xref: /openbmc/linux/drivers/edac/i5400_edac.c (revision b586a59e)
1 /*
2  * Intel 5400 class Memory Controllers kernel module (Seaburg)
3  *
4  * This file may be distributed under the terms of the
5  * GNU General Public License.
6  *
7  * Copyright (c) 2008 by:
8  *	 Ben Woodard <woodard@redhat.com>
9  *	 Mauro Carvalho Chehab
10  *
11  * Red Hat Inc. https://www.redhat.com
12  *
13  * Forked and adapted from the i5000_edac driver which was
14  * written by Douglas Thompson Linux Networx <norsk5@xmission.com>
15  *
16  * This module is based on the following document:
17  *
18  * Intel 5400 Chipset Memory Controller Hub (MCH) - Datasheet
19  * 	http://developer.intel.com/design/chipsets/datashts/313070.htm
20  *
21  * This Memory Controller manages DDR2 FB-DIMMs. It has 2 branches, each with
22  * 2 channels operating in lockstep no-mirror mode. Each channel can have up to
23  * 4 dimm's, each with up to 8GB.
24  *
25  */
26 
27 #include <linux/module.h>
28 #include <linux/init.h>
29 #include <linux/pci.h>
30 #include <linux/pci_ids.h>
31 #include <linux/slab.h>
32 #include <linux/edac.h>
33 #include <linux/mmzone.h>
34 
35 #include "edac_module.h"
36 
37 /*
38  * Alter this version for the I5400 module when modifications are made
39  */
40 #define I5400_REVISION    " Ver: 1.0.0"
41 
42 #define EDAC_MOD_STR      "i5400_edac"
43 
44 #define i5400_printk(level, fmt, arg...) \
45 	edac_printk(level, "i5400", fmt, ##arg)
46 
47 #define i5400_mc_printk(mci, level, fmt, arg...) \
48 	edac_mc_chipset_printk(mci, level, "i5400", fmt, ##arg)
49 
50 /* Limits for i5400 */
51 #define MAX_BRANCHES		2
52 #define CHANNELS_PER_BRANCH	2
53 #define DIMMS_PER_CHANNEL	4
54 #define	MAX_CHANNELS		(MAX_BRANCHES * CHANNELS_PER_BRANCH)
55 
56 /* Device 16,
57  * Function 0: System Address
58  * Function 1: Memory Branch Map, Control, Errors Register
59  * Function 2: FSB Error Registers
60  *
61  * All 3 functions of Device 16 (0,1,2) share the SAME DID and
62  * uses PCI_DEVICE_ID_INTEL_5400_ERR for device 16 (0,1,2),
63  * PCI_DEVICE_ID_INTEL_5400_FBD0 and PCI_DEVICE_ID_INTEL_5400_FBD1
64  * for device 21 (0,1).
65  */
66 
67 	/* OFFSETS for Function 0 */
68 #define		AMBASE			0x48 /* AMB Mem Mapped Reg Region Base */
69 #define		MAXCH			0x56 /* Max Channel Number */
70 #define		MAXDIMMPERCH		0x57 /* Max DIMM PER Channel Number */
71 
72 	/* OFFSETS for Function 1 */
73 #define		TOLM			0x6C
74 #define		REDMEMB			0x7C
75 #define			REC_ECC_LOCATOR_ODD(x)	((x) & 0x3fe00) /* bits [17:9] indicate ODD, [8:0]  indicate EVEN */
76 #define		MIR0			0x80
77 #define		MIR1			0x84
78 #define		AMIR0			0x8c
79 #define		AMIR1			0x90
80 
81 	/* Fatal error registers */
82 #define		FERR_FAT_FBD		0x98	/* also called as FERR_FAT_FB_DIMM at datasheet */
83 #define			FERR_FAT_FBDCHAN (3<<28)	/* channel index where the highest-order error occurred */
84 
85 #define		NERR_FAT_FBD		0x9c
86 #define		FERR_NF_FBD		0xa0	/* also called as FERR_NFAT_FB_DIMM at datasheet */
87 
88 	/* Non-fatal error register */
89 #define		NERR_NF_FBD		0xa4
90 
91 	/* Enable error mask */
92 #define		EMASK_FBD		0xa8
93 
94 #define		ERR0_FBD		0xac
95 #define		ERR1_FBD		0xb0
96 #define		ERR2_FBD		0xb4
97 #define		MCERR_FBD		0xb8
98 
99 	/* No OFFSETS for Device 16 Function 2 */
100 
101 /*
102  * Device 21,
103  * Function 0: Memory Map Branch 0
104  *
105  * Device 22,
106  * Function 0: Memory Map Branch 1
107  */
108 
109 	/* OFFSETS for Function 0 */
110 #define AMBPRESENT_0	0x64
111 #define AMBPRESENT_1	0x66
112 #define MTR0		0x80
113 #define MTR1		0x82
114 #define MTR2		0x84
115 #define MTR3		0x86
116 
117 	/* OFFSETS for Function 1 */
118 #define NRECFGLOG		0x74
119 #define RECFGLOG		0x78
120 #define NRECMEMA		0xbe
121 #define NRECMEMB		0xc0
122 #define NRECFB_DIMMA		0xc4
123 #define NRECFB_DIMMB		0xc8
124 #define NRECFB_DIMMC		0xcc
125 #define NRECFB_DIMMD		0xd0
126 #define NRECFB_DIMME		0xd4
127 #define NRECFB_DIMMF		0xd8
128 #define REDMEMA			0xdC
129 #define RECMEMA			0xf0
130 #define RECMEMB			0xf4
131 #define RECFB_DIMMA		0xf8
132 #define RECFB_DIMMB		0xec
133 #define RECFB_DIMMC		0xf0
134 #define RECFB_DIMMD		0xf4
135 #define RECFB_DIMME		0xf8
136 #define RECFB_DIMMF		0xfC
137 
138 /*
139  * Error indicator bits and masks
140  * Error masks are according with Table 5-17 of i5400 datasheet
141  */
142 
143 enum error_mask {
144 	EMASK_M1  = 1<<0,  /* Memory Write error on non-redundant retry */
145 	EMASK_M2  = 1<<1,  /* Memory or FB-DIMM configuration CRC read error */
146 	EMASK_M3  = 1<<2,  /* Reserved */
147 	EMASK_M4  = 1<<3,  /* Uncorrectable Data ECC on Replay */
148 	EMASK_M5  = 1<<4,  /* Aliased Uncorrectable Non-Mirrored Demand Data ECC */
149 	EMASK_M6  = 1<<5,  /* Unsupported on i5400 */
150 	EMASK_M7  = 1<<6,  /* Aliased Uncorrectable Resilver- or Spare-Copy Data ECC */
151 	EMASK_M8  = 1<<7,  /* Aliased Uncorrectable Patrol Data ECC */
152 	EMASK_M9  = 1<<8,  /* Non-Aliased Uncorrectable Non-Mirrored Demand Data ECC */
153 	EMASK_M10 = 1<<9,  /* Unsupported on i5400 */
154 	EMASK_M11 = 1<<10, /* Non-Aliased Uncorrectable Resilver- or Spare-Copy Data ECC  */
155 	EMASK_M12 = 1<<11, /* Non-Aliased Uncorrectable Patrol Data ECC */
156 	EMASK_M13 = 1<<12, /* Memory Write error on first attempt */
157 	EMASK_M14 = 1<<13, /* FB-DIMM Configuration Write error on first attempt */
158 	EMASK_M15 = 1<<14, /* Memory or FB-DIMM configuration CRC read error */
159 	EMASK_M16 = 1<<15, /* Channel Failed-Over Occurred */
160 	EMASK_M17 = 1<<16, /* Correctable Non-Mirrored Demand Data ECC */
161 	EMASK_M18 = 1<<17, /* Unsupported on i5400 */
162 	EMASK_M19 = 1<<18, /* Correctable Resilver- or Spare-Copy Data ECC */
163 	EMASK_M20 = 1<<19, /* Correctable Patrol Data ECC */
164 	EMASK_M21 = 1<<20, /* FB-DIMM Northbound parity error on FB-DIMM Sync Status */
165 	EMASK_M22 = 1<<21, /* SPD protocol Error */
166 	EMASK_M23 = 1<<22, /* Non-Redundant Fast Reset Timeout */
167 	EMASK_M24 = 1<<23, /* Refresh error */
168 	EMASK_M25 = 1<<24, /* Memory Write error on redundant retry */
169 	EMASK_M26 = 1<<25, /* Redundant Fast Reset Timeout */
170 	EMASK_M27 = 1<<26, /* Correctable Counter Threshold Exceeded */
171 	EMASK_M28 = 1<<27, /* DIMM-Spare Copy Completed */
172 	EMASK_M29 = 1<<28, /* DIMM-Isolation Completed */
173 };
174 
175 /*
176  * Names to translate bit error into something useful
177  */
178 static const char *error_name[] = {
179 	[0]  = "Memory Write error on non-redundant retry",
180 	[1]  = "Memory or FB-DIMM configuration CRC read error",
181 	/* Reserved */
182 	[3]  = "Uncorrectable Data ECC on Replay",
183 	[4]  = "Aliased Uncorrectable Non-Mirrored Demand Data ECC",
184 	/* M6 Unsupported on i5400 */
185 	[6]  = "Aliased Uncorrectable Resilver- or Spare-Copy Data ECC",
186 	[7]  = "Aliased Uncorrectable Patrol Data ECC",
187 	[8]  = "Non-Aliased Uncorrectable Non-Mirrored Demand Data ECC",
188 	/* M10 Unsupported on i5400 */
189 	[10] = "Non-Aliased Uncorrectable Resilver- or Spare-Copy Data ECC",
190 	[11] = "Non-Aliased Uncorrectable Patrol Data ECC",
191 	[12] = "Memory Write error on first attempt",
192 	[13] = "FB-DIMM Configuration Write error on first attempt",
193 	[14] = "Memory or FB-DIMM configuration CRC read error",
194 	[15] = "Channel Failed-Over Occurred",
195 	[16] = "Correctable Non-Mirrored Demand Data ECC",
196 	/* M18 Unsupported on i5400 */
197 	[18] = "Correctable Resilver- or Spare-Copy Data ECC",
198 	[19] = "Correctable Patrol Data ECC",
199 	[20] = "FB-DIMM Northbound parity error on FB-DIMM Sync Status",
200 	[21] = "SPD protocol Error",
201 	[22] = "Non-Redundant Fast Reset Timeout",
202 	[23] = "Refresh error",
203 	[24] = "Memory Write error on redundant retry",
204 	[25] = "Redundant Fast Reset Timeout",
205 	[26] = "Correctable Counter Threshold Exceeded",
206 	[27] = "DIMM-Spare Copy Completed",
207 	[28] = "DIMM-Isolation Completed",
208 };
209 
210 /* Fatal errors */
211 #define ERROR_FAT_MASK		(EMASK_M1 | \
212 				 EMASK_M2 | \
213 				 EMASK_M23)
214 
215 /* Correctable errors */
216 #define ERROR_NF_CORRECTABLE	(EMASK_M27 | \
217 				 EMASK_M20 | \
218 				 EMASK_M19 | \
219 				 EMASK_M18 | \
220 				 EMASK_M17 | \
221 				 EMASK_M16)
222 #define ERROR_NF_DIMM_SPARE	(EMASK_M29 | \
223 				 EMASK_M28)
224 #define ERROR_NF_SPD_PROTOCOL	(EMASK_M22)
225 #define ERROR_NF_NORTH_CRC	(EMASK_M21)
226 
227 /* Recoverable errors */
228 #define ERROR_NF_RECOVERABLE	(EMASK_M26 | \
229 				 EMASK_M25 | \
230 				 EMASK_M24 | \
231 				 EMASK_M15 | \
232 				 EMASK_M14 | \
233 				 EMASK_M13 | \
234 				 EMASK_M12 | \
235 				 EMASK_M11 | \
236 				 EMASK_M9  | \
237 				 EMASK_M8  | \
238 				 EMASK_M7  | \
239 				 EMASK_M5)
240 
241 /* uncorrectable errors */
242 #define ERROR_NF_UNCORRECTABLE	(EMASK_M4)
243 
244 /* mask to all non-fatal errors */
245 #define ERROR_NF_MASK		(ERROR_NF_CORRECTABLE   | \
246 				 ERROR_NF_UNCORRECTABLE | \
247 				 ERROR_NF_RECOVERABLE   | \
248 				 ERROR_NF_DIMM_SPARE    | \
249 				 ERROR_NF_SPD_PROTOCOL  | \
250 				 ERROR_NF_NORTH_CRC)
251 
252 /*
253  * Define error masks for the several registers
254  */
255 
256 /* Enable all fatal and non fatal errors */
257 #define ENABLE_EMASK_ALL	(ERROR_FAT_MASK | ERROR_NF_MASK)
258 
259 /* mask for fatal error registers */
260 #define FERR_FAT_MASK ERROR_FAT_MASK
261 
262 /* masks for non-fatal error register */
to_nf_mask(unsigned int mask)263 static inline int to_nf_mask(unsigned int mask)
264 {
265 	return (mask & EMASK_M29) | (mask >> 3);
266 };
267 
from_nf_ferr(unsigned int mask)268 static inline int from_nf_ferr(unsigned int mask)
269 {
270 	return (mask & EMASK_M29) |		/* Bit 28 */
271 	       (mask & ((1 << 28) - 1) << 3);	/* Bits 0 to 27 */
272 };
273 
274 #define FERR_NF_MASK		to_nf_mask(ERROR_NF_MASK)
275 #define FERR_NF_CORRECTABLE	to_nf_mask(ERROR_NF_CORRECTABLE)
276 #define FERR_NF_DIMM_SPARE	to_nf_mask(ERROR_NF_DIMM_SPARE)
277 #define FERR_NF_SPD_PROTOCOL	to_nf_mask(ERROR_NF_SPD_PROTOCOL)
278 #define FERR_NF_NORTH_CRC	to_nf_mask(ERROR_NF_NORTH_CRC)
279 #define FERR_NF_RECOVERABLE	to_nf_mask(ERROR_NF_RECOVERABLE)
280 #define FERR_NF_UNCORRECTABLE	to_nf_mask(ERROR_NF_UNCORRECTABLE)
281 
282 /*
283  * Defines to extract the various fields from the
284  *	MTRx - Memory Technology Registers
285  */
286 #define MTR_DIMMS_PRESENT(mtr)		((mtr) & (1 << 10))
287 #define MTR_DIMMS_ETHROTTLE(mtr)	((mtr) & (1 << 9))
288 #define MTR_DRAM_WIDTH(mtr)		(((mtr) & (1 << 8)) ? 8 : 4)
289 #define MTR_DRAM_BANKS(mtr)		(((mtr) & (1 << 6)) ? 8 : 4)
290 #define MTR_DRAM_BANKS_ADDR_BITS(mtr)	((MTR_DRAM_BANKS(mtr) == 8) ? 3 : 2)
291 #define MTR_DIMM_RANK(mtr)		(((mtr) >> 5) & 0x1)
292 #define MTR_DIMM_RANK_ADDR_BITS(mtr)	(MTR_DIMM_RANK(mtr) ? 2 : 1)
293 #define MTR_DIMM_ROWS(mtr)		(((mtr) >> 2) & 0x3)
294 #define MTR_DIMM_ROWS_ADDR_BITS(mtr)	(MTR_DIMM_ROWS(mtr) + 13)
295 #define MTR_DIMM_COLS(mtr)		((mtr) & 0x3)
296 #define MTR_DIMM_COLS_ADDR_BITS(mtr)	(MTR_DIMM_COLS(mtr) + 10)
297 
298 /* This applies to FERR_NF_FB-DIMM as well as FERR_FAT_FB-DIMM */
extract_fbdchan_indx(u32 x)299 static inline int extract_fbdchan_indx(u32 x)
300 {
301 	return (x>>28) & 0x3;
302 }
303 
304 /* Device name and register DID (Device ID) */
305 struct i5400_dev_info {
306 	const char *ctl_name;	/* name for this device */
307 	u16 fsb_mapping_errors;	/* DID for the branchmap,control */
308 };
309 
310 /* Table of devices attributes supported by this driver */
311 static const struct i5400_dev_info i5400_devs[] = {
312 	{
313 		.ctl_name = "I5400",
314 		.fsb_mapping_errors = PCI_DEVICE_ID_INTEL_5400_ERR,
315 	},
316 };
317 
318 struct i5400_dimm_info {
319 	int megabytes;		/* size, 0 means not present  */
320 };
321 
322 /* driver private data structure */
323 struct i5400_pvt {
324 	struct pci_dev *system_address;		/* 16.0 */
325 	struct pci_dev *branchmap_werrors;	/* 16.1 */
326 	struct pci_dev *fsb_error_regs;		/* 16.2 */
327 	struct pci_dev *branch_0;		/* 21.0 */
328 	struct pci_dev *branch_1;		/* 22.0 */
329 
330 	u16 tolm;				/* top of low memory */
331 	union {
332 		u64 ambase;				/* AMB BAR */
333 		struct {
334 			u32 ambase_bottom;
335 			u32 ambase_top;
336 		} u __packed;
337 	};
338 
339 	u16 mir0, mir1;
340 
341 	u16 b0_mtr[DIMMS_PER_CHANNEL];	/* Memory Technlogy Reg */
342 	u16 b0_ambpresent0;			/* Branch 0, Channel 0 */
343 	u16 b0_ambpresent1;			/* Brnach 0, Channel 1 */
344 
345 	u16 b1_mtr[DIMMS_PER_CHANNEL];	/* Memory Technlogy Reg */
346 	u16 b1_ambpresent0;			/* Branch 1, Channel 8 */
347 	u16 b1_ambpresent1;			/* Branch 1, Channel 1 */
348 
349 	/* DIMM information matrix, allocating architecture maximums */
350 	struct i5400_dimm_info dimm_info[DIMMS_PER_CHANNEL][MAX_CHANNELS];
351 
352 	/* Actual values for this controller */
353 	int maxch;				/* Max channels */
354 	int maxdimmperch;			/* Max DIMMs per channel */
355 };
356 
357 /* I5400 MCH error information retrieved from Hardware */
358 struct i5400_error_info {
359 	/* These registers are always read from the MC */
360 	u32 ferr_fat_fbd;	/* First Errors Fatal */
361 	u32 nerr_fat_fbd;	/* Next Errors Fatal */
362 	u32 ferr_nf_fbd;	/* First Errors Non-Fatal */
363 	u32 nerr_nf_fbd;	/* Next Errors Non-Fatal */
364 
365 	/* These registers are input ONLY if there was a Recoverable Error */
366 	u32 redmemb;		/* Recoverable Mem Data Error log B */
367 	u16 recmema;		/* Recoverable Mem Error log A */
368 	u32 recmemb;		/* Recoverable Mem Error log B */
369 
370 	/* These registers are input ONLY if there was a Non-Rec Error */
371 	u16 nrecmema;		/* Non-Recoverable Mem log A */
372 	u32 nrecmemb;		/* Non-Recoverable Mem log B */
373 
374 };
375 
376 /* note that nrec_rdwr changed from NRECMEMA to NRECMEMB between the 5000 and
377    5400 better to use an inline function than a macro in this case */
nrec_bank(struct i5400_error_info * info)378 static inline int nrec_bank(struct i5400_error_info *info)
379 {
380 	return ((info->nrecmema) >> 12) & 0x7;
381 }
nrec_rank(struct i5400_error_info * info)382 static inline int nrec_rank(struct i5400_error_info *info)
383 {
384 	return ((info->nrecmema) >> 8) & 0xf;
385 }
nrec_buf_id(struct i5400_error_info * info)386 static inline int nrec_buf_id(struct i5400_error_info *info)
387 {
388 	return ((info->nrecmema)) & 0xff;
389 }
nrec_rdwr(struct i5400_error_info * info)390 static inline int nrec_rdwr(struct i5400_error_info *info)
391 {
392 	return (info->nrecmemb) >> 31;
393 }
394 /* This applies to both NREC and REC string so it can be used with nrec_rdwr
395    and rec_rdwr */
rdwr_str(int rdwr)396 static inline const char *rdwr_str(int rdwr)
397 {
398 	return rdwr ? "Write" : "Read";
399 }
nrec_cas(struct i5400_error_info * info)400 static inline int nrec_cas(struct i5400_error_info *info)
401 {
402 	return ((info->nrecmemb) >> 16) & 0x1fff;
403 }
nrec_ras(struct i5400_error_info * info)404 static inline int nrec_ras(struct i5400_error_info *info)
405 {
406 	return (info->nrecmemb) & 0xffff;
407 }
rec_bank(struct i5400_error_info * info)408 static inline int rec_bank(struct i5400_error_info *info)
409 {
410 	return ((info->recmema) >> 12) & 0x7;
411 }
rec_rank(struct i5400_error_info * info)412 static inline int rec_rank(struct i5400_error_info *info)
413 {
414 	return ((info->recmema) >> 8) & 0xf;
415 }
rec_rdwr(struct i5400_error_info * info)416 static inline int rec_rdwr(struct i5400_error_info *info)
417 {
418 	return (info->recmemb) >> 31;
419 }
rec_cas(struct i5400_error_info * info)420 static inline int rec_cas(struct i5400_error_info *info)
421 {
422 	return ((info->recmemb) >> 16) & 0x1fff;
423 }
rec_ras(struct i5400_error_info * info)424 static inline int rec_ras(struct i5400_error_info *info)
425 {
426 	return (info->recmemb) & 0xffff;
427 }
428 
429 static struct edac_pci_ctl_info *i5400_pci;
430 
431 /*
432  *	i5400_get_error_info	Retrieve the hardware error information from
433  *				the hardware and cache it in the 'info'
434  *				structure
435  */
i5400_get_error_info(struct mem_ctl_info * mci,struct i5400_error_info * info)436 static void i5400_get_error_info(struct mem_ctl_info *mci,
437 				 struct i5400_error_info *info)
438 {
439 	struct i5400_pvt *pvt;
440 	u32 value;
441 
442 	pvt = mci->pvt_info;
443 
444 	/* read in the 1st FATAL error register */
445 	pci_read_config_dword(pvt->branchmap_werrors, FERR_FAT_FBD, &value);
446 
447 	/* Mask only the bits that the doc says are valid
448 	 */
449 	value &= (FERR_FAT_FBDCHAN | FERR_FAT_MASK);
450 
451 	/* If there is an error, then read in the
452 	   NEXT FATAL error register and the Memory Error Log Register A
453 	 */
454 	if (value & FERR_FAT_MASK) {
455 		info->ferr_fat_fbd = value;
456 
457 		/* harvest the various error data we need */
458 		pci_read_config_dword(pvt->branchmap_werrors,
459 				NERR_FAT_FBD, &info->nerr_fat_fbd);
460 		pci_read_config_word(pvt->branchmap_werrors,
461 				NRECMEMA, &info->nrecmema);
462 		pci_read_config_dword(pvt->branchmap_werrors,
463 				NRECMEMB, &info->nrecmemb);
464 
465 		/* Clear the error bits, by writing them back */
466 		pci_write_config_dword(pvt->branchmap_werrors,
467 				FERR_FAT_FBD, value);
468 	} else {
469 		info->ferr_fat_fbd = 0;
470 		info->nerr_fat_fbd = 0;
471 		info->nrecmema = 0;
472 		info->nrecmemb = 0;
473 	}
474 
475 	/* read in the 1st NON-FATAL error register */
476 	pci_read_config_dword(pvt->branchmap_werrors, FERR_NF_FBD, &value);
477 
478 	/* If there is an error, then read in the 1st NON-FATAL error
479 	 * register as well */
480 	if (value & FERR_NF_MASK) {
481 		info->ferr_nf_fbd = value;
482 
483 		/* harvest the various error data we need */
484 		pci_read_config_dword(pvt->branchmap_werrors,
485 				NERR_NF_FBD, &info->nerr_nf_fbd);
486 		pci_read_config_word(pvt->branchmap_werrors,
487 				RECMEMA, &info->recmema);
488 		pci_read_config_dword(pvt->branchmap_werrors,
489 				RECMEMB, &info->recmemb);
490 		pci_read_config_dword(pvt->branchmap_werrors,
491 				REDMEMB, &info->redmemb);
492 
493 		/* Clear the error bits, by writing them back */
494 		pci_write_config_dword(pvt->branchmap_werrors,
495 				FERR_NF_FBD, value);
496 	} else {
497 		info->ferr_nf_fbd = 0;
498 		info->nerr_nf_fbd = 0;
499 		info->recmema = 0;
500 		info->recmemb = 0;
501 		info->redmemb = 0;
502 	}
503 }
504 
505 /*
506  * i5400_proccess_non_recoverable_info(struct mem_ctl_info *mci,
507  * 					struct i5400_error_info *info,
508  * 					int handle_errors);
509  *
510  *	handle the Intel FATAL and unrecoverable errors, if any
511  */
i5400_proccess_non_recoverable_info(struct mem_ctl_info * mci,struct i5400_error_info * info,unsigned long allErrors)512 static void i5400_proccess_non_recoverable_info(struct mem_ctl_info *mci,
513 				    struct i5400_error_info *info,
514 				    unsigned long allErrors)
515 {
516 	char msg[EDAC_MC_LABEL_LEN + 1 + 90 + 80];
517 	int branch;
518 	int channel;
519 	int bank;
520 	int buf_id;
521 	int rank;
522 	int rdwr;
523 	int ras, cas;
524 	int errnum;
525 	char *type = NULL;
526 	enum hw_event_mc_err_type tp_event = HW_EVENT_ERR_UNCORRECTED;
527 
528 	if (!allErrors)
529 		return;		/* if no error, return now */
530 
531 	if (allErrors &  ERROR_FAT_MASK) {
532 		type = "FATAL";
533 		tp_event = HW_EVENT_ERR_FATAL;
534 	} else if (allErrors & FERR_NF_UNCORRECTABLE)
535 		type = "NON-FATAL uncorrected";
536 	else
537 		type = "NON-FATAL recoverable";
538 
539 	/* ONLY ONE of the possible error bits will be set, as per the docs */
540 
541 	branch = extract_fbdchan_indx(info->ferr_fat_fbd);
542 	channel = branch;
543 
544 	/* Use the NON-Recoverable macros to extract data */
545 	bank = nrec_bank(info);
546 	rank = nrec_rank(info);
547 	buf_id = nrec_buf_id(info);
548 	rdwr = nrec_rdwr(info);
549 	ras = nrec_ras(info);
550 	cas = nrec_cas(info);
551 
552 	edac_dbg(0, "\t\t%s DIMM= %d  Channels= %d,%d  (Branch= %d DRAM Bank= %d Buffer ID = %d rdwr= %s ras= %d cas= %d)\n",
553 		 type, rank, channel, channel + 1, branch >> 1, bank,
554 		 buf_id, rdwr_str(rdwr), ras, cas);
555 
556 	/* Only 1 bit will be on */
557 	errnum = find_first_bit(&allErrors, ARRAY_SIZE(error_name));
558 
559 	/* Form out message */
560 	snprintf(msg, sizeof(msg),
561 		 "Bank=%d Buffer ID = %d RAS=%d CAS=%d Err=0x%lx (%s)",
562 		 bank, buf_id, ras, cas, allErrors, error_name[errnum]);
563 
564 	edac_mc_handle_error(tp_event, mci, 1, 0, 0, 0,
565 			     branch >> 1, -1, rank,
566 			     rdwr ? "Write error" : "Read error",
567 			     msg);
568 }
569 
570 /*
571  * i5400_process_fatal_error_info(struct mem_ctl_info *mci,
572  * 				struct i5400_error_info *info,
573  * 				int handle_errors);
574  *
575  *	handle the Intel NON-FATAL errors, if any
576  */
i5400_process_nonfatal_error_info(struct mem_ctl_info * mci,struct i5400_error_info * info)577 static void i5400_process_nonfatal_error_info(struct mem_ctl_info *mci,
578 					struct i5400_error_info *info)
579 {
580 	char msg[EDAC_MC_LABEL_LEN + 1 + 90 + 80];
581 	unsigned long allErrors;
582 	int branch;
583 	int channel;
584 	int bank;
585 	int rank;
586 	int rdwr;
587 	int ras, cas;
588 	int errnum;
589 
590 	/* mask off the Error bits that are possible */
591 	allErrors = from_nf_ferr(info->ferr_nf_fbd & FERR_NF_MASK);
592 	if (!allErrors)
593 		return;		/* if no error, return now */
594 
595 	/* ONLY ONE of the possible error bits will be set, as per the docs */
596 
597 	if (allErrors & (ERROR_NF_UNCORRECTABLE | ERROR_NF_RECOVERABLE)) {
598 		i5400_proccess_non_recoverable_info(mci, info, allErrors);
599 		return;
600 	}
601 
602 	/* Correctable errors */
603 	if (allErrors & ERROR_NF_CORRECTABLE) {
604 		edac_dbg(0, "\tCorrected bits= 0x%lx\n", allErrors);
605 
606 		branch = extract_fbdchan_indx(info->ferr_nf_fbd);
607 
608 		channel = 0;
609 		if (REC_ECC_LOCATOR_ODD(info->redmemb))
610 			channel = 1;
611 
612 		/* Convert channel to be based from zero, instead of
613 		 * from branch base of 0 */
614 		channel += branch;
615 
616 		bank = rec_bank(info);
617 		rank = rec_rank(info);
618 		rdwr = rec_rdwr(info);
619 		ras = rec_ras(info);
620 		cas = rec_cas(info);
621 
622 		/* Only 1 bit will be on */
623 		errnum = find_first_bit(&allErrors, ARRAY_SIZE(error_name));
624 
625 		edac_dbg(0, "\t\tDIMM= %d Channel= %d  (Branch %d DRAM Bank= %d rdwr= %s ras= %d cas= %d)\n",
626 			 rank, channel, branch >> 1, bank,
627 			 rdwr_str(rdwr), ras, cas);
628 
629 		/* Form out message */
630 		snprintf(msg, sizeof(msg),
631 			 "Corrected error (Branch=%d DRAM-Bank=%d RDWR=%s "
632 			 "RAS=%d CAS=%d, CE Err=0x%lx (%s))",
633 			 branch >> 1, bank, rdwr_str(rdwr), ras, cas,
634 			 allErrors, error_name[errnum]);
635 
636 		edac_mc_handle_error(HW_EVENT_ERR_CORRECTED, mci, 1, 0, 0, 0,
637 				     branch >> 1, channel % 2, rank,
638 				     rdwr ? "Write error" : "Read error",
639 				     msg);
640 
641 		return;
642 	}
643 
644 	/* Miscellaneous errors */
645 	errnum = find_first_bit(&allErrors, ARRAY_SIZE(error_name));
646 
647 	branch = extract_fbdchan_indx(info->ferr_nf_fbd);
648 
649 	i5400_mc_printk(mci, KERN_EMERG,
650 			"Non-Fatal misc error (Branch=%d Err=%#lx (%s))",
651 			branch >> 1, allErrors, error_name[errnum]);
652 }
653 
654 /*
655  *	i5400_process_error_info	Process the error info that is
656  *	in the 'info' structure, previously retrieved from hardware
657  */
i5400_process_error_info(struct mem_ctl_info * mci,struct i5400_error_info * info)658 static void i5400_process_error_info(struct mem_ctl_info *mci,
659 				struct i5400_error_info *info)
660 {	u32 allErrors;
661 
662 	/* First handle any fatal errors that occurred */
663 	allErrors = (info->ferr_fat_fbd & FERR_FAT_MASK);
664 	i5400_proccess_non_recoverable_info(mci, info, allErrors);
665 
666 	/* now handle any non-fatal errors that occurred */
667 	i5400_process_nonfatal_error_info(mci, info);
668 }
669 
670 /*
671  *	i5400_clear_error	Retrieve any error from the hardware
672  *				but do NOT process that error.
673  *				Used for 'clearing' out of previous errors
674  *				Called by the Core module.
675  */
i5400_clear_error(struct mem_ctl_info * mci)676 static void i5400_clear_error(struct mem_ctl_info *mci)
677 {
678 	struct i5400_error_info info;
679 
680 	i5400_get_error_info(mci, &info);
681 }
682 
683 /*
684  *	i5400_check_error	Retrieve and process errors reported by the
685  *				hardware. Called by the Core module.
686  */
i5400_check_error(struct mem_ctl_info * mci)687 static void i5400_check_error(struct mem_ctl_info *mci)
688 {
689 	struct i5400_error_info info;
690 
691 	i5400_get_error_info(mci, &info);
692 	i5400_process_error_info(mci, &info);
693 }
694 
695 /*
696  *	i5400_put_devices	'put' all the devices that we have
697  *				reserved via 'get'
698  */
i5400_put_devices(struct mem_ctl_info * mci)699 static void i5400_put_devices(struct mem_ctl_info *mci)
700 {
701 	struct i5400_pvt *pvt;
702 
703 	pvt = mci->pvt_info;
704 
705 	/* Decrement usage count for devices */
706 	pci_dev_put(pvt->branch_1);
707 	pci_dev_put(pvt->branch_0);
708 	pci_dev_put(pvt->fsb_error_regs);
709 	pci_dev_put(pvt->branchmap_werrors);
710 }
711 
712 /*
713  *	i5400_get_devices	Find and perform 'get' operation on the MCH's
714  *			device/functions we want to reference for this driver
715  *
716  *			Need to 'get' device 16 func 1 and func 2
717  */
i5400_get_devices(struct mem_ctl_info * mci,int dev_idx)718 static int i5400_get_devices(struct mem_ctl_info *mci, int dev_idx)
719 {
720 	struct i5400_pvt *pvt;
721 	struct pci_dev *pdev;
722 
723 	pvt = mci->pvt_info;
724 	pvt->branchmap_werrors = NULL;
725 	pvt->fsb_error_regs = NULL;
726 	pvt->branch_0 = NULL;
727 	pvt->branch_1 = NULL;
728 
729 	/* Attempt to 'get' the MCH register we want */
730 	pdev = NULL;
731 	while (1) {
732 		pdev = pci_get_device(PCI_VENDOR_ID_INTEL,
733 				      PCI_DEVICE_ID_INTEL_5400_ERR, pdev);
734 		if (!pdev) {
735 			/* End of list, leave */
736 			i5400_printk(KERN_ERR,
737 				"'system address,Process Bus' "
738 				"device not found:"
739 				"vendor 0x%x device 0x%x ERR func 1 "
740 				"(broken BIOS?)\n",
741 				PCI_VENDOR_ID_INTEL,
742 				PCI_DEVICE_ID_INTEL_5400_ERR);
743 			return -ENODEV;
744 		}
745 
746 		/* Store device 16 func 1 */
747 		if (PCI_FUNC(pdev->devfn) == 1)
748 			break;
749 	}
750 	pvt->branchmap_werrors = pdev;
751 
752 	pdev = NULL;
753 	while (1) {
754 		pdev = pci_get_device(PCI_VENDOR_ID_INTEL,
755 				      PCI_DEVICE_ID_INTEL_5400_ERR, pdev);
756 		if (!pdev) {
757 			/* End of list, leave */
758 			i5400_printk(KERN_ERR,
759 				"'system address,Process Bus' "
760 				"device not found:"
761 				"vendor 0x%x device 0x%x ERR func 2 "
762 				"(broken BIOS?)\n",
763 				PCI_VENDOR_ID_INTEL,
764 				PCI_DEVICE_ID_INTEL_5400_ERR);
765 
766 			pci_dev_put(pvt->branchmap_werrors);
767 			return -ENODEV;
768 		}
769 
770 		/* Store device 16 func 2 */
771 		if (PCI_FUNC(pdev->devfn) == 2)
772 			break;
773 	}
774 	pvt->fsb_error_regs = pdev;
775 
776 	edac_dbg(1, "System Address, processor bus- PCI Bus ID: %s  %x:%x\n",
777 		 pci_name(pvt->system_address),
778 		 pvt->system_address->vendor, pvt->system_address->device);
779 	edac_dbg(1, "Branchmap, control and errors - PCI Bus ID: %s  %x:%x\n",
780 		 pci_name(pvt->branchmap_werrors),
781 		 pvt->branchmap_werrors->vendor,
782 		 pvt->branchmap_werrors->device);
783 	edac_dbg(1, "FSB Error Regs - PCI Bus ID: %s  %x:%x\n",
784 		 pci_name(pvt->fsb_error_regs),
785 		 pvt->fsb_error_regs->vendor, pvt->fsb_error_regs->device);
786 
787 	pvt->branch_0 = pci_get_device(PCI_VENDOR_ID_INTEL,
788 				       PCI_DEVICE_ID_INTEL_5400_FBD0, NULL);
789 	if (!pvt->branch_0) {
790 		i5400_printk(KERN_ERR,
791 			"MC: 'BRANCH 0' device not found:"
792 			"vendor 0x%x device 0x%x Func 0 (broken BIOS?)\n",
793 			PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_5400_FBD0);
794 
795 		pci_dev_put(pvt->fsb_error_regs);
796 		pci_dev_put(pvt->branchmap_werrors);
797 		return -ENODEV;
798 	}
799 
800 	/* If this device claims to have more than 2 channels then
801 	 * fetch Branch 1's information
802 	 */
803 	if (pvt->maxch < CHANNELS_PER_BRANCH)
804 		return 0;
805 
806 	pvt->branch_1 = pci_get_device(PCI_VENDOR_ID_INTEL,
807 				       PCI_DEVICE_ID_INTEL_5400_FBD1, NULL);
808 	if (!pvt->branch_1) {
809 		i5400_printk(KERN_ERR,
810 			"MC: 'BRANCH 1' device not found:"
811 			"vendor 0x%x device 0x%x Func 0 "
812 			"(broken BIOS?)\n",
813 			PCI_VENDOR_ID_INTEL,
814 			PCI_DEVICE_ID_INTEL_5400_FBD1);
815 
816 		pci_dev_put(pvt->branch_0);
817 		pci_dev_put(pvt->fsb_error_regs);
818 		pci_dev_put(pvt->branchmap_werrors);
819 		return -ENODEV;
820 	}
821 
822 	return 0;
823 }
824 
825 /*
826  *	determine_amb_present
827  *
828  *		the information is contained in DIMMS_PER_CHANNEL different
829  *		registers determining which of the DIMMS_PER_CHANNEL requires
830  *              knowing which channel is in question
831  *
832  *	2 branches, each with 2 channels
833  *		b0_ambpresent0 for channel '0'
834  *		b0_ambpresent1 for channel '1'
835  *		b1_ambpresent0 for channel '2'
836  *		b1_ambpresent1 for channel '3'
837  */
determine_amb_present_reg(struct i5400_pvt * pvt,int channel)838 static int determine_amb_present_reg(struct i5400_pvt *pvt, int channel)
839 {
840 	int amb_present;
841 
842 	if (channel < CHANNELS_PER_BRANCH) {
843 		if (channel & 0x1)
844 			amb_present = pvt->b0_ambpresent1;
845 		else
846 			amb_present = pvt->b0_ambpresent0;
847 	} else {
848 		if (channel & 0x1)
849 			amb_present = pvt->b1_ambpresent1;
850 		else
851 			amb_present = pvt->b1_ambpresent0;
852 	}
853 
854 	return amb_present;
855 }
856 
857 /*
858  * determine_mtr(pvt, dimm, channel)
859  *
860  * return the proper MTR register as determine by the dimm and desired channel
861  */
determine_mtr(struct i5400_pvt * pvt,int dimm,int channel)862 static int determine_mtr(struct i5400_pvt *pvt, int dimm, int channel)
863 {
864 	int mtr;
865 	int n;
866 
867 	/* There is one MTR for each slot pair of FB-DIMMs,
868 	   Each slot pair may be at branch 0 or branch 1.
869 	 */
870 	n = dimm;
871 
872 	if (n >= DIMMS_PER_CHANNEL) {
873 		edac_dbg(0, "ERROR: trying to access an invalid dimm: %d\n",
874 			 dimm);
875 		return 0;
876 	}
877 
878 	if (channel < CHANNELS_PER_BRANCH)
879 		mtr = pvt->b0_mtr[n];
880 	else
881 		mtr = pvt->b1_mtr[n];
882 
883 	return mtr;
884 }
885 
886 /*
887  */
decode_mtr(int slot_row,u16 mtr)888 static void decode_mtr(int slot_row, u16 mtr)
889 {
890 	int ans;
891 
892 	ans = MTR_DIMMS_PRESENT(mtr);
893 
894 	edac_dbg(2, "\tMTR%d=0x%x:  DIMMs are %sPresent\n",
895 		 slot_row, mtr, ans ? "" : "NOT ");
896 	if (!ans)
897 		return;
898 
899 	edac_dbg(2, "\t\tWIDTH: x%d\n", MTR_DRAM_WIDTH(mtr));
900 
901 	edac_dbg(2, "\t\tELECTRICAL THROTTLING is %s\n",
902 		 MTR_DIMMS_ETHROTTLE(mtr) ? "enabled" : "disabled");
903 
904 	edac_dbg(2, "\t\tNUMBANK: %d bank(s)\n", MTR_DRAM_BANKS(mtr));
905 	edac_dbg(2, "\t\tNUMRANK: %s\n",
906 		 MTR_DIMM_RANK(mtr) ? "double" : "single");
907 	edac_dbg(2, "\t\tNUMROW: %s\n",
908 		 MTR_DIMM_ROWS(mtr) == 0 ? "8,192 - 13 rows" :
909 		 MTR_DIMM_ROWS(mtr) == 1 ? "16,384 - 14 rows" :
910 		 MTR_DIMM_ROWS(mtr) == 2 ? "32,768 - 15 rows" :
911 		 "65,536 - 16 rows");
912 	edac_dbg(2, "\t\tNUMCOL: %s\n",
913 		 MTR_DIMM_COLS(mtr) == 0 ? "1,024 - 10 columns" :
914 		 MTR_DIMM_COLS(mtr) == 1 ? "2,048 - 11 columns" :
915 		 MTR_DIMM_COLS(mtr) == 2 ? "4,096 - 12 columns" :
916 		 "reserved");
917 }
918 
handle_channel(struct i5400_pvt * pvt,int dimm,int channel,struct i5400_dimm_info * dinfo)919 static void handle_channel(struct i5400_pvt *pvt, int dimm, int channel,
920 			struct i5400_dimm_info *dinfo)
921 {
922 	int mtr;
923 	int amb_present_reg;
924 	int addrBits;
925 
926 	mtr = determine_mtr(pvt, dimm, channel);
927 	if (MTR_DIMMS_PRESENT(mtr)) {
928 		amb_present_reg = determine_amb_present_reg(pvt, channel);
929 
930 		/* Determine if there is a DIMM present in this DIMM slot */
931 		if (amb_present_reg & (1 << dimm)) {
932 			/* Start with the number of bits for a Bank
933 			 * on the DRAM */
934 			addrBits = MTR_DRAM_BANKS_ADDR_BITS(mtr);
935 			/* Add thenumber of ROW bits */
936 			addrBits += MTR_DIMM_ROWS_ADDR_BITS(mtr);
937 			/* add the number of COLUMN bits */
938 			addrBits += MTR_DIMM_COLS_ADDR_BITS(mtr);
939 			/* add the number of RANK bits */
940 			addrBits += MTR_DIMM_RANK(mtr);
941 
942 			addrBits += 6;	/* add 64 bits per DIMM */
943 			addrBits -= 20;	/* divide by 2^^20 */
944 			addrBits -= 3;	/* 8 bits per bytes */
945 
946 			dinfo->megabytes = 1 << addrBits;
947 		}
948 	}
949 }
950 
951 /*
952  *	calculate_dimm_size
953  *
954  *	also will output a DIMM matrix map, if debug is enabled, for viewing
955  *	how the DIMMs are populated
956  */
calculate_dimm_size(struct i5400_pvt * pvt)957 static void calculate_dimm_size(struct i5400_pvt *pvt)
958 {
959 	struct i5400_dimm_info *dinfo;
960 	int dimm, max_dimms;
961 	char *p, *mem_buffer;
962 	int space, n;
963 	int channel, branch;
964 
965 	/* ================= Generate some debug output ================= */
966 	space = PAGE_SIZE;
967 	mem_buffer = p = kmalloc(space, GFP_KERNEL);
968 	if (p == NULL) {
969 		i5400_printk(KERN_ERR, "MC: %s:%s() kmalloc() failed\n",
970 			__FILE__, __func__);
971 		return;
972 	}
973 
974 	/* Scan all the actual DIMMS
975 	 * and calculate the information for each DIMM
976 	 * Start with the highest dimm first, to display it first
977 	 * and work toward the 0th dimm
978 	 */
979 	max_dimms = pvt->maxdimmperch;
980 	for (dimm = max_dimms - 1; dimm >= 0; dimm--) {
981 
982 		/* on an odd dimm, first output a 'boundary' marker,
983 		 * then reset the message buffer  */
984 		if (dimm & 0x1) {
985 			n = snprintf(p, space, "---------------------------"
986 					"-------------------------------");
987 			p += n;
988 			space -= n;
989 			edac_dbg(2, "%s\n", mem_buffer);
990 			p = mem_buffer;
991 			space = PAGE_SIZE;
992 		}
993 		n = snprintf(p, space, "dimm %2d    ", dimm);
994 		p += n;
995 		space -= n;
996 
997 		for (channel = 0; channel < pvt->maxch; channel++) {
998 			dinfo = &pvt->dimm_info[dimm][channel];
999 			handle_channel(pvt, dimm, channel, dinfo);
1000 			n = snprintf(p, space, "%4d MB   | ", dinfo->megabytes);
1001 			p += n;
1002 			space -= n;
1003 		}
1004 		edac_dbg(2, "%s\n", mem_buffer);
1005 		p = mem_buffer;
1006 		space = PAGE_SIZE;
1007 	}
1008 
1009 	/* Output the last bottom 'boundary' marker */
1010 	n = snprintf(p, space, "---------------------------"
1011 			"-------------------------------");
1012 	p += n;
1013 	space -= n;
1014 	edac_dbg(2, "%s\n", mem_buffer);
1015 	p = mem_buffer;
1016 	space = PAGE_SIZE;
1017 
1018 	/* now output the 'channel' labels */
1019 	n = snprintf(p, space, "           ");
1020 	p += n;
1021 	space -= n;
1022 	for (channel = 0; channel < pvt->maxch; channel++) {
1023 		n = snprintf(p, space, "channel %d | ", channel);
1024 		p += n;
1025 		space -= n;
1026 	}
1027 
1028 	space -= n;
1029 	edac_dbg(2, "%s\n", mem_buffer);
1030 	p = mem_buffer;
1031 	space = PAGE_SIZE;
1032 
1033 	n = snprintf(p, space, "           ");
1034 	p += n;
1035 	for (branch = 0; branch < MAX_BRANCHES; branch++) {
1036 		n = snprintf(p, space, "       branch %d       | ", branch);
1037 		p += n;
1038 		space -= n;
1039 	}
1040 
1041 	/* output the last message and free buffer */
1042 	edac_dbg(2, "%s\n", mem_buffer);
1043 	kfree(mem_buffer);
1044 }
1045 
1046 /*
1047  *	i5400_get_mc_regs	read in the necessary registers and
1048  *				cache locally
1049  *
1050  *			Fills in the private data members
1051  */
i5400_get_mc_regs(struct mem_ctl_info * mci)1052 static void i5400_get_mc_regs(struct mem_ctl_info *mci)
1053 {
1054 	struct i5400_pvt *pvt;
1055 	u32 actual_tolm;
1056 	u16 limit;
1057 	int slot_row;
1058 	int way0, way1;
1059 
1060 	pvt = mci->pvt_info;
1061 
1062 	pci_read_config_dword(pvt->system_address, AMBASE,
1063 			&pvt->u.ambase_bottom);
1064 	pci_read_config_dword(pvt->system_address, AMBASE + sizeof(u32),
1065 			&pvt->u.ambase_top);
1066 
1067 	edac_dbg(2, "AMBASE= 0x%lx  MAXCH= %d  MAX-DIMM-Per-CH= %d\n",
1068 		 (long unsigned int)pvt->ambase, pvt->maxch, pvt->maxdimmperch);
1069 
1070 	/* Get the Branch Map regs */
1071 	pci_read_config_word(pvt->branchmap_werrors, TOLM, &pvt->tolm);
1072 	pvt->tolm >>= 12;
1073 	edac_dbg(2, "\nTOLM (number of 256M regions) =%u (0x%x)\n",
1074 		 pvt->tolm, pvt->tolm);
1075 
1076 	actual_tolm = (u32) ((1000l * pvt->tolm) >> (30 - 28));
1077 	edac_dbg(2, "Actual TOLM byte addr=%u.%03u GB (0x%x)\n",
1078 		 actual_tolm/1000, actual_tolm % 1000, pvt->tolm << 28);
1079 
1080 	pci_read_config_word(pvt->branchmap_werrors, MIR0, &pvt->mir0);
1081 	pci_read_config_word(pvt->branchmap_werrors, MIR1, &pvt->mir1);
1082 
1083 	/* Get the MIR[0-1] regs */
1084 	limit = (pvt->mir0 >> 4) & 0x0fff;
1085 	way0 = pvt->mir0 & 0x1;
1086 	way1 = pvt->mir0 & 0x2;
1087 	edac_dbg(2, "MIR0: limit= 0x%x  WAY1= %u  WAY0= %x\n",
1088 		 limit, way1, way0);
1089 	limit = (pvt->mir1 >> 4) & 0xfff;
1090 	way0 = pvt->mir1 & 0x1;
1091 	way1 = pvt->mir1 & 0x2;
1092 	edac_dbg(2, "MIR1: limit= 0x%x  WAY1= %u  WAY0= %x\n",
1093 		 limit, way1, way0);
1094 
1095 	/* Get the set of MTR[0-3] regs by each branch */
1096 	for (slot_row = 0; slot_row < DIMMS_PER_CHANNEL; slot_row++) {
1097 		int where = MTR0 + (slot_row * sizeof(u16));
1098 
1099 		/* Branch 0 set of MTR registers */
1100 		pci_read_config_word(pvt->branch_0, where,
1101 				&pvt->b0_mtr[slot_row]);
1102 
1103 		edac_dbg(2, "MTR%d where=0x%x B0 value=0x%x\n",
1104 			 slot_row, where, pvt->b0_mtr[slot_row]);
1105 
1106 		if (pvt->maxch < CHANNELS_PER_BRANCH) {
1107 			pvt->b1_mtr[slot_row] = 0;
1108 			continue;
1109 		}
1110 
1111 		/* Branch 1 set of MTR registers */
1112 		pci_read_config_word(pvt->branch_1, where,
1113 				&pvt->b1_mtr[slot_row]);
1114 		edac_dbg(2, "MTR%d where=0x%x B1 value=0x%x\n",
1115 			 slot_row, where, pvt->b1_mtr[slot_row]);
1116 	}
1117 
1118 	/* Read and dump branch 0's MTRs */
1119 	edac_dbg(2, "Memory Technology Registers:\n");
1120 	edac_dbg(2, "   Branch 0:\n");
1121 	for (slot_row = 0; slot_row < DIMMS_PER_CHANNEL; slot_row++)
1122 		decode_mtr(slot_row, pvt->b0_mtr[slot_row]);
1123 
1124 	pci_read_config_word(pvt->branch_0, AMBPRESENT_0,
1125 			&pvt->b0_ambpresent0);
1126 	edac_dbg(2, "\t\tAMB-Branch 0-present0 0x%x:\n", pvt->b0_ambpresent0);
1127 	pci_read_config_word(pvt->branch_0, AMBPRESENT_1,
1128 			&pvt->b0_ambpresent1);
1129 	edac_dbg(2, "\t\tAMB-Branch 0-present1 0x%x:\n", pvt->b0_ambpresent1);
1130 
1131 	/* Only if we have 2 branchs (4 channels) */
1132 	if (pvt->maxch < CHANNELS_PER_BRANCH) {
1133 		pvt->b1_ambpresent0 = 0;
1134 		pvt->b1_ambpresent1 = 0;
1135 	} else {
1136 		/* Read and dump  branch 1's MTRs */
1137 		edac_dbg(2, "   Branch 1:\n");
1138 		for (slot_row = 0; slot_row < DIMMS_PER_CHANNEL; slot_row++)
1139 			decode_mtr(slot_row, pvt->b1_mtr[slot_row]);
1140 
1141 		pci_read_config_word(pvt->branch_1, AMBPRESENT_0,
1142 				&pvt->b1_ambpresent0);
1143 		edac_dbg(2, "\t\tAMB-Branch 1-present0 0x%x:\n",
1144 			 pvt->b1_ambpresent0);
1145 		pci_read_config_word(pvt->branch_1, AMBPRESENT_1,
1146 				&pvt->b1_ambpresent1);
1147 		edac_dbg(2, "\t\tAMB-Branch 1-present1 0x%x:\n",
1148 			 pvt->b1_ambpresent1);
1149 	}
1150 
1151 	/* Go and determine the size of each DIMM and place in an
1152 	 * orderly matrix */
1153 	calculate_dimm_size(pvt);
1154 }
1155 
1156 /*
1157  *	i5400_init_dimms	Initialize the 'dimms' table within
1158  *				the mci control	structure with the
1159  *				addressing of memory.
1160  *
1161  *	return:
1162  *		0	success
1163  *		1	no actual memory found on this MC
1164  */
i5400_init_dimms(struct mem_ctl_info * mci)1165 static int i5400_init_dimms(struct mem_ctl_info *mci)
1166 {
1167 	struct i5400_pvt *pvt;
1168 	struct dimm_info *dimm;
1169 	int ndimms;
1170 	int mtr;
1171 	int size_mb;
1172 	int  channel, slot;
1173 
1174 	pvt = mci->pvt_info;
1175 
1176 	ndimms = 0;
1177 
1178 	/*
1179 	 * FIXME: remove  pvt->dimm_info[slot][channel] and use the 3
1180 	 * layers here.
1181 	 */
1182 	for (channel = 0; channel < mci->layers[0].size * mci->layers[1].size;
1183 	     channel++) {
1184 		for (slot = 0; slot < mci->layers[2].size; slot++) {
1185 			mtr = determine_mtr(pvt, slot, channel);
1186 
1187 			/* if no DIMMS on this slot, continue */
1188 			if (!MTR_DIMMS_PRESENT(mtr))
1189 				continue;
1190 
1191 			dimm = edac_get_dimm(mci, channel / 2, channel % 2, slot);
1192 
1193 			size_mb =  pvt->dimm_info[slot][channel].megabytes;
1194 
1195 			edac_dbg(2, "dimm (branch %d channel %d slot %d): %d.%03d GB\n",
1196 				 channel / 2, channel % 2, slot,
1197 				 size_mb / 1000, size_mb % 1000);
1198 
1199 			dimm->nr_pages = size_mb << 8;
1200 			dimm->grain = 8;
1201 			dimm->dtype = MTR_DRAM_WIDTH(mtr) == 8 ?
1202 				      DEV_X8 : DEV_X4;
1203 			dimm->mtype = MEM_FB_DDR2;
1204 			/*
1205 			 * The eccc mechanism is SDDC (aka SECC), with
1206 			 * is similar to Chipkill.
1207 			 */
1208 			dimm->edac_mode = MTR_DRAM_WIDTH(mtr) == 8 ?
1209 					  EDAC_S8ECD8ED : EDAC_S4ECD4ED;
1210 			ndimms++;
1211 		}
1212 	}
1213 
1214 	/*
1215 	 * When just one memory is provided, it should be at location (0,0,0).
1216 	 * With such single-DIMM mode, the SDCC algorithm degrades to SECDEC+.
1217 	 */
1218 	if (ndimms == 1)
1219 		mci->dimms[0]->edac_mode = EDAC_SECDED;
1220 
1221 	return (ndimms == 0);
1222 }
1223 
1224 /*
1225  *	i5400_enable_error_reporting
1226  *			Turn on the memory reporting features of the hardware
1227  */
i5400_enable_error_reporting(struct mem_ctl_info * mci)1228 static void i5400_enable_error_reporting(struct mem_ctl_info *mci)
1229 {
1230 	struct i5400_pvt *pvt;
1231 	u32 fbd_error_mask;
1232 
1233 	pvt = mci->pvt_info;
1234 
1235 	/* Read the FBD Error Mask Register */
1236 	pci_read_config_dword(pvt->branchmap_werrors, EMASK_FBD,
1237 			&fbd_error_mask);
1238 
1239 	/* Enable with a '0' */
1240 	fbd_error_mask &= ~(ENABLE_EMASK_ALL);
1241 
1242 	pci_write_config_dword(pvt->branchmap_werrors, EMASK_FBD,
1243 			fbd_error_mask);
1244 }
1245 
1246 /*
1247  *	i5400_probe1	Probe for ONE instance of device to see if it is
1248  *			present.
1249  *	return:
1250  *		0 for FOUND a device
1251  *		< 0 for error code
1252  */
i5400_probe1(struct pci_dev * pdev,int dev_idx)1253 static int i5400_probe1(struct pci_dev *pdev, int dev_idx)
1254 {
1255 	struct mem_ctl_info *mci;
1256 	struct i5400_pvt *pvt;
1257 	struct edac_mc_layer layers[3];
1258 
1259 	if (dev_idx >= ARRAY_SIZE(i5400_devs))
1260 		return -EINVAL;
1261 
1262 	edac_dbg(0, "MC: pdev bus %u dev=0x%x fn=0x%x\n",
1263 		 pdev->bus->number,
1264 		 PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn));
1265 
1266 	/* We only are looking for func 0 of the set */
1267 	if (PCI_FUNC(pdev->devfn) != 0)
1268 		return -ENODEV;
1269 
1270 	/*
1271 	 * allocate a new MC control structure
1272 	 *
1273 	 * This drivers uses the DIMM slot as "csrow" and the rest as "channel".
1274 	 */
1275 	layers[0].type = EDAC_MC_LAYER_BRANCH;
1276 	layers[0].size = MAX_BRANCHES;
1277 	layers[0].is_virt_csrow = false;
1278 	layers[1].type = EDAC_MC_LAYER_CHANNEL;
1279 	layers[1].size = CHANNELS_PER_BRANCH;
1280 	layers[1].is_virt_csrow = false;
1281 	layers[2].type = EDAC_MC_LAYER_SLOT;
1282 	layers[2].size = DIMMS_PER_CHANNEL;
1283 	layers[2].is_virt_csrow = true;
1284 	mci = edac_mc_alloc(0, ARRAY_SIZE(layers), layers, sizeof(*pvt));
1285 	if (mci == NULL)
1286 		return -ENOMEM;
1287 
1288 	edac_dbg(0, "MC: mci = %p\n", mci);
1289 
1290 	mci->pdev = &pdev->dev;	/* record ptr  to the generic device */
1291 
1292 	pvt = mci->pvt_info;
1293 	pvt->system_address = pdev;	/* Record this device in our private */
1294 	pvt->maxch = MAX_CHANNELS;
1295 	pvt->maxdimmperch = DIMMS_PER_CHANNEL;
1296 
1297 	/* 'get' the pci devices we want to reserve for our use */
1298 	if (i5400_get_devices(mci, dev_idx))
1299 		goto fail0;
1300 
1301 	/* Time to get serious */
1302 	i5400_get_mc_regs(mci);	/* retrieve the hardware registers */
1303 
1304 	mci->mc_idx = 0;
1305 	mci->mtype_cap = MEM_FLAG_FB_DDR2;
1306 	mci->edac_ctl_cap = EDAC_FLAG_NONE;
1307 	mci->edac_cap = EDAC_FLAG_NONE;
1308 	mci->mod_name = "i5400_edac.c";
1309 	mci->ctl_name = i5400_devs[dev_idx].ctl_name;
1310 	mci->dev_name = pci_name(pdev);
1311 	mci->ctl_page_to_phys = NULL;
1312 
1313 	/* Set the function pointer to an actual operation function */
1314 	mci->edac_check = i5400_check_error;
1315 
1316 	/* initialize the MC control structure 'dimms' table
1317 	 * with the mapping and control information */
1318 	if (i5400_init_dimms(mci)) {
1319 		edac_dbg(0, "MC: Setting mci->edac_cap to EDAC_FLAG_NONE because i5400_init_dimms() returned nonzero value\n");
1320 		mci->edac_cap = EDAC_FLAG_NONE;	/* no dimms found */
1321 	} else {
1322 		edac_dbg(1, "MC: Enable error reporting now\n");
1323 		i5400_enable_error_reporting(mci);
1324 	}
1325 
1326 	/* add this new MC control structure to EDAC's list of MCs */
1327 	if (edac_mc_add_mc(mci)) {
1328 		edac_dbg(0, "MC: failed edac_mc_add_mc()\n");
1329 		/* FIXME: perhaps some code should go here that disables error
1330 		 * reporting if we just enabled it
1331 		 */
1332 		goto fail1;
1333 	}
1334 
1335 	i5400_clear_error(mci);
1336 
1337 	/* allocating generic PCI control info */
1338 	i5400_pci = edac_pci_create_generic_ctl(&pdev->dev, EDAC_MOD_STR);
1339 	if (!i5400_pci) {
1340 		printk(KERN_WARNING
1341 			"%s(): Unable to create PCI control\n",
1342 			__func__);
1343 		printk(KERN_WARNING
1344 			"%s(): PCI error report via EDAC not setup\n",
1345 			__func__);
1346 	}
1347 
1348 	return 0;
1349 
1350 	/* Error exit unwinding stack */
1351 fail1:
1352 
1353 	i5400_put_devices(mci);
1354 
1355 fail0:
1356 	edac_mc_free(mci);
1357 	return -ENODEV;
1358 }
1359 
1360 /*
1361  *	i5400_init_one	constructor for one instance of device
1362  *
1363  * 	returns:
1364  *		negative on error
1365  *		count (>= 0)
1366  */
i5400_init_one(struct pci_dev * pdev,const struct pci_device_id * id)1367 static int i5400_init_one(struct pci_dev *pdev, const struct pci_device_id *id)
1368 {
1369 	int rc;
1370 
1371 	edac_dbg(0, "MC:\n");
1372 
1373 	/* wake up device */
1374 	rc = pci_enable_device(pdev);
1375 	if (rc)
1376 		return rc;
1377 
1378 	/* now probe and enable the device */
1379 	return i5400_probe1(pdev, id->driver_data);
1380 }
1381 
1382 /*
1383  *	i5400_remove_one	destructor for one instance of device
1384  *
1385  */
i5400_remove_one(struct pci_dev * pdev)1386 static void i5400_remove_one(struct pci_dev *pdev)
1387 {
1388 	struct mem_ctl_info *mci;
1389 
1390 	edac_dbg(0, "\n");
1391 
1392 	if (i5400_pci)
1393 		edac_pci_release_generic_ctl(i5400_pci);
1394 
1395 	mci = edac_mc_del_mc(&pdev->dev);
1396 	if (!mci)
1397 		return;
1398 
1399 	/* retrieve references to resources, and free those resources */
1400 	i5400_put_devices(mci);
1401 
1402 	pci_disable_device(pdev);
1403 
1404 	edac_mc_free(mci);
1405 }
1406 
1407 /*
1408  *	pci_device_id	table for which devices we are looking for
1409  *
1410  *	The "E500P" device is the first device supported.
1411  */
1412 static const struct pci_device_id i5400_pci_tbl[] = {
1413 	{PCI_DEVICE(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_5400_ERR)},
1414 	{0,}			/* 0 terminated list. */
1415 };
1416 
1417 MODULE_DEVICE_TABLE(pci, i5400_pci_tbl);
1418 
1419 /*
1420  *	i5400_driver	pci_driver structure for this module
1421  *
1422  */
1423 static struct pci_driver i5400_driver = {
1424 	.name = "i5400_edac",
1425 	.probe = i5400_init_one,
1426 	.remove = i5400_remove_one,
1427 	.id_table = i5400_pci_tbl,
1428 };
1429 
1430 /*
1431  *	i5400_init		Module entry function
1432  *			Try to initialize this module for its devices
1433  */
i5400_init(void)1434 static int __init i5400_init(void)
1435 {
1436 	int pci_rc;
1437 
1438 	edac_dbg(2, "MC:\n");
1439 
1440 	/* Ensure that the OPSTATE is set correctly for POLL or NMI */
1441 	opstate_init();
1442 
1443 	pci_rc = pci_register_driver(&i5400_driver);
1444 
1445 	return (pci_rc < 0) ? pci_rc : 0;
1446 }
1447 
1448 /*
1449  *	i5400_exit()	Module exit function
1450  *			Unregister the driver
1451  */
i5400_exit(void)1452 static void __exit i5400_exit(void)
1453 {
1454 	edac_dbg(2, "MC:\n");
1455 	pci_unregister_driver(&i5400_driver);
1456 }
1457 
1458 module_init(i5400_init);
1459 module_exit(i5400_exit);
1460 
1461 MODULE_LICENSE("GPL");
1462 MODULE_AUTHOR("Ben Woodard <woodard@redhat.com>");
1463 MODULE_AUTHOR("Mauro Carvalho Chehab");
1464 MODULE_AUTHOR("Red Hat Inc. (https://www.redhat.com)");
1465 MODULE_DESCRIPTION("MC Driver for Intel I5400 memory controllers - "
1466 		   I5400_REVISION);
1467 
1468 module_param(edac_op_state, int, 0444);
1469 MODULE_PARM_DESC(edac_op_state, "EDAC Error Reporting state: 0=Poll,1=NMI");
1470