1 // SPDX-License-Identifier: GPL-2.0-only
2 /* Intel Sandy Bridge -EN/-EP/-EX Memory Controller kernel module
3 *
4 * This driver supports the memory controllers found on the Intel
5 * processor family Sandy Bridge.
6 *
7 * Copyright (c) 2011 by:
8 * Mauro Carvalho Chehab
9 */
10
11 #include <linux/module.h>
12 #include <linux/init.h>
13 #include <linux/pci.h>
14 #include <linux/pci_ids.h>
15 #include <linux/slab.h>
16 #include <linux/delay.h>
17 #include <linux/edac.h>
18 #include <linux/mmzone.h>
19 #include <linux/smp.h>
20 #include <linux/bitmap.h>
21 #include <linux/math64.h>
22 #include <linux/mod_devicetable.h>
23 #include <asm/cpu_device_id.h>
24 #include <asm/intel-family.h>
25 #include <asm/processor.h>
26 #include <asm/mce.h>
27
28 #include "edac_module.h"
29
30 /* Static vars */
31 static LIST_HEAD(sbridge_edac_list);
32
33 /*
34 * Alter this version for the module when modifications are made
35 */
36 #define SBRIDGE_REVISION " Ver: 1.1.2 "
37 #define EDAC_MOD_STR "sb_edac"
38
39 /*
40 * Debug macros
41 */
42 #define sbridge_printk(level, fmt, arg...) \
43 edac_printk(level, "sbridge", fmt, ##arg)
44
45 #define sbridge_mc_printk(mci, level, fmt, arg...) \
46 edac_mc_chipset_printk(mci, level, "sbridge", fmt, ##arg)
47
48 /*
49 * Get a bit field at register value <v>, from bit <lo> to bit <hi>
50 */
51 #define GET_BITFIELD(v, lo, hi) \
52 (((v) & GENMASK_ULL(hi, lo)) >> (lo))
53
54 /* Devices 12 Function 6, Offsets 0x80 to 0xcc */
55 static const u32 sbridge_dram_rule[] = {
56 0x80, 0x88, 0x90, 0x98, 0xa0,
57 0xa8, 0xb0, 0xb8, 0xc0, 0xc8,
58 };
59
60 static const u32 ibridge_dram_rule[] = {
61 0x60, 0x68, 0x70, 0x78, 0x80,
62 0x88, 0x90, 0x98, 0xa0, 0xa8,
63 0xb0, 0xb8, 0xc0, 0xc8, 0xd0,
64 0xd8, 0xe0, 0xe8, 0xf0, 0xf8,
65 };
66
67 static const u32 knl_dram_rule[] = {
68 0x60, 0x68, 0x70, 0x78, 0x80, /* 0-4 */
69 0x88, 0x90, 0x98, 0xa0, 0xa8, /* 5-9 */
70 0xb0, 0xb8, 0xc0, 0xc8, 0xd0, /* 10-14 */
71 0xd8, 0xe0, 0xe8, 0xf0, 0xf8, /* 15-19 */
72 0x100, 0x108, 0x110, 0x118, /* 20-23 */
73 };
74
75 #define DRAM_RULE_ENABLE(reg) GET_BITFIELD(reg, 0, 0)
76 #define A7MODE(reg) GET_BITFIELD(reg, 26, 26)
77
show_dram_attr(u32 attr)78 static char *show_dram_attr(u32 attr)
79 {
80 switch (attr) {
81 case 0:
82 return "DRAM";
83 case 1:
84 return "MMCFG";
85 case 2:
86 return "NXM";
87 default:
88 return "unknown";
89 }
90 }
91
92 static const u32 sbridge_interleave_list[] = {
93 0x84, 0x8c, 0x94, 0x9c, 0xa4,
94 0xac, 0xb4, 0xbc, 0xc4, 0xcc,
95 };
96
97 static const u32 ibridge_interleave_list[] = {
98 0x64, 0x6c, 0x74, 0x7c, 0x84,
99 0x8c, 0x94, 0x9c, 0xa4, 0xac,
100 0xb4, 0xbc, 0xc4, 0xcc, 0xd4,
101 0xdc, 0xe4, 0xec, 0xf4, 0xfc,
102 };
103
104 static const u32 knl_interleave_list[] = {
105 0x64, 0x6c, 0x74, 0x7c, 0x84, /* 0-4 */
106 0x8c, 0x94, 0x9c, 0xa4, 0xac, /* 5-9 */
107 0xb4, 0xbc, 0xc4, 0xcc, 0xd4, /* 10-14 */
108 0xdc, 0xe4, 0xec, 0xf4, 0xfc, /* 15-19 */
109 0x104, 0x10c, 0x114, 0x11c, /* 20-23 */
110 };
111 #define MAX_INTERLEAVE \
112 (max_t(unsigned int, ARRAY_SIZE(sbridge_interleave_list), \
113 max_t(unsigned int, ARRAY_SIZE(ibridge_interleave_list), \
114 ARRAY_SIZE(knl_interleave_list))))
115
116 struct interleave_pkg {
117 unsigned char start;
118 unsigned char end;
119 };
120
121 static const struct interleave_pkg sbridge_interleave_pkg[] = {
122 { 0, 2 },
123 { 3, 5 },
124 { 8, 10 },
125 { 11, 13 },
126 { 16, 18 },
127 { 19, 21 },
128 { 24, 26 },
129 { 27, 29 },
130 };
131
132 static const struct interleave_pkg ibridge_interleave_pkg[] = {
133 { 0, 3 },
134 { 4, 7 },
135 { 8, 11 },
136 { 12, 15 },
137 { 16, 19 },
138 { 20, 23 },
139 { 24, 27 },
140 { 28, 31 },
141 };
142
sad_pkg(const struct interleave_pkg * table,u32 reg,int interleave)143 static inline int sad_pkg(const struct interleave_pkg *table, u32 reg,
144 int interleave)
145 {
146 return GET_BITFIELD(reg, table[interleave].start,
147 table[interleave].end);
148 }
149
150 /* Devices 12 Function 7 */
151
152 #define TOLM 0x80
153 #define TOHM 0x84
154 #define HASWELL_TOLM 0xd0
155 #define HASWELL_TOHM_0 0xd4
156 #define HASWELL_TOHM_1 0xd8
157 #define KNL_TOLM 0xd0
158 #define KNL_TOHM_0 0xd4
159 #define KNL_TOHM_1 0xd8
160
161 #define GET_TOLM(reg) ((GET_BITFIELD(reg, 0, 3) << 28) | 0x3ffffff)
162 #define GET_TOHM(reg) ((GET_BITFIELD(reg, 0, 20) << 25) | 0x3ffffff)
163
164 /* Device 13 Function 6 */
165
166 #define SAD_TARGET 0xf0
167
168 #define SOURCE_ID(reg) GET_BITFIELD(reg, 9, 11)
169
170 #define SOURCE_ID_KNL(reg) GET_BITFIELD(reg, 12, 14)
171
172 #define SAD_CONTROL 0xf4
173
174 /* Device 14 function 0 */
175
176 static const u32 tad_dram_rule[] = {
177 0x40, 0x44, 0x48, 0x4c,
178 0x50, 0x54, 0x58, 0x5c,
179 0x60, 0x64, 0x68, 0x6c,
180 };
181 #define MAX_TAD ARRAY_SIZE(tad_dram_rule)
182
183 #define TAD_LIMIT(reg) ((GET_BITFIELD(reg, 12, 31) << 26) | 0x3ffffff)
184 #define TAD_SOCK(reg) GET_BITFIELD(reg, 10, 11)
185 #define TAD_CH(reg) GET_BITFIELD(reg, 8, 9)
186 #define TAD_TGT3(reg) GET_BITFIELD(reg, 6, 7)
187 #define TAD_TGT2(reg) GET_BITFIELD(reg, 4, 5)
188 #define TAD_TGT1(reg) GET_BITFIELD(reg, 2, 3)
189 #define TAD_TGT0(reg) GET_BITFIELD(reg, 0, 1)
190
191 /* Device 15, function 0 */
192
193 #define MCMTR 0x7c
194 #define KNL_MCMTR 0x624
195
196 #define IS_ECC_ENABLED(mcmtr) GET_BITFIELD(mcmtr, 2, 2)
197 #define IS_LOCKSTEP_ENABLED(mcmtr) GET_BITFIELD(mcmtr, 1, 1)
198 #define IS_CLOSE_PG(mcmtr) GET_BITFIELD(mcmtr, 0, 0)
199
200 /* Device 15, function 1 */
201
202 #define RASENABLES 0xac
203 #define IS_MIRROR_ENABLED(reg) GET_BITFIELD(reg, 0, 0)
204
205 /* Device 15, functions 2-5 */
206
207 static const int mtr_regs[] = {
208 0x80, 0x84, 0x88,
209 };
210
211 static const int knl_mtr_reg = 0xb60;
212
213 #define RANK_DISABLE(mtr) GET_BITFIELD(mtr, 16, 19)
214 #define IS_DIMM_PRESENT(mtr) GET_BITFIELD(mtr, 14, 14)
215 #define RANK_CNT_BITS(mtr) GET_BITFIELD(mtr, 12, 13)
216 #define RANK_WIDTH_BITS(mtr) GET_BITFIELD(mtr, 2, 4)
217 #define COL_WIDTH_BITS(mtr) GET_BITFIELD(mtr, 0, 1)
218
219 static const u32 tad_ch_nilv_offset[] = {
220 0x90, 0x94, 0x98, 0x9c,
221 0xa0, 0xa4, 0xa8, 0xac,
222 0xb0, 0xb4, 0xb8, 0xbc,
223 };
224 #define CHN_IDX_OFFSET(reg) GET_BITFIELD(reg, 28, 29)
225 #define TAD_OFFSET(reg) (GET_BITFIELD(reg, 6, 25) << 26)
226
227 static const u32 rir_way_limit[] = {
228 0x108, 0x10c, 0x110, 0x114, 0x118,
229 };
230 #define MAX_RIR_RANGES ARRAY_SIZE(rir_way_limit)
231
232 #define IS_RIR_VALID(reg) GET_BITFIELD(reg, 31, 31)
233 #define RIR_WAY(reg) GET_BITFIELD(reg, 28, 29)
234
235 #define MAX_RIR_WAY 8
236
237 static const u32 rir_offset[MAX_RIR_RANGES][MAX_RIR_WAY] = {
238 { 0x120, 0x124, 0x128, 0x12c, 0x130, 0x134, 0x138, 0x13c },
239 { 0x140, 0x144, 0x148, 0x14c, 0x150, 0x154, 0x158, 0x15c },
240 { 0x160, 0x164, 0x168, 0x16c, 0x170, 0x174, 0x178, 0x17c },
241 { 0x180, 0x184, 0x188, 0x18c, 0x190, 0x194, 0x198, 0x19c },
242 { 0x1a0, 0x1a4, 0x1a8, 0x1ac, 0x1b0, 0x1b4, 0x1b8, 0x1bc },
243 };
244
245 #define RIR_RNK_TGT(type, reg) (((type) == BROADWELL) ? \
246 GET_BITFIELD(reg, 20, 23) : GET_BITFIELD(reg, 16, 19))
247
248 #define RIR_OFFSET(type, reg) (((type) == HASWELL || (type) == BROADWELL) ? \
249 GET_BITFIELD(reg, 2, 15) : GET_BITFIELD(reg, 2, 14))
250
251 /* Device 16, functions 2-7 */
252
253 /*
254 * FIXME: Implement the error count reads directly
255 */
256
257 #define RANK_ODD_OV(reg) GET_BITFIELD(reg, 31, 31)
258 #define RANK_ODD_ERR_CNT(reg) GET_BITFIELD(reg, 16, 30)
259 #define RANK_EVEN_OV(reg) GET_BITFIELD(reg, 15, 15)
260 #define RANK_EVEN_ERR_CNT(reg) GET_BITFIELD(reg, 0, 14)
261
262 #if 0 /* Currently unused*/
263 static const u32 correrrcnt[] = {
264 0x104, 0x108, 0x10c, 0x110,
265 };
266
267 static const u32 correrrthrsld[] = {
268 0x11c, 0x120, 0x124, 0x128,
269 };
270 #endif
271
272 #define RANK_ODD_ERR_THRSLD(reg) GET_BITFIELD(reg, 16, 30)
273 #define RANK_EVEN_ERR_THRSLD(reg) GET_BITFIELD(reg, 0, 14)
274
275
276 /* Device 17, function 0 */
277
278 #define SB_RANK_CFG_A 0x0328
279
280 #define IB_RANK_CFG_A 0x0320
281
282 /*
283 * sbridge structs
284 */
285
286 #define NUM_CHANNELS 6 /* Max channels per MC */
287 #define MAX_DIMMS 3 /* Max DIMMS per channel */
288 #define KNL_MAX_CHAS 38 /* KNL max num. of Cache Home Agents */
289 #define KNL_MAX_CHANNELS 6 /* KNL max num. of PCI channels */
290 #define KNL_MAX_EDCS 8 /* Embedded DRAM controllers */
291 #define CHANNEL_UNSPECIFIED 0xf /* Intel IA32 SDM 15-14 */
292
293 enum type {
294 SANDY_BRIDGE,
295 IVY_BRIDGE,
296 HASWELL,
297 BROADWELL,
298 KNIGHTS_LANDING,
299 };
300
301 enum domain {
302 IMC0 = 0,
303 IMC1,
304 SOCK,
305 };
306
307 enum mirroring_mode {
308 NON_MIRRORING,
309 ADDR_RANGE_MIRRORING,
310 FULL_MIRRORING,
311 };
312
313 struct sbridge_pvt;
314 struct sbridge_info {
315 enum type type;
316 u32 mcmtr;
317 u32 rankcfgr;
318 u64 (*get_tolm)(struct sbridge_pvt *pvt);
319 u64 (*get_tohm)(struct sbridge_pvt *pvt);
320 u64 (*rir_limit)(u32 reg);
321 u64 (*sad_limit)(u32 reg);
322 u32 (*interleave_mode)(u32 reg);
323 u32 (*dram_attr)(u32 reg);
324 const u32 *dram_rule;
325 const u32 *interleave_list;
326 const struct interleave_pkg *interleave_pkg;
327 u8 max_sad;
328 u8 (*get_node_id)(struct sbridge_pvt *pvt);
329 u8 (*get_ha)(u8 bank);
330 enum mem_type (*get_memory_type)(struct sbridge_pvt *pvt);
331 enum dev_type (*get_width)(struct sbridge_pvt *pvt, u32 mtr);
332 struct pci_dev *pci_vtd;
333 };
334
335 struct sbridge_channel {
336 u32 ranks;
337 u32 dimms;
338 struct dimm {
339 u32 rowbits;
340 u32 colbits;
341 u32 bank_xor_enable;
342 u32 amap_fine;
343 } dimm[MAX_DIMMS];
344 };
345
346 struct pci_id_descr {
347 int dev_id;
348 int optional;
349 enum domain dom;
350 };
351
352 struct pci_id_table {
353 const struct pci_id_descr *descr;
354 int n_devs_per_imc;
355 int n_devs_per_sock;
356 int n_imcs_per_sock;
357 enum type type;
358 };
359
360 struct sbridge_dev {
361 struct list_head list;
362 int seg;
363 u8 bus, mc;
364 u8 node_id, source_id;
365 struct pci_dev **pdev;
366 enum domain dom;
367 int n_devs;
368 int i_devs;
369 struct mem_ctl_info *mci;
370 };
371
372 struct knl_pvt {
373 struct pci_dev *pci_cha[KNL_MAX_CHAS];
374 struct pci_dev *pci_channel[KNL_MAX_CHANNELS];
375 struct pci_dev *pci_mc0;
376 struct pci_dev *pci_mc1;
377 struct pci_dev *pci_mc0_misc;
378 struct pci_dev *pci_mc1_misc;
379 struct pci_dev *pci_mc_info; /* tolm, tohm */
380 };
381
382 struct sbridge_pvt {
383 /* Devices per socket */
384 struct pci_dev *pci_ddrio;
385 struct pci_dev *pci_sad0, *pci_sad1;
386 struct pci_dev *pci_br0, *pci_br1;
387 /* Devices per memory controller */
388 struct pci_dev *pci_ha, *pci_ta, *pci_ras;
389 struct pci_dev *pci_tad[NUM_CHANNELS];
390
391 struct sbridge_dev *sbridge_dev;
392
393 struct sbridge_info info;
394 struct sbridge_channel channel[NUM_CHANNELS];
395
396 /* Memory type detection */
397 bool is_cur_addr_mirrored, is_lockstep, is_close_pg;
398 bool is_chan_hash;
399 enum mirroring_mode mirror_mode;
400
401 /* Memory description */
402 u64 tolm, tohm;
403 struct knl_pvt knl;
404 };
405
406 #define PCI_DESCR(device_id, opt, domain) \
407 .dev_id = (device_id), \
408 .optional = opt, \
409 .dom = domain
410
411 static const struct pci_id_descr pci_dev_descr_sbridge[] = {
412 /* Processor Home Agent */
413 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_HA0, 0, IMC0) },
414
415 /* Memory controller */
416 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TA, 0, IMC0) },
417 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_RAS, 0, IMC0) },
418 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD0, 0, IMC0) },
419 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD1, 0, IMC0) },
420 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD2, 0, IMC0) },
421 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD3, 0, IMC0) },
422 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_DDRIO, 1, SOCK) },
423
424 /* System Address Decoder */
425 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_SAD0, 0, SOCK) },
426 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_SAD1, 0, SOCK) },
427
428 /* Broadcast Registers */
429 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_BR, 0, SOCK) },
430 };
431
432 #define PCI_ID_TABLE_ENTRY(A, N, M, T) { \
433 .descr = A, \
434 .n_devs_per_imc = N, \
435 .n_devs_per_sock = ARRAY_SIZE(A), \
436 .n_imcs_per_sock = M, \
437 .type = T \
438 }
439
440 static const struct pci_id_table pci_dev_descr_sbridge_table[] = {
441 PCI_ID_TABLE_ENTRY(pci_dev_descr_sbridge, ARRAY_SIZE(pci_dev_descr_sbridge), 1, SANDY_BRIDGE),
442 {0,} /* 0 terminated list. */
443 };
444
445 /* This changes depending if 1HA or 2HA:
446 * 1HA:
447 * 0x0eb8 (17.0) is DDRIO0
448 * 2HA:
449 * 0x0ebc (17.4) is DDRIO0
450 */
451 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_1HA_DDRIO0 0x0eb8
452 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_2HA_DDRIO0 0x0ebc
453
454 /* pci ids */
455 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0 0x0ea0
456 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TA 0x0ea8
457 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_RAS 0x0e71
458 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD0 0x0eaa
459 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD1 0x0eab
460 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD2 0x0eac
461 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD3 0x0ead
462 #define PCI_DEVICE_ID_INTEL_IBRIDGE_SAD 0x0ec8
463 #define PCI_DEVICE_ID_INTEL_IBRIDGE_BR0 0x0ec9
464 #define PCI_DEVICE_ID_INTEL_IBRIDGE_BR1 0x0eca
465 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1 0x0e60
466 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TA 0x0e68
467 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_RAS 0x0e79
468 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD0 0x0e6a
469 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD1 0x0e6b
470 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD2 0x0e6c
471 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD3 0x0e6d
472
473 static const struct pci_id_descr pci_dev_descr_ibridge[] = {
474 /* Processor Home Agent */
475 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0, 0, IMC0) },
476 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1, 1, IMC1) },
477
478 /* Memory controller */
479 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TA, 0, IMC0) },
480 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_RAS, 0, IMC0) },
481 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD0, 0, IMC0) },
482 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD1, 0, IMC0) },
483 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD2, 0, IMC0) },
484 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD3, 0, IMC0) },
485
486 /* Optional, mode 2HA */
487 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TA, 1, IMC1) },
488 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_RAS, 1, IMC1) },
489 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD0, 1, IMC1) },
490 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD1, 1, IMC1) },
491 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD2, 1, IMC1) },
492 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD3, 1, IMC1) },
493
494 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_1HA_DDRIO0, 1, SOCK) },
495 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_2HA_DDRIO0, 1, SOCK) },
496
497 /* System Address Decoder */
498 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_SAD, 0, SOCK) },
499
500 /* Broadcast Registers */
501 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_BR0, 1, SOCK) },
502 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_BR1, 0, SOCK) },
503
504 };
505
506 static const struct pci_id_table pci_dev_descr_ibridge_table[] = {
507 PCI_ID_TABLE_ENTRY(pci_dev_descr_ibridge, 12, 2, IVY_BRIDGE),
508 {0,} /* 0 terminated list. */
509 };
510
511 /* Haswell support */
512 /* EN processor:
513 * - 1 IMC
514 * - 3 DDR3 channels, 2 DPC per channel
515 * EP processor:
516 * - 1 or 2 IMC
517 * - 4 DDR4 channels, 3 DPC per channel
518 * EP 4S processor:
519 * - 2 IMC
520 * - 4 DDR4 channels, 3 DPC per channel
521 * EX processor:
522 * - 2 IMC
523 * - each IMC interfaces with a SMI 2 channel
524 * - each SMI channel interfaces with a scalable memory buffer
525 * - each scalable memory buffer supports 4 DDR3/DDR4 channels, 3 DPC
526 */
527 #define HASWELL_DDRCRCLKCONTROLS 0xa10 /* Ditto on Broadwell */
528 #define HASWELL_HASYSDEFEATURE2 0x84
529 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_VTD_MISC 0x2f28
530 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0 0x2fa0
531 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1 0x2f60
532 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TA 0x2fa8
533 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TM 0x2f71
534 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TA 0x2f68
535 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TM 0x2f79
536 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_CBO_SAD0 0x2ffc
537 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_CBO_SAD1 0x2ffd
538 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD0 0x2faa
539 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD1 0x2fab
540 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD2 0x2fac
541 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD3 0x2fad
542 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD0 0x2f6a
543 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD1 0x2f6b
544 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD2 0x2f6c
545 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD3 0x2f6d
546 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO0 0x2fbd
547 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO1 0x2fbf
548 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO2 0x2fb9
549 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO3 0x2fbb
550 static const struct pci_id_descr pci_dev_descr_haswell[] = {
551 /* first item must be the HA */
552 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0, 0, IMC0) },
553 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1, 1, IMC1) },
554
555 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TA, 0, IMC0) },
556 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TM, 0, IMC0) },
557 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD0, 0, IMC0) },
558 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD1, 0, IMC0) },
559 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD2, 1, IMC0) },
560 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD3, 1, IMC0) },
561
562 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TA, 1, IMC1) },
563 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TM, 1, IMC1) },
564 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD0, 1, IMC1) },
565 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD1, 1, IMC1) },
566 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD2, 1, IMC1) },
567 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD3, 1, IMC1) },
568
569 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_CBO_SAD0, 0, SOCK) },
570 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_CBO_SAD1, 0, SOCK) },
571 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO0, 1, SOCK) },
572 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO1, 1, SOCK) },
573 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO2, 1, SOCK) },
574 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO3, 1, SOCK) },
575 };
576
577 static const struct pci_id_table pci_dev_descr_haswell_table[] = {
578 PCI_ID_TABLE_ENTRY(pci_dev_descr_haswell, 13, 2, HASWELL),
579 {0,} /* 0 terminated list. */
580 };
581
582 /* Knight's Landing Support */
583 /*
584 * KNL's memory channels are swizzled between memory controllers.
585 * MC0 is mapped to CH3,4,5 and MC1 is mapped to CH0,1,2
586 */
587 #define knl_channel_remap(mc, chan) ((mc) ? (chan) : (chan) + 3)
588
589 /* Memory controller, TAD tables, error injection - 2-8-0, 2-9-0 (2 of these) */
590 #define PCI_DEVICE_ID_INTEL_KNL_IMC_MC 0x7840
591 /* DRAM channel stuff; bank addrs, dimmmtr, etc.. 2-8-2 - 2-9-4 (6 of these) */
592 #define PCI_DEVICE_ID_INTEL_KNL_IMC_CHAN 0x7843
593 /* kdrwdbu TAD limits/offsets, MCMTR - 2-10-1, 2-11-1 (2 of these) */
594 #define PCI_DEVICE_ID_INTEL_KNL_IMC_TA 0x7844
595 /* CHA broadcast registers, dram rules - 1-29-0 (1 of these) */
596 #define PCI_DEVICE_ID_INTEL_KNL_IMC_SAD0 0x782a
597 /* SAD target - 1-29-1 (1 of these) */
598 #define PCI_DEVICE_ID_INTEL_KNL_IMC_SAD1 0x782b
599 /* Caching / Home Agent */
600 #define PCI_DEVICE_ID_INTEL_KNL_IMC_CHA 0x782c
601 /* Device with TOLM and TOHM, 0-5-0 (1 of these) */
602 #define PCI_DEVICE_ID_INTEL_KNL_IMC_TOLHM 0x7810
603
604 /*
605 * KNL differs from SB, IB, and Haswell in that it has multiple
606 * instances of the same device with the same device ID, so we handle that
607 * by creating as many copies in the table as we expect to find.
608 * (Like device ID must be grouped together.)
609 */
610
611 static const struct pci_id_descr pci_dev_descr_knl[] = {
612 [0 ... 1] = { PCI_DESCR(PCI_DEVICE_ID_INTEL_KNL_IMC_MC, 0, IMC0)},
613 [2 ... 7] = { PCI_DESCR(PCI_DEVICE_ID_INTEL_KNL_IMC_CHAN, 0, IMC0) },
614 [8] = { PCI_DESCR(PCI_DEVICE_ID_INTEL_KNL_IMC_TA, 0, IMC0) },
615 [9] = { PCI_DESCR(PCI_DEVICE_ID_INTEL_KNL_IMC_TOLHM, 0, IMC0) },
616 [10] = { PCI_DESCR(PCI_DEVICE_ID_INTEL_KNL_IMC_SAD0, 0, SOCK) },
617 [11] = { PCI_DESCR(PCI_DEVICE_ID_INTEL_KNL_IMC_SAD1, 0, SOCK) },
618 [12 ... 49] = { PCI_DESCR(PCI_DEVICE_ID_INTEL_KNL_IMC_CHA, 0, SOCK) },
619 };
620
621 static const struct pci_id_table pci_dev_descr_knl_table[] = {
622 PCI_ID_TABLE_ENTRY(pci_dev_descr_knl, ARRAY_SIZE(pci_dev_descr_knl), 1, KNIGHTS_LANDING),
623 {0,}
624 };
625
626 /*
627 * Broadwell support
628 *
629 * DE processor:
630 * - 1 IMC
631 * - 2 DDR3 channels, 2 DPC per channel
632 * EP processor:
633 * - 1 or 2 IMC
634 * - 4 DDR4 channels, 3 DPC per channel
635 * EP 4S processor:
636 * - 2 IMC
637 * - 4 DDR4 channels, 3 DPC per channel
638 * EX processor:
639 * - 2 IMC
640 * - each IMC interfaces with a SMI 2 channel
641 * - each SMI channel interfaces with a scalable memory buffer
642 * - each scalable memory buffer supports 4 DDR3/DDR4 channels, 3 DPC
643 */
644 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_VTD_MISC 0x6f28
645 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0 0x6fa0
646 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1 0x6f60
647 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TA 0x6fa8
648 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TM 0x6f71
649 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TA 0x6f68
650 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TM 0x6f79
651 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_CBO_SAD0 0x6ffc
652 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_CBO_SAD1 0x6ffd
653 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD0 0x6faa
654 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD1 0x6fab
655 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD2 0x6fac
656 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD3 0x6fad
657 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD0 0x6f6a
658 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD1 0x6f6b
659 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD2 0x6f6c
660 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD3 0x6f6d
661 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_DDRIO0 0x6faf
662
663 static const struct pci_id_descr pci_dev_descr_broadwell[] = {
664 /* first item must be the HA */
665 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0, 0, IMC0) },
666 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1, 1, IMC1) },
667
668 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TA, 0, IMC0) },
669 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TM, 0, IMC0) },
670 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD0, 0, IMC0) },
671 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD1, 0, IMC0) },
672 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD2, 1, IMC0) },
673 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD3, 1, IMC0) },
674
675 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TA, 1, IMC1) },
676 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TM, 1, IMC1) },
677 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD0, 1, IMC1) },
678 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD1, 1, IMC1) },
679 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD2, 1, IMC1) },
680 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD3, 1, IMC1) },
681
682 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_CBO_SAD0, 0, SOCK) },
683 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_CBO_SAD1, 0, SOCK) },
684 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_DDRIO0, 1, SOCK) },
685 };
686
687 static const struct pci_id_table pci_dev_descr_broadwell_table[] = {
688 PCI_ID_TABLE_ENTRY(pci_dev_descr_broadwell, 10, 2, BROADWELL),
689 {0,} /* 0 terminated list. */
690 };
691
692
693 /****************************************************************************
694 Ancillary status routines
695 ****************************************************************************/
696
numrank(enum type type,u32 mtr)697 static inline int numrank(enum type type, u32 mtr)
698 {
699 int ranks = (1 << RANK_CNT_BITS(mtr));
700 int max = 4;
701
702 if (type == HASWELL || type == BROADWELL || type == KNIGHTS_LANDING)
703 max = 8;
704
705 if (ranks > max) {
706 edac_dbg(0, "Invalid number of ranks: %d (max = %i) raw value = %x (%04x)\n",
707 ranks, max, (unsigned int)RANK_CNT_BITS(mtr), mtr);
708 return -EINVAL;
709 }
710
711 return ranks;
712 }
713
numrow(u32 mtr)714 static inline int numrow(u32 mtr)
715 {
716 int rows = (RANK_WIDTH_BITS(mtr) + 12);
717
718 if (rows < 13 || rows > 18) {
719 edac_dbg(0, "Invalid number of rows: %d (should be between 14 and 17) raw value = %x (%04x)\n",
720 rows, (unsigned int)RANK_WIDTH_BITS(mtr), mtr);
721 return -EINVAL;
722 }
723
724 return 1 << rows;
725 }
726
numcol(u32 mtr)727 static inline int numcol(u32 mtr)
728 {
729 int cols = (COL_WIDTH_BITS(mtr) + 10);
730
731 if (cols > 12) {
732 edac_dbg(0, "Invalid number of cols: %d (max = 4) raw value = %x (%04x)\n",
733 cols, (unsigned int)COL_WIDTH_BITS(mtr), mtr);
734 return -EINVAL;
735 }
736
737 return 1 << cols;
738 }
739
get_sbridge_dev(int seg,u8 bus,enum domain dom,int multi_bus,struct sbridge_dev * prev)740 static struct sbridge_dev *get_sbridge_dev(int seg, u8 bus, enum domain dom,
741 int multi_bus,
742 struct sbridge_dev *prev)
743 {
744 struct sbridge_dev *sbridge_dev;
745
746 /*
747 * If we have devices scattered across several busses that pertain
748 * to the same memory controller, we'll lump them all together.
749 */
750 if (multi_bus) {
751 return list_first_entry_or_null(&sbridge_edac_list,
752 struct sbridge_dev, list);
753 }
754
755 sbridge_dev = list_entry(prev ? prev->list.next
756 : sbridge_edac_list.next, struct sbridge_dev, list);
757
758 list_for_each_entry_from(sbridge_dev, &sbridge_edac_list, list) {
759 if ((sbridge_dev->seg == seg) && (sbridge_dev->bus == bus) &&
760 (dom == SOCK || dom == sbridge_dev->dom))
761 return sbridge_dev;
762 }
763
764 return NULL;
765 }
766
alloc_sbridge_dev(int seg,u8 bus,enum domain dom,const struct pci_id_table * table)767 static struct sbridge_dev *alloc_sbridge_dev(int seg, u8 bus, enum domain dom,
768 const struct pci_id_table *table)
769 {
770 struct sbridge_dev *sbridge_dev;
771
772 sbridge_dev = kzalloc(sizeof(*sbridge_dev), GFP_KERNEL);
773 if (!sbridge_dev)
774 return NULL;
775
776 sbridge_dev->pdev = kcalloc(table->n_devs_per_imc,
777 sizeof(*sbridge_dev->pdev),
778 GFP_KERNEL);
779 if (!sbridge_dev->pdev) {
780 kfree(sbridge_dev);
781 return NULL;
782 }
783
784 sbridge_dev->seg = seg;
785 sbridge_dev->bus = bus;
786 sbridge_dev->dom = dom;
787 sbridge_dev->n_devs = table->n_devs_per_imc;
788 list_add_tail(&sbridge_dev->list, &sbridge_edac_list);
789
790 return sbridge_dev;
791 }
792
free_sbridge_dev(struct sbridge_dev * sbridge_dev)793 static void free_sbridge_dev(struct sbridge_dev *sbridge_dev)
794 {
795 list_del(&sbridge_dev->list);
796 kfree(sbridge_dev->pdev);
797 kfree(sbridge_dev);
798 }
799
sbridge_get_tolm(struct sbridge_pvt * pvt)800 static u64 sbridge_get_tolm(struct sbridge_pvt *pvt)
801 {
802 u32 reg;
803
804 /* Address range is 32:28 */
805 pci_read_config_dword(pvt->pci_sad1, TOLM, ®);
806 return GET_TOLM(reg);
807 }
808
sbridge_get_tohm(struct sbridge_pvt * pvt)809 static u64 sbridge_get_tohm(struct sbridge_pvt *pvt)
810 {
811 u32 reg;
812
813 pci_read_config_dword(pvt->pci_sad1, TOHM, ®);
814 return GET_TOHM(reg);
815 }
816
ibridge_get_tolm(struct sbridge_pvt * pvt)817 static u64 ibridge_get_tolm(struct sbridge_pvt *pvt)
818 {
819 u32 reg;
820
821 pci_read_config_dword(pvt->pci_br1, TOLM, ®);
822
823 return GET_TOLM(reg);
824 }
825
ibridge_get_tohm(struct sbridge_pvt * pvt)826 static u64 ibridge_get_tohm(struct sbridge_pvt *pvt)
827 {
828 u32 reg;
829
830 pci_read_config_dword(pvt->pci_br1, TOHM, ®);
831
832 return GET_TOHM(reg);
833 }
834
rir_limit(u32 reg)835 static u64 rir_limit(u32 reg)
836 {
837 return ((u64)GET_BITFIELD(reg, 1, 10) << 29) | 0x1fffffff;
838 }
839
sad_limit(u32 reg)840 static u64 sad_limit(u32 reg)
841 {
842 return (GET_BITFIELD(reg, 6, 25) << 26) | 0x3ffffff;
843 }
844
interleave_mode(u32 reg)845 static u32 interleave_mode(u32 reg)
846 {
847 return GET_BITFIELD(reg, 1, 1);
848 }
849
dram_attr(u32 reg)850 static u32 dram_attr(u32 reg)
851 {
852 return GET_BITFIELD(reg, 2, 3);
853 }
854
knl_sad_limit(u32 reg)855 static u64 knl_sad_limit(u32 reg)
856 {
857 return (GET_BITFIELD(reg, 7, 26) << 26) | 0x3ffffff;
858 }
859
knl_interleave_mode(u32 reg)860 static u32 knl_interleave_mode(u32 reg)
861 {
862 return GET_BITFIELD(reg, 1, 2);
863 }
864
865 static const char * const knl_intlv_mode[] = {
866 "[8:6]", "[10:8]", "[14:12]", "[32:30]"
867 };
868
get_intlv_mode_str(u32 reg,enum type t)869 static const char *get_intlv_mode_str(u32 reg, enum type t)
870 {
871 if (t == KNIGHTS_LANDING)
872 return knl_intlv_mode[knl_interleave_mode(reg)];
873 else
874 return interleave_mode(reg) ? "[8:6]" : "[8:6]XOR[18:16]";
875 }
876
dram_attr_knl(u32 reg)877 static u32 dram_attr_knl(u32 reg)
878 {
879 return GET_BITFIELD(reg, 3, 4);
880 }
881
882
get_memory_type(struct sbridge_pvt * pvt)883 static enum mem_type get_memory_type(struct sbridge_pvt *pvt)
884 {
885 u32 reg;
886 enum mem_type mtype;
887
888 if (pvt->pci_ddrio) {
889 pci_read_config_dword(pvt->pci_ddrio, pvt->info.rankcfgr,
890 ®);
891 if (GET_BITFIELD(reg, 11, 11))
892 /* FIXME: Can also be LRDIMM */
893 mtype = MEM_RDDR3;
894 else
895 mtype = MEM_DDR3;
896 } else
897 mtype = MEM_UNKNOWN;
898
899 return mtype;
900 }
901
haswell_get_memory_type(struct sbridge_pvt * pvt)902 static enum mem_type haswell_get_memory_type(struct sbridge_pvt *pvt)
903 {
904 u32 reg;
905 bool registered = false;
906 enum mem_type mtype = MEM_UNKNOWN;
907
908 if (!pvt->pci_ddrio)
909 goto out;
910
911 pci_read_config_dword(pvt->pci_ddrio,
912 HASWELL_DDRCRCLKCONTROLS, ®);
913 /* Is_Rdimm */
914 if (GET_BITFIELD(reg, 16, 16))
915 registered = true;
916
917 pci_read_config_dword(pvt->pci_ta, MCMTR, ®);
918 if (GET_BITFIELD(reg, 14, 14)) {
919 if (registered)
920 mtype = MEM_RDDR4;
921 else
922 mtype = MEM_DDR4;
923 } else {
924 if (registered)
925 mtype = MEM_RDDR3;
926 else
927 mtype = MEM_DDR3;
928 }
929
930 out:
931 return mtype;
932 }
933
knl_get_width(struct sbridge_pvt * pvt,u32 mtr)934 static enum dev_type knl_get_width(struct sbridge_pvt *pvt, u32 mtr)
935 {
936 /* for KNL value is fixed */
937 return DEV_X16;
938 }
939
sbridge_get_width(struct sbridge_pvt * pvt,u32 mtr)940 static enum dev_type sbridge_get_width(struct sbridge_pvt *pvt, u32 mtr)
941 {
942 /* there's no way to figure out */
943 return DEV_UNKNOWN;
944 }
945
__ibridge_get_width(u32 mtr)946 static enum dev_type __ibridge_get_width(u32 mtr)
947 {
948 enum dev_type type = DEV_UNKNOWN;
949
950 switch (mtr) {
951 case 2:
952 type = DEV_X16;
953 break;
954 case 1:
955 type = DEV_X8;
956 break;
957 case 0:
958 type = DEV_X4;
959 break;
960 }
961
962 return type;
963 }
964
ibridge_get_width(struct sbridge_pvt * pvt,u32 mtr)965 static enum dev_type ibridge_get_width(struct sbridge_pvt *pvt, u32 mtr)
966 {
967 /*
968 * ddr3_width on the documentation but also valid for DDR4 on
969 * Haswell
970 */
971 return __ibridge_get_width(GET_BITFIELD(mtr, 7, 8));
972 }
973
broadwell_get_width(struct sbridge_pvt * pvt,u32 mtr)974 static enum dev_type broadwell_get_width(struct sbridge_pvt *pvt, u32 mtr)
975 {
976 /* ddr3_width on the documentation but also valid for DDR4 */
977 return __ibridge_get_width(GET_BITFIELD(mtr, 8, 9));
978 }
979
knl_get_memory_type(struct sbridge_pvt * pvt)980 static enum mem_type knl_get_memory_type(struct sbridge_pvt *pvt)
981 {
982 /* DDR4 RDIMMS and LRDIMMS are supported */
983 return MEM_RDDR4;
984 }
985
get_node_id(struct sbridge_pvt * pvt)986 static u8 get_node_id(struct sbridge_pvt *pvt)
987 {
988 u32 reg;
989 pci_read_config_dword(pvt->pci_br0, SAD_CONTROL, ®);
990 return GET_BITFIELD(reg, 0, 2);
991 }
992
haswell_get_node_id(struct sbridge_pvt * pvt)993 static u8 haswell_get_node_id(struct sbridge_pvt *pvt)
994 {
995 u32 reg;
996
997 pci_read_config_dword(pvt->pci_sad1, SAD_CONTROL, ®);
998 return GET_BITFIELD(reg, 0, 3);
999 }
1000
knl_get_node_id(struct sbridge_pvt * pvt)1001 static u8 knl_get_node_id(struct sbridge_pvt *pvt)
1002 {
1003 u32 reg;
1004
1005 pci_read_config_dword(pvt->pci_sad1, SAD_CONTROL, ®);
1006 return GET_BITFIELD(reg, 0, 2);
1007 }
1008
1009 /*
1010 * Use the reporting bank number to determine which memory
1011 * controller (also known as "ha" for "home agent"). Sandy
1012 * Bridge only has one memory controller per socket, so the
1013 * answer is always zero.
1014 */
sbridge_get_ha(u8 bank)1015 static u8 sbridge_get_ha(u8 bank)
1016 {
1017 return 0;
1018 }
1019
1020 /*
1021 * On Ivy Bridge, Haswell and Broadwell the error may be in a
1022 * home agent bank (7, 8), or one of the per-channel memory
1023 * controller banks (9 .. 16).
1024 */
ibridge_get_ha(u8 bank)1025 static u8 ibridge_get_ha(u8 bank)
1026 {
1027 switch (bank) {
1028 case 7 ... 8:
1029 return bank - 7;
1030 case 9 ... 16:
1031 return (bank - 9) / 4;
1032 default:
1033 return 0xff;
1034 }
1035 }
1036
1037 /* Not used, but included for safety/symmetry */
knl_get_ha(u8 bank)1038 static u8 knl_get_ha(u8 bank)
1039 {
1040 return 0xff;
1041 }
1042
haswell_get_tolm(struct sbridge_pvt * pvt)1043 static u64 haswell_get_tolm(struct sbridge_pvt *pvt)
1044 {
1045 u32 reg;
1046
1047 pci_read_config_dword(pvt->info.pci_vtd, HASWELL_TOLM, ®);
1048 return (GET_BITFIELD(reg, 26, 31) << 26) | 0x3ffffff;
1049 }
1050
haswell_get_tohm(struct sbridge_pvt * pvt)1051 static u64 haswell_get_tohm(struct sbridge_pvt *pvt)
1052 {
1053 u64 rc;
1054 u32 reg;
1055
1056 pci_read_config_dword(pvt->info.pci_vtd, HASWELL_TOHM_0, ®);
1057 rc = GET_BITFIELD(reg, 26, 31);
1058 pci_read_config_dword(pvt->info.pci_vtd, HASWELL_TOHM_1, ®);
1059 rc = ((reg << 6) | rc) << 26;
1060
1061 return rc | 0x3ffffff;
1062 }
1063
knl_get_tolm(struct sbridge_pvt * pvt)1064 static u64 knl_get_tolm(struct sbridge_pvt *pvt)
1065 {
1066 u32 reg;
1067
1068 pci_read_config_dword(pvt->knl.pci_mc_info, KNL_TOLM, ®);
1069 return (GET_BITFIELD(reg, 26, 31) << 26) | 0x3ffffff;
1070 }
1071
knl_get_tohm(struct sbridge_pvt * pvt)1072 static u64 knl_get_tohm(struct sbridge_pvt *pvt)
1073 {
1074 u64 rc;
1075 u32 reg_lo, reg_hi;
1076
1077 pci_read_config_dword(pvt->knl.pci_mc_info, KNL_TOHM_0, ®_lo);
1078 pci_read_config_dword(pvt->knl.pci_mc_info, KNL_TOHM_1, ®_hi);
1079 rc = ((u64)reg_hi << 32) | reg_lo;
1080 return rc | 0x3ffffff;
1081 }
1082
1083
haswell_rir_limit(u32 reg)1084 static u64 haswell_rir_limit(u32 reg)
1085 {
1086 return (((u64)GET_BITFIELD(reg, 1, 11) + 1) << 29) - 1;
1087 }
1088
sad_pkg_socket(u8 pkg)1089 static inline u8 sad_pkg_socket(u8 pkg)
1090 {
1091 /* on Ivy Bridge, nodeID is SASS, where A is HA and S is node id */
1092 return ((pkg >> 3) << 2) | (pkg & 0x3);
1093 }
1094
sad_pkg_ha(u8 pkg)1095 static inline u8 sad_pkg_ha(u8 pkg)
1096 {
1097 return (pkg >> 2) & 0x1;
1098 }
1099
haswell_chan_hash(int idx,u64 addr)1100 static int haswell_chan_hash(int idx, u64 addr)
1101 {
1102 int i;
1103
1104 /*
1105 * XOR even bits from 12:26 to bit0 of idx,
1106 * odd bits from 13:27 to bit1
1107 */
1108 for (i = 12; i < 28; i += 2)
1109 idx ^= (addr >> i) & 3;
1110
1111 return idx;
1112 }
1113
1114 /* Low bits of TAD limit, and some metadata. */
1115 static const u32 knl_tad_dram_limit_lo[] = {
1116 0x400, 0x500, 0x600, 0x700,
1117 0x800, 0x900, 0xa00, 0xb00,
1118 };
1119
1120 /* Low bits of TAD offset. */
1121 static const u32 knl_tad_dram_offset_lo[] = {
1122 0x404, 0x504, 0x604, 0x704,
1123 0x804, 0x904, 0xa04, 0xb04,
1124 };
1125
1126 /* High 16 bits of TAD limit and offset. */
1127 static const u32 knl_tad_dram_hi[] = {
1128 0x408, 0x508, 0x608, 0x708,
1129 0x808, 0x908, 0xa08, 0xb08,
1130 };
1131
1132 /* Number of ways a tad entry is interleaved. */
1133 static const u32 knl_tad_ways[] = {
1134 8, 6, 4, 3, 2, 1,
1135 };
1136
1137 /*
1138 * Retrieve the n'th Target Address Decode table entry
1139 * from the memory controller's TAD table.
1140 *
1141 * @pvt: driver private data
1142 * @entry: which entry you want to retrieve
1143 * @mc: which memory controller (0 or 1)
1144 * @offset: output tad range offset
1145 * @limit: output address of first byte above tad range
1146 * @ways: output number of interleave ways
1147 *
1148 * The offset value has curious semantics. It's a sort of running total
1149 * of the sizes of all the memory regions that aren't mapped in this
1150 * tad table.
1151 */
knl_get_tad(const struct sbridge_pvt * pvt,const int entry,const int mc,u64 * offset,u64 * limit,int * ways)1152 static int knl_get_tad(const struct sbridge_pvt *pvt,
1153 const int entry,
1154 const int mc,
1155 u64 *offset,
1156 u64 *limit,
1157 int *ways)
1158 {
1159 u32 reg_limit_lo, reg_offset_lo, reg_hi;
1160 struct pci_dev *pci_mc;
1161 int way_id;
1162
1163 switch (mc) {
1164 case 0:
1165 pci_mc = pvt->knl.pci_mc0;
1166 break;
1167 case 1:
1168 pci_mc = pvt->knl.pci_mc1;
1169 break;
1170 default:
1171 WARN_ON(1);
1172 return -EINVAL;
1173 }
1174
1175 pci_read_config_dword(pci_mc,
1176 knl_tad_dram_limit_lo[entry], ®_limit_lo);
1177 pci_read_config_dword(pci_mc,
1178 knl_tad_dram_offset_lo[entry], ®_offset_lo);
1179 pci_read_config_dword(pci_mc,
1180 knl_tad_dram_hi[entry], ®_hi);
1181
1182 /* Is this TAD entry enabled? */
1183 if (!GET_BITFIELD(reg_limit_lo, 0, 0))
1184 return -ENODEV;
1185
1186 way_id = GET_BITFIELD(reg_limit_lo, 3, 5);
1187
1188 if (way_id < ARRAY_SIZE(knl_tad_ways)) {
1189 *ways = knl_tad_ways[way_id];
1190 } else {
1191 *ways = 0;
1192 sbridge_printk(KERN_ERR,
1193 "Unexpected value %d in mc_tad_limit_lo wayness field\n",
1194 way_id);
1195 return -ENODEV;
1196 }
1197
1198 /*
1199 * The least significant 6 bits of base and limit are truncated.
1200 * For limit, we fill the missing bits with 1s.
1201 */
1202 *offset = ((u64) GET_BITFIELD(reg_offset_lo, 6, 31) << 6) |
1203 ((u64) GET_BITFIELD(reg_hi, 0, 15) << 32);
1204 *limit = ((u64) GET_BITFIELD(reg_limit_lo, 6, 31) << 6) | 63 |
1205 ((u64) GET_BITFIELD(reg_hi, 16, 31) << 32);
1206
1207 return 0;
1208 }
1209
1210 /* Determine which memory controller is responsible for a given channel. */
knl_channel_mc(int channel)1211 static int knl_channel_mc(int channel)
1212 {
1213 WARN_ON(channel < 0 || channel >= 6);
1214
1215 return channel < 3 ? 1 : 0;
1216 }
1217
1218 /*
1219 * Get the Nth entry from EDC_ROUTE_TABLE register.
1220 * (This is the per-tile mapping of logical interleave targets to
1221 * physical EDC modules.)
1222 *
1223 * entry 0: 0:2
1224 * 1: 3:5
1225 * 2: 6:8
1226 * 3: 9:11
1227 * 4: 12:14
1228 * 5: 15:17
1229 * 6: 18:20
1230 * 7: 21:23
1231 * reserved: 24:31
1232 */
knl_get_edc_route(int entry,u32 reg)1233 static u32 knl_get_edc_route(int entry, u32 reg)
1234 {
1235 WARN_ON(entry >= KNL_MAX_EDCS);
1236 return GET_BITFIELD(reg, entry*3, (entry*3)+2);
1237 }
1238
1239 /*
1240 * Get the Nth entry from MC_ROUTE_TABLE register.
1241 * (This is the per-tile mapping of logical interleave targets to
1242 * physical DRAM channels modules.)
1243 *
1244 * entry 0: mc 0:2 channel 18:19
1245 * 1: mc 3:5 channel 20:21
1246 * 2: mc 6:8 channel 22:23
1247 * 3: mc 9:11 channel 24:25
1248 * 4: mc 12:14 channel 26:27
1249 * 5: mc 15:17 channel 28:29
1250 * reserved: 30:31
1251 *
1252 * Though we have 3 bits to identify the MC, we should only see
1253 * the values 0 or 1.
1254 */
1255
knl_get_mc_route(int entry,u32 reg)1256 static u32 knl_get_mc_route(int entry, u32 reg)
1257 {
1258 int mc, chan;
1259
1260 WARN_ON(entry >= KNL_MAX_CHANNELS);
1261
1262 mc = GET_BITFIELD(reg, entry*3, (entry*3)+2);
1263 chan = GET_BITFIELD(reg, (entry*2) + 18, (entry*2) + 18 + 1);
1264
1265 return knl_channel_remap(mc, chan);
1266 }
1267
1268 /*
1269 * Render the EDC_ROUTE register in human-readable form.
1270 * Output string s should be at least KNL_MAX_EDCS*2 bytes.
1271 */
knl_show_edc_route(u32 reg,char * s)1272 static void knl_show_edc_route(u32 reg, char *s)
1273 {
1274 int i;
1275
1276 for (i = 0; i < KNL_MAX_EDCS; i++) {
1277 s[i*2] = knl_get_edc_route(i, reg) + '0';
1278 s[i*2+1] = '-';
1279 }
1280
1281 s[KNL_MAX_EDCS*2 - 1] = '\0';
1282 }
1283
1284 /*
1285 * Render the MC_ROUTE register in human-readable form.
1286 * Output string s should be at least KNL_MAX_CHANNELS*2 bytes.
1287 */
knl_show_mc_route(u32 reg,char * s)1288 static void knl_show_mc_route(u32 reg, char *s)
1289 {
1290 int i;
1291
1292 for (i = 0; i < KNL_MAX_CHANNELS; i++) {
1293 s[i*2] = knl_get_mc_route(i, reg) + '0';
1294 s[i*2+1] = '-';
1295 }
1296
1297 s[KNL_MAX_CHANNELS*2 - 1] = '\0';
1298 }
1299
1300 #define KNL_EDC_ROUTE 0xb8
1301 #define KNL_MC_ROUTE 0xb4
1302
1303 /* Is this dram rule backed by regular DRAM in flat mode? */
1304 #define KNL_EDRAM(reg) GET_BITFIELD(reg, 29, 29)
1305
1306 /* Is this dram rule cached? */
1307 #define KNL_CACHEABLE(reg) GET_BITFIELD(reg, 28, 28)
1308
1309 /* Is this rule backed by edc ? */
1310 #define KNL_EDRAM_ONLY(reg) GET_BITFIELD(reg, 29, 29)
1311
1312 /* Is this rule backed by DRAM, cacheable in EDRAM? */
1313 #define KNL_CACHEABLE(reg) GET_BITFIELD(reg, 28, 28)
1314
1315 /* Is this rule mod3? */
1316 #define KNL_MOD3(reg) GET_BITFIELD(reg, 27, 27)
1317
1318 /*
1319 * Figure out how big our RAM modules are.
1320 *
1321 * The DIMMMTR register in KNL doesn't tell us the size of the DIMMs, so we
1322 * have to figure this out from the SAD rules, interleave lists, route tables,
1323 * and TAD rules.
1324 *
1325 * SAD rules can have holes in them (e.g. the 3G-4G hole), so we have to
1326 * inspect the TAD rules to figure out how large the SAD regions really are.
1327 *
1328 * When we know the real size of a SAD region and how many ways it's
1329 * interleaved, we know the individual contribution of each channel to
1330 * TAD is size/ways.
1331 *
1332 * Finally, we have to check whether each channel participates in each SAD
1333 * region.
1334 *
1335 * Fortunately, KNL only supports one DIMM per channel, so once we know how
1336 * much memory the channel uses, we know the DIMM is at least that large.
1337 * (The BIOS might possibly choose not to map all available memory, in which
1338 * case we will underreport the size of the DIMM.)
1339 *
1340 * In theory, we could try to determine the EDC sizes as well, but that would
1341 * only work in flat mode, not in cache mode.
1342 *
1343 * @mc_sizes: Output sizes of channels (must have space for KNL_MAX_CHANNELS
1344 * elements)
1345 */
knl_get_dimm_capacity(struct sbridge_pvt * pvt,u64 * mc_sizes)1346 static int knl_get_dimm_capacity(struct sbridge_pvt *pvt, u64 *mc_sizes)
1347 {
1348 u64 sad_base, sad_limit = 0;
1349 u64 tad_base, tad_size, tad_limit, tad_deadspace, tad_livespace;
1350 int sad_rule = 0;
1351 int tad_rule = 0;
1352 int intrlv_ways, tad_ways;
1353 u32 first_pkg, pkg;
1354 int i;
1355 u64 sad_actual_size[2]; /* sad size accounting for holes, per mc */
1356 u32 dram_rule, interleave_reg;
1357 u32 mc_route_reg[KNL_MAX_CHAS];
1358 u32 edc_route_reg[KNL_MAX_CHAS];
1359 int edram_only;
1360 char edc_route_string[KNL_MAX_EDCS*2];
1361 char mc_route_string[KNL_MAX_CHANNELS*2];
1362 int cur_reg_start;
1363 int mc;
1364 int channel;
1365 int participants[KNL_MAX_CHANNELS];
1366
1367 for (i = 0; i < KNL_MAX_CHANNELS; i++)
1368 mc_sizes[i] = 0;
1369
1370 /* Read the EDC route table in each CHA. */
1371 cur_reg_start = 0;
1372 for (i = 0; i < KNL_MAX_CHAS; i++) {
1373 pci_read_config_dword(pvt->knl.pci_cha[i],
1374 KNL_EDC_ROUTE, &edc_route_reg[i]);
1375
1376 if (i > 0 && edc_route_reg[i] != edc_route_reg[i-1]) {
1377 knl_show_edc_route(edc_route_reg[i-1],
1378 edc_route_string);
1379 if (cur_reg_start == i-1)
1380 edac_dbg(0, "edc route table for CHA %d: %s\n",
1381 cur_reg_start, edc_route_string);
1382 else
1383 edac_dbg(0, "edc route table for CHA %d-%d: %s\n",
1384 cur_reg_start, i-1, edc_route_string);
1385 cur_reg_start = i;
1386 }
1387 }
1388 knl_show_edc_route(edc_route_reg[i-1], edc_route_string);
1389 if (cur_reg_start == i-1)
1390 edac_dbg(0, "edc route table for CHA %d: %s\n",
1391 cur_reg_start, edc_route_string);
1392 else
1393 edac_dbg(0, "edc route table for CHA %d-%d: %s\n",
1394 cur_reg_start, i-1, edc_route_string);
1395
1396 /* Read the MC 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_MC_ROUTE, &mc_route_reg[i]);
1401
1402 if (i > 0 && mc_route_reg[i] != mc_route_reg[i-1]) {
1403 knl_show_mc_route(mc_route_reg[i-1], mc_route_string);
1404 if (cur_reg_start == i-1)
1405 edac_dbg(0, "mc route table for CHA %d: %s\n",
1406 cur_reg_start, mc_route_string);
1407 else
1408 edac_dbg(0, "mc route table for CHA %d-%d: %s\n",
1409 cur_reg_start, i-1, mc_route_string);
1410 cur_reg_start = i;
1411 }
1412 }
1413 knl_show_mc_route(mc_route_reg[i-1], mc_route_string);
1414 if (cur_reg_start == i-1)
1415 edac_dbg(0, "mc route table for CHA %d: %s\n",
1416 cur_reg_start, mc_route_string);
1417 else
1418 edac_dbg(0, "mc route table for CHA %d-%d: %s\n",
1419 cur_reg_start, i-1, mc_route_string);
1420
1421 /* Process DRAM rules */
1422 for (sad_rule = 0; sad_rule < pvt->info.max_sad; sad_rule++) {
1423 /* previous limit becomes the new base */
1424 sad_base = sad_limit;
1425
1426 pci_read_config_dword(pvt->pci_sad0,
1427 pvt->info.dram_rule[sad_rule], &dram_rule);
1428
1429 if (!DRAM_RULE_ENABLE(dram_rule))
1430 break;
1431
1432 edram_only = KNL_EDRAM_ONLY(dram_rule);
1433
1434 sad_limit = pvt->info.sad_limit(dram_rule)+1;
1435
1436 pci_read_config_dword(pvt->pci_sad0,
1437 pvt->info.interleave_list[sad_rule], &interleave_reg);
1438
1439 /*
1440 * Find out how many ways this dram rule is interleaved.
1441 * We stop when we see the first channel again.
1442 */
1443 first_pkg = sad_pkg(pvt->info.interleave_pkg,
1444 interleave_reg, 0);
1445 for (intrlv_ways = 1; intrlv_ways < 8; intrlv_ways++) {
1446 pkg = sad_pkg(pvt->info.interleave_pkg,
1447 interleave_reg, intrlv_ways);
1448
1449 if ((pkg & 0x8) == 0) {
1450 /*
1451 * 0 bit means memory is non-local,
1452 * which KNL doesn't support
1453 */
1454 edac_dbg(0, "Unexpected interleave target %d\n",
1455 pkg);
1456 return -1;
1457 }
1458
1459 if (pkg == first_pkg)
1460 break;
1461 }
1462 if (KNL_MOD3(dram_rule))
1463 intrlv_ways *= 3;
1464
1465 edac_dbg(3, "dram rule %d (base 0x%llx, limit 0x%llx), %d way interleave%s\n",
1466 sad_rule,
1467 sad_base,
1468 sad_limit,
1469 intrlv_ways,
1470 edram_only ? ", EDRAM" : "");
1471
1472 /*
1473 * Find out how big the SAD region really is by iterating
1474 * over TAD tables (SAD regions may contain holes).
1475 * Each memory controller might have a different TAD table, so
1476 * we have to look at both.
1477 *
1478 * Livespace is the memory that's mapped in this TAD table,
1479 * deadspace is the holes (this could be the MMIO hole, or it
1480 * could be memory that's mapped by the other TAD table but
1481 * not this one).
1482 */
1483 for (mc = 0; mc < 2; mc++) {
1484 sad_actual_size[mc] = 0;
1485 tad_livespace = 0;
1486 for (tad_rule = 0;
1487 tad_rule < ARRAY_SIZE(
1488 knl_tad_dram_limit_lo);
1489 tad_rule++) {
1490 if (knl_get_tad(pvt,
1491 tad_rule,
1492 mc,
1493 &tad_deadspace,
1494 &tad_limit,
1495 &tad_ways))
1496 break;
1497
1498 tad_size = (tad_limit+1) -
1499 (tad_livespace + tad_deadspace);
1500 tad_livespace += tad_size;
1501 tad_base = (tad_limit+1) - tad_size;
1502
1503 if (tad_base < sad_base) {
1504 if (tad_limit > sad_base)
1505 edac_dbg(0, "TAD region overlaps lower SAD boundary -- TAD tables may be configured incorrectly.\n");
1506 } else if (tad_base < sad_limit) {
1507 if (tad_limit+1 > sad_limit) {
1508 edac_dbg(0, "TAD region overlaps upper SAD boundary -- TAD tables may be configured incorrectly.\n");
1509 } else {
1510 /* TAD region is completely inside SAD region */
1511 edac_dbg(3, "TAD region %d 0x%llx - 0x%llx (%lld bytes) table%d\n",
1512 tad_rule, tad_base,
1513 tad_limit, tad_size,
1514 mc);
1515 sad_actual_size[mc] += tad_size;
1516 }
1517 }
1518 }
1519 }
1520
1521 for (mc = 0; mc < 2; mc++) {
1522 edac_dbg(3, " total TAD DRAM footprint in table%d : 0x%llx (%lld bytes)\n",
1523 mc, sad_actual_size[mc], sad_actual_size[mc]);
1524 }
1525
1526 /* Ignore EDRAM rule */
1527 if (edram_only)
1528 continue;
1529
1530 /* Figure out which channels participate in interleave. */
1531 for (channel = 0; channel < KNL_MAX_CHANNELS; channel++)
1532 participants[channel] = 0;
1533
1534 /* For each channel, does at least one CHA have
1535 * this channel mapped to the given target?
1536 */
1537 for (channel = 0; channel < KNL_MAX_CHANNELS; channel++) {
1538 int target;
1539 int cha;
1540
1541 for (target = 0; target < KNL_MAX_CHANNELS; target++) {
1542 for (cha = 0; cha < KNL_MAX_CHAS; cha++) {
1543 if (knl_get_mc_route(target,
1544 mc_route_reg[cha]) == channel
1545 && !participants[channel]) {
1546 participants[channel] = 1;
1547 break;
1548 }
1549 }
1550 }
1551 }
1552
1553 for (channel = 0; channel < KNL_MAX_CHANNELS; channel++) {
1554 mc = knl_channel_mc(channel);
1555 if (participants[channel]) {
1556 edac_dbg(4, "mc channel %d contributes %lld bytes via sad entry %d\n",
1557 channel,
1558 sad_actual_size[mc]/intrlv_ways,
1559 sad_rule);
1560 mc_sizes[channel] +=
1561 sad_actual_size[mc]/intrlv_ways;
1562 }
1563 }
1564 }
1565
1566 return 0;
1567 }
1568
get_source_id(struct mem_ctl_info * mci)1569 static void get_source_id(struct mem_ctl_info *mci)
1570 {
1571 struct sbridge_pvt *pvt = mci->pvt_info;
1572 u32 reg;
1573
1574 if (pvt->info.type == HASWELL || pvt->info.type == BROADWELL ||
1575 pvt->info.type == KNIGHTS_LANDING)
1576 pci_read_config_dword(pvt->pci_sad1, SAD_TARGET, ®);
1577 else
1578 pci_read_config_dword(pvt->pci_br0, SAD_TARGET, ®);
1579
1580 if (pvt->info.type == KNIGHTS_LANDING)
1581 pvt->sbridge_dev->source_id = SOURCE_ID_KNL(reg);
1582 else
1583 pvt->sbridge_dev->source_id = SOURCE_ID(reg);
1584 }
1585
__populate_dimms(struct mem_ctl_info * mci,u64 knl_mc_sizes[KNL_MAX_CHANNELS],enum edac_type mode)1586 static int __populate_dimms(struct mem_ctl_info *mci,
1587 u64 knl_mc_sizes[KNL_MAX_CHANNELS],
1588 enum edac_type mode)
1589 {
1590 struct sbridge_pvt *pvt = mci->pvt_info;
1591 int channels = pvt->info.type == KNIGHTS_LANDING ? KNL_MAX_CHANNELS
1592 : NUM_CHANNELS;
1593 unsigned int i, j, banks, ranks, rows, cols, npages;
1594 struct dimm_info *dimm;
1595 enum mem_type mtype;
1596 u64 size;
1597
1598 mtype = pvt->info.get_memory_type(pvt);
1599 if (mtype == MEM_RDDR3 || mtype == MEM_RDDR4)
1600 edac_dbg(0, "Memory is registered\n");
1601 else if (mtype == MEM_UNKNOWN)
1602 edac_dbg(0, "Cannot determine memory type\n");
1603 else
1604 edac_dbg(0, "Memory is unregistered\n");
1605
1606 if (mtype == MEM_DDR4 || mtype == MEM_RDDR4)
1607 banks = 16;
1608 else
1609 banks = 8;
1610
1611 for (i = 0; i < channels; i++) {
1612 u32 mtr, amap = 0;
1613
1614 int max_dimms_per_channel;
1615
1616 if (pvt->info.type == KNIGHTS_LANDING) {
1617 max_dimms_per_channel = 1;
1618 if (!pvt->knl.pci_channel[i])
1619 continue;
1620 } else {
1621 max_dimms_per_channel = ARRAY_SIZE(mtr_regs);
1622 if (!pvt->pci_tad[i])
1623 continue;
1624 pci_read_config_dword(pvt->pci_tad[i], 0x8c, &amap);
1625 }
1626
1627 for (j = 0; j < max_dimms_per_channel; j++) {
1628 dimm = edac_get_dimm(mci, i, j, 0);
1629 if (pvt->info.type == KNIGHTS_LANDING) {
1630 pci_read_config_dword(pvt->knl.pci_channel[i],
1631 knl_mtr_reg, &mtr);
1632 } else {
1633 pci_read_config_dword(pvt->pci_tad[i],
1634 mtr_regs[j], &mtr);
1635 }
1636 edac_dbg(4, "Channel #%d MTR%d = %x\n", i, j, mtr);
1637
1638 if (IS_DIMM_PRESENT(mtr)) {
1639 if (!IS_ECC_ENABLED(pvt->info.mcmtr)) {
1640 sbridge_printk(KERN_ERR, "CPU SrcID #%d, Ha #%d, Channel #%d has DIMMs, but ECC is disabled\n",
1641 pvt->sbridge_dev->source_id,
1642 pvt->sbridge_dev->dom, i);
1643 return -ENODEV;
1644 }
1645 pvt->channel[i].dimms++;
1646
1647 ranks = numrank(pvt->info.type, mtr);
1648
1649 if (pvt->info.type == KNIGHTS_LANDING) {
1650 /* For DDR4, this is fixed. */
1651 cols = 1 << 10;
1652 rows = knl_mc_sizes[i] /
1653 ((u64) cols * ranks * banks * 8);
1654 } else {
1655 rows = numrow(mtr);
1656 cols = numcol(mtr);
1657 }
1658
1659 size = ((u64)rows * cols * banks * ranks) >> (20 - 3);
1660 npages = MiB_TO_PAGES(size);
1661
1662 edac_dbg(0, "mc#%d: ha %d channel %d, dimm %d, %lld MiB (%d pages) bank: %d, rank: %d, row: %#x, col: %#x\n",
1663 pvt->sbridge_dev->mc, pvt->sbridge_dev->dom, i, j,
1664 size, npages,
1665 banks, ranks, rows, cols);
1666
1667 dimm->nr_pages = npages;
1668 dimm->grain = 32;
1669 dimm->dtype = pvt->info.get_width(pvt, mtr);
1670 dimm->mtype = mtype;
1671 dimm->edac_mode = mode;
1672 pvt->channel[i].dimm[j].rowbits = order_base_2(rows);
1673 pvt->channel[i].dimm[j].colbits = order_base_2(cols);
1674 pvt->channel[i].dimm[j].bank_xor_enable =
1675 GET_BITFIELD(pvt->info.mcmtr, 9, 9);
1676 pvt->channel[i].dimm[j].amap_fine = GET_BITFIELD(amap, 0, 0);
1677 snprintf(dimm->label, sizeof(dimm->label),
1678 "CPU_SrcID#%u_Ha#%u_Chan#%u_DIMM#%u",
1679 pvt->sbridge_dev->source_id, pvt->sbridge_dev->dom, i, j);
1680 }
1681 }
1682 }
1683
1684 return 0;
1685 }
1686
get_dimm_config(struct mem_ctl_info * mci)1687 static int get_dimm_config(struct mem_ctl_info *mci)
1688 {
1689 struct sbridge_pvt *pvt = mci->pvt_info;
1690 u64 knl_mc_sizes[KNL_MAX_CHANNELS];
1691 enum edac_type mode;
1692 u32 reg;
1693
1694 pvt->sbridge_dev->node_id = pvt->info.get_node_id(pvt);
1695 edac_dbg(0, "mc#%d: Node ID: %d, source ID: %d\n",
1696 pvt->sbridge_dev->mc,
1697 pvt->sbridge_dev->node_id,
1698 pvt->sbridge_dev->source_id);
1699
1700 /* KNL doesn't support mirroring or lockstep,
1701 * and is always closed page
1702 */
1703 if (pvt->info.type == KNIGHTS_LANDING) {
1704 mode = EDAC_S4ECD4ED;
1705 pvt->mirror_mode = NON_MIRRORING;
1706 pvt->is_cur_addr_mirrored = false;
1707
1708 if (knl_get_dimm_capacity(pvt, knl_mc_sizes) != 0)
1709 return -1;
1710 if (pci_read_config_dword(pvt->pci_ta, KNL_MCMTR, &pvt->info.mcmtr)) {
1711 edac_dbg(0, "Failed to read KNL_MCMTR register\n");
1712 return -ENODEV;
1713 }
1714 } else {
1715 if (pvt->info.type == HASWELL || pvt->info.type == BROADWELL) {
1716 if (pci_read_config_dword(pvt->pci_ha, HASWELL_HASYSDEFEATURE2, ®)) {
1717 edac_dbg(0, "Failed to read HASWELL_HASYSDEFEATURE2 register\n");
1718 return -ENODEV;
1719 }
1720 pvt->is_chan_hash = GET_BITFIELD(reg, 21, 21);
1721 if (GET_BITFIELD(reg, 28, 28)) {
1722 pvt->mirror_mode = ADDR_RANGE_MIRRORING;
1723 edac_dbg(0, "Address range partial memory mirroring is enabled\n");
1724 goto next;
1725 }
1726 }
1727 if (pci_read_config_dword(pvt->pci_ras, RASENABLES, ®)) {
1728 edac_dbg(0, "Failed to read RASENABLES register\n");
1729 return -ENODEV;
1730 }
1731 if (IS_MIRROR_ENABLED(reg)) {
1732 pvt->mirror_mode = FULL_MIRRORING;
1733 edac_dbg(0, "Full memory mirroring is enabled\n");
1734 } else {
1735 pvt->mirror_mode = NON_MIRRORING;
1736 edac_dbg(0, "Memory mirroring is disabled\n");
1737 }
1738
1739 next:
1740 if (pci_read_config_dword(pvt->pci_ta, MCMTR, &pvt->info.mcmtr)) {
1741 edac_dbg(0, "Failed to read MCMTR register\n");
1742 return -ENODEV;
1743 }
1744 if (IS_LOCKSTEP_ENABLED(pvt->info.mcmtr)) {
1745 edac_dbg(0, "Lockstep is enabled\n");
1746 mode = EDAC_S8ECD8ED;
1747 pvt->is_lockstep = true;
1748 } else {
1749 edac_dbg(0, "Lockstep is disabled\n");
1750 mode = EDAC_S4ECD4ED;
1751 pvt->is_lockstep = false;
1752 }
1753 if (IS_CLOSE_PG(pvt->info.mcmtr)) {
1754 edac_dbg(0, "address map is on closed page mode\n");
1755 pvt->is_close_pg = true;
1756 } else {
1757 edac_dbg(0, "address map is on open page mode\n");
1758 pvt->is_close_pg = false;
1759 }
1760 }
1761
1762 return __populate_dimms(mci, knl_mc_sizes, mode);
1763 }
1764
get_memory_layout(const struct mem_ctl_info * mci)1765 static void get_memory_layout(const struct mem_ctl_info *mci)
1766 {
1767 struct sbridge_pvt *pvt = mci->pvt_info;
1768 int i, j, k, n_sads, n_tads, sad_interl;
1769 u32 reg;
1770 u64 limit, prv = 0;
1771 u64 tmp_mb;
1772 u32 gb, mb;
1773 u32 rir_way;
1774
1775 /*
1776 * Step 1) Get TOLM/TOHM ranges
1777 */
1778
1779 pvt->tolm = pvt->info.get_tolm(pvt);
1780 tmp_mb = (1 + pvt->tolm) >> 20;
1781
1782 gb = div_u64_rem(tmp_mb, 1024, &mb);
1783 edac_dbg(0, "TOLM: %u.%03u GB (0x%016Lx)\n",
1784 gb, (mb*1000)/1024, (u64)pvt->tolm);
1785
1786 /* Address range is already 45:25 */
1787 pvt->tohm = pvt->info.get_tohm(pvt);
1788 tmp_mb = (1 + pvt->tohm) >> 20;
1789
1790 gb = div_u64_rem(tmp_mb, 1024, &mb);
1791 edac_dbg(0, "TOHM: %u.%03u GB (0x%016Lx)\n",
1792 gb, (mb*1000)/1024, (u64)pvt->tohm);
1793
1794 /*
1795 * Step 2) Get SAD range and SAD Interleave list
1796 * TAD registers contain the interleave wayness. However, it
1797 * seems simpler to just discover it indirectly, with the
1798 * algorithm bellow.
1799 */
1800 prv = 0;
1801 for (n_sads = 0; n_sads < pvt->info.max_sad; n_sads++) {
1802 /* SAD_LIMIT Address range is 45:26 */
1803 pci_read_config_dword(pvt->pci_sad0, pvt->info.dram_rule[n_sads],
1804 ®);
1805 limit = pvt->info.sad_limit(reg);
1806
1807 if (!DRAM_RULE_ENABLE(reg))
1808 continue;
1809
1810 if (limit <= prv)
1811 break;
1812
1813 tmp_mb = (limit + 1) >> 20;
1814 gb = div_u64_rem(tmp_mb, 1024, &mb);
1815 edac_dbg(0, "SAD#%d %s up to %u.%03u GB (0x%016Lx) Interleave: %s reg=0x%08x\n",
1816 n_sads,
1817 show_dram_attr(pvt->info.dram_attr(reg)),
1818 gb, (mb*1000)/1024,
1819 ((u64)tmp_mb) << 20L,
1820 get_intlv_mode_str(reg, pvt->info.type),
1821 reg);
1822 prv = limit;
1823
1824 pci_read_config_dword(pvt->pci_sad0, pvt->info.interleave_list[n_sads],
1825 ®);
1826 sad_interl = sad_pkg(pvt->info.interleave_pkg, reg, 0);
1827 for (j = 0; j < 8; j++) {
1828 u32 pkg = sad_pkg(pvt->info.interleave_pkg, reg, j);
1829 if (j > 0 && sad_interl == pkg)
1830 break;
1831
1832 edac_dbg(0, "SAD#%d, interleave #%d: %d\n",
1833 n_sads, j, pkg);
1834 }
1835 }
1836
1837 if (pvt->info.type == KNIGHTS_LANDING)
1838 return;
1839
1840 /*
1841 * Step 3) Get TAD range
1842 */
1843 prv = 0;
1844 for (n_tads = 0; n_tads < MAX_TAD; n_tads++) {
1845 pci_read_config_dword(pvt->pci_ha, tad_dram_rule[n_tads], ®);
1846 limit = TAD_LIMIT(reg);
1847 if (limit <= prv)
1848 break;
1849 tmp_mb = (limit + 1) >> 20;
1850
1851 gb = div_u64_rem(tmp_mb, 1024, &mb);
1852 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",
1853 n_tads, gb, (mb*1000)/1024,
1854 ((u64)tmp_mb) << 20L,
1855 (u32)(1 << TAD_SOCK(reg)),
1856 (u32)TAD_CH(reg) + 1,
1857 (u32)TAD_TGT0(reg),
1858 (u32)TAD_TGT1(reg),
1859 (u32)TAD_TGT2(reg),
1860 (u32)TAD_TGT3(reg),
1861 reg);
1862 prv = limit;
1863 }
1864
1865 /*
1866 * Step 4) Get TAD offsets, per each channel
1867 */
1868 for (i = 0; i < NUM_CHANNELS; i++) {
1869 if (!pvt->channel[i].dimms)
1870 continue;
1871 for (j = 0; j < n_tads; j++) {
1872 pci_read_config_dword(pvt->pci_tad[i],
1873 tad_ch_nilv_offset[j],
1874 ®);
1875 tmp_mb = TAD_OFFSET(reg) >> 20;
1876 gb = div_u64_rem(tmp_mb, 1024, &mb);
1877 edac_dbg(0, "TAD CH#%d, offset #%d: %u.%03u GB (0x%016Lx), reg=0x%08x\n",
1878 i, j,
1879 gb, (mb*1000)/1024,
1880 ((u64)tmp_mb) << 20L,
1881 reg);
1882 }
1883 }
1884
1885 /*
1886 * Step 6) Get RIR Wayness/Limit, per each channel
1887 */
1888 for (i = 0; i < NUM_CHANNELS; i++) {
1889 if (!pvt->channel[i].dimms)
1890 continue;
1891 for (j = 0; j < MAX_RIR_RANGES; j++) {
1892 pci_read_config_dword(pvt->pci_tad[i],
1893 rir_way_limit[j],
1894 ®);
1895
1896 if (!IS_RIR_VALID(reg))
1897 continue;
1898
1899 tmp_mb = pvt->info.rir_limit(reg) >> 20;
1900 rir_way = 1 << RIR_WAY(reg);
1901 gb = div_u64_rem(tmp_mb, 1024, &mb);
1902 edac_dbg(0, "CH#%d RIR#%d, limit: %u.%03u GB (0x%016Lx), way: %d, reg=0x%08x\n",
1903 i, j,
1904 gb, (mb*1000)/1024,
1905 ((u64)tmp_mb) << 20L,
1906 rir_way,
1907 reg);
1908
1909 for (k = 0; k < rir_way; k++) {
1910 pci_read_config_dword(pvt->pci_tad[i],
1911 rir_offset[j][k],
1912 ®);
1913 tmp_mb = RIR_OFFSET(pvt->info.type, reg) << 6;
1914
1915 gb = div_u64_rem(tmp_mb, 1024, &mb);
1916 edac_dbg(0, "CH#%d RIR#%d INTL#%d, offset %u.%03u GB (0x%016Lx), tgt: %d, reg=0x%08x\n",
1917 i, j, k,
1918 gb, (mb*1000)/1024,
1919 ((u64)tmp_mb) << 20L,
1920 (u32)RIR_RNK_TGT(pvt->info.type, reg),
1921 reg);
1922 }
1923 }
1924 }
1925 }
1926
get_mci_for_node_id(u8 node_id,u8 ha)1927 static struct mem_ctl_info *get_mci_for_node_id(u8 node_id, u8 ha)
1928 {
1929 struct sbridge_dev *sbridge_dev;
1930
1931 list_for_each_entry(sbridge_dev, &sbridge_edac_list, list) {
1932 if (sbridge_dev->node_id == node_id && sbridge_dev->dom == ha)
1933 return sbridge_dev->mci;
1934 }
1935 return NULL;
1936 }
1937
1938 static u8 sb_close_row[] = {
1939 15, 16, 17, 18, 20, 21, 22, 28, 10, 11, 12, 13, 29, 30, 31, 32, 33
1940 };
1941
1942 static u8 sb_close_column[] = {
1943 3, 4, 5, 14, 19, 23, 24, 25, 26, 27
1944 };
1945
1946 static u8 sb_open_row[] = {
1947 14, 15, 16, 20, 28, 21, 22, 23, 24, 25, 26, 27, 29, 30, 31, 32, 33
1948 };
1949
1950 static u8 sb_open_column[] = {
1951 3, 4, 5, 6, 7, 8, 9, 10, 11, 12
1952 };
1953
1954 static u8 sb_open_fine_column[] = {
1955 3, 4, 5, 7, 8, 9, 10, 11, 12, 13
1956 };
1957
sb_bits(u64 addr,int nbits,u8 * bits)1958 static int sb_bits(u64 addr, int nbits, u8 *bits)
1959 {
1960 int i, res = 0;
1961
1962 for (i = 0; i < nbits; i++)
1963 res |= ((addr >> bits[i]) & 1) << i;
1964 return res;
1965 }
1966
sb_bank_bits(u64 addr,int b0,int b1,int do_xor,int x0,int x1)1967 static int sb_bank_bits(u64 addr, int b0, int b1, int do_xor, int x0, int x1)
1968 {
1969 int ret = GET_BITFIELD(addr, b0, b0) | (GET_BITFIELD(addr, b1, b1) << 1);
1970
1971 if (do_xor)
1972 ret ^= GET_BITFIELD(addr, x0, x0) | (GET_BITFIELD(addr, x1, x1) << 1);
1973
1974 return ret;
1975 }
1976
sb_decode_ddr4(struct mem_ctl_info * mci,int ch,u8 rank,u64 rank_addr,char * msg)1977 static bool sb_decode_ddr4(struct mem_ctl_info *mci, int ch, u8 rank,
1978 u64 rank_addr, char *msg)
1979 {
1980 int dimmno = 0;
1981 int row, col, bank_address, bank_group;
1982 struct sbridge_pvt *pvt;
1983 u32 bg0 = 0, rowbits = 0, colbits = 0;
1984 u32 amap_fine = 0, bank_xor_enable = 0;
1985
1986 dimmno = (rank < 12) ? rank / 4 : 2;
1987 pvt = mci->pvt_info;
1988 amap_fine = pvt->channel[ch].dimm[dimmno].amap_fine;
1989 bg0 = amap_fine ? 6 : 13;
1990 rowbits = pvt->channel[ch].dimm[dimmno].rowbits;
1991 colbits = pvt->channel[ch].dimm[dimmno].colbits;
1992 bank_xor_enable = pvt->channel[ch].dimm[dimmno].bank_xor_enable;
1993
1994 if (pvt->is_lockstep) {
1995 pr_warn_once("LockStep row/column decode is not supported yet!\n");
1996 msg[0] = '\0';
1997 return false;
1998 }
1999
2000 if (pvt->is_close_pg) {
2001 row = sb_bits(rank_addr, rowbits, sb_close_row);
2002 col = sb_bits(rank_addr, colbits, sb_close_column);
2003 col |= 0x400; /* C10 is autoprecharge, always set */
2004 bank_address = sb_bank_bits(rank_addr, 8, 9, bank_xor_enable, 22, 28);
2005 bank_group = sb_bank_bits(rank_addr, 6, 7, bank_xor_enable, 20, 21);
2006 } else {
2007 row = sb_bits(rank_addr, rowbits, sb_open_row);
2008 if (amap_fine)
2009 col = sb_bits(rank_addr, colbits, sb_open_fine_column);
2010 else
2011 col = sb_bits(rank_addr, colbits, sb_open_column);
2012 bank_address = sb_bank_bits(rank_addr, 18, 19, bank_xor_enable, 22, 23);
2013 bank_group = sb_bank_bits(rank_addr, bg0, 17, bank_xor_enable, 20, 21);
2014 }
2015
2016 row &= (1u << rowbits) - 1;
2017
2018 sprintf(msg, "row:0x%x col:0x%x bank_addr:%d bank_group:%d",
2019 row, col, bank_address, bank_group);
2020 return true;
2021 }
2022
sb_decode_ddr3(struct mem_ctl_info * mci,int ch,u8 rank,u64 rank_addr,char * msg)2023 static bool sb_decode_ddr3(struct mem_ctl_info *mci, int ch, u8 rank,
2024 u64 rank_addr, char *msg)
2025 {
2026 pr_warn_once("DDR3 row/column decode not support yet!\n");
2027 msg[0] = '\0';
2028 return false;
2029 }
2030
get_memory_error_data(struct mem_ctl_info * mci,u64 addr,u8 * socket,u8 * ha,long * channel_mask,u8 * rank,char ** area_type,char * msg)2031 static int get_memory_error_data(struct mem_ctl_info *mci,
2032 u64 addr,
2033 u8 *socket, u8 *ha,
2034 long *channel_mask,
2035 u8 *rank,
2036 char **area_type, char *msg)
2037 {
2038 struct mem_ctl_info *new_mci;
2039 struct sbridge_pvt *pvt = mci->pvt_info;
2040 struct pci_dev *pci_ha;
2041 int n_rir, n_sads, n_tads, sad_way, sck_xch;
2042 int sad_interl, idx, base_ch;
2043 int interleave_mode, shiftup = 0;
2044 unsigned int sad_interleave[MAX_INTERLEAVE];
2045 u32 reg, dram_rule;
2046 u8 ch_way, sck_way, pkg, sad_ha = 0, rankid = 0;
2047 u32 tad_offset;
2048 u32 rir_way;
2049 u32 mb, gb;
2050 u64 ch_addr, offset, limit = 0, prv = 0;
2051 u64 rank_addr;
2052 enum mem_type mtype;
2053
2054 /*
2055 * Step 0) Check if the address is at special memory ranges
2056 * The check bellow is probably enough to fill all cases where
2057 * the error is not inside a memory, except for the legacy
2058 * range (e. g. VGA addresses). It is unlikely, however, that the
2059 * memory controller would generate an error on that range.
2060 */
2061 if ((addr > (u64) pvt->tolm) && (addr < (1LL << 32))) {
2062 sprintf(msg, "Error at TOLM area, on addr 0x%08Lx", addr);
2063 return -EINVAL;
2064 }
2065 if (addr >= (u64)pvt->tohm) {
2066 sprintf(msg, "Error at MMIOH area, on addr 0x%016Lx", addr);
2067 return -EINVAL;
2068 }
2069
2070 /*
2071 * Step 1) Get socket
2072 */
2073 for (n_sads = 0; n_sads < pvt->info.max_sad; n_sads++) {
2074 pci_read_config_dword(pvt->pci_sad0, pvt->info.dram_rule[n_sads],
2075 ®);
2076
2077 if (!DRAM_RULE_ENABLE(reg))
2078 continue;
2079
2080 limit = pvt->info.sad_limit(reg);
2081 if (limit <= prv) {
2082 sprintf(msg, "Can't discover the memory socket");
2083 return -EINVAL;
2084 }
2085 if (addr <= limit)
2086 break;
2087 prv = limit;
2088 }
2089 if (n_sads == pvt->info.max_sad) {
2090 sprintf(msg, "Can't discover the memory socket");
2091 return -EINVAL;
2092 }
2093 dram_rule = reg;
2094 *area_type = show_dram_attr(pvt->info.dram_attr(dram_rule));
2095 interleave_mode = pvt->info.interleave_mode(dram_rule);
2096
2097 pci_read_config_dword(pvt->pci_sad0, pvt->info.interleave_list[n_sads],
2098 ®);
2099
2100 if (pvt->info.type == SANDY_BRIDGE) {
2101 sad_interl = sad_pkg(pvt->info.interleave_pkg, reg, 0);
2102 for (sad_way = 0; sad_way < 8; sad_way++) {
2103 u32 pkg = sad_pkg(pvt->info.interleave_pkg, reg, sad_way);
2104 if (sad_way > 0 && sad_interl == pkg)
2105 break;
2106 sad_interleave[sad_way] = pkg;
2107 edac_dbg(0, "SAD interleave #%d: %d\n",
2108 sad_way, sad_interleave[sad_way]);
2109 }
2110 edac_dbg(0, "mc#%d: Error detected on SAD#%d: address 0x%016Lx < 0x%016Lx, Interleave [%d:6]%s\n",
2111 pvt->sbridge_dev->mc,
2112 n_sads,
2113 addr,
2114 limit,
2115 sad_way + 7,
2116 !interleave_mode ? "" : "XOR[18:16]");
2117 if (interleave_mode)
2118 idx = ((addr >> 6) ^ (addr >> 16)) & 7;
2119 else
2120 idx = (addr >> 6) & 7;
2121 switch (sad_way) {
2122 case 1:
2123 idx = 0;
2124 break;
2125 case 2:
2126 idx = idx & 1;
2127 break;
2128 case 4:
2129 idx = idx & 3;
2130 break;
2131 case 8:
2132 break;
2133 default:
2134 sprintf(msg, "Can't discover socket interleave");
2135 return -EINVAL;
2136 }
2137 *socket = sad_interleave[idx];
2138 edac_dbg(0, "SAD interleave index: %d (wayness %d) = CPU socket %d\n",
2139 idx, sad_way, *socket);
2140 } else if (pvt->info.type == HASWELL || pvt->info.type == BROADWELL) {
2141 int bits, a7mode = A7MODE(dram_rule);
2142
2143 if (a7mode) {
2144 /* A7 mode swaps P9 with P6 */
2145 bits = GET_BITFIELD(addr, 7, 8) << 1;
2146 bits |= GET_BITFIELD(addr, 9, 9);
2147 } else
2148 bits = GET_BITFIELD(addr, 6, 8);
2149
2150 if (interleave_mode == 0) {
2151 /* interleave mode will XOR {8,7,6} with {18,17,16} */
2152 idx = GET_BITFIELD(addr, 16, 18);
2153 idx ^= bits;
2154 } else
2155 idx = bits;
2156
2157 pkg = sad_pkg(pvt->info.interleave_pkg, reg, idx);
2158 *socket = sad_pkg_socket(pkg);
2159 sad_ha = sad_pkg_ha(pkg);
2160
2161 if (a7mode) {
2162 /* MCChanShiftUpEnable */
2163 pci_read_config_dword(pvt->pci_ha, HASWELL_HASYSDEFEATURE2, ®);
2164 shiftup = GET_BITFIELD(reg, 22, 22);
2165 }
2166
2167 edac_dbg(0, "SAD interleave package: %d = CPU socket %d, HA %i, shiftup: %i\n",
2168 idx, *socket, sad_ha, shiftup);
2169 } else {
2170 /* Ivy Bridge's SAD mode doesn't support XOR interleave mode */
2171 idx = (addr >> 6) & 7;
2172 pkg = sad_pkg(pvt->info.interleave_pkg, reg, idx);
2173 *socket = sad_pkg_socket(pkg);
2174 sad_ha = sad_pkg_ha(pkg);
2175 edac_dbg(0, "SAD interleave package: %d = CPU socket %d, HA %d\n",
2176 idx, *socket, sad_ha);
2177 }
2178
2179 *ha = sad_ha;
2180
2181 /*
2182 * Move to the proper node structure, in order to access the
2183 * right PCI registers
2184 */
2185 new_mci = get_mci_for_node_id(*socket, sad_ha);
2186 if (!new_mci) {
2187 sprintf(msg, "Struct for socket #%u wasn't initialized",
2188 *socket);
2189 return -EINVAL;
2190 }
2191 mci = new_mci;
2192 pvt = mci->pvt_info;
2193
2194 /*
2195 * Step 2) Get memory channel
2196 */
2197 prv = 0;
2198 pci_ha = pvt->pci_ha;
2199 for (n_tads = 0; n_tads < MAX_TAD; n_tads++) {
2200 pci_read_config_dword(pci_ha, tad_dram_rule[n_tads], ®);
2201 limit = TAD_LIMIT(reg);
2202 if (limit <= prv) {
2203 sprintf(msg, "Can't discover the memory channel");
2204 return -EINVAL;
2205 }
2206 if (addr <= limit)
2207 break;
2208 prv = limit;
2209 }
2210 if (n_tads == MAX_TAD) {
2211 sprintf(msg, "Can't discover the memory channel");
2212 return -EINVAL;
2213 }
2214
2215 ch_way = TAD_CH(reg) + 1;
2216 sck_way = TAD_SOCK(reg);
2217
2218 if (ch_way == 3)
2219 idx = addr >> 6;
2220 else {
2221 idx = (addr >> (6 + sck_way + shiftup)) & 0x3;
2222 if (pvt->is_chan_hash)
2223 idx = haswell_chan_hash(idx, addr);
2224 }
2225 idx = idx % ch_way;
2226
2227 /*
2228 * FIXME: Shouldn't we use CHN_IDX_OFFSET() here, when ch_way == 3 ???
2229 */
2230 switch (idx) {
2231 case 0:
2232 base_ch = TAD_TGT0(reg);
2233 break;
2234 case 1:
2235 base_ch = TAD_TGT1(reg);
2236 break;
2237 case 2:
2238 base_ch = TAD_TGT2(reg);
2239 break;
2240 case 3:
2241 base_ch = TAD_TGT3(reg);
2242 break;
2243 default:
2244 sprintf(msg, "Can't discover the TAD target");
2245 return -EINVAL;
2246 }
2247 *channel_mask = 1 << base_ch;
2248
2249 pci_read_config_dword(pvt->pci_tad[base_ch], tad_ch_nilv_offset[n_tads], &tad_offset);
2250
2251 if (pvt->mirror_mode == FULL_MIRRORING ||
2252 (pvt->mirror_mode == ADDR_RANGE_MIRRORING && n_tads == 0)) {
2253 *channel_mask |= 1 << ((base_ch + 2) % 4);
2254 switch(ch_way) {
2255 case 2:
2256 case 4:
2257 sck_xch = (1 << sck_way) * (ch_way >> 1);
2258 break;
2259 default:
2260 sprintf(msg, "Invalid mirror set. Can't decode addr");
2261 return -EINVAL;
2262 }
2263
2264 pvt->is_cur_addr_mirrored = true;
2265 } else {
2266 sck_xch = (1 << sck_way) * ch_way;
2267 pvt->is_cur_addr_mirrored = false;
2268 }
2269
2270 if (pvt->is_lockstep)
2271 *channel_mask |= 1 << ((base_ch + 1) % 4);
2272
2273 offset = TAD_OFFSET(tad_offset);
2274
2275 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",
2276 n_tads,
2277 addr,
2278 limit,
2279 sck_way,
2280 ch_way,
2281 offset,
2282 idx,
2283 base_ch,
2284 *channel_mask);
2285
2286 /* Calculate channel address */
2287 /* Remove the TAD offset */
2288
2289 if (offset > addr) {
2290 sprintf(msg, "Can't calculate ch addr: TAD offset 0x%08Lx is too high for addr 0x%08Lx!",
2291 offset, addr);
2292 return -EINVAL;
2293 }
2294
2295 ch_addr = addr - offset;
2296 ch_addr >>= (6 + shiftup);
2297 ch_addr /= sck_xch;
2298 ch_addr <<= (6 + shiftup);
2299 ch_addr |= addr & ((1 << (6 + shiftup)) - 1);
2300
2301 /*
2302 * Step 3) Decode rank
2303 */
2304 for (n_rir = 0; n_rir < MAX_RIR_RANGES; n_rir++) {
2305 pci_read_config_dword(pvt->pci_tad[base_ch], rir_way_limit[n_rir], ®);
2306
2307 if (!IS_RIR_VALID(reg))
2308 continue;
2309
2310 limit = pvt->info.rir_limit(reg);
2311 gb = div_u64_rem(limit >> 20, 1024, &mb);
2312 edac_dbg(0, "RIR#%d, limit: %u.%03u GB (0x%016Lx), way: %d\n",
2313 n_rir,
2314 gb, (mb*1000)/1024,
2315 limit,
2316 1 << RIR_WAY(reg));
2317 if (ch_addr <= limit)
2318 break;
2319 }
2320 if (n_rir == MAX_RIR_RANGES) {
2321 sprintf(msg, "Can't discover the memory rank for ch addr 0x%08Lx",
2322 ch_addr);
2323 return -EINVAL;
2324 }
2325 rir_way = RIR_WAY(reg);
2326
2327 if (pvt->is_close_pg)
2328 idx = (ch_addr >> 6);
2329 else
2330 idx = (ch_addr >> 13); /* FIXME: Datasheet says to shift by 15 */
2331 idx %= 1 << rir_way;
2332
2333 pci_read_config_dword(pvt->pci_tad[base_ch], rir_offset[n_rir][idx], ®);
2334 *rank = RIR_RNK_TGT(pvt->info.type, reg);
2335
2336 if (pvt->info.type == BROADWELL) {
2337 if (pvt->is_close_pg)
2338 shiftup = 6;
2339 else
2340 shiftup = 13;
2341
2342 rank_addr = ch_addr >> shiftup;
2343 rank_addr /= (1 << rir_way);
2344 rank_addr <<= shiftup;
2345 rank_addr |= ch_addr & GENMASK_ULL(shiftup - 1, 0);
2346 rank_addr -= RIR_OFFSET(pvt->info.type, reg);
2347
2348 mtype = pvt->info.get_memory_type(pvt);
2349 rankid = *rank;
2350 if (mtype == MEM_DDR4 || mtype == MEM_RDDR4)
2351 sb_decode_ddr4(mci, base_ch, rankid, rank_addr, msg);
2352 else
2353 sb_decode_ddr3(mci, base_ch, rankid, rank_addr, msg);
2354 } else {
2355 msg[0] = '\0';
2356 }
2357
2358 edac_dbg(0, "RIR#%d: channel address 0x%08Lx < 0x%08Lx, RIR interleave %d, index %d\n",
2359 n_rir,
2360 ch_addr,
2361 limit,
2362 rir_way,
2363 idx);
2364
2365 return 0;
2366 }
2367
get_memory_error_data_from_mce(struct mem_ctl_info * mci,const struct mce * m,u8 * socket,u8 * ha,long * channel_mask,char * msg)2368 static int get_memory_error_data_from_mce(struct mem_ctl_info *mci,
2369 const struct mce *m, u8 *socket,
2370 u8 *ha, long *channel_mask,
2371 char *msg)
2372 {
2373 u32 reg, channel = GET_BITFIELD(m->status, 0, 3);
2374 struct mem_ctl_info *new_mci;
2375 struct sbridge_pvt *pvt;
2376 struct pci_dev *pci_ha;
2377 bool tad0;
2378
2379 if (channel >= NUM_CHANNELS) {
2380 sprintf(msg, "Invalid channel 0x%x", channel);
2381 return -EINVAL;
2382 }
2383
2384 pvt = mci->pvt_info;
2385 if (!pvt->info.get_ha) {
2386 sprintf(msg, "No get_ha()");
2387 return -EINVAL;
2388 }
2389 *ha = pvt->info.get_ha(m->bank);
2390 if (*ha != 0 && *ha != 1) {
2391 sprintf(msg, "Impossible bank %d", m->bank);
2392 return -EINVAL;
2393 }
2394
2395 *socket = m->socketid;
2396 new_mci = get_mci_for_node_id(*socket, *ha);
2397 if (!new_mci) {
2398 strcpy(msg, "mci socket got corrupted!");
2399 return -EINVAL;
2400 }
2401
2402 pvt = new_mci->pvt_info;
2403 pci_ha = pvt->pci_ha;
2404 pci_read_config_dword(pci_ha, tad_dram_rule[0], ®);
2405 tad0 = m->addr <= TAD_LIMIT(reg);
2406
2407 *channel_mask = 1 << channel;
2408 if (pvt->mirror_mode == FULL_MIRRORING ||
2409 (pvt->mirror_mode == ADDR_RANGE_MIRRORING && tad0)) {
2410 *channel_mask |= 1 << ((channel + 2) % 4);
2411 pvt->is_cur_addr_mirrored = true;
2412 } else {
2413 pvt->is_cur_addr_mirrored = false;
2414 }
2415
2416 if (pvt->is_lockstep)
2417 *channel_mask |= 1 << ((channel + 1) % 4);
2418
2419 return 0;
2420 }
2421
2422 /****************************************************************************
2423 Device initialization routines: put/get, init/exit
2424 ****************************************************************************/
2425
2426 /*
2427 * sbridge_put_all_devices 'put' all the devices that we have
2428 * reserved via 'get'
2429 */
sbridge_put_devices(struct sbridge_dev * sbridge_dev)2430 static void sbridge_put_devices(struct sbridge_dev *sbridge_dev)
2431 {
2432 int i;
2433
2434 edac_dbg(0, "\n");
2435 for (i = 0; i < sbridge_dev->n_devs; i++) {
2436 struct pci_dev *pdev = sbridge_dev->pdev[i];
2437 if (!pdev)
2438 continue;
2439 edac_dbg(0, "Removing dev %02x:%02x.%d\n",
2440 pdev->bus->number,
2441 PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn));
2442 pci_dev_put(pdev);
2443 }
2444 }
2445
sbridge_put_all_devices(void)2446 static void sbridge_put_all_devices(void)
2447 {
2448 struct sbridge_dev *sbridge_dev, *tmp;
2449
2450 list_for_each_entry_safe(sbridge_dev, tmp, &sbridge_edac_list, list) {
2451 sbridge_put_devices(sbridge_dev);
2452 free_sbridge_dev(sbridge_dev);
2453 }
2454 }
2455
sbridge_get_onedevice(struct pci_dev ** prev,u8 * num_mc,const struct pci_id_table * table,const unsigned devno,const int multi_bus)2456 static int sbridge_get_onedevice(struct pci_dev **prev,
2457 u8 *num_mc,
2458 const struct pci_id_table *table,
2459 const unsigned devno,
2460 const int multi_bus)
2461 {
2462 struct sbridge_dev *sbridge_dev = NULL;
2463 const struct pci_id_descr *dev_descr = &table->descr[devno];
2464 struct pci_dev *pdev = NULL;
2465 int seg = 0;
2466 u8 bus = 0;
2467 int i = 0;
2468
2469 sbridge_printk(KERN_DEBUG,
2470 "Seeking for: PCI ID %04x:%04x\n",
2471 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
2472
2473 pdev = pci_get_device(PCI_VENDOR_ID_INTEL,
2474 dev_descr->dev_id, *prev);
2475
2476 if (!pdev) {
2477 if (*prev) {
2478 *prev = pdev;
2479 return 0;
2480 }
2481
2482 if (dev_descr->optional)
2483 return 0;
2484
2485 /* if the HA wasn't found */
2486 if (devno == 0)
2487 return -ENODEV;
2488
2489 sbridge_printk(KERN_INFO,
2490 "Device not found: %04x:%04x\n",
2491 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
2492
2493 /* End of list, leave */
2494 return -ENODEV;
2495 }
2496 seg = pci_domain_nr(pdev->bus);
2497 bus = pdev->bus->number;
2498
2499 next_imc:
2500 sbridge_dev = get_sbridge_dev(seg, bus, dev_descr->dom,
2501 multi_bus, sbridge_dev);
2502 if (!sbridge_dev) {
2503 /* If the HA1 wasn't found, don't create EDAC second memory controller */
2504 if (dev_descr->dom == IMC1 && devno != 1) {
2505 edac_dbg(0, "Skip IMC1: %04x:%04x (since HA1 was absent)\n",
2506 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
2507 pci_dev_put(pdev);
2508 return 0;
2509 }
2510
2511 if (dev_descr->dom == SOCK)
2512 goto out_imc;
2513
2514 sbridge_dev = alloc_sbridge_dev(seg, bus, dev_descr->dom, table);
2515 if (!sbridge_dev) {
2516 pci_dev_put(pdev);
2517 return -ENOMEM;
2518 }
2519 (*num_mc)++;
2520 }
2521
2522 if (sbridge_dev->pdev[sbridge_dev->i_devs]) {
2523 sbridge_printk(KERN_ERR,
2524 "Duplicated device for %04x:%04x\n",
2525 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
2526 pci_dev_put(pdev);
2527 return -ENODEV;
2528 }
2529
2530 sbridge_dev->pdev[sbridge_dev->i_devs++] = pdev;
2531
2532 /* pdev belongs to more than one IMC, do extra gets */
2533 if (++i > 1)
2534 pci_dev_get(pdev);
2535
2536 if (dev_descr->dom == SOCK && i < table->n_imcs_per_sock)
2537 goto next_imc;
2538
2539 out_imc:
2540 /* Be sure that the device is enabled */
2541 if (unlikely(pci_enable_device(pdev) < 0)) {
2542 sbridge_printk(KERN_ERR,
2543 "Couldn't enable %04x:%04x\n",
2544 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
2545 return -ENODEV;
2546 }
2547
2548 edac_dbg(0, "Detected %04x:%04x\n",
2549 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
2550
2551 /*
2552 * As stated on drivers/pci/search.c, the reference count for
2553 * @from is always decremented if it is not %NULL. So, as we need
2554 * to get all devices up to null, we need to do a get for the device
2555 */
2556 pci_dev_get(pdev);
2557
2558 *prev = pdev;
2559
2560 return 0;
2561 }
2562
2563 /*
2564 * sbridge_get_all_devices - Find and perform 'get' operation on the MCH's
2565 * devices we want to reference for this driver.
2566 * @num_mc: pointer to the memory controllers count, to be incremented in case
2567 * of success.
2568 * @table: model specific table
2569 *
2570 * returns 0 in case of success or error code
2571 */
sbridge_get_all_devices(u8 * num_mc,const struct pci_id_table * table)2572 static int sbridge_get_all_devices(u8 *num_mc,
2573 const struct pci_id_table *table)
2574 {
2575 int i, rc;
2576 struct pci_dev *pdev = NULL;
2577 int allow_dups = 0;
2578 int multi_bus = 0;
2579
2580 if (table->type == KNIGHTS_LANDING)
2581 allow_dups = multi_bus = 1;
2582 while (table && table->descr) {
2583 for (i = 0; i < table->n_devs_per_sock; i++) {
2584 if (!allow_dups || i == 0 ||
2585 table->descr[i].dev_id !=
2586 table->descr[i-1].dev_id) {
2587 pdev = NULL;
2588 }
2589 do {
2590 rc = sbridge_get_onedevice(&pdev, num_mc,
2591 table, i, multi_bus);
2592 if (rc < 0) {
2593 if (i == 0) {
2594 i = table->n_devs_per_sock;
2595 break;
2596 }
2597 sbridge_put_all_devices();
2598 return -ENODEV;
2599 }
2600 } while (pdev && !allow_dups);
2601 }
2602 table++;
2603 }
2604
2605 return 0;
2606 }
2607
2608 /*
2609 * Device IDs for {SBRIDGE,IBRIDGE,HASWELL,BROADWELL}_IMC_HA0_TAD0 are in
2610 * the format: XXXa. So we can convert from a device to the corresponding
2611 * channel like this
2612 */
2613 #define TAD_DEV_TO_CHAN(dev) (((dev) & 0xf) - 0xa)
2614
sbridge_mci_bind_devs(struct mem_ctl_info * mci,struct sbridge_dev * sbridge_dev)2615 static int sbridge_mci_bind_devs(struct mem_ctl_info *mci,
2616 struct sbridge_dev *sbridge_dev)
2617 {
2618 struct sbridge_pvt *pvt = mci->pvt_info;
2619 struct pci_dev *pdev;
2620 u8 saw_chan_mask = 0;
2621 int i;
2622
2623 for (i = 0; i < sbridge_dev->n_devs; i++) {
2624 pdev = sbridge_dev->pdev[i];
2625 if (!pdev)
2626 continue;
2627
2628 switch (pdev->device) {
2629 case PCI_DEVICE_ID_INTEL_SBRIDGE_SAD0:
2630 pvt->pci_sad0 = pdev;
2631 break;
2632 case PCI_DEVICE_ID_INTEL_SBRIDGE_SAD1:
2633 pvt->pci_sad1 = pdev;
2634 break;
2635 case PCI_DEVICE_ID_INTEL_SBRIDGE_BR:
2636 pvt->pci_br0 = pdev;
2637 break;
2638 case PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_HA0:
2639 pvt->pci_ha = pdev;
2640 break;
2641 case PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TA:
2642 pvt->pci_ta = pdev;
2643 break;
2644 case PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_RAS:
2645 pvt->pci_ras = pdev;
2646 break;
2647 case PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD0:
2648 case PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD1:
2649 case PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD2:
2650 case PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD3:
2651 {
2652 int id = TAD_DEV_TO_CHAN(pdev->device);
2653 pvt->pci_tad[id] = pdev;
2654 saw_chan_mask |= 1 << id;
2655 }
2656 break;
2657 case PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_DDRIO:
2658 pvt->pci_ddrio = pdev;
2659 break;
2660 default:
2661 goto error;
2662 }
2663
2664 edac_dbg(0, "Associated PCI %02x:%02x, bus %d with dev = %p\n",
2665 pdev->vendor, pdev->device,
2666 sbridge_dev->bus,
2667 pdev);
2668 }
2669
2670 /* Check if everything were registered */
2671 if (!pvt->pci_sad0 || !pvt->pci_sad1 || !pvt->pci_ha ||
2672 !pvt->pci_ras || !pvt->pci_ta)
2673 goto enodev;
2674
2675 if (saw_chan_mask != 0x0f)
2676 goto enodev;
2677 return 0;
2678
2679 enodev:
2680 sbridge_printk(KERN_ERR, "Some needed devices are missing\n");
2681 return -ENODEV;
2682
2683 error:
2684 sbridge_printk(KERN_ERR, "Unexpected device %02x:%02x\n",
2685 PCI_VENDOR_ID_INTEL, pdev->device);
2686 return -EINVAL;
2687 }
2688
ibridge_mci_bind_devs(struct mem_ctl_info * mci,struct sbridge_dev * sbridge_dev)2689 static int ibridge_mci_bind_devs(struct mem_ctl_info *mci,
2690 struct sbridge_dev *sbridge_dev)
2691 {
2692 struct sbridge_pvt *pvt = mci->pvt_info;
2693 struct pci_dev *pdev;
2694 u8 saw_chan_mask = 0;
2695 int i;
2696
2697 for (i = 0; i < sbridge_dev->n_devs; i++) {
2698 pdev = sbridge_dev->pdev[i];
2699 if (!pdev)
2700 continue;
2701
2702 switch (pdev->device) {
2703 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0:
2704 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1:
2705 pvt->pci_ha = pdev;
2706 break;
2707 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TA:
2708 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TA:
2709 pvt->pci_ta = pdev;
2710 break;
2711 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_RAS:
2712 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_RAS:
2713 pvt->pci_ras = pdev;
2714 break;
2715 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD0:
2716 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD1:
2717 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD2:
2718 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD3:
2719 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD0:
2720 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD1:
2721 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD2:
2722 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD3:
2723 {
2724 int id = TAD_DEV_TO_CHAN(pdev->device);
2725 pvt->pci_tad[id] = pdev;
2726 saw_chan_mask |= 1 << id;
2727 }
2728 break;
2729 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_2HA_DDRIO0:
2730 pvt->pci_ddrio = pdev;
2731 break;
2732 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_1HA_DDRIO0:
2733 pvt->pci_ddrio = pdev;
2734 break;
2735 case PCI_DEVICE_ID_INTEL_IBRIDGE_SAD:
2736 pvt->pci_sad0 = pdev;
2737 break;
2738 case PCI_DEVICE_ID_INTEL_IBRIDGE_BR0:
2739 pvt->pci_br0 = pdev;
2740 break;
2741 case PCI_DEVICE_ID_INTEL_IBRIDGE_BR1:
2742 pvt->pci_br1 = pdev;
2743 break;
2744 default:
2745 goto error;
2746 }
2747
2748 edac_dbg(0, "Associated PCI %02x.%02d.%d with dev = %p\n",
2749 sbridge_dev->bus,
2750 PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn),
2751 pdev);
2752 }
2753
2754 /* Check if everything were registered */
2755 if (!pvt->pci_sad0 || !pvt->pci_ha || !pvt->pci_br0 ||
2756 !pvt->pci_br1 || !pvt->pci_ras || !pvt->pci_ta)
2757 goto enodev;
2758
2759 if (saw_chan_mask != 0x0f && /* -EN/-EX */
2760 saw_chan_mask != 0x03) /* -EP */
2761 goto enodev;
2762 return 0;
2763
2764 enodev:
2765 sbridge_printk(KERN_ERR, "Some needed devices are missing\n");
2766 return -ENODEV;
2767
2768 error:
2769 sbridge_printk(KERN_ERR,
2770 "Unexpected device %02x:%02x\n", PCI_VENDOR_ID_INTEL,
2771 pdev->device);
2772 return -EINVAL;
2773 }
2774
haswell_mci_bind_devs(struct mem_ctl_info * mci,struct sbridge_dev * sbridge_dev)2775 static int haswell_mci_bind_devs(struct mem_ctl_info *mci,
2776 struct sbridge_dev *sbridge_dev)
2777 {
2778 struct sbridge_pvt *pvt = mci->pvt_info;
2779 struct pci_dev *pdev;
2780 u8 saw_chan_mask = 0;
2781 int i;
2782
2783 /* there's only one device per system; not tied to any bus */
2784 if (pvt->info.pci_vtd == NULL)
2785 /* result will be checked later */
2786 pvt->info.pci_vtd = pci_get_device(PCI_VENDOR_ID_INTEL,
2787 PCI_DEVICE_ID_INTEL_HASWELL_IMC_VTD_MISC,
2788 NULL);
2789
2790 for (i = 0; i < sbridge_dev->n_devs; i++) {
2791 pdev = sbridge_dev->pdev[i];
2792 if (!pdev)
2793 continue;
2794
2795 switch (pdev->device) {
2796 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_CBO_SAD0:
2797 pvt->pci_sad0 = pdev;
2798 break;
2799 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_CBO_SAD1:
2800 pvt->pci_sad1 = pdev;
2801 break;
2802 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0:
2803 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1:
2804 pvt->pci_ha = pdev;
2805 break;
2806 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TA:
2807 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TA:
2808 pvt->pci_ta = pdev;
2809 break;
2810 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TM:
2811 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TM:
2812 pvt->pci_ras = pdev;
2813 break;
2814 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD0:
2815 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD1:
2816 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD2:
2817 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD3:
2818 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD0:
2819 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD1:
2820 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD2:
2821 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD3:
2822 {
2823 int id = TAD_DEV_TO_CHAN(pdev->device);
2824 pvt->pci_tad[id] = pdev;
2825 saw_chan_mask |= 1 << id;
2826 }
2827 break;
2828 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO0:
2829 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO1:
2830 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO2:
2831 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO3:
2832 if (!pvt->pci_ddrio)
2833 pvt->pci_ddrio = pdev;
2834 break;
2835 default:
2836 break;
2837 }
2838
2839 edac_dbg(0, "Associated PCI %02x.%02d.%d with dev = %p\n",
2840 sbridge_dev->bus,
2841 PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn),
2842 pdev);
2843 }
2844
2845 /* Check if everything were registered */
2846 if (!pvt->pci_sad0 || !pvt->pci_ha || !pvt->pci_sad1 ||
2847 !pvt->pci_ras || !pvt->pci_ta || !pvt->info.pci_vtd)
2848 goto enodev;
2849
2850 if (saw_chan_mask != 0x0f && /* -EN/-EX */
2851 saw_chan_mask != 0x03) /* -EP */
2852 goto enodev;
2853 return 0;
2854
2855 enodev:
2856 sbridge_printk(KERN_ERR, "Some needed devices are missing\n");
2857 return -ENODEV;
2858 }
2859
broadwell_mci_bind_devs(struct mem_ctl_info * mci,struct sbridge_dev * sbridge_dev)2860 static int broadwell_mci_bind_devs(struct mem_ctl_info *mci,
2861 struct sbridge_dev *sbridge_dev)
2862 {
2863 struct sbridge_pvt *pvt = mci->pvt_info;
2864 struct pci_dev *pdev;
2865 u8 saw_chan_mask = 0;
2866 int i;
2867
2868 /* there's only one device per system; not tied to any bus */
2869 if (pvt->info.pci_vtd == NULL)
2870 /* result will be checked later */
2871 pvt->info.pci_vtd = pci_get_device(PCI_VENDOR_ID_INTEL,
2872 PCI_DEVICE_ID_INTEL_BROADWELL_IMC_VTD_MISC,
2873 NULL);
2874
2875 for (i = 0; i < sbridge_dev->n_devs; i++) {
2876 pdev = sbridge_dev->pdev[i];
2877 if (!pdev)
2878 continue;
2879
2880 switch (pdev->device) {
2881 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_CBO_SAD0:
2882 pvt->pci_sad0 = pdev;
2883 break;
2884 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_CBO_SAD1:
2885 pvt->pci_sad1 = pdev;
2886 break;
2887 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0:
2888 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1:
2889 pvt->pci_ha = pdev;
2890 break;
2891 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TA:
2892 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TA:
2893 pvt->pci_ta = pdev;
2894 break;
2895 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TM:
2896 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TM:
2897 pvt->pci_ras = pdev;
2898 break;
2899 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD0:
2900 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD1:
2901 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD2:
2902 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD3:
2903 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD0:
2904 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD1:
2905 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD2:
2906 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD3:
2907 {
2908 int id = TAD_DEV_TO_CHAN(pdev->device);
2909 pvt->pci_tad[id] = pdev;
2910 saw_chan_mask |= 1 << id;
2911 }
2912 break;
2913 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_DDRIO0:
2914 pvt->pci_ddrio = pdev;
2915 break;
2916 default:
2917 break;
2918 }
2919
2920 edac_dbg(0, "Associated PCI %02x.%02d.%d with dev = %p\n",
2921 sbridge_dev->bus,
2922 PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn),
2923 pdev);
2924 }
2925
2926 /* Check if everything were registered */
2927 if (!pvt->pci_sad0 || !pvt->pci_ha || !pvt->pci_sad1 ||
2928 !pvt->pci_ras || !pvt->pci_ta || !pvt->info.pci_vtd)
2929 goto enodev;
2930
2931 if (saw_chan_mask != 0x0f && /* -EN/-EX */
2932 saw_chan_mask != 0x03) /* -EP */
2933 goto enodev;
2934 return 0;
2935
2936 enodev:
2937 sbridge_printk(KERN_ERR, "Some needed devices are missing\n");
2938 return -ENODEV;
2939 }
2940
knl_mci_bind_devs(struct mem_ctl_info * mci,struct sbridge_dev * sbridge_dev)2941 static int knl_mci_bind_devs(struct mem_ctl_info *mci,
2942 struct sbridge_dev *sbridge_dev)
2943 {
2944 struct sbridge_pvt *pvt = mci->pvt_info;
2945 struct pci_dev *pdev;
2946 int dev, func;
2947
2948 int i;
2949 int devidx;
2950
2951 for (i = 0; i < sbridge_dev->n_devs; i++) {
2952 pdev = sbridge_dev->pdev[i];
2953 if (!pdev)
2954 continue;
2955
2956 /* Extract PCI device and function. */
2957 dev = (pdev->devfn >> 3) & 0x1f;
2958 func = pdev->devfn & 0x7;
2959
2960 switch (pdev->device) {
2961 case PCI_DEVICE_ID_INTEL_KNL_IMC_MC:
2962 if (dev == 8)
2963 pvt->knl.pci_mc0 = pdev;
2964 else if (dev == 9)
2965 pvt->knl.pci_mc1 = pdev;
2966 else {
2967 sbridge_printk(KERN_ERR,
2968 "Memory controller in unexpected place! (dev %d, fn %d)\n",
2969 dev, func);
2970 continue;
2971 }
2972 break;
2973
2974 case PCI_DEVICE_ID_INTEL_KNL_IMC_SAD0:
2975 pvt->pci_sad0 = pdev;
2976 break;
2977
2978 case PCI_DEVICE_ID_INTEL_KNL_IMC_SAD1:
2979 pvt->pci_sad1 = pdev;
2980 break;
2981
2982 case PCI_DEVICE_ID_INTEL_KNL_IMC_CHA:
2983 /* There are one of these per tile, and range from
2984 * 1.14.0 to 1.18.5.
2985 */
2986 devidx = ((dev-14)*8)+func;
2987
2988 if (devidx < 0 || devidx >= KNL_MAX_CHAS) {
2989 sbridge_printk(KERN_ERR,
2990 "Caching and Home Agent in unexpected place! (dev %d, fn %d)\n",
2991 dev, func);
2992 continue;
2993 }
2994
2995 WARN_ON(pvt->knl.pci_cha[devidx] != NULL);
2996
2997 pvt->knl.pci_cha[devidx] = pdev;
2998 break;
2999
3000 case PCI_DEVICE_ID_INTEL_KNL_IMC_CHAN:
3001 devidx = -1;
3002
3003 /*
3004 * MC0 channels 0-2 are device 9 function 2-4,
3005 * MC1 channels 3-5 are device 8 function 2-4.
3006 */
3007
3008 if (dev == 9)
3009 devidx = func-2;
3010 else if (dev == 8)
3011 devidx = 3 + (func-2);
3012
3013 if (devidx < 0 || devidx >= KNL_MAX_CHANNELS) {
3014 sbridge_printk(KERN_ERR,
3015 "DRAM Channel Registers in unexpected place! (dev %d, fn %d)\n",
3016 dev, func);
3017 continue;
3018 }
3019
3020 WARN_ON(pvt->knl.pci_channel[devidx] != NULL);
3021 pvt->knl.pci_channel[devidx] = pdev;
3022 break;
3023
3024 case PCI_DEVICE_ID_INTEL_KNL_IMC_TOLHM:
3025 pvt->knl.pci_mc_info = pdev;
3026 break;
3027
3028 case PCI_DEVICE_ID_INTEL_KNL_IMC_TA:
3029 pvt->pci_ta = pdev;
3030 break;
3031
3032 default:
3033 sbridge_printk(KERN_ERR, "Unexpected device %d\n",
3034 pdev->device);
3035 break;
3036 }
3037 }
3038
3039 if (!pvt->knl.pci_mc0 || !pvt->knl.pci_mc1 ||
3040 !pvt->pci_sad0 || !pvt->pci_sad1 ||
3041 !pvt->pci_ta) {
3042 goto enodev;
3043 }
3044
3045 for (i = 0; i < KNL_MAX_CHANNELS; i++) {
3046 if (!pvt->knl.pci_channel[i]) {
3047 sbridge_printk(KERN_ERR, "Missing channel %d\n", i);
3048 goto enodev;
3049 }
3050 }
3051
3052 for (i = 0; i < KNL_MAX_CHAS; i++) {
3053 if (!pvt->knl.pci_cha[i]) {
3054 sbridge_printk(KERN_ERR, "Missing CHA %d\n", i);
3055 goto enodev;
3056 }
3057 }
3058
3059 return 0;
3060
3061 enodev:
3062 sbridge_printk(KERN_ERR, "Some needed devices are missing\n");
3063 return -ENODEV;
3064 }
3065
3066 /****************************************************************************
3067 Error check routines
3068 ****************************************************************************/
3069
3070 /*
3071 * While Sandy Bridge has error count registers, SMI BIOS read values from
3072 * and resets the counters. So, they are not reliable for the OS to read
3073 * from them. So, we have no option but to just trust on whatever MCE is
3074 * telling us about the errors.
3075 */
sbridge_mce_output_error(struct mem_ctl_info * mci,const struct mce * m)3076 static void sbridge_mce_output_error(struct mem_ctl_info *mci,
3077 const struct mce *m)
3078 {
3079 struct mem_ctl_info *new_mci;
3080 struct sbridge_pvt *pvt = mci->pvt_info;
3081 enum hw_event_mc_err_type tp_event;
3082 char *optype, msg[256], msg_full[512];
3083 bool ripv = GET_BITFIELD(m->mcgstatus, 0, 0);
3084 bool overflow = GET_BITFIELD(m->status, 62, 62);
3085 bool uncorrected_error = GET_BITFIELD(m->status, 61, 61);
3086 bool recoverable;
3087 u32 core_err_cnt = GET_BITFIELD(m->status, 38, 52);
3088 u32 mscod = GET_BITFIELD(m->status, 16, 31);
3089 u32 errcode = GET_BITFIELD(m->status, 0, 15);
3090 u32 channel = GET_BITFIELD(m->status, 0, 3);
3091 u32 optypenum = GET_BITFIELD(m->status, 4, 6);
3092 /*
3093 * Bits 5-0 of MCi_MISC give the least significant bit that is valid.
3094 * A value 6 is for cache line aligned address, a value 12 is for page
3095 * aligned address reported by patrol scrubber.
3096 */
3097 u32 lsb = GET_BITFIELD(m->misc, 0, 5);
3098 long channel_mask, first_channel;
3099 u8 rank = 0xff, socket, ha;
3100 int rc, dimm;
3101 char *area_type = "DRAM";
3102
3103 if (pvt->info.type != SANDY_BRIDGE)
3104 recoverable = true;
3105 else
3106 recoverable = GET_BITFIELD(m->status, 56, 56);
3107
3108 if (uncorrected_error) {
3109 core_err_cnt = 1;
3110 if (ripv) {
3111 tp_event = HW_EVENT_ERR_UNCORRECTED;
3112 } else {
3113 tp_event = HW_EVENT_ERR_FATAL;
3114 }
3115 } else {
3116 tp_event = HW_EVENT_ERR_CORRECTED;
3117 }
3118
3119 /*
3120 * According with Table 15-9 of the Intel Architecture spec vol 3A,
3121 * memory errors should fit in this mask:
3122 * 000f 0000 1mmm cccc (binary)
3123 * where:
3124 * f = Correction Report Filtering Bit. If 1, subsequent errors
3125 * won't be shown
3126 * mmm = error type
3127 * cccc = channel
3128 * If the mask doesn't match, report an error to the parsing logic
3129 */
3130 switch (optypenum) {
3131 case 0:
3132 optype = "generic undef request error";
3133 break;
3134 case 1:
3135 optype = "memory read error";
3136 break;
3137 case 2:
3138 optype = "memory write error";
3139 break;
3140 case 3:
3141 optype = "addr/cmd error";
3142 break;
3143 case 4:
3144 optype = "memory scrubbing error";
3145 break;
3146 default:
3147 optype = "reserved";
3148 break;
3149 }
3150
3151 if (pvt->info.type == KNIGHTS_LANDING) {
3152 if (channel == 14) {
3153 edac_dbg(0, "%s%s err_code:%04x:%04x EDRAM bank %d\n",
3154 overflow ? " OVERFLOW" : "",
3155 (uncorrected_error && recoverable)
3156 ? " recoverable" : "",
3157 mscod, errcode,
3158 m->bank);
3159 } else {
3160 char A = *("A");
3161
3162 /*
3163 * Reported channel is in range 0-2, so we can't map it
3164 * back to mc. To figure out mc we check machine check
3165 * bank register that reported this error.
3166 * bank15 means mc0 and bank16 means mc1.
3167 */
3168 channel = knl_channel_remap(m->bank == 16, channel);
3169 channel_mask = 1 << channel;
3170
3171 snprintf(msg, sizeof(msg),
3172 "%s%s err_code:%04x:%04x channel:%d (DIMM_%c)",
3173 overflow ? " OVERFLOW" : "",
3174 (uncorrected_error && recoverable)
3175 ? " recoverable" : " ",
3176 mscod, errcode, channel, A + channel);
3177 edac_mc_handle_error(tp_event, mci, core_err_cnt,
3178 m->addr >> PAGE_SHIFT, m->addr & ~PAGE_MASK, 0,
3179 channel, 0, -1,
3180 optype, msg);
3181 }
3182 return;
3183 } else if (lsb < 12) {
3184 rc = get_memory_error_data(mci, m->addr, &socket, &ha,
3185 &channel_mask, &rank,
3186 &area_type, msg);
3187 } else {
3188 rc = get_memory_error_data_from_mce(mci, m, &socket, &ha,
3189 &channel_mask, msg);
3190 }
3191
3192 if (rc < 0)
3193 goto err_parsing;
3194 new_mci = get_mci_for_node_id(socket, ha);
3195 if (!new_mci) {
3196 strcpy(msg, "Error: socket got corrupted!");
3197 goto err_parsing;
3198 }
3199 mci = new_mci;
3200 pvt = mci->pvt_info;
3201
3202 first_channel = find_first_bit(&channel_mask, NUM_CHANNELS);
3203
3204 if (rank == 0xff)
3205 dimm = -1;
3206 else if (rank < 4)
3207 dimm = 0;
3208 else if (rank < 8)
3209 dimm = 1;
3210 else
3211 dimm = 2;
3212
3213 /*
3214 * FIXME: On some memory configurations (mirror, lockstep), the
3215 * Memory Controller can't point the error to a single DIMM. The
3216 * EDAC core should be handling the channel mask, in order to point
3217 * to the group of dimm's where the error may be happening.
3218 */
3219 if (!pvt->is_lockstep && !pvt->is_cur_addr_mirrored && !pvt->is_close_pg)
3220 channel = first_channel;
3221 snprintf(msg_full, sizeof(msg_full),
3222 "%s%s area:%s err_code:%04x:%04x socket:%d ha:%d channel_mask:%ld rank:%d %s",
3223 overflow ? " OVERFLOW" : "",
3224 (uncorrected_error && recoverable) ? " recoverable" : "",
3225 area_type,
3226 mscod, errcode,
3227 socket, ha,
3228 channel_mask,
3229 rank, msg);
3230
3231 edac_dbg(0, "%s\n", msg_full);
3232
3233 /* FIXME: need support for channel mask */
3234
3235 if (channel == CHANNEL_UNSPECIFIED)
3236 channel = -1;
3237
3238 /* Call the helper to output message */
3239 edac_mc_handle_error(tp_event, mci, core_err_cnt,
3240 m->addr >> PAGE_SHIFT, m->addr & ~PAGE_MASK, 0,
3241 channel, dimm, -1,
3242 optype, msg_full);
3243 return;
3244 err_parsing:
3245 edac_mc_handle_error(tp_event, mci, core_err_cnt, 0, 0, 0,
3246 -1, -1, -1,
3247 msg, "");
3248
3249 }
3250
3251 /*
3252 * Check that logging is enabled and that this is the right type
3253 * of error for us to handle.
3254 */
sbridge_mce_check_error(struct notifier_block * nb,unsigned long val,void * data)3255 static int sbridge_mce_check_error(struct notifier_block *nb, unsigned long val,
3256 void *data)
3257 {
3258 struct mce *mce = (struct mce *)data;
3259 struct mem_ctl_info *mci;
3260 char *type;
3261
3262 if (mce->kflags & MCE_HANDLED_CEC)
3263 return NOTIFY_DONE;
3264
3265 /*
3266 * Just let mcelog handle it if the error is
3267 * outside the memory controller. A memory error
3268 * is indicated by bit 7 = 1 and bits = 8-11,13-15 = 0.
3269 * bit 12 has an special meaning.
3270 */
3271 if ((mce->status & 0xefff) >> 7 != 1)
3272 return NOTIFY_DONE;
3273
3274 /* Check ADDRV bit in STATUS */
3275 if (!GET_BITFIELD(mce->status, 58, 58))
3276 return NOTIFY_DONE;
3277
3278 /* Check MISCV bit in STATUS */
3279 if (!GET_BITFIELD(mce->status, 59, 59))
3280 return NOTIFY_DONE;
3281
3282 /* Check address type in MISC (physical address only) */
3283 if (GET_BITFIELD(mce->misc, 6, 8) != 2)
3284 return NOTIFY_DONE;
3285
3286 mci = get_mci_for_node_id(mce->socketid, IMC0);
3287 if (!mci)
3288 return NOTIFY_DONE;
3289
3290 if (mce->mcgstatus & MCG_STATUS_MCIP)
3291 type = "Exception";
3292 else
3293 type = "Event";
3294
3295 sbridge_mc_printk(mci, KERN_DEBUG, "HANDLING MCE MEMORY ERROR\n");
3296
3297 sbridge_mc_printk(mci, KERN_DEBUG, "CPU %d: Machine Check %s: %Lx "
3298 "Bank %d: %016Lx\n", mce->extcpu, type,
3299 mce->mcgstatus, mce->bank, mce->status);
3300 sbridge_mc_printk(mci, KERN_DEBUG, "TSC %llx ", mce->tsc);
3301 sbridge_mc_printk(mci, KERN_DEBUG, "ADDR %llx ", mce->addr);
3302 sbridge_mc_printk(mci, KERN_DEBUG, "MISC %llx ", mce->misc);
3303
3304 sbridge_mc_printk(mci, KERN_DEBUG, "PROCESSOR %u:%x TIME %llu SOCKET "
3305 "%u APIC %x\n", mce->cpuvendor, mce->cpuid,
3306 mce->time, mce->socketid, mce->apicid);
3307
3308 sbridge_mce_output_error(mci, mce);
3309
3310 /* Advice mcelog that the error were handled */
3311 mce->kflags |= MCE_HANDLED_EDAC;
3312 return NOTIFY_OK;
3313 }
3314
3315 static struct notifier_block sbridge_mce_dec = {
3316 .notifier_call = sbridge_mce_check_error,
3317 .priority = MCE_PRIO_EDAC,
3318 };
3319
3320 /****************************************************************************
3321 EDAC register/unregister logic
3322 ****************************************************************************/
3323
sbridge_unregister_mci(struct sbridge_dev * sbridge_dev)3324 static void sbridge_unregister_mci(struct sbridge_dev *sbridge_dev)
3325 {
3326 struct mem_ctl_info *mci = sbridge_dev->mci;
3327
3328 if (unlikely(!mci || !mci->pvt_info)) {
3329 edac_dbg(0, "MC: dev = %p\n", &sbridge_dev->pdev[0]->dev);
3330
3331 sbridge_printk(KERN_ERR, "Couldn't find mci handler\n");
3332 return;
3333 }
3334
3335 edac_dbg(0, "MC: mci = %p, dev = %p\n",
3336 mci, &sbridge_dev->pdev[0]->dev);
3337
3338 /* Remove MC sysfs nodes */
3339 edac_mc_del_mc(mci->pdev);
3340
3341 edac_dbg(1, "%s: free mci struct\n", mci->ctl_name);
3342 kfree(mci->ctl_name);
3343 edac_mc_free(mci);
3344 sbridge_dev->mci = NULL;
3345 }
3346
sbridge_register_mci(struct sbridge_dev * sbridge_dev,enum type type)3347 static int sbridge_register_mci(struct sbridge_dev *sbridge_dev, enum type type)
3348 {
3349 struct mem_ctl_info *mci;
3350 struct edac_mc_layer layers[2];
3351 struct sbridge_pvt *pvt;
3352 struct pci_dev *pdev = sbridge_dev->pdev[0];
3353 int rc;
3354
3355 /* allocate a new MC control structure */
3356 layers[0].type = EDAC_MC_LAYER_CHANNEL;
3357 layers[0].size = type == KNIGHTS_LANDING ?
3358 KNL_MAX_CHANNELS : NUM_CHANNELS;
3359 layers[0].is_virt_csrow = false;
3360 layers[1].type = EDAC_MC_LAYER_SLOT;
3361 layers[1].size = type == KNIGHTS_LANDING ? 1 : MAX_DIMMS;
3362 layers[1].is_virt_csrow = true;
3363 mci = edac_mc_alloc(sbridge_dev->mc, ARRAY_SIZE(layers), layers,
3364 sizeof(*pvt));
3365
3366 if (unlikely(!mci))
3367 return -ENOMEM;
3368
3369 edac_dbg(0, "MC: mci = %p, dev = %p\n",
3370 mci, &pdev->dev);
3371
3372 pvt = mci->pvt_info;
3373 memset(pvt, 0, sizeof(*pvt));
3374
3375 /* Associate sbridge_dev and mci for future usage */
3376 pvt->sbridge_dev = sbridge_dev;
3377 sbridge_dev->mci = mci;
3378
3379 mci->mtype_cap = type == KNIGHTS_LANDING ?
3380 MEM_FLAG_DDR4 : MEM_FLAG_DDR3;
3381 mci->edac_ctl_cap = EDAC_FLAG_NONE;
3382 mci->edac_cap = EDAC_FLAG_NONE;
3383 mci->mod_name = EDAC_MOD_STR;
3384 mci->dev_name = pci_name(pdev);
3385 mci->ctl_page_to_phys = NULL;
3386
3387 pvt->info.type = type;
3388 switch (type) {
3389 case IVY_BRIDGE:
3390 pvt->info.rankcfgr = IB_RANK_CFG_A;
3391 pvt->info.get_tolm = ibridge_get_tolm;
3392 pvt->info.get_tohm = ibridge_get_tohm;
3393 pvt->info.dram_rule = ibridge_dram_rule;
3394 pvt->info.get_memory_type = get_memory_type;
3395 pvt->info.get_node_id = get_node_id;
3396 pvt->info.get_ha = ibridge_get_ha;
3397 pvt->info.rir_limit = rir_limit;
3398 pvt->info.sad_limit = sad_limit;
3399 pvt->info.interleave_mode = interleave_mode;
3400 pvt->info.dram_attr = dram_attr;
3401 pvt->info.max_sad = ARRAY_SIZE(ibridge_dram_rule);
3402 pvt->info.interleave_list = ibridge_interleave_list;
3403 pvt->info.interleave_pkg = ibridge_interleave_pkg;
3404 pvt->info.get_width = ibridge_get_width;
3405
3406 /* Store pci devices at mci for faster access */
3407 rc = ibridge_mci_bind_devs(mci, sbridge_dev);
3408 if (unlikely(rc < 0))
3409 goto fail0;
3410 get_source_id(mci);
3411 mci->ctl_name = kasprintf(GFP_KERNEL, "Ivy Bridge SrcID#%d_Ha#%d",
3412 pvt->sbridge_dev->source_id, pvt->sbridge_dev->dom);
3413 break;
3414 case SANDY_BRIDGE:
3415 pvt->info.rankcfgr = SB_RANK_CFG_A;
3416 pvt->info.get_tolm = sbridge_get_tolm;
3417 pvt->info.get_tohm = sbridge_get_tohm;
3418 pvt->info.dram_rule = sbridge_dram_rule;
3419 pvt->info.get_memory_type = get_memory_type;
3420 pvt->info.get_node_id = get_node_id;
3421 pvt->info.get_ha = sbridge_get_ha;
3422 pvt->info.rir_limit = rir_limit;
3423 pvt->info.sad_limit = sad_limit;
3424 pvt->info.interleave_mode = interleave_mode;
3425 pvt->info.dram_attr = dram_attr;
3426 pvt->info.max_sad = ARRAY_SIZE(sbridge_dram_rule);
3427 pvt->info.interleave_list = sbridge_interleave_list;
3428 pvt->info.interleave_pkg = sbridge_interleave_pkg;
3429 pvt->info.get_width = sbridge_get_width;
3430
3431 /* Store pci devices at mci for faster access */
3432 rc = sbridge_mci_bind_devs(mci, sbridge_dev);
3433 if (unlikely(rc < 0))
3434 goto fail0;
3435 get_source_id(mci);
3436 mci->ctl_name = kasprintf(GFP_KERNEL, "Sandy Bridge SrcID#%d_Ha#%d",
3437 pvt->sbridge_dev->source_id, pvt->sbridge_dev->dom);
3438 break;
3439 case HASWELL:
3440 /* rankcfgr isn't used */
3441 pvt->info.get_tolm = haswell_get_tolm;
3442 pvt->info.get_tohm = haswell_get_tohm;
3443 pvt->info.dram_rule = ibridge_dram_rule;
3444 pvt->info.get_memory_type = haswell_get_memory_type;
3445 pvt->info.get_node_id = haswell_get_node_id;
3446 pvt->info.get_ha = ibridge_get_ha;
3447 pvt->info.rir_limit = haswell_rir_limit;
3448 pvt->info.sad_limit = sad_limit;
3449 pvt->info.interleave_mode = interleave_mode;
3450 pvt->info.dram_attr = dram_attr;
3451 pvt->info.max_sad = ARRAY_SIZE(ibridge_dram_rule);
3452 pvt->info.interleave_list = ibridge_interleave_list;
3453 pvt->info.interleave_pkg = ibridge_interleave_pkg;
3454 pvt->info.get_width = ibridge_get_width;
3455
3456 /* Store pci devices at mci for faster access */
3457 rc = haswell_mci_bind_devs(mci, sbridge_dev);
3458 if (unlikely(rc < 0))
3459 goto fail0;
3460 get_source_id(mci);
3461 mci->ctl_name = kasprintf(GFP_KERNEL, "Haswell SrcID#%d_Ha#%d",
3462 pvt->sbridge_dev->source_id, pvt->sbridge_dev->dom);
3463 break;
3464 case BROADWELL:
3465 /* rankcfgr isn't used */
3466 pvt->info.get_tolm = haswell_get_tolm;
3467 pvt->info.get_tohm = haswell_get_tohm;
3468 pvt->info.dram_rule = ibridge_dram_rule;
3469 pvt->info.get_memory_type = haswell_get_memory_type;
3470 pvt->info.get_node_id = haswell_get_node_id;
3471 pvt->info.get_ha = ibridge_get_ha;
3472 pvt->info.rir_limit = haswell_rir_limit;
3473 pvt->info.sad_limit = sad_limit;
3474 pvt->info.interleave_mode = interleave_mode;
3475 pvt->info.dram_attr = dram_attr;
3476 pvt->info.max_sad = ARRAY_SIZE(ibridge_dram_rule);
3477 pvt->info.interleave_list = ibridge_interleave_list;
3478 pvt->info.interleave_pkg = ibridge_interleave_pkg;
3479 pvt->info.get_width = broadwell_get_width;
3480
3481 /* Store pci devices at mci for faster access */
3482 rc = broadwell_mci_bind_devs(mci, sbridge_dev);
3483 if (unlikely(rc < 0))
3484 goto fail0;
3485 get_source_id(mci);
3486 mci->ctl_name = kasprintf(GFP_KERNEL, "Broadwell SrcID#%d_Ha#%d",
3487 pvt->sbridge_dev->source_id, pvt->sbridge_dev->dom);
3488 break;
3489 case KNIGHTS_LANDING:
3490 /* pvt->info.rankcfgr == ??? */
3491 pvt->info.get_tolm = knl_get_tolm;
3492 pvt->info.get_tohm = knl_get_tohm;
3493 pvt->info.dram_rule = knl_dram_rule;
3494 pvt->info.get_memory_type = knl_get_memory_type;
3495 pvt->info.get_node_id = knl_get_node_id;
3496 pvt->info.get_ha = knl_get_ha;
3497 pvt->info.rir_limit = NULL;
3498 pvt->info.sad_limit = knl_sad_limit;
3499 pvt->info.interleave_mode = knl_interleave_mode;
3500 pvt->info.dram_attr = dram_attr_knl;
3501 pvt->info.max_sad = ARRAY_SIZE(knl_dram_rule);
3502 pvt->info.interleave_list = knl_interleave_list;
3503 pvt->info.interleave_pkg = ibridge_interleave_pkg;
3504 pvt->info.get_width = knl_get_width;
3505
3506 rc = knl_mci_bind_devs(mci, sbridge_dev);
3507 if (unlikely(rc < 0))
3508 goto fail0;
3509 get_source_id(mci);
3510 mci->ctl_name = kasprintf(GFP_KERNEL, "Knights Landing SrcID#%d_Ha#%d",
3511 pvt->sbridge_dev->source_id, pvt->sbridge_dev->dom);
3512 break;
3513 }
3514
3515 if (!mci->ctl_name) {
3516 rc = -ENOMEM;
3517 goto fail0;
3518 }
3519
3520 /* Get dimm basic config and the memory layout */
3521 rc = get_dimm_config(mci);
3522 if (rc < 0) {
3523 edac_dbg(0, "MC: failed to get_dimm_config()\n");
3524 goto fail;
3525 }
3526 get_memory_layout(mci);
3527
3528 /* record ptr to the generic device */
3529 mci->pdev = &pdev->dev;
3530
3531 /* add this new MC control structure to EDAC's list of MCs */
3532 if (unlikely(edac_mc_add_mc(mci))) {
3533 edac_dbg(0, "MC: failed edac_mc_add_mc()\n");
3534 rc = -EINVAL;
3535 goto fail;
3536 }
3537
3538 return 0;
3539
3540 fail:
3541 kfree(mci->ctl_name);
3542 fail0:
3543 edac_mc_free(mci);
3544 sbridge_dev->mci = NULL;
3545 return rc;
3546 }
3547
3548 static const struct x86_cpu_id sbridge_cpuids[] = {
3549 X86_MATCH_INTEL_FAM6_MODEL(SANDYBRIDGE_X, &pci_dev_descr_sbridge_table),
3550 X86_MATCH_INTEL_FAM6_MODEL(IVYBRIDGE_X, &pci_dev_descr_ibridge_table),
3551 X86_MATCH_INTEL_FAM6_MODEL(HASWELL_X, &pci_dev_descr_haswell_table),
3552 X86_MATCH_INTEL_FAM6_MODEL(BROADWELL_X, &pci_dev_descr_broadwell_table),
3553 X86_MATCH_INTEL_FAM6_MODEL(BROADWELL_D, &pci_dev_descr_broadwell_table),
3554 X86_MATCH_INTEL_FAM6_MODEL(XEON_PHI_KNL, &pci_dev_descr_knl_table),
3555 X86_MATCH_INTEL_FAM6_MODEL(XEON_PHI_KNM, &pci_dev_descr_knl_table),
3556 { }
3557 };
3558 MODULE_DEVICE_TABLE(x86cpu, sbridge_cpuids);
3559
3560 /*
3561 * sbridge_probe Get all devices and register memory controllers
3562 * present.
3563 * return:
3564 * 0 for FOUND a device
3565 * < 0 for error code
3566 */
3567
sbridge_probe(const struct x86_cpu_id * id)3568 static int sbridge_probe(const struct x86_cpu_id *id)
3569 {
3570 int rc;
3571 u8 mc, num_mc = 0;
3572 struct sbridge_dev *sbridge_dev;
3573 struct pci_id_table *ptable = (struct pci_id_table *)id->driver_data;
3574
3575 /* get the pci devices we want to reserve for our use */
3576 rc = sbridge_get_all_devices(&num_mc, ptable);
3577
3578 if (unlikely(rc < 0)) {
3579 edac_dbg(0, "couldn't get all devices\n");
3580 goto fail0;
3581 }
3582
3583 mc = 0;
3584
3585 list_for_each_entry(sbridge_dev, &sbridge_edac_list, list) {
3586 edac_dbg(0, "Registering MC#%d (%d of %d)\n",
3587 mc, mc + 1, num_mc);
3588
3589 sbridge_dev->mc = mc++;
3590 rc = sbridge_register_mci(sbridge_dev, ptable->type);
3591 if (unlikely(rc < 0))
3592 goto fail1;
3593 }
3594
3595 sbridge_printk(KERN_INFO, "%s\n", SBRIDGE_REVISION);
3596
3597 return 0;
3598
3599 fail1:
3600 list_for_each_entry(sbridge_dev, &sbridge_edac_list, list)
3601 sbridge_unregister_mci(sbridge_dev);
3602
3603 sbridge_put_all_devices();
3604 fail0:
3605 return rc;
3606 }
3607
3608 /*
3609 * sbridge_remove cleanup
3610 *
3611 */
sbridge_remove(void)3612 static void sbridge_remove(void)
3613 {
3614 struct sbridge_dev *sbridge_dev;
3615
3616 edac_dbg(0, "\n");
3617
3618 list_for_each_entry(sbridge_dev, &sbridge_edac_list, list)
3619 sbridge_unregister_mci(sbridge_dev);
3620
3621 /* Release PCI resources */
3622 sbridge_put_all_devices();
3623 }
3624
3625 /*
3626 * sbridge_init Module entry function
3627 * Try to initialize this module for its devices
3628 */
sbridge_init(void)3629 static int __init sbridge_init(void)
3630 {
3631 const struct x86_cpu_id *id;
3632 const char *owner;
3633 int rc;
3634
3635 edac_dbg(2, "\n");
3636
3637 if (ghes_get_devices())
3638 return -EBUSY;
3639
3640 owner = edac_get_owner();
3641 if (owner && strncmp(owner, EDAC_MOD_STR, sizeof(EDAC_MOD_STR)))
3642 return -EBUSY;
3643
3644 if (cpu_feature_enabled(X86_FEATURE_HYPERVISOR))
3645 return -ENODEV;
3646
3647 id = x86_match_cpu(sbridge_cpuids);
3648 if (!id)
3649 return -ENODEV;
3650
3651 /* Ensure that the OPSTATE is set correctly for POLL or NMI */
3652 opstate_init();
3653
3654 rc = sbridge_probe(id);
3655
3656 if (rc >= 0) {
3657 mce_register_decode_chain(&sbridge_mce_dec);
3658 return 0;
3659 }
3660
3661 sbridge_printk(KERN_ERR, "Failed to register device with error %d.\n",
3662 rc);
3663
3664 return rc;
3665 }
3666
3667 /*
3668 * sbridge_exit() Module exit function
3669 * Unregister the driver
3670 */
sbridge_exit(void)3671 static void __exit sbridge_exit(void)
3672 {
3673 edac_dbg(2, "\n");
3674 sbridge_remove();
3675 mce_unregister_decode_chain(&sbridge_mce_dec);
3676 }
3677
3678 module_init(sbridge_init);
3679 module_exit(sbridge_exit);
3680
3681 module_param(edac_op_state, int, 0444);
3682 MODULE_PARM_DESC(edac_op_state, "EDAC Error Reporting state: 0=Poll,1=NMI");
3683
3684 MODULE_LICENSE("GPL");
3685 MODULE_AUTHOR("Mauro Carvalho Chehab");
3686 MODULE_AUTHOR("Red Hat Inc. (https://www.redhat.com)");
3687 MODULE_DESCRIPTION("MC Driver for Intel Sandy Bridge and Ivy Bridge memory controllers - "
3688 SBRIDGE_REVISION);
3689