xref: /openbmc/linux/drivers/edac/i5100_edac.c (revision 54a8b680)
1 /*
2  * Intel 5100 Memory Controllers kernel module
3  *
4  * This file may be distributed under the terms of the
5  * GNU General Public License.
6  *
7  * This module is based on the following document:
8  *
9  * Intel 5100X Chipset Memory Controller Hub (MCH) - Datasheet
10  *      http://download.intel.com/design/chipsets/datashts/318378.pdf
11  *
12  * The intel 5100 has two independent channels. EDAC core currently
13  * can not reflect this configuration so instead the chip-select
14  * rows for each respective channel are laid out one after another,
15  * the first half belonging to channel 0, the second half belonging
16  * to channel 1.
17  *
18  * This driver is for DDR2 DIMMs, and it uses chip select to select among the
19  * several ranks. However, instead of showing memories as ranks, it outputs
20  * them as DIMM's. An internal table creates the association between ranks
21  * and DIMM's.
22  */
23 #include <linux/module.h>
24 #include <linux/init.h>
25 #include <linux/pci.h>
26 #include <linux/pci_ids.h>
27 #include <linux/edac.h>
28 #include <linux/delay.h>
29 #include <linux/mmzone.h>
30 #include <linux/debugfs.h>
31 
32 #include "edac_module.h"
33 
34 /* register addresses */
35 
36 /* device 16, func 1 */
37 #define I5100_MC		0x40	/* Memory Control Register */
38 #define 	I5100_MC_SCRBEN_MASK	(1 << 7)
39 #define 	I5100_MC_SCRBDONE_MASK	(1 << 4)
40 #define I5100_MS		0x44	/* Memory Status Register */
41 #define I5100_SPDDATA		0x48	/* Serial Presence Detect Status Reg */
42 #define I5100_SPDCMD		0x4c	/* Serial Presence Detect Command Reg */
43 #define I5100_TOLM		0x6c	/* Top of Low Memory */
44 #define I5100_MIR0		0x80	/* Memory Interleave Range 0 */
45 #define I5100_MIR1		0x84	/* Memory Interleave Range 1 */
46 #define I5100_AMIR_0		0x8c	/* Adjusted Memory Interleave Range 0 */
47 #define I5100_AMIR_1		0x90	/* Adjusted Memory Interleave Range 1 */
48 #define I5100_FERR_NF_MEM	0xa0	/* MC First Non Fatal Errors */
49 #define		I5100_FERR_NF_MEM_M16ERR_MASK	(1 << 16)
50 #define		I5100_FERR_NF_MEM_M15ERR_MASK	(1 << 15)
51 #define		I5100_FERR_NF_MEM_M14ERR_MASK	(1 << 14)
52 #define		I5100_FERR_NF_MEM_M12ERR_MASK	(1 << 12)
53 #define		I5100_FERR_NF_MEM_M11ERR_MASK	(1 << 11)
54 #define		I5100_FERR_NF_MEM_M10ERR_MASK	(1 << 10)
55 #define		I5100_FERR_NF_MEM_M6ERR_MASK	(1 << 6)
56 #define		I5100_FERR_NF_MEM_M5ERR_MASK	(1 << 5)
57 #define		I5100_FERR_NF_MEM_M4ERR_MASK	(1 << 4)
58 #define		I5100_FERR_NF_MEM_M1ERR_MASK	(1 << 1)
59 #define		I5100_FERR_NF_MEM_ANY_MASK	\
60 			(I5100_FERR_NF_MEM_M16ERR_MASK | \
61 			I5100_FERR_NF_MEM_M15ERR_MASK | \
62 			I5100_FERR_NF_MEM_M14ERR_MASK | \
63 			I5100_FERR_NF_MEM_M12ERR_MASK | \
64 			I5100_FERR_NF_MEM_M11ERR_MASK | \
65 			I5100_FERR_NF_MEM_M10ERR_MASK | \
66 			I5100_FERR_NF_MEM_M6ERR_MASK | \
67 			I5100_FERR_NF_MEM_M5ERR_MASK | \
68 			I5100_FERR_NF_MEM_M4ERR_MASK | \
69 			I5100_FERR_NF_MEM_M1ERR_MASK)
70 #define	I5100_NERR_NF_MEM	0xa4	/* MC Next Non-Fatal Errors */
71 #define I5100_EMASK_MEM		0xa8	/* MC Error Mask Register */
72 #define I5100_MEM0EINJMSK0	0x200	/* Injection Mask0 Register Channel 0 */
73 #define I5100_MEM1EINJMSK0	0x208	/* Injection Mask0 Register Channel 1 */
74 #define		I5100_MEMXEINJMSK0_EINJEN	(1 << 27)
75 #define I5100_MEM0EINJMSK1	0x204	/* Injection Mask1 Register Channel 0 */
76 #define I5100_MEM1EINJMSK1	0x206	/* Injection Mask1 Register Channel 1 */
77 
78 /* Device 19, Function 0 */
79 #define I5100_DINJ0 0x9a
80 
81 /* device 21 and 22, func 0 */
82 #define I5100_MTR_0	0x154	/* Memory Technology Registers 0-3 */
83 #define I5100_DMIR	0x15c	/* DIMM Interleave Range */
84 #define	I5100_VALIDLOG	0x18c	/* Valid Log Markers */
85 #define	I5100_NRECMEMA	0x190	/* Non-Recoverable Memory Error Log Reg A */
86 #define	I5100_NRECMEMB	0x194	/* Non-Recoverable Memory Error Log Reg B */
87 #define	I5100_REDMEMA	0x198	/* Recoverable Memory Data Error Log Reg A */
88 #define	I5100_REDMEMB	0x19c	/* Recoverable Memory Data Error Log Reg B */
89 #define	I5100_RECMEMA	0x1a0	/* Recoverable Memory Error Log Reg A */
90 #define	I5100_RECMEMB	0x1a4	/* Recoverable Memory Error Log Reg B */
91 #define I5100_MTR_4	0x1b0	/* Memory Technology Registers 4,5 */
92 
93 /* bit field accessors */
94 
95 static inline u32 i5100_mc_scrben(u32 mc)
96 {
97 	return mc >> 7 & 1;
98 }
99 
100 static inline u32 i5100_mc_errdeten(u32 mc)
101 {
102 	return mc >> 5 & 1;
103 }
104 
105 static inline u32 i5100_mc_scrbdone(u32 mc)
106 {
107 	return mc >> 4 & 1;
108 }
109 
110 static inline u16 i5100_spddata_rdo(u16 a)
111 {
112 	return a >> 15 & 1;
113 }
114 
115 static inline u16 i5100_spddata_sbe(u16 a)
116 {
117 	return a >> 13 & 1;
118 }
119 
120 static inline u16 i5100_spddata_busy(u16 a)
121 {
122 	return a >> 12 & 1;
123 }
124 
125 static inline u16 i5100_spddata_data(u16 a)
126 {
127 	return a & ((1 << 8) - 1);
128 }
129 
130 static inline u32 i5100_spdcmd_create(u32 dti, u32 ckovrd, u32 sa, u32 ba,
131 				      u32 data, u32 cmd)
132 {
133 	return	((dti & ((1 << 4) - 1))  << 28) |
134 		((ckovrd & 1)            << 27) |
135 		((sa & ((1 << 3) - 1))   << 24) |
136 		((ba & ((1 << 8) - 1))   << 16) |
137 		((data & ((1 << 8) - 1)) <<  8) |
138 		(cmd & 1);
139 }
140 
141 static inline u16 i5100_tolm_tolm(u16 a)
142 {
143 	return a >> 12 & ((1 << 4) - 1);
144 }
145 
146 static inline u16 i5100_mir_limit(u16 a)
147 {
148 	return a >> 4 & ((1 << 12) - 1);
149 }
150 
151 static inline u16 i5100_mir_way1(u16 a)
152 {
153 	return a >> 1 & 1;
154 }
155 
156 static inline u16 i5100_mir_way0(u16 a)
157 {
158 	return a & 1;
159 }
160 
161 static inline u32 i5100_ferr_nf_mem_chan_indx(u32 a)
162 {
163 	return a >> 28 & 1;
164 }
165 
166 static inline u32 i5100_ferr_nf_mem_any(u32 a)
167 {
168 	return a & I5100_FERR_NF_MEM_ANY_MASK;
169 }
170 
171 static inline u32 i5100_nerr_nf_mem_any(u32 a)
172 {
173 	return i5100_ferr_nf_mem_any(a);
174 }
175 
176 static inline u32 i5100_dmir_limit(u32 a)
177 {
178 	return a >> 16 & ((1 << 11) - 1);
179 }
180 
181 static inline u32 i5100_dmir_rank(u32 a, u32 i)
182 {
183 	return a >> (4 * i) & ((1 << 2) - 1);
184 }
185 
186 static inline u16 i5100_mtr_present(u16 a)
187 {
188 	return a >> 10 & 1;
189 }
190 
191 static inline u16 i5100_mtr_ethrottle(u16 a)
192 {
193 	return a >> 9 & 1;
194 }
195 
196 static inline u16 i5100_mtr_width(u16 a)
197 {
198 	return a >> 8 & 1;
199 }
200 
201 static inline u16 i5100_mtr_numbank(u16 a)
202 {
203 	return a >> 6 & 1;
204 }
205 
206 static inline u16 i5100_mtr_numrow(u16 a)
207 {
208 	return a >> 2 & ((1 << 2) - 1);
209 }
210 
211 static inline u16 i5100_mtr_numcol(u16 a)
212 {
213 	return a & ((1 << 2) - 1);
214 }
215 
216 
217 static inline u32 i5100_validlog_redmemvalid(u32 a)
218 {
219 	return a >> 2 & 1;
220 }
221 
222 static inline u32 i5100_validlog_recmemvalid(u32 a)
223 {
224 	return a >> 1 & 1;
225 }
226 
227 static inline u32 i5100_validlog_nrecmemvalid(u32 a)
228 {
229 	return a & 1;
230 }
231 
232 static inline u32 i5100_nrecmema_merr(u32 a)
233 {
234 	return a >> 15 & ((1 << 5) - 1);
235 }
236 
237 static inline u32 i5100_nrecmema_bank(u32 a)
238 {
239 	return a >> 12 & ((1 << 3) - 1);
240 }
241 
242 static inline u32 i5100_nrecmema_rank(u32 a)
243 {
244 	return a >>  8 & ((1 << 3) - 1);
245 }
246 
247 static inline u32 i5100_nrecmemb_cas(u32 a)
248 {
249 	return a >> 16 & ((1 << 13) - 1);
250 }
251 
252 static inline u32 i5100_nrecmemb_ras(u32 a)
253 {
254 	return a & ((1 << 16) - 1);
255 }
256 
257 static inline u32 i5100_recmema_merr(u32 a)
258 {
259 	return i5100_nrecmema_merr(a);
260 }
261 
262 static inline u32 i5100_recmema_bank(u32 a)
263 {
264 	return i5100_nrecmema_bank(a);
265 }
266 
267 static inline u32 i5100_recmema_rank(u32 a)
268 {
269 	return i5100_nrecmema_rank(a);
270 }
271 
272 static inline u32 i5100_recmemb_cas(u32 a)
273 {
274 	return i5100_nrecmemb_cas(a);
275 }
276 
277 static inline u32 i5100_recmemb_ras(u32 a)
278 {
279 	return i5100_nrecmemb_ras(a);
280 }
281 
282 /* some generic limits */
283 #define I5100_MAX_RANKS_PER_CHAN	6
284 #define I5100_CHANNELS			    2
285 #define I5100_MAX_RANKS_PER_DIMM	4
286 #define I5100_DIMM_ADDR_LINES		(6 - 3)	/* 64 bits / 8 bits per byte */
287 #define I5100_MAX_DIMM_SLOTS_PER_CHAN	4
288 #define I5100_MAX_RANK_INTERLEAVE	4
289 #define I5100_MAX_DMIRS			5
290 #define I5100_SCRUB_REFRESH_RATE	(5 * 60 * HZ)
291 
292 struct i5100_priv {
293 	/* ranks on each dimm -- 0 maps to not present -- obtained via SPD */
294 	int dimm_numrank[I5100_CHANNELS][I5100_MAX_DIMM_SLOTS_PER_CHAN];
295 
296 	/*
297 	 * mainboard chip select map -- maps i5100 chip selects to
298 	 * DIMM slot chip selects.  In the case of only 4 ranks per
299 	 * channel, the mapping is fairly obvious but not unique.
300 	 * we map -1 -> NC and assume both channels use the same
301 	 * map...
302 	 *
303 	 */
304 	int dimm_csmap[I5100_MAX_DIMM_SLOTS_PER_CHAN][I5100_MAX_RANKS_PER_DIMM];
305 
306 	/* memory interleave range */
307 	struct {
308 		u64	 limit;
309 		unsigned way[2];
310 	} mir[I5100_CHANNELS];
311 
312 	/* adjusted memory interleave range register */
313 	unsigned amir[I5100_CHANNELS];
314 
315 	/* dimm interleave range */
316 	struct {
317 		unsigned rank[I5100_MAX_RANK_INTERLEAVE];
318 		u64	 limit;
319 	} dmir[I5100_CHANNELS][I5100_MAX_DMIRS];
320 
321 	/* memory technology registers... */
322 	struct {
323 		unsigned present;	/* 0 or 1 */
324 		unsigned ethrottle;	/* 0 or 1 */
325 		unsigned width;		/* 4 or 8 bits  */
326 		unsigned numbank;	/* 2 or 3 lines */
327 		unsigned numrow;	/* 13 .. 16 lines */
328 		unsigned numcol;	/* 11 .. 12 lines */
329 	} mtr[I5100_CHANNELS][I5100_MAX_RANKS_PER_CHAN];
330 
331 	u64 tolm;		/* top of low memory in bytes */
332 	unsigned ranksperchan;	/* number of ranks per channel */
333 
334 	struct pci_dev *mc;	/* device 16 func 1 */
335 	struct pci_dev *einj;	/* device 19 func 0 */
336 	struct pci_dev *ch0mm;	/* device 21 func 0 */
337 	struct pci_dev *ch1mm;	/* device 22 func 0 */
338 
339 	struct delayed_work i5100_scrubbing;
340 	int scrub_enable;
341 
342 	/* Error injection */
343 	u8 inject_channel;
344 	u8 inject_hlinesel;
345 	u8 inject_deviceptr1;
346 	u8 inject_deviceptr2;
347 	u16 inject_eccmask1;
348 	u16 inject_eccmask2;
349 
350 	struct dentry *debugfs;
351 };
352 
353 static struct dentry *i5100_debugfs;
354 
355 /* map a rank/chan to a slot number on the mainboard */
356 static int i5100_rank_to_slot(const struct mem_ctl_info *mci,
357 			      int chan, int rank)
358 {
359 	const struct i5100_priv *priv = mci->pvt_info;
360 	int i;
361 
362 	for (i = 0; i < I5100_MAX_DIMM_SLOTS_PER_CHAN; i++) {
363 		int j;
364 		const int numrank = priv->dimm_numrank[chan][i];
365 
366 		for (j = 0; j < numrank; j++)
367 			if (priv->dimm_csmap[i][j] == rank)
368 				return i * 2 + chan;
369 	}
370 
371 	return -1;
372 }
373 
374 static const char *i5100_err_msg(unsigned err)
375 {
376 	static const char *merrs[] = {
377 		"unknown", /* 0 */
378 		"uncorrectable data ECC on replay", /* 1 */
379 		"unknown", /* 2 */
380 		"unknown", /* 3 */
381 		"aliased uncorrectable demand data ECC", /* 4 */
382 		"aliased uncorrectable spare-copy data ECC", /* 5 */
383 		"aliased uncorrectable patrol data ECC", /* 6 */
384 		"unknown", /* 7 */
385 		"unknown", /* 8 */
386 		"unknown", /* 9 */
387 		"non-aliased uncorrectable demand data ECC", /* 10 */
388 		"non-aliased uncorrectable spare-copy data ECC", /* 11 */
389 		"non-aliased uncorrectable patrol data ECC", /* 12 */
390 		"unknown", /* 13 */
391 		"correctable demand data ECC", /* 14 */
392 		"correctable spare-copy data ECC", /* 15 */
393 		"correctable patrol data ECC", /* 16 */
394 		"unknown", /* 17 */
395 		"SPD protocol error", /* 18 */
396 		"unknown", /* 19 */
397 		"spare copy initiated", /* 20 */
398 		"spare copy completed", /* 21 */
399 	};
400 	unsigned i;
401 
402 	for (i = 0; i < ARRAY_SIZE(merrs); i++)
403 		if (1 << i & err)
404 			return merrs[i];
405 
406 	return "none";
407 }
408 
409 /* convert csrow index into a rank (per channel -- 0..5) */
410 static unsigned int i5100_csrow_to_rank(const struct mem_ctl_info *mci,
411 					unsigned int csrow)
412 {
413 	const struct i5100_priv *priv = mci->pvt_info;
414 
415 	return csrow % priv->ranksperchan;
416 }
417 
418 /* convert csrow index into a channel (0..1) */
419 static unsigned int i5100_csrow_to_chan(const struct mem_ctl_info *mci,
420 					unsigned int csrow)
421 {
422 	const struct i5100_priv *priv = mci->pvt_info;
423 
424 	return csrow / priv->ranksperchan;
425 }
426 
427 static void i5100_handle_ce(struct mem_ctl_info *mci,
428 			    int chan,
429 			    unsigned bank,
430 			    unsigned rank,
431 			    unsigned long syndrome,
432 			    unsigned cas,
433 			    unsigned ras,
434 			    const char *msg)
435 {
436 	char detail[80];
437 
438 	/* Form out message */
439 	snprintf(detail, sizeof(detail),
440 		 "bank %u, cas %u, ras %u\n",
441 		 bank, cas, ras);
442 
443 	edac_mc_handle_error(HW_EVENT_ERR_CORRECTED, mci, 1,
444 			     0, 0, syndrome,
445 			     chan, rank, -1,
446 			     msg, detail);
447 }
448 
449 static void i5100_handle_ue(struct mem_ctl_info *mci,
450 			    int chan,
451 			    unsigned bank,
452 			    unsigned rank,
453 			    unsigned long syndrome,
454 			    unsigned cas,
455 			    unsigned ras,
456 			    const char *msg)
457 {
458 	char detail[80];
459 
460 	/* Form out message */
461 	snprintf(detail, sizeof(detail),
462 		 "bank %u, cas %u, ras %u\n",
463 		 bank, cas, ras);
464 
465 	edac_mc_handle_error(HW_EVENT_ERR_UNCORRECTED, mci, 1,
466 			     0, 0, syndrome,
467 			     chan, rank, -1,
468 			     msg, detail);
469 }
470 
471 static void i5100_read_log(struct mem_ctl_info *mci, int chan,
472 			   u32 ferr, u32 nerr)
473 {
474 	struct i5100_priv *priv = mci->pvt_info;
475 	struct pci_dev *pdev = (chan) ? priv->ch1mm : priv->ch0mm;
476 	u32 dw;
477 	u32 dw2;
478 	unsigned syndrome = 0;
479 	unsigned merr;
480 	unsigned bank;
481 	unsigned rank;
482 	unsigned cas;
483 	unsigned ras;
484 
485 	pci_read_config_dword(pdev, I5100_VALIDLOG, &dw);
486 
487 	if (i5100_validlog_redmemvalid(dw)) {
488 		pci_read_config_dword(pdev, I5100_REDMEMA, &dw2);
489 		syndrome = dw2;
490 		pci_read_config_dword(pdev, I5100_REDMEMB, &dw2);
491 	}
492 
493 	if (i5100_validlog_recmemvalid(dw)) {
494 		const char *msg;
495 
496 		pci_read_config_dword(pdev, I5100_RECMEMA, &dw2);
497 		merr = i5100_recmema_merr(dw2);
498 		bank = i5100_recmema_bank(dw2);
499 		rank = i5100_recmema_rank(dw2);
500 
501 		pci_read_config_dword(pdev, I5100_RECMEMB, &dw2);
502 		cas = i5100_recmemb_cas(dw2);
503 		ras = i5100_recmemb_ras(dw2);
504 
505 		/* FIXME:  not really sure if this is what merr is...
506 		 */
507 		if (!merr)
508 			msg = i5100_err_msg(ferr);
509 		else
510 			msg = i5100_err_msg(nerr);
511 
512 		i5100_handle_ce(mci, chan, bank, rank, syndrome, cas, ras, msg);
513 	}
514 
515 	if (i5100_validlog_nrecmemvalid(dw)) {
516 		const char *msg;
517 
518 		pci_read_config_dword(pdev, I5100_NRECMEMA, &dw2);
519 		merr = i5100_nrecmema_merr(dw2);
520 		bank = i5100_nrecmema_bank(dw2);
521 		rank = i5100_nrecmema_rank(dw2);
522 
523 		pci_read_config_dword(pdev, I5100_NRECMEMB, &dw2);
524 		cas = i5100_nrecmemb_cas(dw2);
525 		ras = i5100_nrecmemb_ras(dw2);
526 
527 		/* FIXME:  not really sure if this is what merr is...
528 		 */
529 		if (!merr)
530 			msg = i5100_err_msg(ferr);
531 		else
532 			msg = i5100_err_msg(nerr);
533 
534 		i5100_handle_ue(mci, chan, bank, rank, syndrome, cas, ras, msg);
535 	}
536 
537 	pci_write_config_dword(pdev, I5100_VALIDLOG, dw);
538 }
539 
540 static void i5100_check_error(struct mem_ctl_info *mci)
541 {
542 	struct i5100_priv *priv = mci->pvt_info;
543 	u32 dw, dw2;
544 
545 	pci_read_config_dword(priv->mc, I5100_FERR_NF_MEM, &dw);
546 	if (i5100_ferr_nf_mem_any(dw)) {
547 
548 		pci_read_config_dword(priv->mc, I5100_NERR_NF_MEM, &dw2);
549 
550 		i5100_read_log(mci, i5100_ferr_nf_mem_chan_indx(dw),
551 			       i5100_ferr_nf_mem_any(dw),
552 			       i5100_nerr_nf_mem_any(dw2));
553 
554 		pci_write_config_dword(priv->mc, I5100_NERR_NF_MEM, dw2);
555 	}
556 	pci_write_config_dword(priv->mc, I5100_FERR_NF_MEM, dw);
557 }
558 
559 /* The i5100 chipset will scrub the entire memory once, then
560  * set a done bit. Continuous scrubbing is achieved by enqueing
561  * delayed work to a workqueue, checking every few minutes if
562  * the scrubbing has completed and if so reinitiating it.
563  */
564 
565 static void i5100_refresh_scrubbing(struct work_struct *work)
566 {
567 	struct delayed_work *i5100_scrubbing = to_delayed_work(work);
568 	struct i5100_priv *priv = container_of(i5100_scrubbing,
569 					       struct i5100_priv,
570 					       i5100_scrubbing);
571 	u32 dw;
572 
573 	pci_read_config_dword(priv->mc, I5100_MC, &dw);
574 
575 	if (priv->scrub_enable) {
576 
577 		pci_read_config_dword(priv->mc, I5100_MC, &dw);
578 
579 		if (i5100_mc_scrbdone(dw)) {
580 			dw |= I5100_MC_SCRBEN_MASK;
581 			pci_write_config_dword(priv->mc, I5100_MC, dw);
582 			pci_read_config_dword(priv->mc, I5100_MC, &dw);
583 		}
584 
585 		schedule_delayed_work(&(priv->i5100_scrubbing),
586 				      I5100_SCRUB_REFRESH_RATE);
587 	}
588 }
589 /*
590  * The bandwidth is based on experimentation, feel free to refine it.
591  */
592 static int i5100_set_scrub_rate(struct mem_ctl_info *mci, u32 bandwidth)
593 {
594 	struct i5100_priv *priv = mci->pvt_info;
595 	u32 dw;
596 
597 	pci_read_config_dword(priv->mc, I5100_MC, &dw);
598 	if (bandwidth) {
599 		priv->scrub_enable = 1;
600 		dw |= I5100_MC_SCRBEN_MASK;
601 		schedule_delayed_work(&(priv->i5100_scrubbing),
602 				      I5100_SCRUB_REFRESH_RATE);
603 	} else {
604 		priv->scrub_enable = 0;
605 		dw &= ~I5100_MC_SCRBEN_MASK;
606 		cancel_delayed_work(&(priv->i5100_scrubbing));
607 	}
608 	pci_write_config_dword(priv->mc, I5100_MC, dw);
609 
610 	pci_read_config_dword(priv->mc, I5100_MC, &dw);
611 
612 	bandwidth = 5900000 * i5100_mc_scrben(dw);
613 
614 	return bandwidth;
615 }
616 
617 static int i5100_get_scrub_rate(struct mem_ctl_info *mci)
618 {
619 	struct i5100_priv *priv = mci->pvt_info;
620 	u32 dw;
621 
622 	pci_read_config_dword(priv->mc, I5100_MC, &dw);
623 
624 	return 5900000 * i5100_mc_scrben(dw);
625 }
626 
627 static struct pci_dev *pci_get_device_func(unsigned vendor,
628 					   unsigned device,
629 					   unsigned func)
630 {
631 	struct pci_dev *ret = NULL;
632 
633 	while (1) {
634 		ret = pci_get_device(vendor, device, ret);
635 
636 		if (!ret)
637 			break;
638 
639 		if (PCI_FUNC(ret->devfn) == func)
640 			break;
641 	}
642 
643 	return ret;
644 }
645 
646 static unsigned long i5100_npages(struct mem_ctl_info *mci, unsigned int csrow)
647 {
648 	struct i5100_priv *priv = mci->pvt_info;
649 	const unsigned int chan_rank = i5100_csrow_to_rank(mci, csrow);
650 	const unsigned int chan = i5100_csrow_to_chan(mci, csrow);
651 	unsigned addr_lines;
652 
653 	/* dimm present? */
654 	if (!priv->mtr[chan][chan_rank].present)
655 		return 0ULL;
656 
657 	addr_lines =
658 		I5100_DIMM_ADDR_LINES +
659 		priv->mtr[chan][chan_rank].numcol +
660 		priv->mtr[chan][chan_rank].numrow +
661 		priv->mtr[chan][chan_rank].numbank;
662 
663 	return (unsigned long)
664 		((unsigned long long) (1ULL << addr_lines) / PAGE_SIZE);
665 }
666 
667 static void i5100_init_mtr(struct mem_ctl_info *mci)
668 {
669 	struct i5100_priv *priv = mci->pvt_info;
670 	struct pci_dev *mms[2] = { priv->ch0mm, priv->ch1mm };
671 	int i;
672 
673 	for (i = 0; i < I5100_CHANNELS; i++) {
674 		int j;
675 		struct pci_dev *pdev = mms[i];
676 
677 		for (j = 0; j < I5100_MAX_RANKS_PER_CHAN; j++) {
678 			const unsigned addr =
679 				(j < 4) ? I5100_MTR_0 + j * 2 :
680 					  I5100_MTR_4 + (j - 4) * 2;
681 			u16 w;
682 
683 			pci_read_config_word(pdev, addr, &w);
684 
685 			priv->mtr[i][j].present = i5100_mtr_present(w);
686 			priv->mtr[i][j].ethrottle = i5100_mtr_ethrottle(w);
687 			priv->mtr[i][j].width = 4 + 4 * i5100_mtr_width(w);
688 			priv->mtr[i][j].numbank = 2 + i5100_mtr_numbank(w);
689 			priv->mtr[i][j].numrow = 13 + i5100_mtr_numrow(w);
690 			priv->mtr[i][j].numcol = 10 + i5100_mtr_numcol(w);
691 		}
692 	}
693 }
694 
695 /*
696  * FIXME: make this into a real i2c adapter (so that dimm-decode
697  * will work)?
698  */
699 static int i5100_read_spd_byte(const struct mem_ctl_info *mci,
700 			       u8 ch, u8 slot, u8 addr, u8 *byte)
701 {
702 	struct i5100_priv *priv = mci->pvt_info;
703 	u16 w;
704 
705 	pci_read_config_word(priv->mc, I5100_SPDDATA, &w);
706 	if (i5100_spddata_busy(w))
707 		return -1;
708 
709 	pci_write_config_dword(priv->mc, I5100_SPDCMD,
710 			       i5100_spdcmd_create(0xa, 1, ch * 4 + slot, addr,
711 						   0, 0));
712 
713 	/* wait up to 100ms */
714 	udelay(100);
715 	while (1) {
716 		pci_read_config_word(priv->mc, I5100_SPDDATA, &w);
717 		if (!i5100_spddata_busy(w))
718 			break;
719 		udelay(100);
720 	}
721 
722 	if (!i5100_spddata_rdo(w) || i5100_spddata_sbe(w))
723 		return -1;
724 
725 	*byte = i5100_spddata_data(w);
726 
727 	return 0;
728 }
729 
730 /*
731  * fill dimm chip select map
732  *
733  * FIXME:
734  *   o not the only way to may chip selects to dimm slots
735  *   o investigate if there is some way to obtain this map from the bios
736  */
737 static void i5100_init_dimm_csmap(struct mem_ctl_info *mci)
738 {
739 	struct i5100_priv *priv = mci->pvt_info;
740 	int i;
741 
742 	for (i = 0; i < I5100_MAX_DIMM_SLOTS_PER_CHAN; i++) {
743 		int j;
744 
745 		for (j = 0; j < I5100_MAX_RANKS_PER_DIMM; j++)
746 			priv->dimm_csmap[i][j] = -1; /* default NC */
747 	}
748 
749 	/* only 2 chip selects per slot... */
750 	if (priv->ranksperchan == 4) {
751 		priv->dimm_csmap[0][0] = 0;
752 		priv->dimm_csmap[0][1] = 3;
753 		priv->dimm_csmap[1][0] = 1;
754 		priv->dimm_csmap[1][1] = 2;
755 		priv->dimm_csmap[2][0] = 2;
756 		priv->dimm_csmap[3][0] = 3;
757 	} else {
758 		priv->dimm_csmap[0][0] = 0;
759 		priv->dimm_csmap[0][1] = 1;
760 		priv->dimm_csmap[1][0] = 2;
761 		priv->dimm_csmap[1][1] = 3;
762 		priv->dimm_csmap[2][0] = 4;
763 		priv->dimm_csmap[2][1] = 5;
764 	}
765 }
766 
767 static void i5100_init_dimm_layout(struct pci_dev *pdev,
768 				   struct mem_ctl_info *mci)
769 {
770 	struct i5100_priv *priv = mci->pvt_info;
771 	int i;
772 
773 	for (i = 0; i < I5100_CHANNELS; i++) {
774 		int j;
775 
776 		for (j = 0; j < I5100_MAX_DIMM_SLOTS_PER_CHAN; j++) {
777 			u8 rank;
778 
779 			if (i5100_read_spd_byte(mci, i, j, 5, &rank) < 0)
780 				priv->dimm_numrank[i][j] = 0;
781 			else
782 				priv->dimm_numrank[i][j] = (rank & 3) + 1;
783 		}
784 	}
785 
786 	i5100_init_dimm_csmap(mci);
787 }
788 
789 static void i5100_init_interleaving(struct pci_dev *pdev,
790 				    struct mem_ctl_info *mci)
791 {
792 	u16 w;
793 	u32 dw;
794 	struct i5100_priv *priv = mci->pvt_info;
795 	struct pci_dev *mms[2] = { priv->ch0mm, priv->ch1mm };
796 	int i;
797 
798 	pci_read_config_word(pdev, I5100_TOLM, &w);
799 	priv->tolm = (u64) i5100_tolm_tolm(w) * 256 * 1024 * 1024;
800 
801 	pci_read_config_word(pdev, I5100_MIR0, &w);
802 	priv->mir[0].limit = (u64) i5100_mir_limit(w) << 28;
803 	priv->mir[0].way[1] = i5100_mir_way1(w);
804 	priv->mir[0].way[0] = i5100_mir_way0(w);
805 
806 	pci_read_config_word(pdev, I5100_MIR1, &w);
807 	priv->mir[1].limit = (u64) i5100_mir_limit(w) << 28;
808 	priv->mir[1].way[1] = i5100_mir_way1(w);
809 	priv->mir[1].way[0] = i5100_mir_way0(w);
810 
811 	pci_read_config_word(pdev, I5100_AMIR_0, &w);
812 	priv->amir[0] = w;
813 	pci_read_config_word(pdev, I5100_AMIR_1, &w);
814 	priv->amir[1] = w;
815 
816 	for (i = 0; i < I5100_CHANNELS; i++) {
817 		int j;
818 
819 		for (j = 0; j < 5; j++) {
820 			int k;
821 
822 			pci_read_config_dword(mms[i], I5100_DMIR + j * 4, &dw);
823 
824 			priv->dmir[i][j].limit =
825 				(u64) i5100_dmir_limit(dw) << 28;
826 			for (k = 0; k < I5100_MAX_RANKS_PER_DIMM; k++)
827 				priv->dmir[i][j].rank[k] =
828 					i5100_dmir_rank(dw, k);
829 		}
830 	}
831 
832 	i5100_init_mtr(mci);
833 }
834 
835 static void i5100_init_csrows(struct mem_ctl_info *mci)
836 {
837 	struct i5100_priv *priv = mci->pvt_info;
838 	struct dimm_info *dimm;
839 
840 	mci_for_each_dimm(mci, dimm) {
841 		const unsigned long npages = i5100_npages(mci, dimm->idx);
842 		const unsigned int chan = i5100_csrow_to_chan(mci, dimm->idx);
843 		const unsigned int rank = i5100_csrow_to_rank(mci, dimm->idx);
844 
845 		if (!npages)
846 			continue;
847 
848 		dimm->nr_pages = npages;
849 		dimm->grain = 32;
850 		dimm->dtype = (priv->mtr[chan][rank].width == 4) ?
851 				DEV_X4 : DEV_X8;
852 		dimm->mtype = MEM_RDDR2;
853 		dimm->edac_mode = EDAC_SECDED;
854 		snprintf(dimm->label, sizeof(dimm->label), "DIMM%u",
855 			 i5100_rank_to_slot(mci, chan, rank));
856 
857 		edac_dbg(2, "dimm channel %d, rank %d, size %ld\n",
858 			 chan, rank, (long)PAGES_TO_MiB(npages));
859 	}
860 }
861 
862 /****************************************************************************
863  *                       Error injection routines
864  ****************************************************************************/
865 
866 static void i5100_do_inject(struct mem_ctl_info *mci)
867 {
868 	struct i5100_priv *priv = mci->pvt_info;
869 	u32 mask0;
870 	u16 mask1;
871 
872 	/* MEM[1:0]EINJMSK0
873 	 * 31    - ADDRMATCHEN
874 	 * 29:28 - HLINESEL
875 	 *         00 Reserved
876 	 *         01 Lower half of cache line
877 	 *         10 Upper half of cache line
878 	 *         11 Both upper and lower parts of cache line
879 	 * 27    - EINJEN
880 	 * 25:19 - XORMASK1 for deviceptr1
881 	 * 9:5   - SEC2RAM for deviceptr2
882 	 * 4:0   - FIR2RAM for deviceptr1
883 	 */
884 	mask0 = ((priv->inject_hlinesel & 0x3) << 28) |
885 		I5100_MEMXEINJMSK0_EINJEN |
886 		((priv->inject_eccmask1 & 0xffff) << 10) |
887 		((priv->inject_deviceptr2 & 0x1f) << 5) |
888 		(priv->inject_deviceptr1 & 0x1f);
889 
890 	/* MEM[1:0]EINJMSK1
891 	 * 15:0  - XORMASK2 for deviceptr2
892 	 */
893 	mask1 = priv->inject_eccmask2;
894 
895 	if (priv->inject_channel == 0) {
896 		pci_write_config_dword(priv->mc, I5100_MEM0EINJMSK0, mask0);
897 		pci_write_config_word(priv->mc, I5100_MEM0EINJMSK1, mask1);
898 	} else {
899 		pci_write_config_dword(priv->mc, I5100_MEM1EINJMSK0, mask0);
900 		pci_write_config_word(priv->mc, I5100_MEM1EINJMSK1, mask1);
901 	}
902 
903 	/* Error Injection Response Function
904 	 * Intel 5100 Memory Controller Hub Chipset (318378) datasheet
905 	 * hints about this register but carry no data about them. All
906 	 * data regarding device 19 is based on experimentation and the
907 	 * Intel 7300 Chipset Memory Controller Hub (318082) datasheet
908 	 * which appears to be accurate for the i5100 in this area.
909 	 *
910 	 * The injection code don't work without setting this register.
911 	 * The register needs to be flipped off then on else the hardware
912 	 * will only preform the first injection.
913 	 *
914 	 * Stop condition bits 7:4
915 	 * 1010 - Stop after one injection
916 	 * 1011 - Never stop injecting faults
917 	 *
918 	 * Start condition bits 3:0
919 	 * 1010 - Never start
920 	 * 1011 - Start immediately
921 	 */
922 	pci_write_config_byte(priv->einj, I5100_DINJ0, 0xaa);
923 	pci_write_config_byte(priv->einj, I5100_DINJ0, 0xab);
924 }
925 
926 #define to_mci(k) container_of(k, struct mem_ctl_info, dev)
927 static ssize_t inject_enable_write(struct file *file, const char __user *data,
928 		size_t count, loff_t *ppos)
929 {
930 	struct device *dev = file->private_data;
931 	struct mem_ctl_info *mci = to_mci(dev);
932 
933 	i5100_do_inject(mci);
934 
935 	return count;
936 }
937 
938 static const struct file_operations i5100_inject_enable_fops = {
939 	.open = simple_open,
940 	.write = inject_enable_write,
941 	.llseek = generic_file_llseek,
942 };
943 
944 static int i5100_setup_debugfs(struct mem_ctl_info *mci)
945 {
946 	struct i5100_priv *priv = mci->pvt_info;
947 
948 	if (!i5100_debugfs)
949 		return -ENODEV;
950 
951 	priv->debugfs = edac_debugfs_create_dir_at(mci->bus->name, i5100_debugfs);
952 
953 	if (!priv->debugfs)
954 		return -ENOMEM;
955 
956 	edac_debugfs_create_x8("inject_channel", S_IRUGO | S_IWUSR, priv->debugfs,
957 				&priv->inject_channel);
958 	edac_debugfs_create_x8("inject_hlinesel", S_IRUGO | S_IWUSR, priv->debugfs,
959 				&priv->inject_hlinesel);
960 	edac_debugfs_create_x8("inject_deviceptr1", S_IRUGO | S_IWUSR, priv->debugfs,
961 				&priv->inject_deviceptr1);
962 	edac_debugfs_create_x8("inject_deviceptr2", S_IRUGO | S_IWUSR, priv->debugfs,
963 				&priv->inject_deviceptr2);
964 	edac_debugfs_create_x16("inject_eccmask1", S_IRUGO | S_IWUSR, priv->debugfs,
965 				&priv->inject_eccmask1);
966 	edac_debugfs_create_x16("inject_eccmask2", S_IRUGO | S_IWUSR, priv->debugfs,
967 				&priv->inject_eccmask2);
968 	edac_debugfs_create_file("inject_enable", S_IWUSR, priv->debugfs,
969 				&mci->dev, &i5100_inject_enable_fops);
970 
971 	return 0;
972 
973 }
974 
975 static int i5100_init_one(struct pci_dev *pdev, const struct pci_device_id *id)
976 {
977 	int rc;
978 	struct mem_ctl_info *mci;
979 	struct edac_mc_layer layers[2];
980 	struct i5100_priv *priv;
981 	struct pci_dev *ch0mm, *ch1mm, *einj;
982 	int ret = 0;
983 	u32 dw;
984 	int ranksperch;
985 
986 	if (PCI_FUNC(pdev->devfn) != 1)
987 		return -ENODEV;
988 
989 	rc = pci_enable_device(pdev);
990 	if (rc < 0) {
991 		ret = rc;
992 		goto bail;
993 	}
994 
995 	/* ECC enabled? */
996 	pci_read_config_dword(pdev, I5100_MC, &dw);
997 	if (!i5100_mc_errdeten(dw)) {
998 		printk(KERN_INFO "i5100_edac: ECC not enabled.\n");
999 		ret = -ENODEV;
1000 		goto bail_pdev;
1001 	}
1002 
1003 	/* figure out how many ranks, from strapped state of 48GB_Mode input */
1004 	pci_read_config_dword(pdev, I5100_MS, &dw);
1005 	ranksperch = !!(dw & (1 << 8)) * 2 + 4;
1006 
1007 	/* enable error reporting... */
1008 	pci_read_config_dword(pdev, I5100_EMASK_MEM, &dw);
1009 	dw &= ~I5100_FERR_NF_MEM_ANY_MASK;
1010 	pci_write_config_dword(pdev, I5100_EMASK_MEM, dw);
1011 
1012 	/* device 21, func 0, Channel 0 Memory Map, Error Flag/Mask, etc... */
1013 	ch0mm = pci_get_device_func(PCI_VENDOR_ID_INTEL,
1014 				    PCI_DEVICE_ID_INTEL_5100_21, 0);
1015 	if (!ch0mm) {
1016 		ret = -ENODEV;
1017 		goto bail_pdev;
1018 	}
1019 
1020 	rc = pci_enable_device(ch0mm);
1021 	if (rc < 0) {
1022 		ret = rc;
1023 		goto bail_ch0;
1024 	}
1025 
1026 	/* device 22, func 0, Channel 1 Memory Map, Error Flag/Mask, etc... */
1027 	ch1mm = pci_get_device_func(PCI_VENDOR_ID_INTEL,
1028 				    PCI_DEVICE_ID_INTEL_5100_22, 0);
1029 	if (!ch1mm) {
1030 		ret = -ENODEV;
1031 		goto bail_disable_ch0;
1032 	}
1033 
1034 	rc = pci_enable_device(ch1mm);
1035 	if (rc < 0) {
1036 		ret = rc;
1037 		goto bail_ch1;
1038 	}
1039 
1040 	layers[0].type = EDAC_MC_LAYER_CHANNEL;
1041 	layers[0].size = 2;
1042 	layers[0].is_virt_csrow = false;
1043 	layers[1].type = EDAC_MC_LAYER_SLOT;
1044 	layers[1].size = ranksperch;
1045 	layers[1].is_virt_csrow = true;
1046 	mci = edac_mc_alloc(0, ARRAY_SIZE(layers), layers,
1047 			    sizeof(*priv));
1048 	if (!mci) {
1049 		ret = -ENOMEM;
1050 		goto bail_disable_ch1;
1051 	}
1052 
1053 
1054 	/* device 19, func 0, Error injection */
1055 	einj = pci_get_device_func(PCI_VENDOR_ID_INTEL,
1056 				    PCI_DEVICE_ID_INTEL_5100_19, 0);
1057 	if (!einj) {
1058 		ret = -ENODEV;
1059 		goto bail_mc_free;
1060 	}
1061 
1062 	rc = pci_enable_device(einj);
1063 	if (rc < 0) {
1064 		ret = rc;
1065 		goto bail_einj;
1066 	}
1067 
1068 	mci->pdev = &pdev->dev;
1069 
1070 	priv = mci->pvt_info;
1071 	priv->ranksperchan = ranksperch;
1072 	priv->mc = pdev;
1073 	priv->ch0mm = ch0mm;
1074 	priv->ch1mm = ch1mm;
1075 	priv->einj = einj;
1076 
1077 	INIT_DELAYED_WORK(&(priv->i5100_scrubbing), i5100_refresh_scrubbing);
1078 
1079 	/* If scrubbing was already enabled by the bios, start maintaining it */
1080 	pci_read_config_dword(pdev, I5100_MC, &dw);
1081 	if (i5100_mc_scrben(dw)) {
1082 		priv->scrub_enable = 1;
1083 		schedule_delayed_work(&(priv->i5100_scrubbing),
1084 				      I5100_SCRUB_REFRESH_RATE);
1085 	}
1086 
1087 	i5100_init_dimm_layout(pdev, mci);
1088 	i5100_init_interleaving(pdev, mci);
1089 
1090 	mci->mtype_cap = MEM_FLAG_FB_DDR2;
1091 	mci->edac_ctl_cap = EDAC_FLAG_SECDED;
1092 	mci->edac_cap = EDAC_FLAG_SECDED;
1093 	mci->mod_name = "i5100_edac.c";
1094 	mci->ctl_name = "i5100";
1095 	mci->dev_name = pci_name(pdev);
1096 	mci->ctl_page_to_phys = NULL;
1097 
1098 	mci->edac_check = i5100_check_error;
1099 	mci->set_sdram_scrub_rate = i5100_set_scrub_rate;
1100 	mci->get_sdram_scrub_rate = i5100_get_scrub_rate;
1101 
1102 	priv->inject_channel = 0;
1103 	priv->inject_hlinesel = 0;
1104 	priv->inject_deviceptr1 = 0;
1105 	priv->inject_deviceptr2 = 0;
1106 	priv->inject_eccmask1 = 0;
1107 	priv->inject_eccmask2 = 0;
1108 
1109 	i5100_init_csrows(mci);
1110 
1111 	/* this strange construction seems to be in every driver, dunno why */
1112 	switch (edac_op_state) {
1113 	case EDAC_OPSTATE_POLL:
1114 	case EDAC_OPSTATE_NMI:
1115 		break;
1116 	default:
1117 		edac_op_state = EDAC_OPSTATE_POLL;
1118 		break;
1119 	}
1120 
1121 	if (edac_mc_add_mc(mci)) {
1122 		ret = -ENODEV;
1123 		goto bail_scrub;
1124 	}
1125 
1126 	i5100_setup_debugfs(mci);
1127 
1128 	return ret;
1129 
1130 bail_scrub:
1131 	priv->scrub_enable = 0;
1132 	cancel_delayed_work_sync(&(priv->i5100_scrubbing));
1133 	pci_disable_device(einj);
1134 
1135 bail_einj:
1136 	pci_dev_put(einj);
1137 
1138 bail_mc_free:
1139 	edac_mc_free(mci);
1140 
1141 bail_disable_ch1:
1142 	pci_disable_device(ch1mm);
1143 
1144 bail_ch1:
1145 	pci_dev_put(ch1mm);
1146 
1147 bail_disable_ch0:
1148 	pci_disable_device(ch0mm);
1149 
1150 bail_ch0:
1151 	pci_dev_put(ch0mm);
1152 
1153 bail_pdev:
1154 	pci_disable_device(pdev);
1155 
1156 bail:
1157 	return ret;
1158 }
1159 
1160 static void i5100_remove_one(struct pci_dev *pdev)
1161 {
1162 	struct mem_ctl_info *mci;
1163 	struct i5100_priv *priv;
1164 
1165 	mci = edac_mc_del_mc(&pdev->dev);
1166 
1167 	if (!mci)
1168 		return;
1169 
1170 	priv = mci->pvt_info;
1171 
1172 	edac_debugfs_remove_recursive(priv->debugfs);
1173 
1174 	priv->scrub_enable = 0;
1175 	cancel_delayed_work_sync(&(priv->i5100_scrubbing));
1176 
1177 	pci_disable_device(pdev);
1178 	pci_disable_device(priv->ch0mm);
1179 	pci_disable_device(priv->ch1mm);
1180 	pci_disable_device(priv->einj);
1181 	pci_dev_put(priv->ch0mm);
1182 	pci_dev_put(priv->ch1mm);
1183 	pci_dev_put(priv->einj);
1184 
1185 	edac_mc_free(mci);
1186 }
1187 
1188 static const struct pci_device_id i5100_pci_tbl[] = {
1189 	/* Device 16, Function 0, Channel 0 Memory Map, Error Flag/Mask, ... */
1190 	{ PCI_DEVICE(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_5100_16) },
1191 	{ 0, }
1192 };
1193 MODULE_DEVICE_TABLE(pci, i5100_pci_tbl);
1194 
1195 static struct pci_driver i5100_driver = {
1196 	.name = KBUILD_BASENAME,
1197 	.probe = i5100_init_one,
1198 	.remove = i5100_remove_one,
1199 	.id_table = i5100_pci_tbl,
1200 };
1201 
1202 static int __init i5100_init(void)
1203 {
1204 	int pci_rc;
1205 
1206 	i5100_debugfs = edac_debugfs_create_dir_at("i5100_edac", NULL);
1207 
1208 	pci_rc = pci_register_driver(&i5100_driver);
1209 	return (pci_rc < 0) ? pci_rc : 0;
1210 }
1211 
1212 static void __exit i5100_exit(void)
1213 {
1214 	edac_debugfs_remove(i5100_debugfs);
1215 
1216 	pci_unregister_driver(&i5100_driver);
1217 }
1218 
1219 module_init(i5100_init);
1220 module_exit(i5100_exit);
1221 
1222 MODULE_LICENSE("GPL");
1223 MODULE_AUTHOR
1224     ("Arthur Jones <ajones@riverbed.com>");
1225 MODULE_DESCRIPTION("MC Driver for Intel I5100 memory controllers");
1226