usb: xhci-plat: properly handle probe deferral for devm_clk_get()
[linux/fpc-iii.git] / drivers / edac / sb_edac.c
blob8bf745d2da7e1750571860d58be00d053579f301
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>
25 #include <asm/mce.h>
27 #include "edac_core.h"
29 /* Static vars */
30 static LIST_HEAD(sbridge_edac_list);
31 static DEFINE_MUTEX(sbridge_edac_lock);
32 static int probed;
35 * Alter this version for the module when modifications are made
37 #define SBRIDGE_REVISION " Ver: 1.1.1 "
38 #define EDAC_MOD_STR "sbridge_edac"
41 * Debug macros
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 static const u32 knl_dram_rule[] = {
69 0x60, 0x68, 0x70, 0x78, 0x80, /* 0-4 */
70 0x88, 0x90, 0x98, 0xa0, 0xa8, /* 5-9 */
71 0xb0, 0xb8, 0xc0, 0xc8, 0xd0, /* 10-14 */
72 0xd8, 0xe0, 0xe8, 0xf0, 0xf8, /* 15-19 */
73 0x100, 0x108, 0x110, 0x118, /* 20-23 */
76 #define DRAM_RULE_ENABLE(reg) GET_BITFIELD(reg, 0, 0)
77 #define A7MODE(reg) GET_BITFIELD(reg, 26, 26)
79 static char *show_dram_attr(u32 attr)
81 switch (attr) {
82 case 0:
83 return "DRAM";
84 case 1:
85 return "MMCFG";
86 case 2:
87 return "NXM";
88 default:
89 return "unknown";
93 static const u32 sbridge_interleave_list[] = {
94 0x84, 0x8c, 0x94, 0x9c, 0xa4,
95 0xac, 0xb4, 0xbc, 0xc4, 0xcc,
98 static const u32 ibridge_interleave_list[] = {
99 0x64, 0x6c, 0x74, 0x7c, 0x84,
100 0x8c, 0x94, 0x9c, 0xa4, 0xac,
101 0xb4, 0xbc, 0xc4, 0xcc, 0xd4,
102 0xdc, 0xe4, 0xec, 0xf4, 0xfc,
105 static const u32 knl_interleave_list[] = {
106 0x64, 0x6c, 0x74, 0x7c, 0x84, /* 0-4 */
107 0x8c, 0x94, 0x9c, 0xa4, 0xac, /* 5-9 */
108 0xb4, 0xbc, 0xc4, 0xcc, 0xd4, /* 10-14 */
109 0xdc, 0xe4, 0xec, 0xf4, 0xfc, /* 15-19 */
110 0x104, 0x10c, 0x114, 0x11c, /* 20-23 */
113 struct interleave_pkg {
114 unsigned char start;
115 unsigned char end;
118 static const struct interleave_pkg sbridge_interleave_pkg[] = {
119 { 0, 2 },
120 { 3, 5 },
121 { 8, 10 },
122 { 11, 13 },
123 { 16, 18 },
124 { 19, 21 },
125 { 24, 26 },
126 { 27, 29 },
129 static const struct interleave_pkg ibridge_interleave_pkg[] = {
130 { 0, 3 },
131 { 4, 7 },
132 { 8, 11 },
133 { 12, 15 },
134 { 16, 19 },
135 { 20, 23 },
136 { 24, 27 },
137 { 28, 31 },
140 static inline int sad_pkg(const struct interleave_pkg *table, u32 reg,
141 int interleave)
143 return GET_BITFIELD(reg, table[interleave].start,
144 table[interleave].end);
147 /* Devices 12 Function 7 */
149 #define TOLM 0x80
150 #define TOHM 0x84
151 #define HASWELL_TOLM 0xd0
152 #define HASWELL_TOHM_0 0xd4
153 #define HASWELL_TOHM_1 0xd8
154 #define KNL_TOLM 0xd0
155 #define KNL_TOHM_0 0xd4
156 #define KNL_TOHM_1 0xd8
158 #define GET_TOLM(reg) ((GET_BITFIELD(reg, 0, 3) << 28) | 0x3ffffff)
159 #define GET_TOHM(reg) ((GET_BITFIELD(reg, 0, 20) << 25) | 0x3ffffff)
161 /* Device 13 Function 6 */
163 #define SAD_TARGET 0xf0
165 #define SOURCE_ID(reg) GET_BITFIELD(reg, 9, 11)
167 #define SOURCE_ID_KNL(reg) GET_BITFIELD(reg, 12, 14)
169 #define SAD_CONTROL 0xf4
171 /* Device 14 function 0 */
173 static const u32 tad_dram_rule[] = {
174 0x40, 0x44, 0x48, 0x4c,
175 0x50, 0x54, 0x58, 0x5c,
176 0x60, 0x64, 0x68, 0x6c,
178 #define MAX_TAD ARRAY_SIZE(tad_dram_rule)
180 #define TAD_LIMIT(reg) ((GET_BITFIELD(reg, 12, 31) << 26) | 0x3ffffff)
181 #define TAD_SOCK(reg) GET_BITFIELD(reg, 10, 11)
182 #define TAD_CH(reg) GET_BITFIELD(reg, 8, 9)
183 #define TAD_TGT3(reg) GET_BITFIELD(reg, 6, 7)
184 #define TAD_TGT2(reg) GET_BITFIELD(reg, 4, 5)
185 #define TAD_TGT1(reg) GET_BITFIELD(reg, 2, 3)
186 #define TAD_TGT0(reg) GET_BITFIELD(reg, 0, 1)
188 /* Device 15, function 0 */
190 #define MCMTR 0x7c
191 #define KNL_MCMTR 0x624
193 #define IS_ECC_ENABLED(mcmtr) GET_BITFIELD(mcmtr, 2, 2)
194 #define IS_LOCKSTEP_ENABLED(mcmtr) GET_BITFIELD(mcmtr, 1, 1)
195 #define IS_CLOSE_PG(mcmtr) GET_BITFIELD(mcmtr, 0, 0)
197 /* Device 15, function 1 */
199 #define RASENABLES 0xac
200 #define IS_MIRROR_ENABLED(reg) GET_BITFIELD(reg, 0, 0)
202 /* Device 15, functions 2-5 */
204 static const int mtr_regs[] = {
205 0x80, 0x84, 0x88,
208 static const int knl_mtr_reg = 0xb60;
210 #define RANK_DISABLE(mtr) GET_BITFIELD(mtr, 16, 19)
211 #define IS_DIMM_PRESENT(mtr) GET_BITFIELD(mtr, 14, 14)
212 #define RANK_CNT_BITS(mtr) GET_BITFIELD(mtr, 12, 13)
213 #define RANK_WIDTH_BITS(mtr) GET_BITFIELD(mtr, 2, 4)
214 #define COL_WIDTH_BITS(mtr) GET_BITFIELD(mtr, 0, 1)
216 static const u32 tad_ch_nilv_offset[] = {
217 0x90, 0x94, 0x98, 0x9c,
218 0xa0, 0xa4, 0xa8, 0xac,
219 0xb0, 0xb4, 0xb8, 0xbc,
221 #define CHN_IDX_OFFSET(reg) GET_BITFIELD(reg, 28, 29)
222 #define TAD_OFFSET(reg) (GET_BITFIELD(reg, 6, 25) << 26)
224 static const u32 rir_way_limit[] = {
225 0x108, 0x10c, 0x110, 0x114, 0x118,
227 #define MAX_RIR_RANGES ARRAY_SIZE(rir_way_limit)
229 #define IS_RIR_VALID(reg) GET_BITFIELD(reg, 31, 31)
230 #define RIR_WAY(reg) GET_BITFIELD(reg, 28, 29)
232 #define MAX_RIR_WAY 8
234 static const u32 rir_offset[MAX_RIR_RANGES][MAX_RIR_WAY] = {
235 { 0x120, 0x124, 0x128, 0x12c, 0x130, 0x134, 0x138, 0x13c },
236 { 0x140, 0x144, 0x148, 0x14c, 0x150, 0x154, 0x158, 0x15c },
237 { 0x160, 0x164, 0x168, 0x16c, 0x170, 0x174, 0x178, 0x17c },
238 { 0x180, 0x184, 0x188, 0x18c, 0x190, 0x194, 0x198, 0x19c },
239 { 0x1a0, 0x1a4, 0x1a8, 0x1ac, 0x1b0, 0x1b4, 0x1b8, 0x1bc },
242 #define RIR_RNK_TGT(reg) GET_BITFIELD(reg, 16, 19)
243 #define RIR_OFFSET(reg) GET_BITFIELD(reg, 2, 14)
245 /* Device 16, functions 2-7 */
248 * FIXME: Implement the error count reads directly
251 static const u32 correrrcnt[] = {
252 0x104, 0x108, 0x10c, 0x110,
255 #define RANK_ODD_OV(reg) GET_BITFIELD(reg, 31, 31)
256 #define RANK_ODD_ERR_CNT(reg) GET_BITFIELD(reg, 16, 30)
257 #define RANK_EVEN_OV(reg) GET_BITFIELD(reg, 15, 15)
258 #define RANK_EVEN_ERR_CNT(reg) GET_BITFIELD(reg, 0, 14)
260 static const u32 correrrthrsld[] = {
261 0x11c, 0x120, 0x124, 0x128,
264 #define RANK_ODD_ERR_THRSLD(reg) GET_BITFIELD(reg, 16, 30)
265 #define RANK_EVEN_ERR_THRSLD(reg) GET_BITFIELD(reg, 0, 14)
268 /* Device 17, function 0 */
270 #define SB_RANK_CFG_A 0x0328
272 #define IB_RANK_CFG_A 0x0320
275 * sbridge structs
278 #define NUM_CHANNELS 8 /* 2MC per socket, four chan per MC */
279 #define MAX_DIMMS 3 /* Max DIMMS per channel */
280 #define KNL_MAX_CHAS 38 /* KNL max num. of Cache Home Agents */
281 #define KNL_MAX_CHANNELS 6 /* KNL max num. of PCI channels */
282 #define KNL_MAX_EDCS 8 /* Embedded DRAM controllers */
283 #define CHANNEL_UNSPECIFIED 0xf /* Intel IA32 SDM 15-14 */
285 enum type {
286 SANDY_BRIDGE,
287 IVY_BRIDGE,
288 HASWELL,
289 BROADWELL,
290 KNIGHTS_LANDING,
293 struct sbridge_pvt;
294 struct sbridge_info {
295 enum type type;
296 u32 mcmtr;
297 u32 rankcfgr;
298 u64 (*get_tolm)(struct sbridge_pvt *pvt);
299 u64 (*get_tohm)(struct sbridge_pvt *pvt);
300 u64 (*rir_limit)(u32 reg);
301 u64 (*sad_limit)(u32 reg);
302 u32 (*interleave_mode)(u32 reg);
303 char* (*show_interleave_mode)(u32 reg);
304 u32 (*dram_attr)(u32 reg);
305 const u32 *dram_rule;
306 const u32 *interleave_list;
307 const struct interleave_pkg *interleave_pkg;
308 u8 max_sad;
309 u8 max_interleave;
310 u8 (*get_node_id)(struct sbridge_pvt *pvt);
311 enum mem_type (*get_memory_type)(struct sbridge_pvt *pvt);
312 enum dev_type (*get_width)(struct sbridge_pvt *pvt, u32 mtr);
313 struct pci_dev *pci_vtd;
316 struct sbridge_channel {
317 u32 ranks;
318 u32 dimms;
321 struct pci_id_descr {
322 int dev_id;
323 int optional;
326 struct pci_id_table {
327 const struct pci_id_descr *descr;
328 int n_devs;
331 struct sbridge_dev {
332 struct list_head list;
333 u8 bus, mc;
334 u8 node_id, source_id;
335 struct pci_dev **pdev;
336 int n_devs;
337 struct mem_ctl_info *mci;
340 struct knl_pvt {
341 struct pci_dev *pci_cha[KNL_MAX_CHAS];
342 struct pci_dev *pci_channel[KNL_MAX_CHANNELS];
343 struct pci_dev *pci_mc0;
344 struct pci_dev *pci_mc1;
345 struct pci_dev *pci_mc0_misc;
346 struct pci_dev *pci_mc1_misc;
347 struct pci_dev *pci_mc_info; /* tolm, tohm */
350 struct sbridge_pvt {
351 struct pci_dev *pci_ta, *pci_ddrio, *pci_ras;
352 struct pci_dev *pci_sad0, *pci_sad1;
353 struct pci_dev *pci_ha0, *pci_ha1;
354 struct pci_dev *pci_br0, *pci_br1;
355 struct pci_dev *pci_ha1_ta;
356 struct pci_dev *pci_tad[NUM_CHANNELS];
358 struct sbridge_dev *sbridge_dev;
360 struct sbridge_info info;
361 struct sbridge_channel channel[NUM_CHANNELS];
363 /* Memory type detection */
364 bool is_mirrored, is_lockstep, is_close_pg;
365 bool is_chan_hash;
367 /* Fifo double buffers */
368 struct mce mce_entry[MCE_LOG_LEN];
369 struct mce mce_outentry[MCE_LOG_LEN];
371 /* Fifo in/out counters */
372 unsigned mce_in, mce_out;
374 /* Count indicator to show errors not got */
375 unsigned mce_overrun;
377 /* Memory description */
378 u64 tolm, tohm;
379 struct knl_pvt knl;
382 #define PCI_DESCR(device_id, opt) \
383 .dev_id = (device_id), \
384 .optional = opt
386 static const struct pci_id_descr pci_dev_descr_sbridge[] = {
387 /* Processor Home Agent */
388 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_HA0, 0) },
390 /* Memory controller */
391 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TA, 0) },
392 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_RAS, 0) },
393 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD0, 0) },
394 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD1, 0) },
395 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD2, 0) },
396 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD3, 0) },
397 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_DDRIO, 1) },
399 /* System Address Decoder */
400 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_SAD0, 0) },
401 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_SAD1, 0) },
403 /* Broadcast Registers */
404 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_BR, 0) },
407 #define PCI_ID_TABLE_ENTRY(A) { .descr=A, .n_devs = ARRAY_SIZE(A) }
408 static const struct pci_id_table pci_dev_descr_sbridge_table[] = {
409 PCI_ID_TABLE_ENTRY(pci_dev_descr_sbridge),
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),
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),
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 CH3,5,6 and MC1 is mapped to CH0,1,2
558 #define knl_channel_remap(channel) ((channel + 3) % 6)
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),
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),
662 {0,} /* 0 terminated list. */
666 * pci_device_id table for which devices we are looking for
668 static const struct pci_device_id sbridge_pci_tbl[] = {
669 {PCI_DEVICE(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_HA0)},
670 {PCI_DEVICE(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TA)},
671 {PCI_DEVICE(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0)},
672 {PCI_DEVICE(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0)},
673 {PCI_DEVICE(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_KNL_IMC_SAD0)},
674 {0,} /* 0 terminated list. */
678 /****************************************************************************
679 Ancillary status routines
680 ****************************************************************************/
682 static inline int numrank(enum type type, u32 mtr)
684 int ranks = (1 << RANK_CNT_BITS(mtr));
685 int max = 4;
687 if (type == HASWELL || type == BROADWELL || type == KNIGHTS_LANDING)
688 max = 8;
690 if (ranks > max) {
691 edac_dbg(0, "Invalid number of ranks: %d (max = %i) raw value = %x (%04x)\n",
692 ranks, max, (unsigned int)RANK_CNT_BITS(mtr), mtr);
693 return -EINVAL;
696 return ranks;
699 static inline int numrow(u32 mtr)
701 int rows = (RANK_WIDTH_BITS(mtr) + 12);
703 if (rows < 13 || rows > 18) {
704 edac_dbg(0, "Invalid number of rows: %d (should be between 14 and 17) raw value = %x (%04x)\n",
705 rows, (unsigned int)RANK_WIDTH_BITS(mtr), mtr);
706 return -EINVAL;
709 return 1 << rows;
712 static inline int numcol(u32 mtr)
714 int cols = (COL_WIDTH_BITS(mtr) + 10);
716 if (cols > 12) {
717 edac_dbg(0, "Invalid number of cols: %d (max = 4) raw value = %x (%04x)\n",
718 cols, (unsigned int)COL_WIDTH_BITS(mtr), mtr);
719 return -EINVAL;
722 return 1 << cols;
725 static struct sbridge_dev *get_sbridge_dev(u8 bus, int multi_bus)
727 struct sbridge_dev *sbridge_dev;
730 * If we have devices scattered across several busses that pertain
731 * to the same memory controller, we'll lump them all together.
733 if (multi_bus) {
734 return list_first_entry_or_null(&sbridge_edac_list,
735 struct sbridge_dev, list);
738 list_for_each_entry(sbridge_dev, &sbridge_edac_list, list) {
739 if (sbridge_dev->bus == bus)
740 return sbridge_dev;
743 return NULL;
746 static struct sbridge_dev *alloc_sbridge_dev(u8 bus,
747 const struct pci_id_table *table)
749 struct sbridge_dev *sbridge_dev;
751 sbridge_dev = kzalloc(sizeof(*sbridge_dev), GFP_KERNEL);
752 if (!sbridge_dev)
753 return NULL;
755 sbridge_dev->pdev = kzalloc(sizeof(*sbridge_dev->pdev) * table->n_devs,
756 GFP_KERNEL);
757 if (!sbridge_dev->pdev) {
758 kfree(sbridge_dev);
759 return NULL;
762 sbridge_dev->bus = bus;
763 sbridge_dev->n_devs = table->n_devs;
764 list_add_tail(&sbridge_dev->list, &sbridge_edac_list);
766 return sbridge_dev;
769 static void free_sbridge_dev(struct sbridge_dev *sbridge_dev)
771 list_del(&sbridge_dev->list);
772 kfree(sbridge_dev->pdev);
773 kfree(sbridge_dev);
776 static u64 sbridge_get_tolm(struct sbridge_pvt *pvt)
778 u32 reg;
780 /* Address range is 32:28 */
781 pci_read_config_dword(pvt->pci_sad1, TOLM, &reg);
782 return GET_TOLM(reg);
785 static u64 sbridge_get_tohm(struct sbridge_pvt *pvt)
787 u32 reg;
789 pci_read_config_dword(pvt->pci_sad1, TOHM, &reg);
790 return GET_TOHM(reg);
793 static u64 ibridge_get_tolm(struct sbridge_pvt *pvt)
795 u32 reg;
797 pci_read_config_dword(pvt->pci_br1, TOLM, &reg);
799 return GET_TOLM(reg);
802 static u64 ibridge_get_tohm(struct sbridge_pvt *pvt)
804 u32 reg;
806 pci_read_config_dword(pvt->pci_br1, TOHM, &reg);
808 return GET_TOHM(reg);
811 static u64 rir_limit(u32 reg)
813 return ((u64)GET_BITFIELD(reg, 1, 10) << 29) | 0x1fffffff;
816 static u64 sad_limit(u32 reg)
818 return (GET_BITFIELD(reg, 6, 25) << 26) | 0x3ffffff;
821 static u32 interleave_mode(u32 reg)
823 return GET_BITFIELD(reg, 1, 1);
826 char *show_interleave_mode(u32 reg)
828 return interleave_mode(reg) ? "8:6" : "[8:6]XOR[18:16]";
831 static u32 dram_attr(u32 reg)
833 return GET_BITFIELD(reg, 2, 3);
836 static u64 knl_sad_limit(u32 reg)
838 return (GET_BITFIELD(reg, 7, 26) << 26) | 0x3ffffff;
841 static u32 knl_interleave_mode(u32 reg)
843 return GET_BITFIELD(reg, 1, 2);
846 static char *knl_show_interleave_mode(u32 reg)
848 char *s;
850 switch (knl_interleave_mode(reg)) {
851 case 0:
852 s = "use address bits [8:6]";
853 break;
854 case 1:
855 s = "use address bits [10:8]";
856 break;
857 case 2:
858 s = "use address bits [14:12]";
859 break;
860 case 3:
861 s = "use address bits [32:30]";
862 break;
863 default:
864 WARN_ON(1);
865 break;
868 return s;
871 static u32 dram_attr_knl(u32 reg)
873 return GET_BITFIELD(reg, 3, 4);
877 static enum mem_type get_memory_type(struct sbridge_pvt *pvt)
879 u32 reg;
880 enum mem_type mtype;
882 if (pvt->pci_ddrio) {
883 pci_read_config_dword(pvt->pci_ddrio, pvt->info.rankcfgr,
884 &reg);
885 if (GET_BITFIELD(reg, 11, 11))
886 /* FIXME: Can also be LRDIMM */
887 mtype = MEM_RDDR3;
888 else
889 mtype = MEM_DDR3;
890 } else
891 mtype = MEM_UNKNOWN;
893 return mtype;
896 static enum mem_type haswell_get_memory_type(struct sbridge_pvt *pvt)
898 u32 reg;
899 bool registered = false;
900 enum mem_type mtype = MEM_UNKNOWN;
902 if (!pvt->pci_ddrio)
903 goto out;
905 pci_read_config_dword(pvt->pci_ddrio,
906 HASWELL_DDRCRCLKCONTROLS, &reg);
907 /* Is_Rdimm */
908 if (GET_BITFIELD(reg, 16, 16))
909 registered = true;
911 pci_read_config_dword(pvt->pci_ta, MCMTR, &reg);
912 if (GET_BITFIELD(reg, 14, 14)) {
913 if (registered)
914 mtype = MEM_RDDR4;
915 else
916 mtype = MEM_DDR4;
917 } else {
918 if (registered)
919 mtype = MEM_RDDR3;
920 else
921 mtype = MEM_DDR3;
924 out:
925 return mtype;
928 static enum dev_type knl_get_width(struct sbridge_pvt *pvt, u32 mtr)
930 /* for KNL value is fixed */
931 return DEV_X16;
934 static enum dev_type sbridge_get_width(struct sbridge_pvt *pvt, u32 mtr)
936 /* there's no way to figure out */
937 return DEV_UNKNOWN;
940 static enum dev_type __ibridge_get_width(u32 mtr)
942 enum dev_type type;
944 switch (mtr) {
945 case 3:
946 type = DEV_UNKNOWN;
947 break;
948 case 2:
949 type = DEV_X16;
950 break;
951 case 1:
952 type = DEV_X8;
953 break;
954 case 0:
955 type = DEV_X4;
956 break;
959 return type;
962 static enum dev_type ibridge_get_width(struct sbridge_pvt *pvt, u32 mtr)
965 * ddr3_width on the documentation but also valid for DDR4 on
966 * Haswell
968 return __ibridge_get_width(GET_BITFIELD(mtr, 7, 8));
971 static enum dev_type broadwell_get_width(struct sbridge_pvt *pvt, u32 mtr)
973 /* ddr3_width on the documentation but also valid for DDR4 */
974 return __ibridge_get_width(GET_BITFIELD(mtr, 8, 9));
977 static enum mem_type knl_get_memory_type(struct sbridge_pvt *pvt)
979 /* DDR4 RDIMMS and LRDIMMS are supported */
980 return MEM_RDDR4;
983 static u8 get_node_id(struct sbridge_pvt *pvt)
985 u32 reg;
986 pci_read_config_dword(pvt->pci_br0, SAD_CONTROL, &reg);
987 return GET_BITFIELD(reg, 0, 2);
990 static u8 haswell_get_node_id(struct sbridge_pvt *pvt)
992 u32 reg;
994 pci_read_config_dword(pvt->pci_sad1, SAD_CONTROL, &reg);
995 return GET_BITFIELD(reg, 0, 3);
998 static u8 knl_get_node_id(struct sbridge_pvt *pvt)
1000 u32 reg;
1002 pci_read_config_dword(pvt->pci_sad1, SAD_CONTROL, &reg);
1003 return GET_BITFIELD(reg, 0, 2);
1007 static u64 haswell_get_tolm(struct sbridge_pvt *pvt)
1009 u32 reg;
1011 pci_read_config_dword(pvt->info.pci_vtd, HASWELL_TOLM, &reg);
1012 return (GET_BITFIELD(reg, 26, 31) << 26) | 0x3ffffff;
1015 static u64 haswell_get_tohm(struct sbridge_pvt *pvt)
1017 u64 rc;
1018 u32 reg;
1020 pci_read_config_dword(pvt->info.pci_vtd, HASWELL_TOHM_0, &reg);
1021 rc = GET_BITFIELD(reg, 26, 31);
1022 pci_read_config_dword(pvt->info.pci_vtd, HASWELL_TOHM_1, &reg);
1023 rc = ((reg << 6) | rc) << 26;
1025 return rc | 0x1ffffff;
1028 static u64 knl_get_tolm(struct sbridge_pvt *pvt)
1030 u32 reg;
1032 pci_read_config_dword(pvt->knl.pci_mc_info, KNL_TOLM, &reg);
1033 return (GET_BITFIELD(reg, 26, 31) << 26) | 0x3ffffff;
1036 static u64 knl_get_tohm(struct sbridge_pvt *pvt)
1038 u64 rc;
1039 u32 reg_lo, reg_hi;
1041 pci_read_config_dword(pvt->knl.pci_mc_info, KNL_TOHM_0, &reg_lo);
1042 pci_read_config_dword(pvt->knl.pci_mc_info, KNL_TOHM_1, &reg_hi);
1043 rc = ((u64)reg_hi << 32) | reg_lo;
1044 return rc | 0x3ffffff;
1048 static u64 haswell_rir_limit(u32 reg)
1050 return (((u64)GET_BITFIELD(reg, 1, 11) + 1) << 29) - 1;
1053 static inline u8 sad_pkg_socket(u8 pkg)
1055 /* on Ivy Bridge, nodeID is SASS, where A is HA and S is node id */
1056 return ((pkg >> 3) << 2) | (pkg & 0x3);
1059 static inline u8 sad_pkg_ha(u8 pkg)
1061 return (pkg >> 2) & 0x1;
1064 static int haswell_chan_hash(int idx, u64 addr)
1066 int i;
1069 * XOR even bits from 12:26 to bit0 of idx,
1070 * odd bits from 13:27 to bit1
1072 for (i = 12; i < 28; i += 2)
1073 idx ^= (addr >> i) & 3;
1075 return idx;
1078 /****************************************************************************
1079 Memory check routines
1080 ****************************************************************************/
1081 static struct pci_dev *get_pdev_same_bus(u8 bus, u32 id)
1083 struct pci_dev *pdev = NULL;
1085 do {
1086 pdev = pci_get_device(PCI_VENDOR_ID_INTEL, id, pdev);
1087 if (pdev && pdev->bus->number == bus)
1088 break;
1089 } while (pdev);
1091 return pdev;
1095 * check_if_ecc_is_active() - Checks if ECC is active
1096 * @bus: Device bus
1097 * @type: Memory controller type
1098 * returns: 0 in case ECC is active, -ENODEV if it can't be determined or
1099 * disabled
1101 static int check_if_ecc_is_active(const u8 bus, enum type type)
1103 struct pci_dev *pdev = NULL;
1104 u32 mcmtr, id;
1106 switch (type) {
1107 case IVY_BRIDGE:
1108 id = PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TA;
1109 break;
1110 case HASWELL:
1111 id = PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TA;
1112 break;
1113 case SANDY_BRIDGE:
1114 id = PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TA;
1115 break;
1116 case BROADWELL:
1117 id = PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TA;
1118 break;
1119 case KNIGHTS_LANDING:
1121 * KNL doesn't group things by bus the same way
1122 * SB/IB/Haswell does.
1124 id = PCI_DEVICE_ID_INTEL_KNL_IMC_TA;
1125 break;
1126 default:
1127 return -ENODEV;
1130 if (type != KNIGHTS_LANDING)
1131 pdev = get_pdev_same_bus(bus, id);
1132 else
1133 pdev = pci_get_device(PCI_VENDOR_ID_INTEL, id, 0);
1135 if (!pdev) {
1136 sbridge_printk(KERN_ERR, "Couldn't find PCI device "
1137 "%04x:%04x! on bus %02d\n",
1138 PCI_VENDOR_ID_INTEL, id, bus);
1139 return -ENODEV;
1142 pci_read_config_dword(pdev,
1143 type == KNIGHTS_LANDING ? KNL_MCMTR : MCMTR, &mcmtr);
1144 if (!IS_ECC_ENABLED(mcmtr)) {
1145 sbridge_printk(KERN_ERR, "ECC is disabled. Aborting\n");
1146 return -ENODEV;
1148 return 0;
1151 /* Low bits of TAD limit, and some metadata. */
1152 static const u32 knl_tad_dram_limit_lo[] = {
1153 0x400, 0x500, 0x600, 0x700,
1154 0x800, 0x900, 0xa00, 0xb00,
1157 /* Low bits of TAD offset. */
1158 static const u32 knl_tad_dram_offset_lo[] = {
1159 0x404, 0x504, 0x604, 0x704,
1160 0x804, 0x904, 0xa04, 0xb04,
1163 /* High 16 bits of TAD limit and offset. */
1164 static const u32 knl_tad_dram_hi[] = {
1165 0x408, 0x508, 0x608, 0x708,
1166 0x808, 0x908, 0xa08, 0xb08,
1169 /* Number of ways a tad entry is interleaved. */
1170 static const u32 knl_tad_ways[] = {
1171 8, 6, 4, 3, 2, 1,
1175 * Retrieve the n'th Target Address Decode table entry
1176 * from the memory controller's TAD table.
1178 * @pvt: driver private data
1179 * @entry: which entry you want to retrieve
1180 * @mc: which memory controller (0 or 1)
1181 * @offset: output tad range offset
1182 * @limit: output address of first byte above tad range
1183 * @ways: output number of interleave ways
1185 * The offset value has curious semantics. It's a sort of running total
1186 * of the sizes of all the memory regions that aren't mapped in this
1187 * tad table.
1189 static int knl_get_tad(const struct sbridge_pvt *pvt,
1190 const int entry,
1191 const int mc,
1192 u64 *offset,
1193 u64 *limit,
1194 int *ways)
1196 u32 reg_limit_lo, reg_offset_lo, reg_hi;
1197 struct pci_dev *pci_mc;
1198 int way_id;
1200 switch (mc) {
1201 case 0:
1202 pci_mc = pvt->knl.pci_mc0;
1203 break;
1204 case 1:
1205 pci_mc = pvt->knl.pci_mc1;
1206 break;
1207 default:
1208 WARN_ON(1);
1209 return -EINVAL;
1212 pci_read_config_dword(pci_mc,
1213 knl_tad_dram_limit_lo[entry], &reg_limit_lo);
1214 pci_read_config_dword(pci_mc,
1215 knl_tad_dram_offset_lo[entry], &reg_offset_lo);
1216 pci_read_config_dword(pci_mc,
1217 knl_tad_dram_hi[entry], &reg_hi);
1219 /* Is this TAD entry enabled? */
1220 if (!GET_BITFIELD(reg_limit_lo, 0, 0))
1221 return -ENODEV;
1223 way_id = GET_BITFIELD(reg_limit_lo, 3, 5);
1225 if (way_id < ARRAY_SIZE(knl_tad_ways)) {
1226 *ways = knl_tad_ways[way_id];
1227 } else {
1228 *ways = 0;
1229 sbridge_printk(KERN_ERR,
1230 "Unexpected value %d in mc_tad_limit_lo wayness field\n",
1231 way_id);
1232 return -ENODEV;
1236 * The least significant 6 bits of base and limit are truncated.
1237 * For limit, we fill the missing bits with 1s.
1239 *offset = ((u64) GET_BITFIELD(reg_offset_lo, 6, 31) << 6) |
1240 ((u64) GET_BITFIELD(reg_hi, 0, 15) << 32);
1241 *limit = ((u64) GET_BITFIELD(reg_limit_lo, 6, 31) << 6) | 63 |
1242 ((u64) GET_BITFIELD(reg_hi, 16, 31) << 32);
1244 return 0;
1247 /* Determine which memory controller is responsible for a given channel. */
1248 static int knl_channel_mc(int channel)
1250 WARN_ON(channel < 0 || channel >= 6);
1252 return channel < 3 ? 1 : 0;
1256 * Get the Nth entry from EDC_ROUTE_TABLE register.
1257 * (This is the per-tile mapping of logical interleave targets to
1258 * physical EDC modules.)
1260 * entry 0: 0:2
1261 * 1: 3:5
1262 * 2: 6:8
1263 * 3: 9:11
1264 * 4: 12:14
1265 * 5: 15:17
1266 * 6: 18:20
1267 * 7: 21:23
1268 * reserved: 24:31
1270 static u32 knl_get_edc_route(int entry, u32 reg)
1272 WARN_ON(entry >= KNL_MAX_EDCS);
1273 return GET_BITFIELD(reg, entry*3, (entry*3)+2);
1277 * Get the Nth entry from MC_ROUTE_TABLE register.
1278 * (This is the per-tile mapping of logical interleave targets to
1279 * physical DRAM channels modules.)
1281 * entry 0: mc 0:2 channel 18:19
1282 * 1: mc 3:5 channel 20:21
1283 * 2: mc 6:8 channel 22:23
1284 * 3: mc 9:11 channel 24:25
1285 * 4: mc 12:14 channel 26:27
1286 * 5: mc 15:17 channel 28:29
1287 * reserved: 30:31
1289 * Though we have 3 bits to identify the MC, we should only see
1290 * the values 0 or 1.
1293 static u32 knl_get_mc_route(int entry, u32 reg)
1295 int mc, chan;
1297 WARN_ON(entry >= KNL_MAX_CHANNELS);
1299 mc = GET_BITFIELD(reg, entry*3, (entry*3)+2);
1300 chan = GET_BITFIELD(reg, (entry*2) + 18, (entry*2) + 18 + 1);
1302 return knl_channel_remap(mc*3 + chan);
1306 * Render the EDC_ROUTE register in human-readable form.
1307 * Output string s should be at least KNL_MAX_EDCS*2 bytes.
1309 static void knl_show_edc_route(u32 reg, char *s)
1311 int i;
1313 for (i = 0; i < KNL_MAX_EDCS; i++) {
1314 s[i*2] = knl_get_edc_route(i, reg) + '0';
1315 s[i*2+1] = '-';
1318 s[KNL_MAX_EDCS*2 - 1] = '\0';
1322 * Render the MC_ROUTE register in human-readable form.
1323 * Output string s should be at least KNL_MAX_CHANNELS*2 bytes.
1325 static void knl_show_mc_route(u32 reg, char *s)
1327 int i;
1329 for (i = 0; i < KNL_MAX_CHANNELS; i++) {
1330 s[i*2] = knl_get_mc_route(i, reg) + '0';
1331 s[i*2+1] = '-';
1334 s[KNL_MAX_CHANNELS*2 - 1] = '\0';
1337 #define KNL_EDC_ROUTE 0xb8
1338 #define KNL_MC_ROUTE 0xb4
1340 /* Is this dram rule backed by regular DRAM in flat mode? */
1341 #define KNL_EDRAM(reg) GET_BITFIELD(reg, 29, 29)
1343 /* Is this dram rule cached? */
1344 #define KNL_CACHEABLE(reg) GET_BITFIELD(reg, 28, 28)
1346 /* Is this rule backed by edc ? */
1347 #define KNL_EDRAM_ONLY(reg) GET_BITFIELD(reg, 29, 29)
1349 /* Is this rule backed by DRAM, cacheable in EDRAM? */
1350 #define KNL_CACHEABLE(reg) GET_BITFIELD(reg, 28, 28)
1352 /* Is this rule mod3? */
1353 #define KNL_MOD3(reg) GET_BITFIELD(reg, 27, 27)
1356 * Figure out how big our RAM modules are.
1358 * The DIMMMTR register in KNL doesn't tell us the size of the DIMMs, so we
1359 * have to figure this out from the SAD rules, interleave lists, route tables,
1360 * and TAD rules.
1362 * SAD rules can have holes in them (e.g. the 3G-4G hole), so we have to
1363 * inspect the TAD rules to figure out how large the SAD regions really are.
1365 * When we know the real size of a SAD region and how many ways it's
1366 * interleaved, we know the individual contribution of each channel to
1367 * TAD is size/ways.
1369 * Finally, we have to check whether each channel participates in each SAD
1370 * region.
1372 * Fortunately, KNL only supports one DIMM per channel, so once we know how
1373 * much memory the channel uses, we know the DIMM is at least that large.
1374 * (The BIOS might possibly choose not to map all available memory, in which
1375 * case we will underreport the size of the DIMM.)
1377 * In theory, we could try to determine the EDC sizes as well, but that would
1378 * only work in flat mode, not in cache mode.
1380 * @mc_sizes: Output sizes of channels (must have space for KNL_MAX_CHANNELS
1381 * elements)
1383 static int knl_get_dimm_capacity(struct sbridge_pvt *pvt, u64 *mc_sizes)
1385 u64 sad_base, sad_size, sad_limit = 0;
1386 u64 tad_base, tad_size, tad_limit, tad_deadspace, tad_livespace;
1387 int sad_rule = 0;
1388 int tad_rule = 0;
1389 int intrlv_ways, tad_ways;
1390 u32 first_pkg, pkg;
1391 int i;
1392 u64 sad_actual_size[2]; /* sad size accounting for holes, per mc */
1393 u32 dram_rule, interleave_reg;
1394 u32 mc_route_reg[KNL_MAX_CHAS];
1395 u32 edc_route_reg[KNL_MAX_CHAS];
1396 int edram_only;
1397 char edc_route_string[KNL_MAX_EDCS*2];
1398 char mc_route_string[KNL_MAX_CHANNELS*2];
1399 int cur_reg_start;
1400 int mc;
1401 int channel;
1402 int way;
1403 int participants[KNL_MAX_CHANNELS];
1404 int participant_count = 0;
1406 for (i = 0; i < KNL_MAX_CHANNELS; i++)
1407 mc_sizes[i] = 0;
1409 /* Read the EDC route table in each CHA. */
1410 cur_reg_start = 0;
1411 for (i = 0; i < KNL_MAX_CHAS; i++) {
1412 pci_read_config_dword(pvt->knl.pci_cha[i],
1413 KNL_EDC_ROUTE, &edc_route_reg[i]);
1415 if (i > 0 && edc_route_reg[i] != edc_route_reg[i-1]) {
1416 knl_show_edc_route(edc_route_reg[i-1],
1417 edc_route_string);
1418 if (cur_reg_start == i-1)
1419 edac_dbg(0, "edc route table for CHA %d: %s\n",
1420 cur_reg_start, edc_route_string);
1421 else
1422 edac_dbg(0, "edc route table for CHA %d-%d: %s\n",
1423 cur_reg_start, i-1, edc_route_string);
1424 cur_reg_start = i;
1427 knl_show_edc_route(edc_route_reg[i-1], edc_route_string);
1428 if (cur_reg_start == i-1)
1429 edac_dbg(0, "edc route table for CHA %d: %s\n",
1430 cur_reg_start, edc_route_string);
1431 else
1432 edac_dbg(0, "edc route table for CHA %d-%d: %s\n",
1433 cur_reg_start, i-1, edc_route_string);
1435 /* Read the MC route table in each CHA. */
1436 cur_reg_start = 0;
1437 for (i = 0; i < KNL_MAX_CHAS; i++) {
1438 pci_read_config_dword(pvt->knl.pci_cha[i],
1439 KNL_MC_ROUTE, &mc_route_reg[i]);
1441 if (i > 0 && mc_route_reg[i] != mc_route_reg[i-1]) {
1442 knl_show_mc_route(mc_route_reg[i-1], mc_route_string);
1443 if (cur_reg_start == i-1)
1444 edac_dbg(0, "mc route table for CHA %d: %s\n",
1445 cur_reg_start, mc_route_string);
1446 else
1447 edac_dbg(0, "mc route table for CHA %d-%d: %s\n",
1448 cur_reg_start, i-1, mc_route_string);
1449 cur_reg_start = i;
1452 knl_show_mc_route(mc_route_reg[i-1], mc_route_string);
1453 if (cur_reg_start == i-1)
1454 edac_dbg(0, "mc route table for CHA %d: %s\n",
1455 cur_reg_start, mc_route_string);
1456 else
1457 edac_dbg(0, "mc route table for CHA %d-%d: %s\n",
1458 cur_reg_start, i-1, mc_route_string);
1460 /* Process DRAM rules */
1461 for (sad_rule = 0; sad_rule < pvt->info.max_sad; sad_rule++) {
1462 /* previous limit becomes the new base */
1463 sad_base = sad_limit;
1465 pci_read_config_dword(pvt->pci_sad0,
1466 pvt->info.dram_rule[sad_rule], &dram_rule);
1468 if (!DRAM_RULE_ENABLE(dram_rule))
1469 break;
1471 edram_only = KNL_EDRAM_ONLY(dram_rule);
1473 sad_limit = pvt->info.sad_limit(dram_rule)+1;
1474 sad_size = sad_limit - sad_base;
1476 pci_read_config_dword(pvt->pci_sad0,
1477 pvt->info.interleave_list[sad_rule], &interleave_reg);
1480 * Find out how many ways this dram rule is interleaved.
1481 * We stop when we see the first channel again.
1483 first_pkg = sad_pkg(pvt->info.interleave_pkg,
1484 interleave_reg, 0);
1485 for (intrlv_ways = 1; intrlv_ways < 8; intrlv_ways++) {
1486 pkg = sad_pkg(pvt->info.interleave_pkg,
1487 interleave_reg, intrlv_ways);
1489 if ((pkg & 0x8) == 0) {
1491 * 0 bit means memory is non-local,
1492 * which KNL doesn't support
1494 edac_dbg(0, "Unexpected interleave target %d\n",
1495 pkg);
1496 return -1;
1499 if (pkg == first_pkg)
1500 break;
1502 if (KNL_MOD3(dram_rule))
1503 intrlv_ways *= 3;
1505 edac_dbg(3, "dram rule %d (base 0x%llx, limit 0x%llx), %d way interleave%s\n",
1506 sad_rule,
1507 sad_base,
1508 sad_limit,
1509 intrlv_ways,
1510 edram_only ? ", EDRAM" : "");
1513 * Find out how big the SAD region really is by iterating
1514 * over TAD tables (SAD regions may contain holes).
1515 * Each memory controller might have a different TAD table, so
1516 * we have to look at both.
1518 * Livespace is the memory that's mapped in this TAD table,
1519 * deadspace is the holes (this could be the MMIO hole, or it
1520 * could be memory that's mapped by the other TAD table but
1521 * not this one).
1523 for (mc = 0; mc < 2; mc++) {
1524 sad_actual_size[mc] = 0;
1525 tad_livespace = 0;
1526 for (tad_rule = 0;
1527 tad_rule < ARRAY_SIZE(
1528 knl_tad_dram_limit_lo);
1529 tad_rule++) {
1530 if (knl_get_tad(pvt,
1531 tad_rule,
1533 &tad_deadspace,
1534 &tad_limit,
1535 &tad_ways))
1536 break;
1538 tad_size = (tad_limit+1) -
1539 (tad_livespace + tad_deadspace);
1540 tad_livespace += tad_size;
1541 tad_base = (tad_limit+1) - tad_size;
1543 if (tad_base < sad_base) {
1544 if (tad_limit > sad_base)
1545 edac_dbg(0, "TAD region overlaps lower SAD boundary -- TAD tables may be configured incorrectly.\n");
1546 } else if (tad_base < sad_limit) {
1547 if (tad_limit+1 > sad_limit) {
1548 edac_dbg(0, "TAD region overlaps upper SAD boundary -- TAD tables may be configured incorrectly.\n");
1549 } else {
1550 /* TAD region is completely inside SAD region */
1551 edac_dbg(3, "TAD region %d 0x%llx - 0x%llx (%lld bytes) table%d\n",
1552 tad_rule, tad_base,
1553 tad_limit, tad_size,
1554 mc);
1555 sad_actual_size[mc] += tad_size;
1558 tad_base = tad_limit+1;
1562 for (mc = 0; mc < 2; mc++) {
1563 edac_dbg(3, " total TAD DRAM footprint in table%d : 0x%llx (%lld bytes)\n",
1564 mc, sad_actual_size[mc], sad_actual_size[mc]);
1567 /* Ignore EDRAM rule */
1568 if (edram_only)
1569 continue;
1571 /* Figure out which channels participate in interleave. */
1572 for (channel = 0; channel < KNL_MAX_CHANNELS; channel++)
1573 participants[channel] = 0;
1575 /* For each channel, does at least one CHA have
1576 * this channel mapped to the given target?
1578 for (channel = 0; channel < KNL_MAX_CHANNELS; channel++) {
1579 for (way = 0; way < intrlv_ways; way++) {
1580 int target;
1581 int cha;
1583 if (KNL_MOD3(dram_rule))
1584 target = way;
1585 else
1586 target = 0x7 & sad_pkg(
1587 pvt->info.interleave_pkg, interleave_reg, way);
1589 for (cha = 0; cha < KNL_MAX_CHAS; cha++) {
1590 if (knl_get_mc_route(target,
1591 mc_route_reg[cha]) == channel
1592 && !participants[channel]) {
1593 participant_count++;
1594 participants[channel] = 1;
1595 break;
1601 if (participant_count != intrlv_ways)
1602 edac_dbg(0, "participant_count (%d) != interleave_ways (%d): DIMM size may be incorrect\n",
1603 participant_count, intrlv_ways);
1605 for (channel = 0; channel < KNL_MAX_CHANNELS; channel++) {
1606 mc = knl_channel_mc(channel);
1607 if (participants[channel]) {
1608 edac_dbg(4, "mc channel %d contributes %lld bytes via sad entry %d\n",
1609 channel,
1610 sad_actual_size[mc]/intrlv_ways,
1611 sad_rule);
1612 mc_sizes[channel] +=
1613 sad_actual_size[mc]/intrlv_ways;
1618 return 0;
1621 static int get_dimm_config(struct mem_ctl_info *mci)
1623 struct sbridge_pvt *pvt = mci->pvt_info;
1624 struct dimm_info *dimm;
1625 unsigned i, j, banks, ranks, rows, cols, npages;
1626 u64 size;
1627 u32 reg;
1628 enum edac_type mode;
1629 enum mem_type mtype;
1630 int channels = pvt->info.type == KNIGHTS_LANDING ?
1631 KNL_MAX_CHANNELS : NUM_CHANNELS;
1632 u64 knl_mc_sizes[KNL_MAX_CHANNELS];
1634 if (pvt->info.type == HASWELL || pvt->info.type == BROADWELL) {
1635 pci_read_config_dword(pvt->pci_ha0, HASWELL_HASYSDEFEATURE2, &reg);
1636 pvt->is_chan_hash = GET_BITFIELD(reg, 21, 21);
1638 if (pvt->info.type == HASWELL || pvt->info.type == BROADWELL ||
1639 pvt->info.type == KNIGHTS_LANDING)
1640 pci_read_config_dword(pvt->pci_sad1, SAD_TARGET, &reg);
1641 else
1642 pci_read_config_dword(pvt->pci_br0, SAD_TARGET, &reg);
1644 if (pvt->info.type == KNIGHTS_LANDING)
1645 pvt->sbridge_dev->source_id = SOURCE_ID_KNL(reg);
1646 else
1647 pvt->sbridge_dev->source_id = SOURCE_ID(reg);
1649 pvt->sbridge_dev->node_id = pvt->info.get_node_id(pvt);
1650 edac_dbg(0, "mc#%d: Node ID: %d, source ID: %d\n",
1651 pvt->sbridge_dev->mc,
1652 pvt->sbridge_dev->node_id,
1653 pvt->sbridge_dev->source_id);
1655 /* KNL doesn't support mirroring or lockstep,
1656 * and is always closed page
1658 if (pvt->info.type == KNIGHTS_LANDING) {
1659 mode = EDAC_S4ECD4ED;
1660 pvt->is_mirrored = false;
1662 if (knl_get_dimm_capacity(pvt, knl_mc_sizes) != 0)
1663 return -1;
1664 } else {
1665 pci_read_config_dword(pvt->pci_ras, RASENABLES, &reg);
1666 if (IS_MIRROR_ENABLED(reg)) {
1667 edac_dbg(0, "Memory mirror is enabled\n");
1668 pvt->is_mirrored = true;
1669 } else {
1670 edac_dbg(0, "Memory mirror is disabled\n");
1671 pvt->is_mirrored = false;
1674 pci_read_config_dword(pvt->pci_ta, MCMTR, &pvt->info.mcmtr);
1675 if (IS_LOCKSTEP_ENABLED(pvt->info.mcmtr)) {
1676 edac_dbg(0, "Lockstep is enabled\n");
1677 mode = EDAC_S8ECD8ED;
1678 pvt->is_lockstep = true;
1679 } else {
1680 edac_dbg(0, "Lockstep is disabled\n");
1681 mode = EDAC_S4ECD4ED;
1682 pvt->is_lockstep = false;
1684 if (IS_CLOSE_PG(pvt->info.mcmtr)) {
1685 edac_dbg(0, "address map is on closed page mode\n");
1686 pvt->is_close_pg = true;
1687 } else {
1688 edac_dbg(0, "address map is on open page mode\n");
1689 pvt->is_close_pg = false;
1693 mtype = pvt->info.get_memory_type(pvt);
1694 if (mtype == MEM_RDDR3 || mtype == MEM_RDDR4)
1695 edac_dbg(0, "Memory is registered\n");
1696 else if (mtype == MEM_UNKNOWN)
1697 edac_dbg(0, "Cannot determine memory type\n");
1698 else
1699 edac_dbg(0, "Memory is unregistered\n");
1701 if (mtype == MEM_DDR4 || mtype == MEM_RDDR4)
1702 banks = 16;
1703 else
1704 banks = 8;
1706 for (i = 0; i < channels; i++) {
1707 u32 mtr;
1709 int max_dimms_per_channel;
1711 if (pvt->info.type == KNIGHTS_LANDING) {
1712 max_dimms_per_channel = 1;
1713 if (!pvt->knl.pci_channel[i])
1714 continue;
1715 } else {
1716 max_dimms_per_channel = ARRAY_SIZE(mtr_regs);
1717 if (!pvt->pci_tad[i])
1718 continue;
1721 for (j = 0; j < max_dimms_per_channel; j++) {
1722 dimm = EDAC_DIMM_PTR(mci->layers, mci->dimms, mci->n_layers,
1723 i, j, 0);
1724 if (pvt->info.type == KNIGHTS_LANDING) {
1725 pci_read_config_dword(pvt->knl.pci_channel[i],
1726 knl_mtr_reg, &mtr);
1727 } else {
1728 pci_read_config_dword(pvt->pci_tad[i],
1729 mtr_regs[j], &mtr);
1731 edac_dbg(4, "Channel #%d MTR%d = %x\n", i, j, mtr);
1732 if (IS_DIMM_PRESENT(mtr)) {
1733 pvt->channel[i].dimms++;
1735 ranks = numrank(pvt->info.type, mtr);
1737 if (pvt->info.type == KNIGHTS_LANDING) {
1738 /* For DDR4, this is fixed. */
1739 cols = 1 << 10;
1740 rows = knl_mc_sizes[i] /
1741 ((u64) cols * ranks * banks * 8);
1742 } else {
1743 rows = numrow(mtr);
1744 cols = numcol(mtr);
1747 size = ((u64)rows * cols * banks * ranks) >> (20 - 3);
1748 npages = MiB_TO_PAGES(size);
1750 edac_dbg(0, "mc#%d: ha %d channel %d, dimm %d, %lld Mb (%d pages) bank: %d, rank: %d, row: %#x, col: %#x\n",
1751 pvt->sbridge_dev->mc, i/4, i%4, j,
1752 size, npages,
1753 banks, ranks, rows, cols);
1755 dimm->nr_pages = npages;
1756 dimm->grain = 32;
1757 dimm->dtype = pvt->info.get_width(pvt, mtr);
1758 dimm->mtype = mtype;
1759 dimm->edac_mode = mode;
1760 snprintf(dimm->label, sizeof(dimm->label),
1761 "CPU_SrcID#%u_Ha#%u_Chan#%u_DIMM#%u",
1762 pvt->sbridge_dev->source_id, i/4, i%4, j);
1767 return 0;
1770 static void get_memory_layout(const struct mem_ctl_info *mci)
1772 struct sbridge_pvt *pvt = mci->pvt_info;
1773 int i, j, k, n_sads, n_tads, sad_interl;
1774 u32 reg;
1775 u64 limit, prv = 0;
1776 u64 tmp_mb;
1777 u32 gb, mb;
1778 u32 rir_way;
1781 * Step 1) Get TOLM/TOHM ranges
1784 pvt->tolm = pvt->info.get_tolm(pvt);
1785 tmp_mb = (1 + pvt->tolm) >> 20;
1787 gb = div_u64_rem(tmp_mb, 1024, &mb);
1788 edac_dbg(0, "TOLM: %u.%03u GB (0x%016Lx)\n",
1789 gb, (mb*1000)/1024, (u64)pvt->tolm);
1791 /* Address range is already 45:25 */
1792 pvt->tohm = pvt->info.get_tohm(pvt);
1793 tmp_mb = (1 + pvt->tohm) >> 20;
1795 gb = div_u64_rem(tmp_mb, 1024, &mb);
1796 edac_dbg(0, "TOHM: %u.%03u GB (0x%016Lx)\n",
1797 gb, (mb*1000)/1024, (u64)pvt->tohm);
1800 * Step 2) Get SAD range and SAD Interleave list
1801 * TAD registers contain the interleave wayness. However, it
1802 * seems simpler to just discover it indirectly, with the
1803 * algorithm bellow.
1805 prv = 0;
1806 for (n_sads = 0; n_sads < pvt->info.max_sad; n_sads++) {
1807 /* SAD_LIMIT Address range is 45:26 */
1808 pci_read_config_dword(pvt->pci_sad0, pvt->info.dram_rule[n_sads],
1809 &reg);
1810 limit = pvt->info.sad_limit(reg);
1812 if (!DRAM_RULE_ENABLE(reg))
1813 continue;
1815 if (limit <= prv)
1816 break;
1818 tmp_mb = (limit + 1) >> 20;
1819 gb = div_u64_rem(tmp_mb, 1024, &mb);
1820 edac_dbg(0, "SAD#%d %s up to %u.%03u GB (0x%016Lx) Interleave: %s reg=0x%08x\n",
1821 n_sads,
1822 show_dram_attr(pvt->info.dram_attr(reg)),
1823 gb, (mb*1000)/1024,
1824 ((u64)tmp_mb) << 20L,
1825 pvt->info.show_interleave_mode(reg),
1826 reg);
1827 prv = limit;
1829 pci_read_config_dword(pvt->pci_sad0, pvt->info.interleave_list[n_sads],
1830 &reg);
1831 sad_interl = sad_pkg(pvt->info.interleave_pkg, reg, 0);
1832 for (j = 0; j < 8; j++) {
1833 u32 pkg = sad_pkg(pvt->info.interleave_pkg, reg, j);
1834 if (j > 0 && sad_interl == pkg)
1835 break;
1837 edac_dbg(0, "SAD#%d, interleave #%d: %d\n",
1838 n_sads, j, pkg);
1842 if (pvt->info.type == KNIGHTS_LANDING)
1843 return;
1846 * Step 3) Get TAD range
1848 prv = 0;
1849 for (n_tads = 0; n_tads < MAX_TAD; n_tads++) {
1850 pci_read_config_dword(pvt->pci_ha0, tad_dram_rule[n_tads],
1851 &reg);
1852 limit = TAD_LIMIT(reg);
1853 if (limit <= prv)
1854 break;
1855 tmp_mb = (limit + 1) >> 20;
1857 gb = div_u64_rem(tmp_mb, 1024, &mb);
1858 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",
1859 n_tads, gb, (mb*1000)/1024,
1860 ((u64)tmp_mb) << 20L,
1861 (u32)(1 << TAD_SOCK(reg)),
1862 (u32)TAD_CH(reg) + 1,
1863 (u32)TAD_TGT0(reg),
1864 (u32)TAD_TGT1(reg),
1865 (u32)TAD_TGT2(reg),
1866 (u32)TAD_TGT3(reg),
1867 reg);
1868 prv = limit;
1872 * Step 4) Get TAD offsets, per each channel
1874 for (i = 0; i < NUM_CHANNELS; i++) {
1875 if (!pvt->channel[i].dimms)
1876 continue;
1877 for (j = 0; j < n_tads; j++) {
1878 pci_read_config_dword(pvt->pci_tad[i],
1879 tad_ch_nilv_offset[j],
1880 &reg);
1881 tmp_mb = TAD_OFFSET(reg) >> 20;
1882 gb = div_u64_rem(tmp_mb, 1024, &mb);
1883 edac_dbg(0, "TAD CH#%d, offset #%d: %u.%03u GB (0x%016Lx), reg=0x%08x\n",
1884 i, j,
1885 gb, (mb*1000)/1024,
1886 ((u64)tmp_mb) << 20L,
1887 reg);
1892 * Step 6) Get RIR Wayness/Limit, per each channel
1894 for (i = 0; i < NUM_CHANNELS; i++) {
1895 if (!pvt->channel[i].dimms)
1896 continue;
1897 for (j = 0; j < MAX_RIR_RANGES; j++) {
1898 pci_read_config_dword(pvt->pci_tad[i],
1899 rir_way_limit[j],
1900 &reg);
1902 if (!IS_RIR_VALID(reg))
1903 continue;
1905 tmp_mb = pvt->info.rir_limit(reg) >> 20;
1906 rir_way = 1 << RIR_WAY(reg);
1907 gb = div_u64_rem(tmp_mb, 1024, &mb);
1908 edac_dbg(0, "CH#%d RIR#%d, limit: %u.%03u GB (0x%016Lx), way: %d, reg=0x%08x\n",
1909 i, j,
1910 gb, (mb*1000)/1024,
1911 ((u64)tmp_mb) << 20L,
1912 rir_way,
1913 reg);
1915 for (k = 0; k < rir_way; k++) {
1916 pci_read_config_dword(pvt->pci_tad[i],
1917 rir_offset[j][k],
1918 &reg);
1919 tmp_mb = RIR_OFFSET(reg) << 6;
1921 gb = div_u64_rem(tmp_mb, 1024, &mb);
1922 edac_dbg(0, "CH#%d RIR#%d INTL#%d, offset %u.%03u GB (0x%016Lx), tgt: %d, reg=0x%08x\n",
1923 i, j, k,
1924 gb, (mb*1000)/1024,
1925 ((u64)tmp_mb) << 20L,
1926 (u32)RIR_RNK_TGT(reg),
1927 reg);
1933 static struct mem_ctl_info *get_mci_for_node_id(u8 node_id)
1935 struct sbridge_dev *sbridge_dev;
1937 list_for_each_entry(sbridge_dev, &sbridge_edac_list, list) {
1938 if (sbridge_dev->node_id == node_id)
1939 return sbridge_dev->mci;
1941 return NULL;
1944 static int get_memory_error_data(struct mem_ctl_info *mci,
1945 u64 addr,
1946 u8 *socket, u8 *ha,
1947 long *channel_mask,
1948 u8 *rank,
1949 char **area_type, char *msg)
1951 struct mem_ctl_info *new_mci;
1952 struct sbridge_pvt *pvt = mci->pvt_info;
1953 struct pci_dev *pci_ha;
1954 int n_rir, n_sads, n_tads, sad_way, sck_xch;
1955 int sad_interl, idx, base_ch;
1956 int interleave_mode, shiftup = 0;
1957 unsigned sad_interleave[pvt->info.max_interleave];
1958 u32 reg, dram_rule;
1959 u8 ch_way, sck_way, pkg, sad_ha = 0, ch_add = 0;
1960 u32 tad_offset;
1961 u32 rir_way;
1962 u32 mb, gb;
1963 u64 ch_addr, offset, limit = 0, prv = 0;
1967 * Step 0) Check if the address is at special memory ranges
1968 * The check bellow is probably enough to fill all cases where
1969 * the error is not inside a memory, except for the legacy
1970 * range (e. g. VGA addresses). It is unlikely, however, that the
1971 * memory controller would generate an error on that range.
1973 if ((addr > (u64) pvt->tolm) && (addr < (1LL << 32))) {
1974 sprintf(msg, "Error at TOLM area, on addr 0x%08Lx", addr);
1975 return -EINVAL;
1977 if (addr >= (u64)pvt->tohm) {
1978 sprintf(msg, "Error at MMIOH area, on addr 0x%016Lx", addr);
1979 return -EINVAL;
1983 * Step 1) Get socket
1985 for (n_sads = 0; n_sads < pvt->info.max_sad; n_sads++) {
1986 pci_read_config_dword(pvt->pci_sad0, pvt->info.dram_rule[n_sads],
1987 &reg);
1989 if (!DRAM_RULE_ENABLE(reg))
1990 continue;
1992 limit = pvt->info.sad_limit(reg);
1993 if (limit <= prv) {
1994 sprintf(msg, "Can't discover the memory socket");
1995 return -EINVAL;
1997 if (addr <= limit)
1998 break;
1999 prv = limit;
2001 if (n_sads == pvt->info.max_sad) {
2002 sprintf(msg, "Can't discover the memory socket");
2003 return -EINVAL;
2005 dram_rule = reg;
2006 *area_type = show_dram_attr(pvt->info.dram_attr(dram_rule));
2007 interleave_mode = pvt->info.interleave_mode(dram_rule);
2009 pci_read_config_dword(pvt->pci_sad0, pvt->info.interleave_list[n_sads],
2010 &reg);
2012 if (pvt->info.type == SANDY_BRIDGE) {
2013 sad_interl = sad_pkg(pvt->info.interleave_pkg, reg, 0);
2014 for (sad_way = 0; sad_way < 8; sad_way++) {
2015 u32 pkg = sad_pkg(pvt->info.interleave_pkg, reg, sad_way);
2016 if (sad_way > 0 && sad_interl == pkg)
2017 break;
2018 sad_interleave[sad_way] = pkg;
2019 edac_dbg(0, "SAD interleave #%d: %d\n",
2020 sad_way, sad_interleave[sad_way]);
2022 edac_dbg(0, "mc#%d: Error detected on SAD#%d: address 0x%016Lx < 0x%016Lx, Interleave [%d:6]%s\n",
2023 pvt->sbridge_dev->mc,
2024 n_sads,
2025 addr,
2026 limit,
2027 sad_way + 7,
2028 !interleave_mode ? "" : "XOR[18:16]");
2029 if (interleave_mode)
2030 idx = ((addr >> 6) ^ (addr >> 16)) & 7;
2031 else
2032 idx = (addr >> 6) & 7;
2033 switch (sad_way) {
2034 case 1:
2035 idx = 0;
2036 break;
2037 case 2:
2038 idx = idx & 1;
2039 break;
2040 case 4:
2041 idx = idx & 3;
2042 break;
2043 case 8:
2044 break;
2045 default:
2046 sprintf(msg, "Can't discover socket interleave");
2047 return -EINVAL;
2049 *socket = sad_interleave[idx];
2050 edac_dbg(0, "SAD interleave index: %d (wayness %d) = CPU socket %d\n",
2051 idx, sad_way, *socket);
2052 } else if (pvt->info.type == HASWELL || pvt->info.type == BROADWELL) {
2053 int bits, a7mode = A7MODE(dram_rule);
2055 if (a7mode) {
2056 /* A7 mode swaps P9 with P6 */
2057 bits = GET_BITFIELD(addr, 7, 8) << 1;
2058 bits |= GET_BITFIELD(addr, 9, 9);
2059 } else
2060 bits = GET_BITFIELD(addr, 6, 8);
2062 if (interleave_mode == 0) {
2063 /* interleave mode will XOR {8,7,6} with {18,17,16} */
2064 idx = GET_BITFIELD(addr, 16, 18);
2065 idx ^= bits;
2066 } else
2067 idx = bits;
2069 pkg = sad_pkg(pvt->info.interleave_pkg, reg, idx);
2070 *socket = sad_pkg_socket(pkg);
2071 sad_ha = sad_pkg_ha(pkg);
2072 if (sad_ha)
2073 ch_add = 4;
2075 if (a7mode) {
2076 /* MCChanShiftUpEnable */
2077 pci_read_config_dword(pvt->pci_ha0,
2078 HASWELL_HASYSDEFEATURE2, &reg);
2079 shiftup = GET_BITFIELD(reg, 22, 22);
2082 edac_dbg(0, "SAD interleave package: %d = CPU socket %d, HA %i, shiftup: %i\n",
2083 idx, *socket, sad_ha, shiftup);
2084 } else {
2085 /* Ivy Bridge's SAD mode doesn't support XOR interleave mode */
2086 idx = (addr >> 6) & 7;
2087 pkg = sad_pkg(pvt->info.interleave_pkg, reg, idx);
2088 *socket = sad_pkg_socket(pkg);
2089 sad_ha = sad_pkg_ha(pkg);
2090 if (sad_ha)
2091 ch_add = 4;
2092 edac_dbg(0, "SAD interleave package: %d = CPU socket %d, HA %d\n",
2093 idx, *socket, sad_ha);
2096 *ha = sad_ha;
2099 * Move to the proper node structure, in order to access the
2100 * right PCI registers
2102 new_mci = get_mci_for_node_id(*socket);
2103 if (!new_mci) {
2104 sprintf(msg, "Struct for socket #%u wasn't initialized",
2105 *socket);
2106 return -EINVAL;
2108 mci = new_mci;
2109 pvt = mci->pvt_info;
2112 * Step 2) Get memory channel
2114 prv = 0;
2115 if (pvt->info.type == SANDY_BRIDGE)
2116 pci_ha = pvt->pci_ha0;
2117 else {
2118 if (sad_ha)
2119 pci_ha = pvt->pci_ha1;
2120 else
2121 pci_ha = pvt->pci_ha0;
2123 for (n_tads = 0; n_tads < MAX_TAD; n_tads++) {
2124 pci_read_config_dword(pci_ha, tad_dram_rule[n_tads], &reg);
2125 limit = TAD_LIMIT(reg);
2126 if (limit <= prv) {
2127 sprintf(msg, "Can't discover the memory channel");
2128 return -EINVAL;
2130 if (addr <= limit)
2131 break;
2132 prv = limit;
2134 if (n_tads == MAX_TAD) {
2135 sprintf(msg, "Can't discover the memory channel");
2136 return -EINVAL;
2139 ch_way = TAD_CH(reg) + 1;
2140 sck_way = TAD_SOCK(reg);
2142 if (ch_way == 3)
2143 idx = addr >> 6;
2144 else {
2145 idx = (addr >> (6 + sck_way + shiftup)) & 0x3;
2146 if (pvt->is_chan_hash)
2147 idx = haswell_chan_hash(idx, addr);
2149 idx = idx % ch_way;
2152 * FIXME: Shouldn't we use CHN_IDX_OFFSET() here, when ch_way == 3 ???
2154 switch (idx) {
2155 case 0:
2156 base_ch = TAD_TGT0(reg);
2157 break;
2158 case 1:
2159 base_ch = TAD_TGT1(reg);
2160 break;
2161 case 2:
2162 base_ch = TAD_TGT2(reg);
2163 break;
2164 case 3:
2165 base_ch = TAD_TGT3(reg);
2166 break;
2167 default:
2168 sprintf(msg, "Can't discover the TAD target");
2169 return -EINVAL;
2171 *channel_mask = 1 << base_ch;
2173 pci_read_config_dword(pvt->pci_tad[ch_add + base_ch],
2174 tad_ch_nilv_offset[n_tads],
2175 &tad_offset);
2177 if (pvt->is_mirrored) {
2178 *channel_mask |= 1 << ((base_ch + 2) % 4);
2179 switch(ch_way) {
2180 case 2:
2181 case 4:
2182 sck_xch = (1 << sck_way) * (ch_way >> 1);
2183 break;
2184 default:
2185 sprintf(msg, "Invalid mirror set. Can't decode addr");
2186 return -EINVAL;
2188 } else
2189 sck_xch = (1 << sck_way) * ch_way;
2191 if (pvt->is_lockstep)
2192 *channel_mask |= 1 << ((base_ch + 1) % 4);
2194 offset = TAD_OFFSET(tad_offset);
2196 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",
2197 n_tads,
2198 addr,
2199 limit,
2200 sck_way,
2201 ch_way,
2202 offset,
2203 idx,
2204 base_ch,
2205 *channel_mask);
2207 /* Calculate channel address */
2208 /* Remove the TAD offset */
2210 if (offset > addr) {
2211 sprintf(msg, "Can't calculate ch addr: TAD offset 0x%08Lx is too high for addr 0x%08Lx!",
2212 offset, addr);
2213 return -EINVAL;
2216 ch_addr = addr - offset;
2217 ch_addr >>= (6 + shiftup);
2218 ch_addr /= sck_xch;
2219 ch_addr <<= (6 + shiftup);
2220 ch_addr |= addr & ((1 << (6 + shiftup)) - 1);
2223 * Step 3) Decode rank
2225 for (n_rir = 0; n_rir < MAX_RIR_RANGES; n_rir++) {
2226 pci_read_config_dword(pvt->pci_tad[ch_add + base_ch],
2227 rir_way_limit[n_rir],
2228 &reg);
2230 if (!IS_RIR_VALID(reg))
2231 continue;
2233 limit = pvt->info.rir_limit(reg);
2234 gb = div_u64_rem(limit >> 20, 1024, &mb);
2235 edac_dbg(0, "RIR#%d, limit: %u.%03u GB (0x%016Lx), way: %d\n",
2236 n_rir,
2237 gb, (mb*1000)/1024,
2238 limit,
2239 1 << RIR_WAY(reg));
2240 if (ch_addr <= limit)
2241 break;
2243 if (n_rir == MAX_RIR_RANGES) {
2244 sprintf(msg, "Can't discover the memory rank for ch addr 0x%08Lx",
2245 ch_addr);
2246 return -EINVAL;
2248 rir_way = RIR_WAY(reg);
2250 if (pvt->is_close_pg)
2251 idx = (ch_addr >> 6);
2252 else
2253 idx = (ch_addr >> 13); /* FIXME: Datasheet says to shift by 15 */
2254 idx %= 1 << rir_way;
2256 pci_read_config_dword(pvt->pci_tad[ch_add + base_ch],
2257 rir_offset[n_rir][idx],
2258 &reg);
2259 *rank = RIR_RNK_TGT(reg);
2261 edac_dbg(0, "RIR#%d: channel address 0x%08Lx < 0x%08Lx, RIR interleave %d, index %d\n",
2262 n_rir,
2263 ch_addr,
2264 limit,
2265 rir_way,
2266 idx);
2268 return 0;
2271 /****************************************************************************
2272 Device initialization routines: put/get, init/exit
2273 ****************************************************************************/
2276 * sbridge_put_all_devices 'put' all the devices that we have
2277 * reserved via 'get'
2279 static void sbridge_put_devices(struct sbridge_dev *sbridge_dev)
2281 int i;
2283 edac_dbg(0, "\n");
2284 for (i = 0; i < sbridge_dev->n_devs; i++) {
2285 struct pci_dev *pdev = sbridge_dev->pdev[i];
2286 if (!pdev)
2287 continue;
2288 edac_dbg(0, "Removing dev %02x:%02x.%d\n",
2289 pdev->bus->number,
2290 PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn));
2291 pci_dev_put(pdev);
2295 static void sbridge_put_all_devices(void)
2297 struct sbridge_dev *sbridge_dev, *tmp;
2299 list_for_each_entry_safe(sbridge_dev, tmp, &sbridge_edac_list, list) {
2300 sbridge_put_devices(sbridge_dev);
2301 free_sbridge_dev(sbridge_dev);
2305 static int sbridge_get_onedevice(struct pci_dev **prev,
2306 u8 *num_mc,
2307 const struct pci_id_table *table,
2308 const unsigned devno,
2309 const int multi_bus)
2311 struct sbridge_dev *sbridge_dev;
2312 const struct pci_id_descr *dev_descr = &table->descr[devno];
2313 struct pci_dev *pdev = NULL;
2314 u8 bus = 0;
2316 sbridge_printk(KERN_DEBUG,
2317 "Seeking for: PCI ID %04x:%04x\n",
2318 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
2320 pdev = pci_get_device(PCI_VENDOR_ID_INTEL,
2321 dev_descr->dev_id, *prev);
2323 if (!pdev) {
2324 if (*prev) {
2325 *prev = pdev;
2326 return 0;
2329 if (dev_descr->optional)
2330 return 0;
2332 /* if the HA wasn't found */
2333 if (devno == 0)
2334 return -ENODEV;
2336 sbridge_printk(KERN_INFO,
2337 "Device not found: %04x:%04x\n",
2338 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
2340 /* End of list, leave */
2341 return -ENODEV;
2343 bus = pdev->bus->number;
2345 sbridge_dev = get_sbridge_dev(bus, multi_bus);
2346 if (!sbridge_dev) {
2347 sbridge_dev = alloc_sbridge_dev(bus, table);
2348 if (!sbridge_dev) {
2349 pci_dev_put(pdev);
2350 return -ENOMEM;
2352 (*num_mc)++;
2355 if (sbridge_dev->pdev[devno]) {
2356 sbridge_printk(KERN_ERR,
2357 "Duplicated device for %04x:%04x\n",
2358 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
2359 pci_dev_put(pdev);
2360 return -ENODEV;
2363 sbridge_dev->pdev[devno] = pdev;
2365 /* Be sure that the device is enabled */
2366 if (unlikely(pci_enable_device(pdev) < 0)) {
2367 sbridge_printk(KERN_ERR,
2368 "Couldn't enable %04x:%04x\n",
2369 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
2370 return -ENODEV;
2373 edac_dbg(0, "Detected %04x:%04x\n",
2374 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
2377 * As stated on drivers/pci/search.c, the reference count for
2378 * @from is always decremented if it is not %NULL. So, as we need
2379 * to get all devices up to null, we need to do a get for the device
2381 pci_dev_get(pdev);
2383 *prev = pdev;
2385 return 0;
2389 * sbridge_get_all_devices - Find and perform 'get' operation on the MCH's
2390 * devices we want to reference for this driver.
2391 * @num_mc: pointer to the memory controllers count, to be incremented in case
2392 * of success.
2393 * @table: model specific table
2394 * @allow_dups: allow for multiple devices to exist with the same device id
2395 * (as implemented, this isn't expected to work correctly in the
2396 * multi-socket case).
2397 * @multi_bus: don't assume devices on different buses belong to different
2398 * memory controllers.
2400 * returns 0 in case of success or error code
2402 static int sbridge_get_all_devices_full(u8 *num_mc,
2403 const struct pci_id_table *table,
2404 int allow_dups,
2405 int multi_bus)
2407 int i, rc;
2408 struct pci_dev *pdev = NULL;
2410 while (table && table->descr) {
2411 for (i = 0; i < table->n_devs; i++) {
2412 if (!allow_dups || i == 0 ||
2413 table->descr[i].dev_id !=
2414 table->descr[i-1].dev_id) {
2415 pdev = NULL;
2417 do {
2418 rc = sbridge_get_onedevice(&pdev, num_mc,
2419 table, i, multi_bus);
2420 if (rc < 0) {
2421 if (i == 0) {
2422 i = table->n_devs;
2423 break;
2425 sbridge_put_all_devices();
2426 return -ENODEV;
2428 } while (pdev && !allow_dups);
2430 table++;
2433 return 0;
2436 #define sbridge_get_all_devices(num_mc, table) \
2437 sbridge_get_all_devices_full(num_mc, table, 0, 0)
2438 #define sbridge_get_all_devices_knl(num_mc, table) \
2439 sbridge_get_all_devices_full(num_mc, table, 1, 1)
2441 static int sbridge_mci_bind_devs(struct mem_ctl_info *mci,
2442 struct sbridge_dev *sbridge_dev)
2444 struct sbridge_pvt *pvt = mci->pvt_info;
2445 struct pci_dev *pdev;
2446 u8 saw_chan_mask = 0;
2447 int i;
2449 for (i = 0; i < sbridge_dev->n_devs; i++) {
2450 pdev = sbridge_dev->pdev[i];
2451 if (!pdev)
2452 continue;
2454 switch (pdev->device) {
2455 case PCI_DEVICE_ID_INTEL_SBRIDGE_SAD0:
2456 pvt->pci_sad0 = pdev;
2457 break;
2458 case PCI_DEVICE_ID_INTEL_SBRIDGE_SAD1:
2459 pvt->pci_sad1 = pdev;
2460 break;
2461 case PCI_DEVICE_ID_INTEL_SBRIDGE_BR:
2462 pvt->pci_br0 = pdev;
2463 break;
2464 case PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_HA0:
2465 pvt->pci_ha0 = pdev;
2466 break;
2467 case PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TA:
2468 pvt->pci_ta = pdev;
2469 break;
2470 case PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_RAS:
2471 pvt->pci_ras = pdev;
2472 break;
2473 case PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD0:
2474 case PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD1:
2475 case PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD2:
2476 case PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD3:
2478 int id = pdev->device - PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD0;
2479 pvt->pci_tad[id] = pdev;
2480 saw_chan_mask |= 1 << id;
2482 break;
2483 case PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_DDRIO:
2484 pvt->pci_ddrio = pdev;
2485 break;
2486 default:
2487 goto error;
2490 edac_dbg(0, "Associated PCI %02x:%02x, bus %d with dev = %p\n",
2491 pdev->vendor, pdev->device,
2492 sbridge_dev->bus,
2493 pdev);
2496 /* Check if everything were registered */
2497 if (!pvt->pci_sad0 || !pvt->pci_sad1 || !pvt->pci_ha0 ||
2498 !pvt-> pci_tad || !pvt->pci_ras || !pvt->pci_ta)
2499 goto enodev;
2501 if (saw_chan_mask != 0x0f)
2502 goto enodev;
2503 return 0;
2505 enodev:
2506 sbridge_printk(KERN_ERR, "Some needed devices are missing\n");
2507 return -ENODEV;
2509 error:
2510 sbridge_printk(KERN_ERR, "Unexpected device %02x:%02x\n",
2511 PCI_VENDOR_ID_INTEL, pdev->device);
2512 return -EINVAL;
2515 static int ibridge_mci_bind_devs(struct mem_ctl_info *mci,
2516 struct sbridge_dev *sbridge_dev)
2518 struct sbridge_pvt *pvt = mci->pvt_info;
2519 struct pci_dev *pdev;
2520 u8 saw_chan_mask = 0;
2521 int i;
2523 for (i = 0; i < sbridge_dev->n_devs; i++) {
2524 pdev = sbridge_dev->pdev[i];
2525 if (!pdev)
2526 continue;
2528 switch (pdev->device) {
2529 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0:
2530 pvt->pci_ha0 = pdev;
2531 break;
2532 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TA:
2533 pvt->pci_ta = pdev;
2534 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_RAS:
2535 pvt->pci_ras = pdev;
2536 break;
2537 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD0:
2538 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD1:
2539 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD2:
2540 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD3:
2542 int id = pdev->device - PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD0;
2543 pvt->pci_tad[id] = pdev;
2544 saw_chan_mask |= 1 << id;
2546 break;
2547 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_2HA_DDRIO0:
2548 pvt->pci_ddrio = pdev;
2549 break;
2550 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_1HA_DDRIO0:
2551 pvt->pci_ddrio = pdev;
2552 break;
2553 case PCI_DEVICE_ID_INTEL_IBRIDGE_SAD:
2554 pvt->pci_sad0 = pdev;
2555 break;
2556 case PCI_DEVICE_ID_INTEL_IBRIDGE_BR0:
2557 pvt->pci_br0 = pdev;
2558 break;
2559 case PCI_DEVICE_ID_INTEL_IBRIDGE_BR1:
2560 pvt->pci_br1 = pdev;
2561 break;
2562 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1:
2563 pvt->pci_ha1 = pdev;
2564 break;
2565 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD0:
2566 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD1:
2567 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD2:
2568 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD3:
2570 int id = pdev->device - PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD0 + 4;
2571 pvt->pci_tad[id] = pdev;
2572 saw_chan_mask |= 1 << id;
2574 break;
2575 default:
2576 goto error;
2579 edac_dbg(0, "Associated PCI %02x.%02d.%d with dev = %p\n",
2580 sbridge_dev->bus,
2581 PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn),
2582 pdev);
2585 /* Check if everything were registered */
2586 if (!pvt->pci_sad0 || !pvt->pci_ha0 || !pvt->pci_br0 ||
2587 !pvt->pci_br1 || !pvt->pci_tad || !pvt->pci_ras ||
2588 !pvt->pci_ta)
2589 goto enodev;
2591 if (saw_chan_mask != 0x0f && /* -EN */
2592 saw_chan_mask != 0x33 && /* -EP */
2593 saw_chan_mask != 0xff) /* -EX */
2594 goto enodev;
2595 return 0;
2597 enodev:
2598 sbridge_printk(KERN_ERR, "Some needed devices are missing\n");
2599 return -ENODEV;
2601 error:
2602 sbridge_printk(KERN_ERR,
2603 "Unexpected device %02x:%02x\n", PCI_VENDOR_ID_INTEL,
2604 pdev->device);
2605 return -EINVAL;
2608 static int haswell_mci_bind_devs(struct mem_ctl_info *mci,
2609 struct sbridge_dev *sbridge_dev)
2611 struct sbridge_pvt *pvt = mci->pvt_info;
2612 struct pci_dev *pdev;
2613 u8 saw_chan_mask = 0;
2614 int i;
2616 /* there's only one device per system; not tied to any bus */
2617 if (pvt->info.pci_vtd == NULL)
2618 /* result will be checked later */
2619 pvt->info.pci_vtd = pci_get_device(PCI_VENDOR_ID_INTEL,
2620 PCI_DEVICE_ID_INTEL_HASWELL_IMC_VTD_MISC,
2621 NULL);
2623 for (i = 0; i < sbridge_dev->n_devs; i++) {
2624 pdev = sbridge_dev->pdev[i];
2625 if (!pdev)
2626 continue;
2628 switch (pdev->device) {
2629 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_CBO_SAD0:
2630 pvt->pci_sad0 = pdev;
2631 break;
2632 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_CBO_SAD1:
2633 pvt->pci_sad1 = pdev;
2634 break;
2635 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0:
2636 pvt->pci_ha0 = pdev;
2637 break;
2638 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TA:
2639 pvt->pci_ta = pdev;
2640 break;
2641 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_THERMAL:
2642 pvt->pci_ras = pdev;
2643 break;
2644 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD0:
2645 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD1:
2646 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD2:
2647 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD3:
2649 int id = pdev->device - PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD0;
2651 pvt->pci_tad[id] = pdev;
2652 saw_chan_mask |= 1 << id;
2654 break;
2655 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD0:
2656 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD1:
2657 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD2:
2658 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD3:
2660 int id = pdev->device - PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD0 + 4;
2662 pvt->pci_tad[id] = pdev;
2663 saw_chan_mask |= 1 << id;
2665 break;
2666 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO0:
2667 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO1:
2668 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO2:
2669 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO3:
2670 if (!pvt->pci_ddrio)
2671 pvt->pci_ddrio = pdev;
2672 break;
2673 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1:
2674 pvt->pci_ha1 = pdev;
2675 break;
2676 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TA:
2677 pvt->pci_ha1_ta = pdev;
2678 break;
2679 default:
2680 break;
2683 edac_dbg(0, "Associated PCI %02x.%02d.%d with dev = %p\n",
2684 sbridge_dev->bus,
2685 PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn),
2686 pdev);
2689 /* Check if everything were registered */
2690 if (!pvt->pci_sad0 || !pvt->pci_ha0 || !pvt->pci_sad1 ||
2691 !pvt->pci_ras || !pvt->pci_ta || !pvt->info.pci_vtd)
2692 goto enodev;
2694 if (saw_chan_mask != 0x0f && /* -EN */
2695 saw_chan_mask != 0x33 && /* -EP */
2696 saw_chan_mask != 0xff) /* -EX */
2697 goto enodev;
2698 return 0;
2700 enodev:
2701 sbridge_printk(KERN_ERR, "Some needed devices are missing\n");
2702 return -ENODEV;
2705 static int broadwell_mci_bind_devs(struct mem_ctl_info *mci,
2706 struct sbridge_dev *sbridge_dev)
2708 struct sbridge_pvt *pvt = mci->pvt_info;
2709 struct pci_dev *pdev;
2710 u8 saw_chan_mask = 0;
2711 int i;
2713 /* there's only one device per system; not tied to any bus */
2714 if (pvt->info.pci_vtd == NULL)
2715 /* result will be checked later */
2716 pvt->info.pci_vtd = pci_get_device(PCI_VENDOR_ID_INTEL,
2717 PCI_DEVICE_ID_INTEL_BROADWELL_IMC_VTD_MISC,
2718 NULL);
2720 for (i = 0; i < sbridge_dev->n_devs; i++) {
2721 pdev = sbridge_dev->pdev[i];
2722 if (!pdev)
2723 continue;
2725 switch (pdev->device) {
2726 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_CBO_SAD0:
2727 pvt->pci_sad0 = pdev;
2728 break;
2729 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_CBO_SAD1:
2730 pvt->pci_sad1 = pdev;
2731 break;
2732 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0:
2733 pvt->pci_ha0 = pdev;
2734 break;
2735 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TA:
2736 pvt->pci_ta = pdev;
2737 break;
2738 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_THERMAL:
2739 pvt->pci_ras = pdev;
2740 break;
2741 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD0:
2742 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD1:
2743 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD2:
2744 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD3:
2746 int id = pdev->device - PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD0;
2747 pvt->pci_tad[id] = pdev;
2748 saw_chan_mask |= 1 << id;
2750 break;
2751 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD0:
2752 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD1:
2753 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD2:
2754 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD3:
2756 int id = pdev->device - PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD0 + 4;
2757 pvt->pci_tad[id] = pdev;
2758 saw_chan_mask |= 1 << id;
2760 break;
2761 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_DDRIO0:
2762 pvt->pci_ddrio = pdev;
2763 break;
2764 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1:
2765 pvt->pci_ha1 = pdev;
2766 break;
2767 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TA:
2768 pvt->pci_ha1_ta = pdev;
2769 break;
2770 default:
2771 break;
2774 edac_dbg(0, "Associated PCI %02x.%02d.%d with dev = %p\n",
2775 sbridge_dev->bus,
2776 PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn),
2777 pdev);
2780 /* Check if everything were registered */
2781 if (!pvt->pci_sad0 || !pvt->pci_ha0 || !pvt->pci_sad1 ||
2782 !pvt->pci_ras || !pvt->pci_ta || !pvt->info.pci_vtd)
2783 goto enodev;
2785 if (saw_chan_mask != 0x0f && /* -EN */
2786 saw_chan_mask != 0x33 && /* -EP */
2787 saw_chan_mask != 0xff) /* -EX */
2788 goto enodev;
2789 return 0;
2791 enodev:
2792 sbridge_printk(KERN_ERR, "Some needed devices are missing\n");
2793 return -ENODEV;
2796 static int knl_mci_bind_devs(struct mem_ctl_info *mci,
2797 struct sbridge_dev *sbridge_dev)
2799 struct sbridge_pvt *pvt = mci->pvt_info;
2800 struct pci_dev *pdev;
2801 int dev, func;
2803 int i;
2804 int devidx;
2806 for (i = 0; i < sbridge_dev->n_devs; i++) {
2807 pdev = sbridge_dev->pdev[i];
2808 if (!pdev)
2809 continue;
2811 /* Extract PCI device and function. */
2812 dev = (pdev->devfn >> 3) & 0x1f;
2813 func = pdev->devfn & 0x7;
2815 switch (pdev->device) {
2816 case PCI_DEVICE_ID_INTEL_KNL_IMC_MC:
2817 if (dev == 8)
2818 pvt->knl.pci_mc0 = pdev;
2819 else if (dev == 9)
2820 pvt->knl.pci_mc1 = pdev;
2821 else {
2822 sbridge_printk(KERN_ERR,
2823 "Memory controller in unexpected place! (dev %d, fn %d)\n",
2824 dev, func);
2825 continue;
2827 break;
2829 case PCI_DEVICE_ID_INTEL_KNL_IMC_SAD0:
2830 pvt->pci_sad0 = pdev;
2831 break;
2833 case PCI_DEVICE_ID_INTEL_KNL_IMC_SAD1:
2834 pvt->pci_sad1 = pdev;
2835 break;
2837 case PCI_DEVICE_ID_INTEL_KNL_IMC_CHA:
2838 /* There are one of these per tile, and range from
2839 * 1.14.0 to 1.18.5.
2841 devidx = ((dev-14)*8)+func;
2843 if (devidx < 0 || devidx >= KNL_MAX_CHAS) {
2844 sbridge_printk(KERN_ERR,
2845 "Caching and Home Agent in unexpected place! (dev %d, fn %d)\n",
2846 dev, func);
2847 continue;
2850 WARN_ON(pvt->knl.pci_cha[devidx] != NULL);
2852 pvt->knl.pci_cha[devidx] = pdev;
2853 break;
2855 case PCI_DEVICE_ID_INTEL_KNL_IMC_CHANNEL:
2856 devidx = -1;
2859 * MC0 channels 0-2 are device 9 function 2-4,
2860 * MC1 channels 3-5 are device 8 function 2-4.
2863 if (dev == 9)
2864 devidx = func-2;
2865 else if (dev == 8)
2866 devidx = 3 + (func-2);
2868 if (devidx < 0 || devidx >= KNL_MAX_CHANNELS) {
2869 sbridge_printk(KERN_ERR,
2870 "DRAM Channel Registers in unexpected place! (dev %d, fn %d)\n",
2871 dev, func);
2872 continue;
2875 WARN_ON(pvt->knl.pci_channel[devidx] != NULL);
2876 pvt->knl.pci_channel[devidx] = pdev;
2877 break;
2879 case PCI_DEVICE_ID_INTEL_KNL_IMC_TOLHM:
2880 pvt->knl.pci_mc_info = pdev;
2881 break;
2883 case PCI_DEVICE_ID_INTEL_KNL_IMC_TA:
2884 pvt->pci_ta = pdev;
2885 break;
2887 default:
2888 sbridge_printk(KERN_ERR, "Unexpected device %d\n",
2889 pdev->device);
2890 break;
2894 if (!pvt->knl.pci_mc0 || !pvt->knl.pci_mc1 ||
2895 !pvt->pci_sad0 || !pvt->pci_sad1 ||
2896 !pvt->pci_ta) {
2897 goto enodev;
2900 for (i = 0; i < KNL_MAX_CHANNELS; i++) {
2901 if (!pvt->knl.pci_channel[i]) {
2902 sbridge_printk(KERN_ERR, "Missing channel %d\n", i);
2903 goto enodev;
2907 for (i = 0; i < KNL_MAX_CHAS; i++) {
2908 if (!pvt->knl.pci_cha[i]) {
2909 sbridge_printk(KERN_ERR, "Missing CHA %d\n", i);
2910 goto enodev;
2914 return 0;
2916 enodev:
2917 sbridge_printk(KERN_ERR, "Some needed devices are missing\n");
2918 return -ENODEV;
2921 /****************************************************************************
2922 Error check routines
2923 ****************************************************************************/
2926 * While Sandy Bridge has error count registers, SMI BIOS read values from
2927 * and resets the counters. So, they are not reliable for the OS to read
2928 * from them. So, we have no option but to just trust on whatever MCE is
2929 * telling us about the errors.
2931 static void sbridge_mce_output_error(struct mem_ctl_info *mci,
2932 const struct mce *m)
2934 struct mem_ctl_info *new_mci;
2935 struct sbridge_pvt *pvt = mci->pvt_info;
2936 enum hw_event_mc_err_type tp_event;
2937 char *type, *optype, msg[256];
2938 bool ripv = GET_BITFIELD(m->mcgstatus, 0, 0);
2939 bool overflow = GET_BITFIELD(m->status, 62, 62);
2940 bool uncorrected_error = GET_BITFIELD(m->status, 61, 61);
2941 bool recoverable;
2942 u32 core_err_cnt = GET_BITFIELD(m->status, 38, 52);
2943 u32 mscod = GET_BITFIELD(m->status, 16, 31);
2944 u32 errcode = GET_BITFIELD(m->status, 0, 15);
2945 u32 channel = GET_BITFIELD(m->status, 0, 3);
2946 u32 optypenum = GET_BITFIELD(m->status, 4, 6);
2947 long channel_mask, first_channel;
2948 u8 rank, socket, ha;
2949 int rc, dimm;
2950 char *area_type = NULL;
2952 if (pvt->info.type != SANDY_BRIDGE)
2953 recoverable = true;
2954 else
2955 recoverable = GET_BITFIELD(m->status, 56, 56);
2957 if (uncorrected_error) {
2958 if (ripv) {
2959 type = "FATAL";
2960 tp_event = HW_EVENT_ERR_FATAL;
2961 } else {
2962 type = "NON_FATAL";
2963 tp_event = HW_EVENT_ERR_UNCORRECTED;
2965 } else {
2966 type = "CORRECTED";
2967 tp_event = HW_EVENT_ERR_CORRECTED;
2971 * According with Table 15-9 of the Intel Architecture spec vol 3A,
2972 * memory errors should fit in this mask:
2973 * 000f 0000 1mmm cccc (binary)
2974 * where:
2975 * f = Correction Report Filtering Bit. If 1, subsequent errors
2976 * won't be shown
2977 * mmm = error type
2978 * cccc = channel
2979 * If the mask doesn't match, report an error to the parsing logic
2981 if (! ((errcode & 0xef80) == 0x80)) {
2982 optype = "Can't parse: it is not a mem";
2983 } else {
2984 switch (optypenum) {
2985 case 0:
2986 optype = "generic undef request error";
2987 break;
2988 case 1:
2989 optype = "memory read error";
2990 break;
2991 case 2:
2992 optype = "memory write error";
2993 break;
2994 case 3:
2995 optype = "addr/cmd error";
2996 break;
2997 case 4:
2998 optype = "memory scrubbing error";
2999 break;
3000 default:
3001 optype = "reserved";
3002 break;
3006 /* Only decode errors with an valid address (ADDRV) */
3007 if (!GET_BITFIELD(m->status, 58, 58))
3008 return;
3010 if (pvt->info.type == KNIGHTS_LANDING) {
3011 if (channel == 14) {
3012 edac_dbg(0, "%s%s err_code:%04x:%04x EDRAM bank %d\n",
3013 overflow ? " OVERFLOW" : "",
3014 (uncorrected_error && recoverable)
3015 ? " recoverable" : "",
3016 mscod, errcode,
3017 m->bank);
3018 } else {
3019 char A = *("A");
3021 channel = knl_channel_remap(channel);
3022 channel_mask = 1 << channel;
3023 snprintf(msg, sizeof(msg),
3024 "%s%s err_code:%04x:%04x channel:%d (DIMM_%c)",
3025 overflow ? " OVERFLOW" : "",
3026 (uncorrected_error && recoverable)
3027 ? " recoverable" : " ",
3028 mscod, errcode, channel, A + channel);
3029 edac_mc_handle_error(tp_event, mci, core_err_cnt,
3030 m->addr >> PAGE_SHIFT, m->addr & ~PAGE_MASK, 0,
3031 channel, 0, -1,
3032 optype, msg);
3034 return;
3035 } else {
3036 rc = get_memory_error_data(mci, m->addr, &socket, &ha,
3037 &channel_mask, &rank, &area_type, msg);
3040 if (rc < 0)
3041 goto err_parsing;
3042 new_mci = get_mci_for_node_id(socket);
3043 if (!new_mci) {
3044 strcpy(msg, "Error: socket got corrupted!");
3045 goto err_parsing;
3047 mci = new_mci;
3048 pvt = mci->pvt_info;
3050 first_channel = find_first_bit(&channel_mask, NUM_CHANNELS);
3052 if (rank < 4)
3053 dimm = 0;
3054 else if (rank < 8)
3055 dimm = 1;
3056 else
3057 dimm = 2;
3061 * FIXME: On some memory configurations (mirror, lockstep), the
3062 * Memory Controller can't point the error to a single DIMM. The
3063 * EDAC core should be handling the channel mask, in order to point
3064 * to the group of dimm's where the error may be happening.
3066 if (!pvt->is_lockstep && !pvt->is_mirrored && !pvt->is_close_pg)
3067 channel = first_channel;
3069 snprintf(msg, sizeof(msg),
3070 "%s%s area:%s err_code:%04x:%04x socket:%d ha:%d channel_mask:%ld rank:%d",
3071 overflow ? " OVERFLOW" : "",
3072 (uncorrected_error && recoverable) ? " recoverable" : "",
3073 area_type,
3074 mscod, errcode,
3075 socket, ha,
3076 channel_mask,
3077 rank);
3079 edac_dbg(0, "%s\n", msg);
3081 /* FIXME: need support for channel mask */
3083 if (channel == CHANNEL_UNSPECIFIED)
3084 channel = -1;
3086 /* Call the helper to output message */
3087 edac_mc_handle_error(tp_event, mci, core_err_cnt,
3088 m->addr >> PAGE_SHIFT, m->addr & ~PAGE_MASK, 0,
3089 4*ha+channel, dimm, -1,
3090 optype, msg);
3091 return;
3092 err_parsing:
3093 edac_mc_handle_error(tp_event, mci, core_err_cnt, 0, 0, 0,
3094 -1, -1, -1,
3095 msg, "");
3100 * sbridge_check_error Retrieve and process errors reported by the
3101 * hardware. Called by the Core module.
3103 static void sbridge_check_error(struct mem_ctl_info *mci)
3105 struct sbridge_pvt *pvt = mci->pvt_info;
3106 int i;
3107 unsigned count = 0;
3108 struct mce *m;
3111 * MCE first step: Copy all mce errors into a temporary buffer
3112 * We use a double buffering here, to reduce the risk of
3113 * loosing an error.
3115 smp_rmb();
3116 count = (pvt->mce_out + MCE_LOG_LEN - pvt->mce_in)
3117 % MCE_LOG_LEN;
3118 if (!count)
3119 return;
3121 m = pvt->mce_outentry;
3122 if (pvt->mce_in + count > MCE_LOG_LEN) {
3123 unsigned l = MCE_LOG_LEN - pvt->mce_in;
3125 memcpy(m, &pvt->mce_entry[pvt->mce_in], sizeof(*m) * l);
3126 smp_wmb();
3127 pvt->mce_in = 0;
3128 count -= l;
3129 m += l;
3131 memcpy(m, &pvt->mce_entry[pvt->mce_in], sizeof(*m) * count);
3132 smp_wmb();
3133 pvt->mce_in += count;
3135 smp_rmb();
3136 if (pvt->mce_overrun) {
3137 sbridge_printk(KERN_ERR, "Lost %d memory errors\n",
3138 pvt->mce_overrun);
3139 smp_wmb();
3140 pvt->mce_overrun = 0;
3144 * MCE second step: parse errors and display
3146 for (i = 0; i < count; i++)
3147 sbridge_mce_output_error(mci, &pvt->mce_outentry[i]);
3151 * sbridge_mce_check_error Replicates mcelog routine to get errors
3152 * This routine simply queues mcelog errors, and
3153 * return. The error itself should be handled later
3154 * by sbridge_check_error.
3155 * WARNING: As this routine should be called at NMI time, extra care should
3156 * be taken to avoid deadlocks, and to be as fast as possible.
3158 static int sbridge_mce_check_error(struct notifier_block *nb, unsigned long val,
3159 void *data)
3161 struct mce *mce = (struct mce *)data;
3162 struct mem_ctl_info *mci;
3163 struct sbridge_pvt *pvt;
3164 char *type;
3166 if (get_edac_report_status() == EDAC_REPORTING_DISABLED)
3167 return NOTIFY_DONE;
3169 mci = get_mci_for_node_id(mce->socketid);
3170 if (!mci)
3171 return NOTIFY_DONE;
3172 pvt = mci->pvt_info;
3175 * Just let mcelog handle it if the error is
3176 * outside the memory controller. A memory error
3177 * is indicated by bit 7 = 1 and bits = 8-11,13-15 = 0.
3178 * bit 12 has an special meaning.
3180 if ((mce->status & 0xefff) >> 7 != 1)
3181 return NOTIFY_DONE;
3183 if (mce->mcgstatus & MCG_STATUS_MCIP)
3184 type = "Exception";
3185 else
3186 type = "Event";
3188 sbridge_mc_printk(mci, KERN_DEBUG, "HANDLING MCE MEMORY ERROR\n");
3190 sbridge_mc_printk(mci, KERN_DEBUG, "CPU %d: Machine Check %s: %Lx "
3191 "Bank %d: %016Lx\n", mce->extcpu, type,
3192 mce->mcgstatus, mce->bank, mce->status);
3193 sbridge_mc_printk(mci, KERN_DEBUG, "TSC %llx ", mce->tsc);
3194 sbridge_mc_printk(mci, KERN_DEBUG, "ADDR %llx ", mce->addr);
3195 sbridge_mc_printk(mci, KERN_DEBUG, "MISC %llx ", mce->misc);
3197 sbridge_mc_printk(mci, KERN_DEBUG, "PROCESSOR %u:%x TIME %llu SOCKET "
3198 "%u APIC %x\n", mce->cpuvendor, mce->cpuid,
3199 mce->time, mce->socketid, mce->apicid);
3201 smp_rmb();
3202 if ((pvt->mce_out + 1) % MCE_LOG_LEN == pvt->mce_in) {
3203 smp_wmb();
3204 pvt->mce_overrun++;
3205 return NOTIFY_DONE;
3208 /* Copy memory error at the ringbuffer */
3209 memcpy(&pvt->mce_entry[pvt->mce_out], mce, sizeof(*mce));
3210 smp_wmb();
3211 pvt->mce_out = (pvt->mce_out + 1) % MCE_LOG_LEN;
3213 /* Handle fatal errors immediately */
3214 if (mce->mcgstatus & 1)
3215 sbridge_check_error(mci);
3217 /* Advice mcelog that the error were handled */
3218 return NOTIFY_STOP;
3221 static struct notifier_block sbridge_mce_dec = {
3222 .notifier_call = sbridge_mce_check_error,
3225 /****************************************************************************
3226 EDAC register/unregister logic
3227 ****************************************************************************/
3229 static void sbridge_unregister_mci(struct sbridge_dev *sbridge_dev)
3231 struct mem_ctl_info *mci = sbridge_dev->mci;
3232 struct sbridge_pvt *pvt;
3234 if (unlikely(!mci || !mci->pvt_info)) {
3235 edac_dbg(0, "MC: dev = %p\n", &sbridge_dev->pdev[0]->dev);
3237 sbridge_printk(KERN_ERR, "Couldn't find mci handler\n");
3238 return;
3241 pvt = mci->pvt_info;
3243 edac_dbg(0, "MC: mci = %p, dev = %p\n",
3244 mci, &sbridge_dev->pdev[0]->dev);
3246 /* Remove MC sysfs nodes */
3247 edac_mc_del_mc(mci->pdev);
3249 edac_dbg(1, "%s: free mci struct\n", mci->ctl_name);
3250 kfree(mci->ctl_name);
3251 edac_mc_free(mci);
3252 sbridge_dev->mci = NULL;
3255 static int sbridge_register_mci(struct sbridge_dev *sbridge_dev, enum type type)
3257 struct mem_ctl_info *mci;
3258 struct edac_mc_layer layers[2];
3259 struct sbridge_pvt *pvt;
3260 struct pci_dev *pdev = sbridge_dev->pdev[0];
3261 int rc;
3263 /* Check the number of active and not disabled channels */
3264 rc = check_if_ecc_is_active(sbridge_dev->bus, type);
3265 if (unlikely(rc < 0))
3266 return rc;
3268 /* allocate a new MC control structure */
3269 layers[0].type = EDAC_MC_LAYER_CHANNEL;
3270 layers[0].size = type == KNIGHTS_LANDING ?
3271 KNL_MAX_CHANNELS : NUM_CHANNELS;
3272 layers[0].is_virt_csrow = false;
3273 layers[1].type = EDAC_MC_LAYER_SLOT;
3274 layers[1].size = type == KNIGHTS_LANDING ? 1 : MAX_DIMMS;
3275 layers[1].is_virt_csrow = true;
3276 mci = edac_mc_alloc(sbridge_dev->mc, ARRAY_SIZE(layers), layers,
3277 sizeof(*pvt));
3279 if (unlikely(!mci))
3280 return -ENOMEM;
3282 edac_dbg(0, "MC: mci = %p, dev = %p\n",
3283 mci, &pdev->dev);
3285 pvt = mci->pvt_info;
3286 memset(pvt, 0, sizeof(*pvt));
3288 /* Associate sbridge_dev and mci for future usage */
3289 pvt->sbridge_dev = sbridge_dev;
3290 sbridge_dev->mci = mci;
3292 mci->mtype_cap = type == KNIGHTS_LANDING ?
3293 MEM_FLAG_DDR4 : MEM_FLAG_DDR3;
3294 mci->edac_ctl_cap = EDAC_FLAG_NONE;
3295 mci->edac_cap = EDAC_FLAG_NONE;
3296 mci->mod_name = "sbridge_edac.c";
3297 mci->mod_ver = SBRIDGE_REVISION;
3298 mci->dev_name = pci_name(pdev);
3299 mci->ctl_page_to_phys = NULL;
3301 /* Set the function pointer to an actual operation function */
3302 mci->edac_check = sbridge_check_error;
3304 pvt->info.type = type;
3305 switch (type) {
3306 case IVY_BRIDGE:
3307 pvt->info.rankcfgr = IB_RANK_CFG_A;
3308 pvt->info.get_tolm = ibridge_get_tolm;
3309 pvt->info.get_tohm = ibridge_get_tohm;
3310 pvt->info.dram_rule = ibridge_dram_rule;
3311 pvt->info.get_memory_type = get_memory_type;
3312 pvt->info.get_node_id = get_node_id;
3313 pvt->info.rir_limit = rir_limit;
3314 pvt->info.sad_limit = sad_limit;
3315 pvt->info.interleave_mode = interleave_mode;
3316 pvt->info.show_interleave_mode = show_interleave_mode;
3317 pvt->info.dram_attr = dram_attr;
3318 pvt->info.max_sad = ARRAY_SIZE(ibridge_dram_rule);
3319 pvt->info.interleave_list = ibridge_interleave_list;
3320 pvt->info.max_interleave = ARRAY_SIZE(ibridge_interleave_list);
3321 pvt->info.interleave_pkg = ibridge_interleave_pkg;
3322 pvt->info.get_width = ibridge_get_width;
3323 mci->ctl_name = kasprintf(GFP_KERNEL, "Ivy Bridge Socket#%d", mci->mc_idx);
3325 /* Store pci devices at mci for faster access */
3326 rc = ibridge_mci_bind_devs(mci, sbridge_dev);
3327 if (unlikely(rc < 0))
3328 goto fail0;
3329 break;
3330 case SANDY_BRIDGE:
3331 pvt->info.rankcfgr = SB_RANK_CFG_A;
3332 pvt->info.get_tolm = sbridge_get_tolm;
3333 pvt->info.get_tohm = sbridge_get_tohm;
3334 pvt->info.dram_rule = sbridge_dram_rule;
3335 pvt->info.get_memory_type = get_memory_type;
3336 pvt->info.get_node_id = get_node_id;
3337 pvt->info.rir_limit = rir_limit;
3338 pvt->info.sad_limit = sad_limit;
3339 pvt->info.interleave_mode = interleave_mode;
3340 pvt->info.show_interleave_mode = show_interleave_mode;
3341 pvt->info.dram_attr = dram_attr;
3342 pvt->info.max_sad = ARRAY_SIZE(sbridge_dram_rule);
3343 pvt->info.interleave_list = sbridge_interleave_list;
3344 pvt->info.max_interleave = ARRAY_SIZE(sbridge_interleave_list);
3345 pvt->info.interleave_pkg = sbridge_interleave_pkg;
3346 pvt->info.get_width = sbridge_get_width;
3347 mci->ctl_name = kasprintf(GFP_KERNEL, "Sandy Bridge Socket#%d", mci->mc_idx);
3349 /* Store pci devices at mci for faster access */
3350 rc = sbridge_mci_bind_devs(mci, sbridge_dev);
3351 if (unlikely(rc < 0))
3352 goto fail0;
3353 break;
3354 case HASWELL:
3355 /* rankcfgr isn't used */
3356 pvt->info.get_tolm = haswell_get_tolm;
3357 pvt->info.get_tohm = haswell_get_tohm;
3358 pvt->info.dram_rule = ibridge_dram_rule;
3359 pvt->info.get_memory_type = haswell_get_memory_type;
3360 pvt->info.get_node_id = haswell_get_node_id;
3361 pvt->info.rir_limit = haswell_rir_limit;
3362 pvt->info.sad_limit = sad_limit;
3363 pvt->info.interleave_mode = interleave_mode;
3364 pvt->info.show_interleave_mode = show_interleave_mode;
3365 pvt->info.dram_attr = dram_attr;
3366 pvt->info.max_sad = ARRAY_SIZE(ibridge_dram_rule);
3367 pvt->info.interleave_list = ibridge_interleave_list;
3368 pvt->info.max_interleave = ARRAY_SIZE(ibridge_interleave_list);
3369 pvt->info.interleave_pkg = ibridge_interleave_pkg;
3370 pvt->info.get_width = ibridge_get_width;
3371 mci->ctl_name = kasprintf(GFP_KERNEL, "Haswell Socket#%d", mci->mc_idx);
3373 /* Store pci devices at mci for faster access */
3374 rc = haswell_mci_bind_devs(mci, sbridge_dev);
3375 if (unlikely(rc < 0))
3376 goto fail0;
3377 break;
3378 case BROADWELL:
3379 /* rankcfgr isn't used */
3380 pvt->info.get_tolm = haswell_get_tolm;
3381 pvt->info.get_tohm = haswell_get_tohm;
3382 pvt->info.dram_rule = ibridge_dram_rule;
3383 pvt->info.get_memory_type = haswell_get_memory_type;
3384 pvt->info.get_node_id = haswell_get_node_id;
3385 pvt->info.rir_limit = haswell_rir_limit;
3386 pvt->info.sad_limit = sad_limit;
3387 pvt->info.interleave_mode = interleave_mode;
3388 pvt->info.show_interleave_mode = show_interleave_mode;
3389 pvt->info.dram_attr = dram_attr;
3390 pvt->info.max_sad = ARRAY_SIZE(ibridge_dram_rule);
3391 pvt->info.interleave_list = ibridge_interleave_list;
3392 pvt->info.max_interleave = ARRAY_SIZE(ibridge_interleave_list);
3393 pvt->info.interleave_pkg = ibridge_interleave_pkg;
3394 pvt->info.get_width = broadwell_get_width;
3395 mci->ctl_name = kasprintf(GFP_KERNEL, "Broadwell Socket#%d", mci->mc_idx);
3397 /* Store pci devices at mci for faster access */
3398 rc = broadwell_mci_bind_devs(mci, sbridge_dev);
3399 if (unlikely(rc < 0))
3400 goto fail0;
3401 break;
3402 case KNIGHTS_LANDING:
3403 /* pvt->info.rankcfgr == ??? */
3404 pvt->info.get_tolm = knl_get_tolm;
3405 pvt->info.get_tohm = knl_get_tohm;
3406 pvt->info.dram_rule = knl_dram_rule;
3407 pvt->info.get_memory_type = knl_get_memory_type;
3408 pvt->info.get_node_id = knl_get_node_id;
3409 pvt->info.rir_limit = NULL;
3410 pvt->info.sad_limit = knl_sad_limit;
3411 pvt->info.interleave_mode = knl_interleave_mode;
3412 pvt->info.show_interleave_mode = knl_show_interleave_mode;
3413 pvt->info.dram_attr = dram_attr_knl;
3414 pvt->info.max_sad = ARRAY_SIZE(knl_dram_rule);
3415 pvt->info.interleave_list = knl_interleave_list;
3416 pvt->info.max_interleave = ARRAY_SIZE(knl_interleave_list);
3417 pvt->info.interleave_pkg = ibridge_interleave_pkg;
3418 pvt->info.get_width = knl_get_width;
3419 mci->ctl_name = kasprintf(GFP_KERNEL,
3420 "Knights Landing Socket#%d", mci->mc_idx);
3422 rc = knl_mci_bind_devs(mci, sbridge_dev);
3423 if (unlikely(rc < 0))
3424 goto fail0;
3425 break;
3428 /* Get dimm basic config and the memory layout */
3429 get_dimm_config(mci);
3430 get_memory_layout(mci);
3432 /* record ptr to the generic device */
3433 mci->pdev = &pdev->dev;
3435 /* add this new MC control structure to EDAC's list of MCs */
3436 if (unlikely(edac_mc_add_mc(mci))) {
3437 edac_dbg(0, "MC: failed edac_mc_add_mc()\n");
3438 rc = -EINVAL;
3439 goto fail0;
3442 return 0;
3444 fail0:
3445 kfree(mci->ctl_name);
3446 edac_mc_free(mci);
3447 sbridge_dev->mci = NULL;
3448 return rc;
3452 * sbridge_probe Probe for ONE instance of device to see if it is
3453 * present.
3454 * return:
3455 * 0 for FOUND a device
3456 * < 0 for error code
3459 static int sbridge_probe(struct pci_dev *pdev, const struct pci_device_id *id)
3461 int rc = -ENODEV;
3462 u8 mc, num_mc = 0;
3463 struct sbridge_dev *sbridge_dev;
3464 enum type type = SANDY_BRIDGE;
3466 /* get the pci devices we want to reserve for our use */
3467 mutex_lock(&sbridge_edac_lock);
3470 * All memory controllers are allocated at the first pass.
3472 if (unlikely(probed >= 1)) {
3473 mutex_unlock(&sbridge_edac_lock);
3474 return -ENODEV;
3476 probed++;
3478 switch (pdev->device) {
3479 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TA:
3480 rc = sbridge_get_all_devices(&num_mc,
3481 pci_dev_descr_ibridge_table);
3482 type = IVY_BRIDGE;
3483 break;
3484 case PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_HA0:
3485 rc = sbridge_get_all_devices(&num_mc,
3486 pci_dev_descr_sbridge_table);
3487 type = SANDY_BRIDGE;
3488 break;
3489 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0:
3490 rc = sbridge_get_all_devices(&num_mc,
3491 pci_dev_descr_haswell_table);
3492 type = HASWELL;
3493 break;
3494 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0:
3495 rc = sbridge_get_all_devices(&num_mc,
3496 pci_dev_descr_broadwell_table);
3497 type = BROADWELL;
3498 break;
3499 case PCI_DEVICE_ID_INTEL_KNL_IMC_SAD0:
3500 rc = sbridge_get_all_devices_knl(&num_mc,
3501 pci_dev_descr_knl_table);
3502 type = KNIGHTS_LANDING;
3503 break;
3505 if (unlikely(rc < 0)) {
3506 edac_dbg(0, "couldn't get all devices for 0x%x\n", pdev->device);
3507 goto fail0;
3510 mc = 0;
3512 list_for_each_entry(sbridge_dev, &sbridge_edac_list, list) {
3513 edac_dbg(0, "Registering MC#%d (%d of %d)\n",
3514 mc, mc + 1, num_mc);
3516 sbridge_dev->mc = mc++;
3517 rc = sbridge_register_mci(sbridge_dev, type);
3518 if (unlikely(rc < 0))
3519 goto fail1;
3522 sbridge_printk(KERN_INFO, "%s\n", SBRIDGE_REVISION);
3524 mutex_unlock(&sbridge_edac_lock);
3525 return 0;
3527 fail1:
3528 list_for_each_entry(sbridge_dev, &sbridge_edac_list, list)
3529 sbridge_unregister_mci(sbridge_dev);
3531 sbridge_put_all_devices();
3532 fail0:
3533 mutex_unlock(&sbridge_edac_lock);
3534 return rc;
3538 * sbridge_remove destructor for one instance of device
3541 static void sbridge_remove(struct pci_dev *pdev)
3543 struct sbridge_dev *sbridge_dev;
3545 edac_dbg(0, "\n");
3548 * we have a trouble here: pdev value for removal will be wrong, since
3549 * it will point to the X58 register used to detect that the machine
3550 * is a Nehalem or upper design. However, due to the way several PCI
3551 * devices are grouped together to provide MC functionality, we need
3552 * to use a different method for releasing the devices
3555 mutex_lock(&sbridge_edac_lock);
3557 if (unlikely(!probed)) {
3558 mutex_unlock(&sbridge_edac_lock);
3559 return;
3562 list_for_each_entry(sbridge_dev, &sbridge_edac_list, list)
3563 sbridge_unregister_mci(sbridge_dev);
3565 /* Release PCI resources */
3566 sbridge_put_all_devices();
3568 probed--;
3570 mutex_unlock(&sbridge_edac_lock);
3573 MODULE_DEVICE_TABLE(pci, sbridge_pci_tbl);
3576 * sbridge_driver pci_driver structure for this module
3579 static struct pci_driver sbridge_driver = {
3580 .name = "sbridge_edac",
3581 .probe = sbridge_probe,
3582 .remove = sbridge_remove,
3583 .id_table = sbridge_pci_tbl,
3587 * sbridge_init Module entry function
3588 * Try to initialize this module for its devices
3590 static int __init sbridge_init(void)
3592 int pci_rc;
3594 edac_dbg(2, "\n");
3596 /* Ensure that the OPSTATE is set correctly for POLL or NMI */
3597 opstate_init();
3599 pci_rc = pci_register_driver(&sbridge_driver);
3600 if (pci_rc >= 0) {
3601 mce_register_decode_chain(&sbridge_mce_dec);
3602 if (get_edac_report_status() == EDAC_REPORTING_DISABLED)
3603 sbridge_printk(KERN_WARNING, "Loading driver, error reporting disabled.\n");
3604 return 0;
3607 sbridge_printk(KERN_ERR, "Failed to register device with error %d.\n",
3608 pci_rc);
3610 return pci_rc;
3614 * sbridge_exit() Module exit function
3615 * Unregister the driver
3617 static void __exit sbridge_exit(void)
3619 edac_dbg(2, "\n");
3620 pci_unregister_driver(&sbridge_driver);
3621 mce_unregister_decode_chain(&sbridge_mce_dec);
3624 module_init(sbridge_init);
3625 module_exit(sbridge_exit);
3627 module_param(edac_op_state, int, 0444);
3628 MODULE_PARM_DESC(edac_op_state, "EDAC Error Reporting state: 0=Poll,1=NMI");
3630 MODULE_LICENSE("GPL");
3631 MODULE_AUTHOR("Mauro Carvalho Chehab");
3632 MODULE_AUTHOR("Red Hat Inc. (http://www.redhat.com)");
3633 MODULE_DESCRIPTION("MC Driver for Intel Sandy Bridge and Ivy Bridge memory controllers - "
3634 SBRIDGE_REVISION);