xref: /linux/drivers/edac/i3000_edac.c (revision cdd38c5f1ce4398ec58fec95904b75824daab7b5)
1 /*
2  * Intel 3000/3010 Memory Controller kernel module
3  * Copyright (C) 2007 Akamai Technologies, Inc.
4  * Shamelessly copied from:
5  * 	Intel D82875P Memory Controller kernel module
6  * 	(C) 2003 Linux Networx (http://lnxi.com)
7  *
8  * This file may be distributed under the terms of the
9  * GNU General Public License.
10  */
11 
12 #include <linux/module.h>
13 #include <linux/init.h>
14 #include <linux/pci.h>
15 #include <linux/pci_ids.h>
16 #include <linux/edac.h>
17 #include "edac_module.h"
18 
19 #define EDAC_MOD_STR		"i3000_edac"
20 
21 #define I3000_RANKS		8
22 #define I3000_RANKS_PER_CHANNEL	4
23 #define I3000_CHANNELS		2
24 
25 /* Intel 3000 register addresses - device 0 function 0 - DRAM Controller */
26 
27 #define I3000_MCHBAR		0x44	/* MCH Memory Mapped Register BAR */
28 #define I3000_MCHBAR_MASK	0xffffc000
29 #define I3000_MMR_WINDOW_SIZE	16384
30 
31 #define I3000_EDEAP	0x70	/* Extended DRAM Error Address Pointer (8b)
32 				 *
33 				 * 7:1   reserved
34 				 * 0     bit 32 of address
35 				 */
36 #define I3000_DEAP	0x58	/* DRAM Error Address Pointer (32b)
37 				 *
38 				 * 31:7  address
39 				 * 6:1   reserved
40 				 * 0     Error channel 0/1
41 				 */
42 #define I3000_DEAP_GRAIN 		(1 << 7)
43 
44 /*
45  * Helper functions to decode the DEAP/EDEAP hardware registers.
46  *
47  * The type promotion here is deliberate; we're deriving an
48  * unsigned long pfn and offset from hardware regs which are u8/u32.
49  */
50 
deap_pfn(u8 edeap,u32 deap)51 static inline unsigned long deap_pfn(u8 edeap, u32 deap)
52 {
53 	deap >>= PAGE_SHIFT;
54 	deap |= (edeap & 1) << (32 - PAGE_SHIFT);
55 	return deap;
56 }
57 
deap_offset(u32 deap)58 static inline unsigned long deap_offset(u32 deap)
59 {
60 	return deap & ~(I3000_DEAP_GRAIN - 1) & ~PAGE_MASK;
61 }
62 
deap_channel(u32 deap)63 static inline int deap_channel(u32 deap)
64 {
65 	return deap & 1;
66 }
67 
68 #define I3000_DERRSYN	0x5c	/* DRAM Error Syndrome (8b)
69 				 *
70 				 *  7:0  DRAM ECC Syndrome
71 				 */
72 
73 #define I3000_ERRSTS	0xc8	/* Error Status Register (16b)
74 				 *
75 				 * 15:12 reserved
76 				 * 11    MCH Thermal Sensor Event
77 				 *         for SMI/SCI/SERR
78 				 * 10    reserved
79 				 *  9    LOCK to non-DRAM Memory Flag (LCKF)
80 				 *  8    Received Refresh Timeout Flag (RRTOF)
81 				 *  7:2  reserved
82 				 *  1    Multi-bit DRAM ECC Error Flag (DMERR)
83 				 *  0    Single-bit DRAM ECC Error Flag (DSERR)
84 				 */
85 #define I3000_ERRSTS_BITS	0x0b03	/* bits which indicate errors */
86 #define I3000_ERRSTS_UE		0x0002
87 #define I3000_ERRSTS_CE		0x0001
88 
89 #define I3000_ERRCMD	0xca	/* Error Command (16b)
90 				 *
91 				 * 15:12 reserved
92 				 * 11    SERR on MCH Thermal Sensor Event
93 				 *         (TSESERR)
94 				 * 10    reserved
95 				 *  9    SERR on LOCK to non-DRAM Memory
96 				 *         (LCKERR)
97 				 *  8    SERR on DRAM Refresh Timeout
98 				 *         (DRTOERR)
99 				 *  7:2  reserved
100 				 *  1    SERR Multi-Bit DRAM ECC Error
101 				 *         (DMERR)
102 				 *  0    SERR on Single-Bit ECC Error
103 				 *         (DSERR)
104 				 */
105 
106 /* Intel  MMIO register space - device 0 function 0 - MMR space */
107 
108 #define I3000_DRB_SHIFT 25	/* 32MiB grain */
109 
110 #define I3000_C0DRB	0x100	/* Channel 0 DRAM Rank Boundary (8b x 4)
111 				 *
112 				 * 7:0   Channel 0 DRAM Rank Boundary Address
113 				 */
114 #define I3000_C1DRB	0x180	/* Channel 1 DRAM Rank Boundary (8b x 4)
115 				 *
116 				 * 7:0   Channel 1 DRAM Rank Boundary Address
117 				 */
118 
119 #define I3000_C0DRA	0x108	/* Channel 0 DRAM Rank Attribute (8b x 2)
120 				 *
121 				 * 7     reserved
122 				 * 6:4   DRAM odd Rank Attribute
123 				 * 3     reserved
124 				 * 2:0   DRAM even Rank Attribute
125 				 *
126 				 * Each attribute defines the page
127 				 * size of the corresponding rank:
128 				 *     000: unpopulated
129 				 *     001: reserved
130 				 *     010: 4 KB
131 				 *     011: 8 KB
132 				 *     100: 16 KB
133 				 *     Others: reserved
134 				 */
135 #define I3000_C1DRA	0x188	/* Channel 1 DRAM Rank Attribute (8b x 2) */
136 
odd_rank_attrib(unsigned char dra)137 static inline unsigned char odd_rank_attrib(unsigned char dra)
138 {
139 	return (dra & 0x70) >> 4;
140 }
141 
even_rank_attrib(unsigned char dra)142 static inline unsigned char even_rank_attrib(unsigned char dra)
143 {
144 	return dra & 0x07;
145 }
146 
147 #define I3000_C0DRC0	0x120	/* DRAM Controller Mode 0 (32b)
148 				 *
149 				 * 31:30 reserved
150 				 * 29    Initialization Complete (IC)
151 				 * 28:11 reserved
152 				 * 10:8  Refresh Mode Select (RMS)
153 				 * 7     reserved
154 				 * 6:4   Mode Select (SMS)
155 				 * 3:2   reserved
156 				 * 1:0   DRAM Type (DT)
157 				 */
158 
159 #define I3000_C0DRC1	0x124	/* DRAM Controller Mode 1 (32b)
160 				 *
161 				 * 31    Enhanced Addressing Enable (ENHADE)
162 				 * 30:0  reserved
163 				 */
164 
165 enum i3000p_chips {
166 	I3000 = 0,
167 };
168 
169 struct i3000_dev_info {
170 	const char *ctl_name;
171 };
172 
173 struct i3000_error_info {
174 	u16 errsts;
175 	u8 derrsyn;
176 	u8 edeap;
177 	u32 deap;
178 	u16 errsts2;
179 };
180 
181 static const struct i3000_dev_info i3000_devs[] = {
182 	[I3000] = {
183 		.ctl_name = "i3000"},
184 };
185 
186 static struct pci_dev *mci_pdev;
187 static int i3000_registered = 1;
188 static struct edac_pci_ctl_info *i3000_pci;
189 
i3000_get_error_info(struct mem_ctl_info * mci,struct i3000_error_info * info)190 static void i3000_get_error_info(struct mem_ctl_info *mci,
191 				 struct i3000_error_info *info)
192 {
193 	struct pci_dev *pdev;
194 
195 	pdev = to_pci_dev(mci->pdev);
196 
197 	/*
198 	 * This is a mess because there is no atomic way to read all the
199 	 * registers at once and the registers can transition from CE being
200 	 * overwritten by UE.
201 	 */
202 	pci_read_config_word(pdev, I3000_ERRSTS, &info->errsts);
203 	if (!(info->errsts & I3000_ERRSTS_BITS))
204 		return;
205 	pci_read_config_byte(pdev, I3000_EDEAP, &info->edeap);
206 	pci_read_config_dword(pdev, I3000_DEAP, &info->deap);
207 	pci_read_config_byte(pdev, I3000_DERRSYN, &info->derrsyn);
208 	pci_read_config_word(pdev, I3000_ERRSTS, &info->errsts2);
209 
210 	/*
211 	 * If the error is the same for both reads then the first set
212 	 * of reads is valid.  If there is a change then there is a CE
213 	 * with no info and the second set of reads is valid and
214 	 * should be UE info.
215 	 */
216 	if ((info->errsts ^ info->errsts2) & I3000_ERRSTS_BITS) {
217 		pci_read_config_byte(pdev, I3000_EDEAP, &info->edeap);
218 		pci_read_config_dword(pdev, I3000_DEAP, &info->deap);
219 		pci_read_config_byte(pdev, I3000_DERRSYN, &info->derrsyn);
220 	}
221 
222 	/*
223 	 * Clear any error bits.
224 	 * (Yes, we really clear bits by writing 1 to them.)
225 	 */
226 	pci_write_bits16(pdev, I3000_ERRSTS, I3000_ERRSTS_BITS,
227 			 I3000_ERRSTS_BITS);
228 }
229 
i3000_process_error_info(struct mem_ctl_info * mci,struct i3000_error_info * info,int handle_errors)230 static int i3000_process_error_info(struct mem_ctl_info *mci,
231 				struct i3000_error_info *info,
232 				int handle_errors)
233 {
234 	int row, multi_chan, channel;
235 	unsigned long pfn, offset;
236 
237 	multi_chan = mci->csrows[0]->nr_channels - 1;
238 
239 	if (!(info->errsts & I3000_ERRSTS_BITS))
240 		return 0;
241 
242 	if (!handle_errors)
243 		return 1;
244 
245 	if ((info->errsts ^ info->errsts2) & I3000_ERRSTS_BITS) {
246 		edac_mc_handle_error(HW_EVENT_ERR_UNCORRECTED, mci, 1, 0, 0, 0,
247 				     -1, -1, -1,
248 				     "UE overwrote CE", "");
249 		info->errsts = info->errsts2;
250 	}
251 
252 	pfn = deap_pfn(info->edeap, info->deap);
253 	offset = deap_offset(info->deap);
254 	channel = deap_channel(info->deap);
255 
256 	row = edac_mc_find_csrow_by_page(mci, pfn);
257 
258 	if (info->errsts & I3000_ERRSTS_UE)
259 		edac_mc_handle_error(HW_EVENT_ERR_UNCORRECTED, mci, 1,
260 				     pfn, offset, 0,
261 				     row, -1, -1,
262 				     "i3000 UE", "");
263 	else
264 		edac_mc_handle_error(HW_EVENT_ERR_CORRECTED, mci, 1,
265 				     pfn, offset, info->derrsyn,
266 				     row, multi_chan ? channel : 0, -1,
267 				     "i3000 CE", "");
268 
269 	return 1;
270 }
271 
i3000_check(struct mem_ctl_info * mci)272 static void i3000_check(struct mem_ctl_info *mci)
273 {
274 	struct i3000_error_info info;
275 
276 	i3000_get_error_info(mci, &info);
277 	i3000_process_error_info(mci, &info, 1);
278 }
279 
i3000_is_interleaved(const unsigned char * c0dra,const unsigned char * c1dra,const unsigned char * c0drb,const unsigned char * c1drb)280 static int i3000_is_interleaved(const unsigned char *c0dra,
281 				const unsigned char *c1dra,
282 				const unsigned char *c0drb,
283 				const unsigned char *c1drb)
284 {
285 	int i;
286 
287 	/*
288 	 * If the channels aren't populated identically then
289 	 * we're not interleaved.
290 	 */
291 	for (i = 0; i < I3000_RANKS_PER_CHANNEL / 2; i++)
292 		if (odd_rank_attrib(c0dra[i]) != odd_rank_attrib(c1dra[i]) ||
293 			even_rank_attrib(c0dra[i]) !=
294 						even_rank_attrib(c1dra[i]))
295 			return 0;
296 
297 	/*
298 	 * If the rank boundaries for the two channels are different
299 	 * then we're not interleaved.
300 	 */
301 	for (i = 0; i < I3000_RANKS_PER_CHANNEL; i++)
302 		if (c0drb[i] != c1drb[i])
303 			return 0;
304 
305 	return 1;
306 }
307 
i3000_probe1(struct pci_dev * pdev,int dev_idx)308 static int i3000_probe1(struct pci_dev *pdev, int dev_idx)
309 {
310 	int rc;
311 	int i, j;
312 	struct mem_ctl_info *mci = NULL;
313 	struct edac_mc_layer layers[2];
314 	unsigned long last_cumul_size, nr_pages;
315 	int interleaved, nr_channels;
316 	unsigned char dra[I3000_RANKS / 2], drb[I3000_RANKS];
317 	unsigned char *c0dra = dra, *c1dra = &dra[I3000_RANKS_PER_CHANNEL / 2];
318 	unsigned char *c0drb = drb, *c1drb = &drb[I3000_RANKS_PER_CHANNEL];
319 	unsigned long mchbar;
320 	void __iomem *window;
321 
322 	edac_dbg(0, "MC:\n");
323 
324 	pci_read_config_dword(pdev, I3000_MCHBAR, (u32 *) & mchbar);
325 	mchbar &= I3000_MCHBAR_MASK;
326 	window = ioremap(mchbar, I3000_MMR_WINDOW_SIZE);
327 	if (!window) {
328 		printk(KERN_ERR "i3000: cannot map mmio space at 0x%lx\n",
329 			mchbar);
330 		return -ENODEV;
331 	}
332 
333 	c0dra[0] = readb(window + I3000_C0DRA + 0);	/* ranks 0,1 */
334 	c0dra[1] = readb(window + I3000_C0DRA + 1);	/* ranks 2,3 */
335 	c1dra[0] = readb(window + I3000_C1DRA + 0);	/* ranks 0,1 */
336 	c1dra[1] = readb(window + I3000_C1DRA + 1);	/* ranks 2,3 */
337 
338 	for (i = 0; i < I3000_RANKS_PER_CHANNEL; i++) {
339 		c0drb[i] = readb(window + I3000_C0DRB + i);
340 		c1drb[i] = readb(window + I3000_C1DRB + i);
341 	}
342 
343 	iounmap(window);
344 
345 	/*
346 	 * Figure out how many channels we have.
347 	 *
348 	 * If we have what the datasheet calls "asymmetric channels"
349 	 * (essentially the same as what was called "virtual single
350 	 * channel mode" in the i82875) then it's a single channel as
351 	 * far as EDAC is concerned.
352 	 */
353 	interleaved = i3000_is_interleaved(c0dra, c1dra, c0drb, c1drb);
354 	nr_channels = interleaved ? 2 : 1;
355 
356 	layers[0].type = EDAC_MC_LAYER_CHIP_SELECT;
357 	layers[0].size = I3000_RANKS / nr_channels;
358 	layers[0].is_virt_csrow = true;
359 	layers[1].type = EDAC_MC_LAYER_CHANNEL;
360 	layers[1].size = nr_channels;
361 	layers[1].is_virt_csrow = false;
362 	mci = edac_mc_alloc(0, ARRAY_SIZE(layers), layers, 0);
363 	if (!mci)
364 		return -ENOMEM;
365 
366 	edac_dbg(3, "MC: init mci\n");
367 
368 	mci->pdev = &pdev->dev;
369 	mci->mtype_cap = MEM_FLAG_DDR2;
370 
371 	mci->edac_ctl_cap = EDAC_FLAG_SECDED;
372 	mci->edac_cap = EDAC_FLAG_SECDED;
373 
374 	mci->mod_name = EDAC_MOD_STR;
375 	mci->ctl_name = i3000_devs[dev_idx].ctl_name;
376 	mci->dev_name = pci_name(pdev);
377 	mci->edac_check = i3000_check;
378 	mci->ctl_page_to_phys = NULL;
379 
380 	/*
381 	 * The dram rank boundary (DRB) reg values are boundary addresses
382 	 * for each DRAM rank with a granularity of 32MB.  DRB regs are
383 	 * cumulative; the last one will contain the total memory
384 	 * contained in all ranks.
385 	 *
386 	 * If we're in interleaved mode then we're only walking through
387 	 * the ranks of controller 0, so we double all the values we see.
388 	 */
389 	for (last_cumul_size = i = 0; i < mci->nr_csrows; i++) {
390 		u8 value;
391 		u32 cumul_size;
392 		struct csrow_info *csrow = mci->csrows[i];
393 
394 		value = drb[i];
395 		cumul_size = value << (I3000_DRB_SHIFT - PAGE_SHIFT);
396 		if (interleaved)
397 			cumul_size <<= 1;
398 		edac_dbg(3, "MC: (%d) cumul_size 0x%x\n", i, cumul_size);
399 		if (cumul_size == last_cumul_size)
400 			continue;
401 
402 		csrow->first_page = last_cumul_size;
403 		csrow->last_page = cumul_size - 1;
404 		nr_pages = cumul_size - last_cumul_size;
405 		last_cumul_size = cumul_size;
406 
407 		for (j = 0; j < nr_channels; j++) {
408 			struct dimm_info *dimm = csrow->channels[j]->dimm;
409 
410 			dimm->nr_pages = nr_pages / nr_channels;
411 			dimm->grain = I3000_DEAP_GRAIN;
412 			dimm->mtype = MEM_DDR2;
413 			dimm->dtype = DEV_UNKNOWN;
414 			dimm->edac_mode = EDAC_UNKNOWN;
415 		}
416 	}
417 
418 	/*
419 	 * Clear any error bits.
420 	 * (Yes, we really clear bits by writing 1 to them.)
421 	 */
422 	pci_write_bits16(pdev, I3000_ERRSTS, I3000_ERRSTS_BITS,
423 			 I3000_ERRSTS_BITS);
424 
425 	rc = -ENODEV;
426 	if (edac_mc_add_mc(mci)) {
427 		edac_dbg(3, "MC: failed edac_mc_add_mc()\n");
428 		goto fail;
429 	}
430 
431 	/* allocating generic PCI control info */
432 	i3000_pci = edac_pci_create_generic_ctl(&pdev->dev, EDAC_MOD_STR);
433 	if (!i3000_pci) {
434 		printk(KERN_WARNING
435 			"%s(): Unable to create PCI control\n",
436 			__func__);
437 		printk(KERN_WARNING
438 			"%s(): PCI error report via EDAC not setup\n",
439 			__func__);
440 	}
441 
442 	/* get this far and it's successful */
443 	edac_dbg(3, "MC: success\n");
444 	return 0;
445 
446 fail:
447 	if (mci)
448 		edac_mc_free(mci);
449 
450 	return rc;
451 }
452 
453 /* returns count (>= 0), or negative on error */
i3000_init_one(struct pci_dev * pdev,const struct pci_device_id * ent)454 static int i3000_init_one(struct pci_dev *pdev, const struct pci_device_id *ent)
455 {
456 	int rc;
457 
458 	edac_dbg(0, "MC:\n");
459 
460 	if (pci_enable_device(pdev) < 0)
461 		return -EIO;
462 
463 	rc = i3000_probe1(pdev, ent->driver_data);
464 	if (!mci_pdev)
465 		mci_pdev = pci_dev_get(pdev);
466 
467 	return rc;
468 }
469 
i3000_remove_one(struct pci_dev * pdev)470 static void i3000_remove_one(struct pci_dev *pdev)
471 {
472 	struct mem_ctl_info *mci;
473 
474 	edac_dbg(0, "\n");
475 
476 	if (i3000_pci)
477 		edac_pci_release_generic_ctl(i3000_pci);
478 
479 	mci = edac_mc_del_mc(&pdev->dev);
480 	if (!mci)
481 		return;
482 
483 	edac_mc_free(mci);
484 }
485 
486 static const struct pci_device_id i3000_pci_tbl[] = {
487 	{
488 	 PCI_VEND_DEV(INTEL, 3000_HB), PCI_ANY_ID, PCI_ANY_ID, 0, 0,
489 	 I3000},
490 	{
491 	 0,
492 	 }			/* 0 terminated list. */
493 };
494 
495 MODULE_DEVICE_TABLE(pci, i3000_pci_tbl);
496 
497 static struct pci_driver i3000_driver = {
498 	.name = EDAC_MOD_STR,
499 	.probe = i3000_init_one,
500 	.remove = i3000_remove_one,
501 	.id_table = i3000_pci_tbl,
502 };
503 
i3000_init(void)504 static int __init i3000_init(void)
505 {
506 	int pci_rc;
507 
508 	edac_dbg(3, "MC:\n");
509 
510 	/* Ensure that the OPSTATE is set correctly for POLL or NMI */
511 	opstate_init();
512 
513 	pci_rc = pci_register_driver(&i3000_driver);
514 	if (pci_rc < 0)
515 		goto fail0;
516 
517 	if (!mci_pdev) {
518 		i3000_registered = 0;
519 		mci_pdev = pci_get_device(PCI_VENDOR_ID_INTEL,
520 					PCI_DEVICE_ID_INTEL_3000_HB, NULL);
521 		if (!mci_pdev) {
522 			edac_dbg(0, "i3000 pci_get_device fail\n");
523 			pci_rc = -ENODEV;
524 			goto fail1;
525 		}
526 
527 		pci_rc = i3000_init_one(mci_pdev, i3000_pci_tbl);
528 		if (pci_rc < 0) {
529 			edac_dbg(0, "i3000 init fail\n");
530 			pci_rc = -ENODEV;
531 			goto fail1;
532 		}
533 	}
534 
535 	return 0;
536 
537 fail1:
538 	pci_unregister_driver(&i3000_driver);
539 
540 fail0:
541 	pci_dev_put(mci_pdev);
542 
543 	return pci_rc;
544 }
545 
i3000_exit(void)546 static void __exit i3000_exit(void)
547 {
548 	edac_dbg(3, "MC:\n");
549 
550 	pci_unregister_driver(&i3000_driver);
551 	if (!i3000_registered) {
552 		i3000_remove_one(mci_pdev);
553 		pci_dev_put(mci_pdev);
554 	}
555 }
556 
557 module_init(i3000_init);
558 module_exit(i3000_exit);
559 
560 MODULE_LICENSE("GPL");
561 MODULE_AUTHOR("Akamai Technologies Arthur Ulfeldt/Jason Uhlenkott");
562 MODULE_DESCRIPTION("MC support for Intel 3000 memory hub controllers");
563 
564 module_param(edac_op_state, int, 0444);
565 MODULE_PARM_DESC(edac_op_state, "EDAC Error Reporting state: 0=Poll,1=NMI");
566