xref: /linux/drivers/edac/x38_edac.c (revision cdd38c5f1ce4398ec58fec95904b75824daab7b5)
1 /*
2  * Intel X38 Memory Controller kernel module
3  * Copyright (C) 2008 Cluster Computing, Inc.
4  *
5  * This file may be distributed under the terms of the
6  * GNU General Public License.
7  *
8  * This file is based on i3200_edac.c
9  *
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 
18 #include <linux/io-64-nonatomic-lo-hi.h>
19 #include "edac_module.h"
20 
21 #define EDAC_MOD_STR		"x38_edac"
22 
23 #define PCI_DEVICE_ID_INTEL_X38_HB	0x29e0
24 
25 #define X38_RANKS		8
26 #define X38_RANKS_PER_CHANNEL	4
27 #define X38_CHANNELS		2
28 
29 /* Intel X38 register addresses - device 0 function 0 - DRAM Controller */
30 
31 #define X38_MCHBAR_LOW	0x48	/* MCH Memory Mapped Register BAR */
32 #define X38_MCHBAR_HIGH	0x4c
33 #define X38_MCHBAR_MASK	0xfffffc000ULL	/* bits 35:14 */
34 #define X38_MMR_WINDOW_SIZE	16384
35 
36 #define X38_TOM	0xa0	/* Top of Memory (16b)
37 				 *
38 				 * 15:10 reserved
39 				 *  9:0  total populated physical memory
40 				 */
41 #define X38_TOM_MASK	0x3ff	/* bits 9:0 */
42 #define X38_TOM_SHIFT 26	/* 64MiB grain */
43 
44 #define X38_ERRSTS	0xc8	/* Error Status Register (16b)
45 				 *
46 				 * 15    reserved
47 				 * 14    Isochronous TBWRR Run Behind FIFO Full
48 				 *       (ITCV)
49 				 * 13    Isochronous TBWRR Run Behind FIFO Put
50 				 *       (ITSTV)
51 				 * 12    reserved
52 				 * 11    MCH Thermal Sensor Event
53 				 *       for SMI/SCI/SERR (GTSE)
54 				 * 10    reserved
55 				 *  9    LOCK to non-DRAM Memory Flag (LCKF)
56 				 *  8    reserved
57 				 *  7    DRAM Throttle Flag (DTF)
58 				 *  6:2  reserved
59 				 *  1    Multi-bit DRAM ECC Error Flag (DMERR)
60 				 *  0    Single-bit DRAM ECC Error Flag (DSERR)
61 				 */
62 #define X38_ERRSTS_UE		0x0002
63 #define X38_ERRSTS_CE		0x0001
64 #define X38_ERRSTS_BITS	(X38_ERRSTS_UE | X38_ERRSTS_CE)
65 
66 
67 /* Intel  MMIO register space - device 0 function 0 - MMR space */
68 
69 #define X38_C0DRB	0x200	/* Channel 0 DRAM Rank Boundary (16b x 4)
70 				 *
71 				 * 15:10 reserved
72 				 *  9:0  Channel 0 DRAM Rank Boundary Address
73 				 */
74 #define X38_C1DRB	0x600	/* Channel 1 DRAM Rank Boundary (16b x 4) */
75 #define X38_DRB_MASK	0x3ff	/* bits 9:0 */
76 #define X38_DRB_SHIFT 26	/* 64MiB grain */
77 
78 #define X38_C0ECCERRLOG 0x280	/* Channel 0 ECC Error Log (64b)
79 				 *
80 				 * 63:48 Error Column Address (ERRCOL)
81 				 * 47:32 Error Row Address (ERRROW)
82 				 * 31:29 Error Bank Address (ERRBANK)
83 				 * 28:27 Error Rank Address (ERRRANK)
84 				 * 26:24 reserved
85 				 * 23:16 Error Syndrome (ERRSYND)
86 				 * 15: 2 reserved
87 				 *    1  Multiple Bit Error Status (MERRSTS)
88 				 *    0  Correctable Error Status (CERRSTS)
89 				 */
90 #define X38_C1ECCERRLOG 0x680	/* Channel 1 ECC Error Log (64b) */
91 #define X38_ECCERRLOG_CE	0x1
92 #define X38_ECCERRLOG_UE	0x2
93 #define X38_ECCERRLOG_RANK_BITS	0x18000000
94 #define X38_ECCERRLOG_SYNDROME_BITS	0xff0000
95 
96 #define X38_CAPID0 0xe0	/* see P.94 of spec for details */
97 
98 static int x38_channel_num;
99 
how_many_channel(struct pci_dev * pdev)100 static int how_many_channel(struct pci_dev *pdev)
101 {
102 	unsigned char capid0_8b; /* 8th byte of CAPID0 */
103 
104 	pci_read_config_byte(pdev, X38_CAPID0 + 8, &capid0_8b);
105 	if (capid0_8b & 0x20) {	/* check DCD: Dual Channel Disable */
106 		edac_dbg(0, "In single channel mode\n");
107 		x38_channel_num = 1;
108 	} else {
109 		edac_dbg(0, "In dual channel mode\n");
110 		x38_channel_num = 2;
111 	}
112 
113 	return x38_channel_num;
114 }
115 
eccerrlog_syndrome(u64 log)116 static unsigned long eccerrlog_syndrome(u64 log)
117 {
118 	return (log & X38_ECCERRLOG_SYNDROME_BITS) >> 16;
119 }
120 
eccerrlog_row(int channel,u64 log)121 static int eccerrlog_row(int channel, u64 log)
122 {
123 	return ((log & X38_ECCERRLOG_RANK_BITS) >> 27) |
124 		(channel * X38_RANKS_PER_CHANNEL);
125 }
126 
127 enum x38_chips {
128 	X38 = 0,
129 };
130 
131 struct x38_dev_info {
132 	const char *ctl_name;
133 };
134 
135 struct x38_error_info {
136 	u16 errsts;
137 	u16 errsts2;
138 	u64 eccerrlog[X38_CHANNELS];
139 };
140 
141 static const struct x38_dev_info x38_devs[] = {
142 	[X38] = {
143 		.ctl_name = "x38"},
144 };
145 
146 static struct pci_dev *mci_pdev;
147 static int x38_registered = 1;
148 
149 
x38_clear_error_info(struct mem_ctl_info * mci)150 static void x38_clear_error_info(struct mem_ctl_info *mci)
151 {
152 	struct pci_dev *pdev;
153 
154 	pdev = to_pci_dev(mci->pdev);
155 
156 	/*
157 	 * Clear any error bits.
158 	 * (Yes, we really clear bits by writing 1 to them.)
159 	 */
160 	pci_write_bits16(pdev, X38_ERRSTS, X38_ERRSTS_BITS,
161 			 X38_ERRSTS_BITS);
162 }
163 
x38_get_and_clear_error_info(struct mem_ctl_info * mci,struct x38_error_info * info)164 static void x38_get_and_clear_error_info(struct mem_ctl_info *mci,
165 				 struct x38_error_info *info)
166 {
167 	struct pci_dev *pdev;
168 	void __iomem *window = mci->pvt_info;
169 
170 	pdev = to_pci_dev(mci->pdev);
171 
172 	/*
173 	 * This is a mess because there is no atomic way to read all the
174 	 * registers at once and the registers can transition from CE being
175 	 * overwritten by UE.
176 	 */
177 	pci_read_config_word(pdev, X38_ERRSTS, &info->errsts);
178 	if (!(info->errsts & X38_ERRSTS_BITS))
179 		return;
180 
181 	info->eccerrlog[0] = lo_hi_readq(window + X38_C0ECCERRLOG);
182 	if (x38_channel_num == 2)
183 		info->eccerrlog[1] = lo_hi_readq(window + X38_C1ECCERRLOG);
184 
185 	pci_read_config_word(pdev, X38_ERRSTS, &info->errsts2);
186 
187 	/*
188 	 * If the error is the same for both reads then the first set
189 	 * of reads is valid.  If there is a change then there is a CE
190 	 * with no info and the second set of reads is valid and
191 	 * should be UE info.
192 	 */
193 	if ((info->errsts ^ info->errsts2) & X38_ERRSTS_BITS) {
194 		info->eccerrlog[0] = lo_hi_readq(window + X38_C0ECCERRLOG);
195 		if (x38_channel_num == 2)
196 			info->eccerrlog[1] =
197 				lo_hi_readq(window + X38_C1ECCERRLOG);
198 	}
199 
200 	x38_clear_error_info(mci);
201 }
202 
x38_process_error_info(struct mem_ctl_info * mci,struct x38_error_info * info)203 static void x38_process_error_info(struct mem_ctl_info *mci,
204 				struct x38_error_info *info)
205 {
206 	int channel;
207 	u64 log;
208 
209 	if (!(info->errsts & X38_ERRSTS_BITS))
210 		return;
211 
212 	if ((info->errsts ^ info->errsts2) & X38_ERRSTS_BITS) {
213 		edac_mc_handle_error(HW_EVENT_ERR_UNCORRECTED, mci, 1, 0, 0, 0,
214 				     -1, -1, -1,
215 				     "UE overwrote CE", "");
216 		info->errsts = info->errsts2;
217 	}
218 
219 	for (channel = 0; channel < x38_channel_num; channel++) {
220 		log = info->eccerrlog[channel];
221 		if (log & X38_ECCERRLOG_UE) {
222 			edac_mc_handle_error(HW_EVENT_ERR_UNCORRECTED, mci, 1,
223 					     0, 0, 0,
224 					     eccerrlog_row(channel, log),
225 					     -1, -1,
226 					     "x38 UE", "");
227 		} else if (log & X38_ECCERRLOG_CE) {
228 			edac_mc_handle_error(HW_EVENT_ERR_CORRECTED, mci, 1,
229 					     0, 0, eccerrlog_syndrome(log),
230 					     eccerrlog_row(channel, log),
231 					     -1, -1,
232 					     "x38 CE", "");
233 		}
234 	}
235 }
236 
x38_check(struct mem_ctl_info * mci)237 static void x38_check(struct mem_ctl_info *mci)
238 {
239 	struct x38_error_info info;
240 
241 	x38_get_and_clear_error_info(mci, &info);
242 	x38_process_error_info(mci, &info);
243 }
244 
x38_map_mchbar(struct pci_dev * pdev)245 static void __iomem *x38_map_mchbar(struct pci_dev *pdev)
246 {
247 	union {
248 		u64 mchbar;
249 		struct {
250 			u32 mchbar_low;
251 			u32 mchbar_high;
252 		};
253 	} u;
254 	void __iomem *window;
255 
256 	pci_read_config_dword(pdev, X38_MCHBAR_LOW, &u.mchbar_low);
257 	pci_write_config_dword(pdev, X38_MCHBAR_LOW, u.mchbar_low | 0x1);
258 	pci_read_config_dword(pdev, X38_MCHBAR_HIGH, &u.mchbar_high);
259 	u.mchbar &= X38_MCHBAR_MASK;
260 
261 	if (u.mchbar != (resource_size_t)u.mchbar) {
262 		printk(KERN_ERR
263 			"x38: mmio space beyond accessible range (0x%llx)\n",
264 			(unsigned long long)u.mchbar);
265 		return NULL;
266 	}
267 
268 	window = ioremap(u.mchbar, X38_MMR_WINDOW_SIZE);
269 	if (!window)
270 		printk(KERN_ERR "x38: cannot map mmio space at 0x%llx\n",
271 			(unsigned long long)u.mchbar);
272 
273 	return window;
274 }
275 
276 
x38_get_drbs(void __iomem * window,u16 drbs[X38_CHANNELS][X38_RANKS_PER_CHANNEL])277 static void x38_get_drbs(void __iomem *window,
278 			u16 drbs[X38_CHANNELS][X38_RANKS_PER_CHANNEL])
279 {
280 	int i;
281 
282 	for (i = 0; i < X38_RANKS_PER_CHANNEL; i++) {
283 		drbs[0][i] = readw(window + X38_C0DRB + 2*i) & X38_DRB_MASK;
284 		drbs[1][i] = readw(window + X38_C1DRB + 2*i) & X38_DRB_MASK;
285 	}
286 }
287 
x38_is_stacked(struct pci_dev * pdev,u16 drbs[X38_CHANNELS][X38_RANKS_PER_CHANNEL])288 static bool x38_is_stacked(struct pci_dev *pdev,
289 			u16 drbs[X38_CHANNELS][X38_RANKS_PER_CHANNEL])
290 {
291 	u16 tom;
292 
293 	pci_read_config_word(pdev, X38_TOM, &tom);
294 	tom &= X38_TOM_MASK;
295 
296 	return drbs[X38_CHANNELS - 1][X38_RANKS_PER_CHANNEL - 1] == tom;
297 }
298 
drb_to_nr_pages(u16 drbs[X38_CHANNELS][X38_RANKS_PER_CHANNEL],bool stacked,int channel,int rank)299 static unsigned long drb_to_nr_pages(
300 			u16 drbs[X38_CHANNELS][X38_RANKS_PER_CHANNEL],
301 			bool stacked, int channel, int rank)
302 {
303 	int n;
304 
305 	n = drbs[channel][rank];
306 	if (rank > 0)
307 		n -= drbs[channel][rank - 1];
308 	if (stacked && (channel == 1) && drbs[channel][rank] ==
309 				drbs[channel][X38_RANKS_PER_CHANNEL - 1]) {
310 		n -= drbs[0][X38_RANKS_PER_CHANNEL - 1];
311 	}
312 
313 	n <<= (X38_DRB_SHIFT - PAGE_SHIFT);
314 	return n;
315 }
316 
x38_probe1(struct pci_dev * pdev,int dev_idx)317 static int x38_probe1(struct pci_dev *pdev, int dev_idx)
318 {
319 	int rc;
320 	int i, j;
321 	struct mem_ctl_info *mci = NULL;
322 	struct edac_mc_layer layers[2];
323 	u16 drbs[X38_CHANNELS][X38_RANKS_PER_CHANNEL];
324 	bool stacked;
325 	void __iomem *window;
326 
327 	edac_dbg(0, "MC:\n");
328 
329 	window = x38_map_mchbar(pdev);
330 	if (!window)
331 		return -ENODEV;
332 
333 	x38_get_drbs(window, drbs);
334 
335 	how_many_channel(pdev);
336 
337 	/* FIXME: unconventional pvt_info usage */
338 	layers[0].type = EDAC_MC_LAYER_CHIP_SELECT;
339 	layers[0].size = X38_RANKS;
340 	layers[0].is_virt_csrow = true;
341 	layers[1].type = EDAC_MC_LAYER_CHANNEL;
342 	layers[1].size = x38_channel_num;
343 	layers[1].is_virt_csrow = false;
344 	mci = edac_mc_alloc(0, ARRAY_SIZE(layers), layers, 0);
345 	if (!mci)
346 		return -ENOMEM;
347 
348 	edac_dbg(3, "MC: init mci\n");
349 
350 	mci->pdev = &pdev->dev;
351 	mci->mtype_cap = MEM_FLAG_DDR2;
352 
353 	mci->edac_ctl_cap = EDAC_FLAG_SECDED;
354 	mci->edac_cap = EDAC_FLAG_SECDED;
355 
356 	mci->mod_name = EDAC_MOD_STR;
357 	mci->ctl_name = x38_devs[dev_idx].ctl_name;
358 	mci->dev_name = pci_name(pdev);
359 	mci->edac_check = x38_check;
360 	mci->ctl_page_to_phys = NULL;
361 	mci->pvt_info = window;
362 
363 	stacked = x38_is_stacked(pdev, drbs);
364 
365 	/*
366 	 * The dram rank boundary (DRB) reg values are boundary addresses
367 	 * for each DRAM rank with a granularity of 64MB.  DRB regs are
368 	 * cumulative; the last one will contain the total memory
369 	 * contained in all ranks.
370 	 */
371 	for (i = 0; i < mci->nr_csrows; i++) {
372 		unsigned long nr_pages;
373 		struct csrow_info *csrow = mci->csrows[i];
374 
375 		nr_pages = drb_to_nr_pages(drbs, stacked,
376 			i / X38_RANKS_PER_CHANNEL,
377 			i % X38_RANKS_PER_CHANNEL);
378 
379 		if (nr_pages == 0)
380 			continue;
381 
382 		for (j = 0; j < x38_channel_num; j++) {
383 			struct dimm_info *dimm = csrow->channels[j]->dimm;
384 
385 			dimm->nr_pages = nr_pages / x38_channel_num;
386 			dimm->grain = nr_pages << PAGE_SHIFT;
387 			dimm->mtype = MEM_DDR2;
388 			dimm->dtype = DEV_UNKNOWN;
389 			dimm->edac_mode = EDAC_UNKNOWN;
390 		}
391 	}
392 
393 	x38_clear_error_info(mci);
394 
395 	rc = -ENODEV;
396 	if (edac_mc_add_mc(mci)) {
397 		edac_dbg(3, "MC: failed edac_mc_add_mc()\n");
398 		goto fail;
399 	}
400 
401 	/* get this far and it's successful */
402 	edac_dbg(3, "MC: success\n");
403 	return 0;
404 
405 fail:
406 	iounmap(window);
407 	if (mci)
408 		edac_mc_free(mci);
409 
410 	return rc;
411 }
412 
x38_init_one(struct pci_dev * pdev,const struct pci_device_id * ent)413 static int x38_init_one(struct pci_dev *pdev, const struct pci_device_id *ent)
414 {
415 	int rc;
416 
417 	edac_dbg(0, "MC:\n");
418 
419 	if (pci_enable_device(pdev) < 0)
420 		return -EIO;
421 
422 	rc = x38_probe1(pdev, ent->driver_data);
423 	if (!mci_pdev)
424 		mci_pdev = pci_dev_get(pdev);
425 
426 	return rc;
427 }
428 
x38_remove_one(struct pci_dev * pdev)429 static void x38_remove_one(struct pci_dev *pdev)
430 {
431 	struct mem_ctl_info *mci;
432 
433 	edac_dbg(0, "\n");
434 
435 	mci = edac_mc_del_mc(&pdev->dev);
436 	if (!mci)
437 		return;
438 
439 	iounmap(mci->pvt_info);
440 
441 	edac_mc_free(mci);
442 }
443 
444 static const struct pci_device_id x38_pci_tbl[] = {
445 	{
446 	 PCI_VEND_DEV(INTEL, X38_HB), PCI_ANY_ID, PCI_ANY_ID, 0, 0,
447 	 X38},
448 	{
449 	 0,
450 	 }			/* 0 terminated list. */
451 };
452 
453 MODULE_DEVICE_TABLE(pci, x38_pci_tbl);
454 
455 static struct pci_driver x38_driver = {
456 	.name = EDAC_MOD_STR,
457 	.probe = x38_init_one,
458 	.remove = x38_remove_one,
459 	.id_table = x38_pci_tbl,
460 };
461 
x38_init(void)462 static int __init x38_init(void)
463 {
464 	int pci_rc;
465 
466 	edac_dbg(3, "MC:\n");
467 
468 	/* Ensure that the OPSTATE is set correctly for POLL or NMI */
469 	opstate_init();
470 
471 	pci_rc = pci_register_driver(&x38_driver);
472 	if (pci_rc < 0)
473 		goto fail0;
474 
475 	if (!mci_pdev) {
476 		x38_registered = 0;
477 		mci_pdev = pci_get_device(PCI_VENDOR_ID_INTEL,
478 					PCI_DEVICE_ID_INTEL_X38_HB, NULL);
479 		if (!mci_pdev) {
480 			edac_dbg(0, "x38 pci_get_device fail\n");
481 			pci_rc = -ENODEV;
482 			goto fail1;
483 		}
484 
485 		pci_rc = x38_init_one(mci_pdev, x38_pci_tbl);
486 		if (pci_rc < 0) {
487 			edac_dbg(0, "x38 init fail\n");
488 			pci_rc = -ENODEV;
489 			goto fail1;
490 		}
491 	}
492 
493 	return 0;
494 
495 fail1:
496 	pci_unregister_driver(&x38_driver);
497 
498 fail0:
499 	pci_dev_put(mci_pdev);
500 
501 	return pci_rc;
502 }
503 
x38_exit(void)504 static void __exit x38_exit(void)
505 {
506 	edac_dbg(3, "MC:\n");
507 
508 	pci_unregister_driver(&x38_driver);
509 	if (!x38_registered) {
510 		x38_remove_one(mci_pdev);
511 		pci_dev_put(mci_pdev);
512 	}
513 }
514 
515 module_init(x38_init);
516 module_exit(x38_exit);
517 
518 MODULE_LICENSE("GPL");
519 MODULE_AUTHOR("Cluster Computing, Inc. Hitoshi Mitake");
520 MODULE_DESCRIPTION("MC support for Intel X38 memory hub controllers");
521 
522 module_param(edac_op_state, int, 0444);
523 MODULE_PARM_DESC(edac_op_state, "EDAC Error Reporting state: 0=Poll,1=NMI");
524