2 * Intel 7300 class Memory Controllers kernel module (Clarksboro)
4 * This file may be distributed under the terms of the
5 * GNU General Public License version 2 only.
7 * Copyright (c) 2010 by:
8 * Mauro Carvalho Chehab <mchehab@redhat.com>
10 * Red Hat Inc. http://www.redhat.com
12 * Intel 7300 Chipset Memory Controller Hub (MCH) - Datasheet
13 * http://www.intel.com/Assets/PDF/datasheet/318082.pdf
15 * TODO: The chipset allow checking for PCI Express errors also. Currently,
16 * the driver covers only memory error errors
18 * This driver uses "csrows" EDAC attribute to represent DIMM slot#
21 #include <linux/module.h>
22 #include <linux/init.h>
23 #include <linux/pci.h>
24 #include <linux/pci_ids.h>
25 #include <linux/slab.h>
26 #include <linux/edac.h>
27 #include <linux/mmzone.h>
29 #include "edac_core.h"
32 * Alter this version for the I7300 module when modifications are made
34 #define I7300_REVISION " Ver: 1.0.0 " __DATE__
36 #define EDAC_MOD_STR "i7300_edac"
38 #define i7300_printk(level, fmt, arg...) \
39 edac_printk(level, "i7300", fmt, ##arg)
41 #define i7300_mc_printk(mci, level, fmt, arg...) \
42 edac_mc_chipset_printk(mci, level, "i7300", fmt, ##arg)
44 /***********************************************
45 * i7300 Limit constants Structs and static vars
46 ***********************************************/
49 * Memory topology is organized as:
50 * Branch 0 - 2 channels: channels 0 and 1 (FDB0 PCI dev 21.0)
51 * Branch 1 - 2 channels: channels 2 and 3 (FDB1 PCI dev 22.0)
52 * Each channel can have to 8 DIMM sets (called as SLOTS)
53 * Slots should generally be filled in pairs
54 * Except on Single Channel mode of operation
55 * just slot 0/channel0 filled on this mode
56 * On normal operation mode, the two channels on a branch should be
57 * filled together for the same SLOT#
58 * When in mirrored mode, Branch 1 replicate memory at Branch 0, so, the four
59 * channels on both branches should be filled
62 /* Limits for i7300 */
64 #define MAX_BRANCHES 2
65 #define MAX_CH_PER_BRANCH 2
66 #define MAX_CHANNELS (MAX_CH_PER_BRANCH * MAX_BRANCHES)
69 #define to_channel(ch, branch) ((((branch)) << 1) | (ch))
71 #define to_csrow(slot, ch, branch) \
72 (to_channel(ch, branch) | ((slot) << 2))
74 /* Device name and register DID (Device ID) */
75 struct i7300_dev_info
{
76 const char *ctl_name
; /* name for this device */
77 u16 fsb_mapping_errors
; /* DID for the branchmap,control */
80 /* Table of devices attributes supported by this driver */
81 static const struct i7300_dev_info i7300_devs
[] = {
84 .fsb_mapping_errors
= PCI_DEVICE_ID_INTEL_I7300_MCH_ERR
,
88 struct i7300_dimm_info
{
89 int megabytes
; /* size, 0 means not present */
92 /* driver private data structure */
94 struct pci_dev
*pci_dev_16_0_fsb_ctlr
; /* 16.0 */
95 struct pci_dev
*pci_dev_16_1_fsb_addr_map
; /* 16.1 */
96 struct pci_dev
*pci_dev_16_2_fsb_err_regs
; /* 16.2 */
97 struct pci_dev
*pci_dev_2x_0_fbd_branch
[MAX_BRANCHES
]; /* 21.0 and 22.0 */
99 u16 tolm
; /* top of low memory */
100 u64 ambase
; /* AMB BAR */
102 u32 mc_settings
; /* Report several settings */
105 u16 mir
[MAX_MIR
]; /* Memory Interleave Reg*/
107 u16 mtr
[MAX_SLOTS
][MAX_BRANCHES
]; /* Memory Technlogy Reg */
108 u16 ambpresent
[MAX_CHANNELS
]; /* AMB present regs */
110 /* DIMM information matrix, allocating architecture maximums */
111 struct i7300_dimm_info dimm_info
[MAX_SLOTS
][MAX_CHANNELS
];
113 /* Temporary buffer for use when preparing error messages */
114 char *tmp_prt_buffer
;
117 /* FIXME: Why do we need to have this static? */
118 static struct edac_pci_ctl_info
*i7300_pci
;
120 /***************************************************
121 * i7300 Register definitions for memory enumeration
122 ***************************************************/
126 * Function 0: System Address (not documented)
127 * Function 1: Memory Branch Map, Control, Errors Register
130 /* OFFSETS for Function 0 */
131 #define AMBASE 0x48 /* AMB Mem Mapped Reg Region Base */
132 #define MAXCH 0x56 /* Max Channel Number */
133 #define MAXDIMMPERCH 0x57 /* Max DIMM PER Channel Number */
135 /* OFFSETS for Function 1 */
136 #define MC_SETTINGS 0x40
137 #define IS_MIRRORED(mc) ((mc) & (1 << 16))
138 #define IS_ECC_ENABLED(mc) ((mc) & (1 << 5))
139 #define IS_RETRY_ENABLED(mc) ((mc) & (1 << 31))
140 #define IS_SCRBALGO_ENHANCED(mc) ((mc) & (1 << 8))
142 #define MC_SETTINGS_A 0x58
143 #define IS_SINGLE_MODE(mca) ((mca) & (1 << 14))
152 * Note: Other Intel EDAC drivers use AMBPRESENT to identify if the available
153 * memory. From datasheet item 7.3.1 (FB-DIMM technology & organization), it
154 * seems that we cannot use this information directly for the same usage.
155 * Each memory slot may have up to 2 AMB interfaces, one for income and another
156 * for outcome interface to the next slot.
157 * For now, the driver just stores the AMB present registers, but rely only at
158 * the MTR info to detect memory.
159 * Datasheet is also not clear about how to map each AMBPRESENT registers to
160 * one of the 4 available channels.
162 #define AMBPRESENT_0 0x64
163 #define AMBPRESENT_1 0x66
165 static const u16 mtr_regs
[MAX_SLOTS
] = {
166 0x80, 0x84, 0x88, 0x8c,
167 0x82, 0x86, 0x8a, 0x8e
171 * Defines to extract the vaious fields from the
172 * MTRx - Memory Technology Registers
174 #define MTR_DIMMS_PRESENT(mtr) ((mtr) & (1 << 8))
175 #define MTR_DIMMS_ETHROTTLE(mtr) ((mtr) & (1 << 7))
176 #define MTR_DRAM_WIDTH(mtr) (((mtr) & (1 << 6)) ? 8 : 4)
177 #define MTR_DRAM_BANKS(mtr) (((mtr) & (1 << 5)) ? 8 : 4)
178 #define MTR_DIMM_RANKS(mtr) (((mtr) & (1 << 4)) ? 1 : 0)
179 #define MTR_DIMM_ROWS(mtr) (((mtr) >> 2) & 0x3)
180 #define MTR_DRAM_BANKS_ADDR_BITS 2
181 #define MTR_DIMM_ROWS_ADDR_BITS(mtr) (MTR_DIMM_ROWS(mtr) + 13)
182 #define MTR_DIMM_COLS(mtr) ((mtr) & 0x3)
183 #define MTR_DIMM_COLS_ADDR_BITS(mtr) (MTR_DIMM_COLS(mtr) + 10)
185 #ifdef CONFIG_EDAC_DEBUG
187 static const char *numrow_toString
[] = {
195 static const char *numcol_toString
[] = {
196 "1,024 - 10 columns",
197 "2,048 - 11 columns",
198 "4,096 - 12 columns",
203 /************************************************
204 * i7300 Register definitions for error detection
205 ************************************************/
208 * Device 16.1: FBD Error Registers
210 #define FERR_FAT_FBD 0x98
211 static const char *ferr_fat_fbd_name
[] = {
212 [22] = "Non-Redundant Fast Reset Timeout",
213 [2] = ">Tmid Thermal event with intelligent throttling disabled",
214 [1] = "Memory or FBD configuration CRC read error",
215 [0] = "Memory Write error on non-redundant retry or "
216 "FBD configuration Write error on retry",
218 #define GET_FBD_FAT_IDX(fbderr) (fbderr & (3 << 28))
219 #define FERR_FAT_FBD_ERR_MASK ((1 << 0) | (1 << 1) | (1 << 2) | (1 << 3))
221 #define FERR_NF_FBD 0xa0
222 static const char *ferr_nf_fbd_name
[] = {
223 [24] = "DIMM-Spare Copy Completed",
224 [23] = "DIMM-Spare Copy Initiated",
225 [22] = "Redundant Fast Reset Timeout",
226 [21] = "Memory Write error on redundant retry",
227 [18] = "SPD protocol Error",
228 [17] = "FBD Northbound parity error on FBD Sync Status",
229 [16] = "Correctable Patrol Data ECC",
230 [15] = "Correctable Resilver- or Spare-Copy Data ECC",
231 [14] = "Correctable Mirrored Demand Data ECC",
232 [13] = "Correctable Non-Mirrored Demand Data ECC",
233 [11] = "Memory or FBD configuration CRC read error",
234 [10] = "FBD Configuration Write error on first attempt",
235 [9] = "Memory Write error on first attempt",
236 [8] = "Non-Aliased Uncorrectable Patrol Data ECC",
237 [7] = "Non-Aliased Uncorrectable Resilver- or Spare-Copy Data ECC",
238 [6] = "Non-Aliased Uncorrectable Mirrored Demand Data ECC",
239 [5] = "Non-Aliased Uncorrectable Non-Mirrored Demand Data ECC",
240 [4] = "Aliased Uncorrectable Patrol Data ECC",
241 [3] = "Aliased Uncorrectable Resilver- or Spare-Copy Data ECC",
242 [2] = "Aliased Uncorrectable Mirrored Demand Data ECC",
243 [1] = "Aliased Uncorrectable Non-Mirrored Demand Data ECC",
244 [0] = "Uncorrectable Data ECC on Replay",
246 #define GET_FBD_NF_IDX(fbderr) (fbderr & (3 << 28))
247 #define FERR_NF_FBD_ERR_MASK ((1 << 24) | (1 << 23) | (1 << 22) | (1 << 21) |\
248 (1 << 18) | (1 << 17) | (1 << 16) | (1 << 15) |\
249 (1 << 14) | (1 << 13) | (1 << 11) | (1 << 10) |\
250 (1 << 9) | (1 << 8) | (1 << 7) | (1 << 6) |\
251 (1 << 5) | (1 << 4) | (1 << 3) | (1 << 2) |\
254 #define EMASK_FBD 0xa8
255 #define EMASK_FBD_ERR_MASK ((1 << 27) | (1 << 26) | (1 << 25) | (1 << 24) |\
256 (1 << 22) | (1 << 21) | (1 << 20) | (1 << 19) |\
257 (1 << 18) | (1 << 17) | (1 << 16) | (1 << 14) |\
258 (1 << 13) | (1 << 12) | (1 << 11) | (1 << 10) |\
259 (1 << 9) | (1 << 8) | (1 << 7) | (1 << 6) |\
260 (1 << 5) | (1 << 4) | (1 << 3) | (1 << 2) |\
264 * Device 16.2: Global Error Registers
267 #define FERR_GLOBAL_HI 0x48
268 static const char *ferr_global_hi_name
[] = {
269 [3] = "FSB 3 Fatal Error",
270 [2] = "FSB 2 Fatal Error",
271 [1] = "FSB 1 Fatal Error",
272 [0] = "FSB 0 Fatal Error",
274 #define ferr_global_hi_is_fatal(errno) 1
276 #define FERR_GLOBAL_LO 0x40
277 static const char *ferr_global_lo_name
[] = {
278 [31] = "Internal MCH Fatal Error",
279 [30] = "Intel QuickData Technology Device Fatal Error",
280 [29] = "FSB1 Fatal Error",
281 [28] = "FSB0 Fatal Error",
282 [27] = "FBD Channel 3 Fatal Error",
283 [26] = "FBD Channel 2 Fatal Error",
284 [25] = "FBD Channel 1 Fatal Error",
285 [24] = "FBD Channel 0 Fatal Error",
286 [23] = "PCI Express Device 7Fatal Error",
287 [22] = "PCI Express Device 6 Fatal Error",
288 [21] = "PCI Express Device 5 Fatal Error",
289 [20] = "PCI Express Device 4 Fatal Error",
290 [19] = "PCI Express Device 3 Fatal Error",
291 [18] = "PCI Express Device 2 Fatal Error",
292 [17] = "PCI Express Device 1 Fatal Error",
293 [16] = "ESI Fatal Error",
294 [15] = "Internal MCH Non-Fatal Error",
295 [14] = "Intel QuickData Technology Device Non Fatal Error",
296 [13] = "FSB1 Non-Fatal Error",
297 [12] = "FSB 0 Non-Fatal Error",
298 [11] = "FBD Channel 3 Non-Fatal Error",
299 [10] = "FBD Channel 2 Non-Fatal Error",
300 [9] = "FBD Channel 1 Non-Fatal Error",
301 [8] = "FBD Channel 0 Non-Fatal Error",
302 [7] = "PCI Express Device 7 Non-Fatal Error",
303 [6] = "PCI Express Device 6 Non-Fatal Error",
304 [5] = "PCI Express Device 5 Non-Fatal Error",
305 [4] = "PCI Express Device 4 Non-Fatal Error",
306 [3] = "PCI Express Device 3 Non-Fatal Error",
307 [2] = "PCI Express Device 2 Non-Fatal Error",
308 [1] = "PCI Express Device 1 Non-Fatal Error",
309 [0] = "ESI Non-Fatal Error",
311 #define ferr_global_lo_is_fatal(errno) ((errno < 16) ? 0 : 1)
313 #define NRECMEMA 0xbe
314 #define NRECMEMA_BANK(v) (((v) >> 12) & 7)
315 #define NRECMEMA_RANK(v) (((v) >> 8) & 15)
317 #define NRECMEMB 0xc0
318 #define NRECMEMB_IS_WR(v) ((v) & (1 << 31))
319 #define NRECMEMB_CAS(v) (((v) >> 16) & 0x1fff)
320 #define NRECMEMB_RAS(v) ((v) & 0xffff)
325 #define IS_SECOND_CH(v) ((v) * (1 << 17))
328 #define RECMEMA_BANK(v) (((v) >> 12) & 7)
329 #define RECMEMA_RANK(v) (((v) >> 8) & 15)
332 #define RECMEMB_IS_WR(v) ((v) & (1 << 31))
333 #define RECMEMB_CAS(v) (((v) >> 16) & 0x1fff)
334 #define RECMEMB_RAS(v) ((v) & 0xffff)
336 /********************************************
337 * i7300 Functions related to error detection
338 ********************************************/
341 * get_err_from_table() - Gets the error message from a table
342 * @table: table name (array of char *)
343 * @size: number of elements at the table
344 * @pos: position of the element to be returned
346 * This is a small routine that gets the pos-th element of a table. If the
347 * element doesn't exist (or it is empty), it returns "reserved".
348 * Instead of calling it directly, the better is to call via the macro
349 * GET_ERR_FROM_TABLE(), that automatically checks the table size via
352 static const char *get_err_from_table(const char *table
[], int size
, int pos
)
354 if (unlikely(pos
>= size
))
357 if (unlikely(!table
[pos
]))
363 #define GET_ERR_FROM_TABLE(table, pos) \
364 get_err_from_table(table, ARRAY_SIZE(table), pos)
367 * i7300_process_error_global() - Retrieve the hardware error information from
368 * the hardware global error registers and
370 * @mci: struct mem_ctl_info pointer
372 static void i7300_process_error_global(struct mem_ctl_info
*mci
)
374 struct i7300_pvt
*pvt
;
376 unsigned long errors
;
377 const char *specific
;
382 /* read in the 1st FATAL error register */
383 pci_read_config_dword(pvt
->pci_dev_16_2_fsb_err_regs
,
384 FERR_GLOBAL_HI
, &value
);
385 if (unlikely(value
)) {
387 errnum
= find_first_bit(&errors
,
388 ARRAY_SIZE(ferr_global_hi_name
));
389 specific
= GET_ERR_FROM_TABLE(ferr_global_hi_name
, errnum
);
390 is_fatal
= ferr_global_hi_is_fatal(errnum
);
392 /* Clear the error bit */
393 pci_write_config_dword(pvt
->pci_dev_16_2_fsb_err_regs
,
394 FERR_GLOBAL_HI
, value
);
399 pci_read_config_dword(pvt
->pci_dev_16_2_fsb_err_regs
,
400 FERR_GLOBAL_LO
, &value
);
401 if (unlikely(value
)) {
403 errnum
= find_first_bit(&errors
,
404 ARRAY_SIZE(ferr_global_lo_name
));
405 specific
= GET_ERR_FROM_TABLE(ferr_global_lo_name
, errnum
);
406 is_fatal
= ferr_global_lo_is_fatal(errnum
);
408 /* Clear the error bit */
409 pci_write_config_dword(pvt
->pci_dev_16_2_fsb_err_regs
,
410 FERR_GLOBAL_LO
, value
);
417 i7300_mc_printk(mci
, KERN_EMERG
, "%s misc error: %s\n",
418 is_fatal
? "Fatal" : "NOT fatal", specific
);
422 * i7300_process_fbd_error() - Retrieve the hardware error information from
423 * the FBD error registers and sends it via
424 * EDAC error API calls
425 * @mci: struct mem_ctl_info pointer
427 static void i7300_process_fbd_error(struct mem_ctl_info
*mci
)
429 struct i7300_pvt
*pvt
;
432 unsigned branch
, channel
, bank
, rank
, cas
, ras
;
435 unsigned long errors
;
436 const char *specific
;
441 /* read in the 1st FATAL error register */
442 pci_read_config_dword(pvt
->pci_dev_16_1_fsb_addr_map
,
443 FERR_FAT_FBD
, &value
);
444 if (unlikely(value
& FERR_FAT_FBD_ERR_MASK
)) {
445 errors
= value
& FERR_FAT_FBD_ERR_MASK
;
446 errnum
= find_first_bit(&errors
,
447 ARRAY_SIZE(ferr_fat_fbd_name
));
448 specific
= GET_ERR_FROM_TABLE(ferr_fat_fbd_name
, errnum
);
450 branch
= (GET_FBD_FAT_IDX(value
) == 2) ? 1 : 0;
451 pci_read_config_word(pvt
->pci_dev_16_1_fsb_addr_map
,
453 bank
= NRECMEMA_BANK(val16
);
454 rank
= NRECMEMA_RANK(val16
);
456 pci_read_config_dword(pvt
->pci_dev_16_1_fsb_addr_map
,
459 is_wr
= NRECMEMB_IS_WR(value
);
460 cas
= NRECMEMB_CAS(value
);
461 ras
= NRECMEMB_RAS(value
);
463 snprintf(pvt
->tmp_prt_buffer
, PAGE_SIZE
,
464 "FATAL (Branch=%d DRAM-Bank=%d %s "
465 "RAS=%d CAS=%d Err=0x%lx (%s))",
467 is_wr
? "RDWR" : "RD",
471 /* Call the helper to output message */
472 edac_mc_handle_fbd_ue(mci
, rank
, branch
<< 1,
474 pvt
->tmp_prt_buffer
);
477 /* read in the 1st NON-FATAL error register */
478 pci_read_config_dword(pvt
->pci_dev_16_1_fsb_addr_map
,
479 FERR_NF_FBD
, &value
);
480 if (unlikely(value
& FERR_NF_FBD_ERR_MASK
)) {
481 errors
= value
& FERR_NF_FBD_ERR_MASK
;
482 errnum
= find_first_bit(&errors
,
483 ARRAY_SIZE(ferr_nf_fbd_name
));
484 specific
= GET_ERR_FROM_TABLE(ferr_nf_fbd_name
, errnum
);
486 /* Clear the error bit */
487 pci_write_config_dword(pvt
->pci_dev_16_2_fsb_err_regs
,
488 FERR_GLOBAL_LO
, value
);
490 pci_read_config_dword(pvt
->pci_dev_16_1_fsb_addr_map
,
493 branch
= (GET_FBD_FAT_IDX(value
) == 2) ? 1 : 0;
494 pci_read_config_word(pvt
->pci_dev_16_1_fsb_addr_map
,
496 bank
= RECMEMA_BANK(val16
);
497 rank
= RECMEMA_RANK(val16
);
499 pci_read_config_dword(pvt
->pci_dev_16_1_fsb_addr_map
,
502 is_wr
= RECMEMB_IS_WR(value
);
503 cas
= RECMEMB_CAS(value
);
504 ras
= RECMEMB_RAS(value
);
506 pci_read_config_dword(pvt
->pci_dev_16_1_fsb_addr_map
,
509 channel
= (branch
<< 1);
510 if (IS_SECOND_CH(value
))
513 /* Form out message */
514 snprintf(pvt
->tmp_prt_buffer
, PAGE_SIZE
,
515 "Corrected error (Branch=%d, Channel %d), "
517 "RAS=%d CAS=%d, CE Err=0x%lx, Syndrome=0x%08x(%s))",
520 is_wr
? "RDWR" : "RD",
522 errors
, syndrome
, specific
);
525 * Call the helper to output message
526 * NOTE: Errors are reported per-branch, and not per-channel
527 * Currently, we don't know how to identify the right
530 edac_mc_handle_fbd_ce(mci
, rank
, channel
,
531 pvt
->tmp_prt_buffer
);
537 * i7300_check_error() - Calls the error checking subroutines
538 * @mci: struct mem_ctl_info pointer
540 static void i7300_check_error(struct mem_ctl_info
*mci
)
542 i7300_process_error_global(mci
);
543 i7300_process_fbd_error(mci
);
547 * i7300_clear_error() - Clears the error registers
548 * @mci: struct mem_ctl_info pointer
550 static void i7300_clear_error(struct mem_ctl_info
*mci
)
552 struct i7300_pvt
*pvt
= mci
->pvt_info
;
555 * All error values are RWC - we need to read and write 1 to the
556 * bit that we want to cleanup
559 /* Clear global error registers */
560 pci_read_config_dword(pvt
->pci_dev_16_2_fsb_err_regs
,
561 FERR_GLOBAL_HI
, &value
);
562 pci_write_config_dword(pvt
->pci_dev_16_2_fsb_err_regs
,
563 FERR_GLOBAL_HI
, value
);
565 pci_read_config_dword(pvt
->pci_dev_16_2_fsb_err_regs
,
566 FERR_GLOBAL_LO
, &value
);
567 pci_write_config_dword(pvt
->pci_dev_16_2_fsb_err_regs
,
568 FERR_GLOBAL_LO
, value
);
570 /* Clear FBD error registers */
571 pci_read_config_dword(pvt
->pci_dev_16_1_fsb_addr_map
,
572 FERR_FAT_FBD
, &value
);
573 pci_write_config_dword(pvt
->pci_dev_16_1_fsb_addr_map
,
574 FERR_FAT_FBD
, value
);
576 pci_read_config_dword(pvt
->pci_dev_16_1_fsb_addr_map
,
577 FERR_NF_FBD
, &value
);
578 pci_write_config_dword(pvt
->pci_dev_16_1_fsb_addr_map
,
583 * i7300_enable_error_reporting() - Enable the memory reporting logic at the
585 * @mci: struct mem_ctl_info pointer
587 static void i7300_enable_error_reporting(struct mem_ctl_info
*mci
)
589 struct i7300_pvt
*pvt
= mci
->pvt_info
;
592 /* Read the FBD Error Mask Register */
593 pci_read_config_dword(pvt
->pci_dev_16_1_fsb_addr_map
,
594 EMASK_FBD
, &fbd_error_mask
);
596 /* Enable with a '0' */
597 fbd_error_mask
&= ~(EMASK_FBD_ERR_MASK
);
599 pci_write_config_dword(pvt
->pci_dev_16_1_fsb_addr_map
,
600 EMASK_FBD
, fbd_error_mask
);
603 /************************************************
604 * i7300 Functions related to memory enumberation
605 ************************************************/
608 * decode_mtr() - Decodes the MTR descriptor, filling the edac structs
609 * @pvt: pointer to the private data struct used by i7300 driver
610 * @slot: DIMM slot (0 to 7)
611 * @ch: Channel number within the branch (0 or 1)
612 * @branch: Branch number (0 or 1)
613 * @dinfo: Pointer to DIMM info where dimm size is stored
614 * @p_csrow: Pointer to the struct csrow_info that corresponds to that element
616 static int decode_mtr(struct i7300_pvt
*pvt
,
617 int slot
, int ch
, int branch
,
618 struct i7300_dimm_info
*dinfo
,
619 struct csrow_info
*p_csrow
,
622 int mtr
, ans
, addrBits
, channel
;
624 channel
= to_channel(ch
, branch
);
626 mtr
= pvt
->mtr
[slot
][branch
];
627 ans
= MTR_DIMMS_PRESENT(mtr
) ? 1 : 0;
629 debugf2("\tMTR%d CH%d: DIMMs are %s (mtr)\n",
631 ans
? "Present" : "NOT Present");
633 /* Determine if there is a DIMM present in this DIMM slot */
637 /* Start with the number of bits for a Bank
639 addrBits
= MTR_DRAM_BANKS_ADDR_BITS
;
640 /* Add thenumber of ROW bits */
641 addrBits
+= MTR_DIMM_ROWS_ADDR_BITS(mtr
);
642 /* add the number of COLUMN bits */
643 addrBits
+= MTR_DIMM_COLS_ADDR_BITS(mtr
);
644 /* add the number of RANK bits */
645 addrBits
+= MTR_DIMM_RANKS(mtr
);
647 addrBits
+= 6; /* add 64 bits per DIMM */
648 addrBits
-= 20; /* divide by 2^^20 */
649 addrBits
-= 3; /* 8 bits per bytes */
651 dinfo
->megabytes
= 1 << addrBits
;
652 *nr_pages
= dinfo
->megabytes
<< 8;
654 debugf2("\t\tWIDTH: x%d\n", MTR_DRAM_WIDTH(mtr
));
656 debugf2("\t\tELECTRICAL THROTTLING is %s\n",
657 MTR_DIMMS_ETHROTTLE(mtr
) ? "enabled" : "disabled");
659 debugf2("\t\tNUMBANK: %d bank(s)\n", MTR_DRAM_BANKS(mtr
));
660 debugf2("\t\tNUMRANK: %s\n", MTR_DIMM_RANKS(mtr
) ? "double" : "single");
661 debugf2("\t\tNUMROW: %s\n", numrow_toString
[MTR_DIMM_ROWS(mtr
)]);
662 debugf2("\t\tNUMCOL: %s\n", numcol_toString
[MTR_DIMM_COLS(mtr
)]);
663 debugf2("\t\tSIZE: %d MB\n", dinfo
->megabytes
);
666 p_csrow
->mtype
= MEM_FB_DDR2
;
667 p_csrow
->csrow_idx
= slot
;
668 p_csrow
->page_mask
= 0;
671 * The type of error detection actually depends of the
672 * mode of operation. When it is just one single memory chip, at
673 * socket 0, channel 0, it uses 8-byte-over-32-byte SECDED+ code.
674 * In normal or mirrored mode, it uses Lockstep mode,
675 * with the possibility of using an extended algorithm for x8 memories
676 * See datasheet Sections 7.3.6 to 7.3.8
679 if (IS_SINGLE_MODE(pvt
->mc_settings_a
)) {
680 p_csrow
->edac_mode
= EDAC_SECDED
;
681 debugf2("\t\tECC code is 8-byte-over-32-byte SECDED+ code\n");
683 debugf2("\t\tECC code is on Lockstep mode\n");
684 if (MTR_DRAM_WIDTH(mtr
) == 8)
685 p_csrow
->edac_mode
= EDAC_S8ECD8ED
;
687 p_csrow
->edac_mode
= EDAC_S4ECD4ED
;
690 /* ask what device type on this row */
691 if (MTR_DRAM_WIDTH(mtr
) == 8) {
692 debugf2("\t\tScrub algorithm for x8 is on %s mode\n",
693 IS_SCRBALGO_ENHANCED(pvt
->mc_settings
) ?
694 "enhanced" : "normal");
696 p_csrow
->dtype
= DEV_X8
;
698 p_csrow
->dtype
= DEV_X4
;
704 * print_dimm_size() - Prints dump of the memory organization
705 * @pvt: pointer to the private data struct used by i7300 driver
707 * Useful for debug. If debug is disabled, this routine do nothing
709 static void print_dimm_size(struct i7300_pvt
*pvt
)
711 #ifdef CONFIG_EDAC_DEBUG
712 struct i7300_dimm_info
*dinfo
;
718 p
= pvt
->tmp_prt_buffer
;
720 n
= snprintf(p
, space
, " ");
723 for (channel
= 0; channel
< MAX_CHANNELS
; channel
++) {
724 n
= snprintf(p
, space
, "channel %d | ", channel
);
728 debugf2("%s\n", pvt
->tmp_prt_buffer
);
729 p
= pvt
->tmp_prt_buffer
;
731 n
= snprintf(p
, space
, "-------------------------------"
732 "------------------------------");
735 debugf2("%s\n", pvt
->tmp_prt_buffer
);
736 p
= pvt
->tmp_prt_buffer
;
739 for (slot
= 0; slot
< MAX_SLOTS
; slot
++) {
740 n
= snprintf(p
, space
, "csrow/SLOT %d ", slot
);
744 for (channel
= 0; channel
< MAX_CHANNELS
; channel
++) {
745 dinfo
= &pvt
->dimm_info
[slot
][channel
];
746 n
= snprintf(p
, space
, "%4d MB | ", dinfo
->megabytes
);
751 debugf2("%s\n", pvt
->tmp_prt_buffer
);
752 p
= pvt
->tmp_prt_buffer
;
756 n
= snprintf(p
, space
, "-------------------------------"
757 "------------------------------");
760 debugf2("%s\n", pvt
->tmp_prt_buffer
);
761 p
= pvt
->tmp_prt_buffer
;
767 * i7300_init_csrows() - Initialize the 'csrows' table within
768 * the mci control structure with the
769 * addressing of memory.
770 * @mci: struct mem_ctl_info pointer
772 static int i7300_init_csrows(struct mem_ctl_info
*mci
)
774 struct i7300_pvt
*pvt
;
775 struct i7300_dimm_info
*dinfo
;
776 struct csrow_info
*p_csrow
;
779 int ch
, branch
, slot
, channel
;
780 u32 last_page
= 0, nr_pages
;
784 debugf2("Memory Technology Registers:\n");
786 /* Get the AMB present registers for the four channels */
787 for (branch
= 0; branch
< MAX_BRANCHES
; branch
++) {
788 /* Read and dump branch 0's MTRs */
789 channel
= to_channel(0, branch
);
790 pci_read_config_word(pvt
->pci_dev_2x_0_fbd_branch
[branch
],
792 &pvt
->ambpresent
[channel
]);
793 debugf2("\t\tAMB-present CH%d = 0x%x:\n",
794 channel
, pvt
->ambpresent
[channel
]);
796 channel
= to_channel(1, branch
);
797 pci_read_config_word(pvt
->pci_dev_2x_0_fbd_branch
[branch
],
799 &pvt
->ambpresent
[channel
]);
800 debugf2("\t\tAMB-present CH%d = 0x%x:\n",
801 channel
, pvt
->ambpresent
[channel
]);
804 /* Get the set of MTR[0-7] regs by each branch */
805 for (slot
= 0; slot
< MAX_SLOTS
; slot
++) {
806 int where
= mtr_regs
[slot
];
807 for (branch
= 0; branch
< MAX_BRANCHES
; branch
++) {
808 pci_read_config_word(pvt
->pci_dev_2x_0_fbd_branch
[branch
],
810 &pvt
->mtr
[slot
][branch
]);
811 for (ch
= 0; ch
< MAX_BRANCHES
; ch
++) {
812 int channel
= to_channel(ch
, branch
);
814 dinfo
= &pvt
->dimm_info
[slot
][channel
];
815 p_csrow
= &mci
->csrows
[slot
];
817 mtr
= decode_mtr(pvt
, slot
, ch
, branch
,
818 dinfo
, p_csrow
, &nr_pages
);
819 /* if no DIMMS on this row, continue */
820 if (!MTR_DIMMS_PRESENT(mtr
))
823 /* Update per_csrow memory count */
824 p_csrow
->nr_pages
+= nr_pages
;
825 p_csrow
->first_page
= last_page
;
826 last_page
+= nr_pages
;
827 p_csrow
->last_page
= last_page
;
838 * decode_mir() - Decodes Memory Interleave Register (MIR) info
839 * @int mir_no: number of the MIR register to decode
840 * @mir: array with the MIR data cached on the driver
842 static void decode_mir(int mir_no
, u16 mir
[MAX_MIR
])
845 debugf2("MIR%d: limit= 0x%x Branch(es) that participate:"
848 (mir
[mir_no
] >> 4) & 0xfff,
849 (mir
[mir_no
] & 1) ? "B0" : "",
850 (mir
[mir_no
] & 2) ? "B1" : "");
854 * i7300_get_mc_regs() - Get the contents of the MC enumeration registers
855 * @mci: struct mem_ctl_info pointer
857 * Data read is cached internally for its usage when needed
859 static int i7300_get_mc_regs(struct mem_ctl_info
*mci
)
861 struct i7300_pvt
*pvt
;
867 pci_read_config_dword(pvt
->pci_dev_16_0_fsb_ctlr
, AMBASE
,
868 (u32
*) &pvt
->ambase
);
870 debugf2("AMBASE= 0x%lx\n", (long unsigned int)pvt
->ambase
);
872 /* Get the Branch Map regs */
873 pci_read_config_word(pvt
->pci_dev_16_1_fsb_addr_map
, TOLM
, &pvt
->tolm
);
875 debugf2("TOLM (number of 256M regions) =%u (0x%x)\n", pvt
->tolm
,
878 actual_tolm
= (u32
) ((1000l * pvt
->tolm
) >> (30 - 28));
879 debugf2("Actual TOLM byte addr=%u.%03u GB (0x%x)\n",
880 actual_tolm
/1000, actual_tolm
% 1000, pvt
->tolm
<< 28);
882 /* Get memory controller settings */
883 pci_read_config_dword(pvt
->pci_dev_16_1_fsb_addr_map
, MC_SETTINGS
,
885 pci_read_config_dword(pvt
->pci_dev_16_1_fsb_addr_map
, MC_SETTINGS_A
,
886 &pvt
->mc_settings_a
);
888 if (IS_SINGLE_MODE(pvt
->mc_settings_a
))
889 debugf0("Memory controller operating on single mode\n");
891 debugf0("Memory controller operating on %s mode\n",
892 IS_MIRRORED(pvt
->mc_settings
) ? "mirrored" : "non-mirrored");
894 debugf0("Error detection is %s\n",
895 IS_ECC_ENABLED(pvt
->mc_settings
) ? "enabled" : "disabled");
896 debugf0("Retry is %s\n",
897 IS_RETRY_ENABLED(pvt
->mc_settings
) ? "enabled" : "disabled");
899 /* Get Memory Interleave Range registers */
900 pci_read_config_word(pvt
->pci_dev_16_1_fsb_addr_map
, MIR0
,
902 pci_read_config_word(pvt
->pci_dev_16_1_fsb_addr_map
, MIR1
,
904 pci_read_config_word(pvt
->pci_dev_16_1_fsb_addr_map
, MIR2
,
907 /* Decode the MIR regs */
908 for (i
= 0; i
< MAX_MIR
; i
++)
909 decode_mir(i
, pvt
->mir
);
911 rc
= i7300_init_csrows(mci
);
915 /* Go and determine the size of each DIMM and place in an
917 print_dimm_size(pvt
);
922 /*************************************************
923 * i7300 Functions related to device probe/release
924 *************************************************/
927 * i7300_put_devices() - Release the PCI devices
928 * @mci: struct mem_ctl_info pointer
930 static void i7300_put_devices(struct mem_ctl_info
*mci
)
932 struct i7300_pvt
*pvt
;
937 /* Decrement usage count for devices */
938 for (branch
= 0; branch
< MAX_CH_PER_BRANCH
; branch
++)
939 pci_dev_put(pvt
->pci_dev_2x_0_fbd_branch
[branch
]);
940 pci_dev_put(pvt
->pci_dev_16_2_fsb_err_regs
);
941 pci_dev_put(pvt
->pci_dev_16_1_fsb_addr_map
);
945 * i7300_get_devices() - Find and perform 'get' operation on the MCH's
946 * device/functions we want to reference for this driver
947 * @mci: struct mem_ctl_info pointer
949 * Access and prepare the several devices for usage:
950 * I7300 devices used by this driver:
951 * Device 16, functions 0,1 and 2: PCI_DEVICE_ID_INTEL_I7300_MCH_ERR
952 * Device 21 function 0: PCI_DEVICE_ID_INTEL_I7300_MCH_FB0
953 * Device 22 function 0: PCI_DEVICE_ID_INTEL_I7300_MCH_FB1
955 static int __devinit
i7300_get_devices(struct mem_ctl_info
*mci
)
957 struct i7300_pvt
*pvt
;
958 struct pci_dev
*pdev
;
962 /* Attempt to 'get' the MCH register we want */
964 while (!pvt
->pci_dev_16_1_fsb_addr_map
||
965 !pvt
->pci_dev_16_2_fsb_err_regs
) {
966 pdev
= pci_get_device(PCI_VENDOR_ID_INTEL
,
967 PCI_DEVICE_ID_INTEL_I7300_MCH_ERR
, pdev
);
969 /* End of list, leave */
970 i7300_printk(KERN_ERR
,
971 "'system address,Process Bus' "
973 "vendor 0x%x device 0x%x ERR funcs "
976 PCI_DEVICE_ID_INTEL_I7300_MCH_ERR
);
980 /* Store device 16 funcs 1 and 2 */
981 switch (PCI_FUNC(pdev
->devfn
)) {
983 pvt
->pci_dev_16_1_fsb_addr_map
= pdev
;
986 pvt
->pci_dev_16_2_fsb_err_regs
= pdev
;
991 debugf1("System Address, processor bus- PCI Bus ID: %s %x:%x\n",
992 pci_name(pvt
->pci_dev_16_0_fsb_ctlr
),
993 pvt
->pci_dev_16_0_fsb_ctlr
->vendor
,
994 pvt
->pci_dev_16_0_fsb_ctlr
->device
);
995 debugf1("Branchmap, control and errors - PCI Bus ID: %s %x:%x\n",
996 pci_name(pvt
->pci_dev_16_1_fsb_addr_map
),
997 pvt
->pci_dev_16_1_fsb_addr_map
->vendor
,
998 pvt
->pci_dev_16_1_fsb_addr_map
->device
);
999 debugf1("FSB Error Regs - PCI Bus ID: %s %x:%x\n",
1000 pci_name(pvt
->pci_dev_16_2_fsb_err_regs
),
1001 pvt
->pci_dev_16_2_fsb_err_regs
->vendor
,
1002 pvt
->pci_dev_16_2_fsb_err_regs
->device
);
1004 pvt
->pci_dev_2x_0_fbd_branch
[0] = pci_get_device(PCI_VENDOR_ID_INTEL
,
1005 PCI_DEVICE_ID_INTEL_I7300_MCH_FB0
,
1007 if (!pvt
->pci_dev_2x_0_fbd_branch
[0]) {
1008 i7300_printk(KERN_ERR
,
1009 "MC: 'BRANCH 0' device not found:"
1010 "vendor 0x%x device 0x%x Func 0 (broken BIOS?)\n",
1011 PCI_VENDOR_ID_INTEL
, PCI_DEVICE_ID_INTEL_I7300_MCH_FB0
);
1015 pvt
->pci_dev_2x_0_fbd_branch
[1] = pci_get_device(PCI_VENDOR_ID_INTEL
,
1016 PCI_DEVICE_ID_INTEL_I7300_MCH_FB1
,
1018 if (!pvt
->pci_dev_2x_0_fbd_branch
[1]) {
1019 i7300_printk(KERN_ERR
,
1020 "MC: 'BRANCH 1' device not found:"
1021 "vendor 0x%x device 0x%x Func 0 "
1023 PCI_VENDOR_ID_INTEL
,
1024 PCI_DEVICE_ID_INTEL_I7300_MCH_FB1
);
1031 i7300_put_devices(mci
);
1036 * i7300_init_one() - Probe for one instance of the device
1037 * @pdev: struct pci_dev pointer
1038 * @id: struct pci_device_id pointer - currently unused
1040 static int __devinit
i7300_init_one(struct pci_dev
*pdev
,
1041 const struct pci_device_id
*id
)
1043 struct mem_ctl_info
*mci
;
1044 struct i7300_pvt
*pvt
;
1046 int num_dimms_per_channel
;
1050 /* wake up device */
1051 rc
= pci_enable_device(pdev
);
1055 debugf0("MC: " __FILE__
": %s(), pdev bus %u dev=0x%x fn=0x%x\n",
1058 PCI_SLOT(pdev
->devfn
), PCI_FUNC(pdev
->devfn
));
1060 /* We only are looking for func 0 of the set */
1061 if (PCI_FUNC(pdev
->devfn
) != 0)
1064 /* As we don't have a motherboard identification routine to determine
1065 * actual number of slots/dimms per channel, we thus utilize the
1066 * resource as specified by the chipset. Thus, we might have
1067 * have more DIMMs per channel than actually on the mobo, but this
1068 * allows the driver to support upto the chipset max, without
1069 * some fancy mobo determination.
1071 num_dimms_per_channel
= MAX_SLOTS
;
1072 num_channels
= MAX_CHANNELS
;
1073 num_csrows
= MAX_SLOTS
* MAX_CHANNELS
;
1075 debugf0("MC: %s(): Number of - Channels= %d DIMMS= %d CSROWS= %d\n",
1076 __func__
, num_channels
, num_dimms_per_channel
, num_csrows
);
1078 /* allocate a new MC control structure */
1079 mci
= edac_mc_alloc(sizeof(*pvt
), num_csrows
, num_channels
, 0);
1084 debugf0("MC: " __FILE__
": %s(): mci = %p\n", __func__
, mci
);
1086 mci
->dev
= &pdev
->dev
; /* record ptr to the generic device */
1088 pvt
= mci
->pvt_info
;
1089 pvt
->pci_dev_16_0_fsb_ctlr
= pdev
; /* Record this device in our private */
1091 pvt
->tmp_prt_buffer
= kmalloc(PAGE_SIZE
, GFP_KERNEL
);
1092 if (!pvt
->tmp_prt_buffer
) {
1097 /* 'get' the pci devices we want to reserve for our use */
1098 if (i7300_get_devices(mci
))
1102 mci
->mtype_cap
= MEM_FLAG_FB_DDR2
;
1103 mci
->edac_ctl_cap
= EDAC_FLAG_NONE
;
1104 mci
->edac_cap
= EDAC_FLAG_NONE
;
1105 mci
->mod_name
= "i7300_edac.c";
1106 mci
->mod_ver
= I7300_REVISION
;
1107 mci
->ctl_name
= i7300_devs
[0].ctl_name
;
1108 mci
->dev_name
= pci_name(pdev
);
1109 mci
->ctl_page_to_phys
= NULL
;
1111 /* Set the function pointer to an actual operation function */
1112 mci
->edac_check
= i7300_check_error
;
1114 /* initialize the MC control structure 'csrows' table
1115 * with the mapping and control information */
1116 if (i7300_get_mc_regs(mci
)) {
1117 debugf0("MC: Setting mci->edac_cap to EDAC_FLAG_NONE\n"
1118 " because i7300_init_csrows() returned nonzero "
1120 mci
->edac_cap
= EDAC_FLAG_NONE
; /* no csrows found */
1122 debugf1("MC: Enable error reporting now\n");
1123 i7300_enable_error_reporting(mci
);
1126 /* add this new MC control structure to EDAC's list of MCs */
1127 if (edac_mc_add_mc(mci
)) {
1128 debugf0("MC: " __FILE__
1129 ": %s(): failed edac_mc_add_mc()\n", __func__
);
1130 /* FIXME: perhaps some code should go here that disables error
1131 * reporting if we just enabled it
1136 i7300_clear_error(mci
);
1138 /* allocating generic PCI control info */
1139 i7300_pci
= edac_pci_create_generic_ctl(&pdev
->dev
, EDAC_MOD_STR
);
1142 "%s(): Unable to create PCI control\n",
1145 "%s(): PCI error report via EDAC not setup\n",
1151 /* Error exit unwinding stack */
1154 i7300_put_devices(mci
);
1157 kfree(pvt
->tmp_prt_buffer
);
1163 * i7300_remove_one() - Remove the driver
1164 * @pdev: struct pci_dev pointer
1166 static void __devexit
i7300_remove_one(struct pci_dev
*pdev
)
1168 struct mem_ctl_info
*mci
;
1171 debugf0(__FILE__
": %s()\n", __func__
);
1174 edac_pci_release_generic_ctl(i7300_pci
);
1176 mci
= edac_mc_del_mc(&pdev
->dev
);
1180 tmp
= ((struct i7300_pvt
*)mci
->pvt_info
)->tmp_prt_buffer
;
1182 /* retrieve references to resources, and free those resources */
1183 i7300_put_devices(mci
);
1190 * pci_device_id: table for which devices we are looking for
1192 * Has only 8086:360c PCI ID
1194 static const struct pci_device_id i7300_pci_tbl
[] __devinitdata
= {
1195 {PCI_DEVICE(PCI_VENDOR_ID_INTEL
, PCI_DEVICE_ID_INTEL_I7300_MCH_ERR
)},
1196 {0,} /* 0 terminated list. */
1199 MODULE_DEVICE_TABLE(pci
, i7300_pci_tbl
);
1202 * i7300_driver: pci_driver structure for this module
1204 static struct pci_driver i7300_driver
= {
1205 .name
= "i7300_edac",
1206 .probe
= i7300_init_one
,
1207 .remove
= __devexit_p(i7300_remove_one
),
1208 .id_table
= i7300_pci_tbl
,
1212 * i7300_init() - Registers the driver
1214 static int __init
i7300_init(void)
1218 debugf2("MC: " __FILE__
": %s()\n", __func__
);
1220 /* Ensure that the OPSTATE is set correctly for POLL or NMI */
1223 pci_rc
= pci_register_driver(&i7300_driver
);
1225 return (pci_rc
< 0) ? pci_rc
: 0;
1229 * i7300_init() - Unregisters the driver
1231 static void __exit
i7300_exit(void)
1233 debugf2("MC: " __FILE__
": %s()\n", __func__
);
1234 pci_unregister_driver(&i7300_driver
);
1237 module_init(i7300_init
);
1238 module_exit(i7300_exit
);
1240 MODULE_LICENSE("GPL");
1241 MODULE_AUTHOR("Mauro Carvalho Chehab <mchehab@redhat.com>");
1242 MODULE_AUTHOR("Red Hat Inc. (http://www.redhat.com)");
1243 MODULE_DESCRIPTION("MC Driver for Intel I7300 memory controllers - "
1246 module_param(edac_op_state
, int, 0444);
1247 MODULE_PARM_DESC(edac_op_state
, "EDAC Error Reporting state: 0=Poll,1=NMI");