2 * Intel 5000(P/V/X) class Memory Controllers kernel module
4 * This file may be distributed under the terms of the
5 * GNU General Public License.
7 * Written by Douglas Thompson Linux Networx (http://lnxi.com)
10 * This module is based on the following document:
12 * Intel 5000X Chipset Memory Controller Hub (MCH) - Datasheet
13 * http://developer.intel.com/design/chipsets/datashts/313070.htm
17 #include <linux/module.h>
18 #include <linux/init.h>
19 #include <linux/pci.h>
20 #include <linux/pci_ids.h>
21 #include <linux/slab.h>
22 #include <linux/edac.h>
23 #include <asm/mmzone.h>
25 #include "edac_core.h"
28 * Alter this version for the I5000 module when modifications are made
30 #define I5000_REVISION " Ver: 2.0.12 " __DATE__
31 #define EDAC_MOD_STR "i5000_edac"
33 #define i5000_printk(level, fmt, arg...) \
34 edac_printk(level, "i5000", fmt, ##arg)
36 #define i5000_mc_printk(mci, level, fmt, arg...) \
37 edac_mc_chipset_printk(mci, level, "i5000", fmt, ##arg)
39 #ifndef PCI_DEVICE_ID_INTEL_FBD_0
40 #define PCI_DEVICE_ID_INTEL_FBD_0 0x25F5
42 #ifndef PCI_DEVICE_ID_INTEL_FBD_1
43 #define PCI_DEVICE_ID_INTEL_FBD_1 0x25F6
47 * Function 0: System Address
48 * Function 1: Memory Branch Map, Control, Errors Register
49 * Function 2: FSB Error Registers
51 * All 3 functions of Device 16 (0,1,2) share the SAME DID
53 #define PCI_DEVICE_ID_INTEL_I5000_DEV16 0x25F0
55 /* OFFSETS for Function 0 */
57 /* OFFSETS for Function 1 */
60 #define MAXDIMMPERCH 0x57
63 #define RED_ECC_LOCATOR(x) ((x) & 0x3FFFF)
64 #define REC_ECC_LOCATOR_EVEN(x) ((x) & 0x001FF)
65 #define REC_ECC_LOCATOR_ODD(x) ((x) & 0x3FE00)
73 #define FERR_FAT_FBD 0x98
74 #define NERR_FAT_FBD 0x9C
75 #define EXTRACT_FBDCHAN_INDX(x) (((x)>>28) & 0x3)
76 #define FERR_FAT_FBDCHAN 0x30000000
77 #define FERR_FAT_M3ERR 0x00000004
78 #define FERR_FAT_M2ERR 0x00000002
79 #define FERR_FAT_M1ERR 0x00000001
80 #define FERR_FAT_MASK (FERR_FAT_M1ERR | \
84 #define FERR_NF_FBD 0xA0
86 /* Thermal and SPD or BFD errors */
87 #define FERR_NF_M28ERR 0x01000000
88 #define FERR_NF_M27ERR 0x00800000
89 #define FERR_NF_M26ERR 0x00400000
90 #define FERR_NF_M25ERR 0x00200000
91 #define FERR_NF_M24ERR 0x00100000
92 #define FERR_NF_M23ERR 0x00080000
93 #define FERR_NF_M22ERR 0x00040000
94 #define FERR_NF_M21ERR 0x00020000
96 /* Correctable errors */
97 #define FERR_NF_M20ERR 0x00010000
98 #define FERR_NF_M19ERR 0x00008000
99 #define FERR_NF_M18ERR 0x00004000
100 #define FERR_NF_M17ERR 0x00002000
102 /* Non-Retry or redundant Retry errors */
103 #define FERR_NF_M16ERR 0x00001000
104 #define FERR_NF_M15ERR 0x00000800
105 #define FERR_NF_M14ERR 0x00000400
106 #define FERR_NF_M13ERR 0x00000200
108 /* Uncorrectable errors */
109 #define FERR_NF_M12ERR 0x00000100
110 #define FERR_NF_M11ERR 0x00000080
111 #define FERR_NF_M10ERR 0x00000040
112 #define FERR_NF_M9ERR 0x00000020
113 #define FERR_NF_M8ERR 0x00000010
114 #define FERR_NF_M7ERR 0x00000008
115 #define FERR_NF_M6ERR 0x00000004
116 #define FERR_NF_M5ERR 0x00000002
117 #define FERR_NF_M4ERR 0x00000001
119 #define FERR_NF_UNCORRECTABLE (FERR_NF_M12ERR | \
127 #define FERR_NF_CORRECTABLE (FERR_NF_M20ERR | \
131 #define FERR_NF_DIMM_SPARE (FERR_NF_M27ERR | \
133 #define FERR_NF_THERMAL (FERR_NF_M26ERR | \
137 #define FERR_NF_SPD_PROTOCOL (FERR_NF_M22ERR)
138 #define FERR_NF_NORTH_CRC (FERR_NF_M21ERR)
139 #define FERR_NF_NON_RETRY (FERR_NF_M13ERR | \
143 #define NERR_NF_FBD 0xA4
144 #define FERR_NF_MASK (FERR_NF_UNCORRECTABLE | \
145 FERR_NF_CORRECTABLE | \
146 FERR_NF_DIMM_SPARE | \
148 FERR_NF_SPD_PROTOCOL | \
149 FERR_NF_NORTH_CRC | \
152 #define EMASK_FBD 0xA8
153 #define EMASK_FBD_M28ERR 0x08000000
154 #define EMASK_FBD_M27ERR 0x04000000
155 #define EMASK_FBD_M26ERR 0x02000000
156 #define EMASK_FBD_M25ERR 0x01000000
157 #define EMASK_FBD_M24ERR 0x00800000
158 #define EMASK_FBD_M23ERR 0x00400000
159 #define EMASK_FBD_M22ERR 0x00200000
160 #define EMASK_FBD_M21ERR 0x00100000
161 #define EMASK_FBD_M20ERR 0x00080000
162 #define EMASK_FBD_M19ERR 0x00040000
163 #define EMASK_FBD_M18ERR 0x00020000
164 #define EMASK_FBD_M17ERR 0x00010000
166 #define EMASK_FBD_M15ERR 0x00004000
167 #define EMASK_FBD_M14ERR 0x00002000
168 #define EMASK_FBD_M13ERR 0x00001000
169 #define EMASK_FBD_M12ERR 0x00000800
170 #define EMASK_FBD_M11ERR 0x00000400
171 #define EMASK_FBD_M10ERR 0x00000200
172 #define EMASK_FBD_M9ERR 0x00000100
173 #define EMASK_FBD_M8ERR 0x00000080
174 #define EMASK_FBD_M7ERR 0x00000040
175 #define EMASK_FBD_M6ERR 0x00000020
176 #define EMASK_FBD_M5ERR 0x00000010
177 #define EMASK_FBD_M4ERR 0x00000008
178 #define EMASK_FBD_M3ERR 0x00000004
179 #define EMASK_FBD_M2ERR 0x00000002
180 #define EMASK_FBD_M1ERR 0x00000001
182 #define ENABLE_EMASK_FBD_FATAL_ERRORS (EMASK_FBD_M1ERR | \
186 #define ENABLE_EMASK_FBD_UNCORRECTABLE (EMASK_FBD_M4ERR | \
195 #define ENABLE_EMASK_FBD_CORRECTABLE (EMASK_FBD_M17ERR | \
199 #define ENABLE_EMASK_FBD_DIMM_SPARE (EMASK_FBD_M27ERR | \
201 #define ENABLE_EMASK_FBD_THERMALS (EMASK_FBD_M26ERR | \
205 #define ENABLE_EMASK_FBD_SPD_PROTOCOL (EMASK_FBD_M22ERR)
206 #define ENABLE_EMASK_FBD_NORTH_CRC (EMASK_FBD_M21ERR)
207 #define ENABLE_EMASK_FBD_NON_RETRY (EMASK_FBD_M15ERR | \
211 #define ENABLE_EMASK_ALL (ENABLE_EMASK_FBD_NON_RETRY | \
212 ENABLE_EMASK_FBD_NORTH_CRC | \
213 ENABLE_EMASK_FBD_SPD_PROTOCOL | \
214 ENABLE_EMASK_FBD_THERMALS | \
215 ENABLE_EMASK_FBD_DIMM_SPARE | \
216 ENABLE_EMASK_FBD_FATAL_ERRORS | \
217 ENABLE_EMASK_FBD_CORRECTABLE | \
218 ENABLE_EMASK_FBD_UNCORRECTABLE)
220 #define ERR0_FBD 0xAC
221 #define ERR1_FBD 0xB0
222 #define ERR2_FBD 0xB4
223 #define MCERR_FBD 0xB8
224 #define NRECMEMA 0xBE
225 #define NREC_BANK(x) (((x)>>12) & 0x7)
226 #define NREC_RDWR(x) (((x)>>11) & 1)
227 #define NREC_RANK(x) (((x)>>8) & 0x7)
228 #define NRECMEMB 0xC0
229 #define NREC_CAS(x) (((x)>>16) & 0xFFFFFF)
230 #define NREC_RAS(x) ((x) & 0x7FFF)
231 #define NRECFGLOG 0xC4
232 #define NREEECFBDA 0xC8
233 #define NREEECFBDB 0xCC
234 #define NREEECFBDC 0xD0
235 #define NREEECFBDD 0xD4
236 #define NREEECFBDE 0xD8
239 #define REC_BANK(x) (((x)>>12) & 0x7)
240 #define REC_RDWR(x) (((x)>>11) & 1)
241 #define REC_RANK(x) (((x)>>8) & 0x7)
243 #define REC_CAS(x) (((x)>>16) & 0xFFFFFF)
244 #define REC_RAS(x) ((x) & 0x7FFF)
245 #define RECFGLOG 0xE8
252 /* OFFSETS for Function 2 */
256 * Function 0: Memory Map Branch 0
259 * Function 0: Memory Map Branch 1
261 #define PCI_DEVICE_ID_I5000_BRANCH_0 0x25F5
262 #define PCI_DEVICE_ID_I5000_BRANCH_1 0x25F6
264 #define AMB_PRESENT_0 0x64
265 #define AMB_PRESENT_1 0x66
272 #define CHANNELS_PER_BRANCH (2)
274 /* Defines to extract the vaious fields from the
275 * MTRx - Memory Technology Registers
277 #define MTR_DIMMS_PRESENT(mtr) ((mtr) & (0x1 << 8))
278 #define MTR_DRAM_WIDTH(mtr) ((((mtr) >> 6) & 0x1) ? 8 : 4)
279 #define MTR_DRAM_BANKS(mtr) ((((mtr) >> 5) & 0x1) ? 8 : 4)
280 #define MTR_DRAM_BANKS_ADDR_BITS(mtr) ((MTR_DRAM_BANKS(mtr) == 8) ? 3 : 2)
281 #define MTR_DIMM_RANK(mtr) (((mtr) >> 4) & 0x1)
282 #define MTR_DIMM_RANK_ADDR_BITS(mtr) (MTR_DIMM_RANK(mtr) ? 2 : 1)
283 #define MTR_DIMM_ROWS(mtr) (((mtr) >> 2) & 0x3)
284 #define MTR_DIMM_ROWS_ADDR_BITS(mtr) (MTR_DIMM_ROWS(mtr) + 13)
285 #define MTR_DIMM_COLS(mtr) ((mtr) & 0x3)
286 #define MTR_DIMM_COLS_ADDR_BITS(mtr) (MTR_DIMM_COLS(mtr) + 10)
288 #ifdef CONFIG_EDAC_DEBUG
289 static char *numrow_toString[] = {
296 static char *numcol_toString[] = {
297 "1,024 - 10 columns",
298 "2,048 - 11 columns",
299 "4,096 - 12 columns",
304 /* Enumeration of supported devices */
307 I5000V = 1, /* future */
308 I5000X = 2 /* future */
311 /* Device name and register DID (Device ID) */
312 struct i5000_dev_info {
313 const char *ctl_name; /* name for this device */
314 u16 fsb_mapping_errors; /* DID for the branchmap,control */
317 /* Table of devices attributes supported by this driver */
318 static const struct i5000_dev_info i5000_devs[] = {
321 .fsb_mapping_errors = PCI_DEVICE_ID_INTEL_I5000_DEV16,
325 struct i5000_dimm_info {
326 int megabytes; /* size, 0 means not present */
330 #define MAX_CHANNELS 6 /* max possible channels */
331 #define MAX_CSROWS (8*2) /* max possible csrows per channel */
333 /* driver private data structure */
335 struct pci_dev *system_address; /* 16.0 */
336 struct pci_dev *branchmap_werrors; /* 16.1 */
337 struct pci_dev *fsb_error_regs; /* 16.2 */
338 struct pci_dev *branch_0; /* 21.0 */
339 struct pci_dev *branch_1; /* 22.0 */
341 u16 tolm; /* top of low memory */
342 u64 ambase; /* AMB BAR */
344 u16 mir0, mir1, mir2;
346 u16 b0_mtr[NUM_MTRS]; /* Memory Technlogy Reg */
347 u16 b0_ambpresent0; /* Branch 0, Channel 0 */
348 u16 b0_ambpresent1; /* Brnach 0, Channel 1 */
350 u16 b1_mtr[NUM_MTRS]; /* Memory Technlogy Reg */
351 u16 b1_ambpresent0; /* Branch 1, Channel 8 */
352 u16 b1_ambpresent1; /* Branch 1, Channel 1 */
354 /* DIMM information matrix, allocating architecture maximums */
355 struct i5000_dimm_info dimm_info[MAX_CSROWS][MAX_CHANNELS];
357 /* Actual values for this controller */
358 int maxch; /* Max channels */
359 int maxdimmperch; /* Max DIMMs per channel */
362 /* I5000 MCH error information retrieved from Hardware */
363 struct i5000_error_info {
365 /* These registers are always read from the MC */
366 u32 ferr_fat_fbd; /* First Errors Fatal */
367 u32 nerr_fat_fbd; /* Next Errors Fatal */
368 u32 ferr_nf_fbd; /* First Errors Non-Fatal */
369 u32 nerr_nf_fbd; /* Next Errors Non-Fatal */
371 /* These registers are input ONLY if there was a Recoverable Error */
372 u32 redmemb; /* Recoverable Mem Data Error log B */
373 u16 recmema; /* Recoverable Mem Error log A */
374 u32 recmemb; /* Recoverable Mem Error log B */
376 /* These registers are input ONLY if there was a
377 * Non-Recoverable Error */
378 u16 nrecmema; /* Non-Recoverable Mem log A */
379 u16 nrecmemb; /* Non-Recoverable Mem log B */
383 static struct edac_pci_ctl_info *i5000_pci;
386 * i5000_get_error_info Retrieve the hardware error information from
387 * the hardware and cache it in the 'info'
390 static void i5000_get_error_info(struct mem_ctl_info *mci,
391 struct i5000_error_info *info)
393 struct i5000_pvt *pvt;
398 /* read in the 1st FATAL error register */
399 pci_read_config_dword(pvt->branchmap_werrors, FERR_FAT_FBD, &value);
401 /* Mask only the bits that the doc says are valid
403 value &= (FERR_FAT_FBDCHAN | FERR_FAT_MASK);
405 /* If there is an error, then read in the */
406 /* NEXT FATAL error register and the Memory Error Log Register A */
407 if (value & FERR_FAT_MASK) {
408 info->ferr_fat_fbd = value;
410 /* harvest the various error data we need */
411 pci_read_config_dword(pvt->branchmap_werrors,
412 NERR_FAT_FBD, &info->nerr_fat_fbd);
413 pci_read_config_word(pvt->branchmap_werrors,
414 NRECMEMA, &info->nrecmema);
415 pci_read_config_word(pvt->branchmap_werrors,
416 NRECMEMB, &info->nrecmemb);
418 /* Clear the error bits, by writing them back */
419 pci_write_config_dword(pvt->branchmap_werrors,
420 FERR_FAT_FBD, value);
422 info->ferr_fat_fbd = 0;
423 info->nerr_fat_fbd = 0;
428 /* read in the 1st NON-FATAL error register */
429 pci_read_config_dword(pvt->branchmap_werrors, FERR_NF_FBD, &value);
431 /* If there is an error, then read in the 1st NON-FATAL error
432 * register as well */
433 if (value & FERR_NF_MASK) {
434 info->ferr_nf_fbd = value;
436 /* harvest the various error data we need */
437 pci_read_config_dword(pvt->branchmap_werrors,
438 NERR_NF_FBD, &info->nerr_nf_fbd);
439 pci_read_config_word(pvt->branchmap_werrors,
440 RECMEMA, &info->recmema);
441 pci_read_config_dword(pvt->branchmap_werrors,
442 RECMEMB, &info->recmemb);
443 pci_read_config_dword(pvt->branchmap_werrors,
444 REDMEMB, &info->redmemb);
446 /* Clear the error bits, by writing them back */
447 pci_write_config_dword(pvt->branchmap_werrors,
450 info->ferr_nf_fbd = 0;
451 info->nerr_nf_fbd = 0;
459 * i5000_process_fatal_error_info(struct mem_ctl_info *mci,
460 * struct i5000_error_info *info,
461 * int handle_errors);
463 * handle the Intel FATAL errors, if any
465 static void i5000_process_fatal_error_info(struct mem_ctl_info *mci,
466 struct i5000_error_info *info,
469 char msg[EDAC_MC_LABEL_LEN + 1 + 90];
478 /* mask off the Error bits that are possible */
479 allErrors = (info->ferr_fat_fbd & FERR_FAT_MASK);
481 return; /* if no error, return now */
483 /* ONLY ONE of the possible error bits will be set, as per the docs */
484 i5000_mc_printk(mci, KERN_ERR,
485 "FATAL ERRORS Found!!! 1st FATAL Err Reg= 0x%x\n",
488 branch = EXTRACT_FBDCHAN_INDX(info->ferr_fat_fbd);
491 /* Use the NON-Recoverable macros to extract data */
492 bank = NREC_BANK(info->nrecmema);
493 rank = NREC_RANK(info->nrecmema);
494 rdwr = NREC_RDWR(info->nrecmema);
495 ras = NREC_RAS(info->nrecmemb);
496 cas = NREC_CAS(info->nrecmemb);
498 debugf0("\t\tCSROW= %d Channels= %d,%d (Branch= %d "
499 "DRAM Bank= %d rdwr= %s ras= %d cas= %d)\n",
500 rank, channel, channel + 1, branch >> 1, bank,
501 rdwr ? "Write" : "Read", ras, cas);
503 /* Only 1 bit will be on */
504 if (allErrors & FERR_FAT_M1ERR) {
505 i5000_mc_printk(mci, KERN_ERR,
506 "Alert on non-redundant retry or fast "
509 } else if (allErrors & FERR_FAT_M2ERR) {
510 i5000_mc_printk(mci, KERN_ERR,
511 "Northbound CRC error on non-redundant "
514 } else if (allErrors & FERR_FAT_M3ERR) {
515 i5000_mc_printk(mci, KERN_ERR,
516 ">Tmid Thermal event with intelligent "
517 "throttling disabled\n");
520 /* Form out message */
521 snprintf(msg, sizeof(msg),
522 "(Branch=%d DRAM-Bank=%d RDWR=%s RAS=%d CAS=%d "
524 branch >> 1, bank, rdwr ? "Write" : "Read", ras, cas,
527 /* Call the helper to output message */
528 edac_mc_handle_fbd_ue(mci, rank, channel, channel + 1, msg);
532 * i5000_process_fatal_error_info(struct mem_ctl_info *mci,
533 * struct i5000_error_info *info,
534 * int handle_errors);
536 * handle the Intel NON-FATAL errors, if any
538 static void i5000_process_nonfatal_error_info(struct mem_ctl_info *mci,
539 struct i5000_error_info *info,
542 char msg[EDAC_MC_LABEL_LEN + 1 + 90];
554 /* mask off the Error bits that are possible */
555 allErrors = (info->ferr_nf_fbd & FERR_NF_MASK);
557 return; /* if no error, return now */
559 /* ONLY ONE of the possible error bits will be set, as per the docs */
560 i5000_mc_printk(mci, KERN_WARNING,
561 "NON-FATAL ERRORS Found!!! 1st NON-FATAL Err "
562 "Reg= 0x%x\n", allErrors);
564 ue_errors = allErrors & FERR_NF_UNCORRECTABLE;
566 debugf0("\tUncorrected bits= 0x%x\n", ue_errors);
568 branch = EXTRACT_FBDCHAN_INDX(info->ferr_nf_fbd);
570 bank = NREC_BANK(info->nrecmema);
571 rank = NREC_RANK(info->nrecmema);
572 rdwr = NREC_RDWR(info->nrecmema);
573 ras = NREC_RAS(info->nrecmemb);
574 cas = NREC_CAS(info->nrecmemb);
577 ("\t\tCSROW= %d Channels= %d,%d (Branch= %d "
578 "DRAM Bank= %d rdwr= %s ras= %d cas= %d)\n",
579 rank, channel, channel + 1, branch >> 1, bank,
580 rdwr ? "Write" : "Read", ras, cas);
582 /* Form out message */
583 snprintf(msg, sizeof(msg),
584 "(Branch=%d DRAM-Bank=%d RDWR=%s RAS=%d "
585 "CAS=%d, UE Err=0x%x)",
586 branch >> 1, bank, rdwr ? "Write" : "Read", ras, cas,
589 /* Call the helper to output message */
590 edac_mc_handle_fbd_ue(mci, rank, channel, channel + 1, msg);
593 /* Check correctable errors */
594 ce_errors = allErrors & FERR_NF_CORRECTABLE;
596 debugf0("\tCorrected bits= 0x%x\n", ce_errors);
598 branch = EXTRACT_FBDCHAN_INDX(info->ferr_nf_fbd);
601 if (REC_ECC_LOCATOR_ODD(info->redmemb))
604 /* Convert channel to be based from zero, instead of
605 * from branch base of 0 */
608 bank = REC_BANK(info->recmema);
609 rank = REC_RANK(info->recmema);
610 rdwr = REC_RDWR(info->recmema);
611 ras = REC_RAS(info->recmemb);
612 cas = REC_CAS(info->recmemb);
614 debugf0("\t\tCSROW= %d Channel= %d (Branch %d "
615 "DRAM Bank= %d rdwr= %s ras= %d cas= %d)\n",
616 rank, channel, branch >> 1, bank,
617 rdwr ? "Write" : "Read", ras, cas);
619 /* Form out message */
620 snprintf(msg, sizeof(msg),
621 "(Branch=%d DRAM-Bank=%d RDWR=%s RAS=%d "
622 "CAS=%d, CE Err=0x%x)", branch >> 1, bank,
623 rdwr ? "Write" : "Read", ras, cas, ce_errors);
625 /* Call the helper to output message */
626 edac_mc_handle_fbd_ce(mci, rank, channel, msg);
629 /* See if any of the thermal errors have fired */
630 misc_errors = allErrors & FERR_NF_THERMAL;
632 i5000_printk(KERN_WARNING, "\tTHERMAL Error, bits= 0x%x\n",
636 /* See if any of the thermal errors have fired */
637 misc_errors = allErrors & FERR_NF_NON_RETRY;
639 i5000_printk(KERN_WARNING, "\tNON-Retry Errors, bits= 0x%x\n",
643 /* See if any of the thermal errors have fired */
644 misc_errors = allErrors & FERR_NF_NORTH_CRC;
646 i5000_printk(KERN_WARNING,
647 "\tNORTHBOUND CRC Error, bits= 0x%x\n",
651 /* See if any of the thermal errors have fired */
652 misc_errors = allErrors & FERR_NF_SPD_PROTOCOL;
654 i5000_printk(KERN_WARNING,
655 "\tSPD Protocol Error, bits= 0x%x\n",
659 /* See if any of the thermal errors have fired */
660 misc_errors = allErrors & FERR_NF_DIMM_SPARE;
662 i5000_printk(KERN_WARNING, "\tDIMM-Spare Error, bits= 0x%x\n",
668 * i5000_process_error_info Process the error info that is
669 * in the 'info' structure, previously retrieved from hardware
671 static void i5000_process_error_info(struct mem_ctl_info *mci,
672 struct i5000_error_info *info,
675 /* First handle any fatal errors that occurred */
676 i5000_process_fatal_error_info(mci, info, handle_errors);
678 /* now handle any non-fatal errors that occurred */
679 i5000_process_nonfatal_error_info(mci, info, handle_errors);
683 * i5000_clear_error Retrieve any error from the hardware
684 * but do NOT process that error.
685 * Used for 'clearing' out of previous errors
686 * Called by the Core module.
688 static void i5000_clear_error(struct mem_ctl_info *mci)
690 struct i5000_error_info info;
692 i5000_get_error_info(mci, &info);
696 * i5000_check_error Retrieve and process errors reported by the
697 * hardware. Called by the Core module.
699 static void i5000_check_error(struct mem_ctl_info *mci)
701 struct i5000_error_info info;
702 debugf4("MC%d: " __FILE__ ": %s()\n", mci->mc_idx, __func__);
703 i5000_get_error_info(mci, &info);
704 i5000_process_error_info(mci, &info, 1);
708 * i5000_get_devices Find and perform 'get' operation on the MCH's
709 * device/functions we want to reference for this driver
711 * Need to 'get' device 16 func 1 and func 2
713 static int i5000_get_devices(struct mem_ctl_info *mci, int dev_idx)
715 //const struct i5000_dev_info *i5000_dev = &i5000_devs[dev_idx];
716 struct i5000_pvt *pvt;
717 struct pci_dev *pdev;
721 /* Attempt to 'get' the MCH register we want */
724 pdev = pci_get_device(PCI_VENDOR_ID_INTEL,
725 PCI_DEVICE_ID_INTEL_I5000_DEV16, pdev);
727 /* End of list, leave */
729 i5000_printk(KERN_ERR,
730 "'system address,Process Bus' "
732 "vendor 0x%x device 0x%x FUNC 1 "
735 PCI_DEVICE_ID_INTEL_I5000_DEV16);
740 /* Scan for device 16 func 1 */
741 if (PCI_FUNC(pdev->devfn) == 1)
745 pvt->branchmap_werrors = pdev;
747 /* Attempt to 'get' the MCH register we want */
750 pdev = pci_get_device(PCI_VENDOR_ID_INTEL,
751 PCI_DEVICE_ID_INTEL_I5000_DEV16, pdev);
754 i5000_printk(KERN_ERR,
755 "MC: 'branchmap,control,errors' "
757 "vendor 0x%x device 0x%x Func 2 "
760 PCI_DEVICE_ID_INTEL_I5000_DEV16);
762 pci_dev_put(pvt->branchmap_werrors);
766 /* Scan for device 16 func 1 */
767 if (PCI_FUNC(pdev->devfn) == 2)
771 pvt->fsb_error_regs = pdev;
773 debugf1("System Address, processor bus- PCI Bus ID: %s %x:%x\n",
774 pci_name(pvt->system_address),
775 pvt->system_address->vendor, pvt->system_address->device);
776 debugf1("Branchmap, control and errors - PCI Bus ID: %s %x:%x\n",
777 pci_name(pvt->branchmap_werrors),
778 pvt->branchmap_werrors->vendor, pvt->branchmap_werrors->device);
779 debugf1("FSB Error Regs - PCI Bus ID: %s %x:%x\n",
780 pci_name(pvt->fsb_error_regs),
781 pvt->fsb_error_regs->vendor, pvt->fsb_error_regs->device);
784 pdev = pci_get_device(PCI_VENDOR_ID_INTEL,
785 PCI_DEVICE_ID_I5000_BRANCH_0, pdev);
788 i5000_printk(KERN_ERR,
789 "MC: 'BRANCH 0' device not found:"
790 "vendor 0x%x device 0x%x Func 0 (broken BIOS?)\n",
791 PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_I5000_BRANCH_0);
793 pci_dev_put(pvt->branchmap_werrors);
794 pci_dev_put(pvt->fsb_error_regs);
798 pvt->branch_0 = pdev;
800 /* If this device claims to have more than 2 channels then
801 * fetch Branch 1's information
803 if (pvt->maxch >= CHANNELS_PER_BRANCH) {
805 pdev = pci_get_device(PCI_VENDOR_ID_INTEL,
806 PCI_DEVICE_ID_I5000_BRANCH_1, pdev);
809 i5000_printk(KERN_ERR,
810 "MC: 'BRANCH 1' device not found:"
811 "vendor 0x%x device 0x%x Func 0 "
814 PCI_DEVICE_ID_I5000_BRANCH_1);
816 pci_dev_put(pvt->branchmap_werrors);
817 pci_dev_put(pvt->fsb_error_regs);
818 pci_dev_put(pvt->branch_0);
822 pvt->branch_1 = pdev;
829 * i5000_put_devices 'put' all the devices that we have
832 static void i5000_put_devices(struct mem_ctl_info *mci)
834 struct i5000_pvt *pvt;
838 pci_dev_put(pvt->branchmap_werrors); /* FUNC 1 */
839 pci_dev_put(pvt->fsb_error_regs); /* FUNC 2 */
840 pci_dev_put(pvt->branch_0); /* DEV 21 */
842 /* Only if more than 2 channels do we release the second branch */
843 if (pvt->maxch >= CHANNELS_PER_BRANCH)
844 pci_dev_put(pvt->branch_1); /* DEV 22 */
848 * determine_amb_resent
850 * the information is contained in NUM_MTRS different registers
851 * determineing which of the NUM_MTRS requires knowing
852 * which channel is in question
854 * 2 branches, each with 2 channels
855 * b0_ambpresent0 for channel '0'
856 * b0_ambpresent1 for channel '1'
857 * b1_ambpresent0 for channel '2'
858 * b1_ambpresent1 for channel '3'
860 static int determine_amb_present_reg(struct i5000_pvt *pvt, int channel)
864 if (channel < CHANNELS_PER_BRANCH) {
866 amb_present = pvt->b0_ambpresent1;
868 amb_present = pvt->b0_ambpresent0;
871 amb_present = pvt->b1_ambpresent1;
873 amb_present = pvt->b1_ambpresent0;
880 * determine_mtr(pvt, csrow, channel)
882 * return the proper MTR register as determine by the csrow and channel desired
884 static int determine_mtr(struct i5000_pvt *pvt, int csrow, int channel)
888 if (channel < CHANNELS_PER_BRANCH)
889 mtr = pvt->b0_mtr[csrow >> 1];
891 mtr = pvt->b1_mtr[csrow >> 1];
898 static void decode_mtr(int slot_row, u16 mtr)
902 ans = MTR_DIMMS_PRESENT(mtr);
904 debugf2("\tMTR%d=0x%x: DIMMs are %s\n", slot_row, mtr,
905 ans ? "Present" : "NOT Present");
909 debugf2("\t\tWIDTH: x%d\n", MTR_DRAM_WIDTH(mtr));
910 debugf2("\t\tNUMBANK: %d bank(s)\n", MTR_DRAM_BANKS(mtr));
911 debugf2("\t\tNUMRANK: %s\n", MTR_DIMM_RANK(mtr) ? "double" : "single");
912 debugf2("\t\tNUMROW: %s\n", numrow_toString[MTR_DIMM_ROWS(mtr)]);
913 debugf2("\t\tNUMCOL: %s\n", numcol_toString[MTR_DIMM_COLS(mtr)]);
916 static void handle_channel(struct i5000_pvt *pvt, int csrow, int channel,
917 struct i5000_dimm_info *dinfo)
923 mtr = determine_mtr(pvt, csrow, channel);
924 if (MTR_DIMMS_PRESENT(mtr)) {
925 amb_present_reg = determine_amb_present_reg(pvt, channel);
927 /* Determine if there is a DIMM present in this DIMM slot */
928 if (amb_present_reg & (1 << (csrow >> 1))) {
929 dinfo->dual_rank = MTR_DIMM_RANK(mtr);
931 if (!((dinfo->dual_rank == 0) &&
932 ((csrow & 0x1) == 0x1))) {
933 /* Start with the number of bits for a Bank
935 addrBits = MTR_DRAM_BANKS_ADDR_BITS(mtr);
936 /* Add thenumber of ROW bits */
937 addrBits += MTR_DIMM_ROWS_ADDR_BITS(mtr);
938 /* add the number of COLUMN bits */
939 addrBits += MTR_DIMM_COLS_ADDR_BITS(mtr);
941 addrBits += 6; /* add 64 bits per DIMM */
942 addrBits -= 20; /* divide by 2^^20 */
943 addrBits -= 3; /* 8 bits per bytes */
945 dinfo->megabytes = 1 << addrBits;
952 * calculate_dimm_size
954 * also will output a DIMM matrix map, if debug is enabled, for viewing
955 * how the DIMMs are populated
957 static void calculate_dimm_size(struct i5000_pvt *pvt)
959 struct i5000_dimm_info *dinfo;
960 int csrow, max_csrows;
961 char *p, *mem_buffer;
965 /* ================= Generate some debug output ================= */
967 mem_buffer = p = kmalloc(space, GFP_KERNEL);
969 i5000_printk(KERN_ERR, "MC: %s:%s() kmalloc() failed\n",
974 n = snprintf(p, space, "\n");
978 /* Scan all the actual CSROWS (which is # of DIMMS * 2)
979 * and calculate the information for each DIMM
980 * Start with the highest csrow first, to display it first
981 * and work toward the 0th csrow
983 max_csrows = pvt->maxdimmperch * 2;
984 for (csrow = max_csrows - 1; csrow >= 0; csrow--) {
986 /* on an odd csrow, first output a 'boundary' marker,
987 * then reset the message buffer */
989 n = snprintf(p, space, "---------------------------"
990 "--------------------------------");
993 debugf2("%s\n", mem_buffer);
997 n = snprintf(p, space, "csrow %2d ", csrow);
1001 for (channel = 0; channel < pvt->maxch; channel++) {
1002 dinfo = &pvt->dimm_info[csrow][channel];
1003 handle_channel(pvt, csrow, channel, dinfo);
1004 n = snprintf(p, space, "%4d MB | ", dinfo->megabytes);
1008 n = snprintf(p, space, "\n");
1013 /* Output the last bottom 'boundary' marker */
1014 n = snprintf(p, space, "---------------------------"
1015 "--------------------------------\n");
1019 /* now output the 'channel' labels */
1020 n = snprintf(p, space, " ");
1023 for (channel = 0; channel < pvt->maxch; channel++) {
1024 n = snprintf(p, space, "channel %d | ", channel);
1028 n = snprintf(p, space, "\n");
1032 /* output the last message and free buffer */
1033 debugf2("%s\n", mem_buffer);
1038 * i5000_get_mc_regs read in the necessary registers and
1041 * Fills in the private data members
1043 static void i5000_get_mc_regs(struct mem_ctl_info *mci)
1045 struct i5000_pvt *pvt;
1053 pvt = mci->pvt_info;
1055 pci_read_config_dword(pvt->system_address, AMBASE,
1056 (u32 *) & pvt->ambase);
1057 pci_read_config_dword(pvt->system_address, AMBASE + sizeof(u32),
1058 ((u32 *) & pvt->ambase) + sizeof(u32));
1060 maxdimmperch = pvt->maxdimmperch;
1063 debugf2("AMBASE= 0x%lx MAXCH= %d MAX-DIMM-Per-CH= %d\n",
1064 (long unsigned int)pvt->ambase, pvt->maxch, pvt->maxdimmperch);
1066 /* Get the Branch Map regs */
1067 pci_read_config_word(pvt->branchmap_werrors, TOLM, &pvt->tolm);
1069 debugf2("\nTOLM (number of 256M regions) =%u (0x%x)\n", pvt->tolm,
1072 actual_tolm = pvt->tolm << 28;
1073 debugf2("Actual TOLM byte addr=%u (0x%x)\n", actual_tolm, actual_tolm);
1075 pci_read_config_word(pvt->branchmap_werrors, MIR0, &pvt->mir0);
1076 pci_read_config_word(pvt->branchmap_werrors, MIR1, &pvt->mir1);
1077 pci_read_config_word(pvt->branchmap_werrors, MIR2, &pvt->mir2);
1079 /* Get the MIR[0-2] regs */
1080 limit = (pvt->mir0 >> 4) & 0x0FFF;
1081 way0 = pvt->mir0 & 0x1;
1082 way1 = pvt->mir0 & 0x2;
1083 debugf2("MIR0: limit= 0x%x WAY1= %u WAY0= %x\n", limit, way1, way0);
1084 limit = (pvt->mir1 >> 4) & 0x0FFF;
1085 way0 = pvt->mir1 & 0x1;
1086 way1 = pvt->mir1 & 0x2;
1087 debugf2("MIR1: limit= 0x%x WAY1= %u WAY0= %x\n", limit, way1, way0);
1088 limit = (pvt->mir2 >> 4) & 0x0FFF;
1089 way0 = pvt->mir2 & 0x1;
1090 way1 = pvt->mir2 & 0x2;
1091 debugf2("MIR2: limit= 0x%x WAY1= %u WAY0= %x\n", limit, way1, way0);
1093 /* Get the MTR[0-3] regs */
1094 for (slot_row = 0; slot_row < NUM_MTRS; slot_row++) {
1095 int where = MTR0 + (slot_row * sizeof(u32));
1097 pci_read_config_word(pvt->branch_0, where,
1098 &pvt->b0_mtr[slot_row]);
1100 debugf2("MTR%d where=0x%x B0 value=0x%x\n", slot_row, where,
1101 pvt->b0_mtr[slot_row]);
1103 if (pvt->maxch >= CHANNELS_PER_BRANCH) {
1104 pci_read_config_word(pvt->branch_1, where,
1105 &pvt->b1_mtr[slot_row]);
1106 debugf2("MTR%d where=0x%x B1 value=0x%x\n", slot_row,
1107 where, pvt->b0_mtr[slot_row]);
1109 pvt->b1_mtr[slot_row] = 0;
1113 /* Read and dump branch 0's MTRs */
1114 debugf2("\nMemory Technology Registers:\n");
1115 debugf2(" Branch 0:\n");
1116 for (slot_row = 0; slot_row < NUM_MTRS; slot_row++) {
1117 decode_mtr(slot_row, pvt->b0_mtr[slot_row]);
1119 pci_read_config_word(pvt->branch_0, AMB_PRESENT_0,
1120 &pvt->b0_ambpresent0);
1121 debugf2("\t\tAMB-Branch 0-present0 0x%x:\n", pvt->b0_ambpresent0);
1122 pci_read_config_word(pvt->branch_0, AMB_PRESENT_1,
1123 &pvt->b0_ambpresent1);
1124 debugf2("\t\tAMB-Branch 0-present1 0x%x:\n", pvt->b0_ambpresent1);
1126 /* Only if we have 2 branchs (4 channels) */
1127 if (pvt->maxch < CHANNELS_PER_BRANCH) {
1128 pvt->b1_ambpresent0 = 0;
1129 pvt->b1_ambpresent1 = 0;
1131 /* Read and dump branch 1's MTRs */
1132 debugf2(" Branch 1:\n");
1133 for (slot_row = 0; slot_row < NUM_MTRS; slot_row++) {
1134 decode_mtr(slot_row, pvt->b1_mtr[slot_row]);
1136 pci_read_config_word(pvt->branch_1, AMB_PRESENT_0,
1137 &pvt->b1_ambpresent0);
1138 debugf2("\t\tAMB-Branch 1-present0 0x%x:\n",
1139 pvt->b1_ambpresent0);
1140 pci_read_config_word(pvt->branch_1, AMB_PRESENT_1,
1141 &pvt->b1_ambpresent1);
1142 debugf2("\t\tAMB-Branch 1-present1 0x%x:\n",
1143 pvt->b1_ambpresent1);
1146 /* Go and determine the size of each DIMM and place in an
1148 calculate_dimm_size(pvt);
1152 * i5000_init_csrows Initialize the 'csrows' table within
1153 * the mci control structure with the
1154 * addressing of memory.
1158 * 1 no actual memory found on this MC
1160 static int i5000_init_csrows(struct mem_ctl_info *mci)
1162 struct i5000_pvt *pvt;
1163 struct csrow_info *p_csrow;
1164 int empty, channel_count;
1171 pvt = mci->pvt_info;
1173 channel_count = pvt->maxch;
1174 max_csrows = pvt->maxdimmperch * 2;
1176 empty = 1; /* Assume NO memory */
1178 for (csrow = 0; csrow < max_csrows; csrow++) {
1179 p_csrow = &mci->csrows[csrow];
1181 p_csrow->csrow_idx = csrow;
1183 /* use branch 0 for the basis */
1184 mtr = pvt->b0_mtr[csrow >> 1];
1186 /* if no DIMMS on this row, continue */
1187 if (!MTR_DIMMS_PRESENT(mtr))
1190 /* FAKE OUT VALUES, FIXME */
1191 p_csrow->first_page = 0 + csrow * 20;
1192 p_csrow->last_page = 9 + csrow * 20;
1193 p_csrow->page_mask = 0xFFF;
1198 for (channel = 0; channel < pvt->maxch; channel++) {
1199 csrow_megs += pvt->dimm_info[csrow][channel].megabytes;
1202 p_csrow->nr_pages = csrow_megs << 8;
1204 /* Assume DDR2 for now */
1205 p_csrow->mtype = MEM_FB_DDR2;
1207 /* ask what device type on this row */
1208 if (MTR_DRAM_WIDTH(mtr))
1209 p_csrow->dtype = DEV_X8;
1211 p_csrow->dtype = DEV_X4;
1213 p_csrow->edac_mode = EDAC_S8ECD8ED;
1222 * i5000_enable_error_reporting
1223 * Turn on the memory reporting features of the hardware
1225 static void i5000_enable_error_reporting(struct mem_ctl_info *mci)
1227 struct i5000_pvt *pvt;
1230 pvt = mci->pvt_info;
1232 /* Read the FBD Error Mask Register */
1233 pci_read_config_dword(pvt->branchmap_werrors, EMASK_FBD,
1236 /* Enable with a '0' */
1237 fbd_error_mask &= ~(ENABLE_EMASK_ALL);
1239 pci_write_config_dword(pvt->branchmap_werrors, EMASK_FBD,
1244 * i5000_get_dimm_and_channel_counts(pdev, &num_csrows, &num_channels)
1246 * ask the device how many channels are present and how many CSROWS
1249 static void i5000_get_dimm_and_channel_counts(struct pci_dev *pdev,
1250 int *num_dimms_per_channel,
1255 /* Need to retrieve just how many channels and dimms per channel are
1256 * supported on this memory controller
1258 pci_read_config_byte(pdev, MAXDIMMPERCH, &value);
1259 *num_dimms_per_channel = (int)value *2;
1261 pci_read_config_byte(pdev, MAXCH, &value);
1262 *num_channels = (int)value;
1266 * i5000_probe1 Probe for ONE instance of device to see if it is
1269 * 0 for FOUND a device
1270 * < 0 for error code
1272 static int i5000_probe1(struct pci_dev *pdev, int dev_idx)
1274 struct mem_ctl_info *mci;
1275 struct i5000_pvt *pvt;
1277 int num_dimms_per_channel;
1280 debugf0("MC: " __FILE__ ": %s(), pdev bus %u dev=0x%x fn=0x%x\n",
1283 PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn));
1285 /* We only are looking for func 0 of the set */
1286 if (PCI_FUNC(pdev->devfn) != 0)
1289 /* Ask the devices for the number of CSROWS and CHANNELS so
1290 * that we can calculate the memory resources, etc
1292 * The Chipset will report what it can handle which will be greater
1293 * or equal to what the motherboard manufacturer will implement.
1295 * As we don't have a motherboard identification routine to determine
1296 * actual number of slots/dimms per channel, we thus utilize the
1297 * resource as specified by the chipset. Thus, we might have
1298 * have more DIMMs per channel than actually on the mobo, but this
1299 * allows the driver to support upto the chipset max, without
1300 * some fancy mobo determination.
1302 i5000_get_dimm_and_channel_counts(pdev, &num_dimms_per_channel,
1304 num_csrows = num_dimms_per_channel * 2;
1306 debugf0("MC: %s(): Number of - Channels= %d DIMMS= %d CSROWS= %d\n",
1307 __func__, num_channels, num_dimms_per_channel, num_csrows);
1309 /* allocate a new MC control structure */
1310 mci = edac_mc_alloc(sizeof(*pvt), num_csrows, num_channels, 0);
1315 debugf0("MC: " __FILE__ ": %s(): mci = %p\n", __func__, mci);
1317 mci->dev = &pdev->dev; /* record ptr to the generic device */
1319 pvt = mci->pvt_info;
1320 pvt->system_address = pdev; /* Record this device in our private */
1321 pvt->maxch = num_channels;
1322 pvt->maxdimmperch = num_dimms_per_channel;
1324 /* 'get' the pci devices we want to reserve for our use */
1325 if (i5000_get_devices(mci, dev_idx))
1328 /* Time to get serious */
1329 i5000_get_mc_regs(mci); /* retrieve the hardware registers */
1332 mci->mtype_cap = MEM_FLAG_FB_DDR2;
1333 mci->edac_ctl_cap = EDAC_FLAG_NONE;
1334 mci->edac_cap = EDAC_FLAG_NONE;
1335 mci->mod_name = "i5000_edac.c";
1336 mci->mod_ver = I5000_REVISION;
1337 mci->ctl_name = i5000_devs[dev_idx].ctl_name;
1338 mci->dev_name = pci_name(pdev);
1339 mci->ctl_page_to_phys = NULL;
1341 /* Set the function pointer to an actual operation function */
1342 mci->edac_check = i5000_check_error;
1344 /* initialize the MC control structure 'csrows' table
1345 * with the mapping and control information */
1346 if (i5000_init_csrows(mci)) {
1347 debugf0("MC: Setting mci->edac_cap to EDAC_FLAG_NONE\n"
1348 " because i5000_init_csrows() returned nonzero "
1350 mci->edac_cap = EDAC_FLAG_NONE; /* no csrows found */
1352 debugf1("MC: Enable error reporting now\n");
1353 i5000_enable_error_reporting(mci);
1356 /* add this new MC control structure to EDAC's list of MCs */
1357 if (edac_mc_add_mc(mci)) {
1358 debugf0("MC: " __FILE__
1359 ": %s(): failed edac_mc_add_mc()\n", __func__);
1360 /* FIXME: perhaps some code should go here that disables error
1361 * reporting if we just enabled it
1366 i5000_clear_error(mci);
1368 /* allocating generic PCI control info */
1369 i5000_pci = edac_pci_create_generic_ctl(&pdev->dev, EDAC_MOD_STR);
1372 "%s(): Unable to create PCI control\n",
1375 "%s(): PCI error report via EDAC not setup\n",
1381 /* Error exit unwinding stack */
1384 i5000_put_devices(mci);
1392 * i5000_init_one constructor for one instance of device
1398 static int __devinit i5000_init_one(struct pci_dev *pdev,
1399 const struct pci_device_id *id)
1403 debugf0("MC: " __FILE__ ": %s()\n", __func__);
1405 /* wake up device */
1406 rc = pci_enable_device(pdev);
1410 /* now probe and enable the device */
1411 return i5000_probe1(pdev, id->driver_data);
1415 * i5000_remove_one destructor for one instance of device
1418 static void __devexit i5000_remove_one(struct pci_dev *pdev)
1420 struct mem_ctl_info *mci;
1422 debugf0(__FILE__ ": %s()\n", __func__);
1425 edac_pci_release_generic_ctl(i5000_pci);
1427 if ((mci = edac_mc_del_mc(&pdev->dev)) == NULL)
1430 /* retrieve references to resources, and free those resources */
1431 i5000_put_devices(mci);
1437 * pci_device_id table for which devices we are looking for
1439 * The "E500P" device is the first device supported.
1441 static const struct pci_device_id i5000_pci_tbl[] __devinitdata = {
1442 {PCI_DEVICE(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_I5000_DEV16),
1443 .driver_data = I5000P},
1445 {0,} /* 0 terminated list. */
1448 MODULE_DEVICE_TABLE(pci, i5000_pci_tbl);
1451 * i5000_driver pci_driver structure for this module
1454 static struct pci_driver i5000_driver = {
1455 .name = KBUILD_BASENAME,
1456 .probe = i5000_init_one,
1457 .remove = __devexit_p(i5000_remove_one),
1458 .id_table = i5000_pci_tbl,
1462 * i5000_init Module entry function
1463 * Try to initialize this module for its devices
1465 static int __init i5000_init(void)
1469 debugf2("MC: " __FILE__ ": %s()\n", __func__);
1471 /* Ensure that the OPSTATE is set correctly for POLL or NMI */
1474 pci_rc = pci_register_driver(&i5000_driver);
1476 return (pci_rc < 0) ? pci_rc : 0;
1480 * i5000_exit() Module exit function
1481 * Unregister the driver
1483 static void __exit i5000_exit(void)
1485 debugf2("MC: " __FILE__ ": %s()\n", __func__);
1486 pci_unregister_driver(&i5000_driver);
1489 module_init(i5000_init);
1490 module_exit(i5000_exit);
1492 MODULE_LICENSE("GPL");
1494 ("Linux Networx (http://lnxi.com) Doug Thompson <norsk5@xmission.com>");
1495 MODULE_DESCRIPTION("MC Driver for Intel I5000 memory controllers - "
1498 module_param(edac_op_state, int, 0444);
1499 MODULE_PARM_DESC(edac_op_state, "EDAC Error Reporting state: 0=Poll,1=NMI");