beceem: use kernel print_hex_dump function
[linux-2.6/x86.git] / drivers / edac / i82975x_edac.c
blob3218819b7286008c908f53869506a73063fa9a57
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/edac.h>
17 #include "edac_core.h"
19 #define I82975X_REVISION " Ver: 1.0.0 " __DATE__
20 #define EDAC_MOD_STR "i82975x_edac"
22 #define i82975x_printk(level, fmt, arg...) \
23 edac_printk(level, "i82975x", fmt, ##arg)
25 #define i82975x_mc_printk(mci, level, fmt, arg...) \
26 edac_mc_chipset_printk(mci, level, "i82975x", fmt, ##arg)
28 #ifndef PCI_DEVICE_ID_INTEL_82975_0
29 #define PCI_DEVICE_ID_INTEL_82975_0 0x277c
30 #endif /* PCI_DEVICE_ID_INTEL_82975_0 */
32 #define I82975X_NR_CSROWS(nr_chans) (8/(nr_chans))
34 /* Intel 82975X register addresses - device 0 function 0 - DRAM Controller */
35 #define I82975X_EAP 0x58 /* Dram Error Address Pointer (32b)
37 * 31:7 128 byte cache-line address
38 * 6:1 reserved
39 * 0 0: CH0; 1: CH1
42 #define I82975X_DERRSYN 0x5c /* Dram Error SYNdrome (8b)
44 * 7:0 DRAM ECC Syndrome
47 #define I82975X_DES 0x5d /* Dram ERRor DeSTination (8b)
48 * 0h: Processor Memory Reads
49 * 1h:7h reserved
50 * More - See Page 65 of Intel DocSheet.
53 #define I82975X_ERRSTS 0xc8 /* Error Status Register (16b)
55 * 15:12 reserved
56 * 11 Thermal Sensor Event
57 * 10 reserved
58 * 9 non-DRAM lock error (ndlock)
59 * 8 Refresh Timeout
60 * 7:2 reserved
61 * 1 ECC UE (multibit DRAM error)
62 * 0 ECC CE (singlebit DRAM error)
65 /* Error Reporting is supported by 3 mechanisms:
66 1. DMI SERR generation ( ERRCMD )
67 2. SMI DMI generation ( SMICMD )
68 3. SCI DMI generation ( SCICMD )
69 NOTE: Only ONE of the three must be enabled
71 #define I82975X_ERRCMD 0xca /* Error Command (16b)
73 * 15:12 reserved
74 * 11 Thermal Sensor Event
75 * 10 reserved
76 * 9 non-DRAM lock error (ndlock)
77 * 8 Refresh Timeout
78 * 7:2 reserved
79 * 1 ECC UE (multibit DRAM error)
80 * 0 ECC CE (singlebit DRAM error)
83 #define I82975X_SMICMD 0xcc /* Error Command (16b)
85 * 15:2 reserved
86 * 1 ECC UE (multibit DRAM error)
87 * 0 ECC CE (singlebit DRAM error)
90 #define I82975X_SCICMD 0xce /* Error Command (16b)
92 * 15:2 reserved
93 * 1 ECC UE (multibit DRAM error)
94 * 0 ECC CE (singlebit DRAM error)
97 #define I82975X_XEAP 0xfc /* Extended Dram Error Address Pointer (8b)
99 * 7:1 reserved
100 * 0 Bit32 of the Dram Error Address
103 #define I82975X_MCHBAR 0x44 /*
105 * 31:14 Base Addr of 16K memory-mapped
106 * configuration space
107 * 13:1 reserverd
108 * 0 mem-mapped config space enable
111 /* NOTE: Following addresses have to indexed using MCHBAR offset (44h, 32b) */
112 /* Intel 82975x memory mapped register space */
114 #define I82975X_DRB_SHIFT 25 /* fixed 32MiB grain */
116 #define I82975X_DRB 0x100 /* DRAM Row Boundary (8b x 8)
118 * 7 set to 1 in highest DRB of
119 * channel if 4GB in ch.
120 * 6:2 upper boundary of rank in
121 * 32MB grains
122 * 1:0 set to 0
124 #define I82975X_DRB_CH0R0 0x100
125 #define I82975X_DRB_CH0R1 0x101
126 #define I82975X_DRB_CH0R2 0x102
127 #define I82975X_DRB_CH0R3 0x103
128 #define I82975X_DRB_CH1R0 0x180
129 #define I82975X_DRB_CH1R1 0x181
130 #define I82975X_DRB_CH1R2 0x182
131 #define I82975X_DRB_CH1R3 0x183
134 #define I82975X_DRA 0x108 /* DRAM Row Attribute (4b x 8)
135 * defines the PAGE SIZE to be used
136 * for the rank
137 * 7 reserved
138 * 6:4 row attr of odd rank, i.e. 1
139 * 3 reserved
140 * 2:0 row attr of even rank, i.e. 0
142 * 000 = unpopulated
143 * 001 = reserved
144 * 010 = 4KiB
145 * 011 = 8KiB
146 * 100 = 16KiB
147 * others = reserved
149 #define I82975X_DRA_CH0R01 0x108
150 #define I82975X_DRA_CH0R23 0x109
151 #define I82975X_DRA_CH1R01 0x188
152 #define I82975X_DRA_CH1R23 0x189
155 #define I82975X_BNKARC 0x10e /* Type of device in each rank - Bank Arch (16b)
157 * 15:8 reserved
158 * 7:6 Rank 3 architecture
159 * 5:4 Rank 2 architecture
160 * 3:2 Rank 1 architecture
161 * 1:0 Rank 0 architecture
163 * 00 => x16 devices; i.e 4 banks
164 * 01 => x8 devices; i.e 8 banks
166 #define I82975X_C0BNKARC 0x10e
167 #define I82975X_C1BNKARC 0x18e
171 #define I82975X_DRC 0x120 /* DRAM Controller Mode0 (32b)
173 * 31:30 reserved
174 * 29 init complete
175 * 28:11 reserved, according to Intel
176 * 22:21 number of channels
177 * 00=1 01=2 in 82875
178 * seems to be ECC mode
179 * bits in 82975 in Asus
180 * P5W
181 * 19:18 Data Integ Mode
182 * 00=none 01=ECC in 82875
183 * 10:8 refresh mode
184 * 7 reserved
185 * 6:4 mode select
186 * 3:2 reserved
187 * 1:0 DRAM type 10=Second Revision
188 * DDR2 SDRAM
189 * 00, 01, 11 reserved
191 #define I82975X_DRC_CH0M0 0x120
192 #define I82975X_DRC_CH1M0 0x1A0
195 #define I82975X_DRC_M1 0x124 /* DRAM Controller Mode1 (32b)
196 * 31 0=Standard Address Map
197 * 1=Enhanced Address Map
198 * 30:0 reserved
201 #define I82975X_DRC_CH0M1 0x124
202 #define I82975X_DRC_CH1M1 0x1A4
204 enum i82975x_chips {
205 I82975X = 0,
208 struct i82975x_pvt {
209 void __iomem *mch_window;
212 struct i82975x_dev_info {
213 const char *ctl_name;
216 struct i82975x_error_info {
217 u16 errsts;
218 u32 eap;
219 u8 des;
220 u8 derrsyn;
221 u16 errsts2;
222 u8 chan; /* the channel is bit 0 of EAP */
223 u8 xeap; /* extended eap bit */
226 static const struct i82975x_dev_info i82975x_devs[] = {
227 [I82975X] = {
228 .ctl_name = "i82975x"
232 static struct pci_dev *mci_pdev; /* init dev: in case that AGP code has
233 * already registered driver
236 static int i82975x_registered = 1;
238 static void i82975x_get_error_info(struct mem_ctl_info *mci,
239 struct i82975x_error_info *info)
241 struct pci_dev *pdev;
243 pdev = to_pci_dev(mci->dev);
246 * This is a mess because there is no atomic way to read all the
247 * registers at once and the registers can transition from CE being
248 * overwritten by UE.
250 pci_read_config_word(pdev, I82975X_ERRSTS, &info->errsts);
251 pci_read_config_dword(pdev, I82975X_EAP, &info->eap);
252 pci_read_config_byte(pdev, I82975X_XEAP, &info->xeap);
253 pci_read_config_byte(pdev, I82975X_DES, &info->des);
254 pci_read_config_byte(pdev, I82975X_DERRSYN, &info->derrsyn);
255 pci_read_config_word(pdev, I82975X_ERRSTS, &info->errsts2);
257 pci_write_bits16(pdev, I82975X_ERRSTS, 0x0003, 0x0003);
260 * If the error is the same then we can for both reads then
261 * the first set of reads is valid. If there is a change then
262 * there is a CE no info and the second set of reads is valid
263 * and should be UE info.
265 if (!(info->errsts2 & 0x0003))
266 return;
268 if ((info->errsts ^ info->errsts2) & 0x0003) {
269 pci_read_config_dword(pdev, I82975X_EAP, &info->eap);
270 pci_read_config_byte(pdev, I82975X_XEAP, &info->xeap);
271 pci_read_config_byte(pdev, I82975X_DES, &info->des);
272 pci_read_config_byte(pdev, I82975X_DERRSYN,
273 &info->derrsyn);
277 static int i82975x_process_error_info(struct mem_ctl_info *mci,
278 struct i82975x_error_info *info, int handle_errors)
280 int row, multi_chan, chan;
282 multi_chan = mci->csrows[0].nr_channels - 1;
284 if (!(info->errsts2 & 0x0003))
285 return 0;
287 if (!handle_errors)
288 return 1;
290 if ((info->errsts ^ info->errsts2) & 0x0003) {
291 edac_mc_handle_ce_no_info(mci, "UE overwrote CE");
292 info->errsts = info->errsts2;
295 chan = info->eap & 1;
296 info->eap >>= 1;
297 if (info->xeap )
298 info->eap |= 0x80000000;
299 info->eap >>= PAGE_SHIFT;
300 row = edac_mc_find_csrow_by_page(mci, info->eap);
302 if (info->errsts & 0x0002)
303 edac_mc_handle_ue(mci, info->eap, 0, row, "i82975x UE");
304 else
305 edac_mc_handle_ce(mci, info->eap, 0, info->derrsyn, row,
306 multi_chan ? chan : 0,
307 "i82975x CE");
309 return 1;
312 static void i82975x_check(struct mem_ctl_info *mci)
314 struct i82975x_error_info info;
316 debugf1("MC%d: %s()\n", mci->mc_idx, __func__);
317 i82975x_get_error_info(mci, &info);
318 i82975x_process_error_info(mci, &info, 1);
321 /* Return 1 if dual channel mode is active. Else return 0. */
322 static int dual_channel_active(void __iomem *mch_window)
325 * We treat interleaved-symmetric configuration as dual-channel - EAP's
326 * bit-0 giving the channel of the error location.
328 * All other configurations are treated as single channel - the EAP's
329 * bit-0 will resolve ok in symmetric area of mixed
330 * (symmetric/asymmetric) configurations
332 u8 drb[4][2];
333 int row;
334 int dualch;
336 for (dualch = 1, row = 0; dualch && (row < 4); row++) {
337 drb[row][0] = readb(mch_window + I82975X_DRB + row);
338 drb[row][1] = readb(mch_window + I82975X_DRB + row + 0x80);
339 dualch = dualch && (drb[row][0] == drb[row][1]);
341 return dualch;
344 static enum dev_type i82975x_dram_type(void __iomem *mch_window, int rank)
347 * ASUS P5W DH either does not program this register or programs
348 * it wrong!
349 * ECC is possible on i92975x ONLY with DEV_X8 which should mean 'val'
350 * for each rank should be 01b - the LSB of the word should be 0x55;
351 * but it reads 0!
353 return DEV_X8;
356 static void i82975x_init_csrows(struct mem_ctl_info *mci,
357 struct pci_dev *pdev, void __iomem *mch_window)
359 struct csrow_info *csrow;
360 unsigned long last_cumul_size;
361 u8 value;
362 u32 cumul_size;
363 int index;
365 last_cumul_size = 0;
368 * 82875 comment:
369 * The dram row boundary (DRB) reg values are boundary address
370 * for each DRAM row with a granularity of 32 or 64MB (single/dual
371 * channel operation). DRB regs are cumulative; therefore DRB7 will
372 * contain the total memory contained in all eight rows.
374 * FIXME:
375 * EDAC currently works for Dual-channel Interleaved configuration.
376 * Other configurations, which the chip supports, need fixing/testing.
380 for (index = 0; index < mci->nr_csrows; index++) {
381 csrow = &mci->csrows[index];
383 value = readb(mch_window + I82975X_DRB + index +
384 ((index >= 4) ? 0x80 : 0));
385 cumul_size = value;
386 cumul_size <<= (I82975X_DRB_SHIFT - PAGE_SHIFT);
387 debugf3("%s(): (%d) cumul_size 0x%x\n", __func__, index,
388 cumul_size);
389 if (cumul_size == last_cumul_size)
390 continue; /* not populated */
392 csrow->first_page = last_cumul_size;
393 csrow->last_page = cumul_size - 1;
394 csrow->nr_pages = cumul_size - last_cumul_size;
395 last_cumul_size = cumul_size;
396 csrow->grain = 1 << 7; /* I82975X_EAP has 128B resolution */
397 csrow->mtype = MEM_DDR; /* i82975x supports only DDR2 */
398 csrow->dtype = i82975x_dram_type(mch_window, index);
399 csrow->edac_mode = EDAC_SECDED; /* only supported */
403 /* #define i82975x_DEBUG_IOMEM */
405 #ifdef i82975x_DEBUG_IOMEM
406 static void i82975x_print_dram_timings(void __iomem *mch_window)
409 * The register meanings are from Intel specs;
410 * (shows 13-5-5-5 for 800-DDR2)
411 * Asus P5W Bios reports 15-5-4-4
412 * What's your religion?
414 static const int caslats[4] = { 5, 4, 3, 6 };
415 u32 dtreg[2];
417 dtreg[0] = readl(mch_window + 0x114);
418 dtreg[1] = readl(mch_window + 0x194);
419 i82975x_printk(KERN_INFO, "DRAM Timings : Ch0 Ch1\n"
420 " RAS Active Min = %d %d\n"
421 " CAS latency = %d %d\n"
422 " RAS to CAS = %d %d\n"
423 " RAS precharge = %d %d\n",
424 (dtreg[0] >> 19 ) & 0x0f,
425 (dtreg[1] >> 19) & 0x0f,
426 caslats[(dtreg[0] >> 8) & 0x03],
427 caslats[(dtreg[1] >> 8) & 0x03],
428 ((dtreg[0] >> 4) & 0x07) + 2,
429 ((dtreg[1] >> 4) & 0x07) + 2,
430 (dtreg[0] & 0x07) + 2,
431 (dtreg[1] & 0x07) + 2
435 #endif
437 static int i82975x_probe1(struct pci_dev *pdev, int dev_idx)
439 int rc = -ENODEV;
440 struct mem_ctl_info *mci;
441 struct i82975x_pvt *pvt;
442 void __iomem *mch_window;
443 u32 mchbar;
444 u32 drc[2];
445 struct i82975x_error_info discard;
446 int chans;
447 #ifdef i82975x_DEBUG_IOMEM
448 u8 c0drb[4];
449 u8 c1drb[4];
450 #endif
452 debugf0("%s()\n", __func__);
454 pci_read_config_dword(pdev, I82975X_MCHBAR, &mchbar);
455 if (!(mchbar & 1)) {
456 debugf3("%s(): failed, MCHBAR disabled!\n", __func__);
457 goto fail0;
459 mchbar &= 0xffffc000; /* bits 31:14 used for 16K window */
460 mch_window = ioremap_nocache(mchbar, 0x1000);
462 #ifdef i82975x_DEBUG_IOMEM
463 i82975x_printk(KERN_INFO, "MCHBAR real = %0x, remapped = %p\n",
464 mchbar, mch_window);
466 c0drb[0] = readb(mch_window + I82975X_DRB_CH0R0);
467 c0drb[1] = readb(mch_window + I82975X_DRB_CH0R1);
468 c0drb[2] = readb(mch_window + I82975X_DRB_CH0R2);
469 c0drb[3] = readb(mch_window + I82975X_DRB_CH0R3);
470 c1drb[0] = readb(mch_window + I82975X_DRB_CH1R0);
471 c1drb[1] = readb(mch_window + I82975X_DRB_CH1R1);
472 c1drb[2] = readb(mch_window + I82975X_DRB_CH1R2);
473 c1drb[3] = readb(mch_window + I82975X_DRB_CH1R3);
474 i82975x_printk(KERN_INFO, "DRBCH0R0 = 0x%02x\n", c0drb[0]);
475 i82975x_printk(KERN_INFO, "DRBCH0R1 = 0x%02x\n", c0drb[1]);
476 i82975x_printk(KERN_INFO, "DRBCH0R2 = 0x%02x\n", c0drb[2]);
477 i82975x_printk(KERN_INFO, "DRBCH0R3 = 0x%02x\n", c0drb[3]);
478 i82975x_printk(KERN_INFO, "DRBCH1R0 = 0x%02x\n", c1drb[0]);
479 i82975x_printk(KERN_INFO, "DRBCH1R1 = 0x%02x\n", c1drb[1]);
480 i82975x_printk(KERN_INFO, "DRBCH1R2 = 0x%02x\n", c1drb[2]);
481 i82975x_printk(KERN_INFO, "DRBCH1R3 = 0x%02x\n", c1drb[3]);
482 #endif
484 drc[0] = readl(mch_window + I82975X_DRC_CH0M0);
485 drc[1] = readl(mch_window + I82975X_DRC_CH1M0);
486 #ifdef i82975x_DEBUG_IOMEM
487 i82975x_printk(KERN_INFO, "DRC_CH0 = %0x, %s\n", drc[0],
488 ((drc[0] >> 21) & 3) == 1 ?
489 "ECC enabled" : "ECC disabled");
490 i82975x_printk(KERN_INFO, "DRC_CH1 = %0x, %s\n", drc[1],
491 ((drc[1] >> 21) & 3) == 1 ?
492 "ECC enabled" : "ECC disabled");
494 i82975x_printk(KERN_INFO, "C0 BNKARC = %0x\n",
495 readw(mch_window + I82975X_C0BNKARC));
496 i82975x_printk(KERN_INFO, "C1 BNKARC = %0x\n",
497 readw(mch_window + I82975X_C1BNKARC));
498 i82975x_print_dram_timings(mch_window);
499 goto fail1;
500 #endif
501 if (!(((drc[0] >> 21) & 3) == 1 || ((drc[1] >> 21) & 3) == 1)) {
502 i82975x_printk(KERN_INFO, "ECC disabled on both channels.\n");
503 goto fail1;
506 chans = dual_channel_active(mch_window) + 1;
508 /* assuming only one controller, index thus is 0 */
509 mci = edac_mc_alloc(sizeof(*pvt), I82975X_NR_CSROWS(chans),
510 chans, 0);
511 if (!mci) {
512 rc = -ENOMEM;
513 goto fail1;
516 debugf3("%s(): init mci\n", __func__);
517 mci->dev = &pdev->dev;
518 mci->mtype_cap = MEM_FLAG_DDR;
519 mci->edac_ctl_cap = EDAC_FLAG_NONE | EDAC_FLAG_SECDED;
520 mci->edac_cap = EDAC_FLAG_NONE | EDAC_FLAG_SECDED;
521 mci->mod_name = EDAC_MOD_STR;
522 mci->mod_ver = I82975X_REVISION;
523 mci->ctl_name = i82975x_devs[dev_idx].ctl_name;
524 mci->edac_check = i82975x_check;
525 mci->ctl_page_to_phys = NULL;
526 debugf3("%s(): init pvt\n", __func__);
527 pvt = (struct i82975x_pvt *) mci->pvt_info;
528 pvt->mch_window = mch_window;
529 i82975x_init_csrows(mci, pdev, mch_window);
530 i82975x_get_error_info(mci, &discard); /* clear counters */
532 /* finalize this instance of memory controller with edac core */
533 if (edac_mc_add_mc(mci)) {
534 debugf3("%s(): failed edac_mc_add_mc()\n", __func__);
535 goto fail2;
538 /* get this far and it's successful */
539 debugf3("%s(): success\n", __func__);
540 return 0;
542 fail2:
543 edac_mc_free(mci);
545 fail1:
546 iounmap(mch_window);
547 fail0:
548 return rc;
551 /* returns count (>= 0), or negative on error */
552 static int __devinit i82975x_init_one(struct pci_dev *pdev,
553 const struct pci_device_id *ent)
555 int rc;
557 debugf0("%s()\n", __func__);
559 if (pci_enable_device(pdev) < 0)
560 return -EIO;
562 rc = i82975x_probe1(pdev, ent->driver_data);
564 if (mci_pdev == NULL)
565 mci_pdev = pci_dev_get(pdev);
567 return rc;
570 static void __devexit i82975x_remove_one(struct pci_dev *pdev)
572 struct mem_ctl_info *mci;
573 struct i82975x_pvt *pvt;
575 debugf0("%s()\n", __func__);
577 mci = edac_mc_del_mc(&pdev->dev);
578 if (mci == NULL)
579 return;
581 pvt = mci->pvt_info;
582 if (pvt->mch_window)
583 iounmap( pvt->mch_window );
585 edac_mc_free(mci);
588 static const struct pci_device_id i82975x_pci_tbl[] __devinitdata = {
590 PCI_VEND_DEV(INTEL, 82975_0), PCI_ANY_ID, PCI_ANY_ID, 0, 0,
591 I82975X
595 } /* 0 terminated list. */
598 MODULE_DEVICE_TABLE(pci, i82975x_pci_tbl);
600 static struct pci_driver i82975x_driver = {
601 .name = EDAC_MOD_STR,
602 .probe = i82975x_init_one,
603 .remove = __devexit_p(i82975x_remove_one),
604 .id_table = i82975x_pci_tbl,
607 static int __init i82975x_init(void)
609 int pci_rc;
611 debugf3("%s()\n", __func__);
613 /* Ensure that the OPSTATE is set correctly for POLL or NMI */
614 opstate_init();
616 pci_rc = pci_register_driver(&i82975x_driver);
617 if (pci_rc < 0)
618 goto fail0;
620 if (mci_pdev == NULL) {
621 mci_pdev = pci_get_device(PCI_VENDOR_ID_INTEL,
622 PCI_DEVICE_ID_INTEL_82975_0, NULL);
624 if (!mci_pdev) {
625 debugf0("i82975x pci_get_device fail\n");
626 pci_rc = -ENODEV;
627 goto fail1;
630 pci_rc = i82975x_init_one(mci_pdev, i82975x_pci_tbl);
632 if (pci_rc < 0) {
633 debugf0("i82975x init fail\n");
634 pci_rc = -ENODEV;
635 goto fail1;
639 return 0;
641 fail1:
642 pci_unregister_driver(&i82975x_driver);
644 fail0:
645 if (mci_pdev != NULL)
646 pci_dev_put(mci_pdev);
648 return pci_rc;
651 static void __exit i82975x_exit(void)
653 debugf3("%s()\n", __func__);
655 pci_unregister_driver(&i82975x_driver);
657 if (!i82975x_registered) {
658 i82975x_remove_one(mci_pdev);
659 pci_dev_put(mci_pdev);
663 module_init(i82975x_init);
664 module_exit(i82975x_exit);
666 MODULE_LICENSE("GPL");
667 MODULE_AUTHOR("Arvind R. <arvind@acarlab.com>");
668 MODULE_DESCRIPTION("MC support for Intel 82975 memory hub controllers");
670 module_param(edac_op_state, int, 0444);
671 MODULE_PARM_DESC(edac_op_state, "EDAC Error Reporting state: 0=Poll,1=NMI");