xref: /openbmc/linux/drivers/edac/i5100_edac.c (revision 3a83e4e6)
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_nrecmema_dm_buf_id(u32 a)
248 {
249 	return a & ((1 << 8) - 1);
250 }
251 
252 static inline u32 i5100_nrecmemb_cas(u32 a)
253 {
254 	return a >> 16 & ((1 << 13) - 1);
255 }
256 
257 static inline u32 i5100_nrecmemb_ras(u32 a)
258 {
259 	return a & ((1 << 16) - 1);
260 }
261 
262 static inline u32 i5100_recmema_merr(u32 a)
263 {
264 	return i5100_nrecmema_merr(a);
265 }
266 
267 static inline u32 i5100_recmema_bank(u32 a)
268 {
269 	return i5100_nrecmema_bank(a);
270 }
271 
272 static inline u32 i5100_recmema_rank(u32 a)
273 {
274 	return i5100_nrecmema_rank(a);
275 }
276 
277 static inline u32 i5100_recmemb_cas(u32 a)
278 {
279 	return i5100_nrecmemb_cas(a);
280 }
281 
282 static inline u32 i5100_recmemb_ras(u32 a)
283 {
284 	return i5100_nrecmemb_ras(a);
285 }
286 
287 /* some generic limits */
288 #define I5100_MAX_RANKS_PER_CHAN	6
289 #define I5100_CHANNELS			    2
290 #define I5100_MAX_RANKS_PER_DIMM	4
291 #define I5100_DIMM_ADDR_LINES		(6 - 3)	/* 64 bits / 8 bits per byte */
292 #define I5100_MAX_DIMM_SLOTS_PER_CHAN	4
293 #define I5100_MAX_RANK_INTERLEAVE	4
294 #define I5100_MAX_DMIRS			5
295 #define I5100_SCRUB_REFRESH_RATE	(5 * 60 * HZ)
296 
297 struct i5100_priv {
298 	/* ranks on each dimm -- 0 maps to not present -- obtained via SPD */
299 	int dimm_numrank[I5100_CHANNELS][I5100_MAX_DIMM_SLOTS_PER_CHAN];
300 
301 	/*
302 	 * mainboard chip select map -- maps i5100 chip selects to
303 	 * DIMM slot chip selects.  In the case of only 4 ranks per
304 	 * channel, the mapping is fairly obvious but not unique.
305 	 * we map -1 -> NC and assume both channels use the same
306 	 * map...
307 	 *
308 	 */
309 	int dimm_csmap[I5100_MAX_DIMM_SLOTS_PER_CHAN][I5100_MAX_RANKS_PER_DIMM];
310 
311 	/* memory interleave range */
312 	struct {
313 		u64	 limit;
314 		unsigned way[2];
315 	} mir[I5100_CHANNELS];
316 
317 	/* adjusted memory interleave range register */
318 	unsigned amir[I5100_CHANNELS];
319 
320 	/* dimm interleave range */
321 	struct {
322 		unsigned rank[I5100_MAX_RANK_INTERLEAVE];
323 		u64	 limit;
324 	} dmir[I5100_CHANNELS][I5100_MAX_DMIRS];
325 
326 	/* memory technology registers... */
327 	struct {
328 		unsigned present;	/* 0 or 1 */
329 		unsigned ethrottle;	/* 0 or 1 */
330 		unsigned width;		/* 4 or 8 bits  */
331 		unsigned numbank;	/* 2 or 3 lines */
332 		unsigned numrow;	/* 13 .. 16 lines */
333 		unsigned numcol;	/* 11 .. 12 lines */
334 	} mtr[I5100_CHANNELS][I5100_MAX_RANKS_PER_CHAN];
335 
336 	u64 tolm;		/* top of low memory in bytes */
337 	unsigned ranksperchan;	/* number of ranks per channel */
338 
339 	struct pci_dev *mc;	/* device 16 func 1 */
340 	struct pci_dev *einj;	/* device 19 func 0 */
341 	struct pci_dev *ch0mm;	/* device 21 func 0 */
342 	struct pci_dev *ch1mm;	/* device 22 func 0 */
343 
344 	struct delayed_work i5100_scrubbing;
345 	int scrub_enable;
346 
347 	/* Error injection */
348 	u8 inject_channel;
349 	u8 inject_hlinesel;
350 	u8 inject_deviceptr1;
351 	u8 inject_deviceptr2;
352 	u16 inject_eccmask1;
353 	u16 inject_eccmask2;
354 
355 	struct dentry *debugfs;
356 };
357 
358 static struct dentry *i5100_debugfs;
359 
360 /* map a rank/chan to a slot number on the mainboard */
361 static int i5100_rank_to_slot(const struct mem_ctl_info *mci,
362 			      int chan, int rank)
363 {
364 	const struct i5100_priv *priv = mci->pvt_info;
365 	int i;
366 
367 	for (i = 0; i < I5100_MAX_DIMM_SLOTS_PER_CHAN; i++) {
368 		int j;
369 		const int numrank = priv->dimm_numrank[chan][i];
370 
371 		for (j = 0; j < numrank; j++)
372 			if (priv->dimm_csmap[i][j] == rank)
373 				return i * 2 + chan;
374 	}
375 
376 	return -1;
377 }
378 
379 static const char *i5100_err_msg(unsigned err)
380 {
381 	static const char *merrs[] = {
382 		"unknown", /* 0 */
383 		"uncorrectable data ECC on replay", /* 1 */
384 		"unknown", /* 2 */
385 		"unknown", /* 3 */
386 		"aliased uncorrectable demand data ECC", /* 4 */
387 		"aliased uncorrectable spare-copy data ECC", /* 5 */
388 		"aliased uncorrectable patrol data ECC", /* 6 */
389 		"unknown", /* 7 */
390 		"unknown", /* 8 */
391 		"unknown", /* 9 */
392 		"non-aliased uncorrectable demand data ECC", /* 10 */
393 		"non-aliased uncorrectable spare-copy data ECC", /* 11 */
394 		"non-aliased uncorrectable patrol data ECC", /* 12 */
395 		"unknown", /* 13 */
396 		"correctable demand data ECC", /* 14 */
397 		"correctable spare-copy data ECC", /* 15 */
398 		"correctable patrol data ECC", /* 16 */
399 		"unknown", /* 17 */
400 		"SPD protocol error", /* 18 */
401 		"unknown", /* 19 */
402 		"spare copy initiated", /* 20 */
403 		"spare copy completed", /* 21 */
404 	};
405 	unsigned i;
406 
407 	for (i = 0; i < ARRAY_SIZE(merrs); i++)
408 		if (1 << i & err)
409 			return merrs[i];
410 
411 	return "none";
412 }
413 
414 /* convert csrow index into a rank (per channel -- 0..5) */
415 static unsigned int i5100_csrow_to_rank(const struct mem_ctl_info *mci,
416 					unsigned int csrow)
417 {
418 	const struct i5100_priv *priv = mci->pvt_info;
419 
420 	return csrow % priv->ranksperchan;
421 }
422 
423 /* convert csrow index into a channel (0..1) */
424 static unsigned int i5100_csrow_to_chan(const struct mem_ctl_info *mci,
425 					unsigned int csrow)
426 {
427 	const struct i5100_priv *priv = mci->pvt_info;
428 
429 	return csrow / priv->ranksperchan;
430 }
431 
432 static void i5100_handle_ce(struct mem_ctl_info *mci,
433 			    int chan,
434 			    unsigned bank,
435 			    unsigned rank,
436 			    unsigned long syndrome,
437 			    unsigned cas,
438 			    unsigned ras,
439 			    const char *msg)
440 {
441 	char detail[80];
442 
443 	/* Form out message */
444 	snprintf(detail, sizeof(detail),
445 		 "bank %u, cas %u, ras %u\n",
446 		 bank, cas, ras);
447 
448 	edac_mc_handle_error(HW_EVENT_ERR_CORRECTED, mci, 1,
449 			     0, 0, syndrome,
450 			     chan, rank, -1,
451 			     msg, detail);
452 }
453 
454 static void i5100_handle_ue(struct mem_ctl_info *mci,
455 			    int chan,
456 			    unsigned bank,
457 			    unsigned rank,
458 			    unsigned long syndrome,
459 			    unsigned cas,
460 			    unsigned ras,
461 			    const char *msg)
462 {
463 	char detail[80];
464 
465 	/* Form out message */
466 	snprintf(detail, sizeof(detail),
467 		 "bank %u, cas %u, ras %u\n",
468 		 bank, cas, ras);
469 
470 	edac_mc_handle_error(HW_EVENT_ERR_UNCORRECTED, mci, 1,
471 			     0, 0, syndrome,
472 			     chan, rank, -1,
473 			     msg, detail);
474 }
475 
476 static void i5100_read_log(struct mem_ctl_info *mci, int chan,
477 			   u32 ferr, u32 nerr)
478 {
479 	struct i5100_priv *priv = mci->pvt_info;
480 	struct pci_dev *pdev = (chan) ? priv->ch1mm : priv->ch0mm;
481 	u32 dw;
482 	u32 dw2;
483 	unsigned syndrome = 0;
484 	unsigned merr;
485 	unsigned bank;
486 	unsigned rank;
487 	unsigned cas;
488 	unsigned ras;
489 
490 	pci_read_config_dword(pdev, I5100_VALIDLOG, &dw);
491 
492 	if (i5100_validlog_redmemvalid(dw)) {
493 		pci_read_config_dword(pdev, I5100_REDMEMA, &dw2);
494 		syndrome = dw2;
495 		pci_read_config_dword(pdev, I5100_REDMEMB, &dw2);
496 	}
497 
498 	if (i5100_validlog_recmemvalid(dw)) {
499 		const char *msg;
500 
501 		pci_read_config_dword(pdev, I5100_RECMEMA, &dw2);
502 		merr = i5100_recmema_merr(dw2);
503 		bank = i5100_recmema_bank(dw2);
504 		rank = i5100_recmema_rank(dw2);
505 
506 		pci_read_config_dword(pdev, I5100_RECMEMB, &dw2);
507 		cas = i5100_recmemb_cas(dw2);
508 		ras = i5100_recmemb_ras(dw2);
509 
510 		/* FIXME:  not really sure if this is what merr is...
511 		 */
512 		if (!merr)
513 			msg = i5100_err_msg(ferr);
514 		else
515 			msg = i5100_err_msg(nerr);
516 
517 		i5100_handle_ce(mci, chan, bank, rank, syndrome, cas, ras, msg);
518 	}
519 
520 	if (i5100_validlog_nrecmemvalid(dw)) {
521 		const char *msg;
522 
523 		pci_read_config_dword(pdev, I5100_NRECMEMA, &dw2);
524 		merr = i5100_nrecmema_merr(dw2);
525 		bank = i5100_nrecmema_bank(dw2);
526 		rank = i5100_nrecmema_rank(dw2);
527 
528 		pci_read_config_dword(pdev, I5100_NRECMEMB, &dw2);
529 		cas = i5100_nrecmemb_cas(dw2);
530 		ras = i5100_nrecmemb_ras(dw2);
531 
532 		/* FIXME:  not really sure if this is what merr is...
533 		 */
534 		if (!merr)
535 			msg = i5100_err_msg(ferr);
536 		else
537 			msg = i5100_err_msg(nerr);
538 
539 		i5100_handle_ue(mci, chan, bank, rank, syndrome, cas, ras, msg);
540 	}
541 
542 	pci_write_config_dword(pdev, I5100_VALIDLOG, dw);
543 }
544 
545 static void i5100_check_error(struct mem_ctl_info *mci)
546 {
547 	struct i5100_priv *priv = mci->pvt_info;
548 	u32 dw, dw2;
549 
550 	pci_read_config_dword(priv->mc, I5100_FERR_NF_MEM, &dw);
551 	if (i5100_ferr_nf_mem_any(dw)) {
552 
553 		pci_read_config_dword(priv->mc, I5100_NERR_NF_MEM, &dw2);
554 
555 		i5100_read_log(mci, i5100_ferr_nf_mem_chan_indx(dw),
556 			       i5100_ferr_nf_mem_any(dw),
557 			       i5100_nerr_nf_mem_any(dw2));
558 
559 		pci_write_config_dword(priv->mc, I5100_NERR_NF_MEM, dw2);
560 	}
561 	pci_write_config_dword(priv->mc, I5100_FERR_NF_MEM, dw);
562 }
563 
564 /* The i5100 chipset will scrub the entire memory once, then
565  * set a done bit. Continuous scrubbing is achieved by enqueing
566  * delayed work to a workqueue, checking every few minutes if
567  * the scrubbing has completed and if so reinitiating it.
568  */
569 
570 static void i5100_refresh_scrubbing(struct work_struct *work)
571 {
572 	struct delayed_work *i5100_scrubbing = to_delayed_work(work);
573 	struct i5100_priv *priv = container_of(i5100_scrubbing,
574 					       struct i5100_priv,
575 					       i5100_scrubbing);
576 	u32 dw;
577 
578 	pci_read_config_dword(priv->mc, I5100_MC, &dw);
579 
580 	if (priv->scrub_enable) {
581 
582 		pci_read_config_dword(priv->mc, I5100_MC, &dw);
583 
584 		if (i5100_mc_scrbdone(dw)) {
585 			dw |= I5100_MC_SCRBEN_MASK;
586 			pci_write_config_dword(priv->mc, I5100_MC, dw);
587 			pci_read_config_dword(priv->mc, I5100_MC, &dw);
588 		}
589 
590 		schedule_delayed_work(&(priv->i5100_scrubbing),
591 				      I5100_SCRUB_REFRESH_RATE);
592 	}
593 }
594 /*
595  * The bandwidth is based on experimentation, feel free to refine it.
596  */
597 static int i5100_set_scrub_rate(struct mem_ctl_info *mci, u32 bandwidth)
598 {
599 	struct i5100_priv *priv = mci->pvt_info;
600 	u32 dw;
601 
602 	pci_read_config_dword(priv->mc, I5100_MC, &dw);
603 	if (bandwidth) {
604 		priv->scrub_enable = 1;
605 		dw |= I5100_MC_SCRBEN_MASK;
606 		schedule_delayed_work(&(priv->i5100_scrubbing),
607 				      I5100_SCRUB_REFRESH_RATE);
608 	} else {
609 		priv->scrub_enable = 0;
610 		dw &= ~I5100_MC_SCRBEN_MASK;
611 		cancel_delayed_work(&(priv->i5100_scrubbing));
612 	}
613 	pci_write_config_dword(priv->mc, I5100_MC, dw);
614 
615 	pci_read_config_dword(priv->mc, I5100_MC, &dw);
616 
617 	bandwidth = 5900000 * i5100_mc_scrben(dw);
618 
619 	return bandwidth;
620 }
621 
622 static int i5100_get_scrub_rate(struct mem_ctl_info *mci)
623 {
624 	struct i5100_priv *priv = mci->pvt_info;
625 	u32 dw;
626 
627 	pci_read_config_dword(priv->mc, I5100_MC, &dw);
628 
629 	return 5900000 * i5100_mc_scrben(dw);
630 }
631 
632 static struct pci_dev *pci_get_device_func(unsigned vendor,
633 					   unsigned device,
634 					   unsigned func)
635 {
636 	struct pci_dev *ret = NULL;
637 
638 	while (1) {
639 		ret = pci_get_device(vendor, device, ret);
640 
641 		if (!ret)
642 			break;
643 
644 		if (PCI_FUNC(ret->devfn) == func)
645 			break;
646 	}
647 
648 	return ret;
649 }
650 
651 static unsigned long i5100_npages(struct mem_ctl_info *mci, unsigned int csrow)
652 {
653 	struct i5100_priv *priv = mci->pvt_info;
654 	const unsigned int chan_rank = i5100_csrow_to_rank(mci, csrow);
655 	const unsigned int chan = i5100_csrow_to_chan(mci, csrow);
656 	unsigned addr_lines;
657 
658 	/* dimm present? */
659 	if (!priv->mtr[chan][chan_rank].present)
660 		return 0ULL;
661 
662 	addr_lines =
663 		I5100_DIMM_ADDR_LINES +
664 		priv->mtr[chan][chan_rank].numcol +
665 		priv->mtr[chan][chan_rank].numrow +
666 		priv->mtr[chan][chan_rank].numbank;
667 
668 	return (unsigned long)
669 		((unsigned long long) (1ULL << addr_lines) / PAGE_SIZE);
670 }
671 
672 static void i5100_init_mtr(struct mem_ctl_info *mci)
673 {
674 	struct i5100_priv *priv = mci->pvt_info;
675 	struct pci_dev *mms[2] = { priv->ch0mm, priv->ch1mm };
676 	int i;
677 
678 	for (i = 0; i < I5100_CHANNELS; i++) {
679 		int j;
680 		struct pci_dev *pdev = mms[i];
681 
682 		for (j = 0; j < I5100_MAX_RANKS_PER_CHAN; j++) {
683 			const unsigned addr =
684 				(j < 4) ? I5100_MTR_0 + j * 2 :
685 					  I5100_MTR_4 + (j - 4) * 2;
686 			u16 w;
687 
688 			pci_read_config_word(pdev, addr, &w);
689 
690 			priv->mtr[i][j].present = i5100_mtr_present(w);
691 			priv->mtr[i][j].ethrottle = i5100_mtr_ethrottle(w);
692 			priv->mtr[i][j].width = 4 + 4 * i5100_mtr_width(w);
693 			priv->mtr[i][j].numbank = 2 + i5100_mtr_numbank(w);
694 			priv->mtr[i][j].numrow = 13 + i5100_mtr_numrow(w);
695 			priv->mtr[i][j].numcol = 10 + i5100_mtr_numcol(w);
696 		}
697 	}
698 }
699 
700 /*
701  * FIXME: make this into a real i2c adapter (so that dimm-decode
702  * will work)?
703  */
704 static int i5100_read_spd_byte(const struct mem_ctl_info *mci,
705 			       u8 ch, u8 slot, u8 addr, u8 *byte)
706 {
707 	struct i5100_priv *priv = mci->pvt_info;
708 	u16 w;
709 
710 	pci_read_config_word(priv->mc, I5100_SPDDATA, &w);
711 	if (i5100_spddata_busy(w))
712 		return -1;
713 
714 	pci_write_config_dword(priv->mc, I5100_SPDCMD,
715 			       i5100_spdcmd_create(0xa, 1, ch * 4 + slot, addr,
716 						   0, 0));
717 
718 	/* wait up to 100ms */
719 	udelay(100);
720 	while (1) {
721 		pci_read_config_word(priv->mc, I5100_SPDDATA, &w);
722 		if (!i5100_spddata_busy(w))
723 			break;
724 		udelay(100);
725 	}
726 
727 	if (!i5100_spddata_rdo(w) || i5100_spddata_sbe(w))
728 		return -1;
729 
730 	*byte = i5100_spddata_data(w);
731 
732 	return 0;
733 }
734 
735 /*
736  * fill dimm chip select map
737  *
738  * FIXME:
739  *   o not the only way to may chip selects to dimm slots
740  *   o investigate if there is some way to obtain this map from the bios
741  */
742 static void i5100_init_dimm_csmap(struct mem_ctl_info *mci)
743 {
744 	struct i5100_priv *priv = mci->pvt_info;
745 	int i;
746 
747 	for (i = 0; i < I5100_MAX_DIMM_SLOTS_PER_CHAN; i++) {
748 		int j;
749 
750 		for (j = 0; j < I5100_MAX_RANKS_PER_DIMM; j++)
751 			priv->dimm_csmap[i][j] = -1; /* default NC */
752 	}
753 
754 	/* only 2 chip selects per slot... */
755 	if (priv->ranksperchan == 4) {
756 		priv->dimm_csmap[0][0] = 0;
757 		priv->dimm_csmap[0][1] = 3;
758 		priv->dimm_csmap[1][0] = 1;
759 		priv->dimm_csmap[1][1] = 2;
760 		priv->dimm_csmap[2][0] = 2;
761 		priv->dimm_csmap[3][0] = 3;
762 	} else {
763 		priv->dimm_csmap[0][0] = 0;
764 		priv->dimm_csmap[0][1] = 1;
765 		priv->dimm_csmap[1][0] = 2;
766 		priv->dimm_csmap[1][1] = 3;
767 		priv->dimm_csmap[2][0] = 4;
768 		priv->dimm_csmap[2][1] = 5;
769 	}
770 }
771 
772 static void i5100_init_dimm_layout(struct pci_dev *pdev,
773 				   struct mem_ctl_info *mci)
774 {
775 	struct i5100_priv *priv = mci->pvt_info;
776 	int i;
777 
778 	for (i = 0; i < I5100_CHANNELS; i++) {
779 		int j;
780 
781 		for (j = 0; j < I5100_MAX_DIMM_SLOTS_PER_CHAN; j++) {
782 			u8 rank;
783 
784 			if (i5100_read_spd_byte(mci, i, j, 5, &rank) < 0)
785 				priv->dimm_numrank[i][j] = 0;
786 			else
787 				priv->dimm_numrank[i][j] = (rank & 3) + 1;
788 		}
789 	}
790 
791 	i5100_init_dimm_csmap(mci);
792 }
793 
794 static void i5100_init_interleaving(struct pci_dev *pdev,
795 				    struct mem_ctl_info *mci)
796 {
797 	u16 w;
798 	u32 dw;
799 	struct i5100_priv *priv = mci->pvt_info;
800 	struct pci_dev *mms[2] = { priv->ch0mm, priv->ch1mm };
801 	int i;
802 
803 	pci_read_config_word(pdev, I5100_TOLM, &w);
804 	priv->tolm = (u64) i5100_tolm_tolm(w) * 256 * 1024 * 1024;
805 
806 	pci_read_config_word(pdev, I5100_MIR0, &w);
807 	priv->mir[0].limit = (u64) i5100_mir_limit(w) << 28;
808 	priv->mir[0].way[1] = i5100_mir_way1(w);
809 	priv->mir[0].way[0] = i5100_mir_way0(w);
810 
811 	pci_read_config_word(pdev, I5100_MIR1, &w);
812 	priv->mir[1].limit = (u64) i5100_mir_limit(w) << 28;
813 	priv->mir[1].way[1] = i5100_mir_way1(w);
814 	priv->mir[1].way[0] = i5100_mir_way0(w);
815 
816 	pci_read_config_word(pdev, I5100_AMIR_0, &w);
817 	priv->amir[0] = w;
818 	pci_read_config_word(pdev, I5100_AMIR_1, &w);
819 	priv->amir[1] = w;
820 
821 	for (i = 0; i < I5100_CHANNELS; i++) {
822 		int j;
823 
824 		for (j = 0; j < 5; j++) {
825 			int k;
826 
827 			pci_read_config_dword(mms[i], I5100_DMIR + j * 4, &dw);
828 
829 			priv->dmir[i][j].limit =
830 				(u64) i5100_dmir_limit(dw) << 28;
831 			for (k = 0; k < I5100_MAX_RANKS_PER_DIMM; k++)
832 				priv->dmir[i][j].rank[k] =
833 					i5100_dmir_rank(dw, k);
834 		}
835 	}
836 
837 	i5100_init_mtr(mci);
838 }
839 
840 static void i5100_init_csrows(struct mem_ctl_info *mci)
841 {
842 	struct i5100_priv *priv = mci->pvt_info;
843 	struct dimm_info *dimm;
844 
845 	mci_for_each_dimm(mci, dimm) {
846 		const unsigned long npages = i5100_npages(mci, dimm->idx);
847 		const unsigned int chan = i5100_csrow_to_chan(mci, dimm->idx);
848 		const unsigned int rank = i5100_csrow_to_rank(mci, dimm->idx);
849 
850 		if (!npages)
851 			continue;
852 
853 		dimm->nr_pages = npages;
854 		dimm->grain = 32;
855 		dimm->dtype = (priv->mtr[chan][rank].width == 4) ?
856 				DEV_X4 : DEV_X8;
857 		dimm->mtype = MEM_RDDR2;
858 		dimm->edac_mode = EDAC_SECDED;
859 		snprintf(dimm->label, sizeof(dimm->label), "DIMM%u",
860 			 i5100_rank_to_slot(mci, chan, rank));
861 
862 		edac_dbg(2, "dimm channel %d, rank %d, size %ld\n",
863 			 chan, rank, (long)PAGES_TO_MiB(npages));
864 	}
865 }
866 
867 /****************************************************************************
868  *                       Error injection routines
869  ****************************************************************************/
870 
871 static void i5100_do_inject(struct mem_ctl_info *mci)
872 {
873 	struct i5100_priv *priv = mci->pvt_info;
874 	u32 mask0;
875 	u16 mask1;
876 
877 	/* MEM[1:0]EINJMSK0
878 	 * 31    - ADDRMATCHEN
879 	 * 29:28 - HLINESEL
880 	 *         00 Reserved
881 	 *         01 Lower half of cache line
882 	 *         10 Upper half of cache line
883 	 *         11 Both upper and lower parts of cache line
884 	 * 27    - EINJEN
885 	 * 25:19 - XORMASK1 for deviceptr1
886 	 * 9:5   - SEC2RAM for deviceptr2
887 	 * 4:0   - FIR2RAM for deviceptr1
888 	 */
889 	mask0 = ((priv->inject_hlinesel & 0x3) << 28) |
890 		I5100_MEMXEINJMSK0_EINJEN |
891 		((priv->inject_eccmask1 & 0xffff) << 10) |
892 		((priv->inject_deviceptr2 & 0x1f) << 5) |
893 		(priv->inject_deviceptr1 & 0x1f);
894 
895 	/* MEM[1:0]EINJMSK1
896 	 * 15:0  - XORMASK2 for deviceptr2
897 	 */
898 	mask1 = priv->inject_eccmask2;
899 
900 	if (priv->inject_channel == 0) {
901 		pci_write_config_dword(priv->mc, I5100_MEM0EINJMSK0, mask0);
902 		pci_write_config_word(priv->mc, I5100_MEM0EINJMSK1, mask1);
903 	} else {
904 		pci_write_config_dword(priv->mc, I5100_MEM1EINJMSK0, mask0);
905 		pci_write_config_word(priv->mc, I5100_MEM1EINJMSK1, mask1);
906 	}
907 
908 	/* Error Injection Response Function
909 	 * Intel 5100 Memory Controller Hub Chipset (318378) datasheet
910 	 * hints about this register but carry no data about them. All
911 	 * data regarding device 19 is based on experimentation and the
912 	 * Intel 7300 Chipset Memory Controller Hub (318082) datasheet
913 	 * which appears to be accurate for the i5100 in this area.
914 	 *
915 	 * The injection code don't work without setting this register.
916 	 * The register needs to be flipped off then on else the hardware
917 	 * will only preform the first injection.
918 	 *
919 	 * Stop condition bits 7:4
920 	 * 1010 - Stop after one injection
921 	 * 1011 - Never stop injecting faults
922 	 *
923 	 * Start condition bits 3:0
924 	 * 1010 - Never start
925 	 * 1011 - Start immediately
926 	 */
927 	pci_write_config_byte(priv->einj, I5100_DINJ0, 0xaa);
928 	pci_write_config_byte(priv->einj, I5100_DINJ0, 0xab);
929 }
930 
931 #define to_mci(k) container_of(k, struct mem_ctl_info, dev)
932 static ssize_t inject_enable_write(struct file *file, const char __user *data,
933 		size_t count, loff_t *ppos)
934 {
935 	struct device *dev = file->private_data;
936 	struct mem_ctl_info *mci = to_mci(dev);
937 
938 	i5100_do_inject(mci);
939 
940 	return count;
941 }
942 
943 static const struct file_operations i5100_inject_enable_fops = {
944 	.open = simple_open,
945 	.write = inject_enable_write,
946 	.llseek = generic_file_llseek,
947 };
948 
949 static int i5100_setup_debugfs(struct mem_ctl_info *mci)
950 {
951 	struct i5100_priv *priv = mci->pvt_info;
952 
953 	if (!i5100_debugfs)
954 		return -ENODEV;
955 
956 	priv->debugfs = edac_debugfs_create_dir_at(mci->bus->name, i5100_debugfs);
957 
958 	if (!priv->debugfs)
959 		return -ENOMEM;
960 
961 	edac_debugfs_create_x8("inject_channel", S_IRUGO | S_IWUSR, priv->debugfs,
962 				&priv->inject_channel);
963 	edac_debugfs_create_x8("inject_hlinesel", S_IRUGO | S_IWUSR, priv->debugfs,
964 				&priv->inject_hlinesel);
965 	edac_debugfs_create_x8("inject_deviceptr1", S_IRUGO | S_IWUSR, priv->debugfs,
966 				&priv->inject_deviceptr1);
967 	edac_debugfs_create_x8("inject_deviceptr2", S_IRUGO | S_IWUSR, priv->debugfs,
968 				&priv->inject_deviceptr2);
969 	edac_debugfs_create_x16("inject_eccmask1", S_IRUGO | S_IWUSR, priv->debugfs,
970 				&priv->inject_eccmask1);
971 	edac_debugfs_create_x16("inject_eccmask2", S_IRUGO | S_IWUSR, priv->debugfs,
972 				&priv->inject_eccmask2);
973 	edac_debugfs_create_file("inject_enable", S_IWUSR, priv->debugfs,
974 				&mci->dev, &i5100_inject_enable_fops);
975 
976 	return 0;
977 
978 }
979 
980 static int i5100_init_one(struct pci_dev *pdev, const struct pci_device_id *id)
981 {
982 	int rc;
983 	struct mem_ctl_info *mci;
984 	struct edac_mc_layer layers[2];
985 	struct i5100_priv *priv;
986 	struct pci_dev *ch0mm, *ch1mm, *einj;
987 	int ret = 0;
988 	u32 dw;
989 	int ranksperch;
990 
991 	if (PCI_FUNC(pdev->devfn) != 1)
992 		return -ENODEV;
993 
994 	rc = pci_enable_device(pdev);
995 	if (rc < 0) {
996 		ret = rc;
997 		goto bail;
998 	}
999 
1000 	/* ECC enabled? */
1001 	pci_read_config_dword(pdev, I5100_MC, &dw);
1002 	if (!i5100_mc_errdeten(dw)) {
1003 		printk(KERN_INFO "i5100_edac: ECC not enabled.\n");
1004 		ret = -ENODEV;
1005 		goto bail_pdev;
1006 	}
1007 
1008 	/* figure out how many ranks, from strapped state of 48GB_Mode input */
1009 	pci_read_config_dword(pdev, I5100_MS, &dw);
1010 	ranksperch = !!(dw & (1 << 8)) * 2 + 4;
1011 
1012 	/* enable error reporting... */
1013 	pci_read_config_dword(pdev, I5100_EMASK_MEM, &dw);
1014 	dw &= ~I5100_FERR_NF_MEM_ANY_MASK;
1015 	pci_write_config_dword(pdev, I5100_EMASK_MEM, dw);
1016 
1017 	/* device 21, func 0, Channel 0 Memory Map, Error Flag/Mask, etc... */
1018 	ch0mm = pci_get_device_func(PCI_VENDOR_ID_INTEL,
1019 				    PCI_DEVICE_ID_INTEL_5100_21, 0);
1020 	if (!ch0mm) {
1021 		ret = -ENODEV;
1022 		goto bail_pdev;
1023 	}
1024 
1025 	rc = pci_enable_device(ch0mm);
1026 	if (rc < 0) {
1027 		ret = rc;
1028 		goto bail_ch0;
1029 	}
1030 
1031 	/* device 22, func 0, Channel 1 Memory Map, Error Flag/Mask, etc... */
1032 	ch1mm = pci_get_device_func(PCI_VENDOR_ID_INTEL,
1033 				    PCI_DEVICE_ID_INTEL_5100_22, 0);
1034 	if (!ch1mm) {
1035 		ret = -ENODEV;
1036 		goto bail_disable_ch0;
1037 	}
1038 
1039 	rc = pci_enable_device(ch1mm);
1040 	if (rc < 0) {
1041 		ret = rc;
1042 		goto bail_ch1;
1043 	}
1044 
1045 	layers[0].type = EDAC_MC_LAYER_CHANNEL;
1046 	layers[0].size = 2;
1047 	layers[0].is_virt_csrow = false;
1048 	layers[1].type = EDAC_MC_LAYER_SLOT;
1049 	layers[1].size = ranksperch;
1050 	layers[1].is_virt_csrow = true;
1051 	mci = edac_mc_alloc(0, ARRAY_SIZE(layers), layers,
1052 			    sizeof(*priv));
1053 	if (!mci) {
1054 		ret = -ENOMEM;
1055 		goto bail_disable_ch1;
1056 	}
1057 
1058 
1059 	/* device 19, func 0, Error injection */
1060 	einj = pci_get_device_func(PCI_VENDOR_ID_INTEL,
1061 				    PCI_DEVICE_ID_INTEL_5100_19, 0);
1062 	if (!einj) {
1063 		ret = -ENODEV;
1064 		goto bail_einj;
1065 	}
1066 
1067 	rc = pci_enable_device(einj);
1068 	if (rc < 0) {
1069 		ret = rc;
1070 		goto bail_disable_einj;
1071 	}
1072 
1073 
1074 	mci->pdev = &pdev->dev;
1075 
1076 	priv = mci->pvt_info;
1077 	priv->ranksperchan = ranksperch;
1078 	priv->mc = pdev;
1079 	priv->ch0mm = ch0mm;
1080 	priv->ch1mm = ch1mm;
1081 	priv->einj = einj;
1082 
1083 	INIT_DELAYED_WORK(&(priv->i5100_scrubbing), i5100_refresh_scrubbing);
1084 
1085 	/* If scrubbing was already enabled by the bios, start maintaining it */
1086 	pci_read_config_dword(pdev, I5100_MC, &dw);
1087 	if (i5100_mc_scrben(dw)) {
1088 		priv->scrub_enable = 1;
1089 		schedule_delayed_work(&(priv->i5100_scrubbing),
1090 				      I5100_SCRUB_REFRESH_RATE);
1091 	}
1092 
1093 	i5100_init_dimm_layout(pdev, mci);
1094 	i5100_init_interleaving(pdev, mci);
1095 
1096 	mci->mtype_cap = MEM_FLAG_FB_DDR2;
1097 	mci->edac_ctl_cap = EDAC_FLAG_SECDED;
1098 	mci->edac_cap = EDAC_FLAG_SECDED;
1099 	mci->mod_name = "i5100_edac.c";
1100 	mci->ctl_name = "i5100";
1101 	mci->dev_name = pci_name(pdev);
1102 	mci->ctl_page_to_phys = NULL;
1103 
1104 	mci->edac_check = i5100_check_error;
1105 	mci->set_sdram_scrub_rate = i5100_set_scrub_rate;
1106 	mci->get_sdram_scrub_rate = i5100_get_scrub_rate;
1107 
1108 	priv->inject_channel = 0;
1109 	priv->inject_hlinesel = 0;
1110 	priv->inject_deviceptr1 = 0;
1111 	priv->inject_deviceptr2 = 0;
1112 	priv->inject_eccmask1 = 0;
1113 	priv->inject_eccmask2 = 0;
1114 
1115 	i5100_init_csrows(mci);
1116 
1117 	/* this strange construction seems to be in every driver, dunno why */
1118 	switch (edac_op_state) {
1119 	case EDAC_OPSTATE_POLL:
1120 	case EDAC_OPSTATE_NMI:
1121 		break;
1122 	default:
1123 		edac_op_state = EDAC_OPSTATE_POLL;
1124 		break;
1125 	}
1126 
1127 	if (edac_mc_add_mc(mci)) {
1128 		ret = -ENODEV;
1129 		goto bail_scrub;
1130 	}
1131 
1132 	i5100_setup_debugfs(mci);
1133 
1134 	return ret;
1135 
1136 bail_scrub:
1137 	priv->scrub_enable = 0;
1138 	cancel_delayed_work_sync(&(priv->i5100_scrubbing));
1139 	edac_mc_free(mci);
1140 
1141 bail_disable_einj:
1142 	pci_disable_device(einj);
1143 
1144 bail_einj:
1145 	pci_dev_put(einj);
1146 
1147 bail_disable_ch1:
1148 	pci_disable_device(ch1mm);
1149 
1150 bail_ch1:
1151 	pci_dev_put(ch1mm);
1152 
1153 bail_disable_ch0:
1154 	pci_disable_device(ch0mm);
1155 
1156 bail_ch0:
1157 	pci_dev_put(ch0mm);
1158 
1159 bail_pdev:
1160 	pci_disable_device(pdev);
1161 
1162 bail:
1163 	return ret;
1164 }
1165 
1166 static void i5100_remove_one(struct pci_dev *pdev)
1167 {
1168 	struct mem_ctl_info *mci;
1169 	struct i5100_priv *priv;
1170 
1171 	mci = edac_mc_del_mc(&pdev->dev);
1172 
1173 	if (!mci)
1174 		return;
1175 
1176 	priv = mci->pvt_info;
1177 
1178 	edac_debugfs_remove_recursive(priv->debugfs);
1179 
1180 	priv->scrub_enable = 0;
1181 	cancel_delayed_work_sync(&(priv->i5100_scrubbing));
1182 
1183 	pci_disable_device(pdev);
1184 	pci_disable_device(priv->ch0mm);
1185 	pci_disable_device(priv->ch1mm);
1186 	pci_disable_device(priv->einj);
1187 	pci_dev_put(priv->ch0mm);
1188 	pci_dev_put(priv->ch1mm);
1189 	pci_dev_put(priv->einj);
1190 
1191 	edac_mc_free(mci);
1192 }
1193 
1194 static const struct pci_device_id i5100_pci_tbl[] = {
1195 	/* Device 16, Function 0, Channel 0 Memory Map, Error Flag/Mask, ... */
1196 	{ PCI_DEVICE(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_5100_16) },
1197 	{ 0, }
1198 };
1199 MODULE_DEVICE_TABLE(pci, i5100_pci_tbl);
1200 
1201 static struct pci_driver i5100_driver = {
1202 	.name = KBUILD_BASENAME,
1203 	.probe = i5100_init_one,
1204 	.remove = i5100_remove_one,
1205 	.id_table = i5100_pci_tbl,
1206 };
1207 
1208 static int __init i5100_init(void)
1209 {
1210 	int pci_rc;
1211 
1212 	i5100_debugfs = edac_debugfs_create_dir_at("i5100_edac", NULL);
1213 
1214 	pci_rc = pci_register_driver(&i5100_driver);
1215 	return (pci_rc < 0) ? pci_rc : 0;
1216 }
1217 
1218 static void __exit i5100_exit(void)
1219 {
1220 	edac_debugfs_remove(i5100_debugfs);
1221 
1222 	pci_unregister_driver(&i5100_driver);
1223 }
1224 
1225 module_init(i5100_init);
1226 module_exit(i5100_exit);
1227 
1228 MODULE_LICENSE("GPL");
1229 MODULE_AUTHOR
1230     ("Arthur Jones <ajones@riverbed.com>");
1231 MODULE_DESCRIPTION("MC Driver for Intel I5100 memory controllers");
1232