Merge branch 'release' of git://lm-sensors.org/kernel/mhoffman/hwmon-2.6
[linux-2.6] / drivers / edac / e7xxx_edac.c
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.
6  *
7  * See "enum e7xxx_chips" below for supported chipsets
8  *
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/
12  *
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)
20  *
21  * $Id: edac_e7xxx.c,v 1.5.2.9 2005/10/05 00:43:44 dsp_llnl Exp $
22  *
23  */
24
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"
32
33 #define E7XXX_REVISION " Ver: 2.0.2 " __DATE__
34 #define EDAC_MOD_STR    "e7xxx_edac"
35
36 #define e7xxx_printk(level, fmt, arg...) \
37         edac_printk(level, "e7xxx", fmt, ##arg)
38
39 #define e7xxx_mc_printk(mci, level, fmt, arg...) \
40         edac_mc_chipset_printk(mci, level, "e7xxx", fmt, ##arg)
41
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 */
45
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 */
49
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 */
53
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 */
57
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 */
61
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 */
65
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 */
69
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 */
73
74 #define E7XXX_NR_CSROWS         8       /* number of csrows */
75 #define E7XXX_NR_DIMMS          8       /* FIXME - is this correct? */
76
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) */
80                                         /*
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
89                                          */
90 #define E7XXX_DRC               0x7C    /* DRAM controller mode reg (32b) */
91                                         /*
92                                          * 22    Number channels 0=1,1=2
93                                          * 19:18 DRB Granularity 32/64MB
94                                          */
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) */
98
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) */
104                                         /*
105                                          * 31:28 Reserved
106                                          * 27:6  CE address (4k block 33:12)
107                                          *  5:0  Reserved
108                                          */
109 #define E7XXX_DRAM_UELOG_ADD    0xB0    /* DRAM first uncorrectable memory */
110                                         /*     error address register (32b) */
111                                         /*
112                                          * 31:28 Reserved
113                                          * 27:6  CE address (4k block 33:12)
114                                          *  5:0  Reserved
115                                          */
116 #define E7XXX_DRAM_CELOG_SYNDROME 0xD0  /* DRAM first correctable memory */
117                                         /*     error syndrome register (16b) */
118
119 enum e7xxx_chips {
120         E7500 = 0,
121         E7501,
122         E7505,
123         E7205,
124 };
125
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;
132 };
133
134 struct e7xxx_dev_info {
135         u16 err_dev;
136         const char *ctl_name;
137 };
138
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;
145 };
146
147 static struct edac_pci_ctl_info *e7xxx_pci;
148
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"},
162 };
163
164 /* FIXME - is this valid for both SECDED and S4ECD4ED? */
165 static inline int e7xxx_find_channel(u16 syndrome)
166 {
167         debugf3("%s()\n", __func__);
168
169         if ((syndrome & 0xff00) == 0)
170                 return 0;
171
172         if ((syndrome & 0x00ff) == 0)
173                 return 1;
174
175         if ((syndrome & 0xf000) == 0 || (syndrome & 0x0f00) == 0)
176                 return 0;
177
178         return 1;
179 }
180
181 static unsigned long ctl_page_to_phys(struct mem_ctl_info *mci,
182                                 unsigned long page)
183 {
184         u32 remap;
185         struct e7xxx_pvt *pvt = (struct e7xxx_pvt *)mci->pvt_info;
186
187         debugf3("%s()\n", __func__);
188
189         if ((page < pvt->tolm) ||
190                 ((page >= 0x100000) && (page < pvt->remapbase)))
191                 return page;
192
193         remap = (page - pvt->tolm) + pvt->remapbase;
194
195         if (remap < pvt->remaplimit)
196                 return remap;
197
198         e7xxx_printk(KERN_ERR, "Invalid page %lx - out of range\n", page);
199         return pvt->tolm - 1;
200 }
201
202 static void process_ce(struct mem_ctl_info *mci, struct e7xxx_error_info *info)
203 {
204         u32 error_1b, page;
205         u16 syndrome;
206         int row;
207         int channel;
208
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");
221 }
222
223 static void process_ce_no_info(struct mem_ctl_info *mci)
224 {
225         debugf3("%s()\n", __func__);
226         edac_mc_handle_ce_no_info(mci, "e7xxx CE log register overflow");
227 }
228
229 static void process_ue(struct mem_ctl_info *mci, struct e7xxx_error_info *info)
230 {
231         u32 error_2b, block_page;
232         int row;
233
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");
241 }
242
243 static void process_ue_no_info(struct mem_ctl_info *mci)
244 {
245         debugf3("%s()\n", __func__);
246         edac_mc_handle_ue_no_info(mci, "e7xxx UE log register overflow");
247 }
248
249 static void e7xxx_get_error_info(struct mem_ctl_info *mci,
250                                  struct e7xxx_error_info *info)
251 {
252         struct e7xxx_pvt *pvt;
253
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);
257
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);
264         }
265
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);
269
270         if (info->dram_ferr & 3)
271                 pci_write_bits8(pvt->bridge_ck, E7XXX_DRAM_FERR, 0x03, 0x03);
272
273         if (info->dram_nerr & 3)
274                 pci_write_bits8(pvt->bridge_ck, E7XXX_DRAM_NERR, 0x03, 0x03);
275 }
276
277 static int e7xxx_process_error_info(struct mem_ctl_info *mci,
278                                 struct e7xxx_error_info *info,
279                                 int handle_errors)
280 {
281         int error_found;
282
283         error_found = 0;
284
285         /* decode and report errors */
286         if (info->dram_ferr & 1) {      /* check first error correctable */
287                 error_found = 1;
288
289                 if (handle_errors)
290                         process_ce(mci, info);
291         }
292
293         if (info->dram_ferr & 2) {      /* check first error uncorrectable */
294                 error_found = 1;
295
296                 if (handle_errors)
297                         process_ue(mci, info);
298         }
299
300         if (info->dram_nerr & 1) {      /* check next error correctable */
301                 error_found = 1;
302
303                 if (handle_errors) {
304                         if (info->dram_ferr & 1)
305                                 process_ce_no_info(mci);
306                         else
307                                 process_ce(mci, info);
308                 }
309         }
310
311         if (info->dram_nerr & 2) {      /* check next error uncorrectable */
312                 error_found = 1;
313
314                 if (handle_errors) {
315                         if (info->dram_ferr & 2)
316                                 process_ue_no_info(mci);
317                         else
318                                 process_ue(mci, info);
319                 }
320         }
321
322         return error_found;
323 }
324
325 static void e7xxx_check(struct mem_ctl_info *mci)
326 {
327         struct e7xxx_error_info info;
328
329         debugf3("%s()\n", __func__);
330         e7xxx_get_error_info(mci, &info);
331         e7xxx_process_error_info(mci, &info, 1);
332 }
333
334 /* Return 1 if dual channel mode is active.  Else return 0. */
335 static inline int dual_channel_active(u32 drc, int dev_idx)
336 {
337         return (dev_idx == E7501) ? ((drc >> 22) & 0x1) : 1;
338 }
339
340 /* Return DRB granularity (0=32mb, 1=64mb). */
341 static inline int drb_granularity(u32 drc, int dev_idx)
342 {
343         /* only e7501 can be single channel */
344         return (dev_idx == E7501) ? ((drc >> 18) & 0x3) : 1;
345 }
346
347 static void e7xxx_init_csrows(struct mem_ctl_info *mci, struct pci_dev *pdev,
348                         int dev_idx, u32 drc)
349 {
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;
356
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;
362
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.
367          */
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];
372
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 */
380
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;
388
389                 /*
390                  * if single channel or x8 devices then SECDED
391                  * if dual channel and x4 then S4ECD4ED
392                  */
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;
400                         }
401                 } else
402                         csrow->edac_mode = EDAC_NONE;
403         }
404 }
405
406 static int e7xxx_probe1(struct pci_dev *pdev, int dev_idx)
407 {
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;
414
415         debugf0("%s(): mci\n", __func__);
416
417         pci_read_config_dword(pdev, E7XXX_DRC, &drc);
418
419         drc_chan = dual_channel_active(drc, dev_idx);
420         mci = edac_mc_alloc(sizeof(*pvt), E7XXX_NR_CSROWS, drc_chan + 1, 0);
421
422         if (mci == NULL)
423                 return -ENOMEM;
424
425         debugf3("%s(): init mci\n", __func__);
426         mci->mtype_cap = MEM_FLAG_RDDR;
427         mci->edac_ctl_cap = EDAC_FLAG_NONE | EDAC_FLAG_SECDED |
428                 EDAC_FLAG_S4ECD4ED;
429         /* FIXME - what if different memory types are in different csrows? */
430         mci->mod_name = EDAC_MOD_STR;
431         mci->mod_ver = E7XXX_REVISION;
432         mci->dev = &pdev->dev;
433         debugf3("%s(): init pvt\n", __func__);
434         pvt = (struct e7xxx_pvt *)mci->pvt_info;
435         pvt->dev_info = &e7xxx_devs[dev_idx];
436         pvt->bridge_ck = pci_get_device(PCI_VENDOR_ID_INTEL,
437                                         pvt->dev_info->err_dev, pvt->bridge_ck);
438
439         if (!pvt->bridge_ck) {
440                 e7xxx_printk(KERN_ERR, "error reporting device not found:"
441                         "vendor %x device 0x%x (broken BIOS?)\n",
442                         PCI_VENDOR_ID_INTEL, e7xxx_devs[dev_idx].err_dev);
443                 goto fail0;
444         }
445
446         debugf3("%s(): more mci init\n", __func__);
447         mci->ctl_name = pvt->dev_info->ctl_name;
448         mci->dev_name = pci_name(pdev);
449         mci->edac_check = e7xxx_check;
450         mci->ctl_page_to_phys = ctl_page_to_phys;
451         e7xxx_init_csrows(mci, pdev, dev_idx, drc);
452         mci->edac_cap |= EDAC_FLAG_NONE;
453         debugf3("%s(): tolm, remapbase, remaplimit\n", __func__);
454         /* load the top of low memory, remap base, and remap limit vars */
455         pci_read_config_word(pdev, E7XXX_TOLM, &pci_data);
456         pvt->tolm = ((u32) pci_data) << 4;
457         pci_read_config_word(pdev, E7XXX_REMAPBASE, &pci_data);
458         pvt->remapbase = ((u32) pci_data) << 14;
459         pci_read_config_word(pdev, E7XXX_REMAPLIMIT, &pci_data);
460         pvt->remaplimit = ((u32) pci_data) << 14;
461         e7xxx_printk(KERN_INFO,
462                 "tolm = %x, remapbase = %x, remaplimit = %x\n", pvt->tolm,
463                 pvt->remapbase, pvt->remaplimit);
464
465         /* clear any pending errors, or initial state bits */
466         e7xxx_get_error_info(mci, &discard);
467
468         /* Here we assume that we will never see multiple instances of this
469          * type of memory controller.  The ID is therefore hardcoded to 0.
470          */
471         if (edac_mc_add_mc(mci)) {
472                 debugf3("%s(): failed edac_mc_add_mc()\n", __func__);
473                 goto fail1;
474         }
475
476         /* allocating generic PCI control info */
477         e7xxx_pci = edac_pci_create_generic_ctl(&pdev->dev, EDAC_MOD_STR);
478         if (!e7xxx_pci) {
479                 printk(KERN_WARNING
480                         "%s(): Unable to create PCI control\n",
481                         __func__);
482                 printk(KERN_WARNING
483                         "%s(): PCI error report via EDAC not setup\n",
484                         __func__);
485         }
486
487         /* get this far and it's successful */
488         debugf3("%s(): success\n", __func__);
489         return 0;
490
491 fail1:
492         pci_dev_put(pvt->bridge_ck);
493
494 fail0:
495         edac_mc_free(mci);
496
497         return -ENODEV;
498 }
499
500 /* returns count (>= 0), or negative on error */
501 static int __devinit e7xxx_init_one(struct pci_dev *pdev,
502                                 const struct pci_device_id *ent)
503 {
504         debugf0("%s()\n", __func__);
505
506         /* wake up and enable device */
507         return pci_enable_device(pdev) ?
508                 -EIO : e7xxx_probe1(pdev, ent->driver_data);
509 }
510
511 static void __devexit e7xxx_remove_one(struct pci_dev *pdev)
512 {
513         struct mem_ctl_info *mci;
514         struct e7xxx_pvt *pvt;
515
516         debugf0("%s()\n", __func__);
517
518         if (e7xxx_pci)
519                 edac_pci_release_generic_ctl(e7xxx_pci);
520
521         if ((mci = edac_mc_del_mc(&pdev->dev)) == NULL)
522                 return;
523
524         pvt = (struct e7xxx_pvt *)mci->pvt_info;
525         pci_dev_put(pvt->bridge_ck);
526         edac_mc_free(mci);
527 }
528
529 static const struct pci_device_id e7xxx_pci_tbl[] __devinitdata = {
530         {
531          PCI_VEND_DEV(INTEL, 7205_0), PCI_ANY_ID, PCI_ANY_ID, 0, 0,
532          E7205},
533         {
534          PCI_VEND_DEV(INTEL, 7500_0), PCI_ANY_ID, PCI_ANY_ID, 0, 0,
535          E7500},
536         {
537          PCI_VEND_DEV(INTEL, 7501_0), PCI_ANY_ID, PCI_ANY_ID, 0, 0,
538          E7501},
539         {
540          PCI_VEND_DEV(INTEL, 7505_0), PCI_ANY_ID, PCI_ANY_ID, 0, 0,
541          E7505},
542         {
543          0,
544          }                      /* 0 terminated list. */
545 };
546
547 MODULE_DEVICE_TABLE(pci, e7xxx_pci_tbl);
548
549 static struct pci_driver e7xxx_driver = {
550         .name = EDAC_MOD_STR,
551         .probe = e7xxx_init_one,
552         .remove = __devexit_p(e7xxx_remove_one),
553         .id_table = e7xxx_pci_tbl,
554 };
555
556 static int __init e7xxx_init(void)
557 {
558        /* Ensure that the OPSTATE is set correctly for POLL or NMI */
559        opstate_init();
560
561         return pci_register_driver(&e7xxx_driver);
562 }
563
564 static void __exit e7xxx_exit(void)
565 {
566         pci_unregister_driver(&e7xxx_driver);
567 }
568
569 module_init(e7xxx_init);
570 module_exit(e7xxx_exit);
571
572 MODULE_LICENSE("GPL");
573 MODULE_AUTHOR("Linux Networx (http://lnxi.com) Thayne Harbaugh et al\n"
574                 "Based on.work by Dan Hollis et al");
575 MODULE_DESCRIPTION("MC support for Intel e7xxx memory controllers");
576 module_param(edac_op_state, int, 0444);
577 MODULE_PARM_DESC(edac_op_state, "EDAC Error Reporting state: 0=Poll,1=NMI");