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/intel-family.h>
27 #include <asm/processor.h>
30 #include "edac_module.h"
33 static LIST_HEAD(sbridge_edac_list);
36 * Alter this version for the module when modifications are made
38 #define SBRIDGE_REVISION " Ver: 1.1.1 "
39 #define EDAC_MOD_STR "sbridge_edac"
44 #define sbridge_printk(level, fmt, arg...) \
45 edac_printk(level, "sbridge", fmt, ##arg)
47 #define sbridge_mc_printk(mci, level, fmt, arg...) \
48 edac_mc_chipset_printk(mci, level, "sbridge", fmt, ##arg)
51 * Get a bit field at register value <v>, from bit <lo> to bit <hi>
53 #define GET_BITFIELD(v, lo, hi) \
54 (((v) & GENMASK_ULL(hi, lo)) >> (lo))
56 /* Devices 12 Function 6, Offsets 0x80 to 0xcc */
57 static const u32 sbridge_dram_rule[] = {
58 0x80, 0x88, 0x90, 0x98, 0xa0,
59 0xa8, 0xb0, 0xb8, 0xc0, 0xc8,
62 static const u32 ibridge_dram_rule[] = {
63 0x60, 0x68, 0x70, 0x78, 0x80,
64 0x88, 0x90, 0x98, 0xa0, 0xa8,
65 0xb0, 0xb8, 0xc0, 0xc8, 0xd0,
66 0xd8, 0xe0, 0xe8, 0xf0, 0xf8,
69 static const u32 knl_dram_rule[] = {
70 0x60, 0x68, 0x70, 0x78, 0x80, /* 0-4 */
71 0x88, 0x90, 0x98, 0xa0, 0xa8, /* 5-9 */
72 0xb0, 0xb8, 0xc0, 0xc8, 0xd0, /* 10-14 */
73 0xd8, 0xe0, 0xe8, 0xf0, 0xf8, /* 15-19 */
74 0x100, 0x108, 0x110, 0x118, /* 20-23 */
77 #define DRAM_RULE_ENABLE(reg) GET_BITFIELD(reg, 0, 0)
78 #define A7MODE(reg) GET_BITFIELD(reg, 26, 26)
80 static char *show_dram_attr(u32 attr)
94 static const u32 sbridge_interleave_list[] = {
95 0x84, 0x8c, 0x94, 0x9c, 0xa4,
96 0xac, 0xb4, 0xbc, 0xc4, 0xcc,
99 static const u32 ibridge_interleave_list[] = {
100 0x64, 0x6c, 0x74, 0x7c, 0x84,
101 0x8c, 0x94, 0x9c, 0xa4, 0xac,
102 0xb4, 0xbc, 0xc4, 0xcc, 0xd4,
103 0xdc, 0xe4, 0xec, 0xf4, 0xfc,
106 static const u32 knl_interleave_list[] = {
107 0x64, 0x6c, 0x74, 0x7c, 0x84, /* 0-4 */
108 0x8c, 0x94, 0x9c, 0xa4, 0xac, /* 5-9 */
109 0xb4, 0xbc, 0xc4, 0xcc, 0xd4, /* 10-14 */
110 0xdc, 0xe4, 0xec, 0xf4, 0xfc, /* 15-19 */
111 0x104, 0x10c, 0x114, 0x11c, /* 20-23 */
114 struct interleave_pkg {
119 static const struct interleave_pkg sbridge_interleave_pkg[] = {
130 static const struct interleave_pkg ibridge_interleave_pkg[] = {
141 static inline int sad_pkg(const struct interleave_pkg *table, u32 reg,
144 return GET_BITFIELD(reg, table[interleave].start,
145 table[interleave].end);
148 /* Devices 12 Function 7 */
152 #define HASWELL_TOLM 0xd0
153 #define HASWELL_TOHM_0 0xd4
154 #define HASWELL_TOHM_1 0xd8
155 #define KNL_TOLM 0xd0
156 #define KNL_TOHM_0 0xd4
157 #define KNL_TOHM_1 0xd8
159 #define GET_TOLM(reg) ((GET_BITFIELD(reg, 0, 3) << 28) | 0x3ffffff)
160 #define GET_TOHM(reg) ((GET_BITFIELD(reg, 0, 20) << 25) | 0x3ffffff)
162 /* Device 13 Function 6 */
164 #define SAD_TARGET 0xf0
166 #define SOURCE_ID(reg) GET_BITFIELD(reg, 9, 11)
168 #define SOURCE_ID_KNL(reg) GET_BITFIELD(reg, 12, 14)
170 #define SAD_CONTROL 0xf4
172 /* Device 14 function 0 */
174 static const u32 tad_dram_rule[] = {
175 0x40, 0x44, 0x48, 0x4c,
176 0x50, 0x54, 0x58, 0x5c,
177 0x60, 0x64, 0x68, 0x6c,
179 #define MAX_TAD ARRAY_SIZE(tad_dram_rule)
181 #define TAD_LIMIT(reg) ((GET_BITFIELD(reg, 12, 31) << 26) | 0x3ffffff)
182 #define TAD_SOCK(reg) GET_BITFIELD(reg, 10, 11)
183 #define TAD_CH(reg) GET_BITFIELD(reg, 8, 9)
184 #define TAD_TGT3(reg) GET_BITFIELD(reg, 6, 7)
185 #define TAD_TGT2(reg) GET_BITFIELD(reg, 4, 5)
186 #define TAD_TGT1(reg) GET_BITFIELD(reg, 2, 3)
187 #define TAD_TGT0(reg) GET_BITFIELD(reg, 0, 1)
189 /* Device 15, function 0 */
192 #define KNL_MCMTR 0x624
194 #define IS_ECC_ENABLED(mcmtr) GET_BITFIELD(mcmtr, 2, 2)
195 #define IS_LOCKSTEP_ENABLED(mcmtr) GET_BITFIELD(mcmtr, 1, 1)
196 #define IS_CLOSE_PG(mcmtr) GET_BITFIELD(mcmtr, 0, 0)
198 /* Device 15, function 1 */
200 #define RASENABLES 0xac
201 #define IS_MIRROR_ENABLED(reg) GET_BITFIELD(reg, 0, 0)
203 /* Device 15, functions 2-5 */
205 static const int mtr_regs[] = {
209 static const int knl_mtr_reg = 0xb60;
211 #define RANK_DISABLE(mtr) GET_BITFIELD(mtr, 16, 19)
212 #define IS_DIMM_PRESENT(mtr) GET_BITFIELD(mtr, 14, 14)
213 #define RANK_CNT_BITS(mtr) GET_BITFIELD(mtr, 12, 13)
214 #define RANK_WIDTH_BITS(mtr) GET_BITFIELD(mtr, 2, 4)
215 #define COL_WIDTH_BITS(mtr) GET_BITFIELD(mtr, 0, 1)
217 static const u32 tad_ch_nilv_offset[] = {
218 0x90, 0x94, 0x98, 0x9c,
219 0xa0, 0xa4, 0xa8, 0xac,
220 0xb0, 0xb4, 0xb8, 0xbc,
222 #define CHN_IDX_OFFSET(reg) GET_BITFIELD(reg, 28, 29)
223 #define TAD_OFFSET(reg) (GET_BITFIELD(reg, 6, 25) << 26)
225 static const u32 rir_way_limit[] = {
226 0x108, 0x10c, 0x110, 0x114, 0x118,
228 #define MAX_RIR_RANGES ARRAY_SIZE(rir_way_limit)
230 #define IS_RIR_VALID(reg) GET_BITFIELD(reg, 31, 31)
231 #define RIR_WAY(reg) GET_BITFIELD(reg, 28, 29)
233 #define MAX_RIR_WAY 8
235 static const u32 rir_offset[MAX_RIR_RANGES][MAX_RIR_WAY] = {
236 { 0x120, 0x124, 0x128, 0x12c, 0x130, 0x134, 0x138, 0x13c },
237 { 0x140, 0x144, 0x148, 0x14c, 0x150, 0x154, 0x158, 0x15c },
238 { 0x160, 0x164, 0x168, 0x16c, 0x170, 0x174, 0x178, 0x17c },
239 { 0x180, 0x184, 0x188, 0x18c, 0x190, 0x194, 0x198, 0x19c },
240 { 0x1a0, 0x1a4, 0x1a8, 0x1ac, 0x1b0, 0x1b4, 0x1b8, 0x1bc },
243 #define RIR_RNK_TGT(type, reg) (((type) == BROADWELL) ? \
244 GET_BITFIELD(reg, 20, 23) : GET_BITFIELD(reg, 16, 19))
246 #define RIR_OFFSET(type, reg) (((type) == HASWELL || (type) == BROADWELL) ? \
247 GET_BITFIELD(reg, 2, 15) : GET_BITFIELD(reg, 2, 14))
249 /* Device 16, functions 2-7 */
252 * FIXME: Implement the error count reads directly
255 static const u32 correrrcnt[] = {
256 0x104, 0x108, 0x10c, 0x110,
259 #define RANK_ODD_OV(reg) GET_BITFIELD(reg, 31, 31)
260 #define RANK_ODD_ERR_CNT(reg) GET_BITFIELD(reg, 16, 30)
261 #define RANK_EVEN_OV(reg) GET_BITFIELD(reg, 15, 15)
262 #define RANK_EVEN_ERR_CNT(reg) GET_BITFIELD(reg, 0, 14)
264 static const u32 correrrthrsld[] = {
265 0x11c, 0x120, 0x124, 0x128,
268 #define RANK_ODD_ERR_THRSLD(reg) GET_BITFIELD(reg, 16, 30)
269 #define RANK_EVEN_ERR_THRSLD(reg) GET_BITFIELD(reg, 0, 14)
272 /* Device 17, function 0 */
274 #define SB_RANK_CFG_A 0x0328
276 #define IB_RANK_CFG_A 0x0320
282 #define NUM_CHANNELS 8 /* 2MC per socket, four chan per MC */
283 #define MAX_DIMMS 3 /* Max DIMMS per channel */
284 #define KNL_MAX_CHAS 38 /* KNL max num. of Cache Home Agents */
285 #define KNL_MAX_CHANNELS 6 /* KNL max num. of PCI channels */
286 #define KNL_MAX_EDCS 8 /* Embedded DRAM controllers */
287 #define CHANNEL_UNSPECIFIED 0xf /* Intel IA32 SDM 15-14 */
298 struct sbridge_info {
302 u64 (*get_tolm)(struct sbridge_pvt *pvt);
303 u64 (*get_tohm)(struct sbridge_pvt *pvt);
304 u64 (*rir_limit)(u32 reg);
305 u64 (*sad_limit)(u32 reg);
306 u32 (*interleave_mode)(u32 reg);
307 u32 (*dram_attr)(u32 reg);
308 const u32 *dram_rule;
309 const u32 *interleave_list;
310 const struct interleave_pkg *interleave_pkg;
313 u8 (*get_node_id)(struct sbridge_pvt *pvt);
314 enum mem_type (*get_memory_type)(struct sbridge_pvt *pvt);
315 enum dev_type (*get_width)(struct sbridge_pvt *pvt, u32 mtr);
316 struct pci_dev *pci_vtd;
319 struct sbridge_channel {
324 struct pci_id_descr {
329 struct pci_id_table {
330 const struct pci_id_descr *descr;
336 struct list_head list;
338 u8 node_id, source_id;
339 struct pci_dev **pdev;
341 struct mem_ctl_info *mci;
345 struct pci_dev *pci_cha[KNL_MAX_CHAS];
346 struct pci_dev *pci_channel[KNL_MAX_CHANNELS];
347 struct pci_dev *pci_mc0;
348 struct pci_dev *pci_mc1;
349 struct pci_dev *pci_mc0_misc;
350 struct pci_dev *pci_mc1_misc;
351 struct pci_dev *pci_mc_info; /* tolm, tohm */
355 struct pci_dev *pci_ta, *pci_ddrio, *pci_ras;
356 struct pci_dev *pci_sad0, *pci_sad1;
357 struct pci_dev *pci_ha0, *pci_ha1;
358 struct pci_dev *pci_br0, *pci_br1;
359 struct pci_dev *pci_ha1_ta;
360 struct pci_dev *pci_tad[NUM_CHANNELS];
362 struct sbridge_dev *sbridge_dev;
364 struct sbridge_info info;
365 struct sbridge_channel channel[NUM_CHANNELS];
367 /* Memory type detection */
368 bool is_mirrored, is_lockstep, is_close_pg;
371 /* Memory description */
376 #define PCI_DESCR(device_id, opt) \
377 .dev_id = (device_id), \
380 static const struct pci_id_descr pci_dev_descr_sbridge[] = {
381 /* Processor Home Agent */
382 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_HA0, 0) },
384 /* Memory controller */
385 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TA, 0) },
386 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_RAS, 0) },
387 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD0, 0) },
388 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD1, 0) },
389 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD2, 0) },
390 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD3, 0) },
391 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_DDRIO, 1) },
393 /* System Address Decoder */
394 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_SAD0, 0) },
395 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_SAD1, 0) },
397 /* Broadcast Registers */
398 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_BR, 0) },
401 #define PCI_ID_TABLE_ENTRY(A, T) { \
403 .n_devs = ARRAY_SIZE(A), \
407 static const struct pci_id_table pci_dev_descr_sbridge_table[] = {
408 PCI_ID_TABLE_ENTRY(pci_dev_descr_sbridge, SANDY_BRIDGE),
409 {0,} /* 0 terminated list. */
412 /* This changes depending if 1HA or 2HA:
414 * 0x0eb8 (17.0) is DDRIO0
416 * 0x0ebc (17.4) is DDRIO0
418 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_1HA_DDRIO0 0x0eb8
419 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_2HA_DDRIO0 0x0ebc
422 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0 0x0ea0
423 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TA 0x0ea8
424 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_RAS 0x0e71
425 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD0 0x0eaa
426 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD1 0x0eab
427 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD2 0x0eac
428 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD3 0x0ead
429 #define PCI_DEVICE_ID_INTEL_IBRIDGE_SAD 0x0ec8
430 #define PCI_DEVICE_ID_INTEL_IBRIDGE_BR0 0x0ec9
431 #define PCI_DEVICE_ID_INTEL_IBRIDGE_BR1 0x0eca
432 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1 0x0e60
433 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TA 0x0e68
434 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_RAS 0x0e79
435 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD0 0x0e6a
436 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD1 0x0e6b
437 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD2 0x0e6c
438 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD3 0x0e6d
440 static const struct pci_id_descr pci_dev_descr_ibridge[] = {
441 /* Processor Home Agent */
442 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0, 0) },
444 /* Memory controller */
445 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TA, 0) },
446 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_RAS, 0) },
447 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD0, 0) },
448 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD1, 0) },
449 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD2, 0) },
450 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD3, 0) },
452 /* System Address Decoder */
453 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_SAD, 0) },
455 /* Broadcast Registers */
456 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_BR0, 1) },
457 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_BR1, 0) },
459 /* Optional, mode 2HA */
460 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1, 1) },
462 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TA, 1) },
463 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_RAS, 1) },
465 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD0, 1) },
466 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD1, 1) },
467 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD2, 1) },
468 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD3, 1) },
470 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_1HA_DDRIO0, 1) },
471 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_2HA_DDRIO0, 1) },
474 static const struct pci_id_table pci_dev_descr_ibridge_table[] = {
475 PCI_ID_TABLE_ENTRY(pci_dev_descr_ibridge, IVY_BRIDGE),
476 {0,} /* 0 terminated list. */
479 /* Haswell support */
482 * - 3 DDR3 channels, 2 DPC per channel
485 * - 4 DDR4 channels, 3 DPC per channel
488 * - 4 DDR4 channels, 3 DPC per channel
491 * - each IMC interfaces with a SMI 2 channel
492 * - each SMI channel interfaces with a scalable memory buffer
493 * - each scalable memory buffer supports 4 DDR3/DDR4 channels, 3 DPC
495 #define HASWELL_DDRCRCLKCONTROLS 0xa10 /* Ditto on Broadwell */
496 #define HASWELL_HASYSDEFEATURE2 0x84
497 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_VTD_MISC 0x2f28
498 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0 0x2fa0
499 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1 0x2f60
500 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TA 0x2fa8
501 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_THERMAL 0x2f71
502 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TA 0x2f68
503 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_THERMAL 0x2f79
504 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_CBO_SAD0 0x2ffc
505 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_CBO_SAD1 0x2ffd
506 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD0 0x2faa
507 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD1 0x2fab
508 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD2 0x2fac
509 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD3 0x2fad
510 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD0 0x2f6a
511 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD1 0x2f6b
512 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD2 0x2f6c
513 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD3 0x2f6d
514 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO0 0x2fbd
515 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO1 0x2fbf
516 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO2 0x2fb9
517 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO3 0x2fbb
518 static const struct pci_id_descr pci_dev_descr_haswell[] = {
519 /* first item must be the HA */
520 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0, 0) },
522 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_CBO_SAD0, 0) },
523 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_CBO_SAD1, 0) },
525 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1, 1) },
527 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TA, 0) },
528 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_THERMAL, 0) },
529 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD0, 0) },
530 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD1, 0) },
531 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD2, 1) },
532 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD3, 1) },
534 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO0, 1) },
535 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO1, 1) },
536 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO2, 1) },
537 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO3, 1) },
539 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TA, 1) },
540 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_THERMAL, 1) },
541 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD0, 1) },
542 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD1, 1) },
543 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD2, 1) },
544 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD3, 1) },
547 static const struct pci_id_table pci_dev_descr_haswell_table[] = {
548 PCI_ID_TABLE_ENTRY(pci_dev_descr_haswell, HASWELL),
549 {0,} /* 0 terminated list. */
552 /* Knight's Landing Support */
554 * KNL's memory channels are swizzled between memory controllers.
555 * MC0 is mapped to CH3,4,5 and MC1 is mapped to CH0,1,2
557 #define knl_channel_remap(mc, chan) ((mc) ? (chan) : (chan) + 3)
559 /* Memory controller, TAD tables, error injection - 2-8-0, 2-9-0 (2 of these) */
560 #define PCI_DEVICE_ID_INTEL_KNL_IMC_MC 0x7840
561 /* DRAM channel stuff; bank addrs, dimmmtr, etc.. 2-8-2 - 2-9-4 (6 of these) */
562 #define PCI_DEVICE_ID_INTEL_KNL_IMC_CHANNEL 0x7843
563 /* kdrwdbu TAD limits/offsets, MCMTR - 2-10-1, 2-11-1 (2 of these) */
564 #define PCI_DEVICE_ID_INTEL_KNL_IMC_TA 0x7844
565 /* CHA broadcast registers, dram rules - 1-29-0 (1 of these) */
566 #define PCI_DEVICE_ID_INTEL_KNL_IMC_SAD0 0x782a
567 /* SAD target - 1-29-1 (1 of these) */
568 #define PCI_DEVICE_ID_INTEL_KNL_IMC_SAD1 0x782b
569 /* Caching / Home Agent */
570 #define PCI_DEVICE_ID_INTEL_KNL_IMC_CHA 0x782c
571 /* Device with TOLM and TOHM, 0-5-0 (1 of these) */
572 #define PCI_DEVICE_ID_INTEL_KNL_IMC_TOLHM 0x7810
575 * KNL differs from SB, IB, and Haswell in that it has multiple
576 * instances of the same device with the same device ID, so we handle that
577 * by creating as many copies in the table as we expect to find.
578 * (Like device ID must be grouped together.)
581 static const struct pci_id_descr pci_dev_descr_knl[] = {
582 [0] = { PCI_DESCR(PCI_DEVICE_ID_INTEL_KNL_IMC_SAD0, 0) },
583 [1] = { PCI_DESCR(PCI_DEVICE_ID_INTEL_KNL_IMC_SAD1, 0) },
584 [2 ... 3] = { PCI_DESCR(PCI_DEVICE_ID_INTEL_KNL_IMC_MC, 0)},
585 [4 ... 41] = { PCI_DESCR(PCI_DEVICE_ID_INTEL_KNL_IMC_CHA, 0) },
586 [42 ... 47] = { PCI_DESCR(PCI_DEVICE_ID_INTEL_KNL_IMC_CHANNEL, 0) },
587 [48] = { PCI_DESCR(PCI_DEVICE_ID_INTEL_KNL_IMC_TA, 0) },
588 [49] = { PCI_DESCR(PCI_DEVICE_ID_INTEL_KNL_IMC_TOLHM, 0) },
591 static const struct pci_id_table pci_dev_descr_knl_table[] = {
592 PCI_ID_TABLE_ENTRY(pci_dev_descr_knl, KNIGHTS_LANDING),
601 * - 2 DDR3 channels, 2 DPC per channel
604 * - 4 DDR4 channels, 3 DPC per channel
607 * - 4 DDR4 channels, 3 DPC per channel
610 * - each IMC interfaces with a SMI 2 channel
611 * - each SMI channel interfaces with a scalable memory buffer
612 * - each scalable memory buffer supports 4 DDR3/DDR4 channels, 3 DPC
614 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_VTD_MISC 0x6f28
615 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0 0x6fa0
616 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1 0x6f60
617 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TA 0x6fa8
618 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_THERMAL 0x6f71
619 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TA 0x6f68
620 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_THERMAL 0x6f79
621 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_CBO_SAD0 0x6ffc
622 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_CBO_SAD1 0x6ffd
623 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD0 0x6faa
624 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD1 0x6fab
625 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD2 0x6fac
626 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD3 0x6fad
627 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD0 0x6f6a
628 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD1 0x6f6b
629 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD2 0x6f6c
630 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD3 0x6f6d
631 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_DDRIO0 0x6faf
633 static const struct pci_id_descr pci_dev_descr_broadwell[] = {
634 /* first item must be the HA */
635 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0, 0) },
637 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_CBO_SAD0, 0) },
638 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_CBO_SAD1, 0) },
640 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1, 1) },
642 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TA, 0) },
643 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_THERMAL, 0) },
644 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD0, 0) },
645 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD1, 0) },
646 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD2, 1) },
647 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD3, 1) },
649 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_DDRIO0, 1) },
651 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TA, 1) },
652 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_THERMAL, 1) },
653 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD0, 1) },
654 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD1, 1) },
655 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD2, 1) },
656 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD3, 1) },
659 static const struct pci_id_table pci_dev_descr_broadwell_table[] = {
660 PCI_ID_TABLE_ENTRY(pci_dev_descr_broadwell, BROADWELL),
661 {0,} /* 0 terminated list. */
665 /****************************************************************************
666 Ancillary status routines
667 ****************************************************************************/
669 static inline int numrank(enum type type, u32 mtr)
671 int ranks = (1 << RANK_CNT_BITS(mtr));
674 if (type == HASWELL || type == BROADWELL || type == KNIGHTS_LANDING)
678 edac_dbg(0, "Invalid number of ranks: %d (max = %i) raw value = %x (%04x)\n",
679 ranks, max, (unsigned int)RANK_CNT_BITS(mtr), mtr);
686 static inline int numrow(u32 mtr)
688 int rows = (RANK_WIDTH_BITS(mtr) + 12);
690 if (rows < 13 || rows > 18) {
691 edac_dbg(0, "Invalid number of rows: %d (should be between 14 and 17) raw value = %x (%04x)\n",
692 rows, (unsigned int)RANK_WIDTH_BITS(mtr), mtr);
699 static inline int numcol(u32 mtr)
701 int cols = (COL_WIDTH_BITS(mtr) + 10);
704 edac_dbg(0, "Invalid number of cols: %d (max = 4) raw value = %x (%04x)\n",
705 cols, (unsigned int)COL_WIDTH_BITS(mtr), mtr);
712 static struct sbridge_dev *get_sbridge_dev(u8 bus, int multi_bus)
714 struct sbridge_dev *sbridge_dev;
717 * If we have devices scattered across several busses that pertain
718 * to the same memory controller, we'll lump them all together.
721 return list_first_entry_or_null(&sbridge_edac_list,
722 struct sbridge_dev, list);
725 list_for_each_entry(sbridge_dev, &sbridge_edac_list, list) {
726 if (sbridge_dev->bus == bus)
733 static struct sbridge_dev *alloc_sbridge_dev(u8 bus,
734 const struct pci_id_table *table)
736 struct sbridge_dev *sbridge_dev;
738 sbridge_dev = kzalloc(sizeof(*sbridge_dev), GFP_KERNEL);
742 sbridge_dev->pdev = kzalloc(sizeof(*sbridge_dev->pdev) * table->n_devs,
744 if (!sbridge_dev->pdev) {
749 sbridge_dev->bus = bus;
750 sbridge_dev->n_devs = table->n_devs;
751 list_add_tail(&sbridge_dev->list, &sbridge_edac_list);
756 static void free_sbridge_dev(struct sbridge_dev *sbridge_dev)
758 list_del(&sbridge_dev->list);
759 kfree(sbridge_dev->pdev);
763 static u64 sbridge_get_tolm(struct sbridge_pvt *pvt)
767 /* Address range is 32:28 */
768 pci_read_config_dword(pvt->pci_sad1, TOLM, ®);
769 return GET_TOLM(reg);
772 static u64 sbridge_get_tohm(struct sbridge_pvt *pvt)
776 pci_read_config_dword(pvt->pci_sad1, TOHM, ®);
777 return GET_TOHM(reg);
780 static u64 ibridge_get_tolm(struct sbridge_pvt *pvt)
784 pci_read_config_dword(pvt->pci_br1, TOLM, ®);
786 return GET_TOLM(reg);
789 static u64 ibridge_get_tohm(struct sbridge_pvt *pvt)
793 pci_read_config_dword(pvt->pci_br1, TOHM, ®);
795 return GET_TOHM(reg);
798 static u64 rir_limit(u32 reg)
800 return ((u64)GET_BITFIELD(reg, 1, 10) << 29) | 0x1fffffff;
803 static u64 sad_limit(u32 reg)
805 return (GET_BITFIELD(reg, 6, 25) << 26) | 0x3ffffff;
808 static u32 interleave_mode(u32 reg)
810 return GET_BITFIELD(reg, 1, 1);
813 static u32 dram_attr(u32 reg)
815 return GET_BITFIELD(reg, 2, 3);
818 static u64 knl_sad_limit(u32 reg)
820 return (GET_BITFIELD(reg, 7, 26) << 26) | 0x3ffffff;
823 static u32 knl_interleave_mode(u32 reg)
825 return GET_BITFIELD(reg, 1, 2);
828 static const char * const knl_intlv_mode[] = {
829 "[8:6]", "[10:8]", "[14:12]", "[32:30]"
832 static const char *get_intlv_mode_str(u32 reg, enum type t)
834 if (t == KNIGHTS_LANDING)
835 return knl_intlv_mode[knl_interleave_mode(reg)];
837 return interleave_mode(reg) ? "[8:6]" : "[8:6]XOR[18:16]";
840 static u32 dram_attr_knl(u32 reg)
842 return GET_BITFIELD(reg, 3, 4);
846 static enum mem_type get_memory_type(struct sbridge_pvt *pvt)
851 if (pvt->pci_ddrio) {
852 pci_read_config_dword(pvt->pci_ddrio, pvt->info.rankcfgr,
854 if (GET_BITFIELD(reg, 11, 11))
855 /* FIXME: Can also be LRDIMM */
865 static enum mem_type haswell_get_memory_type(struct sbridge_pvt *pvt)
868 bool registered = false;
869 enum mem_type mtype = MEM_UNKNOWN;
874 pci_read_config_dword(pvt->pci_ddrio,
875 HASWELL_DDRCRCLKCONTROLS, ®);
877 if (GET_BITFIELD(reg, 16, 16))
880 pci_read_config_dword(pvt->pci_ta, MCMTR, ®);
881 if (GET_BITFIELD(reg, 14, 14)) {
897 static enum dev_type knl_get_width(struct sbridge_pvt *pvt, u32 mtr)
899 /* for KNL value is fixed */
903 static enum dev_type sbridge_get_width(struct sbridge_pvt *pvt, u32 mtr)
905 /* there's no way to figure out */
909 static enum dev_type __ibridge_get_width(u32 mtr)
931 static enum dev_type ibridge_get_width(struct sbridge_pvt *pvt, u32 mtr)
934 * ddr3_width on the documentation but also valid for DDR4 on
937 return __ibridge_get_width(GET_BITFIELD(mtr, 7, 8));
940 static enum dev_type broadwell_get_width(struct sbridge_pvt *pvt, u32 mtr)
942 /* ddr3_width on the documentation but also valid for DDR4 */
943 return __ibridge_get_width(GET_BITFIELD(mtr, 8, 9));
946 static enum mem_type knl_get_memory_type(struct sbridge_pvt *pvt)
948 /* DDR4 RDIMMS and LRDIMMS are supported */
952 static u8 get_node_id(struct sbridge_pvt *pvt)
955 pci_read_config_dword(pvt->pci_br0, SAD_CONTROL, ®);
956 return GET_BITFIELD(reg, 0, 2);
959 static u8 haswell_get_node_id(struct sbridge_pvt *pvt)
963 pci_read_config_dword(pvt->pci_sad1, SAD_CONTROL, ®);
964 return GET_BITFIELD(reg, 0, 3);
967 static u8 knl_get_node_id(struct sbridge_pvt *pvt)
971 pci_read_config_dword(pvt->pci_sad1, SAD_CONTROL, ®);
972 return GET_BITFIELD(reg, 0, 2);
976 static u64 haswell_get_tolm(struct sbridge_pvt *pvt)
980 pci_read_config_dword(pvt->info.pci_vtd, HASWELL_TOLM, ®);
981 return (GET_BITFIELD(reg, 26, 31) << 26) | 0x3ffffff;
984 static u64 haswell_get_tohm(struct sbridge_pvt *pvt)
989 pci_read_config_dword(pvt->info.pci_vtd, HASWELL_TOHM_0, ®);
990 rc = GET_BITFIELD(reg, 26, 31);
991 pci_read_config_dword(pvt->info.pci_vtd, HASWELL_TOHM_1, ®);
992 rc = ((reg << 6) | rc) << 26;
994 return rc | 0x1ffffff;
997 static u64 knl_get_tolm(struct sbridge_pvt *pvt)
1001 pci_read_config_dword(pvt->knl.pci_mc_info, KNL_TOLM, ®);
1002 return (GET_BITFIELD(reg, 26, 31) << 26) | 0x3ffffff;
1005 static u64 knl_get_tohm(struct sbridge_pvt *pvt)
1010 pci_read_config_dword(pvt->knl.pci_mc_info, KNL_TOHM_0, ®_lo);
1011 pci_read_config_dword(pvt->knl.pci_mc_info, KNL_TOHM_1, ®_hi);
1012 rc = ((u64)reg_hi << 32) | reg_lo;
1013 return rc | 0x3ffffff;
1017 static u64 haswell_rir_limit(u32 reg)
1019 return (((u64)GET_BITFIELD(reg, 1, 11) + 1) << 29) - 1;
1022 static inline u8 sad_pkg_socket(u8 pkg)
1024 /* on Ivy Bridge, nodeID is SASS, where A is HA and S is node id */
1025 return ((pkg >> 3) << 2) | (pkg & 0x3);
1028 static inline u8 sad_pkg_ha(u8 pkg)
1030 return (pkg >> 2) & 0x1;
1033 static int haswell_chan_hash(int idx, u64 addr)
1038 * XOR even bits from 12:26 to bit0 of idx,
1039 * odd bits from 13:27 to bit1
1041 for (i = 12; i < 28; i += 2)
1042 idx ^= (addr >> i) & 3;
1047 /****************************************************************************
1048 Memory check routines
1049 ****************************************************************************/
1050 static struct pci_dev *get_pdev_same_bus(u8 bus, u32 id)
1052 struct pci_dev *pdev = NULL;
1055 pdev = pci_get_device(PCI_VENDOR_ID_INTEL, id, pdev);
1056 if (pdev && pdev->bus->number == bus)
1064 * check_if_ecc_is_active() - Checks if ECC is active
1066 * @type: Memory controller type
1067 * returns: 0 in case ECC is active, -ENODEV if it can't be determined or
1070 static int check_if_ecc_is_active(const u8 bus, enum type type)
1072 struct pci_dev *pdev = NULL;
1077 id = PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TA;
1080 id = PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TA;
1083 id = PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TA;
1086 id = PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TA;
1088 case KNIGHTS_LANDING:
1090 * KNL doesn't group things by bus the same way
1091 * SB/IB/Haswell does.
1093 id = PCI_DEVICE_ID_INTEL_KNL_IMC_TA;
1099 if (type != KNIGHTS_LANDING)
1100 pdev = get_pdev_same_bus(bus, id);
1102 pdev = pci_get_device(PCI_VENDOR_ID_INTEL, id, 0);
1105 sbridge_printk(KERN_ERR, "Couldn't find PCI device "
1106 "%04x:%04x! on bus %02d\n",
1107 PCI_VENDOR_ID_INTEL, id, bus);
1111 pci_read_config_dword(pdev,
1112 type == KNIGHTS_LANDING ? KNL_MCMTR : MCMTR, &mcmtr);
1113 if (!IS_ECC_ENABLED(mcmtr)) {
1114 sbridge_printk(KERN_ERR, "ECC is disabled. Aborting\n");
1120 /* Low bits of TAD limit, and some metadata. */
1121 static const u32 knl_tad_dram_limit_lo[] = {
1122 0x400, 0x500, 0x600, 0x700,
1123 0x800, 0x900, 0xa00, 0xb00,
1126 /* Low bits of TAD offset. */
1127 static const u32 knl_tad_dram_offset_lo[] = {
1128 0x404, 0x504, 0x604, 0x704,
1129 0x804, 0x904, 0xa04, 0xb04,
1132 /* High 16 bits of TAD limit and offset. */
1133 static const u32 knl_tad_dram_hi[] = {
1134 0x408, 0x508, 0x608, 0x708,
1135 0x808, 0x908, 0xa08, 0xb08,
1138 /* Number of ways a tad entry is interleaved. */
1139 static const u32 knl_tad_ways[] = {
1144 * Retrieve the n'th Target Address Decode table entry
1145 * from the memory controller's TAD table.
1147 * @pvt: driver private data
1148 * @entry: which entry you want to retrieve
1149 * @mc: which memory controller (0 or 1)
1150 * @offset: output tad range offset
1151 * @limit: output address of first byte above tad range
1152 * @ways: output number of interleave ways
1154 * The offset value has curious semantics. It's a sort of running total
1155 * of the sizes of all the memory regions that aren't mapped in this
1158 static int knl_get_tad(const struct sbridge_pvt *pvt,
1165 u32 reg_limit_lo, reg_offset_lo, reg_hi;
1166 struct pci_dev *pci_mc;
1171 pci_mc = pvt->knl.pci_mc0;
1174 pci_mc = pvt->knl.pci_mc1;
1181 pci_read_config_dword(pci_mc,
1182 knl_tad_dram_limit_lo[entry], ®_limit_lo);
1183 pci_read_config_dword(pci_mc,
1184 knl_tad_dram_offset_lo[entry], ®_offset_lo);
1185 pci_read_config_dword(pci_mc,
1186 knl_tad_dram_hi[entry], ®_hi);
1188 /* Is this TAD entry enabled? */
1189 if (!GET_BITFIELD(reg_limit_lo, 0, 0))
1192 way_id = GET_BITFIELD(reg_limit_lo, 3, 5);
1194 if (way_id < ARRAY_SIZE(knl_tad_ways)) {
1195 *ways = knl_tad_ways[way_id];
1198 sbridge_printk(KERN_ERR,
1199 "Unexpected value %d in mc_tad_limit_lo wayness field\n",
1205 * The least significant 6 bits of base and limit are truncated.
1206 * For limit, we fill the missing bits with 1s.
1208 *offset = ((u64) GET_BITFIELD(reg_offset_lo, 6, 31) << 6) |
1209 ((u64) GET_BITFIELD(reg_hi, 0, 15) << 32);
1210 *limit = ((u64) GET_BITFIELD(reg_limit_lo, 6, 31) << 6) | 63 |
1211 ((u64) GET_BITFIELD(reg_hi, 16, 31) << 32);
1216 /* Determine which memory controller is responsible for a given channel. */
1217 static int knl_channel_mc(int channel)
1219 WARN_ON(channel < 0 || channel >= 6);
1221 return channel < 3 ? 1 : 0;
1225 * Get the Nth entry from EDC_ROUTE_TABLE register.
1226 * (This is the per-tile mapping of logical interleave targets to
1227 * physical EDC modules.)
1239 static u32 knl_get_edc_route(int entry, u32 reg)
1241 WARN_ON(entry >= KNL_MAX_EDCS);
1242 return GET_BITFIELD(reg, entry*3, (entry*3)+2);
1246 * Get the Nth entry from MC_ROUTE_TABLE register.
1247 * (This is the per-tile mapping of logical interleave targets to
1248 * physical DRAM channels modules.)
1250 * entry 0: mc 0:2 channel 18:19
1251 * 1: mc 3:5 channel 20:21
1252 * 2: mc 6:8 channel 22:23
1253 * 3: mc 9:11 channel 24:25
1254 * 4: mc 12:14 channel 26:27
1255 * 5: mc 15:17 channel 28:29
1258 * Though we have 3 bits to identify the MC, we should only see
1259 * the values 0 or 1.
1262 static u32 knl_get_mc_route(int entry, u32 reg)
1266 WARN_ON(entry >= KNL_MAX_CHANNELS);
1268 mc = GET_BITFIELD(reg, entry*3, (entry*3)+2);
1269 chan = GET_BITFIELD(reg, (entry*2) + 18, (entry*2) + 18 + 1);
1271 return knl_channel_remap(mc, chan);
1275 * Render the EDC_ROUTE register in human-readable form.
1276 * Output string s should be at least KNL_MAX_EDCS*2 bytes.
1278 static void knl_show_edc_route(u32 reg, char *s)
1282 for (i = 0; i < KNL_MAX_EDCS; i++) {
1283 s[i*2] = knl_get_edc_route(i, reg) + '0';
1287 s[KNL_MAX_EDCS*2 - 1] = '\0';
1291 * Render the MC_ROUTE register in human-readable form.
1292 * Output string s should be at least KNL_MAX_CHANNELS*2 bytes.
1294 static void knl_show_mc_route(u32 reg, char *s)
1298 for (i = 0; i < KNL_MAX_CHANNELS; i++) {
1299 s[i*2] = knl_get_mc_route(i, reg) + '0';
1303 s[KNL_MAX_CHANNELS*2 - 1] = '\0';
1306 #define KNL_EDC_ROUTE 0xb8
1307 #define KNL_MC_ROUTE 0xb4
1309 /* Is this dram rule backed by regular DRAM in flat mode? */
1310 #define KNL_EDRAM(reg) GET_BITFIELD(reg, 29, 29)
1312 /* Is this dram rule cached? */
1313 #define KNL_CACHEABLE(reg) GET_BITFIELD(reg, 28, 28)
1315 /* Is this rule backed by edc ? */
1316 #define KNL_EDRAM_ONLY(reg) GET_BITFIELD(reg, 29, 29)
1318 /* Is this rule backed by DRAM, cacheable in EDRAM? */
1319 #define KNL_CACHEABLE(reg) GET_BITFIELD(reg, 28, 28)
1321 /* Is this rule mod3? */
1322 #define KNL_MOD3(reg) GET_BITFIELD(reg, 27, 27)
1325 * Figure out how big our RAM modules are.
1327 * The DIMMMTR register in KNL doesn't tell us the size of the DIMMs, so we
1328 * have to figure this out from the SAD rules, interleave lists, route tables,
1331 * SAD rules can have holes in them (e.g. the 3G-4G hole), so we have to
1332 * inspect the TAD rules to figure out how large the SAD regions really are.
1334 * When we know the real size of a SAD region and how many ways it's
1335 * interleaved, we know the individual contribution of each channel to
1338 * Finally, we have to check whether each channel participates in each SAD
1341 * Fortunately, KNL only supports one DIMM per channel, so once we know how
1342 * much memory the channel uses, we know the DIMM is at least that large.
1343 * (The BIOS might possibly choose not to map all available memory, in which
1344 * case we will underreport the size of the DIMM.)
1346 * In theory, we could try to determine the EDC sizes as well, but that would
1347 * only work in flat mode, not in cache mode.
1349 * @mc_sizes: Output sizes of channels (must have space for KNL_MAX_CHANNELS
1352 static int knl_get_dimm_capacity(struct sbridge_pvt *pvt, u64 *mc_sizes)
1354 u64 sad_base, sad_size, sad_limit = 0;
1355 u64 tad_base, tad_size, tad_limit, tad_deadspace, tad_livespace;
1358 int intrlv_ways, tad_ways;
1361 u64 sad_actual_size[2]; /* sad size accounting for holes, per mc */
1362 u32 dram_rule, interleave_reg;
1363 u32 mc_route_reg[KNL_MAX_CHAS];
1364 u32 edc_route_reg[KNL_MAX_CHAS];
1366 char edc_route_string[KNL_MAX_EDCS*2];
1367 char mc_route_string[KNL_MAX_CHANNELS*2];
1372 int participants[KNL_MAX_CHANNELS];
1373 int participant_count = 0;
1375 for (i = 0; i < KNL_MAX_CHANNELS; i++)
1378 /* Read the EDC route table in each CHA. */
1380 for (i = 0; i < KNL_MAX_CHAS; i++) {
1381 pci_read_config_dword(pvt->knl.pci_cha[i],
1382 KNL_EDC_ROUTE, &edc_route_reg[i]);
1384 if (i > 0 && edc_route_reg[i] != edc_route_reg[i-1]) {
1385 knl_show_edc_route(edc_route_reg[i-1],
1387 if (cur_reg_start == i-1)
1388 edac_dbg(0, "edc route table for CHA %d: %s\n",
1389 cur_reg_start, edc_route_string);
1391 edac_dbg(0, "edc route table for CHA %d-%d: %s\n",
1392 cur_reg_start, i-1, edc_route_string);
1396 knl_show_edc_route(edc_route_reg[i-1], edc_route_string);
1397 if (cur_reg_start == i-1)
1398 edac_dbg(0, "edc route table for CHA %d: %s\n",
1399 cur_reg_start, edc_route_string);
1401 edac_dbg(0, "edc route table for CHA %d-%d: %s\n",
1402 cur_reg_start, i-1, edc_route_string);
1404 /* Read the MC route table in each CHA. */
1406 for (i = 0; i < KNL_MAX_CHAS; i++) {
1407 pci_read_config_dword(pvt->knl.pci_cha[i],
1408 KNL_MC_ROUTE, &mc_route_reg[i]);
1410 if (i > 0 && mc_route_reg[i] != mc_route_reg[i-1]) {
1411 knl_show_mc_route(mc_route_reg[i-1], mc_route_string);
1412 if (cur_reg_start == i-1)
1413 edac_dbg(0, "mc route table for CHA %d: %s\n",
1414 cur_reg_start, mc_route_string);
1416 edac_dbg(0, "mc route table for CHA %d-%d: %s\n",
1417 cur_reg_start, i-1, mc_route_string);
1421 knl_show_mc_route(mc_route_reg[i-1], mc_route_string);
1422 if (cur_reg_start == i-1)
1423 edac_dbg(0, "mc route table for CHA %d: %s\n",
1424 cur_reg_start, mc_route_string);
1426 edac_dbg(0, "mc route table for CHA %d-%d: %s\n",
1427 cur_reg_start, i-1, mc_route_string);
1429 /* Process DRAM rules */
1430 for (sad_rule = 0; sad_rule < pvt->info.max_sad; sad_rule++) {
1431 /* previous limit becomes the new base */
1432 sad_base = sad_limit;
1434 pci_read_config_dword(pvt->pci_sad0,
1435 pvt->info.dram_rule[sad_rule], &dram_rule);
1437 if (!DRAM_RULE_ENABLE(dram_rule))
1440 edram_only = KNL_EDRAM_ONLY(dram_rule);
1442 sad_limit = pvt->info.sad_limit(dram_rule)+1;
1443 sad_size = sad_limit - sad_base;
1445 pci_read_config_dword(pvt->pci_sad0,
1446 pvt->info.interleave_list[sad_rule], &interleave_reg);
1449 * Find out how many ways this dram rule is interleaved.
1450 * We stop when we see the first channel again.
1452 first_pkg = sad_pkg(pvt->info.interleave_pkg,
1454 for (intrlv_ways = 1; intrlv_ways < 8; intrlv_ways++) {
1455 pkg = sad_pkg(pvt->info.interleave_pkg,
1456 interleave_reg, intrlv_ways);
1458 if ((pkg & 0x8) == 0) {
1460 * 0 bit means memory is non-local,
1461 * which KNL doesn't support
1463 edac_dbg(0, "Unexpected interleave target %d\n",
1468 if (pkg == first_pkg)
1471 if (KNL_MOD3(dram_rule))
1474 edac_dbg(3, "dram rule %d (base 0x%llx, limit 0x%llx), %d way interleave%s\n",
1479 edram_only ? ", EDRAM" : "");
1482 * Find out how big the SAD region really is by iterating
1483 * over TAD tables (SAD regions may contain holes).
1484 * Each memory controller might have a different TAD table, so
1485 * we have to look at both.
1487 * Livespace is the memory that's mapped in this TAD table,
1488 * deadspace is the holes (this could be the MMIO hole, or it
1489 * could be memory that's mapped by the other TAD table but
1492 for (mc = 0; mc < 2; mc++) {
1493 sad_actual_size[mc] = 0;
1496 tad_rule < ARRAY_SIZE(
1497 knl_tad_dram_limit_lo);
1499 if (knl_get_tad(pvt,
1507 tad_size = (tad_limit+1) -
1508 (tad_livespace + tad_deadspace);
1509 tad_livespace += tad_size;
1510 tad_base = (tad_limit+1) - tad_size;
1512 if (tad_base < sad_base) {
1513 if (tad_limit > sad_base)
1514 edac_dbg(0, "TAD region overlaps lower SAD boundary -- TAD tables may be configured incorrectly.\n");
1515 } else if (tad_base < sad_limit) {
1516 if (tad_limit+1 > sad_limit) {
1517 edac_dbg(0, "TAD region overlaps upper SAD boundary -- TAD tables may be configured incorrectly.\n");
1519 /* TAD region is completely inside SAD region */
1520 edac_dbg(3, "TAD region %d 0x%llx - 0x%llx (%lld bytes) table%d\n",
1522 tad_limit, tad_size,
1524 sad_actual_size[mc] += tad_size;
1527 tad_base = tad_limit+1;
1531 for (mc = 0; mc < 2; mc++) {
1532 edac_dbg(3, " total TAD DRAM footprint in table%d : 0x%llx (%lld bytes)\n",
1533 mc, sad_actual_size[mc], sad_actual_size[mc]);
1536 /* Ignore EDRAM rule */
1540 /* Figure out which channels participate in interleave. */
1541 for (channel = 0; channel < KNL_MAX_CHANNELS; channel++)
1542 participants[channel] = 0;
1544 /* For each channel, does at least one CHA have
1545 * this channel mapped to the given target?
1547 for (channel = 0; channel < KNL_MAX_CHANNELS; channel++) {
1548 for (way = 0; way < intrlv_ways; way++) {
1552 if (KNL_MOD3(dram_rule))
1555 target = 0x7 & sad_pkg(
1556 pvt->info.interleave_pkg, interleave_reg, way);
1558 for (cha = 0; cha < KNL_MAX_CHAS; cha++) {
1559 if (knl_get_mc_route(target,
1560 mc_route_reg[cha]) == channel
1561 && !participants[channel]) {
1562 participant_count++;
1563 participants[channel] = 1;
1570 if (participant_count != intrlv_ways)
1571 edac_dbg(0, "participant_count (%d) != interleave_ways (%d): DIMM size may be incorrect\n",
1572 participant_count, intrlv_ways);
1574 for (channel = 0; channel < KNL_MAX_CHANNELS; channel++) {
1575 mc = knl_channel_mc(channel);
1576 if (participants[channel]) {
1577 edac_dbg(4, "mc channel %d contributes %lld bytes via sad entry %d\n",
1579 sad_actual_size[mc]/intrlv_ways,
1581 mc_sizes[channel] +=
1582 sad_actual_size[mc]/intrlv_ways;
1590 static int get_dimm_config(struct mem_ctl_info *mci)
1592 struct sbridge_pvt *pvt = mci->pvt_info;
1593 struct dimm_info *dimm;
1594 unsigned i, j, banks, ranks, rows, cols, npages;
1597 enum edac_type mode;
1598 enum mem_type mtype;
1599 int channels = pvt->info.type == KNIGHTS_LANDING ?
1600 KNL_MAX_CHANNELS : NUM_CHANNELS;
1601 u64 knl_mc_sizes[KNL_MAX_CHANNELS];
1603 if (pvt->info.type == HASWELL || pvt->info.type == BROADWELL) {
1604 pci_read_config_dword(pvt->pci_ha0, HASWELL_HASYSDEFEATURE2, ®);
1605 pvt->is_chan_hash = GET_BITFIELD(reg, 21, 21);
1607 if (pvt->info.type == HASWELL || pvt->info.type == BROADWELL ||
1608 pvt->info.type == KNIGHTS_LANDING)
1609 pci_read_config_dword(pvt->pci_sad1, SAD_TARGET, ®);
1611 pci_read_config_dword(pvt->pci_br0, SAD_TARGET, ®);
1613 if (pvt->info.type == KNIGHTS_LANDING)
1614 pvt->sbridge_dev->source_id = SOURCE_ID_KNL(reg);
1616 pvt->sbridge_dev->source_id = SOURCE_ID(reg);
1618 pvt->sbridge_dev->node_id = pvt->info.get_node_id(pvt);
1619 edac_dbg(0, "mc#%d: Node ID: %d, source ID: %d\n",
1620 pvt->sbridge_dev->mc,
1621 pvt->sbridge_dev->node_id,
1622 pvt->sbridge_dev->source_id);
1624 /* KNL doesn't support mirroring or lockstep,
1625 * and is always closed page
1627 if (pvt->info.type == KNIGHTS_LANDING) {
1628 mode = EDAC_S4ECD4ED;
1629 pvt->is_mirrored = false;
1631 if (knl_get_dimm_capacity(pvt, knl_mc_sizes) != 0)
1634 pci_read_config_dword(pvt->pci_ras, RASENABLES, ®);
1635 if (IS_MIRROR_ENABLED(reg)) {
1636 edac_dbg(0, "Memory mirror is enabled\n");
1637 pvt->is_mirrored = true;
1639 edac_dbg(0, "Memory mirror is disabled\n");
1640 pvt->is_mirrored = false;
1643 pci_read_config_dword(pvt->pci_ta, MCMTR, &pvt->info.mcmtr);
1644 if (IS_LOCKSTEP_ENABLED(pvt->info.mcmtr)) {
1645 edac_dbg(0, "Lockstep is enabled\n");
1646 mode = EDAC_S8ECD8ED;
1647 pvt->is_lockstep = true;
1649 edac_dbg(0, "Lockstep is disabled\n");
1650 mode = EDAC_S4ECD4ED;
1651 pvt->is_lockstep = false;
1653 if (IS_CLOSE_PG(pvt->info.mcmtr)) {
1654 edac_dbg(0, "address map is on closed page mode\n");
1655 pvt->is_close_pg = true;
1657 edac_dbg(0, "address map is on open page mode\n");
1658 pvt->is_close_pg = false;
1662 mtype = pvt->info.get_memory_type(pvt);
1663 if (mtype == MEM_RDDR3 || mtype == MEM_RDDR4)
1664 edac_dbg(0, "Memory is registered\n");
1665 else if (mtype == MEM_UNKNOWN)
1666 edac_dbg(0, "Cannot determine memory type\n");
1668 edac_dbg(0, "Memory is unregistered\n");
1670 if (mtype == MEM_DDR4 || mtype == MEM_RDDR4)
1675 for (i = 0; i < channels; i++) {
1678 int max_dimms_per_channel;
1680 if (pvt->info.type == KNIGHTS_LANDING) {
1681 max_dimms_per_channel = 1;
1682 if (!pvt->knl.pci_channel[i])
1685 max_dimms_per_channel = ARRAY_SIZE(mtr_regs);
1686 if (!pvt->pci_tad[i])
1690 for (j = 0; j < max_dimms_per_channel; j++) {
1691 dimm = EDAC_DIMM_PTR(mci->layers, mci->dimms, mci->n_layers,
1693 if (pvt->info.type == KNIGHTS_LANDING) {
1694 pci_read_config_dword(pvt->knl.pci_channel[i],
1697 pci_read_config_dword(pvt->pci_tad[i],
1700 edac_dbg(4, "Channel #%d MTR%d = %x\n", i, j, mtr);
1701 if (IS_DIMM_PRESENT(mtr)) {
1702 pvt->channel[i].dimms++;
1704 ranks = numrank(pvt->info.type, mtr);
1706 if (pvt->info.type == KNIGHTS_LANDING) {
1707 /* For DDR4, this is fixed. */
1709 rows = knl_mc_sizes[i] /
1710 ((u64) cols * ranks * banks * 8);
1716 size = ((u64)rows * cols * banks * ranks) >> (20 - 3);
1717 npages = MiB_TO_PAGES(size);
1719 edac_dbg(0, "mc#%d: ha %d channel %d, dimm %d, %lld Mb (%d pages) bank: %d, rank: %d, row: %#x, col: %#x\n",
1720 pvt->sbridge_dev->mc, i/4, i%4, j,
1722 banks, ranks, rows, cols);
1724 dimm->nr_pages = npages;
1726 dimm->dtype = pvt->info.get_width(pvt, mtr);
1727 dimm->mtype = mtype;
1728 dimm->edac_mode = mode;
1729 snprintf(dimm->label, sizeof(dimm->label),
1730 "CPU_SrcID#%u_Ha#%u_Chan#%u_DIMM#%u",
1731 pvt->sbridge_dev->source_id, i/4, i%4, j);
1739 static void get_memory_layout(const struct mem_ctl_info *mci)
1741 struct sbridge_pvt *pvt = mci->pvt_info;
1742 int i, j, k, n_sads, n_tads, sad_interl;
1750 * Step 1) Get TOLM/TOHM ranges
1753 pvt->tolm = pvt->info.get_tolm(pvt);
1754 tmp_mb = (1 + pvt->tolm) >> 20;
1756 gb = div_u64_rem(tmp_mb, 1024, &mb);
1757 edac_dbg(0, "TOLM: %u.%03u GB (0x%016Lx)\n",
1758 gb, (mb*1000)/1024, (u64)pvt->tolm);
1760 /* Address range is already 45:25 */
1761 pvt->tohm = pvt->info.get_tohm(pvt);
1762 tmp_mb = (1 + pvt->tohm) >> 20;
1764 gb = div_u64_rem(tmp_mb, 1024, &mb);
1765 edac_dbg(0, "TOHM: %u.%03u GB (0x%016Lx)\n",
1766 gb, (mb*1000)/1024, (u64)pvt->tohm);
1769 * Step 2) Get SAD range and SAD Interleave list
1770 * TAD registers contain the interleave wayness. However, it
1771 * seems simpler to just discover it indirectly, with the
1775 for (n_sads = 0; n_sads < pvt->info.max_sad; n_sads++) {
1776 /* SAD_LIMIT Address range is 45:26 */
1777 pci_read_config_dword(pvt->pci_sad0, pvt->info.dram_rule[n_sads],
1779 limit = pvt->info.sad_limit(reg);
1781 if (!DRAM_RULE_ENABLE(reg))
1787 tmp_mb = (limit + 1) >> 20;
1788 gb = div_u64_rem(tmp_mb, 1024, &mb);
1789 edac_dbg(0, "SAD#%d %s up to %u.%03u GB (0x%016Lx) Interleave: %s reg=0x%08x\n",
1791 show_dram_attr(pvt->info.dram_attr(reg)),
1793 ((u64)tmp_mb) << 20L,
1794 get_intlv_mode_str(reg, pvt->info.type),
1798 pci_read_config_dword(pvt->pci_sad0, pvt->info.interleave_list[n_sads],
1800 sad_interl = sad_pkg(pvt->info.interleave_pkg, reg, 0);
1801 for (j = 0; j < 8; j++) {
1802 u32 pkg = sad_pkg(pvt->info.interleave_pkg, reg, j);
1803 if (j > 0 && sad_interl == pkg)
1806 edac_dbg(0, "SAD#%d, interleave #%d: %d\n",
1811 if (pvt->info.type == KNIGHTS_LANDING)
1815 * Step 3) Get TAD range
1818 for (n_tads = 0; n_tads < MAX_TAD; n_tads++) {
1819 pci_read_config_dword(pvt->pci_ha0, tad_dram_rule[n_tads],
1821 limit = TAD_LIMIT(reg);
1824 tmp_mb = (limit + 1) >> 20;
1826 gb = div_u64_rem(tmp_mb, 1024, &mb);
1827 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",
1828 n_tads, gb, (mb*1000)/1024,
1829 ((u64)tmp_mb) << 20L,
1830 (u32)(1 << TAD_SOCK(reg)),
1831 (u32)TAD_CH(reg) + 1,
1841 * Step 4) Get TAD offsets, per each channel
1843 for (i = 0; i < NUM_CHANNELS; i++) {
1844 if (!pvt->channel[i].dimms)
1846 for (j = 0; j < n_tads; j++) {
1847 pci_read_config_dword(pvt->pci_tad[i],
1848 tad_ch_nilv_offset[j],
1850 tmp_mb = TAD_OFFSET(reg) >> 20;
1851 gb = div_u64_rem(tmp_mb, 1024, &mb);
1852 edac_dbg(0, "TAD CH#%d, offset #%d: %u.%03u GB (0x%016Lx), reg=0x%08x\n",
1855 ((u64)tmp_mb) << 20L,
1861 * Step 6) Get RIR Wayness/Limit, per each channel
1863 for (i = 0; i < NUM_CHANNELS; i++) {
1864 if (!pvt->channel[i].dimms)
1866 for (j = 0; j < MAX_RIR_RANGES; j++) {
1867 pci_read_config_dword(pvt->pci_tad[i],
1871 if (!IS_RIR_VALID(reg))
1874 tmp_mb = pvt->info.rir_limit(reg) >> 20;
1875 rir_way = 1 << RIR_WAY(reg);
1876 gb = div_u64_rem(tmp_mb, 1024, &mb);
1877 edac_dbg(0, "CH#%d RIR#%d, limit: %u.%03u GB (0x%016Lx), way: %d, reg=0x%08x\n",
1880 ((u64)tmp_mb) << 20L,
1884 for (k = 0; k < rir_way; k++) {
1885 pci_read_config_dword(pvt->pci_tad[i],
1888 tmp_mb = RIR_OFFSET(pvt->info.type, reg) << 6;
1890 gb = div_u64_rem(tmp_mb, 1024, &mb);
1891 edac_dbg(0, "CH#%d RIR#%d INTL#%d, offset %u.%03u GB (0x%016Lx), tgt: %d, reg=0x%08x\n",
1894 ((u64)tmp_mb) << 20L,
1895 (u32)RIR_RNK_TGT(pvt->info.type, reg),
1902 static struct mem_ctl_info *get_mci_for_node_id(u8 node_id)
1904 struct sbridge_dev *sbridge_dev;
1906 list_for_each_entry(sbridge_dev, &sbridge_edac_list, list) {
1907 if (sbridge_dev->node_id == node_id)
1908 return sbridge_dev->mci;
1913 static int get_memory_error_data(struct mem_ctl_info *mci,
1918 char **area_type, char *msg)
1920 struct mem_ctl_info *new_mci;
1921 struct sbridge_pvt *pvt = mci->pvt_info;
1922 struct pci_dev *pci_ha;
1923 int n_rir, n_sads, n_tads, sad_way, sck_xch;
1924 int sad_interl, idx, base_ch;
1925 int interleave_mode, shiftup = 0;
1926 unsigned sad_interleave[pvt->info.max_interleave];
1928 u8 ch_way, sck_way, pkg, sad_ha = 0, ch_add = 0;
1932 u64 ch_addr, offset, limit = 0, prv = 0;
1936 * Step 0) Check if the address is at special memory ranges
1937 * The check bellow is probably enough to fill all cases where
1938 * the error is not inside a memory, except for the legacy
1939 * range (e. g. VGA addresses). It is unlikely, however, that the
1940 * memory controller would generate an error on that range.
1942 if ((addr > (u64) pvt->tolm) && (addr < (1LL << 32))) {
1943 sprintf(msg, "Error at TOLM area, on addr 0x%08Lx", addr);
1946 if (addr >= (u64)pvt->tohm) {
1947 sprintf(msg, "Error at MMIOH area, on addr 0x%016Lx", addr);
1952 * Step 1) Get socket
1954 for (n_sads = 0; n_sads < pvt->info.max_sad; n_sads++) {
1955 pci_read_config_dword(pvt->pci_sad0, pvt->info.dram_rule[n_sads],
1958 if (!DRAM_RULE_ENABLE(reg))
1961 limit = pvt->info.sad_limit(reg);
1963 sprintf(msg, "Can't discover the memory socket");
1970 if (n_sads == pvt->info.max_sad) {
1971 sprintf(msg, "Can't discover the memory socket");
1975 *area_type = show_dram_attr(pvt->info.dram_attr(dram_rule));
1976 interleave_mode = pvt->info.interleave_mode(dram_rule);
1978 pci_read_config_dword(pvt->pci_sad0, pvt->info.interleave_list[n_sads],
1981 if (pvt->info.type == SANDY_BRIDGE) {
1982 sad_interl = sad_pkg(pvt->info.interleave_pkg, reg, 0);
1983 for (sad_way = 0; sad_way < 8; sad_way++) {
1984 u32 pkg = sad_pkg(pvt->info.interleave_pkg, reg, sad_way);
1985 if (sad_way > 0 && sad_interl == pkg)
1987 sad_interleave[sad_way] = pkg;
1988 edac_dbg(0, "SAD interleave #%d: %d\n",
1989 sad_way, sad_interleave[sad_way]);
1991 edac_dbg(0, "mc#%d: Error detected on SAD#%d: address 0x%016Lx < 0x%016Lx, Interleave [%d:6]%s\n",
1992 pvt->sbridge_dev->mc,
1997 !interleave_mode ? "" : "XOR[18:16]");
1998 if (interleave_mode)
1999 idx = ((addr >> 6) ^ (addr >> 16)) & 7;
2001 idx = (addr >> 6) & 7;
2015 sprintf(msg, "Can't discover socket interleave");
2018 *socket = sad_interleave[idx];
2019 edac_dbg(0, "SAD interleave index: %d (wayness %d) = CPU socket %d\n",
2020 idx, sad_way, *socket);
2021 } else if (pvt->info.type == HASWELL || pvt->info.type == BROADWELL) {
2022 int bits, a7mode = A7MODE(dram_rule);
2025 /* A7 mode swaps P9 with P6 */
2026 bits = GET_BITFIELD(addr, 7, 8) << 1;
2027 bits |= GET_BITFIELD(addr, 9, 9);
2029 bits = GET_BITFIELD(addr, 6, 8);
2031 if (interleave_mode == 0) {
2032 /* interleave mode will XOR {8,7,6} with {18,17,16} */
2033 idx = GET_BITFIELD(addr, 16, 18);
2038 pkg = sad_pkg(pvt->info.interleave_pkg, reg, idx);
2039 *socket = sad_pkg_socket(pkg);
2040 sad_ha = sad_pkg_ha(pkg);
2045 /* MCChanShiftUpEnable */
2046 pci_read_config_dword(pvt->pci_ha0,
2047 HASWELL_HASYSDEFEATURE2, ®);
2048 shiftup = GET_BITFIELD(reg, 22, 22);
2051 edac_dbg(0, "SAD interleave package: %d = CPU socket %d, HA %i, shiftup: %i\n",
2052 idx, *socket, sad_ha, shiftup);
2054 /* Ivy Bridge's SAD mode doesn't support XOR interleave mode */
2055 idx = (addr >> 6) & 7;
2056 pkg = sad_pkg(pvt->info.interleave_pkg, reg, idx);
2057 *socket = sad_pkg_socket(pkg);
2058 sad_ha = sad_pkg_ha(pkg);
2061 edac_dbg(0, "SAD interleave package: %d = CPU socket %d, HA %d\n",
2062 idx, *socket, sad_ha);
2068 * Move to the proper node structure, in order to access the
2069 * right PCI registers
2071 new_mci = get_mci_for_node_id(*socket);
2073 sprintf(msg, "Struct for socket #%u wasn't initialized",
2078 pvt = mci->pvt_info;
2081 * Step 2) Get memory channel
2084 if (pvt->info.type == SANDY_BRIDGE)
2085 pci_ha = pvt->pci_ha0;
2088 pci_ha = pvt->pci_ha1;
2090 pci_ha = pvt->pci_ha0;
2092 for (n_tads = 0; n_tads < MAX_TAD; n_tads++) {
2093 pci_read_config_dword(pci_ha, tad_dram_rule[n_tads], ®);
2094 limit = TAD_LIMIT(reg);
2096 sprintf(msg, "Can't discover the memory channel");
2103 if (n_tads == MAX_TAD) {
2104 sprintf(msg, "Can't discover the memory channel");
2108 ch_way = TAD_CH(reg) + 1;
2109 sck_way = TAD_SOCK(reg);
2114 idx = (addr >> (6 + sck_way + shiftup)) & 0x3;
2115 if (pvt->is_chan_hash)
2116 idx = haswell_chan_hash(idx, addr);
2121 * FIXME: Shouldn't we use CHN_IDX_OFFSET() here, when ch_way == 3 ???
2125 base_ch = TAD_TGT0(reg);
2128 base_ch = TAD_TGT1(reg);
2131 base_ch = TAD_TGT2(reg);
2134 base_ch = TAD_TGT3(reg);
2137 sprintf(msg, "Can't discover the TAD target");
2140 *channel_mask = 1 << base_ch;
2142 pci_read_config_dword(pvt->pci_tad[ch_add + base_ch],
2143 tad_ch_nilv_offset[n_tads],
2146 if (pvt->is_mirrored) {
2147 *channel_mask |= 1 << ((base_ch + 2) % 4);
2151 sck_xch = (1 << sck_way) * (ch_way >> 1);
2154 sprintf(msg, "Invalid mirror set. Can't decode addr");
2158 sck_xch = (1 << sck_way) * ch_way;
2160 if (pvt->is_lockstep)
2161 *channel_mask |= 1 << ((base_ch + 1) % 4);
2163 offset = TAD_OFFSET(tad_offset);
2165 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",
2176 /* Calculate channel address */
2177 /* Remove the TAD offset */
2179 if (offset > addr) {
2180 sprintf(msg, "Can't calculate ch addr: TAD offset 0x%08Lx is too high for addr 0x%08Lx!",
2185 ch_addr = addr - offset;
2186 ch_addr >>= (6 + shiftup);
2188 ch_addr <<= (6 + shiftup);
2189 ch_addr |= addr & ((1 << (6 + shiftup)) - 1);
2192 * Step 3) Decode rank
2194 for (n_rir = 0; n_rir < MAX_RIR_RANGES; n_rir++) {
2195 pci_read_config_dword(pvt->pci_tad[ch_add + base_ch],
2196 rir_way_limit[n_rir],
2199 if (!IS_RIR_VALID(reg))
2202 limit = pvt->info.rir_limit(reg);
2203 gb = div_u64_rem(limit >> 20, 1024, &mb);
2204 edac_dbg(0, "RIR#%d, limit: %u.%03u GB (0x%016Lx), way: %d\n",
2209 if (ch_addr <= limit)
2212 if (n_rir == MAX_RIR_RANGES) {
2213 sprintf(msg, "Can't discover the memory rank for ch addr 0x%08Lx",
2217 rir_way = RIR_WAY(reg);
2219 if (pvt->is_close_pg)
2220 idx = (ch_addr >> 6);
2222 idx = (ch_addr >> 13); /* FIXME: Datasheet says to shift by 15 */
2223 idx %= 1 << rir_way;
2225 pci_read_config_dword(pvt->pci_tad[ch_add + base_ch],
2226 rir_offset[n_rir][idx],
2228 *rank = RIR_RNK_TGT(pvt->info.type, reg);
2230 edac_dbg(0, "RIR#%d: channel address 0x%08Lx < 0x%08Lx, RIR interleave %d, index %d\n",
2240 /****************************************************************************
2241 Device initialization routines: put/get, init/exit
2242 ****************************************************************************/
2245 * sbridge_put_all_devices 'put' all the devices that we have
2246 * reserved via 'get'
2248 static void sbridge_put_devices(struct sbridge_dev *sbridge_dev)
2253 for (i = 0; i < sbridge_dev->n_devs; i++) {
2254 struct pci_dev *pdev = sbridge_dev->pdev[i];
2257 edac_dbg(0, "Removing dev %02x:%02x.%d\n",
2259 PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn));
2264 static void sbridge_put_all_devices(void)
2266 struct sbridge_dev *sbridge_dev, *tmp;
2268 list_for_each_entry_safe(sbridge_dev, tmp, &sbridge_edac_list, list) {
2269 sbridge_put_devices(sbridge_dev);
2270 free_sbridge_dev(sbridge_dev);
2274 static int sbridge_get_onedevice(struct pci_dev **prev,
2276 const struct pci_id_table *table,
2277 const unsigned devno,
2278 const int multi_bus)
2280 struct sbridge_dev *sbridge_dev;
2281 const struct pci_id_descr *dev_descr = &table->descr[devno];
2282 struct pci_dev *pdev = NULL;
2285 sbridge_printk(KERN_DEBUG,
2286 "Seeking for: PCI ID %04x:%04x\n",
2287 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
2289 pdev = pci_get_device(PCI_VENDOR_ID_INTEL,
2290 dev_descr->dev_id, *prev);
2298 if (dev_descr->optional)
2301 /* if the HA wasn't found */
2305 sbridge_printk(KERN_INFO,
2306 "Device not found: %04x:%04x\n",
2307 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
2309 /* End of list, leave */
2312 bus = pdev->bus->number;
2314 sbridge_dev = get_sbridge_dev(bus, multi_bus);
2316 sbridge_dev = alloc_sbridge_dev(bus, table);
2324 if (sbridge_dev->pdev[devno]) {
2325 sbridge_printk(KERN_ERR,
2326 "Duplicated device for %04x:%04x\n",
2327 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
2332 sbridge_dev->pdev[devno] = pdev;
2334 /* Be sure that the device is enabled */
2335 if (unlikely(pci_enable_device(pdev) < 0)) {
2336 sbridge_printk(KERN_ERR,
2337 "Couldn't enable %04x:%04x\n",
2338 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
2342 edac_dbg(0, "Detected %04x:%04x\n",
2343 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
2346 * As stated on drivers/pci/search.c, the reference count for
2347 * @from is always decremented if it is not %NULL. So, as we need
2348 * to get all devices up to null, we need to do a get for the device
2358 * sbridge_get_all_devices - Find and perform 'get' operation on the MCH's
2359 * devices we want to reference for this driver.
2360 * @num_mc: pointer to the memory controllers count, to be incremented in case
2362 * @table: model specific table
2364 * returns 0 in case of success or error code
2366 static int sbridge_get_all_devices(u8 *num_mc,
2367 const struct pci_id_table *table)
2370 struct pci_dev *pdev = NULL;
2374 if (table->type == KNIGHTS_LANDING)
2375 allow_dups = multi_bus = 1;
2376 while (table && table->descr) {
2377 for (i = 0; i < table->n_devs; i++) {
2378 if (!allow_dups || i == 0 ||
2379 table->descr[i].dev_id !=
2380 table->descr[i-1].dev_id) {
2384 rc = sbridge_get_onedevice(&pdev, num_mc,
2385 table, i, multi_bus);
2391 sbridge_put_all_devices();
2394 } while (pdev && !allow_dups);
2402 static int sbridge_mci_bind_devs(struct mem_ctl_info *mci,
2403 struct sbridge_dev *sbridge_dev)
2405 struct sbridge_pvt *pvt = mci->pvt_info;
2406 struct pci_dev *pdev;
2407 u8 saw_chan_mask = 0;
2410 for (i = 0; i < sbridge_dev->n_devs; i++) {
2411 pdev = sbridge_dev->pdev[i];
2415 switch (pdev->device) {
2416 case PCI_DEVICE_ID_INTEL_SBRIDGE_SAD0:
2417 pvt->pci_sad0 = pdev;
2419 case PCI_DEVICE_ID_INTEL_SBRIDGE_SAD1:
2420 pvt->pci_sad1 = pdev;
2422 case PCI_DEVICE_ID_INTEL_SBRIDGE_BR:
2423 pvt->pci_br0 = pdev;
2425 case PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_HA0:
2426 pvt->pci_ha0 = pdev;
2428 case PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TA:
2431 case PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_RAS:
2432 pvt->pci_ras = pdev;
2434 case PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD0:
2435 case PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD1:
2436 case PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD2:
2437 case PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD3:
2439 int id = pdev->device - PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD0;
2440 pvt->pci_tad[id] = pdev;
2441 saw_chan_mask |= 1 << id;
2444 case PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_DDRIO:
2445 pvt->pci_ddrio = pdev;
2451 edac_dbg(0, "Associated PCI %02x:%02x, bus %d with dev = %p\n",
2452 pdev->vendor, pdev->device,
2457 /* Check if everything were registered */
2458 if (!pvt->pci_sad0 || !pvt->pci_sad1 || !pvt->pci_ha0 ||
2459 !pvt->pci_ras || !pvt->pci_ta)
2462 if (saw_chan_mask != 0x0f)
2467 sbridge_printk(KERN_ERR, "Some needed devices are missing\n");
2471 sbridge_printk(KERN_ERR, "Unexpected device %02x:%02x\n",
2472 PCI_VENDOR_ID_INTEL, pdev->device);
2476 static int ibridge_mci_bind_devs(struct mem_ctl_info *mci,
2477 struct sbridge_dev *sbridge_dev)
2479 struct sbridge_pvt *pvt = mci->pvt_info;
2480 struct pci_dev *pdev;
2481 u8 saw_chan_mask = 0;
2484 for (i = 0; i < sbridge_dev->n_devs; i++) {
2485 pdev = sbridge_dev->pdev[i];
2489 switch (pdev->device) {
2490 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0:
2491 pvt->pci_ha0 = pdev;
2493 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TA:
2495 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_RAS:
2496 pvt->pci_ras = pdev;
2498 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD0:
2499 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD1:
2500 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD2:
2501 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD3:
2503 int id = pdev->device - PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD0;
2504 pvt->pci_tad[id] = pdev;
2505 saw_chan_mask |= 1 << id;
2508 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_2HA_DDRIO0:
2509 pvt->pci_ddrio = pdev;
2511 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_1HA_DDRIO0:
2512 pvt->pci_ddrio = pdev;
2514 case PCI_DEVICE_ID_INTEL_IBRIDGE_SAD:
2515 pvt->pci_sad0 = pdev;
2517 case PCI_DEVICE_ID_INTEL_IBRIDGE_BR0:
2518 pvt->pci_br0 = pdev;
2520 case PCI_DEVICE_ID_INTEL_IBRIDGE_BR1:
2521 pvt->pci_br1 = pdev;
2523 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1:
2524 pvt->pci_ha1 = pdev;
2526 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD0:
2527 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD1:
2528 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD2:
2529 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD3:
2531 int id = pdev->device - PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD0 + 4;
2532 pvt->pci_tad[id] = pdev;
2533 saw_chan_mask |= 1 << id;
2540 edac_dbg(0, "Associated PCI %02x.%02d.%d with dev = %p\n",
2542 PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn),
2546 /* Check if everything were registered */
2547 if (!pvt->pci_sad0 || !pvt->pci_ha0 || !pvt->pci_br0 ||
2548 !pvt->pci_br1 || !pvt->pci_ras || !pvt->pci_ta)
2551 if (saw_chan_mask != 0x0f && /* -EN */
2552 saw_chan_mask != 0x33 && /* -EP */
2553 saw_chan_mask != 0xff) /* -EX */
2558 sbridge_printk(KERN_ERR, "Some needed devices are missing\n");
2562 sbridge_printk(KERN_ERR,
2563 "Unexpected device %02x:%02x\n", PCI_VENDOR_ID_INTEL,
2568 static int haswell_mci_bind_devs(struct mem_ctl_info *mci,
2569 struct sbridge_dev *sbridge_dev)
2571 struct sbridge_pvt *pvt = mci->pvt_info;
2572 struct pci_dev *pdev;
2573 u8 saw_chan_mask = 0;
2576 /* there's only one device per system; not tied to any bus */
2577 if (pvt->info.pci_vtd == NULL)
2578 /* result will be checked later */
2579 pvt->info.pci_vtd = pci_get_device(PCI_VENDOR_ID_INTEL,
2580 PCI_DEVICE_ID_INTEL_HASWELL_IMC_VTD_MISC,
2583 for (i = 0; i < sbridge_dev->n_devs; i++) {
2584 pdev = sbridge_dev->pdev[i];
2588 switch (pdev->device) {
2589 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_CBO_SAD0:
2590 pvt->pci_sad0 = pdev;
2592 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_CBO_SAD1:
2593 pvt->pci_sad1 = pdev;
2595 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0:
2596 pvt->pci_ha0 = pdev;
2598 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TA:
2601 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_THERMAL:
2602 pvt->pci_ras = pdev;
2604 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD0:
2605 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD1:
2606 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD2:
2607 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD3:
2609 int id = pdev->device - PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD0;
2611 pvt->pci_tad[id] = pdev;
2612 saw_chan_mask |= 1 << id;
2615 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD0:
2616 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD1:
2617 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD2:
2618 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD3:
2620 int id = pdev->device - PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD0 + 4;
2622 pvt->pci_tad[id] = pdev;
2623 saw_chan_mask |= 1 << id;
2626 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO0:
2627 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO1:
2628 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO2:
2629 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO3:
2630 if (!pvt->pci_ddrio)
2631 pvt->pci_ddrio = pdev;
2633 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1:
2634 pvt->pci_ha1 = pdev;
2636 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TA:
2637 pvt->pci_ha1_ta = pdev;
2643 edac_dbg(0, "Associated PCI %02x.%02d.%d with dev = %p\n",
2645 PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn),
2649 /* Check if everything were registered */
2650 if (!pvt->pci_sad0 || !pvt->pci_ha0 || !pvt->pci_sad1 ||
2651 !pvt->pci_ras || !pvt->pci_ta || !pvt->info.pci_vtd)
2654 if (saw_chan_mask != 0x0f && /* -EN */
2655 saw_chan_mask != 0x33 && /* -EP */
2656 saw_chan_mask != 0xff) /* -EX */
2661 sbridge_printk(KERN_ERR, "Some needed devices are missing\n");
2665 static int broadwell_mci_bind_devs(struct mem_ctl_info *mci,
2666 struct sbridge_dev *sbridge_dev)
2668 struct sbridge_pvt *pvt = mci->pvt_info;
2669 struct pci_dev *pdev;
2670 u8 saw_chan_mask = 0;
2673 /* there's only one device per system; not tied to any bus */
2674 if (pvt->info.pci_vtd == NULL)
2675 /* result will be checked later */
2676 pvt->info.pci_vtd = pci_get_device(PCI_VENDOR_ID_INTEL,
2677 PCI_DEVICE_ID_INTEL_BROADWELL_IMC_VTD_MISC,
2680 for (i = 0; i < sbridge_dev->n_devs; i++) {
2681 pdev = sbridge_dev->pdev[i];
2685 switch (pdev->device) {
2686 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_CBO_SAD0:
2687 pvt->pci_sad0 = pdev;
2689 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_CBO_SAD1:
2690 pvt->pci_sad1 = pdev;
2692 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0:
2693 pvt->pci_ha0 = pdev;
2695 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TA:
2698 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_THERMAL:
2699 pvt->pci_ras = pdev;
2701 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD0:
2702 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD1:
2703 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD2:
2704 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD3:
2706 int id = pdev->device - PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD0;
2707 pvt->pci_tad[id] = pdev;
2708 saw_chan_mask |= 1 << id;
2711 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD0:
2712 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD1:
2713 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD2:
2714 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD3:
2716 int id = pdev->device - PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD0 + 4;
2717 pvt->pci_tad[id] = pdev;
2718 saw_chan_mask |= 1 << id;
2721 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_DDRIO0:
2722 pvt->pci_ddrio = pdev;
2724 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1:
2725 pvt->pci_ha1 = pdev;
2727 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TA:
2728 pvt->pci_ha1_ta = pdev;
2734 edac_dbg(0, "Associated PCI %02x.%02d.%d with dev = %p\n",
2736 PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn),
2740 /* Check if everything were registered */
2741 if (!pvt->pci_sad0 || !pvt->pci_ha0 || !pvt->pci_sad1 ||
2742 !pvt->pci_ras || !pvt->pci_ta || !pvt->info.pci_vtd)
2745 if (saw_chan_mask != 0x0f && /* -EN */
2746 saw_chan_mask != 0x33 && /* -EP */
2747 saw_chan_mask != 0xff) /* -EX */
2752 sbridge_printk(KERN_ERR, "Some needed devices are missing\n");
2756 static int knl_mci_bind_devs(struct mem_ctl_info *mci,
2757 struct sbridge_dev *sbridge_dev)
2759 struct sbridge_pvt *pvt = mci->pvt_info;
2760 struct pci_dev *pdev;
2766 for (i = 0; i < sbridge_dev->n_devs; i++) {
2767 pdev = sbridge_dev->pdev[i];
2771 /* Extract PCI device and function. */
2772 dev = (pdev->devfn >> 3) & 0x1f;
2773 func = pdev->devfn & 0x7;
2775 switch (pdev->device) {
2776 case PCI_DEVICE_ID_INTEL_KNL_IMC_MC:
2778 pvt->knl.pci_mc0 = pdev;
2780 pvt->knl.pci_mc1 = pdev;
2782 sbridge_printk(KERN_ERR,
2783 "Memory controller in unexpected place! (dev %d, fn %d)\n",
2789 case PCI_DEVICE_ID_INTEL_KNL_IMC_SAD0:
2790 pvt->pci_sad0 = pdev;
2793 case PCI_DEVICE_ID_INTEL_KNL_IMC_SAD1:
2794 pvt->pci_sad1 = pdev;
2797 case PCI_DEVICE_ID_INTEL_KNL_IMC_CHA:
2798 /* There are one of these per tile, and range from
2801 devidx = ((dev-14)*8)+func;
2803 if (devidx < 0 || devidx >= KNL_MAX_CHAS) {
2804 sbridge_printk(KERN_ERR,
2805 "Caching and Home Agent in unexpected place! (dev %d, fn %d)\n",
2810 WARN_ON(pvt->knl.pci_cha[devidx] != NULL);
2812 pvt->knl.pci_cha[devidx] = pdev;
2815 case PCI_DEVICE_ID_INTEL_KNL_IMC_CHANNEL:
2819 * MC0 channels 0-2 are device 9 function 2-4,
2820 * MC1 channels 3-5 are device 8 function 2-4.
2826 devidx = 3 + (func-2);
2828 if (devidx < 0 || devidx >= KNL_MAX_CHANNELS) {
2829 sbridge_printk(KERN_ERR,
2830 "DRAM Channel Registers in unexpected place! (dev %d, fn %d)\n",
2835 WARN_ON(pvt->knl.pci_channel[devidx] != NULL);
2836 pvt->knl.pci_channel[devidx] = pdev;
2839 case PCI_DEVICE_ID_INTEL_KNL_IMC_TOLHM:
2840 pvt->knl.pci_mc_info = pdev;
2843 case PCI_DEVICE_ID_INTEL_KNL_IMC_TA:
2848 sbridge_printk(KERN_ERR, "Unexpected device %d\n",
2854 if (!pvt->knl.pci_mc0 || !pvt->knl.pci_mc1 ||
2855 !pvt->pci_sad0 || !pvt->pci_sad1 ||
2860 for (i = 0; i < KNL_MAX_CHANNELS; i++) {
2861 if (!pvt->knl.pci_channel[i]) {
2862 sbridge_printk(KERN_ERR, "Missing channel %d\n", i);
2867 for (i = 0; i < KNL_MAX_CHAS; i++) {
2868 if (!pvt->knl.pci_cha[i]) {
2869 sbridge_printk(KERN_ERR, "Missing CHA %d\n", i);
2877 sbridge_printk(KERN_ERR, "Some needed devices are missing\n");
2881 /****************************************************************************
2882 Error check routines
2883 ****************************************************************************/
2886 * While Sandy Bridge has error count registers, SMI BIOS read values from
2887 * and resets the counters. So, they are not reliable for the OS to read
2888 * from them. So, we have no option but to just trust on whatever MCE is
2889 * telling us about the errors.
2891 static void sbridge_mce_output_error(struct mem_ctl_info *mci,
2892 const struct mce *m)
2894 struct mem_ctl_info *new_mci;
2895 struct sbridge_pvt *pvt = mci->pvt_info;
2896 enum hw_event_mc_err_type tp_event;
2897 char *type, *optype, msg[256];
2898 bool ripv = GET_BITFIELD(m->mcgstatus, 0, 0);
2899 bool overflow = GET_BITFIELD(m->status, 62, 62);
2900 bool uncorrected_error = GET_BITFIELD(m->status, 61, 61);
2902 u32 core_err_cnt = GET_BITFIELD(m->status, 38, 52);
2903 u32 mscod = GET_BITFIELD(m->status, 16, 31);
2904 u32 errcode = GET_BITFIELD(m->status, 0, 15);
2905 u32 channel = GET_BITFIELD(m->status, 0, 3);
2906 u32 optypenum = GET_BITFIELD(m->status, 4, 6);
2907 long channel_mask, first_channel;
2908 u8 rank, socket, ha;
2910 char *area_type = NULL;
2912 if (pvt->info.type != SANDY_BRIDGE)
2915 recoverable = GET_BITFIELD(m->status, 56, 56);
2917 if (uncorrected_error) {
2920 tp_event = HW_EVENT_ERR_FATAL;
2923 tp_event = HW_EVENT_ERR_UNCORRECTED;
2927 tp_event = HW_EVENT_ERR_CORRECTED;
2931 * According with Table 15-9 of the Intel Architecture spec vol 3A,
2932 * memory errors should fit in this mask:
2933 * 000f 0000 1mmm cccc (binary)
2935 * f = Correction Report Filtering Bit. If 1, subsequent errors
2939 * If the mask doesn't match, report an error to the parsing logic
2941 if (! ((errcode & 0xef80) == 0x80)) {
2942 optype = "Can't parse: it is not a mem";
2944 switch (optypenum) {
2946 optype = "generic undef request error";
2949 optype = "memory read error";
2952 optype = "memory write error";
2955 optype = "addr/cmd error";
2958 optype = "memory scrubbing error";
2961 optype = "reserved";
2966 /* Only decode errors with an valid address (ADDRV) */
2967 if (!GET_BITFIELD(m->status, 58, 58))
2970 if (pvt->info.type == KNIGHTS_LANDING) {
2971 if (channel == 14) {
2972 edac_dbg(0, "%s%s err_code:%04x:%04x EDRAM bank %d\n",
2973 overflow ? " OVERFLOW" : "",
2974 (uncorrected_error && recoverable)
2975 ? " recoverable" : "",
2982 * Reported channel is in range 0-2, so we can't map it
2983 * back to mc. To figure out mc we check machine check
2984 * bank register that reported this error.
2985 * bank15 means mc0 and bank16 means mc1.
2987 channel = knl_channel_remap(m->bank == 16, channel);
2988 channel_mask = 1 << channel;
2990 snprintf(msg, sizeof(msg),
2991 "%s%s err_code:%04x:%04x channel:%d (DIMM_%c)",
2992 overflow ? " OVERFLOW" : "",
2993 (uncorrected_error && recoverable)
2994 ? " recoverable" : " ",
2995 mscod, errcode, channel, A + channel);
2996 edac_mc_handle_error(tp_event, mci, core_err_cnt,
2997 m->addr >> PAGE_SHIFT, m->addr & ~PAGE_MASK, 0,
3003 rc = get_memory_error_data(mci, m->addr, &socket, &ha,
3004 &channel_mask, &rank, &area_type, msg);
3009 new_mci = get_mci_for_node_id(socket);
3011 strcpy(msg, "Error: socket got corrupted!");
3015 pvt = mci->pvt_info;
3017 first_channel = find_first_bit(&channel_mask, NUM_CHANNELS);
3028 * FIXME: On some memory configurations (mirror, lockstep), the
3029 * Memory Controller can't point the error to a single DIMM. The
3030 * EDAC core should be handling the channel mask, in order to point
3031 * to the group of dimm's where the error may be happening.
3033 if (!pvt->is_lockstep && !pvt->is_mirrored && !pvt->is_close_pg)
3034 channel = first_channel;
3036 snprintf(msg, sizeof(msg),
3037 "%s%s area:%s err_code:%04x:%04x socket:%d ha:%d channel_mask:%ld rank:%d",
3038 overflow ? " OVERFLOW" : "",
3039 (uncorrected_error && recoverable) ? " recoverable" : "",
3046 edac_dbg(0, "%s\n", msg);
3048 /* FIXME: need support for channel mask */
3050 if (channel == CHANNEL_UNSPECIFIED)
3053 /* Call the helper to output message */
3054 edac_mc_handle_error(tp_event, mci, core_err_cnt,
3055 m->addr >> PAGE_SHIFT, m->addr & ~PAGE_MASK, 0,
3056 4*ha+channel, dimm, -1,
3060 edac_mc_handle_error(tp_event, mci, core_err_cnt, 0, 0, 0,
3067 * Check that logging is enabled and that this is the right type
3068 * of error for us to handle.
3070 static int sbridge_mce_check_error(struct notifier_block *nb, unsigned long val,
3073 struct mce *mce = (struct mce *)data;
3074 struct mem_ctl_info *mci;
3075 struct sbridge_pvt *pvt;
3078 if (get_edac_report_status() == EDAC_REPORTING_DISABLED)
3081 mci = get_mci_for_node_id(mce->socketid);
3084 pvt = mci->pvt_info;
3087 * Just let mcelog handle it if the error is
3088 * outside the memory controller. A memory error
3089 * is indicated by bit 7 = 1 and bits = 8-11,13-15 = 0.
3090 * bit 12 has an special meaning.
3092 if ((mce->status & 0xefff) >> 7 != 1)
3095 if (mce->mcgstatus & MCG_STATUS_MCIP)
3100 sbridge_mc_printk(mci, KERN_DEBUG, "HANDLING MCE MEMORY ERROR\n");
3102 sbridge_mc_printk(mci, KERN_DEBUG, "CPU %d: Machine Check %s: %Lx "
3103 "Bank %d: %016Lx\n", mce->extcpu, type,
3104 mce->mcgstatus, mce->bank, mce->status);
3105 sbridge_mc_printk(mci, KERN_DEBUG, "TSC %llx ", mce->tsc);
3106 sbridge_mc_printk(mci, KERN_DEBUG, "ADDR %llx ", mce->addr);
3107 sbridge_mc_printk(mci, KERN_DEBUG, "MISC %llx ", mce->misc);
3109 sbridge_mc_printk(mci, KERN_DEBUG, "PROCESSOR %u:%x TIME %llu SOCKET "
3110 "%u APIC %x\n", mce->cpuvendor, mce->cpuid,
3111 mce->time, mce->socketid, mce->apicid);
3113 sbridge_mce_output_error(mci, mce);
3115 /* Advice mcelog that the error were handled */
3119 static struct notifier_block sbridge_mce_dec = {
3120 .notifier_call = sbridge_mce_check_error,
3123 /****************************************************************************
3124 EDAC register/unregister logic
3125 ****************************************************************************/
3127 static void sbridge_unregister_mci(struct sbridge_dev *sbridge_dev)
3129 struct mem_ctl_info *mci = sbridge_dev->mci;
3130 struct sbridge_pvt *pvt;
3132 if (unlikely(!mci || !mci->pvt_info)) {
3133 edac_dbg(0, "MC: dev = %p\n", &sbridge_dev->pdev[0]->dev);
3135 sbridge_printk(KERN_ERR, "Couldn't find mci handler\n");
3139 pvt = mci->pvt_info;
3141 edac_dbg(0, "MC: mci = %p, dev = %p\n",
3142 mci, &sbridge_dev->pdev[0]->dev);
3144 /* Remove MC sysfs nodes */
3145 edac_mc_del_mc(mci->pdev);
3147 edac_dbg(1, "%s: free mci struct\n", mci->ctl_name);
3148 kfree(mci->ctl_name);
3150 sbridge_dev->mci = NULL;
3153 static int sbridge_register_mci(struct sbridge_dev *sbridge_dev, enum type type)
3155 struct mem_ctl_info *mci;
3156 struct edac_mc_layer layers[2];
3157 struct sbridge_pvt *pvt;
3158 struct pci_dev *pdev = sbridge_dev->pdev[0];
3161 /* Check the number of active and not disabled channels */
3162 rc = check_if_ecc_is_active(sbridge_dev->bus, type);
3163 if (unlikely(rc < 0))
3166 /* allocate a new MC control structure */
3167 layers[0].type = EDAC_MC_LAYER_CHANNEL;
3168 layers[0].size = type == KNIGHTS_LANDING ?
3169 KNL_MAX_CHANNELS : NUM_CHANNELS;
3170 layers[0].is_virt_csrow = false;
3171 layers[1].type = EDAC_MC_LAYER_SLOT;
3172 layers[1].size = type == KNIGHTS_LANDING ? 1 : MAX_DIMMS;
3173 layers[1].is_virt_csrow = true;
3174 mci = edac_mc_alloc(sbridge_dev->mc, ARRAY_SIZE(layers), layers,
3180 edac_dbg(0, "MC: mci = %p, dev = %p\n",
3183 pvt = mci->pvt_info;
3184 memset(pvt, 0, sizeof(*pvt));
3186 /* Associate sbridge_dev and mci for future usage */
3187 pvt->sbridge_dev = sbridge_dev;
3188 sbridge_dev->mci = mci;
3190 mci->mtype_cap = type == KNIGHTS_LANDING ?
3191 MEM_FLAG_DDR4 : MEM_FLAG_DDR3;
3192 mci->edac_ctl_cap = EDAC_FLAG_NONE;
3193 mci->edac_cap = EDAC_FLAG_NONE;
3194 mci->mod_name = "sbridge_edac.c";
3195 mci->mod_ver = SBRIDGE_REVISION;
3196 mci->dev_name = pci_name(pdev);
3197 mci->ctl_page_to_phys = NULL;
3199 pvt->info.type = type;
3202 pvt->info.rankcfgr = IB_RANK_CFG_A;
3203 pvt->info.get_tolm = ibridge_get_tolm;
3204 pvt->info.get_tohm = ibridge_get_tohm;
3205 pvt->info.dram_rule = ibridge_dram_rule;
3206 pvt->info.get_memory_type = get_memory_type;
3207 pvt->info.get_node_id = get_node_id;
3208 pvt->info.rir_limit = rir_limit;
3209 pvt->info.sad_limit = sad_limit;
3210 pvt->info.interleave_mode = interleave_mode;
3211 pvt->info.dram_attr = dram_attr;
3212 pvt->info.max_sad = ARRAY_SIZE(ibridge_dram_rule);
3213 pvt->info.interleave_list = ibridge_interleave_list;
3214 pvt->info.max_interleave = ARRAY_SIZE(ibridge_interleave_list);
3215 pvt->info.interleave_pkg = ibridge_interleave_pkg;
3216 pvt->info.get_width = ibridge_get_width;
3217 mci->ctl_name = kasprintf(GFP_KERNEL, "Ivy Bridge Socket#%d", mci->mc_idx);
3219 /* Store pci devices at mci for faster access */
3220 rc = ibridge_mci_bind_devs(mci, sbridge_dev);
3221 if (unlikely(rc < 0))
3225 pvt->info.rankcfgr = SB_RANK_CFG_A;
3226 pvt->info.get_tolm = sbridge_get_tolm;
3227 pvt->info.get_tohm = sbridge_get_tohm;
3228 pvt->info.dram_rule = sbridge_dram_rule;
3229 pvt->info.get_memory_type = get_memory_type;
3230 pvt->info.get_node_id = get_node_id;
3231 pvt->info.rir_limit = rir_limit;
3232 pvt->info.sad_limit = sad_limit;
3233 pvt->info.interleave_mode = interleave_mode;
3234 pvt->info.dram_attr = dram_attr;
3235 pvt->info.max_sad = ARRAY_SIZE(sbridge_dram_rule);
3236 pvt->info.interleave_list = sbridge_interleave_list;
3237 pvt->info.max_interleave = ARRAY_SIZE(sbridge_interleave_list);
3238 pvt->info.interleave_pkg = sbridge_interleave_pkg;
3239 pvt->info.get_width = sbridge_get_width;
3240 mci->ctl_name = kasprintf(GFP_KERNEL, "Sandy Bridge Socket#%d", mci->mc_idx);
3242 /* Store pci devices at mci for faster access */
3243 rc = sbridge_mci_bind_devs(mci, sbridge_dev);
3244 if (unlikely(rc < 0))
3248 /* rankcfgr isn't used */
3249 pvt->info.get_tolm = haswell_get_tolm;
3250 pvt->info.get_tohm = haswell_get_tohm;
3251 pvt->info.dram_rule = ibridge_dram_rule;
3252 pvt->info.get_memory_type = haswell_get_memory_type;
3253 pvt->info.get_node_id = haswell_get_node_id;
3254 pvt->info.rir_limit = haswell_rir_limit;
3255 pvt->info.sad_limit = sad_limit;
3256 pvt->info.interleave_mode = interleave_mode;
3257 pvt->info.dram_attr = dram_attr;
3258 pvt->info.max_sad = ARRAY_SIZE(ibridge_dram_rule);
3259 pvt->info.interleave_list = ibridge_interleave_list;
3260 pvt->info.max_interleave = ARRAY_SIZE(ibridge_interleave_list);
3261 pvt->info.interleave_pkg = ibridge_interleave_pkg;
3262 pvt->info.get_width = ibridge_get_width;
3263 mci->ctl_name = kasprintf(GFP_KERNEL, "Haswell Socket#%d", mci->mc_idx);
3265 /* Store pci devices at mci for faster access */
3266 rc = haswell_mci_bind_devs(mci, sbridge_dev);
3267 if (unlikely(rc < 0))
3271 /* rankcfgr isn't used */
3272 pvt->info.get_tolm = haswell_get_tolm;
3273 pvt->info.get_tohm = haswell_get_tohm;
3274 pvt->info.dram_rule = ibridge_dram_rule;
3275 pvt->info.get_memory_type = haswell_get_memory_type;
3276 pvt->info.get_node_id = haswell_get_node_id;
3277 pvt->info.rir_limit = haswell_rir_limit;
3278 pvt->info.sad_limit = sad_limit;
3279 pvt->info.interleave_mode = interleave_mode;
3280 pvt->info.dram_attr = dram_attr;
3281 pvt->info.max_sad = ARRAY_SIZE(ibridge_dram_rule);
3282 pvt->info.interleave_list = ibridge_interleave_list;
3283 pvt->info.max_interleave = ARRAY_SIZE(ibridge_interleave_list);
3284 pvt->info.interleave_pkg = ibridge_interleave_pkg;
3285 pvt->info.get_width = broadwell_get_width;
3286 mci->ctl_name = kasprintf(GFP_KERNEL, "Broadwell Socket#%d", mci->mc_idx);
3288 /* Store pci devices at mci for faster access */
3289 rc = broadwell_mci_bind_devs(mci, sbridge_dev);
3290 if (unlikely(rc < 0))
3293 case KNIGHTS_LANDING:
3294 /* pvt->info.rankcfgr == ??? */
3295 pvt->info.get_tolm = knl_get_tolm;
3296 pvt->info.get_tohm = knl_get_tohm;
3297 pvt->info.dram_rule = knl_dram_rule;
3298 pvt->info.get_memory_type = knl_get_memory_type;
3299 pvt->info.get_node_id = knl_get_node_id;
3300 pvt->info.rir_limit = NULL;
3301 pvt->info.sad_limit = knl_sad_limit;
3302 pvt->info.interleave_mode = knl_interleave_mode;
3303 pvt->info.dram_attr = dram_attr_knl;
3304 pvt->info.max_sad = ARRAY_SIZE(knl_dram_rule);
3305 pvt->info.interleave_list = knl_interleave_list;
3306 pvt->info.max_interleave = ARRAY_SIZE(knl_interleave_list);
3307 pvt->info.interleave_pkg = ibridge_interleave_pkg;
3308 pvt->info.get_width = knl_get_width;
3309 mci->ctl_name = kasprintf(GFP_KERNEL,
3310 "Knights Landing Socket#%d", mci->mc_idx);
3312 rc = knl_mci_bind_devs(mci, sbridge_dev);
3313 if (unlikely(rc < 0))
3318 /* Get dimm basic config and the memory layout */
3319 get_dimm_config(mci);
3320 get_memory_layout(mci);
3322 /* record ptr to the generic device */
3323 mci->pdev = &pdev->dev;
3325 /* add this new MC control structure to EDAC's list of MCs */
3326 if (unlikely(edac_mc_add_mc(mci))) {
3327 edac_dbg(0, "MC: failed edac_mc_add_mc()\n");
3335 kfree(mci->ctl_name);
3337 sbridge_dev->mci = NULL;
3341 #define ICPU(model, table) \
3342 { X86_VENDOR_INTEL, 6, model, 0, (unsigned long)&table }
3344 static const struct x86_cpu_id sbridge_cpuids[] = {
3345 ICPU(INTEL_FAM6_SANDYBRIDGE_X, pci_dev_descr_sbridge_table),
3346 ICPU(INTEL_FAM6_IVYBRIDGE_X, pci_dev_descr_ibridge_table),
3347 ICPU(INTEL_FAM6_HASWELL_X, pci_dev_descr_haswell_table),
3348 ICPU(INTEL_FAM6_BROADWELL_X, pci_dev_descr_broadwell_table),
3349 ICPU(INTEL_FAM6_BROADWELL_XEON_D, pci_dev_descr_broadwell_table),
3350 ICPU(INTEL_FAM6_XEON_PHI_KNL, pci_dev_descr_knl_table),
3351 ICPU(INTEL_FAM6_XEON_PHI_KNM, pci_dev_descr_knl_table),
3354 MODULE_DEVICE_TABLE(x86cpu, sbridge_cpuids);
3357 * sbridge_probe Get all devices and register memory controllers
3360 * 0 for FOUND a device
3361 * < 0 for error code
3364 static int sbridge_probe(const struct x86_cpu_id *id)
3368 struct sbridge_dev *sbridge_dev;
3369 struct pci_id_table *ptable = (struct pci_id_table *)id->driver_data;
3371 /* get the pci devices we want to reserve for our use */
3372 rc = sbridge_get_all_devices(&num_mc, ptable);
3374 if (unlikely(rc < 0)) {
3375 edac_dbg(0, "couldn't get all devices\n");
3381 list_for_each_entry(sbridge_dev, &sbridge_edac_list, list) {
3382 edac_dbg(0, "Registering MC#%d (%d of %d)\n",
3383 mc, mc + 1, num_mc);
3385 sbridge_dev->mc = mc++;
3386 rc = sbridge_register_mci(sbridge_dev, ptable->type);
3387 if (unlikely(rc < 0))
3391 sbridge_printk(KERN_INFO, "%s\n", SBRIDGE_REVISION);
3396 list_for_each_entry(sbridge_dev, &sbridge_edac_list, list)
3397 sbridge_unregister_mci(sbridge_dev);
3399 sbridge_put_all_devices();
3405 * sbridge_remove cleanup
3408 static void sbridge_remove(void)
3410 struct sbridge_dev *sbridge_dev;
3414 list_for_each_entry(sbridge_dev, &sbridge_edac_list, list)
3415 sbridge_unregister_mci(sbridge_dev);
3417 /* Release PCI resources */
3418 sbridge_put_all_devices();
3422 * sbridge_init Module entry function
3423 * Try to initialize this module for its devices
3425 static int __init sbridge_init(void)
3427 const struct x86_cpu_id *id;
3432 id = x86_match_cpu(sbridge_cpuids);
3436 /* Ensure that the OPSTATE is set correctly for POLL or NMI */
3439 rc = sbridge_probe(id);
3442 mce_register_decode_chain(&sbridge_mce_dec);
3443 if (get_edac_report_status() == EDAC_REPORTING_DISABLED)
3444 sbridge_printk(KERN_WARNING, "Loading driver, error reporting disabled.\n");
3448 sbridge_printk(KERN_ERR, "Failed to register device with error %d.\n",
3455 * sbridge_exit() Module exit function
3456 * Unregister the driver
3458 static void __exit sbridge_exit(void)
3462 mce_unregister_decode_chain(&sbridge_mce_dec);
3465 module_init(sbridge_init);
3466 module_exit(sbridge_exit);
3468 module_param(edac_op_state, int, 0444);
3469 MODULE_PARM_DESC(edac_op_state, "EDAC Error Reporting state: 0=Poll,1=NMI");
3471 MODULE_LICENSE("GPL");
3472 MODULE_AUTHOR("Mauro Carvalho Chehab");
3473 MODULE_AUTHOR("Red Hat Inc. (http://www.redhat.com)");
3474 MODULE_DESCRIPTION("MC Driver for Intel Sandy Bridge and Ivy Bridge memory controllers - "