sh_eth: fix EESIPR values for SH77{34|63}
[linux/fpc-iii.git] / drivers / edac / sb_edac.c
blob54ae6dc45ab2216343b0c542765537889332a0fb
1 /* Intel Sandy Bridge -EN/-EP/-EX Memory Controller kernel module
3 * This driver supports the memory controllers found on the Intel
4 * processor family Sandy Bridge.
6 * This file may be distributed under the terms of the
7 * GNU General Public License version 2 only.
9 * Copyright (c) 2011 by:
10 * Mauro Carvalho Chehab
13 #include <linux/module.h>
14 #include <linux/init.h>
15 #include <linux/pci.h>
16 #include <linux/pci_ids.h>
17 #include <linux/slab.h>
18 #include <linux/delay.h>
19 #include <linux/edac.h>
20 #include <linux/mmzone.h>
21 #include <linux/smp.h>
22 #include <linux/bitmap.h>
23 #include <linux/math64.h>
24 #include <linux/mod_devicetable.h>
25 #include <asm/cpu_device_id.h>
26 #include <asm/intel-family.h>
27 #include <asm/processor.h>
28 #include <asm/mce.h>
30 #include "edac_module.h"
32 /* Static vars */
33 static LIST_HEAD(sbridge_edac_list);
36 * Alter this version for the module when modifications are made
38 #define SBRIDGE_REVISION " Ver: 1.1.1 "
39 #define EDAC_MOD_STR "sbridge_edac"
42 * Debug macros
44 #define sbridge_printk(level, fmt, arg...) \
45 edac_printk(level, "sbridge", fmt, ##arg)
47 #define sbridge_mc_printk(mci, level, fmt, arg...) \
48 edac_mc_chipset_printk(mci, level, "sbridge", fmt, ##arg)
51 * Get a bit field at register value <v>, from bit <lo> to bit <hi>
53 #define GET_BITFIELD(v, lo, hi) \
54 (((v) & GENMASK_ULL(hi, lo)) >> (lo))
56 /* Devices 12 Function 6, Offsets 0x80 to 0xcc */
57 static const u32 sbridge_dram_rule[] = {
58 0x80, 0x88, 0x90, 0x98, 0xa0,
59 0xa8, 0xb0, 0xb8, 0xc0, 0xc8,
62 static const u32 ibridge_dram_rule[] = {
63 0x60, 0x68, 0x70, 0x78, 0x80,
64 0x88, 0x90, 0x98, 0xa0, 0xa8,
65 0xb0, 0xb8, 0xc0, 0xc8, 0xd0,
66 0xd8, 0xe0, 0xe8, 0xf0, 0xf8,
69 static const u32 knl_dram_rule[] = {
70 0x60, 0x68, 0x70, 0x78, 0x80, /* 0-4 */
71 0x88, 0x90, 0x98, 0xa0, 0xa8, /* 5-9 */
72 0xb0, 0xb8, 0xc0, 0xc8, 0xd0, /* 10-14 */
73 0xd8, 0xe0, 0xe8, 0xf0, 0xf8, /* 15-19 */
74 0x100, 0x108, 0x110, 0x118, /* 20-23 */
77 #define DRAM_RULE_ENABLE(reg) GET_BITFIELD(reg, 0, 0)
78 #define A7MODE(reg) GET_BITFIELD(reg, 26, 26)
80 static char *show_dram_attr(u32 attr)
82 switch (attr) {
83 case 0:
84 return "DRAM";
85 case 1:
86 return "MMCFG";
87 case 2:
88 return "NXM";
89 default:
90 return "unknown";
94 static const u32 sbridge_interleave_list[] = {
95 0x84, 0x8c, 0x94, 0x9c, 0xa4,
96 0xac, 0xb4, 0xbc, 0xc4, 0xcc,
99 static const u32 ibridge_interleave_list[] = {
100 0x64, 0x6c, 0x74, 0x7c, 0x84,
101 0x8c, 0x94, 0x9c, 0xa4, 0xac,
102 0xb4, 0xbc, 0xc4, 0xcc, 0xd4,
103 0xdc, 0xe4, 0xec, 0xf4, 0xfc,
106 static const u32 knl_interleave_list[] = {
107 0x64, 0x6c, 0x74, 0x7c, 0x84, /* 0-4 */
108 0x8c, 0x94, 0x9c, 0xa4, 0xac, /* 5-9 */
109 0xb4, 0xbc, 0xc4, 0xcc, 0xd4, /* 10-14 */
110 0xdc, 0xe4, 0xec, 0xf4, 0xfc, /* 15-19 */
111 0x104, 0x10c, 0x114, 0x11c, /* 20-23 */
114 struct interleave_pkg {
115 unsigned char start;
116 unsigned char end;
119 static const struct interleave_pkg sbridge_interleave_pkg[] = {
120 { 0, 2 },
121 { 3, 5 },
122 { 8, 10 },
123 { 11, 13 },
124 { 16, 18 },
125 { 19, 21 },
126 { 24, 26 },
127 { 27, 29 },
130 static const struct interleave_pkg ibridge_interleave_pkg[] = {
131 { 0, 3 },
132 { 4, 7 },
133 { 8, 11 },
134 { 12, 15 },
135 { 16, 19 },
136 { 20, 23 },
137 { 24, 27 },
138 { 28, 31 },
141 static inline int sad_pkg(const struct interleave_pkg *table, u32 reg,
142 int interleave)
144 return GET_BITFIELD(reg, table[interleave].start,
145 table[interleave].end);
148 /* Devices 12 Function 7 */
150 #define TOLM 0x80
151 #define TOHM 0x84
152 #define HASWELL_TOLM 0xd0
153 #define HASWELL_TOHM_0 0xd4
154 #define HASWELL_TOHM_1 0xd8
155 #define KNL_TOLM 0xd0
156 #define KNL_TOHM_0 0xd4
157 #define KNL_TOHM_1 0xd8
159 #define GET_TOLM(reg) ((GET_BITFIELD(reg, 0, 3) << 28) | 0x3ffffff)
160 #define GET_TOHM(reg) ((GET_BITFIELD(reg, 0, 20) << 25) | 0x3ffffff)
162 /* Device 13 Function 6 */
164 #define SAD_TARGET 0xf0
166 #define SOURCE_ID(reg) GET_BITFIELD(reg, 9, 11)
168 #define SOURCE_ID_KNL(reg) GET_BITFIELD(reg, 12, 14)
170 #define SAD_CONTROL 0xf4
172 /* Device 14 function 0 */
174 static const u32 tad_dram_rule[] = {
175 0x40, 0x44, 0x48, 0x4c,
176 0x50, 0x54, 0x58, 0x5c,
177 0x60, 0x64, 0x68, 0x6c,
179 #define MAX_TAD ARRAY_SIZE(tad_dram_rule)
181 #define TAD_LIMIT(reg) ((GET_BITFIELD(reg, 12, 31) << 26) | 0x3ffffff)
182 #define TAD_SOCK(reg) GET_BITFIELD(reg, 10, 11)
183 #define TAD_CH(reg) GET_BITFIELD(reg, 8, 9)
184 #define TAD_TGT3(reg) GET_BITFIELD(reg, 6, 7)
185 #define TAD_TGT2(reg) GET_BITFIELD(reg, 4, 5)
186 #define TAD_TGT1(reg) GET_BITFIELD(reg, 2, 3)
187 #define TAD_TGT0(reg) GET_BITFIELD(reg, 0, 1)
189 /* Device 15, function 0 */
191 #define MCMTR 0x7c
192 #define KNL_MCMTR 0x624
194 #define IS_ECC_ENABLED(mcmtr) GET_BITFIELD(mcmtr, 2, 2)
195 #define IS_LOCKSTEP_ENABLED(mcmtr) GET_BITFIELD(mcmtr, 1, 1)
196 #define IS_CLOSE_PG(mcmtr) GET_BITFIELD(mcmtr, 0, 0)
198 /* Device 15, function 1 */
200 #define RASENABLES 0xac
201 #define IS_MIRROR_ENABLED(reg) GET_BITFIELD(reg, 0, 0)
203 /* Device 15, functions 2-5 */
205 static const int mtr_regs[] = {
206 0x80, 0x84, 0x88,
209 static const int knl_mtr_reg = 0xb60;
211 #define RANK_DISABLE(mtr) GET_BITFIELD(mtr, 16, 19)
212 #define IS_DIMM_PRESENT(mtr) GET_BITFIELD(mtr, 14, 14)
213 #define RANK_CNT_BITS(mtr) GET_BITFIELD(mtr, 12, 13)
214 #define RANK_WIDTH_BITS(mtr) GET_BITFIELD(mtr, 2, 4)
215 #define COL_WIDTH_BITS(mtr) GET_BITFIELD(mtr, 0, 1)
217 static const u32 tad_ch_nilv_offset[] = {
218 0x90, 0x94, 0x98, 0x9c,
219 0xa0, 0xa4, 0xa8, 0xac,
220 0xb0, 0xb4, 0xb8, 0xbc,
222 #define CHN_IDX_OFFSET(reg) GET_BITFIELD(reg, 28, 29)
223 #define TAD_OFFSET(reg) (GET_BITFIELD(reg, 6, 25) << 26)
225 static const u32 rir_way_limit[] = {
226 0x108, 0x10c, 0x110, 0x114, 0x118,
228 #define MAX_RIR_RANGES ARRAY_SIZE(rir_way_limit)
230 #define IS_RIR_VALID(reg) GET_BITFIELD(reg, 31, 31)
231 #define RIR_WAY(reg) GET_BITFIELD(reg, 28, 29)
233 #define MAX_RIR_WAY 8
235 static const u32 rir_offset[MAX_RIR_RANGES][MAX_RIR_WAY] = {
236 { 0x120, 0x124, 0x128, 0x12c, 0x130, 0x134, 0x138, 0x13c },
237 { 0x140, 0x144, 0x148, 0x14c, 0x150, 0x154, 0x158, 0x15c },
238 { 0x160, 0x164, 0x168, 0x16c, 0x170, 0x174, 0x178, 0x17c },
239 { 0x180, 0x184, 0x188, 0x18c, 0x190, 0x194, 0x198, 0x19c },
240 { 0x1a0, 0x1a4, 0x1a8, 0x1ac, 0x1b0, 0x1b4, 0x1b8, 0x1bc },
243 #define RIR_RNK_TGT(type, reg) (((type) == BROADWELL) ? \
244 GET_BITFIELD(reg, 20, 23) : GET_BITFIELD(reg, 16, 19))
246 #define RIR_OFFSET(type, reg) (((type) == HASWELL || (type) == BROADWELL) ? \
247 GET_BITFIELD(reg, 2, 15) : GET_BITFIELD(reg, 2, 14))
249 /* Device 16, functions 2-7 */
252 * FIXME: Implement the error count reads directly
255 static const u32 correrrcnt[] = {
256 0x104, 0x108, 0x10c, 0x110,
259 #define RANK_ODD_OV(reg) GET_BITFIELD(reg, 31, 31)
260 #define RANK_ODD_ERR_CNT(reg) GET_BITFIELD(reg, 16, 30)
261 #define RANK_EVEN_OV(reg) GET_BITFIELD(reg, 15, 15)
262 #define RANK_EVEN_ERR_CNT(reg) GET_BITFIELD(reg, 0, 14)
264 static const u32 correrrthrsld[] = {
265 0x11c, 0x120, 0x124, 0x128,
268 #define RANK_ODD_ERR_THRSLD(reg) GET_BITFIELD(reg, 16, 30)
269 #define RANK_EVEN_ERR_THRSLD(reg) GET_BITFIELD(reg, 0, 14)
272 /* Device 17, function 0 */
274 #define SB_RANK_CFG_A 0x0328
276 #define IB_RANK_CFG_A 0x0320
279 * sbridge structs
282 #define NUM_CHANNELS 8 /* 2MC per socket, four chan per MC */
283 #define MAX_DIMMS 3 /* Max DIMMS per channel */
284 #define KNL_MAX_CHAS 38 /* KNL max num. of Cache Home Agents */
285 #define KNL_MAX_CHANNELS 6 /* KNL max num. of PCI channels */
286 #define KNL_MAX_EDCS 8 /* Embedded DRAM controllers */
287 #define CHANNEL_UNSPECIFIED 0xf /* Intel IA32 SDM 15-14 */
289 enum type {
290 SANDY_BRIDGE,
291 IVY_BRIDGE,
292 HASWELL,
293 BROADWELL,
294 KNIGHTS_LANDING,
297 struct sbridge_pvt;
298 struct sbridge_info {
299 enum type type;
300 u32 mcmtr;
301 u32 rankcfgr;
302 u64 (*get_tolm)(struct sbridge_pvt *pvt);
303 u64 (*get_tohm)(struct sbridge_pvt *pvt);
304 u64 (*rir_limit)(u32 reg);
305 u64 (*sad_limit)(u32 reg);
306 u32 (*interleave_mode)(u32 reg);
307 char* (*show_interleave_mode)(u32 reg);
308 u32 (*dram_attr)(u32 reg);
309 const u32 *dram_rule;
310 const u32 *interleave_list;
311 const struct interleave_pkg *interleave_pkg;
312 u8 max_sad;
313 u8 max_interleave;
314 u8 (*get_node_id)(struct sbridge_pvt *pvt);
315 enum mem_type (*get_memory_type)(struct sbridge_pvt *pvt);
316 enum dev_type (*get_width)(struct sbridge_pvt *pvt, u32 mtr);
317 struct pci_dev *pci_vtd;
320 struct sbridge_channel {
321 u32 ranks;
322 u32 dimms;
325 struct pci_id_descr {
326 int dev_id;
327 int optional;
330 struct pci_id_table {
331 const struct pci_id_descr *descr;
332 int n_devs;
333 enum type type;
336 struct sbridge_dev {
337 struct list_head list;
338 u8 bus, mc;
339 u8 node_id, source_id;
340 struct pci_dev **pdev;
341 int n_devs;
342 struct mem_ctl_info *mci;
345 struct knl_pvt {
346 struct pci_dev *pci_cha[KNL_MAX_CHAS];
347 struct pci_dev *pci_channel[KNL_MAX_CHANNELS];
348 struct pci_dev *pci_mc0;
349 struct pci_dev *pci_mc1;
350 struct pci_dev *pci_mc0_misc;
351 struct pci_dev *pci_mc1_misc;
352 struct pci_dev *pci_mc_info; /* tolm, tohm */
355 struct sbridge_pvt {
356 struct pci_dev *pci_ta, *pci_ddrio, *pci_ras;
357 struct pci_dev *pci_sad0, *pci_sad1;
358 struct pci_dev *pci_ha0, *pci_ha1;
359 struct pci_dev *pci_br0, *pci_br1;
360 struct pci_dev *pci_ha1_ta;
361 struct pci_dev *pci_tad[NUM_CHANNELS];
363 struct sbridge_dev *sbridge_dev;
365 struct sbridge_info info;
366 struct sbridge_channel channel[NUM_CHANNELS];
368 /* Memory type detection */
369 bool is_mirrored, is_lockstep, is_close_pg;
370 bool is_chan_hash;
372 /* Memory description */
373 u64 tolm, tohm;
374 struct knl_pvt knl;
377 #define PCI_DESCR(device_id, opt) \
378 .dev_id = (device_id), \
379 .optional = opt
381 static const struct pci_id_descr pci_dev_descr_sbridge[] = {
382 /* Processor Home Agent */
383 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_HA0, 0) },
385 /* Memory controller */
386 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TA, 0) },
387 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_RAS, 0) },
388 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD0, 0) },
389 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD1, 0) },
390 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD2, 0) },
391 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD3, 0) },
392 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_DDRIO, 1) },
394 /* System Address Decoder */
395 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_SAD0, 0) },
396 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_SAD1, 0) },
398 /* Broadcast Registers */
399 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_BR, 0) },
402 #define PCI_ID_TABLE_ENTRY(A, T) { \
403 .descr = A, \
404 .n_devs = ARRAY_SIZE(A), \
405 .type = T \
408 static const struct pci_id_table pci_dev_descr_sbridge_table[] = {
409 PCI_ID_TABLE_ENTRY(pci_dev_descr_sbridge, SANDY_BRIDGE),
410 {0,} /* 0 terminated list. */
413 /* This changes depending if 1HA or 2HA:
414 * 1HA:
415 * 0x0eb8 (17.0) is DDRIO0
416 * 2HA:
417 * 0x0ebc (17.4) is DDRIO0
419 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_1HA_DDRIO0 0x0eb8
420 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_2HA_DDRIO0 0x0ebc
422 /* pci ids */
423 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0 0x0ea0
424 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TA 0x0ea8
425 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_RAS 0x0e71
426 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD0 0x0eaa
427 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD1 0x0eab
428 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD2 0x0eac
429 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD3 0x0ead
430 #define PCI_DEVICE_ID_INTEL_IBRIDGE_SAD 0x0ec8
431 #define PCI_DEVICE_ID_INTEL_IBRIDGE_BR0 0x0ec9
432 #define PCI_DEVICE_ID_INTEL_IBRIDGE_BR1 0x0eca
433 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1 0x0e60
434 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TA 0x0e68
435 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_RAS 0x0e79
436 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD0 0x0e6a
437 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD1 0x0e6b
438 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD2 0x0e6c
439 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD3 0x0e6d
441 static const struct pci_id_descr pci_dev_descr_ibridge[] = {
442 /* Processor Home Agent */
443 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0, 0) },
445 /* Memory controller */
446 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TA, 0) },
447 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_RAS, 0) },
448 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD0, 0) },
449 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD1, 0) },
450 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD2, 0) },
451 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD3, 0) },
453 /* System Address Decoder */
454 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_SAD, 0) },
456 /* Broadcast Registers */
457 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_BR0, 1) },
458 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_BR1, 0) },
460 /* Optional, mode 2HA */
461 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1, 1) },
462 #if 0
463 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TA, 1) },
464 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_RAS, 1) },
465 #endif
466 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD0, 1) },
467 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD1, 1) },
468 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD2, 1) },
469 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD3, 1) },
471 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_1HA_DDRIO0, 1) },
472 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_2HA_DDRIO0, 1) },
475 static const struct pci_id_table pci_dev_descr_ibridge_table[] = {
476 PCI_ID_TABLE_ENTRY(pci_dev_descr_ibridge, IVY_BRIDGE),
477 {0,} /* 0 terminated list. */
480 /* Haswell support */
481 /* EN processor:
482 * - 1 IMC
483 * - 3 DDR3 channels, 2 DPC per channel
484 * EP processor:
485 * - 1 or 2 IMC
486 * - 4 DDR4 channels, 3 DPC per channel
487 * EP 4S processor:
488 * - 2 IMC
489 * - 4 DDR4 channels, 3 DPC per channel
490 * EX processor:
491 * - 2 IMC
492 * - each IMC interfaces with a SMI 2 channel
493 * - each SMI channel interfaces with a scalable memory buffer
494 * - each scalable memory buffer supports 4 DDR3/DDR4 channels, 3 DPC
496 #define HASWELL_DDRCRCLKCONTROLS 0xa10 /* Ditto on Broadwell */
497 #define HASWELL_HASYSDEFEATURE2 0x84
498 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_VTD_MISC 0x2f28
499 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0 0x2fa0
500 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1 0x2f60
501 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TA 0x2fa8
502 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_THERMAL 0x2f71
503 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TA 0x2f68
504 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_THERMAL 0x2f79
505 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_CBO_SAD0 0x2ffc
506 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_CBO_SAD1 0x2ffd
507 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD0 0x2faa
508 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD1 0x2fab
509 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD2 0x2fac
510 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD3 0x2fad
511 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD0 0x2f6a
512 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD1 0x2f6b
513 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD2 0x2f6c
514 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD3 0x2f6d
515 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO0 0x2fbd
516 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO1 0x2fbf
517 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO2 0x2fb9
518 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO3 0x2fbb
519 static const struct pci_id_descr pci_dev_descr_haswell[] = {
520 /* first item must be the HA */
521 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0, 0) },
523 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_CBO_SAD0, 0) },
524 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_CBO_SAD1, 0) },
526 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1, 1) },
528 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TA, 0) },
529 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_THERMAL, 0) },
530 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD0, 0) },
531 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD1, 0) },
532 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD2, 1) },
533 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD3, 1) },
535 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO0, 1) },
536 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO1, 1) },
537 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO2, 1) },
538 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO3, 1) },
540 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TA, 1) },
541 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_THERMAL, 1) },
542 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD0, 1) },
543 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD1, 1) },
544 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD2, 1) },
545 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD3, 1) },
548 static const struct pci_id_table pci_dev_descr_haswell_table[] = {
549 PCI_ID_TABLE_ENTRY(pci_dev_descr_haswell, HASWELL),
550 {0,} /* 0 terminated list. */
553 /* Knight's Landing Support */
555 * KNL's memory channels are swizzled between memory controllers.
556 * MC0 is mapped to CH3,4,5 and MC1 is mapped to CH0,1,2
558 #define knl_channel_remap(mc, chan) ((mc) ? (chan) : (chan) + 3)
560 /* Memory controller, TAD tables, error injection - 2-8-0, 2-9-0 (2 of these) */
561 #define PCI_DEVICE_ID_INTEL_KNL_IMC_MC 0x7840
562 /* DRAM channel stuff; bank addrs, dimmmtr, etc.. 2-8-2 - 2-9-4 (6 of these) */
563 #define PCI_DEVICE_ID_INTEL_KNL_IMC_CHANNEL 0x7843
564 /* kdrwdbu TAD limits/offsets, MCMTR - 2-10-1, 2-11-1 (2 of these) */
565 #define PCI_DEVICE_ID_INTEL_KNL_IMC_TA 0x7844
566 /* CHA broadcast registers, dram rules - 1-29-0 (1 of these) */
567 #define PCI_DEVICE_ID_INTEL_KNL_IMC_SAD0 0x782a
568 /* SAD target - 1-29-1 (1 of these) */
569 #define PCI_DEVICE_ID_INTEL_KNL_IMC_SAD1 0x782b
570 /* Caching / Home Agent */
571 #define PCI_DEVICE_ID_INTEL_KNL_IMC_CHA 0x782c
572 /* Device with TOLM and TOHM, 0-5-0 (1 of these) */
573 #define PCI_DEVICE_ID_INTEL_KNL_IMC_TOLHM 0x7810
576 * KNL differs from SB, IB, and Haswell in that it has multiple
577 * instances of the same device with the same device ID, so we handle that
578 * by creating as many copies in the table as we expect to find.
579 * (Like device ID must be grouped together.)
582 static const struct pci_id_descr pci_dev_descr_knl[] = {
583 [0] = { PCI_DESCR(PCI_DEVICE_ID_INTEL_KNL_IMC_SAD0, 0) },
584 [1] = { PCI_DESCR(PCI_DEVICE_ID_INTEL_KNL_IMC_SAD1, 0) },
585 [2 ... 3] = { PCI_DESCR(PCI_DEVICE_ID_INTEL_KNL_IMC_MC, 0)},
586 [4 ... 41] = { PCI_DESCR(PCI_DEVICE_ID_INTEL_KNL_IMC_CHA, 0) },
587 [42 ... 47] = { PCI_DESCR(PCI_DEVICE_ID_INTEL_KNL_IMC_CHANNEL, 0) },
588 [48] = { PCI_DESCR(PCI_DEVICE_ID_INTEL_KNL_IMC_TA, 0) },
589 [49] = { PCI_DESCR(PCI_DEVICE_ID_INTEL_KNL_IMC_TOLHM, 0) },
592 static const struct pci_id_table pci_dev_descr_knl_table[] = {
593 PCI_ID_TABLE_ENTRY(pci_dev_descr_knl, KNIGHTS_LANDING),
594 {0,}
598 * Broadwell support
600 * DE processor:
601 * - 1 IMC
602 * - 2 DDR3 channels, 2 DPC per channel
603 * EP processor:
604 * - 1 or 2 IMC
605 * - 4 DDR4 channels, 3 DPC per channel
606 * EP 4S processor:
607 * - 2 IMC
608 * - 4 DDR4 channels, 3 DPC per channel
609 * EX processor:
610 * - 2 IMC
611 * - each IMC interfaces with a SMI 2 channel
612 * - each SMI channel interfaces with a scalable memory buffer
613 * - each scalable memory buffer supports 4 DDR3/DDR4 channels, 3 DPC
615 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_VTD_MISC 0x6f28
616 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0 0x6fa0
617 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1 0x6f60
618 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TA 0x6fa8
619 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_THERMAL 0x6f71
620 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TA 0x6f68
621 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_THERMAL 0x6f79
622 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_CBO_SAD0 0x6ffc
623 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_CBO_SAD1 0x6ffd
624 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD0 0x6faa
625 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD1 0x6fab
626 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD2 0x6fac
627 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD3 0x6fad
628 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD0 0x6f6a
629 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD1 0x6f6b
630 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD2 0x6f6c
631 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD3 0x6f6d
632 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_DDRIO0 0x6faf
634 static const struct pci_id_descr pci_dev_descr_broadwell[] = {
635 /* first item must be the HA */
636 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0, 0) },
638 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_CBO_SAD0, 0) },
639 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_CBO_SAD1, 0) },
641 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1, 1) },
643 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TA, 0) },
644 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_THERMAL, 0) },
645 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD0, 0) },
646 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD1, 0) },
647 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD2, 1) },
648 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD3, 1) },
650 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_DDRIO0, 1) },
652 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TA, 1) },
653 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_THERMAL, 1) },
654 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD0, 1) },
655 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD1, 1) },
656 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD2, 1) },
657 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD3, 1) },
660 static const struct pci_id_table pci_dev_descr_broadwell_table[] = {
661 PCI_ID_TABLE_ENTRY(pci_dev_descr_broadwell, BROADWELL),
662 {0,} /* 0 terminated list. */
666 /****************************************************************************
667 Ancillary status routines
668 ****************************************************************************/
670 static inline int numrank(enum type type, u32 mtr)
672 int ranks = (1 << RANK_CNT_BITS(mtr));
673 int max = 4;
675 if (type == HASWELL || type == BROADWELL || type == KNIGHTS_LANDING)
676 max = 8;
678 if (ranks > max) {
679 edac_dbg(0, "Invalid number of ranks: %d (max = %i) raw value = %x (%04x)\n",
680 ranks, max, (unsigned int)RANK_CNT_BITS(mtr), mtr);
681 return -EINVAL;
684 return ranks;
687 static inline int numrow(u32 mtr)
689 int rows = (RANK_WIDTH_BITS(mtr) + 12);
691 if (rows < 13 || rows > 18) {
692 edac_dbg(0, "Invalid number of rows: %d (should be between 14 and 17) raw value = %x (%04x)\n",
693 rows, (unsigned int)RANK_WIDTH_BITS(mtr), mtr);
694 return -EINVAL;
697 return 1 << rows;
700 static inline int numcol(u32 mtr)
702 int cols = (COL_WIDTH_BITS(mtr) + 10);
704 if (cols > 12) {
705 edac_dbg(0, "Invalid number of cols: %d (max = 4) raw value = %x (%04x)\n",
706 cols, (unsigned int)COL_WIDTH_BITS(mtr), mtr);
707 return -EINVAL;
710 return 1 << cols;
713 static struct sbridge_dev *get_sbridge_dev(u8 bus, int multi_bus)
715 struct sbridge_dev *sbridge_dev;
718 * If we have devices scattered across several busses that pertain
719 * to the same memory controller, we'll lump them all together.
721 if (multi_bus) {
722 return list_first_entry_or_null(&sbridge_edac_list,
723 struct sbridge_dev, list);
726 list_for_each_entry(sbridge_dev, &sbridge_edac_list, list) {
727 if (sbridge_dev->bus == bus)
728 return sbridge_dev;
731 return NULL;
734 static struct sbridge_dev *alloc_sbridge_dev(u8 bus,
735 const struct pci_id_table *table)
737 struct sbridge_dev *sbridge_dev;
739 sbridge_dev = kzalloc(sizeof(*sbridge_dev), GFP_KERNEL);
740 if (!sbridge_dev)
741 return NULL;
743 sbridge_dev->pdev = kzalloc(sizeof(*sbridge_dev->pdev) * table->n_devs,
744 GFP_KERNEL);
745 if (!sbridge_dev->pdev) {
746 kfree(sbridge_dev);
747 return NULL;
750 sbridge_dev->bus = bus;
751 sbridge_dev->n_devs = table->n_devs;
752 list_add_tail(&sbridge_dev->list, &sbridge_edac_list);
754 return sbridge_dev;
757 static void free_sbridge_dev(struct sbridge_dev *sbridge_dev)
759 list_del(&sbridge_dev->list);
760 kfree(sbridge_dev->pdev);
761 kfree(sbridge_dev);
764 static u64 sbridge_get_tolm(struct sbridge_pvt *pvt)
766 u32 reg;
768 /* Address range is 32:28 */
769 pci_read_config_dword(pvt->pci_sad1, TOLM, &reg);
770 return GET_TOLM(reg);
773 static u64 sbridge_get_tohm(struct sbridge_pvt *pvt)
775 u32 reg;
777 pci_read_config_dword(pvt->pci_sad1, TOHM, &reg);
778 return GET_TOHM(reg);
781 static u64 ibridge_get_tolm(struct sbridge_pvt *pvt)
783 u32 reg;
785 pci_read_config_dword(pvt->pci_br1, TOLM, &reg);
787 return GET_TOLM(reg);
790 static u64 ibridge_get_tohm(struct sbridge_pvt *pvt)
792 u32 reg;
794 pci_read_config_dword(pvt->pci_br1, TOHM, &reg);
796 return GET_TOHM(reg);
799 static u64 rir_limit(u32 reg)
801 return ((u64)GET_BITFIELD(reg, 1, 10) << 29) | 0x1fffffff;
804 static u64 sad_limit(u32 reg)
806 return (GET_BITFIELD(reg, 6, 25) << 26) | 0x3ffffff;
809 static u32 interleave_mode(u32 reg)
811 return GET_BITFIELD(reg, 1, 1);
814 char *show_interleave_mode(u32 reg)
816 return interleave_mode(reg) ? "8:6" : "[8:6]XOR[18:16]";
819 static u32 dram_attr(u32 reg)
821 return GET_BITFIELD(reg, 2, 3);
824 static u64 knl_sad_limit(u32 reg)
826 return (GET_BITFIELD(reg, 7, 26) << 26) | 0x3ffffff;
829 static u32 knl_interleave_mode(u32 reg)
831 return GET_BITFIELD(reg, 1, 2);
834 static char *knl_show_interleave_mode(u32 reg)
836 char *s;
838 switch (knl_interleave_mode(reg)) {
839 case 0:
840 s = "use address bits [8:6]";
841 break;
842 case 1:
843 s = "use address bits [10:8]";
844 break;
845 case 2:
846 s = "use address bits [14:12]";
847 break;
848 case 3:
849 s = "use address bits [32:30]";
850 break;
851 default:
852 WARN_ON(1);
853 break;
856 return s;
859 static u32 dram_attr_knl(u32 reg)
861 return GET_BITFIELD(reg, 3, 4);
865 static enum mem_type get_memory_type(struct sbridge_pvt *pvt)
867 u32 reg;
868 enum mem_type mtype;
870 if (pvt->pci_ddrio) {
871 pci_read_config_dword(pvt->pci_ddrio, pvt->info.rankcfgr,
872 &reg);
873 if (GET_BITFIELD(reg, 11, 11))
874 /* FIXME: Can also be LRDIMM */
875 mtype = MEM_RDDR3;
876 else
877 mtype = MEM_DDR3;
878 } else
879 mtype = MEM_UNKNOWN;
881 return mtype;
884 static enum mem_type haswell_get_memory_type(struct sbridge_pvt *pvt)
886 u32 reg;
887 bool registered = false;
888 enum mem_type mtype = MEM_UNKNOWN;
890 if (!pvt->pci_ddrio)
891 goto out;
893 pci_read_config_dword(pvt->pci_ddrio,
894 HASWELL_DDRCRCLKCONTROLS, &reg);
895 /* Is_Rdimm */
896 if (GET_BITFIELD(reg, 16, 16))
897 registered = true;
899 pci_read_config_dword(pvt->pci_ta, MCMTR, &reg);
900 if (GET_BITFIELD(reg, 14, 14)) {
901 if (registered)
902 mtype = MEM_RDDR4;
903 else
904 mtype = MEM_DDR4;
905 } else {
906 if (registered)
907 mtype = MEM_RDDR3;
908 else
909 mtype = MEM_DDR3;
912 out:
913 return mtype;
916 static enum dev_type knl_get_width(struct sbridge_pvt *pvt, u32 mtr)
918 /* for KNL value is fixed */
919 return DEV_X16;
922 static enum dev_type sbridge_get_width(struct sbridge_pvt *pvt, u32 mtr)
924 /* there's no way to figure out */
925 return DEV_UNKNOWN;
928 static enum dev_type __ibridge_get_width(u32 mtr)
930 enum dev_type type;
932 switch (mtr) {
933 case 3:
934 type = DEV_UNKNOWN;
935 break;
936 case 2:
937 type = DEV_X16;
938 break;
939 case 1:
940 type = DEV_X8;
941 break;
942 case 0:
943 type = DEV_X4;
944 break;
947 return type;
950 static enum dev_type ibridge_get_width(struct sbridge_pvt *pvt, u32 mtr)
953 * ddr3_width on the documentation but also valid for DDR4 on
954 * Haswell
956 return __ibridge_get_width(GET_BITFIELD(mtr, 7, 8));
959 static enum dev_type broadwell_get_width(struct sbridge_pvt *pvt, u32 mtr)
961 /* ddr3_width on the documentation but also valid for DDR4 */
962 return __ibridge_get_width(GET_BITFIELD(mtr, 8, 9));
965 static enum mem_type knl_get_memory_type(struct sbridge_pvt *pvt)
967 /* DDR4 RDIMMS and LRDIMMS are supported */
968 return MEM_RDDR4;
971 static u8 get_node_id(struct sbridge_pvt *pvt)
973 u32 reg;
974 pci_read_config_dword(pvt->pci_br0, SAD_CONTROL, &reg);
975 return GET_BITFIELD(reg, 0, 2);
978 static u8 haswell_get_node_id(struct sbridge_pvt *pvt)
980 u32 reg;
982 pci_read_config_dword(pvt->pci_sad1, SAD_CONTROL, &reg);
983 return GET_BITFIELD(reg, 0, 3);
986 static u8 knl_get_node_id(struct sbridge_pvt *pvt)
988 u32 reg;
990 pci_read_config_dword(pvt->pci_sad1, SAD_CONTROL, &reg);
991 return GET_BITFIELD(reg, 0, 2);
995 static u64 haswell_get_tolm(struct sbridge_pvt *pvt)
997 u32 reg;
999 pci_read_config_dword(pvt->info.pci_vtd, HASWELL_TOLM, &reg);
1000 return (GET_BITFIELD(reg, 26, 31) << 26) | 0x3ffffff;
1003 static u64 haswell_get_tohm(struct sbridge_pvt *pvt)
1005 u64 rc;
1006 u32 reg;
1008 pci_read_config_dword(pvt->info.pci_vtd, HASWELL_TOHM_0, &reg);
1009 rc = GET_BITFIELD(reg, 26, 31);
1010 pci_read_config_dword(pvt->info.pci_vtd, HASWELL_TOHM_1, &reg);
1011 rc = ((reg << 6) | rc) << 26;
1013 return rc | 0x1ffffff;
1016 static u64 knl_get_tolm(struct sbridge_pvt *pvt)
1018 u32 reg;
1020 pci_read_config_dword(pvt->knl.pci_mc_info, KNL_TOLM, &reg);
1021 return (GET_BITFIELD(reg, 26, 31) << 26) | 0x3ffffff;
1024 static u64 knl_get_tohm(struct sbridge_pvt *pvt)
1026 u64 rc;
1027 u32 reg_lo, reg_hi;
1029 pci_read_config_dword(pvt->knl.pci_mc_info, KNL_TOHM_0, &reg_lo);
1030 pci_read_config_dword(pvt->knl.pci_mc_info, KNL_TOHM_1, &reg_hi);
1031 rc = ((u64)reg_hi << 32) | reg_lo;
1032 return rc | 0x3ffffff;
1036 static u64 haswell_rir_limit(u32 reg)
1038 return (((u64)GET_BITFIELD(reg, 1, 11) + 1) << 29) - 1;
1041 static inline u8 sad_pkg_socket(u8 pkg)
1043 /* on Ivy Bridge, nodeID is SASS, where A is HA and S is node id */
1044 return ((pkg >> 3) << 2) | (pkg & 0x3);
1047 static inline u8 sad_pkg_ha(u8 pkg)
1049 return (pkg >> 2) & 0x1;
1052 static int haswell_chan_hash(int idx, u64 addr)
1054 int i;
1057 * XOR even bits from 12:26 to bit0 of idx,
1058 * odd bits from 13:27 to bit1
1060 for (i = 12; i < 28; i += 2)
1061 idx ^= (addr >> i) & 3;
1063 return idx;
1066 /****************************************************************************
1067 Memory check routines
1068 ****************************************************************************/
1069 static struct pci_dev *get_pdev_same_bus(u8 bus, u32 id)
1071 struct pci_dev *pdev = NULL;
1073 do {
1074 pdev = pci_get_device(PCI_VENDOR_ID_INTEL, id, pdev);
1075 if (pdev && pdev->bus->number == bus)
1076 break;
1077 } while (pdev);
1079 return pdev;
1083 * check_if_ecc_is_active() - Checks if ECC is active
1084 * @bus: Device bus
1085 * @type: Memory controller type
1086 * returns: 0 in case ECC is active, -ENODEV if it can't be determined or
1087 * disabled
1089 static int check_if_ecc_is_active(const u8 bus, enum type type)
1091 struct pci_dev *pdev = NULL;
1092 u32 mcmtr, id;
1094 switch (type) {
1095 case IVY_BRIDGE:
1096 id = PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TA;
1097 break;
1098 case HASWELL:
1099 id = PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TA;
1100 break;
1101 case SANDY_BRIDGE:
1102 id = PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TA;
1103 break;
1104 case BROADWELL:
1105 id = PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TA;
1106 break;
1107 case KNIGHTS_LANDING:
1109 * KNL doesn't group things by bus the same way
1110 * SB/IB/Haswell does.
1112 id = PCI_DEVICE_ID_INTEL_KNL_IMC_TA;
1113 break;
1114 default:
1115 return -ENODEV;
1118 if (type != KNIGHTS_LANDING)
1119 pdev = get_pdev_same_bus(bus, id);
1120 else
1121 pdev = pci_get_device(PCI_VENDOR_ID_INTEL, id, 0);
1123 if (!pdev) {
1124 sbridge_printk(KERN_ERR, "Couldn't find PCI device "
1125 "%04x:%04x! on bus %02d\n",
1126 PCI_VENDOR_ID_INTEL, id, bus);
1127 return -ENODEV;
1130 pci_read_config_dword(pdev,
1131 type == KNIGHTS_LANDING ? KNL_MCMTR : MCMTR, &mcmtr);
1132 if (!IS_ECC_ENABLED(mcmtr)) {
1133 sbridge_printk(KERN_ERR, "ECC is disabled. Aborting\n");
1134 return -ENODEV;
1136 return 0;
1139 /* Low bits of TAD limit, and some metadata. */
1140 static const u32 knl_tad_dram_limit_lo[] = {
1141 0x400, 0x500, 0x600, 0x700,
1142 0x800, 0x900, 0xa00, 0xb00,
1145 /* Low bits of TAD offset. */
1146 static const u32 knl_tad_dram_offset_lo[] = {
1147 0x404, 0x504, 0x604, 0x704,
1148 0x804, 0x904, 0xa04, 0xb04,
1151 /* High 16 bits of TAD limit and offset. */
1152 static const u32 knl_tad_dram_hi[] = {
1153 0x408, 0x508, 0x608, 0x708,
1154 0x808, 0x908, 0xa08, 0xb08,
1157 /* Number of ways a tad entry is interleaved. */
1158 static const u32 knl_tad_ways[] = {
1159 8, 6, 4, 3, 2, 1,
1163 * Retrieve the n'th Target Address Decode table entry
1164 * from the memory controller's TAD table.
1166 * @pvt: driver private data
1167 * @entry: which entry you want to retrieve
1168 * @mc: which memory controller (0 or 1)
1169 * @offset: output tad range offset
1170 * @limit: output address of first byte above tad range
1171 * @ways: output number of interleave ways
1173 * The offset value has curious semantics. It's a sort of running total
1174 * of the sizes of all the memory regions that aren't mapped in this
1175 * tad table.
1177 static int knl_get_tad(const struct sbridge_pvt *pvt,
1178 const int entry,
1179 const int mc,
1180 u64 *offset,
1181 u64 *limit,
1182 int *ways)
1184 u32 reg_limit_lo, reg_offset_lo, reg_hi;
1185 struct pci_dev *pci_mc;
1186 int way_id;
1188 switch (mc) {
1189 case 0:
1190 pci_mc = pvt->knl.pci_mc0;
1191 break;
1192 case 1:
1193 pci_mc = pvt->knl.pci_mc1;
1194 break;
1195 default:
1196 WARN_ON(1);
1197 return -EINVAL;
1200 pci_read_config_dword(pci_mc,
1201 knl_tad_dram_limit_lo[entry], &reg_limit_lo);
1202 pci_read_config_dword(pci_mc,
1203 knl_tad_dram_offset_lo[entry], &reg_offset_lo);
1204 pci_read_config_dword(pci_mc,
1205 knl_tad_dram_hi[entry], &reg_hi);
1207 /* Is this TAD entry enabled? */
1208 if (!GET_BITFIELD(reg_limit_lo, 0, 0))
1209 return -ENODEV;
1211 way_id = GET_BITFIELD(reg_limit_lo, 3, 5);
1213 if (way_id < ARRAY_SIZE(knl_tad_ways)) {
1214 *ways = knl_tad_ways[way_id];
1215 } else {
1216 *ways = 0;
1217 sbridge_printk(KERN_ERR,
1218 "Unexpected value %d in mc_tad_limit_lo wayness field\n",
1219 way_id);
1220 return -ENODEV;
1224 * The least significant 6 bits of base and limit are truncated.
1225 * For limit, we fill the missing bits with 1s.
1227 *offset = ((u64) GET_BITFIELD(reg_offset_lo, 6, 31) << 6) |
1228 ((u64) GET_BITFIELD(reg_hi, 0, 15) << 32);
1229 *limit = ((u64) GET_BITFIELD(reg_limit_lo, 6, 31) << 6) | 63 |
1230 ((u64) GET_BITFIELD(reg_hi, 16, 31) << 32);
1232 return 0;
1235 /* Determine which memory controller is responsible for a given channel. */
1236 static int knl_channel_mc(int channel)
1238 WARN_ON(channel < 0 || channel >= 6);
1240 return channel < 3 ? 1 : 0;
1244 * Get the Nth entry from EDC_ROUTE_TABLE register.
1245 * (This is the per-tile mapping of logical interleave targets to
1246 * physical EDC modules.)
1248 * entry 0: 0:2
1249 * 1: 3:5
1250 * 2: 6:8
1251 * 3: 9:11
1252 * 4: 12:14
1253 * 5: 15:17
1254 * 6: 18:20
1255 * 7: 21:23
1256 * reserved: 24:31
1258 static u32 knl_get_edc_route(int entry, u32 reg)
1260 WARN_ON(entry >= KNL_MAX_EDCS);
1261 return GET_BITFIELD(reg, entry*3, (entry*3)+2);
1265 * Get the Nth entry from MC_ROUTE_TABLE register.
1266 * (This is the per-tile mapping of logical interleave targets to
1267 * physical DRAM channels modules.)
1269 * entry 0: mc 0:2 channel 18:19
1270 * 1: mc 3:5 channel 20:21
1271 * 2: mc 6:8 channel 22:23
1272 * 3: mc 9:11 channel 24:25
1273 * 4: mc 12:14 channel 26:27
1274 * 5: mc 15:17 channel 28:29
1275 * reserved: 30:31
1277 * Though we have 3 bits to identify the MC, we should only see
1278 * the values 0 or 1.
1281 static u32 knl_get_mc_route(int entry, u32 reg)
1283 int mc, chan;
1285 WARN_ON(entry >= KNL_MAX_CHANNELS);
1287 mc = GET_BITFIELD(reg, entry*3, (entry*3)+2);
1288 chan = GET_BITFIELD(reg, (entry*2) + 18, (entry*2) + 18 + 1);
1290 return knl_channel_remap(mc, chan);
1294 * Render the EDC_ROUTE register in human-readable form.
1295 * Output string s should be at least KNL_MAX_EDCS*2 bytes.
1297 static void knl_show_edc_route(u32 reg, char *s)
1299 int i;
1301 for (i = 0; i < KNL_MAX_EDCS; i++) {
1302 s[i*2] = knl_get_edc_route(i, reg) + '0';
1303 s[i*2+1] = '-';
1306 s[KNL_MAX_EDCS*2 - 1] = '\0';
1310 * Render the MC_ROUTE register in human-readable form.
1311 * Output string s should be at least KNL_MAX_CHANNELS*2 bytes.
1313 static void knl_show_mc_route(u32 reg, char *s)
1315 int i;
1317 for (i = 0; i < KNL_MAX_CHANNELS; i++) {
1318 s[i*2] = knl_get_mc_route(i, reg) + '0';
1319 s[i*2+1] = '-';
1322 s[KNL_MAX_CHANNELS*2 - 1] = '\0';
1325 #define KNL_EDC_ROUTE 0xb8
1326 #define KNL_MC_ROUTE 0xb4
1328 /* Is this dram rule backed by regular DRAM in flat mode? */
1329 #define KNL_EDRAM(reg) GET_BITFIELD(reg, 29, 29)
1331 /* Is this dram rule cached? */
1332 #define KNL_CACHEABLE(reg) GET_BITFIELD(reg, 28, 28)
1334 /* Is this rule backed by edc ? */
1335 #define KNL_EDRAM_ONLY(reg) GET_BITFIELD(reg, 29, 29)
1337 /* Is this rule backed by DRAM, cacheable in EDRAM? */
1338 #define KNL_CACHEABLE(reg) GET_BITFIELD(reg, 28, 28)
1340 /* Is this rule mod3? */
1341 #define KNL_MOD3(reg) GET_BITFIELD(reg, 27, 27)
1344 * Figure out how big our RAM modules are.
1346 * The DIMMMTR register in KNL doesn't tell us the size of the DIMMs, so we
1347 * have to figure this out from the SAD rules, interleave lists, route tables,
1348 * and TAD rules.
1350 * SAD rules can have holes in them (e.g. the 3G-4G hole), so we have to
1351 * inspect the TAD rules to figure out how large the SAD regions really are.
1353 * When we know the real size of a SAD region and how many ways it's
1354 * interleaved, we know the individual contribution of each channel to
1355 * TAD is size/ways.
1357 * Finally, we have to check whether each channel participates in each SAD
1358 * region.
1360 * Fortunately, KNL only supports one DIMM per channel, so once we know how
1361 * much memory the channel uses, we know the DIMM is at least that large.
1362 * (The BIOS might possibly choose not to map all available memory, in which
1363 * case we will underreport the size of the DIMM.)
1365 * In theory, we could try to determine the EDC sizes as well, but that would
1366 * only work in flat mode, not in cache mode.
1368 * @mc_sizes: Output sizes of channels (must have space for KNL_MAX_CHANNELS
1369 * elements)
1371 static int knl_get_dimm_capacity(struct sbridge_pvt *pvt, u64 *mc_sizes)
1373 u64 sad_base, sad_size, sad_limit = 0;
1374 u64 tad_base, tad_size, tad_limit, tad_deadspace, tad_livespace;
1375 int sad_rule = 0;
1376 int tad_rule = 0;
1377 int intrlv_ways, tad_ways;
1378 u32 first_pkg, pkg;
1379 int i;
1380 u64 sad_actual_size[2]; /* sad size accounting for holes, per mc */
1381 u32 dram_rule, interleave_reg;
1382 u32 mc_route_reg[KNL_MAX_CHAS];
1383 u32 edc_route_reg[KNL_MAX_CHAS];
1384 int edram_only;
1385 char edc_route_string[KNL_MAX_EDCS*2];
1386 char mc_route_string[KNL_MAX_CHANNELS*2];
1387 int cur_reg_start;
1388 int mc;
1389 int channel;
1390 int way;
1391 int participants[KNL_MAX_CHANNELS];
1392 int participant_count = 0;
1394 for (i = 0; i < KNL_MAX_CHANNELS; i++)
1395 mc_sizes[i] = 0;
1397 /* Read the EDC route table in each CHA. */
1398 cur_reg_start = 0;
1399 for (i = 0; i < KNL_MAX_CHAS; i++) {
1400 pci_read_config_dword(pvt->knl.pci_cha[i],
1401 KNL_EDC_ROUTE, &edc_route_reg[i]);
1403 if (i > 0 && edc_route_reg[i] != edc_route_reg[i-1]) {
1404 knl_show_edc_route(edc_route_reg[i-1],
1405 edc_route_string);
1406 if (cur_reg_start == i-1)
1407 edac_dbg(0, "edc route table for CHA %d: %s\n",
1408 cur_reg_start, edc_route_string);
1409 else
1410 edac_dbg(0, "edc route table for CHA %d-%d: %s\n",
1411 cur_reg_start, i-1, edc_route_string);
1412 cur_reg_start = i;
1415 knl_show_edc_route(edc_route_reg[i-1], edc_route_string);
1416 if (cur_reg_start == i-1)
1417 edac_dbg(0, "edc route table for CHA %d: %s\n",
1418 cur_reg_start, edc_route_string);
1419 else
1420 edac_dbg(0, "edc route table for CHA %d-%d: %s\n",
1421 cur_reg_start, i-1, edc_route_string);
1423 /* Read the MC route table in each CHA. */
1424 cur_reg_start = 0;
1425 for (i = 0; i < KNL_MAX_CHAS; i++) {
1426 pci_read_config_dword(pvt->knl.pci_cha[i],
1427 KNL_MC_ROUTE, &mc_route_reg[i]);
1429 if (i > 0 && mc_route_reg[i] != mc_route_reg[i-1]) {
1430 knl_show_mc_route(mc_route_reg[i-1], mc_route_string);
1431 if (cur_reg_start == i-1)
1432 edac_dbg(0, "mc route table for CHA %d: %s\n",
1433 cur_reg_start, mc_route_string);
1434 else
1435 edac_dbg(0, "mc route table for CHA %d-%d: %s\n",
1436 cur_reg_start, i-1, mc_route_string);
1437 cur_reg_start = i;
1440 knl_show_mc_route(mc_route_reg[i-1], mc_route_string);
1441 if (cur_reg_start == i-1)
1442 edac_dbg(0, "mc route table for CHA %d: %s\n",
1443 cur_reg_start, mc_route_string);
1444 else
1445 edac_dbg(0, "mc route table for CHA %d-%d: %s\n",
1446 cur_reg_start, i-1, mc_route_string);
1448 /* Process DRAM rules */
1449 for (sad_rule = 0; sad_rule < pvt->info.max_sad; sad_rule++) {
1450 /* previous limit becomes the new base */
1451 sad_base = sad_limit;
1453 pci_read_config_dword(pvt->pci_sad0,
1454 pvt->info.dram_rule[sad_rule], &dram_rule);
1456 if (!DRAM_RULE_ENABLE(dram_rule))
1457 break;
1459 edram_only = KNL_EDRAM_ONLY(dram_rule);
1461 sad_limit = pvt->info.sad_limit(dram_rule)+1;
1462 sad_size = sad_limit - sad_base;
1464 pci_read_config_dword(pvt->pci_sad0,
1465 pvt->info.interleave_list[sad_rule], &interleave_reg);
1468 * Find out how many ways this dram rule is interleaved.
1469 * We stop when we see the first channel again.
1471 first_pkg = sad_pkg(pvt->info.interleave_pkg,
1472 interleave_reg, 0);
1473 for (intrlv_ways = 1; intrlv_ways < 8; intrlv_ways++) {
1474 pkg = sad_pkg(pvt->info.interleave_pkg,
1475 interleave_reg, intrlv_ways);
1477 if ((pkg & 0x8) == 0) {
1479 * 0 bit means memory is non-local,
1480 * which KNL doesn't support
1482 edac_dbg(0, "Unexpected interleave target %d\n",
1483 pkg);
1484 return -1;
1487 if (pkg == first_pkg)
1488 break;
1490 if (KNL_MOD3(dram_rule))
1491 intrlv_ways *= 3;
1493 edac_dbg(3, "dram rule %d (base 0x%llx, limit 0x%llx), %d way interleave%s\n",
1494 sad_rule,
1495 sad_base,
1496 sad_limit,
1497 intrlv_ways,
1498 edram_only ? ", EDRAM" : "");
1501 * Find out how big the SAD region really is by iterating
1502 * over TAD tables (SAD regions may contain holes).
1503 * Each memory controller might have a different TAD table, so
1504 * we have to look at both.
1506 * Livespace is the memory that's mapped in this TAD table,
1507 * deadspace is the holes (this could be the MMIO hole, or it
1508 * could be memory that's mapped by the other TAD table but
1509 * not this one).
1511 for (mc = 0; mc < 2; mc++) {
1512 sad_actual_size[mc] = 0;
1513 tad_livespace = 0;
1514 for (tad_rule = 0;
1515 tad_rule < ARRAY_SIZE(
1516 knl_tad_dram_limit_lo);
1517 tad_rule++) {
1518 if (knl_get_tad(pvt,
1519 tad_rule,
1521 &tad_deadspace,
1522 &tad_limit,
1523 &tad_ways))
1524 break;
1526 tad_size = (tad_limit+1) -
1527 (tad_livespace + tad_deadspace);
1528 tad_livespace += tad_size;
1529 tad_base = (tad_limit+1) - tad_size;
1531 if (tad_base < sad_base) {
1532 if (tad_limit > sad_base)
1533 edac_dbg(0, "TAD region overlaps lower SAD boundary -- TAD tables may be configured incorrectly.\n");
1534 } else if (tad_base < sad_limit) {
1535 if (tad_limit+1 > sad_limit) {
1536 edac_dbg(0, "TAD region overlaps upper SAD boundary -- TAD tables may be configured incorrectly.\n");
1537 } else {
1538 /* TAD region is completely inside SAD region */
1539 edac_dbg(3, "TAD region %d 0x%llx - 0x%llx (%lld bytes) table%d\n",
1540 tad_rule, tad_base,
1541 tad_limit, tad_size,
1542 mc);
1543 sad_actual_size[mc] += tad_size;
1546 tad_base = tad_limit+1;
1550 for (mc = 0; mc < 2; mc++) {
1551 edac_dbg(3, " total TAD DRAM footprint in table%d : 0x%llx (%lld bytes)\n",
1552 mc, sad_actual_size[mc], sad_actual_size[mc]);
1555 /* Ignore EDRAM rule */
1556 if (edram_only)
1557 continue;
1559 /* Figure out which channels participate in interleave. */
1560 for (channel = 0; channel < KNL_MAX_CHANNELS; channel++)
1561 participants[channel] = 0;
1563 /* For each channel, does at least one CHA have
1564 * this channel mapped to the given target?
1566 for (channel = 0; channel < KNL_MAX_CHANNELS; channel++) {
1567 for (way = 0; way < intrlv_ways; way++) {
1568 int target;
1569 int cha;
1571 if (KNL_MOD3(dram_rule))
1572 target = way;
1573 else
1574 target = 0x7 & sad_pkg(
1575 pvt->info.interleave_pkg, interleave_reg, way);
1577 for (cha = 0; cha < KNL_MAX_CHAS; cha++) {
1578 if (knl_get_mc_route(target,
1579 mc_route_reg[cha]) == channel
1580 && !participants[channel]) {
1581 participant_count++;
1582 participants[channel] = 1;
1583 break;
1589 if (participant_count != intrlv_ways)
1590 edac_dbg(0, "participant_count (%d) != interleave_ways (%d): DIMM size may be incorrect\n",
1591 participant_count, intrlv_ways);
1593 for (channel = 0; channel < KNL_MAX_CHANNELS; channel++) {
1594 mc = knl_channel_mc(channel);
1595 if (participants[channel]) {
1596 edac_dbg(4, "mc channel %d contributes %lld bytes via sad entry %d\n",
1597 channel,
1598 sad_actual_size[mc]/intrlv_ways,
1599 sad_rule);
1600 mc_sizes[channel] +=
1601 sad_actual_size[mc]/intrlv_ways;
1606 return 0;
1609 static int get_dimm_config(struct mem_ctl_info *mci)
1611 struct sbridge_pvt *pvt = mci->pvt_info;
1612 struct dimm_info *dimm;
1613 unsigned i, j, banks, ranks, rows, cols, npages;
1614 u64 size;
1615 u32 reg;
1616 enum edac_type mode;
1617 enum mem_type mtype;
1618 int channels = pvt->info.type == KNIGHTS_LANDING ?
1619 KNL_MAX_CHANNELS : NUM_CHANNELS;
1620 u64 knl_mc_sizes[KNL_MAX_CHANNELS];
1622 if (pvt->info.type == HASWELL || pvt->info.type == BROADWELL) {
1623 pci_read_config_dword(pvt->pci_ha0, HASWELL_HASYSDEFEATURE2, &reg);
1624 pvt->is_chan_hash = GET_BITFIELD(reg, 21, 21);
1626 if (pvt->info.type == HASWELL || pvt->info.type == BROADWELL ||
1627 pvt->info.type == KNIGHTS_LANDING)
1628 pci_read_config_dword(pvt->pci_sad1, SAD_TARGET, &reg);
1629 else
1630 pci_read_config_dword(pvt->pci_br0, SAD_TARGET, &reg);
1632 if (pvt->info.type == KNIGHTS_LANDING)
1633 pvt->sbridge_dev->source_id = SOURCE_ID_KNL(reg);
1634 else
1635 pvt->sbridge_dev->source_id = SOURCE_ID(reg);
1637 pvt->sbridge_dev->node_id = pvt->info.get_node_id(pvt);
1638 edac_dbg(0, "mc#%d: Node ID: %d, source ID: %d\n",
1639 pvt->sbridge_dev->mc,
1640 pvt->sbridge_dev->node_id,
1641 pvt->sbridge_dev->source_id);
1643 /* KNL doesn't support mirroring or lockstep,
1644 * and is always closed page
1646 if (pvt->info.type == KNIGHTS_LANDING) {
1647 mode = EDAC_S4ECD4ED;
1648 pvt->is_mirrored = false;
1650 if (knl_get_dimm_capacity(pvt, knl_mc_sizes) != 0)
1651 return -1;
1652 } else {
1653 pci_read_config_dword(pvt->pci_ras, RASENABLES, &reg);
1654 if (IS_MIRROR_ENABLED(reg)) {
1655 edac_dbg(0, "Memory mirror is enabled\n");
1656 pvt->is_mirrored = true;
1657 } else {
1658 edac_dbg(0, "Memory mirror is disabled\n");
1659 pvt->is_mirrored = false;
1662 pci_read_config_dword(pvt->pci_ta, MCMTR, &pvt->info.mcmtr);
1663 if (IS_LOCKSTEP_ENABLED(pvt->info.mcmtr)) {
1664 edac_dbg(0, "Lockstep is enabled\n");
1665 mode = EDAC_S8ECD8ED;
1666 pvt->is_lockstep = true;
1667 } else {
1668 edac_dbg(0, "Lockstep is disabled\n");
1669 mode = EDAC_S4ECD4ED;
1670 pvt->is_lockstep = false;
1672 if (IS_CLOSE_PG(pvt->info.mcmtr)) {
1673 edac_dbg(0, "address map is on closed page mode\n");
1674 pvt->is_close_pg = true;
1675 } else {
1676 edac_dbg(0, "address map is on open page mode\n");
1677 pvt->is_close_pg = false;
1681 mtype = pvt->info.get_memory_type(pvt);
1682 if (mtype == MEM_RDDR3 || mtype == MEM_RDDR4)
1683 edac_dbg(0, "Memory is registered\n");
1684 else if (mtype == MEM_UNKNOWN)
1685 edac_dbg(0, "Cannot determine memory type\n");
1686 else
1687 edac_dbg(0, "Memory is unregistered\n");
1689 if (mtype == MEM_DDR4 || mtype == MEM_RDDR4)
1690 banks = 16;
1691 else
1692 banks = 8;
1694 for (i = 0; i < channels; i++) {
1695 u32 mtr;
1697 int max_dimms_per_channel;
1699 if (pvt->info.type == KNIGHTS_LANDING) {
1700 max_dimms_per_channel = 1;
1701 if (!pvt->knl.pci_channel[i])
1702 continue;
1703 } else {
1704 max_dimms_per_channel = ARRAY_SIZE(mtr_regs);
1705 if (!pvt->pci_tad[i])
1706 continue;
1709 for (j = 0; j < max_dimms_per_channel; j++) {
1710 dimm = EDAC_DIMM_PTR(mci->layers, mci->dimms, mci->n_layers,
1711 i, j, 0);
1712 if (pvt->info.type == KNIGHTS_LANDING) {
1713 pci_read_config_dword(pvt->knl.pci_channel[i],
1714 knl_mtr_reg, &mtr);
1715 } else {
1716 pci_read_config_dword(pvt->pci_tad[i],
1717 mtr_regs[j], &mtr);
1719 edac_dbg(4, "Channel #%d MTR%d = %x\n", i, j, mtr);
1720 if (IS_DIMM_PRESENT(mtr)) {
1721 pvt->channel[i].dimms++;
1723 ranks = numrank(pvt->info.type, mtr);
1725 if (pvt->info.type == KNIGHTS_LANDING) {
1726 /* For DDR4, this is fixed. */
1727 cols = 1 << 10;
1728 rows = knl_mc_sizes[i] /
1729 ((u64) cols * ranks * banks * 8);
1730 } else {
1731 rows = numrow(mtr);
1732 cols = numcol(mtr);
1735 size = ((u64)rows * cols * banks * ranks) >> (20 - 3);
1736 npages = MiB_TO_PAGES(size);
1738 edac_dbg(0, "mc#%d: ha %d channel %d, dimm %d, %lld Mb (%d pages) bank: %d, rank: %d, row: %#x, col: %#x\n",
1739 pvt->sbridge_dev->mc, i/4, i%4, j,
1740 size, npages,
1741 banks, ranks, rows, cols);
1743 dimm->nr_pages = npages;
1744 dimm->grain = 32;
1745 dimm->dtype = pvt->info.get_width(pvt, mtr);
1746 dimm->mtype = mtype;
1747 dimm->edac_mode = mode;
1748 snprintf(dimm->label, sizeof(dimm->label),
1749 "CPU_SrcID#%u_Ha#%u_Chan#%u_DIMM#%u",
1750 pvt->sbridge_dev->source_id, i/4, i%4, j);
1755 return 0;
1758 static void get_memory_layout(const struct mem_ctl_info *mci)
1760 struct sbridge_pvt *pvt = mci->pvt_info;
1761 int i, j, k, n_sads, n_tads, sad_interl;
1762 u32 reg;
1763 u64 limit, prv = 0;
1764 u64 tmp_mb;
1765 u32 gb, mb;
1766 u32 rir_way;
1769 * Step 1) Get TOLM/TOHM ranges
1772 pvt->tolm = pvt->info.get_tolm(pvt);
1773 tmp_mb = (1 + pvt->tolm) >> 20;
1775 gb = div_u64_rem(tmp_mb, 1024, &mb);
1776 edac_dbg(0, "TOLM: %u.%03u GB (0x%016Lx)\n",
1777 gb, (mb*1000)/1024, (u64)pvt->tolm);
1779 /* Address range is already 45:25 */
1780 pvt->tohm = pvt->info.get_tohm(pvt);
1781 tmp_mb = (1 + pvt->tohm) >> 20;
1783 gb = div_u64_rem(tmp_mb, 1024, &mb);
1784 edac_dbg(0, "TOHM: %u.%03u GB (0x%016Lx)\n",
1785 gb, (mb*1000)/1024, (u64)pvt->tohm);
1788 * Step 2) Get SAD range and SAD Interleave list
1789 * TAD registers contain the interleave wayness. However, it
1790 * seems simpler to just discover it indirectly, with the
1791 * algorithm bellow.
1793 prv = 0;
1794 for (n_sads = 0; n_sads < pvt->info.max_sad; n_sads++) {
1795 /* SAD_LIMIT Address range is 45:26 */
1796 pci_read_config_dword(pvt->pci_sad0, pvt->info.dram_rule[n_sads],
1797 &reg);
1798 limit = pvt->info.sad_limit(reg);
1800 if (!DRAM_RULE_ENABLE(reg))
1801 continue;
1803 if (limit <= prv)
1804 break;
1806 tmp_mb = (limit + 1) >> 20;
1807 gb = div_u64_rem(tmp_mb, 1024, &mb);
1808 edac_dbg(0, "SAD#%d %s up to %u.%03u GB (0x%016Lx) Interleave: %s reg=0x%08x\n",
1809 n_sads,
1810 show_dram_attr(pvt->info.dram_attr(reg)),
1811 gb, (mb*1000)/1024,
1812 ((u64)tmp_mb) << 20L,
1813 pvt->info.show_interleave_mode(reg),
1814 reg);
1815 prv = limit;
1817 pci_read_config_dword(pvt->pci_sad0, pvt->info.interleave_list[n_sads],
1818 &reg);
1819 sad_interl = sad_pkg(pvt->info.interleave_pkg, reg, 0);
1820 for (j = 0; j < 8; j++) {
1821 u32 pkg = sad_pkg(pvt->info.interleave_pkg, reg, j);
1822 if (j > 0 && sad_interl == pkg)
1823 break;
1825 edac_dbg(0, "SAD#%d, interleave #%d: %d\n",
1826 n_sads, j, pkg);
1830 if (pvt->info.type == KNIGHTS_LANDING)
1831 return;
1834 * Step 3) Get TAD range
1836 prv = 0;
1837 for (n_tads = 0; n_tads < MAX_TAD; n_tads++) {
1838 pci_read_config_dword(pvt->pci_ha0, tad_dram_rule[n_tads],
1839 &reg);
1840 limit = TAD_LIMIT(reg);
1841 if (limit <= prv)
1842 break;
1843 tmp_mb = (limit + 1) >> 20;
1845 gb = div_u64_rem(tmp_mb, 1024, &mb);
1846 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",
1847 n_tads, gb, (mb*1000)/1024,
1848 ((u64)tmp_mb) << 20L,
1849 (u32)(1 << TAD_SOCK(reg)),
1850 (u32)TAD_CH(reg) + 1,
1851 (u32)TAD_TGT0(reg),
1852 (u32)TAD_TGT1(reg),
1853 (u32)TAD_TGT2(reg),
1854 (u32)TAD_TGT3(reg),
1855 reg);
1856 prv = limit;
1860 * Step 4) Get TAD offsets, per each channel
1862 for (i = 0; i < NUM_CHANNELS; i++) {
1863 if (!pvt->channel[i].dimms)
1864 continue;
1865 for (j = 0; j < n_tads; j++) {
1866 pci_read_config_dword(pvt->pci_tad[i],
1867 tad_ch_nilv_offset[j],
1868 &reg);
1869 tmp_mb = TAD_OFFSET(reg) >> 20;
1870 gb = div_u64_rem(tmp_mb, 1024, &mb);
1871 edac_dbg(0, "TAD CH#%d, offset #%d: %u.%03u GB (0x%016Lx), reg=0x%08x\n",
1872 i, j,
1873 gb, (mb*1000)/1024,
1874 ((u64)tmp_mb) << 20L,
1875 reg);
1880 * Step 6) Get RIR Wayness/Limit, per each channel
1882 for (i = 0; i < NUM_CHANNELS; i++) {
1883 if (!pvt->channel[i].dimms)
1884 continue;
1885 for (j = 0; j < MAX_RIR_RANGES; j++) {
1886 pci_read_config_dword(pvt->pci_tad[i],
1887 rir_way_limit[j],
1888 &reg);
1890 if (!IS_RIR_VALID(reg))
1891 continue;
1893 tmp_mb = pvt->info.rir_limit(reg) >> 20;
1894 rir_way = 1 << RIR_WAY(reg);
1895 gb = div_u64_rem(tmp_mb, 1024, &mb);
1896 edac_dbg(0, "CH#%d RIR#%d, limit: %u.%03u GB (0x%016Lx), way: %d, reg=0x%08x\n",
1897 i, j,
1898 gb, (mb*1000)/1024,
1899 ((u64)tmp_mb) << 20L,
1900 rir_way,
1901 reg);
1903 for (k = 0; k < rir_way; k++) {
1904 pci_read_config_dword(pvt->pci_tad[i],
1905 rir_offset[j][k],
1906 &reg);
1907 tmp_mb = RIR_OFFSET(pvt->info.type, reg) << 6;
1909 gb = div_u64_rem(tmp_mb, 1024, &mb);
1910 edac_dbg(0, "CH#%d RIR#%d INTL#%d, offset %u.%03u GB (0x%016Lx), tgt: %d, reg=0x%08x\n",
1911 i, j, k,
1912 gb, (mb*1000)/1024,
1913 ((u64)tmp_mb) << 20L,
1914 (u32)RIR_RNK_TGT(pvt->info.type, reg),
1915 reg);
1921 static struct mem_ctl_info *get_mci_for_node_id(u8 node_id)
1923 struct sbridge_dev *sbridge_dev;
1925 list_for_each_entry(sbridge_dev, &sbridge_edac_list, list) {
1926 if (sbridge_dev->node_id == node_id)
1927 return sbridge_dev->mci;
1929 return NULL;
1932 static int get_memory_error_data(struct mem_ctl_info *mci,
1933 u64 addr,
1934 u8 *socket, u8 *ha,
1935 long *channel_mask,
1936 u8 *rank,
1937 char **area_type, char *msg)
1939 struct mem_ctl_info *new_mci;
1940 struct sbridge_pvt *pvt = mci->pvt_info;
1941 struct pci_dev *pci_ha;
1942 int n_rir, n_sads, n_tads, sad_way, sck_xch;
1943 int sad_interl, idx, base_ch;
1944 int interleave_mode, shiftup = 0;
1945 unsigned sad_interleave[pvt->info.max_interleave];
1946 u32 reg, dram_rule;
1947 u8 ch_way, sck_way, pkg, sad_ha = 0, ch_add = 0;
1948 u32 tad_offset;
1949 u32 rir_way;
1950 u32 mb, gb;
1951 u64 ch_addr, offset, limit = 0, prv = 0;
1955 * Step 0) Check if the address is at special memory ranges
1956 * The check bellow is probably enough to fill all cases where
1957 * the error is not inside a memory, except for the legacy
1958 * range (e. g. VGA addresses). It is unlikely, however, that the
1959 * memory controller would generate an error on that range.
1961 if ((addr > (u64) pvt->tolm) && (addr < (1LL << 32))) {
1962 sprintf(msg, "Error at TOLM area, on addr 0x%08Lx", addr);
1963 return -EINVAL;
1965 if (addr >= (u64)pvt->tohm) {
1966 sprintf(msg, "Error at MMIOH area, on addr 0x%016Lx", addr);
1967 return -EINVAL;
1971 * Step 1) Get socket
1973 for (n_sads = 0; n_sads < pvt->info.max_sad; n_sads++) {
1974 pci_read_config_dword(pvt->pci_sad0, pvt->info.dram_rule[n_sads],
1975 &reg);
1977 if (!DRAM_RULE_ENABLE(reg))
1978 continue;
1980 limit = pvt->info.sad_limit(reg);
1981 if (limit <= prv) {
1982 sprintf(msg, "Can't discover the memory socket");
1983 return -EINVAL;
1985 if (addr <= limit)
1986 break;
1987 prv = limit;
1989 if (n_sads == pvt->info.max_sad) {
1990 sprintf(msg, "Can't discover the memory socket");
1991 return -EINVAL;
1993 dram_rule = reg;
1994 *area_type = show_dram_attr(pvt->info.dram_attr(dram_rule));
1995 interleave_mode = pvt->info.interleave_mode(dram_rule);
1997 pci_read_config_dword(pvt->pci_sad0, pvt->info.interleave_list[n_sads],
1998 &reg);
2000 if (pvt->info.type == SANDY_BRIDGE) {
2001 sad_interl = sad_pkg(pvt->info.interleave_pkg, reg, 0);
2002 for (sad_way = 0; sad_way < 8; sad_way++) {
2003 u32 pkg = sad_pkg(pvt->info.interleave_pkg, reg, sad_way);
2004 if (sad_way > 0 && sad_interl == pkg)
2005 break;
2006 sad_interleave[sad_way] = pkg;
2007 edac_dbg(0, "SAD interleave #%d: %d\n",
2008 sad_way, sad_interleave[sad_way]);
2010 edac_dbg(0, "mc#%d: Error detected on SAD#%d: address 0x%016Lx < 0x%016Lx, Interleave [%d:6]%s\n",
2011 pvt->sbridge_dev->mc,
2012 n_sads,
2013 addr,
2014 limit,
2015 sad_way + 7,
2016 !interleave_mode ? "" : "XOR[18:16]");
2017 if (interleave_mode)
2018 idx = ((addr >> 6) ^ (addr >> 16)) & 7;
2019 else
2020 idx = (addr >> 6) & 7;
2021 switch (sad_way) {
2022 case 1:
2023 idx = 0;
2024 break;
2025 case 2:
2026 idx = idx & 1;
2027 break;
2028 case 4:
2029 idx = idx & 3;
2030 break;
2031 case 8:
2032 break;
2033 default:
2034 sprintf(msg, "Can't discover socket interleave");
2035 return -EINVAL;
2037 *socket = sad_interleave[idx];
2038 edac_dbg(0, "SAD interleave index: %d (wayness %d) = CPU socket %d\n",
2039 idx, sad_way, *socket);
2040 } else if (pvt->info.type == HASWELL || pvt->info.type == BROADWELL) {
2041 int bits, a7mode = A7MODE(dram_rule);
2043 if (a7mode) {
2044 /* A7 mode swaps P9 with P6 */
2045 bits = GET_BITFIELD(addr, 7, 8) << 1;
2046 bits |= GET_BITFIELD(addr, 9, 9);
2047 } else
2048 bits = GET_BITFIELD(addr, 6, 8);
2050 if (interleave_mode == 0) {
2051 /* interleave mode will XOR {8,7,6} with {18,17,16} */
2052 idx = GET_BITFIELD(addr, 16, 18);
2053 idx ^= bits;
2054 } else
2055 idx = bits;
2057 pkg = sad_pkg(pvt->info.interleave_pkg, reg, idx);
2058 *socket = sad_pkg_socket(pkg);
2059 sad_ha = sad_pkg_ha(pkg);
2060 if (sad_ha)
2061 ch_add = 4;
2063 if (a7mode) {
2064 /* MCChanShiftUpEnable */
2065 pci_read_config_dword(pvt->pci_ha0,
2066 HASWELL_HASYSDEFEATURE2, &reg);
2067 shiftup = GET_BITFIELD(reg, 22, 22);
2070 edac_dbg(0, "SAD interleave package: %d = CPU socket %d, HA %i, shiftup: %i\n",
2071 idx, *socket, sad_ha, shiftup);
2072 } else {
2073 /* Ivy Bridge's SAD mode doesn't support XOR interleave mode */
2074 idx = (addr >> 6) & 7;
2075 pkg = sad_pkg(pvt->info.interleave_pkg, reg, idx);
2076 *socket = sad_pkg_socket(pkg);
2077 sad_ha = sad_pkg_ha(pkg);
2078 if (sad_ha)
2079 ch_add = 4;
2080 edac_dbg(0, "SAD interleave package: %d = CPU socket %d, HA %d\n",
2081 idx, *socket, sad_ha);
2084 *ha = sad_ha;
2087 * Move to the proper node structure, in order to access the
2088 * right PCI registers
2090 new_mci = get_mci_for_node_id(*socket);
2091 if (!new_mci) {
2092 sprintf(msg, "Struct for socket #%u wasn't initialized",
2093 *socket);
2094 return -EINVAL;
2096 mci = new_mci;
2097 pvt = mci->pvt_info;
2100 * Step 2) Get memory channel
2102 prv = 0;
2103 if (pvt->info.type == SANDY_BRIDGE)
2104 pci_ha = pvt->pci_ha0;
2105 else {
2106 if (sad_ha)
2107 pci_ha = pvt->pci_ha1;
2108 else
2109 pci_ha = pvt->pci_ha0;
2111 for (n_tads = 0; n_tads < MAX_TAD; n_tads++) {
2112 pci_read_config_dword(pci_ha, tad_dram_rule[n_tads], &reg);
2113 limit = TAD_LIMIT(reg);
2114 if (limit <= prv) {
2115 sprintf(msg, "Can't discover the memory channel");
2116 return -EINVAL;
2118 if (addr <= limit)
2119 break;
2120 prv = limit;
2122 if (n_tads == MAX_TAD) {
2123 sprintf(msg, "Can't discover the memory channel");
2124 return -EINVAL;
2127 ch_way = TAD_CH(reg) + 1;
2128 sck_way = TAD_SOCK(reg);
2130 if (ch_way == 3)
2131 idx = addr >> 6;
2132 else {
2133 idx = (addr >> (6 + sck_way + shiftup)) & 0x3;
2134 if (pvt->is_chan_hash)
2135 idx = haswell_chan_hash(idx, addr);
2137 idx = idx % ch_way;
2140 * FIXME: Shouldn't we use CHN_IDX_OFFSET() here, when ch_way == 3 ???
2142 switch (idx) {
2143 case 0:
2144 base_ch = TAD_TGT0(reg);
2145 break;
2146 case 1:
2147 base_ch = TAD_TGT1(reg);
2148 break;
2149 case 2:
2150 base_ch = TAD_TGT2(reg);
2151 break;
2152 case 3:
2153 base_ch = TAD_TGT3(reg);
2154 break;
2155 default:
2156 sprintf(msg, "Can't discover the TAD target");
2157 return -EINVAL;
2159 *channel_mask = 1 << base_ch;
2161 pci_read_config_dword(pvt->pci_tad[ch_add + base_ch],
2162 tad_ch_nilv_offset[n_tads],
2163 &tad_offset);
2165 if (pvt->is_mirrored) {
2166 *channel_mask |= 1 << ((base_ch + 2) % 4);
2167 switch(ch_way) {
2168 case 2:
2169 case 4:
2170 sck_xch = (1 << sck_way) * (ch_way >> 1);
2171 break;
2172 default:
2173 sprintf(msg, "Invalid mirror set. Can't decode addr");
2174 return -EINVAL;
2176 } else
2177 sck_xch = (1 << sck_way) * ch_way;
2179 if (pvt->is_lockstep)
2180 *channel_mask |= 1 << ((base_ch + 1) % 4);
2182 offset = TAD_OFFSET(tad_offset);
2184 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",
2185 n_tads,
2186 addr,
2187 limit,
2188 sck_way,
2189 ch_way,
2190 offset,
2191 idx,
2192 base_ch,
2193 *channel_mask);
2195 /* Calculate channel address */
2196 /* Remove the TAD offset */
2198 if (offset > addr) {
2199 sprintf(msg, "Can't calculate ch addr: TAD offset 0x%08Lx is too high for addr 0x%08Lx!",
2200 offset, addr);
2201 return -EINVAL;
2204 ch_addr = addr - offset;
2205 ch_addr >>= (6 + shiftup);
2206 ch_addr /= sck_xch;
2207 ch_addr <<= (6 + shiftup);
2208 ch_addr |= addr & ((1 << (6 + shiftup)) - 1);
2211 * Step 3) Decode rank
2213 for (n_rir = 0; n_rir < MAX_RIR_RANGES; n_rir++) {
2214 pci_read_config_dword(pvt->pci_tad[ch_add + base_ch],
2215 rir_way_limit[n_rir],
2216 &reg);
2218 if (!IS_RIR_VALID(reg))
2219 continue;
2221 limit = pvt->info.rir_limit(reg);
2222 gb = div_u64_rem(limit >> 20, 1024, &mb);
2223 edac_dbg(0, "RIR#%d, limit: %u.%03u GB (0x%016Lx), way: %d\n",
2224 n_rir,
2225 gb, (mb*1000)/1024,
2226 limit,
2227 1 << RIR_WAY(reg));
2228 if (ch_addr <= limit)
2229 break;
2231 if (n_rir == MAX_RIR_RANGES) {
2232 sprintf(msg, "Can't discover the memory rank for ch addr 0x%08Lx",
2233 ch_addr);
2234 return -EINVAL;
2236 rir_way = RIR_WAY(reg);
2238 if (pvt->is_close_pg)
2239 idx = (ch_addr >> 6);
2240 else
2241 idx = (ch_addr >> 13); /* FIXME: Datasheet says to shift by 15 */
2242 idx %= 1 << rir_way;
2244 pci_read_config_dword(pvt->pci_tad[ch_add + base_ch],
2245 rir_offset[n_rir][idx],
2246 &reg);
2247 *rank = RIR_RNK_TGT(pvt->info.type, reg);
2249 edac_dbg(0, "RIR#%d: channel address 0x%08Lx < 0x%08Lx, RIR interleave %d, index %d\n",
2250 n_rir,
2251 ch_addr,
2252 limit,
2253 rir_way,
2254 idx);
2256 return 0;
2259 /****************************************************************************
2260 Device initialization routines: put/get, init/exit
2261 ****************************************************************************/
2264 * sbridge_put_all_devices 'put' all the devices that we have
2265 * reserved via 'get'
2267 static void sbridge_put_devices(struct sbridge_dev *sbridge_dev)
2269 int i;
2271 edac_dbg(0, "\n");
2272 for (i = 0; i < sbridge_dev->n_devs; i++) {
2273 struct pci_dev *pdev = sbridge_dev->pdev[i];
2274 if (!pdev)
2275 continue;
2276 edac_dbg(0, "Removing dev %02x:%02x.%d\n",
2277 pdev->bus->number,
2278 PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn));
2279 pci_dev_put(pdev);
2283 static void sbridge_put_all_devices(void)
2285 struct sbridge_dev *sbridge_dev, *tmp;
2287 list_for_each_entry_safe(sbridge_dev, tmp, &sbridge_edac_list, list) {
2288 sbridge_put_devices(sbridge_dev);
2289 free_sbridge_dev(sbridge_dev);
2293 static int sbridge_get_onedevice(struct pci_dev **prev,
2294 u8 *num_mc,
2295 const struct pci_id_table *table,
2296 const unsigned devno,
2297 const int multi_bus)
2299 struct sbridge_dev *sbridge_dev;
2300 const struct pci_id_descr *dev_descr = &table->descr[devno];
2301 struct pci_dev *pdev = NULL;
2302 u8 bus = 0;
2304 sbridge_printk(KERN_DEBUG,
2305 "Seeking for: PCI ID %04x:%04x\n",
2306 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
2308 pdev = pci_get_device(PCI_VENDOR_ID_INTEL,
2309 dev_descr->dev_id, *prev);
2311 if (!pdev) {
2312 if (*prev) {
2313 *prev = pdev;
2314 return 0;
2317 if (dev_descr->optional)
2318 return 0;
2320 /* if the HA wasn't found */
2321 if (devno == 0)
2322 return -ENODEV;
2324 sbridge_printk(KERN_INFO,
2325 "Device not found: %04x:%04x\n",
2326 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
2328 /* End of list, leave */
2329 return -ENODEV;
2331 bus = pdev->bus->number;
2333 sbridge_dev = get_sbridge_dev(bus, multi_bus);
2334 if (!sbridge_dev) {
2335 sbridge_dev = alloc_sbridge_dev(bus, table);
2336 if (!sbridge_dev) {
2337 pci_dev_put(pdev);
2338 return -ENOMEM;
2340 (*num_mc)++;
2343 if (sbridge_dev->pdev[devno]) {
2344 sbridge_printk(KERN_ERR,
2345 "Duplicated device for %04x:%04x\n",
2346 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
2347 pci_dev_put(pdev);
2348 return -ENODEV;
2351 sbridge_dev->pdev[devno] = pdev;
2353 /* Be sure that the device is enabled */
2354 if (unlikely(pci_enable_device(pdev) < 0)) {
2355 sbridge_printk(KERN_ERR,
2356 "Couldn't enable %04x:%04x\n",
2357 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
2358 return -ENODEV;
2361 edac_dbg(0, "Detected %04x:%04x\n",
2362 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
2365 * As stated on drivers/pci/search.c, the reference count for
2366 * @from is always decremented if it is not %NULL. So, as we need
2367 * to get all devices up to null, we need to do a get for the device
2369 pci_dev_get(pdev);
2371 *prev = pdev;
2373 return 0;
2377 * sbridge_get_all_devices - Find and perform 'get' operation on the MCH's
2378 * devices we want to reference for this driver.
2379 * @num_mc: pointer to the memory controllers count, to be incremented in case
2380 * of success.
2381 * @table: model specific table
2383 * returns 0 in case of success or error code
2385 static int sbridge_get_all_devices(u8 *num_mc,
2386 const struct pci_id_table *table)
2388 int i, rc;
2389 struct pci_dev *pdev = NULL;
2390 int allow_dups = 0;
2391 int multi_bus = 0;
2393 if (table->type == KNIGHTS_LANDING)
2394 allow_dups = multi_bus = 1;
2395 while (table && table->descr) {
2396 for (i = 0; i < table->n_devs; i++) {
2397 if (!allow_dups || i == 0 ||
2398 table->descr[i].dev_id !=
2399 table->descr[i-1].dev_id) {
2400 pdev = NULL;
2402 do {
2403 rc = sbridge_get_onedevice(&pdev, num_mc,
2404 table, i, multi_bus);
2405 if (rc < 0) {
2406 if (i == 0) {
2407 i = table->n_devs;
2408 break;
2410 sbridge_put_all_devices();
2411 return -ENODEV;
2413 } while (pdev && !allow_dups);
2415 table++;
2418 return 0;
2421 static int sbridge_mci_bind_devs(struct mem_ctl_info *mci,
2422 struct sbridge_dev *sbridge_dev)
2424 struct sbridge_pvt *pvt = mci->pvt_info;
2425 struct pci_dev *pdev;
2426 u8 saw_chan_mask = 0;
2427 int i;
2429 for (i = 0; i < sbridge_dev->n_devs; i++) {
2430 pdev = sbridge_dev->pdev[i];
2431 if (!pdev)
2432 continue;
2434 switch (pdev->device) {
2435 case PCI_DEVICE_ID_INTEL_SBRIDGE_SAD0:
2436 pvt->pci_sad0 = pdev;
2437 break;
2438 case PCI_DEVICE_ID_INTEL_SBRIDGE_SAD1:
2439 pvt->pci_sad1 = pdev;
2440 break;
2441 case PCI_DEVICE_ID_INTEL_SBRIDGE_BR:
2442 pvt->pci_br0 = pdev;
2443 break;
2444 case PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_HA0:
2445 pvt->pci_ha0 = pdev;
2446 break;
2447 case PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TA:
2448 pvt->pci_ta = pdev;
2449 break;
2450 case PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_RAS:
2451 pvt->pci_ras = pdev;
2452 break;
2453 case PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD0:
2454 case PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD1:
2455 case PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD2:
2456 case PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD3:
2458 int id = pdev->device - PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD0;
2459 pvt->pci_tad[id] = pdev;
2460 saw_chan_mask |= 1 << id;
2462 break;
2463 case PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_DDRIO:
2464 pvt->pci_ddrio = pdev;
2465 break;
2466 default:
2467 goto error;
2470 edac_dbg(0, "Associated PCI %02x:%02x, bus %d with dev = %p\n",
2471 pdev->vendor, pdev->device,
2472 sbridge_dev->bus,
2473 pdev);
2476 /* Check if everything were registered */
2477 if (!pvt->pci_sad0 || !pvt->pci_sad1 || !pvt->pci_ha0 ||
2478 !pvt->pci_ras || !pvt->pci_ta)
2479 goto enodev;
2481 if (saw_chan_mask != 0x0f)
2482 goto enodev;
2483 return 0;
2485 enodev:
2486 sbridge_printk(KERN_ERR, "Some needed devices are missing\n");
2487 return -ENODEV;
2489 error:
2490 sbridge_printk(KERN_ERR, "Unexpected device %02x:%02x\n",
2491 PCI_VENDOR_ID_INTEL, pdev->device);
2492 return -EINVAL;
2495 static int ibridge_mci_bind_devs(struct mem_ctl_info *mci,
2496 struct sbridge_dev *sbridge_dev)
2498 struct sbridge_pvt *pvt = mci->pvt_info;
2499 struct pci_dev *pdev;
2500 u8 saw_chan_mask = 0;
2501 int i;
2503 for (i = 0; i < sbridge_dev->n_devs; i++) {
2504 pdev = sbridge_dev->pdev[i];
2505 if (!pdev)
2506 continue;
2508 switch (pdev->device) {
2509 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0:
2510 pvt->pci_ha0 = pdev;
2511 break;
2512 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TA:
2513 pvt->pci_ta = pdev;
2514 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_RAS:
2515 pvt->pci_ras = pdev;
2516 break;
2517 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD0:
2518 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD1:
2519 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD2:
2520 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD3:
2522 int id = pdev->device - PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD0;
2523 pvt->pci_tad[id] = pdev;
2524 saw_chan_mask |= 1 << id;
2526 break;
2527 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_2HA_DDRIO0:
2528 pvt->pci_ddrio = pdev;
2529 break;
2530 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_1HA_DDRIO0:
2531 pvt->pci_ddrio = pdev;
2532 break;
2533 case PCI_DEVICE_ID_INTEL_IBRIDGE_SAD:
2534 pvt->pci_sad0 = pdev;
2535 break;
2536 case PCI_DEVICE_ID_INTEL_IBRIDGE_BR0:
2537 pvt->pci_br0 = pdev;
2538 break;
2539 case PCI_DEVICE_ID_INTEL_IBRIDGE_BR1:
2540 pvt->pci_br1 = pdev;
2541 break;
2542 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1:
2543 pvt->pci_ha1 = pdev;
2544 break;
2545 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD0:
2546 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD1:
2547 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD2:
2548 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD3:
2550 int id = pdev->device - PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD0 + 4;
2551 pvt->pci_tad[id] = pdev;
2552 saw_chan_mask |= 1 << id;
2554 break;
2555 default:
2556 goto error;
2559 edac_dbg(0, "Associated PCI %02x.%02d.%d with dev = %p\n",
2560 sbridge_dev->bus,
2561 PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn),
2562 pdev);
2565 /* Check if everything were registered */
2566 if (!pvt->pci_sad0 || !pvt->pci_ha0 || !pvt->pci_br0 ||
2567 !pvt->pci_br1 || !pvt->pci_ras || !pvt->pci_ta)
2568 goto enodev;
2570 if (saw_chan_mask != 0x0f && /* -EN */
2571 saw_chan_mask != 0x33 && /* -EP */
2572 saw_chan_mask != 0xff) /* -EX */
2573 goto enodev;
2574 return 0;
2576 enodev:
2577 sbridge_printk(KERN_ERR, "Some needed devices are missing\n");
2578 return -ENODEV;
2580 error:
2581 sbridge_printk(KERN_ERR,
2582 "Unexpected device %02x:%02x\n", PCI_VENDOR_ID_INTEL,
2583 pdev->device);
2584 return -EINVAL;
2587 static int haswell_mci_bind_devs(struct mem_ctl_info *mci,
2588 struct sbridge_dev *sbridge_dev)
2590 struct sbridge_pvt *pvt = mci->pvt_info;
2591 struct pci_dev *pdev;
2592 u8 saw_chan_mask = 0;
2593 int i;
2595 /* there's only one device per system; not tied to any bus */
2596 if (pvt->info.pci_vtd == NULL)
2597 /* result will be checked later */
2598 pvt->info.pci_vtd = pci_get_device(PCI_VENDOR_ID_INTEL,
2599 PCI_DEVICE_ID_INTEL_HASWELL_IMC_VTD_MISC,
2600 NULL);
2602 for (i = 0; i < sbridge_dev->n_devs; i++) {
2603 pdev = sbridge_dev->pdev[i];
2604 if (!pdev)
2605 continue;
2607 switch (pdev->device) {
2608 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_CBO_SAD0:
2609 pvt->pci_sad0 = pdev;
2610 break;
2611 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_CBO_SAD1:
2612 pvt->pci_sad1 = pdev;
2613 break;
2614 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0:
2615 pvt->pci_ha0 = pdev;
2616 break;
2617 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TA:
2618 pvt->pci_ta = pdev;
2619 break;
2620 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_THERMAL:
2621 pvt->pci_ras = pdev;
2622 break;
2623 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD0:
2624 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD1:
2625 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD2:
2626 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD3:
2628 int id = pdev->device - PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD0;
2630 pvt->pci_tad[id] = pdev;
2631 saw_chan_mask |= 1 << id;
2633 break;
2634 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD0:
2635 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD1:
2636 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD2:
2637 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD3:
2639 int id = pdev->device - PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD0 + 4;
2641 pvt->pci_tad[id] = pdev;
2642 saw_chan_mask |= 1 << id;
2644 break;
2645 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO0:
2646 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO1:
2647 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO2:
2648 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO3:
2649 if (!pvt->pci_ddrio)
2650 pvt->pci_ddrio = pdev;
2651 break;
2652 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1:
2653 pvt->pci_ha1 = pdev;
2654 break;
2655 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TA:
2656 pvt->pci_ha1_ta = pdev;
2657 break;
2658 default:
2659 break;
2662 edac_dbg(0, "Associated PCI %02x.%02d.%d with dev = %p\n",
2663 sbridge_dev->bus,
2664 PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn),
2665 pdev);
2668 /* Check if everything were registered */
2669 if (!pvt->pci_sad0 || !pvt->pci_ha0 || !pvt->pci_sad1 ||
2670 !pvt->pci_ras || !pvt->pci_ta || !pvt->info.pci_vtd)
2671 goto enodev;
2673 if (saw_chan_mask != 0x0f && /* -EN */
2674 saw_chan_mask != 0x33 && /* -EP */
2675 saw_chan_mask != 0xff) /* -EX */
2676 goto enodev;
2677 return 0;
2679 enodev:
2680 sbridge_printk(KERN_ERR, "Some needed devices are missing\n");
2681 return -ENODEV;
2684 static int broadwell_mci_bind_devs(struct mem_ctl_info *mci,
2685 struct sbridge_dev *sbridge_dev)
2687 struct sbridge_pvt *pvt = mci->pvt_info;
2688 struct pci_dev *pdev;
2689 u8 saw_chan_mask = 0;
2690 int i;
2692 /* there's only one device per system; not tied to any bus */
2693 if (pvt->info.pci_vtd == NULL)
2694 /* result will be checked later */
2695 pvt->info.pci_vtd = pci_get_device(PCI_VENDOR_ID_INTEL,
2696 PCI_DEVICE_ID_INTEL_BROADWELL_IMC_VTD_MISC,
2697 NULL);
2699 for (i = 0; i < sbridge_dev->n_devs; i++) {
2700 pdev = sbridge_dev->pdev[i];
2701 if (!pdev)
2702 continue;
2704 switch (pdev->device) {
2705 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_CBO_SAD0:
2706 pvt->pci_sad0 = pdev;
2707 break;
2708 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_CBO_SAD1:
2709 pvt->pci_sad1 = pdev;
2710 break;
2711 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0:
2712 pvt->pci_ha0 = pdev;
2713 break;
2714 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TA:
2715 pvt->pci_ta = pdev;
2716 break;
2717 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_THERMAL:
2718 pvt->pci_ras = pdev;
2719 break;
2720 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD0:
2721 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD1:
2722 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD2:
2723 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD3:
2725 int id = pdev->device - PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD0;
2726 pvt->pci_tad[id] = pdev;
2727 saw_chan_mask |= 1 << id;
2729 break;
2730 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD0:
2731 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD1:
2732 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD2:
2733 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD3:
2735 int id = pdev->device - PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD0 + 4;
2736 pvt->pci_tad[id] = pdev;
2737 saw_chan_mask |= 1 << id;
2739 break;
2740 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_DDRIO0:
2741 pvt->pci_ddrio = pdev;
2742 break;
2743 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1:
2744 pvt->pci_ha1 = pdev;
2745 break;
2746 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TA:
2747 pvt->pci_ha1_ta = pdev;
2748 break;
2749 default:
2750 break;
2753 edac_dbg(0, "Associated PCI %02x.%02d.%d with dev = %p\n",
2754 sbridge_dev->bus,
2755 PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn),
2756 pdev);
2759 /* Check if everything were registered */
2760 if (!pvt->pci_sad0 || !pvt->pci_ha0 || !pvt->pci_sad1 ||
2761 !pvt->pci_ras || !pvt->pci_ta || !pvt->info.pci_vtd)
2762 goto enodev;
2764 if (saw_chan_mask != 0x0f && /* -EN */
2765 saw_chan_mask != 0x33 && /* -EP */
2766 saw_chan_mask != 0xff) /* -EX */
2767 goto enodev;
2768 return 0;
2770 enodev:
2771 sbridge_printk(KERN_ERR, "Some needed devices are missing\n");
2772 return -ENODEV;
2775 static int knl_mci_bind_devs(struct mem_ctl_info *mci,
2776 struct sbridge_dev *sbridge_dev)
2778 struct sbridge_pvt *pvt = mci->pvt_info;
2779 struct pci_dev *pdev;
2780 int dev, func;
2782 int i;
2783 int devidx;
2785 for (i = 0; i < sbridge_dev->n_devs; i++) {
2786 pdev = sbridge_dev->pdev[i];
2787 if (!pdev)
2788 continue;
2790 /* Extract PCI device and function. */
2791 dev = (pdev->devfn >> 3) & 0x1f;
2792 func = pdev->devfn & 0x7;
2794 switch (pdev->device) {
2795 case PCI_DEVICE_ID_INTEL_KNL_IMC_MC:
2796 if (dev == 8)
2797 pvt->knl.pci_mc0 = pdev;
2798 else if (dev == 9)
2799 pvt->knl.pci_mc1 = pdev;
2800 else {
2801 sbridge_printk(KERN_ERR,
2802 "Memory controller in unexpected place! (dev %d, fn %d)\n",
2803 dev, func);
2804 continue;
2806 break;
2808 case PCI_DEVICE_ID_INTEL_KNL_IMC_SAD0:
2809 pvt->pci_sad0 = pdev;
2810 break;
2812 case PCI_DEVICE_ID_INTEL_KNL_IMC_SAD1:
2813 pvt->pci_sad1 = pdev;
2814 break;
2816 case PCI_DEVICE_ID_INTEL_KNL_IMC_CHA:
2817 /* There are one of these per tile, and range from
2818 * 1.14.0 to 1.18.5.
2820 devidx = ((dev-14)*8)+func;
2822 if (devidx < 0 || devidx >= KNL_MAX_CHAS) {
2823 sbridge_printk(KERN_ERR,
2824 "Caching and Home Agent in unexpected place! (dev %d, fn %d)\n",
2825 dev, func);
2826 continue;
2829 WARN_ON(pvt->knl.pci_cha[devidx] != NULL);
2831 pvt->knl.pci_cha[devidx] = pdev;
2832 break;
2834 case PCI_DEVICE_ID_INTEL_KNL_IMC_CHANNEL:
2835 devidx = -1;
2838 * MC0 channels 0-2 are device 9 function 2-4,
2839 * MC1 channels 3-5 are device 8 function 2-4.
2842 if (dev == 9)
2843 devidx = func-2;
2844 else if (dev == 8)
2845 devidx = 3 + (func-2);
2847 if (devidx < 0 || devidx >= KNL_MAX_CHANNELS) {
2848 sbridge_printk(KERN_ERR,
2849 "DRAM Channel Registers in unexpected place! (dev %d, fn %d)\n",
2850 dev, func);
2851 continue;
2854 WARN_ON(pvt->knl.pci_channel[devidx] != NULL);
2855 pvt->knl.pci_channel[devidx] = pdev;
2856 break;
2858 case PCI_DEVICE_ID_INTEL_KNL_IMC_TOLHM:
2859 pvt->knl.pci_mc_info = pdev;
2860 break;
2862 case PCI_DEVICE_ID_INTEL_KNL_IMC_TA:
2863 pvt->pci_ta = pdev;
2864 break;
2866 default:
2867 sbridge_printk(KERN_ERR, "Unexpected device %d\n",
2868 pdev->device);
2869 break;
2873 if (!pvt->knl.pci_mc0 || !pvt->knl.pci_mc1 ||
2874 !pvt->pci_sad0 || !pvt->pci_sad1 ||
2875 !pvt->pci_ta) {
2876 goto enodev;
2879 for (i = 0; i < KNL_MAX_CHANNELS; i++) {
2880 if (!pvt->knl.pci_channel[i]) {
2881 sbridge_printk(KERN_ERR, "Missing channel %d\n", i);
2882 goto enodev;
2886 for (i = 0; i < KNL_MAX_CHAS; i++) {
2887 if (!pvt->knl.pci_cha[i]) {
2888 sbridge_printk(KERN_ERR, "Missing CHA %d\n", i);
2889 goto enodev;
2893 return 0;
2895 enodev:
2896 sbridge_printk(KERN_ERR, "Some needed devices are missing\n");
2897 return -ENODEV;
2900 /****************************************************************************
2901 Error check routines
2902 ****************************************************************************/
2905 * While Sandy Bridge has error count registers, SMI BIOS read values from
2906 * and resets the counters. So, they are not reliable for the OS to read
2907 * from them. So, we have no option but to just trust on whatever MCE is
2908 * telling us about the errors.
2910 static void sbridge_mce_output_error(struct mem_ctl_info *mci,
2911 const struct mce *m)
2913 struct mem_ctl_info *new_mci;
2914 struct sbridge_pvt *pvt = mci->pvt_info;
2915 enum hw_event_mc_err_type tp_event;
2916 char *type, *optype, msg[256];
2917 bool ripv = GET_BITFIELD(m->mcgstatus, 0, 0);
2918 bool overflow = GET_BITFIELD(m->status, 62, 62);
2919 bool uncorrected_error = GET_BITFIELD(m->status, 61, 61);
2920 bool recoverable;
2921 u32 core_err_cnt = GET_BITFIELD(m->status, 38, 52);
2922 u32 mscod = GET_BITFIELD(m->status, 16, 31);
2923 u32 errcode = GET_BITFIELD(m->status, 0, 15);
2924 u32 channel = GET_BITFIELD(m->status, 0, 3);
2925 u32 optypenum = GET_BITFIELD(m->status, 4, 6);
2926 long channel_mask, first_channel;
2927 u8 rank, socket, ha;
2928 int rc, dimm;
2929 char *area_type = NULL;
2931 if (pvt->info.type != SANDY_BRIDGE)
2932 recoverable = true;
2933 else
2934 recoverable = GET_BITFIELD(m->status, 56, 56);
2936 if (uncorrected_error) {
2937 if (ripv) {
2938 type = "FATAL";
2939 tp_event = HW_EVENT_ERR_FATAL;
2940 } else {
2941 type = "NON_FATAL";
2942 tp_event = HW_EVENT_ERR_UNCORRECTED;
2944 } else {
2945 type = "CORRECTED";
2946 tp_event = HW_EVENT_ERR_CORRECTED;
2950 * According with Table 15-9 of the Intel Architecture spec vol 3A,
2951 * memory errors should fit in this mask:
2952 * 000f 0000 1mmm cccc (binary)
2953 * where:
2954 * f = Correction Report Filtering Bit. If 1, subsequent errors
2955 * won't be shown
2956 * mmm = error type
2957 * cccc = channel
2958 * If the mask doesn't match, report an error to the parsing logic
2960 if (! ((errcode & 0xef80) == 0x80)) {
2961 optype = "Can't parse: it is not a mem";
2962 } else {
2963 switch (optypenum) {
2964 case 0:
2965 optype = "generic undef request error";
2966 break;
2967 case 1:
2968 optype = "memory read error";
2969 break;
2970 case 2:
2971 optype = "memory write error";
2972 break;
2973 case 3:
2974 optype = "addr/cmd error";
2975 break;
2976 case 4:
2977 optype = "memory scrubbing error";
2978 break;
2979 default:
2980 optype = "reserved";
2981 break;
2985 /* Only decode errors with an valid address (ADDRV) */
2986 if (!GET_BITFIELD(m->status, 58, 58))
2987 return;
2989 if (pvt->info.type == KNIGHTS_LANDING) {
2990 if (channel == 14) {
2991 edac_dbg(0, "%s%s err_code:%04x:%04x EDRAM bank %d\n",
2992 overflow ? " OVERFLOW" : "",
2993 (uncorrected_error && recoverable)
2994 ? " recoverable" : "",
2995 mscod, errcode,
2996 m->bank);
2997 } else {
2998 char A = *("A");
3001 * Reported channel is in range 0-2, so we can't map it
3002 * back to mc. To figure out mc we check machine check
3003 * bank register that reported this error.
3004 * bank15 means mc0 and bank16 means mc1.
3006 channel = knl_channel_remap(m->bank == 16, channel);
3007 channel_mask = 1 << channel;
3009 snprintf(msg, sizeof(msg),
3010 "%s%s err_code:%04x:%04x channel:%d (DIMM_%c)",
3011 overflow ? " OVERFLOW" : "",
3012 (uncorrected_error && recoverable)
3013 ? " recoverable" : " ",
3014 mscod, errcode, channel, A + channel);
3015 edac_mc_handle_error(tp_event, mci, core_err_cnt,
3016 m->addr >> PAGE_SHIFT, m->addr & ~PAGE_MASK, 0,
3017 channel, 0, -1,
3018 optype, msg);
3020 return;
3021 } else {
3022 rc = get_memory_error_data(mci, m->addr, &socket, &ha,
3023 &channel_mask, &rank, &area_type, msg);
3026 if (rc < 0)
3027 goto err_parsing;
3028 new_mci = get_mci_for_node_id(socket);
3029 if (!new_mci) {
3030 strcpy(msg, "Error: socket got corrupted!");
3031 goto err_parsing;
3033 mci = new_mci;
3034 pvt = mci->pvt_info;
3036 first_channel = find_first_bit(&channel_mask, NUM_CHANNELS);
3038 if (rank < 4)
3039 dimm = 0;
3040 else if (rank < 8)
3041 dimm = 1;
3042 else
3043 dimm = 2;
3047 * FIXME: On some memory configurations (mirror, lockstep), the
3048 * Memory Controller can't point the error to a single DIMM. The
3049 * EDAC core should be handling the channel mask, in order to point
3050 * to the group of dimm's where the error may be happening.
3052 if (!pvt->is_lockstep && !pvt->is_mirrored && !pvt->is_close_pg)
3053 channel = first_channel;
3055 snprintf(msg, sizeof(msg),
3056 "%s%s area:%s err_code:%04x:%04x socket:%d ha:%d channel_mask:%ld rank:%d",
3057 overflow ? " OVERFLOW" : "",
3058 (uncorrected_error && recoverable) ? " recoverable" : "",
3059 area_type,
3060 mscod, errcode,
3061 socket, ha,
3062 channel_mask,
3063 rank);
3065 edac_dbg(0, "%s\n", msg);
3067 /* FIXME: need support for channel mask */
3069 if (channel == CHANNEL_UNSPECIFIED)
3070 channel = -1;
3072 /* Call the helper to output message */
3073 edac_mc_handle_error(tp_event, mci, core_err_cnt,
3074 m->addr >> PAGE_SHIFT, m->addr & ~PAGE_MASK, 0,
3075 4*ha+channel, dimm, -1,
3076 optype, msg);
3077 return;
3078 err_parsing:
3079 edac_mc_handle_error(tp_event, mci, core_err_cnt, 0, 0, 0,
3080 -1, -1, -1,
3081 msg, "");
3086 * Check that logging is enabled and that this is the right type
3087 * of error for us to handle.
3089 static int sbridge_mce_check_error(struct notifier_block *nb, unsigned long val,
3090 void *data)
3092 struct mce *mce = (struct mce *)data;
3093 struct mem_ctl_info *mci;
3094 struct sbridge_pvt *pvt;
3095 char *type;
3097 if (get_edac_report_status() == EDAC_REPORTING_DISABLED)
3098 return NOTIFY_DONE;
3100 mci = get_mci_for_node_id(mce->socketid);
3101 if (!mci)
3102 return NOTIFY_DONE;
3103 pvt = mci->pvt_info;
3106 * Just let mcelog handle it if the error is
3107 * outside the memory controller. A memory error
3108 * is indicated by bit 7 = 1 and bits = 8-11,13-15 = 0.
3109 * bit 12 has an special meaning.
3111 if ((mce->status & 0xefff) >> 7 != 1)
3112 return NOTIFY_DONE;
3114 if (mce->mcgstatus & MCG_STATUS_MCIP)
3115 type = "Exception";
3116 else
3117 type = "Event";
3119 sbridge_mc_printk(mci, KERN_DEBUG, "HANDLING MCE MEMORY ERROR\n");
3121 sbridge_mc_printk(mci, KERN_DEBUG, "CPU %d: Machine Check %s: %Lx "
3122 "Bank %d: %016Lx\n", mce->extcpu, type,
3123 mce->mcgstatus, mce->bank, mce->status);
3124 sbridge_mc_printk(mci, KERN_DEBUG, "TSC %llx ", mce->tsc);
3125 sbridge_mc_printk(mci, KERN_DEBUG, "ADDR %llx ", mce->addr);
3126 sbridge_mc_printk(mci, KERN_DEBUG, "MISC %llx ", mce->misc);
3128 sbridge_mc_printk(mci, KERN_DEBUG, "PROCESSOR %u:%x TIME %llu SOCKET "
3129 "%u APIC %x\n", mce->cpuvendor, mce->cpuid,
3130 mce->time, mce->socketid, mce->apicid);
3132 sbridge_mce_output_error(mci, mce);
3134 /* Advice mcelog that the error were handled */
3135 return NOTIFY_STOP;
3138 static struct notifier_block sbridge_mce_dec = {
3139 .notifier_call = sbridge_mce_check_error,
3142 /****************************************************************************
3143 EDAC register/unregister logic
3144 ****************************************************************************/
3146 static void sbridge_unregister_mci(struct sbridge_dev *sbridge_dev)
3148 struct mem_ctl_info *mci = sbridge_dev->mci;
3149 struct sbridge_pvt *pvt;
3151 if (unlikely(!mci || !mci->pvt_info)) {
3152 edac_dbg(0, "MC: dev = %p\n", &sbridge_dev->pdev[0]->dev);
3154 sbridge_printk(KERN_ERR, "Couldn't find mci handler\n");
3155 return;
3158 pvt = mci->pvt_info;
3160 edac_dbg(0, "MC: mci = %p, dev = %p\n",
3161 mci, &sbridge_dev->pdev[0]->dev);
3163 /* Remove MC sysfs nodes */
3164 edac_mc_del_mc(mci->pdev);
3166 edac_dbg(1, "%s: free mci struct\n", mci->ctl_name);
3167 kfree(mci->ctl_name);
3168 edac_mc_free(mci);
3169 sbridge_dev->mci = NULL;
3172 static int sbridge_register_mci(struct sbridge_dev *sbridge_dev, enum type type)
3174 struct mem_ctl_info *mci;
3175 struct edac_mc_layer layers[2];
3176 struct sbridge_pvt *pvt;
3177 struct pci_dev *pdev = sbridge_dev->pdev[0];
3178 int rc;
3180 /* Check the number of active and not disabled channels */
3181 rc = check_if_ecc_is_active(sbridge_dev->bus, type);
3182 if (unlikely(rc < 0))
3183 return rc;
3185 /* allocate a new MC control structure */
3186 layers[0].type = EDAC_MC_LAYER_CHANNEL;
3187 layers[0].size = type == KNIGHTS_LANDING ?
3188 KNL_MAX_CHANNELS : NUM_CHANNELS;
3189 layers[0].is_virt_csrow = false;
3190 layers[1].type = EDAC_MC_LAYER_SLOT;
3191 layers[1].size = type == KNIGHTS_LANDING ? 1 : MAX_DIMMS;
3192 layers[1].is_virt_csrow = true;
3193 mci = edac_mc_alloc(sbridge_dev->mc, ARRAY_SIZE(layers), layers,
3194 sizeof(*pvt));
3196 if (unlikely(!mci))
3197 return -ENOMEM;
3199 edac_dbg(0, "MC: mci = %p, dev = %p\n",
3200 mci, &pdev->dev);
3202 pvt = mci->pvt_info;
3203 memset(pvt, 0, sizeof(*pvt));
3205 /* Associate sbridge_dev and mci for future usage */
3206 pvt->sbridge_dev = sbridge_dev;
3207 sbridge_dev->mci = mci;
3209 mci->mtype_cap = type == KNIGHTS_LANDING ?
3210 MEM_FLAG_DDR4 : MEM_FLAG_DDR3;
3211 mci->edac_ctl_cap = EDAC_FLAG_NONE;
3212 mci->edac_cap = EDAC_FLAG_NONE;
3213 mci->mod_name = "sbridge_edac.c";
3214 mci->mod_ver = SBRIDGE_REVISION;
3215 mci->dev_name = pci_name(pdev);
3216 mci->ctl_page_to_phys = NULL;
3218 pvt->info.type = type;
3219 switch (type) {
3220 case IVY_BRIDGE:
3221 pvt->info.rankcfgr = IB_RANK_CFG_A;
3222 pvt->info.get_tolm = ibridge_get_tolm;
3223 pvt->info.get_tohm = ibridge_get_tohm;
3224 pvt->info.dram_rule = ibridge_dram_rule;
3225 pvt->info.get_memory_type = get_memory_type;
3226 pvt->info.get_node_id = get_node_id;
3227 pvt->info.rir_limit = rir_limit;
3228 pvt->info.sad_limit = sad_limit;
3229 pvt->info.interleave_mode = interleave_mode;
3230 pvt->info.show_interleave_mode = show_interleave_mode;
3231 pvt->info.dram_attr = dram_attr;
3232 pvt->info.max_sad = ARRAY_SIZE(ibridge_dram_rule);
3233 pvt->info.interleave_list = ibridge_interleave_list;
3234 pvt->info.max_interleave = ARRAY_SIZE(ibridge_interleave_list);
3235 pvt->info.interleave_pkg = ibridge_interleave_pkg;
3236 pvt->info.get_width = ibridge_get_width;
3237 mci->ctl_name = kasprintf(GFP_KERNEL, "Ivy Bridge Socket#%d", mci->mc_idx);
3239 /* Store pci devices at mci for faster access */
3240 rc = ibridge_mci_bind_devs(mci, sbridge_dev);
3241 if (unlikely(rc < 0))
3242 goto fail0;
3243 break;
3244 case SANDY_BRIDGE:
3245 pvt->info.rankcfgr = SB_RANK_CFG_A;
3246 pvt->info.get_tolm = sbridge_get_tolm;
3247 pvt->info.get_tohm = sbridge_get_tohm;
3248 pvt->info.dram_rule = sbridge_dram_rule;
3249 pvt->info.get_memory_type = get_memory_type;
3250 pvt->info.get_node_id = get_node_id;
3251 pvt->info.rir_limit = rir_limit;
3252 pvt->info.sad_limit = sad_limit;
3253 pvt->info.interleave_mode = interleave_mode;
3254 pvt->info.show_interleave_mode = show_interleave_mode;
3255 pvt->info.dram_attr = dram_attr;
3256 pvt->info.max_sad = ARRAY_SIZE(sbridge_dram_rule);
3257 pvt->info.interleave_list = sbridge_interleave_list;
3258 pvt->info.max_interleave = ARRAY_SIZE(sbridge_interleave_list);
3259 pvt->info.interleave_pkg = sbridge_interleave_pkg;
3260 pvt->info.get_width = sbridge_get_width;
3261 mci->ctl_name = kasprintf(GFP_KERNEL, "Sandy Bridge Socket#%d", mci->mc_idx);
3263 /* Store pci devices at mci for faster access */
3264 rc = sbridge_mci_bind_devs(mci, sbridge_dev);
3265 if (unlikely(rc < 0))
3266 goto fail0;
3267 break;
3268 case HASWELL:
3269 /* rankcfgr isn't used */
3270 pvt->info.get_tolm = haswell_get_tolm;
3271 pvt->info.get_tohm = haswell_get_tohm;
3272 pvt->info.dram_rule = ibridge_dram_rule;
3273 pvt->info.get_memory_type = haswell_get_memory_type;
3274 pvt->info.get_node_id = haswell_get_node_id;
3275 pvt->info.rir_limit = haswell_rir_limit;
3276 pvt->info.sad_limit = sad_limit;
3277 pvt->info.interleave_mode = interleave_mode;
3278 pvt->info.show_interleave_mode = show_interleave_mode;
3279 pvt->info.dram_attr = dram_attr;
3280 pvt->info.max_sad = ARRAY_SIZE(ibridge_dram_rule);
3281 pvt->info.interleave_list = ibridge_interleave_list;
3282 pvt->info.max_interleave = ARRAY_SIZE(ibridge_interleave_list);
3283 pvt->info.interleave_pkg = ibridge_interleave_pkg;
3284 pvt->info.get_width = ibridge_get_width;
3285 mci->ctl_name = kasprintf(GFP_KERNEL, "Haswell Socket#%d", mci->mc_idx);
3287 /* Store pci devices at mci for faster access */
3288 rc = haswell_mci_bind_devs(mci, sbridge_dev);
3289 if (unlikely(rc < 0))
3290 goto fail0;
3291 break;
3292 case BROADWELL:
3293 /* rankcfgr isn't used */
3294 pvt->info.get_tolm = haswell_get_tolm;
3295 pvt->info.get_tohm = haswell_get_tohm;
3296 pvt->info.dram_rule = ibridge_dram_rule;
3297 pvt->info.get_memory_type = haswell_get_memory_type;
3298 pvt->info.get_node_id = haswell_get_node_id;
3299 pvt->info.rir_limit = haswell_rir_limit;
3300 pvt->info.sad_limit = sad_limit;
3301 pvt->info.interleave_mode = interleave_mode;
3302 pvt->info.show_interleave_mode = show_interleave_mode;
3303 pvt->info.dram_attr = dram_attr;
3304 pvt->info.max_sad = ARRAY_SIZE(ibridge_dram_rule);
3305 pvt->info.interleave_list = ibridge_interleave_list;
3306 pvt->info.max_interleave = ARRAY_SIZE(ibridge_interleave_list);
3307 pvt->info.interleave_pkg = ibridge_interleave_pkg;
3308 pvt->info.get_width = broadwell_get_width;
3309 mci->ctl_name = kasprintf(GFP_KERNEL, "Broadwell Socket#%d", mci->mc_idx);
3311 /* Store pci devices at mci for faster access */
3312 rc = broadwell_mci_bind_devs(mci, sbridge_dev);
3313 if (unlikely(rc < 0))
3314 goto fail0;
3315 break;
3316 case KNIGHTS_LANDING:
3317 /* pvt->info.rankcfgr == ??? */
3318 pvt->info.get_tolm = knl_get_tolm;
3319 pvt->info.get_tohm = knl_get_tohm;
3320 pvt->info.dram_rule = knl_dram_rule;
3321 pvt->info.get_memory_type = knl_get_memory_type;
3322 pvt->info.get_node_id = knl_get_node_id;
3323 pvt->info.rir_limit = NULL;
3324 pvt->info.sad_limit = knl_sad_limit;
3325 pvt->info.interleave_mode = knl_interleave_mode;
3326 pvt->info.show_interleave_mode = knl_show_interleave_mode;
3327 pvt->info.dram_attr = dram_attr_knl;
3328 pvt->info.max_sad = ARRAY_SIZE(knl_dram_rule);
3329 pvt->info.interleave_list = knl_interleave_list;
3330 pvt->info.max_interleave = ARRAY_SIZE(knl_interleave_list);
3331 pvt->info.interleave_pkg = ibridge_interleave_pkg;
3332 pvt->info.get_width = knl_get_width;
3333 mci->ctl_name = kasprintf(GFP_KERNEL,
3334 "Knights Landing Socket#%d", mci->mc_idx);
3336 rc = knl_mci_bind_devs(mci, sbridge_dev);
3337 if (unlikely(rc < 0))
3338 goto fail0;
3339 break;
3342 /* Get dimm basic config and the memory layout */
3343 get_dimm_config(mci);
3344 get_memory_layout(mci);
3346 /* record ptr to the generic device */
3347 mci->pdev = &pdev->dev;
3349 /* add this new MC control structure to EDAC's list of MCs */
3350 if (unlikely(edac_mc_add_mc(mci))) {
3351 edac_dbg(0, "MC: failed edac_mc_add_mc()\n");
3352 rc = -EINVAL;
3353 goto fail0;
3356 return 0;
3358 fail0:
3359 kfree(mci->ctl_name);
3360 edac_mc_free(mci);
3361 sbridge_dev->mci = NULL;
3362 return rc;
3365 #define ICPU(model, table) \
3366 { X86_VENDOR_INTEL, 6, model, 0, (unsigned long)&table }
3368 static const struct x86_cpu_id sbridge_cpuids[] = {
3369 ICPU(INTEL_FAM6_SANDYBRIDGE_X, pci_dev_descr_sbridge_table),
3370 ICPU(INTEL_FAM6_IVYBRIDGE_X, pci_dev_descr_ibridge_table),
3371 ICPU(INTEL_FAM6_HASWELL_X, pci_dev_descr_haswell_table),
3372 ICPU(INTEL_FAM6_BROADWELL_X, pci_dev_descr_broadwell_table),
3373 ICPU(INTEL_FAM6_BROADWELL_XEON_D, pci_dev_descr_broadwell_table),
3374 ICPU(INTEL_FAM6_XEON_PHI_KNL, pci_dev_descr_knl_table),
3375 ICPU(INTEL_FAM6_XEON_PHI_KNM, pci_dev_descr_knl_table),
3378 MODULE_DEVICE_TABLE(x86cpu, sbridge_cpuids);
3381 * sbridge_probe Get all devices and register memory controllers
3382 * present.
3383 * return:
3384 * 0 for FOUND a device
3385 * < 0 for error code
3388 static int sbridge_probe(const struct x86_cpu_id *id)
3390 int rc = -ENODEV;
3391 u8 mc, num_mc = 0;
3392 struct sbridge_dev *sbridge_dev;
3393 struct pci_id_table *ptable = (struct pci_id_table *)id->driver_data;
3395 /* get the pci devices we want to reserve for our use */
3396 rc = sbridge_get_all_devices(&num_mc, ptable);
3398 if (unlikely(rc < 0)) {
3399 edac_dbg(0, "couldn't get all devices\n");
3400 goto fail0;
3403 mc = 0;
3405 list_for_each_entry(sbridge_dev, &sbridge_edac_list, list) {
3406 edac_dbg(0, "Registering MC#%d (%d of %d)\n",
3407 mc, mc + 1, num_mc);
3409 sbridge_dev->mc = mc++;
3410 rc = sbridge_register_mci(sbridge_dev, ptable->type);
3411 if (unlikely(rc < 0))
3412 goto fail1;
3415 sbridge_printk(KERN_INFO, "%s\n", SBRIDGE_REVISION);
3417 return 0;
3419 fail1:
3420 list_for_each_entry(sbridge_dev, &sbridge_edac_list, list)
3421 sbridge_unregister_mci(sbridge_dev);
3423 sbridge_put_all_devices();
3424 fail0:
3425 return rc;
3429 * sbridge_remove cleanup
3432 static void sbridge_remove(void)
3434 struct sbridge_dev *sbridge_dev;
3436 edac_dbg(0, "\n");
3438 list_for_each_entry(sbridge_dev, &sbridge_edac_list, list)
3439 sbridge_unregister_mci(sbridge_dev);
3441 /* Release PCI resources */
3442 sbridge_put_all_devices();
3446 * sbridge_init Module entry function
3447 * Try to initialize this module for its devices
3449 static int __init sbridge_init(void)
3451 const struct x86_cpu_id *id;
3452 int rc;
3454 edac_dbg(2, "\n");
3456 id = x86_match_cpu(sbridge_cpuids);
3457 if (!id)
3458 return -ENODEV;
3460 /* Ensure that the OPSTATE is set correctly for POLL or NMI */
3461 opstate_init();
3463 rc = sbridge_probe(id);
3465 if (rc >= 0) {
3466 mce_register_decode_chain(&sbridge_mce_dec);
3467 if (get_edac_report_status() == EDAC_REPORTING_DISABLED)
3468 sbridge_printk(KERN_WARNING, "Loading driver, error reporting disabled.\n");
3469 return 0;
3472 sbridge_printk(KERN_ERR, "Failed to register device with error %d.\n",
3473 rc);
3475 return rc;
3479 * sbridge_exit() Module exit function
3480 * Unregister the driver
3482 static void __exit sbridge_exit(void)
3484 edac_dbg(2, "\n");
3485 sbridge_remove();
3486 mce_unregister_decode_chain(&sbridge_mce_dec);
3489 module_init(sbridge_init);
3490 module_exit(sbridge_exit);
3492 module_param(edac_op_state, int, 0444);
3493 MODULE_PARM_DESC(edac_op_state, "EDAC Error Reporting state: 0=Poll,1=NMI");
3495 MODULE_LICENSE("GPL");
3496 MODULE_AUTHOR("Mauro Carvalho Chehab");
3497 MODULE_AUTHOR("Red Hat Inc. (http://www.redhat.com)");
3498 MODULE_DESCRIPTION("MC Driver for Intel Sandy Bridge and Ivy Bridge memory controllers - "
3499 SBRIDGE_REVISION);