ALSA: dice: fix kernel NULL pointer dereference due to invalid calculation for array...
[linux/fpc-iii.git] / drivers / edac / sb_edac.c
blob872100215ca00f0f4703c025b8e52e637c85c710
1 /* Intel Sandy Bridge -EN/-EP/-EX Memory Controller kernel module
3 * This driver supports the memory controllers found on the Intel
4 * processor family Sandy Bridge.
6 * This file may be distributed under the terms of the
7 * GNU General Public License version 2 only.
9 * Copyright (c) 2011 by:
10 * Mauro Carvalho Chehab
13 #include <linux/module.h>
14 #include <linux/init.h>
15 #include <linux/pci.h>
16 #include <linux/pci_ids.h>
17 #include <linux/slab.h>
18 #include <linux/delay.h>
19 #include <linux/edac.h>
20 #include <linux/mmzone.h>
21 #include <linux/smp.h>
22 #include <linux/bitmap.h>
23 #include <linux/math64.h>
24 #include <linux/mod_devicetable.h>
25 #include <asm/cpu_device_id.h>
26 #include <asm/intel-family.h>
27 #include <asm/processor.h>
28 #include <asm/mce.h>
30 #include "edac_module.h"
32 /* Static vars */
33 static LIST_HEAD(sbridge_edac_list);
36 * Alter this version for the module when modifications are made
38 #define SBRIDGE_REVISION " Ver: 1.1.2 "
39 #define EDAC_MOD_STR "sb_edac"
42 * Debug macros
44 #define sbridge_printk(level, fmt, arg...) \
45 edac_printk(level, "sbridge", fmt, ##arg)
47 #define sbridge_mc_printk(mci, level, fmt, arg...) \
48 edac_mc_chipset_printk(mci, level, "sbridge", fmt, ##arg)
51 * Get a bit field at register value <v>, from bit <lo> to bit <hi>
53 #define GET_BITFIELD(v, lo, hi) \
54 (((v) & GENMASK_ULL(hi, lo)) >> (lo))
56 /* Devices 12 Function 6, Offsets 0x80 to 0xcc */
57 static const u32 sbridge_dram_rule[] = {
58 0x80, 0x88, 0x90, 0x98, 0xa0,
59 0xa8, 0xb0, 0xb8, 0xc0, 0xc8,
62 static const u32 ibridge_dram_rule[] = {
63 0x60, 0x68, 0x70, 0x78, 0x80,
64 0x88, 0x90, 0x98, 0xa0, 0xa8,
65 0xb0, 0xb8, 0xc0, 0xc8, 0xd0,
66 0xd8, 0xe0, 0xe8, 0xf0, 0xf8,
69 static const u32 knl_dram_rule[] = {
70 0x60, 0x68, 0x70, 0x78, 0x80, /* 0-4 */
71 0x88, 0x90, 0x98, 0xa0, 0xa8, /* 5-9 */
72 0xb0, 0xb8, 0xc0, 0xc8, 0xd0, /* 10-14 */
73 0xd8, 0xe0, 0xe8, 0xf0, 0xf8, /* 15-19 */
74 0x100, 0x108, 0x110, 0x118, /* 20-23 */
77 #define DRAM_RULE_ENABLE(reg) GET_BITFIELD(reg, 0, 0)
78 #define A7MODE(reg) GET_BITFIELD(reg, 26, 26)
80 static char *show_dram_attr(u32 attr)
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";
94 static const u32 sbridge_interleave_list[] = {
95 0x84, 0x8c, 0x94, 0x9c, 0xa4,
96 0xac, 0xb4, 0xbc, 0xc4, 0xcc,
99 static const u32 ibridge_interleave_list[] = {
100 0x64, 0x6c, 0x74, 0x7c, 0x84,
101 0x8c, 0x94, 0x9c, 0xa4, 0xac,
102 0xb4, 0xbc, 0xc4, 0xcc, 0xd4,
103 0xdc, 0xe4, 0xec, 0xf4, 0xfc,
106 static const u32 knl_interleave_list[] = {
107 0x64, 0x6c, 0x74, 0x7c, 0x84, /* 0-4 */
108 0x8c, 0x94, 0x9c, 0xa4, 0xac, /* 5-9 */
109 0xb4, 0xbc, 0xc4, 0xcc, 0xd4, /* 10-14 */
110 0xdc, 0xe4, 0xec, 0xf4, 0xfc, /* 15-19 */
111 0x104, 0x10c, 0x114, 0x11c, /* 20-23 */
114 struct interleave_pkg {
115 unsigned char start;
116 unsigned char end;
119 static const struct interleave_pkg sbridge_interleave_pkg[] = {
120 { 0, 2 },
121 { 3, 5 },
122 { 8, 10 },
123 { 11, 13 },
124 { 16, 18 },
125 { 19, 21 },
126 { 24, 26 },
127 { 27, 29 },
130 static const struct interleave_pkg ibridge_interleave_pkg[] = {
131 { 0, 3 },
132 { 4, 7 },
133 { 8, 11 },
134 { 12, 15 },
135 { 16, 19 },
136 { 20, 23 },
137 { 24, 27 },
138 { 28, 31 },
141 static inline int sad_pkg(const struct interleave_pkg *table, u32 reg,
142 int interleave)
144 return GET_BITFIELD(reg, table[interleave].start,
145 table[interleave].end);
148 /* Devices 12 Function 7 */
150 #define TOLM 0x80
151 #define TOHM 0x84
152 #define HASWELL_TOLM 0xd0
153 #define HASWELL_TOHM_0 0xd4
154 #define HASWELL_TOHM_1 0xd8
155 #define KNL_TOLM 0xd0
156 #define KNL_TOHM_0 0xd4
157 #define KNL_TOHM_1 0xd8
159 #define GET_TOLM(reg) ((GET_BITFIELD(reg, 0, 3) << 28) | 0x3ffffff)
160 #define GET_TOHM(reg) ((GET_BITFIELD(reg, 0, 20) << 25) | 0x3ffffff)
162 /* Device 13 Function 6 */
164 #define SAD_TARGET 0xf0
166 #define SOURCE_ID(reg) GET_BITFIELD(reg, 9, 11)
168 #define SOURCE_ID_KNL(reg) GET_BITFIELD(reg, 12, 14)
170 #define SAD_CONTROL 0xf4
172 /* Device 14 function 0 */
174 static const u32 tad_dram_rule[] = {
175 0x40, 0x44, 0x48, 0x4c,
176 0x50, 0x54, 0x58, 0x5c,
177 0x60, 0x64, 0x68, 0x6c,
179 #define MAX_TAD ARRAY_SIZE(tad_dram_rule)
181 #define TAD_LIMIT(reg) ((GET_BITFIELD(reg, 12, 31) << 26) | 0x3ffffff)
182 #define TAD_SOCK(reg) GET_BITFIELD(reg, 10, 11)
183 #define TAD_CH(reg) GET_BITFIELD(reg, 8, 9)
184 #define TAD_TGT3(reg) GET_BITFIELD(reg, 6, 7)
185 #define TAD_TGT2(reg) GET_BITFIELD(reg, 4, 5)
186 #define TAD_TGT1(reg) GET_BITFIELD(reg, 2, 3)
187 #define TAD_TGT0(reg) GET_BITFIELD(reg, 0, 1)
189 /* Device 15, function 0 */
191 #define MCMTR 0x7c
192 #define KNL_MCMTR 0x624
194 #define IS_ECC_ENABLED(mcmtr) GET_BITFIELD(mcmtr, 2, 2)
195 #define IS_LOCKSTEP_ENABLED(mcmtr) GET_BITFIELD(mcmtr, 1, 1)
196 #define IS_CLOSE_PG(mcmtr) GET_BITFIELD(mcmtr, 0, 0)
198 /* Device 15, function 1 */
200 #define RASENABLES 0xac
201 #define IS_MIRROR_ENABLED(reg) GET_BITFIELD(reg, 0, 0)
203 /* Device 15, functions 2-5 */
205 static const int mtr_regs[] = {
206 0x80, 0x84, 0x88,
209 static const int knl_mtr_reg = 0xb60;
211 #define RANK_DISABLE(mtr) GET_BITFIELD(mtr, 16, 19)
212 #define IS_DIMM_PRESENT(mtr) GET_BITFIELD(mtr, 14, 14)
213 #define RANK_CNT_BITS(mtr) GET_BITFIELD(mtr, 12, 13)
214 #define RANK_WIDTH_BITS(mtr) GET_BITFIELD(mtr, 2, 4)
215 #define COL_WIDTH_BITS(mtr) GET_BITFIELD(mtr, 0, 1)
217 static const u32 tad_ch_nilv_offset[] = {
218 0x90, 0x94, 0x98, 0x9c,
219 0xa0, 0xa4, 0xa8, 0xac,
220 0xb0, 0xb4, 0xb8, 0xbc,
222 #define CHN_IDX_OFFSET(reg) GET_BITFIELD(reg, 28, 29)
223 #define TAD_OFFSET(reg) (GET_BITFIELD(reg, 6, 25) << 26)
225 static const u32 rir_way_limit[] = {
226 0x108, 0x10c, 0x110, 0x114, 0x118,
228 #define MAX_RIR_RANGES ARRAY_SIZE(rir_way_limit)
230 #define IS_RIR_VALID(reg) GET_BITFIELD(reg, 31, 31)
231 #define RIR_WAY(reg) GET_BITFIELD(reg, 28, 29)
233 #define MAX_RIR_WAY 8
235 static const u32 rir_offset[MAX_RIR_RANGES][MAX_RIR_WAY] = {
236 { 0x120, 0x124, 0x128, 0x12c, 0x130, 0x134, 0x138, 0x13c },
237 { 0x140, 0x144, 0x148, 0x14c, 0x150, 0x154, 0x158, 0x15c },
238 { 0x160, 0x164, 0x168, 0x16c, 0x170, 0x174, 0x178, 0x17c },
239 { 0x180, 0x184, 0x188, 0x18c, 0x190, 0x194, 0x198, 0x19c },
240 { 0x1a0, 0x1a4, 0x1a8, 0x1ac, 0x1b0, 0x1b4, 0x1b8, 0x1bc },
243 #define RIR_RNK_TGT(type, reg) (((type) == BROADWELL) ? \
244 GET_BITFIELD(reg, 20, 23) : GET_BITFIELD(reg, 16, 19))
246 #define RIR_OFFSET(type, reg) (((type) == HASWELL || (type) == BROADWELL) ? \
247 GET_BITFIELD(reg, 2, 15) : GET_BITFIELD(reg, 2, 14))
249 /* Device 16, functions 2-7 */
252 * FIXME: Implement the error count reads directly
255 static const u32 correrrcnt[] = {
256 0x104, 0x108, 0x10c, 0x110,
259 #define RANK_ODD_OV(reg) GET_BITFIELD(reg, 31, 31)
260 #define RANK_ODD_ERR_CNT(reg) GET_BITFIELD(reg, 16, 30)
261 #define RANK_EVEN_OV(reg) GET_BITFIELD(reg, 15, 15)
262 #define RANK_EVEN_ERR_CNT(reg) GET_BITFIELD(reg, 0, 14)
264 static const u32 correrrthrsld[] = {
265 0x11c, 0x120, 0x124, 0x128,
268 #define RANK_ODD_ERR_THRSLD(reg) GET_BITFIELD(reg, 16, 30)
269 #define RANK_EVEN_ERR_THRSLD(reg) GET_BITFIELD(reg, 0, 14)
272 /* Device 17, function 0 */
274 #define SB_RANK_CFG_A 0x0328
276 #define IB_RANK_CFG_A 0x0320
279 * sbridge structs
282 #define NUM_CHANNELS 6 /* Max channels per MC */
283 #define MAX_DIMMS 3 /* Max DIMMS per channel */
284 #define KNL_MAX_CHAS 38 /* KNL max num. of Cache Home Agents */
285 #define KNL_MAX_CHANNELS 6 /* KNL max num. of PCI channels */
286 #define KNL_MAX_EDCS 8 /* Embedded DRAM controllers */
287 #define CHANNEL_UNSPECIFIED 0xf /* Intel IA32 SDM 15-14 */
289 enum type {
290 SANDY_BRIDGE,
291 IVY_BRIDGE,
292 HASWELL,
293 BROADWELL,
294 KNIGHTS_LANDING,
297 enum domain {
298 IMC0 = 0,
299 IMC1,
300 SOCK,
303 enum mirroring_mode {
304 NON_MIRRORING,
305 ADDR_RANGE_MIRRORING,
306 FULL_MIRRORING,
309 struct sbridge_pvt;
310 struct sbridge_info {
311 enum type type;
312 u32 mcmtr;
313 u32 rankcfgr;
314 u64 (*get_tolm)(struct sbridge_pvt *pvt);
315 u64 (*get_tohm)(struct sbridge_pvt *pvt);
316 u64 (*rir_limit)(u32 reg);
317 u64 (*sad_limit)(u32 reg);
318 u32 (*interleave_mode)(u32 reg);
319 u32 (*dram_attr)(u32 reg);
320 const u32 *dram_rule;
321 const u32 *interleave_list;
322 const struct interleave_pkg *interleave_pkg;
323 u8 max_sad;
324 u8 max_interleave;
325 u8 (*get_node_id)(struct sbridge_pvt *pvt);
326 enum mem_type (*get_memory_type)(struct sbridge_pvt *pvt);
327 enum dev_type (*get_width)(struct sbridge_pvt *pvt, u32 mtr);
328 struct pci_dev *pci_vtd;
331 struct sbridge_channel {
332 u32 ranks;
333 u32 dimms;
336 struct pci_id_descr {
337 int dev_id;
338 int optional;
339 enum domain dom;
342 struct pci_id_table {
343 const struct pci_id_descr *descr;
344 int n_devs_per_imc;
345 int n_devs_per_sock;
346 int n_imcs_per_sock;
347 enum type type;
350 struct sbridge_dev {
351 struct list_head list;
352 u8 bus, mc;
353 u8 node_id, source_id;
354 struct pci_dev **pdev;
355 enum domain dom;
356 int n_devs;
357 int i_devs;
358 struct mem_ctl_info *mci;
361 struct knl_pvt {
362 struct pci_dev *pci_cha[KNL_MAX_CHAS];
363 struct pci_dev *pci_channel[KNL_MAX_CHANNELS];
364 struct pci_dev *pci_mc0;
365 struct pci_dev *pci_mc1;
366 struct pci_dev *pci_mc0_misc;
367 struct pci_dev *pci_mc1_misc;
368 struct pci_dev *pci_mc_info; /* tolm, tohm */
371 struct sbridge_pvt {
372 /* Devices per socket */
373 struct pci_dev *pci_ddrio;
374 struct pci_dev *pci_sad0, *pci_sad1;
375 struct pci_dev *pci_br0, *pci_br1;
376 /* Devices per memory controller */
377 struct pci_dev *pci_ha, *pci_ta, *pci_ras;
378 struct pci_dev *pci_tad[NUM_CHANNELS];
380 struct sbridge_dev *sbridge_dev;
382 struct sbridge_info info;
383 struct sbridge_channel channel[NUM_CHANNELS];
385 /* Memory type detection */
386 bool is_cur_addr_mirrored, is_lockstep, is_close_pg;
387 bool is_chan_hash;
388 enum mirroring_mode mirror_mode;
390 /* Memory description */
391 u64 tolm, tohm;
392 struct knl_pvt knl;
395 #define PCI_DESCR(device_id, opt, domain) \
396 .dev_id = (device_id), \
397 .optional = opt, \
398 .dom = domain
400 static const struct pci_id_descr pci_dev_descr_sbridge[] = {
401 /* Processor Home Agent */
402 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_HA0, 0, IMC0) },
404 /* Memory controller */
405 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TA, 0, IMC0) },
406 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_RAS, 0, IMC0) },
407 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD0, 0, IMC0) },
408 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD1, 0, IMC0) },
409 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD2, 0, IMC0) },
410 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD3, 0, IMC0) },
411 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_DDRIO, 1, SOCK) },
413 /* System Address Decoder */
414 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_SAD0, 0, SOCK) },
415 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_SAD1, 0, SOCK) },
417 /* Broadcast Registers */
418 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_BR, 0, SOCK) },
421 #define PCI_ID_TABLE_ENTRY(A, N, M, T) { \
422 .descr = A, \
423 .n_devs_per_imc = N, \
424 .n_devs_per_sock = ARRAY_SIZE(A), \
425 .n_imcs_per_sock = M, \
426 .type = T \
429 static const struct pci_id_table pci_dev_descr_sbridge_table[] = {
430 PCI_ID_TABLE_ENTRY(pci_dev_descr_sbridge, ARRAY_SIZE(pci_dev_descr_sbridge), 1, SANDY_BRIDGE),
431 {0,} /* 0 terminated list. */
434 /* This changes depending if 1HA or 2HA:
435 * 1HA:
436 * 0x0eb8 (17.0) is DDRIO0
437 * 2HA:
438 * 0x0ebc (17.4) is DDRIO0
440 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_1HA_DDRIO0 0x0eb8
441 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_2HA_DDRIO0 0x0ebc
443 /* pci ids */
444 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0 0x0ea0
445 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TA 0x0ea8
446 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_RAS 0x0e71
447 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD0 0x0eaa
448 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD1 0x0eab
449 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD2 0x0eac
450 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD3 0x0ead
451 #define PCI_DEVICE_ID_INTEL_IBRIDGE_SAD 0x0ec8
452 #define PCI_DEVICE_ID_INTEL_IBRIDGE_BR0 0x0ec9
453 #define PCI_DEVICE_ID_INTEL_IBRIDGE_BR1 0x0eca
454 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1 0x0e60
455 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TA 0x0e68
456 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_RAS 0x0e79
457 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD0 0x0e6a
458 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD1 0x0e6b
459 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD2 0x0e6c
460 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD3 0x0e6d
462 static const struct pci_id_descr pci_dev_descr_ibridge[] = {
463 /* Processor Home Agent */
464 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0, 0, IMC0) },
465 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1, 1, IMC1) },
467 /* Memory controller */
468 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TA, 0, IMC0) },
469 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_RAS, 0, IMC0) },
470 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD0, 0, IMC0) },
471 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD1, 0, IMC0) },
472 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD2, 0, IMC0) },
473 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD3, 0, IMC0) },
475 /* Optional, mode 2HA */
476 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TA, 1, IMC1) },
477 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_RAS, 1, IMC1) },
478 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD0, 1, IMC1) },
479 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD1, 1, IMC1) },
480 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD2, 1, IMC1) },
481 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD3, 1, IMC1) },
483 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_1HA_DDRIO0, 1, SOCK) },
484 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_2HA_DDRIO0, 1, SOCK) },
486 /* System Address Decoder */
487 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_SAD, 0, SOCK) },
489 /* Broadcast Registers */
490 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_BR0, 1, SOCK) },
491 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_BR1, 0, SOCK) },
495 static const struct pci_id_table pci_dev_descr_ibridge_table[] = {
496 PCI_ID_TABLE_ENTRY(pci_dev_descr_ibridge, 12, 2, IVY_BRIDGE),
497 {0,} /* 0 terminated list. */
500 /* Haswell support */
501 /* EN processor:
502 * - 1 IMC
503 * - 3 DDR3 channels, 2 DPC per channel
504 * EP processor:
505 * - 1 or 2 IMC
506 * - 4 DDR4 channels, 3 DPC per channel
507 * EP 4S processor:
508 * - 2 IMC
509 * - 4 DDR4 channels, 3 DPC per channel
510 * EX processor:
511 * - 2 IMC
512 * - each IMC interfaces with a SMI 2 channel
513 * - each SMI channel interfaces with a scalable memory buffer
514 * - each scalable memory buffer supports 4 DDR3/DDR4 channels, 3 DPC
516 #define HASWELL_DDRCRCLKCONTROLS 0xa10 /* Ditto on Broadwell */
517 #define HASWELL_HASYSDEFEATURE2 0x84
518 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_VTD_MISC 0x2f28
519 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0 0x2fa0
520 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1 0x2f60
521 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TA 0x2fa8
522 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TM 0x2f71
523 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TA 0x2f68
524 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TM 0x2f79
525 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_CBO_SAD0 0x2ffc
526 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_CBO_SAD1 0x2ffd
527 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD0 0x2faa
528 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD1 0x2fab
529 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD2 0x2fac
530 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD3 0x2fad
531 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD0 0x2f6a
532 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD1 0x2f6b
533 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD2 0x2f6c
534 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD3 0x2f6d
535 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO0 0x2fbd
536 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO1 0x2fbf
537 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO2 0x2fb9
538 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO3 0x2fbb
539 static const struct pci_id_descr pci_dev_descr_haswell[] = {
540 /* first item must be the HA */
541 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0, 0, IMC0) },
542 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1, 1, IMC1) },
544 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TA, 0, IMC0) },
545 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TM, 0, IMC0) },
546 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD0, 0, IMC0) },
547 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD1, 0, IMC0) },
548 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD2, 1, IMC0) },
549 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD3, 1, IMC0) },
551 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TA, 1, IMC1) },
552 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TM, 1, IMC1) },
553 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD0, 1, IMC1) },
554 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD1, 1, IMC1) },
555 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD2, 1, IMC1) },
556 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD3, 1, IMC1) },
558 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_CBO_SAD0, 0, SOCK) },
559 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_CBO_SAD1, 0, SOCK) },
560 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO0, 1, SOCK) },
561 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO1, 1, SOCK) },
562 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO2, 1, SOCK) },
563 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO3, 1, SOCK) },
566 static const struct pci_id_table pci_dev_descr_haswell_table[] = {
567 PCI_ID_TABLE_ENTRY(pci_dev_descr_haswell, 13, 2, HASWELL),
568 {0,} /* 0 terminated list. */
571 /* Knight's Landing Support */
573 * KNL's memory channels are swizzled between memory controllers.
574 * MC0 is mapped to CH3,4,5 and MC1 is mapped to CH0,1,2
576 #define knl_channel_remap(mc, chan) ((mc) ? (chan) : (chan) + 3)
578 /* Memory controller, TAD tables, error injection - 2-8-0, 2-9-0 (2 of these) */
579 #define PCI_DEVICE_ID_INTEL_KNL_IMC_MC 0x7840
580 /* DRAM channel stuff; bank addrs, dimmmtr, etc.. 2-8-2 - 2-9-4 (6 of these) */
581 #define PCI_DEVICE_ID_INTEL_KNL_IMC_CHAN 0x7843
582 /* kdrwdbu TAD limits/offsets, MCMTR - 2-10-1, 2-11-1 (2 of these) */
583 #define PCI_DEVICE_ID_INTEL_KNL_IMC_TA 0x7844
584 /* CHA broadcast registers, dram rules - 1-29-0 (1 of these) */
585 #define PCI_DEVICE_ID_INTEL_KNL_IMC_SAD0 0x782a
586 /* SAD target - 1-29-1 (1 of these) */
587 #define PCI_DEVICE_ID_INTEL_KNL_IMC_SAD1 0x782b
588 /* Caching / Home Agent */
589 #define PCI_DEVICE_ID_INTEL_KNL_IMC_CHA 0x782c
590 /* Device with TOLM and TOHM, 0-5-0 (1 of these) */
591 #define PCI_DEVICE_ID_INTEL_KNL_IMC_TOLHM 0x7810
594 * KNL differs from SB, IB, and Haswell in that it has multiple
595 * instances of the same device with the same device ID, so we handle that
596 * by creating as many copies in the table as we expect to find.
597 * (Like device ID must be grouped together.)
600 static const struct pci_id_descr pci_dev_descr_knl[] = {
601 [0 ... 1] = { PCI_DESCR(PCI_DEVICE_ID_INTEL_KNL_IMC_MC, 0, IMC0)},
602 [2 ... 7] = { PCI_DESCR(PCI_DEVICE_ID_INTEL_KNL_IMC_CHAN, 0, IMC0) },
603 [8] = { PCI_DESCR(PCI_DEVICE_ID_INTEL_KNL_IMC_TA, 0, IMC0) },
604 [9] = { PCI_DESCR(PCI_DEVICE_ID_INTEL_KNL_IMC_TOLHM, 0, IMC0) },
605 [10] = { PCI_DESCR(PCI_DEVICE_ID_INTEL_KNL_IMC_SAD0, 0, SOCK) },
606 [11] = { PCI_DESCR(PCI_DEVICE_ID_INTEL_KNL_IMC_SAD1, 0, SOCK) },
607 [12 ... 49] = { PCI_DESCR(PCI_DEVICE_ID_INTEL_KNL_IMC_CHA, 0, SOCK) },
610 static const struct pci_id_table pci_dev_descr_knl_table[] = {
611 PCI_ID_TABLE_ENTRY(pci_dev_descr_knl, ARRAY_SIZE(pci_dev_descr_knl), 1, KNIGHTS_LANDING),
612 {0,}
616 * Broadwell support
618 * DE processor:
619 * - 1 IMC
620 * - 2 DDR3 channels, 2 DPC per channel
621 * EP processor:
622 * - 1 or 2 IMC
623 * - 4 DDR4 channels, 3 DPC per channel
624 * EP 4S processor:
625 * - 2 IMC
626 * - 4 DDR4 channels, 3 DPC per channel
627 * EX processor:
628 * - 2 IMC
629 * - each IMC interfaces with a SMI 2 channel
630 * - each SMI channel interfaces with a scalable memory buffer
631 * - each scalable memory buffer supports 4 DDR3/DDR4 channels, 3 DPC
633 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_VTD_MISC 0x6f28
634 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0 0x6fa0
635 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1 0x6f60
636 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TA 0x6fa8
637 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TM 0x6f71
638 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TA 0x6f68
639 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TM 0x6f79
640 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_CBO_SAD0 0x6ffc
641 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_CBO_SAD1 0x6ffd
642 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD0 0x6faa
643 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD1 0x6fab
644 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD2 0x6fac
645 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD3 0x6fad
646 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD0 0x6f6a
647 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD1 0x6f6b
648 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD2 0x6f6c
649 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD3 0x6f6d
650 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_DDRIO0 0x6faf
652 static const struct pci_id_descr pci_dev_descr_broadwell[] = {
653 /* first item must be the HA */
654 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0, 0, IMC0) },
655 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1, 1, IMC1) },
657 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TA, 0, IMC0) },
658 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TM, 0, IMC0) },
659 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD0, 0, IMC0) },
660 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD1, 0, IMC0) },
661 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD2, 1, IMC0) },
662 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD3, 1, IMC0) },
664 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TA, 1, IMC1) },
665 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TM, 1, IMC1) },
666 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD0, 1, IMC1) },
667 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD1, 1, IMC1) },
668 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD2, 1, IMC1) },
669 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD3, 1, IMC1) },
671 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_CBO_SAD0, 0, SOCK) },
672 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_CBO_SAD1, 0, SOCK) },
673 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_DDRIO0, 1, SOCK) },
676 static const struct pci_id_table pci_dev_descr_broadwell_table[] = {
677 PCI_ID_TABLE_ENTRY(pci_dev_descr_broadwell, 10, 2, BROADWELL),
678 {0,} /* 0 terminated list. */
682 /****************************************************************************
683 Ancillary status routines
684 ****************************************************************************/
686 static inline int numrank(enum type type, u32 mtr)
688 int ranks = (1 << RANK_CNT_BITS(mtr));
689 int max = 4;
691 if (type == HASWELL || type == BROADWELL || type == KNIGHTS_LANDING)
692 max = 8;
694 if (ranks > max) {
695 edac_dbg(0, "Invalid number of ranks: %d (max = %i) raw value = %x (%04x)\n",
696 ranks, max, (unsigned int)RANK_CNT_BITS(mtr), mtr);
697 return -EINVAL;
700 return ranks;
703 static inline int numrow(u32 mtr)
705 int rows = (RANK_WIDTH_BITS(mtr) + 12);
707 if (rows < 13 || rows > 18) {
708 edac_dbg(0, "Invalid number of rows: %d (should be between 14 and 17) raw value = %x (%04x)\n",
709 rows, (unsigned int)RANK_WIDTH_BITS(mtr), mtr);
710 return -EINVAL;
713 return 1 << rows;
716 static inline int numcol(u32 mtr)
718 int cols = (COL_WIDTH_BITS(mtr) + 10);
720 if (cols > 12) {
721 edac_dbg(0, "Invalid number of cols: %d (max = 4) raw value = %x (%04x)\n",
722 cols, (unsigned int)COL_WIDTH_BITS(mtr), mtr);
723 return -EINVAL;
726 return 1 << cols;
729 static struct sbridge_dev *get_sbridge_dev(u8 bus, enum domain dom, int multi_bus,
730 struct sbridge_dev *prev)
732 struct sbridge_dev *sbridge_dev;
735 * If we have devices scattered across several busses that pertain
736 * to the same memory controller, we'll lump them all together.
738 if (multi_bus) {
739 return list_first_entry_or_null(&sbridge_edac_list,
740 struct sbridge_dev, list);
743 sbridge_dev = list_entry(prev ? prev->list.next
744 : sbridge_edac_list.next, struct sbridge_dev, list);
746 list_for_each_entry_from(sbridge_dev, &sbridge_edac_list, list) {
747 if (sbridge_dev->bus == bus && (dom == SOCK || dom == sbridge_dev->dom))
748 return sbridge_dev;
751 return NULL;
754 static struct sbridge_dev *alloc_sbridge_dev(u8 bus, enum domain dom,
755 const struct pci_id_table *table)
757 struct sbridge_dev *sbridge_dev;
759 sbridge_dev = kzalloc(sizeof(*sbridge_dev), GFP_KERNEL);
760 if (!sbridge_dev)
761 return NULL;
763 sbridge_dev->pdev = kcalloc(table->n_devs_per_imc,
764 sizeof(*sbridge_dev->pdev),
765 GFP_KERNEL);
766 if (!sbridge_dev->pdev) {
767 kfree(sbridge_dev);
768 return NULL;
771 sbridge_dev->bus = bus;
772 sbridge_dev->dom = dom;
773 sbridge_dev->n_devs = table->n_devs_per_imc;
774 list_add_tail(&sbridge_dev->list, &sbridge_edac_list);
776 return sbridge_dev;
779 static void free_sbridge_dev(struct sbridge_dev *sbridge_dev)
781 list_del(&sbridge_dev->list);
782 kfree(sbridge_dev->pdev);
783 kfree(sbridge_dev);
786 static u64 sbridge_get_tolm(struct sbridge_pvt *pvt)
788 u32 reg;
790 /* Address range is 32:28 */
791 pci_read_config_dword(pvt->pci_sad1, TOLM, &reg);
792 return GET_TOLM(reg);
795 static u64 sbridge_get_tohm(struct sbridge_pvt *pvt)
797 u32 reg;
799 pci_read_config_dword(pvt->pci_sad1, TOHM, &reg);
800 return GET_TOHM(reg);
803 static u64 ibridge_get_tolm(struct sbridge_pvt *pvt)
805 u32 reg;
807 pci_read_config_dword(pvt->pci_br1, TOLM, &reg);
809 return GET_TOLM(reg);
812 static u64 ibridge_get_tohm(struct sbridge_pvt *pvt)
814 u32 reg;
816 pci_read_config_dword(pvt->pci_br1, TOHM, &reg);
818 return GET_TOHM(reg);
821 static u64 rir_limit(u32 reg)
823 return ((u64)GET_BITFIELD(reg, 1, 10) << 29) | 0x1fffffff;
826 static u64 sad_limit(u32 reg)
828 return (GET_BITFIELD(reg, 6, 25) << 26) | 0x3ffffff;
831 static u32 interleave_mode(u32 reg)
833 return GET_BITFIELD(reg, 1, 1);
836 static u32 dram_attr(u32 reg)
838 return GET_BITFIELD(reg, 2, 3);
841 static u64 knl_sad_limit(u32 reg)
843 return (GET_BITFIELD(reg, 7, 26) << 26) | 0x3ffffff;
846 static u32 knl_interleave_mode(u32 reg)
848 return GET_BITFIELD(reg, 1, 2);
851 static const char * const knl_intlv_mode[] = {
852 "[8:6]", "[10:8]", "[14:12]", "[32:30]"
855 static const char *get_intlv_mode_str(u32 reg, enum type t)
857 if (t == KNIGHTS_LANDING)
858 return knl_intlv_mode[knl_interleave_mode(reg)];
859 else
860 return interleave_mode(reg) ? "[8:6]" : "[8:6]XOR[18:16]";
863 static u32 dram_attr_knl(u32 reg)
865 return GET_BITFIELD(reg, 3, 4);
869 static enum mem_type get_memory_type(struct sbridge_pvt *pvt)
871 u32 reg;
872 enum mem_type mtype;
874 if (pvt->pci_ddrio) {
875 pci_read_config_dword(pvt->pci_ddrio, pvt->info.rankcfgr,
876 &reg);
877 if (GET_BITFIELD(reg, 11, 11))
878 /* FIXME: Can also be LRDIMM */
879 mtype = MEM_RDDR3;
880 else
881 mtype = MEM_DDR3;
882 } else
883 mtype = MEM_UNKNOWN;
885 return mtype;
888 static enum mem_type haswell_get_memory_type(struct sbridge_pvt *pvt)
890 u32 reg;
891 bool registered = false;
892 enum mem_type mtype = MEM_UNKNOWN;
894 if (!pvt->pci_ddrio)
895 goto out;
897 pci_read_config_dword(pvt->pci_ddrio,
898 HASWELL_DDRCRCLKCONTROLS, &reg);
899 /* Is_Rdimm */
900 if (GET_BITFIELD(reg, 16, 16))
901 registered = true;
903 pci_read_config_dword(pvt->pci_ta, MCMTR, &reg);
904 if (GET_BITFIELD(reg, 14, 14)) {
905 if (registered)
906 mtype = MEM_RDDR4;
907 else
908 mtype = MEM_DDR4;
909 } else {
910 if (registered)
911 mtype = MEM_RDDR3;
912 else
913 mtype = MEM_DDR3;
916 out:
917 return mtype;
920 static enum dev_type knl_get_width(struct sbridge_pvt *pvt, u32 mtr)
922 /* for KNL value is fixed */
923 return DEV_X16;
926 static enum dev_type sbridge_get_width(struct sbridge_pvt *pvt, u32 mtr)
928 /* there's no way to figure out */
929 return DEV_UNKNOWN;
932 static enum dev_type __ibridge_get_width(u32 mtr)
934 enum dev_type type;
936 switch (mtr) {
937 case 3:
938 type = DEV_UNKNOWN;
939 break;
940 case 2:
941 type = DEV_X16;
942 break;
943 case 1:
944 type = DEV_X8;
945 break;
946 case 0:
947 type = DEV_X4;
948 break;
951 return type;
954 static enum dev_type ibridge_get_width(struct sbridge_pvt *pvt, u32 mtr)
957 * ddr3_width on the documentation but also valid for DDR4 on
958 * Haswell
960 return __ibridge_get_width(GET_BITFIELD(mtr, 7, 8));
963 static enum dev_type broadwell_get_width(struct sbridge_pvt *pvt, u32 mtr)
965 /* ddr3_width on the documentation but also valid for DDR4 */
966 return __ibridge_get_width(GET_BITFIELD(mtr, 8, 9));
969 static enum mem_type knl_get_memory_type(struct sbridge_pvt *pvt)
971 /* DDR4 RDIMMS and LRDIMMS are supported */
972 return MEM_RDDR4;
975 static u8 get_node_id(struct sbridge_pvt *pvt)
977 u32 reg;
978 pci_read_config_dword(pvt->pci_br0, SAD_CONTROL, &reg);
979 return GET_BITFIELD(reg, 0, 2);
982 static u8 haswell_get_node_id(struct sbridge_pvt *pvt)
984 u32 reg;
986 pci_read_config_dword(pvt->pci_sad1, SAD_CONTROL, &reg);
987 return GET_BITFIELD(reg, 0, 3);
990 static u8 knl_get_node_id(struct sbridge_pvt *pvt)
992 u32 reg;
994 pci_read_config_dword(pvt->pci_sad1, SAD_CONTROL, &reg);
995 return GET_BITFIELD(reg, 0, 2);
999 static u64 haswell_get_tolm(struct sbridge_pvt *pvt)
1001 u32 reg;
1003 pci_read_config_dword(pvt->info.pci_vtd, HASWELL_TOLM, &reg);
1004 return (GET_BITFIELD(reg, 26, 31) << 26) | 0x3ffffff;
1007 static u64 haswell_get_tohm(struct sbridge_pvt *pvt)
1009 u64 rc;
1010 u32 reg;
1012 pci_read_config_dword(pvt->info.pci_vtd, HASWELL_TOHM_0, &reg);
1013 rc = GET_BITFIELD(reg, 26, 31);
1014 pci_read_config_dword(pvt->info.pci_vtd, HASWELL_TOHM_1, &reg);
1015 rc = ((reg << 6) | rc) << 26;
1017 return rc | 0x1ffffff;
1020 static u64 knl_get_tolm(struct sbridge_pvt *pvt)
1022 u32 reg;
1024 pci_read_config_dword(pvt->knl.pci_mc_info, KNL_TOLM, &reg);
1025 return (GET_BITFIELD(reg, 26, 31) << 26) | 0x3ffffff;
1028 static u64 knl_get_tohm(struct sbridge_pvt *pvt)
1030 u64 rc;
1031 u32 reg_lo, reg_hi;
1033 pci_read_config_dword(pvt->knl.pci_mc_info, KNL_TOHM_0, &reg_lo);
1034 pci_read_config_dword(pvt->knl.pci_mc_info, KNL_TOHM_1, &reg_hi);
1035 rc = ((u64)reg_hi << 32) | reg_lo;
1036 return rc | 0x3ffffff;
1040 static u64 haswell_rir_limit(u32 reg)
1042 return (((u64)GET_BITFIELD(reg, 1, 11) + 1) << 29) - 1;
1045 static inline u8 sad_pkg_socket(u8 pkg)
1047 /* on Ivy Bridge, nodeID is SASS, where A is HA and S is node id */
1048 return ((pkg >> 3) << 2) | (pkg & 0x3);
1051 static inline u8 sad_pkg_ha(u8 pkg)
1053 return (pkg >> 2) & 0x1;
1056 static int haswell_chan_hash(int idx, u64 addr)
1058 int i;
1061 * XOR even bits from 12:26 to bit0 of idx,
1062 * odd bits from 13:27 to bit1
1064 for (i = 12; i < 28; i += 2)
1065 idx ^= (addr >> i) & 3;
1067 return idx;
1070 /* Low bits of TAD limit, and some metadata. */
1071 static const u32 knl_tad_dram_limit_lo[] = {
1072 0x400, 0x500, 0x600, 0x700,
1073 0x800, 0x900, 0xa00, 0xb00,
1076 /* Low bits of TAD offset. */
1077 static const u32 knl_tad_dram_offset_lo[] = {
1078 0x404, 0x504, 0x604, 0x704,
1079 0x804, 0x904, 0xa04, 0xb04,
1082 /* High 16 bits of TAD limit and offset. */
1083 static const u32 knl_tad_dram_hi[] = {
1084 0x408, 0x508, 0x608, 0x708,
1085 0x808, 0x908, 0xa08, 0xb08,
1088 /* Number of ways a tad entry is interleaved. */
1089 static const u32 knl_tad_ways[] = {
1090 8, 6, 4, 3, 2, 1,
1094 * Retrieve the n'th Target Address Decode table entry
1095 * from the memory controller's TAD table.
1097 * @pvt: driver private data
1098 * @entry: which entry you want to retrieve
1099 * @mc: which memory controller (0 or 1)
1100 * @offset: output tad range offset
1101 * @limit: output address of first byte above tad range
1102 * @ways: output number of interleave ways
1104 * The offset value has curious semantics. It's a sort of running total
1105 * of the sizes of all the memory regions that aren't mapped in this
1106 * tad table.
1108 static int knl_get_tad(const struct sbridge_pvt *pvt,
1109 const int entry,
1110 const int mc,
1111 u64 *offset,
1112 u64 *limit,
1113 int *ways)
1115 u32 reg_limit_lo, reg_offset_lo, reg_hi;
1116 struct pci_dev *pci_mc;
1117 int way_id;
1119 switch (mc) {
1120 case 0:
1121 pci_mc = pvt->knl.pci_mc0;
1122 break;
1123 case 1:
1124 pci_mc = pvt->knl.pci_mc1;
1125 break;
1126 default:
1127 WARN_ON(1);
1128 return -EINVAL;
1131 pci_read_config_dword(pci_mc,
1132 knl_tad_dram_limit_lo[entry], &reg_limit_lo);
1133 pci_read_config_dword(pci_mc,
1134 knl_tad_dram_offset_lo[entry], &reg_offset_lo);
1135 pci_read_config_dword(pci_mc,
1136 knl_tad_dram_hi[entry], &reg_hi);
1138 /* Is this TAD entry enabled? */
1139 if (!GET_BITFIELD(reg_limit_lo, 0, 0))
1140 return -ENODEV;
1142 way_id = GET_BITFIELD(reg_limit_lo, 3, 5);
1144 if (way_id < ARRAY_SIZE(knl_tad_ways)) {
1145 *ways = knl_tad_ways[way_id];
1146 } else {
1147 *ways = 0;
1148 sbridge_printk(KERN_ERR,
1149 "Unexpected value %d in mc_tad_limit_lo wayness field\n",
1150 way_id);
1151 return -ENODEV;
1155 * The least significant 6 bits of base and limit are truncated.
1156 * For limit, we fill the missing bits with 1s.
1158 *offset = ((u64) GET_BITFIELD(reg_offset_lo, 6, 31) << 6) |
1159 ((u64) GET_BITFIELD(reg_hi, 0, 15) << 32);
1160 *limit = ((u64) GET_BITFIELD(reg_limit_lo, 6, 31) << 6) | 63 |
1161 ((u64) GET_BITFIELD(reg_hi, 16, 31) << 32);
1163 return 0;
1166 /* Determine which memory controller is responsible for a given channel. */
1167 static int knl_channel_mc(int channel)
1169 WARN_ON(channel < 0 || channel >= 6);
1171 return channel < 3 ? 1 : 0;
1175 * Get the Nth entry from EDC_ROUTE_TABLE register.
1176 * (This is the per-tile mapping of logical interleave targets to
1177 * physical EDC modules.)
1179 * entry 0: 0:2
1180 * 1: 3:5
1181 * 2: 6:8
1182 * 3: 9:11
1183 * 4: 12:14
1184 * 5: 15:17
1185 * 6: 18:20
1186 * 7: 21:23
1187 * reserved: 24:31
1189 static u32 knl_get_edc_route(int entry, u32 reg)
1191 WARN_ON(entry >= KNL_MAX_EDCS);
1192 return GET_BITFIELD(reg, entry*3, (entry*3)+2);
1196 * Get the Nth entry from MC_ROUTE_TABLE register.
1197 * (This is the per-tile mapping of logical interleave targets to
1198 * physical DRAM channels modules.)
1200 * entry 0: mc 0:2 channel 18:19
1201 * 1: mc 3:5 channel 20:21
1202 * 2: mc 6:8 channel 22:23
1203 * 3: mc 9:11 channel 24:25
1204 * 4: mc 12:14 channel 26:27
1205 * 5: mc 15:17 channel 28:29
1206 * reserved: 30:31
1208 * Though we have 3 bits to identify the MC, we should only see
1209 * the values 0 or 1.
1212 static u32 knl_get_mc_route(int entry, u32 reg)
1214 int mc, chan;
1216 WARN_ON(entry >= KNL_MAX_CHANNELS);
1218 mc = GET_BITFIELD(reg, entry*3, (entry*3)+2);
1219 chan = GET_BITFIELD(reg, (entry*2) + 18, (entry*2) + 18 + 1);
1221 return knl_channel_remap(mc, chan);
1225 * Render the EDC_ROUTE register in human-readable form.
1226 * Output string s should be at least KNL_MAX_EDCS*2 bytes.
1228 static void knl_show_edc_route(u32 reg, char *s)
1230 int i;
1232 for (i = 0; i < KNL_MAX_EDCS; i++) {
1233 s[i*2] = knl_get_edc_route(i, reg) + '0';
1234 s[i*2+1] = '-';
1237 s[KNL_MAX_EDCS*2 - 1] = '\0';
1241 * Render the MC_ROUTE register in human-readable form.
1242 * Output string s should be at least KNL_MAX_CHANNELS*2 bytes.
1244 static void knl_show_mc_route(u32 reg, char *s)
1246 int i;
1248 for (i = 0; i < KNL_MAX_CHANNELS; i++) {
1249 s[i*2] = knl_get_mc_route(i, reg) + '0';
1250 s[i*2+1] = '-';
1253 s[KNL_MAX_CHANNELS*2 - 1] = '\0';
1256 #define KNL_EDC_ROUTE 0xb8
1257 #define KNL_MC_ROUTE 0xb4
1259 /* Is this dram rule backed by regular DRAM in flat mode? */
1260 #define KNL_EDRAM(reg) GET_BITFIELD(reg, 29, 29)
1262 /* Is this dram rule cached? */
1263 #define KNL_CACHEABLE(reg) GET_BITFIELD(reg, 28, 28)
1265 /* Is this rule backed by edc ? */
1266 #define KNL_EDRAM_ONLY(reg) GET_BITFIELD(reg, 29, 29)
1268 /* Is this rule backed by DRAM, cacheable in EDRAM? */
1269 #define KNL_CACHEABLE(reg) GET_BITFIELD(reg, 28, 28)
1271 /* Is this rule mod3? */
1272 #define KNL_MOD3(reg) GET_BITFIELD(reg, 27, 27)
1275 * Figure out how big our RAM modules are.
1277 * The DIMMMTR register in KNL doesn't tell us the size of the DIMMs, so we
1278 * have to figure this out from the SAD rules, interleave lists, route tables,
1279 * and TAD rules.
1281 * SAD rules can have holes in them (e.g. the 3G-4G hole), so we have to
1282 * inspect the TAD rules to figure out how large the SAD regions really are.
1284 * When we know the real size of a SAD region and how many ways it's
1285 * interleaved, we know the individual contribution of each channel to
1286 * TAD is size/ways.
1288 * Finally, we have to check whether each channel participates in each SAD
1289 * region.
1291 * Fortunately, KNL only supports one DIMM per channel, so once we know how
1292 * much memory the channel uses, we know the DIMM is at least that large.
1293 * (The BIOS might possibly choose not to map all available memory, in which
1294 * case we will underreport the size of the DIMM.)
1296 * In theory, we could try to determine the EDC sizes as well, but that would
1297 * only work in flat mode, not in cache mode.
1299 * @mc_sizes: Output sizes of channels (must have space for KNL_MAX_CHANNELS
1300 * elements)
1302 static int knl_get_dimm_capacity(struct sbridge_pvt *pvt, u64 *mc_sizes)
1304 u64 sad_base, sad_size, sad_limit = 0;
1305 u64 tad_base, tad_size, tad_limit, tad_deadspace, tad_livespace;
1306 int sad_rule = 0;
1307 int tad_rule = 0;
1308 int intrlv_ways, tad_ways;
1309 u32 first_pkg, pkg;
1310 int i;
1311 u64 sad_actual_size[2]; /* sad size accounting for holes, per mc */
1312 u32 dram_rule, interleave_reg;
1313 u32 mc_route_reg[KNL_MAX_CHAS];
1314 u32 edc_route_reg[KNL_MAX_CHAS];
1315 int edram_only;
1316 char edc_route_string[KNL_MAX_EDCS*2];
1317 char mc_route_string[KNL_MAX_CHANNELS*2];
1318 int cur_reg_start;
1319 int mc;
1320 int channel;
1321 int participants[KNL_MAX_CHANNELS];
1323 for (i = 0; i < KNL_MAX_CHANNELS; i++)
1324 mc_sizes[i] = 0;
1326 /* Read the EDC route table in each CHA. */
1327 cur_reg_start = 0;
1328 for (i = 0; i < KNL_MAX_CHAS; i++) {
1329 pci_read_config_dword(pvt->knl.pci_cha[i],
1330 KNL_EDC_ROUTE, &edc_route_reg[i]);
1332 if (i > 0 && edc_route_reg[i] != edc_route_reg[i-1]) {
1333 knl_show_edc_route(edc_route_reg[i-1],
1334 edc_route_string);
1335 if (cur_reg_start == i-1)
1336 edac_dbg(0, "edc route table for CHA %d: %s\n",
1337 cur_reg_start, edc_route_string);
1338 else
1339 edac_dbg(0, "edc route table for CHA %d-%d: %s\n",
1340 cur_reg_start, i-1, edc_route_string);
1341 cur_reg_start = i;
1344 knl_show_edc_route(edc_route_reg[i-1], edc_route_string);
1345 if (cur_reg_start == i-1)
1346 edac_dbg(0, "edc route table for CHA %d: %s\n",
1347 cur_reg_start, edc_route_string);
1348 else
1349 edac_dbg(0, "edc route table for CHA %d-%d: %s\n",
1350 cur_reg_start, i-1, edc_route_string);
1352 /* Read the MC route table in each CHA. */
1353 cur_reg_start = 0;
1354 for (i = 0; i < KNL_MAX_CHAS; i++) {
1355 pci_read_config_dword(pvt->knl.pci_cha[i],
1356 KNL_MC_ROUTE, &mc_route_reg[i]);
1358 if (i > 0 && mc_route_reg[i] != mc_route_reg[i-1]) {
1359 knl_show_mc_route(mc_route_reg[i-1], mc_route_string);
1360 if (cur_reg_start == i-1)
1361 edac_dbg(0, "mc route table for CHA %d: %s\n",
1362 cur_reg_start, mc_route_string);
1363 else
1364 edac_dbg(0, "mc route table for CHA %d-%d: %s\n",
1365 cur_reg_start, i-1, mc_route_string);
1366 cur_reg_start = i;
1369 knl_show_mc_route(mc_route_reg[i-1], mc_route_string);
1370 if (cur_reg_start == i-1)
1371 edac_dbg(0, "mc route table for CHA %d: %s\n",
1372 cur_reg_start, mc_route_string);
1373 else
1374 edac_dbg(0, "mc route table for CHA %d-%d: %s\n",
1375 cur_reg_start, i-1, mc_route_string);
1377 /* Process DRAM rules */
1378 for (sad_rule = 0; sad_rule < pvt->info.max_sad; sad_rule++) {
1379 /* previous limit becomes the new base */
1380 sad_base = sad_limit;
1382 pci_read_config_dword(pvt->pci_sad0,
1383 pvt->info.dram_rule[sad_rule], &dram_rule);
1385 if (!DRAM_RULE_ENABLE(dram_rule))
1386 break;
1388 edram_only = KNL_EDRAM_ONLY(dram_rule);
1390 sad_limit = pvt->info.sad_limit(dram_rule)+1;
1391 sad_size = sad_limit - sad_base;
1393 pci_read_config_dword(pvt->pci_sad0,
1394 pvt->info.interleave_list[sad_rule], &interleave_reg);
1397 * Find out how many ways this dram rule is interleaved.
1398 * We stop when we see the first channel again.
1400 first_pkg = sad_pkg(pvt->info.interleave_pkg,
1401 interleave_reg, 0);
1402 for (intrlv_ways = 1; intrlv_ways < 8; intrlv_ways++) {
1403 pkg = sad_pkg(pvt->info.interleave_pkg,
1404 interleave_reg, intrlv_ways);
1406 if ((pkg & 0x8) == 0) {
1408 * 0 bit means memory is non-local,
1409 * which KNL doesn't support
1411 edac_dbg(0, "Unexpected interleave target %d\n",
1412 pkg);
1413 return -1;
1416 if (pkg == first_pkg)
1417 break;
1419 if (KNL_MOD3(dram_rule))
1420 intrlv_ways *= 3;
1422 edac_dbg(3, "dram rule %d (base 0x%llx, limit 0x%llx), %d way interleave%s\n",
1423 sad_rule,
1424 sad_base,
1425 sad_limit,
1426 intrlv_ways,
1427 edram_only ? ", EDRAM" : "");
1430 * Find out how big the SAD region really is by iterating
1431 * over TAD tables (SAD regions may contain holes).
1432 * Each memory controller might have a different TAD table, so
1433 * we have to look at both.
1435 * Livespace is the memory that's mapped in this TAD table,
1436 * deadspace is the holes (this could be the MMIO hole, or it
1437 * could be memory that's mapped by the other TAD table but
1438 * not this one).
1440 for (mc = 0; mc < 2; mc++) {
1441 sad_actual_size[mc] = 0;
1442 tad_livespace = 0;
1443 for (tad_rule = 0;
1444 tad_rule < ARRAY_SIZE(
1445 knl_tad_dram_limit_lo);
1446 tad_rule++) {
1447 if (knl_get_tad(pvt,
1448 tad_rule,
1450 &tad_deadspace,
1451 &tad_limit,
1452 &tad_ways))
1453 break;
1455 tad_size = (tad_limit+1) -
1456 (tad_livespace + tad_deadspace);
1457 tad_livespace += tad_size;
1458 tad_base = (tad_limit+1) - tad_size;
1460 if (tad_base < sad_base) {
1461 if (tad_limit > sad_base)
1462 edac_dbg(0, "TAD region overlaps lower SAD boundary -- TAD tables may be configured incorrectly.\n");
1463 } else if (tad_base < sad_limit) {
1464 if (tad_limit+1 > sad_limit) {
1465 edac_dbg(0, "TAD region overlaps upper SAD boundary -- TAD tables may be configured incorrectly.\n");
1466 } else {
1467 /* TAD region is completely inside SAD region */
1468 edac_dbg(3, "TAD region %d 0x%llx - 0x%llx (%lld bytes) table%d\n",
1469 tad_rule, tad_base,
1470 tad_limit, tad_size,
1471 mc);
1472 sad_actual_size[mc] += tad_size;
1475 tad_base = tad_limit+1;
1479 for (mc = 0; mc < 2; mc++) {
1480 edac_dbg(3, " total TAD DRAM footprint in table%d : 0x%llx (%lld bytes)\n",
1481 mc, sad_actual_size[mc], sad_actual_size[mc]);
1484 /* Ignore EDRAM rule */
1485 if (edram_only)
1486 continue;
1488 /* Figure out which channels participate in interleave. */
1489 for (channel = 0; channel < KNL_MAX_CHANNELS; channel++)
1490 participants[channel] = 0;
1492 /* For each channel, does at least one CHA have
1493 * this channel mapped to the given target?
1495 for (channel = 0; channel < KNL_MAX_CHANNELS; channel++) {
1496 int target;
1497 int cha;
1499 for (target = 0; target < KNL_MAX_CHANNELS; target++) {
1500 for (cha = 0; cha < KNL_MAX_CHAS; cha++) {
1501 if (knl_get_mc_route(target,
1502 mc_route_reg[cha]) == channel
1503 && !participants[channel]) {
1504 participants[channel] = 1;
1505 break;
1511 for (channel = 0; channel < KNL_MAX_CHANNELS; channel++) {
1512 mc = knl_channel_mc(channel);
1513 if (participants[channel]) {
1514 edac_dbg(4, "mc channel %d contributes %lld bytes via sad entry %d\n",
1515 channel,
1516 sad_actual_size[mc]/intrlv_ways,
1517 sad_rule);
1518 mc_sizes[channel] +=
1519 sad_actual_size[mc]/intrlv_ways;
1524 return 0;
1527 static void get_source_id(struct mem_ctl_info *mci)
1529 struct sbridge_pvt *pvt = mci->pvt_info;
1530 u32 reg;
1532 if (pvt->info.type == HASWELL || pvt->info.type == BROADWELL ||
1533 pvt->info.type == KNIGHTS_LANDING)
1534 pci_read_config_dword(pvt->pci_sad1, SAD_TARGET, &reg);
1535 else
1536 pci_read_config_dword(pvt->pci_br0, SAD_TARGET, &reg);
1538 if (pvt->info.type == KNIGHTS_LANDING)
1539 pvt->sbridge_dev->source_id = SOURCE_ID_KNL(reg);
1540 else
1541 pvt->sbridge_dev->source_id = SOURCE_ID(reg);
1544 static int __populate_dimms(struct mem_ctl_info *mci,
1545 u64 knl_mc_sizes[KNL_MAX_CHANNELS],
1546 enum edac_type mode)
1548 struct sbridge_pvt *pvt = mci->pvt_info;
1549 int channels = pvt->info.type == KNIGHTS_LANDING ? KNL_MAX_CHANNELS
1550 : NUM_CHANNELS;
1551 unsigned int i, j, banks, ranks, rows, cols, npages;
1552 struct dimm_info *dimm;
1553 enum mem_type mtype;
1554 u64 size;
1556 mtype = pvt->info.get_memory_type(pvt);
1557 if (mtype == MEM_RDDR3 || mtype == MEM_RDDR4)
1558 edac_dbg(0, "Memory is registered\n");
1559 else if (mtype == MEM_UNKNOWN)
1560 edac_dbg(0, "Cannot determine memory type\n");
1561 else
1562 edac_dbg(0, "Memory is unregistered\n");
1564 if (mtype == MEM_DDR4 || mtype == MEM_RDDR4)
1565 banks = 16;
1566 else
1567 banks = 8;
1569 for (i = 0; i < channels; i++) {
1570 u32 mtr;
1572 int max_dimms_per_channel;
1574 if (pvt->info.type == KNIGHTS_LANDING) {
1575 max_dimms_per_channel = 1;
1576 if (!pvt->knl.pci_channel[i])
1577 continue;
1578 } else {
1579 max_dimms_per_channel = ARRAY_SIZE(mtr_regs);
1580 if (!pvt->pci_tad[i])
1581 continue;
1584 for (j = 0; j < max_dimms_per_channel; j++) {
1585 dimm = EDAC_DIMM_PTR(mci->layers, mci->dimms, mci->n_layers, i, j, 0);
1586 if (pvt->info.type == KNIGHTS_LANDING) {
1587 pci_read_config_dword(pvt->knl.pci_channel[i],
1588 knl_mtr_reg, &mtr);
1589 } else {
1590 pci_read_config_dword(pvt->pci_tad[i],
1591 mtr_regs[j], &mtr);
1593 edac_dbg(4, "Channel #%d MTR%d = %x\n", i, j, mtr);
1594 if (IS_DIMM_PRESENT(mtr)) {
1595 if (!IS_ECC_ENABLED(pvt->info.mcmtr)) {
1596 sbridge_printk(KERN_ERR, "CPU SrcID #%d, Ha #%d, Channel #%d has DIMMs, but ECC is disabled\n",
1597 pvt->sbridge_dev->source_id,
1598 pvt->sbridge_dev->dom, i);
1599 return -ENODEV;
1601 pvt->channel[i].dimms++;
1603 ranks = numrank(pvt->info.type, mtr);
1605 if (pvt->info.type == KNIGHTS_LANDING) {
1606 /* For DDR4, this is fixed. */
1607 cols = 1 << 10;
1608 rows = knl_mc_sizes[i] /
1609 ((u64) cols * ranks * banks * 8);
1610 } else {
1611 rows = numrow(mtr);
1612 cols = numcol(mtr);
1615 size = ((u64)rows * cols * banks * ranks) >> (20 - 3);
1616 npages = MiB_TO_PAGES(size);
1618 edac_dbg(0, "mc#%d: ha %d channel %d, dimm %d, %lld Mb (%d pages) bank: %d, rank: %d, row: %#x, col: %#x\n",
1619 pvt->sbridge_dev->mc, pvt->sbridge_dev->dom, i, j,
1620 size, npages,
1621 banks, ranks, rows, cols);
1623 dimm->nr_pages = npages;
1624 dimm->grain = 32;
1625 dimm->dtype = pvt->info.get_width(pvt, mtr);
1626 dimm->mtype = mtype;
1627 dimm->edac_mode = mode;
1628 snprintf(dimm->label, sizeof(dimm->label),
1629 "CPU_SrcID#%u_Ha#%u_Chan#%u_DIMM#%u",
1630 pvt->sbridge_dev->source_id, pvt->sbridge_dev->dom, i, j);
1635 return 0;
1638 static int get_dimm_config(struct mem_ctl_info *mci)
1640 struct sbridge_pvt *pvt = mci->pvt_info;
1641 u64 knl_mc_sizes[KNL_MAX_CHANNELS];
1642 enum edac_type mode;
1643 u32 reg;
1645 pvt->sbridge_dev->node_id = pvt->info.get_node_id(pvt);
1646 edac_dbg(0, "mc#%d: Node ID: %d, source ID: %d\n",
1647 pvt->sbridge_dev->mc,
1648 pvt->sbridge_dev->node_id,
1649 pvt->sbridge_dev->source_id);
1651 /* KNL doesn't support mirroring or lockstep,
1652 * and is always closed page
1654 if (pvt->info.type == KNIGHTS_LANDING) {
1655 mode = EDAC_S4ECD4ED;
1656 pvt->mirror_mode = NON_MIRRORING;
1657 pvt->is_cur_addr_mirrored = false;
1659 if (knl_get_dimm_capacity(pvt, knl_mc_sizes) != 0)
1660 return -1;
1661 if (pci_read_config_dword(pvt->pci_ta, KNL_MCMTR, &pvt->info.mcmtr)) {
1662 edac_dbg(0, "Failed to read KNL_MCMTR register\n");
1663 return -ENODEV;
1665 } else {
1666 if (pvt->info.type == HASWELL || pvt->info.type == BROADWELL) {
1667 if (pci_read_config_dword(pvt->pci_ha, HASWELL_HASYSDEFEATURE2, &reg)) {
1668 edac_dbg(0, "Failed to read HASWELL_HASYSDEFEATURE2 register\n");
1669 return -ENODEV;
1671 pvt->is_chan_hash = GET_BITFIELD(reg, 21, 21);
1672 if (GET_BITFIELD(reg, 28, 28)) {
1673 pvt->mirror_mode = ADDR_RANGE_MIRRORING;
1674 edac_dbg(0, "Address range partial memory mirroring is enabled\n");
1675 goto next;
1678 if (pci_read_config_dword(pvt->pci_ras, RASENABLES, &reg)) {
1679 edac_dbg(0, "Failed to read RASENABLES register\n");
1680 return -ENODEV;
1682 if (IS_MIRROR_ENABLED(reg)) {
1683 pvt->mirror_mode = FULL_MIRRORING;
1684 edac_dbg(0, "Full memory mirroring is enabled\n");
1685 } else {
1686 pvt->mirror_mode = NON_MIRRORING;
1687 edac_dbg(0, "Memory mirroring is disabled\n");
1690 next:
1691 if (pci_read_config_dword(pvt->pci_ta, MCMTR, &pvt->info.mcmtr)) {
1692 edac_dbg(0, "Failed to read MCMTR register\n");
1693 return -ENODEV;
1695 if (IS_LOCKSTEP_ENABLED(pvt->info.mcmtr)) {
1696 edac_dbg(0, "Lockstep is enabled\n");
1697 mode = EDAC_S8ECD8ED;
1698 pvt->is_lockstep = true;
1699 } else {
1700 edac_dbg(0, "Lockstep is disabled\n");
1701 mode = EDAC_S4ECD4ED;
1702 pvt->is_lockstep = false;
1704 if (IS_CLOSE_PG(pvt->info.mcmtr)) {
1705 edac_dbg(0, "address map is on closed page mode\n");
1706 pvt->is_close_pg = true;
1707 } else {
1708 edac_dbg(0, "address map is on open page mode\n");
1709 pvt->is_close_pg = false;
1713 return __populate_dimms(mci, knl_mc_sizes, mode);
1716 static void get_memory_layout(const struct mem_ctl_info *mci)
1718 struct sbridge_pvt *pvt = mci->pvt_info;
1719 int i, j, k, n_sads, n_tads, sad_interl;
1720 u32 reg;
1721 u64 limit, prv = 0;
1722 u64 tmp_mb;
1723 u32 gb, mb;
1724 u32 rir_way;
1727 * Step 1) Get TOLM/TOHM ranges
1730 pvt->tolm = pvt->info.get_tolm(pvt);
1731 tmp_mb = (1 + pvt->tolm) >> 20;
1733 gb = div_u64_rem(tmp_mb, 1024, &mb);
1734 edac_dbg(0, "TOLM: %u.%03u GB (0x%016Lx)\n",
1735 gb, (mb*1000)/1024, (u64)pvt->tolm);
1737 /* Address range is already 45:25 */
1738 pvt->tohm = pvt->info.get_tohm(pvt);
1739 tmp_mb = (1 + pvt->tohm) >> 20;
1741 gb = div_u64_rem(tmp_mb, 1024, &mb);
1742 edac_dbg(0, "TOHM: %u.%03u GB (0x%016Lx)\n",
1743 gb, (mb*1000)/1024, (u64)pvt->tohm);
1746 * Step 2) Get SAD range and SAD Interleave list
1747 * TAD registers contain the interleave wayness. However, it
1748 * seems simpler to just discover it indirectly, with the
1749 * algorithm bellow.
1751 prv = 0;
1752 for (n_sads = 0; n_sads < pvt->info.max_sad; n_sads++) {
1753 /* SAD_LIMIT Address range is 45:26 */
1754 pci_read_config_dword(pvt->pci_sad0, pvt->info.dram_rule[n_sads],
1755 &reg);
1756 limit = pvt->info.sad_limit(reg);
1758 if (!DRAM_RULE_ENABLE(reg))
1759 continue;
1761 if (limit <= prv)
1762 break;
1764 tmp_mb = (limit + 1) >> 20;
1765 gb = div_u64_rem(tmp_mb, 1024, &mb);
1766 edac_dbg(0, "SAD#%d %s up to %u.%03u GB (0x%016Lx) Interleave: %s reg=0x%08x\n",
1767 n_sads,
1768 show_dram_attr(pvt->info.dram_attr(reg)),
1769 gb, (mb*1000)/1024,
1770 ((u64)tmp_mb) << 20L,
1771 get_intlv_mode_str(reg, pvt->info.type),
1772 reg);
1773 prv = limit;
1775 pci_read_config_dword(pvt->pci_sad0, pvt->info.interleave_list[n_sads],
1776 &reg);
1777 sad_interl = sad_pkg(pvt->info.interleave_pkg, reg, 0);
1778 for (j = 0; j < 8; j++) {
1779 u32 pkg = sad_pkg(pvt->info.interleave_pkg, reg, j);
1780 if (j > 0 && sad_interl == pkg)
1781 break;
1783 edac_dbg(0, "SAD#%d, interleave #%d: %d\n",
1784 n_sads, j, pkg);
1788 if (pvt->info.type == KNIGHTS_LANDING)
1789 return;
1792 * Step 3) Get TAD range
1794 prv = 0;
1795 for (n_tads = 0; n_tads < MAX_TAD; n_tads++) {
1796 pci_read_config_dword(pvt->pci_ha, tad_dram_rule[n_tads], &reg);
1797 limit = TAD_LIMIT(reg);
1798 if (limit <= prv)
1799 break;
1800 tmp_mb = (limit + 1) >> 20;
1802 gb = div_u64_rem(tmp_mb, 1024, &mb);
1803 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",
1804 n_tads, gb, (mb*1000)/1024,
1805 ((u64)tmp_mb) << 20L,
1806 (u32)(1 << TAD_SOCK(reg)),
1807 (u32)TAD_CH(reg) + 1,
1808 (u32)TAD_TGT0(reg),
1809 (u32)TAD_TGT1(reg),
1810 (u32)TAD_TGT2(reg),
1811 (u32)TAD_TGT3(reg),
1812 reg);
1813 prv = limit;
1817 * Step 4) Get TAD offsets, per each channel
1819 for (i = 0; i < NUM_CHANNELS; i++) {
1820 if (!pvt->channel[i].dimms)
1821 continue;
1822 for (j = 0; j < n_tads; j++) {
1823 pci_read_config_dword(pvt->pci_tad[i],
1824 tad_ch_nilv_offset[j],
1825 &reg);
1826 tmp_mb = TAD_OFFSET(reg) >> 20;
1827 gb = div_u64_rem(tmp_mb, 1024, &mb);
1828 edac_dbg(0, "TAD CH#%d, offset #%d: %u.%03u GB (0x%016Lx), reg=0x%08x\n",
1829 i, j,
1830 gb, (mb*1000)/1024,
1831 ((u64)tmp_mb) << 20L,
1832 reg);
1837 * Step 6) Get RIR Wayness/Limit, per each channel
1839 for (i = 0; i < NUM_CHANNELS; i++) {
1840 if (!pvt->channel[i].dimms)
1841 continue;
1842 for (j = 0; j < MAX_RIR_RANGES; j++) {
1843 pci_read_config_dword(pvt->pci_tad[i],
1844 rir_way_limit[j],
1845 &reg);
1847 if (!IS_RIR_VALID(reg))
1848 continue;
1850 tmp_mb = pvt->info.rir_limit(reg) >> 20;
1851 rir_way = 1 << RIR_WAY(reg);
1852 gb = div_u64_rem(tmp_mb, 1024, &mb);
1853 edac_dbg(0, "CH#%d RIR#%d, limit: %u.%03u GB (0x%016Lx), way: %d, reg=0x%08x\n",
1854 i, j,
1855 gb, (mb*1000)/1024,
1856 ((u64)tmp_mb) << 20L,
1857 rir_way,
1858 reg);
1860 for (k = 0; k < rir_way; k++) {
1861 pci_read_config_dword(pvt->pci_tad[i],
1862 rir_offset[j][k],
1863 &reg);
1864 tmp_mb = RIR_OFFSET(pvt->info.type, reg) << 6;
1866 gb = div_u64_rem(tmp_mb, 1024, &mb);
1867 edac_dbg(0, "CH#%d RIR#%d INTL#%d, offset %u.%03u GB (0x%016Lx), tgt: %d, reg=0x%08x\n",
1868 i, j, k,
1869 gb, (mb*1000)/1024,
1870 ((u64)tmp_mb) << 20L,
1871 (u32)RIR_RNK_TGT(pvt->info.type, reg),
1872 reg);
1878 static struct mem_ctl_info *get_mci_for_node_id(u8 node_id, u8 ha)
1880 struct sbridge_dev *sbridge_dev;
1882 list_for_each_entry(sbridge_dev, &sbridge_edac_list, list) {
1883 if (sbridge_dev->node_id == node_id && sbridge_dev->dom == ha)
1884 return sbridge_dev->mci;
1886 return NULL;
1889 static int get_memory_error_data(struct mem_ctl_info *mci,
1890 u64 addr,
1891 u8 *socket, u8 *ha,
1892 long *channel_mask,
1893 u8 *rank,
1894 char **area_type, char *msg)
1896 struct mem_ctl_info *new_mci;
1897 struct sbridge_pvt *pvt = mci->pvt_info;
1898 struct pci_dev *pci_ha;
1899 int n_rir, n_sads, n_tads, sad_way, sck_xch;
1900 int sad_interl, idx, base_ch;
1901 int interleave_mode, shiftup = 0;
1902 unsigned sad_interleave[pvt->info.max_interleave];
1903 u32 reg, dram_rule;
1904 u8 ch_way, sck_way, pkg, sad_ha = 0;
1905 u32 tad_offset;
1906 u32 rir_way;
1907 u32 mb, gb;
1908 u64 ch_addr, offset, limit = 0, prv = 0;
1912 * Step 0) Check if the address is at special memory ranges
1913 * The check bellow is probably enough to fill all cases where
1914 * the error is not inside a memory, except for the legacy
1915 * range (e. g. VGA addresses). It is unlikely, however, that the
1916 * memory controller would generate an error on that range.
1918 if ((addr > (u64) pvt->tolm) && (addr < (1LL << 32))) {
1919 sprintf(msg, "Error at TOLM area, on addr 0x%08Lx", addr);
1920 return -EINVAL;
1922 if (addr >= (u64)pvt->tohm) {
1923 sprintf(msg, "Error at MMIOH area, on addr 0x%016Lx", addr);
1924 return -EINVAL;
1928 * Step 1) Get socket
1930 for (n_sads = 0; n_sads < pvt->info.max_sad; n_sads++) {
1931 pci_read_config_dword(pvt->pci_sad0, pvt->info.dram_rule[n_sads],
1932 &reg);
1934 if (!DRAM_RULE_ENABLE(reg))
1935 continue;
1937 limit = pvt->info.sad_limit(reg);
1938 if (limit <= prv) {
1939 sprintf(msg, "Can't discover the memory socket");
1940 return -EINVAL;
1942 if (addr <= limit)
1943 break;
1944 prv = limit;
1946 if (n_sads == pvt->info.max_sad) {
1947 sprintf(msg, "Can't discover the memory socket");
1948 return -EINVAL;
1950 dram_rule = reg;
1951 *area_type = show_dram_attr(pvt->info.dram_attr(dram_rule));
1952 interleave_mode = pvt->info.interleave_mode(dram_rule);
1954 pci_read_config_dword(pvt->pci_sad0, pvt->info.interleave_list[n_sads],
1955 &reg);
1957 if (pvt->info.type == SANDY_BRIDGE) {
1958 sad_interl = sad_pkg(pvt->info.interleave_pkg, reg, 0);
1959 for (sad_way = 0; sad_way < 8; sad_way++) {
1960 u32 pkg = sad_pkg(pvt->info.interleave_pkg, reg, sad_way);
1961 if (sad_way > 0 && sad_interl == pkg)
1962 break;
1963 sad_interleave[sad_way] = pkg;
1964 edac_dbg(0, "SAD interleave #%d: %d\n",
1965 sad_way, sad_interleave[sad_way]);
1967 edac_dbg(0, "mc#%d: Error detected on SAD#%d: address 0x%016Lx < 0x%016Lx, Interleave [%d:6]%s\n",
1968 pvt->sbridge_dev->mc,
1969 n_sads,
1970 addr,
1971 limit,
1972 sad_way + 7,
1973 !interleave_mode ? "" : "XOR[18:16]");
1974 if (interleave_mode)
1975 idx = ((addr >> 6) ^ (addr >> 16)) & 7;
1976 else
1977 idx = (addr >> 6) & 7;
1978 switch (sad_way) {
1979 case 1:
1980 idx = 0;
1981 break;
1982 case 2:
1983 idx = idx & 1;
1984 break;
1985 case 4:
1986 idx = idx & 3;
1987 break;
1988 case 8:
1989 break;
1990 default:
1991 sprintf(msg, "Can't discover socket interleave");
1992 return -EINVAL;
1994 *socket = sad_interleave[idx];
1995 edac_dbg(0, "SAD interleave index: %d (wayness %d) = CPU socket %d\n",
1996 idx, sad_way, *socket);
1997 } else if (pvt->info.type == HASWELL || pvt->info.type == BROADWELL) {
1998 int bits, a7mode = A7MODE(dram_rule);
2000 if (a7mode) {
2001 /* A7 mode swaps P9 with P6 */
2002 bits = GET_BITFIELD(addr, 7, 8) << 1;
2003 bits |= GET_BITFIELD(addr, 9, 9);
2004 } else
2005 bits = GET_BITFIELD(addr, 6, 8);
2007 if (interleave_mode == 0) {
2008 /* interleave mode will XOR {8,7,6} with {18,17,16} */
2009 idx = GET_BITFIELD(addr, 16, 18);
2010 idx ^= bits;
2011 } else
2012 idx = bits;
2014 pkg = sad_pkg(pvt->info.interleave_pkg, reg, idx);
2015 *socket = sad_pkg_socket(pkg);
2016 sad_ha = sad_pkg_ha(pkg);
2018 if (a7mode) {
2019 /* MCChanShiftUpEnable */
2020 pci_read_config_dword(pvt->pci_ha, HASWELL_HASYSDEFEATURE2, &reg);
2021 shiftup = GET_BITFIELD(reg, 22, 22);
2024 edac_dbg(0, "SAD interleave package: %d = CPU socket %d, HA %i, shiftup: %i\n",
2025 idx, *socket, sad_ha, shiftup);
2026 } else {
2027 /* Ivy Bridge's SAD mode doesn't support XOR interleave mode */
2028 idx = (addr >> 6) & 7;
2029 pkg = sad_pkg(pvt->info.interleave_pkg, reg, idx);
2030 *socket = sad_pkg_socket(pkg);
2031 sad_ha = sad_pkg_ha(pkg);
2032 edac_dbg(0, "SAD interleave package: %d = CPU socket %d, HA %d\n",
2033 idx, *socket, sad_ha);
2036 *ha = sad_ha;
2039 * Move to the proper node structure, in order to access the
2040 * right PCI registers
2042 new_mci = get_mci_for_node_id(*socket, sad_ha);
2043 if (!new_mci) {
2044 sprintf(msg, "Struct for socket #%u wasn't initialized",
2045 *socket);
2046 return -EINVAL;
2048 mci = new_mci;
2049 pvt = mci->pvt_info;
2052 * Step 2) Get memory channel
2054 prv = 0;
2055 pci_ha = pvt->pci_ha;
2056 for (n_tads = 0; n_tads < MAX_TAD; n_tads++) {
2057 pci_read_config_dword(pci_ha, tad_dram_rule[n_tads], &reg);
2058 limit = TAD_LIMIT(reg);
2059 if (limit <= prv) {
2060 sprintf(msg, "Can't discover the memory channel");
2061 return -EINVAL;
2063 if (addr <= limit)
2064 break;
2065 prv = limit;
2067 if (n_tads == MAX_TAD) {
2068 sprintf(msg, "Can't discover the memory channel");
2069 return -EINVAL;
2072 ch_way = TAD_CH(reg) + 1;
2073 sck_way = TAD_SOCK(reg);
2075 if (ch_way == 3)
2076 idx = addr >> 6;
2077 else {
2078 idx = (addr >> (6 + sck_way + shiftup)) & 0x3;
2079 if (pvt->is_chan_hash)
2080 idx = haswell_chan_hash(idx, addr);
2082 idx = idx % ch_way;
2085 * FIXME: Shouldn't we use CHN_IDX_OFFSET() here, when ch_way == 3 ???
2087 switch (idx) {
2088 case 0:
2089 base_ch = TAD_TGT0(reg);
2090 break;
2091 case 1:
2092 base_ch = TAD_TGT1(reg);
2093 break;
2094 case 2:
2095 base_ch = TAD_TGT2(reg);
2096 break;
2097 case 3:
2098 base_ch = TAD_TGT3(reg);
2099 break;
2100 default:
2101 sprintf(msg, "Can't discover the TAD target");
2102 return -EINVAL;
2104 *channel_mask = 1 << base_ch;
2106 pci_read_config_dword(pvt->pci_tad[base_ch], tad_ch_nilv_offset[n_tads], &tad_offset);
2108 if (pvt->mirror_mode == FULL_MIRRORING ||
2109 (pvt->mirror_mode == ADDR_RANGE_MIRRORING && n_tads == 0)) {
2110 *channel_mask |= 1 << ((base_ch + 2) % 4);
2111 switch(ch_way) {
2112 case 2:
2113 case 4:
2114 sck_xch = (1 << sck_way) * (ch_way >> 1);
2115 break;
2116 default:
2117 sprintf(msg, "Invalid mirror set. Can't decode addr");
2118 return -EINVAL;
2121 pvt->is_cur_addr_mirrored = true;
2122 } else {
2123 sck_xch = (1 << sck_way) * ch_way;
2124 pvt->is_cur_addr_mirrored = false;
2127 if (pvt->is_lockstep)
2128 *channel_mask |= 1 << ((base_ch + 1) % 4);
2130 offset = TAD_OFFSET(tad_offset);
2132 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",
2133 n_tads,
2134 addr,
2135 limit,
2136 sck_way,
2137 ch_way,
2138 offset,
2139 idx,
2140 base_ch,
2141 *channel_mask);
2143 /* Calculate channel address */
2144 /* Remove the TAD offset */
2146 if (offset > addr) {
2147 sprintf(msg, "Can't calculate ch addr: TAD offset 0x%08Lx is too high for addr 0x%08Lx!",
2148 offset, addr);
2149 return -EINVAL;
2152 ch_addr = addr - offset;
2153 ch_addr >>= (6 + shiftup);
2154 ch_addr /= sck_xch;
2155 ch_addr <<= (6 + shiftup);
2156 ch_addr |= addr & ((1 << (6 + shiftup)) - 1);
2159 * Step 3) Decode rank
2161 for (n_rir = 0; n_rir < MAX_RIR_RANGES; n_rir++) {
2162 pci_read_config_dword(pvt->pci_tad[base_ch], rir_way_limit[n_rir], &reg);
2164 if (!IS_RIR_VALID(reg))
2165 continue;
2167 limit = pvt->info.rir_limit(reg);
2168 gb = div_u64_rem(limit >> 20, 1024, &mb);
2169 edac_dbg(0, "RIR#%d, limit: %u.%03u GB (0x%016Lx), way: %d\n",
2170 n_rir,
2171 gb, (mb*1000)/1024,
2172 limit,
2173 1 << RIR_WAY(reg));
2174 if (ch_addr <= limit)
2175 break;
2177 if (n_rir == MAX_RIR_RANGES) {
2178 sprintf(msg, "Can't discover the memory rank for ch addr 0x%08Lx",
2179 ch_addr);
2180 return -EINVAL;
2182 rir_way = RIR_WAY(reg);
2184 if (pvt->is_close_pg)
2185 idx = (ch_addr >> 6);
2186 else
2187 idx = (ch_addr >> 13); /* FIXME: Datasheet says to shift by 15 */
2188 idx %= 1 << rir_way;
2190 pci_read_config_dword(pvt->pci_tad[base_ch], rir_offset[n_rir][idx], &reg);
2191 *rank = RIR_RNK_TGT(pvt->info.type, reg);
2193 edac_dbg(0, "RIR#%d: channel address 0x%08Lx < 0x%08Lx, RIR interleave %d, index %d\n",
2194 n_rir,
2195 ch_addr,
2196 limit,
2197 rir_way,
2198 idx);
2200 return 0;
2203 /****************************************************************************
2204 Device initialization routines: put/get, init/exit
2205 ****************************************************************************/
2208 * sbridge_put_all_devices 'put' all the devices that we have
2209 * reserved via 'get'
2211 static void sbridge_put_devices(struct sbridge_dev *sbridge_dev)
2213 int i;
2215 edac_dbg(0, "\n");
2216 for (i = 0; i < sbridge_dev->n_devs; i++) {
2217 struct pci_dev *pdev = sbridge_dev->pdev[i];
2218 if (!pdev)
2219 continue;
2220 edac_dbg(0, "Removing dev %02x:%02x.%d\n",
2221 pdev->bus->number,
2222 PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn));
2223 pci_dev_put(pdev);
2227 static void sbridge_put_all_devices(void)
2229 struct sbridge_dev *sbridge_dev, *tmp;
2231 list_for_each_entry_safe(sbridge_dev, tmp, &sbridge_edac_list, list) {
2232 sbridge_put_devices(sbridge_dev);
2233 free_sbridge_dev(sbridge_dev);
2237 static int sbridge_get_onedevice(struct pci_dev **prev,
2238 u8 *num_mc,
2239 const struct pci_id_table *table,
2240 const unsigned devno,
2241 const int multi_bus)
2243 struct sbridge_dev *sbridge_dev = NULL;
2244 const struct pci_id_descr *dev_descr = &table->descr[devno];
2245 struct pci_dev *pdev = NULL;
2246 u8 bus = 0;
2247 int i = 0;
2249 sbridge_printk(KERN_DEBUG,
2250 "Seeking for: PCI ID %04x:%04x\n",
2251 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
2253 pdev = pci_get_device(PCI_VENDOR_ID_INTEL,
2254 dev_descr->dev_id, *prev);
2256 if (!pdev) {
2257 if (*prev) {
2258 *prev = pdev;
2259 return 0;
2262 if (dev_descr->optional)
2263 return 0;
2265 /* if the HA wasn't found */
2266 if (devno == 0)
2267 return -ENODEV;
2269 sbridge_printk(KERN_INFO,
2270 "Device not found: %04x:%04x\n",
2271 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
2273 /* End of list, leave */
2274 return -ENODEV;
2276 bus = pdev->bus->number;
2278 next_imc:
2279 sbridge_dev = get_sbridge_dev(bus, dev_descr->dom, multi_bus, sbridge_dev);
2280 if (!sbridge_dev) {
2281 /* If the HA1 wasn't found, don't create EDAC second memory controller */
2282 if (dev_descr->dom == IMC1 && devno != 1) {
2283 edac_dbg(0, "Skip IMC1: %04x:%04x (since HA1 was absent)\n",
2284 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
2285 pci_dev_put(pdev);
2286 return 0;
2289 if (dev_descr->dom == SOCK)
2290 goto out_imc;
2292 sbridge_dev = alloc_sbridge_dev(bus, dev_descr->dom, table);
2293 if (!sbridge_dev) {
2294 pci_dev_put(pdev);
2295 return -ENOMEM;
2297 (*num_mc)++;
2300 if (sbridge_dev->pdev[sbridge_dev->i_devs]) {
2301 sbridge_printk(KERN_ERR,
2302 "Duplicated device for %04x:%04x\n",
2303 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
2304 pci_dev_put(pdev);
2305 return -ENODEV;
2308 sbridge_dev->pdev[sbridge_dev->i_devs++] = pdev;
2310 /* pdev belongs to more than one IMC, do extra gets */
2311 if (++i > 1)
2312 pci_dev_get(pdev);
2314 if (dev_descr->dom == SOCK && i < table->n_imcs_per_sock)
2315 goto next_imc;
2317 out_imc:
2318 /* Be sure that the device is enabled */
2319 if (unlikely(pci_enable_device(pdev) < 0)) {
2320 sbridge_printk(KERN_ERR,
2321 "Couldn't enable %04x:%04x\n",
2322 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
2323 return -ENODEV;
2326 edac_dbg(0, "Detected %04x:%04x\n",
2327 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
2330 * As stated on drivers/pci/search.c, the reference count for
2331 * @from is always decremented if it is not %NULL. So, as we need
2332 * to get all devices up to null, we need to do a get for the device
2334 pci_dev_get(pdev);
2336 *prev = pdev;
2338 return 0;
2342 * sbridge_get_all_devices - Find and perform 'get' operation on the MCH's
2343 * devices we want to reference for this driver.
2344 * @num_mc: pointer to the memory controllers count, to be incremented in case
2345 * of success.
2346 * @table: model specific table
2348 * returns 0 in case of success or error code
2350 static int sbridge_get_all_devices(u8 *num_mc,
2351 const struct pci_id_table *table)
2353 int i, rc;
2354 struct pci_dev *pdev = NULL;
2355 int allow_dups = 0;
2356 int multi_bus = 0;
2358 if (table->type == KNIGHTS_LANDING)
2359 allow_dups = multi_bus = 1;
2360 while (table && table->descr) {
2361 for (i = 0; i < table->n_devs_per_sock; i++) {
2362 if (!allow_dups || i == 0 ||
2363 table->descr[i].dev_id !=
2364 table->descr[i-1].dev_id) {
2365 pdev = NULL;
2367 do {
2368 rc = sbridge_get_onedevice(&pdev, num_mc,
2369 table, i, multi_bus);
2370 if (rc < 0) {
2371 if (i == 0) {
2372 i = table->n_devs_per_sock;
2373 break;
2375 sbridge_put_all_devices();
2376 return -ENODEV;
2378 } while (pdev && !allow_dups);
2380 table++;
2383 return 0;
2387 * Device IDs for {SBRIDGE,IBRIDGE,HASWELL,BROADWELL}_IMC_HA0_TAD0 are in
2388 * the format: XXXa. So we can convert from a device to the corresponding
2389 * channel like this
2391 #define TAD_DEV_TO_CHAN(dev) (((dev) & 0xf) - 0xa)
2393 static int sbridge_mci_bind_devs(struct mem_ctl_info *mci,
2394 struct sbridge_dev *sbridge_dev)
2396 struct sbridge_pvt *pvt = mci->pvt_info;
2397 struct pci_dev *pdev;
2398 u8 saw_chan_mask = 0;
2399 int i;
2401 for (i = 0; i < sbridge_dev->n_devs; i++) {
2402 pdev = sbridge_dev->pdev[i];
2403 if (!pdev)
2404 continue;
2406 switch (pdev->device) {
2407 case PCI_DEVICE_ID_INTEL_SBRIDGE_SAD0:
2408 pvt->pci_sad0 = pdev;
2409 break;
2410 case PCI_DEVICE_ID_INTEL_SBRIDGE_SAD1:
2411 pvt->pci_sad1 = pdev;
2412 break;
2413 case PCI_DEVICE_ID_INTEL_SBRIDGE_BR:
2414 pvt->pci_br0 = pdev;
2415 break;
2416 case PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_HA0:
2417 pvt->pci_ha = pdev;
2418 break;
2419 case PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TA:
2420 pvt->pci_ta = pdev;
2421 break;
2422 case PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_RAS:
2423 pvt->pci_ras = pdev;
2424 break;
2425 case PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD0:
2426 case PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD1:
2427 case PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD2:
2428 case PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD3:
2430 int id = TAD_DEV_TO_CHAN(pdev->device);
2431 pvt->pci_tad[id] = pdev;
2432 saw_chan_mask |= 1 << id;
2434 break;
2435 case PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_DDRIO:
2436 pvt->pci_ddrio = pdev;
2437 break;
2438 default:
2439 goto error;
2442 edac_dbg(0, "Associated PCI %02x:%02x, bus %d with dev = %p\n",
2443 pdev->vendor, pdev->device,
2444 sbridge_dev->bus,
2445 pdev);
2448 /* Check if everything were registered */
2449 if (!pvt->pci_sad0 || !pvt->pci_sad1 || !pvt->pci_ha ||
2450 !pvt->pci_ras || !pvt->pci_ta)
2451 goto enodev;
2453 if (saw_chan_mask != 0x0f)
2454 goto enodev;
2455 return 0;
2457 enodev:
2458 sbridge_printk(KERN_ERR, "Some needed devices are missing\n");
2459 return -ENODEV;
2461 error:
2462 sbridge_printk(KERN_ERR, "Unexpected device %02x:%02x\n",
2463 PCI_VENDOR_ID_INTEL, pdev->device);
2464 return -EINVAL;
2467 static int ibridge_mci_bind_devs(struct mem_ctl_info *mci,
2468 struct sbridge_dev *sbridge_dev)
2470 struct sbridge_pvt *pvt = mci->pvt_info;
2471 struct pci_dev *pdev;
2472 u8 saw_chan_mask = 0;
2473 int i;
2475 for (i = 0; i < sbridge_dev->n_devs; i++) {
2476 pdev = sbridge_dev->pdev[i];
2477 if (!pdev)
2478 continue;
2480 switch (pdev->device) {
2481 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0:
2482 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1:
2483 pvt->pci_ha = pdev;
2484 break;
2485 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TA:
2486 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TA:
2487 pvt->pci_ta = pdev;
2488 break;
2489 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_RAS:
2490 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_RAS:
2491 pvt->pci_ras = pdev;
2492 break;
2493 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD0:
2494 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD1:
2495 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD2:
2496 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD3:
2497 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD0:
2498 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD1:
2499 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD2:
2500 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD3:
2502 int id = TAD_DEV_TO_CHAN(pdev->device);
2503 pvt->pci_tad[id] = pdev;
2504 saw_chan_mask |= 1 << id;
2506 break;
2507 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_2HA_DDRIO0:
2508 pvt->pci_ddrio = pdev;
2509 break;
2510 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_1HA_DDRIO0:
2511 pvt->pci_ddrio = pdev;
2512 break;
2513 case PCI_DEVICE_ID_INTEL_IBRIDGE_SAD:
2514 pvt->pci_sad0 = pdev;
2515 break;
2516 case PCI_DEVICE_ID_INTEL_IBRIDGE_BR0:
2517 pvt->pci_br0 = pdev;
2518 break;
2519 case PCI_DEVICE_ID_INTEL_IBRIDGE_BR1:
2520 pvt->pci_br1 = pdev;
2521 break;
2522 default:
2523 goto error;
2526 edac_dbg(0, "Associated PCI %02x.%02d.%d with dev = %p\n",
2527 sbridge_dev->bus,
2528 PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn),
2529 pdev);
2532 /* Check if everything were registered */
2533 if (!pvt->pci_sad0 || !pvt->pci_ha || !pvt->pci_br0 ||
2534 !pvt->pci_br1 || !pvt->pci_ras || !pvt->pci_ta)
2535 goto enodev;
2537 if (saw_chan_mask != 0x0f && /* -EN/-EX */
2538 saw_chan_mask != 0x03) /* -EP */
2539 goto enodev;
2540 return 0;
2542 enodev:
2543 sbridge_printk(KERN_ERR, "Some needed devices are missing\n");
2544 return -ENODEV;
2546 error:
2547 sbridge_printk(KERN_ERR,
2548 "Unexpected device %02x:%02x\n", PCI_VENDOR_ID_INTEL,
2549 pdev->device);
2550 return -EINVAL;
2553 static int haswell_mci_bind_devs(struct mem_ctl_info *mci,
2554 struct sbridge_dev *sbridge_dev)
2556 struct sbridge_pvt *pvt = mci->pvt_info;
2557 struct pci_dev *pdev;
2558 u8 saw_chan_mask = 0;
2559 int i;
2561 /* there's only one device per system; not tied to any bus */
2562 if (pvt->info.pci_vtd == NULL)
2563 /* result will be checked later */
2564 pvt->info.pci_vtd = pci_get_device(PCI_VENDOR_ID_INTEL,
2565 PCI_DEVICE_ID_INTEL_HASWELL_IMC_VTD_MISC,
2566 NULL);
2568 for (i = 0; i < sbridge_dev->n_devs; i++) {
2569 pdev = sbridge_dev->pdev[i];
2570 if (!pdev)
2571 continue;
2573 switch (pdev->device) {
2574 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_CBO_SAD0:
2575 pvt->pci_sad0 = pdev;
2576 break;
2577 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_CBO_SAD1:
2578 pvt->pci_sad1 = pdev;
2579 break;
2580 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0:
2581 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1:
2582 pvt->pci_ha = pdev;
2583 break;
2584 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TA:
2585 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TA:
2586 pvt->pci_ta = pdev;
2587 break;
2588 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TM:
2589 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TM:
2590 pvt->pci_ras = pdev;
2591 break;
2592 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD0:
2593 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD1:
2594 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD2:
2595 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD3:
2596 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD0:
2597 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD1:
2598 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD2:
2599 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD3:
2601 int id = TAD_DEV_TO_CHAN(pdev->device);
2602 pvt->pci_tad[id] = pdev;
2603 saw_chan_mask |= 1 << id;
2605 break;
2606 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO0:
2607 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO1:
2608 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO2:
2609 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO3:
2610 if (!pvt->pci_ddrio)
2611 pvt->pci_ddrio = pdev;
2612 break;
2613 default:
2614 break;
2617 edac_dbg(0, "Associated PCI %02x.%02d.%d with dev = %p\n",
2618 sbridge_dev->bus,
2619 PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn),
2620 pdev);
2623 /* Check if everything were registered */
2624 if (!pvt->pci_sad0 || !pvt->pci_ha || !pvt->pci_sad1 ||
2625 !pvt->pci_ras || !pvt->pci_ta || !pvt->info.pci_vtd)
2626 goto enodev;
2628 if (saw_chan_mask != 0x0f && /* -EN/-EX */
2629 saw_chan_mask != 0x03) /* -EP */
2630 goto enodev;
2631 return 0;
2633 enodev:
2634 sbridge_printk(KERN_ERR, "Some needed devices are missing\n");
2635 return -ENODEV;
2638 static int broadwell_mci_bind_devs(struct mem_ctl_info *mci,
2639 struct sbridge_dev *sbridge_dev)
2641 struct sbridge_pvt *pvt = mci->pvt_info;
2642 struct pci_dev *pdev;
2643 u8 saw_chan_mask = 0;
2644 int i;
2646 /* there's only one device per system; not tied to any bus */
2647 if (pvt->info.pci_vtd == NULL)
2648 /* result will be checked later */
2649 pvt->info.pci_vtd = pci_get_device(PCI_VENDOR_ID_INTEL,
2650 PCI_DEVICE_ID_INTEL_BROADWELL_IMC_VTD_MISC,
2651 NULL);
2653 for (i = 0; i < sbridge_dev->n_devs; i++) {
2654 pdev = sbridge_dev->pdev[i];
2655 if (!pdev)
2656 continue;
2658 switch (pdev->device) {
2659 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_CBO_SAD0:
2660 pvt->pci_sad0 = pdev;
2661 break;
2662 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_CBO_SAD1:
2663 pvt->pci_sad1 = pdev;
2664 break;
2665 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0:
2666 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1:
2667 pvt->pci_ha = pdev;
2668 break;
2669 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TA:
2670 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TA:
2671 pvt->pci_ta = pdev;
2672 break;
2673 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TM:
2674 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TM:
2675 pvt->pci_ras = pdev;
2676 break;
2677 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD0:
2678 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD1:
2679 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD2:
2680 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD3:
2681 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD0:
2682 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD1:
2683 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD2:
2684 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD3:
2686 int id = TAD_DEV_TO_CHAN(pdev->device);
2687 pvt->pci_tad[id] = pdev;
2688 saw_chan_mask |= 1 << id;
2690 break;
2691 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_DDRIO0:
2692 pvt->pci_ddrio = pdev;
2693 break;
2694 default:
2695 break;
2698 edac_dbg(0, "Associated PCI %02x.%02d.%d with dev = %p\n",
2699 sbridge_dev->bus,
2700 PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn),
2701 pdev);
2704 /* Check if everything were registered */
2705 if (!pvt->pci_sad0 || !pvt->pci_ha || !pvt->pci_sad1 ||
2706 !pvt->pci_ras || !pvt->pci_ta || !pvt->info.pci_vtd)
2707 goto enodev;
2709 if (saw_chan_mask != 0x0f && /* -EN/-EX */
2710 saw_chan_mask != 0x03) /* -EP */
2711 goto enodev;
2712 return 0;
2714 enodev:
2715 sbridge_printk(KERN_ERR, "Some needed devices are missing\n");
2716 return -ENODEV;
2719 static int knl_mci_bind_devs(struct mem_ctl_info *mci,
2720 struct sbridge_dev *sbridge_dev)
2722 struct sbridge_pvt *pvt = mci->pvt_info;
2723 struct pci_dev *pdev;
2724 int dev, func;
2726 int i;
2727 int devidx;
2729 for (i = 0; i < sbridge_dev->n_devs; i++) {
2730 pdev = sbridge_dev->pdev[i];
2731 if (!pdev)
2732 continue;
2734 /* Extract PCI device and function. */
2735 dev = (pdev->devfn >> 3) & 0x1f;
2736 func = pdev->devfn & 0x7;
2738 switch (pdev->device) {
2739 case PCI_DEVICE_ID_INTEL_KNL_IMC_MC:
2740 if (dev == 8)
2741 pvt->knl.pci_mc0 = pdev;
2742 else if (dev == 9)
2743 pvt->knl.pci_mc1 = pdev;
2744 else {
2745 sbridge_printk(KERN_ERR,
2746 "Memory controller in unexpected place! (dev %d, fn %d)\n",
2747 dev, func);
2748 continue;
2750 break;
2752 case PCI_DEVICE_ID_INTEL_KNL_IMC_SAD0:
2753 pvt->pci_sad0 = pdev;
2754 break;
2756 case PCI_DEVICE_ID_INTEL_KNL_IMC_SAD1:
2757 pvt->pci_sad1 = pdev;
2758 break;
2760 case PCI_DEVICE_ID_INTEL_KNL_IMC_CHA:
2761 /* There are one of these per tile, and range from
2762 * 1.14.0 to 1.18.5.
2764 devidx = ((dev-14)*8)+func;
2766 if (devidx < 0 || devidx >= KNL_MAX_CHAS) {
2767 sbridge_printk(KERN_ERR,
2768 "Caching and Home Agent in unexpected place! (dev %d, fn %d)\n",
2769 dev, func);
2770 continue;
2773 WARN_ON(pvt->knl.pci_cha[devidx] != NULL);
2775 pvt->knl.pci_cha[devidx] = pdev;
2776 break;
2778 case PCI_DEVICE_ID_INTEL_KNL_IMC_CHAN:
2779 devidx = -1;
2782 * MC0 channels 0-2 are device 9 function 2-4,
2783 * MC1 channels 3-5 are device 8 function 2-4.
2786 if (dev == 9)
2787 devidx = func-2;
2788 else if (dev == 8)
2789 devidx = 3 + (func-2);
2791 if (devidx < 0 || devidx >= KNL_MAX_CHANNELS) {
2792 sbridge_printk(KERN_ERR,
2793 "DRAM Channel Registers in unexpected place! (dev %d, fn %d)\n",
2794 dev, func);
2795 continue;
2798 WARN_ON(pvt->knl.pci_channel[devidx] != NULL);
2799 pvt->knl.pci_channel[devidx] = pdev;
2800 break;
2802 case PCI_DEVICE_ID_INTEL_KNL_IMC_TOLHM:
2803 pvt->knl.pci_mc_info = pdev;
2804 break;
2806 case PCI_DEVICE_ID_INTEL_KNL_IMC_TA:
2807 pvt->pci_ta = pdev;
2808 break;
2810 default:
2811 sbridge_printk(KERN_ERR, "Unexpected device %d\n",
2812 pdev->device);
2813 break;
2817 if (!pvt->knl.pci_mc0 || !pvt->knl.pci_mc1 ||
2818 !pvt->pci_sad0 || !pvt->pci_sad1 ||
2819 !pvt->pci_ta) {
2820 goto enodev;
2823 for (i = 0; i < KNL_MAX_CHANNELS; i++) {
2824 if (!pvt->knl.pci_channel[i]) {
2825 sbridge_printk(KERN_ERR, "Missing channel %d\n", i);
2826 goto enodev;
2830 for (i = 0; i < KNL_MAX_CHAS; i++) {
2831 if (!pvt->knl.pci_cha[i]) {
2832 sbridge_printk(KERN_ERR, "Missing CHA %d\n", i);
2833 goto enodev;
2837 return 0;
2839 enodev:
2840 sbridge_printk(KERN_ERR, "Some needed devices are missing\n");
2841 return -ENODEV;
2844 /****************************************************************************
2845 Error check routines
2846 ****************************************************************************/
2849 * While Sandy Bridge has error count registers, SMI BIOS read values from
2850 * and resets the counters. So, they are not reliable for the OS to read
2851 * from them. So, we have no option but to just trust on whatever MCE is
2852 * telling us about the errors.
2854 static void sbridge_mce_output_error(struct mem_ctl_info *mci,
2855 const struct mce *m)
2857 struct mem_ctl_info *new_mci;
2858 struct sbridge_pvt *pvt = mci->pvt_info;
2859 enum hw_event_mc_err_type tp_event;
2860 char *type, *optype, msg[256];
2861 bool ripv = GET_BITFIELD(m->mcgstatus, 0, 0);
2862 bool overflow = GET_BITFIELD(m->status, 62, 62);
2863 bool uncorrected_error = GET_BITFIELD(m->status, 61, 61);
2864 bool recoverable;
2865 u32 core_err_cnt = GET_BITFIELD(m->status, 38, 52);
2866 u32 mscod = GET_BITFIELD(m->status, 16, 31);
2867 u32 errcode = GET_BITFIELD(m->status, 0, 15);
2868 u32 channel = GET_BITFIELD(m->status, 0, 3);
2869 u32 optypenum = GET_BITFIELD(m->status, 4, 6);
2870 long channel_mask, first_channel;
2871 u8 rank, socket, ha;
2872 int rc, dimm;
2873 char *area_type = NULL;
2875 if (pvt->info.type != SANDY_BRIDGE)
2876 recoverable = true;
2877 else
2878 recoverable = GET_BITFIELD(m->status, 56, 56);
2880 if (uncorrected_error) {
2881 if (ripv) {
2882 type = "FATAL";
2883 tp_event = HW_EVENT_ERR_FATAL;
2884 } else {
2885 type = "NON_FATAL";
2886 tp_event = HW_EVENT_ERR_UNCORRECTED;
2888 } else {
2889 type = "CORRECTED";
2890 tp_event = HW_EVENT_ERR_CORRECTED;
2894 * According with Table 15-9 of the Intel Architecture spec vol 3A,
2895 * memory errors should fit in this mask:
2896 * 000f 0000 1mmm cccc (binary)
2897 * where:
2898 * f = Correction Report Filtering Bit. If 1, subsequent errors
2899 * won't be shown
2900 * mmm = error type
2901 * cccc = channel
2902 * If the mask doesn't match, report an error to the parsing logic
2904 if (! ((errcode & 0xef80) == 0x80)) {
2905 optype = "Can't parse: it is not a mem";
2906 } else {
2907 switch (optypenum) {
2908 case 0:
2909 optype = "generic undef request error";
2910 break;
2911 case 1:
2912 optype = "memory read error";
2913 break;
2914 case 2:
2915 optype = "memory write error";
2916 break;
2917 case 3:
2918 optype = "addr/cmd error";
2919 break;
2920 case 4:
2921 optype = "memory scrubbing error";
2922 break;
2923 default:
2924 optype = "reserved";
2925 break;
2929 /* Only decode errors with an valid address (ADDRV) */
2930 if (!GET_BITFIELD(m->status, 58, 58))
2931 return;
2933 if (pvt->info.type == KNIGHTS_LANDING) {
2934 if (channel == 14) {
2935 edac_dbg(0, "%s%s err_code:%04x:%04x EDRAM bank %d\n",
2936 overflow ? " OVERFLOW" : "",
2937 (uncorrected_error && recoverable)
2938 ? " recoverable" : "",
2939 mscod, errcode,
2940 m->bank);
2941 } else {
2942 char A = *("A");
2945 * Reported channel is in range 0-2, so we can't map it
2946 * back to mc. To figure out mc we check machine check
2947 * bank register that reported this error.
2948 * bank15 means mc0 and bank16 means mc1.
2950 channel = knl_channel_remap(m->bank == 16, channel);
2951 channel_mask = 1 << channel;
2953 snprintf(msg, sizeof(msg),
2954 "%s%s err_code:%04x:%04x channel:%d (DIMM_%c)",
2955 overflow ? " OVERFLOW" : "",
2956 (uncorrected_error && recoverable)
2957 ? " recoverable" : " ",
2958 mscod, errcode, channel, A + channel);
2959 edac_mc_handle_error(tp_event, mci, core_err_cnt,
2960 m->addr >> PAGE_SHIFT, m->addr & ~PAGE_MASK, 0,
2961 channel, 0, -1,
2962 optype, msg);
2964 return;
2965 } else {
2966 rc = get_memory_error_data(mci, m->addr, &socket, &ha,
2967 &channel_mask, &rank, &area_type, msg);
2970 if (rc < 0)
2971 goto err_parsing;
2972 new_mci = get_mci_for_node_id(socket, ha);
2973 if (!new_mci) {
2974 strcpy(msg, "Error: socket got corrupted!");
2975 goto err_parsing;
2977 mci = new_mci;
2978 pvt = mci->pvt_info;
2980 first_channel = find_first_bit(&channel_mask, NUM_CHANNELS);
2982 if (rank < 4)
2983 dimm = 0;
2984 else if (rank < 8)
2985 dimm = 1;
2986 else
2987 dimm = 2;
2991 * FIXME: On some memory configurations (mirror, lockstep), the
2992 * Memory Controller can't point the error to a single DIMM. The
2993 * EDAC core should be handling the channel mask, in order to point
2994 * to the group of dimm's where the error may be happening.
2996 if (!pvt->is_lockstep && !pvt->is_cur_addr_mirrored && !pvt->is_close_pg)
2997 channel = first_channel;
2999 snprintf(msg, sizeof(msg),
3000 "%s%s area:%s err_code:%04x:%04x socket:%d ha:%d channel_mask:%ld rank:%d",
3001 overflow ? " OVERFLOW" : "",
3002 (uncorrected_error && recoverable) ? " recoverable" : "",
3003 area_type,
3004 mscod, errcode,
3005 socket, ha,
3006 channel_mask,
3007 rank);
3009 edac_dbg(0, "%s\n", msg);
3011 /* FIXME: need support for channel mask */
3013 if (channel == CHANNEL_UNSPECIFIED)
3014 channel = -1;
3016 /* Call the helper to output message */
3017 edac_mc_handle_error(tp_event, mci, core_err_cnt,
3018 m->addr >> PAGE_SHIFT, m->addr & ~PAGE_MASK, 0,
3019 channel, dimm, -1,
3020 optype, msg);
3021 return;
3022 err_parsing:
3023 edac_mc_handle_error(tp_event, mci, core_err_cnt, 0, 0, 0,
3024 -1, -1, -1,
3025 msg, "");
3030 * Check that logging is enabled and that this is the right type
3031 * of error for us to handle.
3033 static int sbridge_mce_check_error(struct notifier_block *nb, unsigned long val,
3034 void *data)
3036 struct mce *mce = (struct mce *)data;
3037 struct mem_ctl_info *mci;
3038 struct sbridge_pvt *pvt;
3039 char *type;
3041 if (edac_get_report_status() == EDAC_REPORTING_DISABLED)
3042 return NOTIFY_DONE;
3044 mci = get_mci_for_node_id(mce->socketid, IMC0);
3045 if (!mci)
3046 return NOTIFY_DONE;
3047 pvt = mci->pvt_info;
3050 * Just let mcelog handle it if the error is
3051 * outside the memory controller. A memory error
3052 * is indicated by bit 7 = 1 and bits = 8-11,13-15 = 0.
3053 * bit 12 has an special meaning.
3055 if ((mce->status & 0xefff) >> 7 != 1)
3056 return NOTIFY_DONE;
3058 if (mce->mcgstatus & MCG_STATUS_MCIP)
3059 type = "Exception";
3060 else
3061 type = "Event";
3063 sbridge_mc_printk(mci, KERN_DEBUG, "HANDLING MCE MEMORY ERROR\n");
3065 sbridge_mc_printk(mci, KERN_DEBUG, "CPU %d: Machine Check %s: %Lx "
3066 "Bank %d: %016Lx\n", mce->extcpu, type,
3067 mce->mcgstatus, mce->bank, mce->status);
3068 sbridge_mc_printk(mci, KERN_DEBUG, "TSC %llx ", mce->tsc);
3069 sbridge_mc_printk(mci, KERN_DEBUG, "ADDR %llx ", mce->addr);
3070 sbridge_mc_printk(mci, KERN_DEBUG, "MISC %llx ", mce->misc);
3072 sbridge_mc_printk(mci, KERN_DEBUG, "PROCESSOR %u:%x TIME %llu SOCKET "
3073 "%u APIC %x\n", mce->cpuvendor, mce->cpuid,
3074 mce->time, mce->socketid, mce->apicid);
3076 sbridge_mce_output_error(mci, mce);
3078 /* Advice mcelog that the error were handled */
3079 return NOTIFY_STOP;
3082 static struct notifier_block sbridge_mce_dec = {
3083 .notifier_call = sbridge_mce_check_error,
3084 .priority = MCE_PRIO_EDAC,
3087 /****************************************************************************
3088 EDAC register/unregister logic
3089 ****************************************************************************/
3091 static void sbridge_unregister_mci(struct sbridge_dev *sbridge_dev)
3093 struct mem_ctl_info *mci = sbridge_dev->mci;
3094 struct sbridge_pvt *pvt;
3096 if (unlikely(!mci || !mci->pvt_info)) {
3097 edac_dbg(0, "MC: dev = %p\n", &sbridge_dev->pdev[0]->dev);
3099 sbridge_printk(KERN_ERR, "Couldn't find mci handler\n");
3100 return;
3103 pvt = mci->pvt_info;
3105 edac_dbg(0, "MC: mci = %p, dev = %p\n",
3106 mci, &sbridge_dev->pdev[0]->dev);
3108 /* Remove MC sysfs nodes */
3109 edac_mc_del_mc(mci->pdev);
3111 edac_dbg(1, "%s: free mci struct\n", mci->ctl_name);
3112 kfree(mci->ctl_name);
3113 edac_mc_free(mci);
3114 sbridge_dev->mci = NULL;
3117 static int sbridge_register_mci(struct sbridge_dev *sbridge_dev, enum type type)
3119 struct mem_ctl_info *mci;
3120 struct edac_mc_layer layers[2];
3121 struct sbridge_pvt *pvt;
3122 struct pci_dev *pdev = sbridge_dev->pdev[0];
3123 int rc;
3125 /* allocate a new MC control structure */
3126 layers[0].type = EDAC_MC_LAYER_CHANNEL;
3127 layers[0].size = type == KNIGHTS_LANDING ?
3128 KNL_MAX_CHANNELS : NUM_CHANNELS;
3129 layers[0].is_virt_csrow = false;
3130 layers[1].type = EDAC_MC_LAYER_SLOT;
3131 layers[1].size = type == KNIGHTS_LANDING ? 1 : MAX_DIMMS;
3132 layers[1].is_virt_csrow = true;
3133 mci = edac_mc_alloc(sbridge_dev->mc, ARRAY_SIZE(layers), layers,
3134 sizeof(*pvt));
3136 if (unlikely(!mci))
3137 return -ENOMEM;
3139 edac_dbg(0, "MC: mci = %p, dev = %p\n",
3140 mci, &pdev->dev);
3142 pvt = mci->pvt_info;
3143 memset(pvt, 0, sizeof(*pvt));
3145 /* Associate sbridge_dev and mci for future usage */
3146 pvt->sbridge_dev = sbridge_dev;
3147 sbridge_dev->mci = mci;
3149 mci->mtype_cap = type == KNIGHTS_LANDING ?
3150 MEM_FLAG_DDR4 : MEM_FLAG_DDR3;
3151 mci->edac_ctl_cap = EDAC_FLAG_NONE;
3152 mci->edac_cap = EDAC_FLAG_NONE;
3153 mci->mod_name = EDAC_MOD_STR;
3154 mci->dev_name = pci_name(pdev);
3155 mci->ctl_page_to_phys = NULL;
3157 pvt->info.type = type;
3158 switch (type) {
3159 case IVY_BRIDGE:
3160 pvt->info.rankcfgr = IB_RANK_CFG_A;
3161 pvt->info.get_tolm = ibridge_get_tolm;
3162 pvt->info.get_tohm = ibridge_get_tohm;
3163 pvt->info.dram_rule = ibridge_dram_rule;
3164 pvt->info.get_memory_type = get_memory_type;
3165 pvt->info.get_node_id = get_node_id;
3166 pvt->info.rir_limit = rir_limit;
3167 pvt->info.sad_limit = sad_limit;
3168 pvt->info.interleave_mode = interleave_mode;
3169 pvt->info.dram_attr = dram_attr;
3170 pvt->info.max_sad = ARRAY_SIZE(ibridge_dram_rule);
3171 pvt->info.interleave_list = ibridge_interleave_list;
3172 pvt->info.max_interleave = ARRAY_SIZE(ibridge_interleave_list);
3173 pvt->info.interleave_pkg = ibridge_interleave_pkg;
3174 pvt->info.get_width = ibridge_get_width;
3176 /* Store pci devices at mci for faster access */
3177 rc = ibridge_mci_bind_devs(mci, sbridge_dev);
3178 if (unlikely(rc < 0))
3179 goto fail0;
3180 get_source_id(mci);
3181 mci->ctl_name = kasprintf(GFP_KERNEL, "Ivy Bridge SrcID#%d_Ha#%d",
3182 pvt->sbridge_dev->source_id, pvt->sbridge_dev->dom);
3183 break;
3184 case SANDY_BRIDGE:
3185 pvt->info.rankcfgr = SB_RANK_CFG_A;
3186 pvt->info.get_tolm = sbridge_get_tolm;
3187 pvt->info.get_tohm = sbridge_get_tohm;
3188 pvt->info.dram_rule = sbridge_dram_rule;
3189 pvt->info.get_memory_type = get_memory_type;
3190 pvt->info.get_node_id = get_node_id;
3191 pvt->info.rir_limit = rir_limit;
3192 pvt->info.sad_limit = sad_limit;
3193 pvt->info.interleave_mode = interleave_mode;
3194 pvt->info.dram_attr = dram_attr;
3195 pvt->info.max_sad = ARRAY_SIZE(sbridge_dram_rule);
3196 pvt->info.interleave_list = sbridge_interleave_list;
3197 pvt->info.max_interleave = ARRAY_SIZE(sbridge_interleave_list);
3198 pvt->info.interleave_pkg = sbridge_interleave_pkg;
3199 pvt->info.get_width = sbridge_get_width;
3201 /* Store pci devices at mci for faster access */
3202 rc = sbridge_mci_bind_devs(mci, sbridge_dev);
3203 if (unlikely(rc < 0))
3204 goto fail0;
3205 get_source_id(mci);
3206 mci->ctl_name = kasprintf(GFP_KERNEL, "Sandy Bridge SrcID#%d_Ha#%d",
3207 pvt->sbridge_dev->source_id, pvt->sbridge_dev->dom);
3208 break;
3209 case HASWELL:
3210 /* rankcfgr isn't used */
3211 pvt->info.get_tolm = haswell_get_tolm;
3212 pvt->info.get_tohm = haswell_get_tohm;
3213 pvt->info.dram_rule = ibridge_dram_rule;
3214 pvt->info.get_memory_type = haswell_get_memory_type;
3215 pvt->info.get_node_id = haswell_get_node_id;
3216 pvt->info.rir_limit = haswell_rir_limit;
3217 pvt->info.sad_limit = sad_limit;
3218 pvt->info.interleave_mode = interleave_mode;
3219 pvt->info.dram_attr = dram_attr;
3220 pvt->info.max_sad = ARRAY_SIZE(ibridge_dram_rule);
3221 pvt->info.interleave_list = ibridge_interleave_list;
3222 pvt->info.max_interleave = ARRAY_SIZE(ibridge_interleave_list);
3223 pvt->info.interleave_pkg = ibridge_interleave_pkg;
3224 pvt->info.get_width = ibridge_get_width;
3226 /* Store pci devices at mci for faster access */
3227 rc = haswell_mci_bind_devs(mci, sbridge_dev);
3228 if (unlikely(rc < 0))
3229 goto fail0;
3230 get_source_id(mci);
3231 mci->ctl_name = kasprintf(GFP_KERNEL, "Haswell SrcID#%d_Ha#%d",
3232 pvt->sbridge_dev->source_id, pvt->sbridge_dev->dom);
3233 break;
3234 case BROADWELL:
3235 /* rankcfgr isn't used */
3236 pvt->info.get_tolm = haswell_get_tolm;
3237 pvt->info.get_tohm = haswell_get_tohm;
3238 pvt->info.dram_rule = ibridge_dram_rule;
3239 pvt->info.get_memory_type = haswell_get_memory_type;
3240 pvt->info.get_node_id = haswell_get_node_id;
3241 pvt->info.rir_limit = haswell_rir_limit;
3242 pvt->info.sad_limit = sad_limit;
3243 pvt->info.interleave_mode = interleave_mode;
3244 pvt->info.dram_attr = dram_attr;
3245 pvt->info.max_sad = ARRAY_SIZE(ibridge_dram_rule);
3246 pvt->info.interleave_list = ibridge_interleave_list;
3247 pvt->info.max_interleave = ARRAY_SIZE(ibridge_interleave_list);
3248 pvt->info.interleave_pkg = ibridge_interleave_pkg;
3249 pvt->info.get_width = broadwell_get_width;
3251 /* Store pci devices at mci for faster access */
3252 rc = broadwell_mci_bind_devs(mci, sbridge_dev);
3253 if (unlikely(rc < 0))
3254 goto fail0;
3255 get_source_id(mci);
3256 mci->ctl_name = kasprintf(GFP_KERNEL, "Broadwell SrcID#%d_Ha#%d",
3257 pvt->sbridge_dev->source_id, pvt->sbridge_dev->dom);
3258 break;
3259 case KNIGHTS_LANDING:
3260 /* pvt->info.rankcfgr == ??? */
3261 pvt->info.get_tolm = knl_get_tolm;
3262 pvt->info.get_tohm = knl_get_tohm;
3263 pvt->info.dram_rule = knl_dram_rule;
3264 pvt->info.get_memory_type = knl_get_memory_type;
3265 pvt->info.get_node_id = knl_get_node_id;
3266 pvt->info.rir_limit = NULL;
3267 pvt->info.sad_limit = knl_sad_limit;
3268 pvt->info.interleave_mode = knl_interleave_mode;
3269 pvt->info.dram_attr = dram_attr_knl;
3270 pvt->info.max_sad = ARRAY_SIZE(knl_dram_rule);
3271 pvt->info.interleave_list = knl_interleave_list;
3272 pvt->info.max_interleave = ARRAY_SIZE(knl_interleave_list);
3273 pvt->info.interleave_pkg = ibridge_interleave_pkg;
3274 pvt->info.get_width = knl_get_width;
3276 rc = knl_mci_bind_devs(mci, sbridge_dev);
3277 if (unlikely(rc < 0))
3278 goto fail0;
3279 get_source_id(mci);
3280 mci->ctl_name = kasprintf(GFP_KERNEL, "Knights Landing SrcID#%d_Ha#%d",
3281 pvt->sbridge_dev->source_id, pvt->sbridge_dev->dom);
3282 break;
3285 if (!mci->ctl_name) {
3286 rc = -ENOMEM;
3287 goto fail0;
3290 /* Get dimm basic config and the memory layout */
3291 rc = get_dimm_config(mci);
3292 if (rc < 0) {
3293 edac_dbg(0, "MC: failed to get_dimm_config()\n");
3294 goto fail;
3296 get_memory_layout(mci);
3298 /* record ptr to the generic device */
3299 mci->pdev = &pdev->dev;
3301 /* add this new MC control structure to EDAC's list of MCs */
3302 if (unlikely(edac_mc_add_mc(mci))) {
3303 edac_dbg(0, "MC: failed edac_mc_add_mc()\n");
3304 rc = -EINVAL;
3305 goto fail;
3308 return 0;
3310 fail:
3311 kfree(mci->ctl_name);
3312 fail0:
3313 edac_mc_free(mci);
3314 sbridge_dev->mci = NULL;
3315 return rc;
3318 #define ICPU(model, table) \
3319 { X86_VENDOR_INTEL, 6, model, 0, (unsigned long)&table }
3321 static const struct x86_cpu_id sbridge_cpuids[] = {
3322 ICPU(INTEL_FAM6_SANDYBRIDGE_X, pci_dev_descr_sbridge_table),
3323 ICPU(INTEL_FAM6_IVYBRIDGE_X, pci_dev_descr_ibridge_table),
3324 ICPU(INTEL_FAM6_HASWELL_X, pci_dev_descr_haswell_table),
3325 ICPU(INTEL_FAM6_BROADWELL_X, pci_dev_descr_broadwell_table),
3326 ICPU(INTEL_FAM6_BROADWELL_XEON_D, pci_dev_descr_broadwell_table),
3327 ICPU(INTEL_FAM6_XEON_PHI_KNL, pci_dev_descr_knl_table),
3328 ICPU(INTEL_FAM6_XEON_PHI_KNM, pci_dev_descr_knl_table),
3331 MODULE_DEVICE_TABLE(x86cpu, sbridge_cpuids);
3334 * sbridge_probe Get all devices and register memory controllers
3335 * present.
3336 * return:
3337 * 0 for FOUND a device
3338 * < 0 for error code
3341 static int sbridge_probe(const struct x86_cpu_id *id)
3343 int rc = -ENODEV;
3344 u8 mc, num_mc = 0;
3345 struct sbridge_dev *sbridge_dev;
3346 struct pci_id_table *ptable = (struct pci_id_table *)id->driver_data;
3348 /* get the pci devices we want to reserve for our use */
3349 rc = sbridge_get_all_devices(&num_mc, ptable);
3351 if (unlikely(rc < 0)) {
3352 edac_dbg(0, "couldn't get all devices\n");
3353 goto fail0;
3356 mc = 0;
3358 list_for_each_entry(sbridge_dev, &sbridge_edac_list, list) {
3359 edac_dbg(0, "Registering MC#%d (%d of %d)\n",
3360 mc, mc + 1, num_mc);
3362 sbridge_dev->mc = mc++;
3363 rc = sbridge_register_mci(sbridge_dev, ptable->type);
3364 if (unlikely(rc < 0))
3365 goto fail1;
3368 sbridge_printk(KERN_INFO, "%s\n", SBRIDGE_REVISION);
3370 return 0;
3372 fail1:
3373 list_for_each_entry(sbridge_dev, &sbridge_edac_list, list)
3374 sbridge_unregister_mci(sbridge_dev);
3376 sbridge_put_all_devices();
3377 fail0:
3378 return rc;
3382 * sbridge_remove cleanup
3385 static void sbridge_remove(void)
3387 struct sbridge_dev *sbridge_dev;
3389 edac_dbg(0, "\n");
3391 list_for_each_entry(sbridge_dev, &sbridge_edac_list, list)
3392 sbridge_unregister_mci(sbridge_dev);
3394 /* Release PCI resources */
3395 sbridge_put_all_devices();
3399 * sbridge_init Module entry function
3400 * Try to initialize this module for its devices
3402 static int __init sbridge_init(void)
3404 const struct x86_cpu_id *id;
3405 const char *owner;
3406 int rc;
3408 edac_dbg(2, "\n");
3410 owner = edac_get_owner();
3411 if (owner && strncmp(owner, EDAC_MOD_STR, sizeof(EDAC_MOD_STR)))
3412 return -EBUSY;
3414 id = x86_match_cpu(sbridge_cpuids);
3415 if (!id)
3416 return -ENODEV;
3418 /* Ensure that the OPSTATE is set correctly for POLL or NMI */
3419 opstate_init();
3421 rc = sbridge_probe(id);
3423 if (rc >= 0) {
3424 mce_register_decode_chain(&sbridge_mce_dec);
3425 if (edac_get_report_status() == EDAC_REPORTING_DISABLED)
3426 sbridge_printk(KERN_WARNING, "Loading driver, error reporting disabled.\n");
3427 return 0;
3430 sbridge_printk(KERN_ERR, "Failed to register device with error %d.\n",
3431 rc);
3433 return rc;
3437 * sbridge_exit() Module exit function
3438 * Unregister the driver
3440 static void __exit sbridge_exit(void)
3442 edac_dbg(2, "\n");
3443 sbridge_remove();
3444 mce_unregister_decode_chain(&sbridge_mce_dec);
3447 module_init(sbridge_init);
3448 module_exit(sbridge_exit);
3450 module_param(edac_op_state, int, 0444);
3451 MODULE_PARM_DESC(edac_op_state, "EDAC Error Reporting state: 0=Poll,1=NMI");
3453 MODULE_LICENSE("GPL");
3454 MODULE_AUTHOR("Mauro Carvalho Chehab");
3455 MODULE_AUTHOR("Red Hat Inc. (http://www.redhat.com)");
3456 MODULE_DESCRIPTION("MC Driver for Intel Sandy Bridge and Ivy Bridge memory controllers - "
3457 SBRIDGE_REVISION);