various bugfixes
[cor_2_6_31.git] / drivers / edac / i82975x_edac.c
blob2eed3ea2cf621303320aec118f0ed3bcfcd9764f
1 /*
2 * Intel 82975X Memory Controller kernel module
3 * (C) 2007 aCarLab (India) Pvt. Ltd. (http://acarlab.com)
4 * (C) 2007 jetzbroadband (http://jetzbroadband.com)
5 * This file may be distributed under the terms of the
6 * GNU General Public License.
8 * Written by Arvind R.
9 * Copied from i82875p_edac.c source:
12 #include <linux/module.h>
13 #include <linux/init.h>
14 #include <linux/pci.h>
15 #include <linux/pci_ids.h>
16 #include <linux/slab.h>
17 #include <linux/edac.h>
18 #include "edac_core.h"
20 #define I82975X_REVISION " Ver: 1.0.0 " __DATE__
21 #define EDAC_MOD_STR "i82975x_edac"
23 #define i82975x_printk(level, fmt, arg...) \
24 edac_printk(level, "i82975x", fmt, ##arg)
26 #define i82975x_mc_printk(mci, level, fmt, arg...) \
27 edac_mc_chipset_printk(mci, level, "i82975x", fmt, ##arg)
29 #ifndef PCI_DEVICE_ID_INTEL_82975_0
30 #define PCI_DEVICE_ID_INTEL_82975_0 0x277c
31 #endif /* PCI_DEVICE_ID_INTEL_82975_0 */
33 #define I82975X_NR_CSROWS(nr_chans) (8/(nr_chans))
35 /* Intel 82975X register addresses - device 0 function 0 - DRAM Controller */
36 #define I82975X_EAP 0x58 /* Dram Error Address Pointer (32b)
38 * 31:7 128 byte cache-line address
39 * 6:1 reserved
40 * 0 0: CH0; 1: CH1
43 #define I82975X_DERRSYN 0x5c /* Dram Error SYNdrome (8b)
45 * 7:0 DRAM ECC Syndrome
48 #define I82975X_DES 0x5d /* Dram ERRor DeSTination (8b)
49 * 0h: Processor Memory Reads
50 * 1h:7h reserved
51 * More - See Page 65 of Intel DocSheet.
54 #define I82975X_ERRSTS 0xc8 /* Error Status Register (16b)
56 * 15:12 reserved
57 * 11 Thermal Sensor Event
58 * 10 reserved
59 * 9 non-DRAM lock error (ndlock)
60 * 8 Refresh Timeout
61 * 7:2 reserved
62 * 1 ECC UE (multibit DRAM error)
63 * 0 ECC CE (singlebit DRAM error)
66 /* Error Reporting is supported by 3 mechanisms:
67 1. DMI SERR generation ( ERRCMD )
68 2. SMI DMI generation ( SMICMD )
69 3. SCI DMI generation ( SCICMD )
70 NOTE: Only ONE of the three must be enabled
72 #define I82975X_ERRCMD 0xca /* Error Command (16b)
74 * 15:12 reserved
75 * 11 Thermal Sensor Event
76 * 10 reserved
77 * 9 non-DRAM lock error (ndlock)
78 * 8 Refresh Timeout
79 * 7:2 reserved
80 * 1 ECC UE (multibit DRAM error)
81 * 0 ECC CE (singlebit DRAM error)
84 #define I82975X_SMICMD 0xcc /* Error Command (16b)
86 * 15:2 reserved
87 * 1 ECC UE (multibit DRAM error)
88 * 0 ECC CE (singlebit DRAM error)
91 #define I82975X_SCICMD 0xce /* Error Command (16b)
93 * 15:2 reserved
94 * 1 ECC UE (multibit DRAM error)
95 * 0 ECC CE (singlebit DRAM error)
98 #define I82975X_XEAP 0xfc /* Extended Dram Error Address Pointer (8b)
100 * 7:1 reserved
101 * 0 Bit32 of the Dram Error Address
104 #define I82975X_MCHBAR 0x44 /*
106 * 31:14 Base Addr of 16K memory-mapped
107 * configuration space
108 * 13:1 reserverd
109 * 0 mem-mapped config space enable
112 /* NOTE: Following addresses have to indexed using MCHBAR offset (44h, 32b) */
113 /* Intel 82975x memory mapped register space */
115 #define I82975X_DRB_SHIFT 25 /* fixed 32MiB grain */
117 #define I82975X_DRB 0x100 /* DRAM Row Boundary (8b x 8)
119 * 7 set to 1 in highest DRB of
120 * channel if 4GB in ch.
121 * 6:2 upper boundary of rank in
122 * 32MB grains
123 * 1:0 set to 0
125 #define I82975X_DRB_CH0R0 0x100
126 #define I82975X_DRB_CH0R1 0x101
127 #define I82975X_DRB_CH0R2 0x102
128 #define I82975X_DRB_CH0R3 0x103
129 #define I82975X_DRB_CH1R0 0x180
130 #define I82975X_DRB_CH1R1 0x181
131 #define I82975X_DRB_CH1R2 0x182
132 #define I82975X_DRB_CH1R3 0x183
135 #define I82975X_DRA 0x108 /* DRAM Row Attribute (4b x 8)
136 * defines the PAGE SIZE to be used
137 * for the rank
138 * 7 reserved
139 * 6:4 row attr of odd rank, i.e. 1
140 * 3 reserved
141 * 2:0 row attr of even rank, i.e. 0
143 * 000 = unpopulated
144 * 001 = reserved
145 * 010 = 4KiB
146 * 011 = 8KiB
147 * 100 = 16KiB
148 * others = reserved
150 #define I82975X_DRA_CH0R01 0x108
151 #define I82975X_DRA_CH0R23 0x109
152 #define I82975X_DRA_CH1R01 0x188
153 #define I82975X_DRA_CH1R23 0x189
156 #define I82975X_BNKARC 0x10e /* Type of device in each rank - Bank Arch (16b)
158 * 15:8 reserved
159 * 7:6 Rank 3 architecture
160 * 5:4 Rank 2 architecture
161 * 3:2 Rank 1 architecture
162 * 1:0 Rank 0 architecture
164 * 00 => x16 devices; i.e 4 banks
165 * 01 => x8 devices; i.e 8 banks
167 #define I82975X_C0BNKARC 0x10e
168 #define I82975X_C1BNKARC 0x18e
172 #define I82975X_DRC 0x120 /* DRAM Controller Mode0 (32b)
174 * 31:30 reserved
175 * 29 init complete
176 * 28:11 reserved, according to Intel
177 * 22:21 number of channels
178 * 00=1 01=2 in 82875
179 * seems to be ECC mode
180 * bits in 82975 in Asus
181 * P5W
182 * 19:18 Data Integ Mode
183 * 00=none 01=ECC in 82875
184 * 10:8 refresh mode
185 * 7 reserved
186 * 6:4 mode select
187 * 3:2 reserved
188 * 1:0 DRAM type 10=Second Revision
189 * DDR2 SDRAM
190 * 00, 01, 11 reserved
192 #define I82975X_DRC_CH0M0 0x120
193 #define I82975X_DRC_CH1M0 0x1A0
196 #define I82975X_DRC_M1 0x124 /* DRAM Controller Mode1 (32b)
197 * 31 0=Standard Address Map
198 * 1=Enhanced Address Map
199 * 30:0 reserved
202 #define I82975X_DRC_CH0M1 0x124
203 #define I82975X_DRC_CH1M1 0x1A4
205 enum i82975x_chips {
206 I82975X = 0,
209 struct i82975x_pvt {
210 void __iomem *mch_window;
213 struct i82975x_dev_info {
214 const char *ctl_name;
217 struct i82975x_error_info {
218 u16 errsts;
219 u32 eap;
220 u8 des;
221 u8 derrsyn;
222 u16 errsts2;
223 u8 chan; /* the channel is bit 0 of EAP */
224 u8 xeap; /* extended eap bit */
227 static const struct i82975x_dev_info i82975x_devs[] = {
228 [I82975X] = {
229 .ctl_name = "i82975x"
233 static struct pci_dev *mci_pdev; /* init dev: in case that AGP code has
234 * already registered driver
237 static int i82975x_registered = 1;
239 static void i82975x_get_error_info(struct mem_ctl_info *mci,
240 struct i82975x_error_info *info)
242 struct pci_dev *pdev;
244 pdev = to_pci_dev(mci->dev);
247 * This is a mess because there is no atomic way to read all the
248 * registers at once and the registers can transition from CE being
249 * overwritten by UE.
251 pci_read_config_word(pdev, I82975X_ERRSTS, &info->errsts);
252 pci_read_config_dword(pdev, I82975X_EAP, &info->eap);
253 pci_read_config_byte(pdev, I82975X_XEAP, &info->xeap);
254 pci_read_config_byte(pdev, I82975X_DES, &info->des);
255 pci_read_config_byte(pdev, I82975X_DERRSYN, &info->derrsyn);
256 pci_read_config_word(pdev, I82975X_ERRSTS, &info->errsts2);
258 pci_write_bits16(pdev, I82975X_ERRSTS, 0x0003, 0x0003);
261 * If the error is the same then we can for both reads then
262 * the first set of reads is valid. If there is a change then
263 * there is a CE no info and the second set of reads is valid
264 * and should be UE info.
266 if (!(info->errsts2 & 0x0003))
267 return;
269 if ((info->errsts ^ info->errsts2) & 0x0003) {
270 pci_read_config_dword(pdev, I82975X_EAP, &info->eap);
271 pci_read_config_byte(pdev, I82975X_XEAP, &info->xeap);
272 pci_read_config_byte(pdev, I82975X_DES, &info->des);
273 pci_read_config_byte(pdev, I82975X_DERRSYN,
274 &info->derrsyn);
278 static int i82975x_process_error_info(struct mem_ctl_info *mci,
279 struct i82975x_error_info *info, int handle_errors)
281 int row, multi_chan, chan;
283 multi_chan = mci->csrows[0].nr_channels - 1;
285 if (!(info->errsts2 & 0x0003))
286 return 0;
288 if (!handle_errors)
289 return 1;
291 if ((info->errsts ^ info->errsts2) & 0x0003) {
292 edac_mc_handle_ce_no_info(mci, "UE overwrote CE");
293 info->errsts = info->errsts2;
296 chan = info->eap & 1;
297 info->eap >>= 1;
298 if (info->xeap )
299 info->eap |= 0x80000000;
300 info->eap >>= PAGE_SHIFT;
301 row = edac_mc_find_csrow_by_page(mci, info->eap);
303 if (info->errsts & 0x0002)
304 edac_mc_handle_ue(mci, info->eap, 0, row, "i82975x UE");
305 else
306 edac_mc_handle_ce(mci, info->eap, 0, info->derrsyn, row,
307 multi_chan ? chan : 0,
308 "i82975x CE");
310 return 1;
313 static void i82975x_check(struct mem_ctl_info *mci)
315 struct i82975x_error_info info;
317 debugf1("MC%d: %s()\n", mci->mc_idx, __func__);
318 i82975x_get_error_info(mci, &info);
319 i82975x_process_error_info(mci, &info, 1);
322 /* Return 1 if dual channel mode is active. Else return 0. */
323 static int dual_channel_active(void __iomem *mch_window)
326 * We treat interleaved-symmetric configuration as dual-channel - EAP's
327 * bit-0 giving the channel of the error location.
329 * All other configurations are treated as single channel - the EAP's
330 * bit-0 will resolve ok in symmetric area of mixed
331 * (symmetric/asymmetric) configurations
333 u8 drb[4][2];
334 int row;
335 int dualch;
337 for (dualch = 1, row = 0; dualch && (row < 4); row++) {
338 drb[row][0] = readb(mch_window + I82975X_DRB + row);
339 drb[row][1] = readb(mch_window + I82975X_DRB + row + 0x80);
340 dualch = dualch && (drb[row][0] == drb[row][1]);
342 return dualch;
345 static enum dev_type i82975x_dram_type(void __iomem *mch_window, int rank)
348 * ASUS P5W DH either does not program this register or programs
349 * it wrong!
350 * ECC is possible on i92975x ONLY with DEV_X8 which should mean 'val'
351 * for each rank should be 01b - the LSB of the word should be 0x55;
352 * but it reads 0!
354 return DEV_X8;
357 static void i82975x_init_csrows(struct mem_ctl_info *mci,
358 struct pci_dev *pdev, void __iomem *mch_window)
360 struct csrow_info *csrow;
361 unsigned long last_cumul_size;
362 u8 value;
363 u32 cumul_size;
364 int index;
366 last_cumul_size = 0;
369 * 82875 comment:
370 * The dram row boundary (DRB) reg values are boundary address
371 * for each DRAM row with a granularity of 32 or 64MB (single/dual
372 * channel operation). DRB regs are cumulative; therefore DRB7 will
373 * contain the total memory contained in all eight rows.
375 * FIXME:
376 * EDAC currently works for Dual-channel Interleaved configuration.
377 * Other configurations, which the chip supports, need fixing/testing.
381 for (index = 0; index < mci->nr_csrows; index++) {
382 csrow = &mci->csrows[index];
384 value = readb(mch_window + I82975X_DRB + index +
385 ((index >= 4) ? 0x80 : 0));
386 cumul_size = value;
387 cumul_size <<= (I82975X_DRB_SHIFT - PAGE_SHIFT);
388 debugf3("%s(): (%d) cumul_size 0x%x\n", __func__, index,
389 cumul_size);
390 if (cumul_size == last_cumul_size)
391 continue; /* not populated */
393 csrow->first_page = last_cumul_size;
394 csrow->last_page = cumul_size - 1;
395 csrow->nr_pages = cumul_size - last_cumul_size;
396 last_cumul_size = cumul_size;
397 csrow->grain = 1 << 7; /* I82975X_EAP has 128B resolution */
398 csrow->mtype = MEM_DDR; /* i82975x supports only DDR2 */
399 csrow->dtype = i82975x_dram_type(mch_window, index);
400 csrow->edac_mode = EDAC_SECDED; /* only supported */
404 /* #define i82975x_DEBUG_IOMEM */
406 #ifdef i82975x_DEBUG_IOMEM
407 static void i82975x_print_dram_timings(void __iomem *mch_window)
410 * The register meanings are from Intel specs;
411 * (shows 13-5-5-5 for 800-DDR2)
412 * Asus P5W Bios reports 15-5-4-4
413 * What's your religion?
415 static const int caslats[4] = { 5, 4, 3, 6 };
416 u32 dtreg[2];
418 dtreg[0] = readl(mch_window + 0x114);
419 dtreg[1] = readl(mch_window + 0x194);
420 i82975x_printk(KERN_INFO, "DRAM Timings : Ch0 Ch1\n"
421 " RAS Active Min = %d %d\n"
422 " CAS latency = %d %d\n"
423 " RAS to CAS = %d %d\n"
424 " RAS precharge = %d %d\n",
425 (dtreg[0] >> 19 ) & 0x0f,
426 (dtreg[1] >> 19) & 0x0f,
427 caslats[(dtreg[0] >> 8) & 0x03],
428 caslats[(dtreg[1] >> 8) & 0x03],
429 ((dtreg[0] >> 4) & 0x07) + 2,
430 ((dtreg[1] >> 4) & 0x07) + 2,
431 (dtreg[0] & 0x07) + 2,
432 (dtreg[1] & 0x07) + 2
436 #endif
438 static int i82975x_probe1(struct pci_dev *pdev, int dev_idx)
440 int rc = -ENODEV;
441 struct mem_ctl_info *mci;
442 struct i82975x_pvt *pvt;
443 void __iomem *mch_window;
444 u32 mchbar;
445 u32 drc[2];
446 struct i82975x_error_info discard;
447 int chans;
448 #ifdef i82975x_DEBUG_IOMEM
449 u8 c0drb[4];
450 u8 c1drb[4];
451 #endif
453 debugf0("%s()\n", __func__);
455 pci_read_config_dword(pdev, I82975X_MCHBAR, &mchbar);
456 if (!(mchbar & 1)) {
457 debugf3("%s(): failed, MCHBAR disabled!\n", __func__);
458 goto fail0;
460 mchbar &= 0xffffc000; /* bits 31:14 used for 16K window */
461 mch_window = ioremap_nocache(mchbar, 0x1000);
463 #ifdef i82975x_DEBUG_IOMEM
464 i82975x_printk(KERN_INFO, "MCHBAR real = %0x, remapped = %p\n",
465 mchbar, mch_window);
467 c0drb[0] = readb(mch_window + I82975X_DRB_CH0R0);
468 c0drb[1] = readb(mch_window + I82975X_DRB_CH0R1);
469 c0drb[2] = readb(mch_window + I82975X_DRB_CH0R2);
470 c0drb[3] = readb(mch_window + I82975X_DRB_CH0R3);
471 c1drb[0] = readb(mch_window + I82975X_DRB_CH1R0);
472 c1drb[1] = readb(mch_window + I82975X_DRB_CH1R1);
473 c1drb[2] = readb(mch_window + I82975X_DRB_CH1R2);
474 c1drb[3] = readb(mch_window + I82975X_DRB_CH1R3);
475 i82975x_printk(KERN_INFO, "DRBCH0R0 = 0x%02x\n", c0drb[0]);
476 i82975x_printk(KERN_INFO, "DRBCH0R1 = 0x%02x\n", c0drb[1]);
477 i82975x_printk(KERN_INFO, "DRBCH0R2 = 0x%02x\n", c0drb[2]);
478 i82975x_printk(KERN_INFO, "DRBCH0R3 = 0x%02x\n", c0drb[3]);
479 i82975x_printk(KERN_INFO, "DRBCH1R0 = 0x%02x\n", c1drb[0]);
480 i82975x_printk(KERN_INFO, "DRBCH1R1 = 0x%02x\n", c1drb[1]);
481 i82975x_printk(KERN_INFO, "DRBCH1R2 = 0x%02x\n", c1drb[2]);
482 i82975x_printk(KERN_INFO, "DRBCH1R3 = 0x%02x\n", c1drb[3]);
483 #endif
485 drc[0] = readl(mch_window + I82975X_DRC_CH0M0);
486 drc[1] = readl(mch_window + I82975X_DRC_CH1M0);
487 #ifdef i82975x_DEBUG_IOMEM
488 i82975x_printk(KERN_INFO, "DRC_CH0 = %0x, %s\n", drc[0],
489 ((drc[0] >> 21) & 3) == 1 ?
490 "ECC enabled" : "ECC disabled");
491 i82975x_printk(KERN_INFO, "DRC_CH1 = %0x, %s\n", drc[1],
492 ((drc[1] >> 21) & 3) == 1 ?
493 "ECC enabled" : "ECC disabled");
495 i82975x_printk(KERN_INFO, "C0 BNKARC = %0x\n",
496 readw(mch_window + I82975X_C0BNKARC));
497 i82975x_printk(KERN_INFO, "C1 BNKARC = %0x\n",
498 readw(mch_window + I82975X_C1BNKARC));
499 i82975x_print_dram_timings(mch_window);
500 goto fail1;
501 #endif
502 if (!(((drc[0] >> 21) & 3) == 1 || ((drc[1] >> 21) & 3) == 1)) {
503 i82975x_printk(KERN_INFO, "ECC disabled on both channels.\n");
504 goto fail1;
507 chans = dual_channel_active(mch_window) + 1;
509 /* assuming only one controller, index thus is 0 */
510 mci = edac_mc_alloc(sizeof(*pvt), I82975X_NR_CSROWS(chans),
511 chans, 0);
512 if (!mci) {
513 rc = -ENOMEM;
514 goto fail1;
517 debugf3("%s(): init mci\n", __func__);
518 mci->dev = &pdev->dev;
519 mci->mtype_cap = MEM_FLAG_DDR;
520 mci->edac_ctl_cap = EDAC_FLAG_NONE | EDAC_FLAG_SECDED;
521 mci->edac_cap = EDAC_FLAG_NONE | EDAC_FLAG_SECDED;
522 mci->mod_name = EDAC_MOD_STR;
523 mci->mod_ver = I82975X_REVISION;
524 mci->ctl_name = i82975x_devs[dev_idx].ctl_name;
525 mci->edac_check = i82975x_check;
526 mci->ctl_page_to_phys = NULL;
527 debugf3("%s(): init pvt\n", __func__);
528 pvt = (struct i82975x_pvt *) mci->pvt_info;
529 pvt->mch_window = mch_window;
530 i82975x_init_csrows(mci, pdev, mch_window);
531 i82975x_get_error_info(mci, &discard); /* clear counters */
533 /* finalize this instance of memory controller with edac core */
534 if (edac_mc_add_mc(mci)) {
535 debugf3("%s(): failed edac_mc_add_mc()\n", __func__);
536 goto fail2;
539 /* get this far and it's successful */
540 debugf3("%s(): success\n", __func__);
541 return 0;
543 fail2:
544 edac_mc_free(mci);
546 fail1:
547 iounmap(mch_window);
548 fail0:
549 return rc;
552 /* returns count (>= 0), or negative on error */
553 static int __devinit i82975x_init_one(struct pci_dev *pdev,
554 const struct pci_device_id *ent)
556 int rc;
558 debugf0("%s()\n", __func__);
560 if (pci_enable_device(pdev) < 0)
561 return -EIO;
563 rc = i82975x_probe1(pdev, ent->driver_data);
565 if (mci_pdev == NULL)
566 mci_pdev = pci_dev_get(pdev);
568 return rc;
571 static void __devexit i82975x_remove_one(struct pci_dev *pdev)
573 struct mem_ctl_info *mci;
574 struct i82975x_pvt *pvt;
576 debugf0("%s()\n", __func__);
578 mci = edac_mc_del_mc(&pdev->dev);
579 if (mci == NULL)
580 return;
582 pvt = mci->pvt_info;
583 if (pvt->mch_window)
584 iounmap( pvt->mch_window );
586 edac_mc_free(mci);
589 static const struct pci_device_id i82975x_pci_tbl[] __devinitdata = {
591 PCI_VEND_DEV(INTEL, 82975_0), PCI_ANY_ID, PCI_ANY_ID, 0, 0,
592 I82975X
596 } /* 0 terminated list. */
599 MODULE_DEVICE_TABLE(pci, i82975x_pci_tbl);
601 static struct pci_driver i82975x_driver = {
602 .name = EDAC_MOD_STR,
603 .probe = i82975x_init_one,
604 .remove = __devexit_p(i82975x_remove_one),
605 .id_table = i82975x_pci_tbl,
608 static int __init i82975x_init(void)
610 int pci_rc;
612 debugf3("%s()\n", __func__);
614 /* Ensure that the OPSTATE is set correctly for POLL or NMI */
615 opstate_init();
617 pci_rc = pci_register_driver(&i82975x_driver);
618 if (pci_rc < 0)
619 goto fail0;
621 if (mci_pdev == NULL) {
622 mci_pdev = pci_get_device(PCI_VENDOR_ID_INTEL,
623 PCI_DEVICE_ID_INTEL_82975_0, NULL);
625 if (!mci_pdev) {
626 debugf0("i82975x pci_get_device fail\n");
627 pci_rc = -ENODEV;
628 goto fail1;
631 pci_rc = i82975x_init_one(mci_pdev, i82975x_pci_tbl);
633 if (pci_rc < 0) {
634 debugf0("i82975x init fail\n");
635 pci_rc = -ENODEV;
636 goto fail1;
640 return 0;
642 fail1:
643 pci_unregister_driver(&i82975x_driver);
645 fail0:
646 if (mci_pdev != NULL)
647 pci_dev_put(mci_pdev);
649 return pci_rc;
652 static void __exit i82975x_exit(void)
654 debugf3("%s()\n", __func__);
656 pci_unregister_driver(&i82975x_driver);
658 if (!i82975x_registered) {
659 i82975x_remove_one(mci_pdev);
660 pci_dev_put(mci_pdev);
664 module_init(i82975x_init);
665 module_exit(i82975x_exit);
667 MODULE_LICENSE("GPL");
668 MODULE_AUTHOR("Arvind R. <arvind@acarlab.com>");
669 MODULE_DESCRIPTION("MC support for Intel 82975 memory hub controllers");
671 module_param(edac_op_state, int, 0444);
672 MODULE_PARM_DESC(edac_op_state, "EDAC Error Reporting state: 0=Poll,1=NMI");