i5100_edac.c 25.6 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11
/*
 * Intel 5100 Memory Controllers kernel module
 *
 * This file may be distributed under the terms of the
 * GNU General Public License.
 *
 * This module is based on the following document:
 *
 * Intel 5100X Chipset Memory Controller Hub (MCH) - Datasheet
 *      http://download.intel.com/design/chipsets/datashts/318378.pdf
 *
N
Nils Carlson 已提交
12 13
 * The intel 5100 has two independent channels. EDAC core currently
 * can not reflect this configuration so instead the chip-select
L
Lucas De Marchi 已提交
14
 * rows for each respective channel are laid out one after another,
N
Nils Carlson 已提交
15 16
 * the first half belonging to channel 0, the second half belonging
 * to channel 1.
17 18 19 20 21
 *
 * This driver is for DDR2 DIMMs, and it uses chip select to select among the
 * several ranks. However, instead of showing memories as ranks, it outputs
 * them as DIMM's. An internal table creates the association between ranks
 * and DIMM's.
22 23 24 25 26 27 28 29 30 31 32
 */
#include <linux/module.h>
#include <linux/init.h>
#include <linux/pci.h>
#include <linux/pci_ids.h>
#include <linux/edac.h>
#include <linux/delay.h>
#include <linux/mmzone.h>

#include "edac_core.h"

A
Arthur Jones 已提交
33
/* register addresses */
34 35

/* device 16, func 1 */
36
#define I5100_MC		0x40	/* Memory Control Register */
N
Nils Carlson 已提交
37 38
#define 	I5100_MC_SCRBEN_MASK	(1 << 7)
#define 	I5100_MC_SCRBDONE_MASK	(1 << 4)
39 40 41 42 43 44 45 46 47 48 49 50
#define I5100_MS		0x44	/* Memory Status Register */
#define I5100_SPDDATA		0x48	/* Serial Presence Detect Status Reg */
#define I5100_SPDCMD		0x4c	/* Serial Presence Detect Command Reg */
#define I5100_TOLM		0x6c	/* Top of Low Memory */
#define I5100_MIR0		0x80	/* Memory Interleave Range 0 */
#define I5100_MIR1		0x84	/* Memory Interleave Range 1 */
#define I5100_AMIR_0		0x8c	/* Adjusted Memory Interleave Range 0 */
#define I5100_AMIR_1		0x90	/* Adjusted Memory Interleave Range 1 */
#define I5100_FERR_NF_MEM	0xa0	/* MC First Non Fatal Errors */
#define		I5100_FERR_NF_MEM_M16ERR_MASK	(1 << 16)
#define		I5100_FERR_NF_MEM_M15ERR_MASK	(1 << 15)
#define		I5100_FERR_NF_MEM_M14ERR_MASK	(1 << 14)
A
Arthur Jones 已提交
51 52 53 54 55 56
#define		I5100_FERR_NF_MEM_M12ERR_MASK	(1 << 12)
#define		I5100_FERR_NF_MEM_M11ERR_MASK	(1 << 11)
#define		I5100_FERR_NF_MEM_M10ERR_MASK	(1 << 10)
#define		I5100_FERR_NF_MEM_M6ERR_MASK	(1 << 6)
#define		I5100_FERR_NF_MEM_M5ERR_MASK	(1 << 5)
#define		I5100_FERR_NF_MEM_M4ERR_MASK	(1 << 4)
57
#define		I5100_FERR_NF_MEM_M1ERR_MASK	(1 << 1)
58 59 60
#define		I5100_FERR_NF_MEM_ANY_MASK	\
			(I5100_FERR_NF_MEM_M16ERR_MASK | \
			I5100_FERR_NF_MEM_M15ERR_MASK | \
A
Arthur Jones 已提交
61 62 63 64 65 66 67 68
			I5100_FERR_NF_MEM_M14ERR_MASK | \
			I5100_FERR_NF_MEM_M12ERR_MASK | \
			I5100_FERR_NF_MEM_M11ERR_MASK | \
			I5100_FERR_NF_MEM_M10ERR_MASK | \
			I5100_FERR_NF_MEM_M6ERR_MASK | \
			I5100_FERR_NF_MEM_M5ERR_MASK | \
			I5100_FERR_NF_MEM_M4ERR_MASK | \
			I5100_FERR_NF_MEM_M1ERR_MASK)
69
#define	I5100_NERR_NF_MEM	0xa4	/* MC Next Non-Fatal Errors */
A
Arthur Jones 已提交
70
#define I5100_EMASK_MEM		0xa8	/* MC Error Mask Register */
71 72 73 74 75 76 77 78 79 80 81

/* device 21 and 22, func 0 */
#define I5100_MTR_0	0x154	/* Memory Technology Registers 0-3 */
#define I5100_DMIR	0x15c	/* DIMM Interleave Range */
#define	I5100_VALIDLOG	0x18c	/* Valid Log Markers */
#define	I5100_NRECMEMA	0x190	/* Non-Recoverable Memory Error Log Reg A */
#define	I5100_NRECMEMB	0x194	/* Non-Recoverable Memory Error Log Reg B */
#define	I5100_REDMEMA	0x198	/* Recoverable Memory Data Error Log Reg A */
#define	I5100_REDMEMB	0x19c	/* Recoverable Memory Data Error Log Reg B */
#define	I5100_RECMEMA	0x1a0	/* Recoverable Memory Error Log Reg A */
#define	I5100_RECMEMB	0x1a4	/* Recoverable Memory Error Log Reg B */
A
Arthur Jones 已提交
82 83 84 85
#define I5100_MTR_4	0x1b0	/* Memory Technology Registers 4,5 */

/* bit field accessors */

N
Nils Carlson 已提交
86 87 88 89 90
static inline u32 i5100_mc_scrben(u32 mc)
{
	return mc >> 7 & 1;
}

A
Arthur Jones 已提交
91 92 93 94 95
static inline u32 i5100_mc_errdeten(u32 mc)
{
	return mc >> 5 & 1;
}

N
Nils Carlson 已提交
96 97 98 99 100
static inline u32 i5100_mc_scrbdone(u32 mc)
{
	return mc >> 4 & 1;
}

A
Arthur Jones 已提交
101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286
static inline u16 i5100_spddata_rdo(u16 a)
{
	return a >> 15 & 1;
}

static inline u16 i5100_spddata_sbe(u16 a)
{
	return a >> 13 & 1;
}

static inline u16 i5100_spddata_busy(u16 a)
{
	return a >> 12 & 1;
}

static inline u16 i5100_spddata_data(u16 a)
{
	return a & ((1 << 8) - 1);
}

static inline u32 i5100_spdcmd_create(u32 dti, u32 ckovrd, u32 sa, u32 ba,
				      u32 data, u32 cmd)
{
	return	((dti & ((1 << 4) - 1))  << 28) |
		((ckovrd & 1)            << 27) |
		((sa & ((1 << 3) - 1))   << 24) |
		((ba & ((1 << 8) - 1))   << 16) |
		((data & ((1 << 8) - 1)) <<  8) |
		(cmd & 1);
}

static inline u16 i5100_tolm_tolm(u16 a)
{
	return a >> 12 & ((1 << 4) - 1);
}

static inline u16 i5100_mir_limit(u16 a)
{
	return a >> 4 & ((1 << 12) - 1);
}

static inline u16 i5100_mir_way1(u16 a)
{
	return a >> 1 & 1;
}

static inline u16 i5100_mir_way0(u16 a)
{
	return a & 1;
}

static inline u32 i5100_ferr_nf_mem_chan_indx(u32 a)
{
	return a >> 28 & 1;
}

static inline u32 i5100_ferr_nf_mem_any(u32 a)
{
	return a & I5100_FERR_NF_MEM_ANY_MASK;
}

static inline u32 i5100_nerr_nf_mem_any(u32 a)
{
	return i5100_ferr_nf_mem_any(a);
}

static inline u32 i5100_dmir_limit(u32 a)
{
	return a >> 16 & ((1 << 11) - 1);
}

static inline u32 i5100_dmir_rank(u32 a, u32 i)
{
	return a >> (4 * i) & ((1 << 2) - 1);
}

static inline u16 i5100_mtr_present(u16 a)
{
	return a >> 10 & 1;
}

static inline u16 i5100_mtr_ethrottle(u16 a)
{
	return a >> 9 & 1;
}

static inline u16 i5100_mtr_width(u16 a)
{
	return a >> 8 & 1;
}

static inline u16 i5100_mtr_numbank(u16 a)
{
	return a >> 6 & 1;
}

static inline u16 i5100_mtr_numrow(u16 a)
{
	return a >> 2 & ((1 << 2) - 1);
}

static inline u16 i5100_mtr_numcol(u16 a)
{
	return a & ((1 << 2) - 1);
}


static inline u32 i5100_validlog_redmemvalid(u32 a)
{
	return a >> 2 & 1;
}

static inline u32 i5100_validlog_recmemvalid(u32 a)
{
	return a >> 1 & 1;
}

static inline u32 i5100_validlog_nrecmemvalid(u32 a)
{
	return a & 1;
}

static inline u32 i5100_nrecmema_merr(u32 a)
{
	return a >> 15 & ((1 << 5) - 1);
}

static inline u32 i5100_nrecmema_bank(u32 a)
{
	return a >> 12 & ((1 << 3) - 1);
}

static inline u32 i5100_nrecmema_rank(u32 a)
{
	return a >>  8 & ((1 << 3) - 1);
}

static inline u32 i5100_nrecmema_dm_buf_id(u32 a)
{
	return a & ((1 << 8) - 1);
}

static inline u32 i5100_nrecmemb_cas(u32 a)
{
	return a >> 16 & ((1 << 13) - 1);
}

static inline u32 i5100_nrecmemb_ras(u32 a)
{
	return a & ((1 << 16) - 1);
}

static inline u32 i5100_redmemb_ecc_locator(u32 a)
{
	return a & ((1 << 18) - 1);
}

static inline u32 i5100_recmema_merr(u32 a)
{
	return i5100_nrecmema_merr(a);
}

static inline u32 i5100_recmema_bank(u32 a)
{
	return i5100_nrecmema_bank(a);
}

static inline u32 i5100_recmema_rank(u32 a)
{
	return i5100_nrecmema_rank(a);
}

static inline u32 i5100_recmema_dm_buf_id(u32 a)
{
	return i5100_nrecmema_dm_buf_id(a);
}

static inline u32 i5100_recmemb_cas(u32 a)
{
	return i5100_nrecmemb_cas(a);
}

static inline u32 i5100_recmemb_ras(u32 a)
{
	return i5100_nrecmemb_ras(a);
}
287 288

/* some generic limits */
289 290
#define I5100_MAX_RANKS_PER_CHAN	6
#define I5100_CHANNELS			    2
291 292
#define I5100_MAX_RANKS_PER_DIMM	4
#define I5100_DIMM_ADDR_LINES		(6 - 3)	/* 64 bits / 8 bits per byte */
293
#define I5100_MAX_DIMM_SLOTS_PER_CHAN	4
294 295
#define I5100_MAX_RANK_INTERLEAVE	4
#define I5100_MAX_DMIRS			5
N
Nils Carlson 已提交
296
#define I5100_SCRUB_REFRESH_RATE	(5 * 60 * HZ)
297 298 299

struct i5100_priv {
	/* ranks on each dimm -- 0 maps to not present -- obtained via SPD */
300
	int dimm_numrank[I5100_CHANNELS][I5100_MAX_DIMM_SLOTS_PER_CHAN];
301 302 303 304

	/*
	 * mainboard chip select map -- maps i5100 chip selects to
	 * DIMM slot chip selects.  In the case of only 4 ranks per
305 306
	 * channel, the mapping is fairly obvious but not unique.
	 * we map -1 -> NC and assume both channels use the same
307 308 309
	 * map...
	 *
	 */
310
	int dimm_csmap[I5100_MAX_DIMM_SLOTS_PER_CHAN][I5100_MAX_RANKS_PER_DIMM];
311 312 313 314 315

	/* memory interleave range */
	struct {
		u64	 limit;
		unsigned way[2];
316
	} mir[I5100_CHANNELS];
317 318

	/* adjusted memory interleave range register */
319
	unsigned amir[I5100_CHANNELS];
320 321 322 323 324

	/* dimm interleave range */
	struct {
		unsigned rank[I5100_MAX_RANK_INTERLEAVE];
		u64	 limit;
325
	} dmir[I5100_CHANNELS][I5100_MAX_DMIRS];
326 327 328 329 330 331 332 333 334

	/* memory technology registers... */
	struct {
		unsigned present;	/* 0 or 1 */
		unsigned ethrottle;	/* 0 or 1 */
		unsigned width;		/* 4 or 8 bits  */
		unsigned numbank;	/* 2 or 3 lines */
		unsigned numrow;	/* 13 .. 16 lines */
		unsigned numcol;	/* 11 .. 12 lines */
335
	} mtr[I5100_CHANNELS][I5100_MAX_RANKS_PER_CHAN];
336 337

	u64 tolm;		/* top of low memory in bytes */
338
	unsigned ranksperchan;	/* number of ranks per channel */
339 340 341 342

	struct pci_dev *mc;	/* device 16 func 1 */
	struct pci_dev *ch0mm;	/* device 21 func 0 */
	struct pci_dev *ch1mm;	/* device 22 func 0 */
N
Nils Carlson 已提交
343 344 345

	struct delayed_work i5100_scrubbing;
	int scrub_enable;
346 347
};

348
/* map a rank/chan to a slot number on the mainboard */
349
static int i5100_rank_to_slot(const struct mem_ctl_info *mci,
350
			      int chan, int rank)
351 352 353 354
{
	const struct i5100_priv *priv = mci->pvt_info;
	int i;

355
	for (i = 0; i < I5100_MAX_DIMM_SLOTS_PER_CHAN; i++) {
356
		int j;
357
		const int numrank = priv->dimm_numrank[chan][i];
358 359 360

		for (j = 0; j < numrank; j++)
			if (priv->dimm_csmap[i][j] == rank)
361
				return i * 2 + chan;
362 363 364 365 366 367 368
	}

	return -1;
}

static const char *i5100_err_msg(unsigned err)
{
A
Arthur Jones 已提交
369
	static const char *merrs[] = {
370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401
		"unknown", /* 0 */
		"uncorrectable data ECC on replay", /* 1 */
		"unknown", /* 2 */
		"unknown", /* 3 */
		"aliased uncorrectable demand data ECC", /* 4 */
		"aliased uncorrectable spare-copy data ECC", /* 5 */
		"aliased uncorrectable patrol data ECC", /* 6 */
		"unknown", /* 7 */
		"unknown", /* 8 */
		"unknown", /* 9 */
		"non-aliased uncorrectable demand data ECC", /* 10 */
		"non-aliased uncorrectable spare-copy data ECC", /* 11 */
		"non-aliased uncorrectable patrol data ECC", /* 12 */
		"unknown", /* 13 */
		"correctable demand data ECC", /* 14 */
		"correctable spare-copy data ECC", /* 15 */
		"correctable patrol data ECC", /* 16 */
		"unknown", /* 17 */
		"SPD protocol error", /* 18 */
		"unknown", /* 19 */
		"spare copy initiated", /* 20 */
		"spare copy completed", /* 21 */
	};
	unsigned i;

	for (i = 0; i < ARRAY_SIZE(merrs); i++)
		if (1 << i & err)
			return merrs[i];

	return "none";
}

402
/* convert csrow index into a rank (per channel -- 0..5) */
403 404 405 406
static int i5100_csrow_to_rank(const struct mem_ctl_info *mci, int csrow)
{
	const struct i5100_priv *priv = mci->pvt_info;

407
	return csrow % priv->ranksperchan;
408 409
}

410 411
/* convert csrow index into a channel (0..1) */
static int i5100_csrow_to_chan(const struct mem_ctl_info *mci, int csrow)
412 413 414
{
	const struct i5100_priv *priv = mci->pvt_info;

415
	return csrow / priv->ranksperchan;
416 417 418
}

static void i5100_handle_ce(struct mem_ctl_info *mci,
419
			    int chan,
420 421 422 423 424 425 426
			    unsigned bank,
			    unsigned rank,
			    unsigned long syndrome,
			    unsigned cas,
			    unsigned ras,
			    const char *msg)
{
427
	char detail[80];
428

429 430 431 432
	/* Form out message */
	snprintf(detail, sizeof(detail),
		 "bank %u, cas %u, ras %u\n",
		 bank, cas, ras);
433

434 435 436 437
	edac_mc_handle_error(HW_EVENT_ERR_CORRECTED, mci,
			     0, 0, syndrome,
			     chan, rank, -1,
			     msg, detail, NULL);
438 439 440
}

static void i5100_handle_ue(struct mem_ctl_info *mci,
441
			    int chan,
442 443 444 445 446 447 448
			    unsigned bank,
			    unsigned rank,
			    unsigned long syndrome,
			    unsigned cas,
			    unsigned ras,
			    const char *msg)
{
449
	char detail[80];
450

451 452 453 454
	/* Form out message */
	snprintf(detail, sizeof(detail),
		 "bank %u, cas %u, ras %u\n",
		 bank, cas, ras);
455

456 457 458 459
	edac_mc_handle_error(HW_EVENT_ERR_UNCORRECTED, mci,
			     0, 0, syndrome,
			     chan, rank, -1,
			     msg, detail, NULL);
460 461
}

462
static void i5100_read_log(struct mem_ctl_info *mci, int chan,
463 464 465
			   u32 ferr, u32 nerr)
{
	struct i5100_priv *priv = mci->pvt_info;
466
	struct pci_dev *pdev = (chan) ? priv->ch1mm : priv->ch0mm;
467 468 469 470 471 472 473 474 475 476 477 478
	u32 dw;
	u32 dw2;
	unsigned syndrome = 0;
	unsigned ecc_loc = 0;
	unsigned merr;
	unsigned bank;
	unsigned rank;
	unsigned cas;
	unsigned ras;

	pci_read_config_dword(pdev, I5100_VALIDLOG, &dw);

A
Arthur Jones 已提交
479
	if (i5100_validlog_redmemvalid(dw)) {
480
		pci_read_config_dword(pdev, I5100_REDMEMA, &dw2);
A
Arthur Jones 已提交
481
		syndrome = dw2;
482
		pci_read_config_dword(pdev, I5100_REDMEMB, &dw2);
A
Arthur Jones 已提交
483
		ecc_loc = i5100_redmemb_ecc_locator(dw2);
484 485
	}

A
Arthur Jones 已提交
486
	if (i5100_validlog_recmemvalid(dw)) {
487 488 489
		const char *msg;

		pci_read_config_dword(pdev, I5100_RECMEMA, &dw2);
A
Arthur Jones 已提交
490 491 492
		merr = i5100_recmema_merr(dw2);
		bank = i5100_recmema_bank(dw2);
		rank = i5100_recmema_rank(dw2);
493 494

		pci_read_config_dword(pdev, I5100_RECMEMB, &dw2);
A
Arthur Jones 已提交
495 496
		cas = i5100_recmemb_cas(dw2);
		ras = i5100_recmemb_ras(dw2);
497 498 499 500 501 502 503 504

		/* FIXME:  not really sure if this is what merr is...
		 */
		if (!merr)
			msg = i5100_err_msg(ferr);
		else
			msg = i5100_err_msg(nerr);

505
		i5100_handle_ce(mci, chan, bank, rank, syndrome, cas, ras, msg);
506 507
	}

A
Arthur Jones 已提交
508
	if (i5100_validlog_nrecmemvalid(dw)) {
509 510 511
		const char *msg;

		pci_read_config_dword(pdev, I5100_NRECMEMA, &dw2);
A
Arthur Jones 已提交
512 513 514
		merr = i5100_nrecmema_merr(dw2);
		bank = i5100_nrecmema_bank(dw2);
		rank = i5100_nrecmema_rank(dw2);
515 516

		pci_read_config_dword(pdev, I5100_NRECMEMB, &dw2);
A
Arthur Jones 已提交
517 518
		cas = i5100_nrecmemb_cas(dw2);
		ras = i5100_nrecmemb_ras(dw2);
519 520 521 522 523 524 525 526

		/* FIXME:  not really sure if this is what merr is...
		 */
		if (!merr)
			msg = i5100_err_msg(ferr);
		else
			msg = i5100_err_msg(nerr);

527
		i5100_handle_ue(mci, chan, bank, rank, syndrome, cas, ras, msg);
528 529 530 531 532 533 534 535
	}

	pci_write_config_dword(pdev, I5100_VALIDLOG, dw);
}

static void i5100_check_error(struct mem_ctl_info *mci)
{
	struct i5100_priv *priv = mci->pvt_info;
536
	u32 dw, dw2;
537 538

	pci_read_config_dword(priv->mc, I5100_FERR_NF_MEM, &dw);
A
Arthur Jones 已提交
539
	if (i5100_ferr_nf_mem_any(dw)) {
540 541 542

		pci_read_config_dword(priv->mc, I5100_NERR_NF_MEM, &dw2);

A
Arthur Jones 已提交
543 544 545
		i5100_read_log(mci, i5100_ferr_nf_mem_chan_indx(dw),
			       i5100_ferr_nf_mem_any(dw),
			       i5100_nerr_nf_mem_any(dw2));
546 547

		pci_write_config_dword(priv->mc, I5100_NERR_NF_MEM, dw2);
548
	}
549
	pci_write_config_dword(priv->mc, I5100_FERR_NF_MEM, dw);
550 551
}

N
Nils Carlson 已提交
552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586
/* The i5100 chipset will scrub the entire memory once, then
 * set a done bit. Continuous scrubbing is achieved by enqueing
 * delayed work to a workqueue, checking every few minutes if
 * the scrubbing has completed and if so reinitiating it.
 */

static void i5100_refresh_scrubbing(struct work_struct *work)
{
	struct delayed_work *i5100_scrubbing = container_of(work,
							    struct delayed_work,
							    work);
	struct i5100_priv *priv = container_of(i5100_scrubbing,
					       struct i5100_priv,
					       i5100_scrubbing);
	u32 dw;

	pci_read_config_dword(priv->mc, I5100_MC, &dw);

	if (priv->scrub_enable) {

		pci_read_config_dword(priv->mc, I5100_MC, &dw);

		if (i5100_mc_scrbdone(dw)) {
			dw |= I5100_MC_SCRBEN_MASK;
			pci_write_config_dword(priv->mc, I5100_MC, dw);
			pci_read_config_dword(priv->mc, I5100_MC, &dw);
		}

		schedule_delayed_work(&(priv->i5100_scrubbing),
				      I5100_SCRUB_REFRESH_RATE);
	}
}
/*
 * The bandwidth is based on experimentation, feel free to refine it.
 */
587
static int i5100_set_scrub_rate(struct mem_ctl_info *mci, u32 bandwidth)
N
Nils Carlson 已提交
588 589 590 591 592
{
	struct i5100_priv *priv = mci->pvt_info;
	u32 dw;

	pci_read_config_dword(priv->mc, I5100_MC, &dw);
593
	if (bandwidth) {
N
Nils Carlson 已提交
594 595 596 597 598 599 600 601 602 603 604 605 606
		priv->scrub_enable = 1;
		dw |= I5100_MC_SCRBEN_MASK;
		schedule_delayed_work(&(priv->i5100_scrubbing),
				      I5100_SCRUB_REFRESH_RATE);
	} else {
		priv->scrub_enable = 0;
		dw &= ~I5100_MC_SCRBEN_MASK;
		cancel_delayed_work(&(priv->i5100_scrubbing));
	}
	pci_write_config_dword(priv->mc, I5100_MC, dw);

	pci_read_config_dword(priv->mc, I5100_MC, &dw);

607
	bandwidth = 5900000 * i5100_mc_scrben(dw);
N
Nils Carlson 已提交
608

609
	return bandwidth;
N
Nils Carlson 已提交
610 611
}

612
static int i5100_get_scrub_rate(struct mem_ctl_info *mci)
N
Nils Carlson 已提交
613 614 615 616 617 618
{
	struct i5100_priv *priv = mci->pvt_info;
	u32 dw;

	pci_read_config_dword(priv->mc, I5100_MC, &dw);

619
	return 5900000 * i5100_mc_scrben(dw);
N
Nils Carlson 已提交
620 621
}

622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644
static struct pci_dev *pci_get_device_func(unsigned vendor,
					   unsigned device,
					   unsigned func)
{
	struct pci_dev *ret = NULL;

	while (1) {
		ret = pci_get_device(vendor, device, ret);

		if (!ret)
			break;

		if (PCI_FUNC(ret->devfn) == func)
			break;
	}

	return ret;
}

static unsigned long __devinit i5100_npages(struct mem_ctl_info *mci,
					    int csrow)
{
	struct i5100_priv *priv = mci->pvt_info;
645 646
	const unsigned chan_rank = i5100_csrow_to_rank(mci, csrow);
	const unsigned chan = i5100_csrow_to_chan(mci, csrow);
647 648 649
	unsigned addr_lines;

	/* dimm present? */
650
	if (!priv->mtr[chan][chan_rank].present)
651 652 653 654
		return 0ULL;

	addr_lines =
		I5100_DIMM_ADDR_LINES +
655 656 657
		priv->mtr[chan][chan_rank].numcol +
		priv->mtr[chan][chan_rank].numrow +
		priv->mtr[chan][chan_rank].numbank;
658 659 660 661 662 663 664 665 666 667 668

	return (unsigned long)
		((unsigned long long) (1ULL << addr_lines) / PAGE_SIZE);
}

static void __devinit i5100_init_mtr(struct mem_ctl_info *mci)
{
	struct i5100_priv *priv = mci->pvt_info;
	struct pci_dev *mms[2] = { priv->ch0mm, priv->ch1mm };
	int i;

669
	for (i = 0; i < I5100_CHANNELS; i++) {
670 671 672
		int j;
		struct pci_dev *pdev = mms[i];

673
		for (j = 0; j < I5100_MAX_RANKS_PER_CHAN; j++) {
674 675 676 677 678 679 680
			const unsigned addr =
				(j < 4) ? I5100_MTR_0 + j * 2 :
					  I5100_MTR_4 + (j - 4) * 2;
			u16 w;

			pci_read_config_word(pdev, addr, &w);

A
Arthur Jones 已提交
681 682 683 684 685 686
			priv->mtr[i][j].present = i5100_mtr_present(w);
			priv->mtr[i][j].ethrottle = i5100_mtr_ethrottle(w);
			priv->mtr[i][j].width = 4 + 4 * i5100_mtr_width(w);
			priv->mtr[i][j].numbank = 2 + i5100_mtr_numbank(w);
			priv->mtr[i][j].numrow = 13 + i5100_mtr_numrow(w);
			priv->mtr[i][j].numcol = 10 + i5100_mtr_numcol(w);
687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702
		}
	}
}

/*
 * FIXME: make this into a real i2c adapter (so that dimm-decode
 * will work)?
 */
static int i5100_read_spd_byte(const struct mem_ctl_info *mci,
			       u8 ch, u8 slot, u8 addr, u8 *byte)
{
	struct i5100_priv *priv = mci->pvt_info;
	u16 w;
	unsigned long et;

	pci_read_config_word(priv->mc, I5100_SPDDATA, &w);
A
Arthur Jones 已提交
703
	if (i5100_spddata_busy(w))
704 705
		return -1;

A
Arthur Jones 已提交
706 707 708
	pci_write_config_dword(priv->mc, I5100_SPDCMD,
			       i5100_spdcmd_create(0xa, 1, ch * 4 + slot, addr,
						   0, 0));
709 710 711 712 713 714

	/* wait up to 100ms */
	et = jiffies + HZ / 10;
	udelay(100);
	while (1) {
		pci_read_config_word(priv->mc, I5100_SPDDATA, &w);
A
Arthur Jones 已提交
715
		if (!i5100_spddata_busy(w))
716 717 718 719
			break;
		udelay(100);
	}

A
Arthur Jones 已提交
720
	if (!i5100_spddata_rdo(w) || i5100_spddata_sbe(w))
721 722
		return -1;

A
Arthur Jones 已提交
723
	*byte = i5100_spddata_data(w);
724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739

	return 0;
}

/*
 * fill dimm chip select map
 *
 * FIXME:
 *   o not the only way to may chip selects to dimm slots
 *   o investigate if there is some way to obtain this map from the bios
 */
static void __devinit i5100_init_dimm_csmap(struct mem_ctl_info *mci)
{
	struct i5100_priv *priv = mci->pvt_info;
	int i;

740
	for (i = 0; i < I5100_MAX_DIMM_SLOTS_PER_CHAN; i++) {
741 742 743 744 745 746 747
		int j;

		for (j = 0; j < I5100_MAX_RANKS_PER_DIMM; j++)
			priv->dimm_csmap[i][j] = -1; /* default NC */
	}

	/* only 2 chip selects per slot... */
N
Nils Carlson 已提交
748 749 750 751 752 753 754 755 756 757 758 759 760 761 762
	if (priv->ranksperchan == 4) {
		priv->dimm_csmap[0][0] = 0;
		priv->dimm_csmap[0][1] = 3;
		priv->dimm_csmap[1][0] = 1;
		priv->dimm_csmap[1][1] = 2;
		priv->dimm_csmap[2][0] = 2;
		priv->dimm_csmap[3][0] = 3;
	} else {
		priv->dimm_csmap[0][0] = 0;
		priv->dimm_csmap[0][1] = 1;
		priv->dimm_csmap[1][0] = 2;
		priv->dimm_csmap[1][1] = 3;
		priv->dimm_csmap[2][0] = 4;
		priv->dimm_csmap[2][1] = 5;
	}
763 764 765 766 767 768 769 770
}

static void __devinit i5100_init_dimm_layout(struct pci_dev *pdev,
					     struct mem_ctl_info *mci)
{
	struct i5100_priv *priv = mci->pvt_info;
	int i;

771
	for (i = 0; i < I5100_CHANNELS; i++) {
772 773
		int j;

774
		for (j = 0; j < I5100_MAX_DIMM_SLOTS_PER_CHAN; j++) {
775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796
			u8 rank;

			if (i5100_read_spd_byte(mci, i, j, 5, &rank) < 0)
				priv->dimm_numrank[i][j] = 0;
			else
				priv->dimm_numrank[i][j] = (rank & 3) + 1;
		}
	}

	i5100_init_dimm_csmap(mci);
}

static void __devinit i5100_init_interleaving(struct pci_dev *pdev,
					      struct mem_ctl_info *mci)
{
	u16 w;
	u32 dw;
	struct i5100_priv *priv = mci->pvt_info;
	struct pci_dev *mms[2] = { priv->ch0mm, priv->ch1mm };
	int i;

	pci_read_config_word(pdev, I5100_TOLM, &w);
A
Arthur Jones 已提交
797
	priv->tolm = (u64) i5100_tolm_tolm(w) * 256 * 1024 * 1024;
798 799

	pci_read_config_word(pdev, I5100_MIR0, &w);
A
Arthur Jones 已提交
800 801 802
	priv->mir[0].limit = (u64) i5100_mir_limit(w) << 28;
	priv->mir[0].way[1] = i5100_mir_way1(w);
	priv->mir[0].way[0] = i5100_mir_way0(w);
803 804

	pci_read_config_word(pdev, I5100_MIR1, &w);
A
Arthur Jones 已提交
805 806 807
	priv->mir[1].limit = (u64) i5100_mir_limit(w) << 28;
	priv->mir[1].way[1] = i5100_mir_way1(w);
	priv->mir[1].way[0] = i5100_mir_way0(w);
808 809 810 811 812 813

	pci_read_config_word(pdev, I5100_AMIR_0, &w);
	priv->amir[0] = w;
	pci_read_config_word(pdev, I5100_AMIR_1, &w);
	priv->amir[1] = w;

814
	for (i = 0; i < I5100_CHANNELS; i++) {
815 816 817 818 819 820 821 822
		int j;

		for (j = 0; j < 5; j++) {
			int k;

			pci_read_config_dword(mms[i], I5100_DMIR + j * 4, &dw);

			priv->dmir[i][j].limit =
A
Arthur Jones 已提交
823
				(u64) i5100_dmir_limit(dw) << 28;
824 825
			for (k = 0; k < I5100_MAX_RANKS_PER_DIMM; k++)
				priv->dmir[i][j].rank[k] =
A
Arthur Jones 已提交
826
					i5100_dmir_rank(dw, k);
827 828 829 830 831 832 833 834 835 836 837
		}
	}

	i5100_init_mtr(mci);
}

static void __devinit i5100_init_csrows(struct mem_ctl_info *mci)
{
	int i;
	struct i5100_priv *priv = mci->pvt_info;

838 839
	for (i = 0; i < mci->tot_dimms; i++) {
		struct dimm_info *dimm;
840
		const unsigned long npages = i5100_npages(mci, i);
841
		const unsigned chan = i5100_csrow_to_chan(mci, i);
842 843 844 845 846
		const unsigned rank = i5100_csrow_to_rank(mci, i);

		if (!npages)
			continue;

847 848
		dimm = EDAC_DIMM_PTR(mci->layers, mci->dimms, mci->n_layers,
			       chan, rank, 0);
849

850 851 852 853
		dimm->nr_pages = npages;
		if (npages) {
			dimm->grain = 32;
			dimm->dtype = (priv->mtr[chan][rank].width == 4) ?
854
					DEV_X4 : DEV_X8;
855 856 857 858 859 860
			dimm->mtype = MEM_RDDR2;
			dimm->edac_mode = EDAC_SECDED;
			snprintf(dimm->label, sizeof(dimm->label),
				"DIMM%u",
				i5100_rank_to_slot(mci, chan, rank));
		}
861

862 863
		edac_dbg(2, "dimm channel %d, rank %d, size %ld\n",
			 chan, rank, (long)PAGES_TO_MiB(npages));
864 865 866 867 868 869 870 871
	}
}

static int __devinit i5100_init_one(struct pci_dev *pdev,
				    const struct pci_device_id *id)
{
	int rc;
	struct mem_ctl_info *mci;
872
	struct edac_mc_layer layers[2];
873 874 875 876 877 878 879 880 881 882 883 884 885 886 887
	struct i5100_priv *priv;
	struct pci_dev *ch0mm, *ch1mm;
	int ret = 0;
	u32 dw;
	int ranksperch;

	if (PCI_FUNC(pdev->devfn) != 1)
		return -ENODEV;

	rc = pci_enable_device(pdev);
	if (rc < 0) {
		ret = rc;
		goto bail;
	}

888 889
	/* ECC enabled? */
	pci_read_config_dword(pdev, I5100_MC, &dw);
A
Arthur Jones 已提交
890
	if (!i5100_mc_errdeten(dw)) {
891 892
		printk(KERN_INFO "i5100_edac: ECC not enabled.\n");
		ret = -ENODEV;
A
Arthur Jones 已提交
893
		goto bail_pdev;
894 895
	}

896 897 898 899
	/* figure out how many ranks, from strapped state of 48GB_Mode input */
	pci_read_config_dword(pdev, I5100_MS, &dw);
	ranksperch = !!(dw & (1 << 8)) * 2 + 4;

A
Arthur Jones 已提交
900 901 902 903 904
	/* enable error reporting... */
	pci_read_config_dword(pdev, I5100_EMASK_MEM, &dw);
	dw &= ~I5100_FERR_NF_MEM_ANY_MASK;
	pci_write_config_dword(pdev, I5100_EMASK_MEM, dw);

905 906 907
	/* device 21, func 0, Channel 0 Memory Map, Error Flag/Mask, etc... */
	ch0mm = pci_get_device_func(PCI_VENDOR_ID_INTEL,
				    PCI_DEVICE_ID_INTEL_5100_21, 0);
A
Arthur Jones 已提交
908 909 910 911
	if (!ch0mm) {
		ret = -ENODEV;
		goto bail_pdev;
	}
912 913 914 915 916 917 918 919 920 921 922 923

	rc = pci_enable_device(ch0mm);
	if (rc < 0) {
		ret = rc;
		goto bail_ch0;
	}

	/* device 22, func 0, Channel 1 Memory Map, Error Flag/Mask, etc... */
	ch1mm = pci_get_device_func(PCI_VENDOR_ID_INTEL,
				    PCI_DEVICE_ID_INTEL_5100_22, 0);
	if (!ch1mm) {
		ret = -ENODEV;
A
Arthur Jones 已提交
924
		goto bail_disable_ch0;
925 926 927 928 929 930 931 932
	}

	rc = pci_enable_device(ch1mm);
	if (rc < 0) {
		ret = rc;
		goto bail_ch1;
	}

933 934 935 936 937 938
	layers[0].type = EDAC_MC_LAYER_CHANNEL;
	layers[0].size = 2;
	layers[0].is_virt_csrow = false;
	layers[1].type = EDAC_MC_LAYER_SLOT;
	layers[1].size = ranksperch;
	layers[1].is_virt_csrow = true;
939
	mci = edac_mc_alloc(0, ARRAY_SIZE(layers), layers,
940
			    sizeof(*priv));
941 942
	if (!mci) {
		ret = -ENOMEM;
A
Arthur Jones 已提交
943
		goto bail_disable_ch1;
944 945
	}

946
	mci->pdev = &pdev->dev;
947 948

	priv = mci->pvt_info;
949
	priv->ranksperchan = ranksperch;
950 951 952 953
	priv->mc = pdev;
	priv->ch0mm = ch0mm;
	priv->ch1mm = ch1mm;

N
Nils Carlson 已提交
954 955 956 957 958 959 960 961 962 963
	INIT_DELAYED_WORK(&(priv->i5100_scrubbing), i5100_refresh_scrubbing);

	/* If scrubbing was already enabled by the bios, start maintaining it */
	pci_read_config_dword(pdev, I5100_MC, &dw);
	if (i5100_mc_scrben(dw)) {
		priv->scrub_enable = 1;
		schedule_delayed_work(&(priv->i5100_scrubbing),
				      I5100_SCRUB_REFRESH_RATE);
	}

964 965 966 967 968 969 970 971 972 973
	i5100_init_dimm_layout(pdev, mci);
	i5100_init_interleaving(pdev, mci);

	mci->mtype_cap = MEM_FLAG_FB_DDR2;
	mci->edac_ctl_cap = EDAC_FLAG_SECDED;
	mci->edac_cap = EDAC_FLAG_SECDED;
	mci->mod_name = "i5100_edac.c";
	mci->mod_ver = "not versioned";
	mci->ctl_name = "i5100";
	mci->dev_name = pci_name(pdev);
A
Arthur Jones 已提交
974
	mci->ctl_page_to_phys = NULL;
975 976

	mci->edac_check = i5100_check_error;
N
Nils Carlson 已提交
977 978
	mci->set_sdram_scrub_rate = i5100_set_scrub_rate;
	mci->get_sdram_scrub_rate = i5100_get_scrub_rate;
979 980 981 982 983 984 985 986 987 988 989 990 991 992 993

	i5100_init_csrows(mci);

	/* this strange construction seems to be in every driver, dunno why */
	switch (edac_op_state) {
	case EDAC_OPSTATE_POLL:
	case EDAC_OPSTATE_NMI:
		break;
	default:
		edac_op_state = EDAC_OPSTATE_POLL;
		break;
	}

	if (edac_mc_add_mc(mci)) {
		ret = -ENODEV;
N
Nils Carlson 已提交
994
		goto bail_scrub;
995 996
	}

A
Arthur Jones 已提交
997
	return ret;
998

N
Nils Carlson 已提交
999 1000 1001
bail_scrub:
	priv->scrub_enable = 0;
	cancel_delayed_work_sync(&(priv->i5100_scrubbing));
1002 1003
	edac_mc_free(mci);

A
Arthur Jones 已提交
1004 1005 1006
bail_disable_ch1:
	pci_disable_device(ch1mm);

1007 1008 1009
bail_ch1:
	pci_dev_put(ch1mm);

A
Arthur Jones 已提交
1010 1011 1012
bail_disable_ch0:
	pci_disable_device(ch0mm);

1013 1014 1015
bail_ch0:
	pci_dev_put(ch0mm);

A
Arthur Jones 已提交
1016 1017 1018
bail_pdev:
	pci_disable_device(pdev);

1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033
bail:
	return ret;
}

static void __devexit i5100_remove_one(struct pci_dev *pdev)
{
	struct mem_ctl_info *mci;
	struct i5100_priv *priv;

	mci = edac_mc_del_mc(&pdev->dev);

	if (!mci)
		return;

	priv = mci->pvt_info;
N
Nils Carlson 已提交
1034 1035 1036 1037

	priv->scrub_enable = 0;
	cancel_delayed_work_sync(&(priv->i5100_scrubbing));

A
Arthur Jones 已提交
1038 1039 1040
	pci_disable_device(pdev);
	pci_disable_device(priv->ch0mm);
	pci_disable_device(priv->ch1mm);
1041 1042 1043 1044 1045 1046
	pci_dev_put(priv->ch0mm);
	pci_dev_put(priv->ch1mm);

	edac_mc_free(mci);
}

1047
static DEFINE_PCI_DEVICE_TABLE(i5100_pci_tbl) = {
1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081
	/* Device 16, Function 0, Channel 0 Memory Map, Error Flag/Mask, ... */
	{ PCI_DEVICE(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_5100_16) },
	{ 0, }
};
MODULE_DEVICE_TABLE(pci, i5100_pci_tbl);

static struct pci_driver i5100_driver = {
	.name = KBUILD_BASENAME,
	.probe = i5100_init_one,
	.remove = __devexit_p(i5100_remove_one),
	.id_table = i5100_pci_tbl,
};

static int __init i5100_init(void)
{
	int pci_rc;

	pci_rc = pci_register_driver(&i5100_driver);

	return (pci_rc < 0) ? pci_rc : 0;
}

static void __exit i5100_exit(void)
{
	pci_unregister_driver(&i5100_driver);
}

module_init(i5100_init);
module_exit(i5100_exit);

MODULE_LICENSE("GPL");
MODULE_AUTHOR
    ("Arthur Jones <ajones@riverbed.com>");
MODULE_DESCRIPTION("MC Driver for Intel I5100 memory controllers");