1 /* Intel Sandy Bridge -EN/-EP/-EX Memory Controller kernel module
2  *
3  * This driver supports the memory controllers found on the Intel
4  * processor family Sandy Bridge.
5  *
6  * This file may be distributed under the terms of the
7  * GNU General Public License version 2 only.
8  *
9  * Copyright (c) 2011 by:
10  *	 Mauro Carvalho Chehab
11  */
12 
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>
26 
27 #include "edac_core.h"
28 
29 /* Static vars */
30 static LIST_HEAD(sbridge_edac_list);
31 static DEFINE_MUTEX(sbridge_edac_lock);
32 static int probed;
33 
34 /*
35  * Alter this version for the module when modifications are made
36  */
37 #define SBRIDGE_REVISION    " Ver: 1.1.0 "
38 #define EDAC_MOD_STR      "sbridge_edac"
39 
40 /*
41  * Debug macros
42  */
43 #define sbridge_printk(level, fmt, arg...)			\
44 	edac_printk(level, "sbridge", fmt, ##arg)
45 
46 #define sbridge_mc_printk(mci, level, fmt, arg...)		\
47 	edac_mc_chipset_printk(mci, level, "sbridge", fmt, ##arg)
48 
49 /*
50  * Get a bit field at register value <v>, from bit <lo> to bit <hi>
51  */
52 #define GET_BITFIELD(v, lo, hi)	\
53 	(((v) & GENMASK_ULL(hi, lo)) >> (lo))
54 
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,
59 };
60 
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,
66 };
67 
68 #define SAD_LIMIT(reg)		((GET_BITFIELD(reg, 6, 25) << 26) | 0x3ffffff)
69 #define DRAM_ATTR(reg)		GET_BITFIELD(reg, 2,  3)
70 #define INTERLEAVE_MODE(reg)	GET_BITFIELD(reg, 1,  1)
71 #define DRAM_RULE_ENABLE(reg)	GET_BITFIELD(reg, 0,  0)
72 #define A7MODE(reg)		GET_BITFIELD(reg, 26, 26)
73 
get_dram_attr(u32 reg)74 static char *get_dram_attr(u32 reg)
75 {
76 	switch(DRAM_ATTR(reg)) {
77 		case 0:
78 			return "DRAM";
79 		case 1:
80 			return "MMCFG";
81 		case 2:
82 			return "NXM";
83 		default:
84 			return "unknown";
85 	}
86 }
87 
88 static const u32 sbridge_interleave_list[] = {
89 	0x84, 0x8c, 0x94, 0x9c, 0xa4,
90 	0xac, 0xb4, 0xbc, 0xc4, 0xcc,
91 };
92 
93 static const u32 ibridge_interleave_list[] = {
94 	0x64, 0x6c, 0x74, 0x7c, 0x84,
95 	0x8c, 0x94, 0x9c, 0xa4, 0xac,
96 	0xb4, 0xbc, 0xc4, 0xcc, 0xd4,
97 	0xdc, 0xe4, 0xec, 0xf4, 0xfc,
98 };
99 
100 struct interleave_pkg {
101 	unsigned char start;
102 	unsigned char end;
103 };
104 
105 static const struct interleave_pkg sbridge_interleave_pkg[] = {
106 	{ 0, 2 },
107 	{ 3, 5 },
108 	{ 8, 10 },
109 	{ 11, 13 },
110 	{ 16, 18 },
111 	{ 19, 21 },
112 	{ 24, 26 },
113 	{ 27, 29 },
114 };
115 
116 static const struct interleave_pkg ibridge_interleave_pkg[] = {
117 	{ 0, 3 },
118 	{ 4, 7 },
119 	{ 8, 11 },
120 	{ 12, 15 },
121 	{ 16, 19 },
122 	{ 20, 23 },
123 	{ 24, 27 },
124 	{ 28, 31 },
125 };
126 
sad_pkg(const struct interleave_pkg * table,u32 reg,int interleave)127 static inline int sad_pkg(const struct interleave_pkg *table, u32 reg,
128 			  int interleave)
129 {
130 	return GET_BITFIELD(reg, table[interleave].start,
131 			    table[interleave].end);
132 }
133 
134 /* Devices 12 Function 7 */
135 
136 #define TOLM		0x80
137 #define	TOHM		0x84
138 #define HASWELL_TOLM	0xd0
139 #define HASWELL_TOHM_0	0xd4
140 #define HASWELL_TOHM_1	0xd8
141 
142 #define GET_TOLM(reg)		((GET_BITFIELD(reg, 0,  3) << 28) | 0x3ffffff)
143 #define GET_TOHM(reg)		((GET_BITFIELD(reg, 0, 20) << 25) | 0x3ffffff)
144 
145 /* Device 13 Function 6 */
146 
147 #define SAD_TARGET	0xf0
148 
149 #define SOURCE_ID(reg)		GET_BITFIELD(reg, 9, 11)
150 
151 #define SAD_CONTROL	0xf4
152 
153 /* Device 14 function 0 */
154 
155 static const u32 tad_dram_rule[] = {
156 	0x40, 0x44, 0x48, 0x4c,
157 	0x50, 0x54, 0x58, 0x5c,
158 	0x60, 0x64, 0x68, 0x6c,
159 };
160 #define MAX_TAD	ARRAY_SIZE(tad_dram_rule)
161 
162 #define TAD_LIMIT(reg)		((GET_BITFIELD(reg, 12, 31) << 26) | 0x3ffffff)
163 #define TAD_SOCK(reg)		GET_BITFIELD(reg, 10, 11)
164 #define TAD_CH(reg)		GET_BITFIELD(reg,  8,  9)
165 #define TAD_TGT3(reg)		GET_BITFIELD(reg,  6,  7)
166 #define TAD_TGT2(reg)		GET_BITFIELD(reg,  4,  5)
167 #define TAD_TGT1(reg)		GET_BITFIELD(reg,  2,  3)
168 #define TAD_TGT0(reg)		GET_BITFIELD(reg,  0,  1)
169 
170 /* Device 15, function 0 */
171 
172 #define MCMTR			0x7c
173 
174 #define IS_ECC_ENABLED(mcmtr)		GET_BITFIELD(mcmtr, 2, 2)
175 #define IS_LOCKSTEP_ENABLED(mcmtr)	GET_BITFIELD(mcmtr, 1, 1)
176 #define IS_CLOSE_PG(mcmtr)		GET_BITFIELD(mcmtr, 0, 0)
177 
178 /* Device 15, function 1 */
179 
180 #define RASENABLES		0xac
181 #define IS_MIRROR_ENABLED(reg)		GET_BITFIELD(reg, 0, 0)
182 
183 /* Device 15, functions 2-5 */
184 
185 static const int mtr_regs[] = {
186 	0x80, 0x84, 0x88,
187 };
188 
189 #define RANK_DISABLE(mtr)		GET_BITFIELD(mtr, 16, 19)
190 #define IS_DIMM_PRESENT(mtr)		GET_BITFIELD(mtr, 14, 14)
191 #define RANK_CNT_BITS(mtr)		GET_BITFIELD(mtr, 12, 13)
192 #define RANK_WIDTH_BITS(mtr)		GET_BITFIELD(mtr, 2, 4)
193 #define COL_WIDTH_BITS(mtr)		GET_BITFIELD(mtr, 0, 1)
194 
195 static const u32 tad_ch_nilv_offset[] = {
196 	0x90, 0x94, 0x98, 0x9c,
197 	0xa0, 0xa4, 0xa8, 0xac,
198 	0xb0, 0xb4, 0xb8, 0xbc,
199 };
200 #define CHN_IDX_OFFSET(reg)		GET_BITFIELD(reg, 28, 29)
201 #define TAD_OFFSET(reg)			(GET_BITFIELD(reg,  6, 25) << 26)
202 
203 static const u32 rir_way_limit[] = {
204 	0x108, 0x10c, 0x110, 0x114, 0x118,
205 };
206 #define MAX_RIR_RANGES ARRAY_SIZE(rir_way_limit)
207 
208 #define IS_RIR_VALID(reg)	GET_BITFIELD(reg, 31, 31)
209 #define RIR_WAY(reg)		GET_BITFIELD(reg, 28, 29)
210 
211 #define MAX_RIR_WAY	8
212 
213 static const u32 rir_offset[MAX_RIR_RANGES][MAX_RIR_WAY] = {
214 	{ 0x120, 0x124, 0x128, 0x12c, 0x130, 0x134, 0x138, 0x13c },
215 	{ 0x140, 0x144, 0x148, 0x14c, 0x150, 0x154, 0x158, 0x15c },
216 	{ 0x160, 0x164, 0x168, 0x16c, 0x170, 0x174, 0x178, 0x17c },
217 	{ 0x180, 0x184, 0x188, 0x18c, 0x190, 0x194, 0x198, 0x19c },
218 	{ 0x1a0, 0x1a4, 0x1a8, 0x1ac, 0x1b0, 0x1b4, 0x1b8, 0x1bc },
219 };
220 
221 #define RIR_RNK_TGT(type, reg) (((type) == BROADWELL) ? \
222 	GET_BITFIELD(reg, 20, 23) : GET_BITFIELD(reg, 16, 19))
223 
224 #define RIR_OFFSET(type, reg) (((type) == HASWELL || (type) == BROADWELL) ? \
225 	GET_BITFIELD(reg,  2, 15) : GET_BITFIELD(reg,  2, 14))
226 
227 /* Device 16, functions 2-7 */
228 
229 /*
230  * FIXME: Implement the error count reads directly
231  */
232 
233 static const u32 correrrcnt[] = {
234 	0x104, 0x108, 0x10c, 0x110,
235 };
236 
237 #define RANK_ODD_OV(reg)		GET_BITFIELD(reg, 31, 31)
238 #define RANK_ODD_ERR_CNT(reg)		GET_BITFIELD(reg, 16, 30)
239 #define RANK_EVEN_OV(reg)		GET_BITFIELD(reg, 15, 15)
240 #define RANK_EVEN_ERR_CNT(reg)		GET_BITFIELD(reg,  0, 14)
241 
242 static const u32 correrrthrsld[] = {
243 	0x11c, 0x120, 0x124, 0x128,
244 };
245 
246 #define RANK_ODD_ERR_THRSLD(reg)	GET_BITFIELD(reg, 16, 30)
247 #define RANK_EVEN_ERR_THRSLD(reg)	GET_BITFIELD(reg,  0, 14)
248 
249 
250 /* Device 17, function 0 */
251 
252 #define SB_RANK_CFG_A		0x0328
253 
254 #define IB_RANK_CFG_A		0x0320
255 
256 /*
257  * sbridge structs
258  */
259 
260 #define NUM_CHANNELS		4
261 #define MAX_DIMMS		3	/* Max DIMMS per channel */
262 #define CHANNEL_UNSPECIFIED	0xf	/* Intel IA32 SDM 15-14 */
263 
264 enum type {
265 	SANDY_BRIDGE,
266 	IVY_BRIDGE,
267 	HASWELL,
268 	BROADWELL,
269 };
270 
271 struct sbridge_pvt;
272 struct sbridge_info {
273 	enum type	type;
274 	u32		mcmtr;
275 	u32		rankcfgr;
276 	u64		(*get_tolm)(struct sbridge_pvt *pvt);
277 	u64		(*get_tohm)(struct sbridge_pvt *pvt);
278 	u64		(*rir_limit)(u32 reg);
279 	const u32	*dram_rule;
280 	const u32	*interleave_list;
281 	const struct interleave_pkg *interleave_pkg;
282 	u8		max_sad;
283 	u8		max_interleave;
284 	u8		(*get_node_id)(struct sbridge_pvt *pvt);
285 	enum mem_type	(*get_memory_type)(struct sbridge_pvt *pvt);
286 	struct pci_dev	*pci_vtd;
287 };
288 
289 struct sbridge_channel {
290 	u32		ranks;
291 	u32		dimms;
292 };
293 
294 struct pci_id_descr {
295 	int			dev_id;
296 	int			optional;
297 };
298 
299 struct pci_id_table {
300 	const struct pci_id_descr	*descr;
301 	int				n_devs;
302 };
303 
304 struct sbridge_dev {
305 	struct list_head	list;
306 	u8			bus, mc;
307 	u8			node_id, source_id;
308 	struct pci_dev		**pdev;
309 	int			n_devs;
310 	struct mem_ctl_info	*mci;
311 };
312 
313 struct sbridge_pvt {
314 	struct pci_dev		*pci_ta, *pci_ddrio, *pci_ras;
315 	struct pci_dev		*pci_sad0, *pci_sad1;
316 	struct pci_dev		*pci_ha0, *pci_ha1;
317 	struct pci_dev		*pci_br0, *pci_br1;
318 	struct pci_dev		*pci_ha1_ta;
319 	struct pci_dev		*pci_tad[NUM_CHANNELS];
320 
321 	struct sbridge_dev	*sbridge_dev;
322 
323 	struct sbridge_info	info;
324 	struct sbridge_channel	channel[NUM_CHANNELS];
325 
326 	/* Memory type detection */
327 	bool			is_mirrored, is_lockstep, is_close_pg;
328 
329 	/* Fifo double buffers */
330 	struct mce		mce_entry[MCE_LOG_LEN];
331 	struct mce		mce_outentry[MCE_LOG_LEN];
332 
333 	/* Fifo in/out counters */
334 	unsigned		mce_in, mce_out;
335 
336 	/* Count indicator to show errors not got */
337 	unsigned		mce_overrun;
338 
339 	/* Memory description */
340 	u64			tolm, tohm;
341 };
342 
343 #define PCI_DESCR(device_id, opt)	\
344 	.dev_id = (device_id),		\
345 	.optional = opt
346 
347 static const struct pci_id_descr pci_dev_descr_sbridge[] = {
348 		/* Processor Home Agent */
349 	{ PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_HA0, 0)	},
350 
351 		/* Memory controller */
352 	{ PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TA, 0)	},
353 	{ PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_RAS, 0)	},
354 	{ PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD0, 0)	},
355 	{ PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD1, 0)	},
356 	{ PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD2, 0)	},
357 	{ PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD3, 0)	},
358 	{ PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_DDRIO, 1)	},
359 
360 		/* System Address Decoder */
361 	{ PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_SAD0, 0)	},
362 	{ PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_SAD1, 0)	},
363 
364 		/* Broadcast Registers */
365 	{ PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_BR, 0)		},
366 };
367 
368 #define PCI_ID_TABLE_ENTRY(A) { .descr=A, .n_devs = ARRAY_SIZE(A) }
369 static const struct pci_id_table pci_dev_descr_sbridge_table[] = {
370 	PCI_ID_TABLE_ENTRY(pci_dev_descr_sbridge),
371 	{0,}			/* 0 terminated list. */
372 };
373 
374 /* This changes depending if 1HA or 2HA:
375  * 1HA:
376  *	0x0eb8 (17.0) is DDRIO0
377  * 2HA:
378  *	0x0ebc (17.4) is DDRIO0
379  */
380 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_1HA_DDRIO0	0x0eb8
381 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_2HA_DDRIO0	0x0ebc
382 
383 /* pci ids */
384 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0		0x0ea0
385 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TA		0x0ea8
386 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_RAS		0x0e71
387 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD0	0x0eaa
388 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD1	0x0eab
389 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD2	0x0eac
390 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD3	0x0ead
391 #define PCI_DEVICE_ID_INTEL_IBRIDGE_SAD			0x0ec8
392 #define PCI_DEVICE_ID_INTEL_IBRIDGE_BR0			0x0ec9
393 #define PCI_DEVICE_ID_INTEL_IBRIDGE_BR1			0x0eca
394 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1		0x0e60
395 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TA		0x0e68
396 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_RAS		0x0e79
397 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD0	0x0e6a
398 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD1	0x0e6b
399 
400 static const struct pci_id_descr pci_dev_descr_ibridge[] = {
401 		/* Processor Home Agent */
402 	{ PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0, 0)		},
403 
404 		/* Memory controller */
405 	{ PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TA, 0)		},
406 	{ PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_RAS, 0)		},
407 	{ PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD0, 0)	},
408 	{ PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD1, 0)	},
409 	{ PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD2, 0)	},
410 	{ PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD3, 0)	},
411 
412 		/* System Address Decoder */
413 	{ PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_SAD, 0)			},
414 
415 		/* Broadcast Registers */
416 	{ PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_BR0, 1)			},
417 	{ PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_BR1, 0)			},
418 
419 		/* Optional, mode 2HA */
420 	{ PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1, 1)		},
421 #if 0
422 	{ PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TA, 1)	},
423 	{ PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_RAS, 1)	},
424 #endif
425 	{ PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD0, 1)	},
426 	{ PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD1, 1)	},
427 
428 	{ PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_1HA_DDRIO0, 1)	},
429 	{ PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_2HA_DDRIO0, 1)	},
430 };
431 
432 static const struct pci_id_table pci_dev_descr_ibridge_table[] = {
433 	PCI_ID_TABLE_ENTRY(pci_dev_descr_ibridge),
434 	{0,}			/* 0 terminated list. */
435 };
436 
437 /* Haswell support */
438 /* EN processor:
439  *	- 1 IMC
440  *	- 3 DDR3 channels, 2 DPC per channel
441  * EP processor:
442  *	- 1 or 2 IMC
443  *	- 4 DDR4 channels, 3 DPC per channel
444  * EP 4S processor:
445  *	- 2 IMC
446  *	- 4 DDR4 channels, 3 DPC per channel
447  * EX processor:
448  *	- 2 IMC
449  *	- each IMC interfaces with a SMI 2 channel
450  *	- each SMI channel interfaces with a scalable memory buffer
451  *	- each scalable memory buffer supports 4 DDR3/DDR4 channels, 3 DPC
452  */
453 #define HASWELL_DDRCRCLKCONTROLS 0xa10 /* Ditto on Broadwell */
454 #define HASWELL_HASYSDEFEATURE2 0x84
455 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_VTD_MISC 0x2f28
456 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0	0x2fa0
457 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1	0x2f60
458 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TA	0x2fa8
459 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_THERMAL 0x2f71
460 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TA	0x2f68
461 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_THERMAL 0x2f79
462 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_CBO_SAD0 0x2ffc
463 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_CBO_SAD1 0x2ffd
464 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD0 0x2faa
465 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD1 0x2fab
466 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD2 0x2fac
467 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD3 0x2fad
468 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD0 0x2f6a
469 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD1 0x2f6b
470 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD2 0x2f6c
471 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD3 0x2f6d
472 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO0 0x2fbd
473 static const struct pci_id_descr pci_dev_descr_haswell[] = {
474 	/* first item must be the HA */
475 	{ PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0, 0)		},
476 
477 	{ PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_CBO_SAD0, 0)	},
478 	{ PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_CBO_SAD1, 0)	},
479 
480 	{ PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1, 1)		},
481 
482 	{ PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TA, 0)		},
483 	{ PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_THERMAL, 0)	},
484 	{ PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD0, 0)	},
485 	{ PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD1, 0)	},
486 	{ PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD2, 1)	},
487 	{ PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD3, 1)	},
488 
489 	{ PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO0, 1)		},
490 
491 	{ PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TA, 1)		},
492 	{ PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_THERMAL, 1)	},
493 	{ PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD0, 1)	},
494 	{ PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD1, 1)	},
495 	{ PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD2, 1)	},
496 	{ PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD3, 1)	},
497 };
498 
499 static const struct pci_id_table pci_dev_descr_haswell_table[] = {
500 	PCI_ID_TABLE_ENTRY(pci_dev_descr_haswell),
501 	{0,}			/* 0 terminated list. */
502 };
503 
504 /*
505  * Broadwell support
506  *
507  * DE processor:
508  *	- 1 IMC
509  *	- 2 DDR3 channels, 2 DPC per channel
510  */
511 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_VTD_MISC 0x6f28
512 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0	0x6fa0
513 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TA	0x6fa8
514 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_THERMAL 0x6f71
515 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_CBO_SAD0 0x6ffc
516 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_CBO_SAD1 0x6ffd
517 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD0 0x6faa
518 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD1 0x6fab
519 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD2 0x6fac
520 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD3 0x6fad
521 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_DDRIO0 0x6faf
522 
523 static const struct pci_id_descr pci_dev_descr_broadwell[] = {
524 	/* first item must be the HA */
525 	{ PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0, 0)		},
526 
527 	{ PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_CBO_SAD0, 0)	},
528 	{ PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_CBO_SAD1, 0)	},
529 
530 	{ PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TA, 0)	},
531 	{ PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_THERMAL, 0)	},
532 	{ PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD0, 0)	},
533 	{ PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD1, 0)	},
534 	{ PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD2, 0)	},
535 	{ PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD3, 0)	},
536 	{ PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_DDRIO0, 1)	},
537 };
538 
539 static const struct pci_id_table pci_dev_descr_broadwell_table[] = {
540 	PCI_ID_TABLE_ENTRY(pci_dev_descr_broadwell),
541 	{0,}			/* 0 terminated list. */
542 };
543 
544 /*
545  *	pci_device_id	table for which devices we are looking for
546  */
547 static const struct pci_device_id sbridge_pci_tbl[] = {
548 	{PCI_DEVICE(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_HA0)},
549 	{PCI_DEVICE(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TA)},
550 	{PCI_DEVICE(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0)},
551 	{PCI_DEVICE(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0)},
552 	{0,}			/* 0 terminated list. */
553 };
554 
555 
556 /****************************************************************************
557 			Ancillary status routines
558  ****************************************************************************/
559 
numrank(enum type type,u32 mtr)560 static inline int numrank(enum type type, u32 mtr)
561 {
562 	int ranks = (1 << RANK_CNT_BITS(mtr));
563 	int max = 4;
564 
565 	if (type == HASWELL)
566 		max = 8;
567 
568 	if (ranks > max) {
569 		edac_dbg(0, "Invalid number of ranks: %d (max = %i) raw value = %x (%04x)\n",
570 			 ranks, max, (unsigned int)RANK_CNT_BITS(mtr), mtr);
571 		return -EINVAL;
572 	}
573 
574 	return ranks;
575 }
576 
numrow(u32 mtr)577 static inline int numrow(u32 mtr)
578 {
579 	int rows = (RANK_WIDTH_BITS(mtr) + 12);
580 
581 	if (rows < 13 || rows > 18) {
582 		edac_dbg(0, "Invalid number of rows: %d (should be between 14 and 17) raw value = %x (%04x)\n",
583 			 rows, (unsigned int)RANK_WIDTH_BITS(mtr), mtr);
584 		return -EINVAL;
585 	}
586 
587 	return 1 << rows;
588 }
589 
numcol(u32 mtr)590 static inline int numcol(u32 mtr)
591 {
592 	int cols = (COL_WIDTH_BITS(mtr) + 10);
593 
594 	if (cols > 12) {
595 		edac_dbg(0, "Invalid number of cols: %d (max = 4) raw value = %x (%04x)\n",
596 			 cols, (unsigned int)COL_WIDTH_BITS(mtr), mtr);
597 		return -EINVAL;
598 	}
599 
600 	return 1 << cols;
601 }
602 
get_sbridge_dev(u8 bus)603 static struct sbridge_dev *get_sbridge_dev(u8 bus)
604 {
605 	struct sbridge_dev *sbridge_dev;
606 
607 	list_for_each_entry(sbridge_dev, &sbridge_edac_list, list) {
608 		if (sbridge_dev->bus == bus)
609 			return sbridge_dev;
610 	}
611 
612 	return NULL;
613 }
614 
alloc_sbridge_dev(u8 bus,const struct pci_id_table * table)615 static struct sbridge_dev *alloc_sbridge_dev(u8 bus,
616 					   const struct pci_id_table *table)
617 {
618 	struct sbridge_dev *sbridge_dev;
619 
620 	sbridge_dev = kzalloc(sizeof(*sbridge_dev), GFP_KERNEL);
621 	if (!sbridge_dev)
622 		return NULL;
623 
624 	sbridge_dev->pdev = kzalloc(sizeof(*sbridge_dev->pdev) * table->n_devs,
625 				   GFP_KERNEL);
626 	if (!sbridge_dev->pdev) {
627 		kfree(sbridge_dev);
628 		return NULL;
629 	}
630 
631 	sbridge_dev->bus = bus;
632 	sbridge_dev->n_devs = table->n_devs;
633 	list_add_tail(&sbridge_dev->list, &sbridge_edac_list);
634 
635 	return sbridge_dev;
636 }
637 
free_sbridge_dev(struct sbridge_dev * sbridge_dev)638 static void free_sbridge_dev(struct sbridge_dev *sbridge_dev)
639 {
640 	list_del(&sbridge_dev->list);
641 	kfree(sbridge_dev->pdev);
642 	kfree(sbridge_dev);
643 }
644 
sbridge_get_tolm(struct sbridge_pvt * pvt)645 static u64 sbridge_get_tolm(struct sbridge_pvt *pvt)
646 {
647 	u32 reg;
648 
649 	/* Address range is 32:28 */
650 	pci_read_config_dword(pvt->pci_sad1, TOLM, &reg);
651 	return GET_TOLM(reg);
652 }
653 
sbridge_get_tohm(struct sbridge_pvt * pvt)654 static u64 sbridge_get_tohm(struct sbridge_pvt *pvt)
655 {
656 	u32 reg;
657 
658 	pci_read_config_dword(pvt->pci_sad1, TOHM, &reg);
659 	return GET_TOHM(reg);
660 }
661 
ibridge_get_tolm(struct sbridge_pvt * pvt)662 static u64 ibridge_get_tolm(struct sbridge_pvt *pvt)
663 {
664 	u32 reg;
665 
666 	pci_read_config_dword(pvt->pci_br1, TOLM, &reg);
667 
668 	return GET_TOLM(reg);
669 }
670 
ibridge_get_tohm(struct sbridge_pvt * pvt)671 static u64 ibridge_get_tohm(struct sbridge_pvt *pvt)
672 {
673 	u32 reg;
674 
675 	pci_read_config_dword(pvt->pci_br1, TOHM, &reg);
676 
677 	return GET_TOHM(reg);
678 }
679 
rir_limit(u32 reg)680 static u64 rir_limit(u32 reg)
681 {
682 	return ((u64)GET_BITFIELD(reg,  1, 10) << 29) | 0x1fffffff;
683 }
684 
get_memory_type(struct sbridge_pvt * pvt)685 static enum mem_type get_memory_type(struct sbridge_pvt *pvt)
686 {
687 	u32 reg;
688 	enum mem_type mtype;
689 
690 	if (pvt->pci_ddrio) {
691 		pci_read_config_dword(pvt->pci_ddrio, pvt->info.rankcfgr,
692 				      &reg);
693 		if (GET_BITFIELD(reg, 11, 11))
694 			/* FIXME: Can also be LRDIMM */
695 			mtype = MEM_RDDR3;
696 		else
697 			mtype = MEM_DDR3;
698 	} else
699 		mtype = MEM_UNKNOWN;
700 
701 	return mtype;
702 }
703 
haswell_get_memory_type(struct sbridge_pvt * pvt)704 static enum mem_type haswell_get_memory_type(struct sbridge_pvt *pvt)
705 {
706 	u32 reg;
707 	bool registered = false;
708 	enum mem_type mtype = MEM_UNKNOWN;
709 
710 	if (!pvt->pci_ddrio)
711 		goto out;
712 
713 	pci_read_config_dword(pvt->pci_ddrio,
714 			      HASWELL_DDRCRCLKCONTROLS, &reg);
715 	/* Is_Rdimm */
716 	if (GET_BITFIELD(reg, 16, 16))
717 		registered = true;
718 
719 	pci_read_config_dword(pvt->pci_ta, MCMTR, &reg);
720 	if (GET_BITFIELD(reg, 14, 14)) {
721 		if (registered)
722 			mtype = MEM_RDDR4;
723 		else
724 			mtype = MEM_DDR4;
725 	} else {
726 		if (registered)
727 			mtype = MEM_RDDR3;
728 		else
729 			mtype = MEM_DDR3;
730 	}
731 
732 out:
733 	return mtype;
734 }
735 
get_node_id(struct sbridge_pvt * pvt)736 static u8 get_node_id(struct sbridge_pvt *pvt)
737 {
738 	u32 reg;
739 	pci_read_config_dword(pvt->pci_br0, SAD_CONTROL, &reg);
740 	return GET_BITFIELD(reg, 0, 2);
741 }
742 
haswell_get_node_id(struct sbridge_pvt * pvt)743 static u8 haswell_get_node_id(struct sbridge_pvt *pvt)
744 {
745 	u32 reg;
746 
747 	pci_read_config_dword(pvt->pci_sad1, SAD_CONTROL, &reg);
748 	return GET_BITFIELD(reg, 0, 3);
749 }
750 
haswell_get_tolm(struct sbridge_pvt * pvt)751 static u64 haswell_get_tolm(struct sbridge_pvt *pvt)
752 {
753 	u32 reg;
754 
755 	pci_read_config_dword(pvt->info.pci_vtd, HASWELL_TOLM, &reg);
756 	return (GET_BITFIELD(reg, 26, 31) << 26) | 0x3ffffff;
757 }
758 
haswell_get_tohm(struct sbridge_pvt * pvt)759 static u64 haswell_get_tohm(struct sbridge_pvt *pvt)
760 {
761 	u64 rc;
762 	u32 reg;
763 
764 	pci_read_config_dword(pvt->info.pci_vtd, HASWELL_TOHM_0, &reg);
765 	rc = GET_BITFIELD(reg, 26, 31);
766 	pci_read_config_dword(pvt->info.pci_vtd, HASWELL_TOHM_1, &reg);
767 	rc = ((reg << 6) | rc) << 26;
768 
769 	return rc | 0x1ffffff;
770 }
771 
haswell_rir_limit(u32 reg)772 static u64 haswell_rir_limit(u32 reg)
773 {
774 	return (((u64)GET_BITFIELD(reg,  1, 11) + 1) << 29) - 1;
775 }
776 
sad_pkg_socket(u8 pkg)777 static inline u8 sad_pkg_socket(u8 pkg)
778 {
779 	/* on Ivy Bridge, nodeID is SASS, where A is HA and S is node id */
780 	return ((pkg >> 3) << 2) | (pkg & 0x3);
781 }
782 
sad_pkg_ha(u8 pkg)783 static inline u8 sad_pkg_ha(u8 pkg)
784 {
785 	return (pkg >> 2) & 0x1;
786 }
787 
788 /****************************************************************************
789 			Memory check routines
790  ****************************************************************************/
get_pdev_same_bus(u8 bus,u32 id)791 static struct pci_dev *get_pdev_same_bus(u8 bus, u32 id)
792 {
793 	struct pci_dev *pdev = NULL;
794 
795 	do {
796 		pdev = pci_get_device(PCI_VENDOR_ID_INTEL, id, pdev);
797 		if (pdev && pdev->bus->number == bus)
798 			break;
799 	} while (pdev);
800 
801 	return pdev;
802 }
803 
804 /**
805  * check_if_ecc_is_active() - Checks if ECC is active
806  * @bus:	Device bus
807  * @type:	Memory controller type
808  * returns: 0 in case ECC is active, -ENODEV if it can't be determined or
809  *	    disabled
810  */
check_if_ecc_is_active(const u8 bus,enum type type)811 static int check_if_ecc_is_active(const u8 bus, enum type type)
812 {
813 	struct pci_dev *pdev = NULL;
814 	u32 mcmtr, id;
815 
816 	switch (type) {
817 	case IVY_BRIDGE:
818 		id = PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TA;
819 		break;
820 	case HASWELL:
821 		id = PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TA;
822 		break;
823 	case SANDY_BRIDGE:
824 		id = PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TA;
825 		break;
826 	case BROADWELL:
827 		id = PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TA;
828 		break;
829 	default:
830 		return -ENODEV;
831 	}
832 
833 	pdev = get_pdev_same_bus(bus, id);
834 	if (!pdev) {
835 		sbridge_printk(KERN_ERR, "Couldn't find PCI device "
836 					"%04x:%04x! on bus %02d\n",
837 					PCI_VENDOR_ID_INTEL, id, bus);
838 		return -ENODEV;
839 	}
840 
841 	pci_read_config_dword(pdev, MCMTR, &mcmtr);
842 	if (!IS_ECC_ENABLED(mcmtr)) {
843 		sbridge_printk(KERN_ERR, "ECC is disabled. Aborting\n");
844 		return -ENODEV;
845 	}
846 	return 0;
847 }
848 
get_dimm_config(struct mem_ctl_info * mci)849 static int get_dimm_config(struct mem_ctl_info *mci)
850 {
851 	struct sbridge_pvt *pvt = mci->pvt_info;
852 	struct dimm_info *dimm;
853 	unsigned i, j, banks, ranks, rows, cols, npages;
854 	u64 size;
855 	u32 reg;
856 	enum edac_type mode;
857 	enum mem_type mtype;
858 
859 	if (pvt->info.type == HASWELL || pvt->info.type == BROADWELL)
860 		pci_read_config_dword(pvt->pci_sad1, SAD_TARGET, &reg);
861 	else
862 		pci_read_config_dword(pvt->pci_br0, SAD_TARGET, &reg);
863 
864 	pvt->sbridge_dev->source_id = SOURCE_ID(reg);
865 
866 	pvt->sbridge_dev->node_id = pvt->info.get_node_id(pvt);
867 	edac_dbg(0, "mc#%d: Node ID: %d, source ID: %d\n",
868 		 pvt->sbridge_dev->mc,
869 		 pvt->sbridge_dev->node_id,
870 		 pvt->sbridge_dev->source_id);
871 
872 	pci_read_config_dword(pvt->pci_ras, RASENABLES, &reg);
873 	if (IS_MIRROR_ENABLED(reg)) {
874 		edac_dbg(0, "Memory mirror is enabled\n");
875 		pvt->is_mirrored = true;
876 	} else {
877 		edac_dbg(0, "Memory mirror is disabled\n");
878 		pvt->is_mirrored = false;
879 	}
880 
881 	pci_read_config_dword(pvt->pci_ta, MCMTR, &pvt->info.mcmtr);
882 	if (IS_LOCKSTEP_ENABLED(pvt->info.mcmtr)) {
883 		edac_dbg(0, "Lockstep is enabled\n");
884 		mode = EDAC_S8ECD8ED;
885 		pvt->is_lockstep = true;
886 	} else {
887 		edac_dbg(0, "Lockstep is disabled\n");
888 		mode = EDAC_S4ECD4ED;
889 		pvt->is_lockstep = false;
890 	}
891 	if (IS_CLOSE_PG(pvt->info.mcmtr)) {
892 		edac_dbg(0, "address map is on closed page mode\n");
893 		pvt->is_close_pg = true;
894 	} else {
895 		edac_dbg(0, "address map is on open page mode\n");
896 		pvt->is_close_pg = false;
897 	}
898 
899 	mtype = pvt->info.get_memory_type(pvt);
900 	if (mtype == MEM_RDDR3 || mtype == MEM_RDDR4)
901 		edac_dbg(0, "Memory is registered\n");
902 	else if (mtype == MEM_UNKNOWN)
903 		edac_dbg(0, "Cannot determine memory type\n");
904 	else
905 		edac_dbg(0, "Memory is unregistered\n");
906 
907 	if (mtype == MEM_DDR4 || mtype == MEM_RDDR4)
908 		banks = 16;
909 	else
910 		banks = 8;
911 
912 	for (i = 0; i < NUM_CHANNELS; i++) {
913 		u32 mtr;
914 
915 		for (j = 0; j < ARRAY_SIZE(mtr_regs); j++) {
916 			dimm = EDAC_DIMM_PTR(mci->layers, mci->dimms, mci->n_layers,
917 				       i, j, 0);
918 			pci_read_config_dword(pvt->pci_tad[i],
919 					      mtr_regs[j], &mtr);
920 			edac_dbg(4, "Channel #%d  MTR%d = %x\n", i, j, mtr);
921 			if (IS_DIMM_PRESENT(mtr)) {
922 				pvt->channel[i].dimms++;
923 
924 				ranks = numrank(pvt->info.type, mtr);
925 				rows = numrow(mtr);
926 				cols = numcol(mtr);
927 
928 				size = ((u64)rows * cols * banks * ranks) >> (20 - 3);
929 				npages = MiB_TO_PAGES(size);
930 
931 				edac_dbg(0, "mc#%d: channel %d, dimm %d, %Ld Mb (%d pages) bank: %d, rank: %d, row: %#x, col: %#x\n",
932 					 pvt->sbridge_dev->mc, i, j,
933 					 size, npages,
934 					 banks, ranks, rows, cols);
935 
936 				dimm->nr_pages = npages;
937 				dimm->grain = 32;
938 				switch (banks) {
939 				case 16:
940 					dimm->dtype = DEV_X16;
941 					break;
942 				case 8:
943 					dimm->dtype = DEV_X8;
944 					break;
945 				case 4:
946 					dimm->dtype = DEV_X4;
947 					break;
948 				}
949 				dimm->mtype = mtype;
950 				dimm->edac_mode = mode;
951 				snprintf(dimm->label, sizeof(dimm->label),
952 					 "CPU_SrcID#%u_Channel#%u_DIMM#%u",
953 					 pvt->sbridge_dev->source_id, i, j);
954 			}
955 		}
956 	}
957 
958 	return 0;
959 }
960 
get_memory_layout(const struct mem_ctl_info * mci)961 static void get_memory_layout(const struct mem_ctl_info *mci)
962 {
963 	struct sbridge_pvt *pvt = mci->pvt_info;
964 	int i, j, k, n_sads, n_tads, sad_interl;
965 	u32 reg;
966 	u64 limit, prv = 0;
967 	u64 tmp_mb;
968 	u32 gb, mb;
969 	u32 rir_way;
970 
971 	/*
972 	 * Step 1) Get TOLM/TOHM ranges
973 	 */
974 
975 	pvt->tolm = pvt->info.get_tolm(pvt);
976 	tmp_mb = (1 + pvt->tolm) >> 20;
977 
978 	gb = div_u64_rem(tmp_mb, 1024, &mb);
979 	edac_dbg(0, "TOLM: %u.%03u GB (0x%016Lx)\n",
980 		gb, (mb*1000)/1024, (u64)pvt->tolm);
981 
982 	/* Address range is already 45:25 */
983 	pvt->tohm = pvt->info.get_tohm(pvt);
984 	tmp_mb = (1 + pvt->tohm) >> 20;
985 
986 	gb = div_u64_rem(tmp_mb, 1024, &mb);
987 	edac_dbg(0, "TOHM: %u.%03u GB (0x%016Lx)\n",
988 		gb, (mb*1000)/1024, (u64)pvt->tohm);
989 
990 	/*
991 	 * Step 2) Get SAD range and SAD Interleave list
992 	 * TAD registers contain the interleave wayness. However, it
993 	 * seems simpler to just discover it indirectly, with the
994 	 * algorithm bellow.
995 	 */
996 	prv = 0;
997 	for (n_sads = 0; n_sads < pvt->info.max_sad; n_sads++) {
998 		/* SAD_LIMIT Address range is 45:26 */
999 		pci_read_config_dword(pvt->pci_sad0, pvt->info.dram_rule[n_sads],
1000 				      &reg);
1001 		limit = SAD_LIMIT(reg);
1002 
1003 		if (!DRAM_RULE_ENABLE(reg))
1004 			continue;
1005 
1006 		if (limit <= prv)
1007 			break;
1008 
1009 		tmp_mb = (limit + 1) >> 20;
1010 		gb = div_u64_rem(tmp_mb, 1024, &mb);
1011 		edac_dbg(0, "SAD#%d %s up to %u.%03u GB (0x%016Lx) Interleave: %s reg=0x%08x\n",
1012 			 n_sads,
1013 			 get_dram_attr(reg),
1014 			 gb, (mb*1000)/1024,
1015 			 ((u64)tmp_mb) << 20L,
1016 			 INTERLEAVE_MODE(reg) ? "8:6" : "[8:6]XOR[18:16]",
1017 			 reg);
1018 		prv = limit;
1019 
1020 		pci_read_config_dword(pvt->pci_sad0, pvt->info.interleave_list[n_sads],
1021 				      &reg);
1022 		sad_interl = sad_pkg(pvt->info.interleave_pkg, reg, 0);
1023 		for (j = 0; j < 8; j++) {
1024 			u32 pkg = sad_pkg(pvt->info.interleave_pkg, reg, j);
1025 			if (j > 0 && sad_interl == pkg)
1026 				break;
1027 
1028 			edac_dbg(0, "SAD#%d, interleave #%d: %d\n",
1029 				 n_sads, j, pkg);
1030 		}
1031 	}
1032 
1033 	/*
1034 	 * Step 3) Get TAD range
1035 	 */
1036 	prv = 0;
1037 	for (n_tads = 0; n_tads < MAX_TAD; n_tads++) {
1038 		pci_read_config_dword(pvt->pci_ha0, tad_dram_rule[n_tads],
1039 				      &reg);
1040 		limit = TAD_LIMIT(reg);
1041 		if (limit <= prv)
1042 			break;
1043 		tmp_mb = (limit + 1) >> 20;
1044 
1045 		gb = div_u64_rem(tmp_mb, 1024, &mb);
1046 		edac_dbg(0, "TAD#%d: up to %u.%03u GB (0x%016Lx), socket interleave %d, memory interleave %d, TGT: %d, %d, %d, %d, reg=0x%08x\n",
1047 			 n_tads, gb, (mb*1000)/1024,
1048 			 ((u64)tmp_mb) << 20L,
1049 			 (u32)(1 << TAD_SOCK(reg)),
1050 			 (u32)TAD_CH(reg) + 1,
1051 			 (u32)TAD_TGT0(reg),
1052 			 (u32)TAD_TGT1(reg),
1053 			 (u32)TAD_TGT2(reg),
1054 			 (u32)TAD_TGT3(reg),
1055 			 reg);
1056 		prv = limit;
1057 	}
1058 
1059 	/*
1060 	 * Step 4) Get TAD offsets, per each channel
1061 	 */
1062 	for (i = 0; i < NUM_CHANNELS; i++) {
1063 		if (!pvt->channel[i].dimms)
1064 			continue;
1065 		for (j = 0; j < n_tads; j++) {
1066 			pci_read_config_dword(pvt->pci_tad[i],
1067 					      tad_ch_nilv_offset[j],
1068 					      &reg);
1069 			tmp_mb = TAD_OFFSET(reg) >> 20;
1070 			gb = div_u64_rem(tmp_mb, 1024, &mb);
1071 			edac_dbg(0, "TAD CH#%d, offset #%d: %u.%03u GB (0x%016Lx), reg=0x%08x\n",
1072 				 i, j,
1073 				 gb, (mb*1000)/1024,
1074 				 ((u64)tmp_mb) << 20L,
1075 				 reg);
1076 		}
1077 	}
1078 
1079 	/*
1080 	 * Step 6) Get RIR Wayness/Limit, per each channel
1081 	 */
1082 	for (i = 0; i < NUM_CHANNELS; i++) {
1083 		if (!pvt->channel[i].dimms)
1084 			continue;
1085 		for (j = 0; j < MAX_RIR_RANGES; j++) {
1086 			pci_read_config_dword(pvt->pci_tad[i],
1087 					      rir_way_limit[j],
1088 					      &reg);
1089 
1090 			if (!IS_RIR_VALID(reg))
1091 				continue;
1092 
1093 			tmp_mb = pvt->info.rir_limit(reg) >> 20;
1094 			rir_way = 1 << RIR_WAY(reg);
1095 			gb = div_u64_rem(tmp_mb, 1024, &mb);
1096 			edac_dbg(0, "CH#%d RIR#%d, limit: %u.%03u GB (0x%016Lx), way: %d, reg=0x%08x\n",
1097 				 i, j,
1098 				 gb, (mb*1000)/1024,
1099 				 ((u64)tmp_mb) << 20L,
1100 				 rir_way,
1101 				 reg);
1102 
1103 			for (k = 0; k < rir_way; k++) {
1104 				pci_read_config_dword(pvt->pci_tad[i],
1105 						      rir_offset[j][k],
1106 						      &reg);
1107 				tmp_mb = RIR_OFFSET(pvt->info.type, reg) << 6;
1108 
1109 				gb = div_u64_rem(tmp_mb, 1024, &mb);
1110 				edac_dbg(0, "CH#%d RIR#%d INTL#%d, offset %u.%03u GB (0x%016Lx), tgt: %d, reg=0x%08x\n",
1111 					 i, j, k,
1112 					 gb, (mb*1000)/1024,
1113 					 ((u64)tmp_mb) << 20L,
1114 					 (u32)RIR_RNK_TGT(pvt->info.type, reg),
1115 					 reg);
1116 			}
1117 		}
1118 	}
1119 }
1120 
get_mci_for_node_id(u8 node_id)1121 static struct mem_ctl_info *get_mci_for_node_id(u8 node_id)
1122 {
1123 	struct sbridge_dev *sbridge_dev;
1124 
1125 	list_for_each_entry(sbridge_dev, &sbridge_edac_list, list) {
1126 		if (sbridge_dev->node_id == node_id)
1127 			return sbridge_dev->mci;
1128 	}
1129 	return NULL;
1130 }
1131 
get_memory_error_data(struct mem_ctl_info * mci,u64 addr,u8 * socket,long * channel_mask,u8 * rank,char ** area_type,char * msg)1132 static int get_memory_error_data(struct mem_ctl_info *mci,
1133 				 u64 addr,
1134 				 u8 *socket,
1135 				 long *channel_mask,
1136 				 u8 *rank,
1137 				 char **area_type, char *msg)
1138 {
1139 	struct mem_ctl_info	*new_mci;
1140 	struct sbridge_pvt *pvt = mci->pvt_info;
1141 	struct pci_dev		*pci_ha;
1142 	int			n_rir, n_sads, n_tads, sad_way, sck_xch;
1143 	int			sad_interl, idx, base_ch;
1144 	int			interleave_mode, shiftup = 0;
1145 	unsigned		sad_interleave[pvt->info.max_interleave];
1146 	u32			reg, dram_rule;
1147 	u8			ch_way, sck_way, pkg, sad_ha = 0;
1148 	u32			tad_offset;
1149 	u32			rir_way;
1150 	u32			mb, gb;
1151 	u64			ch_addr, offset, limit = 0, prv = 0;
1152 
1153 
1154 	/*
1155 	 * Step 0) Check if the address is at special memory ranges
1156 	 * The check bellow is probably enough to fill all cases where
1157 	 * the error is not inside a memory, except for the legacy
1158 	 * range (e. g. VGA addresses). It is unlikely, however, that the
1159 	 * memory controller would generate an error on that range.
1160 	 */
1161 	if ((addr > (u64) pvt->tolm) && (addr < (1LL << 32))) {
1162 		sprintf(msg, "Error at TOLM area, on addr 0x%08Lx", addr);
1163 		return -EINVAL;
1164 	}
1165 	if (addr >= (u64)pvt->tohm) {
1166 		sprintf(msg, "Error at MMIOH area, on addr 0x%016Lx", addr);
1167 		return -EINVAL;
1168 	}
1169 
1170 	/*
1171 	 * Step 1) Get socket
1172 	 */
1173 	for (n_sads = 0; n_sads < pvt->info.max_sad; n_sads++) {
1174 		pci_read_config_dword(pvt->pci_sad0, pvt->info.dram_rule[n_sads],
1175 				      &reg);
1176 
1177 		if (!DRAM_RULE_ENABLE(reg))
1178 			continue;
1179 
1180 		limit = SAD_LIMIT(reg);
1181 		if (limit <= prv) {
1182 			sprintf(msg, "Can't discover the memory socket");
1183 			return -EINVAL;
1184 		}
1185 		if  (addr <= limit)
1186 			break;
1187 		prv = limit;
1188 	}
1189 	if (n_sads == pvt->info.max_sad) {
1190 		sprintf(msg, "Can't discover the memory socket");
1191 		return -EINVAL;
1192 	}
1193 	dram_rule = reg;
1194 	*area_type = get_dram_attr(dram_rule);
1195 	interleave_mode = INTERLEAVE_MODE(dram_rule);
1196 
1197 	pci_read_config_dword(pvt->pci_sad0, pvt->info.interleave_list[n_sads],
1198 			      &reg);
1199 
1200 	if (pvt->info.type == SANDY_BRIDGE) {
1201 		sad_interl = sad_pkg(pvt->info.interleave_pkg, reg, 0);
1202 		for (sad_way = 0; sad_way < 8; sad_way++) {
1203 			u32 pkg = sad_pkg(pvt->info.interleave_pkg, reg, sad_way);
1204 			if (sad_way > 0 && sad_interl == pkg)
1205 				break;
1206 			sad_interleave[sad_way] = pkg;
1207 			edac_dbg(0, "SAD interleave #%d: %d\n",
1208 				 sad_way, sad_interleave[sad_way]);
1209 		}
1210 		edac_dbg(0, "mc#%d: Error detected on SAD#%d: address 0x%016Lx < 0x%016Lx, Interleave [%d:6]%s\n",
1211 			 pvt->sbridge_dev->mc,
1212 			 n_sads,
1213 			 addr,
1214 			 limit,
1215 			 sad_way + 7,
1216 			 !interleave_mode ? "" : "XOR[18:16]");
1217 		if (interleave_mode)
1218 			idx = ((addr >> 6) ^ (addr >> 16)) & 7;
1219 		else
1220 			idx = (addr >> 6) & 7;
1221 		switch (sad_way) {
1222 		case 1:
1223 			idx = 0;
1224 			break;
1225 		case 2:
1226 			idx = idx & 1;
1227 			break;
1228 		case 4:
1229 			idx = idx & 3;
1230 			break;
1231 		case 8:
1232 			break;
1233 		default:
1234 			sprintf(msg, "Can't discover socket interleave");
1235 			return -EINVAL;
1236 		}
1237 		*socket = sad_interleave[idx];
1238 		edac_dbg(0, "SAD interleave index: %d (wayness %d) = CPU socket %d\n",
1239 			 idx, sad_way, *socket);
1240 	} else if (pvt->info.type == HASWELL || pvt->info.type == BROADWELL) {
1241 		int bits, a7mode = A7MODE(dram_rule);
1242 
1243 		if (a7mode) {
1244 			/* A7 mode swaps P9 with P6 */
1245 			bits = GET_BITFIELD(addr, 7, 8) << 1;
1246 			bits |= GET_BITFIELD(addr, 9, 9);
1247 		} else
1248 			bits = GET_BITFIELD(addr, 7, 9);
1249 
1250 		if (interleave_mode) {
1251 			/* interleave mode will XOR {8,7,6} with {18,17,16} */
1252 			idx = GET_BITFIELD(addr, 16, 18);
1253 			idx ^= bits;
1254 		} else
1255 			idx = bits;
1256 
1257 		pkg = sad_pkg(pvt->info.interleave_pkg, reg, idx);
1258 		*socket = sad_pkg_socket(pkg);
1259 		sad_ha = sad_pkg_ha(pkg);
1260 
1261 		if (a7mode) {
1262 			/* MCChanShiftUpEnable */
1263 			pci_read_config_dword(pvt->pci_ha0,
1264 					      HASWELL_HASYSDEFEATURE2, &reg);
1265 			shiftup = GET_BITFIELD(reg, 22, 22);
1266 		}
1267 
1268 		edac_dbg(0, "SAD interleave package: %d = CPU socket %d, HA %i, shiftup: %i\n",
1269 			 idx, *socket, sad_ha, shiftup);
1270 	} else {
1271 		/* Ivy Bridge's SAD mode doesn't support XOR interleave mode */
1272 		idx = (addr >> 6) & 7;
1273 		pkg = sad_pkg(pvt->info.interleave_pkg, reg, idx);
1274 		*socket = sad_pkg_socket(pkg);
1275 		sad_ha = sad_pkg_ha(pkg);
1276 		edac_dbg(0, "SAD interleave package: %d = CPU socket %d, HA %d\n",
1277 			 idx, *socket, sad_ha);
1278 	}
1279 
1280 	/*
1281 	 * Move to the proper node structure, in order to access the
1282 	 * right PCI registers
1283 	 */
1284 	new_mci = get_mci_for_node_id(*socket);
1285 	if (!new_mci) {
1286 		sprintf(msg, "Struct for socket #%u wasn't initialized",
1287 			*socket);
1288 		return -EINVAL;
1289 	}
1290 	mci = new_mci;
1291 	pvt = mci->pvt_info;
1292 
1293 	/*
1294 	 * Step 2) Get memory channel
1295 	 */
1296 	prv = 0;
1297 	if (pvt->info.type == SANDY_BRIDGE)
1298 		pci_ha = pvt->pci_ha0;
1299 	else {
1300 		if (sad_ha)
1301 			pci_ha = pvt->pci_ha1;
1302 		else
1303 			pci_ha = pvt->pci_ha0;
1304 	}
1305 	for (n_tads = 0; n_tads < MAX_TAD; n_tads++) {
1306 		pci_read_config_dword(pci_ha, tad_dram_rule[n_tads], &reg);
1307 		limit = TAD_LIMIT(reg);
1308 		if (limit <= prv) {
1309 			sprintf(msg, "Can't discover the memory channel");
1310 			return -EINVAL;
1311 		}
1312 		if  (addr <= limit)
1313 			break;
1314 		prv = limit;
1315 	}
1316 	if (n_tads == MAX_TAD) {
1317 		sprintf(msg, "Can't discover the memory channel");
1318 		return -EINVAL;
1319 	}
1320 
1321 	ch_way = TAD_CH(reg) + 1;
1322 	sck_way = TAD_SOCK(reg);
1323 
1324 	if (ch_way == 3)
1325 		idx = addr >> 6;
1326 	else
1327 		idx = (addr >> (6 + sck_way + shiftup)) & 0x3;
1328 	idx = idx % ch_way;
1329 
1330 	/*
1331 	 * FIXME: Shouldn't we use CHN_IDX_OFFSET() here, when ch_way == 3 ???
1332 	 */
1333 	switch (idx) {
1334 	case 0:
1335 		base_ch = TAD_TGT0(reg);
1336 		break;
1337 	case 1:
1338 		base_ch = TAD_TGT1(reg);
1339 		break;
1340 	case 2:
1341 		base_ch = TAD_TGT2(reg);
1342 		break;
1343 	case 3:
1344 		base_ch = TAD_TGT3(reg);
1345 		break;
1346 	default:
1347 		sprintf(msg, "Can't discover the TAD target");
1348 		return -EINVAL;
1349 	}
1350 	*channel_mask = 1 << base_ch;
1351 
1352 	pci_read_config_dword(pvt->pci_tad[base_ch],
1353 				tad_ch_nilv_offset[n_tads],
1354 				&tad_offset);
1355 
1356 	if (pvt->is_mirrored) {
1357 		*channel_mask |= 1 << ((base_ch + 2) % 4);
1358 		switch(ch_way) {
1359 		case 2:
1360 		case 4:
1361 			sck_xch = (1 << sck_way) * (ch_way >> 1);
1362 			break;
1363 		default:
1364 			sprintf(msg, "Invalid mirror set. Can't decode addr");
1365 			return -EINVAL;
1366 		}
1367 	} else
1368 		sck_xch = (1 << sck_way) * ch_way;
1369 
1370 	if (pvt->is_lockstep)
1371 		*channel_mask |= 1 << ((base_ch + 1) % 4);
1372 
1373 	offset = TAD_OFFSET(tad_offset);
1374 
1375 	edac_dbg(0, "TAD#%d: address 0x%016Lx < 0x%016Lx, socket interleave %d, channel interleave %d (offset 0x%08Lx), index %d, base ch: %d, ch mask: 0x%02lx\n",
1376 		 n_tads,
1377 		 addr,
1378 		 limit,
1379 		 sck_way,
1380 		 ch_way,
1381 		 offset,
1382 		 idx,
1383 		 base_ch,
1384 		 *channel_mask);
1385 
1386 	/* Calculate channel address */
1387 	/* Remove the TAD offset */
1388 
1389 	if (offset > addr) {
1390 		sprintf(msg, "Can't calculate ch addr: TAD offset 0x%08Lx is too high for addr 0x%08Lx!",
1391 			offset, addr);
1392 		return -EINVAL;
1393 	}
1394 
1395 	ch_addr = addr - offset;
1396 	ch_addr >>= (6 + shiftup);
1397 	ch_addr /= sck_xch;
1398 	ch_addr <<= (6 + shiftup);
1399 	ch_addr |= addr & ((1 << (6 + shiftup)) - 1);
1400 
1401 	/*
1402 	 * Step 3) Decode rank
1403 	 */
1404 	for (n_rir = 0; n_rir < MAX_RIR_RANGES; n_rir++) {
1405 		pci_read_config_dword(pvt->pci_tad[base_ch],
1406 				      rir_way_limit[n_rir],
1407 				      &reg);
1408 
1409 		if (!IS_RIR_VALID(reg))
1410 			continue;
1411 
1412 		limit = pvt->info.rir_limit(reg);
1413 		gb = div_u64_rem(limit >> 20, 1024, &mb);
1414 		edac_dbg(0, "RIR#%d, limit: %u.%03u GB (0x%016Lx), way: %d\n",
1415 			 n_rir,
1416 			 gb, (mb*1000)/1024,
1417 			 limit,
1418 			 1 << RIR_WAY(reg));
1419 		if  (ch_addr <= limit)
1420 			break;
1421 	}
1422 	if (n_rir == MAX_RIR_RANGES) {
1423 		sprintf(msg, "Can't discover the memory rank for ch addr 0x%08Lx",
1424 			ch_addr);
1425 		return -EINVAL;
1426 	}
1427 	rir_way = RIR_WAY(reg);
1428 
1429 	if (pvt->is_close_pg)
1430 		idx = (ch_addr >> 6);
1431 	else
1432 		idx = (ch_addr >> 13);	/* FIXME: Datasheet says to shift by 15 */
1433 	idx %= 1 << rir_way;
1434 
1435 	pci_read_config_dword(pvt->pci_tad[base_ch],
1436 			      rir_offset[n_rir][idx],
1437 			      &reg);
1438 	*rank = RIR_RNK_TGT(pvt->info.type, reg);
1439 
1440 	edac_dbg(0, "RIR#%d: channel address 0x%08Lx < 0x%08Lx, RIR interleave %d, index %d\n",
1441 		 n_rir,
1442 		 ch_addr,
1443 		 limit,
1444 		 rir_way,
1445 		 idx);
1446 
1447 	return 0;
1448 }
1449 
1450 /****************************************************************************
1451 	Device initialization routines: put/get, init/exit
1452  ****************************************************************************/
1453 
1454 /*
1455  *	sbridge_put_all_devices	'put' all the devices that we have
1456  *				reserved via 'get'
1457  */
sbridge_put_devices(struct sbridge_dev * sbridge_dev)1458 static void sbridge_put_devices(struct sbridge_dev *sbridge_dev)
1459 {
1460 	int i;
1461 
1462 	edac_dbg(0, "\n");
1463 	for (i = 0; i < sbridge_dev->n_devs; i++) {
1464 		struct pci_dev *pdev = sbridge_dev->pdev[i];
1465 		if (!pdev)
1466 			continue;
1467 		edac_dbg(0, "Removing dev %02x:%02x.%d\n",
1468 			 pdev->bus->number,
1469 			 PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn));
1470 		pci_dev_put(pdev);
1471 	}
1472 }
1473 
sbridge_put_all_devices(void)1474 static void sbridge_put_all_devices(void)
1475 {
1476 	struct sbridge_dev *sbridge_dev, *tmp;
1477 
1478 	list_for_each_entry_safe(sbridge_dev, tmp, &sbridge_edac_list, list) {
1479 		sbridge_put_devices(sbridge_dev);
1480 		free_sbridge_dev(sbridge_dev);
1481 	}
1482 }
1483 
sbridge_get_onedevice(struct pci_dev ** prev,u8 * num_mc,const struct pci_id_table * table,const unsigned devno)1484 static int sbridge_get_onedevice(struct pci_dev **prev,
1485 				 u8 *num_mc,
1486 				 const struct pci_id_table *table,
1487 				 const unsigned devno)
1488 {
1489 	struct sbridge_dev *sbridge_dev;
1490 	const struct pci_id_descr *dev_descr = &table->descr[devno];
1491 	struct pci_dev *pdev = NULL;
1492 	u8 bus = 0;
1493 
1494 	sbridge_printk(KERN_DEBUG,
1495 		"Seeking for: PCI ID %04x:%04x\n",
1496 		PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
1497 
1498 	pdev = pci_get_device(PCI_VENDOR_ID_INTEL,
1499 			      dev_descr->dev_id, *prev);
1500 
1501 	if (!pdev) {
1502 		if (*prev) {
1503 			*prev = pdev;
1504 			return 0;
1505 		}
1506 
1507 		if (dev_descr->optional)
1508 			return 0;
1509 
1510 		/* if the HA wasn't found */
1511 		if (devno == 0)
1512 			return -ENODEV;
1513 
1514 		sbridge_printk(KERN_INFO,
1515 			"Device not found: %04x:%04x\n",
1516 			PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
1517 
1518 		/* End of list, leave */
1519 		return -ENODEV;
1520 	}
1521 	bus = pdev->bus->number;
1522 
1523 	sbridge_dev = get_sbridge_dev(bus);
1524 	if (!sbridge_dev) {
1525 		sbridge_dev = alloc_sbridge_dev(bus, table);
1526 		if (!sbridge_dev) {
1527 			pci_dev_put(pdev);
1528 			return -ENOMEM;
1529 		}
1530 		(*num_mc)++;
1531 	}
1532 
1533 	if (sbridge_dev->pdev[devno]) {
1534 		sbridge_printk(KERN_ERR,
1535 			"Duplicated device for %04x:%04x\n",
1536 			PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
1537 		pci_dev_put(pdev);
1538 		return -ENODEV;
1539 	}
1540 
1541 	sbridge_dev->pdev[devno] = pdev;
1542 
1543 	/* Be sure that the device is enabled */
1544 	if (unlikely(pci_enable_device(pdev) < 0)) {
1545 		sbridge_printk(KERN_ERR,
1546 			"Couldn't enable %04x:%04x\n",
1547 			PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
1548 		return -ENODEV;
1549 	}
1550 
1551 	edac_dbg(0, "Detected %04x:%04x\n",
1552 		 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
1553 
1554 	/*
1555 	 * As stated on drivers/pci/search.c, the reference count for
1556 	 * @from is always decremented if it is not %NULL. So, as we need
1557 	 * to get all devices up to null, we need to do a get for the device
1558 	 */
1559 	pci_dev_get(pdev);
1560 
1561 	*prev = pdev;
1562 
1563 	return 0;
1564 }
1565 
1566 /*
1567  * sbridge_get_all_devices - Find and perform 'get' operation on the MCH's
1568  *			     devices we want to reference for this driver.
1569  * @num_mc: pointer to the memory controllers count, to be incremented in case
1570  *	    of success.
1571  * @table: model specific table
1572  *
1573  * returns 0 in case of success or error code
1574  */
sbridge_get_all_devices(u8 * num_mc,const struct pci_id_table * table)1575 static int sbridge_get_all_devices(u8 *num_mc,
1576 				   const struct pci_id_table *table)
1577 {
1578 	int i, rc;
1579 	struct pci_dev *pdev = NULL;
1580 
1581 	while (table && table->descr) {
1582 		for (i = 0; i < table->n_devs; i++) {
1583 			pdev = NULL;
1584 			do {
1585 				rc = sbridge_get_onedevice(&pdev, num_mc,
1586 							   table, i);
1587 				if (rc < 0) {
1588 					if (i == 0) {
1589 						i = table->n_devs;
1590 						break;
1591 					}
1592 					sbridge_put_all_devices();
1593 					return -ENODEV;
1594 				}
1595 			} while (pdev);
1596 		}
1597 		table++;
1598 	}
1599 
1600 	return 0;
1601 }
1602 
sbridge_mci_bind_devs(struct mem_ctl_info * mci,struct sbridge_dev * sbridge_dev)1603 static int sbridge_mci_bind_devs(struct mem_ctl_info *mci,
1604 				 struct sbridge_dev *sbridge_dev)
1605 {
1606 	struct sbridge_pvt *pvt = mci->pvt_info;
1607 	struct pci_dev *pdev;
1608 	u8 saw_chan_mask = 0;
1609 	int i;
1610 
1611 	for (i = 0; i < sbridge_dev->n_devs; i++) {
1612 		pdev = sbridge_dev->pdev[i];
1613 		if (!pdev)
1614 			continue;
1615 
1616 		switch (pdev->device) {
1617 		case PCI_DEVICE_ID_INTEL_SBRIDGE_SAD0:
1618 			pvt->pci_sad0 = pdev;
1619 			break;
1620 		case PCI_DEVICE_ID_INTEL_SBRIDGE_SAD1:
1621 			pvt->pci_sad1 = pdev;
1622 			break;
1623 		case PCI_DEVICE_ID_INTEL_SBRIDGE_BR:
1624 			pvt->pci_br0 = pdev;
1625 			break;
1626 		case PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_HA0:
1627 			pvt->pci_ha0 = pdev;
1628 			break;
1629 		case PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TA:
1630 			pvt->pci_ta = pdev;
1631 			break;
1632 		case PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_RAS:
1633 			pvt->pci_ras = pdev;
1634 			break;
1635 		case PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD0:
1636 		case PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD1:
1637 		case PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD2:
1638 		case PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD3:
1639 		{
1640 			int id = pdev->device - PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD0;
1641 			pvt->pci_tad[id] = pdev;
1642 			saw_chan_mask |= 1 << id;
1643 		}
1644 			break;
1645 		case PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_DDRIO:
1646 			pvt->pci_ddrio = pdev;
1647 			break;
1648 		default:
1649 			goto error;
1650 		}
1651 
1652 		edac_dbg(0, "Associated PCI %02x:%02x, bus %d with dev = %p\n",
1653 			 pdev->vendor, pdev->device,
1654 			 sbridge_dev->bus,
1655 			 pdev);
1656 	}
1657 
1658 	/* Check if everything were registered */
1659 	if (!pvt->pci_sad0 || !pvt->pci_sad1 || !pvt->pci_ha0 ||
1660 	    !pvt-> pci_tad || !pvt->pci_ras  || !pvt->pci_ta)
1661 		goto enodev;
1662 
1663 	if (saw_chan_mask != 0x0f)
1664 		goto enodev;
1665 	return 0;
1666 
1667 enodev:
1668 	sbridge_printk(KERN_ERR, "Some needed devices are missing\n");
1669 	return -ENODEV;
1670 
1671 error:
1672 	sbridge_printk(KERN_ERR, "Unexpected device %02x:%02x\n",
1673 		       PCI_VENDOR_ID_INTEL, pdev->device);
1674 	return -EINVAL;
1675 }
1676 
ibridge_mci_bind_devs(struct mem_ctl_info * mci,struct sbridge_dev * sbridge_dev)1677 static int ibridge_mci_bind_devs(struct mem_ctl_info *mci,
1678 				 struct sbridge_dev *sbridge_dev)
1679 {
1680 	struct sbridge_pvt *pvt = mci->pvt_info;
1681 	struct pci_dev *pdev, *tmp;
1682 	int i;
1683 	bool mode_2ha = false;
1684 
1685 	tmp = pci_get_device(PCI_VENDOR_ID_INTEL,
1686 			     PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1, NULL);
1687 	if (tmp) {
1688 		mode_2ha = true;
1689 		pci_dev_put(tmp);
1690 	}
1691 
1692 	for (i = 0; i < sbridge_dev->n_devs; i++) {
1693 		pdev = sbridge_dev->pdev[i];
1694 		if (!pdev)
1695 			continue;
1696 
1697 		switch (pdev->device) {
1698 		case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0:
1699 			pvt->pci_ha0 = pdev;
1700 			break;
1701 		case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TA:
1702 			pvt->pci_ta = pdev;
1703 		case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_RAS:
1704 			pvt->pci_ras = pdev;
1705 			break;
1706 		case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD2:
1707 		case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD3:
1708 			/* if we have 2 HAs active, channels 2 and 3
1709 			 * are in other device */
1710 			if (mode_2ha)
1711 				break;
1712 			/* fall through */
1713 		case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD0:
1714 		case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD1:
1715 		{
1716 			int id = pdev->device - PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD0;
1717 			pvt->pci_tad[id] = pdev;
1718 		}
1719 			break;
1720 		case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_2HA_DDRIO0:
1721 			pvt->pci_ddrio = pdev;
1722 			break;
1723 		case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_1HA_DDRIO0:
1724 			if (!mode_2ha)
1725 				pvt->pci_ddrio = pdev;
1726 			break;
1727 		case PCI_DEVICE_ID_INTEL_IBRIDGE_SAD:
1728 			pvt->pci_sad0 = pdev;
1729 			break;
1730 		case PCI_DEVICE_ID_INTEL_IBRIDGE_BR0:
1731 			pvt->pci_br0 = pdev;
1732 			break;
1733 		case PCI_DEVICE_ID_INTEL_IBRIDGE_BR1:
1734 			pvt->pci_br1 = pdev;
1735 			break;
1736 		case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1:
1737 			pvt->pci_ha1 = pdev;
1738 			break;
1739 		case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD0:
1740 		case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD1:
1741 		{
1742 			int id = pdev->device - PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD0 + 2;
1743 
1744 			/* we shouldn't have this device if we have just one
1745 			 * HA present */
1746 			WARN_ON(!mode_2ha);
1747 			pvt->pci_tad[id] = pdev;
1748 		}
1749 			break;
1750 		default:
1751 			goto error;
1752 		}
1753 
1754 		edac_dbg(0, "Associated PCI %02x.%02d.%d with dev = %p\n",
1755 			 sbridge_dev->bus,
1756 			 PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn),
1757 			 pdev);
1758 	}
1759 
1760 	/* Check if everything were registered */
1761 	if (!pvt->pci_sad0 || !pvt->pci_ha0 || !pvt->pci_br0 ||
1762 	    !pvt->pci_br1 || !pvt->pci_tad || !pvt->pci_ras  ||
1763 	    !pvt->pci_ta)
1764 		goto enodev;
1765 
1766 	for (i = 0; i < NUM_CHANNELS; i++) {
1767 		if (!pvt->pci_tad[i])
1768 			goto enodev;
1769 	}
1770 	return 0;
1771 
1772 enodev:
1773 	sbridge_printk(KERN_ERR, "Some needed devices are missing\n");
1774 	return -ENODEV;
1775 
1776 error:
1777 	sbridge_printk(KERN_ERR,
1778 		       "Unexpected device %02x:%02x\n", PCI_VENDOR_ID_INTEL,
1779 			pdev->device);
1780 	return -EINVAL;
1781 }
1782 
haswell_mci_bind_devs(struct mem_ctl_info * mci,struct sbridge_dev * sbridge_dev)1783 static int haswell_mci_bind_devs(struct mem_ctl_info *mci,
1784 				 struct sbridge_dev *sbridge_dev)
1785 {
1786 	struct sbridge_pvt *pvt = mci->pvt_info;
1787 	struct pci_dev *pdev, *tmp;
1788 	int i;
1789 	bool mode_2ha = false;
1790 
1791 	tmp = pci_get_device(PCI_VENDOR_ID_INTEL,
1792 			     PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1, NULL);
1793 	if (tmp) {
1794 		mode_2ha = true;
1795 		pci_dev_put(tmp);
1796 	}
1797 
1798 	/* there's only one device per system; not tied to any bus */
1799 	if (pvt->info.pci_vtd == NULL)
1800 		/* result will be checked later */
1801 		pvt->info.pci_vtd = pci_get_device(PCI_VENDOR_ID_INTEL,
1802 						   PCI_DEVICE_ID_INTEL_HASWELL_IMC_VTD_MISC,
1803 						   NULL);
1804 
1805 	for (i = 0; i < sbridge_dev->n_devs; i++) {
1806 		pdev = sbridge_dev->pdev[i];
1807 		if (!pdev)
1808 			continue;
1809 
1810 		switch (pdev->device) {
1811 		case PCI_DEVICE_ID_INTEL_HASWELL_IMC_CBO_SAD0:
1812 			pvt->pci_sad0 = pdev;
1813 			break;
1814 		case PCI_DEVICE_ID_INTEL_HASWELL_IMC_CBO_SAD1:
1815 			pvt->pci_sad1 = pdev;
1816 			break;
1817 		case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0:
1818 			pvt->pci_ha0 = pdev;
1819 			break;
1820 		case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TA:
1821 			pvt->pci_ta = pdev;
1822 			break;
1823 		case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_THERMAL:
1824 			pvt->pci_ras = pdev;
1825 			break;
1826 		case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD0:
1827 			pvt->pci_tad[0] = pdev;
1828 			break;
1829 		case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD1:
1830 			pvt->pci_tad[1] = pdev;
1831 			break;
1832 		case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD2:
1833 			if (!mode_2ha)
1834 				pvt->pci_tad[2] = pdev;
1835 			break;
1836 		case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD3:
1837 			if (!mode_2ha)
1838 				pvt->pci_tad[3] = pdev;
1839 			break;
1840 		case PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO0:
1841 			pvt->pci_ddrio = pdev;
1842 			break;
1843 		case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1:
1844 			pvt->pci_ha1 = pdev;
1845 			break;
1846 		case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TA:
1847 			pvt->pci_ha1_ta = pdev;
1848 			break;
1849 		case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD0:
1850 			if (mode_2ha)
1851 				pvt->pci_tad[2] = pdev;
1852 			break;
1853 		case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD1:
1854 			if (mode_2ha)
1855 				pvt->pci_tad[3] = pdev;
1856 			break;
1857 		default:
1858 			break;
1859 		}
1860 
1861 		edac_dbg(0, "Associated PCI %02x.%02d.%d with dev = %p\n",
1862 			 sbridge_dev->bus,
1863 			 PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn),
1864 			 pdev);
1865 	}
1866 
1867 	/* Check if everything were registered */
1868 	if (!pvt->pci_sad0 || !pvt->pci_ha0 || !pvt->pci_sad1 ||
1869 	    !pvt->pci_ras  || !pvt->pci_ta || !pvt->info.pci_vtd)
1870 		goto enodev;
1871 
1872 	for (i = 0; i < NUM_CHANNELS; i++) {
1873 		if (!pvt->pci_tad[i])
1874 			goto enodev;
1875 	}
1876 	return 0;
1877 
1878 enodev:
1879 	sbridge_printk(KERN_ERR, "Some needed devices are missing\n");
1880 	return -ENODEV;
1881 }
1882 
broadwell_mci_bind_devs(struct mem_ctl_info * mci,struct sbridge_dev * sbridge_dev)1883 static int broadwell_mci_bind_devs(struct mem_ctl_info *mci,
1884 				 struct sbridge_dev *sbridge_dev)
1885 {
1886 	struct sbridge_pvt *pvt = mci->pvt_info;
1887 	struct pci_dev *pdev;
1888 	int i;
1889 
1890 	/* there's only one device per system; not tied to any bus */
1891 	if (pvt->info.pci_vtd == NULL)
1892 		/* result will be checked later */
1893 		pvt->info.pci_vtd = pci_get_device(PCI_VENDOR_ID_INTEL,
1894 						   PCI_DEVICE_ID_INTEL_BROADWELL_IMC_VTD_MISC,
1895 						   NULL);
1896 
1897 	for (i = 0; i < sbridge_dev->n_devs; i++) {
1898 		pdev = sbridge_dev->pdev[i];
1899 		if (!pdev)
1900 			continue;
1901 
1902 		switch (pdev->device) {
1903 		case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_CBO_SAD0:
1904 			pvt->pci_sad0 = pdev;
1905 			break;
1906 		case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_CBO_SAD1:
1907 			pvt->pci_sad1 = pdev;
1908 			break;
1909 		case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0:
1910 			pvt->pci_ha0 = pdev;
1911 			break;
1912 		case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TA:
1913 			pvt->pci_ta = pdev;
1914 			break;
1915 		case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_THERMAL:
1916 			pvt->pci_ras = pdev;
1917 			break;
1918 		case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD0:
1919 			pvt->pci_tad[0] = pdev;
1920 			break;
1921 		case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD1:
1922 			pvt->pci_tad[1] = pdev;
1923 			break;
1924 		case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD2:
1925 			pvt->pci_tad[2] = pdev;
1926 			break;
1927 		case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD3:
1928 			pvt->pci_tad[3] = pdev;
1929 			break;
1930 		case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_DDRIO0:
1931 			pvt->pci_ddrio = pdev;
1932 			break;
1933 		default:
1934 			break;
1935 		}
1936 
1937 		edac_dbg(0, "Associated PCI %02x.%02d.%d with dev = %p\n",
1938 			 sbridge_dev->bus,
1939 			 PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn),
1940 			 pdev);
1941 	}
1942 
1943 	/* Check if everything were registered */
1944 	if (!pvt->pci_sad0 || !pvt->pci_ha0 || !pvt->pci_sad1 ||
1945 	    !pvt->pci_ras  || !pvt->pci_ta || !pvt->info.pci_vtd)
1946 		goto enodev;
1947 
1948 	for (i = 0; i < NUM_CHANNELS; i++) {
1949 		if (!pvt->pci_tad[i])
1950 			goto enodev;
1951 	}
1952 	return 0;
1953 
1954 enodev:
1955 	sbridge_printk(KERN_ERR, "Some needed devices are missing\n");
1956 	return -ENODEV;
1957 }
1958 
1959 /****************************************************************************
1960 			Error check routines
1961  ****************************************************************************/
1962 
1963 /*
1964  * While Sandy Bridge has error count registers, SMI BIOS read values from
1965  * and resets the counters. So, they are not reliable for the OS to read
1966  * from them. So, we have no option but to just trust on whatever MCE is
1967  * telling us about the errors.
1968  */
sbridge_mce_output_error(struct mem_ctl_info * mci,const struct mce * m)1969 static void sbridge_mce_output_error(struct mem_ctl_info *mci,
1970 				    const struct mce *m)
1971 {
1972 	struct mem_ctl_info *new_mci;
1973 	struct sbridge_pvt *pvt = mci->pvt_info;
1974 	enum hw_event_mc_err_type tp_event;
1975 	char *type, *optype, msg[256];
1976 	bool ripv = GET_BITFIELD(m->mcgstatus, 0, 0);
1977 	bool overflow = GET_BITFIELD(m->status, 62, 62);
1978 	bool uncorrected_error = GET_BITFIELD(m->status, 61, 61);
1979 	bool recoverable;
1980 	u32 core_err_cnt = GET_BITFIELD(m->status, 38, 52);
1981 	u32 mscod = GET_BITFIELD(m->status, 16, 31);
1982 	u32 errcode = GET_BITFIELD(m->status, 0, 15);
1983 	u32 channel = GET_BITFIELD(m->status, 0, 3);
1984 	u32 optypenum = GET_BITFIELD(m->status, 4, 6);
1985 	long channel_mask, first_channel;
1986 	u8  rank, socket;
1987 	int rc, dimm;
1988 	char *area_type = NULL;
1989 
1990 	if (pvt->info.type == IVY_BRIDGE)
1991 		recoverable = true;
1992 	else
1993 		recoverable = GET_BITFIELD(m->status, 56, 56);
1994 
1995 	if (uncorrected_error) {
1996 		if (ripv) {
1997 			type = "FATAL";
1998 			tp_event = HW_EVENT_ERR_FATAL;
1999 		} else {
2000 			type = "NON_FATAL";
2001 			tp_event = HW_EVENT_ERR_UNCORRECTED;
2002 		}
2003 	} else {
2004 		type = "CORRECTED";
2005 		tp_event = HW_EVENT_ERR_CORRECTED;
2006 	}
2007 
2008 	/*
2009 	 * According with Table 15-9 of the Intel Architecture spec vol 3A,
2010 	 * memory errors should fit in this mask:
2011 	 *	000f 0000 1mmm cccc (binary)
2012 	 * where:
2013 	 *	f = Correction Report Filtering Bit. If 1, subsequent errors
2014 	 *	    won't be shown
2015 	 *	mmm = error type
2016 	 *	cccc = channel
2017 	 * If the mask doesn't match, report an error to the parsing logic
2018 	 */
2019 	if (! ((errcode & 0xef80) == 0x80)) {
2020 		optype = "Can't parse: it is not a mem";
2021 	} else {
2022 		switch (optypenum) {
2023 		case 0:
2024 			optype = "generic undef request error";
2025 			break;
2026 		case 1:
2027 			optype = "memory read error";
2028 			break;
2029 		case 2:
2030 			optype = "memory write error";
2031 			break;
2032 		case 3:
2033 			optype = "addr/cmd error";
2034 			break;
2035 		case 4:
2036 			optype = "memory scrubbing error";
2037 			break;
2038 		default:
2039 			optype = "reserved";
2040 			break;
2041 		}
2042 	}
2043 
2044 	/* Only decode errors with an valid address (ADDRV) */
2045 	if (!GET_BITFIELD(m->status, 58, 58))
2046 		return;
2047 
2048 	rc = get_memory_error_data(mci, m->addr, &socket,
2049 				   &channel_mask, &rank, &area_type, msg);
2050 	if (rc < 0)
2051 		goto err_parsing;
2052 	new_mci = get_mci_for_node_id(socket);
2053 	if (!new_mci) {
2054 		strcpy(msg, "Error: socket got corrupted!");
2055 		goto err_parsing;
2056 	}
2057 	mci = new_mci;
2058 	pvt = mci->pvt_info;
2059 
2060 	first_channel = find_first_bit(&channel_mask, NUM_CHANNELS);
2061 
2062 	if (rank < 4)
2063 		dimm = 0;
2064 	else if (rank < 8)
2065 		dimm = 1;
2066 	else
2067 		dimm = 2;
2068 
2069 
2070 	/*
2071 	 * FIXME: On some memory configurations (mirror, lockstep), the
2072 	 * Memory Controller can't point the error to a single DIMM. The
2073 	 * EDAC core should be handling the channel mask, in order to point
2074 	 * to the group of dimm's where the error may be happening.
2075 	 */
2076 	if (!pvt->is_lockstep && !pvt->is_mirrored && !pvt->is_close_pg)
2077 		channel = first_channel;
2078 
2079 	snprintf(msg, sizeof(msg),
2080 		 "%s%s area:%s err_code:%04x:%04x socket:%d channel_mask:%ld rank:%d",
2081 		 overflow ? " OVERFLOW" : "",
2082 		 (uncorrected_error && recoverable) ? " recoverable" : "",
2083 		 area_type,
2084 		 mscod, errcode,
2085 		 socket,
2086 		 channel_mask,
2087 		 rank);
2088 
2089 	edac_dbg(0, "%s\n", msg);
2090 
2091 	/* FIXME: need support for channel mask */
2092 
2093 	if (channel == CHANNEL_UNSPECIFIED)
2094 		channel = -1;
2095 
2096 	/* Call the helper to output message */
2097 	edac_mc_handle_error(tp_event, mci, core_err_cnt,
2098 			     m->addr >> PAGE_SHIFT, m->addr & ~PAGE_MASK, 0,
2099 			     channel, dimm, -1,
2100 			     optype, msg);
2101 	return;
2102 err_parsing:
2103 	edac_mc_handle_error(tp_event, mci, core_err_cnt, 0, 0, 0,
2104 			     -1, -1, -1,
2105 			     msg, "");
2106 
2107 }
2108 
2109 /*
2110  *	sbridge_check_error	Retrieve and process errors reported by the
2111  *				hardware. Called by the Core module.
2112  */
sbridge_check_error(struct mem_ctl_info * mci)2113 static void sbridge_check_error(struct mem_ctl_info *mci)
2114 {
2115 	struct sbridge_pvt *pvt = mci->pvt_info;
2116 	int i;
2117 	unsigned count = 0;
2118 	struct mce *m;
2119 
2120 	/*
2121 	 * MCE first step: Copy all mce errors into a temporary buffer
2122 	 * We use a double buffering here, to reduce the risk of
2123 	 * loosing an error.
2124 	 */
2125 	smp_rmb();
2126 	count = (pvt->mce_out + MCE_LOG_LEN - pvt->mce_in)
2127 		% MCE_LOG_LEN;
2128 	if (!count)
2129 		return;
2130 
2131 	m = pvt->mce_outentry;
2132 	if (pvt->mce_in + count > MCE_LOG_LEN) {
2133 		unsigned l = MCE_LOG_LEN - pvt->mce_in;
2134 
2135 		memcpy(m, &pvt->mce_entry[pvt->mce_in], sizeof(*m) * l);
2136 		smp_wmb();
2137 		pvt->mce_in = 0;
2138 		count -= l;
2139 		m += l;
2140 	}
2141 	memcpy(m, &pvt->mce_entry[pvt->mce_in], sizeof(*m) * count);
2142 	smp_wmb();
2143 	pvt->mce_in += count;
2144 
2145 	smp_rmb();
2146 	if (pvt->mce_overrun) {
2147 		sbridge_printk(KERN_ERR, "Lost %d memory errors\n",
2148 			      pvt->mce_overrun);
2149 		smp_wmb();
2150 		pvt->mce_overrun = 0;
2151 	}
2152 
2153 	/*
2154 	 * MCE second step: parse errors and display
2155 	 */
2156 	for (i = 0; i < count; i++)
2157 		sbridge_mce_output_error(mci, &pvt->mce_outentry[i]);
2158 }
2159 
2160 /*
2161  * sbridge_mce_check_error	Replicates mcelog routine to get errors
2162  *				This routine simply queues mcelog errors, and
2163  *				return. The error itself should be handled later
2164  *				by sbridge_check_error.
2165  * WARNING: As this routine should be called at NMI time, extra care should
2166  * be taken to avoid deadlocks, and to be as fast as possible.
2167  */
sbridge_mce_check_error(struct notifier_block * nb,unsigned long val,void * data)2168 static int sbridge_mce_check_error(struct notifier_block *nb, unsigned long val,
2169 				   void *data)
2170 {
2171 	struct mce *mce = (struct mce *)data;
2172 	struct mem_ctl_info *mci;
2173 	struct sbridge_pvt *pvt;
2174 	char *type;
2175 
2176 	if (get_edac_report_status() == EDAC_REPORTING_DISABLED)
2177 		return NOTIFY_DONE;
2178 
2179 	mci = get_mci_for_node_id(mce->socketid);
2180 	if (!mci)
2181 		return NOTIFY_DONE;
2182 	pvt = mci->pvt_info;
2183 
2184 	/*
2185 	 * Just let mcelog handle it if the error is
2186 	 * outside the memory controller. A memory error
2187 	 * is indicated by bit 7 = 1 and bits = 8-11,13-15 = 0.
2188 	 * bit 12 has an special meaning.
2189 	 */
2190 	if ((mce->status & 0xefff) >> 7 != 1)
2191 		return NOTIFY_DONE;
2192 
2193 	if (mce->mcgstatus & MCG_STATUS_MCIP)
2194 		type = "Exception";
2195 	else
2196 		type = "Event";
2197 
2198 	sbridge_mc_printk(mci, KERN_DEBUG, "HANDLING MCE MEMORY ERROR\n");
2199 
2200 	sbridge_mc_printk(mci, KERN_DEBUG, "CPU %d: Machine Check %s: %Lx "
2201 			  "Bank %d: %016Lx\n", mce->extcpu, type,
2202 			  mce->mcgstatus, mce->bank, mce->status);
2203 	sbridge_mc_printk(mci, KERN_DEBUG, "TSC %llx ", mce->tsc);
2204 	sbridge_mc_printk(mci, KERN_DEBUG, "ADDR %llx ", mce->addr);
2205 	sbridge_mc_printk(mci, KERN_DEBUG, "MISC %llx ", mce->misc);
2206 
2207 	sbridge_mc_printk(mci, KERN_DEBUG, "PROCESSOR %u:%x TIME %llu SOCKET "
2208 			  "%u APIC %x\n", mce->cpuvendor, mce->cpuid,
2209 			  mce->time, mce->socketid, mce->apicid);
2210 
2211 	smp_rmb();
2212 	if ((pvt->mce_out + 1) % MCE_LOG_LEN == pvt->mce_in) {
2213 		smp_wmb();
2214 		pvt->mce_overrun++;
2215 		return NOTIFY_DONE;
2216 	}
2217 
2218 	/* Copy memory error at the ringbuffer */
2219 	memcpy(&pvt->mce_entry[pvt->mce_out], mce, sizeof(*mce));
2220 	smp_wmb();
2221 	pvt->mce_out = (pvt->mce_out + 1) % MCE_LOG_LEN;
2222 
2223 	/* Handle fatal errors immediately */
2224 	if (mce->mcgstatus & 1)
2225 		sbridge_check_error(mci);
2226 
2227 	/* Advice mcelog that the error were handled */
2228 	return NOTIFY_STOP;
2229 }
2230 
2231 static struct notifier_block sbridge_mce_dec = {
2232 	.notifier_call      = sbridge_mce_check_error,
2233 };
2234 
2235 /****************************************************************************
2236 			EDAC register/unregister logic
2237  ****************************************************************************/
2238 
sbridge_unregister_mci(struct sbridge_dev * sbridge_dev)2239 static void sbridge_unregister_mci(struct sbridge_dev *sbridge_dev)
2240 {
2241 	struct mem_ctl_info *mci = sbridge_dev->mci;
2242 	struct sbridge_pvt *pvt;
2243 
2244 	if (unlikely(!mci || !mci->pvt_info)) {
2245 		edac_dbg(0, "MC: dev = %p\n", &sbridge_dev->pdev[0]->dev);
2246 
2247 		sbridge_printk(KERN_ERR, "Couldn't find mci handler\n");
2248 		return;
2249 	}
2250 
2251 	pvt = mci->pvt_info;
2252 
2253 	edac_dbg(0, "MC: mci = %p, dev = %p\n",
2254 		 mci, &sbridge_dev->pdev[0]->dev);
2255 
2256 	/* Remove MC sysfs nodes */
2257 	edac_mc_del_mc(mci->pdev);
2258 
2259 	edac_dbg(1, "%s: free mci struct\n", mci->ctl_name);
2260 	kfree(mci->ctl_name);
2261 	edac_mc_free(mci);
2262 	sbridge_dev->mci = NULL;
2263 }
2264 
sbridge_register_mci(struct sbridge_dev * sbridge_dev,enum type type)2265 static int sbridge_register_mci(struct sbridge_dev *sbridge_dev, enum type type)
2266 {
2267 	struct mem_ctl_info *mci;
2268 	struct edac_mc_layer layers[2];
2269 	struct sbridge_pvt *pvt;
2270 	struct pci_dev *pdev = sbridge_dev->pdev[0];
2271 	int rc;
2272 
2273 	/* Check the number of active and not disabled channels */
2274 	rc = check_if_ecc_is_active(sbridge_dev->bus, type);
2275 	if (unlikely(rc < 0))
2276 		return rc;
2277 
2278 	/* allocate a new MC control structure */
2279 	layers[0].type = EDAC_MC_LAYER_CHANNEL;
2280 	layers[0].size = NUM_CHANNELS;
2281 	layers[0].is_virt_csrow = false;
2282 	layers[1].type = EDAC_MC_LAYER_SLOT;
2283 	layers[1].size = MAX_DIMMS;
2284 	layers[1].is_virt_csrow = true;
2285 	mci = edac_mc_alloc(sbridge_dev->mc, ARRAY_SIZE(layers), layers,
2286 			    sizeof(*pvt));
2287 
2288 	if (unlikely(!mci))
2289 		return -ENOMEM;
2290 
2291 	edac_dbg(0, "MC: mci = %p, dev = %p\n",
2292 		 mci, &pdev->dev);
2293 
2294 	pvt = mci->pvt_info;
2295 	memset(pvt, 0, sizeof(*pvt));
2296 
2297 	/* Associate sbridge_dev and mci for future usage */
2298 	pvt->sbridge_dev = sbridge_dev;
2299 	sbridge_dev->mci = mci;
2300 
2301 	mci->mtype_cap = MEM_FLAG_DDR3;
2302 	mci->edac_ctl_cap = EDAC_FLAG_NONE;
2303 	mci->edac_cap = EDAC_FLAG_NONE;
2304 	mci->mod_name = "sbridge_edac.c";
2305 	mci->mod_ver = SBRIDGE_REVISION;
2306 	mci->dev_name = pci_name(pdev);
2307 	mci->ctl_page_to_phys = NULL;
2308 
2309 	/* Set the function pointer to an actual operation function */
2310 	mci->edac_check = sbridge_check_error;
2311 
2312 	pvt->info.type = type;
2313 	switch (type) {
2314 	case IVY_BRIDGE:
2315 		pvt->info.rankcfgr = IB_RANK_CFG_A;
2316 		pvt->info.get_tolm = ibridge_get_tolm;
2317 		pvt->info.get_tohm = ibridge_get_tohm;
2318 		pvt->info.dram_rule = ibridge_dram_rule;
2319 		pvt->info.get_memory_type = get_memory_type;
2320 		pvt->info.get_node_id = get_node_id;
2321 		pvt->info.rir_limit = rir_limit;
2322 		pvt->info.max_sad = ARRAY_SIZE(ibridge_dram_rule);
2323 		pvt->info.interleave_list = ibridge_interleave_list;
2324 		pvt->info.max_interleave = ARRAY_SIZE(ibridge_interleave_list);
2325 		pvt->info.interleave_pkg = ibridge_interleave_pkg;
2326 		mci->ctl_name = kasprintf(GFP_KERNEL, "Ivy Bridge Socket#%d", mci->mc_idx);
2327 
2328 		/* Store pci devices at mci for faster access */
2329 		rc = ibridge_mci_bind_devs(mci, sbridge_dev);
2330 		if (unlikely(rc < 0))
2331 			goto fail0;
2332 		break;
2333 	case SANDY_BRIDGE:
2334 		pvt->info.rankcfgr = SB_RANK_CFG_A;
2335 		pvt->info.get_tolm = sbridge_get_tolm;
2336 		pvt->info.get_tohm = sbridge_get_tohm;
2337 		pvt->info.dram_rule = sbridge_dram_rule;
2338 		pvt->info.get_memory_type = get_memory_type;
2339 		pvt->info.get_node_id = get_node_id;
2340 		pvt->info.rir_limit = rir_limit;
2341 		pvt->info.max_sad = ARRAY_SIZE(sbridge_dram_rule);
2342 		pvt->info.interleave_list = sbridge_interleave_list;
2343 		pvt->info.max_interleave = ARRAY_SIZE(sbridge_interleave_list);
2344 		pvt->info.interleave_pkg = sbridge_interleave_pkg;
2345 		mci->ctl_name = kasprintf(GFP_KERNEL, "Sandy Bridge Socket#%d", mci->mc_idx);
2346 
2347 		/* Store pci devices at mci for faster access */
2348 		rc = sbridge_mci_bind_devs(mci, sbridge_dev);
2349 		if (unlikely(rc < 0))
2350 			goto fail0;
2351 		break;
2352 	case HASWELL:
2353 		/* rankcfgr isn't used */
2354 		pvt->info.get_tolm = haswell_get_tolm;
2355 		pvt->info.get_tohm = haswell_get_tohm;
2356 		pvt->info.dram_rule = ibridge_dram_rule;
2357 		pvt->info.get_memory_type = haswell_get_memory_type;
2358 		pvt->info.get_node_id = haswell_get_node_id;
2359 		pvt->info.rir_limit = haswell_rir_limit;
2360 		pvt->info.max_sad = ARRAY_SIZE(ibridge_dram_rule);
2361 		pvt->info.interleave_list = ibridge_interleave_list;
2362 		pvt->info.max_interleave = ARRAY_SIZE(ibridge_interleave_list);
2363 		pvt->info.interleave_pkg = ibridge_interleave_pkg;
2364 		mci->ctl_name = kasprintf(GFP_KERNEL, "Haswell Socket#%d", mci->mc_idx);
2365 
2366 		/* Store pci devices at mci for faster access */
2367 		rc = haswell_mci_bind_devs(mci, sbridge_dev);
2368 		if (unlikely(rc < 0))
2369 			goto fail0;
2370 		break;
2371 	case BROADWELL:
2372 		/* rankcfgr isn't used */
2373 		pvt->info.get_tolm = haswell_get_tolm;
2374 		pvt->info.get_tohm = haswell_get_tohm;
2375 		pvt->info.dram_rule = ibridge_dram_rule;
2376 		pvt->info.get_memory_type = haswell_get_memory_type;
2377 		pvt->info.get_node_id = haswell_get_node_id;
2378 		pvt->info.rir_limit = haswell_rir_limit;
2379 		pvt->info.max_sad = ARRAY_SIZE(ibridge_dram_rule);
2380 		pvt->info.interleave_list = ibridge_interleave_list;
2381 		pvt->info.max_interleave = ARRAY_SIZE(ibridge_interleave_list);
2382 		pvt->info.interleave_pkg = ibridge_interleave_pkg;
2383 		mci->ctl_name = kasprintf(GFP_KERNEL, "Broadwell Socket#%d", mci->mc_idx);
2384 
2385 		/* Store pci devices at mci for faster access */
2386 		rc = broadwell_mci_bind_devs(mci, sbridge_dev);
2387 		if (unlikely(rc < 0))
2388 			goto fail0;
2389 		break;
2390 	}
2391 
2392 	/* Get dimm basic config and the memory layout */
2393 	get_dimm_config(mci);
2394 	get_memory_layout(mci);
2395 
2396 	/* record ptr to the generic device */
2397 	mci->pdev = &pdev->dev;
2398 
2399 	/* add this new MC control structure to EDAC's list of MCs */
2400 	if (unlikely(edac_mc_add_mc(mci))) {
2401 		edac_dbg(0, "MC: failed edac_mc_add_mc()\n");
2402 		rc = -EINVAL;
2403 		goto fail0;
2404 	}
2405 
2406 	return 0;
2407 
2408 fail0:
2409 	kfree(mci->ctl_name);
2410 	edac_mc_free(mci);
2411 	sbridge_dev->mci = NULL;
2412 	return rc;
2413 }
2414 
2415 /*
2416  *	sbridge_probe	Probe for ONE instance of device to see if it is
2417  *			present.
2418  *	return:
2419  *		0 for FOUND a device
2420  *		< 0 for error code
2421  */
2422 
sbridge_probe(struct pci_dev * pdev,const struct pci_device_id * id)2423 static int sbridge_probe(struct pci_dev *pdev, const struct pci_device_id *id)
2424 {
2425 	int rc = -ENODEV;
2426 	u8 mc, num_mc = 0;
2427 	struct sbridge_dev *sbridge_dev;
2428 	enum type type = SANDY_BRIDGE;
2429 
2430 	/* get the pci devices we want to reserve for our use */
2431 	mutex_lock(&sbridge_edac_lock);
2432 
2433 	/*
2434 	 * All memory controllers are allocated at the first pass.
2435 	 */
2436 	if (unlikely(probed >= 1)) {
2437 		mutex_unlock(&sbridge_edac_lock);
2438 		return -ENODEV;
2439 	}
2440 	probed++;
2441 
2442 	switch (pdev->device) {
2443 	case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TA:
2444 		rc = sbridge_get_all_devices(&num_mc, pci_dev_descr_ibridge_table);
2445 		type = IVY_BRIDGE;
2446 		break;
2447 	case PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_HA0:
2448 		rc = sbridge_get_all_devices(&num_mc, pci_dev_descr_sbridge_table);
2449 		type = SANDY_BRIDGE;
2450 		break;
2451 	case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0:
2452 		rc = sbridge_get_all_devices(&num_mc, pci_dev_descr_haswell_table);
2453 		type = HASWELL;
2454 		break;
2455 	case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0:
2456 		rc = sbridge_get_all_devices(&num_mc, pci_dev_descr_broadwell_table);
2457 		type = BROADWELL;
2458 		break;
2459 	}
2460 	if (unlikely(rc < 0)) {
2461 		edac_dbg(0, "couldn't get all devices for 0x%x\n", pdev->device);
2462 		goto fail0;
2463 	}
2464 
2465 	mc = 0;
2466 
2467 	list_for_each_entry(sbridge_dev, &sbridge_edac_list, list) {
2468 		edac_dbg(0, "Registering MC#%d (%d of %d)\n",
2469 			 mc, mc + 1, num_mc);
2470 
2471 		sbridge_dev->mc = mc++;
2472 		rc = sbridge_register_mci(sbridge_dev, type);
2473 		if (unlikely(rc < 0))
2474 			goto fail1;
2475 	}
2476 
2477 	sbridge_printk(KERN_INFO, "%s\n", SBRIDGE_REVISION);
2478 
2479 	mutex_unlock(&sbridge_edac_lock);
2480 	return 0;
2481 
2482 fail1:
2483 	list_for_each_entry(sbridge_dev, &sbridge_edac_list, list)
2484 		sbridge_unregister_mci(sbridge_dev);
2485 
2486 	sbridge_put_all_devices();
2487 fail0:
2488 	mutex_unlock(&sbridge_edac_lock);
2489 	return rc;
2490 }
2491 
2492 /*
2493  *	sbridge_remove	destructor for one instance of device
2494  *
2495  */
sbridge_remove(struct pci_dev * pdev)2496 static void sbridge_remove(struct pci_dev *pdev)
2497 {
2498 	struct sbridge_dev *sbridge_dev;
2499 
2500 	edac_dbg(0, "\n");
2501 
2502 	/*
2503 	 * we have a trouble here: pdev value for removal will be wrong, since
2504 	 * it will point to the X58 register used to detect that the machine
2505 	 * is a Nehalem or upper design. However, due to the way several PCI
2506 	 * devices are grouped together to provide MC functionality, we need
2507 	 * to use a different method for releasing the devices
2508 	 */
2509 
2510 	mutex_lock(&sbridge_edac_lock);
2511 
2512 	if (unlikely(!probed)) {
2513 		mutex_unlock(&sbridge_edac_lock);
2514 		return;
2515 	}
2516 
2517 	list_for_each_entry(sbridge_dev, &sbridge_edac_list, list)
2518 		sbridge_unregister_mci(sbridge_dev);
2519 
2520 	/* Release PCI resources */
2521 	sbridge_put_all_devices();
2522 
2523 	probed--;
2524 
2525 	mutex_unlock(&sbridge_edac_lock);
2526 }
2527 
2528 MODULE_DEVICE_TABLE(pci, sbridge_pci_tbl);
2529 
2530 /*
2531  *	sbridge_driver	pci_driver structure for this module
2532  *
2533  */
2534 static struct pci_driver sbridge_driver = {
2535 	.name     = "sbridge_edac",
2536 	.probe    = sbridge_probe,
2537 	.remove   = sbridge_remove,
2538 	.id_table = sbridge_pci_tbl,
2539 };
2540 
2541 /*
2542  *	sbridge_init		Module entry function
2543  *			Try to initialize this module for its devices
2544  */
sbridge_init(void)2545 static int __init sbridge_init(void)
2546 {
2547 	int pci_rc;
2548 
2549 	edac_dbg(2, "\n");
2550 
2551 	/* Ensure that the OPSTATE is set correctly for POLL or NMI */
2552 	opstate_init();
2553 
2554 	pci_rc = pci_register_driver(&sbridge_driver);
2555 	if (pci_rc >= 0) {
2556 		mce_register_decode_chain(&sbridge_mce_dec);
2557 		if (get_edac_report_status() == EDAC_REPORTING_DISABLED)
2558 			sbridge_printk(KERN_WARNING, "Loading driver, error reporting disabled.\n");
2559 		return 0;
2560 	}
2561 
2562 	sbridge_printk(KERN_ERR, "Failed to register device with error %d.\n",
2563 		      pci_rc);
2564 
2565 	return pci_rc;
2566 }
2567 
2568 /*
2569  *	sbridge_exit()	Module exit function
2570  *			Unregister the driver
2571  */
sbridge_exit(void)2572 static void __exit sbridge_exit(void)
2573 {
2574 	edac_dbg(2, "\n");
2575 	pci_unregister_driver(&sbridge_driver);
2576 	mce_unregister_decode_chain(&sbridge_mce_dec);
2577 }
2578 
2579 module_init(sbridge_init);
2580 module_exit(sbridge_exit);
2581 
2582 module_param(edac_op_state, int, 0444);
2583 MODULE_PARM_DESC(edac_op_state, "EDAC Error Reporting state: 0=Poll,1=NMI");
2584 
2585 MODULE_LICENSE("GPL");
2586 MODULE_AUTHOR("Mauro Carvalho Chehab");
2587 MODULE_AUTHOR("Red Hat Inc. (http://www.redhat.com)");
2588 MODULE_DESCRIPTION("MC Driver for Intel Sandy Bridge and Ivy Bridge memory controllers - "
2589 		   SBRIDGE_REVISION);
2590