TAD_TGT2(reg);
2143 base_ch = TAD_TGT3(reg);
2146 sprintf(msg, "Can't discover the TAD target");
2149 *channel_mask = 1 << base_ch;
2151 pci_read_config_dword(pvt->pci_tad[ch_add + base_ch],
2152 tad_ch_nilv_offset[n_tads],
2155 if (pvt->is_mirrored) {
2156 *channel_mask |= 1 << ((base_ch + 2) % 4);
2160 sck_xch = (1 << sck_way) * (ch_way >> 1);
2163 sprintf(msg, "Invalid mirror set. Can't decode addr");
2167 sck_xch = (1 << sck_way) * ch_way;
2169 if (pvt->is_lockstep)
2170 *channel_mask |= 1 << ((base_ch + 1) % 4);
2172 offset = TAD_OFFSET(tad_offset);
2174 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",
2185 /* Calculate channel address */
2186 /* Remove the TAD offset */
2188 if (offset > addr) {
2189 sprintf(msg, "Can't calculate ch addr: TAD offset 0x%08Lx is too high for addr 0x%08Lx!",
2194 ch_addr = addr - offset;
2195 ch_addr >>= (6 + shiftup);
2197 ch_addr <<= (6 + shiftup);
2198 ch_addr |= addr & ((1 << (6 + shiftup)) - 1);
2201 * Step 3) Decode rank
2203 for (n_rir = 0; n_rir < MAX_RIR_RANGES; n_rir++) {
2204 pci_read_config_dword(pvt->pci_tad[ch_add + base_ch],
2205 rir_way_limit[n_rir],
2208 if (!IS_RIR_VALID(reg))
2211 limit = pvt->info.rir_limit(reg);
2212 gb = div_u64_rem(limit >> 20, 1024, &mb);
2213 edac_dbg(0, "RIR#%d, limit: %u.%03u GB (0x%016Lx), way: %d\n",
2218 if (ch_addr <= limit)
2221 if (n_rir == MAX_RIR_RANGES) {
2222 sprintf(msg, "Can't discover the memory rank for ch addr 0x%08Lx",
2226 rir_way = RIR_WAY(reg);
2228 if (pvt->is_close_pg)
2229 idx = (ch_addr >> 6);
2231 idx = (ch_addr >> 13); /* FIXME: Datasheet says to shift by 15 */
2232 idx %= 1 << rir_way;
2234 pci_read_config_dword(pvt->pci_tad[ch_add + base_ch],
2235 rir_offset[n_rir][idx],
2237 *rank = RIR_RNK_TGT(reg);
2239 edac_dbg(0, "RIR#%d: channel address 0x%08Lx < 0x%08Lx, RIR interleave %d, index %d\n",
2249 /****************************************************************************
2250 Device initialization routines: put/get, init/exit
2251 ****************************************************************************/
2254 * sbridge_put_all_devices 'put' all the devices that we have
2255 * reserved via 'get'
2257 static void sbridge_put_devices(struct sbridge_dev *sbridge_dev)
2262 for (i = 0; i < sbridge_dev->n_devs; i++) {
2263 struct pci_dev *pdev = sbridge_dev->pdev[i];
2266 edac_dbg(0, "Removing dev %02x:%02x.%d\n",
2268 PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn));
2273 static void sbridge_put_all_devices(void)
2275 struct sbridge_dev *sbridge_dev, *tmp;
2277 list_for_each_entry_safe(sbridge_dev, tmp, &sbridge_edac_list, list) {
2278 sbridge_put_devices(sbridge_dev);
2279 free_sbridge_dev(sbridge_dev);
2283 static int sbridge_get_onedevice(struct pci_dev **prev,
2285 const struct pci_id_table *table,
2286 const unsigned devno,
2287 const int multi_bus)
2289 struct sbridge_dev *sbridge_dev;
2290 const struct pci_id_descr *dev_descr = &table->descr[devno];
2291 struct pci_dev *pdev = NULL;
2294 sbridge_printk(KERN_DEBUG,
2295 "Seeking for: PCI ID %04x:%04x\n",
2296 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
2298 pdev = pci_get_device(PCI_VENDOR_ID_INTEL,
2299 dev_descr->dev_id, *prev);
2307 if (dev_descr->optional)
2310 /* if the HA wasn't found */
2314 sbridge_printk(KERN_INFO,
2315 "Device not found: %04x:%04x\n",
2316 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
2318 /* End of list, leave */
2321 bus = pdev->bus->number;
2323 sbridge_dev = get_sbridge_dev(bus, multi_bus);
2325 sbridge_dev = alloc_sbridge_dev(bus, table);
2333 if (sbridge_dev->pdev[devno]) {
2334 sbridge_printk(KERN_ERR,
2335 "Duplicated device for %04x:%04x\n",
2336 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
2341 sbridge_dev->pdev[devno] = pdev;
2343 /* Be sure that the device is enabled */
2344 if (unlikely(pci_enable_device(pdev) < 0)) {
2345 sbridge_printk(KERN_ERR,
2346 "Couldn't enable %04x:%04x\n",
2347 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
2351 edac_dbg(0, "Detected %04x:%04x\n",
2352 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
2355 * As stated on drivers/pci/search.c, the reference count for
2356 * @from is always decremented if it is not %NULL. So, as we need
2357 * to get all devices up to null, we need to do a get for the device
2367 * sbridge_get_all_devices - Find and perform 'get' operation on the MCH's
2368 * devices we want to reference for this driver.
2369 * @num_mc: pointer to the memory controllers count, to be incremented in case
2371 * @table: model specific table
2372 * @allow_dups: allow for multiple devices to exist with the same device id
2373 * (as implemented, this isn't expected to work correctly in the
2374 * multi-socket case).
2375 * @multi_bus: don't assume devices on different buses belong to different
2376 * memory controllers.
2378 * returns 0 in case of success or error code
2380 static int sbridge_get_all_devices_full(u8 *num_mc,
2381 const struct pci_id_table *table,
2386 struct pci_dev *pdev = NULL;
2388 while (table && table->descr) {
2389 for (i = 0; i < table->n_devs; i++) {
2390 if (!allow_dups || i == 0 ||
2391 table->descr[i].dev_id !=
2392 table->descr[i-1].dev_id) {
2396 rc = sbridge_get_onedevice(&pdev, num_mc,
2397 table, i, multi_bus);
2403 sbridge_put_all_devices();
2406 } while (pdev && !allow_dups);
2414 #define sbridge_get_all_devices(num_mc, table) \
2415 sbridge_get_all_devices_full(num_mc, table, 0, 0)
2416 #define sbridge_get_all_devices_knl(num_mc, table) \
2417 sbridge_get_all_devices_full(num_mc, table, 1, 1)
2419 static int sbridge_mci_bind_devs(struct mem_ctl_info *mci,
2420 struct sbridge_dev *sbridge_dev)
2422 struct sbridge_pvt *pvt = mci->pvt_info;
2423 struct pci_dev *pdev;
2424 u8 saw_chan_mask = 0;
2427 for (i = 0; i < sbridge_dev->n_devs; i++) {
2428 pdev = sbridge_dev->pdev[i];
2432 switch (pdev->device) {
2433 case PCI_DEVICE_ID_INTEL_SBRIDGE_SAD0:
2434 pvt->pci_sad0 = pdev;
2436 case PCI_DEVICE_ID_INTEL_SBRIDGE_SAD1:
2437 pvt->pci_sad1 = pdev;
2439 case PCI_DEVICE_ID_INTEL_SBRIDGE_BR:
2440 pvt->pci_br0 = pdev;
2442 case PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_HA0:
2443 pvt->pci_ha0 = pdev;
2445 case PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TA:
2448 case PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_RAS:
2449 pvt->pci_ras = pdev;
2451 case PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD0:
2452 case PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD1:
2453 case PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD2:
2454 case PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD3:
2456 int id = pdev->device - PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD0;
2457 pvt->pci_tad[id] = pdev;
2458 saw_chan_mask |= 1 << id;
2461 case PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_DDRIO:
2462 pvt->pci_ddrio = pdev;
2468 edac_dbg(0, "Associated PCI %02x:%02x, bus %d with dev = %p\n",
2469 pdev->vendor, pdev->device,
2474 /* Check if everything were registered */
2475 if (!pvt->pci_sad0 || !pvt->pci_sad1 || !pvt->pci_ha0 ||
2476 !pvt-> pci_tad || !pvt->pci_ras || !pvt->pci_ta)
2479 if (saw_chan_mask != 0x0f)
2484 sbridge_printk(KERN_ERR, "Some needed devices are missing\n");
2488 sbridge_printk(KERN_ERR, "Unexpected device %02x:%02x\n",
2489 PCI_VENDOR_ID_INTEL, pdev->device);
2493 static int ibridge_mci_bind_devs(struct mem_ctl_info *mci,
2494 struct sbridge_dev *sbridge_dev)
2496 struct sbridge_pvt *pvt = mci->pvt_info;
2497 struct pci_dev *pdev;
2498 u8 saw_chan_mask = 0;
2501 for (i = 0; i < sbridge_dev->n_devs; i++) {
2502 pdev = sbridge_dev->pdev[i];
2506 switch (pdev->device) {
2507 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0:
2508 pvt->pci_ha0 = pdev;
2510 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TA:
2512 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_RAS:
2513 pvt->pci_ras = pdev;
2515 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD0:
2516 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD1:
2517 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD2:
2518 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD3:
2520 int id = pdev->device - PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD0;
2521 pvt->pci_tad[id] = pdev;
2522 saw_chan_mask |= 1 << id;
2525 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_2HA_DDRIO0:
2526 pvt->pci_ddrio = pdev;
2528 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_1HA_DDRIO0:
2529 pvt->pci_ddrio = pdev;
2531 case PCI_DEVICE_ID_INTEL_IBRIDGE_SAD:
2532 pvt->pci_sad0 = pdev;
2534 case PCI_DEVICE_ID_INTEL_IBRIDGE_BR0:
2535 pvt->pci_br0 = pdev;
2537 case PCI_DEVICE_ID_INTEL_IBRIDGE_BR1:
2538 pvt->pci_br1 = pdev;
2540 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1:
2541 pvt->pci_ha1 = pdev;
2543 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD0:
2544 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD1:
2545 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD2:
2546 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD3:
2548 int id = pdev->device - PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD0 + 4;
2549 pvt->pci_tad[id] = pdev;
2550 saw_chan_mask |= 1 << id;
2557 edac_dbg(0, "Associated PCI %02x.%02d.%d with dev = %p\n",
2559 PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn),
2563 /* Check if everything were registered */
2564 if (!pvt->pci_sad0 || !pvt->pci_ha0 || !pvt->pci_br0 ||
2565 !pvt->pci_br1 || !pvt->pci_tad || !pvt->pci_ras ||
2569 if (saw_chan_mask != 0x0f && /* -EN */
2570 saw_chan_mask != 0x33 && /* -EP */
2571 saw_chan_mask != 0xff) /* -EX */
2576 sbridge_printk(KERN_ERR, "Some needed devices are missing\n");
2580 sbridge_printk(KERN_ERR,
2581 "Unexpected device %02x:%02x\n", PCI_VENDOR_ID_INTEL,
2586 static int haswell_mci_bind_devs(struct mem_ctl_info *mci,
2587 struct sbridge_dev *sbridge_dev)
2589 struct sbridge_pvt *pvt = mci->pvt_info;
2590 struct pci_dev *pdev;
2591 u8 saw_chan_mask = 0;
2594 /* there's only one device per system; not tied to any bus */
2595 if (pvt->info.pci_vtd == NULL)
2596 /* result will be checked later */
2597 pvt->info.pci_vtd = pci_get_device(PCI_VENDOR_ID_INTEL,
2598 PCI_DEVICE_ID_INTEL_HASWELL_IMC_VTD_MISC,
2601 for (i = 0; i < sbridge_dev->n_devs; i++) {
2602 pdev = sbridge_dev->pdev[i];
2606 switch (pdev->device) {
2607 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_CBO_SAD0:
2608 pvt->pci_sad0 = pdev;
2610 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_CBO_SAD1:
2611 pvt->pci_sad1 = pdev;
2613 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0:
2614 pvt->pci_ha0 = pdev;
2616 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TA:
2619 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_THERMAL:
2620 pvt->pci_ras = pdev;
2622 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD0:
2623 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD1:
2624 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD2:
2625 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD3:
2627 int id = pdev->device - PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD0;
2629 pvt->pci_tad[id] = pdev;
2630 saw_chan_mask |= 1 << id;
2633 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD0:
2634 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD1:
2635 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD2:
2636 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD3:
2638 int id = pdev->device - PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD0 + 4;
2640 pvt->pci_tad[id] = pdev;
2641 saw_chan_mask |= 1 << id;
2644 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO0:
2645 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO1:
2646 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO2:
2647 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO3:
2648 if (!pvt->pci_ddrio)
2649 pvt->pci_ddrio = pdev;
2651 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1:
2652 pvt->pci_ha1 = pdev;
2654 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TA:
2655 pvt->pci_ha1_ta = pdev;
2661 edac_dbg(0, "Associated PCI %02x.%02d.%d with dev = %p\n",
2663 PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn),
2667 /* Check if everything were registered */
2668 if (!pvt->pci_sad0 || !pvt->pci_ha0 || !pvt->pci_sad1 ||
2669 !pvt->pci_ras || !pvt->pci_ta || !pvt->info.pci_vtd)
2672 if (saw_chan_mask != 0x0f && /* -EN */
2673 saw_chan_mask != 0x33 && /* -EP */
2674 saw_chan_mask != 0xff) /* -EX */
2679 sbridge_printk(KERN_ERR, "Some needed devices are missing\n");
2683 static int broadwell_mci_bind_devs(struct mem_ctl_info *mci,
2684 struct sbridge_dev *sbridge_dev)
2686 struct sbridge_pvt *pvt = mci->pvt_info;
2687 struct pci_dev *pdev;
2688 u8 saw_chan_mask = 0;
2691 /* there's only one device per system; not tied to any bus */
2692 if (pvt->info.pci_vtd == NULL)
2693 /* result will be checked later */
2694 pvt->info.pci_vtd = pci_get_device(PCI_VENDOR_ID_INTEL,
2695 PCI_DEVICE_ID_INTEL_BROADWELL_IMC_VTD_MISC,
2698 for (i = 0; i < sbridge_dev->n_devs; i++) {
2699 pdev = sbridge_dev->pdev[i];
2703 switch (pdev->device) {
2704 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_CBO_SAD0:
2705 pvt->pci_sad0 = pdev;
2707 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_CBO_SAD1:
2708 pvt->pci_sad1 = pdev;
2710 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0:
2711 pvt->pci_ha0 = pdev;
2713 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TA:
2716 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_THERMAL:
2717 pvt->pci_ras = pdev;
2719 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD0:
2720 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD1:
2721 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD2:
2722 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD3:
2724 int id = pdev->device - PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD0;
2725 pvt->pci_tad[id] = pdev;
2726 saw_chan_mask |= 1 << id;
2729 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD0:
2730 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD1:
2731 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD2:
2732 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD3:
2734 int id = pdev->device - PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD0 + 4;
2735 pvt->pci_tad[id] = pdev;
2736 saw_chan_mask |= 1 << id;
2739 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_DDRIO0:
2740 pvt->pci_ddrio = pdev;
2742 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1:
2743 pvt->pci_ha1 = pdev;
2745 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TA:
2746 pvt->pci_ha1_ta = pdev;
2752 edac_dbg(0, "Associated PCI %02x.%02d.%d with dev = %p\n",
2754 PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn),
2758 /* Check if everything were registered */
2759 if (!pvt->pci_sad0 || !pvt->pci_ha0 || !pvt->pci_sad1 ||
2760 !pvt->pci_ras || !pvt->pci_ta || !pvt->info.pci_vtd)
2763 if (saw_chan_mask != 0x0f && /* -EN */
2764 saw_chan_mask != 0x33 && /* -EP */
2765 saw_chan_mask != 0xff) /* -EX */
2770 sbridge_printk(KERN_ERR, "Some needed devices are missing\n");
2774 static int knl_mci_bind_devs(struct mem_ctl_info *mci,
2775 struct sbridge_dev *sbridge_dev)
2777 struct sbridge_pvt *pvt = mci->pvt_info;
2778 struct pci_dev *pdev;
2784 for (i = 0; i < sbridge_dev->n_devs; i++) {
2785 pdev = sbridge_dev->pdev[i];
2789 /* Extract PCI device and function. */
2790 dev = (pdev->devfn >> 3) & 0x1f;
2791 func = pdev->devfn & 0x7;
2793 switch (pdev->device) {
2794 case PCI_DEVICE_ID_INTEL_KNL_IMC_MC:
2796 pvt->knl.pci_mc0 = pdev;
2798 pvt->knl.pci_mc1 = pdev;
2800 sbridge_printk(KERN_ERR,
2801 "Memory controller in unexpected place! (dev %d, fn %d)\n",
2807 case PCI_DEVICE_ID_INTEL_KNL_IMC_SAD0:
2808 pvt->pci_sad0 = pdev;
2811 case PCI_DEVICE_ID_INTEL_KNL_IMC_SAD1:
2812 pvt->pci_sad1 = pdev;
2815 case PCI_DEVICE_ID_INTEL_KNL_IMC_CHA:
2816 /* There are one of these per tile, and range from
2819 devidx = ((dev-14)*8)+func;
2821 if (devidx < 0 || devidx >= KNL_MAX_CHAS) {
2822 sbridge_printk(KERN_ERR,
2823 "Caching and Home Agent in unexpected place! (dev %d, fn %d)\n",
2828 WARN_ON(pvt->knl.pci_cha[devidx] != NULL);
2830 pvt->knl.pci_cha[devidx] = pdev;
2833 case PCI_DEVICE_ID_INTEL_KNL_IMC_CHANNEL:
2837 * MC0 channels 0-2 are device 9 function 2-4,
2838 * MC1 channels 3-5 are device 8 function 2-4.
2844 devidx = 3 + (func-2);
2846 if (devidx < 0 || devidx >= KNL_MAX_CHANNELS) {
2847 sbridge_printk(KERN_ERR,
2848 "DRAM Channel Registers in unexpected place! (dev %d, fn %d)\n",
2853 WARN_ON(pvt->knl.pci_channel[devidx] != NULL);
2854 pvt->knl.pci_channel[devidx] = pdev;
2857 case PCI_DEVICE_ID_INTEL_KNL_IMC_TOLHM:
2858 pvt->knl.pci_mc_info = pdev;
2861 case PCI_DEVICE_ID_INTEL_KNL_IMC_TA:
2866 sbridge_printk(KERN_ERR, "Unexpected device %d\n",
2872 if (!pvt->knl.pci_mc0 || !pvt->knl.pci_mc1 ||
2873 !pvt->pci_sad0 || !pvt->pci_sad1 ||
2878 for (i = 0; i < KNL_MAX_CHANNELS; i++) {
2879 if (!pvt->knl.pci_channel[i]) {
2880 sbridge_printk(KERN_ERR, "Missing channel %d\n", i);
2885 for (i = 0; i < KNL_MAX_CHAS; i++) {
2886 if (!pvt->knl.pci_cha[i]) {
2887 sbridge_printk(KERN_ERR, "Missing CHA %d\n", i);
2895 sbridge_printk(KERN_ERR, "Some needed devices are missing\n");
2899 /****************************************************************************
2900 Error check routines
2901 ****************************************************************************/
2904 * While Sandy Bridge has error count registers, SMI BIOS read values from
2905 * and resets the counters. So, they are not reliable for the OS to read
2906 * from them. So, we have no option but to just trust on whatever MCE is
2907 * telling us about the errors.
2909 static void sbridge_mce_output_error(struct mem_ctl_info *mci,
2910 const struct mce *m)
2912 struct mem_ctl_info *new_mci;
2913 struct sbridge_pvt *pvt = mci->pvt_info;
2914 enum hw_event_mc_err_type tp_event;
2915 char *type, *optype, msg[256];
2916 bool ripv = GET_BITFIELD(m->mcgstatus, 0, 0);
2917 bool overflow = GET_BITFIELD(m->status, 62, 62);
2918 bool uncorrected_error = GET_BITFIELD(m->status, 61, 61);
2920 u32 core_err_cnt = GET_BITFIELD(m->status, 38, 52);
2921 u32 mscod = GET_BITFIELD(m->status, 16, 31);
2922 u32 errcode = GET_BITFIELD(m->status, 0, 15);
2923 u32 channel = GET_BITFIELD(m->status, 0, 3);
2924 u32 optypenum = GET_BITFIELD(m->status, 4, 6);
2925 long channel_mask, first_channel;
2926 u8 rank, socket, ha;
2928 char *area_type = NULL;
2930 if (pvt->info.type != SANDY_BRIDGE)
2933 recoverable = GET_BITFIELD(m->status, 56, 56);
2935 if (uncorrected_error) {
2938 tp_event = HW_EVENT_ERR_FATAL;
2941 tp_event = HW_EVENT_ERR_UNCORRECTED;
2945 tp_event = HW_EVENT_ERR_CORRECTED;
2949 * According with Table 15-9 of the Intel Architecture spec vol 3A,
2950 * memory errors should fit in this mask:
2951 * 000f 0000 1mmm cccc (binary)
2953 * f = Correction Report Filtering Bit. If 1, subsequent errors
2957 * If the mask doesn't match, report an error to the parsing logic
2959 if (! ((errcode & 0xef80) == 0x80)) {
2960 optype = "Can't parse: it is not a mem";
2962 switch (optypenum) {
2964 optype = "generic undef request error";
2967 optype = "memory read error";
2970 optype = "memory write error";
2973 optype = "addr/cmd error";
2976 optype = "memory scrubbing error";
2979 optype = "reserved";
2984 /* Only decode errors with an valid address (ADDRV) */
2985 if (!GET_BITFIELD(m->status, 58, 58))
2988 if (pvt->info.type == KNIGHTS_LANDING) {
2989 if (channel == 14) {
2990 edac_dbg(0, "%s%s err_code:%04x:%04x EDRAM bank %d\n",
2991 overflow ? " OVERFLOW" : "",
2992 (uncorrected_error && recoverable)
2993 ? " recoverable" : "",
2999 channel = knl_channel_remap(channel);
3000 channel_mask = 1 << channel;
3001 snprintf(msg, sizeof(msg),
3002 "%s%s err_code:%04x:%04x channel:%d (DIMM_%c)",
3003 overflow ? " OVERFLOW" : "",
3004 (uncorrected_error && recoverable)
3005 ? " recoverable" : " ",
3006 mscod, errcode, channel, A + channel);
3007 edac_mc_handle_error(tp_event, mci, core_err_cnt,
3008 m->addr >> PAGE_SHIFT, m->addr & ~PAGE_MASK, 0,
3014 rc = get_memory_error_data(mci, m->addr, &socket, &ha,
3015 &channel_mask, &rank, &area_type, msg);
3020 new_mci = get_mci_for_node_id(socket);
3022 strcpy(msg, "Error: socket got corrupted!");
3026 pvt = mci->pvt_info;
3028 first_channel = find_first_bit(&channel_mask, NUM_CHANNELS);
3039 * FIXME: On some memory configurations (mirror, lockstep), the
3040 * Memory Controller can't point the error to a single DIMM. The
3041 * EDAC core should be handling the channel mask, in order to point
3042 * to the group of dimm's where the error may be happening.
3044 if (!pvt->is_lockstep && !pvt->is_mirrored && !pvt->is_close_pg)
3045 channel = first_channel;
3047 snprintf(msg, sizeof(msg),
3048 "%s%s area:%s err_code:%04x:%04x socket:%d ha:%d channel_mask:%ld rank:%d",
3049 overflow ? " OVERFLOW" : "",
3050 (uncorrected_error && recoverable) ? " recoverable" : "",
3057 edac_dbg(0, "%s\n", msg);
3059 /* FIXME: need support for channel mask */
3061 if (channel == CHANNEL_UNSPECIFIED)
3064 /* Call the helper to output message */
3065 edac_mc_handle_error(tp_event, mci, core_err_cnt,
3066 m->addr >> PAGE_SHIFT, m->addr & ~PAGE_MASK, 0,
3067 4*ha+channel, dimm, -1,
3071 edac_mc_handle_error(tp_event, mci, core_err_cnt, 0, 0, 0,
3078 * Check that logging is enabled and that this is the right type
3079 * of error for us to handle.
3081 static int sbridge_mce_check_error(struct notifier_block *nb, unsigned long val,
3084 struct mce *mce = (struct mce *)data;
3085 struct mem_ctl_info *mci;
3086 struct sbridge_pvt *pvt;
3089 if (get_edac_report_status() == EDAC_REPORTING_DISABLED)
3092 mci = get_mci_for_node_id(mce->socketid);
3095 pvt = mci->pvt_info;
3098 * Just let mcelog handle it if the error is
3099 * outside the memory controller. A memory error
3100 * is indicated by bit 7 = 1 and bits = 8-11,13-15 = 0.
3101 * bit 12 has an special meaning.
3103 if ((mce->status & 0xefff) >> 7 != 1)
3106 if (mce->mcgstatus & MCG_STATUS_MCIP)
3111 sbridge_mc_printk(mci, KERN_DEBUG, "HANDLING MCE MEMORY ERROR\n");
3113 sbridge_mc_printk(mci, KERN_DEBUG, "CPU %d: Machine Check %s: %Lx "
3114 "Bank %d: %016Lx\n", mce->extcpu, type,
3115 mce->mcgstatus, mce->bank, mce->status);
3116 sbridge_mc_printk(mci, KERN_DEBUG, "TSC %llx ", mce->tsc);
3117 sbridge_mc_printk(mci, KERN_DEBUG, "ADDR %llx ", mce->addr);
3118 sbridge_mc_printk(mci, KERN_DEBUG, "MISC %llx ", mce->misc);
3120 sbridge_mc_printk(mci, KERN_DEBUG, "PROCESSOR %u:%x TIME %llu SOCKET "
3121 "%u APIC %x\n", mce->cpuvendor, mce->cpuid,
3122 mce->time, mce->socketid, mce->apicid);
3124 sbridge_mce_output_error(mci, mce);
3126 /* Advice mcelog that the error were handled */
3130 static struct notifier_block sbridge_mce_dec = {
3131 .notifier_call = sbridge_mce_check_error,
3134 /****************************************************************************
3135 EDAC register/unregister logic
3136 ****************************************************************************/
3138 static void sbridge_unregister_mci(struct sbridge_dev *sbridge_dev)
3140 struct mem_ctl_info *mci = sbridge_dev->mci;
3141 struct sbridge_pvt *pvt;
3143 if (unlikely(!mci || !mci->pvt_info)) {
3144 edac_dbg(0, "MC: dev = %p\n", &sbridge_dev->pdev[0]->dev);
3146 sbridge_printk(KERN_ERR, "Couldn't find mci handler\n");
3150 pvt = mci->pvt_info;
3152 edac_dbg(0, "MC: mci = %p, dev = %p\n",
3153 mci, &sbridge_dev->pdev[0]->dev);
3155 /* Remove MC sysfs nodes */
3156 edac_mc_del_mc(mci->pdev);
3158 edac_dbg(1, "%s: free mci struct\n", mci->ctl_name);
3159 kfree(mci->ctl_name);
3161 sbridge_dev->mci = NULL;
3164 static int sbridge_register_mci(struct sbridge_dev *sbridge_dev, enum type type)
3166 struct mem_ctl_info *mci;
3167 struct edac_mc_layer layers[2];
3168 struct sbridge_pvt *pvt;
3169 struct pci_dev *pdev = sbridge_dev->pdev[0];
3172 /* Check the number of active and not disabled channels */
3173 rc = check_if_ecc_is_active(sbridge_dev->bus, type);
3174 if (unlikely(rc < 0))
3177 /* allocate a new MC control structure */
3178 layers[0].type = EDAC_MC_LAYER_CHANNEL;
3179 layers[0].size = type == KNIGHTS_LANDING ?
3180 KNL_MAX_CHANNELS : NUM_CHANNELS;
3181 layers[0].is_virt_csrow = false;
3182 layers[1].type = EDAC_MC_LAYER_SLOT;
3183 layers[1].size = type == KNIGHTS_LANDING ? 1 : MAX_DIMMS;
3184 layers[1].is_virt_csrow = true;
3185 mci = edac_mc_alloc(sbridge_dev->mc, ARRAY_SIZE(layers), layers,
3191 edac_dbg(0, "MC: mci = %p, dev = %p\n",
3194 pvt = mci->pvt_info;
3195 memset(pvt, 0, sizeof(*pvt));
3197 /* Associate sbridge_dev and mci for future usage */
3198 pvt->sbridge_dev = sbridge_dev;
3199 sbridge_dev->mci = mci;
3201 mci->mtype_cap = type == KNIGHTS_LANDING ?
3202 MEM_FLAG_DDR4 : MEM_FLAG_DDR3;
3203 mci->edac_ctl_cap = EDAC_FLAG_NONE;
3204 mci->edac_cap = EDAC_FLAG_NONE;
3205 mci->mod_name = "sbridge_edac.c";
3206 mci->mod_ver = SBRIDGE_REVISION;
3207 mci->dev_name = pci_name(pdev);
3208 mci->ctl_page_to_phys = NULL;
3210 pvt->info.type = type;
3213 pvt->info.rankcfgr = IB_RANK_CFG_A;
3214 pvt->info.get_tolm = ibridge_get_tolm;
3215 pvt->info.get_tohm = ibridge_get_tohm;
3216 pvt->info.dram_rule = ibridge_dram_rule;
3217 pvt->info.get_memory_type = get_memory_type;
3218 pvt->info.get_node_id = get_node_id;
3219 pvt->info.rir_limit = rir_limit;
3220 pvt->info.sad_limit = sad_limit;
3221 pvt->info.interleave_mode = interleave_mode;
3222 pvt->info.show_interleave_mode = show_interleave_mode;
3223 pvt->info.dram_attr = dram_attr;
3224 pvt->info.max_sad = ARRAY_SIZE(ibridge_dram_rule);
3225 pvt->info.interleave_list = ibridge_interleave_list;
3226 pvt->info.max_interleave = ARRAY_SIZE(ibridge_interleave_list);
3227 pvt->info.interleave_pkg = ibridge_interleave_pkg;
3228 pvt->info.get_width = ibridge_get_width;
3229 mci->ctl_name = kasprintf(GFP_KERNEL, "Ivy Bridge Socket#%d", mci->mc_idx);
3231 /* Store pci devices at mci for faster access */
3232 rc = ibridge_mci_bind_devs(mci, sbridge_dev);
3233 if (unlikely(rc < 0))
3237 pvt->info.rankcfgr = SB_RANK_CFG_A;
3238 pvt->info.get_tolm = sbridge_get_tolm;
3239 pvt->info.get_tohm = sbridge_get_tohm;
3240 pvt->info.dram_rule = sbridge_dram_rule;
3241 pvt->info.get_memory_type = get_memory_type;
3242 pvt->info.get_node_id = get_node_id;
3243 pvt->info.rir_limit = rir_limit;
3244 pvt->info.sad_limit = sad_limit;
3245 pvt->info.interleave_mode = interleave_mode;
3246 pvt->info.show_interleave_mode = show_interleave_mode;
3247 pvt->info.dram_attr = dram_attr;
3248 pvt->info.max_sad = ARRAY_SIZE(sbridge_dram_rule);
3249 pvt->info.interleave_list = sbridge_interleave_list;
3250 pvt->info.max_interleave = ARRAY_SIZE(sbridge_interleave_list);
3251 pvt->info.interleave_pkg = sbridge_interleave_pkg;
3252 pvt->info.get_width = sbridge_get_width;
3253 mci->ctl_name = kasprintf(GFP_KERNEL, "Sandy Bridge Socket#%d", mci->mc_idx);
3255 /* Store pci devices at mci for faster access */
3256 rc = sbridge_mci_bind_devs(mci, sbridge_dev);
3257 if (unlikely(rc < 0))
3261 /* rankcfgr isn't used */
3262 pvt->info.get_tolm = haswell_get_tolm;
3263 pvt->info.get_tohm = haswell_get_tohm;
3264 pvt->info.dram_rule = ibridge_dram_rule;
3265 pvt->info.get_memory_type = haswell_get_memory_type;
3266 pvt->info.get_node_id = haswell_get_node_id;
3267 pvt->info.rir_limit = haswell_rir_limit;
3268 pvt->info.sad_limit = sad_limit;
3269 pvt->info.interleave_mode = interleave_mode;
3270 pvt->info.show_interleave_mode = show_interleave_mode;
3271 pvt->info.dram_attr = dram_attr;
3272 pvt->info.max_sad = ARRAY_SIZE(ibridge_dram_rule);
3273 pvt->info.interleave_list = ibridge_interleave_list;
3274 pvt->info.max_interleave = ARRAY_SIZE(ibridge_interleave_list);
3275 pvt->info.interleave_pkg = ibridge_interleave_pkg;
3276 pvt->info.get_width = ibridge_get_width;
3277 mci->ctl_name = kasprintf(GFP_KERNEL, "Haswell Socket#%d", mci->mc_idx);
3279 /* Store pci devices at mci for faster access */
3280 rc = haswell_mci_bind_devs(mci, sbridge_dev);
3281 if (unlikely(rc < 0))
3285 /* rankcfgr isn't used */
3286 pvt->info.get_tolm = haswell_get_tolm;
3287 pvt->info.get_tohm = haswell_get_tohm;
3288 pvt->info.dram_rule = ibridge_dram_rule;
3289 pvt->info.get_memory_type = haswell_get_memory_type;
3290 pvt->info.get_node_id = haswell_get_node_id;
3291 pvt->info.rir_limit = haswell_rir_limit;
3292 pvt->info.sad_limit = sad_limit;
3293 pvt->info.interleave_mode = interleave_mode;
3294 pvt->info.show_interleave_mode = show_interleave_mode;
3295 pvt->info.dram_attr = dram_attr;
3296 pvt->info.max_sad = ARRAY_SIZE(ibridge_dram_rule);
3297 pvt->info.interleave_list = ibridge_interleave_list;
3298 pvt->info.max_interleave = ARRAY_SIZE(ibridge_interleave_list);
3299 pvt->info.interleave_pkg = ibridge_interleave_pkg;
3300 pvt->info.get_width = broadwell_get_width;
3301 mci->ctl_name = kasprintf(GFP_KERNEL, "Broadwell Socket#%d", mci->mc_idx);
3303 /* Store pci devices at mci for faster access */
3304 rc = broadwell_mci_bind_devs(mci, sbridge_dev);
3305 if (unlikely(rc < 0))
3308 case KNIGHTS_LANDING:
3309 /* pvt->info.rankcfgr == ??? */
3310 pvt->info.get_tolm = knl_get_tolm;
3311 pvt->info.get_tohm = knl_get_tohm;
3312 pvt->info.dram_rule = knl_dram_rule;
3313 pvt->info.get_memory_type = knl_get_memory_type;
3314 pvt->info.get_node_id = knl_get_node_id;
3315 pvt->info.rir_limit = NULL;
3316 pvt->info.sad_limit = knl_sad_limit;
3317 pvt->info.interleave_mode = knl_interleave_mode;
3318 pvt->info.show_interleave_mode = knl_show_interleave_mode;
3319 pvt->info.dram_attr = dram_attr_knl;
3320 pvt->info.max_sad = ARRAY_SIZE(knl_dram_rule);
3321 pvt->info.interleave_list = knl_interleave_list;
3322 pvt->info.max_interleave = ARRAY_SIZE(knl_interleave_list);
3323 pvt->info.interleave_pkg = ibridge_interleave_pkg;
3324 pvt->info.get_width = knl_get_width;
3325 mci->ctl_name = kasprintf(GFP_KERNEL,
3326 "Knights Landing Socket#%d", mci->mc_idx);
3328 rc = knl_mci_bind_devs(mci, sbridge_dev);
3329 if (unlikely(rc < 0))
3334 /* Get dimm basic config and the memory layout */
3335 get_dimm_config(mci);
3336 get_memory_layout(mci);
3338 /* record ptr to the generic device */
3339 mci->pdev = &pdev->dev;
3341 /* add this new MC control structure to EDAC's list of MCs */
3342 if (unlikely(edac_mc_add_mc(mci))) {
3343 edac_dbg(0, "MC: failed edac_mc_add_mc()\n");
3351 kfree(mci->ctl_name);
3353 sbridge_dev->mci = NULL;
3357 #define ICPU(model, table) \
3358 { X86_VENDOR_INTEL, 6, model, 0, (unsigned long)&table }
3360 /* Order here must match "enum type" */
3361 static const struct x86_cpu_id sbridge_cpuids[] = {
3362 ICPU(0x2d, pci_dev_descr_sbridge_table), /* SANDY_BRIDGE */
3363 ICPU(0x3e, pci_dev_descr_ibridge_table), /* IVY_BRIDGE */
3364 ICPU(0x3f, pci_dev_descr_haswell_table), /* HASWELL */
3365 ICPU(0x4f, pci_dev_descr_broadwell_table), /* BROADWELL */
3366 ICPU(0x57, pci_dev_descr_knl_table), /* KNIGHTS_LANDING */
3369 MODULE_DEVICE_TABLE(x86cpu, sbridge_cpuids);
3372 * sbridge_probe Get all devices and register memory controllers
3375 * 0 for FOUND a device
3376 * < 0 for error code
3379 static int sbridge_probe(const struct x86_cpu_id *id)
3383 struct sbridge_dev *sbridge_dev;
3384 struct pci_id_table *ptable = (struct pci_id_table *)id->driver_data;
3386 /* get the pci devices we want to reserve for our use */
3387 rc = sbridge_get_all_devices(&num_mc, ptable);
3389 if (unlikely(rc < 0)) {
3390 edac_dbg(0, "couldn't get all devices\n");
3396 list_for_each_entry(sbridge_dev, &sbridge_edac_list, list) {
3397 edac_dbg(0, "Registering MC#%d (%d of %d)\n",
3398 mc, mc + 1, num_mc);
3400 sbridge_dev->mc = mc++;
3401 rc = sbridge_register_mci(sbridge_dev, id - sbridge_cpuids);
3402 if (unlikely(rc < 0))
3406 sbridge_printk(KERN_INFO, "%s\n", SBRIDGE_REVISION);
3411 list_for_each_entry(sbridge_dev, &sbridge_edac_list, list)
3412 sbridge_unregister_mci(sbridge_dev);
3414 sbridge_put_all_devices();
3420 * sbridge_remove cleanup
3423 static void sbridge_remove(void)
3425 struct sbridge_dev *sbridge_dev;
3429 list_for_each_entry(sbridge_dev, &sbridge_edac_list, list)
3430 sbridge_unregister_mci(sbridge_dev);
3432 /* Release PCI resources */
3433 sbridge_put_all_devices();
3437 * sbridge_init Module entry function
3438 * Try to initialize this module for its devices
3440 static int __init sbridge_init(void)
3442 const struct x86_cpu_id *id;
3447 id = x86_match_cpu(sbridge_cpuids);
3451 /* Ensure that the OPSTATE is set correctly for POLL or NMI */
3454 rc = sbridge_probe(id);
3457 mce_register_decode_chain(&sbridge_mce_dec);
3458 if (get_edac_report_status() == EDAC_REPORTING_DISABLED)
3459 sbridge_printk(KERN_WARNING, "Loading driver, error reporting disabled.\n");
3463 sbridge_printk(KERN_ERR, "Failed to register device with error %d.\n",
3470 * sbridge_exit() Module exit function
3471 * Unregister the driver
3473 static void __exit sbridge_exit(void)
3477 mce_unregister_decode_chain(&sbridge_mce_dec);
3480 module_init(sbridge_init);
3481 module_exit(sbridge_exit);
3483 module_param(edac_op_state, int, 0444);
3484 MODULE_PARM_DESC(edac_op_state, "EDAC Error Reporting state: 0=Poll,1=NMI");
3486 MODULE_LICENSE("GPL");
3487 MODULE_AUTHOR("Mauro Carvalho Chehab");
3488 MODULE_AUTHOR("Red Hat Inc. (http://www.redhat.com)");
3489 MODULE_DESCRIPTION("MC Driver for Intel Sandy Bridge and Ivy Bridge memory controllers - "