gpio: rcar: Fix runtime PM imbalance on error
[linux/fpc-iii.git] / drivers / edac / sb_edac.c
blob7d51c82be62ba3cf9d379c53c67ee7b4caa53f08
1 // SPDX-License-Identifier: GPL-2.0-only
2 /* Intel Sandy Bridge -EN/-EP/-EX Memory Controller kernel module
4 * This driver supports the memory controllers found on the Intel
5 * processor family Sandy Bridge.
7 * Copyright (c) 2011 by:
8 * Mauro Carvalho Chehab
9 */
11 #include <linux/module.h>
12 #include <linux/init.h>
13 #include <linux/pci.h>
14 #include <linux/pci_ids.h>
15 #include <linux/slab.h>
16 #include <linux/delay.h>
17 #include <linux/edac.h>
18 #include <linux/mmzone.h>
19 #include <linux/smp.h>
20 #include <linux/bitmap.h>
21 #include <linux/math64.h>
22 #include <linux/mod_devicetable.h>
23 #include <asm/cpu_device_id.h>
24 #include <asm/intel-family.h>
25 #include <asm/processor.h>
26 #include <asm/mce.h>
28 #include "edac_module.h"
30 /* Static vars */
31 static LIST_HEAD(sbridge_edac_list);
34 * Alter this version for the module when modifications are made
36 #define SBRIDGE_REVISION " Ver: 1.1.2 "
37 #define EDAC_MOD_STR "sb_edac"
40 * Debug macros
42 #define sbridge_printk(level, fmt, arg...) \
43 edac_printk(level, "sbridge", fmt, ##arg)
45 #define sbridge_mc_printk(mci, level, fmt, arg...) \
46 edac_mc_chipset_printk(mci, level, "sbridge", fmt, ##arg)
49 * Get a bit field at register value <v>, from bit <lo> to bit <hi>
51 #define GET_BITFIELD(v, lo, hi) \
52 (((v) & GENMASK_ULL(hi, lo)) >> (lo))
54 /* Devices 12 Function 6, Offsets 0x80 to 0xcc */
55 static const u32 sbridge_dram_rule[] = {
56 0x80, 0x88, 0x90, 0x98, 0xa0,
57 0xa8, 0xb0, 0xb8, 0xc0, 0xc8,
60 static const u32 ibridge_dram_rule[] = {
61 0x60, 0x68, 0x70, 0x78, 0x80,
62 0x88, 0x90, 0x98, 0xa0, 0xa8,
63 0xb0, 0xb8, 0xc0, 0xc8, 0xd0,
64 0xd8, 0xe0, 0xe8, 0xf0, 0xf8,
67 static const u32 knl_dram_rule[] = {
68 0x60, 0x68, 0x70, 0x78, 0x80, /* 0-4 */
69 0x88, 0x90, 0x98, 0xa0, 0xa8, /* 5-9 */
70 0xb0, 0xb8, 0xc0, 0xc8, 0xd0, /* 10-14 */
71 0xd8, 0xe0, 0xe8, 0xf0, 0xf8, /* 15-19 */
72 0x100, 0x108, 0x110, 0x118, /* 20-23 */
75 #define DRAM_RULE_ENABLE(reg) GET_BITFIELD(reg, 0, 0)
76 #define A7MODE(reg) GET_BITFIELD(reg, 26, 26)
78 static char *show_dram_attr(u32 attr)
80 switch (attr) {
81 case 0:
82 return "DRAM";
83 case 1:
84 return "MMCFG";
85 case 2:
86 return "NXM";
87 default:
88 return "unknown";
92 static const u32 sbridge_interleave_list[] = {
93 0x84, 0x8c, 0x94, 0x9c, 0xa4,
94 0xac, 0xb4, 0xbc, 0xc4, 0xcc,
97 static const u32 ibridge_interleave_list[] = {
98 0x64, 0x6c, 0x74, 0x7c, 0x84,
99 0x8c, 0x94, 0x9c, 0xa4, 0xac,
100 0xb4, 0xbc, 0xc4, 0xcc, 0xd4,
101 0xdc, 0xe4, 0xec, 0xf4, 0xfc,
104 static const u32 knl_interleave_list[] = {
105 0x64, 0x6c, 0x74, 0x7c, 0x84, /* 0-4 */
106 0x8c, 0x94, 0x9c, 0xa4, 0xac, /* 5-9 */
107 0xb4, 0xbc, 0xc4, 0xcc, 0xd4, /* 10-14 */
108 0xdc, 0xe4, 0xec, 0xf4, 0xfc, /* 15-19 */
109 0x104, 0x10c, 0x114, 0x11c, /* 20-23 */
111 #define MAX_INTERLEAVE \
112 (max_t(unsigned int, ARRAY_SIZE(sbridge_interleave_list), \
113 max_t(unsigned int, ARRAY_SIZE(ibridge_interleave_list), \
114 ARRAY_SIZE(knl_interleave_list))))
116 struct interleave_pkg {
117 unsigned char start;
118 unsigned char end;
121 static const struct interleave_pkg sbridge_interleave_pkg[] = {
122 { 0, 2 },
123 { 3, 5 },
124 { 8, 10 },
125 { 11, 13 },
126 { 16, 18 },
127 { 19, 21 },
128 { 24, 26 },
129 { 27, 29 },
132 static const struct interleave_pkg ibridge_interleave_pkg[] = {
133 { 0, 3 },
134 { 4, 7 },
135 { 8, 11 },
136 { 12, 15 },
137 { 16, 19 },
138 { 20, 23 },
139 { 24, 27 },
140 { 28, 31 },
143 static inline int sad_pkg(const struct interleave_pkg *table, u32 reg,
144 int interleave)
146 return GET_BITFIELD(reg, table[interleave].start,
147 table[interleave].end);
150 /* Devices 12 Function 7 */
152 #define TOLM 0x80
153 #define TOHM 0x84
154 #define HASWELL_TOLM 0xd0
155 #define HASWELL_TOHM_0 0xd4
156 #define HASWELL_TOHM_1 0xd8
157 #define KNL_TOLM 0xd0
158 #define KNL_TOHM_0 0xd4
159 #define KNL_TOHM_1 0xd8
161 #define GET_TOLM(reg) ((GET_BITFIELD(reg, 0, 3) << 28) | 0x3ffffff)
162 #define GET_TOHM(reg) ((GET_BITFIELD(reg, 0, 20) << 25) | 0x3ffffff)
164 /* Device 13 Function 6 */
166 #define SAD_TARGET 0xf0
168 #define SOURCE_ID(reg) GET_BITFIELD(reg, 9, 11)
170 #define SOURCE_ID_KNL(reg) GET_BITFIELD(reg, 12, 14)
172 #define SAD_CONTROL 0xf4
174 /* Device 14 function 0 */
176 static const u32 tad_dram_rule[] = {
177 0x40, 0x44, 0x48, 0x4c,
178 0x50, 0x54, 0x58, 0x5c,
179 0x60, 0x64, 0x68, 0x6c,
181 #define MAX_TAD ARRAY_SIZE(tad_dram_rule)
183 #define TAD_LIMIT(reg) ((GET_BITFIELD(reg, 12, 31) << 26) | 0x3ffffff)
184 #define TAD_SOCK(reg) GET_BITFIELD(reg, 10, 11)
185 #define TAD_CH(reg) GET_BITFIELD(reg, 8, 9)
186 #define TAD_TGT3(reg) GET_BITFIELD(reg, 6, 7)
187 #define TAD_TGT2(reg) GET_BITFIELD(reg, 4, 5)
188 #define TAD_TGT1(reg) GET_BITFIELD(reg, 2, 3)
189 #define TAD_TGT0(reg) GET_BITFIELD(reg, 0, 1)
191 /* Device 15, function 0 */
193 #define MCMTR 0x7c
194 #define KNL_MCMTR 0x624
196 #define IS_ECC_ENABLED(mcmtr) GET_BITFIELD(mcmtr, 2, 2)
197 #define IS_LOCKSTEP_ENABLED(mcmtr) GET_BITFIELD(mcmtr, 1, 1)
198 #define IS_CLOSE_PG(mcmtr) GET_BITFIELD(mcmtr, 0, 0)
200 /* Device 15, function 1 */
202 #define RASENABLES 0xac
203 #define IS_MIRROR_ENABLED(reg) GET_BITFIELD(reg, 0, 0)
205 /* Device 15, functions 2-5 */
207 static const int mtr_regs[] = {
208 0x80, 0x84, 0x88,
211 static const int knl_mtr_reg = 0xb60;
213 #define RANK_DISABLE(mtr) GET_BITFIELD(mtr, 16, 19)
214 #define IS_DIMM_PRESENT(mtr) GET_BITFIELD(mtr, 14, 14)
215 #define RANK_CNT_BITS(mtr) GET_BITFIELD(mtr, 12, 13)
216 #define RANK_WIDTH_BITS(mtr) GET_BITFIELD(mtr, 2, 4)
217 #define COL_WIDTH_BITS(mtr) GET_BITFIELD(mtr, 0, 1)
219 static const u32 tad_ch_nilv_offset[] = {
220 0x90, 0x94, 0x98, 0x9c,
221 0xa0, 0xa4, 0xa8, 0xac,
222 0xb0, 0xb4, 0xb8, 0xbc,
224 #define CHN_IDX_OFFSET(reg) GET_BITFIELD(reg, 28, 29)
225 #define TAD_OFFSET(reg) (GET_BITFIELD(reg, 6, 25) << 26)
227 static const u32 rir_way_limit[] = {
228 0x108, 0x10c, 0x110, 0x114, 0x118,
230 #define MAX_RIR_RANGES ARRAY_SIZE(rir_way_limit)
232 #define IS_RIR_VALID(reg) GET_BITFIELD(reg, 31, 31)
233 #define RIR_WAY(reg) GET_BITFIELD(reg, 28, 29)
235 #define MAX_RIR_WAY 8
237 static const u32 rir_offset[MAX_RIR_RANGES][MAX_RIR_WAY] = {
238 { 0x120, 0x124, 0x128, 0x12c, 0x130, 0x134, 0x138, 0x13c },
239 { 0x140, 0x144, 0x148, 0x14c, 0x150, 0x154, 0x158, 0x15c },
240 { 0x160, 0x164, 0x168, 0x16c, 0x170, 0x174, 0x178, 0x17c },
241 { 0x180, 0x184, 0x188, 0x18c, 0x190, 0x194, 0x198, 0x19c },
242 { 0x1a0, 0x1a4, 0x1a8, 0x1ac, 0x1b0, 0x1b4, 0x1b8, 0x1bc },
245 #define RIR_RNK_TGT(type, reg) (((type) == BROADWELL) ? \
246 GET_BITFIELD(reg, 20, 23) : GET_BITFIELD(reg, 16, 19))
248 #define RIR_OFFSET(type, reg) (((type) == HASWELL || (type) == BROADWELL) ? \
249 GET_BITFIELD(reg, 2, 15) : GET_BITFIELD(reg, 2, 14))
251 /* Device 16, functions 2-7 */
254 * FIXME: Implement the error count reads directly
257 #define RANK_ODD_OV(reg) GET_BITFIELD(reg, 31, 31)
258 #define RANK_ODD_ERR_CNT(reg) GET_BITFIELD(reg, 16, 30)
259 #define RANK_EVEN_OV(reg) GET_BITFIELD(reg, 15, 15)
260 #define RANK_EVEN_ERR_CNT(reg) GET_BITFIELD(reg, 0, 14)
262 #if 0 /* Currently unused*/
263 static const u32 correrrcnt[] = {
264 0x104, 0x108, 0x10c, 0x110,
267 static const u32 correrrthrsld[] = {
268 0x11c, 0x120, 0x124, 0x128,
270 #endif
272 #define RANK_ODD_ERR_THRSLD(reg) GET_BITFIELD(reg, 16, 30)
273 #define RANK_EVEN_ERR_THRSLD(reg) GET_BITFIELD(reg, 0, 14)
276 /* Device 17, function 0 */
278 #define SB_RANK_CFG_A 0x0328
280 #define IB_RANK_CFG_A 0x0320
283 * sbridge structs
286 #define NUM_CHANNELS 6 /* Max channels per MC */
287 #define MAX_DIMMS 3 /* Max DIMMS per channel */
288 #define KNL_MAX_CHAS 38 /* KNL max num. of Cache Home Agents */
289 #define KNL_MAX_CHANNELS 6 /* KNL max num. of PCI channels */
290 #define KNL_MAX_EDCS 8 /* Embedded DRAM controllers */
291 #define CHANNEL_UNSPECIFIED 0xf /* Intel IA32 SDM 15-14 */
293 enum type {
294 SANDY_BRIDGE,
295 IVY_BRIDGE,
296 HASWELL,
297 BROADWELL,
298 KNIGHTS_LANDING,
301 enum domain {
302 IMC0 = 0,
303 IMC1,
304 SOCK,
307 enum mirroring_mode {
308 NON_MIRRORING,
309 ADDR_RANGE_MIRRORING,
310 FULL_MIRRORING,
313 struct sbridge_pvt;
314 struct sbridge_info {
315 enum type type;
316 u32 mcmtr;
317 u32 rankcfgr;
318 u64 (*get_tolm)(struct sbridge_pvt *pvt);
319 u64 (*get_tohm)(struct sbridge_pvt *pvt);
320 u64 (*rir_limit)(u32 reg);
321 u64 (*sad_limit)(u32 reg);
322 u32 (*interleave_mode)(u32 reg);
323 u32 (*dram_attr)(u32 reg);
324 const u32 *dram_rule;
325 const u32 *interleave_list;
326 const struct interleave_pkg *interleave_pkg;
327 u8 max_sad;
328 u8 (*get_node_id)(struct sbridge_pvt *pvt);
329 u8 (*get_ha)(u8 bank);
330 enum mem_type (*get_memory_type)(struct sbridge_pvt *pvt);
331 enum dev_type (*get_width)(struct sbridge_pvt *pvt, u32 mtr);
332 struct pci_dev *pci_vtd;
335 struct sbridge_channel {
336 u32 ranks;
337 u32 dimms;
340 struct pci_id_descr {
341 int dev_id;
342 int optional;
343 enum domain dom;
346 struct pci_id_table {
347 const struct pci_id_descr *descr;
348 int n_devs_per_imc;
349 int n_devs_per_sock;
350 int n_imcs_per_sock;
351 enum type type;
354 struct sbridge_dev {
355 struct list_head list;
356 int seg;
357 u8 bus, mc;
358 u8 node_id, source_id;
359 struct pci_dev **pdev;
360 enum domain dom;
361 int n_devs;
362 int i_devs;
363 struct mem_ctl_info *mci;
366 struct knl_pvt {
367 struct pci_dev *pci_cha[KNL_MAX_CHAS];
368 struct pci_dev *pci_channel[KNL_MAX_CHANNELS];
369 struct pci_dev *pci_mc0;
370 struct pci_dev *pci_mc1;
371 struct pci_dev *pci_mc0_misc;
372 struct pci_dev *pci_mc1_misc;
373 struct pci_dev *pci_mc_info; /* tolm, tohm */
376 struct sbridge_pvt {
377 /* Devices per socket */
378 struct pci_dev *pci_ddrio;
379 struct pci_dev *pci_sad0, *pci_sad1;
380 struct pci_dev *pci_br0, *pci_br1;
381 /* Devices per memory controller */
382 struct pci_dev *pci_ha, *pci_ta, *pci_ras;
383 struct pci_dev *pci_tad[NUM_CHANNELS];
385 struct sbridge_dev *sbridge_dev;
387 struct sbridge_info info;
388 struct sbridge_channel channel[NUM_CHANNELS];
390 /* Memory type detection */
391 bool is_cur_addr_mirrored, is_lockstep, is_close_pg;
392 bool is_chan_hash;
393 enum mirroring_mode mirror_mode;
395 /* Memory description */
396 u64 tolm, tohm;
397 struct knl_pvt knl;
400 #define PCI_DESCR(device_id, opt, domain) \
401 .dev_id = (device_id), \
402 .optional = opt, \
403 .dom = domain
405 static const struct pci_id_descr pci_dev_descr_sbridge[] = {
406 /* Processor Home Agent */
407 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_HA0, 0, IMC0) },
409 /* Memory controller */
410 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TA, 0, IMC0) },
411 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_RAS, 0, IMC0) },
412 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD0, 0, IMC0) },
413 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD1, 0, IMC0) },
414 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD2, 0, IMC0) },
415 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD3, 0, IMC0) },
416 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_DDRIO, 1, SOCK) },
418 /* System Address Decoder */
419 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_SAD0, 0, SOCK) },
420 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_SAD1, 0, SOCK) },
422 /* Broadcast Registers */
423 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_BR, 0, SOCK) },
426 #define PCI_ID_TABLE_ENTRY(A, N, M, T) { \
427 .descr = A, \
428 .n_devs_per_imc = N, \
429 .n_devs_per_sock = ARRAY_SIZE(A), \
430 .n_imcs_per_sock = M, \
431 .type = T \
434 static const struct pci_id_table pci_dev_descr_sbridge_table[] = {
435 PCI_ID_TABLE_ENTRY(pci_dev_descr_sbridge, ARRAY_SIZE(pci_dev_descr_sbridge), 1, SANDY_BRIDGE),
436 {0,} /* 0 terminated list. */
439 /* This changes depending if 1HA or 2HA:
440 * 1HA:
441 * 0x0eb8 (17.0) is DDRIO0
442 * 2HA:
443 * 0x0ebc (17.4) is DDRIO0
445 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_1HA_DDRIO0 0x0eb8
446 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_2HA_DDRIO0 0x0ebc
448 /* pci ids */
449 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0 0x0ea0
450 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TA 0x0ea8
451 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_RAS 0x0e71
452 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD0 0x0eaa
453 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD1 0x0eab
454 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD2 0x0eac
455 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD3 0x0ead
456 #define PCI_DEVICE_ID_INTEL_IBRIDGE_SAD 0x0ec8
457 #define PCI_DEVICE_ID_INTEL_IBRIDGE_BR0 0x0ec9
458 #define PCI_DEVICE_ID_INTEL_IBRIDGE_BR1 0x0eca
459 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1 0x0e60
460 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TA 0x0e68
461 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_RAS 0x0e79
462 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD0 0x0e6a
463 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD1 0x0e6b
464 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD2 0x0e6c
465 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD3 0x0e6d
467 static const struct pci_id_descr pci_dev_descr_ibridge[] = {
468 /* Processor Home Agent */
469 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0, 0, IMC0) },
470 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1, 1, IMC1) },
472 /* Memory controller */
473 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TA, 0, IMC0) },
474 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_RAS, 0, IMC0) },
475 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD0, 0, IMC0) },
476 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD1, 0, IMC0) },
477 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD2, 0, IMC0) },
478 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD3, 0, IMC0) },
480 /* Optional, mode 2HA */
481 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TA, 1, IMC1) },
482 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_RAS, 1, IMC1) },
483 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD0, 1, IMC1) },
484 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD1, 1, IMC1) },
485 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD2, 1, IMC1) },
486 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD3, 1, IMC1) },
488 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_1HA_DDRIO0, 1, SOCK) },
489 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_2HA_DDRIO0, 1, SOCK) },
491 /* System Address Decoder */
492 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_SAD, 0, SOCK) },
494 /* Broadcast Registers */
495 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_BR0, 1, SOCK) },
496 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_BR1, 0, SOCK) },
500 static const struct pci_id_table pci_dev_descr_ibridge_table[] = {
501 PCI_ID_TABLE_ENTRY(pci_dev_descr_ibridge, 12, 2, IVY_BRIDGE),
502 {0,} /* 0 terminated list. */
505 /* Haswell support */
506 /* EN processor:
507 * - 1 IMC
508 * - 3 DDR3 channels, 2 DPC per channel
509 * EP processor:
510 * - 1 or 2 IMC
511 * - 4 DDR4 channels, 3 DPC per channel
512 * EP 4S processor:
513 * - 2 IMC
514 * - 4 DDR4 channels, 3 DPC per channel
515 * EX processor:
516 * - 2 IMC
517 * - each IMC interfaces with a SMI 2 channel
518 * - each SMI channel interfaces with a scalable memory buffer
519 * - each scalable memory buffer supports 4 DDR3/DDR4 channels, 3 DPC
521 #define HASWELL_DDRCRCLKCONTROLS 0xa10 /* Ditto on Broadwell */
522 #define HASWELL_HASYSDEFEATURE2 0x84
523 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_VTD_MISC 0x2f28
524 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0 0x2fa0
525 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1 0x2f60
526 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TA 0x2fa8
527 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TM 0x2f71
528 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TA 0x2f68
529 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TM 0x2f79
530 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_CBO_SAD0 0x2ffc
531 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_CBO_SAD1 0x2ffd
532 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD0 0x2faa
533 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD1 0x2fab
534 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD2 0x2fac
535 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD3 0x2fad
536 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD0 0x2f6a
537 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD1 0x2f6b
538 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD2 0x2f6c
539 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD3 0x2f6d
540 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO0 0x2fbd
541 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO1 0x2fbf
542 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO2 0x2fb9
543 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO3 0x2fbb
544 static const struct pci_id_descr pci_dev_descr_haswell[] = {
545 /* first item must be the HA */
546 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0, 0, IMC0) },
547 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1, 1, IMC1) },
549 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TA, 0, IMC0) },
550 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TM, 0, IMC0) },
551 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD0, 0, IMC0) },
552 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD1, 0, IMC0) },
553 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD2, 1, IMC0) },
554 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD3, 1, IMC0) },
556 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TA, 1, IMC1) },
557 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TM, 1, IMC1) },
558 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD0, 1, IMC1) },
559 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD1, 1, IMC1) },
560 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD2, 1, IMC1) },
561 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD3, 1, IMC1) },
563 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_CBO_SAD0, 0, SOCK) },
564 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_CBO_SAD1, 0, SOCK) },
565 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO0, 1, SOCK) },
566 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO1, 1, SOCK) },
567 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO2, 1, SOCK) },
568 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO3, 1, SOCK) },
571 static const struct pci_id_table pci_dev_descr_haswell_table[] = {
572 PCI_ID_TABLE_ENTRY(pci_dev_descr_haswell, 13, 2, HASWELL),
573 {0,} /* 0 terminated list. */
576 /* Knight's Landing Support */
578 * KNL's memory channels are swizzled between memory controllers.
579 * MC0 is mapped to CH3,4,5 and MC1 is mapped to CH0,1,2
581 #define knl_channel_remap(mc, chan) ((mc) ? (chan) : (chan) + 3)
583 /* Memory controller, TAD tables, error injection - 2-8-0, 2-9-0 (2 of these) */
584 #define PCI_DEVICE_ID_INTEL_KNL_IMC_MC 0x7840
585 /* DRAM channel stuff; bank addrs, dimmmtr, etc.. 2-8-2 - 2-9-4 (6 of these) */
586 #define PCI_DEVICE_ID_INTEL_KNL_IMC_CHAN 0x7843
587 /* kdrwdbu TAD limits/offsets, MCMTR - 2-10-1, 2-11-1 (2 of these) */
588 #define PCI_DEVICE_ID_INTEL_KNL_IMC_TA 0x7844
589 /* CHA broadcast registers, dram rules - 1-29-0 (1 of these) */
590 #define PCI_DEVICE_ID_INTEL_KNL_IMC_SAD0 0x782a
591 /* SAD target - 1-29-1 (1 of these) */
592 #define PCI_DEVICE_ID_INTEL_KNL_IMC_SAD1 0x782b
593 /* Caching / Home Agent */
594 #define PCI_DEVICE_ID_INTEL_KNL_IMC_CHA 0x782c
595 /* Device with TOLM and TOHM, 0-5-0 (1 of these) */
596 #define PCI_DEVICE_ID_INTEL_KNL_IMC_TOLHM 0x7810
599 * KNL differs from SB, IB, and Haswell in that it has multiple
600 * instances of the same device with the same device ID, so we handle that
601 * by creating as many copies in the table as we expect to find.
602 * (Like device ID must be grouped together.)
605 static const struct pci_id_descr pci_dev_descr_knl[] = {
606 [0 ... 1] = { PCI_DESCR(PCI_DEVICE_ID_INTEL_KNL_IMC_MC, 0, IMC0)},
607 [2 ... 7] = { PCI_DESCR(PCI_DEVICE_ID_INTEL_KNL_IMC_CHAN, 0, IMC0) },
608 [8] = { PCI_DESCR(PCI_DEVICE_ID_INTEL_KNL_IMC_TA, 0, IMC0) },
609 [9] = { PCI_DESCR(PCI_DEVICE_ID_INTEL_KNL_IMC_TOLHM, 0, IMC0) },
610 [10] = { PCI_DESCR(PCI_DEVICE_ID_INTEL_KNL_IMC_SAD0, 0, SOCK) },
611 [11] = { PCI_DESCR(PCI_DEVICE_ID_INTEL_KNL_IMC_SAD1, 0, SOCK) },
612 [12 ... 49] = { PCI_DESCR(PCI_DEVICE_ID_INTEL_KNL_IMC_CHA, 0, SOCK) },
615 static const struct pci_id_table pci_dev_descr_knl_table[] = {
616 PCI_ID_TABLE_ENTRY(pci_dev_descr_knl, ARRAY_SIZE(pci_dev_descr_knl), 1, KNIGHTS_LANDING),
617 {0,}
621 * Broadwell support
623 * DE processor:
624 * - 1 IMC
625 * - 2 DDR3 channels, 2 DPC per channel
626 * EP processor:
627 * - 1 or 2 IMC
628 * - 4 DDR4 channels, 3 DPC per channel
629 * EP 4S processor:
630 * - 2 IMC
631 * - 4 DDR4 channels, 3 DPC per channel
632 * EX processor:
633 * - 2 IMC
634 * - each IMC interfaces with a SMI 2 channel
635 * - each SMI channel interfaces with a scalable memory buffer
636 * - each scalable memory buffer supports 4 DDR3/DDR4 channels, 3 DPC
638 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_VTD_MISC 0x6f28
639 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0 0x6fa0
640 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1 0x6f60
641 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TA 0x6fa8
642 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TM 0x6f71
643 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TA 0x6f68
644 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TM 0x6f79
645 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_CBO_SAD0 0x6ffc
646 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_CBO_SAD1 0x6ffd
647 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD0 0x6faa
648 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD1 0x6fab
649 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD2 0x6fac
650 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD3 0x6fad
651 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD0 0x6f6a
652 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD1 0x6f6b
653 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD2 0x6f6c
654 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD3 0x6f6d
655 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_DDRIO0 0x6faf
657 static const struct pci_id_descr pci_dev_descr_broadwell[] = {
658 /* first item must be the HA */
659 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0, 0, IMC0) },
660 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1, 1, IMC1) },
662 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TA, 0, IMC0) },
663 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TM, 0, IMC0) },
664 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD0, 0, IMC0) },
665 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD1, 0, IMC0) },
666 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD2, 1, IMC0) },
667 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD3, 1, IMC0) },
669 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TA, 1, IMC1) },
670 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TM, 1, IMC1) },
671 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD0, 1, IMC1) },
672 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD1, 1, IMC1) },
673 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD2, 1, IMC1) },
674 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD3, 1, IMC1) },
676 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_CBO_SAD0, 0, SOCK) },
677 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_CBO_SAD1, 0, SOCK) },
678 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_DDRIO0, 1, SOCK) },
681 static const struct pci_id_table pci_dev_descr_broadwell_table[] = {
682 PCI_ID_TABLE_ENTRY(pci_dev_descr_broadwell, 10, 2, BROADWELL),
683 {0,} /* 0 terminated list. */
687 /****************************************************************************
688 Ancillary status routines
689 ****************************************************************************/
691 static inline int numrank(enum type type, u32 mtr)
693 int ranks = (1 << RANK_CNT_BITS(mtr));
694 int max = 4;
696 if (type == HASWELL || type == BROADWELL || type == KNIGHTS_LANDING)
697 max = 8;
699 if (ranks > max) {
700 edac_dbg(0, "Invalid number of ranks: %d (max = %i) raw value = %x (%04x)\n",
701 ranks, max, (unsigned int)RANK_CNT_BITS(mtr), mtr);
702 return -EINVAL;
705 return ranks;
708 static inline int numrow(u32 mtr)
710 int rows = (RANK_WIDTH_BITS(mtr) + 12);
712 if (rows < 13 || rows > 18) {
713 edac_dbg(0, "Invalid number of rows: %d (should be between 14 and 17) raw value = %x (%04x)\n",
714 rows, (unsigned int)RANK_WIDTH_BITS(mtr), mtr);
715 return -EINVAL;
718 return 1 << rows;
721 static inline int numcol(u32 mtr)
723 int cols = (COL_WIDTH_BITS(mtr) + 10);
725 if (cols > 12) {
726 edac_dbg(0, "Invalid number of cols: %d (max = 4) raw value = %x (%04x)\n",
727 cols, (unsigned int)COL_WIDTH_BITS(mtr), mtr);
728 return -EINVAL;
731 return 1 << cols;
734 static struct sbridge_dev *get_sbridge_dev(int seg, u8 bus, enum domain dom,
735 int multi_bus,
736 struct sbridge_dev *prev)
738 struct sbridge_dev *sbridge_dev;
741 * If we have devices scattered across several busses that pertain
742 * to the same memory controller, we'll lump them all together.
744 if (multi_bus) {
745 return list_first_entry_or_null(&sbridge_edac_list,
746 struct sbridge_dev, list);
749 sbridge_dev = list_entry(prev ? prev->list.next
750 : sbridge_edac_list.next, struct sbridge_dev, list);
752 list_for_each_entry_from(sbridge_dev, &sbridge_edac_list, list) {
753 if ((sbridge_dev->seg == seg) && (sbridge_dev->bus == bus) &&
754 (dom == SOCK || dom == sbridge_dev->dom))
755 return sbridge_dev;
758 return NULL;
761 static struct sbridge_dev *alloc_sbridge_dev(int seg, u8 bus, enum domain dom,
762 const struct pci_id_table *table)
764 struct sbridge_dev *sbridge_dev;
766 sbridge_dev = kzalloc(sizeof(*sbridge_dev), GFP_KERNEL);
767 if (!sbridge_dev)
768 return NULL;
770 sbridge_dev->pdev = kcalloc(table->n_devs_per_imc,
771 sizeof(*sbridge_dev->pdev),
772 GFP_KERNEL);
773 if (!sbridge_dev->pdev) {
774 kfree(sbridge_dev);
775 return NULL;
778 sbridge_dev->seg = seg;
779 sbridge_dev->bus = bus;
780 sbridge_dev->dom = dom;
781 sbridge_dev->n_devs = table->n_devs_per_imc;
782 list_add_tail(&sbridge_dev->list, &sbridge_edac_list);
784 return sbridge_dev;
787 static void free_sbridge_dev(struct sbridge_dev *sbridge_dev)
789 list_del(&sbridge_dev->list);
790 kfree(sbridge_dev->pdev);
791 kfree(sbridge_dev);
794 static u64 sbridge_get_tolm(struct sbridge_pvt *pvt)
796 u32 reg;
798 /* Address range is 32:28 */
799 pci_read_config_dword(pvt->pci_sad1, TOLM, &reg);
800 return GET_TOLM(reg);
803 static u64 sbridge_get_tohm(struct sbridge_pvt *pvt)
805 u32 reg;
807 pci_read_config_dword(pvt->pci_sad1, TOHM, &reg);
808 return GET_TOHM(reg);
811 static u64 ibridge_get_tolm(struct sbridge_pvt *pvt)
813 u32 reg;
815 pci_read_config_dword(pvt->pci_br1, TOLM, &reg);
817 return GET_TOLM(reg);
820 static u64 ibridge_get_tohm(struct sbridge_pvt *pvt)
822 u32 reg;
824 pci_read_config_dword(pvt->pci_br1, TOHM, &reg);
826 return GET_TOHM(reg);
829 static u64 rir_limit(u32 reg)
831 return ((u64)GET_BITFIELD(reg, 1, 10) << 29) | 0x1fffffff;
834 static u64 sad_limit(u32 reg)
836 return (GET_BITFIELD(reg, 6, 25) << 26) | 0x3ffffff;
839 static u32 interleave_mode(u32 reg)
841 return GET_BITFIELD(reg, 1, 1);
844 static u32 dram_attr(u32 reg)
846 return GET_BITFIELD(reg, 2, 3);
849 static u64 knl_sad_limit(u32 reg)
851 return (GET_BITFIELD(reg, 7, 26) << 26) | 0x3ffffff;
854 static u32 knl_interleave_mode(u32 reg)
856 return GET_BITFIELD(reg, 1, 2);
859 static const char * const knl_intlv_mode[] = {
860 "[8:6]", "[10:8]", "[14:12]", "[32:30]"
863 static const char *get_intlv_mode_str(u32 reg, enum type t)
865 if (t == KNIGHTS_LANDING)
866 return knl_intlv_mode[knl_interleave_mode(reg)];
867 else
868 return interleave_mode(reg) ? "[8:6]" : "[8:6]XOR[18:16]";
871 static u32 dram_attr_knl(u32 reg)
873 return GET_BITFIELD(reg, 3, 4);
877 static enum mem_type get_memory_type(struct sbridge_pvt *pvt)
879 u32 reg;
880 enum mem_type mtype;
882 if (pvt->pci_ddrio) {
883 pci_read_config_dword(pvt->pci_ddrio, pvt->info.rankcfgr,
884 &reg);
885 if (GET_BITFIELD(reg, 11, 11))
886 /* FIXME: Can also be LRDIMM */
887 mtype = MEM_RDDR3;
888 else
889 mtype = MEM_DDR3;
890 } else
891 mtype = MEM_UNKNOWN;
893 return mtype;
896 static enum mem_type haswell_get_memory_type(struct sbridge_pvt *pvt)
898 u32 reg;
899 bool registered = false;
900 enum mem_type mtype = MEM_UNKNOWN;
902 if (!pvt->pci_ddrio)
903 goto out;
905 pci_read_config_dword(pvt->pci_ddrio,
906 HASWELL_DDRCRCLKCONTROLS, &reg);
907 /* Is_Rdimm */
908 if (GET_BITFIELD(reg, 16, 16))
909 registered = true;
911 pci_read_config_dword(pvt->pci_ta, MCMTR, &reg);
912 if (GET_BITFIELD(reg, 14, 14)) {
913 if (registered)
914 mtype = MEM_RDDR4;
915 else
916 mtype = MEM_DDR4;
917 } else {
918 if (registered)
919 mtype = MEM_RDDR3;
920 else
921 mtype = MEM_DDR3;
924 out:
925 return mtype;
928 static enum dev_type knl_get_width(struct sbridge_pvt *pvt, u32 mtr)
930 /* for KNL value is fixed */
931 return DEV_X16;
934 static enum dev_type sbridge_get_width(struct sbridge_pvt *pvt, u32 mtr)
936 /* there's no way to figure out */
937 return DEV_UNKNOWN;
940 static enum dev_type __ibridge_get_width(u32 mtr)
942 enum dev_type type;
944 switch (mtr) {
945 case 3:
946 type = DEV_UNKNOWN;
947 break;
948 case 2:
949 type = DEV_X16;
950 break;
951 case 1:
952 type = DEV_X8;
953 break;
954 case 0:
955 type = DEV_X4;
956 break;
959 return type;
962 static enum dev_type ibridge_get_width(struct sbridge_pvt *pvt, u32 mtr)
965 * ddr3_width on the documentation but also valid for DDR4 on
966 * Haswell
968 return __ibridge_get_width(GET_BITFIELD(mtr, 7, 8));
971 static enum dev_type broadwell_get_width(struct sbridge_pvt *pvt, u32 mtr)
973 /* ddr3_width on the documentation but also valid for DDR4 */
974 return __ibridge_get_width(GET_BITFIELD(mtr, 8, 9));
977 static enum mem_type knl_get_memory_type(struct sbridge_pvt *pvt)
979 /* DDR4 RDIMMS and LRDIMMS are supported */
980 return MEM_RDDR4;
983 static u8 get_node_id(struct sbridge_pvt *pvt)
985 u32 reg;
986 pci_read_config_dword(pvt->pci_br0, SAD_CONTROL, &reg);
987 return GET_BITFIELD(reg, 0, 2);
990 static u8 haswell_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, 3);
998 static u8 knl_get_node_id(struct sbridge_pvt *pvt)
1000 u32 reg;
1002 pci_read_config_dword(pvt->pci_sad1, SAD_CONTROL, &reg);
1003 return GET_BITFIELD(reg, 0, 2);
1007 * Use the reporting bank number to determine which memory
1008 * controller (also known as "ha" for "home agent"). Sandy
1009 * Bridge only has one memory controller per socket, so the
1010 * answer is always zero.
1012 static u8 sbridge_get_ha(u8 bank)
1014 return 0;
1018 * On Ivy Bridge, Haswell and Broadwell the error may be in a
1019 * home agent bank (7, 8), or one of the per-channel memory
1020 * controller banks (9 .. 16).
1022 static u8 ibridge_get_ha(u8 bank)
1024 switch (bank) {
1025 case 7 ... 8:
1026 return bank - 7;
1027 case 9 ... 16:
1028 return (bank - 9) / 4;
1029 default:
1030 return 0xff;
1034 /* Not used, but included for safety/symmetry */
1035 static u8 knl_get_ha(u8 bank)
1037 return 0xff;
1040 static u64 haswell_get_tolm(struct sbridge_pvt *pvt)
1042 u32 reg;
1044 pci_read_config_dword(pvt->info.pci_vtd, HASWELL_TOLM, &reg);
1045 return (GET_BITFIELD(reg, 26, 31) << 26) | 0x3ffffff;
1048 static u64 haswell_get_tohm(struct sbridge_pvt *pvt)
1050 u64 rc;
1051 u32 reg;
1053 pci_read_config_dword(pvt->info.pci_vtd, HASWELL_TOHM_0, &reg);
1054 rc = GET_BITFIELD(reg, 26, 31);
1055 pci_read_config_dword(pvt->info.pci_vtd, HASWELL_TOHM_1, &reg);
1056 rc = ((reg << 6) | rc) << 26;
1058 return rc | 0x1ffffff;
1061 static u64 knl_get_tolm(struct sbridge_pvt *pvt)
1063 u32 reg;
1065 pci_read_config_dword(pvt->knl.pci_mc_info, KNL_TOLM, &reg);
1066 return (GET_BITFIELD(reg, 26, 31) << 26) | 0x3ffffff;
1069 static u64 knl_get_tohm(struct sbridge_pvt *pvt)
1071 u64 rc;
1072 u32 reg_lo, reg_hi;
1074 pci_read_config_dword(pvt->knl.pci_mc_info, KNL_TOHM_0, &reg_lo);
1075 pci_read_config_dword(pvt->knl.pci_mc_info, KNL_TOHM_1, &reg_hi);
1076 rc = ((u64)reg_hi << 32) | reg_lo;
1077 return rc | 0x3ffffff;
1081 static u64 haswell_rir_limit(u32 reg)
1083 return (((u64)GET_BITFIELD(reg, 1, 11) + 1) << 29) - 1;
1086 static inline u8 sad_pkg_socket(u8 pkg)
1088 /* on Ivy Bridge, nodeID is SASS, where A is HA and S is node id */
1089 return ((pkg >> 3) << 2) | (pkg & 0x3);
1092 static inline u8 sad_pkg_ha(u8 pkg)
1094 return (pkg >> 2) & 0x1;
1097 static int haswell_chan_hash(int idx, u64 addr)
1099 int i;
1102 * XOR even bits from 12:26 to bit0 of idx,
1103 * odd bits from 13:27 to bit1
1105 for (i = 12; i < 28; i += 2)
1106 idx ^= (addr >> i) & 3;
1108 return idx;
1111 /* Low bits of TAD limit, and some metadata. */
1112 static const u32 knl_tad_dram_limit_lo[] = {
1113 0x400, 0x500, 0x600, 0x700,
1114 0x800, 0x900, 0xa00, 0xb00,
1117 /* Low bits of TAD offset. */
1118 static const u32 knl_tad_dram_offset_lo[] = {
1119 0x404, 0x504, 0x604, 0x704,
1120 0x804, 0x904, 0xa04, 0xb04,
1123 /* High 16 bits of TAD limit and offset. */
1124 static const u32 knl_tad_dram_hi[] = {
1125 0x408, 0x508, 0x608, 0x708,
1126 0x808, 0x908, 0xa08, 0xb08,
1129 /* Number of ways a tad entry is interleaved. */
1130 static const u32 knl_tad_ways[] = {
1131 8, 6, 4, 3, 2, 1,
1135 * Retrieve the n'th Target Address Decode table entry
1136 * from the memory controller's TAD table.
1138 * @pvt: driver private data
1139 * @entry: which entry you want to retrieve
1140 * @mc: which memory controller (0 or 1)
1141 * @offset: output tad range offset
1142 * @limit: output address of first byte above tad range
1143 * @ways: output number of interleave ways
1145 * The offset value has curious semantics. It's a sort of running total
1146 * of the sizes of all the memory regions that aren't mapped in this
1147 * tad table.
1149 static int knl_get_tad(const struct sbridge_pvt *pvt,
1150 const int entry,
1151 const int mc,
1152 u64 *offset,
1153 u64 *limit,
1154 int *ways)
1156 u32 reg_limit_lo, reg_offset_lo, reg_hi;
1157 struct pci_dev *pci_mc;
1158 int way_id;
1160 switch (mc) {
1161 case 0:
1162 pci_mc = pvt->knl.pci_mc0;
1163 break;
1164 case 1:
1165 pci_mc = pvt->knl.pci_mc1;
1166 break;
1167 default:
1168 WARN_ON(1);
1169 return -EINVAL;
1172 pci_read_config_dword(pci_mc,
1173 knl_tad_dram_limit_lo[entry], &reg_limit_lo);
1174 pci_read_config_dword(pci_mc,
1175 knl_tad_dram_offset_lo[entry], &reg_offset_lo);
1176 pci_read_config_dword(pci_mc,
1177 knl_tad_dram_hi[entry], &reg_hi);
1179 /* Is this TAD entry enabled? */
1180 if (!GET_BITFIELD(reg_limit_lo, 0, 0))
1181 return -ENODEV;
1183 way_id = GET_BITFIELD(reg_limit_lo, 3, 5);
1185 if (way_id < ARRAY_SIZE(knl_tad_ways)) {
1186 *ways = knl_tad_ways[way_id];
1187 } else {
1188 *ways = 0;
1189 sbridge_printk(KERN_ERR,
1190 "Unexpected value %d in mc_tad_limit_lo wayness field\n",
1191 way_id);
1192 return -ENODEV;
1196 * The least significant 6 bits of base and limit are truncated.
1197 * For limit, we fill the missing bits with 1s.
1199 *offset = ((u64) GET_BITFIELD(reg_offset_lo, 6, 31) << 6) |
1200 ((u64) GET_BITFIELD(reg_hi, 0, 15) << 32);
1201 *limit = ((u64) GET_BITFIELD(reg_limit_lo, 6, 31) << 6) | 63 |
1202 ((u64) GET_BITFIELD(reg_hi, 16, 31) << 32);
1204 return 0;
1207 /* Determine which memory controller is responsible for a given channel. */
1208 static int knl_channel_mc(int channel)
1210 WARN_ON(channel < 0 || channel >= 6);
1212 return channel < 3 ? 1 : 0;
1216 * Get the Nth entry from EDC_ROUTE_TABLE register.
1217 * (This is the per-tile mapping of logical interleave targets to
1218 * physical EDC modules.)
1220 * entry 0: 0:2
1221 * 1: 3:5
1222 * 2: 6:8
1223 * 3: 9:11
1224 * 4: 12:14
1225 * 5: 15:17
1226 * 6: 18:20
1227 * 7: 21:23
1228 * reserved: 24:31
1230 static u32 knl_get_edc_route(int entry, u32 reg)
1232 WARN_ON(entry >= KNL_MAX_EDCS);
1233 return GET_BITFIELD(reg, entry*3, (entry*3)+2);
1237 * Get the Nth entry from MC_ROUTE_TABLE register.
1238 * (This is the per-tile mapping of logical interleave targets to
1239 * physical DRAM channels modules.)
1241 * entry 0: mc 0:2 channel 18:19
1242 * 1: mc 3:5 channel 20:21
1243 * 2: mc 6:8 channel 22:23
1244 * 3: mc 9:11 channel 24:25
1245 * 4: mc 12:14 channel 26:27
1246 * 5: mc 15:17 channel 28:29
1247 * reserved: 30:31
1249 * Though we have 3 bits to identify the MC, we should only see
1250 * the values 0 or 1.
1253 static u32 knl_get_mc_route(int entry, u32 reg)
1255 int mc, chan;
1257 WARN_ON(entry >= KNL_MAX_CHANNELS);
1259 mc = GET_BITFIELD(reg, entry*3, (entry*3)+2);
1260 chan = GET_BITFIELD(reg, (entry*2) + 18, (entry*2) + 18 + 1);
1262 return knl_channel_remap(mc, chan);
1266 * Render the EDC_ROUTE register in human-readable form.
1267 * Output string s should be at least KNL_MAX_EDCS*2 bytes.
1269 static void knl_show_edc_route(u32 reg, char *s)
1271 int i;
1273 for (i = 0; i < KNL_MAX_EDCS; i++) {
1274 s[i*2] = knl_get_edc_route(i, reg) + '0';
1275 s[i*2+1] = '-';
1278 s[KNL_MAX_EDCS*2 - 1] = '\0';
1282 * Render the MC_ROUTE register in human-readable form.
1283 * Output string s should be at least KNL_MAX_CHANNELS*2 bytes.
1285 static void knl_show_mc_route(u32 reg, char *s)
1287 int i;
1289 for (i = 0; i < KNL_MAX_CHANNELS; i++) {
1290 s[i*2] = knl_get_mc_route(i, reg) + '0';
1291 s[i*2+1] = '-';
1294 s[KNL_MAX_CHANNELS*2 - 1] = '\0';
1297 #define KNL_EDC_ROUTE 0xb8
1298 #define KNL_MC_ROUTE 0xb4
1300 /* Is this dram rule backed by regular DRAM in flat mode? */
1301 #define KNL_EDRAM(reg) GET_BITFIELD(reg, 29, 29)
1303 /* Is this dram rule cached? */
1304 #define KNL_CACHEABLE(reg) GET_BITFIELD(reg, 28, 28)
1306 /* Is this rule backed by edc ? */
1307 #define KNL_EDRAM_ONLY(reg) GET_BITFIELD(reg, 29, 29)
1309 /* Is this rule backed by DRAM, cacheable in EDRAM? */
1310 #define KNL_CACHEABLE(reg) GET_BITFIELD(reg, 28, 28)
1312 /* Is this rule mod3? */
1313 #define KNL_MOD3(reg) GET_BITFIELD(reg, 27, 27)
1316 * Figure out how big our RAM modules are.
1318 * The DIMMMTR register in KNL doesn't tell us the size of the DIMMs, so we
1319 * have to figure this out from the SAD rules, interleave lists, route tables,
1320 * and TAD rules.
1322 * SAD rules can have holes in them (e.g. the 3G-4G hole), so we have to
1323 * inspect the TAD rules to figure out how large the SAD regions really are.
1325 * When we know the real size of a SAD region and how many ways it's
1326 * interleaved, we know the individual contribution of each channel to
1327 * TAD is size/ways.
1329 * Finally, we have to check whether each channel participates in each SAD
1330 * region.
1332 * Fortunately, KNL only supports one DIMM per channel, so once we know how
1333 * much memory the channel uses, we know the DIMM is at least that large.
1334 * (The BIOS might possibly choose not to map all available memory, in which
1335 * case we will underreport the size of the DIMM.)
1337 * In theory, we could try to determine the EDC sizes as well, but that would
1338 * only work in flat mode, not in cache mode.
1340 * @mc_sizes: Output sizes of channels (must have space for KNL_MAX_CHANNELS
1341 * elements)
1343 static int knl_get_dimm_capacity(struct sbridge_pvt *pvt, u64 *mc_sizes)
1345 u64 sad_base, sad_limit = 0;
1346 u64 tad_base, tad_size, tad_limit, tad_deadspace, tad_livespace;
1347 int sad_rule = 0;
1348 int tad_rule = 0;
1349 int intrlv_ways, tad_ways;
1350 u32 first_pkg, pkg;
1351 int i;
1352 u64 sad_actual_size[2]; /* sad size accounting for holes, per mc */
1353 u32 dram_rule, interleave_reg;
1354 u32 mc_route_reg[KNL_MAX_CHAS];
1355 u32 edc_route_reg[KNL_MAX_CHAS];
1356 int edram_only;
1357 char edc_route_string[KNL_MAX_EDCS*2];
1358 char mc_route_string[KNL_MAX_CHANNELS*2];
1359 int cur_reg_start;
1360 int mc;
1361 int channel;
1362 int participants[KNL_MAX_CHANNELS];
1364 for (i = 0; i < KNL_MAX_CHANNELS; i++)
1365 mc_sizes[i] = 0;
1367 /* Read the EDC route table in each CHA. */
1368 cur_reg_start = 0;
1369 for (i = 0; i < KNL_MAX_CHAS; i++) {
1370 pci_read_config_dword(pvt->knl.pci_cha[i],
1371 KNL_EDC_ROUTE, &edc_route_reg[i]);
1373 if (i > 0 && edc_route_reg[i] != edc_route_reg[i-1]) {
1374 knl_show_edc_route(edc_route_reg[i-1],
1375 edc_route_string);
1376 if (cur_reg_start == i-1)
1377 edac_dbg(0, "edc route table for CHA %d: %s\n",
1378 cur_reg_start, edc_route_string);
1379 else
1380 edac_dbg(0, "edc route table for CHA %d-%d: %s\n",
1381 cur_reg_start, i-1, edc_route_string);
1382 cur_reg_start = i;
1385 knl_show_edc_route(edc_route_reg[i-1], edc_route_string);
1386 if (cur_reg_start == i-1)
1387 edac_dbg(0, "edc route table for CHA %d: %s\n",
1388 cur_reg_start, edc_route_string);
1389 else
1390 edac_dbg(0, "edc route table for CHA %d-%d: %s\n",
1391 cur_reg_start, i-1, edc_route_string);
1393 /* Read the MC route table in each CHA. */
1394 cur_reg_start = 0;
1395 for (i = 0; i < KNL_MAX_CHAS; i++) {
1396 pci_read_config_dword(pvt->knl.pci_cha[i],
1397 KNL_MC_ROUTE, &mc_route_reg[i]);
1399 if (i > 0 && mc_route_reg[i] != mc_route_reg[i-1]) {
1400 knl_show_mc_route(mc_route_reg[i-1], mc_route_string);
1401 if (cur_reg_start == i-1)
1402 edac_dbg(0, "mc route table for CHA %d: %s\n",
1403 cur_reg_start, mc_route_string);
1404 else
1405 edac_dbg(0, "mc route table for CHA %d-%d: %s\n",
1406 cur_reg_start, i-1, mc_route_string);
1407 cur_reg_start = i;
1410 knl_show_mc_route(mc_route_reg[i-1], mc_route_string);
1411 if (cur_reg_start == i-1)
1412 edac_dbg(0, "mc route table for CHA %d: %s\n",
1413 cur_reg_start, mc_route_string);
1414 else
1415 edac_dbg(0, "mc route table for CHA %d-%d: %s\n",
1416 cur_reg_start, i-1, mc_route_string);
1418 /* Process DRAM rules */
1419 for (sad_rule = 0; sad_rule < pvt->info.max_sad; sad_rule++) {
1420 /* previous limit becomes the new base */
1421 sad_base = sad_limit;
1423 pci_read_config_dword(pvt->pci_sad0,
1424 pvt->info.dram_rule[sad_rule], &dram_rule);
1426 if (!DRAM_RULE_ENABLE(dram_rule))
1427 break;
1429 edram_only = KNL_EDRAM_ONLY(dram_rule);
1431 sad_limit = pvt->info.sad_limit(dram_rule)+1;
1433 pci_read_config_dword(pvt->pci_sad0,
1434 pvt->info.interleave_list[sad_rule], &interleave_reg);
1437 * Find out how many ways this dram rule is interleaved.
1438 * We stop when we see the first channel again.
1440 first_pkg = sad_pkg(pvt->info.interleave_pkg,
1441 interleave_reg, 0);
1442 for (intrlv_ways = 1; intrlv_ways < 8; intrlv_ways++) {
1443 pkg = sad_pkg(pvt->info.interleave_pkg,
1444 interleave_reg, intrlv_ways);
1446 if ((pkg & 0x8) == 0) {
1448 * 0 bit means memory is non-local,
1449 * which KNL doesn't support
1451 edac_dbg(0, "Unexpected interleave target %d\n",
1452 pkg);
1453 return -1;
1456 if (pkg == first_pkg)
1457 break;
1459 if (KNL_MOD3(dram_rule))
1460 intrlv_ways *= 3;
1462 edac_dbg(3, "dram rule %d (base 0x%llx, limit 0x%llx), %d way interleave%s\n",
1463 sad_rule,
1464 sad_base,
1465 sad_limit,
1466 intrlv_ways,
1467 edram_only ? ", EDRAM" : "");
1470 * Find out how big the SAD region really is by iterating
1471 * over TAD tables (SAD regions may contain holes).
1472 * Each memory controller might have a different TAD table, so
1473 * we have to look at both.
1475 * Livespace is the memory that's mapped in this TAD table,
1476 * deadspace is the holes (this could be the MMIO hole, or it
1477 * could be memory that's mapped by the other TAD table but
1478 * not this one).
1480 for (mc = 0; mc < 2; mc++) {
1481 sad_actual_size[mc] = 0;
1482 tad_livespace = 0;
1483 for (tad_rule = 0;
1484 tad_rule < ARRAY_SIZE(
1485 knl_tad_dram_limit_lo);
1486 tad_rule++) {
1487 if (knl_get_tad(pvt,
1488 tad_rule,
1490 &tad_deadspace,
1491 &tad_limit,
1492 &tad_ways))
1493 break;
1495 tad_size = (tad_limit+1) -
1496 (tad_livespace + tad_deadspace);
1497 tad_livespace += tad_size;
1498 tad_base = (tad_limit+1) - tad_size;
1500 if (tad_base < sad_base) {
1501 if (tad_limit > sad_base)
1502 edac_dbg(0, "TAD region overlaps lower SAD boundary -- TAD tables may be configured incorrectly.\n");
1503 } else if (tad_base < sad_limit) {
1504 if (tad_limit+1 > sad_limit) {
1505 edac_dbg(0, "TAD region overlaps upper SAD boundary -- TAD tables may be configured incorrectly.\n");
1506 } else {
1507 /* TAD region is completely inside SAD region */
1508 edac_dbg(3, "TAD region %d 0x%llx - 0x%llx (%lld bytes) table%d\n",
1509 tad_rule, tad_base,
1510 tad_limit, tad_size,
1511 mc);
1512 sad_actual_size[mc] += tad_size;
1518 for (mc = 0; mc < 2; mc++) {
1519 edac_dbg(3, " total TAD DRAM footprint in table%d : 0x%llx (%lld bytes)\n",
1520 mc, sad_actual_size[mc], sad_actual_size[mc]);
1523 /* Ignore EDRAM rule */
1524 if (edram_only)
1525 continue;
1527 /* Figure out which channels participate in interleave. */
1528 for (channel = 0; channel < KNL_MAX_CHANNELS; channel++)
1529 participants[channel] = 0;
1531 /* For each channel, does at least one CHA have
1532 * this channel mapped to the given target?
1534 for (channel = 0; channel < KNL_MAX_CHANNELS; channel++) {
1535 int target;
1536 int cha;
1538 for (target = 0; target < KNL_MAX_CHANNELS; target++) {
1539 for (cha = 0; cha < KNL_MAX_CHAS; cha++) {
1540 if (knl_get_mc_route(target,
1541 mc_route_reg[cha]) == channel
1542 && !participants[channel]) {
1543 participants[channel] = 1;
1544 break;
1550 for (channel = 0; channel < KNL_MAX_CHANNELS; channel++) {
1551 mc = knl_channel_mc(channel);
1552 if (participants[channel]) {
1553 edac_dbg(4, "mc channel %d contributes %lld bytes via sad entry %d\n",
1554 channel,
1555 sad_actual_size[mc]/intrlv_ways,
1556 sad_rule);
1557 mc_sizes[channel] +=
1558 sad_actual_size[mc]/intrlv_ways;
1563 return 0;
1566 static void get_source_id(struct mem_ctl_info *mci)
1568 struct sbridge_pvt *pvt = mci->pvt_info;
1569 u32 reg;
1571 if (pvt->info.type == HASWELL || pvt->info.type == BROADWELL ||
1572 pvt->info.type == KNIGHTS_LANDING)
1573 pci_read_config_dword(pvt->pci_sad1, SAD_TARGET, &reg);
1574 else
1575 pci_read_config_dword(pvt->pci_br0, SAD_TARGET, &reg);
1577 if (pvt->info.type == KNIGHTS_LANDING)
1578 pvt->sbridge_dev->source_id = SOURCE_ID_KNL(reg);
1579 else
1580 pvt->sbridge_dev->source_id = SOURCE_ID(reg);
1583 static int __populate_dimms(struct mem_ctl_info *mci,
1584 u64 knl_mc_sizes[KNL_MAX_CHANNELS],
1585 enum edac_type mode)
1587 struct sbridge_pvt *pvt = mci->pvt_info;
1588 int channels = pvt->info.type == KNIGHTS_LANDING ? KNL_MAX_CHANNELS
1589 : NUM_CHANNELS;
1590 unsigned int i, j, banks, ranks, rows, cols, npages;
1591 struct dimm_info *dimm;
1592 enum mem_type mtype;
1593 u64 size;
1595 mtype = pvt->info.get_memory_type(pvt);
1596 if (mtype == MEM_RDDR3 || mtype == MEM_RDDR4)
1597 edac_dbg(0, "Memory is registered\n");
1598 else if (mtype == MEM_UNKNOWN)
1599 edac_dbg(0, "Cannot determine memory type\n");
1600 else
1601 edac_dbg(0, "Memory is unregistered\n");
1603 if (mtype == MEM_DDR4 || mtype == MEM_RDDR4)
1604 banks = 16;
1605 else
1606 banks = 8;
1608 for (i = 0; i < channels; i++) {
1609 u32 mtr;
1611 int max_dimms_per_channel;
1613 if (pvt->info.type == KNIGHTS_LANDING) {
1614 max_dimms_per_channel = 1;
1615 if (!pvt->knl.pci_channel[i])
1616 continue;
1617 } else {
1618 max_dimms_per_channel = ARRAY_SIZE(mtr_regs);
1619 if (!pvt->pci_tad[i])
1620 continue;
1623 for (j = 0; j < max_dimms_per_channel; j++) {
1624 dimm = edac_get_dimm(mci, i, j, 0);
1625 if (pvt->info.type == KNIGHTS_LANDING) {
1626 pci_read_config_dword(pvt->knl.pci_channel[i],
1627 knl_mtr_reg, &mtr);
1628 } else {
1629 pci_read_config_dword(pvt->pci_tad[i],
1630 mtr_regs[j], &mtr);
1632 edac_dbg(4, "Channel #%d MTR%d = %x\n", i, j, mtr);
1633 if (IS_DIMM_PRESENT(mtr)) {
1634 if (!IS_ECC_ENABLED(pvt->info.mcmtr)) {
1635 sbridge_printk(KERN_ERR, "CPU SrcID #%d, Ha #%d, Channel #%d has DIMMs, but ECC is disabled\n",
1636 pvt->sbridge_dev->source_id,
1637 pvt->sbridge_dev->dom, i);
1638 return -ENODEV;
1640 pvt->channel[i].dimms++;
1642 ranks = numrank(pvt->info.type, mtr);
1644 if (pvt->info.type == KNIGHTS_LANDING) {
1645 /* For DDR4, this is fixed. */
1646 cols = 1 << 10;
1647 rows = knl_mc_sizes[i] /
1648 ((u64) cols * ranks * banks * 8);
1649 } else {
1650 rows = numrow(mtr);
1651 cols = numcol(mtr);
1654 size = ((u64)rows * cols * banks * ranks) >> (20 - 3);
1655 npages = MiB_TO_PAGES(size);
1657 edac_dbg(0, "mc#%d: ha %d channel %d, dimm %d, %lld MiB (%d pages) bank: %d, rank: %d, row: %#x, col: %#x\n",
1658 pvt->sbridge_dev->mc, pvt->sbridge_dev->dom, i, j,
1659 size, npages,
1660 banks, ranks, rows, cols);
1662 dimm->nr_pages = npages;
1663 dimm->grain = 32;
1664 dimm->dtype = pvt->info.get_width(pvt, mtr);
1665 dimm->mtype = mtype;
1666 dimm->edac_mode = mode;
1667 snprintf(dimm->label, sizeof(dimm->label),
1668 "CPU_SrcID#%u_Ha#%u_Chan#%u_DIMM#%u",
1669 pvt->sbridge_dev->source_id, pvt->sbridge_dev->dom, i, j);
1674 return 0;
1677 static int get_dimm_config(struct mem_ctl_info *mci)
1679 struct sbridge_pvt *pvt = mci->pvt_info;
1680 u64 knl_mc_sizes[KNL_MAX_CHANNELS];
1681 enum edac_type mode;
1682 u32 reg;
1684 pvt->sbridge_dev->node_id = pvt->info.get_node_id(pvt);
1685 edac_dbg(0, "mc#%d: Node ID: %d, source ID: %d\n",
1686 pvt->sbridge_dev->mc,
1687 pvt->sbridge_dev->node_id,
1688 pvt->sbridge_dev->source_id);
1690 /* KNL doesn't support mirroring or lockstep,
1691 * and is always closed page
1693 if (pvt->info.type == KNIGHTS_LANDING) {
1694 mode = EDAC_S4ECD4ED;
1695 pvt->mirror_mode = NON_MIRRORING;
1696 pvt->is_cur_addr_mirrored = false;
1698 if (knl_get_dimm_capacity(pvt, knl_mc_sizes) != 0)
1699 return -1;
1700 if (pci_read_config_dword(pvt->pci_ta, KNL_MCMTR, &pvt->info.mcmtr)) {
1701 edac_dbg(0, "Failed to read KNL_MCMTR register\n");
1702 return -ENODEV;
1704 } else {
1705 if (pvt->info.type == HASWELL || pvt->info.type == BROADWELL) {
1706 if (pci_read_config_dword(pvt->pci_ha, HASWELL_HASYSDEFEATURE2, &reg)) {
1707 edac_dbg(0, "Failed to read HASWELL_HASYSDEFEATURE2 register\n");
1708 return -ENODEV;
1710 pvt->is_chan_hash = GET_BITFIELD(reg, 21, 21);
1711 if (GET_BITFIELD(reg, 28, 28)) {
1712 pvt->mirror_mode = ADDR_RANGE_MIRRORING;
1713 edac_dbg(0, "Address range partial memory mirroring is enabled\n");
1714 goto next;
1717 if (pci_read_config_dword(pvt->pci_ras, RASENABLES, &reg)) {
1718 edac_dbg(0, "Failed to read RASENABLES register\n");
1719 return -ENODEV;
1721 if (IS_MIRROR_ENABLED(reg)) {
1722 pvt->mirror_mode = FULL_MIRRORING;
1723 edac_dbg(0, "Full memory mirroring is enabled\n");
1724 } else {
1725 pvt->mirror_mode = NON_MIRRORING;
1726 edac_dbg(0, "Memory mirroring is disabled\n");
1729 next:
1730 if (pci_read_config_dword(pvt->pci_ta, MCMTR, &pvt->info.mcmtr)) {
1731 edac_dbg(0, "Failed to read MCMTR register\n");
1732 return -ENODEV;
1734 if (IS_LOCKSTEP_ENABLED(pvt->info.mcmtr)) {
1735 edac_dbg(0, "Lockstep is enabled\n");
1736 mode = EDAC_S8ECD8ED;
1737 pvt->is_lockstep = true;
1738 } else {
1739 edac_dbg(0, "Lockstep is disabled\n");
1740 mode = EDAC_S4ECD4ED;
1741 pvt->is_lockstep = false;
1743 if (IS_CLOSE_PG(pvt->info.mcmtr)) {
1744 edac_dbg(0, "address map is on closed page mode\n");
1745 pvt->is_close_pg = true;
1746 } else {
1747 edac_dbg(0, "address map is on open page mode\n");
1748 pvt->is_close_pg = false;
1752 return __populate_dimms(mci, knl_mc_sizes, mode);
1755 static void get_memory_layout(const struct mem_ctl_info *mci)
1757 struct sbridge_pvt *pvt = mci->pvt_info;
1758 int i, j, k, n_sads, n_tads, sad_interl;
1759 u32 reg;
1760 u64 limit, prv = 0;
1761 u64 tmp_mb;
1762 u32 gb, mb;
1763 u32 rir_way;
1766 * Step 1) Get TOLM/TOHM ranges
1769 pvt->tolm = pvt->info.get_tolm(pvt);
1770 tmp_mb = (1 + pvt->tolm) >> 20;
1772 gb = div_u64_rem(tmp_mb, 1024, &mb);
1773 edac_dbg(0, "TOLM: %u.%03u GB (0x%016Lx)\n",
1774 gb, (mb*1000)/1024, (u64)pvt->tolm);
1776 /* Address range is already 45:25 */
1777 pvt->tohm = pvt->info.get_tohm(pvt);
1778 tmp_mb = (1 + pvt->tohm) >> 20;
1780 gb = div_u64_rem(tmp_mb, 1024, &mb);
1781 edac_dbg(0, "TOHM: %u.%03u GB (0x%016Lx)\n",
1782 gb, (mb*1000)/1024, (u64)pvt->tohm);
1785 * Step 2) Get SAD range and SAD Interleave list
1786 * TAD registers contain the interleave wayness. However, it
1787 * seems simpler to just discover it indirectly, with the
1788 * algorithm bellow.
1790 prv = 0;
1791 for (n_sads = 0; n_sads < pvt->info.max_sad; n_sads++) {
1792 /* SAD_LIMIT Address range is 45:26 */
1793 pci_read_config_dword(pvt->pci_sad0, pvt->info.dram_rule[n_sads],
1794 &reg);
1795 limit = pvt->info.sad_limit(reg);
1797 if (!DRAM_RULE_ENABLE(reg))
1798 continue;
1800 if (limit <= prv)
1801 break;
1803 tmp_mb = (limit + 1) >> 20;
1804 gb = div_u64_rem(tmp_mb, 1024, &mb);
1805 edac_dbg(0, "SAD#%d %s up to %u.%03u GB (0x%016Lx) Interleave: %s reg=0x%08x\n",
1806 n_sads,
1807 show_dram_attr(pvt->info.dram_attr(reg)),
1808 gb, (mb*1000)/1024,
1809 ((u64)tmp_mb) << 20L,
1810 get_intlv_mode_str(reg, pvt->info.type),
1811 reg);
1812 prv = limit;
1814 pci_read_config_dword(pvt->pci_sad0, pvt->info.interleave_list[n_sads],
1815 &reg);
1816 sad_interl = sad_pkg(pvt->info.interleave_pkg, reg, 0);
1817 for (j = 0; j < 8; j++) {
1818 u32 pkg = sad_pkg(pvt->info.interleave_pkg, reg, j);
1819 if (j > 0 && sad_interl == pkg)
1820 break;
1822 edac_dbg(0, "SAD#%d, interleave #%d: %d\n",
1823 n_sads, j, pkg);
1827 if (pvt->info.type == KNIGHTS_LANDING)
1828 return;
1831 * Step 3) Get TAD range
1833 prv = 0;
1834 for (n_tads = 0; n_tads < MAX_TAD; n_tads++) {
1835 pci_read_config_dword(pvt->pci_ha, tad_dram_rule[n_tads], &reg);
1836 limit = TAD_LIMIT(reg);
1837 if (limit <= prv)
1838 break;
1839 tmp_mb = (limit + 1) >> 20;
1841 gb = div_u64_rem(tmp_mb, 1024, &mb);
1842 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",
1843 n_tads, gb, (mb*1000)/1024,
1844 ((u64)tmp_mb) << 20L,
1845 (u32)(1 << TAD_SOCK(reg)),
1846 (u32)TAD_CH(reg) + 1,
1847 (u32)TAD_TGT0(reg),
1848 (u32)TAD_TGT1(reg),
1849 (u32)TAD_TGT2(reg),
1850 (u32)TAD_TGT3(reg),
1851 reg);
1852 prv = limit;
1856 * Step 4) Get TAD offsets, per each channel
1858 for (i = 0; i < NUM_CHANNELS; i++) {
1859 if (!pvt->channel[i].dimms)
1860 continue;
1861 for (j = 0; j < n_tads; j++) {
1862 pci_read_config_dword(pvt->pci_tad[i],
1863 tad_ch_nilv_offset[j],
1864 &reg);
1865 tmp_mb = TAD_OFFSET(reg) >> 20;
1866 gb = div_u64_rem(tmp_mb, 1024, &mb);
1867 edac_dbg(0, "TAD CH#%d, offset #%d: %u.%03u GB (0x%016Lx), reg=0x%08x\n",
1868 i, j,
1869 gb, (mb*1000)/1024,
1870 ((u64)tmp_mb) << 20L,
1871 reg);
1876 * Step 6) Get RIR Wayness/Limit, per each channel
1878 for (i = 0; i < NUM_CHANNELS; i++) {
1879 if (!pvt->channel[i].dimms)
1880 continue;
1881 for (j = 0; j < MAX_RIR_RANGES; j++) {
1882 pci_read_config_dword(pvt->pci_tad[i],
1883 rir_way_limit[j],
1884 &reg);
1886 if (!IS_RIR_VALID(reg))
1887 continue;
1889 tmp_mb = pvt->info.rir_limit(reg) >> 20;
1890 rir_way = 1 << RIR_WAY(reg);
1891 gb = div_u64_rem(tmp_mb, 1024, &mb);
1892 edac_dbg(0, "CH#%d RIR#%d, limit: %u.%03u GB (0x%016Lx), way: %d, reg=0x%08x\n",
1893 i, j,
1894 gb, (mb*1000)/1024,
1895 ((u64)tmp_mb) << 20L,
1896 rir_way,
1897 reg);
1899 for (k = 0; k < rir_way; k++) {
1900 pci_read_config_dword(pvt->pci_tad[i],
1901 rir_offset[j][k],
1902 &reg);
1903 tmp_mb = RIR_OFFSET(pvt->info.type, reg) << 6;
1905 gb = div_u64_rem(tmp_mb, 1024, &mb);
1906 edac_dbg(0, "CH#%d RIR#%d INTL#%d, offset %u.%03u GB (0x%016Lx), tgt: %d, reg=0x%08x\n",
1907 i, j, k,
1908 gb, (mb*1000)/1024,
1909 ((u64)tmp_mb) << 20L,
1910 (u32)RIR_RNK_TGT(pvt->info.type, reg),
1911 reg);
1917 static struct mem_ctl_info *get_mci_for_node_id(u8 node_id, u8 ha)
1919 struct sbridge_dev *sbridge_dev;
1921 list_for_each_entry(sbridge_dev, &sbridge_edac_list, list) {
1922 if (sbridge_dev->node_id == node_id && sbridge_dev->dom == ha)
1923 return sbridge_dev->mci;
1925 return NULL;
1928 static int get_memory_error_data(struct mem_ctl_info *mci,
1929 u64 addr,
1930 u8 *socket, u8 *ha,
1931 long *channel_mask,
1932 u8 *rank,
1933 char **area_type, char *msg)
1935 struct mem_ctl_info *new_mci;
1936 struct sbridge_pvt *pvt = mci->pvt_info;
1937 struct pci_dev *pci_ha;
1938 int n_rir, n_sads, n_tads, sad_way, sck_xch;
1939 int sad_interl, idx, base_ch;
1940 int interleave_mode, shiftup = 0;
1941 unsigned int sad_interleave[MAX_INTERLEAVE];
1942 u32 reg, dram_rule;
1943 u8 ch_way, sck_way, pkg, sad_ha = 0;
1944 u32 tad_offset;
1945 u32 rir_way;
1946 u32 mb, gb;
1947 u64 ch_addr, offset, limit = 0, prv = 0;
1951 * Step 0) Check if the address is at special memory ranges
1952 * The check bellow is probably enough to fill all cases where
1953 * the error is not inside a memory, except for the legacy
1954 * range (e. g. VGA addresses). It is unlikely, however, that the
1955 * memory controller would generate an error on that range.
1957 if ((addr > (u64) pvt->tolm) && (addr < (1LL << 32))) {
1958 sprintf(msg, "Error at TOLM area, on addr 0x%08Lx", addr);
1959 return -EINVAL;
1961 if (addr >= (u64)pvt->tohm) {
1962 sprintf(msg, "Error at MMIOH area, on addr 0x%016Lx", addr);
1963 return -EINVAL;
1967 * Step 1) Get socket
1969 for (n_sads = 0; n_sads < pvt->info.max_sad; n_sads++) {
1970 pci_read_config_dword(pvt->pci_sad0, pvt->info.dram_rule[n_sads],
1971 &reg);
1973 if (!DRAM_RULE_ENABLE(reg))
1974 continue;
1976 limit = pvt->info.sad_limit(reg);
1977 if (limit <= prv) {
1978 sprintf(msg, "Can't discover the memory socket");
1979 return -EINVAL;
1981 if (addr <= limit)
1982 break;
1983 prv = limit;
1985 if (n_sads == pvt->info.max_sad) {
1986 sprintf(msg, "Can't discover the memory socket");
1987 return -EINVAL;
1989 dram_rule = reg;
1990 *area_type = show_dram_attr(pvt->info.dram_attr(dram_rule));
1991 interleave_mode = pvt->info.interleave_mode(dram_rule);
1993 pci_read_config_dword(pvt->pci_sad0, pvt->info.interleave_list[n_sads],
1994 &reg);
1996 if (pvt->info.type == SANDY_BRIDGE) {
1997 sad_interl = sad_pkg(pvt->info.interleave_pkg, reg, 0);
1998 for (sad_way = 0; sad_way < 8; sad_way++) {
1999 u32 pkg = sad_pkg(pvt->info.interleave_pkg, reg, sad_way);
2000 if (sad_way > 0 && sad_interl == pkg)
2001 break;
2002 sad_interleave[sad_way] = pkg;
2003 edac_dbg(0, "SAD interleave #%d: %d\n",
2004 sad_way, sad_interleave[sad_way]);
2006 edac_dbg(0, "mc#%d: Error detected on SAD#%d: address 0x%016Lx < 0x%016Lx, Interleave [%d:6]%s\n",
2007 pvt->sbridge_dev->mc,
2008 n_sads,
2009 addr,
2010 limit,
2011 sad_way + 7,
2012 !interleave_mode ? "" : "XOR[18:16]");
2013 if (interleave_mode)
2014 idx = ((addr >> 6) ^ (addr >> 16)) & 7;
2015 else
2016 idx = (addr >> 6) & 7;
2017 switch (sad_way) {
2018 case 1:
2019 idx = 0;
2020 break;
2021 case 2:
2022 idx = idx & 1;
2023 break;
2024 case 4:
2025 idx = idx & 3;
2026 break;
2027 case 8:
2028 break;
2029 default:
2030 sprintf(msg, "Can't discover socket interleave");
2031 return -EINVAL;
2033 *socket = sad_interleave[idx];
2034 edac_dbg(0, "SAD interleave index: %d (wayness %d) = CPU socket %d\n",
2035 idx, sad_way, *socket);
2036 } else if (pvt->info.type == HASWELL || pvt->info.type == BROADWELL) {
2037 int bits, a7mode = A7MODE(dram_rule);
2039 if (a7mode) {
2040 /* A7 mode swaps P9 with P6 */
2041 bits = GET_BITFIELD(addr, 7, 8) << 1;
2042 bits |= GET_BITFIELD(addr, 9, 9);
2043 } else
2044 bits = GET_BITFIELD(addr, 6, 8);
2046 if (interleave_mode == 0) {
2047 /* interleave mode will XOR {8,7,6} with {18,17,16} */
2048 idx = GET_BITFIELD(addr, 16, 18);
2049 idx ^= bits;
2050 } else
2051 idx = bits;
2053 pkg = sad_pkg(pvt->info.interleave_pkg, reg, idx);
2054 *socket = sad_pkg_socket(pkg);
2055 sad_ha = sad_pkg_ha(pkg);
2057 if (a7mode) {
2058 /* MCChanShiftUpEnable */
2059 pci_read_config_dword(pvt->pci_ha, HASWELL_HASYSDEFEATURE2, &reg);
2060 shiftup = GET_BITFIELD(reg, 22, 22);
2063 edac_dbg(0, "SAD interleave package: %d = CPU socket %d, HA %i, shiftup: %i\n",
2064 idx, *socket, sad_ha, shiftup);
2065 } else {
2066 /* Ivy Bridge's SAD mode doesn't support XOR interleave mode */
2067 idx = (addr >> 6) & 7;
2068 pkg = sad_pkg(pvt->info.interleave_pkg, reg, idx);
2069 *socket = sad_pkg_socket(pkg);
2070 sad_ha = sad_pkg_ha(pkg);
2071 edac_dbg(0, "SAD interleave package: %d = CPU socket %d, HA %d\n",
2072 idx, *socket, sad_ha);
2075 *ha = sad_ha;
2078 * Move to the proper node structure, in order to access the
2079 * right PCI registers
2081 new_mci = get_mci_for_node_id(*socket, sad_ha);
2082 if (!new_mci) {
2083 sprintf(msg, "Struct for socket #%u wasn't initialized",
2084 *socket);
2085 return -EINVAL;
2087 mci = new_mci;
2088 pvt = mci->pvt_info;
2091 * Step 2) Get memory channel
2093 prv = 0;
2094 pci_ha = pvt->pci_ha;
2095 for (n_tads = 0; n_tads < MAX_TAD; n_tads++) {
2096 pci_read_config_dword(pci_ha, tad_dram_rule[n_tads], &reg);
2097 limit = TAD_LIMIT(reg);
2098 if (limit <= prv) {
2099 sprintf(msg, "Can't discover the memory channel");
2100 return -EINVAL;
2102 if (addr <= limit)
2103 break;
2104 prv = limit;
2106 if (n_tads == MAX_TAD) {
2107 sprintf(msg, "Can't discover the memory channel");
2108 return -EINVAL;
2111 ch_way = TAD_CH(reg) + 1;
2112 sck_way = TAD_SOCK(reg);
2114 if (ch_way == 3)
2115 idx = addr >> 6;
2116 else {
2117 idx = (addr >> (6 + sck_way + shiftup)) & 0x3;
2118 if (pvt->is_chan_hash)
2119 idx = haswell_chan_hash(idx, addr);
2121 idx = idx % ch_way;
2124 * FIXME: Shouldn't we use CHN_IDX_OFFSET() here, when ch_way == 3 ???
2126 switch (idx) {
2127 case 0:
2128 base_ch = TAD_TGT0(reg);
2129 break;
2130 case 1:
2131 base_ch = TAD_TGT1(reg);
2132 break;
2133 case 2:
2134 base_ch = TAD_TGT2(reg);
2135 break;
2136 case 3:
2137 base_ch = TAD_TGT3(reg);
2138 break;
2139 default:
2140 sprintf(msg, "Can't discover the TAD target");
2141 return -EINVAL;
2143 *channel_mask = 1 << base_ch;
2145 pci_read_config_dword(pvt->pci_tad[base_ch], tad_ch_nilv_offset[n_tads], &tad_offset);
2147 if (pvt->mirror_mode == FULL_MIRRORING ||
2148 (pvt->mirror_mode == ADDR_RANGE_MIRRORING && n_tads == 0)) {
2149 *channel_mask |= 1 << ((base_ch + 2) % 4);
2150 switch(ch_way) {
2151 case 2:
2152 case 4:
2153 sck_xch = (1 << sck_way) * (ch_way >> 1);
2154 break;
2155 default:
2156 sprintf(msg, "Invalid mirror set. Can't decode addr");
2157 return -EINVAL;
2160 pvt->is_cur_addr_mirrored = true;
2161 } else {
2162 sck_xch = (1 << sck_way) * ch_way;
2163 pvt->is_cur_addr_mirrored = false;
2166 if (pvt->is_lockstep)
2167 *channel_mask |= 1 << ((base_ch + 1) % 4);
2169 offset = TAD_OFFSET(tad_offset);
2171 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",
2172 n_tads,
2173 addr,
2174 limit,
2175 sck_way,
2176 ch_way,
2177 offset,
2178 idx,
2179 base_ch,
2180 *channel_mask);
2182 /* Calculate channel address */
2183 /* Remove the TAD offset */
2185 if (offset > addr) {
2186 sprintf(msg, "Can't calculate ch addr: TAD offset 0x%08Lx is too high for addr 0x%08Lx!",
2187 offset, addr);
2188 return -EINVAL;
2191 ch_addr = addr - offset;
2192 ch_addr >>= (6 + shiftup);
2193 ch_addr /= sck_xch;
2194 ch_addr <<= (6 + shiftup);
2195 ch_addr |= addr & ((1 << (6 + shiftup)) - 1);
2198 * Step 3) Decode rank
2200 for (n_rir = 0; n_rir < MAX_RIR_RANGES; n_rir++) {
2201 pci_read_config_dword(pvt->pci_tad[base_ch], rir_way_limit[n_rir], &reg);
2203 if (!IS_RIR_VALID(reg))
2204 continue;
2206 limit = pvt->info.rir_limit(reg);
2207 gb = div_u64_rem(limit >> 20, 1024, &mb);
2208 edac_dbg(0, "RIR#%d, limit: %u.%03u GB (0x%016Lx), way: %d\n",
2209 n_rir,
2210 gb, (mb*1000)/1024,
2211 limit,
2212 1 << RIR_WAY(reg));
2213 if (ch_addr <= limit)
2214 break;
2216 if (n_rir == MAX_RIR_RANGES) {
2217 sprintf(msg, "Can't discover the memory rank for ch addr 0x%08Lx",
2218 ch_addr);
2219 return -EINVAL;
2221 rir_way = RIR_WAY(reg);
2223 if (pvt->is_close_pg)
2224 idx = (ch_addr >> 6);
2225 else
2226 idx = (ch_addr >> 13); /* FIXME: Datasheet says to shift by 15 */
2227 idx %= 1 << rir_way;
2229 pci_read_config_dword(pvt->pci_tad[base_ch], rir_offset[n_rir][idx], &reg);
2230 *rank = RIR_RNK_TGT(pvt->info.type, reg);
2232 edac_dbg(0, "RIR#%d: channel address 0x%08Lx < 0x%08Lx, RIR interleave %d, index %d\n",
2233 n_rir,
2234 ch_addr,
2235 limit,
2236 rir_way,
2237 idx);
2239 return 0;
2242 static int get_memory_error_data_from_mce(struct mem_ctl_info *mci,
2243 const struct mce *m, u8 *socket,
2244 u8 *ha, long *channel_mask,
2245 char *msg)
2247 u32 reg, channel = GET_BITFIELD(m->status, 0, 3);
2248 struct mem_ctl_info *new_mci;
2249 struct sbridge_pvt *pvt;
2250 struct pci_dev *pci_ha;
2251 bool tad0;
2253 if (channel >= NUM_CHANNELS) {
2254 sprintf(msg, "Invalid channel 0x%x", channel);
2255 return -EINVAL;
2258 pvt = mci->pvt_info;
2259 if (!pvt->info.get_ha) {
2260 sprintf(msg, "No get_ha()");
2261 return -EINVAL;
2263 *ha = pvt->info.get_ha(m->bank);
2264 if (*ha != 0 && *ha != 1) {
2265 sprintf(msg, "Impossible bank %d", m->bank);
2266 return -EINVAL;
2269 *socket = m->socketid;
2270 new_mci = get_mci_for_node_id(*socket, *ha);
2271 if (!new_mci) {
2272 strcpy(msg, "mci socket got corrupted!");
2273 return -EINVAL;
2276 pvt = new_mci->pvt_info;
2277 pci_ha = pvt->pci_ha;
2278 pci_read_config_dword(pci_ha, tad_dram_rule[0], &reg);
2279 tad0 = m->addr <= TAD_LIMIT(reg);
2281 *channel_mask = 1 << channel;
2282 if (pvt->mirror_mode == FULL_MIRRORING ||
2283 (pvt->mirror_mode == ADDR_RANGE_MIRRORING && tad0)) {
2284 *channel_mask |= 1 << ((channel + 2) % 4);
2285 pvt->is_cur_addr_mirrored = true;
2286 } else {
2287 pvt->is_cur_addr_mirrored = false;
2290 if (pvt->is_lockstep)
2291 *channel_mask |= 1 << ((channel + 1) % 4);
2293 return 0;
2296 /****************************************************************************
2297 Device initialization routines: put/get, init/exit
2298 ****************************************************************************/
2301 * sbridge_put_all_devices 'put' all the devices that we have
2302 * reserved via 'get'
2304 static void sbridge_put_devices(struct sbridge_dev *sbridge_dev)
2306 int i;
2308 edac_dbg(0, "\n");
2309 for (i = 0; i < sbridge_dev->n_devs; i++) {
2310 struct pci_dev *pdev = sbridge_dev->pdev[i];
2311 if (!pdev)
2312 continue;
2313 edac_dbg(0, "Removing dev %02x:%02x.%d\n",
2314 pdev->bus->number,
2315 PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn));
2316 pci_dev_put(pdev);
2320 static void sbridge_put_all_devices(void)
2322 struct sbridge_dev *sbridge_dev, *tmp;
2324 list_for_each_entry_safe(sbridge_dev, tmp, &sbridge_edac_list, list) {
2325 sbridge_put_devices(sbridge_dev);
2326 free_sbridge_dev(sbridge_dev);
2330 static int sbridge_get_onedevice(struct pci_dev **prev,
2331 u8 *num_mc,
2332 const struct pci_id_table *table,
2333 const unsigned devno,
2334 const int multi_bus)
2336 struct sbridge_dev *sbridge_dev = NULL;
2337 const struct pci_id_descr *dev_descr = &table->descr[devno];
2338 struct pci_dev *pdev = NULL;
2339 int seg = 0;
2340 u8 bus = 0;
2341 int i = 0;
2343 sbridge_printk(KERN_DEBUG,
2344 "Seeking for: PCI ID %04x:%04x\n",
2345 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
2347 pdev = pci_get_device(PCI_VENDOR_ID_INTEL,
2348 dev_descr->dev_id, *prev);
2350 if (!pdev) {
2351 if (*prev) {
2352 *prev = pdev;
2353 return 0;
2356 if (dev_descr->optional)
2357 return 0;
2359 /* if the HA wasn't found */
2360 if (devno == 0)
2361 return -ENODEV;
2363 sbridge_printk(KERN_INFO,
2364 "Device not found: %04x:%04x\n",
2365 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
2367 /* End of list, leave */
2368 return -ENODEV;
2370 seg = pci_domain_nr(pdev->bus);
2371 bus = pdev->bus->number;
2373 next_imc:
2374 sbridge_dev = get_sbridge_dev(seg, bus, dev_descr->dom,
2375 multi_bus, sbridge_dev);
2376 if (!sbridge_dev) {
2377 /* If the HA1 wasn't found, don't create EDAC second memory controller */
2378 if (dev_descr->dom == IMC1 && devno != 1) {
2379 edac_dbg(0, "Skip IMC1: %04x:%04x (since HA1 was absent)\n",
2380 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
2381 pci_dev_put(pdev);
2382 return 0;
2385 if (dev_descr->dom == SOCK)
2386 goto out_imc;
2388 sbridge_dev = alloc_sbridge_dev(seg, bus, dev_descr->dom, table);
2389 if (!sbridge_dev) {
2390 pci_dev_put(pdev);
2391 return -ENOMEM;
2393 (*num_mc)++;
2396 if (sbridge_dev->pdev[sbridge_dev->i_devs]) {
2397 sbridge_printk(KERN_ERR,
2398 "Duplicated device for %04x:%04x\n",
2399 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
2400 pci_dev_put(pdev);
2401 return -ENODEV;
2404 sbridge_dev->pdev[sbridge_dev->i_devs++] = pdev;
2406 /* pdev belongs to more than one IMC, do extra gets */
2407 if (++i > 1)
2408 pci_dev_get(pdev);
2410 if (dev_descr->dom == SOCK && i < table->n_imcs_per_sock)
2411 goto next_imc;
2413 out_imc:
2414 /* Be sure that the device is enabled */
2415 if (unlikely(pci_enable_device(pdev) < 0)) {
2416 sbridge_printk(KERN_ERR,
2417 "Couldn't enable %04x:%04x\n",
2418 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
2419 return -ENODEV;
2422 edac_dbg(0, "Detected %04x:%04x\n",
2423 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
2426 * As stated on drivers/pci/search.c, the reference count for
2427 * @from is always decremented if it is not %NULL. So, as we need
2428 * to get all devices up to null, we need to do a get for the device
2430 pci_dev_get(pdev);
2432 *prev = pdev;
2434 return 0;
2438 * sbridge_get_all_devices - Find and perform 'get' operation on the MCH's
2439 * devices we want to reference for this driver.
2440 * @num_mc: pointer to the memory controllers count, to be incremented in case
2441 * of success.
2442 * @table: model specific table
2444 * returns 0 in case of success or error code
2446 static int sbridge_get_all_devices(u8 *num_mc,
2447 const struct pci_id_table *table)
2449 int i, rc;
2450 struct pci_dev *pdev = NULL;
2451 int allow_dups = 0;
2452 int multi_bus = 0;
2454 if (table->type == KNIGHTS_LANDING)
2455 allow_dups = multi_bus = 1;
2456 while (table && table->descr) {
2457 for (i = 0; i < table->n_devs_per_sock; i++) {
2458 if (!allow_dups || i == 0 ||
2459 table->descr[i].dev_id !=
2460 table->descr[i-1].dev_id) {
2461 pdev = NULL;
2463 do {
2464 rc = sbridge_get_onedevice(&pdev, num_mc,
2465 table, i, multi_bus);
2466 if (rc < 0) {
2467 if (i == 0) {
2468 i = table->n_devs_per_sock;
2469 break;
2471 sbridge_put_all_devices();
2472 return -ENODEV;
2474 } while (pdev && !allow_dups);
2476 table++;
2479 return 0;
2483 * Device IDs for {SBRIDGE,IBRIDGE,HASWELL,BROADWELL}_IMC_HA0_TAD0 are in
2484 * the format: XXXa. So we can convert from a device to the corresponding
2485 * channel like this
2487 #define TAD_DEV_TO_CHAN(dev) (((dev) & 0xf) - 0xa)
2489 static int sbridge_mci_bind_devs(struct mem_ctl_info *mci,
2490 struct sbridge_dev *sbridge_dev)
2492 struct sbridge_pvt *pvt = mci->pvt_info;
2493 struct pci_dev *pdev;
2494 u8 saw_chan_mask = 0;
2495 int i;
2497 for (i = 0; i < sbridge_dev->n_devs; i++) {
2498 pdev = sbridge_dev->pdev[i];
2499 if (!pdev)
2500 continue;
2502 switch (pdev->device) {
2503 case PCI_DEVICE_ID_INTEL_SBRIDGE_SAD0:
2504 pvt->pci_sad0 = pdev;
2505 break;
2506 case PCI_DEVICE_ID_INTEL_SBRIDGE_SAD1:
2507 pvt->pci_sad1 = pdev;
2508 break;
2509 case PCI_DEVICE_ID_INTEL_SBRIDGE_BR:
2510 pvt->pci_br0 = pdev;
2511 break;
2512 case PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_HA0:
2513 pvt->pci_ha = pdev;
2514 break;
2515 case PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TA:
2516 pvt->pci_ta = pdev;
2517 break;
2518 case PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_RAS:
2519 pvt->pci_ras = pdev;
2520 break;
2521 case PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD0:
2522 case PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD1:
2523 case PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD2:
2524 case PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD3:
2526 int id = TAD_DEV_TO_CHAN(pdev->device);
2527 pvt->pci_tad[id] = pdev;
2528 saw_chan_mask |= 1 << id;
2530 break;
2531 case PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_DDRIO:
2532 pvt->pci_ddrio = pdev;
2533 break;
2534 default:
2535 goto error;
2538 edac_dbg(0, "Associated PCI %02x:%02x, bus %d with dev = %p\n",
2539 pdev->vendor, pdev->device,
2540 sbridge_dev->bus,
2541 pdev);
2544 /* Check if everything were registered */
2545 if (!pvt->pci_sad0 || !pvt->pci_sad1 || !pvt->pci_ha ||
2546 !pvt->pci_ras || !pvt->pci_ta)
2547 goto enodev;
2549 if (saw_chan_mask != 0x0f)
2550 goto enodev;
2551 return 0;
2553 enodev:
2554 sbridge_printk(KERN_ERR, "Some needed devices are missing\n");
2555 return -ENODEV;
2557 error:
2558 sbridge_printk(KERN_ERR, "Unexpected device %02x:%02x\n",
2559 PCI_VENDOR_ID_INTEL, pdev->device);
2560 return -EINVAL;
2563 static int ibridge_mci_bind_devs(struct mem_ctl_info *mci,
2564 struct sbridge_dev *sbridge_dev)
2566 struct sbridge_pvt *pvt = mci->pvt_info;
2567 struct pci_dev *pdev;
2568 u8 saw_chan_mask = 0;
2569 int i;
2571 for (i = 0; i < sbridge_dev->n_devs; i++) {
2572 pdev = sbridge_dev->pdev[i];
2573 if (!pdev)
2574 continue;
2576 switch (pdev->device) {
2577 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0:
2578 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1:
2579 pvt->pci_ha = pdev;
2580 break;
2581 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TA:
2582 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TA:
2583 pvt->pci_ta = pdev;
2584 break;
2585 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_RAS:
2586 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_RAS:
2587 pvt->pci_ras = pdev;
2588 break;
2589 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD0:
2590 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD1:
2591 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD2:
2592 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD3:
2593 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD0:
2594 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD1:
2595 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD2:
2596 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD3:
2598 int id = TAD_DEV_TO_CHAN(pdev->device);
2599 pvt->pci_tad[id] = pdev;
2600 saw_chan_mask |= 1 << id;
2602 break;
2603 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_2HA_DDRIO0:
2604 pvt->pci_ddrio = pdev;
2605 break;
2606 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_1HA_DDRIO0:
2607 pvt->pci_ddrio = pdev;
2608 break;
2609 case PCI_DEVICE_ID_INTEL_IBRIDGE_SAD:
2610 pvt->pci_sad0 = pdev;
2611 break;
2612 case PCI_DEVICE_ID_INTEL_IBRIDGE_BR0:
2613 pvt->pci_br0 = pdev;
2614 break;
2615 case PCI_DEVICE_ID_INTEL_IBRIDGE_BR1:
2616 pvt->pci_br1 = pdev;
2617 break;
2618 default:
2619 goto error;
2622 edac_dbg(0, "Associated PCI %02x.%02d.%d with dev = %p\n",
2623 sbridge_dev->bus,
2624 PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn),
2625 pdev);
2628 /* Check if everything were registered */
2629 if (!pvt->pci_sad0 || !pvt->pci_ha || !pvt->pci_br0 ||
2630 !pvt->pci_br1 || !pvt->pci_ras || !pvt->pci_ta)
2631 goto enodev;
2633 if (saw_chan_mask != 0x0f && /* -EN/-EX */
2634 saw_chan_mask != 0x03) /* -EP */
2635 goto enodev;
2636 return 0;
2638 enodev:
2639 sbridge_printk(KERN_ERR, "Some needed devices are missing\n");
2640 return -ENODEV;
2642 error:
2643 sbridge_printk(KERN_ERR,
2644 "Unexpected device %02x:%02x\n", PCI_VENDOR_ID_INTEL,
2645 pdev->device);
2646 return -EINVAL;
2649 static int haswell_mci_bind_devs(struct mem_ctl_info *mci,
2650 struct sbridge_dev *sbridge_dev)
2652 struct sbridge_pvt *pvt = mci->pvt_info;
2653 struct pci_dev *pdev;
2654 u8 saw_chan_mask = 0;
2655 int i;
2657 /* there's only one device per system; not tied to any bus */
2658 if (pvt->info.pci_vtd == NULL)
2659 /* result will be checked later */
2660 pvt->info.pci_vtd = pci_get_device(PCI_VENDOR_ID_INTEL,
2661 PCI_DEVICE_ID_INTEL_HASWELL_IMC_VTD_MISC,
2662 NULL);
2664 for (i = 0; i < sbridge_dev->n_devs; i++) {
2665 pdev = sbridge_dev->pdev[i];
2666 if (!pdev)
2667 continue;
2669 switch (pdev->device) {
2670 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_CBO_SAD0:
2671 pvt->pci_sad0 = pdev;
2672 break;
2673 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_CBO_SAD1:
2674 pvt->pci_sad1 = pdev;
2675 break;
2676 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0:
2677 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1:
2678 pvt->pci_ha = pdev;
2679 break;
2680 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TA:
2681 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TA:
2682 pvt->pci_ta = pdev;
2683 break;
2684 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TM:
2685 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TM:
2686 pvt->pci_ras = pdev;
2687 break;
2688 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD0:
2689 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD1:
2690 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD2:
2691 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD3:
2692 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD0:
2693 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD1:
2694 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD2:
2695 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD3:
2697 int id = TAD_DEV_TO_CHAN(pdev->device);
2698 pvt->pci_tad[id] = pdev;
2699 saw_chan_mask |= 1 << id;
2701 break;
2702 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO0:
2703 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO1:
2704 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO2:
2705 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO3:
2706 if (!pvt->pci_ddrio)
2707 pvt->pci_ddrio = pdev;
2708 break;
2709 default:
2710 break;
2713 edac_dbg(0, "Associated PCI %02x.%02d.%d with dev = %p\n",
2714 sbridge_dev->bus,
2715 PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn),
2716 pdev);
2719 /* Check if everything were registered */
2720 if (!pvt->pci_sad0 || !pvt->pci_ha || !pvt->pci_sad1 ||
2721 !pvt->pci_ras || !pvt->pci_ta || !pvt->info.pci_vtd)
2722 goto enodev;
2724 if (saw_chan_mask != 0x0f && /* -EN/-EX */
2725 saw_chan_mask != 0x03) /* -EP */
2726 goto enodev;
2727 return 0;
2729 enodev:
2730 sbridge_printk(KERN_ERR, "Some needed devices are missing\n");
2731 return -ENODEV;
2734 static int broadwell_mci_bind_devs(struct mem_ctl_info *mci,
2735 struct sbridge_dev *sbridge_dev)
2737 struct sbridge_pvt *pvt = mci->pvt_info;
2738 struct pci_dev *pdev;
2739 u8 saw_chan_mask = 0;
2740 int i;
2742 /* there's only one device per system; not tied to any bus */
2743 if (pvt->info.pci_vtd == NULL)
2744 /* result will be checked later */
2745 pvt->info.pci_vtd = pci_get_device(PCI_VENDOR_ID_INTEL,
2746 PCI_DEVICE_ID_INTEL_BROADWELL_IMC_VTD_MISC,
2747 NULL);
2749 for (i = 0; i < sbridge_dev->n_devs; i++) {
2750 pdev = sbridge_dev->pdev[i];
2751 if (!pdev)
2752 continue;
2754 switch (pdev->device) {
2755 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_CBO_SAD0:
2756 pvt->pci_sad0 = pdev;
2757 break;
2758 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_CBO_SAD1:
2759 pvt->pci_sad1 = pdev;
2760 break;
2761 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0:
2762 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1:
2763 pvt->pci_ha = pdev;
2764 break;
2765 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TA:
2766 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TA:
2767 pvt->pci_ta = pdev;
2768 break;
2769 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TM:
2770 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TM:
2771 pvt->pci_ras = pdev;
2772 break;
2773 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD0:
2774 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD1:
2775 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD2:
2776 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD3:
2777 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD0:
2778 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD1:
2779 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD2:
2780 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD3:
2782 int id = TAD_DEV_TO_CHAN(pdev->device);
2783 pvt->pci_tad[id] = pdev;
2784 saw_chan_mask |= 1 << id;
2786 break;
2787 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_DDRIO0:
2788 pvt->pci_ddrio = pdev;
2789 break;
2790 default:
2791 break;
2794 edac_dbg(0, "Associated PCI %02x.%02d.%d with dev = %p\n",
2795 sbridge_dev->bus,
2796 PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn),
2797 pdev);
2800 /* Check if everything were registered */
2801 if (!pvt->pci_sad0 || !pvt->pci_ha || !pvt->pci_sad1 ||
2802 !pvt->pci_ras || !pvt->pci_ta || !pvt->info.pci_vtd)
2803 goto enodev;
2805 if (saw_chan_mask != 0x0f && /* -EN/-EX */
2806 saw_chan_mask != 0x03) /* -EP */
2807 goto enodev;
2808 return 0;
2810 enodev:
2811 sbridge_printk(KERN_ERR, "Some needed devices are missing\n");
2812 return -ENODEV;
2815 static int knl_mci_bind_devs(struct mem_ctl_info *mci,
2816 struct sbridge_dev *sbridge_dev)
2818 struct sbridge_pvt *pvt = mci->pvt_info;
2819 struct pci_dev *pdev;
2820 int dev, func;
2822 int i;
2823 int devidx;
2825 for (i = 0; i < sbridge_dev->n_devs; i++) {
2826 pdev = sbridge_dev->pdev[i];
2827 if (!pdev)
2828 continue;
2830 /* Extract PCI device and function. */
2831 dev = (pdev->devfn >> 3) & 0x1f;
2832 func = pdev->devfn & 0x7;
2834 switch (pdev->device) {
2835 case PCI_DEVICE_ID_INTEL_KNL_IMC_MC:
2836 if (dev == 8)
2837 pvt->knl.pci_mc0 = pdev;
2838 else if (dev == 9)
2839 pvt->knl.pci_mc1 = pdev;
2840 else {
2841 sbridge_printk(KERN_ERR,
2842 "Memory controller in unexpected place! (dev %d, fn %d)\n",
2843 dev, func);
2844 continue;
2846 break;
2848 case PCI_DEVICE_ID_INTEL_KNL_IMC_SAD0:
2849 pvt->pci_sad0 = pdev;
2850 break;
2852 case PCI_DEVICE_ID_INTEL_KNL_IMC_SAD1:
2853 pvt->pci_sad1 = pdev;
2854 break;
2856 case PCI_DEVICE_ID_INTEL_KNL_IMC_CHA:
2857 /* There are one of these per tile, and range from
2858 * 1.14.0 to 1.18.5.
2860 devidx = ((dev-14)*8)+func;
2862 if (devidx < 0 || devidx >= KNL_MAX_CHAS) {
2863 sbridge_printk(KERN_ERR,
2864 "Caching and Home Agent in unexpected place! (dev %d, fn %d)\n",
2865 dev, func);
2866 continue;
2869 WARN_ON(pvt->knl.pci_cha[devidx] != NULL);
2871 pvt->knl.pci_cha[devidx] = pdev;
2872 break;
2874 case PCI_DEVICE_ID_INTEL_KNL_IMC_CHAN:
2875 devidx = -1;
2878 * MC0 channels 0-2 are device 9 function 2-4,
2879 * MC1 channels 3-5 are device 8 function 2-4.
2882 if (dev == 9)
2883 devidx = func-2;
2884 else if (dev == 8)
2885 devidx = 3 + (func-2);
2887 if (devidx < 0 || devidx >= KNL_MAX_CHANNELS) {
2888 sbridge_printk(KERN_ERR,
2889 "DRAM Channel Registers in unexpected place! (dev %d, fn %d)\n",
2890 dev, func);
2891 continue;
2894 WARN_ON(pvt->knl.pci_channel[devidx] != NULL);
2895 pvt->knl.pci_channel[devidx] = pdev;
2896 break;
2898 case PCI_DEVICE_ID_INTEL_KNL_IMC_TOLHM:
2899 pvt->knl.pci_mc_info = pdev;
2900 break;
2902 case PCI_DEVICE_ID_INTEL_KNL_IMC_TA:
2903 pvt->pci_ta = pdev;
2904 break;
2906 default:
2907 sbridge_printk(KERN_ERR, "Unexpected device %d\n",
2908 pdev->device);
2909 break;
2913 if (!pvt->knl.pci_mc0 || !pvt->knl.pci_mc1 ||
2914 !pvt->pci_sad0 || !pvt->pci_sad1 ||
2915 !pvt->pci_ta) {
2916 goto enodev;
2919 for (i = 0; i < KNL_MAX_CHANNELS; i++) {
2920 if (!pvt->knl.pci_channel[i]) {
2921 sbridge_printk(KERN_ERR, "Missing channel %d\n", i);
2922 goto enodev;
2926 for (i = 0; i < KNL_MAX_CHAS; i++) {
2927 if (!pvt->knl.pci_cha[i]) {
2928 sbridge_printk(KERN_ERR, "Missing CHA %d\n", i);
2929 goto enodev;
2933 return 0;
2935 enodev:
2936 sbridge_printk(KERN_ERR, "Some needed devices are missing\n");
2937 return -ENODEV;
2940 /****************************************************************************
2941 Error check routines
2942 ****************************************************************************/
2945 * While Sandy Bridge has error count registers, SMI BIOS read values from
2946 * and resets the counters. So, they are not reliable for the OS to read
2947 * from them. So, we have no option but to just trust on whatever MCE is
2948 * telling us about the errors.
2950 static void sbridge_mce_output_error(struct mem_ctl_info *mci,
2951 const struct mce *m)
2953 struct mem_ctl_info *new_mci;
2954 struct sbridge_pvt *pvt = mci->pvt_info;
2955 enum hw_event_mc_err_type tp_event;
2956 char *optype, msg[256];
2957 bool ripv = GET_BITFIELD(m->mcgstatus, 0, 0);
2958 bool overflow = GET_BITFIELD(m->status, 62, 62);
2959 bool uncorrected_error = GET_BITFIELD(m->status, 61, 61);
2960 bool recoverable;
2961 u32 core_err_cnt = GET_BITFIELD(m->status, 38, 52);
2962 u32 mscod = GET_BITFIELD(m->status, 16, 31);
2963 u32 errcode = GET_BITFIELD(m->status, 0, 15);
2964 u32 channel = GET_BITFIELD(m->status, 0, 3);
2965 u32 optypenum = GET_BITFIELD(m->status, 4, 6);
2967 * Bits 5-0 of MCi_MISC give the least significant bit that is valid.
2968 * A value 6 is for cache line aligned address, a value 12 is for page
2969 * aligned address reported by patrol scrubber.
2971 u32 lsb = GET_BITFIELD(m->misc, 0, 5);
2972 long channel_mask, first_channel;
2973 u8 rank = 0xff, socket, ha;
2974 int rc, dimm;
2975 char *area_type = "DRAM";
2977 if (pvt->info.type != SANDY_BRIDGE)
2978 recoverable = true;
2979 else
2980 recoverable = GET_BITFIELD(m->status, 56, 56);
2982 if (uncorrected_error) {
2983 core_err_cnt = 1;
2984 if (ripv) {
2985 tp_event = HW_EVENT_ERR_FATAL;
2986 } else {
2987 tp_event = HW_EVENT_ERR_UNCORRECTED;
2989 } else {
2990 tp_event = HW_EVENT_ERR_CORRECTED;
2994 * According with Table 15-9 of the Intel Architecture spec vol 3A,
2995 * memory errors should fit in this mask:
2996 * 000f 0000 1mmm cccc (binary)
2997 * where:
2998 * f = Correction Report Filtering Bit. If 1, subsequent errors
2999 * won't be shown
3000 * mmm = error type
3001 * cccc = channel
3002 * If the mask doesn't match, report an error to the parsing logic
3004 switch (optypenum) {
3005 case 0:
3006 optype = "generic undef request error";
3007 break;
3008 case 1:
3009 optype = "memory read error";
3010 break;
3011 case 2:
3012 optype = "memory write error";
3013 break;
3014 case 3:
3015 optype = "addr/cmd error";
3016 break;
3017 case 4:
3018 optype = "memory scrubbing error";
3019 break;
3020 default:
3021 optype = "reserved";
3022 break;
3025 if (pvt->info.type == KNIGHTS_LANDING) {
3026 if (channel == 14) {
3027 edac_dbg(0, "%s%s err_code:%04x:%04x EDRAM bank %d\n",
3028 overflow ? " OVERFLOW" : "",
3029 (uncorrected_error && recoverable)
3030 ? " recoverable" : "",
3031 mscod, errcode,
3032 m->bank);
3033 } else {
3034 char A = *("A");
3037 * Reported channel is in range 0-2, so we can't map it
3038 * back to mc. To figure out mc we check machine check
3039 * bank register that reported this error.
3040 * bank15 means mc0 and bank16 means mc1.
3042 channel = knl_channel_remap(m->bank == 16, channel);
3043 channel_mask = 1 << channel;
3045 snprintf(msg, sizeof(msg),
3046 "%s%s err_code:%04x:%04x channel:%d (DIMM_%c)",
3047 overflow ? " OVERFLOW" : "",
3048 (uncorrected_error && recoverable)
3049 ? " recoverable" : " ",
3050 mscod, errcode, channel, A + channel);
3051 edac_mc_handle_error(tp_event, mci, core_err_cnt,
3052 m->addr >> PAGE_SHIFT, m->addr & ~PAGE_MASK, 0,
3053 channel, 0, -1,
3054 optype, msg);
3056 return;
3057 } else if (lsb < 12) {
3058 rc = get_memory_error_data(mci, m->addr, &socket, &ha,
3059 &channel_mask, &rank,
3060 &area_type, msg);
3061 } else {
3062 rc = get_memory_error_data_from_mce(mci, m, &socket, &ha,
3063 &channel_mask, msg);
3066 if (rc < 0)
3067 goto err_parsing;
3068 new_mci = get_mci_for_node_id(socket, ha);
3069 if (!new_mci) {
3070 strcpy(msg, "Error: socket got corrupted!");
3071 goto err_parsing;
3073 mci = new_mci;
3074 pvt = mci->pvt_info;
3076 first_channel = find_first_bit(&channel_mask, NUM_CHANNELS);
3078 if (rank == 0xff)
3079 dimm = -1;
3080 else if (rank < 4)
3081 dimm = 0;
3082 else if (rank < 8)
3083 dimm = 1;
3084 else
3085 dimm = 2;
3088 * FIXME: On some memory configurations (mirror, lockstep), the
3089 * Memory Controller can't point the error to a single DIMM. The
3090 * EDAC core should be handling the channel mask, in order to point
3091 * to the group of dimm's where the error may be happening.
3093 if (!pvt->is_lockstep && !pvt->is_cur_addr_mirrored && !pvt->is_close_pg)
3094 channel = first_channel;
3096 snprintf(msg, sizeof(msg),
3097 "%s%s area:%s err_code:%04x:%04x socket:%d ha:%d channel_mask:%ld rank:%d",
3098 overflow ? " OVERFLOW" : "",
3099 (uncorrected_error && recoverable) ? " recoverable" : "",
3100 area_type,
3101 mscod, errcode,
3102 socket, ha,
3103 channel_mask,
3104 rank);
3106 edac_dbg(0, "%s\n", msg);
3108 /* FIXME: need support for channel mask */
3110 if (channel == CHANNEL_UNSPECIFIED)
3111 channel = -1;
3113 /* Call the helper to output message */
3114 edac_mc_handle_error(tp_event, mci, core_err_cnt,
3115 m->addr >> PAGE_SHIFT, m->addr & ~PAGE_MASK, 0,
3116 channel, dimm, -1,
3117 optype, msg);
3118 return;
3119 err_parsing:
3120 edac_mc_handle_error(tp_event, mci, core_err_cnt, 0, 0, 0,
3121 -1, -1, -1,
3122 msg, "");
3127 * Check that logging is enabled and that this is the right type
3128 * of error for us to handle.
3130 static int sbridge_mce_check_error(struct notifier_block *nb, unsigned long val,
3131 void *data)
3133 struct mce *mce = (struct mce *)data;
3134 struct mem_ctl_info *mci;
3135 char *type;
3137 if (edac_get_report_status() == EDAC_REPORTING_DISABLED)
3138 return NOTIFY_DONE;
3141 * Just let mcelog handle it if the error is
3142 * outside the memory controller. A memory error
3143 * is indicated by bit 7 = 1 and bits = 8-11,13-15 = 0.
3144 * bit 12 has an special meaning.
3146 if ((mce->status & 0xefff) >> 7 != 1)
3147 return NOTIFY_DONE;
3149 /* Check ADDRV bit in STATUS */
3150 if (!GET_BITFIELD(mce->status, 58, 58))
3151 return NOTIFY_DONE;
3153 /* Check MISCV bit in STATUS */
3154 if (!GET_BITFIELD(mce->status, 59, 59))
3155 return NOTIFY_DONE;
3157 /* Check address type in MISC (physical address only) */
3158 if (GET_BITFIELD(mce->misc, 6, 8) != 2)
3159 return NOTIFY_DONE;
3161 mci = get_mci_for_node_id(mce->socketid, IMC0);
3162 if (!mci)
3163 return NOTIFY_DONE;
3165 if (mce->mcgstatus & MCG_STATUS_MCIP)
3166 type = "Exception";
3167 else
3168 type = "Event";
3170 sbridge_mc_printk(mci, KERN_DEBUG, "HANDLING MCE MEMORY ERROR\n");
3172 sbridge_mc_printk(mci, KERN_DEBUG, "CPU %d: Machine Check %s: %Lx "
3173 "Bank %d: %016Lx\n", mce->extcpu, type,
3174 mce->mcgstatus, mce->bank, mce->status);
3175 sbridge_mc_printk(mci, KERN_DEBUG, "TSC %llx ", mce->tsc);
3176 sbridge_mc_printk(mci, KERN_DEBUG, "ADDR %llx ", mce->addr);
3177 sbridge_mc_printk(mci, KERN_DEBUG, "MISC %llx ", mce->misc);
3179 sbridge_mc_printk(mci, KERN_DEBUG, "PROCESSOR %u:%x TIME %llu SOCKET "
3180 "%u APIC %x\n", mce->cpuvendor, mce->cpuid,
3181 mce->time, mce->socketid, mce->apicid);
3183 sbridge_mce_output_error(mci, mce);
3185 /* Advice mcelog that the error were handled */
3186 return NOTIFY_STOP;
3189 static struct notifier_block sbridge_mce_dec = {
3190 .notifier_call = sbridge_mce_check_error,
3191 .priority = MCE_PRIO_EDAC,
3194 /****************************************************************************
3195 EDAC register/unregister logic
3196 ****************************************************************************/
3198 static void sbridge_unregister_mci(struct sbridge_dev *sbridge_dev)
3200 struct mem_ctl_info *mci = sbridge_dev->mci;
3202 if (unlikely(!mci || !mci->pvt_info)) {
3203 edac_dbg(0, "MC: dev = %p\n", &sbridge_dev->pdev[0]->dev);
3205 sbridge_printk(KERN_ERR, "Couldn't find mci handler\n");
3206 return;
3209 edac_dbg(0, "MC: mci = %p, dev = %p\n",
3210 mci, &sbridge_dev->pdev[0]->dev);
3212 /* Remove MC sysfs nodes */
3213 edac_mc_del_mc(mci->pdev);
3215 edac_dbg(1, "%s: free mci struct\n", mci->ctl_name);
3216 kfree(mci->ctl_name);
3217 edac_mc_free(mci);
3218 sbridge_dev->mci = NULL;
3221 static int sbridge_register_mci(struct sbridge_dev *sbridge_dev, enum type type)
3223 struct mem_ctl_info *mci;
3224 struct edac_mc_layer layers[2];
3225 struct sbridge_pvt *pvt;
3226 struct pci_dev *pdev = sbridge_dev->pdev[0];
3227 int rc;
3229 /* allocate a new MC control structure */
3230 layers[0].type = EDAC_MC_LAYER_CHANNEL;
3231 layers[0].size = type == KNIGHTS_LANDING ?
3232 KNL_MAX_CHANNELS : NUM_CHANNELS;
3233 layers[0].is_virt_csrow = false;
3234 layers[1].type = EDAC_MC_LAYER_SLOT;
3235 layers[1].size = type == KNIGHTS_LANDING ? 1 : MAX_DIMMS;
3236 layers[1].is_virt_csrow = true;
3237 mci = edac_mc_alloc(sbridge_dev->mc, ARRAY_SIZE(layers), layers,
3238 sizeof(*pvt));
3240 if (unlikely(!mci))
3241 return -ENOMEM;
3243 edac_dbg(0, "MC: mci = %p, dev = %p\n",
3244 mci, &pdev->dev);
3246 pvt = mci->pvt_info;
3247 memset(pvt, 0, sizeof(*pvt));
3249 /* Associate sbridge_dev and mci for future usage */
3250 pvt->sbridge_dev = sbridge_dev;
3251 sbridge_dev->mci = mci;
3253 mci->mtype_cap = type == KNIGHTS_LANDING ?
3254 MEM_FLAG_DDR4 : MEM_FLAG_DDR3;
3255 mci->edac_ctl_cap = EDAC_FLAG_NONE;
3256 mci->edac_cap = EDAC_FLAG_NONE;
3257 mci->mod_name = EDAC_MOD_STR;
3258 mci->dev_name = pci_name(pdev);
3259 mci->ctl_page_to_phys = NULL;
3261 pvt->info.type = type;
3262 switch (type) {
3263 case IVY_BRIDGE:
3264 pvt->info.rankcfgr = IB_RANK_CFG_A;
3265 pvt->info.get_tolm = ibridge_get_tolm;
3266 pvt->info.get_tohm = ibridge_get_tohm;
3267 pvt->info.dram_rule = ibridge_dram_rule;
3268 pvt->info.get_memory_type = get_memory_type;
3269 pvt->info.get_node_id = get_node_id;
3270 pvt->info.get_ha = ibridge_get_ha;
3271 pvt->info.rir_limit = rir_limit;
3272 pvt->info.sad_limit = sad_limit;
3273 pvt->info.interleave_mode = interleave_mode;
3274 pvt->info.dram_attr = dram_attr;
3275 pvt->info.max_sad = ARRAY_SIZE(ibridge_dram_rule);
3276 pvt->info.interleave_list = ibridge_interleave_list;
3277 pvt->info.interleave_pkg = ibridge_interleave_pkg;
3278 pvt->info.get_width = ibridge_get_width;
3280 /* Store pci devices at mci for faster access */
3281 rc = ibridge_mci_bind_devs(mci, sbridge_dev);
3282 if (unlikely(rc < 0))
3283 goto fail0;
3284 get_source_id(mci);
3285 mci->ctl_name = kasprintf(GFP_KERNEL, "Ivy Bridge SrcID#%d_Ha#%d",
3286 pvt->sbridge_dev->source_id, pvt->sbridge_dev->dom);
3287 break;
3288 case SANDY_BRIDGE:
3289 pvt->info.rankcfgr = SB_RANK_CFG_A;
3290 pvt->info.get_tolm = sbridge_get_tolm;
3291 pvt->info.get_tohm = sbridge_get_tohm;
3292 pvt->info.dram_rule = sbridge_dram_rule;
3293 pvt->info.get_memory_type = get_memory_type;
3294 pvt->info.get_node_id = get_node_id;
3295 pvt->info.get_ha = sbridge_get_ha;
3296 pvt->info.rir_limit = rir_limit;
3297 pvt->info.sad_limit = sad_limit;
3298 pvt->info.interleave_mode = interleave_mode;
3299 pvt->info.dram_attr = dram_attr;
3300 pvt->info.max_sad = ARRAY_SIZE(sbridge_dram_rule);
3301 pvt->info.interleave_list = sbridge_interleave_list;
3302 pvt->info.interleave_pkg = sbridge_interleave_pkg;
3303 pvt->info.get_width = sbridge_get_width;
3305 /* Store pci devices at mci for faster access */
3306 rc = sbridge_mci_bind_devs(mci, sbridge_dev);
3307 if (unlikely(rc < 0))
3308 goto fail0;
3309 get_source_id(mci);
3310 mci->ctl_name = kasprintf(GFP_KERNEL, "Sandy Bridge SrcID#%d_Ha#%d",
3311 pvt->sbridge_dev->source_id, pvt->sbridge_dev->dom);
3312 break;
3313 case HASWELL:
3314 /* rankcfgr isn't used */
3315 pvt->info.get_tolm = haswell_get_tolm;
3316 pvt->info.get_tohm = haswell_get_tohm;
3317 pvt->info.dram_rule = ibridge_dram_rule;
3318 pvt->info.get_memory_type = haswell_get_memory_type;
3319 pvt->info.get_node_id = haswell_get_node_id;
3320 pvt->info.get_ha = ibridge_get_ha;
3321 pvt->info.rir_limit = haswell_rir_limit;
3322 pvt->info.sad_limit = sad_limit;
3323 pvt->info.interleave_mode = interleave_mode;
3324 pvt->info.dram_attr = dram_attr;
3325 pvt->info.max_sad = ARRAY_SIZE(ibridge_dram_rule);
3326 pvt->info.interleave_list = ibridge_interleave_list;
3327 pvt->info.interleave_pkg = ibridge_interleave_pkg;
3328 pvt->info.get_width = ibridge_get_width;
3330 /* Store pci devices at mci for faster access */
3331 rc = haswell_mci_bind_devs(mci, sbridge_dev);
3332 if (unlikely(rc < 0))
3333 goto fail0;
3334 get_source_id(mci);
3335 mci->ctl_name = kasprintf(GFP_KERNEL, "Haswell SrcID#%d_Ha#%d",
3336 pvt->sbridge_dev->source_id, pvt->sbridge_dev->dom);
3337 break;
3338 case BROADWELL:
3339 /* rankcfgr isn't used */
3340 pvt->info.get_tolm = haswell_get_tolm;
3341 pvt->info.get_tohm = haswell_get_tohm;
3342 pvt->info.dram_rule = ibridge_dram_rule;
3343 pvt->info.get_memory_type = haswell_get_memory_type;
3344 pvt->info.get_node_id = haswell_get_node_id;
3345 pvt->info.get_ha = ibridge_get_ha;
3346 pvt->info.rir_limit = haswell_rir_limit;
3347 pvt->info.sad_limit = sad_limit;
3348 pvt->info.interleave_mode = interleave_mode;
3349 pvt->info.dram_attr = dram_attr;
3350 pvt->info.max_sad = ARRAY_SIZE(ibridge_dram_rule);
3351 pvt->info.interleave_list = ibridge_interleave_list;
3352 pvt->info.interleave_pkg = ibridge_interleave_pkg;
3353 pvt->info.get_width = broadwell_get_width;
3355 /* Store pci devices at mci for faster access */
3356 rc = broadwell_mci_bind_devs(mci, sbridge_dev);
3357 if (unlikely(rc < 0))
3358 goto fail0;
3359 get_source_id(mci);
3360 mci->ctl_name = kasprintf(GFP_KERNEL, "Broadwell SrcID#%d_Ha#%d",
3361 pvt->sbridge_dev->source_id, pvt->sbridge_dev->dom);
3362 break;
3363 case KNIGHTS_LANDING:
3364 /* pvt->info.rankcfgr == ??? */
3365 pvt->info.get_tolm = knl_get_tolm;
3366 pvt->info.get_tohm = knl_get_tohm;
3367 pvt->info.dram_rule = knl_dram_rule;
3368 pvt->info.get_memory_type = knl_get_memory_type;
3369 pvt->info.get_node_id = knl_get_node_id;
3370 pvt->info.get_ha = knl_get_ha;
3371 pvt->info.rir_limit = NULL;
3372 pvt->info.sad_limit = knl_sad_limit;
3373 pvt->info.interleave_mode = knl_interleave_mode;
3374 pvt->info.dram_attr = dram_attr_knl;
3375 pvt->info.max_sad = ARRAY_SIZE(knl_dram_rule);
3376 pvt->info.interleave_list = knl_interleave_list;
3377 pvt->info.interleave_pkg = ibridge_interleave_pkg;
3378 pvt->info.get_width = knl_get_width;
3380 rc = knl_mci_bind_devs(mci, sbridge_dev);
3381 if (unlikely(rc < 0))
3382 goto fail0;
3383 get_source_id(mci);
3384 mci->ctl_name = kasprintf(GFP_KERNEL, "Knights Landing SrcID#%d_Ha#%d",
3385 pvt->sbridge_dev->source_id, pvt->sbridge_dev->dom);
3386 break;
3389 if (!mci->ctl_name) {
3390 rc = -ENOMEM;
3391 goto fail0;
3394 /* Get dimm basic config and the memory layout */
3395 rc = get_dimm_config(mci);
3396 if (rc < 0) {
3397 edac_dbg(0, "MC: failed to get_dimm_config()\n");
3398 goto fail;
3400 get_memory_layout(mci);
3402 /* record ptr to the generic device */
3403 mci->pdev = &pdev->dev;
3405 /* add this new MC control structure to EDAC's list of MCs */
3406 if (unlikely(edac_mc_add_mc(mci))) {
3407 edac_dbg(0, "MC: failed edac_mc_add_mc()\n");
3408 rc = -EINVAL;
3409 goto fail;
3412 return 0;
3414 fail:
3415 kfree(mci->ctl_name);
3416 fail0:
3417 edac_mc_free(mci);
3418 sbridge_dev->mci = NULL;
3419 return rc;
3422 static const struct x86_cpu_id sbridge_cpuids[] = {
3423 X86_MATCH_INTEL_FAM6_MODEL(SANDYBRIDGE_X, &pci_dev_descr_sbridge_table),
3424 X86_MATCH_INTEL_FAM6_MODEL(IVYBRIDGE_X, &pci_dev_descr_ibridge_table),
3425 X86_MATCH_INTEL_FAM6_MODEL(HASWELL_X, &pci_dev_descr_haswell_table),
3426 X86_MATCH_INTEL_FAM6_MODEL(BROADWELL_X, &pci_dev_descr_broadwell_table),
3427 X86_MATCH_INTEL_FAM6_MODEL(BROADWELL_D, &pci_dev_descr_broadwell_table),
3428 X86_MATCH_INTEL_FAM6_MODEL(XEON_PHI_KNL, &pci_dev_descr_knl_table),
3429 X86_MATCH_INTEL_FAM6_MODEL(XEON_PHI_KNM, &pci_dev_descr_knl_table),
3432 MODULE_DEVICE_TABLE(x86cpu, sbridge_cpuids);
3435 * sbridge_probe Get all devices and register memory controllers
3436 * present.
3437 * return:
3438 * 0 for FOUND a device
3439 * < 0 for error code
3442 static int sbridge_probe(const struct x86_cpu_id *id)
3444 int rc = -ENODEV;
3445 u8 mc, num_mc = 0;
3446 struct sbridge_dev *sbridge_dev;
3447 struct pci_id_table *ptable = (struct pci_id_table *)id->driver_data;
3449 /* get the pci devices we want to reserve for our use */
3450 rc = sbridge_get_all_devices(&num_mc, ptable);
3452 if (unlikely(rc < 0)) {
3453 edac_dbg(0, "couldn't get all devices\n");
3454 goto fail0;
3457 mc = 0;
3459 list_for_each_entry(sbridge_dev, &sbridge_edac_list, list) {
3460 edac_dbg(0, "Registering MC#%d (%d of %d)\n",
3461 mc, mc + 1, num_mc);
3463 sbridge_dev->mc = mc++;
3464 rc = sbridge_register_mci(sbridge_dev, ptable->type);
3465 if (unlikely(rc < 0))
3466 goto fail1;
3469 sbridge_printk(KERN_INFO, "%s\n", SBRIDGE_REVISION);
3471 return 0;
3473 fail1:
3474 list_for_each_entry(sbridge_dev, &sbridge_edac_list, list)
3475 sbridge_unregister_mci(sbridge_dev);
3477 sbridge_put_all_devices();
3478 fail0:
3479 return rc;
3483 * sbridge_remove cleanup
3486 static void sbridge_remove(void)
3488 struct sbridge_dev *sbridge_dev;
3490 edac_dbg(0, "\n");
3492 list_for_each_entry(sbridge_dev, &sbridge_edac_list, list)
3493 sbridge_unregister_mci(sbridge_dev);
3495 /* Release PCI resources */
3496 sbridge_put_all_devices();
3500 * sbridge_init Module entry function
3501 * Try to initialize this module for its devices
3503 static int __init sbridge_init(void)
3505 const struct x86_cpu_id *id;
3506 const char *owner;
3507 int rc;
3509 edac_dbg(2, "\n");
3511 owner = edac_get_owner();
3512 if (owner && strncmp(owner, EDAC_MOD_STR, sizeof(EDAC_MOD_STR)))
3513 return -EBUSY;
3515 id = x86_match_cpu(sbridge_cpuids);
3516 if (!id)
3517 return -ENODEV;
3519 /* Ensure that the OPSTATE is set correctly for POLL or NMI */
3520 opstate_init();
3522 rc = sbridge_probe(id);
3524 if (rc >= 0) {
3525 mce_register_decode_chain(&sbridge_mce_dec);
3526 if (edac_get_report_status() == EDAC_REPORTING_DISABLED)
3527 sbridge_printk(KERN_WARNING, "Loading driver, error reporting disabled.\n");
3528 return 0;
3531 sbridge_printk(KERN_ERR, "Failed to register device with error %d.\n",
3532 rc);
3534 return rc;
3538 * sbridge_exit() Module exit function
3539 * Unregister the driver
3541 static void __exit sbridge_exit(void)
3543 edac_dbg(2, "\n");
3544 sbridge_remove();
3545 mce_unregister_decode_chain(&sbridge_mce_dec);
3548 module_init(sbridge_init);
3549 module_exit(sbridge_exit);
3551 module_param(edac_op_state, int, 0444);
3552 MODULE_PARM_DESC(edac_op_state, "EDAC Error Reporting state: 0=Poll,1=NMI");
3554 MODULE_LICENSE("GPL");
3555 MODULE_AUTHOR("Mauro Carvalho Chehab");
3556 MODULE_AUTHOR("Red Hat Inc. (http://www.redhat.com)");
3557 MODULE_DESCRIPTION("MC Driver for Intel Sandy Bridge and Ivy Bridge memory controllers - "
3558 SBRIDGE_REVISION);