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 <asm/processor.h>
27 #include "edac_core.h"
30 static LIST_HEAD(sbridge_edac_list
);
31 static DEFINE_MUTEX(sbridge_edac_lock
);
35 * Alter this version for the module when modifications are made
37 #define SBRIDGE_REVISION " Ver: 1.1.0 "
38 #define EDAC_MOD_STR "sbridge_edac"
43 #define sbridge_printk(level, fmt, arg...) \
44 edac_printk(level, "sbridge", fmt, ##arg)
46 #define sbridge_mc_printk(mci, level, fmt, arg...) \
47 edac_mc_chipset_printk(mci, level, "sbridge", fmt, ##arg)
50 * Get a bit field at register value <v>, from bit <lo> to bit <hi>
52 #define GET_BITFIELD(v, lo, hi) \
53 (((v) & GENMASK_ULL(hi, lo)) >> (lo))
55 /* Devices 12 Function 6, Offsets 0x80 to 0xcc */
56 static const u32 sbridge_dram_rule
[] = {
57 0x80, 0x88, 0x90, 0x98, 0xa0,
58 0xa8, 0xb0, 0xb8, 0xc0, 0xc8,
61 static const u32 ibridge_dram_rule
[] = {
62 0x60, 0x68, 0x70, 0x78, 0x80,
63 0x88, 0x90, 0x98, 0xa0, 0xa8,
64 0xb0, 0xb8, 0xc0, 0xc8, 0xd0,
65 0xd8, 0xe0, 0xe8, 0xf0, 0xf8,
68 #define SAD_LIMIT(reg) ((GET_BITFIELD(reg, 6, 25) << 26) | 0x3ffffff)
69 #define DRAM_ATTR(reg) GET_BITFIELD(reg, 2, 3)
70 #define INTERLEAVE_MODE(reg) GET_BITFIELD(reg, 1, 1)
71 #define DRAM_RULE_ENABLE(reg) GET_BITFIELD(reg, 0, 0)
72 #define A7MODE(reg) GET_BITFIELD(reg, 26, 26)
74 static char *get_dram_attr(u32 reg
)
76 switch(DRAM_ATTR(reg
)) {
88 static const u32 sbridge_interleave_list
[] = {
89 0x84, 0x8c, 0x94, 0x9c, 0xa4,
90 0xac, 0xb4, 0xbc, 0xc4, 0xcc,
93 static const u32 ibridge_interleave_list
[] = {
94 0x64, 0x6c, 0x74, 0x7c, 0x84,
95 0x8c, 0x94, 0x9c, 0xa4, 0xac,
96 0xb4, 0xbc, 0xc4, 0xcc, 0xd4,
97 0xdc, 0xe4, 0xec, 0xf4, 0xfc,
100 struct interleave_pkg
{
105 static const struct interleave_pkg sbridge_interleave_pkg
[] = {
116 static const struct interleave_pkg ibridge_interleave_pkg
[] = {
127 static inline int sad_pkg(const struct interleave_pkg
*table
, u32 reg
,
130 return GET_BITFIELD(reg
, table
[interleave
].start
,
131 table
[interleave
].end
);
134 /* Devices 12 Function 7 */
138 #define HASWELL_TOLM 0xd0
139 #define HASWELL_TOHM_0 0xd4
140 #define HASWELL_TOHM_1 0xd8
142 #define GET_TOLM(reg) ((GET_BITFIELD(reg, 0, 3) << 28) | 0x3ffffff)
143 #define GET_TOHM(reg) ((GET_BITFIELD(reg, 0, 20) << 25) | 0x3ffffff)
145 /* Device 13 Function 6 */
147 #define SAD_TARGET 0xf0
149 #define SOURCE_ID(reg) GET_BITFIELD(reg, 9, 11)
151 #define SAD_CONTROL 0xf4
153 /* Device 14 function 0 */
155 static const u32 tad_dram_rule
[] = {
156 0x40, 0x44, 0x48, 0x4c,
157 0x50, 0x54, 0x58, 0x5c,
158 0x60, 0x64, 0x68, 0x6c,
160 #define MAX_TAD ARRAY_SIZE(tad_dram_rule)
162 #define TAD_LIMIT(reg) ((GET_BITFIELD(reg, 12, 31) << 26) | 0x3ffffff)
163 #define TAD_SOCK(reg) GET_BITFIELD(reg, 10, 11)
164 #define TAD_CH(reg) GET_BITFIELD(reg, 8, 9)
165 #define TAD_TGT3(reg) GET_BITFIELD(reg, 6, 7)
166 #define TAD_TGT2(reg) GET_BITFIELD(reg, 4, 5)
167 #define TAD_TGT1(reg) GET_BITFIELD(reg, 2, 3)
168 #define TAD_TGT0(reg) GET_BITFIELD(reg, 0, 1)
170 /* Device 15, function 0 */
174 #define IS_ECC_ENABLED(mcmtr) GET_BITFIELD(mcmtr, 2, 2)
175 #define IS_LOCKSTEP_ENABLED(mcmtr) GET_BITFIELD(mcmtr, 1, 1)
176 #define IS_CLOSE_PG(mcmtr) GET_BITFIELD(mcmtr, 0, 0)
178 /* Device 15, function 1 */
180 #define RASENABLES 0xac
181 #define IS_MIRROR_ENABLED(reg) GET_BITFIELD(reg, 0, 0)
183 /* Device 15, functions 2-5 */
185 static const int mtr_regs
[] = {
189 #define RANK_DISABLE(mtr) GET_BITFIELD(mtr, 16, 19)
190 #define IS_DIMM_PRESENT(mtr) GET_BITFIELD(mtr, 14, 14)
191 #define RANK_CNT_BITS(mtr) GET_BITFIELD(mtr, 12, 13)
192 #define RANK_WIDTH_BITS(mtr) GET_BITFIELD(mtr, 2, 4)
193 #define COL_WIDTH_BITS(mtr) GET_BITFIELD(mtr, 0, 1)
195 static const u32 tad_ch_nilv_offset
[] = {
196 0x90, 0x94, 0x98, 0x9c,
197 0xa0, 0xa4, 0xa8, 0xac,
198 0xb0, 0xb4, 0xb8, 0xbc,
200 #define CHN_IDX_OFFSET(reg) GET_BITFIELD(reg, 28, 29)
201 #define TAD_OFFSET(reg) (GET_BITFIELD(reg, 6, 25) << 26)
203 static const u32 rir_way_limit
[] = {
204 0x108, 0x10c, 0x110, 0x114, 0x118,
206 #define MAX_RIR_RANGES ARRAY_SIZE(rir_way_limit)
208 #define IS_RIR_VALID(reg) GET_BITFIELD(reg, 31, 31)
209 #define RIR_WAY(reg) GET_BITFIELD(reg, 28, 29)
211 #define MAX_RIR_WAY 8
213 static const u32 rir_offset
[MAX_RIR_RANGES
][MAX_RIR_WAY
] = {
214 { 0x120, 0x124, 0x128, 0x12c, 0x130, 0x134, 0x138, 0x13c },
215 { 0x140, 0x144, 0x148, 0x14c, 0x150, 0x154, 0x158, 0x15c },
216 { 0x160, 0x164, 0x168, 0x16c, 0x170, 0x174, 0x178, 0x17c },
217 { 0x180, 0x184, 0x188, 0x18c, 0x190, 0x194, 0x198, 0x19c },
218 { 0x1a0, 0x1a4, 0x1a8, 0x1ac, 0x1b0, 0x1b4, 0x1b8, 0x1bc },
221 #define RIR_RNK_TGT(type, reg) (((type) == BROADWELL) ? \
222 GET_BITFIELD(reg, 20, 23) : GET_BITFIELD(reg, 16, 19))
224 #define RIR_OFFSET(type, reg) (((type) == HASWELL || (type) == BROADWELL) ? \
225 GET_BITFIELD(reg, 2, 15) : GET_BITFIELD(reg, 2, 14))
227 /* Device 16, functions 2-7 */
230 * FIXME: Implement the error count reads directly
233 static const u32 correrrcnt
[] = {
234 0x104, 0x108, 0x10c, 0x110,
237 #define RANK_ODD_OV(reg) GET_BITFIELD(reg, 31, 31)
238 #define RANK_ODD_ERR_CNT(reg) GET_BITFIELD(reg, 16, 30)
239 #define RANK_EVEN_OV(reg) GET_BITFIELD(reg, 15, 15)
240 #define RANK_EVEN_ERR_CNT(reg) GET_BITFIELD(reg, 0, 14)
242 static const u32 correrrthrsld
[] = {
243 0x11c, 0x120, 0x124, 0x128,
246 #define RANK_ODD_ERR_THRSLD(reg) GET_BITFIELD(reg, 16, 30)
247 #define RANK_EVEN_ERR_THRSLD(reg) GET_BITFIELD(reg, 0, 14)
250 /* Device 17, function 0 */
252 #define SB_RANK_CFG_A 0x0328
254 #define IB_RANK_CFG_A 0x0320
260 #define NUM_CHANNELS 4
261 #define MAX_DIMMS 3 /* Max DIMMS per channel */
262 #define CHANNEL_UNSPECIFIED 0xf /* Intel IA32 SDM 15-14 */
272 struct sbridge_info
{
276 u64 (*get_tolm
)(struct sbridge_pvt
*pvt
);
277 u64 (*get_tohm
)(struct sbridge_pvt
*pvt
);
278 u64 (*rir_limit
)(u32 reg
);
279 const u32
*dram_rule
;
280 const u32
*interleave_list
;
281 const struct interleave_pkg
*interleave_pkg
;
284 u8 (*get_node_id
)(struct sbridge_pvt
*pvt
);
285 enum mem_type (*get_memory_type
)(struct sbridge_pvt
*pvt
);
286 struct pci_dev
*pci_vtd
;
289 struct sbridge_channel
{
294 struct pci_id_descr
{
299 struct pci_id_table
{
300 const struct pci_id_descr
*descr
;
305 struct list_head list
;
307 u8 node_id
, source_id
;
308 struct pci_dev
**pdev
;
310 struct mem_ctl_info
*mci
;
314 struct pci_dev
*pci_ta
, *pci_ddrio
, *pci_ras
;
315 struct pci_dev
*pci_sad0
, *pci_sad1
;
316 struct pci_dev
*pci_ha0
, *pci_ha1
;
317 struct pci_dev
*pci_br0
, *pci_br1
;
318 struct pci_dev
*pci_ha1_ta
;
319 struct pci_dev
*pci_tad
[NUM_CHANNELS
];
321 struct sbridge_dev
*sbridge_dev
;
323 struct sbridge_info info
;
324 struct sbridge_channel channel
[NUM_CHANNELS
];
326 /* Memory type detection */
327 bool is_mirrored
, is_lockstep
, is_close_pg
;
329 /* Fifo double buffers */
330 struct mce mce_entry
[MCE_LOG_LEN
];
331 struct mce mce_outentry
[MCE_LOG_LEN
];
333 /* Fifo in/out counters */
334 unsigned mce_in
, mce_out
;
336 /* Count indicator to show errors not got */
337 unsigned mce_overrun
;
339 /* Memory description */
343 #define PCI_DESCR(device_id, opt) \
344 .dev_id = (device_id), \
347 static const struct pci_id_descr pci_dev_descr_sbridge
[] = {
348 /* Processor Home Agent */
349 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_HA0
, 0) },
351 /* Memory controller */
352 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TA
, 0) },
353 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_RAS
, 0) },
354 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD0
, 0) },
355 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD1
, 0) },
356 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD2
, 0) },
357 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD3
, 0) },
358 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_DDRIO
, 1) },
360 /* System Address Decoder */
361 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_SAD0
, 0) },
362 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_SAD1
, 0) },
364 /* Broadcast Registers */
365 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_BR
, 0) },
368 #define PCI_ID_TABLE_ENTRY(A) { .descr=A, .n_devs = ARRAY_SIZE(A) }
369 static const struct pci_id_table pci_dev_descr_sbridge_table
[] = {
370 PCI_ID_TABLE_ENTRY(pci_dev_descr_sbridge
),
371 {0,} /* 0 terminated list. */
374 /* This changes depending if 1HA or 2HA:
376 * 0x0eb8 (17.0) is DDRIO0
378 * 0x0ebc (17.4) is DDRIO0
380 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_1HA_DDRIO0 0x0eb8
381 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_2HA_DDRIO0 0x0ebc
384 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0 0x0ea0
385 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TA 0x0ea8
386 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_RAS 0x0e71
387 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD0 0x0eaa
388 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD1 0x0eab
389 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD2 0x0eac
390 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD3 0x0ead
391 #define PCI_DEVICE_ID_INTEL_IBRIDGE_SAD 0x0ec8
392 #define PCI_DEVICE_ID_INTEL_IBRIDGE_BR0 0x0ec9
393 #define PCI_DEVICE_ID_INTEL_IBRIDGE_BR1 0x0eca
394 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1 0x0e60
395 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TA 0x0e68
396 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_RAS 0x0e79
397 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD0 0x0e6a
398 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD1 0x0e6b
400 static const struct pci_id_descr pci_dev_descr_ibridge
[] = {
401 /* Processor Home Agent */
402 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0
, 0) },
404 /* Memory controller */
405 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TA
, 0) },
406 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_RAS
, 0) },
407 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD0
, 0) },
408 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD1
, 0) },
409 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD2
, 0) },
410 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD3
, 0) },
412 /* System Address Decoder */
413 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_SAD
, 0) },
415 /* Broadcast Registers */
416 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_BR0
, 1) },
417 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_BR1
, 0) },
419 /* Optional, mode 2HA */
420 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1
, 1) },
422 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TA
, 1) },
423 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_RAS
, 1) },
425 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD0
, 1) },
426 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD1
, 1) },
428 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_1HA_DDRIO0
, 1) },
429 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_2HA_DDRIO0
, 1) },
432 static const struct pci_id_table pci_dev_descr_ibridge_table
[] = {
433 PCI_ID_TABLE_ENTRY(pci_dev_descr_ibridge
),
434 {0,} /* 0 terminated list. */
437 /* Haswell support */
440 * - 3 DDR3 channels, 2 DPC per channel
443 * - 4 DDR4 channels, 3 DPC per channel
446 * - 4 DDR4 channels, 3 DPC per channel
449 * - each IMC interfaces with a SMI 2 channel
450 * - each SMI channel interfaces with a scalable memory buffer
451 * - each scalable memory buffer supports 4 DDR3/DDR4 channels, 3 DPC
453 #define HASWELL_DDRCRCLKCONTROLS 0xa10 /* Ditto on Broadwell */
454 #define HASWELL_HASYSDEFEATURE2 0x84
455 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_VTD_MISC 0x2f28
456 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0 0x2fa0
457 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1 0x2f60
458 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TA 0x2fa8
459 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_THERMAL 0x2f71
460 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TA 0x2f68
461 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_THERMAL 0x2f79
462 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_CBO_SAD0 0x2ffc
463 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_CBO_SAD1 0x2ffd
464 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD0 0x2faa
465 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD1 0x2fab
466 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD2 0x2fac
467 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD3 0x2fad
468 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD0 0x2f6a
469 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD1 0x2f6b
470 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD2 0x2f6c
471 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD3 0x2f6d
472 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO0 0x2fbd
473 static const struct pci_id_descr pci_dev_descr_haswell
[] = {
474 /* first item must be the HA */
475 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0
, 0) },
477 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_CBO_SAD0
, 0) },
478 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_CBO_SAD1
, 0) },
480 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1
, 1) },
482 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TA
, 0) },
483 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_THERMAL
, 0) },
484 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD0
, 0) },
485 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD1
, 0) },
486 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD2
, 1) },
487 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD3
, 1) },
489 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO0
, 1) },
491 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TA
, 1) },
492 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_THERMAL
, 1) },
493 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD0
, 1) },
494 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD1
, 1) },
495 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD2
, 1) },
496 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD3
, 1) },
499 static const struct pci_id_table pci_dev_descr_haswell_table
[] = {
500 PCI_ID_TABLE_ENTRY(pci_dev_descr_haswell
),
501 {0,} /* 0 terminated list. */
509 * - 2 DDR3 channels, 2 DPC per channel
511 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_VTD_MISC 0x6f28
512 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0 0x6fa0
513 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TA 0x6fa8
514 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_THERMAL 0x6f71
515 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_CBO_SAD0 0x6ffc
516 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_CBO_SAD1 0x6ffd
517 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD0 0x6faa
518 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD1 0x6fab
519 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD2 0x6fac
520 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD3 0x6fad
521 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_DDRIO0 0x6faf
523 static const struct pci_id_descr pci_dev_descr_broadwell
[] = {
524 /* first item must be the HA */
525 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0
, 0) },
527 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_CBO_SAD0
, 0) },
528 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_CBO_SAD1
, 0) },
530 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TA
, 0) },
531 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_THERMAL
, 0) },
532 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD0
, 0) },
533 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD1
, 0) },
534 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD2
, 0) },
535 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD3
, 0) },
536 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_DDRIO0
, 1) },
539 static const struct pci_id_table pci_dev_descr_broadwell_table
[] = {
540 PCI_ID_TABLE_ENTRY(pci_dev_descr_broadwell
),
541 {0,} /* 0 terminated list. */
545 * pci_device_id table for which devices we are looking for
547 static const struct pci_device_id sbridge_pci_tbl
[] = {
548 {PCI_DEVICE(PCI_VENDOR_ID_INTEL
, PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_HA0
)},
549 {PCI_DEVICE(PCI_VENDOR_ID_INTEL
, PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TA
)},
550 {PCI_DEVICE(PCI_VENDOR_ID_INTEL
, PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0
)},
551 {PCI_DEVICE(PCI_VENDOR_ID_INTEL
, PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0
)},
552 {0,} /* 0 terminated list. */
556 /****************************************************************************
557 Ancillary status routines
558 ****************************************************************************/
560 static inline int numrank(enum type type
, u32 mtr
)
562 int ranks
= (1 << RANK_CNT_BITS(mtr
));
569 edac_dbg(0, "Invalid number of ranks: %d (max = %i) raw value = %x (%04x)\n",
570 ranks
, max
, (unsigned int)RANK_CNT_BITS(mtr
), mtr
);
577 static inline int numrow(u32 mtr
)
579 int rows
= (RANK_WIDTH_BITS(mtr
) + 12);
581 if (rows
< 13 || rows
> 18) {
582 edac_dbg(0, "Invalid number of rows: %d (should be between 14 and 17) raw value = %x (%04x)\n",
583 rows
, (unsigned int)RANK_WIDTH_BITS(mtr
), mtr
);
590 static inline int numcol(u32 mtr
)
592 int cols
= (COL_WIDTH_BITS(mtr
) + 10);
595 edac_dbg(0, "Invalid number of cols: %d (max = 4) raw value = %x (%04x)\n",
596 cols
, (unsigned int)COL_WIDTH_BITS(mtr
), mtr
);
603 static struct sbridge_dev
*get_sbridge_dev(u8 bus
)
605 struct sbridge_dev
*sbridge_dev
;
607 list_for_each_entry(sbridge_dev
, &sbridge_edac_list
, list
) {
608 if (sbridge_dev
->bus
== bus
)
615 static struct sbridge_dev
*alloc_sbridge_dev(u8 bus
,
616 const struct pci_id_table
*table
)
618 struct sbridge_dev
*sbridge_dev
;
620 sbridge_dev
= kzalloc(sizeof(*sbridge_dev
), GFP_KERNEL
);
624 sbridge_dev
->pdev
= kzalloc(sizeof(*sbridge_dev
->pdev
) * table
->n_devs
,
626 if (!sbridge_dev
->pdev
) {
631 sbridge_dev
->bus
= bus
;
632 sbridge_dev
->n_devs
= table
->n_devs
;
633 list_add_tail(&sbridge_dev
->list
, &sbridge_edac_list
);
638 static void free_sbridge_dev(struct sbridge_dev
*sbridge_dev
)
640 list_del(&sbridge_dev
->list
);
641 kfree(sbridge_dev
->pdev
);
645 static u64
sbridge_get_tolm(struct sbridge_pvt
*pvt
)
649 /* Address range is 32:28 */
650 pci_read_config_dword(pvt
->pci_sad1
, TOLM
, ®
);
651 return GET_TOLM(reg
);
654 static u64
sbridge_get_tohm(struct sbridge_pvt
*pvt
)
658 pci_read_config_dword(pvt
->pci_sad1
, TOHM
, ®
);
659 return GET_TOHM(reg
);
662 static u64
ibridge_get_tolm(struct sbridge_pvt
*pvt
)
666 pci_read_config_dword(pvt
->pci_br1
, TOLM
, ®
);
668 return GET_TOLM(reg
);
671 static u64
ibridge_get_tohm(struct sbridge_pvt
*pvt
)
675 pci_read_config_dword(pvt
->pci_br1
, TOHM
, ®
);
677 return GET_TOHM(reg
);
680 static u64
rir_limit(u32 reg
)
682 return ((u64
)GET_BITFIELD(reg
, 1, 10) << 29) | 0x1fffffff;
685 static enum mem_type
get_memory_type(struct sbridge_pvt
*pvt
)
690 if (pvt
->pci_ddrio
) {
691 pci_read_config_dword(pvt
->pci_ddrio
, pvt
->info
.rankcfgr
,
693 if (GET_BITFIELD(reg
, 11, 11))
694 /* FIXME: Can also be LRDIMM */
704 static enum mem_type
haswell_get_memory_type(struct sbridge_pvt
*pvt
)
707 bool registered
= false;
708 enum mem_type mtype
= MEM_UNKNOWN
;
713 pci_read_config_dword(pvt
->pci_ddrio
,
714 HASWELL_DDRCRCLKCONTROLS
, ®
);
716 if (GET_BITFIELD(reg
, 16, 16))
719 pci_read_config_dword(pvt
->pci_ta
, MCMTR
, ®
);
720 if (GET_BITFIELD(reg
, 14, 14)) {
736 static u8
get_node_id(struct sbridge_pvt
*pvt
)
739 pci_read_config_dword(pvt
->pci_br0
, SAD_CONTROL
, ®
);
740 return GET_BITFIELD(reg
, 0, 2);
743 static u8
haswell_get_node_id(struct sbridge_pvt
*pvt
)
747 pci_read_config_dword(pvt
->pci_sad1
, SAD_CONTROL
, ®
);
748 return GET_BITFIELD(reg
, 0, 3);
751 static u64
haswell_get_tolm(struct sbridge_pvt
*pvt
)
755 pci_read_config_dword(pvt
->info
.pci_vtd
, HASWELL_TOLM
, ®
);
756 return (GET_BITFIELD(reg
, 26, 31) << 26) | 0x3ffffff;
759 static u64
haswell_get_tohm(struct sbridge_pvt
*pvt
)
764 pci_read_config_dword(pvt
->info
.pci_vtd
, HASWELL_TOHM_0
, ®
);
765 rc
= GET_BITFIELD(reg
, 26, 31);
766 pci_read_config_dword(pvt
->info
.pci_vtd
, HASWELL_TOHM_1
, ®
);
767 rc
= ((reg
<< 6) | rc
) << 26;
769 return rc
| 0x1ffffff;
772 static u64
haswell_rir_limit(u32 reg
)
774 return (((u64
)GET_BITFIELD(reg
, 1, 11) + 1) << 29) - 1;
777 static inline u8
sad_pkg_socket(u8 pkg
)
779 /* on Ivy Bridge, nodeID is SASS, where A is HA and S is node id */
780 return ((pkg
>> 3) << 2) | (pkg
& 0x3);
783 static inline u8
sad_pkg_ha(u8 pkg
)
785 return (pkg
>> 2) & 0x1;
788 /****************************************************************************
789 Memory check routines
790 ****************************************************************************/
791 static struct pci_dev
*get_pdev_same_bus(u8 bus
, u32 id
)
793 struct pci_dev
*pdev
= NULL
;
796 pdev
= pci_get_device(PCI_VENDOR_ID_INTEL
, id
, pdev
);
797 if (pdev
&& pdev
->bus
->number
== bus
)
805 * check_if_ecc_is_active() - Checks if ECC is active
807 * @type: Memory controller type
808 * returns: 0 in case ECC is active, -ENODEV if it can't be determined or
811 static int check_if_ecc_is_active(const u8 bus
, enum type type
)
813 struct pci_dev
*pdev
= NULL
;
818 id
= PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TA
;
821 id
= PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TA
;
824 id
= PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TA
;
827 id
= PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TA
;
833 pdev
= get_pdev_same_bus(bus
, id
);
835 sbridge_printk(KERN_ERR
, "Couldn't find PCI device "
836 "%04x:%04x! on bus %02d\n",
837 PCI_VENDOR_ID_INTEL
, id
, bus
);
841 pci_read_config_dword(pdev
, MCMTR
, &mcmtr
);
842 if (!IS_ECC_ENABLED(mcmtr
)) {
843 sbridge_printk(KERN_ERR
, "ECC is disabled. Aborting\n");
849 static int get_dimm_config(struct mem_ctl_info
*mci
)
851 struct sbridge_pvt
*pvt
= mci
->pvt_info
;
852 struct dimm_info
*dimm
;
853 unsigned i
, j
, banks
, ranks
, rows
, cols
, npages
;
859 if (pvt
->info
.type
== HASWELL
|| pvt
->info
.type
== BROADWELL
)
860 pci_read_config_dword(pvt
->pci_sad1
, SAD_TARGET
, ®
);
862 pci_read_config_dword(pvt
->pci_br0
, SAD_TARGET
, ®
);
864 pvt
->sbridge_dev
->source_id
= SOURCE_ID(reg
);
866 pvt
->sbridge_dev
->node_id
= pvt
->info
.get_node_id(pvt
);
867 edac_dbg(0, "mc#%d: Node ID: %d, source ID: %d\n",
868 pvt
->sbridge_dev
->mc
,
869 pvt
->sbridge_dev
->node_id
,
870 pvt
->sbridge_dev
->source_id
);
872 pci_read_config_dword(pvt
->pci_ras
, RASENABLES
, ®
);
873 if (IS_MIRROR_ENABLED(reg
)) {
874 edac_dbg(0, "Memory mirror is enabled\n");
875 pvt
->is_mirrored
= true;
877 edac_dbg(0, "Memory mirror is disabled\n");
878 pvt
->is_mirrored
= false;
881 pci_read_config_dword(pvt
->pci_ta
, MCMTR
, &pvt
->info
.mcmtr
);
882 if (IS_LOCKSTEP_ENABLED(pvt
->info
.mcmtr
)) {
883 edac_dbg(0, "Lockstep is enabled\n");
884 mode
= EDAC_S8ECD8ED
;
885 pvt
->is_lockstep
= true;
887 edac_dbg(0, "Lockstep is disabled\n");
888 mode
= EDAC_S4ECD4ED
;
889 pvt
->is_lockstep
= false;
891 if (IS_CLOSE_PG(pvt
->info
.mcmtr
)) {
892 edac_dbg(0, "address map is on closed page mode\n");
893 pvt
->is_close_pg
= true;
895 edac_dbg(0, "address map is on open page mode\n");
896 pvt
->is_close_pg
= false;
899 mtype
= pvt
->info
.get_memory_type(pvt
);
900 if (mtype
== MEM_RDDR3
|| mtype
== MEM_RDDR4
)
901 edac_dbg(0, "Memory is registered\n");
902 else if (mtype
== MEM_UNKNOWN
)
903 edac_dbg(0, "Cannot determine memory type\n");
905 edac_dbg(0, "Memory is unregistered\n");
907 if (mtype
== MEM_DDR4
|| mtype
== MEM_RDDR4
)
912 for (i
= 0; i
< NUM_CHANNELS
; i
++) {
915 for (j
= 0; j
< ARRAY_SIZE(mtr_regs
); j
++) {
916 dimm
= EDAC_DIMM_PTR(mci
->layers
, mci
->dimms
, mci
->n_layers
,
918 pci_read_config_dword(pvt
->pci_tad
[i
],
920 edac_dbg(4, "Channel #%d MTR%d = %x\n", i
, j
, mtr
);
921 if (IS_DIMM_PRESENT(mtr
)) {
922 pvt
->channel
[i
].dimms
++;
924 ranks
= numrank(pvt
->info
.type
, mtr
);
928 size
= ((u64
)rows
* cols
* banks
* ranks
) >> (20 - 3);
929 npages
= MiB_TO_PAGES(size
);
931 edac_dbg(0, "mc#%d: channel %d, dimm %d, %Ld Mb (%d pages) bank: %d, rank: %d, row: %#x, col: %#x\n",
932 pvt
->sbridge_dev
->mc
, i
, j
,
934 banks
, ranks
, rows
, cols
);
936 dimm
->nr_pages
= npages
;
940 dimm
->dtype
= DEV_X16
;
943 dimm
->dtype
= DEV_X8
;
946 dimm
->dtype
= DEV_X4
;
950 dimm
->edac_mode
= mode
;
951 snprintf(dimm
->label
, sizeof(dimm
->label
),
952 "CPU_SrcID#%u_Channel#%u_DIMM#%u",
953 pvt
->sbridge_dev
->source_id
, i
, j
);
961 static void get_memory_layout(const struct mem_ctl_info
*mci
)
963 struct sbridge_pvt
*pvt
= mci
->pvt_info
;
964 int i
, j
, k
, n_sads
, n_tads
, sad_interl
;
972 * Step 1) Get TOLM/TOHM ranges
975 pvt
->tolm
= pvt
->info
.get_tolm(pvt
);
976 tmp_mb
= (1 + pvt
->tolm
) >> 20;
978 gb
= div_u64_rem(tmp_mb
, 1024, &mb
);
979 edac_dbg(0, "TOLM: %u.%03u GB (0x%016Lx)\n",
980 gb
, (mb
*1000)/1024, (u64
)pvt
->tolm
);
982 /* Address range is already 45:25 */
983 pvt
->tohm
= pvt
->info
.get_tohm(pvt
);
984 tmp_mb
= (1 + pvt
->tohm
) >> 20;
986 gb
= div_u64_rem(tmp_mb
, 1024, &mb
);
987 edac_dbg(0, "TOHM: %u.%03u GB (0x%016Lx)\n",
988 gb
, (mb
*1000)/1024, (u64
)pvt
->tohm
);
991 * Step 2) Get SAD range and SAD Interleave list
992 * TAD registers contain the interleave wayness. However, it
993 * seems simpler to just discover it indirectly, with the
997 for (n_sads
= 0; n_sads
< pvt
->info
.max_sad
; n_sads
++) {
998 /* SAD_LIMIT Address range is 45:26 */
999 pci_read_config_dword(pvt
->pci_sad0
, pvt
->info
.dram_rule
[n_sads
],
1001 limit
= SAD_LIMIT(reg
);
1003 if (!DRAM_RULE_ENABLE(reg
))
1009 tmp_mb
= (limit
+ 1) >> 20;
1010 gb
= div_u64_rem(tmp_mb
, 1024, &mb
);
1011 edac_dbg(0, "SAD#%d %s up to %u.%03u GB (0x%016Lx) Interleave: %s reg=0x%08x\n",
1015 ((u64
)tmp_mb
) << 20L,
1016 INTERLEAVE_MODE(reg
) ? "8:6" : "[8:6]XOR[18:16]",
1020 pci_read_config_dword(pvt
->pci_sad0
, pvt
->info
.interleave_list
[n_sads
],
1022 sad_interl
= sad_pkg(pvt
->info
.interleave_pkg
, reg
, 0);
1023 for (j
= 0; j
< 8; j
++) {
1024 u32 pkg
= sad_pkg(pvt
->info
.interleave_pkg
, reg
, j
);
1025 if (j
> 0 && sad_interl
== pkg
)
1028 edac_dbg(0, "SAD#%d, interleave #%d: %d\n",
1034 * Step 3) Get TAD range
1037 for (n_tads
= 0; n_tads
< MAX_TAD
; n_tads
++) {
1038 pci_read_config_dword(pvt
->pci_ha0
, tad_dram_rule
[n_tads
],
1040 limit
= TAD_LIMIT(reg
);
1043 tmp_mb
= (limit
+ 1) >> 20;
1045 gb
= div_u64_rem(tmp_mb
, 1024, &mb
);
1046 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",
1047 n_tads
, gb
, (mb
*1000)/1024,
1048 ((u64
)tmp_mb
) << 20L,
1049 (u32
)(1 << TAD_SOCK(reg
)),
1050 (u32
)TAD_CH(reg
) + 1,
1060 * Step 4) Get TAD offsets, per each channel
1062 for (i
= 0; i
< NUM_CHANNELS
; i
++) {
1063 if (!pvt
->channel
[i
].dimms
)
1065 for (j
= 0; j
< n_tads
; j
++) {
1066 pci_read_config_dword(pvt
->pci_tad
[i
],
1067 tad_ch_nilv_offset
[j
],
1069 tmp_mb
= TAD_OFFSET(reg
) >> 20;
1070 gb
= div_u64_rem(tmp_mb
, 1024, &mb
);
1071 edac_dbg(0, "TAD CH#%d, offset #%d: %u.%03u GB (0x%016Lx), reg=0x%08x\n",
1074 ((u64
)tmp_mb
) << 20L,
1080 * Step 6) Get RIR Wayness/Limit, per each channel
1082 for (i
= 0; i
< NUM_CHANNELS
; i
++) {
1083 if (!pvt
->channel
[i
].dimms
)
1085 for (j
= 0; j
< MAX_RIR_RANGES
; j
++) {
1086 pci_read_config_dword(pvt
->pci_tad
[i
],
1090 if (!IS_RIR_VALID(reg
))
1093 tmp_mb
= pvt
->info
.rir_limit(reg
) >> 20;
1094 rir_way
= 1 << RIR_WAY(reg
);
1095 gb
= div_u64_rem(tmp_mb
, 1024, &mb
);
1096 edac_dbg(0, "CH#%d RIR#%d, limit: %u.%03u GB (0x%016Lx), way: %d, reg=0x%08x\n",
1099 ((u64
)tmp_mb
) << 20L,
1103 for (k
= 0; k
< rir_way
; k
++) {
1104 pci_read_config_dword(pvt
->pci_tad
[i
],
1107 tmp_mb
= RIR_OFFSET(pvt
->info
.type
, reg
) << 6;
1109 gb
= div_u64_rem(tmp_mb
, 1024, &mb
);
1110 edac_dbg(0, "CH#%d RIR#%d INTL#%d, offset %u.%03u GB (0x%016Lx), tgt: %d, reg=0x%08x\n",
1113 ((u64
)tmp_mb
) << 20L,
1114 (u32
)RIR_RNK_TGT(pvt
->info
.type
, reg
),
1121 static struct mem_ctl_info
*get_mci_for_node_id(u8 node_id
)
1123 struct sbridge_dev
*sbridge_dev
;
1125 list_for_each_entry(sbridge_dev
, &sbridge_edac_list
, list
) {
1126 if (sbridge_dev
->node_id
== node_id
)
1127 return sbridge_dev
->mci
;
1132 static int get_memory_error_data(struct mem_ctl_info
*mci
,
1137 char **area_type
, char *msg
)
1139 struct mem_ctl_info
*new_mci
;
1140 struct sbridge_pvt
*pvt
= mci
->pvt_info
;
1141 struct pci_dev
*pci_ha
;
1142 int n_rir
, n_sads
, n_tads
, sad_way
, sck_xch
;
1143 int sad_interl
, idx
, base_ch
;
1144 int interleave_mode
, shiftup
= 0;
1145 unsigned sad_interleave
[pvt
->info
.max_interleave
];
1147 u8 ch_way
, sck_way
, pkg
, sad_ha
= 0;
1151 u64 ch_addr
, offset
, limit
= 0, prv
= 0;
1155 * Step 0) Check if the address is at special memory ranges
1156 * The check bellow is probably enough to fill all cases where
1157 * the error is not inside a memory, except for the legacy
1158 * range (e. g. VGA addresses). It is unlikely, however, that the
1159 * memory controller would generate an error on that range.
1161 if ((addr
> (u64
) pvt
->tolm
) && (addr
< (1LL << 32))) {
1162 sprintf(msg
, "Error at TOLM area, on addr 0x%08Lx", addr
);
1165 if (addr
>= (u64
)pvt
->tohm
) {
1166 sprintf(msg
, "Error at MMIOH area, on addr 0x%016Lx", addr
);
1171 * Step 1) Get socket
1173 for (n_sads
= 0; n_sads
< pvt
->info
.max_sad
; n_sads
++) {
1174 pci_read_config_dword(pvt
->pci_sad0
, pvt
->info
.dram_rule
[n_sads
],
1177 if (!DRAM_RULE_ENABLE(reg
))
1180 limit
= SAD_LIMIT(reg
);
1182 sprintf(msg
, "Can't discover the memory socket");
1189 if (n_sads
== pvt
->info
.max_sad
) {
1190 sprintf(msg
, "Can't discover the memory socket");
1194 *area_type
= get_dram_attr(dram_rule
);
1195 interleave_mode
= INTERLEAVE_MODE(dram_rule
);
1197 pci_read_config_dword(pvt
->pci_sad0
, pvt
->info
.interleave_list
[n_sads
],
1200 if (pvt
->info
.type
== SANDY_BRIDGE
) {
1201 sad_interl
= sad_pkg(pvt
->info
.interleave_pkg
, reg
, 0);
1202 for (sad_way
= 0; sad_way
< 8; sad_way
++) {
1203 u32 pkg
= sad_pkg(pvt
->info
.interleave_pkg
, reg
, sad_way
);
1204 if (sad_way
> 0 && sad_interl
== pkg
)
1206 sad_interleave
[sad_way
] = pkg
;
1207 edac_dbg(0, "SAD interleave #%d: %d\n",
1208 sad_way
, sad_interleave
[sad_way
]);
1210 edac_dbg(0, "mc#%d: Error detected on SAD#%d: address 0x%016Lx < 0x%016Lx, Interleave [%d:6]%s\n",
1211 pvt
->sbridge_dev
->mc
,
1216 !interleave_mode
? "" : "XOR[18:16]");
1217 if (interleave_mode
)
1218 idx
= ((addr
>> 6) ^ (addr
>> 16)) & 7;
1220 idx
= (addr
>> 6) & 7;
1234 sprintf(msg
, "Can't discover socket interleave");
1237 *socket
= sad_interleave
[idx
];
1238 edac_dbg(0, "SAD interleave index: %d (wayness %d) = CPU socket %d\n",
1239 idx
, sad_way
, *socket
);
1240 } else if (pvt
->info
.type
== HASWELL
|| pvt
->info
.type
== BROADWELL
) {
1241 int bits
, a7mode
= A7MODE(dram_rule
);
1244 /* A7 mode swaps P9 with P6 */
1245 bits
= GET_BITFIELD(addr
, 7, 8) << 1;
1246 bits
|= GET_BITFIELD(addr
, 9, 9);
1248 bits
= GET_BITFIELD(addr
, 7, 9);
1250 if (interleave_mode
) {
1251 /* interleave mode will XOR {8,7,6} with {18,17,16} */
1252 idx
= GET_BITFIELD(addr
, 16, 18);
1257 pkg
= sad_pkg(pvt
->info
.interleave_pkg
, reg
, idx
);
1258 *socket
= sad_pkg_socket(pkg
);
1259 sad_ha
= sad_pkg_ha(pkg
);
1262 /* MCChanShiftUpEnable */
1263 pci_read_config_dword(pvt
->pci_ha0
,
1264 HASWELL_HASYSDEFEATURE2
, ®
);
1265 shiftup
= GET_BITFIELD(reg
, 22, 22);
1268 edac_dbg(0, "SAD interleave package: %d = CPU socket %d, HA %i, shiftup: %i\n",
1269 idx
, *socket
, sad_ha
, shiftup
);
1271 /* Ivy Bridge's SAD mode doesn't support XOR interleave mode */
1272 idx
= (addr
>> 6) & 7;
1273 pkg
= sad_pkg(pvt
->info
.interleave_pkg
, reg
, idx
);
1274 *socket
= sad_pkg_socket(pkg
);
1275 sad_ha
= sad_pkg_ha(pkg
);
1276 edac_dbg(0, "SAD interleave package: %d = CPU socket %d, HA %d\n",
1277 idx
, *socket
, sad_ha
);
1281 * Move to the proper node structure, in order to access the
1282 * right PCI registers
1284 new_mci
= get_mci_for_node_id(*socket
);
1286 sprintf(msg
, "Struct for socket #%u wasn't initialized",
1291 pvt
= mci
->pvt_info
;
1294 * Step 2) Get memory channel
1297 if (pvt
->info
.type
== SANDY_BRIDGE
)
1298 pci_ha
= pvt
->pci_ha0
;
1301 pci_ha
= pvt
->pci_ha1
;
1303 pci_ha
= pvt
->pci_ha0
;
1305 for (n_tads
= 0; n_tads
< MAX_TAD
; n_tads
++) {
1306 pci_read_config_dword(pci_ha
, tad_dram_rule
[n_tads
], ®
);
1307 limit
= TAD_LIMIT(reg
);
1309 sprintf(msg
, "Can't discover the memory channel");
1316 if (n_tads
== MAX_TAD
) {
1317 sprintf(msg
, "Can't discover the memory channel");
1321 ch_way
= TAD_CH(reg
) + 1;
1322 sck_way
= TAD_SOCK(reg
);
1327 idx
= (addr
>> (6 + sck_way
+ shiftup
)) & 0x3;
1331 * FIXME: Shouldn't we use CHN_IDX_OFFSET() here, when ch_way == 3 ???
1335 base_ch
= TAD_TGT0(reg
);
1338 base_ch
= TAD_TGT1(reg
);
1341 base_ch
= TAD_TGT2(reg
);
1344 base_ch
= TAD_TGT3(reg
);
1347 sprintf(msg
, "Can't discover the TAD target");
1350 *channel_mask
= 1 << base_ch
;
1352 pci_read_config_dword(pvt
->pci_tad
[base_ch
],
1353 tad_ch_nilv_offset
[n_tads
],
1356 if (pvt
->is_mirrored
) {
1357 *channel_mask
|= 1 << ((base_ch
+ 2) % 4);
1361 sck_xch
= (1 << sck_way
) * (ch_way
>> 1);
1364 sprintf(msg
, "Invalid mirror set. Can't decode addr");
1368 sck_xch
= (1 << sck_way
) * ch_way
;
1370 if (pvt
->is_lockstep
)
1371 *channel_mask
|= 1 << ((base_ch
+ 1) % 4);
1373 offset
= TAD_OFFSET(tad_offset
);
1375 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",
1386 /* Calculate channel address */
1387 /* Remove the TAD offset */
1389 if (offset
> addr
) {
1390 sprintf(msg
, "Can't calculate ch addr: TAD offset 0x%08Lx is too high for addr 0x%08Lx!",
1395 ch_addr
= addr
- offset
;
1396 ch_addr
>>= (6 + shiftup
);
1398 ch_addr
<<= (6 + shiftup
);
1399 ch_addr
|= addr
& ((1 << (6 + shiftup
)) - 1);
1402 * Step 3) Decode rank
1404 for (n_rir
= 0; n_rir
< MAX_RIR_RANGES
; n_rir
++) {
1405 pci_read_config_dword(pvt
->pci_tad
[base_ch
],
1406 rir_way_limit
[n_rir
],
1409 if (!IS_RIR_VALID(reg
))
1412 limit
= pvt
->info
.rir_limit(reg
);
1413 gb
= div_u64_rem(limit
>> 20, 1024, &mb
);
1414 edac_dbg(0, "RIR#%d, limit: %u.%03u GB (0x%016Lx), way: %d\n",
1419 if (ch_addr
<= limit
)
1422 if (n_rir
== MAX_RIR_RANGES
) {
1423 sprintf(msg
, "Can't discover the memory rank for ch addr 0x%08Lx",
1427 rir_way
= RIR_WAY(reg
);
1429 if (pvt
->is_close_pg
)
1430 idx
= (ch_addr
>> 6);
1432 idx
= (ch_addr
>> 13); /* FIXME: Datasheet says to shift by 15 */
1433 idx
%= 1 << rir_way
;
1435 pci_read_config_dword(pvt
->pci_tad
[base_ch
],
1436 rir_offset
[n_rir
][idx
],
1438 *rank
= RIR_RNK_TGT(pvt
->info
.type
, reg
);
1440 edac_dbg(0, "RIR#%d: channel address 0x%08Lx < 0x%08Lx, RIR interleave %d, index %d\n",
1450 /****************************************************************************
1451 Device initialization routines: put/get, init/exit
1452 ****************************************************************************/
1455 * sbridge_put_all_devices 'put' all the devices that we have
1456 * reserved via 'get'
1458 static void sbridge_put_devices(struct sbridge_dev
*sbridge_dev
)
1463 for (i
= 0; i
< sbridge_dev
->n_devs
; i
++) {
1464 struct pci_dev
*pdev
= sbridge_dev
->pdev
[i
];
1467 edac_dbg(0, "Removing dev %02x:%02x.%d\n",
1469 PCI_SLOT(pdev
->devfn
), PCI_FUNC(pdev
->devfn
));
1474 static void sbridge_put_all_devices(void)
1476 struct sbridge_dev
*sbridge_dev
, *tmp
;
1478 list_for_each_entry_safe(sbridge_dev
, tmp
, &sbridge_edac_list
, list
) {
1479 sbridge_put_devices(sbridge_dev
);
1480 free_sbridge_dev(sbridge_dev
);
1484 static int sbridge_get_onedevice(struct pci_dev
**prev
,
1486 const struct pci_id_table
*table
,
1487 const unsigned devno
)
1489 struct sbridge_dev
*sbridge_dev
;
1490 const struct pci_id_descr
*dev_descr
= &table
->descr
[devno
];
1491 struct pci_dev
*pdev
= NULL
;
1494 sbridge_printk(KERN_DEBUG
,
1495 "Seeking for: PCI ID %04x:%04x\n",
1496 PCI_VENDOR_ID_INTEL
, dev_descr
->dev_id
);
1498 pdev
= pci_get_device(PCI_VENDOR_ID_INTEL
,
1499 dev_descr
->dev_id
, *prev
);
1507 if (dev_descr
->optional
)
1510 /* if the HA wasn't found */
1514 sbridge_printk(KERN_INFO
,
1515 "Device not found: %04x:%04x\n",
1516 PCI_VENDOR_ID_INTEL
, dev_descr
->dev_id
);
1518 /* End of list, leave */
1521 bus
= pdev
->bus
->number
;
1523 sbridge_dev
= get_sbridge_dev(bus
);
1525 sbridge_dev
= alloc_sbridge_dev(bus
, table
);
1533 if (sbridge_dev
->pdev
[devno
]) {
1534 sbridge_printk(KERN_ERR
,
1535 "Duplicated device for %04x:%04x\n",
1536 PCI_VENDOR_ID_INTEL
, dev_descr
->dev_id
);
1541 sbridge_dev
->pdev
[devno
] = pdev
;
1543 /* Be sure that the device is enabled */
1544 if (unlikely(pci_enable_device(pdev
) < 0)) {
1545 sbridge_printk(KERN_ERR
,
1546 "Couldn't enable %04x:%04x\n",
1547 PCI_VENDOR_ID_INTEL
, dev_descr
->dev_id
);
1551 edac_dbg(0, "Detected %04x:%04x\n",
1552 PCI_VENDOR_ID_INTEL
, dev_descr
->dev_id
);
1555 * As stated on drivers/pci/search.c, the reference count for
1556 * @from is always decremented if it is not %NULL. So, as we need
1557 * to get all devices up to null, we need to do a get for the device
1567 * sbridge_get_all_devices - Find and perform 'get' operation on the MCH's
1568 * devices we want to reference for this driver.
1569 * @num_mc: pointer to the memory controllers count, to be incremented in case
1571 * @table: model specific table
1573 * returns 0 in case of success or error code
1575 static int sbridge_get_all_devices(u8
*num_mc
,
1576 const struct pci_id_table
*table
)
1579 struct pci_dev
*pdev
= NULL
;
1581 while (table
&& table
->descr
) {
1582 for (i
= 0; i
< table
->n_devs
; i
++) {
1585 rc
= sbridge_get_onedevice(&pdev
, num_mc
,
1592 sbridge_put_all_devices();
1603 static int sbridge_mci_bind_devs(struct mem_ctl_info
*mci
,
1604 struct sbridge_dev
*sbridge_dev
)
1606 struct sbridge_pvt
*pvt
= mci
->pvt_info
;
1607 struct pci_dev
*pdev
;
1608 u8 saw_chan_mask
= 0;
1611 for (i
= 0; i
< sbridge_dev
->n_devs
; i
++) {
1612 pdev
= sbridge_dev
->pdev
[i
];
1616 switch (pdev
->device
) {
1617 case PCI_DEVICE_ID_INTEL_SBRIDGE_SAD0
:
1618 pvt
->pci_sad0
= pdev
;
1620 case PCI_DEVICE_ID_INTEL_SBRIDGE_SAD1
:
1621 pvt
->pci_sad1
= pdev
;
1623 case PCI_DEVICE_ID_INTEL_SBRIDGE_BR
:
1624 pvt
->pci_br0
= pdev
;
1626 case PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_HA0
:
1627 pvt
->pci_ha0
= pdev
;
1629 case PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TA
:
1632 case PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_RAS
:
1633 pvt
->pci_ras
= pdev
;
1635 case PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD0
:
1636 case PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD1
:
1637 case PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD2
:
1638 case PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD3
:
1640 int id
= pdev
->device
- PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD0
;
1641 pvt
->pci_tad
[id
] = pdev
;
1642 saw_chan_mask
|= 1 << id
;
1645 case PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_DDRIO
:
1646 pvt
->pci_ddrio
= pdev
;
1652 edac_dbg(0, "Associated PCI %02x:%02x, bus %d with dev = %p\n",
1653 pdev
->vendor
, pdev
->device
,
1658 /* Check if everything were registered */
1659 if (!pvt
->pci_sad0
|| !pvt
->pci_sad1
|| !pvt
->pci_ha0
||
1660 !pvt
-> pci_tad
|| !pvt
->pci_ras
|| !pvt
->pci_ta
)
1663 if (saw_chan_mask
!= 0x0f)
1668 sbridge_printk(KERN_ERR
, "Some needed devices are missing\n");
1672 sbridge_printk(KERN_ERR
, "Unexpected device %02x:%02x\n",
1673 PCI_VENDOR_ID_INTEL
, pdev
->device
);
1677 static int ibridge_mci_bind_devs(struct mem_ctl_info
*mci
,
1678 struct sbridge_dev
*sbridge_dev
)
1680 struct sbridge_pvt
*pvt
= mci
->pvt_info
;
1681 struct pci_dev
*pdev
, *tmp
;
1683 bool mode_2ha
= false;
1685 tmp
= pci_get_device(PCI_VENDOR_ID_INTEL
,
1686 PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1
, NULL
);
1692 for (i
= 0; i
< sbridge_dev
->n_devs
; i
++) {
1693 pdev
= sbridge_dev
->pdev
[i
];
1697 switch (pdev
->device
) {
1698 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0
:
1699 pvt
->pci_ha0
= pdev
;
1701 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TA
:
1703 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_RAS
:
1704 pvt
->pci_ras
= pdev
;
1706 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD2
:
1707 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD3
:
1708 /* if we have 2 HAs active, channels 2 and 3
1709 * are in other device */
1713 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD0
:
1714 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD1
:
1716 int id
= pdev
->device
- PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD0
;
1717 pvt
->pci_tad
[id
] = pdev
;
1720 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_2HA_DDRIO0
:
1721 pvt
->pci_ddrio
= pdev
;
1723 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_1HA_DDRIO0
:
1725 pvt
->pci_ddrio
= pdev
;
1727 case PCI_DEVICE_ID_INTEL_IBRIDGE_SAD
:
1728 pvt
->pci_sad0
= pdev
;
1730 case PCI_DEVICE_ID_INTEL_IBRIDGE_BR0
:
1731 pvt
->pci_br0
= pdev
;
1733 case PCI_DEVICE_ID_INTEL_IBRIDGE_BR1
:
1734 pvt
->pci_br1
= pdev
;
1736 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1
:
1737 pvt
->pci_ha1
= pdev
;
1739 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD0
:
1740 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD1
:
1742 int id
= pdev
->device
- PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD0
+ 2;
1744 /* we shouldn't have this device if we have just one
1747 pvt
->pci_tad
[id
] = pdev
;
1754 edac_dbg(0, "Associated PCI %02x.%02d.%d with dev = %p\n",
1756 PCI_SLOT(pdev
->devfn
), PCI_FUNC(pdev
->devfn
),
1760 /* Check if everything were registered */
1761 if (!pvt
->pci_sad0
|| !pvt
->pci_ha0
|| !pvt
->pci_br0
||
1762 !pvt
->pci_br1
|| !pvt
->pci_tad
|| !pvt
->pci_ras
||
1766 for (i
= 0; i
< NUM_CHANNELS
; i
++) {
1767 if (!pvt
->pci_tad
[i
])
1773 sbridge_printk(KERN_ERR
, "Some needed devices are missing\n");
1777 sbridge_printk(KERN_ERR
,
1778 "Unexpected device %02x:%02x\n", PCI_VENDOR_ID_INTEL
,
1783 static int haswell_mci_bind_devs(struct mem_ctl_info
*mci
,
1784 struct sbridge_dev
*sbridge_dev
)
1786 struct sbridge_pvt
*pvt
= mci
->pvt_info
;
1787 struct pci_dev
*pdev
, *tmp
;
1789 bool mode_2ha
= false;
1791 tmp
= pci_get_device(PCI_VENDOR_ID_INTEL
,
1792 PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1
, NULL
);
1798 /* there's only one device per system; not tied to any bus */
1799 if (pvt
->info
.pci_vtd
== NULL
)
1800 /* result will be checked later */
1801 pvt
->info
.pci_vtd
= pci_get_device(PCI_VENDOR_ID_INTEL
,
1802 PCI_DEVICE_ID_INTEL_HASWELL_IMC_VTD_MISC
,
1805 for (i
= 0; i
< sbridge_dev
->n_devs
; i
++) {
1806 pdev
= sbridge_dev
->pdev
[i
];
1810 switch (pdev
->device
) {
1811 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_CBO_SAD0
:
1812 pvt
->pci_sad0
= pdev
;
1814 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_CBO_SAD1
:
1815 pvt
->pci_sad1
= pdev
;
1817 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0
:
1818 pvt
->pci_ha0
= pdev
;
1820 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TA
:
1823 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_THERMAL
:
1824 pvt
->pci_ras
= pdev
;
1826 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD0
:
1827 pvt
->pci_tad
[0] = pdev
;
1829 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD1
:
1830 pvt
->pci_tad
[1] = pdev
;
1832 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD2
:
1834 pvt
->pci_tad
[2] = pdev
;
1836 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD3
:
1838 pvt
->pci_tad
[3] = pdev
;
1840 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO0
:
1841 pvt
->pci_ddrio
= pdev
;
1843 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1
:
1844 pvt
->pci_ha1
= pdev
;
1846 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TA
:
1847 pvt
->pci_ha1_ta
= pdev
;
1849 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD0
:
1851 pvt
->pci_tad
[2] = pdev
;
1853 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD1
:
1855 pvt
->pci_tad
[3] = pdev
;
1861 edac_dbg(0, "Associated PCI %02x.%02d.%d with dev = %p\n",
1863 PCI_SLOT(pdev
->devfn
), PCI_FUNC(pdev
->devfn
),
1867 /* Check if everything were registered */
1868 if (!pvt
->pci_sad0
|| !pvt
->pci_ha0
|| !pvt
->pci_sad1
||
1869 !pvt
->pci_ras
|| !pvt
->pci_ta
|| !pvt
->info
.pci_vtd
)
1872 for (i
= 0; i
< NUM_CHANNELS
; i
++) {
1873 if (!pvt
->pci_tad
[i
])
1879 sbridge_printk(KERN_ERR
, "Some needed devices are missing\n");
1883 static int broadwell_mci_bind_devs(struct mem_ctl_info
*mci
,
1884 struct sbridge_dev
*sbridge_dev
)
1886 struct sbridge_pvt
*pvt
= mci
->pvt_info
;
1887 struct pci_dev
*pdev
;
1890 /* there's only one device per system; not tied to any bus */
1891 if (pvt
->info
.pci_vtd
== NULL
)
1892 /* result will be checked later */
1893 pvt
->info
.pci_vtd
= pci_get_device(PCI_VENDOR_ID_INTEL
,
1894 PCI_DEVICE_ID_INTEL_BROADWELL_IMC_VTD_MISC
,
1897 for (i
= 0; i
< sbridge_dev
->n_devs
; i
++) {
1898 pdev
= sbridge_dev
->pdev
[i
];
1902 switch (pdev
->device
) {
1903 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_CBO_SAD0
:
1904 pvt
->pci_sad0
= pdev
;
1906 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_CBO_SAD1
:
1907 pvt
->pci_sad1
= pdev
;
1909 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0
:
1910 pvt
->pci_ha0
= pdev
;
1912 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TA
:
1915 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_THERMAL
:
1916 pvt
->pci_ras
= pdev
;
1918 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD0
:
1919 pvt
->pci_tad
[0] = pdev
;
1921 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD1
:
1922 pvt
->pci_tad
[1] = pdev
;
1924 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD2
:
1925 pvt
->pci_tad
[2] = pdev
;
1927 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD3
:
1928 pvt
->pci_tad
[3] = pdev
;
1930 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_DDRIO0
:
1931 pvt
->pci_ddrio
= pdev
;
1937 edac_dbg(0, "Associated PCI %02x.%02d.%d with dev = %p\n",
1939 PCI_SLOT(pdev
->devfn
), PCI_FUNC(pdev
->devfn
),
1943 /* Check if everything were registered */
1944 if (!pvt
->pci_sad0
|| !pvt
->pci_ha0
|| !pvt
->pci_sad1
||
1945 !pvt
->pci_ras
|| !pvt
->pci_ta
|| !pvt
->info
.pci_vtd
)
1948 for (i
= 0; i
< NUM_CHANNELS
; i
++) {
1949 if (!pvt
->pci_tad
[i
])
1955 sbridge_printk(KERN_ERR
, "Some needed devices are missing\n");
1959 /****************************************************************************
1960 Error check routines
1961 ****************************************************************************/
1964 * While Sandy Bridge has error count registers, SMI BIOS read values from
1965 * and resets the counters. So, they are not reliable for the OS to read
1966 * from them. So, we have no option but to just trust on whatever MCE is
1967 * telling us about the errors.
1969 static void sbridge_mce_output_error(struct mem_ctl_info
*mci
,
1970 const struct mce
*m
)
1972 struct mem_ctl_info
*new_mci
;
1973 struct sbridge_pvt
*pvt
= mci
->pvt_info
;
1974 enum hw_event_mc_err_type tp_event
;
1975 char *type
, *optype
, msg
[256];
1976 bool ripv
= GET_BITFIELD(m
->mcgstatus
, 0, 0);
1977 bool overflow
= GET_BITFIELD(m
->status
, 62, 62);
1978 bool uncorrected_error
= GET_BITFIELD(m
->status
, 61, 61);
1980 u32 core_err_cnt
= GET_BITFIELD(m
->status
, 38, 52);
1981 u32 mscod
= GET_BITFIELD(m
->status
, 16, 31);
1982 u32 errcode
= GET_BITFIELD(m
->status
, 0, 15);
1983 u32 channel
= GET_BITFIELD(m
->status
, 0, 3);
1984 u32 optypenum
= GET_BITFIELD(m
->status
, 4, 6);
1985 long channel_mask
, first_channel
;
1988 char *area_type
= NULL
;
1990 if (pvt
->info
.type
== IVY_BRIDGE
)
1993 recoverable
= GET_BITFIELD(m
->status
, 56, 56);
1995 if (uncorrected_error
) {
1998 tp_event
= HW_EVENT_ERR_FATAL
;
2001 tp_event
= HW_EVENT_ERR_UNCORRECTED
;
2005 tp_event
= HW_EVENT_ERR_CORRECTED
;
2009 * According with Table 15-9 of the Intel Architecture spec vol 3A,
2010 * memory errors should fit in this mask:
2011 * 000f 0000 1mmm cccc (binary)
2013 * f = Correction Report Filtering Bit. If 1, subsequent errors
2017 * If the mask doesn't match, report an error to the parsing logic
2019 if (! ((errcode
& 0xef80) == 0x80)) {
2020 optype
= "Can't parse: it is not a mem";
2022 switch (optypenum
) {
2024 optype
= "generic undef request error";
2027 optype
= "memory read error";
2030 optype
= "memory write error";
2033 optype
= "addr/cmd error";
2036 optype
= "memory scrubbing error";
2039 optype
= "reserved";
2044 /* Only decode errors with an valid address (ADDRV) */
2045 if (!GET_BITFIELD(m
->status
, 58, 58))
2048 rc
= get_memory_error_data(mci
, m
->addr
, &socket
,
2049 &channel_mask
, &rank
, &area_type
, msg
);
2052 new_mci
= get_mci_for_node_id(socket
);
2054 strcpy(msg
, "Error: socket got corrupted!");
2058 pvt
= mci
->pvt_info
;
2060 first_channel
= find_first_bit(&channel_mask
, NUM_CHANNELS
);
2071 * FIXME: On some memory configurations (mirror, lockstep), the
2072 * Memory Controller can't point the error to a single DIMM. The
2073 * EDAC core should be handling the channel mask, in order to point
2074 * to the group of dimm's where the error may be happening.
2076 if (!pvt
->is_lockstep
&& !pvt
->is_mirrored
&& !pvt
->is_close_pg
)
2077 channel
= first_channel
;
2079 snprintf(msg
, sizeof(msg
),
2080 "%s%s area:%s err_code:%04x:%04x socket:%d channel_mask:%ld rank:%d",
2081 overflow
? " OVERFLOW" : "",
2082 (uncorrected_error
&& recoverable
) ? " recoverable" : "",
2089 edac_dbg(0, "%s\n", msg
);
2091 /* FIXME: need support for channel mask */
2093 if (channel
== CHANNEL_UNSPECIFIED
)
2096 /* Call the helper to output message */
2097 edac_mc_handle_error(tp_event
, mci
, core_err_cnt
,
2098 m
->addr
>> PAGE_SHIFT
, m
->addr
& ~PAGE_MASK
, 0,
2103 edac_mc_handle_error(tp_event
, mci
, core_err_cnt
, 0, 0, 0,
2110 * sbridge_check_error Retrieve and process errors reported by the
2111 * hardware. Called by the Core module.
2113 static void sbridge_check_error(struct mem_ctl_info
*mci
)
2115 struct sbridge_pvt
*pvt
= mci
->pvt_info
;
2121 * MCE first step: Copy all mce errors into a temporary buffer
2122 * We use a double buffering here, to reduce the risk of
2126 count
= (pvt
->mce_out
+ MCE_LOG_LEN
- pvt
->mce_in
)
2131 m
= pvt
->mce_outentry
;
2132 if (pvt
->mce_in
+ count
> MCE_LOG_LEN
) {
2133 unsigned l
= MCE_LOG_LEN
- pvt
->mce_in
;
2135 memcpy(m
, &pvt
->mce_entry
[pvt
->mce_in
], sizeof(*m
) * l
);
2141 memcpy(m
, &pvt
->mce_entry
[pvt
->mce_in
], sizeof(*m
) * count
);
2143 pvt
->mce_in
+= count
;
2146 if (pvt
->mce_overrun
) {
2147 sbridge_printk(KERN_ERR
, "Lost %d memory errors\n",
2150 pvt
->mce_overrun
= 0;
2154 * MCE second step: parse errors and display
2156 for (i
= 0; i
< count
; i
++)
2157 sbridge_mce_output_error(mci
, &pvt
->mce_outentry
[i
]);
2161 * sbridge_mce_check_error Replicates mcelog routine to get errors
2162 * This routine simply queues mcelog errors, and
2163 * return. The error itself should be handled later
2164 * by sbridge_check_error.
2165 * WARNING: As this routine should be called at NMI time, extra care should
2166 * be taken to avoid deadlocks, and to be as fast as possible.
2168 static int sbridge_mce_check_error(struct notifier_block
*nb
, unsigned long val
,
2171 struct mce
*mce
= (struct mce
*)data
;
2172 struct mem_ctl_info
*mci
;
2173 struct sbridge_pvt
*pvt
;
2176 if (get_edac_report_status() == EDAC_REPORTING_DISABLED
)
2179 mci
= get_mci_for_node_id(mce
->socketid
);
2182 pvt
= mci
->pvt_info
;
2185 * Just let mcelog handle it if the error is
2186 * outside the memory controller. A memory error
2187 * is indicated by bit 7 = 1 and bits = 8-11,13-15 = 0.
2188 * bit 12 has an special meaning.
2190 if ((mce
->status
& 0xefff) >> 7 != 1)
2193 if (mce
->mcgstatus
& MCG_STATUS_MCIP
)
2198 sbridge_mc_printk(mci
, KERN_DEBUG
, "HANDLING MCE MEMORY ERROR\n");
2200 sbridge_mc_printk(mci
, KERN_DEBUG
, "CPU %d: Machine Check %s: %Lx "
2201 "Bank %d: %016Lx\n", mce
->extcpu
, type
,
2202 mce
->mcgstatus
, mce
->bank
, mce
->status
);
2203 sbridge_mc_printk(mci
, KERN_DEBUG
, "TSC %llx ", mce
->tsc
);
2204 sbridge_mc_printk(mci
, KERN_DEBUG
, "ADDR %llx ", mce
->addr
);
2205 sbridge_mc_printk(mci
, KERN_DEBUG
, "MISC %llx ", mce
->misc
);
2207 sbridge_mc_printk(mci
, KERN_DEBUG
, "PROCESSOR %u:%x TIME %llu SOCKET "
2208 "%u APIC %x\n", mce
->cpuvendor
, mce
->cpuid
,
2209 mce
->time
, mce
->socketid
, mce
->apicid
);
2212 if ((pvt
->mce_out
+ 1) % MCE_LOG_LEN
== pvt
->mce_in
) {
2218 /* Copy memory error at the ringbuffer */
2219 memcpy(&pvt
->mce_entry
[pvt
->mce_out
], mce
, sizeof(*mce
));
2221 pvt
->mce_out
= (pvt
->mce_out
+ 1) % MCE_LOG_LEN
;
2223 /* Handle fatal errors immediately */
2224 if (mce
->mcgstatus
& 1)
2225 sbridge_check_error(mci
);
2227 /* Advice mcelog that the error were handled */
2231 static struct notifier_block sbridge_mce_dec
= {
2232 .notifier_call
= sbridge_mce_check_error
,
2235 /****************************************************************************
2236 EDAC register/unregister logic
2237 ****************************************************************************/
2239 static void sbridge_unregister_mci(struct sbridge_dev
*sbridge_dev
)
2241 struct mem_ctl_info
*mci
= sbridge_dev
->mci
;
2242 struct sbridge_pvt
*pvt
;
2244 if (unlikely(!mci
|| !mci
->pvt_info
)) {
2245 edac_dbg(0, "MC: dev = %p\n", &sbridge_dev
->pdev
[0]->dev
);
2247 sbridge_printk(KERN_ERR
, "Couldn't find mci handler\n");
2251 pvt
= mci
->pvt_info
;
2253 edac_dbg(0, "MC: mci = %p, dev = %p\n",
2254 mci
, &sbridge_dev
->pdev
[0]->dev
);
2256 /* Remove MC sysfs nodes */
2257 edac_mc_del_mc(mci
->pdev
);
2259 edac_dbg(1, "%s: free mci struct\n", mci
->ctl_name
);
2260 kfree(mci
->ctl_name
);
2262 sbridge_dev
->mci
= NULL
;
2265 static int sbridge_register_mci(struct sbridge_dev
*sbridge_dev
, enum type type
)
2267 struct mem_ctl_info
*mci
;
2268 struct edac_mc_layer layers
[2];
2269 struct sbridge_pvt
*pvt
;
2270 struct pci_dev
*pdev
= sbridge_dev
->pdev
[0];
2273 /* Check the number of active and not disabled channels */
2274 rc
= check_if_ecc_is_active(sbridge_dev
->bus
, type
);
2275 if (unlikely(rc
< 0))
2278 /* allocate a new MC control structure */
2279 layers
[0].type
= EDAC_MC_LAYER_CHANNEL
;
2280 layers
[0].size
= NUM_CHANNELS
;
2281 layers
[0].is_virt_csrow
= false;
2282 layers
[1].type
= EDAC_MC_LAYER_SLOT
;
2283 layers
[1].size
= MAX_DIMMS
;
2284 layers
[1].is_virt_csrow
= true;
2285 mci
= edac_mc_alloc(sbridge_dev
->mc
, ARRAY_SIZE(layers
), layers
,
2291 edac_dbg(0, "MC: mci = %p, dev = %p\n",
2294 pvt
= mci
->pvt_info
;
2295 memset(pvt
, 0, sizeof(*pvt
));
2297 /* Associate sbridge_dev and mci for future usage */
2298 pvt
->sbridge_dev
= sbridge_dev
;
2299 sbridge_dev
->mci
= mci
;
2301 mci
->mtype_cap
= MEM_FLAG_DDR3
;
2302 mci
->edac_ctl_cap
= EDAC_FLAG_NONE
;
2303 mci
->edac_cap
= EDAC_FLAG_NONE
;
2304 mci
->mod_name
= "sbridge_edac.c";
2305 mci
->mod_ver
= SBRIDGE_REVISION
;
2306 mci
->dev_name
= pci_name(pdev
);
2307 mci
->ctl_page_to_phys
= NULL
;
2309 /* Set the function pointer to an actual operation function */
2310 mci
->edac_check
= sbridge_check_error
;
2312 pvt
->info
.type
= type
;
2315 pvt
->info
.rankcfgr
= IB_RANK_CFG_A
;
2316 pvt
->info
.get_tolm
= ibridge_get_tolm
;
2317 pvt
->info
.get_tohm
= ibridge_get_tohm
;
2318 pvt
->info
.dram_rule
= ibridge_dram_rule
;
2319 pvt
->info
.get_memory_type
= get_memory_type
;
2320 pvt
->info
.get_node_id
= get_node_id
;
2321 pvt
->info
.rir_limit
= rir_limit
;
2322 pvt
->info
.max_sad
= ARRAY_SIZE(ibridge_dram_rule
);
2323 pvt
->info
.interleave_list
= ibridge_interleave_list
;
2324 pvt
->info
.max_interleave
= ARRAY_SIZE(ibridge_interleave_list
);
2325 pvt
->info
.interleave_pkg
= ibridge_interleave_pkg
;
2326 mci
->ctl_name
= kasprintf(GFP_KERNEL
, "Ivy Bridge Socket#%d", mci
->mc_idx
);
2328 /* Store pci devices at mci for faster access */
2329 rc
= ibridge_mci_bind_devs(mci
, sbridge_dev
);
2330 if (unlikely(rc
< 0))
2334 pvt
->info
.rankcfgr
= SB_RANK_CFG_A
;
2335 pvt
->info
.get_tolm
= sbridge_get_tolm
;
2336 pvt
->info
.get_tohm
= sbridge_get_tohm
;
2337 pvt
->info
.dram_rule
= sbridge_dram_rule
;
2338 pvt
->info
.get_memory_type
= get_memory_type
;
2339 pvt
->info
.get_node_id
= get_node_id
;
2340 pvt
->info
.rir_limit
= rir_limit
;
2341 pvt
->info
.max_sad
= ARRAY_SIZE(sbridge_dram_rule
);
2342 pvt
->info
.interleave_list
= sbridge_interleave_list
;
2343 pvt
->info
.max_interleave
= ARRAY_SIZE(sbridge_interleave_list
);
2344 pvt
->info
.interleave_pkg
= sbridge_interleave_pkg
;
2345 mci
->ctl_name
= kasprintf(GFP_KERNEL
, "Sandy Bridge Socket#%d", mci
->mc_idx
);
2347 /* Store pci devices at mci for faster access */
2348 rc
= sbridge_mci_bind_devs(mci
, sbridge_dev
);
2349 if (unlikely(rc
< 0))
2353 /* rankcfgr isn't used */
2354 pvt
->info
.get_tolm
= haswell_get_tolm
;
2355 pvt
->info
.get_tohm
= haswell_get_tohm
;
2356 pvt
->info
.dram_rule
= ibridge_dram_rule
;
2357 pvt
->info
.get_memory_type
= haswell_get_memory_type
;
2358 pvt
->info
.get_node_id
= haswell_get_node_id
;
2359 pvt
->info
.rir_limit
= haswell_rir_limit
;
2360 pvt
->info
.max_sad
= ARRAY_SIZE(ibridge_dram_rule
);
2361 pvt
->info
.interleave_list
= ibridge_interleave_list
;
2362 pvt
->info
.max_interleave
= ARRAY_SIZE(ibridge_interleave_list
);
2363 pvt
->info
.interleave_pkg
= ibridge_interleave_pkg
;
2364 mci
->ctl_name
= kasprintf(GFP_KERNEL
, "Haswell Socket#%d", mci
->mc_idx
);
2366 /* Store pci devices at mci for faster access */
2367 rc
= haswell_mci_bind_devs(mci
, sbridge_dev
);
2368 if (unlikely(rc
< 0))
2372 /* rankcfgr isn't used */
2373 pvt
->info
.get_tolm
= haswell_get_tolm
;
2374 pvt
->info
.get_tohm
= haswell_get_tohm
;
2375 pvt
->info
.dram_rule
= ibridge_dram_rule
;
2376 pvt
->info
.get_memory_type
= haswell_get_memory_type
;
2377 pvt
->info
.get_node_id
= haswell_get_node_id
;
2378 pvt
->info
.rir_limit
= haswell_rir_limit
;
2379 pvt
->info
.max_sad
= ARRAY_SIZE(ibridge_dram_rule
);
2380 pvt
->info
.interleave_list
= ibridge_interleave_list
;
2381 pvt
->info
.max_interleave
= ARRAY_SIZE(ibridge_interleave_list
);
2382 pvt
->info
.interleave_pkg
= ibridge_interleave_pkg
;
2383 mci
->ctl_name
= kasprintf(GFP_KERNEL
, "Broadwell Socket#%d", mci
->mc_idx
);
2385 /* Store pci devices at mci for faster access */
2386 rc
= broadwell_mci_bind_devs(mci
, sbridge_dev
);
2387 if (unlikely(rc
< 0))
2392 /* Get dimm basic config and the memory layout */
2393 get_dimm_config(mci
);
2394 get_memory_layout(mci
);
2396 /* record ptr to the generic device */
2397 mci
->pdev
= &pdev
->dev
;
2399 /* add this new MC control structure to EDAC's list of MCs */
2400 if (unlikely(edac_mc_add_mc(mci
))) {
2401 edac_dbg(0, "MC: failed edac_mc_add_mc()\n");
2409 kfree(mci
->ctl_name
);
2411 sbridge_dev
->mci
= NULL
;
2416 * sbridge_probe Probe for ONE instance of device to see if it is
2419 * 0 for FOUND a device
2420 * < 0 for error code
2423 static int sbridge_probe(struct pci_dev
*pdev
, const struct pci_device_id
*id
)
2427 struct sbridge_dev
*sbridge_dev
;
2428 enum type type
= SANDY_BRIDGE
;
2430 /* get the pci devices we want to reserve for our use */
2431 mutex_lock(&sbridge_edac_lock
);
2434 * All memory controllers are allocated at the first pass.
2436 if (unlikely(probed
>= 1)) {
2437 mutex_unlock(&sbridge_edac_lock
);
2442 switch (pdev
->device
) {
2443 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TA
:
2444 rc
= sbridge_get_all_devices(&num_mc
, pci_dev_descr_ibridge_table
);
2447 case PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_HA0
:
2448 rc
= sbridge_get_all_devices(&num_mc
, pci_dev_descr_sbridge_table
);
2449 type
= SANDY_BRIDGE
;
2451 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0
:
2452 rc
= sbridge_get_all_devices(&num_mc
, pci_dev_descr_haswell_table
);
2455 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0
:
2456 rc
= sbridge_get_all_devices(&num_mc
, pci_dev_descr_broadwell_table
);
2460 if (unlikely(rc
< 0)) {
2461 edac_dbg(0, "couldn't get all devices for 0x%x\n", pdev
->device
);
2467 list_for_each_entry(sbridge_dev
, &sbridge_edac_list
, list
) {
2468 edac_dbg(0, "Registering MC#%d (%d of %d)\n",
2469 mc
, mc
+ 1, num_mc
);
2471 sbridge_dev
->mc
= mc
++;
2472 rc
= sbridge_register_mci(sbridge_dev
, type
);
2473 if (unlikely(rc
< 0))
2477 sbridge_printk(KERN_INFO
, "%s\n", SBRIDGE_REVISION
);
2479 mutex_unlock(&sbridge_edac_lock
);
2483 list_for_each_entry(sbridge_dev
, &sbridge_edac_list
, list
)
2484 sbridge_unregister_mci(sbridge_dev
);
2486 sbridge_put_all_devices();
2488 mutex_unlock(&sbridge_edac_lock
);
2493 * sbridge_remove destructor for one instance of device
2496 static void sbridge_remove(struct pci_dev
*pdev
)
2498 struct sbridge_dev
*sbridge_dev
;
2503 * we have a trouble here: pdev value for removal will be wrong, since
2504 * it will point to the X58 register used to detect that the machine
2505 * is a Nehalem or upper design. However, due to the way several PCI
2506 * devices are grouped together to provide MC functionality, we need
2507 * to use a different method for releasing the devices
2510 mutex_lock(&sbridge_edac_lock
);
2512 if (unlikely(!probed
)) {
2513 mutex_unlock(&sbridge_edac_lock
);
2517 list_for_each_entry(sbridge_dev
, &sbridge_edac_list
, list
)
2518 sbridge_unregister_mci(sbridge_dev
);
2520 /* Release PCI resources */
2521 sbridge_put_all_devices();
2525 mutex_unlock(&sbridge_edac_lock
);
2528 MODULE_DEVICE_TABLE(pci
, sbridge_pci_tbl
);
2531 * sbridge_driver pci_driver structure for this module
2534 static struct pci_driver sbridge_driver
= {
2535 .name
= "sbridge_edac",
2536 .probe
= sbridge_probe
,
2537 .remove
= sbridge_remove
,
2538 .id_table
= sbridge_pci_tbl
,
2542 * sbridge_init Module entry function
2543 * Try to initialize this module for its devices
2545 static int __init
sbridge_init(void)
2551 /* Ensure that the OPSTATE is set correctly for POLL or NMI */
2554 pci_rc
= pci_register_driver(&sbridge_driver
);
2556 mce_register_decode_chain(&sbridge_mce_dec
);
2557 if (get_edac_report_status() == EDAC_REPORTING_DISABLED
)
2558 sbridge_printk(KERN_WARNING
, "Loading driver, error reporting disabled.\n");
2562 sbridge_printk(KERN_ERR
, "Failed to register device with error %d.\n",
2569 * sbridge_exit() Module exit function
2570 * Unregister the driver
2572 static void __exit
sbridge_exit(void)
2575 pci_unregister_driver(&sbridge_driver
);
2576 mce_unregister_decode_chain(&sbridge_mce_dec
);
2579 module_init(sbridge_init
);
2580 module_exit(sbridge_exit
);
2582 module_param(edac_op_state
, int, 0444);
2583 MODULE_PARM_DESC(edac_op_state
, "EDAC Error Reporting state: 0=Poll,1=NMI");
2585 MODULE_LICENSE("GPL");
2586 MODULE_AUTHOR("Mauro Carvalho Chehab");
2587 MODULE_AUTHOR("Red Hat Inc. (http://www.redhat.com)");
2588 MODULE_DESCRIPTION("MC Driver for Intel Sandy Bridge and Ivy Bridge memory controllers - "