Committer: Michael Beasley <mike@snafu.setup>
[mikesnafu-overlay.git] / drivers / edac / e7xxx_edac.c
blob96ecc492664122cb5025d9996353077642310957
1 /*
2 * Intel e7xxx Memory Controller kernel module
3 * (C) 2003 Linux Networx (http://lnxi.com)
4 * This file may be distributed under the terms of the
5 * GNU General Public License.
7 * See "enum e7xxx_chips" below for supported chipsets
9 * Written by Thayne Harbaugh
10 * Based on work by Dan Hollis <goemon at anime dot net> and others.
11 * http://www.anime.net/~goemon/linux-ecc/
13 * Contributors:
14 * Eric Biederman (Linux Networx)
15 * Tom Zimmerman (Linux Networx)
16 * Jim Garlick (Lawrence Livermore National Labs)
17 * Dave Peterson (Lawrence Livermore National Labs)
18 * That One Guy (Some other place)
19 * Wang Zhenyu (intel.com)
21 * $Id: edac_e7xxx.c,v 1.5.2.9 2005/10/05 00:43:44 dsp_llnl Exp $
25 #include <linux/module.h>
26 #include <linux/init.h>
27 #include <linux/pci.h>
28 #include <linux/pci_ids.h>
29 #include <linux/slab.h>
30 #include <linux/edac.h>
31 #include "edac_core.h"
33 #define E7XXX_REVISION " Ver: 2.0.2 " __DATE__
34 #define EDAC_MOD_STR "e7xxx_edac"
36 #define e7xxx_printk(level, fmt, arg...) \
37 edac_printk(level, "e7xxx", fmt, ##arg)
39 #define e7xxx_mc_printk(mci, level, fmt, arg...) \
40 edac_mc_chipset_printk(mci, level, "e7xxx", fmt, ##arg)
42 #ifndef PCI_DEVICE_ID_INTEL_7205_0
43 #define PCI_DEVICE_ID_INTEL_7205_0 0x255d
44 #endif /* PCI_DEVICE_ID_INTEL_7205_0 */
46 #ifndef PCI_DEVICE_ID_INTEL_7205_1_ERR
47 #define PCI_DEVICE_ID_INTEL_7205_1_ERR 0x2551
48 #endif /* PCI_DEVICE_ID_INTEL_7205_1_ERR */
50 #ifndef PCI_DEVICE_ID_INTEL_7500_0
51 #define PCI_DEVICE_ID_INTEL_7500_0 0x2540
52 #endif /* PCI_DEVICE_ID_INTEL_7500_0 */
54 #ifndef PCI_DEVICE_ID_INTEL_7500_1_ERR
55 #define PCI_DEVICE_ID_INTEL_7500_1_ERR 0x2541
56 #endif /* PCI_DEVICE_ID_INTEL_7500_1_ERR */
58 #ifndef PCI_DEVICE_ID_INTEL_7501_0
59 #define PCI_DEVICE_ID_INTEL_7501_0 0x254c
60 #endif /* PCI_DEVICE_ID_INTEL_7501_0 */
62 #ifndef PCI_DEVICE_ID_INTEL_7501_1_ERR
63 #define PCI_DEVICE_ID_INTEL_7501_1_ERR 0x2541
64 #endif /* PCI_DEVICE_ID_INTEL_7501_1_ERR */
66 #ifndef PCI_DEVICE_ID_INTEL_7505_0
67 #define PCI_DEVICE_ID_INTEL_7505_0 0x2550
68 #endif /* PCI_DEVICE_ID_INTEL_7505_0 */
70 #ifndef PCI_DEVICE_ID_INTEL_7505_1_ERR
71 #define PCI_DEVICE_ID_INTEL_7505_1_ERR 0x2551
72 #endif /* PCI_DEVICE_ID_INTEL_7505_1_ERR */
74 #define E7XXX_NR_CSROWS 8 /* number of csrows */
75 #define E7XXX_NR_DIMMS 8 /* FIXME - is this correct? */
77 /* E7XXX register addresses - device 0 function 0 */
78 #define E7XXX_DRB 0x60 /* DRAM row boundary register (8b) */
79 #define E7XXX_DRA 0x70 /* DRAM row attribute register (8b) */
81 * 31 Device width row 7 0=x8 1=x4
82 * 27 Device width row 6
83 * 23 Device width row 5
84 * 19 Device width row 4
85 * 15 Device width row 3
86 * 11 Device width row 2
87 * 7 Device width row 1
88 * 3 Device width row 0
90 #define E7XXX_DRC 0x7C /* DRAM controller mode reg (32b) */
92 * 22 Number channels 0=1,1=2
93 * 19:18 DRB Granularity 32/64MB
95 #define E7XXX_TOLM 0xC4 /* DRAM top of low memory reg (16b) */
96 #define E7XXX_REMAPBASE 0xC6 /* DRAM remap base address reg (16b) */
97 #define E7XXX_REMAPLIMIT 0xC8 /* DRAM remap limit address reg (16b) */
99 /* E7XXX register addresses - device 0 function 1 */
100 #define E7XXX_DRAM_FERR 0x80 /* DRAM first error register (8b) */
101 #define E7XXX_DRAM_NERR 0x82 /* DRAM next error register (8b) */
102 #define E7XXX_DRAM_CELOG_ADD 0xA0 /* DRAM first correctable memory */
103 /* error address register (32b) */
105 * 31:28 Reserved
106 * 27:6 CE address (4k block 33:12)
107 * 5:0 Reserved
109 #define E7XXX_DRAM_UELOG_ADD 0xB0 /* DRAM first uncorrectable memory */
110 /* error address register (32b) */
112 * 31:28 Reserved
113 * 27:6 CE address (4k block 33:12)
114 * 5:0 Reserved
116 #define E7XXX_DRAM_CELOG_SYNDROME 0xD0 /* DRAM first correctable memory */
117 /* error syndrome register (16b) */
119 enum e7xxx_chips {
120 E7500 = 0,
121 E7501,
122 E7505,
123 E7205,
126 struct e7xxx_pvt {
127 struct pci_dev *bridge_ck;
128 u32 tolm;
129 u32 remapbase;
130 u32 remaplimit;
131 const struct e7xxx_dev_info *dev_info;
134 struct e7xxx_dev_info {
135 u16 err_dev;
136 const char *ctl_name;
139 struct e7xxx_error_info {
140 u8 dram_ferr;
141 u8 dram_nerr;
142 u32 dram_celog_add;
143 u16 dram_celog_syndrome;
144 u32 dram_uelog_add;
147 static struct edac_pci_ctl_info *e7xxx_pci;
149 static const struct e7xxx_dev_info e7xxx_devs[] = {
150 [E7500] = {
151 .err_dev = PCI_DEVICE_ID_INTEL_7500_1_ERR,
152 .ctl_name = "E7500"},
153 [E7501] = {
154 .err_dev = PCI_DEVICE_ID_INTEL_7501_1_ERR,
155 .ctl_name = "E7501"},
156 [E7505] = {
157 .err_dev = PCI_DEVICE_ID_INTEL_7505_1_ERR,
158 .ctl_name = "E7505"},
159 [E7205] = {
160 .err_dev = PCI_DEVICE_ID_INTEL_7205_1_ERR,
161 .ctl_name = "E7205"},
164 /* FIXME - is this valid for both SECDED and S4ECD4ED? */
165 static inline int e7xxx_find_channel(u16 syndrome)
167 debugf3("%s()\n", __func__);
169 if ((syndrome & 0xff00) == 0)
170 return 0;
172 if ((syndrome & 0x00ff) == 0)
173 return 1;
175 if ((syndrome & 0xf000) == 0 || (syndrome & 0x0f00) == 0)
176 return 0;
178 return 1;
181 static unsigned long ctl_page_to_phys(struct mem_ctl_info *mci,
182 unsigned long page)
184 u32 remap;
185 struct e7xxx_pvt *pvt = (struct e7xxx_pvt *)mci->pvt_info;
187 debugf3("%s()\n", __func__);
189 if ((page < pvt->tolm) ||
190 ((page >= 0x100000) && (page < pvt->remapbase)))
191 return page;
193 remap = (page - pvt->tolm) + pvt->remapbase;
195 if (remap < pvt->remaplimit)
196 return remap;
198 e7xxx_printk(KERN_ERR, "Invalid page %lx - out of range\n", page);
199 return pvt->tolm - 1;
202 static void process_ce(struct mem_ctl_info *mci, struct e7xxx_error_info *info)
204 u32 error_1b, page;
205 u16 syndrome;
206 int row;
207 int channel;
209 debugf3("%s()\n", __func__);
210 /* read the error address */
211 error_1b = info->dram_celog_add;
212 /* FIXME - should use PAGE_SHIFT */
213 page = error_1b >> 6; /* convert the address to 4k page */
214 /* read the syndrome */
215 syndrome = info->dram_celog_syndrome;
216 /* FIXME - check for -1 */
217 row = edac_mc_find_csrow_by_page(mci, page);
218 /* convert syndrome to channel */
219 channel = e7xxx_find_channel(syndrome);
220 edac_mc_handle_ce(mci, page, 0, syndrome, row, channel, "e7xxx CE");
223 static void process_ce_no_info(struct mem_ctl_info *mci)
225 debugf3("%s()\n", __func__);
226 edac_mc_handle_ce_no_info(mci, "e7xxx CE log register overflow");
229 static void process_ue(struct mem_ctl_info *mci, struct e7xxx_error_info *info)
231 u32 error_2b, block_page;
232 int row;
234 debugf3("%s()\n", __func__);
235 /* read the error address */
236 error_2b = info->dram_uelog_add;
237 /* FIXME - should use PAGE_SHIFT */
238 block_page = error_2b >> 6; /* convert to 4k address */
239 row = edac_mc_find_csrow_by_page(mci, block_page);
240 edac_mc_handle_ue(mci, block_page, 0, row, "e7xxx UE");
243 static void process_ue_no_info(struct mem_ctl_info *mci)
245 debugf3("%s()\n", __func__);
246 edac_mc_handle_ue_no_info(mci, "e7xxx UE log register overflow");
249 static void e7xxx_get_error_info(struct mem_ctl_info *mci,
250 struct e7xxx_error_info *info)
252 struct e7xxx_pvt *pvt;
254 pvt = (struct e7xxx_pvt *)mci->pvt_info;
255 pci_read_config_byte(pvt->bridge_ck, E7XXX_DRAM_FERR, &info->dram_ferr);
256 pci_read_config_byte(pvt->bridge_ck, E7XXX_DRAM_NERR, &info->dram_nerr);
258 if ((info->dram_ferr & 1) || (info->dram_nerr & 1)) {
259 pci_read_config_dword(pvt->bridge_ck, E7XXX_DRAM_CELOG_ADD,
260 &info->dram_celog_add);
261 pci_read_config_word(pvt->bridge_ck,
262 E7XXX_DRAM_CELOG_SYNDROME,
263 &info->dram_celog_syndrome);
266 if ((info->dram_ferr & 2) || (info->dram_nerr & 2))
267 pci_read_config_dword(pvt->bridge_ck, E7XXX_DRAM_UELOG_ADD,
268 &info->dram_uelog_add);
270 if (info->dram_ferr & 3)
271 pci_write_bits8(pvt->bridge_ck, E7XXX_DRAM_FERR, 0x03, 0x03);
273 if (info->dram_nerr & 3)
274 pci_write_bits8(pvt->bridge_ck, E7XXX_DRAM_NERR, 0x03, 0x03);
277 static int e7xxx_process_error_info(struct mem_ctl_info *mci,
278 struct e7xxx_error_info *info,
279 int handle_errors)
281 int error_found;
283 error_found = 0;
285 /* decode and report errors */
286 if (info->dram_ferr & 1) { /* check first error correctable */
287 error_found = 1;
289 if (handle_errors)
290 process_ce(mci, info);
293 if (info->dram_ferr & 2) { /* check first error uncorrectable */
294 error_found = 1;
296 if (handle_errors)
297 process_ue(mci, info);
300 if (info->dram_nerr & 1) { /* check next error correctable */
301 error_found = 1;
303 if (handle_errors) {
304 if (info->dram_ferr & 1)
305 process_ce_no_info(mci);
306 else
307 process_ce(mci, info);
311 if (info->dram_nerr & 2) { /* check next error uncorrectable */
312 error_found = 1;
314 if (handle_errors) {
315 if (info->dram_ferr & 2)
316 process_ue_no_info(mci);
317 else
318 process_ue(mci, info);
322 return error_found;
325 static void e7xxx_check(struct mem_ctl_info *mci)
327 struct e7xxx_error_info info;
329 debugf3("%s()\n", __func__);
330 e7xxx_get_error_info(mci, &info);
331 e7xxx_process_error_info(mci, &info, 1);
334 /* Return 1 if dual channel mode is active. Else return 0. */
335 static inline int dual_channel_active(u32 drc, int dev_idx)
337 return (dev_idx == E7501) ? ((drc >> 22) & 0x1) : 1;
340 /* Return DRB granularity (0=32mb, 1=64mb). */
341 static inline int drb_granularity(u32 drc, int dev_idx)
343 /* only e7501 can be single channel */
344 return (dev_idx == E7501) ? ((drc >> 18) & 0x3) : 1;
347 static void e7xxx_init_csrows(struct mem_ctl_info *mci, struct pci_dev *pdev,
348 int dev_idx, u32 drc)
350 unsigned long last_cumul_size;
351 int index;
352 u8 value;
353 u32 dra, cumul_size;
354 int drc_chan, drc_drbg, drc_ddim, mem_dev;
355 struct csrow_info *csrow;
357 pci_read_config_dword(pdev, E7XXX_DRA, &dra);
358 drc_chan = dual_channel_active(drc, dev_idx);
359 drc_drbg = drb_granularity(drc, dev_idx);
360 drc_ddim = (drc >> 20) & 0x3;
361 last_cumul_size = 0;
363 /* The dram row boundary (DRB) reg values are boundary address
364 * for each DRAM row with a granularity of 32 or 64MB (single/dual
365 * channel operation). DRB regs are cumulative; therefore DRB7 will
366 * contain the total memory contained in all eight rows.
368 for (index = 0; index < mci->nr_csrows; index++) {
369 /* mem_dev 0=x8, 1=x4 */
370 mem_dev = (dra >> (index * 4 + 3)) & 0x1;
371 csrow = &mci->csrows[index];
373 pci_read_config_byte(pdev, E7XXX_DRB + index, &value);
374 /* convert a 64 or 32 MiB DRB to a page size. */
375 cumul_size = value << (25 + drc_drbg - PAGE_SHIFT);
376 debugf3("%s(): (%d) cumul_size 0x%x\n", __func__, index,
377 cumul_size);
378 if (cumul_size == last_cumul_size)
379 continue; /* not populated */
381 csrow->first_page = last_cumul_size;
382 csrow->last_page = cumul_size - 1;
383 csrow->nr_pages = cumul_size - last_cumul_size;
384 last_cumul_size = cumul_size;
385 csrow->grain = 1 << 12; /* 4KiB - resolution of CELOG */
386 csrow->mtype = MEM_RDDR; /* only one type supported */
387 csrow->dtype = mem_dev ? DEV_X4 : DEV_X8;
390 * if single channel or x8 devices then SECDED
391 * if dual channel and x4 then S4ECD4ED
393 if (drc_ddim) {
394 if (drc_chan && mem_dev) {
395 csrow->edac_mode = EDAC_S4ECD4ED;
396 mci->edac_cap |= EDAC_FLAG_S4ECD4ED;
397 } else {
398 csrow->edac_mode = EDAC_SECDED;
399 mci->edac_cap |= EDAC_FLAG_SECDED;
401 } else
402 csrow->edac_mode = EDAC_NONE;
406 static int e7xxx_probe1(struct pci_dev *pdev, int dev_idx)
408 u16 pci_data;
409 struct mem_ctl_info *mci = NULL;
410 struct e7xxx_pvt *pvt = NULL;
411 u32 drc;
412 int drc_chan;
413 struct e7xxx_error_info discard;
415 debugf0("%s(): mci\n", __func__);
417 /* make sure error reporting method is sane */
418 switch (edac_op_state) {
419 case EDAC_OPSTATE_POLL:
420 case EDAC_OPSTATE_NMI:
421 break;
422 default:
423 edac_op_state = EDAC_OPSTATE_POLL;
424 break;
427 pci_read_config_dword(pdev, E7XXX_DRC, &drc);
429 drc_chan = dual_channel_active(drc, dev_idx);
430 mci = edac_mc_alloc(sizeof(*pvt), E7XXX_NR_CSROWS, drc_chan + 1, 0);
432 if (mci == NULL)
433 return -ENOMEM;
435 debugf3("%s(): init mci\n", __func__);
436 mci->mtype_cap = MEM_FLAG_RDDR;
437 mci->edac_ctl_cap = EDAC_FLAG_NONE | EDAC_FLAG_SECDED |
438 EDAC_FLAG_S4ECD4ED;
439 /* FIXME - what if different memory types are in different csrows? */
440 mci->mod_name = EDAC_MOD_STR;
441 mci->mod_ver = E7XXX_REVISION;
442 mci->dev = &pdev->dev;
443 debugf3("%s(): init pvt\n", __func__);
444 pvt = (struct e7xxx_pvt *)mci->pvt_info;
445 pvt->dev_info = &e7xxx_devs[dev_idx];
446 pvt->bridge_ck = pci_get_device(PCI_VENDOR_ID_INTEL,
447 pvt->dev_info->err_dev, pvt->bridge_ck);
449 if (!pvt->bridge_ck) {
450 e7xxx_printk(KERN_ERR, "error reporting device not found:"
451 "vendor %x device 0x%x (broken BIOS?)\n",
452 PCI_VENDOR_ID_INTEL, e7xxx_devs[dev_idx].err_dev);
453 goto fail0;
456 debugf3("%s(): more mci init\n", __func__);
457 mci->ctl_name = pvt->dev_info->ctl_name;
458 mci->dev_name = pci_name(pdev);
459 mci->edac_check = e7xxx_check;
460 mci->ctl_page_to_phys = ctl_page_to_phys;
461 e7xxx_init_csrows(mci, pdev, dev_idx, drc);
462 mci->edac_cap |= EDAC_FLAG_NONE;
463 debugf3("%s(): tolm, remapbase, remaplimit\n", __func__);
464 /* load the top of low memory, remap base, and remap limit vars */
465 pci_read_config_word(pdev, E7XXX_TOLM, &pci_data);
466 pvt->tolm = ((u32) pci_data) << 4;
467 pci_read_config_word(pdev, E7XXX_REMAPBASE, &pci_data);
468 pvt->remapbase = ((u32) pci_data) << 14;
469 pci_read_config_word(pdev, E7XXX_REMAPLIMIT, &pci_data);
470 pvt->remaplimit = ((u32) pci_data) << 14;
471 e7xxx_printk(KERN_INFO,
472 "tolm = %x, remapbase = %x, remaplimit = %x\n", pvt->tolm,
473 pvt->remapbase, pvt->remaplimit);
475 /* clear any pending errors, or initial state bits */
476 e7xxx_get_error_info(mci, &discard);
478 /* Here we assume that we will never see multiple instances of this
479 * type of memory controller. The ID is therefore hardcoded to 0.
481 if (edac_mc_add_mc(mci)) {
482 debugf3("%s(): failed edac_mc_add_mc()\n", __func__);
483 goto fail1;
486 /* allocating generic PCI control info */
487 e7xxx_pci = edac_pci_create_generic_ctl(&pdev->dev, EDAC_MOD_STR);
488 if (!e7xxx_pci) {
489 printk(KERN_WARNING
490 "%s(): Unable to create PCI control\n",
491 __func__);
492 printk(KERN_WARNING
493 "%s(): PCI error report via EDAC not setup\n",
494 __func__);
497 /* get this far and it's successful */
498 debugf3("%s(): success\n", __func__);
499 return 0;
501 fail1:
502 pci_dev_put(pvt->bridge_ck);
504 fail0:
505 edac_mc_free(mci);
507 return -ENODEV;
510 /* returns count (>= 0), or negative on error */
511 static int __devinit e7xxx_init_one(struct pci_dev *pdev,
512 const struct pci_device_id *ent)
514 debugf0("%s()\n", __func__);
516 /* wake up and enable device */
517 return pci_enable_device(pdev) ?
518 -EIO : e7xxx_probe1(pdev, ent->driver_data);
521 static void __devexit e7xxx_remove_one(struct pci_dev *pdev)
523 struct mem_ctl_info *mci;
524 struct e7xxx_pvt *pvt;
526 debugf0("%s()\n", __func__);
528 if (e7xxx_pci)
529 edac_pci_release_generic_ctl(e7xxx_pci);
531 if ((mci = edac_mc_del_mc(&pdev->dev)) == NULL)
532 return;
534 pvt = (struct e7xxx_pvt *)mci->pvt_info;
535 pci_dev_put(pvt->bridge_ck);
536 edac_mc_free(mci);
539 static const struct pci_device_id e7xxx_pci_tbl[] __devinitdata = {
541 PCI_VEND_DEV(INTEL, 7205_0), PCI_ANY_ID, PCI_ANY_ID, 0, 0,
542 E7205},
544 PCI_VEND_DEV(INTEL, 7500_0), PCI_ANY_ID, PCI_ANY_ID, 0, 0,
545 E7500},
547 PCI_VEND_DEV(INTEL, 7501_0), PCI_ANY_ID, PCI_ANY_ID, 0, 0,
548 E7501},
550 PCI_VEND_DEV(INTEL, 7505_0), PCI_ANY_ID, PCI_ANY_ID, 0, 0,
551 E7505},
554 } /* 0 terminated list. */
557 MODULE_DEVICE_TABLE(pci, e7xxx_pci_tbl);
559 static struct pci_driver e7xxx_driver = {
560 .name = EDAC_MOD_STR,
561 .probe = e7xxx_init_one,
562 .remove = __devexit_p(e7xxx_remove_one),
563 .id_table = e7xxx_pci_tbl,
566 static int __init e7xxx_init(void)
568 return pci_register_driver(&e7xxx_driver);
571 static void __exit e7xxx_exit(void)
573 pci_unregister_driver(&e7xxx_driver);
576 module_init(e7xxx_init);
577 module_exit(e7xxx_exit);
579 MODULE_LICENSE("GPL");
580 MODULE_AUTHOR("Linux Networx (http://lnxi.com) Thayne Harbaugh et al\n"
581 "Based on.work by Dan Hollis et al");
582 MODULE_DESCRIPTION("MC support for Intel e7xxx memory controllers");
583 module_param(edac_op_state, int, 0444);
584 MODULE_PARM_DESC(edac_op_state, "EDAC Error Reporting state: 0=Poll,1=NMI");