Linux 4.9.121
[linux/fpc-iii.git] / drivers / edac / sb_edac.c
blob3c47e6361d8127faeacce4d8b88df28d849a1eb2
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>
27 #include <asm/mce.h>
29 #include "edac_core.h"
31 /* Static vars */
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"
41 * Debug macros
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)
81 switch (attr) {
82 case 0:
83 return "DRAM";
84 case 1:
85 return "MMCFG";
86 case 2:
87 return "NXM";
88 default:
89 return "unknown";
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 {
114 unsigned char start;
115 unsigned char end;
118 static const struct interleave_pkg sbridge_interleave_pkg[] = {
119 { 0, 2 },
120 { 3, 5 },
121 { 8, 10 },
122 { 11, 13 },
123 { 16, 18 },
124 { 19, 21 },
125 { 24, 26 },
126 { 27, 29 },
129 static const struct interleave_pkg ibridge_interleave_pkg[] = {
130 { 0, 3 },
131 { 4, 7 },
132 { 8, 11 },
133 { 12, 15 },
134 { 16, 19 },
135 { 20, 23 },
136 { 24, 27 },
137 { 28, 31 },
140 static inline int sad_pkg(const struct interleave_pkg *table, u32 reg,
141 int interleave)
143 return GET_BITFIELD(reg, table[interleave].start,
144 table[interleave].end);
147 /* Devices 12 Function 7 */
149 #define TOLM 0x80
150 #define TOHM 0x84
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 */
190 #define MCMTR 0x7c
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[] = {
205 0x80, 0x84, 0x88,
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(type, reg) (((type) == BROADWELL) ? \
243 GET_BITFIELD(reg, 20, 23) : GET_BITFIELD(reg, 16, 19))
245 #define RIR_OFFSET(type, reg) (((type) == HASWELL || (type) == BROADWELL) ? \
246 GET_BITFIELD(reg, 2, 15) : GET_BITFIELD(reg, 2, 14))
248 /* Device 16, functions 2-7 */
251 * FIXME: Implement the error count reads directly
254 static const u32 correrrcnt[] = {
255 0x104, 0x108, 0x10c, 0x110,
258 #define RANK_ODD_OV(reg) GET_BITFIELD(reg, 31, 31)
259 #define RANK_ODD_ERR_CNT(reg) GET_BITFIELD(reg, 16, 30)
260 #define RANK_EVEN_OV(reg) GET_BITFIELD(reg, 15, 15)
261 #define RANK_EVEN_ERR_CNT(reg) GET_BITFIELD(reg, 0, 14)
263 static const u32 correrrthrsld[] = {
264 0x11c, 0x120, 0x124, 0x128,
267 #define RANK_ODD_ERR_THRSLD(reg) GET_BITFIELD(reg, 16, 30)
268 #define RANK_EVEN_ERR_THRSLD(reg) GET_BITFIELD(reg, 0, 14)
271 /* Device 17, function 0 */
273 #define SB_RANK_CFG_A 0x0328
275 #define IB_RANK_CFG_A 0x0320
278 * sbridge structs
281 #define NUM_CHANNELS 8 /* 2MC per socket, four chan per MC */
282 #define MAX_DIMMS 3 /* Max DIMMS per channel */
283 #define KNL_MAX_CHAS 38 /* KNL max num. of Cache Home Agents */
284 #define KNL_MAX_CHANNELS 6 /* KNL max num. of PCI channels */
285 #define KNL_MAX_EDCS 8 /* Embedded DRAM controllers */
286 #define CHANNEL_UNSPECIFIED 0xf /* Intel IA32 SDM 15-14 */
288 enum type {
289 SANDY_BRIDGE,
290 IVY_BRIDGE,
291 HASWELL,
292 BROADWELL,
293 KNIGHTS_LANDING,
296 struct sbridge_pvt;
297 struct sbridge_info {
298 enum type type;
299 u32 mcmtr;
300 u32 rankcfgr;
301 u64 (*get_tolm)(struct sbridge_pvt *pvt);
302 u64 (*get_tohm)(struct sbridge_pvt *pvt);
303 u64 (*rir_limit)(u32 reg);
304 u64 (*sad_limit)(u32 reg);
305 u32 (*interleave_mode)(u32 reg);
306 char* (*show_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;
311 u8 max_sad;
312 u8 max_interleave;
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 {
320 u32 ranks;
321 u32 dimms;
324 struct pci_id_descr {
325 int dev_id;
326 int optional;
329 struct pci_id_table {
330 const struct pci_id_descr *descr;
331 int n_devs;
332 enum type type;
335 struct sbridge_dev {
336 struct list_head list;
337 u8 bus, mc;
338 u8 node_id, source_id;
339 struct pci_dev **pdev;
340 int n_devs;
341 struct mem_ctl_info *mci;
344 struct knl_pvt {
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 */
354 struct sbridge_pvt {
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;
369 bool is_chan_hash;
371 /* Memory description */
372 u64 tolm, tohm;
373 struct knl_pvt knl;
376 #define PCI_DESCR(device_id, opt) \
377 .dev_id = (device_id), \
378 .optional = opt
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) { \
402 .descr = A, \
403 .n_devs = ARRAY_SIZE(A), \
404 .type = T \
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:
413 * 1HA:
414 * 0x0eb8 (17.0) is DDRIO0
415 * 2HA:
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
421 /* pci ids */
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) },
461 #if 0
462 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TA, 1) },
463 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_RAS, 1) },
464 #endif
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 */
480 /* EN processor:
481 * - 1 IMC
482 * - 3 DDR3 channels, 2 DPC per channel
483 * EP processor:
484 * - 1 or 2 IMC
485 * - 4 DDR4 channels, 3 DPC per channel
486 * EP 4S processor:
487 * - 2 IMC
488 * - 4 DDR4 channels, 3 DPC per channel
489 * EX processor:
490 * - 2 IMC
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),
593 {0,}
597 * Broadwell support
599 * DE processor:
600 * - 1 IMC
601 * - 2 DDR3 channels, 2 DPC per channel
602 * EP processor:
603 * - 1 or 2 IMC
604 * - 4 DDR4 channels, 3 DPC per channel
605 * EP 4S processor:
606 * - 2 IMC
607 * - 4 DDR4 channels, 3 DPC per channel
608 * EX processor:
609 * - 2 IMC
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));
672 int max = 4;
674 if (type == HASWELL || type == BROADWELL || type == KNIGHTS_LANDING)
675 max = 8;
677 if (ranks > max) {
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);
680 return -EINVAL;
683 return ranks;
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);
693 return -EINVAL;
696 return 1 << rows;
699 static inline int numcol(u32 mtr)
701 int cols = (COL_WIDTH_BITS(mtr) + 10);
703 if (cols > 12) {
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);
706 return -EINVAL;
709 return 1 << cols;
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.
720 if (multi_bus) {
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)
727 return sbridge_dev;
730 return NULL;
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);
739 if (!sbridge_dev)
740 return NULL;
742 sbridge_dev->pdev = kzalloc(sizeof(*sbridge_dev->pdev) * table->n_devs,
743 GFP_KERNEL);
744 if (!sbridge_dev->pdev) {
745 kfree(sbridge_dev);
746 return NULL;
749 sbridge_dev->bus = bus;
750 sbridge_dev->n_devs = table->n_devs;
751 list_add_tail(&sbridge_dev->list, &sbridge_edac_list);
753 return sbridge_dev;
756 static void free_sbridge_dev(struct sbridge_dev *sbridge_dev)
758 list_del(&sbridge_dev->list);
759 kfree(sbridge_dev->pdev);
760 kfree(sbridge_dev);
763 static u64 sbridge_get_tolm(struct sbridge_pvt *pvt)
765 u32 reg;
767 /* Address range is 32:28 */
768 pci_read_config_dword(pvt->pci_sad1, TOLM, &reg);
769 return GET_TOLM(reg);
772 static u64 sbridge_get_tohm(struct sbridge_pvt *pvt)
774 u32 reg;
776 pci_read_config_dword(pvt->pci_sad1, TOHM, &reg);
777 return GET_TOHM(reg);
780 static u64 ibridge_get_tolm(struct sbridge_pvt *pvt)
782 u32 reg;
784 pci_read_config_dword(pvt->pci_br1, TOLM, &reg);
786 return GET_TOLM(reg);
789 static u64 ibridge_get_tohm(struct sbridge_pvt *pvt)
791 u32 reg;
793 pci_read_config_dword(pvt->pci_br1, TOHM, &reg);
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 char *show_interleave_mode(u32 reg)
815 return interleave_mode(reg) ? "8:6" : "[8:6]XOR[18:16]";
818 static u32 dram_attr(u32 reg)
820 return GET_BITFIELD(reg, 2, 3);
823 static u64 knl_sad_limit(u32 reg)
825 return (GET_BITFIELD(reg, 7, 26) << 26) | 0x3ffffff;
828 static u32 knl_interleave_mode(u32 reg)
830 return GET_BITFIELD(reg, 1, 2);
833 static char *knl_show_interleave_mode(u32 reg)
835 char *s;
837 switch (knl_interleave_mode(reg)) {
838 case 0:
839 s = "use address bits [8:6]";
840 break;
841 case 1:
842 s = "use address bits [10:8]";
843 break;
844 case 2:
845 s = "use address bits [14:12]";
846 break;
847 case 3:
848 s = "use address bits [32:30]";
849 break;
850 default:
851 WARN_ON(1);
852 break;
855 return s;
858 static u32 dram_attr_knl(u32 reg)
860 return GET_BITFIELD(reg, 3, 4);
864 static enum mem_type get_memory_type(struct sbridge_pvt *pvt)
866 u32 reg;
867 enum mem_type mtype;
869 if (pvt->pci_ddrio) {
870 pci_read_config_dword(pvt->pci_ddrio, pvt->info.rankcfgr,
871 &reg);
872 if (GET_BITFIELD(reg, 11, 11))
873 /* FIXME: Can also be LRDIMM */
874 mtype = MEM_RDDR3;
875 else
876 mtype = MEM_DDR3;
877 } else
878 mtype = MEM_UNKNOWN;
880 return mtype;
883 static enum mem_type haswell_get_memory_type(struct sbridge_pvt *pvt)
885 u32 reg;
886 bool registered = false;
887 enum mem_type mtype = MEM_UNKNOWN;
889 if (!pvt->pci_ddrio)
890 goto out;
892 pci_read_config_dword(pvt->pci_ddrio,
893 HASWELL_DDRCRCLKCONTROLS, &reg);
894 /* Is_Rdimm */
895 if (GET_BITFIELD(reg, 16, 16))
896 registered = true;
898 pci_read_config_dword(pvt->pci_ta, MCMTR, &reg);
899 if (GET_BITFIELD(reg, 14, 14)) {
900 if (registered)
901 mtype = MEM_RDDR4;
902 else
903 mtype = MEM_DDR4;
904 } else {
905 if (registered)
906 mtype = MEM_RDDR3;
907 else
908 mtype = MEM_DDR3;
911 out:
912 return mtype;
915 static enum dev_type knl_get_width(struct sbridge_pvt *pvt, u32 mtr)
917 /* for KNL value is fixed */
918 return DEV_X16;
921 static enum dev_type sbridge_get_width(struct sbridge_pvt *pvt, u32 mtr)
923 /* there's no way to figure out */
924 return DEV_UNKNOWN;
927 static enum dev_type __ibridge_get_width(u32 mtr)
929 enum dev_type type;
931 switch (mtr) {
932 case 3:
933 type = DEV_UNKNOWN;
934 break;
935 case 2:
936 type = DEV_X16;
937 break;
938 case 1:
939 type = DEV_X8;
940 break;
941 case 0:
942 type = DEV_X4;
943 break;
946 return type;
949 static enum dev_type ibridge_get_width(struct sbridge_pvt *pvt, u32 mtr)
952 * ddr3_width on the documentation but also valid for DDR4 on
953 * Haswell
955 return __ibridge_get_width(GET_BITFIELD(mtr, 7, 8));
958 static enum dev_type broadwell_get_width(struct sbridge_pvt *pvt, u32 mtr)
960 /* ddr3_width on the documentation but also valid for DDR4 */
961 return __ibridge_get_width(GET_BITFIELD(mtr, 8, 9));
964 static enum mem_type knl_get_memory_type(struct sbridge_pvt *pvt)
966 /* DDR4 RDIMMS and LRDIMMS are supported */
967 return MEM_RDDR4;
970 static u8 get_node_id(struct sbridge_pvt *pvt)
972 u32 reg;
973 pci_read_config_dword(pvt->pci_br0, SAD_CONTROL, &reg);
974 return GET_BITFIELD(reg, 0, 2);
977 static u8 haswell_get_node_id(struct sbridge_pvt *pvt)
979 u32 reg;
981 pci_read_config_dword(pvt->pci_sad1, SAD_CONTROL, &reg);
982 return GET_BITFIELD(reg, 0, 3);
985 static u8 knl_get_node_id(struct sbridge_pvt *pvt)
987 u32 reg;
989 pci_read_config_dword(pvt->pci_sad1, SAD_CONTROL, &reg);
990 return GET_BITFIELD(reg, 0, 2);
994 static u64 haswell_get_tolm(struct sbridge_pvt *pvt)
996 u32 reg;
998 pci_read_config_dword(pvt->info.pci_vtd, HASWELL_TOLM, &reg);
999 return (GET_BITFIELD(reg, 26, 31) << 26) | 0x3ffffff;
1002 static u64 haswell_get_tohm(struct sbridge_pvt *pvt)
1004 u64 rc;
1005 u32 reg;
1007 pci_read_config_dword(pvt->info.pci_vtd, HASWELL_TOHM_0, &reg);
1008 rc = GET_BITFIELD(reg, 26, 31);
1009 pci_read_config_dword(pvt->info.pci_vtd, HASWELL_TOHM_1, &reg);
1010 rc = ((reg << 6) | rc) << 26;
1012 return rc | 0x1ffffff;
1015 static u64 knl_get_tolm(struct sbridge_pvt *pvt)
1017 u32 reg;
1019 pci_read_config_dword(pvt->knl.pci_mc_info, KNL_TOLM, &reg);
1020 return (GET_BITFIELD(reg, 26, 31) << 26) | 0x3ffffff;
1023 static u64 knl_get_tohm(struct sbridge_pvt *pvt)
1025 u64 rc;
1026 u32 reg_lo, reg_hi;
1028 pci_read_config_dword(pvt->knl.pci_mc_info, KNL_TOHM_0, &reg_lo);
1029 pci_read_config_dword(pvt->knl.pci_mc_info, KNL_TOHM_1, &reg_hi);
1030 rc = ((u64)reg_hi << 32) | reg_lo;
1031 return rc | 0x3ffffff;
1035 static u64 haswell_rir_limit(u32 reg)
1037 return (((u64)GET_BITFIELD(reg, 1, 11) + 1) << 29) - 1;
1040 static inline u8 sad_pkg_socket(u8 pkg)
1042 /* on Ivy Bridge, nodeID is SASS, where A is HA and S is node id */
1043 return ((pkg >> 3) << 2) | (pkg & 0x3);
1046 static inline u8 sad_pkg_ha(u8 pkg)
1048 return (pkg >> 2) & 0x1;
1051 static int haswell_chan_hash(int idx, u64 addr)
1053 int i;
1056 * XOR even bits from 12:26 to bit0 of idx,
1057 * odd bits from 13:27 to bit1
1059 for (i = 12; i < 28; i += 2)
1060 idx ^= (addr >> i) & 3;
1062 return idx;
1065 /****************************************************************************
1066 Memory check routines
1067 ****************************************************************************/
1068 static struct pci_dev *get_pdev_same_bus(u8 bus, u32 id)
1070 struct pci_dev *pdev = NULL;
1072 do {
1073 pdev = pci_get_device(PCI_VENDOR_ID_INTEL, id, pdev);
1074 if (pdev && pdev->bus->number == bus)
1075 break;
1076 } while (pdev);
1078 return pdev;
1082 * check_if_ecc_is_active() - Checks if ECC is active
1083 * @bus: Device bus
1084 * @type: Memory controller type
1085 * returns: 0 in case ECC is active, -ENODEV if it can't be determined or
1086 * disabled
1088 static int check_if_ecc_is_active(const u8 bus, enum type type)
1090 struct pci_dev *pdev = NULL;
1091 u32 mcmtr, id;
1093 switch (type) {
1094 case IVY_BRIDGE:
1095 id = PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TA;
1096 break;
1097 case HASWELL:
1098 id = PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TA;
1099 break;
1100 case SANDY_BRIDGE:
1101 id = PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TA;
1102 break;
1103 case BROADWELL:
1104 id = PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TA;
1105 break;
1106 case KNIGHTS_LANDING:
1108 * KNL doesn't group things by bus the same way
1109 * SB/IB/Haswell does.
1111 id = PCI_DEVICE_ID_INTEL_KNL_IMC_TA;
1112 break;
1113 default:
1114 return -ENODEV;
1117 if (type != KNIGHTS_LANDING)
1118 pdev = get_pdev_same_bus(bus, id);
1119 else
1120 pdev = pci_get_device(PCI_VENDOR_ID_INTEL, id, 0);
1122 if (!pdev) {
1123 sbridge_printk(KERN_ERR, "Couldn't find PCI device "
1124 "%04x:%04x! on bus %02d\n",
1125 PCI_VENDOR_ID_INTEL, id, bus);
1126 return -ENODEV;
1129 pci_read_config_dword(pdev,
1130 type == KNIGHTS_LANDING ? KNL_MCMTR : MCMTR, &mcmtr);
1131 if (!IS_ECC_ENABLED(mcmtr)) {
1132 sbridge_printk(KERN_ERR, "ECC is disabled. Aborting\n");
1133 return -ENODEV;
1135 return 0;
1138 /* Low bits of TAD limit, and some metadata. */
1139 static const u32 knl_tad_dram_limit_lo[] = {
1140 0x400, 0x500, 0x600, 0x700,
1141 0x800, 0x900, 0xa00, 0xb00,
1144 /* Low bits of TAD offset. */
1145 static const u32 knl_tad_dram_offset_lo[] = {
1146 0x404, 0x504, 0x604, 0x704,
1147 0x804, 0x904, 0xa04, 0xb04,
1150 /* High 16 bits of TAD limit and offset. */
1151 static const u32 knl_tad_dram_hi[] = {
1152 0x408, 0x508, 0x608, 0x708,
1153 0x808, 0x908, 0xa08, 0xb08,
1156 /* Number of ways a tad entry is interleaved. */
1157 static const u32 knl_tad_ways[] = {
1158 8, 6, 4, 3, 2, 1,
1162 * Retrieve the n'th Target Address Decode table entry
1163 * from the memory controller's TAD table.
1165 * @pvt: driver private data
1166 * @entry: which entry you want to retrieve
1167 * @mc: which memory controller (0 or 1)
1168 * @offset: output tad range offset
1169 * @limit: output address of first byte above tad range
1170 * @ways: output number of interleave ways
1172 * The offset value has curious semantics. It's a sort of running total
1173 * of the sizes of all the memory regions that aren't mapped in this
1174 * tad table.
1176 static int knl_get_tad(const struct sbridge_pvt *pvt,
1177 const int entry,
1178 const int mc,
1179 u64 *offset,
1180 u64 *limit,
1181 int *ways)
1183 u32 reg_limit_lo, reg_offset_lo, reg_hi;
1184 struct pci_dev *pci_mc;
1185 int way_id;
1187 switch (mc) {
1188 case 0:
1189 pci_mc = pvt->knl.pci_mc0;
1190 break;
1191 case 1:
1192 pci_mc = pvt->knl.pci_mc1;
1193 break;
1194 default:
1195 WARN_ON(1);
1196 return -EINVAL;
1199 pci_read_config_dword(pci_mc,
1200 knl_tad_dram_limit_lo[entry], &reg_limit_lo);
1201 pci_read_config_dword(pci_mc,
1202 knl_tad_dram_offset_lo[entry], &reg_offset_lo);
1203 pci_read_config_dword(pci_mc,
1204 knl_tad_dram_hi[entry], &reg_hi);
1206 /* Is this TAD entry enabled? */
1207 if (!GET_BITFIELD(reg_limit_lo, 0, 0))
1208 return -ENODEV;
1210 way_id = GET_BITFIELD(reg_limit_lo, 3, 5);
1212 if (way_id < ARRAY_SIZE(knl_tad_ways)) {
1213 *ways = knl_tad_ways[way_id];
1214 } else {
1215 *ways = 0;
1216 sbridge_printk(KERN_ERR,
1217 "Unexpected value %d in mc_tad_limit_lo wayness field\n",
1218 way_id);
1219 return -ENODEV;
1223 * The least significant 6 bits of base and limit are truncated.
1224 * For limit, we fill the missing bits with 1s.
1226 *offset = ((u64) GET_BITFIELD(reg_offset_lo, 6, 31) << 6) |
1227 ((u64) GET_BITFIELD(reg_hi, 0, 15) << 32);
1228 *limit = ((u64) GET_BITFIELD(reg_limit_lo, 6, 31) << 6) | 63 |
1229 ((u64) GET_BITFIELD(reg_hi, 16, 31) << 32);
1231 return 0;
1234 /* Determine which memory controller is responsible for a given channel. */
1235 static int knl_channel_mc(int channel)
1237 WARN_ON(channel < 0 || channel >= 6);
1239 return channel < 3 ? 1 : 0;
1243 * Get the Nth entry from EDC_ROUTE_TABLE register.
1244 * (This is the per-tile mapping of logical interleave targets to
1245 * physical EDC modules.)
1247 * entry 0: 0:2
1248 * 1: 3:5
1249 * 2: 6:8
1250 * 3: 9:11
1251 * 4: 12:14
1252 * 5: 15:17
1253 * 6: 18:20
1254 * 7: 21:23
1255 * reserved: 24:31
1257 static u32 knl_get_edc_route(int entry, u32 reg)
1259 WARN_ON(entry >= KNL_MAX_EDCS);
1260 return GET_BITFIELD(reg, entry*3, (entry*3)+2);
1264 * Get the Nth entry from MC_ROUTE_TABLE register.
1265 * (This is the per-tile mapping of logical interleave targets to
1266 * physical DRAM channels modules.)
1268 * entry 0: mc 0:2 channel 18:19
1269 * 1: mc 3:5 channel 20:21
1270 * 2: mc 6:8 channel 22:23
1271 * 3: mc 9:11 channel 24:25
1272 * 4: mc 12:14 channel 26:27
1273 * 5: mc 15:17 channel 28:29
1274 * reserved: 30:31
1276 * Though we have 3 bits to identify the MC, we should only see
1277 * the values 0 or 1.
1280 static u32 knl_get_mc_route(int entry, u32 reg)
1282 int mc, chan;
1284 WARN_ON(entry >= KNL_MAX_CHANNELS);
1286 mc = GET_BITFIELD(reg, entry*3, (entry*3)+2);
1287 chan = GET_BITFIELD(reg, (entry*2) + 18, (entry*2) + 18 + 1);
1289 return knl_channel_remap(mc, chan);
1293 * Render the EDC_ROUTE register in human-readable form.
1294 * Output string s should be at least KNL_MAX_EDCS*2 bytes.
1296 static void knl_show_edc_route(u32 reg, char *s)
1298 int i;
1300 for (i = 0; i < KNL_MAX_EDCS; i++) {
1301 s[i*2] = knl_get_edc_route(i, reg) + '0';
1302 s[i*2+1] = '-';
1305 s[KNL_MAX_EDCS*2 - 1] = '\0';
1309 * Render the MC_ROUTE register in human-readable form.
1310 * Output string s should be at least KNL_MAX_CHANNELS*2 bytes.
1312 static void knl_show_mc_route(u32 reg, char *s)
1314 int i;
1316 for (i = 0; i < KNL_MAX_CHANNELS; i++) {
1317 s[i*2] = knl_get_mc_route(i, reg) + '0';
1318 s[i*2+1] = '-';
1321 s[KNL_MAX_CHANNELS*2 - 1] = '\0';
1324 #define KNL_EDC_ROUTE 0xb8
1325 #define KNL_MC_ROUTE 0xb4
1327 /* Is this dram rule backed by regular DRAM in flat mode? */
1328 #define KNL_EDRAM(reg) GET_BITFIELD(reg, 29, 29)
1330 /* Is this dram rule cached? */
1331 #define KNL_CACHEABLE(reg) GET_BITFIELD(reg, 28, 28)
1333 /* Is this rule backed by edc ? */
1334 #define KNL_EDRAM_ONLY(reg) GET_BITFIELD(reg, 29, 29)
1336 /* Is this rule backed by DRAM, cacheable in EDRAM? */
1337 #define KNL_CACHEABLE(reg) GET_BITFIELD(reg, 28, 28)
1339 /* Is this rule mod3? */
1340 #define KNL_MOD3(reg) GET_BITFIELD(reg, 27, 27)
1343 * Figure out how big our RAM modules are.
1345 * The DIMMMTR register in KNL doesn't tell us the size of the DIMMs, so we
1346 * have to figure this out from the SAD rules, interleave lists, route tables,
1347 * and TAD rules.
1349 * SAD rules can have holes in them (e.g. the 3G-4G hole), so we have to
1350 * inspect the TAD rules to figure out how large the SAD regions really are.
1352 * When we know the real size of a SAD region and how many ways it's
1353 * interleaved, we know the individual contribution of each channel to
1354 * TAD is size/ways.
1356 * Finally, we have to check whether each channel participates in each SAD
1357 * region.
1359 * Fortunately, KNL only supports one DIMM per channel, so once we know how
1360 * much memory the channel uses, we know the DIMM is at least that large.
1361 * (The BIOS might possibly choose not to map all available memory, in which
1362 * case we will underreport the size of the DIMM.)
1364 * In theory, we could try to determine the EDC sizes as well, but that would
1365 * only work in flat mode, not in cache mode.
1367 * @mc_sizes: Output sizes of channels (must have space for KNL_MAX_CHANNELS
1368 * elements)
1370 static int knl_get_dimm_capacity(struct sbridge_pvt *pvt, u64 *mc_sizes)
1372 u64 sad_base, sad_size, sad_limit = 0;
1373 u64 tad_base, tad_size, tad_limit, tad_deadspace, tad_livespace;
1374 int sad_rule = 0;
1375 int tad_rule = 0;
1376 int intrlv_ways, tad_ways;
1377 u32 first_pkg, pkg;
1378 int i;
1379 u64 sad_actual_size[2]; /* sad size accounting for holes, per mc */
1380 u32 dram_rule, interleave_reg;
1381 u32 mc_route_reg[KNL_MAX_CHAS];
1382 u32 edc_route_reg[KNL_MAX_CHAS];
1383 int edram_only;
1384 char edc_route_string[KNL_MAX_EDCS*2];
1385 char mc_route_string[KNL_MAX_CHANNELS*2];
1386 int cur_reg_start;
1387 int mc;
1388 int channel;
1389 int way;
1390 int participants[KNL_MAX_CHANNELS];
1391 int participant_count = 0;
1393 for (i = 0; i < KNL_MAX_CHANNELS; i++)
1394 mc_sizes[i] = 0;
1396 /* Read the EDC route table in each CHA. */
1397 cur_reg_start = 0;
1398 for (i = 0; i < KNL_MAX_CHAS; i++) {
1399 pci_read_config_dword(pvt->knl.pci_cha[i],
1400 KNL_EDC_ROUTE, &edc_route_reg[i]);
1402 if (i > 0 && edc_route_reg[i] != edc_route_reg[i-1]) {
1403 knl_show_edc_route(edc_route_reg[i-1],
1404 edc_route_string);
1405 if (cur_reg_start == i-1)
1406 edac_dbg(0, "edc route table for CHA %d: %s\n",
1407 cur_reg_start, edc_route_string);
1408 else
1409 edac_dbg(0, "edc route table for CHA %d-%d: %s\n",
1410 cur_reg_start, i-1, edc_route_string);
1411 cur_reg_start = i;
1414 knl_show_edc_route(edc_route_reg[i-1], edc_route_string);
1415 if (cur_reg_start == i-1)
1416 edac_dbg(0, "edc route table for CHA %d: %s\n",
1417 cur_reg_start, edc_route_string);
1418 else
1419 edac_dbg(0, "edc route table for CHA %d-%d: %s\n",
1420 cur_reg_start, i-1, edc_route_string);
1422 /* Read the MC route table in each CHA. */
1423 cur_reg_start = 0;
1424 for (i = 0; i < KNL_MAX_CHAS; i++) {
1425 pci_read_config_dword(pvt->knl.pci_cha[i],
1426 KNL_MC_ROUTE, &mc_route_reg[i]);
1428 if (i > 0 && mc_route_reg[i] != mc_route_reg[i-1]) {
1429 knl_show_mc_route(mc_route_reg[i-1], mc_route_string);
1430 if (cur_reg_start == i-1)
1431 edac_dbg(0, "mc route table for CHA %d: %s\n",
1432 cur_reg_start, mc_route_string);
1433 else
1434 edac_dbg(0, "mc route table for CHA %d-%d: %s\n",
1435 cur_reg_start, i-1, mc_route_string);
1436 cur_reg_start = i;
1439 knl_show_mc_route(mc_route_reg[i-1], mc_route_string);
1440 if (cur_reg_start == i-1)
1441 edac_dbg(0, "mc route table for CHA %d: %s\n",
1442 cur_reg_start, mc_route_string);
1443 else
1444 edac_dbg(0, "mc route table for CHA %d-%d: %s\n",
1445 cur_reg_start, i-1, mc_route_string);
1447 /* Process DRAM rules */
1448 for (sad_rule = 0; sad_rule < pvt->info.max_sad; sad_rule++) {
1449 /* previous limit becomes the new base */
1450 sad_base = sad_limit;
1452 pci_read_config_dword(pvt->pci_sad0,
1453 pvt->info.dram_rule[sad_rule], &dram_rule);
1455 if (!DRAM_RULE_ENABLE(dram_rule))
1456 break;
1458 edram_only = KNL_EDRAM_ONLY(dram_rule);
1460 sad_limit = pvt->info.sad_limit(dram_rule)+1;
1461 sad_size = sad_limit - sad_base;
1463 pci_read_config_dword(pvt->pci_sad0,
1464 pvt->info.interleave_list[sad_rule], &interleave_reg);
1467 * Find out how many ways this dram rule is interleaved.
1468 * We stop when we see the first channel again.
1470 first_pkg = sad_pkg(pvt->info.interleave_pkg,
1471 interleave_reg, 0);
1472 for (intrlv_ways = 1; intrlv_ways < 8; intrlv_ways++) {
1473 pkg = sad_pkg(pvt->info.interleave_pkg,
1474 interleave_reg, intrlv_ways);
1476 if ((pkg & 0x8) == 0) {
1478 * 0 bit means memory is non-local,
1479 * which KNL doesn't support
1481 edac_dbg(0, "Unexpected interleave target %d\n",
1482 pkg);
1483 return -1;
1486 if (pkg == first_pkg)
1487 break;
1489 if (KNL_MOD3(dram_rule))
1490 intrlv_ways *= 3;
1492 edac_dbg(3, "dram rule %d (base 0x%llx, limit 0x%llx), %d way interleave%s\n",
1493 sad_rule,
1494 sad_base,
1495 sad_limit,
1496 intrlv_ways,
1497 edram_only ? ", EDRAM" : "");
1500 * Find out how big the SAD region really is by iterating
1501 * over TAD tables (SAD regions may contain holes).
1502 * Each memory controller might have a different TAD table, so
1503 * we have to look at both.
1505 * Livespace is the memory that's mapped in this TAD table,
1506 * deadspace is the holes (this could be the MMIO hole, or it
1507 * could be memory that's mapped by the other TAD table but
1508 * not this one).
1510 for (mc = 0; mc < 2; mc++) {
1511 sad_actual_size[mc] = 0;
1512 tad_livespace = 0;
1513 for (tad_rule = 0;
1514 tad_rule < ARRAY_SIZE(
1515 knl_tad_dram_limit_lo);
1516 tad_rule++) {
1517 if (knl_get_tad(pvt,
1518 tad_rule,
1520 &tad_deadspace,
1521 &tad_limit,
1522 &tad_ways))
1523 break;
1525 tad_size = (tad_limit+1) -
1526 (tad_livespace + tad_deadspace);
1527 tad_livespace += tad_size;
1528 tad_base = (tad_limit+1) - tad_size;
1530 if (tad_base < sad_base) {
1531 if (tad_limit > sad_base)
1532 edac_dbg(0, "TAD region overlaps lower SAD boundary -- TAD tables may be configured incorrectly.\n");
1533 } else if (tad_base < sad_limit) {
1534 if (tad_limit+1 > sad_limit) {
1535 edac_dbg(0, "TAD region overlaps upper SAD boundary -- TAD tables may be configured incorrectly.\n");
1536 } else {
1537 /* TAD region is completely inside SAD region */
1538 edac_dbg(3, "TAD region %d 0x%llx - 0x%llx (%lld bytes) table%d\n",
1539 tad_rule, tad_base,
1540 tad_limit, tad_size,
1541 mc);
1542 sad_actual_size[mc] += tad_size;
1545 tad_base = tad_limit+1;
1549 for (mc = 0; mc < 2; mc++) {
1550 edac_dbg(3, " total TAD DRAM footprint in table%d : 0x%llx (%lld bytes)\n",
1551 mc, sad_actual_size[mc], sad_actual_size[mc]);
1554 /* Ignore EDRAM rule */
1555 if (edram_only)
1556 continue;
1558 /* Figure out which channels participate in interleave. */
1559 for (channel = 0; channel < KNL_MAX_CHANNELS; channel++)
1560 participants[channel] = 0;
1562 /* For each channel, does at least one CHA have
1563 * this channel mapped to the given target?
1565 for (channel = 0; channel < KNL_MAX_CHANNELS; channel++) {
1566 for (way = 0; way < intrlv_ways; way++) {
1567 int target;
1568 int cha;
1570 if (KNL_MOD3(dram_rule))
1571 target = way;
1572 else
1573 target = 0x7 & sad_pkg(
1574 pvt->info.interleave_pkg, interleave_reg, way);
1576 for (cha = 0; cha < KNL_MAX_CHAS; cha++) {
1577 if (knl_get_mc_route(target,
1578 mc_route_reg[cha]) == channel
1579 && !participants[channel]) {
1580 participant_count++;
1581 participants[channel] = 1;
1582 break;
1588 if (participant_count != intrlv_ways)
1589 edac_dbg(0, "participant_count (%d) != interleave_ways (%d): DIMM size may be incorrect\n",
1590 participant_count, intrlv_ways);
1592 for (channel = 0; channel < KNL_MAX_CHANNELS; channel++) {
1593 mc = knl_channel_mc(channel);
1594 if (participants[channel]) {
1595 edac_dbg(4, "mc channel %d contributes %lld bytes via sad entry %d\n",
1596 channel,
1597 sad_actual_size[mc]/intrlv_ways,
1598 sad_rule);
1599 mc_sizes[channel] +=
1600 sad_actual_size[mc]/intrlv_ways;
1605 return 0;
1608 static int get_dimm_config(struct mem_ctl_info *mci)
1610 struct sbridge_pvt *pvt = mci->pvt_info;
1611 struct dimm_info *dimm;
1612 unsigned i, j, banks, ranks, rows, cols, npages;
1613 u64 size;
1614 u32 reg;
1615 enum edac_type mode;
1616 enum mem_type mtype;
1617 int channels = pvt->info.type == KNIGHTS_LANDING ?
1618 KNL_MAX_CHANNELS : NUM_CHANNELS;
1619 u64 knl_mc_sizes[KNL_MAX_CHANNELS];
1621 if (pvt->info.type == HASWELL || pvt->info.type == BROADWELL) {
1622 pci_read_config_dword(pvt->pci_ha0, HASWELL_HASYSDEFEATURE2, &reg);
1623 pvt->is_chan_hash = GET_BITFIELD(reg, 21, 21);
1625 if (pvt->info.type == HASWELL || pvt->info.type == BROADWELL ||
1626 pvt->info.type == KNIGHTS_LANDING)
1627 pci_read_config_dword(pvt->pci_sad1, SAD_TARGET, &reg);
1628 else
1629 pci_read_config_dword(pvt->pci_br0, SAD_TARGET, &reg);
1631 if (pvt->info.type == KNIGHTS_LANDING)
1632 pvt->sbridge_dev->source_id = SOURCE_ID_KNL(reg);
1633 else
1634 pvt->sbridge_dev->source_id = SOURCE_ID(reg);
1636 pvt->sbridge_dev->node_id = pvt->info.get_node_id(pvt);
1637 edac_dbg(0, "mc#%d: Node ID: %d, source ID: %d\n",
1638 pvt->sbridge_dev->mc,
1639 pvt->sbridge_dev->node_id,
1640 pvt->sbridge_dev->source_id);
1642 /* KNL doesn't support mirroring or lockstep,
1643 * and is always closed page
1645 if (pvt->info.type == KNIGHTS_LANDING) {
1646 mode = EDAC_S4ECD4ED;
1647 pvt->is_mirrored = false;
1649 if (knl_get_dimm_capacity(pvt, knl_mc_sizes) != 0)
1650 return -1;
1651 } else {
1652 pci_read_config_dword(pvt->pci_ras, RASENABLES, &reg);
1653 if (IS_MIRROR_ENABLED(reg)) {
1654 edac_dbg(0, "Memory mirror is enabled\n");
1655 pvt->is_mirrored = true;
1656 } else {
1657 edac_dbg(0, "Memory mirror is disabled\n");
1658 pvt->is_mirrored = false;
1661 pci_read_config_dword(pvt->pci_ta, MCMTR, &pvt->info.mcmtr);
1662 if (IS_LOCKSTEP_ENABLED(pvt->info.mcmtr)) {
1663 edac_dbg(0, "Lockstep is enabled\n");
1664 mode = EDAC_S8ECD8ED;
1665 pvt->is_lockstep = true;
1666 } else {
1667 edac_dbg(0, "Lockstep is disabled\n");
1668 mode = EDAC_S4ECD4ED;
1669 pvt->is_lockstep = false;
1671 if (IS_CLOSE_PG(pvt->info.mcmtr)) {
1672 edac_dbg(0, "address map is on closed page mode\n");
1673 pvt->is_close_pg = true;
1674 } else {
1675 edac_dbg(0, "address map is on open page mode\n");
1676 pvt->is_close_pg = false;
1680 mtype = pvt->info.get_memory_type(pvt);
1681 if (mtype == MEM_RDDR3 || mtype == MEM_RDDR4)
1682 edac_dbg(0, "Memory is registered\n");
1683 else if (mtype == MEM_UNKNOWN)
1684 edac_dbg(0, "Cannot determine memory type\n");
1685 else
1686 edac_dbg(0, "Memory is unregistered\n");
1688 if (mtype == MEM_DDR4 || mtype == MEM_RDDR4)
1689 banks = 16;
1690 else
1691 banks = 8;
1693 for (i = 0; i < channels; i++) {
1694 u32 mtr;
1696 int max_dimms_per_channel;
1698 if (pvt->info.type == KNIGHTS_LANDING) {
1699 max_dimms_per_channel = 1;
1700 if (!pvt->knl.pci_channel[i])
1701 continue;
1702 } else {
1703 max_dimms_per_channel = ARRAY_SIZE(mtr_regs);
1704 if (!pvt->pci_tad[i])
1705 continue;
1708 for (j = 0; j < max_dimms_per_channel; j++) {
1709 dimm = EDAC_DIMM_PTR(mci->layers, mci->dimms, mci->n_layers,
1710 i, j, 0);
1711 if (pvt->info.type == KNIGHTS_LANDING) {
1712 pci_read_config_dword(pvt->knl.pci_channel[i],
1713 knl_mtr_reg, &mtr);
1714 } else {
1715 pci_read_config_dword(pvt->pci_tad[i],
1716 mtr_regs[j], &mtr);
1718 edac_dbg(4, "Channel #%d MTR%d = %x\n", i, j, mtr);
1719 if (IS_DIMM_PRESENT(mtr)) {
1720 pvt->channel[i].dimms++;
1722 ranks = numrank(pvt->info.type, mtr);
1724 if (pvt->info.type == KNIGHTS_LANDING) {
1725 /* For DDR4, this is fixed. */
1726 cols = 1 << 10;
1727 rows = knl_mc_sizes[i] /
1728 ((u64) cols * ranks * banks * 8);
1729 } else {
1730 rows = numrow(mtr);
1731 cols = numcol(mtr);
1734 size = ((u64)rows * cols * banks * ranks) >> (20 - 3);
1735 npages = MiB_TO_PAGES(size);
1737 edac_dbg(0, "mc#%d: ha %d channel %d, dimm %d, %lld Mb (%d pages) bank: %d, rank: %d, row: %#x, col: %#x\n",
1738 pvt->sbridge_dev->mc, i/4, i%4, j,
1739 size, npages,
1740 banks, ranks, rows, cols);
1742 dimm->nr_pages = npages;
1743 dimm->grain = 32;
1744 dimm->dtype = pvt->info.get_width(pvt, mtr);
1745 dimm->mtype = mtype;
1746 dimm->edac_mode = mode;
1747 snprintf(dimm->label, sizeof(dimm->label),
1748 "CPU_SrcID#%u_Ha#%u_Chan#%u_DIMM#%u",
1749 pvt->sbridge_dev->source_id, i/4, i%4, j);
1754 return 0;
1757 static void get_memory_layout(const struct mem_ctl_info *mci)
1759 struct sbridge_pvt *pvt = mci->pvt_info;
1760 int i, j, k, n_sads, n_tads, sad_interl;
1761 u32 reg;
1762 u64 limit, prv = 0;
1763 u64 tmp_mb;
1764 u32 gb, mb;
1765 u32 rir_way;
1768 * Step 1) Get TOLM/TOHM ranges
1771 pvt->tolm = pvt->info.get_tolm(pvt);
1772 tmp_mb = (1 + pvt->tolm) >> 20;
1774 gb = div_u64_rem(tmp_mb, 1024, &mb);
1775 edac_dbg(0, "TOLM: %u.%03u GB (0x%016Lx)\n",
1776 gb, (mb*1000)/1024, (u64)pvt->tolm);
1778 /* Address range is already 45:25 */
1779 pvt->tohm = pvt->info.get_tohm(pvt);
1780 tmp_mb = (1 + pvt->tohm) >> 20;
1782 gb = div_u64_rem(tmp_mb, 1024, &mb);
1783 edac_dbg(0, "TOHM: %u.%03u GB (0x%016Lx)\n",
1784 gb, (mb*1000)/1024, (u64)pvt->tohm);
1787 * Step 2) Get SAD range and SAD Interleave list
1788 * TAD registers contain the interleave wayness. However, it
1789 * seems simpler to just discover it indirectly, with the
1790 * algorithm bellow.
1792 prv = 0;
1793 for (n_sads = 0; n_sads < pvt->info.max_sad; n_sads++) {
1794 /* SAD_LIMIT Address range is 45:26 */
1795 pci_read_config_dword(pvt->pci_sad0, pvt->info.dram_rule[n_sads],
1796 &reg);
1797 limit = pvt->info.sad_limit(reg);
1799 if (!DRAM_RULE_ENABLE(reg))
1800 continue;
1802 if (limit <= prv)
1803 break;
1805 tmp_mb = (limit + 1) >> 20;
1806 gb = div_u64_rem(tmp_mb, 1024, &mb);
1807 edac_dbg(0, "SAD#%d %s up to %u.%03u GB (0x%016Lx) Interleave: %s reg=0x%08x\n",
1808 n_sads,
1809 show_dram_attr(pvt->info.dram_attr(reg)),
1810 gb, (mb*1000)/1024,
1811 ((u64)tmp_mb) << 20L,
1812 pvt->info.show_interleave_mode(reg),
1813 reg);
1814 prv = limit;
1816 pci_read_config_dword(pvt->pci_sad0, pvt->info.interleave_list[n_sads],
1817 &reg);
1818 sad_interl = sad_pkg(pvt->info.interleave_pkg, reg, 0);
1819 for (j = 0; j < 8; j++) {
1820 u32 pkg = sad_pkg(pvt->info.interleave_pkg, reg, j);
1821 if (j > 0 && sad_interl == pkg)
1822 break;
1824 edac_dbg(0, "SAD#%d, interleave #%d: %d\n",
1825 n_sads, j, pkg);
1829 if (pvt->info.type == KNIGHTS_LANDING)
1830 return;
1833 * Step 3) Get TAD range
1835 prv = 0;
1836 for (n_tads = 0; n_tads < MAX_TAD; n_tads++) {
1837 pci_read_config_dword(pvt->pci_ha0, tad_dram_rule[n_tads],
1838 &reg);
1839 limit = TAD_LIMIT(reg);
1840 if (limit <= prv)
1841 break;
1842 tmp_mb = (limit + 1) >> 20;
1844 gb = div_u64_rem(tmp_mb, 1024, &mb);
1845 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",
1846 n_tads, gb, (mb*1000)/1024,
1847 ((u64)tmp_mb) << 20L,
1848 (u32)(1 << TAD_SOCK(reg)),
1849 (u32)TAD_CH(reg) + 1,
1850 (u32)TAD_TGT0(reg),
1851 (u32)TAD_TGT1(reg),
1852 (u32)TAD_TGT2(reg),
1853 (u32)TAD_TGT3(reg),
1854 reg);
1855 prv = limit;
1859 * Step 4) Get TAD offsets, per each channel
1861 for (i = 0; i < NUM_CHANNELS; i++) {
1862 if (!pvt->channel[i].dimms)
1863 continue;
1864 for (j = 0; j < n_tads; j++) {
1865 pci_read_config_dword(pvt->pci_tad[i],
1866 tad_ch_nilv_offset[j],
1867 &reg);
1868 tmp_mb = TAD_OFFSET(reg) >> 20;
1869 gb = div_u64_rem(tmp_mb, 1024, &mb);
1870 edac_dbg(0, "TAD CH#%d, offset #%d: %u.%03u GB (0x%016Lx), reg=0x%08x\n",
1871 i, j,
1872 gb, (mb*1000)/1024,
1873 ((u64)tmp_mb) << 20L,
1874 reg);
1879 * Step 6) Get RIR Wayness/Limit, per each channel
1881 for (i = 0; i < NUM_CHANNELS; i++) {
1882 if (!pvt->channel[i].dimms)
1883 continue;
1884 for (j = 0; j < MAX_RIR_RANGES; j++) {
1885 pci_read_config_dword(pvt->pci_tad[i],
1886 rir_way_limit[j],
1887 &reg);
1889 if (!IS_RIR_VALID(reg))
1890 continue;
1892 tmp_mb = pvt->info.rir_limit(reg) >> 20;
1893 rir_way = 1 << RIR_WAY(reg);
1894 gb = div_u64_rem(tmp_mb, 1024, &mb);
1895 edac_dbg(0, "CH#%d RIR#%d, limit: %u.%03u GB (0x%016Lx), way: %d, reg=0x%08x\n",
1896 i, j,
1897 gb, (mb*1000)/1024,
1898 ((u64)tmp_mb) << 20L,
1899 rir_way,
1900 reg);
1902 for (k = 0; k < rir_way; k++) {
1903 pci_read_config_dword(pvt->pci_tad[i],
1904 rir_offset[j][k],
1905 &reg);
1906 tmp_mb = RIR_OFFSET(pvt->info.type, reg) << 6;
1908 gb = div_u64_rem(tmp_mb, 1024, &mb);
1909 edac_dbg(0, "CH#%d RIR#%d INTL#%d, offset %u.%03u GB (0x%016Lx), tgt: %d, reg=0x%08x\n",
1910 i, j, k,
1911 gb, (mb*1000)/1024,
1912 ((u64)tmp_mb) << 20L,
1913 (u32)RIR_RNK_TGT(pvt->info.type, reg),
1914 reg);
1920 static struct mem_ctl_info *get_mci_for_node_id(u8 node_id)
1922 struct sbridge_dev *sbridge_dev;
1924 list_for_each_entry(sbridge_dev, &sbridge_edac_list, list) {
1925 if (sbridge_dev->node_id == node_id)
1926 return sbridge_dev->mci;
1928 return NULL;
1931 static int get_memory_error_data(struct mem_ctl_info *mci,
1932 u64 addr,
1933 u8 *socket, u8 *ha,
1934 long *channel_mask,
1935 u8 *rank,
1936 char **area_type, char *msg)
1938 struct mem_ctl_info *new_mci;
1939 struct sbridge_pvt *pvt = mci->pvt_info;
1940 struct pci_dev *pci_ha;
1941 int n_rir, n_sads, n_tads, sad_way, sck_xch;
1942 int sad_interl, idx, base_ch;
1943 int interleave_mode, shiftup = 0;
1944 unsigned sad_interleave[pvt->info.max_interleave];
1945 u32 reg, dram_rule;
1946 u8 ch_way, sck_way, pkg, sad_ha = 0, ch_add = 0;
1947 u32 tad_offset;
1948 u32 rir_way;
1949 u32 mb, gb;
1950 u64 ch_addr, offset, limit = 0, prv = 0;
1954 * Step 0) Check if the address is at special memory ranges
1955 * The check bellow is probably enough to fill all cases where
1956 * the error is not inside a memory, except for the legacy
1957 * range (e. g. VGA addresses). It is unlikely, however, that the
1958 * memory controller would generate an error on that range.
1960 if ((addr > (u64) pvt->tolm) && (addr < (1LL << 32))) {
1961 sprintf(msg, "Error at TOLM area, on addr 0x%08Lx", addr);
1962 return -EINVAL;
1964 if (addr >= (u64)pvt->tohm) {
1965 sprintf(msg, "Error at MMIOH area, on addr 0x%016Lx", addr);
1966 return -EINVAL;
1970 * Step 1) Get socket
1972 for (n_sads = 0; n_sads < pvt->info.max_sad; n_sads++) {
1973 pci_read_config_dword(pvt->pci_sad0, pvt->info.dram_rule[n_sads],
1974 &reg);
1976 if (!DRAM_RULE_ENABLE(reg))
1977 continue;
1979 limit = pvt->info.sad_limit(reg);
1980 if (limit <= prv) {
1981 sprintf(msg, "Can't discover the memory socket");
1982 return -EINVAL;
1984 if (addr <= limit)
1985 break;
1986 prv = limit;
1988 if (n_sads == pvt->info.max_sad) {
1989 sprintf(msg, "Can't discover the memory socket");
1990 return -EINVAL;
1992 dram_rule = reg;
1993 *area_type = show_dram_attr(pvt->info.dram_attr(dram_rule));
1994 interleave_mode = pvt->info.interleave_mode(dram_rule);
1996 pci_read_config_dword(pvt->pci_sad0, pvt->info.interleave_list[n_sads],
1997 &reg);
1999 if (pvt->info.type == SANDY_BRIDGE) {
2000 sad_interl = sad_pkg(pvt->info.interleave_pkg, reg, 0);
2001 for (sad_way = 0; sad_way < 8; sad_way++) {
2002 u32 pkg = sad_pkg(pvt->info.interleave_pkg, reg, sad_way);
2003 if (sad_way > 0 && sad_interl == pkg)
2004 break;
2005 sad_interleave[sad_way] = pkg;
2006 edac_dbg(0, "SAD interleave #%d: %d\n",
2007 sad_way, sad_interleave[sad_way]);
2009 edac_dbg(0, "mc#%d: Error detected on SAD#%d: address 0x%016Lx < 0x%016Lx, Interleave [%d:6]%s\n",
2010 pvt->sbridge_dev->mc,
2011 n_sads,
2012 addr,
2013 limit,
2014 sad_way + 7,
2015 !interleave_mode ? "" : "XOR[18:16]");
2016 if (interleave_mode)
2017 idx = ((addr >> 6) ^ (addr >> 16)) & 7;
2018 else
2019 idx = (addr >> 6) & 7;
2020 switch (sad_way) {
2021 case 1:
2022 idx = 0;
2023 break;
2024 case 2:
2025 idx = idx & 1;
2026 break;
2027 case 4:
2028 idx = idx & 3;
2029 break;
2030 case 8:
2031 break;
2032 default:
2033 sprintf(msg, "Can't discover socket interleave");
2034 return -EINVAL;
2036 *socket = sad_interleave[idx];
2037 edac_dbg(0, "SAD interleave index: %d (wayness %d) = CPU socket %d\n",
2038 idx, sad_way, *socket);
2039 } else if (pvt->info.type == HASWELL || pvt->info.type == BROADWELL) {
2040 int bits, a7mode = A7MODE(dram_rule);
2042 if (a7mode) {
2043 /* A7 mode swaps P9 with P6 */
2044 bits = GET_BITFIELD(addr, 7, 8) << 1;
2045 bits |= GET_BITFIELD(addr, 9, 9);
2046 } else
2047 bits = GET_BITFIELD(addr, 6, 8);
2049 if (interleave_mode == 0) {
2050 /* interleave mode will XOR {8,7,6} with {18,17,16} */
2051 idx = GET_BITFIELD(addr, 16, 18);
2052 idx ^= bits;
2053 } else
2054 idx = bits;
2056 pkg = sad_pkg(pvt->info.interleave_pkg, reg, idx);
2057 *socket = sad_pkg_socket(pkg);
2058 sad_ha = sad_pkg_ha(pkg);
2059 if (sad_ha)
2060 ch_add = 4;
2062 if (a7mode) {
2063 /* MCChanShiftUpEnable */
2064 pci_read_config_dword(pvt->pci_ha0,
2065 HASWELL_HASYSDEFEATURE2, &reg);
2066 shiftup = GET_BITFIELD(reg, 22, 22);
2069 edac_dbg(0, "SAD interleave package: %d = CPU socket %d, HA %i, shiftup: %i\n",
2070 idx, *socket, sad_ha, shiftup);
2071 } else {
2072 /* Ivy Bridge's SAD mode doesn't support XOR interleave mode */
2073 idx = (addr >> 6) & 7;
2074 pkg = sad_pkg(pvt->info.interleave_pkg, reg, idx);
2075 *socket = sad_pkg_socket(pkg);
2076 sad_ha = sad_pkg_ha(pkg);
2077 if (sad_ha)
2078 ch_add = 4;
2079 edac_dbg(0, "SAD interleave package: %d = CPU socket %d, HA %d\n",
2080 idx, *socket, sad_ha);
2083 *ha = sad_ha;
2086 * Move to the proper node structure, in order to access the
2087 * right PCI registers
2089 new_mci = get_mci_for_node_id(*socket);
2090 if (!new_mci) {
2091 sprintf(msg, "Struct for socket #%u wasn't initialized",
2092 *socket);
2093 return -EINVAL;
2095 mci = new_mci;
2096 pvt = mci->pvt_info;
2099 * Step 2) Get memory channel
2101 prv = 0;
2102 if (pvt->info.type == SANDY_BRIDGE)
2103 pci_ha = pvt->pci_ha0;
2104 else {
2105 if (sad_ha)
2106 pci_ha = pvt->pci_ha1;
2107 else
2108 pci_ha = pvt->pci_ha0;
2110 for (n_tads = 0; n_tads < MAX_TAD; n_tads++) {
2111 pci_read_config_dword(pci_ha, tad_dram_rule[n_tads], &reg);
2112 limit = TAD_LIMIT(reg);
2113 if (limit <= prv) {
2114 sprintf(msg, "Can't discover the memory channel");
2115 return -EINVAL;
2117 if (addr <= limit)
2118 break;
2119 prv = limit;
2121 if (n_tads == MAX_TAD) {
2122 sprintf(msg, "Can't discover the memory channel");
2123 return -EINVAL;
2126 ch_way = TAD_CH(reg) + 1;
2127 sck_way = TAD_SOCK(reg);
2129 if (ch_way == 3)
2130 idx = addr >> 6;
2131 else {
2132 idx = (addr >> (6 + sck_way + shiftup)) & 0x3;
2133 if (pvt->is_chan_hash)
2134 idx = haswell_chan_hash(idx, addr);
2136 idx = idx % ch_way;
2139 * FIXME: Shouldn't we use CHN_IDX_OFFSET() here, when ch_way == 3 ???
2141 switch (idx) {
2142 case 0:
2143 base_ch = TAD_TGT0(reg);
2144 break;
2145 case 1:
2146 base_ch = TAD_TGT1(reg);
2147 break;
2148 case 2:
2149 base_ch = TAD_TGT2(reg);
2150 break;
2151 case 3:
2152 base_ch = TAD_TGT3(reg);
2153 break;
2154 default:
2155 sprintf(msg, "Can't discover the TAD target");
2156 return -EINVAL;
2158 *channel_mask = 1 << base_ch;
2160 pci_read_config_dword(pvt->pci_tad[ch_add + base_ch],
2161 tad_ch_nilv_offset[n_tads],
2162 &tad_offset);
2164 if (pvt->is_mirrored) {
2165 *channel_mask |= 1 << ((base_ch + 2) % 4);
2166 switch(ch_way) {
2167 case 2:
2168 case 4:
2169 sck_xch = (1 << sck_way) * (ch_way >> 1);
2170 break;
2171 default:
2172 sprintf(msg, "Invalid mirror set. Can't decode addr");
2173 return -EINVAL;
2175 } else
2176 sck_xch = (1 << sck_way) * ch_way;
2178 if (pvt->is_lockstep)
2179 *channel_mask |= 1 << ((base_ch + 1) % 4);
2181 offset = TAD_OFFSET(tad_offset);
2183 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",
2184 n_tads,
2185 addr,
2186 limit,
2187 sck_way,
2188 ch_way,
2189 offset,
2190 idx,
2191 base_ch,
2192 *channel_mask);
2194 /* Calculate channel address */
2195 /* Remove the TAD offset */
2197 if (offset > addr) {
2198 sprintf(msg, "Can't calculate ch addr: TAD offset 0x%08Lx is too high for addr 0x%08Lx!",
2199 offset, addr);
2200 return -EINVAL;
2203 ch_addr = addr - offset;
2204 ch_addr >>= (6 + shiftup);
2205 ch_addr /= sck_xch;
2206 ch_addr <<= (6 + shiftup);
2207 ch_addr |= addr & ((1 << (6 + shiftup)) - 1);
2210 * Step 3) Decode rank
2212 for (n_rir = 0; n_rir < MAX_RIR_RANGES; n_rir++) {
2213 pci_read_config_dword(pvt->pci_tad[ch_add + base_ch],
2214 rir_way_limit[n_rir],
2215 &reg);
2217 if (!IS_RIR_VALID(reg))
2218 continue;
2220 limit = pvt->info.rir_limit(reg);
2221 gb = div_u64_rem(limit >> 20, 1024, &mb);
2222 edac_dbg(0, "RIR#%d, limit: %u.%03u GB (0x%016Lx), way: %d\n",
2223 n_rir,
2224 gb, (mb*1000)/1024,
2225 limit,
2226 1 << RIR_WAY(reg));
2227 if (ch_addr <= limit)
2228 break;
2230 if (n_rir == MAX_RIR_RANGES) {
2231 sprintf(msg, "Can't discover the memory rank for ch addr 0x%08Lx",
2232 ch_addr);
2233 return -EINVAL;
2235 rir_way = RIR_WAY(reg);
2237 if (pvt->is_close_pg)
2238 idx = (ch_addr >> 6);
2239 else
2240 idx = (ch_addr >> 13); /* FIXME: Datasheet says to shift by 15 */
2241 idx %= 1 << rir_way;
2243 pci_read_config_dword(pvt->pci_tad[ch_add + base_ch],
2244 rir_offset[n_rir][idx],
2245 &reg);
2246 *rank = RIR_RNK_TGT(pvt->info.type, reg);
2248 edac_dbg(0, "RIR#%d: channel address 0x%08Lx < 0x%08Lx, RIR interleave %d, index %d\n",
2249 n_rir,
2250 ch_addr,
2251 limit,
2252 rir_way,
2253 idx);
2255 return 0;
2258 /****************************************************************************
2259 Device initialization routines: put/get, init/exit
2260 ****************************************************************************/
2263 * sbridge_put_all_devices 'put' all the devices that we have
2264 * reserved via 'get'
2266 static void sbridge_put_devices(struct sbridge_dev *sbridge_dev)
2268 int i;
2270 edac_dbg(0, "\n");
2271 for (i = 0; i < sbridge_dev->n_devs; i++) {
2272 struct pci_dev *pdev = sbridge_dev->pdev[i];
2273 if (!pdev)
2274 continue;
2275 edac_dbg(0, "Removing dev %02x:%02x.%d\n",
2276 pdev->bus->number,
2277 PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn));
2278 pci_dev_put(pdev);
2282 static void sbridge_put_all_devices(void)
2284 struct sbridge_dev *sbridge_dev, *tmp;
2286 list_for_each_entry_safe(sbridge_dev, tmp, &sbridge_edac_list, list) {
2287 sbridge_put_devices(sbridge_dev);
2288 free_sbridge_dev(sbridge_dev);
2292 static int sbridge_get_onedevice(struct pci_dev **prev,
2293 u8 *num_mc,
2294 const struct pci_id_table *table,
2295 const unsigned devno,
2296 const int multi_bus)
2298 struct sbridge_dev *sbridge_dev;
2299 const struct pci_id_descr *dev_descr = &table->descr[devno];
2300 struct pci_dev *pdev = NULL;
2301 u8 bus = 0;
2303 sbridge_printk(KERN_DEBUG,
2304 "Seeking for: PCI ID %04x:%04x\n",
2305 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
2307 pdev = pci_get_device(PCI_VENDOR_ID_INTEL,
2308 dev_descr->dev_id, *prev);
2310 if (!pdev) {
2311 if (*prev) {
2312 *prev = pdev;
2313 return 0;
2316 if (dev_descr->optional)
2317 return 0;
2319 /* if the HA wasn't found */
2320 if (devno == 0)
2321 return -ENODEV;
2323 sbridge_printk(KERN_INFO,
2324 "Device not found: %04x:%04x\n",
2325 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
2327 /* End of list, leave */
2328 return -ENODEV;
2330 bus = pdev->bus->number;
2332 sbridge_dev = get_sbridge_dev(bus, multi_bus);
2333 if (!sbridge_dev) {
2334 sbridge_dev = alloc_sbridge_dev(bus, table);
2335 if (!sbridge_dev) {
2336 pci_dev_put(pdev);
2337 return -ENOMEM;
2339 (*num_mc)++;
2342 if (sbridge_dev->pdev[devno]) {
2343 sbridge_printk(KERN_ERR,
2344 "Duplicated device for %04x:%04x\n",
2345 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
2346 pci_dev_put(pdev);
2347 return -ENODEV;
2350 sbridge_dev->pdev[devno] = pdev;
2352 /* Be sure that the device is enabled */
2353 if (unlikely(pci_enable_device(pdev) < 0)) {
2354 sbridge_printk(KERN_ERR,
2355 "Couldn't enable %04x:%04x\n",
2356 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
2357 return -ENODEV;
2360 edac_dbg(0, "Detected %04x:%04x\n",
2361 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
2364 * As stated on drivers/pci/search.c, the reference count for
2365 * @from is always decremented if it is not %NULL. So, as we need
2366 * to get all devices up to null, we need to do a get for the device
2368 pci_dev_get(pdev);
2370 *prev = pdev;
2372 return 0;
2376 * sbridge_get_all_devices - Find and perform 'get' operation on the MCH's
2377 * devices we want to reference for this driver.
2378 * @num_mc: pointer to the memory controllers count, to be incremented in case
2379 * of success.
2380 * @table: model specific table
2382 * returns 0 in case of success or error code
2384 static int sbridge_get_all_devices(u8 *num_mc,
2385 const struct pci_id_table *table)
2387 int i, rc;
2388 struct pci_dev *pdev = NULL;
2389 int allow_dups = 0;
2390 int multi_bus = 0;
2392 if (table->type == KNIGHTS_LANDING)
2393 allow_dups = multi_bus = 1;
2394 while (table && table->descr) {
2395 for (i = 0; i < table->n_devs; i++) {
2396 if (!allow_dups || i == 0 ||
2397 table->descr[i].dev_id !=
2398 table->descr[i-1].dev_id) {
2399 pdev = NULL;
2401 do {
2402 rc = sbridge_get_onedevice(&pdev, num_mc,
2403 table, i, multi_bus);
2404 if (rc < 0) {
2405 if (i == 0) {
2406 i = table->n_devs;
2407 break;
2409 sbridge_put_all_devices();
2410 return -ENODEV;
2412 } while (pdev && !allow_dups);
2414 table++;
2417 return 0;
2420 static int sbridge_mci_bind_devs(struct mem_ctl_info *mci,
2421 struct sbridge_dev *sbridge_dev)
2423 struct sbridge_pvt *pvt = mci->pvt_info;
2424 struct pci_dev *pdev;
2425 u8 saw_chan_mask = 0;
2426 int i;
2428 for (i = 0; i < sbridge_dev->n_devs; i++) {
2429 pdev = sbridge_dev->pdev[i];
2430 if (!pdev)
2431 continue;
2433 switch (pdev->device) {
2434 case PCI_DEVICE_ID_INTEL_SBRIDGE_SAD0:
2435 pvt->pci_sad0 = pdev;
2436 break;
2437 case PCI_DEVICE_ID_INTEL_SBRIDGE_SAD1:
2438 pvt->pci_sad1 = pdev;
2439 break;
2440 case PCI_DEVICE_ID_INTEL_SBRIDGE_BR:
2441 pvt->pci_br0 = pdev;
2442 break;
2443 case PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_HA0:
2444 pvt->pci_ha0 = pdev;
2445 break;
2446 case PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TA:
2447 pvt->pci_ta = pdev;
2448 break;
2449 case PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_RAS:
2450 pvt->pci_ras = pdev;
2451 break;
2452 case PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD0:
2453 case PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD1:
2454 case PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD2:
2455 case PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD3:
2457 int id = pdev->device - PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD0;
2458 pvt->pci_tad[id] = pdev;
2459 saw_chan_mask |= 1 << id;
2461 break;
2462 case PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_DDRIO:
2463 pvt->pci_ddrio = pdev;
2464 break;
2465 default:
2466 goto error;
2469 edac_dbg(0, "Associated PCI %02x:%02x, bus %d with dev = %p\n",
2470 pdev->vendor, pdev->device,
2471 sbridge_dev->bus,
2472 pdev);
2475 /* Check if everything were registered */
2476 if (!pvt->pci_sad0 || !pvt->pci_sad1 || !pvt->pci_ha0 ||
2477 !pvt->pci_ras || !pvt->pci_ta)
2478 goto enodev;
2480 if (saw_chan_mask != 0x0f)
2481 goto enodev;
2482 return 0;
2484 enodev:
2485 sbridge_printk(KERN_ERR, "Some needed devices are missing\n");
2486 return -ENODEV;
2488 error:
2489 sbridge_printk(KERN_ERR, "Unexpected device %02x:%02x\n",
2490 PCI_VENDOR_ID_INTEL, pdev->device);
2491 return -EINVAL;
2494 static int ibridge_mci_bind_devs(struct mem_ctl_info *mci,
2495 struct sbridge_dev *sbridge_dev)
2497 struct sbridge_pvt *pvt = mci->pvt_info;
2498 struct pci_dev *pdev;
2499 u8 saw_chan_mask = 0;
2500 int i;
2502 for (i = 0; i < sbridge_dev->n_devs; i++) {
2503 pdev = sbridge_dev->pdev[i];
2504 if (!pdev)
2505 continue;
2507 switch (pdev->device) {
2508 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0:
2509 pvt->pci_ha0 = pdev;
2510 break;
2511 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TA:
2512 pvt->pci_ta = pdev;
2513 break;
2514 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_RAS:
2515 pvt->pci_ras = pdev;
2516 break;
2517 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD0:
2518 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD1:
2519 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD2:
2520 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD3:
2522 int id = pdev->device - PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD0;
2523 pvt->pci_tad[id] = pdev;
2524 saw_chan_mask |= 1 << id;
2526 break;
2527 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_2HA_DDRIO0:
2528 pvt->pci_ddrio = pdev;
2529 break;
2530 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_1HA_DDRIO0:
2531 pvt->pci_ddrio = pdev;
2532 break;
2533 case PCI_DEVICE_ID_INTEL_IBRIDGE_SAD:
2534 pvt->pci_sad0 = pdev;
2535 break;
2536 case PCI_DEVICE_ID_INTEL_IBRIDGE_BR0:
2537 pvt->pci_br0 = pdev;
2538 break;
2539 case PCI_DEVICE_ID_INTEL_IBRIDGE_BR1:
2540 pvt->pci_br1 = pdev;
2541 break;
2542 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1:
2543 pvt->pci_ha1 = pdev;
2544 break;
2545 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD0:
2546 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD1:
2547 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD2:
2548 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD3:
2550 int id = pdev->device - PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD0 + 4;
2551 pvt->pci_tad[id] = pdev;
2552 saw_chan_mask |= 1 << id;
2554 break;
2555 default:
2556 goto error;
2559 edac_dbg(0, "Associated PCI %02x.%02d.%d with dev = %p\n",
2560 sbridge_dev->bus,
2561 PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn),
2562 pdev);
2565 /* Check if everything were registered */
2566 if (!pvt->pci_sad0 || !pvt->pci_ha0 || !pvt->pci_br0 ||
2567 !pvt->pci_br1 || !pvt->pci_ras || !pvt->pci_ta)
2568 goto enodev;
2570 if (saw_chan_mask != 0x0f && /* -EN */
2571 saw_chan_mask != 0x33 && /* -EP */
2572 saw_chan_mask != 0xff) /* -EX */
2573 goto enodev;
2574 return 0;
2576 enodev:
2577 sbridge_printk(KERN_ERR, "Some needed devices are missing\n");
2578 return -ENODEV;
2580 error:
2581 sbridge_printk(KERN_ERR,
2582 "Unexpected device %02x:%02x\n", PCI_VENDOR_ID_INTEL,
2583 pdev->device);
2584 return -EINVAL;
2587 static int haswell_mci_bind_devs(struct mem_ctl_info *mci,
2588 struct sbridge_dev *sbridge_dev)
2590 struct sbridge_pvt *pvt = mci->pvt_info;
2591 struct pci_dev *pdev;
2592 u8 saw_chan_mask = 0;
2593 int i;
2595 /* there's only one device per system; not tied to any bus */
2596 if (pvt->info.pci_vtd == NULL)
2597 /* result will be checked later */
2598 pvt->info.pci_vtd = pci_get_device(PCI_VENDOR_ID_INTEL,
2599 PCI_DEVICE_ID_INTEL_HASWELL_IMC_VTD_MISC,
2600 NULL);
2602 for (i = 0; i < sbridge_dev->n_devs; i++) {
2603 pdev = sbridge_dev->pdev[i];
2604 if (!pdev)
2605 continue;
2607 switch (pdev->device) {
2608 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_CBO_SAD0:
2609 pvt->pci_sad0 = pdev;
2610 break;
2611 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_CBO_SAD1:
2612 pvt->pci_sad1 = pdev;
2613 break;
2614 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0:
2615 pvt->pci_ha0 = pdev;
2616 break;
2617 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TA:
2618 pvt->pci_ta = pdev;
2619 break;
2620 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_THERMAL:
2621 pvt->pci_ras = pdev;
2622 break;
2623 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD0:
2624 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD1:
2625 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD2:
2626 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD3:
2628 int id = pdev->device - PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD0;
2630 pvt->pci_tad[id] = pdev;
2631 saw_chan_mask |= 1 << id;
2633 break;
2634 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD0:
2635 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD1:
2636 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD2:
2637 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD3:
2639 int id = pdev->device - PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD0 + 4;
2641 pvt->pci_tad[id] = pdev;
2642 saw_chan_mask |= 1 << id;
2644 break;
2645 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO0:
2646 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO1:
2647 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO2:
2648 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO3:
2649 if (!pvt->pci_ddrio)
2650 pvt->pci_ddrio = pdev;
2651 break;
2652 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1:
2653 pvt->pci_ha1 = pdev;
2654 break;
2655 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TA:
2656 pvt->pci_ha1_ta = pdev;
2657 break;
2658 default:
2659 break;
2662 edac_dbg(0, "Associated PCI %02x.%02d.%d with dev = %p\n",
2663 sbridge_dev->bus,
2664 PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn),
2665 pdev);
2668 /* Check if everything were registered */
2669 if (!pvt->pci_sad0 || !pvt->pci_ha0 || !pvt->pci_sad1 ||
2670 !pvt->pci_ras || !pvt->pci_ta || !pvt->info.pci_vtd)
2671 goto enodev;
2673 if (saw_chan_mask != 0x0f && /* -EN */
2674 saw_chan_mask != 0x33 && /* -EP */
2675 saw_chan_mask != 0xff) /* -EX */
2676 goto enodev;
2677 return 0;
2679 enodev:
2680 sbridge_printk(KERN_ERR, "Some needed devices are missing\n");
2681 return -ENODEV;
2684 static int broadwell_mci_bind_devs(struct mem_ctl_info *mci,
2685 struct sbridge_dev *sbridge_dev)
2687 struct sbridge_pvt *pvt = mci->pvt_info;
2688 struct pci_dev *pdev;
2689 u8 saw_chan_mask = 0;
2690 int i;
2692 /* there's only one device per system; not tied to any bus */
2693 if (pvt->info.pci_vtd == NULL)
2694 /* result will be checked later */
2695 pvt->info.pci_vtd = pci_get_device(PCI_VENDOR_ID_INTEL,
2696 PCI_DEVICE_ID_INTEL_BROADWELL_IMC_VTD_MISC,
2697 NULL);
2699 for (i = 0; i < sbridge_dev->n_devs; i++) {
2700 pdev = sbridge_dev->pdev[i];
2701 if (!pdev)
2702 continue;
2704 switch (pdev->device) {
2705 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_CBO_SAD0:
2706 pvt->pci_sad0 = pdev;
2707 break;
2708 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_CBO_SAD1:
2709 pvt->pci_sad1 = pdev;
2710 break;
2711 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0:
2712 pvt->pci_ha0 = pdev;
2713 break;
2714 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TA:
2715 pvt->pci_ta = pdev;
2716 break;
2717 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_THERMAL:
2718 pvt->pci_ras = pdev;
2719 break;
2720 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD0:
2721 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD1:
2722 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD2:
2723 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD3:
2725 int id = pdev->device - PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD0;
2726 pvt->pci_tad[id] = pdev;
2727 saw_chan_mask |= 1 << id;
2729 break;
2730 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD0:
2731 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD1:
2732 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD2:
2733 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD3:
2735 int id = pdev->device - PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD0 + 4;
2736 pvt->pci_tad[id] = pdev;
2737 saw_chan_mask |= 1 << id;
2739 break;
2740 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_DDRIO0:
2741 pvt->pci_ddrio = pdev;
2742 break;
2743 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1:
2744 pvt->pci_ha1 = pdev;
2745 break;
2746 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TA:
2747 pvt->pci_ha1_ta = pdev;
2748 break;
2749 default:
2750 break;
2753 edac_dbg(0, "Associated PCI %02x.%02d.%d with dev = %p\n",
2754 sbridge_dev->bus,
2755 PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn),
2756 pdev);
2759 /* Check if everything were registered */
2760 if (!pvt->pci_sad0 || !pvt->pci_ha0 || !pvt->pci_sad1 ||
2761 !pvt->pci_ras || !pvt->pci_ta || !pvt->info.pci_vtd)
2762 goto enodev;
2764 if (saw_chan_mask != 0x0f && /* -EN */
2765 saw_chan_mask != 0x33 && /* -EP */
2766 saw_chan_mask != 0xff) /* -EX */
2767 goto enodev;
2768 return 0;
2770 enodev:
2771 sbridge_printk(KERN_ERR, "Some needed devices are missing\n");
2772 return -ENODEV;
2775 static int knl_mci_bind_devs(struct mem_ctl_info *mci,
2776 struct sbridge_dev *sbridge_dev)
2778 struct sbridge_pvt *pvt = mci->pvt_info;
2779 struct pci_dev *pdev;
2780 int dev, func;
2782 int i;
2783 int devidx;
2785 for (i = 0; i < sbridge_dev->n_devs; i++) {
2786 pdev = sbridge_dev->pdev[i];
2787 if (!pdev)
2788 continue;
2790 /* Extract PCI device and function. */
2791 dev = (pdev->devfn >> 3) & 0x1f;
2792 func = pdev->devfn & 0x7;
2794 switch (pdev->device) {
2795 case PCI_DEVICE_ID_INTEL_KNL_IMC_MC:
2796 if (dev == 8)
2797 pvt->knl.pci_mc0 = pdev;
2798 else if (dev == 9)
2799 pvt->knl.pci_mc1 = pdev;
2800 else {
2801 sbridge_printk(KERN_ERR,
2802 "Memory controller in unexpected place! (dev %d, fn %d)\n",
2803 dev, func);
2804 continue;
2806 break;
2808 case PCI_DEVICE_ID_INTEL_KNL_IMC_SAD0:
2809 pvt->pci_sad0 = pdev;
2810 break;
2812 case PCI_DEVICE_ID_INTEL_KNL_IMC_SAD1:
2813 pvt->pci_sad1 = pdev;
2814 break;
2816 case PCI_DEVICE_ID_INTEL_KNL_IMC_CHA:
2817 /* There are one of these per tile, and range from
2818 * 1.14.0 to 1.18.5.
2820 devidx = ((dev-14)*8)+func;
2822 if (devidx < 0 || devidx >= KNL_MAX_CHAS) {
2823 sbridge_printk(KERN_ERR,
2824 "Caching and Home Agent in unexpected place! (dev %d, fn %d)\n",
2825 dev, func);
2826 continue;
2829 WARN_ON(pvt->knl.pci_cha[devidx] != NULL);
2831 pvt->knl.pci_cha[devidx] = pdev;
2832 break;
2834 case PCI_DEVICE_ID_INTEL_KNL_IMC_CHANNEL:
2835 devidx = -1;
2838 * MC0 channels 0-2 are device 9 function 2-4,
2839 * MC1 channels 3-5 are device 8 function 2-4.
2842 if (dev == 9)
2843 devidx = func-2;
2844 else if (dev == 8)
2845 devidx = 3 + (func-2);
2847 if (devidx < 0 || devidx >= KNL_MAX_CHANNELS) {
2848 sbridge_printk(KERN_ERR,
2849 "DRAM Channel Registers in unexpected place! (dev %d, fn %d)\n",
2850 dev, func);
2851 continue;
2854 WARN_ON(pvt->knl.pci_channel[devidx] != NULL);
2855 pvt->knl.pci_channel[devidx] = pdev;
2856 break;
2858 case PCI_DEVICE_ID_INTEL_KNL_IMC_TOLHM:
2859 pvt->knl.pci_mc_info = pdev;
2860 break;
2862 case PCI_DEVICE_ID_INTEL_KNL_IMC_TA:
2863 pvt->pci_ta = pdev;
2864 break;
2866 default:
2867 sbridge_printk(KERN_ERR, "Unexpected device %d\n",
2868 pdev->device);
2869 break;
2873 if (!pvt->knl.pci_mc0 || !pvt->knl.pci_mc1 ||
2874 !pvt->pci_sad0 || !pvt->pci_sad1 ||
2875 !pvt->pci_ta) {
2876 goto enodev;
2879 for (i = 0; i < KNL_MAX_CHANNELS; i++) {
2880 if (!pvt->knl.pci_channel[i]) {
2881 sbridge_printk(KERN_ERR, "Missing channel %d\n", i);
2882 goto enodev;
2886 for (i = 0; i < KNL_MAX_CHAS; i++) {
2887 if (!pvt->knl.pci_cha[i]) {
2888 sbridge_printk(KERN_ERR, "Missing CHA %d\n", i);
2889 goto enodev;
2893 return 0;
2895 enodev:
2896 sbridge_printk(KERN_ERR, "Some needed devices are missing\n");
2897 return -ENODEV;
2900 /****************************************************************************
2901 Error check routines
2902 ****************************************************************************/
2905 * While Sandy Bridge has error count registers, SMI BIOS read values from
2906 * and resets the counters. So, they are not reliable for the OS to read
2907 * from them. So, we have no option but to just trust on whatever MCE is
2908 * telling us about the errors.
2910 static void sbridge_mce_output_error(struct mem_ctl_info *mci,
2911 const struct mce *m)
2913 struct mem_ctl_info *new_mci;
2914 struct sbridge_pvt *pvt = mci->pvt_info;
2915 enum hw_event_mc_err_type tp_event;
2916 char *type, *optype, msg[256];
2917 bool ripv = GET_BITFIELD(m->mcgstatus, 0, 0);
2918 bool overflow = GET_BITFIELD(m->status, 62, 62);
2919 bool uncorrected_error = GET_BITFIELD(m->status, 61, 61);
2920 bool recoverable;
2921 u32 core_err_cnt = GET_BITFIELD(m->status, 38, 52);
2922 u32 mscod = GET_BITFIELD(m->status, 16, 31);
2923 u32 errcode = GET_BITFIELD(m->status, 0, 15);
2924 u32 channel = GET_BITFIELD(m->status, 0, 3);
2925 u32 optypenum = GET_BITFIELD(m->status, 4, 6);
2926 long channel_mask, first_channel;
2927 u8 rank, socket, ha;
2928 int rc, dimm;
2929 char *area_type = NULL;
2931 if (pvt->info.type != SANDY_BRIDGE)
2932 recoverable = true;
2933 else
2934 recoverable = GET_BITFIELD(m->status, 56, 56);
2936 if (uncorrected_error) {
2937 if (ripv) {
2938 type = "FATAL";
2939 tp_event = HW_EVENT_ERR_FATAL;
2940 } else {
2941 type = "NON_FATAL";
2942 tp_event = HW_EVENT_ERR_UNCORRECTED;
2944 } else {
2945 type = "CORRECTED";
2946 tp_event = HW_EVENT_ERR_CORRECTED;
2950 * According with Table 15-9 of the Intel Architecture spec vol 3A,
2951 * memory errors should fit in this mask:
2952 * 000f 0000 1mmm cccc (binary)
2953 * where:
2954 * f = Correction Report Filtering Bit. If 1, subsequent errors
2955 * won't be shown
2956 * mmm = error type
2957 * cccc = channel
2958 * If the mask doesn't match, report an error to the parsing logic
2960 if (! ((errcode & 0xef80) == 0x80)) {
2961 optype = "Can't parse: it is not a mem";
2962 } else {
2963 switch (optypenum) {
2964 case 0:
2965 optype = "generic undef request error";
2966 break;
2967 case 1:
2968 optype = "memory read error";
2969 break;
2970 case 2:
2971 optype = "memory write error";
2972 break;
2973 case 3:
2974 optype = "addr/cmd error";
2975 break;
2976 case 4:
2977 optype = "memory scrubbing error";
2978 break;
2979 default:
2980 optype = "reserved";
2981 break;
2985 /* Only decode errors with an valid address (ADDRV) */
2986 if (!GET_BITFIELD(m->status, 58, 58))
2987 return;
2989 if (pvt->info.type == KNIGHTS_LANDING) {
2990 if (channel == 14) {
2991 edac_dbg(0, "%s%s err_code:%04x:%04x EDRAM bank %d\n",
2992 overflow ? " OVERFLOW" : "",
2993 (uncorrected_error && recoverable)
2994 ? " recoverable" : "",
2995 mscod, errcode,
2996 m->bank);
2997 } else {
2998 char A = *("A");
3001 * Reported channel is in range 0-2, so we can't map it
3002 * back to mc. To figure out mc we check machine check
3003 * bank register that reported this error.
3004 * bank15 means mc0 and bank16 means mc1.
3006 channel = knl_channel_remap(m->bank == 16, channel);
3007 channel_mask = 1 << channel;
3009 snprintf(msg, sizeof(msg),
3010 "%s%s err_code:%04x:%04x channel:%d (DIMM_%c)",
3011 overflow ? " OVERFLOW" : "",
3012 (uncorrected_error && recoverable)
3013 ? " recoverable" : " ",
3014 mscod, errcode, channel, A + channel);
3015 edac_mc_handle_error(tp_event, mci, core_err_cnt,
3016 m->addr >> PAGE_SHIFT, m->addr & ~PAGE_MASK, 0,
3017 channel, 0, -1,
3018 optype, msg);
3020 return;
3021 } else {
3022 rc = get_memory_error_data(mci, m->addr, &socket, &ha,
3023 &channel_mask, &rank, &area_type, msg);
3026 if (rc < 0)
3027 goto err_parsing;
3028 new_mci = get_mci_for_node_id(socket);
3029 if (!new_mci) {
3030 strcpy(msg, "Error: socket got corrupted!");
3031 goto err_parsing;
3033 mci = new_mci;
3034 pvt = mci->pvt_info;
3036 first_channel = find_first_bit(&channel_mask, NUM_CHANNELS);
3038 if (rank < 4)
3039 dimm = 0;
3040 else if (rank < 8)
3041 dimm = 1;
3042 else
3043 dimm = 2;
3047 * FIXME: On some memory configurations (mirror, lockstep), the
3048 * Memory Controller can't point the error to a single DIMM. The
3049 * EDAC core should be handling the channel mask, in order to point
3050 * to the group of dimm's where the error may be happening.
3052 if (!pvt->is_lockstep && !pvt->is_mirrored && !pvt->is_close_pg)
3053 channel = first_channel;
3055 snprintf(msg, sizeof(msg),
3056 "%s%s area:%s err_code:%04x:%04x socket:%d ha:%d channel_mask:%ld rank:%d",
3057 overflow ? " OVERFLOW" : "",
3058 (uncorrected_error && recoverable) ? " recoverable" : "",
3059 area_type,
3060 mscod, errcode,
3061 socket, ha,
3062 channel_mask,
3063 rank);
3065 edac_dbg(0, "%s\n", msg);
3067 /* FIXME: need support for channel mask */
3069 if (channel == CHANNEL_UNSPECIFIED)
3070 channel = -1;
3072 /* Call the helper to output message */
3073 edac_mc_handle_error(tp_event, mci, core_err_cnt,
3074 m->addr >> PAGE_SHIFT, m->addr & ~PAGE_MASK, 0,
3075 4*ha+channel, dimm, -1,
3076 optype, msg);
3077 return;
3078 err_parsing:
3079 edac_mc_handle_error(tp_event, mci, core_err_cnt, 0, 0, 0,
3080 -1, -1, -1,
3081 msg, "");
3086 * Check that logging is enabled and that this is the right type
3087 * of error for us to handle.
3089 static int sbridge_mce_check_error(struct notifier_block *nb, unsigned long val,
3090 void *data)
3092 struct mce *mce = (struct mce *)data;
3093 struct mem_ctl_info *mci;
3094 struct sbridge_pvt *pvt;
3095 char *type;
3097 if (get_edac_report_status() == EDAC_REPORTING_DISABLED)
3098 return NOTIFY_DONE;
3100 mci = get_mci_for_node_id(mce->socketid);
3101 if (!mci)
3102 return NOTIFY_DONE;
3103 pvt = mci->pvt_info;
3106 * Just let mcelog handle it if the error is
3107 * outside the memory controller. A memory error
3108 * is indicated by bit 7 = 1 and bits = 8-11,13-15 = 0.
3109 * bit 12 has an special meaning.
3111 if ((mce->status & 0xefff) >> 7 != 1)
3112 return NOTIFY_DONE;
3114 if (mce->mcgstatus & MCG_STATUS_MCIP)
3115 type = "Exception";
3116 else
3117 type = "Event";
3119 sbridge_mc_printk(mci, KERN_DEBUG, "HANDLING MCE MEMORY ERROR\n");
3121 sbridge_mc_printk(mci, KERN_DEBUG, "CPU %d: Machine Check %s: %Lx "
3122 "Bank %d: %016Lx\n", mce->extcpu, type,
3123 mce->mcgstatus, mce->bank, mce->status);
3124 sbridge_mc_printk(mci, KERN_DEBUG, "TSC %llx ", mce->tsc);
3125 sbridge_mc_printk(mci, KERN_DEBUG, "ADDR %llx ", mce->addr);
3126 sbridge_mc_printk(mci, KERN_DEBUG, "MISC %llx ", mce->misc);
3128 sbridge_mc_printk(mci, KERN_DEBUG, "PROCESSOR %u:%x TIME %llu SOCKET "
3129 "%u APIC %x\n", mce->cpuvendor, mce->cpuid,
3130 mce->time, mce->socketid, mce->apicid);
3132 sbridge_mce_output_error(mci, mce);
3134 /* Advice mcelog that the error were handled */
3135 return NOTIFY_STOP;
3138 static struct notifier_block sbridge_mce_dec = {
3139 .notifier_call = sbridge_mce_check_error,
3142 /****************************************************************************
3143 EDAC register/unregister logic
3144 ****************************************************************************/
3146 static void sbridge_unregister_mci(struct sbridge_dev *sbridge_dev)
3148 struct mem_ctl_info *mci = sbridge_dev->mci;
3149 struct sbridge_pvt *pvt;
3151 if (unlikely(!mci || !mci->pvt_info)) {
3152 edac_dbg(0, "MC: dev = %p\n", &sbridge_dev->pdev[0]->dev);
3154 sbridge_printk(KERN_ERR, "Couldn't find mci handler\n");
3155 return;
3158 pvt = mci->pvt_info;
3160 edac_dbg(0, "MC: mci = %p, dev = %p\n",
3161 mci, &sbridge_dev->pdev[0]->dev);
3163 /* Remove MC sysfs nodes */
3164 edac_mc_del_mc(mci->pdev);
3166 edac_dbg(1, "%s: free mci struct\n", mci->ctl_name);
3167 kfree(mci->ctl_name);
3168 edac_mc_free(mci);
3169 sbridge_dev->mci = NULL;
3172 static int sbridge_register_mci(struct sbridge_dev *sbridge_dev, enum type type)
3174 struct mem_ctl_info *mci;
3175 struct edac_mc_layer layers[2];
3176 struct sbridge_pvt *pvt;
3177 struct pci_dev *pdev = sbridge_dev->pdev[0];
3178 int rc;
3180 /* Check the number of active and not disabled channels */
3181 rc = check_if_ecc_is_active(sbridge_dev->bus, type);
3182 if (unlikely(rc < 0))
3183 return rc;
3185 /* allocate a new MC control structure */
3186 layers[0].type = EDAC_MC_LAYER_CHANNEL;
3187 layers[0].size = type == KNIGHTS_LANDING ?
3188 KNL_MAX_CHANNELS : NUM_CHANNELS;
3189 layers[0].is_virt_csrow = false;
3190 layers[1].type = EDAC_MC_LAYER_SLOT;
3191 layers[1].size = type == KNIGHTS_LANDING ? 1 : MAX_DIMMS;
3192 layers[1].is_virt_csrow = true;
3193 mci = edac_mc_alloc(sbridge_dev->mc, ARRAY_SIZE(layers), layers,
3194 sizeof(*pvt));
3196 if (unlikely(!mci))
3197 return -ENOMEM;
3199 edac_dbg(0, "MC: mci = %p, dev = %p\n",
3200 mci, &pdev->dev);
3202 pvt = mci->pvt_info;
3203 memset(pvt, 0, sizeof(*pvt));
3205 /* Associate sbridge_dev and mci for future usage */
3206 pvt->sbridge_dev = sbridge_dev;
3207 sbridge_dev->mci = mci;
3209 mci->mtype_cap = type == KNIGHTS_LANDING ?
3210 MEM_FLAG_DDR4 : MEM_FLAG_DDR3;
3211 mci->edac_ctl_cap = EDAC_FLAG_NONE;
3212 mci->edac_cap = EDAC_FLAG_NONE;
3213 mci->mod_name = "sbridge_edac.c";
3214 mci->mod_ver = SBRIDGE_REVISION;
3215 mci->dev_name = pci_name(pdev);
3216 mci->ctl_page_to_phys = NULL;
3218 pvt->info.type = type;
3219 switch (type) {
3220 case IVY_BRIDGE:
3221 pvt->info.rankcfgr = IB_RANK_CFG_A;
3222 pvt->info.get_tolm = ibridge_get_tolm;
3223 pvt->info.get_tohm = ibridge_get_tohm;
3224 pvt->info.dram_rule = ibridge_dram_rule;
3225 pvt->info.get_memory_type = get_memory_type;
3226 pvt->info.get_node_id = get_node_id;
3227 pvt->info.rir_limit = rir_limit;
3228 pvt->info.sad_limit = sad_limit;
3229 pvt->info.interleave_mode = interleave_mode;
3230 pvt->info.show_interleave_mode = show_interleave_mode;
3231 pvt->info.dram_attr = dram_attr;
3232 pvt->info.max_sad = ARRAY_SIZE(ibridge_dram_rule);
3233 pvt->info.interleave_list = ibridge_interleave_list;
3234 pvt->info.max_interleave = ARRAY_SIZE(ibridge_interleave_list);
3235 pvt->info.interleave_pkg = ibridge_interleave_pkg;
3236 pvt->info.get_width = ibridge_get_width;
3237 mci->ctl_name = kasprintf(GFP_KERNEL, "Ivy Bridge Socket#%d", mci->mc_idx);
3239 /* Store pci devices at mci for faster access */
3240 rc = ibridge_mci_bind_devs(mci, sbridge_dev);
3241 if (unlikely(rc < 0))
3242 goto fail0;
3243 break;
3244 case SANDY_BRIDGE:
3245 pvt->info.rankcfgr = SB_RANK_CFG_A;
3246 pvt->info.get_tolm = sbridge_get_tolm;
3247 pvt->info.get_tohm = sbridge_get_tohm;
3248 pvt->info.dram_rule = sbridge_dram_rule;
3249 pvt->info.get_memory_type = get_memory_type;
3250 pvt->info.get_node_id = get_node_id;
3251 pvt->info.rir_limit = rir_limit;
3252 pvt->info.sad_limit = sad_limit;
3253 pvt->info.interleave_mode = interleave_mode;
3254 pvt->info.show_interleave_mode = show_interleave_mode;
3255 pvt->info.dram_attr = dram_attr;
3256 pvt->info.max_sad = ARRAY_SIZE(sbridge_dram_rule);
3257 pvt->info.interleave_list = sbridge_interleave_list;
3258 pvt->info.max_interleave = ARRAY_SIZE(sbridge_interleave_list);
3259 pvt->info.interleave_pkg = sbridge_interleave_pkg;
3260 pvt->info.get_width = sbridge_get_width;
3261 mci->ctl_name = kasprintf(GFP_KERNEL, "Sandy Bridge Socket#%d", mci->mc_idx);
3263 /* Store pci devices at mci for faster access */
3264 rc = sbridge_mci_bind_devs(mci, sbridge_dev);
3265 if (unlikely(rc < 0))
3266 goto fail0;
3267 break;
3268 case HASWELL:
3269 /* rankcfgr isn't used */
3270 pvt->info.get_tolm = haswell_get_tolm;
3271 pvt->info.get_tohm = haswell_get_tohm;
3272 pvt->info.dram_rule = ibridge_dram_rule;
3273 pvt->info.get_memory_type = haswell_get_memory_type;
3274 pvt->info.get_node_id = haswell_get_node_id;
3275 pvt->info.rir_limit = haswell_rir_limit;
3276 pvt->info.sad_limit = sad_limit;
3277 pvt->info.interleave_mode = interleave_mode;
3278 pvt->info.show_interleave_mode = show_interleave_mode;
3279 pvt->info.dram_attr = dram_attr;
3280 pvt->info.max_sad = ARRAY_SIZE(ibridge_dram_rule);
3281 pvt->info.interleave_list = ibridge_interleave_list;
3282 pvt->info.max_interleave = ARRAY_SIZE(ibridge_interleave_list);
3283 pvt->info.interleave_pkg = ibridge_interleave_pkg;
3284 pvt->info.get_width = ibridge_get_width;
3285 mci->ctl_name = kasprintf(GFP_KERNEL, "Haswell Socket#%d", mci->mc_idx);
3287 /* Store pci devices at mci for faster access */
3288 rc = haswell_mci_bind_devs(mci, sbridge_dev);
3289 if (unlikely(rc < 0))
3290 goto fail0;
3291 break;
3292 case BROADWELL:
3293 /* rankcfgr isn't used */
3294 pvt->info.get_tolm = haswell_get_tolm;
3295 pvt->info.get_tohm = haswell_get_tohm;
3296 pvt->info.dram_rule = ibridge_dram_rule;
3297 pvt->info.get_memory_type = haswell_get_memory_type;
3298 pvt->info.get_node_id = haswell_get_node_id;
3299 pvt->info.rir_limit = haswell_rir_limit;
3300 pvt->info.sad_limit = sad_limit;
3301 pvt->info.interleave_mode = interleave_mode;
3302 pvt->info.show_interleave_mode = show_interleave_mode;
3303 pvt->info.dram_attr = dram_attr;
3304 pvt->info.max_sad = ARRAY_SIZE(ibridge_dram_rule);
3305 pvt->info.interleave_list = ibridge_interleave_list;
3306 pvt->info.max_interleave = ARRAY_SIZE(ibridge_interleave_list);
3307 pvt->info.interleave_pkg = ibridge_interleave_pkg;
3308 pvt->info.get_width = broadwell_get_width;
3309 mci->ctl_name = kasprintf(GFP_KERNEL, "Broadwell Socket#%d", mci->mc_idx);
3311 /* Store pci devices at mci for faster access */
3312 rc = broadwell_mci_bind_devs(mci, sbridge_dev);
3313 if (unlikely(rc < 0))
3314 goto fail0;
3315 break;
3316 case KNIGHTS_LANDING:
3317 /* pvt->info.rankcfgr == ??? */
3318 pvt->info.get_tolm = knl_get_tolm;
3319 pvt->info.get_tohm = knl_get_tohm;
3320 pvt->info.dram_rule = knl_dram_rule;
3321 pvt->info.get_memory_type = knl_get_memory_type;
3322 pvt->info.get_node_id = knl_get_node_id;
3323 pvt->info.rir_limit = NULL;
3324 pvt->info.sad_limit = knl_sad_limit;
3325 pvt->info.interleave_mode = knl_interleave_mode;
3326 pvt->info.show_interleave_mode = knl_show_interleave_mode;
3327 pvt->info.dram_attr = dram_attr_knl;
3328 pvt->info.max_sad = ARRAY_SIZE(knl_dram_rule);
3329 pvt->info.interleave_list = knl_interleave_list;
3330 pvt->info.max_interleave = ARRAY_SIZE(knl_interleave_list);
3331 pvt->info.interleave_pkg = ibridge_interleave_pkg;
3332 pvt->info.get_width = knl_get_width;
3333 mci->ctl_name = kasprintf(GFP_KERNEL,
3334 "Knights Landing Socket#%d", mci->mc_idx);
3336 rc = knl_mci_bind_devs(mci, sbridge_dev);
3337 if (unlikely(rc < 0))
3338 goto fail0;
3339 break;
3342 /* Get dimm basic config and the memory layout */
3343 get_dimm_config(mci);
3344 get_memory_layout(mci);
3346 /* record ptr to the generic device */
3347 mci->pdev = &pdev->dev;
3349 /* add this new MC control structure to EDAC's list of MCs */
3350 if (unlikely(edac_mc_add_mc(mci))) {
3351 edac_dbg(0, "MC: failed edac_mc_add_mc()\n");
3352 rc = -EINVAL;
3353 goto fail0;
3356 return 0;
3358 fail0:
3359 kfree(mci->ctl_name);
3360 edac_mc_free(mci);
3361 sbridge_dev->mci = NULL;
3362 return rc;
3365 #define ICPU(model, table) \
3366 { X86_VENDOR_INTEL, 6, model, 0, (unsigned long)&table }
3368 static const struct x86_cpu_id sbridge_cpuids[] = {
3369 ICPU(0x2d, pci_dev_descr_sbridge_table), /* SANDY_BRIDGE */
3370 ICPU(0x3e, pci_dev_descr_ibridge_table), /* IVY_BRIDGE */
3371 ICPU(0x3f, pci_dev_descr_haswell_table), /* HASWELL */
3372 ICPU(0x4f, pci_dev_descr_broadwell_table), /* BROADWELL */
3373 ICPU(0x56, pci_dev_descr_broadwell_table), /* BROADWELL-DE */
3374 ICPU(0x57, pci_dev_descr_knl_table), /* KNIGHTS_LANDING */
3377 MODULE_DEVICE_TABLE(x86cpu, sbridge_cpuids);
3380 * sbridge_probe Get all devices and register memory controllers
3381 * present.
3382 * return:
3383 * 0 for FOUND a device
3384 * < 0 for error code
3387 static int sbridge_probe(const struct x86_cpu_id *id)
3389 int rc = -ENODEV;
3390 u8 mc, num_mc = 0;
3391 struct sbridge_dev *sbridge_dev;
3392 struct pci_id_table *ptable = (struct pci_id_table *)id->driver_data;
3394 /* get the pci devices we want to reserve for our use */
3395 rc = sbridge_get_all_devices(&num_mc, ptable);
3397 if (unlikely(rc < 0)) {
3398 edac_dbg(0, "couldn't get all devices\n");
3399 goto fail0;
3402 mc = 0;
3404 list_for_each_entry(sbridge_dev, &sbridge_edac_list, list) {
3405 edac_dbg(0, "Registering MC#%d (%d of %d)\n",
3406 mc, mc + 1, num_mc);
3408 sbridge_dev->mc = mc++;
3409 rc = sbridge_register_mci(sbridge_dev, ptable->type);
3410 if (unlikely(rc < 0))
3411 goto fail1;
3414 sbridge_printk(KERN_INFO, "%s\n", SBRIDGE_REVISION);
3416 return 0;
3418 fail1:
3419 list_for_each_entry(sbridge_dev, &sbridge_edac_list, list)
3420 sbridge_unregister_mci(sbridge_dev);
3422 sbridge_put_all_devices();
3423 fail0:
3424 return rc;
3428 * sbridge_remove cleanup
3431 static void sbridge_remove(void)
3433 struct sbridge_dev *sbridge_dev;
3435 edac_dbg(0, "\n");
3437 list_for_each_entry(sbridge_dev, &sbridge_edac_list, list)
3438 sbridge_unregister_mci(sbridge_dev);
3440 /* Release PCI resources */
3441 sbridge_put_all_devices();
3445 * sbridge_init Module entry function
3446 * Try to initialize this module for its devices
3448 static int __init sbridge_init(void)
3450 const struct x86_cpu_id *id;
3451 int rc;
3453 edac_dbg(2, "\n");
3455 id = x86_match_cpu(sbridge_cpuids);
3456 if (!id)
3457 return -ENODEV;
3459 /* Ensure that the OPSTATE is set correctly for POLL or NMI */
3460 opstate_init();
3462 rc = sbridge_probe(id);
3464 if (rc >= 0) {
3465 mce_register_decode_chain(&sbridge_mce_dec);
3466 if (get_edac_report_status() == EDAC_REPORTING_DISABLED)
3467 sbridge_printk(KERN_WARNING, "Loading driver, error reporting disabled.\n");
3468 return 0;
3471 sbridge_printk(KERN_ERR, "Failed to register device with error %d.\n",
3472 rc);
3474 return rc;
3478 * sbridge_exit() Module exit function
3479 * Unregister the driver
3481 static void __exit sbridge_exit(void)
3483 edac_dbg(2, "\n");
3484 sbridge_remove();
3485 mce_unregister_decode_chain(&sbridge_mce_dec);
3488 module_init(sbridge_init);
3489 module_exit(sbridge_exit);
3491 module_param(edac_op_state, int, 0444);
3492 MODULE_PARM_DESC(edac_op_state, "EDAC Error Reporting state: 0=Poll,1=NMI");
3494 MODULE_LICENSE("GPL");
3495 MODULE_AUTHOR("Mauro Carvalho Chehab");
3496 MODULE_AUTHOR("Red Hat Inc. (http://www.redhat.com)");
3497 MODULE_DESCRIPTION("MC Driver for Intel Sandy Bridge and Ivy Bridge memory controllers - "
3498 SBRIDGE_REVISION);