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 static char sb_msg[256];
33 static char sb_msg_full[512];
34
35 /*
36 * Alter this version for the module when modifications are made
37 */
38 #define SBRIDGE_REVISION " Ver: 1.1.2 "
39 #define EDAC_MOD_STR "sb_edac"
40
41 /*
42 * Debug macros
43 */
44 #define sbridge_printk(level, fmt, arg...) \
45 edac_printk(level, "sbridge", fmt, ##arg)
46
47 #define sbridge_mc_printk(mci, level, fmt, arg...) \
48 edac_mc_chipset_printk(mci, level, "sbridge", fmt, ##arg)
49
50 /*
51 * Get a bit field at register value <v>, from bit <lo> to bit <hi>
52 */
53 #define GET_BITFIELD(v, lo, hi) \
54 (((v) & GENMASK_ULL(hi, lo)) >> (lo))
55
56 /* Devices 12 Function 6, Offsets 0x80 to 0xcc */
57 static const u32 sbridge_dram_rule[] = {
58 0x80, 0x88, 0x90, 0x98, 0xa0,
59 0xa8, 0xb0, 0xb8, 0xc0, 0xc8,
60 };
61
62 static const u32 ibridge_dram_rule[] = {
63 0x60, 0x68, 0x70, 0x78, 0x80,
64 0x88, 0x90, 0x98, 0xa0, 0xa8,
65 0xb0, 0xb8, 0xc0, 0xc8, 0xd0,
66 0xd8, 0xe0, 0xe8, 0xf0, 0xf8,
67 };
68
69 static const u32 knl_dram_rule[] = {
70 0x60, 0x68, 0x70, 0x78, 0x80, /* 0-4 */
71 0x88, 0x90, 0x98, 0xa0, 0xa8, /* 5-9 */
72 0xb0, 0xb8, 0xc0, 0xc8, 0xd0, /* 10-14 */
73 0xd8, 0xe0, 0xe8, 0xf0, 0xf8, /* 15-19 */
74 0x100, 0x108, 0x110, 0x118, /* 20-23 */
75 };
76
77 #define DRAM_RULE_ENABLE(reg) GET_BITFIELD(reg, 0, 0)
78 #define A7MODE(reg) GET_BITFIELD(reg, 26, 26)
79
show_dram_attr(u32 attr)80 static char *show_dram_attr(u32 attr)
81 {
82 switch (attr) {
83 case 0:
84 return "DRAM";
85 case 1:
86 return "MMCFG";
87 case 2:
88 return "NXM";
89 default:
90 return "unknown";
91 }
92 }
93
94 static const u32 sbridge_interleave_list[] = {
95 0x84, 0x8c, 0x94, 0x9c, 0xa4,
96 0xac, 0xb4, 0xbc, 0xc4, 0xcc,
97 };
98
99 static const u32 ibridge_interleave_list[] = {
100 0x64, 0x6c, 0x74, 0x7c, 0x84,
101 0x8c, 0x94, 0x9c, 0xa4, 0xac,
102 0xb4, 0xbc, 0xc4, 0xcc, 0xd4,
103 0xdc, 0xe4, 0xec, 0xf4, 0xfc,
104 };
105
106 static const u32 knl_interleave_list[] = {
107 0x64, 0x6c, 0x74, 0x7c, 0x84, /* 0-4 */
108 0x8c, 0x94, 0x9c, 0xa4, 0xac, /* 5-9 */
109 0xb4, 0xbc, 0xc4, 0xcc, 0xd4, /* 10-14 */
110 0xdc, 0xe4, 0xec, 0xf4, 0xfc, /* 15-19 */
111 0x104, 0x10c, 0x114, 0x11c, /* 20-23 */
112 };
113 #define MAX_INTERLEAVE \
114 (MAX_T(unsigned int, ARRAY_SIZE(sbridge_interleave_list), \
115 MAX_T(unsigned int, ARRAY_SIZE(ibridge_interleave_list), \
116 ARRAY_SIZE(knl_interleave_list))))
117
118 struct interleave_pkg {
119 unsigned char start;
120 unsigned char end;
121 };
122
123 static const struct interleave_pkg sbridge_interleave_pkg[] = {
124 { 0, 2 },
125 { 3, 5 },
126 { 8, 10 },
127 { 11, 13 },
128 { 16, 18 },
129 { 19, 21 },
130 { 24, 26 },
131 { 27, 29 },
132 };
133
134 static const struct interleave_pkg ibridge_interleave_pkg[] = {
135 { 0, 3 },
136 { 4, 7 },
137 { 8, 11 },
138 { 12, 15 },
139 { 16, 19 },
140 { 20, 23 },
141 { 24, 27 },
142 { 28, 31 },
143 };
144
sad_pkg(const struct interleave_pkg * table,u32 reg,int interleave)145 static inline int sad_pkg(const struct interleave_pkg *table, u32 reg,
146 int interleave)
147 {
148 return GET_BITFIELD(reg, table[interleave].start,
149 table[interleave].end);
150 }
151
152 /* Devices 12 Function 7 */
153
154 #define TOLM 0x80
155 #define TOHM 0x84
156 #define HASWELL_TOLM 0xd0
157 #define HASWELL_TOHM_0 0xd4
158 #define HASWELL_TOHM_1 0xd8
159 #define KNL_TOLM 0xd0
160 #define KNL_TOHM_0 0xd4
161 #define KNL_TOHM_1 0xd8
162
163 #define GET_TOLM(reg) ((GET_BITFIELD(reg, 0, 3) << 28) | 0x3ffffff)
164 #define GET_TOHM(reg) ((GET_BITFIELD(reg, 0, 20) << 25) | 0x3ffffff)
165
166 /* Device 13 Function 6 */
167
168 #define SAD_TARGET 0xf0
169
170 #define SOURCE_ID(reg) GET_BITFIELD(reg, 9, 11)
171
172 #define SOURCE_ID_KNL(reg) GET_BITFIELD(reg, 12, 14)
173
174 #define SAD_CONTROL 0xf4
175
176 /* Device 14 function 0 */
177
178 static const u32 tad_dram_rule[] = {
179 0x40, 0x44, 0x48, 0x4c,
180 0x50, 0x54, 0x58, 0x5c,
181 0x60, 0x64, 0x68, 0x6c,
182 };
183 #define MAX_TAD ARRAY_SIZE(tad_dram_rule)
184
185 #define TAD_LIMIT(reg) ((GET_BITFIELD(reg, 12, 31) << 26) | 0x3ffffff)
186 #define TAD_SOCK(reg) GET_BITFIELD(reg, 10, 11)
187 #define TAD_CH(reg) GET_BITFIELD(reg, 8, 9)
188 #define TAD_TGT3(reg) GET_BITFIELD(reg, 6, 7)
189 #define TAD_TGT2(reg) GET_BITFIELD(reg, 4, 5)
190 #define TAD_TGT1(reg) GET_BITFIELD(reg, 2, 3)
191 #define TAD_TGT0(reg) GET_BITFIELD(reg, 0, 1)
192
193 /* Device 15, function 0 */
194
195 #define MCMTR 0x7c
196 #define KNL_MCMTR 0x624
197
198 #define IS_ECC_ENABLED(mcmtr) GET_BITFIELD(mcmtr, 2, 2)
199 #define IS_LOCKSTEP_ENABLED(mcmtr) GET_BITFIELD(mcmtr, 1, 1)
200 #define IS_CLOSE_PG(mcmtr) GET_BITFIELD(mcmtr, 0, 0)
201
202 /* Device 15, function 1 */
203
204 #define RASENABLES 0xac
205 #define IS_MIRROR_ENABLED(reg) GET_BITFIELD(reg, 0, 0)
206
207 /* Device 15, functions 2-5 */
208
209 static const int mtr_regs[] = {
210 0x80, 0x84, 0x88,
211 };
212
213 static const int knl_mtr_reg = 0xb60;
214
215 #define RANK_DISABLE(mtr) GET_BITFIELD(mtr, 16, 19)
216 #define IS_DIMM_PRESENT(mtr) GET_BITFIELD(mtr, 14, 14)
217 #define RANK_CNT_BITS(mtr) GET_BITFIELD(mtr, 12, 13)
218 #define RANK_WIDTH_BITS(mtr) GET_BITFIELD(mtr, 2, 4)
219 #define COL_WIDTH_BITS(mtr) GET_BITFIELD(mtr, 0, 1)
220
221 static const u32 tad_ch_nilv_offset[] = {
222 0x90, 0x94, 0x98, 0x9c,
223 0xa0, 0xa4, 0xa8, 0xac,
224 0xb0, 0xb4, 0xb8, 0xbc,
225 };
226 #define CHN_IDX_OFFSET(reg) GET_BITFIELD(reg, 28, 29)
227 #define TAD_OFFSET(reg) (GET_BITFIELD(reg, 6, 25) << 26)
228
229 static const u32 rir_way_limit[] = {
230 0x108, 0x10c, 0x110, 0x114, 0x118,
231 };
232 #define MAX_RIR_RANGES ARRAY_SIZE(rir_way_limit)
233
234 #define IS_RIR_VALID(reg) GET_BITFIELD(reg, 31, 31)
235 #define RIR_WAY(reg) GET_BITFIELD(reg, 28, 29)
236
237 #define MAX_RIR_WAY 8
238
239 static const u32 rir_offset[MAX_RIR_RANGES][MAX_RIR_WAY] = {
240 { 0x120, 0x124, 0x128, 0x12c, 0x130, 0x134, 0x138, 0x13c },
241 { 0x140, 0x144, 0x148, 0x14c, 0x150, 0x154, 0x158, 0x15c },
242 { 0x160, 0x164, 0x168, 0x16c, 0x170, 0x174, 0x178, 0x17c },
243 { 0x180, 0x184, 0x188, 0x18c, 0x190, 0x194, 0x198, 0x19c },
244 { 0x1a0, 0x1a4, 0x1a8, 0x1ac, 0x1b0, 0x1b4, 0x1b8, 0x1bc },
245 };
246
247 #define RIR_RNK_TGT(type, reg) (((type) == BROADWELL) ? \
248 GET_BITFIELD(reg, 20, 23) : GET_BITFIELD(reg, 16, 19))
249
250 #define RIR_OFFSET(type, reg) (((type) == HASWELL || (type) == BROADWELL) ? \
251 GET_BITFIELD(reg, 2, 15) : GET_BITFIELD(reg, 2, 14))
252
253 /* Device 16, functions 2-7 */
254
255 /*
256 * FIXME: Implement the error count reads directly
257 */
258
259 #define RANK_ODD_OV(reg) GET_BITFIELD(reg, 31, 31)
260 #define RANK_ODD_ERR_CNT(reg) GET_BITFIELD(reg, 16, 30)
261 #define RANK_EVEN_OV(reg) GET_BITFIELD(reg, 15, 15)
262 #define RANK_EVEN_ERR_CNT(reg) GET_BITFIELD(reg, 0, 14)
263
264 #if 0 /* Currently unused*/
265 static const u32 correrrcnt[] = {
266 0x104, 0x108, 0x10c, 0x110,
267 };
268
269 static const u32 correrrthrsld[] = {
270 0x11c, 0x120, 0x124, 0x128,
271 };
272 #endif
273
274 #define RANK_ODD_ERR_THRSLD(reg) GET_BITFIELD(reg, 16, 30)
275 #define RANK_EVEN_ERR_THRSLD(reg) GET_BITFIELD(reg, 0, 14)
276
277
278 /* Device 17, function 0 */
279
280 #define SB_RANK_CFG_A 0x0328
281
282 #define IB_RANK_CFG_A 0x0320
283
284 /*
285 * sbridge structs
286 */
287
288 #define NUM_CHANNELS 6 /* Max channels per MC */
289 #define MAX_DIMMS 3 /* Max DIMMS per channel */
290 #define KNL_MAX_CHAS 38 /* KNL max num. of Cache Home Agents */
291 #define KNL_MAX_CHANNELS 6 /* KNL max num. of PCI channels */
292 #define KNL_MAX_EDCS 8 /* Embedded DRAM controllers */
293 #define CHANNEL_UNSPECIFIED 0xf /* Intel IA32 SDM 15-14 */
294
295 enum type {
296 SANDY_BRIDGE,
297 IVY_BRIDGE,
298 HASWELL,
299 BROADWELL,
300 KNIGHTS_LANDING,
301 };
302
303 enum domain {
304 IMC0 = 0,
305 IMC1,
306 SOCK,
307 };
308
309 enum mirroring_mode {
310 NON_MIRRORING,
311 ADDR_RANGE_MIRRORING,
312 FULL_MIRRORING,
313 };
314
315 struct sbridge_pvt;
316 struct sbridge_info {
317 enum type type;
318 u32 mcmtr;
319 u32 rankcfgr;
320 u64 (*get_tolm)(struct sbridge_pvt *pvt);
321 u64 (*get_tohm)(struct sbridge_pvt *pvt);
322 u64 (*rir_limit)(u32 reg);
323 u64 (*sad_limit)(u32 reg);
324 u32 (*interleave_mode)(u32 reg);
325 u32 (*dram_attr)(u32 reg);
326 const u32 *dram_rule;
327 const u32 *interleave_list;
328 const struct interleave_pkg *interleave_pkg;
329 u8 max_sad;
330 u8 (*get_node_id)(struct sbridge_pvt *pvt);
331 u8 (*get_ha)(u8 bank);
332 enum mem_type (*get_memory_type)(struct sbridge_pvt *pvt);
333 enum dev_type (*get_width)(struct sbridge_pvt *pvt, u32 mtr);
334 struct pci_dev *pci_vtd;
335 };
336
337 struct sbridge_channel {
338 u32 ranks;
339 u32 dimms;
340 struct dimm {
341 u32 rowbits;
342 u32 colbits;
343 u32 bank_xor_enable;
344 u32 amap_fine;
345 } dimm[MAX_DIMMS];
346 };
347
348 struct pci_id_descr {
349 int dev_id;
350 int optional;
351 enum domain dom;
352 };
353
354 struct pci_id_table {
355 const struct pci_id_descr *descr;
356 int n_devs_per_imc;
357 int n_devs_per_sock;
358 int n_imcs_per_sock;
359 enum type type;
360 };
361
362 struct sbridge_dev {
363 struct list_head list;
364 int seg;
365 u8 bus, mc;
366 u8 node_id, source_id;
367 struct pci_dev **pdev;
368 enum domain dom;
369 int n_devs;
370 int i_devs;
371 struct mem_ctl_info *mci;
372 };
373
374 struct knl_pvt {
375 struct pci_dev *pci_cha[KNL_MAX_CHAS];
376 struct pci_dev *pci_channel[KNL_MAX_CHANNELS];
377 struct pci_dev *pci_mc0;
378 struct pci_dev *pci_mc1;
379 struct pci_dev *pci_mc0_misc;
380 struct pci_dev *pci_mc1_misc;
381 struct pci_dev *pci_mc_info; /* tolm, tohm */
382 };
383
384 struct sbridge_pvt {
385 /* Devices per socket */
386 struct pci_dev *pci_ddrio;
387 struct pci_dev *pci_sad0, *pci_sad1;
388 struct pci_dev *pci_br0, *pci_br1;
389 /* Devices per memory controller */
390 struct pci_dev *pci_ha, *pci_ta, *pci_ras;
391 struct pci_dev *pci_tad[NUM_CHANNELS];
392
393 struct sbridge_dev *sbridge_dev;
394
395 struct sbridge_info info;
396 struct sbridge_channel channel[NUM_CHANNELS];
397
398 /* Memory type detection */
399 bool is_cur_addr_mirrored, is_lockstep, is_close_pg;
400 bool is_chan_hash;
401 enum mirroring_mode mirror_mode;
402
403 /* Memory description */
404 u64 tolm, tohm;
405 struct knl_pvt knl;
406 };
407
408 #define PCI_DESCR(device_id, opt, domain) \
409 .dev_id = (device_id), \
410 .optional = opt, \
411 .dom = domain
412
413 static const struct pci_id_descr pci_dev_descr_sbridge[] = {
414 /* Processor Home Agent */
415 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_HA0, 0, IMC0) },
416
417 /* Memory controller */
418 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TA, 0, IMC0) },
419 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_RAS, 0, IMC0) },
420 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD0, 0, IMC0) },
421 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD1, 0, IMC0) },
422 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD2, 0, IMC0) },
423 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD3, 0, IMC0) },
424 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_DDRIO, 1, SOCK) },
425
426 /* System Address Decoder */
427 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_SAD0, 0, SOCK) },
428 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_SAD1, 0, SOCK) },
429
430 /* Broadcast Registers */
431 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_BR, 0, SOCK) },
432 };
433
434 #define PCI_ID_TABLE_ENTRY(A, N, M, T) { \
435 .descr = A, \
436 .n_devs_per_imc = N, \
437 .n_devs_per_sock = ARRAY_SIZE(A), \
438 .n_imcs_per_sock = M, \
439 .type = T \
440 }
441
442 static const struct pci_id_table pci_dev_descr_sbridge_table[] = {
443 PCI_ID_TABLE_ENTRY(pci_dev_descr_sbridge, ARRAY_SIZE(pci_dev_descr_sbridge), 1, SANDY_BRIDGE),
444 { NULL, }
445 };
446
447 /* This changes depending if 1HA or 2HA:
448 * 1HA:
449 * 0x0eb8 (17.0) is DDRIO0
450 * 2HA:
451 * 0x0ebc (17.4) is DDRIO0
452 */
453 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_1HA_DDRIO0 0x0eb8
454 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_2HA_DDRIO0 0x0ebc
455
456 /* pci ids */
457 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0 0x0ea0
458 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TA 0x0ea8
459 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_RAS 0x0e71
460 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD0 0x0eaa
461 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD1 0x0eab
462 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD2 0x0eac
463 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD3 0x0ead
464 #define PCI_DEVICE_ID_INTEL_IBRIDGE_SAD 0x0ec8
465 #define PCI_DEVICE_ID_INTEL_IBRIDGE_BR0 0x0ec9
466 #define PCI_DEVICE_ID_INTEL_IBRIDGE_BR1 0x0eca
467 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1 0x0e60
468 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TA 0x0e68
469 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_RAS 0x0e79
470 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD0 0x0e6a
471 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD1 0x0e6b
472 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD2 0x0e6c
473 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD3 0x0e6d
474
475 static const struct pci_id_descr pci_dev_descr_ibridge[] = {
476 /* Processor Home Agent */
477 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0, 0, IMC0) },
478 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1, 1, IMC1) },
479
480 /* Memory controller */
481 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TA, 0, IMC0) },
482 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_RAS, 0, IMC0) },
483 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD0, 0, IMC0) },
484 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD1, 0, IMC0) },
485 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD2, 0, IMC0) },
486 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD3, 0, IMC0) },
487
488 /* Optional, mode 2HA */
489 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TA, 1, IMC1) },
490 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_RAS, 1, IMC1) },
491 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD0, 1, IMC1) },
492 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD1, 1, IMC1) },
493 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD2, 1, IMC1) },
494 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD3, 1, IMC1) },
495
496 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_1HA_DDRIO0, 1, SOCK) },
497 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_2HA_DDRIO0, 1, SOCK) },
498
499 /* System Address Decoder */
500 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_SAD, 0, SOCK) },
501
502 /* Broadcast Registers */
503 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_BR0, 1, SOCK) },
504 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_BR1, 0, SOCK) },
505
506 };
507
508 static const struct pci_id_table pci_dev_descr_ibridge_table[] = {
509 PCI_ID_TABLE_ENTRY(pci_dev_descr_ibridge, 12, 2, IVY_BRIDGE),
510 { NULL, }
511 };
512
513 /* Haswell support */
514 /* EN processor:
515 * - 1 IMC
516 * - 3 DDR3 channels, 2 DPC per channel
517 * EP processor:
518 * - 1 or 2 IMC
519 * - 4 DDR4 channels, 3 DPC per channel
520 * EP 4S processor:
521 * - 2 IMC
522 * - 4 DDR4 channels, 3 DPC per channel
523 * EX processor:
524 * - 2 IMC
525 * - each IMC interfaces with a SMI 2 channel
526 * - each SMI channel interfaces with a scalable memory buffer
527 * - each scalable memory buffer supports 4 DDR3/DDR4 channels, 3 DPC
528 */
529 #define HASWELL_DDRCRCLKCONTROLS 0xa10 /* Ditto on Broadwell */
530 #define HASWELL_HASYSDEFEATURE2 0x84
531 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_VTD_MISC 0x2f28
532 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0 0x2fa0
533 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1 0x2f60
534 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TA 0x2fa8
535 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TM 0x2f71
536 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TA 0x2f68
537 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TM 0x2f79
538 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_CBO_SAD0 0x2ffc
539 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_CBO_SAD1 0x2ffd
540 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD0 0x2faa
541 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD1 0x2fab
542 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD2 0x2fac
543 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD3 0x2fad
544 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD0 0x2f6a
545 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD1 0x2f6b
546 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD2 0x2f6c
547 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD3 0x2f6d
548 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO0 0x2fbd
549 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO1 0x2fbf
550 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO2 0x2fb9
551 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO3 0x2fbb
552 static const struct pci_id_descr pci_dev_descr_haswell[] = {
553 /* first item must be the HA */
554 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0, 0, IMC0) },
555 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1, 1, IMC1) },
556
557 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TA, 0, IMC0) },
558 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TM, 0, IMC0) },
559 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD0, 0, IMC0) },
560 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD1, 0, IMC0) },
561 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD2, 1, IMC0) },
562 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD3, 1, IMC0) },
563
564 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TA, 1, IMC1) },
565 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TM, 1, IMC1) },
566 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD0, 1, IMC1) },
567 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD1, 1, IMC1) },
568 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD2, 1, IMC1) },
569 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD3, 1, IMC1) },
570
571 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_CBO_SAD0, 0, SOCK) },
572 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_CBO_SAD1, 0, SOCK) },
573 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO0, 1, SOCK) },
574 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO1, 1, SOCK) },
575 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO2, 1, SOCK) },
576 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO3, 1, SOCK) },
577 };
578
579 static const struct pci_id_table pci_dev_descr_haswell_table[] = {
580 PCI_ID_TABLE_ENTRY(pci_dev_descr_haswell, 13, 2, HASWELL),
581 { NULL, }
582 };
583
584 /* Knight's Landing Support */
585 /*
586 * KNL's memory channels are swizzled between memory controllers.
587 * MC0 is mapped to CH3,4,5 and MC1 is mapped to CH0,1,2
588 */
589 #define knl_channel_remap(mc, chan) ((mc) ? (chan) : (chan) + 3)
590
591 /* Memory controller, TAD tables, error injection - 2-8-0, 2-9-0 (2 of these) */
592 #define PCI_DEVICE_ID_INTEL_KNL_IMC_MC 0x7840
593 /* DRAM channel stuff; bank addrs, dimmmtr, etc.. 2-8-2 - 2-9-4 (6 of these) */
594 #define PCI_DEVICE_ID_INTEL_KNL_IMC_CHAN 0x7843
595 /* kdrwdbu TAD limits/offsets, MCMTR - 2-10-1, 2-11-1 (2 of these) */
596 #define PCI_DEVICE_ID_INTEL_KNL_IMC_TA 0x7844
597 /* CHA broadcast registers, dram rules - 1-29-0 (1 of these) */
598 #define PCI_DEVICE_ID_INTEL_KNL_IMC_SAD0 0x782a
599 /* SAD target - 1-29-1 (1 of these) */
600 #define PCI_DEVICE_ID_INTEL_KNL_IMC_SAD1 0x782b
601 /* Caching / Home Agent */
602 #define PCI_DEVICE_ID_INTEL_KNL_IMC_CHA 0x782c
603 /* Device with TOLM and TOHM, 0-5-0 (1 of these) */
604 #define PCI_DEVICE_ID_INTEL_KNL_IMC_TOLHM 0x7810
605
606 /*
607 * KNL differs from SB, IB, and Haswell in that it has multiple
608 * instances of the same device with the same device ID, so we handle that
609 * by creating as many copies in the table as we expect to find.
610 * (Like device ID must be grouped together.)
611 */
612
613 static const struct pci_id_descr pci_dev_descr_knl[] = {
614 [0 ... 1] = { PCI_DESCR(PCI_DEVICE_ID_INTEL_KNL_IMC_MC, 0, IMC0)},
615 [2 ... 7] = { PCI_DESCR(PCI_DEVICE_ID_INTEL_KNL_IMC_CHAN, 0, IMC0) },
616 [8] = { PCI_DESCR(PCI_DEVICE_ID_INTEL_KNL_IMC_TA, 0, IMC0) },
617 [9] = { PCI_DESCR(PCI_DEVICE_ID_INTEL_KNL_IMC_TOLHM, 0, IMC0) },
618 [10] = { PCI_DESCR(PCI_DEVICE_ID_INTEL_KNL_IMC_SAD0, 0, SOCK) },
619 [11] = { PCI_DESCR(PCI_DEVICE_ID_INTEL_KNL_IMC_SAD1, 0, SOCK) },
620 [12 ... 49] = { PCI_DESCR(PCI_DEVICE_ID_INTEL_KNL_IMC_CHA, 0, SOCK) },
621 };
622
623 static const struct pci_id_table pci_dev_descr_knl_table[] = {
624 PCI_ID_TABLE_ENTRY(pci_dev_descr_knl, ARRAY_SIZE(pci_dev_descr_knl), 1, KNIGHTS_LANDING),
625 { NULL, }
626 };
627
628 /*
629 * Broadwell support
630 *
631 * DE processor:
632 * - 1 IMC
633 * - 2 DDR3 channels, 2 DPC per channel
634 * EP processor:
635 * - 1 or 2 IMC
636 * - 4 DDR4 channels, 3 DPC per channel
637 * EP 4S processor:
638 * - 2 IMC
639 * - 4 DDR4 channels, 3 DPC per channel
640 * EX processor:
641 * - 2 IMC
642 * - each IMC interfaces with a SMI 2 channel
643 * - each SMI channel interfaces with a scalable memory buffer
644 * - each scalable memory buffer supports 4 DDR3/DDR4 channels, 3 DPC
645 */
646 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_VTD_MISC 0x6f28
647 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0 0x6fa0
648 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1 0x6f60
649 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TA 0x6fa8
650 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TM 0x6f71
651 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TA 0x6f68
652 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TM 0x6f79
653 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_CBO_SAD0 0x6ffc
654 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_CBO_SAD1 0x6ffd
655 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD0 0x6faa
656 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD1 0x6fab
657 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD2 0x6fac
658 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD3 0x6fad
659 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD0 0x6f6a
660 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD1 0x6f6b
661 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD2 0x6f6c
662 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD3 0x6f6d
663 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_DDRIO0 0x6faf
664
665 static const struct pci_id_descr pci_dev_descr_broadwell[] = {
666 /* first item must be the HA */
667 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0, 0, IMC0) },
668 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1, 1, IMC1) },
669
670 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TA, 0, IMC0) },
671 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TM, 0, IMC0) },
672 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD0, 0, IMC0) },
673 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD1, 0, IMC0) },
674 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD2, 1, IMC0) },
675 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD3, 1, IMC0) },
676
677 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TA, 1, IMC1) },
678 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TM, 1, IMC1) },
679 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD0, 1, IMC1) },
680 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD1, 1, IMC1) },
681 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD2, 1, IMC1) },
682 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD3, 1, IMC1) },
683
684 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_CBO_SAD0, 0, SOCK) },
685 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_CBO_SAD1, 0, SOCK) },
686 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_DDRIO0, 1, SOCK) },
687 };
688
689 static const struct pci_id_table pci_dev_descr_broadwell_table[] = {
690 PCI_ID_TABLE_ENTRY(pci_dev_descr_broadwell, 10, 2, BROADWELL),
691 { NULL, }
692 };
693
694
695 /****************************************************************************
696 Ancillary status routines
697 ****************************************************************************/
698
numrank(enum type type,u32 mtr)699 static inline int numrank(enum type type, u32 mtr)
700 {
701 int ranks = (1 << RANK_CNT_BITS(mtr));
702 int max = 4;
703
704 if (type == HASWELL || type == BROADWELL || type == KNIGHTS_LANDING)
705 max = 8;
706
707 if (ranks > max) {
708 edac_dbg(0, "Invalid number of ranks: %d (max = %i) raw value = %x (%04x)\n",
709 ranks, max, (unsigned int)RANK_CNT_BITS(mtr), mtr);
710 return -EINVAL;
711 }
712
713 return ranks;
714 }
715
numrow(u32 mtr)716 static inline int numrow(u32 mtr)
717 {
718 int rows = (RANK_WIDTH_BITS(mtr) + 12);
719
720 if (rows < 13 || rows > 18) {
721 edac_dbg(0, "Invalid number of rows: %d (should be between 14 and 17) raw value = %x (%04x)\n",
722 rows, (unsigned int)RANK_WIDTH_BITS(mtr), mtr);
723 return -EINVAL;
724 }
725
726 return 1 << rows;
727 }
728
numcol(u32 mtr)729 static inline int numcol(u32 mtr)
730 {
731 int cols = (COL_WIDTH_BITS(mtr) + 10);
732
733 if (cols > 12) {
734 edac_dbg(0, "Invalid number of cols: %d (max = 4) raw value = %x (%04x)\n",
735 cols, (unsigned int)COL_WIDTH_BITS(mtr), mtr);
736 return -EINVAL;
737 }
738
739 return 1 << cols;
740 }
741
get_sbridge_dev(int seg,u8 bus,enum domain dom,int multi_bus,struct sbridge_dev * prev)742 static struct sbridge_dev *get_sbridge_dev(int seg, u8 bus, enum domain dom,
743 int multi_bus,
744 struct sbridge_dev *prev)
745 {
746 struct sbridge_dev *sbridge_dev;
747
748 /*
749 * If we have devices scattered across several busses that pertain
750 * to the same memory controller, we'll lump them all together.
751 */
752 if (multi_bus) {
753 return list_first_entry_or_null(&sbridge_edac_list,
754 struct sbridge_dev, list);
755 }
756
757 sbridge_dev = list_entry(prev ? prev->list.next
758 : sbridge_edac_list.next, struct sbridge_dev, list);
759
760 list_for_each_entry_from(sbridge_dev, &sbridge_edac_list, list) {
761 if ((sbridge_dev->seg == seg) && (sbridge_dev->bus == bus) &&
762 (dom == SOCK || dom == sbridge_dev->dom))
763 return sbridge_dev;
764 }
765
766 return NULL;
767 }
768
alloc_sbridge_dev(int seg,u8 bus,enum domain dom,const struct pci_id_table * table)769 static struct sbridge_dev *alloc_sbridge_dev(int seg, u8 bus, enum domain dom,
770 const struct pci_id_table *table)
771 {
772 struct sbridge_dev *sbridge_dev;
773
774 sbridge_dev = kzalloc(sizeof(*sbridge_dev), GFP_KERNEL);
775 if (!sbridge_dev)
776 return NULL;
777
778 sbridge_dev->pdev = kcalloc(table->n_devs_per_imc,
779 sizeof(*sbridge_dev->pdev),
780 GFP_KERNEL);
781 if (!sbridge_dev->pdev) {
782 kfree(sbridge_dev);
783 return NULL;
784 }
785
786 sbridge_dev->seg = seg;
787 sbridge_dev->bus = bus;
788 sbridge_dev->dom = dom;
789 sbridge_dev->n_devs = table->n_devs_per_imc;
790 list_add_tail(&sbridge_dev->list, &sbridge_edac_list);
791
792 return sbridge_dev;
793 }
794
free_sbridge_dev(struct sbridge_dev * sbridge_dev)795 static void free_sbridge_dev(struct sbridge_dev *sbridge_dev)
796 {
797 list_del(&sbridge_dev->list);
798 kfree(sbridge_dev->pdev);
799 kfree(sbridge_dev);
800 }
801
sbridge_get_tolm(struct sbridge_pvt * pvt)802 static u64 sbridge_get_tolm(struct sbridge_pvt *pvt)
803 {
804 u32 reg;
805
806 /* Address range is 32:28 */
807 pci_read_config_dword(pvt->pci_sad1, TOLM, ®);
808 return GET_TOLM(reg);
809 }
810
sbridge_get_tohm(struct sbridge_pvt * pvt)811 static u64 sbridge_get_tohm(struct sbridge_pvt *pvt)
812 {
813 u32 reg;
814
815 pci_read_config_dword(pvt->pci_sad1, TOHM, ®);
816 return GET_TOHM(reg);
817 }
818
ibridge_get_tolm(struct sbridge_pvt * pvt)819 static u64 ibridge_get_tolm(struct sbridge_pvt *pvt)
820 {
821 u32 reg;
822
823 pci_read_config_dword(pvt->pci_br1, TOLM, ®);
824
825 return GET_TOLM(reg);
826 }
827
ibridge_get_tohm(struct sbridge_pvt * pvt)828 static u64 ibridge_get_tohm(struct sbridge_pvt *pvt)
829 {
830 u32 reg;
831
832 pci_read_config_dword(pvt->pci_br1, TOHM, ®);
833
834 return GET_TOHM(reg);
835 }
836
rir_limit(u32 reg)837 static u64 rir_limit(u32 reg)
838 {
839 return ((u64)GET_BITFIELD(reg, 1, 10) << 29) | 0x1fffffff;
840 }
841
sad_limit(u32 reg)842 static u64 sad_limit(u32 reg)
843 {
844 return (GET_BITFIELD(reg, 6, 25) << 26) | 0x3ffffff;
845 }
846
interleave_mode(u32 reg)847 static u32 interleave_mode(u32 reg)
848 {
849 return GET_BITFIELD(reg, 1, 1);
850 }
851
dram_attr(u32 reg)852 static u32 dram_attr(u32 reg)
853 {
854 return GET_BITFIELD(reg, 2, 3);
855 }
856
knl_sad_limit(u32 reg)857 static u64 knl_sad_limit(u32 reg)
858 {
859 return (GET_BITFIELD(reg, 7, 26) << 26) | 0x3ffffff;
860 }
861
knl_interleave_mode(u32 reg)862 static u32 knl_interleave_mode(u32 reg)
863 {
864 return GET_BITFIELD(reg, 1, 2);
865 }
866
867 static const char * const knl_intlv_mode[] = {
868 "[8:6]", "[10:8]", "[14:12]", "[32:30]"
869 };
870
get_intlv_mode_str(u32 reg,enum type t)871 static const char *get_intlv_mode_str(u32 reg, enum type t)
872 {
873 if (t == KNIGHTS_LANDING)
874 return knl_intlv_mode[knl_interleave_mode(reg)];
875 else
876 return interleave_mode(reg) ? "[8:6]" : "[8:6]XOR[18:16]";
877 }
878
dram_attr_knl(u32 reg)879 static u32 dram_attr_knl(u32 reg)
880 {
881 return GET_BITFIELD(reg, 3, 4);
882 }
883
884
get_memory_type(struct sbridge_pvt * pvt)885 static enum mem_type get_memory_type(struct sbridge_pvt *pvt)
886 {
887 u32 reg;
888 enum mem_type mtype;
889
890 if (pvt->pci_ddrio) {
891 pci_read_config_dword(pvt->pci_ddrio, pvt->info.rankcfgr,
892 ®);
893 if (GET_BITFIELD(reg, 11, 11))
894 /* FIXME: Can also be LRDIMM */
895 mtype = MEM_RDDR3;
896 else
897 mtype = MEM_DDR3;
898 } else
899 mtype = MEM_UNKNOWN;
900
901 return mtype;
902 }
903
haswell_get_memory_type(struct sbridge_pvt * pvt)904 static enum mem_type haswell_get_memory_type(struct sbridge_pvt *pvt)
905 {
906 u32 reg;
907 bool registered = false;
908 enum mem_type mtype = MEM_UNKNOWN;
909
910 if (!pvt->pci_ddrio)
911 goto out;
912
913 pci_read_config_dword(pvt->pci_ddrio,
914 HASWELL_DDRCRCLKCONTROLS, ®);
915 /* Is_Rdimm */
916 if (GET_BITFIELD(reg, 16, 16))
917 registered = true;
918
919 pci_read_config_dword(pvt->pci_ta, MCMTR, ®);
920 if (GET_BITFIELD(reg, 14, 14)) {
921 if (registered)
922 mtype = MEM_RDDR4;
923 else
924 mtype = MEM_DDR4;
925 } else {
926 if (registered)
927 mtype = MEM_RDDR3;
928 else
929 mtype = MEM_DDR3;
930 }
931
932 out:
933 return mtype;
934 }
935
knl_get_width(struct sbridge_pvt * pvt,u32 mtr)936 static enum dev_type knl_get_width(struct sbridge_pvt *pvt, u32 mtr)
937 {
938 /* for KNL value is fixed */
939 return DEV_X16;
940 }
941
sbridge_get_width(struct sbridge_pvt * pvt,u32 mtr)942 static enum dev_type sbridge_get_width(struct sbridge_pvt *pvt, u32 mtr)
943 {
944 /* there's no way to figure out */
945 return DEV_UNKNOWN;
946 }
947
__ibridge_get_width(u32 mtr)948 static enum dev_type __ibridge_get_width(u32 mtr)
949 {
950 enum dev_type type = DEV_UNKNOWN;
951
952 switch (mtr) {
953 case 2:
954 type = DEV_X16;
955 break;
956 case 1:
957 type = DEV_X8;
958 break;
959 case 0:
960 type = DEV_X4;
961 break;
962 }
963
964 return type;
965 }
966
ibridge_get_width(struct sbridge_pvt * pvt,u32 mtr)967 static enum dev_type ibridge_get_width(struct sbridge_pvt *pvt, u32 mtr)
968 {
969 /*
970 * ddr3_width on the documentation but also valid for DDR4 on
971 * Haswell
972 */
973 return __ibridge_get_width(GET_BITFIELD(mtr, 7, 8));
974 }
975
broadwell_get_width(struct sbridge_pvt * pvt,u32 mtr)976 static enum dev_type broadwell_get_width(struct sbridge_pvt *pvt, u32 mtr)
977 {
978 /* ddr3_width on the documentation but also valid for DDR4 */
979 return __ibridge_get_width(GET_BITFIELD(mtr, 8, 9));
980 }
981
knl_get_memory_type(struct sbridge_pvt * pvt)982 static enum mem_type knl_get_memory_type(struct sbridge_pvt *pvt)
983 {
984 /* DDR4 RDIMMS and LRDIMMS are supported */
985 return MEM_RDDR4;
986 }
987
get_node_id(struct sbridge_pvt * pvt)988 static u8 get_node_id(struct sbridge_pvt *pvt)
989 {
990 u32 reg;
991 pci_read_config_dword(pvt->pci_br0, SAD_CONTROL, ®);
992 return GET_BITFIELD(reg, 0, 2);
993 }
994
haswell_get_node_id(struct sbridge_pvt * pvt)995 static u8 haswell_get_node_id(struct sbridge_pvt *pvt)
996 {
997 u32 reg;
998
999 pci_read_config_dword(pvt->pci_sad1, SAD_CONTROL, ®);
1000 return GET_BITFIELD(reg, 0, 3);
1001 }
1002
knl_get_node_id(struct sbridge_pvt * pvt)1003 static u8 knl_get_node_id(struct sbridge_pvt *pvt)
1004 {
1005 u32 reg;
1006
1007 pci_read_config_dword(pvt->pci_sad1, SAD_CONTROL, ®);
1008 return GET_BITFIELD(reg, 0, 2);
1009 }
1010
1011 /*
1012 * Use the reporting bank number to determine which memory
1013 * controller (also known as "ha" for "home agent"). Sandy
1014 * Bridge only has one memory controller per socket, so the
1015 * answer is always zero.
1016 */
sbridge_get_ha(u8 bank)1017 static u8 sbridge_get_ha(u8 bank)
1018 {
1019 return 0;
1020 }
1021
1022 /*
1023 * On Ivy Bridge, Haswell and Broadwell the error may be in a
1024 * home agent bank (7, 8), or one of the per-channel memory
1025 * controller banks (9 .. 16).
1026 */
ibridge_get_ha(u8 bank)1027 static u8 ibridge_get_ha(u8 bank)
1028 {
1029 switch (bank) {
1030 case 7 ... 8:
1031 return bank - 7;
1032 case 9 ... 16:
1033 return (bank - 9) / 4;
1034 default:
1035 return 0xff;
1036 }
1037 }
1038
1039 /* Not used, but included for safety/symmetry */
knl_get_ha(u8 bank)1040 static u8 knl_get_ha(u8 bank)
1041 {
1042 return 0xff;
1043 }
1044
haswell_get_tolm(struct sbridge_pvt * pvt)1045 static u64 haswell_get_tolm(struct sbridge_pvt *pvt)
1046 {
1047 u32 reg;
1048
1049 pci_read_config_dword(pvt->info.pci_vtd, HASWELL_TOLM, ®);
1050 return (GET_BITFIELD(reg, 26, 31) << 26) | 0x3ffffff;
1051 }
1052
haswell_get_tohm(struct sbridge_pvt * pvt)1053 static u64 haswell_get_tohm(struct sbridge_pvt *pvt)
1054 {
1055 u64 rc;
1056 u32 reg;
1057
1058 pci_read_config_dword(pvt->info.pci_vtd, HASWELL_TOHM_0, ®);
1059 rc = GET_BITFIELD(reg, 26, 31);
1060 pci_read_config_dword(pvt->info.pci_vtd, HASWELL_TOHM_1, ®);
1061 rc = ((reg << 6) | rc) << 26;
1062
1063 return rc | 0x3ffffff;
1064 }
1065
knl_get_tolm(struct sbridge_pvt * pvt)1066 static u64 knl_get_tolm(struct sbridge_pvt *pvt)
1067 {
1068 u32 reg;
1069
1070 pci_read_config_dword(pvt->knl.pci_mc_info, KNL_TOLM, ®);
1071 return (GET_BITFIELD(reg, 26, 31) << 26) | 0x3ffffff;
1072 }
1073
knl_get_tohm(struct sbridge_pvt * pvt)1074 static u64 knl_get_tohm(struct sbridge_pvt *pvt)
1075 {
1076 u64 rc;
1077 u32 reg_lo, reg_hi;
1078
1079 pci_read_config_dword(pvt->knl.pci_mc_info, KNL_TOHM_0, ®_lo);
1080 pci_read_config_dword(pvt->knl.pci_mc_info, KNL_TOHM_1, ®_hi);
1081 rc = ((u64)reg_hi << 32) | reg_lo;
1082 return rc | 0x3ffffff;
1083 }
1084
1085
haswell_rir_limit(u32 reg)1086 static u64 haswell_rir_limit(u32 reg)
1087 {
1088 return (((u64)GET_BITFIELD(reg, 1, 11) + 1) << 29) - 1;
1089 }
1090
sad_pkg_socket(u8 pkg)1091 static inline u8 sad_pkg_socket(u8 pkg)
1092 {
1093 /* on Ivy Bridge, nodeID is SASS, where A is HA and S is node id */
1094 return ((pkg >> 3) << 2) | (pkg & 0x3);
1095 }
1096
sad_pkg_ha(u8 pkg)1097 static inline u8 sad_pkg_ha(u8 pkg)
1098 {
1099 return (pkg >> 2) & 0x1;
1100 }
1101
haswell_chan_hash(int idx,u64 addr)1102 static int haswell_chan_hash(int idx, u64 addr)
1103 {
1104 int i;
1105
1106 /*
1107 * XOR even bits from 12:26 to bit0 of idx,
1108 * odd bits from 13:27 to bit1
1109 */
1110 for (i = 12; i < 28; i += 2)
1111 idx ^= (addr >> i) & 3;
1112
1113 return idx;
1114 }
1115
1116 /* Low bits of TAD limit, and some metadata. */
1117 static const u32 knl_tad_dram_limit_lo[] = {
1118 0x400, 0x500, 0x600, 0x700,
1119 0x800, 0x900, 0xa00, 0xb00,
1120 };
1121
1122 /* Low bits of TAD offset. */
1123 static const u32 knl_tad_dram_offset_lo[] = {
1124 0x404, 0x504, 0x604, 0x704,
1125 0x804, 0x904, 0xa04, 0xb04,
1126 };
1127
1128 /* High 16 bits of TAD limit and offset. */
1129 static const u32 knl_tad_dram_hi[] = {
1130 0x408, 0x508, 0x608, 0x708,
1131 0x808, 0x908, 0xa08, 0xb08,
1132 };
1133
1134 /* Number of ways a tad entry is interleaved. */
1135 static const u32 knl_tad_ways[] = {
1136 8, 6, 4, 3, 2, 1,
1137 };
1138
1139 /*
1140 * Retrieve the n'th Target Address Decode table entry
1141 * from the memory controller's TAD table.
1142 *
1143 * @pvt: driver private data
1144 * @entry: which entry you want to retrieve
1145 * @mc: which memory controller (0 or 1)
1146 * @offset: output tad range offset
1147 * @limit: output address of first byte above tad range
1148 * @ways: output number of interleave ways
1149 *
1150 * The offset value has curious semantics. It's a sort of running total
1151 * of the sizes of all the memory regions that aren't mapped in this
1152 * tad table.
1153 */
knl_get_tad(const struct sbridge_pvt * pvt,const int entry,const int mc,u64 * offset,u64 * limit,int * ways)1154 static int knl_get_tad(const struct sbridge_pvt *pvt,
1155 const int entry,
1156 const int mc,
1157 u64 *offset,
1158 u64 *limit,
1159 int *ways)
1160 {
1161 u32 reg_limit_lo, reg_offset_lo, reg_hi;
1162 struct pci_dev *pci_mc;
1163 int way_id;
1164
1165 switch (mc) {
1166 case 0:
1167 pci_mc = pvt->knl.pci_mc0;
1168 break;
1169 case 1:
1170 pci_mc = pvt->knl.pci_mc1;
1171 break;
1172 default:
1173 WARN_ON(1);
1174 return -EINVAL;
1175 }
1176
1177 pci_read_config_dword(pci_mc,
1178 knl_tad_dram_limit_lo[entry], ®_limit_lo);
1179 pci_read_config_dword(pci_mc,
1180 knl_tad_dram_offset_lo[entry], ®_offset_lo);
1181 pci_read_config_dword(pci_mc,
1182 knl_tad_dram_hi[entry], ®_hi);
1183
1184 /* Is this TAD entry enabled? */
1185 if (!GET_BITFIELD(reg_limit_lo, 0, 0))
1186 return -ENODEV;
1187
1188 way_id = GET_BITFIELD(reg_limit_lo, 3, 5);
1189
1190 if (way_id < ARRAY_SIZE(knl_tad_ways)) {
1191 *ways = knl_tad_ways[way_id];
1192 } else {
1193 *ways = 0;
1194 sbridge_printk(KERN_ERR,
1195 "Unexpected value %d in mc_tad_limit_lo wayness field\n",
1196 way_id);
1197 return -ENODEV;
1198 }
1199
1200 /*
1201 * The least significant 6 bits of base and limit are truncated.
1202 * For limit, we fill the missing bits with 1s.
1203 */
1204 *offset = ((u64) GET_BITFIELD(reg_offset_lo, 6, 31) << 6) |
1205 ((u64) GET_BITFIELD(reg_hi, 0, 15) << 32);
1206 *limit = ((u64) GET_BITFIELD(reg_limit_lo, 6, 31) << 6) | 63 |
1207 ((u64) GET_BITFIELD(reg_hi, 16, 31) << 32);
1208
1209 return 0;
1210 }
1211
1212 /* Determine which memory controller is responsible for a given channel. */
knl_channel_mc(int channel)1213 static int knl_channel_mc(int channel)
1214 {
1215 WARN_ON(channel < 0 || channel >= 6);
1216
1217 return channel < 3 ? 1 : 0;
1218 }
1219
1220 /*
1221 * Get the Nth entry from EDC_ROUTE_TABLE register.
1222 * (This is the per-tile mapping of logical interleave targets to
1223 * physical EDC modules.)
1224 *
1225 * entry 0: 0:2
1226 * 1: 3:5
1227 * 2: 6:8
1228 * 3: 9:11
1229 * 4: 12:14
1230 * 5: 15:17
1231 * 6: 18:20
1232 * 7: 21:23
1233 * reserved: 24:31
1234 */
knl_get_edc_route(int entry,u32 reg)1235 static u32 knl_get_edc_route(int entry, u32 reg)
1236 {
1237 WARN_ON(entry >= KNL_MAX_EDCS);
1238 return GET_BITFIELD(reg, entry*3, (entry*3)+2);
1239 }
1240
1241 /*
1242 * Get the Nth entry from MC_ROUTE_TABLE register.
1243 * (This is the per-tile mapping of logical interleave targets to
1244 * physical DRAM channels modules.)
1245 *
1246 * entry 0: mc 0:2 channel 18:19
1247 * 1: mc 3:5 channel 20:21
1248 * 2: mc 6:8 channel 22:23
1249 * 3: mc 9:11 channel 24:25
1250 * 4: mc 12:14 channel 26:27
1251 * 5: mc 15:17 channel 28:29
1252 * reserved: 30:31
1253 *
1254 * Though we have 3 bits to identify the MC, we should only see
1255 * the values 0 or 1.
1256 */
1257
knl_get_mc_route(int entry,u32 reg)1258 static u32 knl_get_mc_route(int entry, u32 reg)
1259 {
1260 int mc, chan;
1261
1262 WARN_ON(entry >= KNL_MAX_CHANNELS);
1263
1264 mc = GET_BITFIELD(reg, entry*3, (entry*3)+2);
1265 chan = GET_BITFIELD(reg, (entry*2) + 18, (entry*2) + 18 + 1);
1266
1267 return knl_channel_remap(mc, chan);
1268 }
1269
1270 /*
1271 * Render the EDC_ROUTE register in human-readable form.
1272 * Output string s should be at least KNL_MAX_EDCS*2 bytes.
1273 */
knl_show_edc_route(u32 reg,char * s)1274 static void knl_show_edc_route(u32 reg, char *s)
1275 {
1276 int i;
1277
1278 for (i = 0; i < KNL_MAX_EDCS; i++) {
1279 s[i*2] = knl_get_edc_route(i, reg) + '0';
1280 s[i*2+1] = '-';
1281 }
1282
1283 s[KNL_MAX_EDCS*2 - 1] = '\0';
1284 }
1285
1286 /*
1287 * Render the MC_ROUTE register in human-readable form.
1288 * Output string s should be at least KNL_MAX_CHANNELS*2 bytes.
1289 */
knl_show_mc_route(u32 reg,char * s)1290 static void knl_show_mc_route(u32 reg, char *s)
1291 {
1292 int i;
1293
1294 for (i = 0; i < KNL_MAX_CHANNELS; i++) {
1295 s[i*2] = knl_get_mc_route(i, reg) + '0';
1296 s[i*2+1] = '-';
1297 }
1298
1299 s[KNL_MAX_CHANNELS*2 - 1] = '\0';
1300 }
1301
1302 #define KNL_EDC_ROUTE 0xb8
1303 #define KNL_MC_ROUTE 0xb4
1304
1305 /* Is this dram rule backed by regular DRAM in flat mode? */
1306 #define KNL_EDRAM(reg) GET_BITFIELD(reg, 29, 29)
1307
1308 /* Is this dram rule cached? */
1309 #define KNL_CACHEABLE(reg) GET_BITFIELD(reg, 28, 28)
1310
1311 /* Is this rule backed by edc ? */
1312 #define KNL_EDRAM_ONLY(reg) GET_BITFIELD(reg, 29, 29)
1313
1314 /* Is this rule backed by DRAM, cacheable in EDRAM? */
1315 #define KNL_CACHEABLE(reg) GET_BITFIELD(reg, 28, 28)
1316
1317 /* Is this rule mod3? */
1318 #define KNL_MOD3(reg) GET_BITFIELD(reg, 27, 27)
1319
1320 /*
1321 * Figure out how big our RAM modules are.
1322 *
1323 * The DIMMMTR register in KNL doesn't tell us the size of the DIMMs, so we
1324 * have to figure this out from the SAD rules, interleave lists, route tables,
1325 * and TAD rules.
1326 *
1327 * SAD rules can have holes in them (e.g. the 3G-4G hole), so we have to
1328 * inspect the TAD rules to figure out how large the SAD regions really are.
1329 *
1330 * When we know the real size of a SAD region and how many ways it's
1331 * interleaved, we know the individual contribution of each channel to
1332 * TAD is size/ways.
1333 *
1334 * Finally, we have to check whether each channel participates in each SAD
1335 * region.
1336 *
1337 * Fortunately, KNL only supports one DIMM per channel, so once we know how
1338 * much memory the channel uses, we know the DIMM is at least that large.
1339 * (The BIOS might possibly choose not to map all available memory, in which
1340 * case we will underreport the size of the DIMM.)
1341 *
1342 * In theory, we could try to determine the EDC sizes as well, but that would
1343 * only work in flat mode, not in cache mode.
1344 *
1345 * @mc_sizes: Output sizes of channels (must have space for KNL_MAX_CHANNELS
1346 * elements)
1347 */
knl_get_dimm_capacity(struct sbridge_pvt * pvt,u64 * mc_sizes)1348 static int knl_get_dimm_capacity(struct sbridge_pvt *pvt, u64 *mc_sizes)
1349 {
1350 u64 sad_base, sad_limit = 0;
1351 u64 tad_base, tad_size, tad_limit, tad_deadspace, tad_livespace;
1352 int sad_rule = 0;
1353 int tad_rule = 0;
1354 int intrlv_ways, tad_ways;
1355 u32 first_pkg, pkg;
1356 int i;
1357 u64 sad_actual_size[2]; /* sad size accounting for holes, per mc */
1358 u32 dram_rule, interleave_reg;
1359 u32 mc_route_reg[KNL_MAX_CHAS];
1360 u32 edc_route_reg[KNL_MAX_CHAS];
1361 int edram_only;
1362 char edc_route_string[KNL_MAX_EDCS*2];
1363 char mc_route_string[KNL_MAX_CHANNELS*2];
1364 int cur_reg_start;
1365 int mc;
1366 int channel;
1367 int participants[KNL_MAX_CHANNELS];
1368
1369 for (i = 0; i < KNL_MAX_CHANNELS; i++)
1370 mc_sizes[i] = 0;
1371
1372 /* Read the EDC route table in each CHA. */
1373 cur_reg_start = 0;
1374 for (i = 0; i < KNL_MAX_CHAS; i++) {
1375 pci_read_config_dword(pvt->knl.pci_cha[i],
1376 KNL_EDC_ROUTE, &edc_route_reg[i]);
1377
1378 if (i > 0 && edc_route_reg[i] != edc_route_reg[i-1]) {
1379 knl_show_edc_route(edc_route_reg[i-1],
1380 edc_route_string);
1381 if (cur_reg_start == i-1)
1382 edac_dbg(0, "edc route table for CHA %d: %s\n",
1383 cur_reg_start, edc_route_string);
1384 else
1385 edac_dbg(0, "edc route table for CHA %d-%d: %s\n",
1386 cur_reg_start, i-1, edc_route_string);
1387 cur_reg_start = i;
1388 }
1389 }
1390 knl_show_edc_route(edc_route_reg[i-1], edc_route_string);
1391 if (cur_reg_start == i-1)
1392 edac_dbg(0, "edc route table for CHA %d: %s\n",
1393 cur_reg_start, edc_route_string);
1394 else
1395 edac_dbg(0, "edc route table for CHA %d-%d: %s\n",
1396 cur_reg_start, i-1, edc_route_string);
1397
1398 /* Read the MC route table in each CHA. */
1399 cur_reg_start = 0;
1400 for (i = 0; i < KNL_MAX_CHAS; i++) {
1401 pci_read_config_dword(pvt->knl.pci_cha[i],
1402 KNL_MC_ROUTE, &mc_route_reg[i]);
1403
1404 if (i > 0 && mc_route_reg[i] != mc_route_reg[i-1]) {
1405 knl_show_mc_route(mc_route_reg[i-1], mc_route_string);
1406 if (cur_reg_start == i-1)
1407 edac_dbg(0, "mc route table for CHA %d: %s\n",
1408 cur_reg_start, mc_route_string);
1409 else
1410 edac_dbg(0, "mc route table for CHA %d-%d: %s\n",
1411 cur_reg_start, i-1, mc_route_string);
1412 cur_reg_start = i;
1413 }
1414 }
1415 knl_show_mc_route(mc_route_reg[i-1], mc_route_string);
1416 if (cur_reg_start == i-1)
1417 edac_dbg(0, "mc route table for CHA %d: %s\n",
1418 cur_reg_start, mc_route_string);
1419 else
1420 edac_dbg(0, "mc route table for CHA %d-%d: %s\n",
1421 cur_reg_start, i-1, mc_route_string);
1422
1423 /* Process DRAM rules */
1424 for (sad_rule = 0; sad_rule < pvt->info.max_sad; sad_rule++) {
1425 /* previous limit becomes the new base */
1426 sad_base = sad_limit;
1427
1428 pci_read_config_dword(pvt->pci_sad0,
1429 pvt->info.dram_rule[sad_rule], &dram_rule);
1430
1431 if (!DRAM_RULE_ENABLE(dram_rule))
1432 break;
1433
1434 edram_only = KNL_EDRAM_ONLY(dram_rule);
1435
1436 sad_limit = pvt->info.sad_limit(dram_rule)+1;
1437
1438 pci_read_config_dword(pvt->pci_sad0,
1439 pvt->info.interleave_list[sad_rule], &interleave_reg);
1440
1441 /*
1442 * Find out how many ways this dram rule is interleaved.
1443 * We stop when we see the first channel again.
1444 */
1445 first_pkg = sad_pkg(pvt->info.interleave_pkg,
1446 interleave_reg, 0);
1447 for (intrlv_ways = 1; intrlv_ways < 8; intrlv_ways++) {
1448 pkg = sad_pkg(pvt->info.interleave_pkg,
1449 interleave_reg, intrlv_ways);
1450
1451 if ((pkg & 0x8) == 0) {
1452 /*
1453 * 0 bit means memory is non-local,
1454 * which KNL doesn't support
1455 */
1456 edac_dbg(0, "Unexpected interleave target %d\n",
1457 pkg);
1458 return -1;
1459 }
1460
1461 if (pkg == first_pkg)
1462 break;
1463 }
1464 if (KNL_MOD3(dram_rule))
1465 intrlv_ways *= 3;
1466
1467 edac_dbg(3, "dram rule %d (base 0x%llx, limit 0x%llx), %d way interleave%s\n",
1468 sad_rule,
1469 sad_base,
1470 sad_limit,
1471 intrlv_ways,
1472 edram_only ? ", EDRAM" : "");
1473
1474 /*
1475 * Find out how big the SAD region really is by iterating
1476 * over TAD tables (SAD regions may contain holes).
1477 * Each memory controller might have a different TAD table, so
1478 * we have to look at both.
1479 *
1480 * Livespace is the memory that's mapped in this TAD table,
1481 * deadspace is the holes (this could be the MMIO hole, or it
1482 * could be memory that's mapped by the other TAD table but
1483 * not this one).
1484 */
1485 for (mc = 0; mc < 2; mc++) {
1486 sad_actual_size[mc] = 0;
1487 tad_livespace = 0;
1488 for (tad_rule = 0;
1489 tad_rule < ARRAY_SIZE(
1490 knl_tad_dram_limit_lo);
1491 tad_rule++) {
1492 if (knl_get_tad(pvt,
1493 tad_rule,
1494 mc,
1495 &tad_deadspace,
1496 &tad_limit,
1497 &tad_ways))
1498 break;
1499
1500 tad_size = (tad_limit+1) -
1501 (tad_livespace + tad_deadspace);
1502 tad_livespace += tad_size;
1503 tad_base = (tad_limit+1) - tad_size;
1504
1505 if (tad_base < sad_base) {
1506 if (tad_limit > sad_base)
1507 edac_dbg(0, "TAD region overlaps lower SAD boundary -- TAD tables may be configured incorrectly.\n");
1508 } else if (tad_base < sad_limit) {
1509 if (tad_limit+1 > sad_limit) {
1510 edac_dbg(0, "TAD region overlaps upper SAD boundary -- TAD tables may be configured incorrectly.\n");
1511 } else {
1512 /* TAD region is completely inside SAD region */
1513 edac_dbg(3, "TAD region %d 0x%llx - 0x%llx (%lld bytes) table%d\n",
1514 tad_rule, tad_base,
1515 tad_limit, tad_size,
1516 mc);
1517 sad_actual_size[mc] += tad_size;
1518 }
1519 }
1520 }
1521 }
1522
1523 for (mc = 0; mc < 2; mc++) {
1524 edac_dbg(3, " total TAD DRAM footprint in table%d : 0x%llx (%lld bytes)\n",
1525 mc, sad_actual_size[mc], sad_actual_size[mc]);
1526 }
1527
1528 /* Ignore EDRAM rule */
1529 if (edram_only)
1530 continue;
1531
1532 /* Figure out which channels participate in interleave. */
1533 for (channel = 0; channel < KNL_MAX_CHANNELS; channel++)
1534 participants[channel] = 0;
1535
1536 /* For each channel, does at least one CHA have
1537 * this channel mapped to the given target?
1538 */
1539 for (channel = 0; channel < KNL_MAX_CHANNELS; channel++) {
1540 int target;
1541 int cha;
1542
1543 for (target = 0; target < KNL_MAX_CHANNELS; target++) {
1544 for (cha = 0; cha < KNL_MAX_CHAS; cha++) {
1545 if (knl_get_mc_route(target,
1546 mc_route_reg[cha]) == channel
1547 && !participants[channel]) {
1548 participants[channel] = 1;
1549 break;
1550 }
1551 }
1552 }
1553 }
1554
1555 for (channel = 0; channel < KNL_MAX_CHANNELS; channel++) {
1556 mc = knl_channel_mc(channel);
1557 if (participants[channel]) {
1558 edac_dbg(4, "mc channel %d contributes %lld bytes via sad entry %d\n",
1559 channel,
1560 sad_actual_size[mc]/intrlv_ways,
1561 sad_rule);
1562 mc_sizes[channel] +=
1563 sad_actual_size[mc]/intrlv_ways;
1564 }
1565 }
1566 }
1567
1568 return 0;
1569 }
1570
get_source_id(struct mem_ctl_info * mci)1571 static void get_source_id(struct mem_ctl_info *mci)
1572 {
1573 struct sbridge_pvt *pvt = mci->pvt_info;
1574 u32 reg;
1575
1576 if (pvt->info.type == HASWELL || pvt->info.type == BROADWELL ||
1577 pvt->info.type == KNIGHTS_LANDING)
1578 pci_read_config_dword(pvt->pci_sad1, SAD_TARGET, ®);
1579 else
1580 pci_read_config_dword(pvt->pci_br0, SAD_TARGET, ®);
1581
1582 if (pvt->info.type == KNIGHTS_LANDING)
1583 pvt->sbridge_dev->source_id = SOURCE_ID_KNL(reg);
1584 else
1585 pvt->sbridge_dev->source_id = SOURCE_ID(reg);
1586 }
1587
__populate_dimms(struct mem_ctl_info * mci,u64 knl_mc_sizes[KNL_MAX_CHANNELS],enum edac_type mode)1588 static int __populate_dimms(struct mem_ctl_info *mci,
1589 u64 knl_mc_sizes[KNL_MAX_CHANNELS],
1590 enum edac_type mode)
1591 {
1592 struct sbridge_pvt *pvt = mci->pvt_info;
1593 int channels = pvt->info.type == KNIGHTS_LANDING ? KNL_MAX_CHANNELS
1594 : NUM_CHANNELS;
1595 unsigned int i, j, banks, ranks, rows, cols, npages;
1596 struct dimm_info *dimm;
1597 enum mem_type mtype;
1598 u64 size;
1599
1600 mtype = pvt->info.get_memory_type(pvt);
1601 if (mtype == MEM_RDDR3 || mtype == MEM_RDDR4)
1602 edac_dbg(0, "Memory is registered\n");
1603 else if (mtype == MEM_UNKNOWN)
1604 edac_dbg(0, "Cannot determine memory type\n");
1605 else
1606 edac_dbg(0, "Memory is unregistered\n");
1607
1608 if (mtype == MEM_DDR4 || mtype == MEM_RDDR4)
1609 banks = 16;
1610 else
1611 banks = 8;
1612
1613 for (i = 0; i < channels; i++) {
1614 u32 mtr, amap = 0;
1615
1616 int max_dimms_per_channel;
1617
1618 if (pvt->info.type == KNIGHTS_LANDING) {
1619 max_dimms_per_channel = 1;
1620 if (!pvt->knl.pci_channel[i])
1621 continue;
1622 } else {
1623 max_dimms_per_channel = ARRAY_SIZE(mtr_regs);
1624 if (!pvt->pci_tad[i])
1625 continue;
1626 pci_read_config_dword(pvt->pci_tad[i], 0x8c, &amap);
1627 }
1628
1629 for (j = 0; j < max_dimms_per_channel; j++) {
1630 dimm = edac_get_dimm(mci, i, j, 0);
1631 if (pvt->info.type == KNIGHTS_LANDING) {
1632 pci_read_config_dword(pvt->knl.pci_channel[i],
1633 knl_mtr_reg, &mtr);
1634 } else {
1635 pci_read_config_dword(pvt->pci_tad[i],
1636 mtr_regs[j], &mtr);
1637 }
1638 edac_dbg(4, "Channel #%d MTR%d = %x\n", i, j, mtr);
1639
1640 if (IS_DIMM_PRESENT(mtr)) {
1641 if (!IS_ECC_ENABLED(pvt->info.mcmtr)) {
1642 sbridge_printk(KERN_ERR, "CPU SrcID #%d, Ha #%d, Channel #%d has DIMMs, but ECC is disabled\n",
1643 pvt->sbridge_dev->source_id,
1644 pvt->sbridge_dev->dom, i);
1645 return -ENODEV;
1646 }
1647 pvt->channel[i].dimms++;
1648
1649 ranks = numrank(pvt->info.type, mtr);
1650
1651 if (pvt->info.type == KNIGHTS_LANDING) {
1652 /* For DDR4, this is fixed. */
1653 cols = 1 << 10;
1654 rows = knl_mc_sizes[i] /
1655 ((u64) cols * ranks * banks * 8);
1656 } else {
1657 rows = numrow(mtr);
1658 cols = numcol(mtr);
1659 }
1660
1661 size = ((u64)rows * cols * banks * ranks) >> (20 - 3);
1662 npages = MiB_TO_PAGES(size);
1663
1664 edac_dbg(0, "mc#%d: ha %d channel %d, dimm %d, %lld MiB (%d pages) bank: %d, rank: %d, row: %#x, col: %#x\n",
1665 pvt->sbridge_dev->mc, pvt->sbridge_dev->dom, i, j,
1666 size, npages,
1667 banks, ranks, rows, cols);
1668
1669 dimm->nr_pages = npages;
1670 dimm->grain = 32;
1671 dimm->dtype = pvt->info.get_width(pvt, mtr);
1672 dimm->mtype = mtype;
1673 dimm->edac_mode = mode;
1674 pvt->channel[i].dimm[j].rowbits = order_base_2(rows);
1675 pvt->channel[i].dimm[j].colbits = order_base_2(cols);
1676 pvt->channel[i].dimm[j].bank_xor_enable =
1677 GET_BITFIELD(pvt->info.mcmtr, 9, 9);
1678 pvt->channel[i].dimm[j].amap_fine = GET_BITFIELD(amap, 0, 0);
1679 snprintf(dimm->label, sizeof(dimm->label),
1680 "CPU_SrcID#%u_Ha#%u_Chan#%u_DIMM#%u",
1681 pvt->sbridge_dev->source_id, pvt->sbridge_dev->dom, i, j);
1682 }
1683 }
1684 }
1685
1686 return 0;
1687 }
1688
get_dimm_config(struct mem_ctl_info * mci)1689 static int get_dimm_config(struct mem_ctl_info *mci)
1690 {
1691 struct sbridge_pvt *pvt = mci->pvt_info;
1692 u64 knl_mc_sizes[KNL_MAX_CHANNELS];
1693 enum edac_type mode;
1694 u32 reg;
1695
1696 pvt->sbridge_dev->node_id = pvt->info.get_node_id(pvt);
1697 edac_dbg(0, "mc#%d: Node ID: %d, source ID: %d\n",
1698 pvt->sbridge_dev->mc,
1699 pvt->sbridge_dev->node_id,
1700 pvt->sbridge_dev->source_id);
1701
1702 /* KNL doesn't support mirroring or lockstep,
1703 * and is always closed page
1704 */
1705 if (pvt->info.type == KNIGHTS_LANDING) {
1706 mode = EDAC_S4ECD4ED;
1707 pvt->mirror_mode = NON_MIRRORING;
1708 pvt->is_cur_addr_mirrored = false;
1709
1710 if (knl_get_dimm_capacity(pvt, knl_mc_sizes) != 0)
1711 return -1;
1712 if (pci_read_config_dword(pvt->pci_ta, KNL_MCMTR, &pvt->info.mcmtr)) {
1713 edac_dbg(0, "Failed to read KNL_MCMTR register\n");
1714 return -ENODEV;
1715 }
1716 } else {
1717 if (pvt->info.type == HASWELL || pvt->info.type == BROADWELL) {
1718 if (pci_read_config_dword(pvt->pci_ha, HASWELL_HASYSDEFEATURE2, ®)) {
1719 edac_dbg(0, "Failed to read HASWELL_HASYSDEFEATURE2 register\n");
1720 return -ENODEV;
1721 }
1722 pvt->is_chan_hash = GET_BITFIELD(reg, 21, 21);
1723 if (GET_BITFIELD(reg, 28, 28)) {
1724 pvt->mirror_mode = ADDR_RANGE_MIRRORING;
1725 edac_dbg(0, "Address range partial memory mirroring is enabled\n");
1726 goto next;
1727 }
1728 }
1729 if (pci_read_config_dword(pvt->pci_ras, RASENABLES, ®)) {
1730 edac_dbg(0, "Failed to read RASENABLES register\n");
1731 return -ENODEV;
1732 }
1733 if (IS_MIRROR_ENABLED(reg)) {
1734 pvt->mirror_mode = FULL_MIRRORING;
1735 edac_dbg(0, "Full memory mirroring is enabled\n");
1736 } else {
1737 pvt->mirror_mode = NON_MIRRORING;
1738 edac_dbg(0, "Memory mirroring is disabled\n");
1739 }
1740
1741 next:
1742 if (pci_read_config_dword(pvt->pci_ta, MCMTR, &pvt->info.mcmtr)) {
1743 edac_dbg(0, "Failed to read MCMTR register\n");
1744 return -ENODEV;
1745 }
1746 if (IS_LOCKSTEP_ENABLED(pvt->info.mcmtr)) {
1747 edac_dbg(0, "Lockstep is enabled\n");
1748 mode = EDAC_S8ECD8ED;
1749 pvt->is_lockstep = true;
1750 } else {
1751 edac_dbg(0, "Lockstep is disabled\n");
1752 mode = EDAC_S4ECD4ED;
1753 pvt->is_lockstep = false;
1754 }
1755 if (IS_CLOSE_PG(pvt->info.mcmtr)) {
1756 edac_dbg(0, "address map is on closed page mode\n");
1757 pvt->is_close_pg = true;
1758 } else {
1759 edac_dbg(0, "address map is on open page mode\n");
1760 pvt->is_close_pg = false;
1761 }
1762 }
1763
1764 return __populate_dimms(mci, knl_mc_sizes, mode);
1765 }
1766
get_memory_layout(const struct mem_ctl_info * mci)1767 static void get_memory_layout(const struct mem_ctl_info *mci)
1768 {
1769 struct sbridge_pvt *pvt = mci->pvt_info;
1770 int i, j, k, n_sads, n_tads, sad_interl;
1771 u32 reg;
1772 u64 limit, prv = 0;
1773 u64 tmp_mb;
1774 u32 gb, mb;
1775 u32 rir_way;
1776
1777 /*
1778 * Step 1) Get TOLM/TOHM ranges
1779 */
1780
1781 pvt->tolm = pvt->info.get_tolm(pvt);
1782 tmp_mb = (1 + pvt->tolm) >> 20;
1783
1784 gb = div_u64_rem(tmp_mb, 1024, &mb);
1785 edac_dbg(0, "TOLM: %u.%03u GB (0x%016Lx)\n",
1786 gb, (mb*1000)/1024, (u64)pvt->tolm);
1787
1788 /* Address range is already 45:25 */
1789 pvt->tohm = pvt->info.get_tohm(pvt);
1790 tmp_mb = (1 + pvt->tohm) >> 20;
1791
1792 gb = div_u64_rem(tmp_mb, 1024, &mb);
1793 edac_dbg(0, "TOHM: %u.%03u GB (0x%016Lx)\n",
1794 gb, (mb*1000)/1024, (u64)pvt->tohm);
1795
1796 /*
1797 * Step 2) Get SAD range and SAD Interleave list
1798 * TAD registers contain the interleave wayness. However, it
1799 * seems simpler to just discover it indirectly, with the
1800 * algorithm bellow.
1801 */
1802 prv = 0;
1803 for (n_sads = 0; n_sads < pvt->info.max_sad; n_sads++) {
1804 /* SAD_LIMIT Address range is 45:26 */
1805 pci_read_config_dword(pvt->pci_sad0, pvt->info.dram_rule[n_sads],
1806 ®);
1807 limit = pvt->info.sad_limit(reg);
1808
1809 if (!DRAM_RULE_ENABLE(reg))
1810 continue;
1811
1812 if (limit <= prv)
1813 break;
1814
1815 tmp_mb = (limit + 1) >> 20;
1816 gb = div_u64_rem(tmp_mb, 1024, &mb);
1817 edac_dbg(0, "SAD#%d %s up to %u.%03u GB (0x%016Lx) Interleave: %s reg=0x%08x\n",
1818 n_sads,
1819 show_dram_attr(pvt->info.dram_attr(reg)),
1820 gb, (mb*1000)/1024,
1821 ((u64)tmp_mb) << 20L,
1822 get_intlv_mode_str(reg, pvt->info.type),
1823 reg);
1824 prv = limit;
1825
1826 pci_read_config_dword(pvt->pci_sad0, pvt->info.interleave_list[n_sads],
1827 ®);
1828 sad_interl = sad_pkg(pvt->info.interleave_pkg, reg, 0);
1829 for (j = 0; j < 8; j++) {
1830 u32 pkg = sad_pkg(pvt->info.interleave_pkg, reg, j);
1831 if (j > 0 && sad_interl == pkg)
1832 break;
1833
1834 edac_dbg(0, "SAD#%d, interleave #%d: %d\n",
1835 n_sads, j, pkg);
1836 }
1837 }
1838
1839 if (pvt->info.type == KNIGHTS_LANDING)
1840 return;
1841
1842 /*
1843 * Step 3) Get TAD range
1844 */
1845 prv = 0;
1846 for (n_tads = 0; n_tads < MAX_TAD; n_tads++) {
1847 pci_read_config_dword(pvt->pci_ha, tad_dram_rule[n_tads], ®);
1848 limit = TAD_LIMIT(reg);
1849 if (limit <= prv)
1850 break;
1851 tmp_mb = (limit + 1) >> 20;
1852
1853 gb = div_u64_rem(tmp_mb, 1024, &mb);
1854 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",
1855 n_tads, gb, (mb*1000)/1024,
1856 ((u64)tmp_mb) << 20L,
1857 (u32)(1 << TAD_SOCK(reg)),
1858 (u32)TAD_CH(reg) + 1,
1859 (u32)TAD_TGT0(reg),
1860 (u32)TAD_TGT1(reg),
1861 (u32)TAD_TGT2(reg),
1862 (u32)TAD_TGT3(reg),
1863 reg);
1864 prv = limit;
1865 }
1866
1867 /*
1868 * Step 4) Get TAD offsets, per each channel
1869 */
1870 for (i = 0; i < NUM_CHANNELS; i++) {
1871 if (!pvt->channel[i].dimms)
1872 continue;
1873 for (j = 0; j < n_tads; j++) {
1874 pci_read_config_dword(pvt->pci_tad[i],
1875 tad_ch_nilv_offset[j],
1876 ®);
1877 tmp_mb = TAD_OFFSET(reg) >> 20;
1878 gb = div_u64_rem(tmp_mb, 1024, &mb);
1879 edac_dbg(0, "TAD CH#%d, offset #%d: %u.%03u GB (0x%016Lx), reg=0x%08x\n",
1880 i, j,
1881 gb, (mb*1000)/1024,
1882 ((u64)tmp_mb) << 20L,
1883 reg);
1884 }
1885 }
1886
1887 /*
1888 * Step 6) Get RIR Wayness/Limit, per each channel
1889 */
1890 for (i = 0; i < NUM_CHANNELS; i++) {
1891 if (!pvt->channel[i].dimms)
1892 continue;
1893 for (j = 0; j < MAX_RIR_RANGES; j++) {
1894 pci_read_config_dword(pvt->pci_tad[i],
1895 rir_way_limit[j],
1896 ®);
1897
1898 if (!IS_RIR_VALID(reg))
1899 continue;
1900
1901 tmp_mb = pvt->info.rir_limit(reg) >> 20;
1902 rir_way = 1 << RIR_WAY(reg);
1903 gb = div_u64_rem(tmp_mb, 1024, &mb);
1904 edac_dbg(0, "CH#%d RIR#%d, limit: %u.%03u GB (0x%016Lx), way: %d, reg=0x%08x\n",
1905 i, j,
1906 gb, (mb*1000)/1024,
1907 ((u64)tmp_mb) << 20L,
1908 rir_way,
1909 reg);
1910
1911 for (k = 0; k < rir_way; k++) {
1912 pci_read_config_dword(pvt->pci_tad[i],
1913 rir_offset[j][k],
1914 ®);
1915 tmp_mb = RIR_OFFSET(pvt->info.type, reg) << 6;
1916
1917 gb = div_u64_rem(tmp_mb, 1024, &mb);
1918 edac_dbg(0, "CH#%d RIR#%d INTL#%d, offset %u.%03u GB (0x%016Lx), tgt: %d, reg=0x%08x\n",
1919 i, j, k,
1920 gb, (mb*1000)/1024,
1921 ((u64)tmp_mb) << 20L,
1922 (u32)RIR_RNK_TGT(pvt->info.type, reg),
1923 reg);
1924 }
1925 }
1926 }
1927 }
1928
get_mci_for_node_id(u8 node_id,u8 ha)1929 static struct mem_ctl_info *get_mci_for_node_id(u8 node_id, u8 ha)
1930 {
1931 struct sbridge_dev *sbridge_dev;
1932
1933 list_for_each_entry(sbridge_dev, &sbridge_edac_list, list) {
1934 if (sbridge_dev->node_id == node_id && sbridge_dev->dom == ha)
1935 return sbridge_dev->mci;
1936 }
1937 return NULL;
1938 }
1939
1940 static u8 sb_close_row[] = {
1941 15, 16, 17, 18, 20, 21, 22, 28, 10, 11, 12, 13, 29, 30, 31, 32, 33
1942 };
1943
1944 static u8 sb_close_column[] = {
1945 3, 4, 5, 14, 19, 23, 24, 25, 26, 27
1946 };
1947
1948 static u8 sb_open_row[] = {
1949 14, 15, 16, 20, 28, 21, 22, 23, 24, 25, 26, 27, 29, 30, 31, 32, 33
1950 };
1951
1952 static u8 sb_open_column[] = {
1953 3, 4, 5, 6, 7, 8, 9, 10, 11, 12
1954 };
1955
1956 static u8 sb_open_fine_column[] = {
1957 3, 4, 5, 7, 8, 9, 10, 11, 12, 13
1958 };
1959
sb_bits(u64 addr,int nbits,u8 * bits)1960 static int sb_bits(u64 addr, int nbits, u8 *bits)
1961 {
1962 int i, res = 0;
1963
1964 for (i = 0; i < nbits; i++)
1965 res |= ((addr >> bits[i]) & 1) << i;
1966 return res;
1967 }
1968
sb_bank_bits(u64 addr,int b0,int b1,int do_xor,int x0,int x1)1969 static int sb_bank_bits(u64 addr, int b0, int b1, int do_xor, int x0, int x1)
1970 {
1971 int ret = GET_BITFIELD(addr, b0, b0) | (GET_BITFIELD(addr, b1, b1) << 1);
1972
1973 if (do_xor)
1974 ret ^= GET_BITFIELD(addr, x0, x0) | (GET_BITFIELD(addr, x1, x1) << 1);
1975
1976 return ret;
1977 }
1978
sb_decode_ddr4(struct mem_ctl_info * mci,int ch,u8 rank,u64 rank_addr,char * msg)1979 static bool sb_decode_ddr4(struct mem_ctl_info *mci, int ch, u8 rank,
1980 u64 rank_addr, char *msg)
1981 {
1982 int dimmno = 0;
1983 int row, col, bank_address, bank_group;
1984 struct sbridge_pvt *pvt;
1985 u32 bg0 = 0, rowbits = 0, colbits = 0;
1986 u32 amap_fine = 0, bank_xor_enable = 0;
1987
1988 dimmno = (rank < 12) ? rank / 4 : 2;
1989 pvt = mci->pvt_info;
1990 amap_fine = pvt->channel[ch].dimm[dimmno].amap_fine;
1991 bg0 = amap_fine ? 6 : 13;
1992 rowbits = pvt->channel[ch].dimm[dimmno].rowbits;
1993 colbits = pvt->channel[ch].dimm[dimmno].colbits;
1994 bank_xor_enable = pvt->channel[ch].dimm[dimmno].bank_xor_enable;
1995
1996 if (pvt->is_lockstep) {
1997 pr_warn_once("LockStep row/column decode is not supported yet!\n");
1998 msg[0] = '\0';
1999 return false;
2000 }
2001
2002 if (pvt->is_close_pg) {
2003 row = sb_bits(rank_addr, rowbits, sb_close_row);
2004 col = sb_bits(rank_addr, colbits, sb_close_column);
2005 col |= 0x400; /* C10 is autoprecharge, always set */
2006 bank_address = sb_bank_bits(rank_addr, 8, 9, bank_xor_enable, 22, 28);
2007 bank_group = sb_bank_bits(rank_addr, 6, 7, bank_xor_enable, 20, 21);
2008 } else {
2009 row = sb_bits(rank_addr, rowbits, sb_open_row);
2010 if (amap_fine)
2011 col = sb_bits(rank_addr, colbits, sb_open_fine_column);
2012 else
2013 col = sb_bits(rank_addr, colbits, sb_open_column);
2014 bank_address = sb_bank_bits(rank_addr, 18, 19, bank_xor_enable, 22, 23);
2015 bank_group = sb_bank_bits(rank_addr, bg0, 17, bank_xor_enable, 20, 21);
2016 }
2017
2018 row &= (1u << rowbits) - 1;
2019
2020 sprintf(msg, "row:0x%x col:0x%x bank_addr:%d bank_group:%d",
2021 row, col, bank_address, bank_group);
2022 return true;
2023 }
2024
sb_decode_ddr3(struct mem_ctl_info * mci,int ch,u8 rank,u64 rank_addr,char * msg)2025 static bool sb_decode_ddr3(struct mem_ctl_info *mci, int ch, u8 rank,
2026 u64 rank_addr, char *msg)
2027 {
2028 pr_warn_once("DDR3 row/column decode not support yet!\n");
2029 msg[0] = '\0';
2030 return false;
2031 }
2032
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)2033 static int get_memory_error_data(struct mem_ctl_info *mci,
2034 u64 addr,
2035 u8 *socket, u8 *ha,
2036 long *channel_mask,
2037 u8 *rank,
2038 char **area_type, char *msg)
2039 {
2040 struct mem_ctl_info *new_mci;
2041 struct sbridge_pvt *pvt = mci->pvt_info;
2042 struct pci_dev *pci_ha;
2043 int n_rir, n_sads, n_tads, sad_way, sck_xch;
2044 int sad_interl, idx, base_ch;
2045 int interleave_mode, shiftup = 0;
2046 unsigned int sad_interleave[MAX_INTERLEAVE];
2047 u32 reg, dram_rule;
2048 u8 ch_way, sck_way, pkg, sad_ha = 0, rankid = 0;
2049 u32 tad_offset;
2050 u32 rir_way;
2051 u32 mb, gb;
2052 u64 ch_addr, offset, limit = 0, prv = 0;
2053 u64 rank_addr;
2054 enum mem_type mtype;
2055
2056 /*
2057 * Step 0) Check if the address is at special memory ranges
2058 * The check bellow is probably enough to fill all cases where
2059 * the error is not inside a memory, except for the legacy
2060 * range (e. g. VGA addresses). It is unlikely, however, that the
2061 * memory controller would generate an error on that range.
2062 */
2063 if ((addr > (u64) pvt->tolm) && (addr < (1LL << 32))) {
2064 sprintf(msg, "Error at TOLM area, on addr 0x%08Lx", addr);
2065 return -EINVAL;
2066 }
2067 if (addr >= (u64)pvt->tohm) {
2068 sprintf(msg, "Error at MMIOH area, on addr 0x%016Lx", addr);
2069 return -EINVAL;
2070 }
2071
2072 /*
2073 * Step 1) Get socket
2074 */
2075 for (n_sads = 0; n_sads < pvt->info.max_sad; n_sads++) {
2076 pci_read_config_dword(pvt->pci_sad0, pvt->info.dram_rule[n_sads],
2077 ®);
2078
2079 if (!DRAM_RULE_ENABLE(reg))
2080 continue;
2081
2082 limit = pvt->info.sad_limit(reg);
2083 if (limit <= prv) {
2084 sprintf(msg, "Can't discover the memory socket");
2085 return -EINVAL;
2086 }
2087 if (addr <= limit)
2088 break;
2089 prv = limit;
2090 }
2091 if (n_sads == pvt->info.max_sad) {
2092 sprintf(msg, "Can't discover the memory socket");
2093 return -EINVAL;
2094 }
2095 dram_rule = reg;
2096 *area_type = show_dram_attr(pvt->info.dram_attr(dram_rule));
2097 interleave_mode = pvt->info.interleave_mode(dram_rule);
2098
2099 pci_read_config_dword(pvt->pci_sad0, pvt->info.interleave_list[n_sads],
2100 ®);
2101
2102 if (pvt->info.type == SANDY_BRIDGE) {
2103 sad_interl = sad_pkg(pvt->info.interleave_pkg, reg, 0);
2104 for (sad_way = 0; sad_way < 8; sad_way++) {
2105 u32 pkg = sad_pkg(pvt->info.interleave_pkg, reg, sad_way);
2106 if (sad_way > 0 && sad_interl == pkg)
2107 break;
2108 sad_interleave[sad_way] = pkg;
2109 edac_dbg(0, "SAD interleave #%d: %d\n",
2110 sad_way, sad_interleave[sad_way]);
2111 }
2112 edac_dbg(0, "mc#%d: Error detected on SAD#%d: address 0x%016Lx < 0x%016Lx, Interleave [%d:6]%s\n",
2113 pvt->sbridge_dev->mc,
2114 n_sads,
2115 addr,
2116 limit,
2117 sad_way + 7,
2118 !interleave_mode ? "" : "XOR[18:16]");
2119 if (interleave_mode)
2120 idx = ((addr >> 6) ^ (addr >> 16)) & 7;
2121 else
2122 idx = (addr >> 6) & 7;
2123 switch (sad_way) {
2124 case 1:
2125 idx = 0;
2126 break;
2127 case 2:
2128 idx = idx & 1;
2129 break;
2130 case 4:
2131 idx = idx & 3;
2132 break;
2133 case 8:
2134 break;
2135 default:
2136 sprintf(msg, "Can't discover socket interleave");
2137 return -EINVAL;
2138 }
2139 *socket = sad_interleave[idx];
2140 edac_dbg(0, "SAD interleave index: %d (wayness %d) = CPU socket %d\n",
2141 idx, sad_way, *socket);
2142 } else if (pvt->info.type == HASWELL || pvt->info.type == BROADWELL) {
2143 int bits, a7mode = A7MODE(dram_rule);
2144
2145 if (a7mode) {
2146 /* A7 mode swaps P9 with P6 */
2147 bits = GET_BITFIELD(addr, 7, 8) << 1;
2148 bits |= GET_BITFIELD(addr, 9, 9);
2149 } else
2150 bits = GET_BITFIELD(addr, 6, 8);
2151
2152 if (interleave_mode == 0) {
2153 /* interleave mode will XOR {8,7,6} with {18,17,16} */
2154 idx = GET_BITFIELD(addr, 16, 18);
2155 idx ^= bits;
2156 } else
2157 idx = bits;
2158
2159 pkg = sad_pkg(pvt->info.interleave_pkg, reg, idx);
2160 *socket = sad_pkg_socket(pkg);
2161 sad_ha = sad_pkg_ha(pkg);
2162
2163 if (a7mode) {
2164 /* MCChanShiftUpEnable */
2165 pci_read_config_dword(pvt->pci_ha, HASWELL_HASYSDEFEATURE2, ®);
2166 shiftup = GET_BITFIELD(reg, 22, 22);
2167 }
2168
2169 edac_dbg(0, "SAD interleave package: %d = CPU socket %d, HA %i, shiftup: %i\n",
2170 idx, *socket, sad_ha, shiftup);
2171 } else {
2172 /* Ivy Bridge's SAD mode doesn't support XOR interleave mode */
2173 idx = (addr >> 6) & 7;
2174 pkg = sad_pkg(pvt->info.interleave_pkg, reg, idx);
2175 *socket = sad_pkg_socket(pkg);
2176 sad_ha = sad_pkg_ha(pkg);
2177 edac_dbg(0, "SAD interleave package: %d = CPU socket %d, HA %d\n",
2178 idx, *socket, sad_ha);
2179 }
2180
2181 *ha = sad_ha;
2182
2183 /*
2184 * Move to the proper node structure, in order to access the
2185 * right PCI registers
2186 */
2187 new_mci = get_mci_for_node_id(*socket, sad_ha);
2188 if (!new_mci) {
2189 sprintf(msg, "Struct for socket #%u wasn't initialized",
2190 *socket);
2191 return -EINVAL;
2192 }
2193 mci = new_mci;
2194 pvt = mci->pvt_info;
2195
2196 /*
2197 * Step 2) Get memory channel
2198 */
2199 prv = 0;
2200 pci_ha = pvt->pci_ha;
2201 for (n_tads = 0; n_tads < MAX_TAD; n_tads++) {
2202 pci_read_config_dword(pci_ha, tad_dram_rule[n_tads], ®);
2203 limit = TAD_LIMIT(reg);
2204 if (limit <= prv) {
2205 sprintf(msg, "Can't discover the memory channel");
2206 return -EINVAL;
2207 }
2208 if (addr <= limit)
2209 break;
2210 prv = limit;
2211 }
2212 if (n_tads == MAX_TAD) {
2213 sprintf(msg, "Can't discover the memory channel");
2214 return -EINVAL;
2215 }
2216
2217 ch_way = TAD_CH(reg) + 1;
2218 sck_way = TAD_SOCK(reg);
2219
2220 if (ch_way == 3)
2221 idx = addr >> 6;
2222 else {
2223 idx = (addr >> (6 + sck_way + shiftup)) & 0x3;
2224 if (pvt->is_chan_hash)
2225 idx = haswell_chan_hash(idx, addr);
2226 }
2227 idx = idx % ch_way;
2228
2229 /*
2230 * FIXME: Shouldn't we use CHN_IDX_OFFSET() here, when ch_way == 3 ???
2231 */
2232 switch (idx) {
2233 case 0:
2234 base_ch = TAD_TGT0(reg);
2235 break;
2236 case 1:
2237 base_ch = TAD_TGT1(reg);
2238 break;
2239 case 2:
2240 base_ch = TAD_TGT2(reg);
2241 break;
2242 case 3:
2243 base_ch = TAD_TGT3(reg);
2244 break;
2245 default:
2246 sprintf(msg, "Can't discover the TAD target");
2247 return -EINVAL;
2248 }
2249 *channel_mask = 1 << base_ch;
2250
2251 pci_read_config_dword(pvt->pci_tad[base_ch], tad_ch_nilv_offset[n_tads], &tad_offset);
2252
2253 if (pvt->mirror_mode == FULL_MIRRORING ||
2254 (pvt->mirror_mode == ADDR_RANGE_MIRRORING && n_tads == 0)) {
2255 *channel_mask |= 1 << ((base_ch + 2) % 4);
2256 switch(ch_way) {
2257 case 2:
2258 case 4:
2259 sck_xch = (1 << sck_way) * (ch_way >> 1);
2260 break;
2261 default:
2262 sprintf(msg, "Invalid mirror set. Can't decode addr");
2263 return -EINVAL;
2264 }
2265
2266 pvt->is_cur_addr_mirrored = true;
2267 } else {
2268 sck_xch = (1 << sck_way) * ch_way;
2269 pvt->is_cur_addr_mirrored = false;
2270 }
2271
2272 if (pvt->is_lockstep)
2273 *channel_mask |= 1 << ((base_ch + 1) % 4);
2274
2275 offset = TAD_OFFSET(tad_offset);
2276
2277 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",
2278 n_tads,
2279 addr,
2280 limit,
2281 sck_way,
2282 ch_way,
2283 offset,
2284 idx,
2285 base_ch,
2286 *channel_mask);
2287
2288 /* Calculate channel address */
2289 /* Remove the TAD offset */
2290
2291 if (offset > addr) {
2292 sprintf(msg, "Can't calculate ch addr: TAD offset 0x%08Lx is too high for addr 0x%08Lx!",
2293 offset, addr);
2294 return -EINVAL;
2295 }
2296
2297 ch_addr = addr - offset;
2298 ch_addr >>= (6 + shiftup);
2299 ch_addr /= sck_xch;
2300 ch_addr <<= (6 + shiftup);
2301 ch_addr |= addr & ((1 << (6 + shiftup)) - 1);
2302
2303 /*
2304 * Step 3) Decode rank
2305 */
2306 for (n_rir = 0; n_rir < MAX_RIR_RANGES; n_rir++) {
2307 pci_read_config_dword(pvt->pci_tad[base_ch], rir_way_limit[n_rir], ®);
2308
2309 if (!IS_RIR_VALID(reg))
2310 continue;
2311
2312 limit = pvt->info.rir_limit(reg);
2313 gb = div_u64_rem(limit >> 20, 1024, &mb);
2314 edac_dbg(0, "RIR#%d, limit: %u.%03u GB (0x%016Lx), way: %d\n",
2315 n_rir,
2316 gb, (mb*1000)/1024,
2317 limit,
2318 1 << RIR_WAY(reg));
2319 if (ch_addr <= limit)
2320 break;
2321 }
2322 if (n_rir == MAX_RIR_RANGES) {
2323 sprintf(msg, "Can't discover the memory rank for ch addr 0x%08Lx",
2324 ch_addr);
2325 return -EINVAL;
2326 }
2327 rir_way = RIR_WAY(reg);
2328
2329 if (pvt->is_close_pg)
2330 idx = (ch_addr >> 6);
2331 else
2332 idx = (ch_addr >> 13); /* FIXME: Datasheet says to shift by 15 */
2333 idx %= 1 << rir_way;
2334
2335 pci_read_config_dword(pvt->pci_tad[base_ch], rir_offset[n_rir][idx], ®);
2336 *rank = RIR_RNK_TGT(pvt->info.type, reg);
2337
2338 if (pvt->info.type == BROADWELL) {
2339 if (pvt->is_close_pg)
2340 shiftup = 6;
2341 else
2342 shiftup = 13;
2343
2344 rank_addr = ch_addr >> shiftup;
2345 rank_addr /= (1 << rir_way);
2346 rank_addr <<= shiftup;
2347 rank_addr |= ch_addr & GENMASK_ULL(shiftup - 1, 0);
2348 rank_addr -= RIR_OFFSET(pvt->info.type, reg);
2349
2350 mtype = pvt->info.get_memory_type(pvt);
2351 rankid = *rank;
2352 if (mtype == MEM_DDR4 || mtype == MEM_RDDR4)
2353 sb_decode_ddr4(mci, base_ch, rankid, rank_addr, msg);
2354 else
2355 sb_decode_ddr3(mci, base_ch, rankid, rank_addr, msg);
2356 } else {
2357 msg[0] = '\0';
2358 }
2359
2360 edac_dbg(0, "RIR#%d: channel address 0x%08Lx < 0x%08Lx, RIR interleave %d, index %d\n",
2361 n_rir,
2362 ch_addr,
2363 limit,
2364 rir_way,
2365 idx);
2366
2367 return 0;
2368 }
2369
get_memory_error_data_from_mce(struct mem_ctl_info * mci,const struct mce * m,u8 * socket,u8 * ha,long * channel_mask,char * msg)2370 static int get_memory_error_data_from_mce(struct mem_ctl_info *mci,
2371 const struct mce *m, u8 *socket,
2372 u8 *ha, long *channel_mask,
2373 char *msg)
2374 {
2375 u32 reg, channel = GET_BITFIELD(m->status, 0, 3);
2376 struct mem_ctl_info *new_mci;
2377 struct sbridge_pvt *pvt;
2378 struct pci_dev *pci_ha;
2379 bool tad0;
2380
2381 if (channel >= NUM_CHANNELS) {
2382 sprintf(msg, "Invalid channel 0x%x", channel);
2383 return -EINVAL;
2384 }
2385
2386 pvt = mci->pvt_info;
2387 if (!pvt->info.get_ha) {
2388 sprintf(msg, "No get_ha()");
2389 return -EINVAL;
2390 }
2391 *ha = pvt->info.get_ha(m->bank);
2392 if (*ha != 0 && *ha != 1) {
2393 sprintf(msg, "Impossible bank %d", m->bank);
2394 return -EINVAL;
2395 }
2396
2397 *socket = m->socketid;
2398 new_mci = get_mci_for_node_id(*socket, *ha);
2399 if (!new_mci) {
2400 strcpy(msg, "mci socket got corrupted!");
2401 return -EINVAL;
2402 }
2403
2404 pvt = new_mci->pvt_info;
2405 pci_ha = pvt->pci_ha;
2406 pci_read_config_dword(pci_ha, tad_dram_rule[0], ®);
2407 tad0 = m->addr <= TAD_LIMIT(reg);
2408
2409 *channel_mask = 1 << channel;
2410 if (pvt->mirror_mode == FULL_MIRRORING ||
2411 (pvt->mirror_mode == ADDR_RANGE_MIRRORING && tad0)) {
2412 *channel_mask |= 1 << ((channel + 2) % 4);
2413 pvt->is_cur_addr_mirrored = true;
2414 } else {
2415 pvt->is_cur_addr_mirrored = false;
2416 }
2417
2418 if (pvt->is_lockstep)
2419 *channel_mask |= 1 << ((channel + 1) % 4);
2420
2421 return 0;
2422 }
2423
2424 /****************************************************************************
2425 Device initialization routines: put/get, init/exit
2426 ****************************************************************************/
2427
2428 /*
2429 * sbridge_put_all_devices 'put' all the devices that we have
2430 * reserved via 'get'
2431 */
sbridge_put_devices(struct sbridge_dev * sbridge_dev)2432 static void sbridge_put_devices(struct sbridge_dev *sbridge_dev)
2433 {
2434 int i;
2435
2436 edac_dbg(0, "\n");
2437 for (i = 0; i < sbridge_dev->n_devs; i++) {
2438 struct pci_dev *pdev = sbridge_dev->pdev[i];
2439 if (!pdev)
2440 continue;
2441 edac_dbg(0, "Removing dev %02x:%02x.%d\n",
2442 pdev->bus->number,
2443 PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn));
2444 pci_dev_put(pdev);
2445 }
2446 }
2447
sbridge_put_all_devices(void)2448 static void sbridge_put_all_devices(void)
2449 {
2450 struct sbridge_dev *sbridge_dev, *tmp;
2451
2452 list_for_each_entry_safe(sbridge_dev, tmp, &sbridge_edac_list, list) {
2453 sbridge_put_devices(sbridge_dev);
2454 free_sbridge_dev(sbridge_dev);
2455 }
2456 }
2457
sbridge_get_onedevice(struct pci_dev ** prev,u8 * num_mc,const struct pci_id_table * table,const unsigned devno,const int multi_bus)2458 static int sbridge_get_onedevice(struct pci_dev **prev,
2459 u8 *num_mc,
2460 const struct pci_id_table *table,
2461 const unsigned devno,
2462 const int multi_bus)
2463 {
2464 struct sbridge_dev *sbridge_dev = NULL;
2465 const struct pci_id_descr *dev_descr = &table->descr[devno];
2466 struct pci_dev *pdev = NULL;
2467 int seg = 0;
2468 u8 bus = 0;
2469 int i = 0;
2470
2471 sbridge_printk(KERN_DEBUG,
2472 "Seeking for: PCI ID %04x:%04x\n",
2473 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
2474
2475 pdev = pci_get_device(PCI_VENDOR_ID_INTEL,
2476 dev_descr->dev_id, *prev);
2477
2478 if (!pdev) {
2479 if (*prev) {
2480 *prev = pdev;
2481 return 0;
2482 }
2483
2484 if (dev_descr->optional)
2485 return 0;
2486
2487 /* if the HA wasn't found */
2488 if (devno == 0)
2489 return -ENODEV;
2490
2491 sbridge_printk(KERN_INFO,
2492 "Device not found: %04x:%04x\n",
2493 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
2494
2495 /* End of list, leave */
2496 return -ENODEV;
2497 }
2498 seg = pci_domain_nr(pdev->bus);
2499 bus = pdev->bus->number;
2500
2501 next_imc:
2502 sbridge_dev = get_sbridge_dev(seg, bus, dev_descr->dom,
2503 multi_bus, sbridge_dev);
2504 if (!sbridge_dev) {
2505 /* If the HA1 wasn't found, don't create EDAC second memory controller */
2506 if (dev_descr->dom == IMC1 && devno != 1) {
2507 edac_dbg(0, "Skip IMC1: %04x:%04x (since HA1 was absent)\n",
2508 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
2509 pci_dev_put(pdev);
2510 return 0;
2511 }
2512
2513 if (dev_descr->dom == SOCK)
2514 goto out_imc;
2515
2516 sbridge_dev = alloc_sbridge_dev(seg, bus, dev_descr->dom, table);
2517 if (!sbridge_dev) {
2518 pci_dev_put(pdev);
2519 return -ENOMEM;
2520 }
2521 (*num_mc)++;
2522 }
2523
2524 if (sbridge_dev->pdev[sbridge_dev->i_devs]) {
2525 sbridge_printk(KERN_ERR,
2526 "Duplicated device for %04x:%04x\n",
2527 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
2528 pci_dev_put(pdev);
2529 return -ENODEV;
2530 }
2531
2532 sbridge_dev->pdev[sbridge_dev->i_devs++] = pdev;
2533
2534 /* pdev belongs to more than one IMC, do extra gets */
2535 if (++i > 1)
2536 pci_dev_get(pdev);
2537
2538 if (dev_descr->dom == SOCK && i < table->n_imcs_per_sock)
2539 goto next_imc;
2540
2541 out_imc:
2542 /* Be sure that the device is enabled */
2543 if (unlikely(pci_enable_device(pdev) < 0)) {
2544 sbridge_printk(KERN_ERR,
2545 "Couldn't enable %04x:%04x\n",
2546 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
2547 return -ENODEV;
2548 }
2549
2550 edac_dbg(0, "Detected %04x:%04x\n",
2551 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
2552
2553 /*
2554 * As stated on drivers/pci/search.c, the reference count for
2555 * @from is always decremented if it is not %NULL. So, as we need
2556 * to get all devices up to null, we need to do a get for the device
2557 */
2558 pci_dev_get(pdev);
2559
2560 *prev = pdev;
2561
2562 return 0;
2563 }
2564
2565 /*
2566 * sbridge_get_all_devices - Find and perform 'get' operation on the MCH's
2567 * devices we want to reference for this driver.
2568 * @num_mc: pointer to the memory controllers count, to be incremented in case
2569 * of success.
2570 * @table: model specific table
2571 *
2572 * returns 0 in case of success or error code
2573 */
sbridge_get_all_devices(u8 * num_mc,const struct pci_id_table * table)2574 static int sbridge_get_all_devices(u8 *num_mc,
2575 const struct pci_id_table *table)
2576 {
2577 int i, rc;
2578 struct pci_dev *pdev = NULL;
2579 int allow_dups = 0;
2580 int multi_bus = 0;
2581
2582 if (table->type == KNIGHTS_LANDING)
2583 allow_dups = multi_bus = 1;
2584 while (table && table->descr) {
2585 for (i = 0; i < table->n_devs_per_sock; i++) {
2586 if (!allow_dups || i == 0 ||
2587 table->descr[i].dev_id !=
2588 table->descr[i-1].dev_id) {
2589 pdev = NULL;
2590 }
2591 do {
2592 rc = sbridge_get_onedevice(&pdev, num_mc,
2593 table, i, multi_bus);
2594 if (rc < 0) {
2595 if (i == 0) {
2596 i = table->n_devs_per_sock;
2597 break;
2598 }
2599 sbridge_put_all_devices();
2600 return -ENODEV;
2601 }
2602 } while (pdev && !allow_dups);
2603 }
2604 table++;
2605 }
2606
2607 return 0;
2608 }
2609
2610 /*
2611 * Device IDs for {SBRIDGE,IBRIDGE,HASWELL,BROADWELL}_IMC_HA0_TAD0 are in
2612 * the format: XXXa. So we can convert from a device to the corresponding
2613 * channel like this
2614 */
2615 #define TAD_DEV_TO_CHAN(dev) (((dev) & 0xf) - 0xa)
2616
sbridge_mci_bind_devs(struct mem_ctl_info * mci,struct sbridge_dev * sbridge_dev)2617 static int sbridge_mci_bind_devs(struct mem_ctl_info *mci,
2618 struct sbridge_dev *sbridge_dev)
2619 {
2620 struct sbridge_pvt *pvt = mci->pvt_info;
2621 struct pci_dev *pdev;
2622 u8 saw_chan_mask = 0;
2623 int i;
2624
2625 for (i = 0; i < sbridge_dev->n_devs; i++) {
2626 pdev = sbridge_dev->pdev[i];
2627 if (!pdev)
2628 continue;
2629
2630 switch (pdev->device) {
2631 case PCI_DEVICE_ID_INTEL_SBRIDGE_SAD0:
2632 pvt->pci_sad0 = pdev;
2633 break;
2634 case PCI_DEVICE_ID_INTEL_SBRIDGE_SAD1:
2635 pvt->pci_sad1 = pdev;
2636 break;
2637 case PCI_DEVICE_ID_INTEL_SBRIDGE_BR:
2638 pvt->pci_br0 = pdev;
2639 break;
2640 case PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_HA0:
2641 pvt->pci_ha = pdev;
2642 break;
2643 case PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TA:
2644 pvt->pci_ta = pdev;
2645 break;
2646 case PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_RAS:
2647 pvt->pci_ras = pdev;
2648 break;
2649 case PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD0:
2650 case PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD1:
2651 case PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD2:
2652 case PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD3:
2653 {
2654 int id = TAD_DEV_TO_CHAN(pdev->device);
2655 pvt->pci_tad[id] = pdev;
2656 saw_chan_mask |= 1 << id;
2657 }
2658 break;
2659 case PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_DDRIO:
2660 pvt->pci_ddrio = pdev;
2661 break;
2662 default:
2663 goto error;
2664 }
2665
2666 edac_dbg(0, "Associated PCI %02x:%02x, bus %d with dev = %p\n",
2667 pdev->vendor, pdev->device,
2668 sbridge_dev->bus,
2669 pdev);
2670 }
2671
2672 /* Check if everything were registered */
2673 if (!pvt->pci_sad0 || !pvt->pci_sad1 || !pvt->pci_ha ||
2674 !pvt->pci_ras || !pvt->pci_ta)
2675 goto enodev;
2676
2677 if (saw_chan_mask != 0x0f)
2678 goto enodev;
2679 return 0;
2680
2681 enodev:
2682 sbridge_printk(KERN_ERR, "Some needed devices are missing\n");
2683 return -ENODEV;
2684
2685 error:
2686 sbridge_printk(KERN_ERR, "Unexpected device %02x:%02x\n",
2687 PCI_VENDOR_ID_INTEL, pdev->device);
2688 return -EINVAL;
2689 }
2690
ibridge_mci_bind_devs(struct mem_ctl_info * mci,struct sbridge_dev * sbridge_dev)2691 static int ibridge_mci_bind_devs(struct mem_ctl_info *mci,
2692 struct sbridge_dev *sbridge_dev)
2693 {
2694 struct sbridge_pvt *pvt = mci->pvt_info;
2695 struct pci_dev *pdev;
2696 u8 saw_chan_mask = 0;
2697 int i;
2698
2699 for (i = 0; i < sbridge_dev->n_devs; i++) {
2700 pdev = sbridge_dev->pdev[i];
2701 if (!pdev)
2702 continue;
2703
2704 switch (pdev->device) {
2705 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0:
2706 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1:
2707 pvt->pci_ha = pdev;
2708 break;
2709 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TA:
2710 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TA:
2711 pvt->pci_ta = pdev;
2712 break;
2713 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_RAS:
2714 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_RAS:
2715 pvt->pci_ras = pdev;
2716 break;
2717 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD0:
2718 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD1:
2719 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD2:
2720 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD3:
2721 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD0:
2722 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD1:
2723 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD2:
2724 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD3:
2725 {
2726 int id = TAD_DEV_TO_CHAN(pdev->device);
2727 pvt->pci_tad[id] = pdev;
2728 saw_chan_mask |= 1 << id;
2729 }
2730 break;
2731 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_2HA_DDRIO0:
2732 pvt->pci_ddrio = pdev;
2733 break;
2734 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_1HA_DDRIO0:
2735 pvt->pci_ddrio = pdev;
2736 break;
2737 case PCI_DEVICE_ID_INTEL_IBRIDGE_SAD:
2738 pvt->pci_sad0 = pdev;
2739 break;
2740 case PCI_DEVICE_ID_INTEL_IBRIDGE_BR0:
2741 pvt->pci_br0 = pdev;
2742 break;
2743 case PCI_DEVICE_ID_INTEL_IBRIDGE_BR1:
2744 pvt->pci_br1 = pdev;
2745 break;
2746 default:
2747 goto error;
2748 }
2749
2750 edac_dbg(0, "Associated PCI %02x.%02d.%d with dev = %p\n",
2751 sbridge_dev->bus,
2752 PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn),
2753 pdev);
2754 }
2755
2756 /* Check if everything were registered */
2757 if (!pvt->pci_sad0 || !pvt->pci_ha || !pvt->pci_br0 ||
2758 !pvt->pci_br1 || !pvt->pci_ras || !pvt->pci_ta)
2759 goto enodev;
2760
2761 if (saw_chan_mask != 0x0f && /* -EN/-EX */
2762 saw_chan_mask != 0x03) /* -EP */
2763 goto enodev;
2764 return 0;
2765
2766 enodev:
2767 sbridge_printk(KERN_ERR, "Some needed devices are missing\n");
2768 return -ENODEV;
2769
2770 error:
2771 sbridge_printk(KERN_ERR,
2772 "Unexpected device %02x:%02x\n", PCI_VENDOR_ID_INTEL,
2773 pdev->device);
2774 return -EINVAL;
2775 }
2776
haswell_mci_bind_devs(struct mem_ctl_info * mci,struct sbridge_dev * sbridge_dev)2777 static int haswell_mci_bind_devs(struct mem_ctl_info *mci,
2778 struct sbridge_dev *sbridge_dev)
2779 {
2780 struct sbridge_pvt *pvt = mci->pvt_info;
2781 struct pci_dev *pdev;
2782 u8 saw_chan_mask = 0;
2783 int i;
2784
2785 /* there's only one device per system; not tied to any bus */
2786 if (pvt->info.pci_vtd == NULL)
2787 /* result will be checked later */
2788 pvt->info.pci_vtd = pci_get_device(PCI_VENDOR_ID_INTEL,
2789 PCI_DEVICE_ID_INTEL_HASWELL_IMC_VTD_MISC,
2790 NULL);
2791
2792 for (i = 0; i < sbridge_dev->n_devs; i++) {
2793 pdev = sbridge_dev->pdev[i];
2794 if (!pdev)
2795 continue;
2796
2797 switch (pdev->device) {
2798 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_CBO_SAD0:
2799 pvt->pci_sad0 = pdev;
2800 break;
2801 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_CBO_SAD1:
2802 pvt->pci_sad1 = pdev;
2803 break;
2804 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0:
2805 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1:
2806 pvt->pci_ha = pdev;
2807 break;
2808 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TA:
2809 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TA:
2810 pvt->pci_ta = pdev;
2811 break;
2812 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TM:
2813 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TM:
2814 pvt->pci_ras = pdev;
2815 break;
2816 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD0:
2817 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD1:
2818 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD2:
2819 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD3:
2820 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD0:
2821 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD1:
2822 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD2:
2823 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD3:
2824 {
2825 int id = TAD_DEV_TO_CHAN(pdev->device);
2826 pvt->pci_tad[id] = pdev;
2827 saw_chan_mask |= 1 << id;
2828 }
2829 break;
2830 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO0:
2831 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO1:
2832 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO2:
2833 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO3:
2834 if (!pvt->pci_ddrio)
2835 pvt->pci_ddrio = pdev;
2836 break;
2837 default:
2838 break;
2839 }
2840
2841 edac_dbg(0, "Associated PCI %02x.%02d.%d with dev = %p\n",
2842 sbridge_dev->bus,
2843 PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn),
2844 pdev);
2845 }
2846
2847 /* Check if everything were registered */
2848 if (!pvt->pci_sad0 || !pvt->pci_ha || !pvt->pci_sad1 ||
2849 !pvt->pci_ras || !pvt->pci_ta || !pvt->info.pci_vtd)
2850 goto enodev;
2851
2852 if (saw_chan_mask != 0x0f && /* -EN/-EX */
2853 saw_chan_mask != 0x03) /* -EP */
2854 goto enodev;
2855 return 0;
2856
2857 enodev:
2858 sbridge_printk(KERN_ERR, "Some needed devices are missing\n");
2859 return -ENODEV;
2860 }
2861
broadwell_mci_bind_devs(struct mem_ctl_info * mci,struct sbridge_dev * sbridge_dev)2862 static int broadwell_mci_bind_devs(struct mem_ctl_info *mci,
2863 struct sbridge_dev *sbridge_dev)
2864 {
2865 struct sbridge_pvt *pvt = mci->pvt_info;
2866 struct pci_dev *pdev;
2867 u8 saw_chan_mask = 0;
2868 int i;
2869
2870 /* there's only one device per system; not tied to any bus */
2871 if (pvt->info.pci_vtd == NULL)
2872 /* result will be checked later */
2873 pvt->info.pci_vtd = pci_get_device(PCI_VENDOR_ID_INTEL,
2874 PCI_DEVICE_ID_INTEL_BROADWELL_IMC_VTD_MISC,
2875 NULL);
2876
2877 for (i = 0; i < sbridge_dev->n_devs; i++) {
2878 pdev = sbridge_dev->pdev[i];
2879 if (!pdev)
2880 continue;
2881
2882 switch (pdev->device) {
2883 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_CBO_SAD0:
2884 pvt->pci_sad0 = pdev;
2885 break;
2886 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_CBO_SAD1:
2887 pvt->pci_sad1 = pdev;
2888 break;
2889 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0:
2890 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1:
2891 pvt->pci_ha = pdev;
2892 break;
2893 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TA:
2894 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TA:
2895 pvt->pci_ta = pdev;
2896 break;
2897 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TM:
2898 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TM:
2899 pvt->pci_ras = pdev;
2900 break;
2901 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD0:
2902 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD1:
2903 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD2:
2904 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD3:
2905 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD0:
2906 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD1:
2907 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD2:
2908 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD3:
2909 {
2910 int id = TAD_DEV_TO_CHAN(pdev->device);
2911 pvt->pci_tad[id] = pdev;
2912 saw_chan_mask |= 1 << id;
2913 }
2914 break;
2915 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_DDRIO0:
2916 pvt->pci_ddrio = pdev;
2917 break;
2918 default:
2919 break;
2920 }
2921
2922 edac_dbg(0, "Associated PCI %02x.%02d.%d with dev = %p\n",
2923 sbridge_dev->bus,
2924 PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn),
2925 pdev);
2926 }
2927
2928 /* Check if everything were registered */
2929 if (!pvt->pci_sad0 || !pvt->pci_ha || !pvt->pci_sad1 ||
2930 !pvt->pci_ras || !pvt->pci_ta || !pvt->info.pci_vtd)
2931 goto enodev;
2932
2933 if (saw_chan_mask != 0x0f && /* -EN/-EX */
2934 saw_chan_mask != 0x03) /* -EP */
2935 goto enodev;
2936 return 0;
2937
2938 enodev:
2939 sbridge_printk(KERN_ERR, "Some needed devices are missing\n");
2940 return -ENODEV;
2941 }
2942
knl_mci_bind_devs(struct mem_ctl_info * mci,struct sbridge_dev * sbridge_dev)2943 static int knl_mci_bind_devs(struct mem_ctl_info *mci,
2944 struct sbridge_dev *sbridge_dev)
2945 {
2946 struct sbridge_pvt *pvt = mci->pvt_info;
2947 struct pci_dev *pdev;
2948 int dev, func;
2949
2950 int i;
2951 int devidx;
2952
2953 for (i = 0; i < sbridge_dev->n_devs; i++) {
2954 pdev = sbridge_dev->pdev[i];
2955 if (!pdev)
2956 continue;
2957
2958 /* Extract PCI device and function. */
2959 dev = (pdev->devfn >> 3) & 0x1f;
2960 func = pdev->devfn & 0x7;
2961
2962 switch (pdev->device) {
2963 case PCI_DEVICE_ID_INTEL_KNL_IMC_MC:
2964 if (dev == 8)
2965 pvt->knl.pci_mc0 = pdev;
2966 else if (dev == 9)
2967 pvt->knl.pci_mc1 = pdev;
2968 else {
2969 sbridge_printk(KERN_ERR,
2970 "Memory controller in unexpected place! (dev %d, fn %d)\n",
2971 dev, func);
2972 continue;
2973 }
2974 break;
2975
2976 case PCI_DEVICE_ID_INTEL_KNL_IMC_SAD0:
2977 pvt->pci_sad0 = pdev;
2978 break;
2979
2980 case PCI_DEVICE_ID_INTEL_KNL_IMC_SAD1:
2981 pvt->pci_sad1 = pdev;
2982 break;
2983
2984 case PCI_DEVICE_ID_INTEL_KNL_IMC_CHA:
2985 /* There are one of these per tile, and range from
2986 * 1.14.0 to 1.18.5.
2987 */
2988 devidx = ((dev-14)*8)+func;
2989
2990 if (devidx < 0 || devidx >= KNL_MAX_CHAS) {
2991 sbridge_printk(KERN_ERR,
2992 "Caching and Home Agent in unexpected place! (dev %d, fn %d)\n",
2993 dev, func);
2994 continue;
2995 }
2996
2997 WARN_ON(pvt->knl.pci_cha[devidx] != NULL);
2998
2999 pvt->knl.pci_cha[devidx] = pdev;
3000 break;
3001
3002 case PCI_DEVICE_ID_INTEL_KNL_IMC_CHAN:
3003 devidx = -1;
3004
3005 /*
3006 * MC0 channels 0-2 are device 9 function 2-4,
3007 * MC1 channels 3-5 are device 8 function 2-4.
3008 */
3009
3010 if (dev == 9)
3011 devidx = func-2;
3012 else if (dev == 8)
3013 devidx = 3 + (func-2);
3014
3015 if (devidx < 0 || devidx >= KNL_MAX_CHANNELS) {
3016 sbridge_printk(KERN_ERR,
3017 "DRAM Channel Registers in unexpected place! (dev %d, fn %d)\n",
3018 dev, func);
3019 continue;
3020 }
3021
3022 WARN_ON(pvt->knl.pci_channel[devidx] != NULL);
3023 pvt->knl.pci_channel[devidx] = pdev;
3024 break;
3025
3026 case PCI_DEVICE_ID_INTEL_KNL_IMC_TOLHM:
3027 pvt->knl.pci_mc_info = pdev;
3028 break;
3029
3030 case PCI_DEVICE_ID_INTEL_KNL_IMC_TA:
3031 pvt->pci_ta = pdev;
3032 break;
3033
3034 default:
3035 sbridge_printk(KERN_ERR, "Unexpected device %d\n",
3036 pdev->device);
3037 break;
3038 }
3039 }
3040
3041 if (!pvt->knl.pci_mc0 || !pvt->knl.pci_mc1 ||
3042 !pvt->pci_sad0 || !pvt->pci_sad1 ||
3043 !pvt->pci_ta) {
3044 goto enodev;
3045 }
3046
3047 for (i = 0; i < KNL_MAX_CHANNELS; i++) {
3048 if (!pvt->knl.pci_channel[i]) {
3049 sbridge_printk(KERN_ERR, "Missing channel %d\n", i);
3050 goto enodev;
3051 }
3052 }
3053
3054 for (i = 0; i < KNL_MAX_CHAS; i++) {
3055 if (!pvt->knl.pci_cha[i]) {
3056 sbridge_printk(KERN_ERR, "Missing CHA %d\n", i);
3057 goto enodev;
3058 }
3059 }
3060
3061 return 0;
3062
3063 enodev:
3064 sbridge_printk(KERN_ERR, "Some needed devices are missing\n");
3065 return -ENODEV;
3066 }
3067
3068 /****************************************************************************
3069 Error check routines
3070 ****************************************************************************/
3071
3072 /*
3073 * While Sandy Bridge has error count registers, SMI BIOS read values from
3074 * and resets the counters. So, they are not reliable for the OS to read
3075 * from them. So, we have no option but to just trust on whatever MCE is
3076 * telling us about the errors.
3077 */
sbridge_mce_output_error(struct mem_ctl_info * mci,const struct mce * m)3078 static void sbridge_mce_output_error(struct mem_ctl_info *mci,
3079 const struct mce *m)
3080 {
3081 struct mem_ctl_info *new_mci;
3082 struct sbridge_pvt *pvt = mci->pvt_info;
3083 enum hw_event_mc_err_type tp_event;
3084 bool ripv = GET_BITFIELD(m->mcgstatus, 0, 0);
3085 bool overflow = GET_BITFIELD(m->status, 62, 62);
3086 bool uncorrected_error = GET_BITFIELD(m->status, 61, 61);
3087 bool recoverable;
3088 u32 core_err_cnt = GET_BITFIELD(m->status, 38, 52);
3089 u32 mscod = GET_BITFIELD(m->status, 16, 31);
3090 u32 errcode = GET_BITFIELD(m->status, 0, 15);
3091 u32 channel = GET_BITFIELD(m->status, 0, 3);
3092 u32 optypenum = GET_BITFIELD(m->status, 4, 6);
3093 /*
3094 * Bits 5-0 of MCi_MISC give the least significant bit that is valid.
3095 * A value 6 is for cache line aligned address, a value 12 is for page
3096 * aligned address reported by patrol scrubber.
3097 */
3098 u32 lsb = GET_BITFIELD(m->misc, 0, 5);
3099 char *optype, *area_type = "DRAM";
3100 long channel_mask, first_channel;
3101 u8 rank = 0xff, socket, ha;
3102 int rc, dimm;
3103
3104 if (pvt->info.type != SANDY_BRIDGE)
3105 recoverable = true;
3106 else
3107 recoverable = GET_BITFIELD(m->status, 56, 56);
3108
3109 if (uncorrected_error) {
3110 core_err_cnt = 1;
3111 if (ripv) {
3112 tp_event = HW_EVENT_ERR_UNCORRECTED;
3113 } else {
3114 tp_event = HW_EVENT_ERR_FATAL;
3115 }
3116 } else {
3117 tp_event = HW_EVENT_ERR_CORRECTED;
3118 }
3119
3120 /*
3121 * According with Table 15-9 of the Intel Architecture spec vol 3A,
3122 * memory errors should fit in this mask:
3123 * 000f 0000 1mmm cccc (binary)
3124 * where:
3125 * f = Correction Report Filtering Bit. If 1, subsequent errors
3126 * won't be shown
3127 * mmm = error type
3128 * cccc = channel
3129 * If the mask doesn't match, report an error to the parsing logic
3130 */
3131 switch (optypenum) {
3132 case 0:
3133 optype = "generic undef request error";
3134 break;
3135 case 1:
3136 optype = "memory read error";
3137 break;
3138 case 2:
3139 optype = "memory write error";
3140 break;
3141 case 3:
3142 optype = "addr/cmd error";
3143 break;
3144 case 4:
3145 optype = "memory scrubbing error";
3146 break;
3147 default:
3148 optype = "reserved";
3149 break;
3150 }
3151
3152 if (pvt->info.type == KNIGHTS_LANDING) {
3153 if (channel == 14) {
3154 edac_dbg(0, "%s%s err_code:%04x:%04x EDRAM bank %d\n",
3155 overflow ? " OVERFLOW" : "",
3156 (uncorrected_error && recoverable)
3157 ? " recoverable" : "",
3158 mscod, errcode,
3159 m->bank);
3160 } else {
3161 char A = *("A");
3162
3163 /*
3164 * Reported channel is in range 0-2, so we can't map it
3165 * back to mc. To figure out mc we check machine check
3166 * bank register that reported this error.
3167 * bank15 means mc0 and bank16 means mc1.
3168 */
3169 channel = knl_channel_remap(m->bank == 16, channel);
3170 channel_mask = 1 << channel;
3171
3172 snprintf(sb_msg, sizeof(sb_msg),
3173 "%s%s err_code:%04x:%04x channel:%d (DIMM_%c)",
3174 overflow ? " OVERFLOW" : "",
3175 (uncorrected_error && recoverable)
3176 ? " recoverable" : " ",
3177 mscod, errcode, channel, A + channel);
3178 edac_mc_handle_error(tp_event, mci, core_err_cnt,
3179 m->addr >> PAGE_SHIFT, m->addr & ~PAGE_MASK, 0,
3180 channel, 0, -1,
3181 optype, sb_msg);
3182 }
3183 return;
3184 } else if (lsb < 12) {
3185 rc = get_memory_error_data(mci, m->addr, &socket, &ha,
3186 &channel_mask, &rank,
3187 &area_type, sb_msg);
3188 } else {
3189 rc = get_memory_error_data_from_mce(mci, m, &socket, &ha,
3190 &channel_mask, sb_msg);
3191 }
3192
3193 if (rc < 0)
3194 goto err_parsing;
3195 new_mci = get_mci_for_node_id(socket, ha);
3196 if (!new_mci) {
3197 strscpy(sb_msg, "Error: socket got corrupted!");
3198 goto err_parsing;
3199 }
3200 mci = new_mci;
3201 pvt = mci->pvt_info;
3202
3203 first_channel = find_first_bit(&channel_mask, NUM_CHANNELS);
3204
3205 if (rank == 0xff)
3206 dimm = -1;
3207 else if (rank < 4)
3208 dimm = 0;
3209 else if (rank < 8)
3210 dimm = 1;
3211 else
3212 dimm = 2;
3213
3214 /*
3215 * FIXME: On some memory configurations (mirror, lockstep), the
3216 * Memory Controller can't point the error to a single DIMM. The
3217 * EDAC core should be handling the channel mask, in order to point
3218 * to the group of dimm's where the error may be happening.
3219 */
3220 if (!pvt->is_lockstep && !pvt->is_cur_addr_mirrored && !pvt->is_close_pg)
3221 channel = first_channel;
3222 snprintf(sb_msg_full, sizeof(sb_msg_full),
3223 "%s%s area:%s err_code:%04x:%04x socket:%d ha:%d channel_mask:%ld rank:%d %s",
3224 overflow ? " OVERFLOW" : "",
3225 (uncorrected_error && recoverable) ? " recoverable" : "",
3226 area_type,
3227 mscod, errcode,
3228 socket, ha,
3229 channel_mask,
3230 rank, sb_msg);
3231
3232 edac_dbg(0, "%s\n", sb_msg_full);
3233
3234 /* FIXME: need support for channel mask */
3235
3236 if (channel == CHANNEL_UNSPECIFIED)
3237 channel = -1;
3238
3239 /* Call the helper to output message */
3240 edac_mc_handle_error(tp_event, mci, core_err_cnt,
3241 m->addr >> PAGE_SHIFT, m->addr & ~PAGE_MASK, 0,
3242 channel, dimm, -1,
3243 optype, sb_msg_full);
3244 return;
3245 err_parsing:
3246 edac_mc_handle_error(tp_event, mci, core_err_cnt, 0, 0, 0,
3247 -1, -1, -1,
3248 sb_msg, "");
3249
3250 }
3251
3252 /*
3253 * Check that logging is enabled and that this is the right type
3254 * of error for us to handle.
3255 */
sbridge_mce_check_error(struct notifier_block * nb,unsigned long val,void * data)3256 static int sbridge_mce_check_error(struct notifier_block *nb, unsigned long val,
3257 void *data)
3258 {
3259 struct mce *mce = (struct mce *)data;
3260 struct mem_ctl_info *mci;
3261 char *type;
3262
3263 if (mce->kflags & MCE_HANDLED_CEC)
3264 return NOTIFY_DONE;
3265
3266 /*
3267 * Just let mcelog handle it if the error is
3268 * outside the memory controller. A memory error
3269 * is indicated by bit 7 = 1 and bits = 8-11,13-15 = 0.
3270 * bit 12 has an special meaning.
3271 */
3272 if ((mce->status & 0xefff) >> 7 != 1)
3273 return NOTIFY_DONE;
3274
3275 /* Check ADDRV bit in STATUS */
3276 if (!GET_BITFIELD(mce->status, 58, 58))
3277 return NOTIFY_DONE;
3278
3279 /* Check MISCV bit in STATUS */
3280 if (!GET_BITFIELD(mce->status, 59, 59))
3281 return NOTIFY_DONE;
3282
3283 /* Check address type in MISC (physical address only) */
3284 if (GET_BITFIELD(mce->misc, 6, 8) != 2)
3285 return NOTIFY_DONE;
3286
3287 mci = get_mci_for_node_id(mce->socketid, IMC0);
3288 if (!mci)
3289 return NOTIFY_DONE;
3290
3291 if (mce->mcgstatus & MCG_STATUS_MCIP)
3292 type = "Exception";
3293 else
3294 type = "Event";
3295
3296 sbridge_mc_printk(mci, KERN_DEBUG, "HANDLING MCE MEMORY ERROR\n");
3297
3298 sbridge_mc_printk(mci, KERN_DEBUG, "CPU %d: Machine Check %s: %Lx "
3299 "Bank %d: %016Lx\n", mce->extcpu, type,
3300 mce->mcgstatus, mce->bank, mce->status);
3301 sbridge_mc_printk(mci, KERN_DEBUG, "TSC %llx ", mce->tsc);
3302 sbridge_mc_printk(mci, KERN_DEBUG, "ADDR %llx ", mce->addr);
3303 sbridge_mc_printk(mci, KERN_DEBUG, "MISC %llx ", mce->misc);
3304
3305 sbridge_mc_printk(mci, KERN_DEBUG, "PROCESSOR %u:%x TIME %llu SOCKET "
3306 "%u APIC %x\n", mce->cpuvendor, mce->cpuid,
3307 mce->time, mce->socketid, mce->apicid);
3308
3309 sbridge_mce_output_error(mci, mce);
3310
3311 /* Advice mcelog that the error were handled */
3312 mce->kflags |= MCE_HANDLED_EDAC;
3313 return NOTIFY_OK;
3314 }
3315
3316 static struct notifier_block sbridge_mce_dec = {
3317 .notifier_call = sbridge_mce_check_error,
3318 .priority = MCE_PRIO_EDAC,
3319 };
3320
3321 /****************************************************************************
3322 EDAC register/unregister logic
3323 ****************************************************************************/
3324
sbridge_unregister_mci(struct sbridge_dev * sbridge_dev)3325 static void sbridge_unregister_mci(struct sbridge_dev *sbridge_dev)
3326 {
3327 struct mem_ctl_info *mci = sbridge_dev->mci;
3328
3329 if (unlikely(!mci || !mci->pvt_info)) {
3330 edac_dbg(0, "MC: dev = %p\n", &sbridge_dev->pdev[0]->dev);
3331
3332 sbridge_printk(KERN_ERR, "Couldn't find mci handler\n");
3333 return;
3334 }
3335
3336 edac_dbg(0, "MC: mci = %p, dev = %p\n",
3337 mci, &sbridge_dev->pdev[0]->dev);
3338
3339 /* Remove MC sysfs nodes */
3340 edac_mc_del_mc(mci->pdev);
3341
3342 edac_dbg(1, "%s: free mci struct\n", mci->ctl_name);
3343 kfree(mci->ctl_name);
3344 edac_mc_free(mci);
3345 sbridge_dev->mci = NULL;
3346 }
3347
sbridge_register_mci(struct sbridge_dev * sbridge_dev,enum type type)3348 static int sbridge_register_mci(struct sbridge_dev *sbridge_dev, enum type type)
3349 {
3350 struct mem_ctl_info *mci;
3351 struct edac_mc_layer layers[2];
3352 struct sbridge_pvt *pvt;
3353 struct pci_dev *pdev = sbridge_dev->pdev[0];
3354 int rc;
3355
3356 /* allocate a new MC control structure */
3357 layers[0].type = EDAC_MC_LAYER_CHANNEL;
3358 layers[0].size = type == KNIGHTS_LANDING ?
3359 KNL_MAX_CHANNELS : NUM_CHANNELS;
3360 layers[0].is_virt_csrow = false;
3361 layers[1].type = EDAC_MC_LAYER_SLOT;
3362 layers[1].size = type == KNIGHTS_LANDING ? 1 : MAX_DIMMS;
3363 layers[1].is_virt_csrow = true;
3364 mci = edac_mc_alloc(sbridge_dev->mc, ARRAY_SIZE(layers), layers,
3365 sizeof(*pvt));
3366
3367 if (unlikely(!mci))
3368 return -ENOMEM;
3369
3370 edac_dbg(0, "MC: mci = %p, dev = %p\n",
3371 mci, &pdev->dev);
3372
3373 pvt = mci->pvt_info;
3374 memset(pvt, 0, sizeof(*pvt));
3375
3376 /* Associate sbridge_dev and mci for future usage */
3377 pvt->sbridge_dev = sbridge_dev;
3378 sbridge_dev->mci = mci;
3379
3380 mci->mtype_cap = type == KNIGHTS_LANDING ?
3381 MEM_FLAG_DDR4 : MEM_FLAG_DDR3;
3382 mci->edac_ctl_cap = EDAC_FLAG_NONE;
3383 mci->edac_cap = EDAC_FLAG_NONE;
3384 mci->mod_name = EDAC_MOD_STR;
3385 mci->dev_name = pci_name(pdev);
3386 mci->ctl_page_to_phys = NULL;
3387
3388 pvt->info.type = type;
3389 switch (type) {
3390 case IVY_BRIDGE:
3391 pvt->info.rankcfgr = IB_RANK_CFG_A;
3392 pvt->info.get_tolm = ibridge_get_tolm;
3393 pvt->info.get_tohm = ibridge_get_tohm;
3394 pvt->info.dram_rule = ibridge_dram_rule;
3395 pvt->info.get_memory_type = get_memory_type;
3396 pvt->info.get_node_id = get_node_id;
3397 pvt->info.get_ha = ibridge_get_ha;
3398 pvt->info.rir_limit = rir_limit;
3399 pvt->info.sad_limit = sad_limit;
3400 pvt->info.interleave_mode = interleave_mode;
3401 pvt->info.dram_attr = dram_attr;
3402 pvt->info.max_sad = ARRAY_SIZE(ibridge_dram_rule);
3403 pvt->info.interleave_list = ibridge_interleave_list;
3404 pvt->info.interleave_pkg = ibridge_interleave_pkg;
3405 pvt->info.get_width = ibridge_get_width;
3406
3407 /* Store pci devices at mci for faster access */
3408 rc = ibridge_mci_bind_devs(mci, sbridge_dev);
3409 if (unlikely(rc < 0))
3410 goto fail0;
3411 get_source_id(mci);
3412 mci->ctl_name = kasprintf(GFP_KERNEL, "Ivy Bridge SrcID#%d_Ha#%d",
3413 pvt->sbridge_dev->source_id, pvt->sbridge_dev->dom);
3414 break;
3415 case SANDY_BRIDGE:
3416 pvt->info.rankcfgr = SB_RANK_CFG_A;
3417 pvt->info.get_tolm = sbridge_get_tolm;
3418 pvt->info.get_tohm = sbridge_get_tohm;
3419 pvt->info.dram_rule = sbridge_dram_rule;
3420 pvt->info.get_memory_type = get_memory_type;
3421 pvt->info.get_node_id = get_node_id;
3422 pvt->info.get_ha = sbridge_get_ha;
3423 pvt->info.rir_limit = rir_limit;
3424 pvt->info.sad_limit = sad_limit;
3425 pvt->info.interleave_mode = interleave_mode;
3426 pvt->info.dram_attr = dram_attr;
3427 pvt->info.max_sad = ARRAY_SIZE(sbridge_dram_rule);
3428 pvt->info.interleave_list = sbridge_interleave_list;
3429 pvt->info.interleave_pkg = sbridge_interleave_pkg;
3430 pvt->info.get_width = sbridge_get_width;
3431
3432 /* Store pci devices at mci for faster access */
3433 rc = sbridge_mci_bind_devs(mci, sbridge_dev);
3434 if (unlikely(rc < 0))
3435 goto fail0;
3436 get_source_id(mci);
3437 mci->ctl_name = kasprintf(GFP_KERNEL, "Sandy Bridge SrcID#%d_Ha#%d",
3438 pvt->sbridge_dev->source_id, pvt->sbridge_dev->dom);
3439 break;
3440 case HASWELL:
3441 /* rankcfgr isn't used */
3442 pvt->info.get_tolm = haswell_get_tolm;
3443 pvt->info.get_tohm = haswell_get_tohm;
3444 pvt->info.dram_rule = ibridge_dram_rule;
3445 pvt->info.get_memory_type = haswell_get_memory_type;
3446 pvt->info.get_node_id = haswell_get_node_id;
3447 pvt->info.get_ha = ibridge_get_ha;
3448 pvt->info.rir_limit = haswell_rir_limit;
3449 pvt->info.sad_limit = sad_limit;
3450 pvt->info.interleave_mode = interleave_mode;
3451 pvt->info.dram_attr = dram_attr;
3452 pvt->info.max_sad = ARRAY_SIZE(ibridge_dram_rule);
3453 pvt->info.interleave_list = ibridge_interleave_list;
3454 pvt->info.interleave_pkg = ibridge_interleave_pkg;
3455 pvt->info.get_width = ibridge_get_width;
3456
3457 /* Store pci devices at mci for faster access */
3458 rc = haswell_mci_bind_devs(mci, sbridge_dev);
3459 if (unlikely(rc < 0))
3460 goto fail0;
3461 get_source_id(mci);
3462 mci->ctl_name = kasprintf(GFP_KERNEL, "Haswell SrcID#%d_Ha#%d",
3463 pvt->sbridge_dev->source_id, pvt->sbridge_dev->dom);
3464 break;
3465 case BROADWELL:
3466 /* rankcfgr isn't used */
3467 pvt->info.get_tolm = haswell_get_tolm;
3468 pvt->info.get_tohm = haswell_get_tohm;
3469 pvt->info.dram_rule = ibridge_dram_rule;
3470 pvt->info.get_memory_type = haswell_get_memory_type;
3471 pvt->info.get_node_id = haswell_get_node_id;
3472 pvt->info.get_ha = ibridge_get_ha;
3473 pvt->info.rir_limit = haswell_rir_limit;
3474 pvt->info.sad_limit = sad_limit;
3475 pvt->info.interleave_mode = interleave_mode;
3476 pvt->info.dram_attr = dram_attr;
3477 pvt->info.max_sad = ARRAY_SIZE(ibridge_dram_rule);
3478 pvt->info.interleave_list = ibridge_interleave_list;
3479 pvt->info.interleave_pkg = ibridge_interleave_pkg;
3480 pvt->info.get_width = broadwell_get_width;
3481
3482 /* Store pci devices at mci for faster access */
3483 rc = broadwell_mci_bind_devs(mci, sbridge_dev);
3484 if (unlikely(rc < 0))
3485 goto fail0;
3486 get_source_id(mci);
3487 mci->ctl_name = kasprintf(GFP_KERNEL, "Broadwell SrcID#%d_Ha#%d",
3488 pvt->sbridge_dev->source_id, pvt->sbridge_dev->dom);
3489 break;
3490 case KNIGHTS_LANDING:
3491 /* pvt->info.rankcfgr == ??? */
3492 pvt->info.get_tolm = knl_get_tolm;
3493 pvt->info.get_tohm = knl_get_tohm;
3494 pvt->info.dram_rule = knl_dram_rule;
3495 pvt->info.get_memory_type = knl_get_memory_type;
3496 pvt->info.get_node_id = knl_get_node_id;
3497 pvt->info.get_ha = knl_get_ha;
3498 pvt->info.rir_limit = NULL;
3499 pvt->info.sad_limit = knl_sad_limit;
3500 pvt->info.interleave_mode = knl_interleave_mode;
3501 pvt->info.dram_attr = dram_attr_knl;
3502 pvt->info.max_sad = ARRAY_SIZE(knl_dram_rule);
3503 pvt->info.interleave_list = knl_interleave_list;
3504 pvt->info.interleave_pkg = ibridge_interleave_pkg;
3505 pvt->info.get_width = knl_get_width;
3506
3507 rc = knl_mci_bind_devs(mci, sbridge_dev);
3508 if (unlikely(rc < 0))
3509 goto fail0;
3510 get_source_id(mci);
3511 mci->ctl_name = kasprintf(GFP_KERNEL, "Knights Landing SrcID#%d_Ha#%d",
3512 pvt->sbridge_dev->source_id, pvt->sbridge_dev->dom);
3513 break;
3514 }
3515
3516 if (!mci->ctl_name) {
3517 rc = -ENOMEM;
3518 goto fail0;
3519 }
3520
3521 /* Get dimm basic config and the memory layout */
3522 rc = get_dimm_config(mci);
3523 if (rc < 0) {
3524 edac_dbg(0, "MC: failed to get_dimm_config()\n");
3525 goto fail;
3526 }
3527 get_memory_layout(mci);
3528
3529 /* record ptr to the generic device */
3530 mci->pdev = &pdev->dev;
3531
3532 /* add this new MC control structure to EDAC's list of MCs */
3533 if (unlikely(edac_mc_add_mc(mci))) {
3534 edac_dbg(0, "MC: failed edac_mc_add_mc()\n");
3535 rc = -EINVAL;
3536 goto fail;
3537 }
3538
3539 return 0;
3540
3541 fail:
3542 kfree(mci->ctl_name);
3543 fail0:
3544 edac_mc_free(mci);
3545 sbridge_dev->mci = NULL;
3546 return rc;
3547 }
3548
3549 static const struct x86_cpu_id sbridge_cpuids[] = {
3550 X86_MATCH_VFM(INTEL_SANDYBRIDGE_X, &pci_dev_descr_sbridge_table),
3551 X86_MATCH_VFM(INTEL_IVYBRIDGE_X, &pci_dev_descr_ibridge_table),
3552 X86_MATCH_VFM(INTEL_HASWELL_X, &pci_dev_descr_haswell_table),
3553 X86_MATCH_VFM(INTEL_BROADWELL_X, &pci_dev_descr_broadwell_table),
3554 X86_MATCH_VFM(INTEL_BROADWELL_D, &pci_dev_descr_broadwell_table),
3555 X86_MATCH_VFM(INTEL_XEON_PHI_KNL, &pci_dev_descr_knl_table),
3556 X86_MATCH_VFM(INTEL_XEON_PHI_KNM, &pci_dev_descr_knl_table),
3557 { }
3558 };
3559 MODULE_DEVICE_TABLE(x86cpu, sbridge_cpuids);
3560
3561 /*
3562 * sbridge_probe Get all devices and register memory controllers
3563 * present.
3564 * return:
3565 * 0 for FOUND a device
3566 * < 0 for error code
3567 */
3568
sbridge_probe(const struct x86_cpu_id * id)3569 static int sbridge_probe(const struct x86_cpu_id *id)
3570 {
3571 int rc;
3572 u8 mc, num_mc = 0;
3573 struct sbridge_dev *sbridge_dev;
3574 struct pci_id_table *ptable = (struct pci_id_table *)id->driver_data;
3575
3576 /* get the pci devices we want to reserve for our use */
3577 rc = sbridge_get_all_devices(&num_mc, ptable);
3578
3579 if (unlikely(rc < 0)) {
3580 edac_dbg(0, "couldn't get all devices\n");
3581 goto fail0;
3582 }
3583
3584 mc = 0;
3585
3586 list_for_each_entry(sbridge_dev, &sbridge_edac_list, list) {
3587 edac_dbg(0, "Registering MC#%d (%d of %d)\n",
3588 mc, mc + 1, num_mc);
3589
3590 sbridge_dev->mc = mc++;
3591 rc = sbridge_register_mci(sbridge_dev, ptable->type);
3592 if (unlikely(rc < 0))
3593 goto fail1;
3594 }
3595
3596 sbridge_printk(KERN_INFO, "%s\n", SBRIDGE_REVISION);
3597
3598 return 0;
3599
3600 fail1:
3601 list_for_each_entry(sbridge_dev, &sbridge_edac_list, list)
3602 sbridge_unregister_mci(sbridge_dev);
3603
3604 sbridge_put_all_devices();
3605 fail0:
3606 return rc;
3607 }
3608
3609 /*
3610 * sbridge_remove cleanup
3611 *
3612 */
sbridge_remove(void)3613 static void sbridge_remove(void)
3614 {
3615 struct sbridge_dev *sbridge_dev;
3616
3617 edac_dbg(0, "\n");
3618
3619 list_for_each_entry(sbridge_dev, &sbridge_edac_list, list)
3620 sbridge_unregister_mci(sbridge_dev);
3621
3622 /* Release PCI resources */
3623 sbridge_put_all_devices();
3624 }
3625
3626 /*
3627 * sbridge_init Module entry function
3628 * Try to initialize this module for its devices
3629 */
sbridge_init(void)3630 static int __init sbridge_init(void)
3631 {
3632 const struct x86_cpu_id *id;
3633 const char *owner;
3634 int rc;
3635
3636 edac_dbg(2, "\n");
3637
3638 if (ghes_get_devices())
3639 return -EBUSY;
3640
3641 owner = edac_get_owner();
3642 if (owner && strncmp(owner, EDAC_MOD_STR, sizeof(EDAC_MOD_STR)))
3643 return -EBUSY;
3644
3645 if (cpu_feature_enabled(X86_FEATURE_HYPERVISOR))
3646 return -ENODEV;
3647
3648 id = x86_match_cpu(sbridge_cpuids);
3649 if (!id)
3650 return -ENODEV;
3651
3652 /* Ensure that the OPSTATE is set correctly for POLL or NMI */
3653 opstate_init();
3654
3655 rc = sbridge_probe(id);
3656
3657 if (rc >= 0) {
3658 mce_register_decode_chain(&sbridge_mce_dec);
3659 return 0;
3660 }
3661
3662 sbridge_printk(KERN_ERR, "Failed to register device with error %d.\n",
3663 rc);
3664
3665 return rc;
3666 }
3667
3668 /*
3669 * sbridge_exit() Module exit function
3670 * Unregister the driver
3671 */
sbridge_exit(void)3672 static void __exit sbridge_exit(void)
3673 {
3674 edac_dbg(2, "\n");
3675 sbridge_remove();
3676 mce_unregister_decode_chain(&sbridge_mce_dec);
3677 }
3678
3679 module_init(sbridge_init);
3680 module_exit(sbridge_exit);
3681
3682 module_param(edac_op_state, int, 0444);
3683 MODULE_PARM_DESC(edac_op_state, "EDAC Error Reporting state: 0=Poll,1=NMI");
3684
3685 MODULE_LICENSE("GPL");
3686 MODULE_AUTHOR("Mauro Carvalho Chehab");
3687 MODULE_AUTHOR("Red Hat Inc. (https://www.redhat.com)");
3688 MODULE_DESCRIPTION("MC Driver for Intel Sandy Bridge and Ivy Bridge memory controllers - "
3689 SBRIDGE_REVISION);
3690