Commit | Line | Data |
---|---|---|
eebf11a0 MCC |
1 | /* Intel Sandy Bridge -EN/-EP/-EX Memory Controller kernel module |
2 | * | |
3 | * This driver supports the memory controllers found on the Intel | |
4 | * processor family Sandy Bridge. | |
5 | * | |
6 | * This file may be distributed under the terms of the | |
7 | * GNU General Public License version 2 only. | |
8 | * | |
9 | * Copyright (c) 2011 by: | |
10 | * Mauro Carvalho Chehab <mchehab@redhat.com> | |
11 | */ | |
12 | ||
13 | #include <linux/module.h> | |
14 | #include <linux/init.h> | |
15 | #include <linux/pci.h> | |
16 | #include <linux/pci_ids.h> | |
17 | #include <linux/slab.h> | |
18 | #include <linux/delay.h> | |
19 | #include <linux/edac.h> | |
20 | #include <linux/mmzone.h> | |
eebf11a0 MCC |
21 | #include <linux/smp.h> |
22 | #include <linux/bitmap.h> | |
5b889e37 | 23 | #include <linux/math64.h> |
eebf11a0 | 24 | #include <asm/processor.h> |
3d78c9af | 25 | #include <asm/mce.h> |
eebf11a0 MCC |
26 | |
27 | #include "edac_core.h" | |
28 | ||
29 | /* Static vars */ | |
30 | static LIST_HEAD(sbridge_edac_list); | |
31 | static DEFINE_MUTEX(sbridge_edac_lock); | |
32 | static int probed; | |
33 | ||
34 | /* | |
35 | * Alter this version for the module when modifications are made | |
36 | */ | |
37 | #define SBRIDGE_REVISION " Ver: 1.0.0 " | |
38 | #define EDAC_MOD_STR "sbridge_edac" | |
39 | ||
40 | /* | |
41 | * Debug macros | |
42 | */ | |
43 | #define sbridge_printk(level, fmt, arg...) \ | |
44 | edac_printk(level, "sbridge", fmt, ##arg) | |
45 | ||
46 | #define sbridge_mc_printk(mci, level, fmt, arg...) \ | |
47 | edac_mc_chipset_printk(mci, level, "sbridge", fmt, ##arg) | |
48 | ||
49 | /* | |
50 | * Get a bit field at register value <v>, from bit <lo> to bit <hi> | |
51 | */ | |
52 | #define GET_BITFIELD(v, lo, hi) \ | |
53 | (((v) & ((1ULL << ((hi) - (lo) + 1)) - 1) << (lo)) >> (lo)) | |
54 | ||
55 | /* | |
56 | * sbridge Memory Controller Registers | |
57 | */ | |
58 | ||
59 | /* | |
60 | * FIXME: For now, let's order by device function, as it makes | |
15ed103a | 61 | * easier for driver's development process. This table should be |
eebf11a0 MCC |
62 | * moved to pci_id.h when submitted upstream |
63 | */ | |
64 | #define PCI_DEVICE_ID_INTEL_SBRIDGE_SAD0 0x3cf4 /* 12.6 */ | |
65 | #define PCI_DEVICE_ID_INTEL_SBRIDGE_SAD1 0x3cf6 /* 12.7 */ | |
66 | #define PCI_DEVICE_ID_INTEL_SBRIDGE_BR 0x3cf5 /* 13.6 */ | |
67 | #define PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_HA0 0x3ca0 /* 14.0 */ | |
68 | #define PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TA 0x3ca8 /* 15.0 */ | |
69 | #define PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_RAS 0x3c71 /* 15.1 */ | |
70 | #define PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD0 0x3caa /* 15.2 */ | |
71 | #define PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD1 0x3cab /* 15.3 */ | |
72 | #define PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD2 0x3cac /* 15.4 */ | |
73 | #define PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD3 0x3cad /* 15.5 */ | |
74 | #define PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_DDRIO 0x3cb8 /* 17.0 */ | |
75 | ||
76 | /* | |
77 | * Currently, unused, but will be needed in the future | |
78 | * implementations, as they hold the error counters | |
79 | */ | |
80 | #define PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_ERR0 0x3c72 /* 16.2 */ | |
81 | #define PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_ERR1 0x3c73 /* 16.3 */ | |
82 | #define PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_ERR2 0x3c76 /* 16.6 */ | |
83 | #define PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_ERR3 0x3c77 /* 16.7 */ | |
84 | ||
85 | /* Devices 12 Function 6, Offsets 0x80 to 0xcc */ | |
86 | static const u32 dram_rule[] = { | |
87 | 0x80, 0x88, 0x90, 0x98, 0xa0, | |
88 | 0xa8, 0xb0, 0xb8, 0xc0, 0xc8, | |
89 | }; | |
90 | #define MAX_SAD ARRAY_SIZE(dram_rule) | |
91 | ||
92 | #define SAD_LIMIT(reg) ((GET_BITFIELD(reg, 6, 25) << 26) | 0x3ffffff) | |
93 | #define DRAM_ATTR(reg) GET_BITFIELD(reg, 2, 3) | |
94 | #define INTERLEAVE_MODE(reg) GET_BITFIELD(reg, 1, 1) | |
95 | #define DRAM_RULE_ENABLE(reg) GET_BITFIELD(reg, 0, 0) | |
96 | ||
97 | static char *get_dram_attr(u32 reg) | |
98 | { | |
99 | switch(DRAM_ATTR(reg)) { | |
100 | case 0: | |
101 | return "DRAM"; | |
102 | case 1: | |
103 | return "MMCFG"; | |
104 | case 2: | |
105 | return "NXM"; | |
106 | default: | |
107 | return "unknown"; | |
108 | } | |
109 | } | |
110 | ||
111 | static const u32 interleave_list[] = { | |
112 | 0x84, 0x8c, 0x94, 0x9c, 0xa4, | |
113 | 0xac, 0xb4, 0xbc, 0xc4, 0xcc, | |
114 | }; | |
115 | #define MAX_INTERLEAVE ARRAY_SIZE(interleave_list) | |
116 | ||
117 | #define SAD_PKG0(reg) GET_BITFIELD(reg, 0, 2) | |
118 | #define SAD_PKG1(reg) GET_BITFIELD(reg, 3, 5) | |
119 | #define SAD_PKG2(reg) GET_BITFIELD(reg, 8, 10) | |
120 | #define SAD_PKG3(reg) GET_BITFIELD(reg, 11, 13) | |
121 | #define SAD_PKG4(reg) GET_BITFIELD(reg, 16, 18) | |
122 | #define SAD_PKG5(reg) GET_BITFIELD(reg, 19, 21) | |
123 | #define SAD_PKG6(reg) GET_BITFIELD(reg, 24, 26) | |
124 | #define SAD_PKG7(reg) GET_BITFIELD(reg, 27, 29) | |
125 | ||
126 | static inline int sad_pkg(u32 reg, int interleave) | |
127 | { | |
128 | switch (interleave) { | |
129 | case 0: | |
130 | return SAD_PKG0(reg); | |
131 | case 1: | |
132 | return SAD_PKG1(reg); | |
133 | case 2: | |
134 | return SAD_PKG2(reg); | |
135 | case 3: | |
136 | return SAD_PKG3(reg); | |
137 | case 4: | |
138 | return SAD_PKG4(reg); | |
139 | case 5: | |
140 | return SAD_PKG5(reg); | |
141 | case 6: | |
142 | return SAD_PKG6(reg); | |
143 | case 7: | |
144 | return SAD_PKG7(reg); | |
145 | default: | |
146 | return -EINVAL; | |
147 | } | |
148 | } | |
149 | ||
150 | /* Devices 12 Function 7 */ | |
151 | ||
152 | #define TOLM 0x80 | |
153 | #define TOHM 0x84 | |
154 | ||
155 | #define GET_TOLM(reg) ((GET_BITFIELD(reg, 0, 3) << 28) | 0x3ffffff) | |
156 | #define GET_TOHM(reg) ((GET_BITFIELD(reg, 0, 20) << 25) | 0x3ffffff) | |
157 | ||
158 | /* Device 13 Function 6 */ | |
159 | ||
160 | #define SAD_TARGET 0xf0 | |
161 | ||
162 | #define SOURCE_ID(reg) GET_BITFIELD(reg, 9, 11) | |
163 | ||
164 | #define SAD_CONTROL 0xf4 | |
165 | ||
166 | #define NODE_ID(reg) GET_BITFIELD(reg, 0, 2) | |
167 | ||
168 | /* Device 14 function 0 */ | |
169 | ||
170 | static const u32 tad_dram_rule[] = { | |
171 | 0x40, 0x44, 0x48, 0x4c, | |
172 | 0x50, 0x54, 0x58, 0x5c, | |
173 | 0x60, 0x64, 0x68, 0x6c, | |
174 | }; | |
175 | #define MAX_TAD ARRAY_SIZE(tad_dram_rule) | |
176 | ||
177 | #define TAD_LIMIT(reg) ((GET_BITFIELD(reg, 12, 31) << 26) | 0x3ffffff) | |
178 | #define TAD_SOCK(reg) GET_BITFIELD(reg, 10, 11) | |
179 | #define TAD_CH(reg) GET_BITFIELD(reg, 8, 9) | |
180 | #define TAD_TGT3(reg) GET_BITFIELD(reg, 6, 7) | |
181 | #define TAD_TGT2(reg) GET_BITFIELD(reg, 4, 5) | |
182 | #define TAD_TGT1(reg) GET_BITFIELD(reg, 2, 3) | |
183 | #define TAD_TGT0(reg) GET_BITFIELD(reg, 0, 1) | |
184 | ||
185 | /* Device 15, function 0 */ | |
186 | ||
187 | #define MCMTR 0x7c | |
188 | ||
189 | #define IS_ECC_ENABLED(mcmtr) GET_BITFIELD(mcmtr, 2, 2) | |
190 | #define IS_LOCKSTEP_ENABLED(mcmtr) GET_BITFIELD(mcmtr, 1, 1) | |
191 | #define IS_CLOSE_PG(mcmtr) GET_BITFIELD(mcmtr, 0, 0) | |
192 | ||
193 | /* Device 15, function 1 */ | |
194 | ||
195 | #define RASENABLES 0xac | |
196 | #define IS_MIRROR_ENABLED(reg) GET_BITFIELD(reg, 0, 0) | |
197 | ||
198 | /* Device 15, functions 2-5 */ | |
199 | ||
200 | static const int mtr_regs[] = { | |
201 | 0x80, 0x84, 0x88, | |
202 | }; | |
203 | ||
204 | #define RANK_DISABLE(mtr) GET_BITFIELD(mtr, 16, 19) | |
205 | #define IS_DIMM_PRESENT(mtr) GET_BITFIELD(mtr, 14, 14) | |
206 | #define RANK_CNT_BITS(mtr) GET_BITFIELD(mtr, 12, 13) | |
207 | #define RANK_WIDTH_BITS(mtr) GET_BITFIELD(mtr, 2, 4) | |
208 | #define COL_WIDTH_BITS(mtr) GET_BITFIELD(mtr, 0, 1) | |
209 | ||
210 | static const u32 tad_ch_nilv_offset[] = { | |
211 | 0x90, 0x94, 0x98, 0x9c, | |
212 | 0xa0, 0xa4, 0xa8, 0xac, | |
213 | 0xb0, 0xb4, 0xb8, 0xbc, | |
214 | }; | |
215 | #define CHN_IDX_OFFSET(reg) GET_BITFIELD(reg, 28, 29) | |
216 | #define TAD_OFFSET(reg) (GET_BITFIELD(reg, 6, 25) << 26) | |
217 | ||
218 | static const u32 rir_way_limit[] = { | |
219 | 0x108, 0x10c, 0x110, 0x114, 0x118, | |
220 | }; | |
221 | #define MAX_RIR_RANGES ARRAY_SIZE(rir_way_limit) | |
222 | ||
223 | #define IS_RIR_VALID(reg) GET_BITFIELD(reg, 31, 31) | |
224 | #define RIR_WAY(reg) GET_BITFIELD(reg, 28, 29) | |
225 | #define RIR_LIMIT(reg) ((GET_BITFIELD(reg, 1, 10) << 29)| 0x1fffffff) | |
226 | ||
227 | #define MAX_RIR_WAY 8 | |
228 | ||
229 | static const u32 rir_offset[MAX_RIR_RANGES][MAX_RIR_WAY] = { | |
230 | { 0x120, 0x124, 0x128, 0x12c, 0x130, 0x134, 0x138, 0x13c }, | |
231 | { 0x140, 0x144, 0x148, 0x14c, 0x150, 0x154, 0x158, 0x15c }, | |
232 | { 0x160, 0x164, 0x168, 0x16c, 0x170, 0x174, 0x178, 0x17c }, | |
233 | { 0x180, 0x184, 0x188, 0x18c, 0x190, 0x194, 0x198, 0x19c }, | |
234 | { 0x1a0, 0x1a4, 0x1a8, 0x1ac, 0x1b0, 0x1b4, 0x1b8, 0x1bc }, | |
235 | }; | |
236 | ||
237 | #define RIR_RNK_TGT(reg) GET_BITFIELD(reg, 16, 19) | |
238 | #define RIR_OFFSET(reg) GET_BITFIELD(reg, 2, 14) | |
239 | ||
240 | /* Device 16, functions 2-7 */ | |
241 | ||
242 | /* | |
243 | * FIXME: Implement the error count reads directly | |
244 | */ | |
245 | ||
246 | static const u32 correrrcnt[] = { | |
247 | 0x104, 0x108, 0x10c, 0x110, | |
248 | }; | |
249 | ||
250 | #define RANK_ODD_OV(reg) GET_BITFIELD(reg, 31, 31) | |
251 | #define RANK_ODD_ERR_CNT(reg) GET_BITFIELD(reg, 16, 30) | |
252 | #define RANK_EVEN_OV(reg) GET_BITFIELD(reg, 15, 15) | |
253 | #define RANK_EVEN_ERR_CNT(reg) GET_BITFIELD(reg, 0, 14) | |
254 | ||
255 | static const u32 correrrthrsld[] = { | |
256 | 0x11c, 0x120, 0x124, 0x128, | |
257 | }; | |
258 | ||
259 | #define RANK_ODD_ERR_THRSLD(reg) GET_BITFIELD(reg, 16, 30) | |
260 | #define RANK_EVEN_ERR_THRSLD(reg) GET_BITFIELD(reg, 0, 14) | |
261 | ||
262 | ||
263 | /* Device 17, function 0 */ | |
264 | ||
265 | #define RANK_CFG_A 0x0328 | |
266 | ||
267 | #define IS_RDIMM_ENABLED(reg) GET_BITFIELD(reg, 11, 11) | |
268 | ||
269 | /* | |
270 | * sbridge structs | |
271 | */ | |
272 | ||
ebe7e30a SJ |
273 | #define NUM_CHANNELS 4 |
274 | #define MAX_DIMMS 3 /* Max DIMMS per channel */ | |
275 | #define CHANNEL_UNSPECIFIED 0xf /* Intel IA32 SDM 15-14 */ | |
eebf11a0 MCC |
276 | |
277 | struct sbridge_info { | |
278 | u32 mcmtr; | |
279 | }; | |
280 | ||
281 | struct sbridge_channel { | |
282 | u32 ranks; | |
283 | u32 dimms; | |
284 | }; | |
285 | ||
286 | struct pci_id_descr { | |
287 | int dev; | |
288 | int func; | |
289 | int dev_id; | |
290 | int optional; | |
291 | }; | |
292 | ||
293 | struct pci_id_table { | |
294 | const struct pci_id_descr *descr; | |
295 | int n_devs; | |
296 | }; | |
297 | ||
298 | struct sbridge_dev { | |
299 | struct list_head list; | |
300 | u8 bus, mc; | |
301 | u8 node_id, source_id; | |
302 | struct pci_dev **pdev; | |
303 | int n_devs; | |
304 | struct mem_ctl_info *mci; | |
305 | }; | |
306 | ||
307 | struct sbridge_pvt { | |
308 | struct pci_dev *pci_ta, *pci_ddrio, *pci_ras; | |
309 | struct pci_dev *pci_sad0, *pci_sad1, *pci_ha0; | |
310 | struct pci_dev *pci_br; | |
311 | struct pci_dev *pci_tad[NUM_CHANNELS]; | |
312 | ||
313 | struct sbridge_dev *sbridge_dev; | |
314 | ||
315 | struct sbridge_info info; | |
316 | struct sbridge_channel channel[NUM_CHANNELS]; | |
317 | ||
eebf11a0 MCC |
318 | /* Memory type detection */ |
319 | bool is_mirrored, is_lockstep, is_close_pg; | |
320 | ||
eebf11a0 MCC |
321 | /* Fifo double buffers */ |
322 | struct mce mce_entry[MCE_LOG_LEN]; | |
323 | struct mce mce_outentry[MCE_LOG_LEN]; | |
324 | ||
325 | /* Fifo in/out counters */ | |
326 | unsigned mce_in, mce_out; | |
327 | ||
328 | /* Count indicator to show errors not got */ | |
329 | unsigned mce_overrun; | |
330 | ||
331 | /* Memory description */ | |
332 | u64 tolm, tohm; | |
333 | }; | |
334 | ||
de4772c6 TL |
335 | #define PCI_DESCR(device, function, device_id, opt) \ |
336 | .dev = (device), \ | |
337 | .func = (function), \ | |
338 | .dev_id = (device_id), \ | |
339 | .optional = opt | |
eebf11a0 MCC |
340 | |
341 | static const struct pci_id_descr pci_dev_descr_sbridge[] = { | |
342 | /* Processor Home Agent */ | |
de4772c6 | 343 | { PCI_DESCR(14, 0, PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_HA0, 0) }, |
eebf11a0 MCC |
344 | |
345 | /* Memory controller */ | |
de4772c6 TL |
346 | { PCI_DESCR(15, 0, PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TA, 0) }, |
347 | { PCI_DESCR(15, 1, PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_RAS, 0) }, | |
348 | { PCI_DESCR(15, 2, PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD0, 0) }, | |
349 | { PCI_DESCR(15, 3, PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD1, 0) }, | |
350 | { PCI_DESCR(15, 4, PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD2, 0) }, | |
351 | { PCI_DESCR(15, 5, PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD3, 0) }, | |
352 | { PCI_DESCR(17, 0, PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_DDRIO, 1) }, | |
eebf11a0 MCC |
353 | |
354 | /* System Address Decoder */ | |
de4772c6 TL |
355 | { PCI_DESCR(12, 6, PCI_DEVICE_ID_INTEL_SBRIDGE_SAD0, 0) }, |
356 | { PCI_DESCR(12, 7, PCI_DEVICE_ID_INTEL_SBRIDGE_SAD1, 0) }, | |
eebf11a0 MCC |
357 | |
358 | /* Broadcast Registers */ | |
de4772c6 | 359 | { PCI_DESCR(13, 6, PCI_DEVICE_ID_INTEL_SBRIDGE_BR, 0) }, |
eebf11a0 MCC |
360 | }; |
361 | ||
362 | #define PCI_ID_TABLE_ENTRY(A) { .descr=A, .n_devs = ARRAY_SIZE(A) } | |
363 | static const struct pci_id_table pci_dev_descr_sbridge_table[] = { | |
364 | PCI_ID_TABLE_ENTRY(pci_dev_descr_sbridge), | |
365 | {0,} /* 0 terminated list. */ | |
366 | }; | |
367 | ||
368 | /* | |
369 | * pci_device_id table for which devices we are looking for | |
370 | */ | |
36c46f31 | 371 | static DEFINE_PCI_DEVICE_TABLE(sbridge_pci_tbl) = { |
eebf11a0 MCC |
372 | {PCI_DEVICE(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TA)}, |
373 | {0,} /* 0 terminated list. */ | |
374 | }; | |
375 | ||
376 | ||
377 | /**************************************************************************** | |
15ed103a | 378 | Ancillary status routines |
eebf11a0 MCC |
379 | ****************************************************************************/ |
380 | ||
381 | static inline int numrank(u32 mtr) | |
382 | { | |
383 | int ranks = (1 << RANK_CNT_BITS(mtr)); | |
384 | ||
385 | if (ranks > 4) { | |
956b9ba1 JP |
386 | edac_dbg(0, "Invalid number of ranks: %d (max = 4) raw value = %x (%04x)\n", |
387 | ranks, (unsigned int)RANK_CNT_BITS(mtr), mtr); | |
eebf11a0 MCC |
388 | return -EINVAL; |
389 | } | |
390 | ||
391 | return ranks; | |
392 | } | |
393 | ||
394 | static inline int numrow(u32 mtr) | |
395 | { | |
396 | int rows = (RANK_WIDTH_BITS(mtr) + 12); | |
397 | ||
398 | if (rows < 13 || rows > 18) { | |
956b9ba1 JP |
399 | edac_dbg(0, "Invalid number of rows: %d (should be between 14 and 17) raw value = %x (%04x)\n", |
400 | rows, (unsigned int)RANK_WIDTH_BITS(mtr), mtr); | |
eebf11a0 MCC |
401 | return -EINVAL; |
402 | } | |
403 | ||
404 | return 1 << rows; | |
405 | } | |
406 | ||
407 | static inline int numcol(u32 mtr) | |
408 | { | |
409 | int cols = (COL_WIDTH_BITS(mtr) + 10); | |
410 | ||
411 | if (cols > 12) { | |
956b9ba1 JP |
412 | edac_dbg(0, "Invalid number of cols: %d (max = 4) raw value = %x (%04x)\n", |
413 | cols, (unsigned int)COL_WIDTH_BITS(mtr), mtr); | |
eebf11a0 MCC |
414 | return -EINVAL; |
415 | } | |
416 | ||
417 | return 1 << cols; | |
418 | } | |
419 | ||
420 | static struct sbridge_dev *get_sbridge_dev(u8 bus) | |
421 | { | |
422 | struct sbridge_dev *sbridge_dev; | |
423 | ||
424 | list_for_each_entry(sbridge_dev, &sbridge_edac_list, list) { | |
425 | if (sbridge_dev->bus == bus) | |
426 | return sbridge_dev; | |
427 | } | |
428 | ||
429 | return NULL; | |
430 | } | |
431 | ||
432 | static struct sbridge_dev *alloc_sbridge_dev(u8 bus, | |
433 | const struct pci_id_table *table) | |
434 | { | |
435 | struct sbridge_dev *sbridge_dev; | |
436 | ||
437 | sbridge_dev = kzalloc(sizeof(*sbridge_dev), GFP_KERNEL); | |
438 | if (!sbridge_dev) | |
439 | return NULL; | |
440 | ||
441 | sbridge_dev->pdev = kzalloc(sizeof(*sbridge_dev->pdev) * table->n_devs, | |
442 | GFP_KERNEL); | |
443 | if (!sbridge_dev->pdev) { | |
444 | kfree(sbridge_dev); | |
445 | return NULL; | |
446 | } | |
447 | ||
448 | sbridge_dev->bus = bus; | |
449 | sbridge_dev->n_devs = table->n_devs; | |
450 | list_add_tail(&sbridge_dev->list, &sbridge_edac_list); | |
451 | ||
452 | return sbridge_dev; | |
453 | } | |
454 | ||
455 | static void free_sbridge_dev(struct sbridge_dev *sbridge_dev) | |
456 | { | |
457 | list_del(&sbridge_dev->list); | |
458 | kfree(sbridge_dev->pdev); | |
459 | kfree(sbridge_dev); | |
460 | } | |
461 | ||
462 | /**************************************************************************** | |
463 | Memory check routines | |
464 | ****************************************************************************/ | |
465 | static struct pci_dev *get_pdev_slot_func(u8 bus, unsigned slot, | |
466 | unsigned func) | |
467 | { | |
468 | struct sbridge_dev *sbridge_dev = get_sbridge_dev(bus); | |
469 | int i; | |
470 | ||
471 | if (!sbridge_dev) | |
472 | return NULL; | |
473 | ||
474 | for (i = 0; i < sbridge_dev->n_devs; i++) { | |
475 | if (!sbridge_dev->pdev[i]) | |
476 | continue; | |
477 | ||
478 | if (PCI_SLOT(sbridge_dev->pdev[i]->devfn) == slot && | |
479 | PCI_FUNC(sbridge_dev->pdev[i]->devfn) == func) { | |
956b9ba1 JP |
480 | edac_dbg(1, "Associated %02x.%02x.%d with %p\n", |
481 | bus, slot, func, sbridge_dev->pdev[i]); | |
eebf11a0 MCC |
482 | return sbridge_dev->pdev[i]; |
483 | } | |
484 | } | |
485 | ||
486 | return NULL; | |
487 | } | |
488 | ||
489 | /** | |
c36e3e77 | 490 | * check_if_ecc_is_active() - Checks if ECC is active |
eebf11a0 | 491 | * bus: Device bus |
eebf11a0 | 492 | */ |
c36e3e77 | 493 | static int check_if_ecc_is_active(const u8 bus) |
eebf11a0 MCC |
494 | { |
495 | struct pci_dev *pdev = NULL; | |
eebf11a0 MCC |
496 | u32 mcmtr; |
497 | ||
eebf11a0 MCC |
498 | pdev = get_pdev_slot_func(bus, 15, 0); |
499 | if (!pdev) { | |
500 | sbridge_printk(KERN_ERR, "Couldn't find PCI device " | |
501 | "%2x.%02d.%d!!!\n", | |
502 | bus, 15, 0); | |
503 | return -ENODEV; | |
504 | } | |
505 | ||
506 | pci_read_config_dword(pdev, MCMTR, &mcmtr); | |
507 | if (!IS_ECC_ENABLED(mcmtr)) { | |
508 | sbridge_printk(KERN_ERR, "ECC is disabled. Aborting\n"); | |
509 | return -ENODEV; | |
510 | } | |
eebf11a0 MCC |
511 | return 0; |
512 | } | |
513 | ||
084a4fcc | 514 | static int get_dimm_config(struct mem_ctl_info *mci) |
eebf11a0 MCC |
515 | { |
516 | struct sbridge_pvt *pvt = mci->pvt_info; | |
c36e3e77 | 517 | struct dimm_info *dimm; |
deb09dda MCC |
518 | unsigned i, j, banks, ranks, rows, cols, npages; |
519 | u64 size; | |
eebf11a0 MCC |
520 | u32 reg; |
521 | enum edac_type mode; | |
c6e13b52 | 522 | enum mem_type mtype; |
eebf11a0 MCC |
523 | |
524 | pci_read_config_dword(pvt->pci_br, SAD_TARGET, ®); | |
525 | pvt->sbridge_dev->source_id = SOURCE_ID(reg); | |
526 | ||
527 | pci_read_config_dword(pvt->pci_br, SAD_CONTROL, ®); | |
528 | pvt->sbridge_dev->node_id = NODE_ID(reg); | |
956b9ba1 JP |
529 | edac_dbg(0, "mc#%d: Node ID: %d, source ID: %d\n", |
530 | pvt->sbridge_dev->mc, | |
531 | pvt->sbridge_dev->node_id, | |
532 | pvt->sbridge_dev->source_id); | |
eebf11a0 MCC |
533 | |
534 | pci_read_config_dword(pvt->pci_ras, RASENABLES, ®); | |
535 | if (IS_MIRROR_ENABLED(reg)) { | |
956b9ba1 | 536 | edac_dbg(0, "Memory mirror is enabled\n"); |
eebf11a0 MCC |
537 | pvt->is_mirrored = true; |
538 | } else { | |
956b9ba1 | 539 | edac_dbg(0, "Memory mirror is disabled\n"); |
eebf11a0 MCC |
540 | pvt->is_mirrored = false; |
541 | } | |
542 | ||
543 | pci_read_config_dword(pvt->pci_ta, MCMTR, &pvt->info.mcmtr); | |
544 | if (IS_LOCKSTEP_ENABLED(pvt->info.mcmtr)) { | |
956b9ba1 | 545 | edac_dbg(0, "Lockstep is enabled\n"); |
eebf11a0 MCC |
546 | mode = EDAC_S8ECD8ED; |
547 | pvt->is_lockstep = true; | |
548 | } else { | |
956b9ba1 | 549 | edac_dbg(0, "Lockstep is disabled\n"); |
eebf11a0 MCC |
550 | mode = EDAC_S4ECD4ED; |
551 | pvt->is_lockstep = false; | |
552 | } | |
553 | if (IS_CLOSE_PG(pvt->info.mcmtr)) { | |
956b9ba1 | 554 | edac_dbg(0, "address map is on closed page mode\n"); |
eebf11a0 MCC |
555 | pvt->is_close_pg = true; |
556 | } else { | |
956b9ba1 | 557 | edac_dbg(0, "address map is on open page mode\n"); |
eebf11a0 MCC |
558 | pvt->is_close_pg = false; |
559 | } | |
560 | ||
de4772c6 TL |
561 | if (pvt->pci_ddrio) { |
562 | pci_read_config_dword(pvt->pci_ddrio, RANK_CFG_A, ®); | |
563 | if (IS_RDIMM_ENABLED(reg)) { | |
564 | /* FIXME: Can also be LRDIMM */ | |
565 | edac_dbg(0, "Memory is registered\n"); | |
566 | mtype = MEM_RDDR3; | |
567 | } else { | |
568 | edac_dbg(0, "Memory is unregistered\n"); | |
569 | mtype = MEM_DDR3; | |
570 | } | |
eebf11a0 | 571 | } else { |
de4772c6 TL |
572 | edac_dbg(0, "Cannot determine memory type\n"); |
573 | mtype = MEM_UNKNOWN; | |
eebf11a0 MCC |
574 | } |
575 | ||
576 | /* On all supported DDR3 DIMM types, there are 8 banks available */ | |
577 | banks = 8; | |
578 | ||
579 | for (i = 0; i < NUM_CHANNELS; i++) { | |
580 | u32 mtr; | |
581 | ||
582 | for (j = 0; j < ARRAY_SIZE(mtr_regs); j++) { | |
c36e3e77 MCC |
583 | dimm = EDAC_DIMM_PTR(mci->layers, mci->dimms, mci->n_layers, |
584 | i, j, 0); | |
eebf11a0 MCC |
585 | pci_read_config_dword(pvt->pci_tad[i], |
586 | mtr_regs[j], &mtr); | |
956b9ba1 | 587 | edac_dbg(4, "Channel #%d MTR%d = %x\n", i, j, mtr); |
eebf11a0 MCC |
588 | if (IS_DIMM_PRESENT(mtr)) { |
589 | pvt->channel[i].dimms++; | |
590 | ||
591 | ranks = numrank(mtr); | |
592 | rows = numrow(mtr); | |
593 | cols = numcol(mtr); | |
594 | ||
595 | /* DDR3 has 8 I/O banks */ | |
deb09dda | 596 | size = ((u64)rows * cols * banks * ranks) >> (20 - 3); |
eebf11a0 MCC |
597 | npages = MiB_TO_PAGES(size); |
598 | ||
deb09dda | 599 | edac_dbg(0, "mc#%d: channel %d, dimm %d, %Ld Mb (%d pages) bank: %d, rank: %d, row: %#x, col: %#x\n", |
956b9ba1 JP |
600 | pvt->sbridge_dev->mc, i, j, |
601 | size, npages, | |
602 | banks, ranks, rows, cols); | |
eebf11a0 | 603 | |
a895bf8b | 604 | dimm->nr_pages = npages; |
084a4fcc MCC |
605 | dimm->grain = 32; |
606 | dimm->dtype = (banks == 8) ? DEV_X8 : DEV_X4; | |
607 | dimm->mtype = mtype; | |
608 | dimm->edac_mode = mode; | |
609 | snprintf(dimm->label, sizeof(dimm->label), | |
eebf11a0 MCC |
610 | "CPU_SrcID#%u_Channel#%u_DIMM#%u", |
611 | pvt->sbridge_dev->source_id, i, j); | |
eebf11a0 MCC |
612 | } |
613 | } | |
614 | } | |
615 | ||
616 | return 0; | |
617 | } | |
618 | ||
619 | static void get_memory_layout(const struct mem_ctl_info *mci) | |
620 | { | |
621 | struct sbridge_pvt *pvt = mci->pvt_info; | |
622 | int i, j, k, n_sads, n_tads, sad_interl; | |
623 | u32 reg; | |
624 | u64 limit, prv = 0; | |
625 | u64 tmp_mb; | |
5b889e37 | 626 | u32 mb, kb; |
eebf11a0 MCC |
627 | u32 rir_way; |
628 | ||
629 | /* | |
630 | * Step 1) Get TOLM/TOHM ranges | |
631 | */ | |
632 | ||
633 | /* Address range is 32:28 */ | |
634 | pci_read_config_dword(pvt->pci_sad1, TOLM, | |
635 | ®); | |
636 | pvt->tolm = GET_TOLM(reg); | |
637 | tmp_mb = (1 + pvt->tolm) >> 20; | |
638 | ||
5b889e37 | 639 | mb = div_u64_rem(tmp_mb, 1000, &kb); |
956b9ba1 | 640 | edac_dbg(0, "TOLM: %u.%03u GB (0x%016Lx)\n", mb, kb, (u64)pvt->tolm); |
eebf11a0 MCC |
641 | |
642 | /* Address range is already 45:25 */ | |
643 | pci_read_config_dword(pvt->pci_sad1, TOHM, | |
644 | ®); | |
645 | pvt->tohm = GET_TOHM(reg); | |
646 | tmp_mb = (1 + pvt->tohm) >> 20; | |
647 | ||
5b889e37 | 648 | mb = div_u64_rem(tmp_mb, 1000, &kb); |
da14d93d | 649 | edac_dbg(0, "TOHM: %u.%03u GB (0x%016Lx)\n", mb, kb, (u64)pvt->tohm); |
eebf11a0 MCC |
650 | |
651 | /* | |
652 | * Step 2) Get SAD range and SAD Interleave list | |
653 | * TAD registers contain the interleave wayness. However, it | |
654 | * seems simpler to just discover it indirectly, with the | |
655 | * algorithm bellow. | |
656 | */ | |
657 | prv = 0; | |
658 | for (n_sads = 0; n_sads < MAX_SAD; n_sads++) { | |
659 | /* SAD_LIMIT Address range is 45:26 */ | |
660 | pci_read_config_dword(pvt->pci_sad0, dram_rule[n_sads], | |
661 | ®); | |
662 | limit = SAD_LIMIT(reg); | |
663 | ||
664 | if (!DRAM_RULE_ENABLE(reg)) | |
665 | continue; | |
666 | ||
667 | if (limit <= prv) | |
668 | break; | |
669 | ||
670 | tmp_mb = (limit + 1) >> 20; | |
5b889e37 | 671 | mb = div_u64_rem(tmp_mb, 1000, &kb); |
956b9ba1 JP |
672 | edac_dbg(0, "SAD#%d %s up to %u.%03u GB (0x%016Lx) Interleave: %s reg=0x%08x\n", |
673 | n_sads, | |
674 | get_dram_attr(reg), | |
675 | mb, kb, | |
676 | ((u64)tmp_mb) << 20L, | |
677 | INTERLEAVE_MODE(reg) ? "8:6" : "[8:6]XOR[18:16]", | |
678 | reg); | |
eebf11a0 MCC |
679 | prv = limit; |
680 | ||
681 | pci_read_config_dword(pvt->pci_sad0, interleave_list[n_sads], | |
682 | ®); | |
683 | sad_interl = sad_pkg(reg, 0); | |
684 | for (j = 0; j < 8; j++) { | |
685 | if (j > 0 && sad_interl == sad_pkg(reg, j)) | |
686 | break; | |
687 | ||
956b9ba1 JP |
688 | edac_dbg(0, "SAD#%d, interleave #%d: %d\n", |
689 | n_sads, j, sad_pkg(reg, j)); | |
eebf11a0 MCC |
690 | } |
691 | } | |
692 | ||
693 | /* | |
694 | * Step 3) Get TAD range | |
695 | */ | |
696 | prv = 0; | |
697 | for (n_tads = 0; n_tads < MAX_TAD; n_tads++) { | |
698 | pci_read_config_dword(pvt->pci_ha0, tad_dram_rule[n_tads], | |
699 | ®); | |
700 | limit = TAD_LIMIT(reg); | |
701 | if (limit <= prv) | |
702 | break; | |
703 | tmp_mb = (limit + 1) >> 20; | |
704 | ||
5b889e37 | 705 | mb = div_u64_rem(tmp_mb, 1000, &kb); |
956b9ba1 JP |
706 | edac_dbg(0, "TAD#%d: up to %u.%03u GB (0x%016Lx), socket interleave %d, memory interleave %d, TGT: %d, %d, %d, %d, reg=0x%08x\n", |
707 | n_tads, mb, kb, | |
708 | ((u64)tmp_mb) << 20L, | |
709 | (u32)TAD_SOCK(reg), | |
710 | (u32)TAD_CH(reg), | |
711 | (u32)TAD_TGT0(reg), | |
712 | (u32)TAD_TGT1(reg), | |
713 | (u32)TAD_TGT2(reg), | |
714 | (u32)TAD_TGT3(reg), | |
715 | reg); | |
7fae0db4 | 716 | prv = limit; |
eebf11a0 MCC |
717 | } |
718 | ||
719 | /* | |
720 | * Step 4) Get TAD offsets, per each channel | |
721 | */ | |
722 | for (i = 0; i < NUM_CHANNELS; i++) { | |
723 | if (!pvt->channel[i].dimms) | |
724 | continue; | |
725 | for (j = 0; j < n_tads; j++) { | |
726 | pci_read_config_dword(pvt->pci_tad[i], | |
727 | tad_ch_nilv_offset[j], | |
728 | ®); | |
729 | tmp_mb = TAD_OFFSET(reg) >> 20; | |
5b889e37 | 730 | mb = div_u64_rem(tmp_mb, 1000, &kb); |
956b9ba1 JP |
731 | edac_dbg(0, "TAD CH#%d, offset #%d: %u.%03u GB (0x%016Lx), reg=0x%08x\n", |
732 | i, j, | |
733 | mb, kb, | |
734 | ((u64)tmp_mb) << 20L, | |
735 | reg); | |
eebf11a0 MCC |
736 | } |
737 | } | |
738 | ||
739 | /* | |
740 | * Step 6) Get RIR Wayness/Limit, per each channel | |
741 | */ | |
742 | for (i = 0; i < NUM_CHANNELS; i++) { | |
743 | if (!pvt->channel[i].dimms) | |
744 | continue; | |
745 | for (j = 0; j < MAX_RIR_RANGES; j++) { | |
746 | pci_read_config_dword(pvt->pci_tad[i], | |
747 | rir_way_limit[j], | |
748 | ®); | |
749 | ||
750 | if (!IS_RIR_VALID(reg)) | |
751 | continue; | |
752 | ||
753 | tmp_mb = RIR_LIMIT(reg) >> 20; | |
754 | rir_way = 1 << RIR_WAY(reg); | |
5b889e37 | 755 | mb = div_u64_rem(tmp_mb, 1000, &kb); |
956b9ba1 JP |
756 | edac_dbg(0, "CH#%d RIR#%d, limit: %u.%03u GB (0x%016Lx), way: %d, reg=0x%08x\n", |
757 | i, j, | |
758 | mb, kb, | |
759 | ((u64)tmp_mb) << 20L, | |
760 | rir_way, | |
761 | reg); | |
eebf11a0 MCC |
762 | |
763 | for (k = 0; k < rir_way; k++) { | |
764 | pci_read_config_dword(pvt->pci_tad[i], | |
765 | rir_offset[j][k], | |
766 | ®); | |
767 | tmp_mb = RIR_OFFSET(reg) << 6; | |
768 | ||
5b889e37 | 769 | mb = div_u64_rem(tmp_mb, 1000, &kb); |
956b9ba1 JP |
770 | edac_dbg(0, "CH#%d RIR#%d INTL#%d, offset %u.%03u GB (0x%016Lx), tgt: %d, reg=0x%08x\n", |
771 | i, j, k, | |
772 | mb, kb, | |
773 | ((u64)tmp_mb) << 20L, | |
774 | (u32)RIR_RNK_TGT(reg), | |
775 | reg); | |
eebf11a0 MCC |
776 | } |
777 | } | |
778 | } | |
779 | } | |
780 | ||
781 | struct mem_ctl_info *get_mci_for_node_id(u8 node_id) | |
782 | { | |
783 | struct sbridge_dev *sbridge_dev; | |
784 | ||
785 | list_for_each_entry(sbridge_dev, &sbridge_edac_list, list) { | |
786 | if (sbridge_dev->node_id == node_id) | |
787 | return sbridge_dev->mci; | |
788 | } | |
789 | return NULL; | |
790 | } | |
791 | ||
792 | static int get_memory_error_data(struct mem_ctl_info *mci, | |
793 | u64 addr, | |
794 | u8 *socket, | |
795 | long *channel_mask, | |
796 | u8 *rank, | |
e17a2f42 | 797 | char **area_type, char *msg) |
eebf11a0 MCC |
798 | { |
799 | struct mem_ctl_info *new_mci; | |
800 | struct sbridge_pvt *pvt = mci->pvt_info; | |
eebf11a0 MCC |
801 | int n_rir, n_sads, n_tads, sad_way, sck_xch; |
802 | int sad_interl, idx, base_ch; | |
803 | int interleave_mode; | |
804 | unsigned sad_interleave[MAX_INTERLEAVE]; | |
805 | u32 reg; | |
806 | u8 ch_way,sck_way; | |
807 | u32 tad_offset; | |
808 | u32 rir_way; | |
5b889e37 | 809 | u32 mb, kb; |
eebf11a0 MCC |
810 | u64 ch_addr, offset, limit, prv = 0; |
811 | ||
812 | ||
813 | /* | |
814 | * Step 0) Check if the address is at special memory ranges | |
815 | * The check bellow is probably enough to fill all cases where | |
816 | * the error is not inside a memory, except for the legacy | |
817 | * range (e. g. VGA addresses). It is unlikely, however, that the | |
818 | * memory controller would generate an error on that range. | |
819 | */ | |
5b889e37 | 820 | if ((addr > (u64) pvt->tolm) && (addr < (1LL << 32))) { |
eebf11a0 | 821 | sprintf(msg, "Error at TOLM area, on addr 0x%08Lx", addr); |
eebf11a0 MCC |
822 | return -EINVAL; |
823 | } | |
824 | if (addr >= (u64)pvt->tohm) { | |
825 | sprintf(msg, "Error at MMIOH area, on addr 0x%016Lx", addr); | |
eebf11a0 MCC |
826 | return -EINVAL; |
827 | } | |
828 | ||
829 | /* | |
830 | * Step 1) Get socket | |
831 | */ | |
832 | for (n_sads = 0; n_sads < MAX_SAD; n_sads++) { | |
833 | pci_read_config_dword(pvt->pci_sad0, dram_rule[n_sads], | |
834 | ®); | |
835 | ||
836 | if (!DRAM_RULE_ENABLE(reg)) | |
837 | continue; | |
838 | ||
839 | limit = SAD_LIMIT(reg); | |
840 | if (limit <= prv) { | |
841 | sprintf(msg, "Can't discover the memory socket"); | |
eebf11a0 MCC |
842 | return -EINVAL; |
843 | } | |
844 | if (addr <= limit) | |
845 | break; | |
846 | prv = limit; | |
847 | } | |
848 | if (n_sads == MAX_SAD) { | |
849 | sprintf(msg, "Can't discover the memory socket"); | |
eebf11a0 MCC |
850 | return -EINVAL; |
851 | } | |
e17a2f42 | 852 | *area_type = get_dram_attr(reg); |
eebf11a0 MCC |
853 | interleave_mode = INTERLEAVE_MODE(reg); |
854 | ||
855 | pci_read_config_dword(pvt->pci_sad0, interleave_list[n_sads], | |
856 | ®); | |
857 | sad_interl = sad_pkg(reg, 0); | |
858 | for (sad_way = 0; sad_way < 8; sad_way++) { | |
859 | if (sad_way > 0 && sad_interl == sad_pkg(reg, sad_way)) | |
860 | break; | |
861 | sad_interleave[sad_way] = sad_pkg(reg, sad_way); | |
956b9ba1 JP |
862 | edac_dbg(0, "SAD interleave #%d: %d\n", |
863 | sad_way, sad_interleave[sad_way]); | |
eebf11a0 | 864 | } |
956b9ba1 JP |
865 | edac_dbg(0, "mc#%d: Error detected on SAD#%d: address 0x%016Lx < 0x%016Lx, Interleave [%d:6]%s\n", |
866 | pvt->sbridge_dev->mc, | |
867 | n_sads, | |
868 | addr, | |
869 | limit, | |
870 | sad_way + 7, | |
871 | interleave_mode ? "" : "XOR[18:16]"); | |
eebf11a0 MCC |
872 | if (interleave_mode) |
873 | idx = ((addr >> 6) ^ (addr >> 16)) & 7; | |
874 | else | |
875 | idx = (addr >> 6) & 7; | |
876 | switch (sad_way) { | |
877 | case 1: | |
878 | idx = 0; | |
879 | break; | |
880 | case 2: | |
881 | idx = idx & 1; | |
882 | break; | |
883 | case 4: | |
884 | idx = idx & 3; | |
885 | break; | |
886 | case 8: | |
887 | break; | |
888 | default: | |
889 | sprintf(msg, "Can't discover socket interleave"); | |
eebf11a0 MCC |
890 | return -EINVAL; |
891 | } | |
892 | *socket = sad_interleave[idx]; | |
956b9ba1 JP |
893 | edac_dbg(0, "SAD interleave index: %d (wayness %d) = CPU socket %d\n", |
894 | idx, sad_way, *socket); | |
eebf11a0 MCC |
895 | |
896 | /* | |
897 | * Move to the proper node structure, in order to access the | |
898 | * right PCI registers | |
899 | */ | |
900 | new_mci = get_mci_for_node_id(*socket); | |
901 | if (!new_mci) { | |
902 | sprintf(msg, "Struct for socket #%u wasn't initialized", | |
903 | *socket); | |
eebf11a0 MCC |
904 | return -EINVAL; |
905 | } | |
906 | mci = new_mci; | |
907 | pvt = mci->pvt_info; | |
908 | ||
909 | /* | |
910 | * Step 2) Get memory channel | |
911 | */ | |
912 | prv = 0; | |
913 | for (n_tads = 0; n_tads < MAX_TAD; n_tads++) { | |
914 | pci_read_config_dword(pvt->pci_ha0, tad_dram_rule[n_tads], | |
915 | ®); | |
916 | limit = TAD_LIMIT(reg); | |
917 | if (limit <= prv) { | |
918 | sprintf(msg, "Can't discover the memory channel"); | |
eebf11a0 MCC |
919 | return -EINVAL; |
920 | } | |
921 | if (addr <= limit) | |
922 | break; | |
923 | prv = limit; | |
924 | } | |
925 | ch_way = TAD_CH(reg) + 1; | |
926 | sck_way = TAD_SOCK(reg) + 1; | |
927 | /* | |
928 | * FIXME: Is it right to always use channel 0 for offsets? | |
929 | */ | |
930 | pci_read_config_dword(pvt->pci_tad[0], | |
931 | tad_ch_nilv_offset[n_tads], | |
932 | &tad_offset); | |
933 | ||
934 | if (ch_way == 3) | |
935 | idx = addr >> 6; | |
936 | else | |
937 | idx = addr >> (6 + sck_way); | |
938 | idx = idx % ch_way; | |
939 | ||
940 | /* | |
941 | * FIXME: Shouldn't we use CHN_IDX_OFFSET() here, when ch_way == 3 ??? | |
942 | */ | |
943 | switch (idx) { | |
944 | case 0: | |
945 | base_ch = TAD_TGT0(reg); | |
946 | break; | |
947 | case 1: | |
948 | base_ch = TAD_TGT1(reg); | |
949 | break; | |
950 | case 2: | |
951 | base_ch = TAD_TGT2(reg); | |
952 | break; | |
953 | case 3: | |
954 | base_ch = TAD_TGT3(reg); | |
955 | break; | |
956 | default: | |
957 | sprintf(msg, "Can't discover the TAD target"); | |
eebf11a0 MCC |
958 | return -EINVAL; |
959 | } | |
960 | *channel_mask = 1 << base_ch; | |
961 | ||
962 | if (pvt->is_mirrored) { | |
963 | *channel_mask |= 1 << ((base_ch + 2) % 4); | |
964 | switch(ch_way) { | |
965 | case 2: | |
966 | case 4: | |
967 | sck_xch = 1 << sck_way * (ch_way >> 1); | |
968 | break; | |
969 | default: | |
970 | sprintf(msg, "Invalid mirror set. Can't decode addr"); | |
eebf11a0 MCC |
971 | return -EINVAL; |
972 | } | |
973 | } else | |
974 | sck_xch = (1 << sck_way) * ch_way; | |
975 | ||
976 | if (pvt->is_lockstep) | |
977 | *channel_mask |= 1 << ((base_ch + 1) % 4); | |
978 | ||
979 | offset = TAD_OFFSET(tad_offset); | |
980 | ||
956b9ba1 JP |
981 | edac_dbg(0, "TAD#%d: address 0x%016Lx < 0x%016Lx, socket interleave %d, channel interleave %d (offset 0x%08Lx), index %d, base ch: %d, ch mask: 0x%02lx\n", |
982 | n_tads, | |
983 | addr, | |
984 | limit, | |
985 | (u32)TAD_SOCK(reg), | |
986 | ch_way, | |
987 | offset, | |
988 | idx, | |
989 | base_ch, | |
990 | *channel_mask); | |
eebf11a0 MCC |
991 | |
992 | /* Calculate channel address */ | |
993 | /* Remove the TAD offset */ | |
994 | ||
995 | if (offset > addr) { | |
996 | sprintf(msg, "Can't calculate ch addr: TAD offset 0x%08Lx is too high for addr 0x%08Lx!", | |
997 | offset, addr); | |
eebf11a0 MCC |
998 | return -EINVAL; |
999 | } | |
1000 | addr -= offset; | |
1001 | /* Store the low bits [0:6] of the addr */ | |
1002 | ch_addr = addr & 0x7f; | |
1003 | /* Remove socket wayness and remove 6 bits */ | |
1004 | addr >>= 6; | |
5b889e37 | 1005 | addr = div_u64(addr, sck_xch); |
eebf11a0 MCC |
1006 | #if 0 |
1007 | /* Divide by channel way */ | |
1008 | addr = addr / ch_way; | |
1009 | #endif | |
1010 | /* Recover the last 6 bits */ | |
1011 | ch_addr |= addr << 6; | |
1012 | ||
1013 | /* | |
1014 | * Step 3) Decode rank | |
1015 | */ | |
1016 | for (n_rir = 0; n_rir < MAX_RIR_RANGES; n_rir++) { | |
1017 | pci_read_config_dword(pvt->pci_tad[base_ch], | |
1018 | rir_way_limit[n_rir], | |
1019 | ®); | |
1020 | ||
1021 | if (!IS_RIR_VALID(reg)) | |
1022 | continue; | |
1023 | ||
1024 | limit = RIR_LIMIT(reg); | |
5b889e37 | 1025 | mb = div_u64_rem(limit >> 20, 1000, &kb); |
956b9ba1 JP |
1026 | edac_dbg(0, "RIR#%d, limit: %u.%03u GB (0x%016Lx), way: %d\n", |
1027 | n_rir, | |
1028 | mb, kb, | |
1029 | limit, | |
1030 | 1 << RIR_WAY(reg)); | |
eebf11a0 MCC |
1031 | if (ch_addr <= limit) |
1032 | break; | |
1033 | } | |
1034 | if (n_rir == MAX_RIR_RANGES) { | |
1035 | sprintf(msg, "Can't discover the memory rank for ch addr 0x%08Lx", | |
1036 | ch_addr); | |
eebf11a0 MCC |
1037 | return -EINVAL; |
1038 | } | |
1039 | rir_way = RIR_WAY(reg); | |
1040 | if (pvt->is_close_pg) | |
1041 | idx = (ch_addr >> 6); | |
1042 | else | |
1043 | idx = (ch_addr >> 13); /* FIXME: Datasheet says to shift by 15 */ | |
1044 | idx %= 1 << rir_way; | |
1045 | ||
1046 | pci_read_config_dword(pvt->pci_tad[base_ch], | |
1047 | rir_offset[n_rir][idx], | |
1048 | ®); | |
1049 | *rank = RIR_RNK_TGT(reg); | |
1050 | ||
956b9ba1 JP |
1051 | edac_dbg(0, "RIR#%d: channel address 0x%08Lx < 0x%08Lx, RIR interleave %d, index %d\n", |
1052 | n_rir, | |
1053 | ch_addr, | |
1054 | limit, | |
1055 | rir_way, | |
1056 | idx); | |
eebf11a0 MCC |
1057 | |
1058 | return 0; | |
1059 | } | |
1060 | ||
1061 | /**************************************************************************** | |
1062 | Device initialization routines: put/get, init/exit | |
1063 | ****************************************************************************/ | |
1064 | ||
1065 | /* | |
1066 | * sbridge_put_all_devices 'put' all the devices that we have | |
1067 | * reserved via 'get' | |
1068 | */ | |
1069 | static void sbridge_put_devices(struct sbridge_dev *sbridge_dev) | |
1070 | { | |
1071 | int i; | |
1072 | ||
956b9ba1 | 1073 | edac_dbg(0, "\n"); |
eebf11a0 MCC |
1074 | for (i = 0; i < sbridge_dev->n_devs; i++) { |
1075 | struct pci_dev *pdev = sbridge_dev->pdev[i]; | |
1076 | if (!pdev) | |
1077 | continue; | |
956b9ba1 JP |
1078 | edac_dbg(0, "Removing dev %02x:%02x.%d\n", |
1079 | pdev->bus->number, | |
1080 | PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn)); | |
eebf11a0 MCC |
1081 | pci_dev_put(pdev); |
1082 | } | |
1083 | } | |
1084 | ||
1085 | static void sbridge_put_all_devices(void) | |
1086 | { | |
1087 | struct sbridge_dev *sbridge_dev, *tmp; | |
1088 | ||
1089 | list_for_each_entry_safe(sbridge_dev, tmp, &sbridge_edac_list, list) { | |
1090 | sbridge_put_devices(sbridge_dev); | |
1091 | free_sbridge_dev(sbridge_dev); | |
1092 | } | |
1093 | } | |
1094 | ||
1095 | /* | |
1096 | * sbridge_get_all_devices Find and perform 'get' operation on the MCH's | |
1097 | * device/functions we want to reference for this driver | |
1098 | * | |
1099 | * Need to 'get' device 16 func 1 and func 2 | |
1100 | */ | |
1101 | static int sbridge_get_onedevice(struct pci_dev **prev, | |
1102 | u8 *num_mc, | |
1103 | const struct pci_id_table *table, | |
1104 | const unsigned devno) | |
1105 | { | |
1106 | struct sbridge_dev *sbridge_dev; | |
1107 | const struct pci_id_descr *dev_descr = &table->descr[devno]; | |
1108 | ||
1109 | struct pci_dev *pdev = NULL; | |
1110 | u8 bus = 0; | |
1111 | ||
1112 | sbridge_printk(KERN_INFO, | |
1113 | "Seeking for: dev %02x.%d PCI ID %04x:%04x\n", | |
1114 | dev_descr->dev, dev_descr->func, | |
1115 | PCI_VENDOR_ID_INTEL, dev_descr->dev_id); | |
1116 | ||
1117 | pdev = pci_get_device(PCI_VENDOR_ID_INTEL, | |
1118 | dev_descr->dev_id, *prev); | |
1119 | ||
1120 | if (!pdev) { | |
1121 | if (*prev) { | |
1122 | *prev = pdev; | |
1123 | return 0; | |
1124 | } | |
1125 | ||
1126 | if (dev_descr->optional) | |
1127 | return 0; | |
1128 | ||
1129 | if (devno == 0) | |
1130 | return -ENODEV; | |
1131 | ||
1132 | sbridge_printk(KERN_INFO, | |
1133 | "Device not found: dev %02x.%d PCI ID %04x:%04x\n", | |
1134 | dev_descr->dev, dev_descr->func, | |
1135 | PCI_VENDOR_ID_INTEL, dev_descr->dev_id); | |
1136 | ||
1137 | /* End of list, leave */ | |
1138 | return -ENODEV; | |
1139 | } | |
1140 | bus = pdev->bus->number; | |
1141 | ||
1142 | sbridge_dev = get_sbridge_dev(bus); | |
1143 | if (!sbridge_dev) { | |
1144 | sbridge_dev = alloc_sbridge_dev(bus, table); | |
1145 | if (!sbridge_dev) { | |
1146 | pci_dev_put(pdev); | |
1147 | return -ENOMEM; | |
1148 | } | |
1149 | (*num_mc)++; | |
1150 | } | |
1151 | ||
1152 | if (sbridge_dev->pdev[devno]) { | |
1153 | sbridge_printk(KERN_ERR, | |
1154 | "Duplicated device for " | |
1155 | "dev %02x:%d.%d PCI ID %04x:%04x\n", | |
1156 | bus, dev_descr->dev, dev_descr->func, | |
1157 | PCI_VENDOR_ID_INTEL, dev_descr->dev_id); | |
1158 | pci_dev_put(pdev); | |
1159 | return -ENODEV; | |
1160 | } | |
1161 | ||
1162 | sbridge_dev->pdev[devno] = pdev; | |
1163 | ||
1164 | /* Sanity check */ | |
1165 | if (unlikely(PCI_SLOT(pdev->devfn) != dev_descr->dev || | |
1166 | PCI_FUNC(pdev->devfn) != dev_descr->func)) { | |
1167 | sbridge_printk(KERN_ERR, | |
1168 | "Device PCI ID %04x:%04x " | |
1169 | "has dev %02x:%d.%d instead of dev %02x:%02x.%d\n", | |
1170 | PCI_VENDOR_ID_INTEL, dev_descr->dev_id, | |
1171 | bus, PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn), | |
1172 | bus, dev_descr->dev, dev_descr->func); | |
1173 | return -ENODEV; | |
1174 | } | |
1175 | ||
1176 | /* Be sure that the device is enabled */ | |
1177 | if (unlikely(pci_enable_device(pdev) < 0)) { | |
1178 | sbridge_printk(KERN_ERR, | |
1179 | "Couldn't enable " | |
1180 | "dev %02x:%d.%d PCI ID %04x:%04x\n", | |
1181 | bus, dev_descr->dev, dev_descr->func, | |
1182 | PCI_VENDOR_ID_INTEL, dev_descr->dev_id); | |
1183 | return -ENODEV; | |
1184 | } | |
1185 | ||
956b9ba1 JP |
1186 | edac_dbg(0, "Detected dev %02x:%d.%d PCI ID %04x:%04x\n", |
1187 | bus, dev_descr->dev, dev_descr->func, | |
1188 | PCI_VENDOR_ID_INTEL, dev_descr->dev_id); | |
eebf11a0 MCC |
1189 | |
1190 | /* | |
1191 | * As stated on drivers/pci/search.c, the reference count for | |
1192 | * @from is always decremented if it is not %NULL. So, as we need | |
1193 | * to get all devices up to null, we need to do a get for the device | |
1194 | */ | |
1195 | pci_dev_get(pdev); | |
1196 | ||
1197 | *prev = pdev; | |
1198 | ||
1199 | return 0; | |
1200 | } | |
1201 | ||
1202 | static int sbridge_get_all_devices(u8 *num_mc) | |
1203 | { | |
1204 | int i, rc; | |
1205 | struct pci_dev *pdev = NULL; | |
1206 | const struct pci_id_table *table = pci_dev_descr_sbridge_table; | |
1207 | ||
1208 | while (table && table->descr) { | |
1209 | for (i = 0; i < table->n_devs; i++) { | |
1210 | pdev = NULL; | |
1211 | do { | |
1212 | rc = sbridge_get_onedevice(&pdev, num_mc, | |
1213 | table, i); | |
1214 | if (rc < 0) { | |
1215 | if (i == 0) { | |
1216 | i = table->n_devs; | |
1217 | break; | |
1218 | } | |
1219 | sbridge_put_all_devices(); | |
1220 | return -ENODEV; | |
1221 | } | |
1222 | } while (pdev); | |
1223 | } | |
1224 | table++; | |
1225 | } | |
1226 | ||
1227 | return 0; | |
1228 | } | |
1229 | ||
1230 | static int mci_bind_devs(struct mem_ctl_info *mci, | |
1231 | struct sbridge_dev *sbridge_dev) | |
1232 | { | |
1233 | struct sbridge_pvt *pvt = mci->pvt_info; | |
1234 | struct pci_dev *pdev; | |
1235 | int i, func, slot; | |
1236 | ||
1237 | for (i = 0; i < sbridge_dev->n_devs; i++) { | |
1238 | pdev = sbridge_dev->pdev[i]; | |
1239 | if (!pdev) | |
1240 | continue; | |
1241 | slot = PCI_SLOT(pdev->devfn); | |
1242 | func = PCI_FUNC(pdev->devfn); | |
1243 | switch (slot) { | |
1244 | case 12: | |
1245 | switch (func) { | |
1246 | case 6: | |
1247 | pvt->pci_sad0 = pdev; | |
1248 | break; | |
1249 | case 7: | |
1250 | pvt->pci_sad1 = pdev; | |
1251 | break; | |
1252 | default: | |
1253 | goto error; | |
1254 | } | |
1255 | break; | |
1256 | case 13: | |
1257 | switch (func) { | |
1258 | case 6: | |
1259 | pvt->pci_br = pdev; | |
1260 | break; | |
1261 | default: | |
1262 | goto error; | |
1263 | } | |
1264 | break; | |
1265 | case 14: | |
1266 | switch (func) { | |
1267 | case 0: | |
1268 | pvt->pci_ha0 = pdev; | |
1269 | break; | |
1270 | default: | |
1271 | goto error; | |
1272 | } | |
1273 | break; | |
1274 | case 15: | |
1275 | switch (func) { | |
1276 | case 0: | |
1277 | pvt->pci_ta = pdev; | |
1278 | break; | |
1279 | case 1: | |
1280 | pvt->pci_ras = pdev; | |
1281 | break; | |
1282 | case 2: | |
1283 | case 3: | |
1284 | case 4: | |
1285 | case 5: | |
1286 | pvt->pci_tad[func - 2] = pdev; | |
1287 | break; | |
1288 | default: | |
1289 | goto error; | |
1290 | } | |
1291 | break; | |
1292 | case 17: | |
1293 | switch (func) { | |
1294 | case 0: | |
1295 | pvt->pci_ddrio = pdev; | |
1296 | break; | |
1297 | default: | |
1298 | goto error; | |
1299 | } | |
1300 | break; | |
1301 | default: | |
1302 | goto error; | |
1303 | } | |
1304 | ||
956b9ba1 JP |
1305 | edac_dbg(0, "Associated PCI %02x.%02d.%d with dev = %p\n", |
1306 | sbridge_dev->bus, | |
1307 | PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn), | |
1308 | pdev); | |
eebf11a0 MCC |
1309 | } |
1310 | ||
1311 | /* Check if everything were registered */ | |
1312 | if (!pvt->pci_sad0 || !pvt->pci_sad1 || !pvt->pci_ha0 || | |
de4772c6 | 1313 | !pvt-> pci_tad || !pvt->pci_ras || !pvt->pci_ta) |
eebf11a0 MCC |
1314 | goto enodev; |
1315 | ||
1316 | for (i = 0; i < NUM_CHANNELS; i++) { | |
1317 | if (!pvt->pci_tad[i]) | |
1318 | goto enodev; | |
1319 | } | |
1320 | return 0; | |
1321 | ||
1322 | enodev: | |
1323 | sbridge_printk(KERN_ERR, "Some needed devices are missing\n"); | |
1324 | return -ENODEV; | |
1325 | ||
1326 | error: | |
1327 | sbridge_printk(KERN_ERR, "Device %d, function %d " | |
1328 | "is out of the expected range\n", | |
1329 | slot, func); | |
1330 | return -EINVAL; | |
1331 | } | |
1332 | ||
1333 | /**************************************************************************** | |
1334 | Error check routines | |
1335 | ****************************************************************************/ | |
1336 | ||
1337 | /* | |
1338 | * While Sandy Bridge has error count registers, SMI BIOS read values from | |
1339 | * and resets the counters. So, they are not reliable for the OS to read | |
1340 | * from them. So, we have no option but to just trust on whatever MCE is | |
1341 | * telling us about the errors. | |
1342 | */ | |
1343 | static void sbridge_mce_output_error(struct mem_ctl_info *mci, | |
1344 | const struct mce *m) | |
1345 | { | |
1346 | struct mem_ctl_info *new_mci; | |
1347 | struct sbridge_pvt *pvt = mci->pvt_info; | |
c36e3e77 | 1348 | enum hw_event_mc_err_type tp_event; |
e17a2f42 | 1349 | char *type, *optype, msg[256]; |
eebf11a0 MCC |
1350 | bool ripv = GET_BITFIELD(m->mcgstatus, 0, 0); |
1351 | bool overflow = GET_BITFIELD(m->status, 62, 62); | |
1352 | bool uncorrected_error = GET_BITFIELD(m->status, 61, 61); | |
1353 | bool recoverable = GET_BITFIELD(m->status, 56, 56); | |
1354 | u32 core_err_cnt = GET_BITFIELD(m->status, 38, 52); | |
1355 | u32 mscod = GET_BITFIELD(m->status, 16, 31); | |
1356 | u32 errcode = GET_BITFIELD(m->status, 0, 15); | |
1357 | u32 channel = GET_BITFIELD(m->status, 0, 3); | |
1358 | u32 optypenum = GET_BITFIELD(m->status, 4, 6); | |
1359 | long channel_mask, first_channel; | |
1360 | u8 rank, socket; | |
c36e3e77 | 1361 | int rc, dimm; |
e17a2f42 | 1362 | char *area_type = NULL; |
eebf11a0 | 1363 | |
c36e3e77 MCC |
1364 | if (uncorrected_error) { |
1365 | if (ripv) { | |
1366 | type = "FATAL"; | |
1367 | tp_event = HW_EVENT_ERR_FATAL; | |
1368 | } else { | |
1369 | type = "NON_FATAL"; | |
1370 | tp_event = HW_EVENT_ERR_UNCORRECTED; | |
1371 | } | |
1372 | } else { | |
1373 | type = "CORRECTED"; | |
1374 | tp_event = HW_EVENT_ERR_CORRECTED; | |
1375 | } | |
eebf11a0 MCC |
1376 | |
1377 | /* | |
15ed103a | 1378 | * According with Table 15-9 of the Intel Architecture spec vol 3A, |
eebf11a0 MCC |
1379 | * memory errors should fit in this mask: |
1380 | * 000f 0000 1mmm cccc (binary) | |
1381 | * where: | |
1382 | * f = Correction Report Filtering Bit. If 1, subsequent errors | |
1383 | * won't be shown | |
1384 | * mmm = error type | |
1385 | * cccc = channel | |
1386 | * If the mask doesn't match, report an error to the parsing logic | |
1387 | */ | |
1388 | if (! ((errcode & 0xef80) == 0x80)) { | |
1389 | optype = "Can't parse: it is not a mem"; | |
1390 | } else { | |
1391 | switch (optypenum) { | |
1392 | case 0: | |
c36e3e77 | 1393 | optype = "generic undef request error"; |
eebf11a0 MCC |
1394 | break; |
1395 | case 1: | |
c36e3e77 | 1396 | optype = "memory read error"; |
eebf11a0 MCC |
1397 | break; |
1398 | case 2: | |
c36e3e77 | 1399 | optype = "memory write error"; |
eebf11a0 MCC |
1400 | break; |
1401 | case 3: | |
c36e3e77 | 1402 | optype = "addr/cmd error"; |
eebf11a0 MCC |
1403 | break; |
1404 | case 4: | |
c36e3e77 | 1405 | optype = "memory scrubbing error"; |
eebf11a0 MCC |
1406 | break; |
1407 | default: | |
1408 | optype = "reserved"; | |
1409 | break; | |
1410 | } | |
1411 | } | |
1412 | ||
1413 | rc = get_memory_error_data(mci, m->addr, &socket, | |
e17a2f42 | 1414 | &channel_mask, &rank, &area_type, msg); |
eebf11a0 | 1415 | if (rc < 0) |
c36e3e77 | 1416 | goto err_parsing; |
eebf11a0 MCC |
1417 | new_mci = get_mci_for_node_id(socket); |
1418 | if (!new_mci) { | |
c36e3e77 MCC |
1419 | strcpy(msg, "Error: socket got corrupted!"); |
1420 | goto err_parsing; | |
eebf11a0 MCC |
1421 | } |
1422 | mci = new_mci; | |
1423 | pvt = mci->pvt_info; | |
1424 | ||
1425 | first_channel = find_first_bit(&channel_mask, NUM_CHANNELS); | |
1426 | ||
1427 | if (rank < 4) | |
1428 | dimm = 0; | |
1429 | else if (rank < 8) | |
1430 | dimm = 1; | |
1431 | else | |
1432 | dimm = 2; | |
1433 | ||
eebf11a0 MCC |
1434 | |
1435 | /* | |
e17a2f42 MCC |
1436 | * FIXME: On some memory configurations (mirror, lockstep), the |
1437 | * Memory Controller can't point the error to a single DIMM. The | |
1438 | * EDAC core should be handling the channel mask, in order to point | |
1439 | * to the group of dimm's where the error may be happening. | |
eebf11a0 | 1440 | */ |
c36e3e77 | 1441 | snprintf(msg, sizeof(msg), |
c1053839 | 1442 | "%s%s area:%s err_code:%04x:%04x socket:%d channel_mask:%ld rank:%d", |
e17a2f42 MCC |
1443 | overflow ? " OVERFLOW" : "", |
1444 | (uncorrected_error && recoverable) ? " recoverable" : "", | |
1445 | area_type, | |
1446 | mscod, errcode, | |
1447 | socket, | |
1448 | channel_mask, | |
1449 | rank); | |
eebf11a0 | 1450 | |
956b9ba1 | 1451 | edac_dbg(0, "%s\n", msg); |
eebf11a0 | 1452 | |
c36e3e77 MCC |
1453 | /* FIXME: need support for channel mask */ |
1454 | ||
ebe7e30a SJ |
1455 | if (channel == CHANNEL_UNSPECIFIED) |
1456 | channel = -1; | |
1457 | ||
eebf11a0 | 1458 | /* Call the helper to output message */ |
c1053839 | 1459 | edac_mc_handle_error(tp_event, mci, core_err_cnt, |
c36e3e77 MCC |
1460 | m->addr >> PAGE_SHIFT, m->addr & ~PAGE_MASK, 0, |
1461 | channel, dimm, -1, | |
03f7eae8 | 1462 | optype, msg); |
c36e3e77 MCC |
1463 | return; |
1464 | err_parsing: | |
c1053839 | 1465 | edac_mc_handle_error(tp_event, mci, core_err_cnt, 0, 0, 0, |
c36e3e77 | 1466 | -1, -1, -1, |
03f7eae8 | 1467 | msg, ""); |
eebf11a0 | 1468 | |
eebf11a0 MCC |
1469 | } |
1470 | ||
1471 | /* | |
1472 | * sbridge_check_error Retrieve and process errors reported by the | |
1473 | * hardware. Called by the Core module. | |
1474 | */ | |
1475 | static void sbridge_check_error(struct mem_ctl_info *mci) | |
1476 | { | |
1477 | struct sbridge_pvt *pvt = mci->pvt_info; | |
1478 | int i; | |
1479 | unsigned count = 0; | |
1480 | struct mce *m; | |
1481 | ||
1482 | /* | |
1483 | * MCE first step: Copy all mce errors into a temporary buffer | |
1484 | * We use a double buffering here, to reduce the risk of | |
1485 | * loosing an error. | |
1486 | */ | |
1487 | smp_rmb(); | |
1488 | count = (pvt->mce_out + MCE_LOG_LEN - pvt->mce_in) | |
1489 | % MCE_LOG_LEN; | |
1490 | if (!count) | |
1491 | return; | |
1492 | ||
1493 | m = pvt->mce_outentry; | |
1494 | if (pvt->mce_in + count > MCE_LOG_LEN) { | |
1495 | unsigned l = MCE_LOG_LEN - pvt->mce_in; | |
1496 | ||
1497 | memcpy(m, &pvt->mce_entry[pvt->mce_in], sizeof(*m) * l); | |
1498 | smp_wmb(); | |
1499 | pvt->mce_in = 0; | |
1500 | count -= l; | |
1501 | m += l; | |
1502 | } | |
1503 | memcpy(m, &pvt->mce_entry[pvt->mce_in], sizeof(*m) * count); | |
1504 | smp_wmb(); | |
1505 | pvt->mce_in += count; | |
1506 | ||
1507 | smp_rmb(); | |
1508 | if (pvt->mce_overrun) { | |
1509 | sbridge_printk(KERN_ERR, "Lost %d memory errors\n", | |
1510 | pvt->mce_overrun); | |
1511 | smp_wmb(); | |
1512 | pvt->mce_overrun = 0; | |
1513 | } | |
1514 | ||
1515 | /* | |
1516 | * MCE second step: parse errors and display | |
1517 | */ | |
1518 | for (i = 0; i < count; i++) | |
1519 | sbridge_mce_output_error(mci, &pvt->mce_outentry[i]); | |
1520 | } | |
1521 | ||
1522 | /* | |
1523 | * sbridge_mce_check_error Replicates mcelog routine to get errors | |
1524 | * This routine simply queues mcelog errors, and | |
1525 | * return. The error itself should be handled later | |
1526 | * by sbridge_check_error. | |
1527 | * WARNING: As this routine should be called at NMI time, extra care should | |
1528 | * be taken to avoid deadlocks, and to be as fast as possible. | |
1529 | */ | |
3d78c9af MCC |
1530 | static int sbridge_mce_check_error(struct notifier_block *nb, unsigned long val, |
1531 | void *data) | |
eebf11a0 | 1532 | { |
3d78c9af MCC |
1533 | struct mce *mce = (struct mce *)data; |
1534 | struct mem_ctl_info *mci; | |
1535 | struct sbridge_pvt *pvt; | |
1536 | ||
1537 | mci = get_mci_for_node_id(mce->socketid); | |
1538 | if (!mci) | |
1539 | return NOTIFY_BAD; | |
1540 | pvt = mci->pvt_info; | |
eebf11a0 MCC |
1541 | |
1542 | /* | |
1543 | * Just let mcelog handle it if the error is | |
1544 | * outside the memory controller. A memory error | |
1545 | * is indicated by bit 7 = 1 and bits = 8-11,13-15 = 0. | |
1546 | * bit 12 has an special meaning. | |
1547 | */ | |
1548 | if ((mce->status & 0xefff) >> 7 != 1) | |
3d78c9af | 1549 | return NOTIFY_DONE; |
eebf11a0 MCC |
1550 | |
1551 | printk("sbridge: HANDLING MCE MEMORY ERROR\n"); | |
1552 | ||
1553 | printk("CPU %d: Machine Check Exception: %Lx Bank %d: %016Lx\n", | |
1554 | mce->extcpu, mce->mcgstatus, mce->bank, mce->status); | |
1555 | printk("TSC %llx ", mce->tsc); | |
1556 | printk("ADDR %llx ", mce->addr); | |
1557 | printk("MISC %llx ", mce->misc); | |
1558 | ||
1559 | printk("PROCESSOR %u:%x TIME %llu SOCKET %u APIC %x\n", | |
1560 | mce->cpuvendor, mce->cpuid, mce->time, | |
1561 | mce->socketid, mce->apicid); | |
1562 | ||
eebf11a0 MCC |
1563 | /* Only handle if it is the right mc controller */ |
1564 | if (cpu_data(mce->cpu).phys_proc_id != pvt->sbridge_dev->mc) | |
3d78c9af | 1565 | return NOTIFY_DONE; |
eebf11a0 MCC |
1566 | |
1567 | smp_rmb(); | |
1568 | if ((pvt->mce_out + 1) % MCE_LOG_LEN == pvt->mce_in) { | |
1569 | smp_wmb(); | |
1570 | pvt->mce_overrun++; | |
3d78c9af | 1571 | return NOTIFY_DONE; |
eebf11a0 MCC |
1572 | } |
1573 | ||
1574 | /* Copy memory error at the ringbuffer */ | |
1575 | memcpy(&pvt->mce_entry[pvt->mce_out], mce, sizeof(*mce)); | |
1576 | smp_wmb(); | |
1577 | pvt->mce_out = (pvt->mce_out + 1) % MCE_LOG_LEN; | |
1578 | ||
1579 | /* Handle fatal errors immediately */ | |
1580 | if (mce->mcgstatus & 1) | |
1581 | sbridge_check_error(mci); | |
1582 | ||
1583 | /* Advice mcelog that the error were handled */ | |
3d78c9af | 1584 | return NOTIFY_STOP; |
eebf11a0 MCC |
1585 | } |
1586 | ||
3d78c9af MCC |
1587 | static struct notifier_block sbridge_mce_dec = { |
1588 | .notifier_call = sbridge_mce_check_error, | |
1589 | }; | |
1590 | ||
eebf11a0 MCC |
1591 | /**************************************************************************** |
1592 | EDAC register/unregister logic | |
1593 | ****************************************************************************/ | |
1594 | ||
1595 | static void sbridge_unregister_mci(struct sbridge_dev *sbridge_dev) | |
1596 | { | |
1597 | struct mem_ctl_info *mci = sbridge_dev->mci; | |
1598 | struct sbridge_pvt *pvt; | |
1599 | ||
1600 | if (unlikely(!mci || !mci->pvt_info)) { | |
956b9ba1 | 1601 | edac_dbg(0, "MC: dev = %p\n", &sbridge_dev->pdev[0]->dev); |
eebf11a0 MCC |
1602 | |
1603 | sbridge_printk(KERN_ERR, "Couldn't find mci handler\n"); | |
1604 | return; | |
1605 | } | |
1606 | ||
1607 | pvt = mci->pvt_info; | |
1608 | ||
956b9ba1 JP |
1609 | edac_dbg(0, "MC: mci = %p, dev = %p\n", |
1610 | mci, &sbridge_dev->pdev[0]->dev); | |
eebf11a0 | 1611 | |
eebf11a0 | 1612 | /* Remove MC sysfs nodes */ |
fd687502 | 1613 | edac_mc_del_mc(mci->pdev); |
eebf11a0 | 1614 | |
956b9ba1 | 1615 | edac_dbg(1, "%s: free mci struct\n", mci->ctl_name); |
eebf11a0 MCC |
1616 | kfree(mci->ctl_name); |
1617 | edac_mc_free(mci); | |
1618 | sbridge_dev->mci = NULL; | |
1619 | } | |
1620 | ||
1621 | static int sbridge_register_mci(struct sbridge_dev *sbridge_dev) | |
1622 | { | |
1623 | struct mem_ctl_info *mci; | |
c36e3e77 | 1624 | struct edac_mc_layer layers[2]; |
eebf11a0 | 1625 | struct sbridge_pvt *pvt; |
c36e3e77 | 1626 | int rc; |
eebf11a0 MCC |
1627 | |
1628 | /* Check the number of active and not disabled channels */ | |
c36e3e77 | 1629 | rc = check_if_ecc_is_active(sbridge_dev->bus); |
eebf11a0 MCC |
1630 | if (unlikely(rc < 0)) |
1631 | return rc; | |
1632 | ||
1633 | /* allocate a new MC control structure */ | |
c36e3e77 MCC |
1634 | layers[0].type = EDAC_MC_LAYER_CHANNEL; |
1635 | layers[0].size = NUM_CHANNELS; | |
1636 | layers[0].is_virt_csrow = false; | |
1637 | layers[1].type = EDAC_MC_LAYER_SLOT; | |
1638 | layers[1].size = MAX_DIMMS; | |
1639 | layers[1].is_virt_csrow = true; | |
ca0907b9 | 1640 | mci = edac_mc_alloc(sbridge_dev->mc, ARRAY_SIZE(layers), layers, |
c36e3e77 MCC |
1641 | sizeof(*pvt)); |
1642 | ||
eebf11a0 MCC |
1643 | if (unlikely(!mci)) |
1644 | return -ENOMEM; | |
1645 | ||
956b9ba1 JP |
1646 | edac_dbg(0, "MC: mci = %p, dev = %p\n", |
1647 | mci, &sbridge_dev->pdev[0]->dev); | |
eebf11a0 MCC |
1648 | |
1649 | pvt = mci->pvt_info; | |
1650 | memset(pvt, 0, sizeof(*pvt)); | |
1651 | ||
1652 | /* Associate sbridge_dev and mci for future usage */ | |
1653 | pvt->sbridge_dev = sbridge_dev; | |
1654 | sbridge_dev->mci = mci; | |
1655 | ||
1656 | mci->mtype_cap = MEM_FLAG_DDR3; | |
1657 | mci->edac_ctl_cap = EDAC_FLAG_NONE; | |
1658 | mci->edac_cap = EDAC_FLAG_NONE; | |
1659 | mci->mod_name = "sbridge_edac.c"; | |
1660 | mci->mod_ver = SBRIDGE_REVISION; | |
1661 | mci->ctl_name = kasprintf(GFP_KERNEL, "Sandy Bridge Socket#%d", mci->mc_idx); | |
1662 | mci->dev_name = pci_name(sbridge_dev->pdev[0]); | |
1663 | mci->ctl_page_to_phys = NULL; | |
1664 | ||
1665 | /* Set the function pointer to an actual operation function */ | |
1666 | mci->edac_check = sbridge_check_error; | |
1667 | ||
1668 | /* Store pci devices at mci for faster access */ | |
1669 | rc = mci_bind_devs(mci, sbridge_dev); | |
1670 | if (unlikely(rc < 0)) | |
1671 | goto fail0; | |
1672 | ||
1673 | /* Get dimm basic config and the memory layout */ | |
1674 | get_dimm_config(mci); | |
1675 | get_memory_layout(mci); | |
1676 | ||
1677 | /* record ptr to the generic device */ | |
fd687502 | 1678 | mci->pdev = &sbridge_dev->pdev[0]->dev; |
eebf11a0 MCC |
1679 | |
1680 | /* add this new MC control structure to EDAC's list of MCs */ | |
1681 | if (unlikely(edac_mc_add_mc(mci))) { | |
956b9ba1 | 1682 | edac_dbg(0, "MC: failed edac_mc_add_mc()\n"); |
eebf11a0 MCC |
1683 | rc = -EINVAL; |
1684 | goto fail0; | |
1685 | } | |
1686 | ||
eebf11a0 | 1687 | return 0; |
eebf11a0 MCC |
1688 | |
1689 | fail0: | |
1690 | kfree(mci->ctl_name); | |
1691 | edac_mc_free(mci); | |
1692 | sbridge_dev->mci = NULL; | |
1693 | return rc; | |
1694 | } | |
1695 | ||
1696 | /* | |
1697 | * sbridge_probe Probe for ONE instance of device to see if it is | |
1698 | * present. | |
1699 | * return: | |
1700 | * 0 for FOUND a device | |
1701 | * < 0 for error code | |
1702 | */ | |
1703 | ||
9b3c6e85 | 1704 | static int sbridge_probe(struct pci_dev *pdev, const struct pci_device_id *id) |
eebf11a0 MCC |
1705 | { |
1706 | int rc; | |
1707 | u8 mc, num_mc = 0; | |
1708 | struct sbridge_dev *sbridge_dev; | |
1709 | ||
1710 | /* get the pci devices we want to reserve for our use */ | |
1711 | mutex_lock(&sbridge_edac_lock); | |
1712 | ||
1713 | /* | |
1714 | * All memory controllers are allocated at the first pass. | |
1715 | */ | |
1716 | if (unlikely(probed >= 1)) { | |
1717 | mutex_unlock(&sbridge_edac_lock); | |
1718 | return -ENODEV; | |
1719 | } | |
1720 | probed++; | |
1721 | ||
1722 | rc = sbridge_get_all_devices(&num_mc); | |
1723 | if (unlikely(rc < 0)) | |
1724 | goto fail0; | |
1725 | mc = 0; | |
1726 | ||
1727 | list_for_each_entry(sbridge_dev, &sbridge_edac_list, list) { | |
956b9ba1 JP |
1728 | edac_dbg(0, "Registering MC#%d (%d of %d)\n", |
1729 | mc, mc + 1, num_mc); | |
eebf11a0 MCC |
1730 | sbridge_dev->mc = mc++; |
1731 | rc = sbridge_register_mci(sbridge_dev); | |
1732 | if (unlikely(rc < 0)) | |
1733 | goto fail1; | |
1734 | } | |
1735 | ||
1736 | sbridge_printk(KERN_INFO, "Driver loaded.\n"); | |
1737 | ||
1738 | mutex_unlock(&sbridge_edac_lock); | |
1739 | return 0; | |
1740 | ||
1741 | fail1: | |
1742 | list_for_each_entry(sbridge_dev, &sbridge_edac_list, list) | |
1743 | sbridge_unregister_mci(sbridge_dev); | |
1744 | ||
1745 | sbridge_put_all_devices(); | |
1746 | fail0: | |
1747 | mutex_unlock(&sbridge_edac_lock); | |
1748 | return rc; | |
1749 | } | |
1750 | ||
1751 | /* | |
1752 | * sbridge_remove destructor for one instance of device | |
1753 | * | |
1754 | */ | |
9b3c6e85 | 1755 | static void sbridge_remove(struct pci_dev *pdev) |
eebf11a0 MCC |
1756 | { |
1757 | struct sbridge_dev *sbridge_dev; | |
1758 | ||
956b9ba1 | 1759 | edac_dbg(0, "\n"); |
eebf11a0 MCC |
1760 | |
1761 | /* | |
1762 | * we have a trouble here: pdev value for removal will be wrong, since | |
1763 | * it will point to the X58 register used to detect that the machine | |
1764 | * is a Nehalem or upper design. However, due to the way several PCI | |
1765 | * devices are grouped together to provide MC functionality, we need | |
1766 | * to use a different method for releasing the devices | |
1767 | */ | |
1768 | ||
1769 | mutex_lock(&sbridge_edac_lock); | |
1770 | ||
1771 | if (unlikely(!probed)) { | |
1772 | mutex_unlock(&sbridge_edac_lock); | |
1773 | return; | |
1774 | } | |
1775 | ||
1776 | list_for_each_entry(sbridge_dev, &sbridge_edac_list, list) | |
1777 | sbridge_unregister_mci(sbridge_dev); | |
1778 | ||
1779 | /* Release PCI resources */ | |
1780 | sbridge_put_all_devices(); | |
1781 | ||
1782 | probed--; | |
1783 | ||
1784 | mutex_unlock(&sbridge_edac_lock); | |
1785 | } | |
1786 | ||
1787 | MODULE_DEVICE_TABLE(pci, sbridge_pci_tbl); | |
1788 | ||
1789 | /* | |
1790 | * sbridge_driver pci_driver structure for this module | |
1791 | * | |
1792 | */ | |
1793 | static struct pci_driver sbridge_driver = { | |
1794 | .name = "sbridge_edac", | |
1795 | .probe = sbridge_probe, | |
9b3c6e85 | 1796 | .remove = sbridge_remove, |
eebf11a0 MCC |
1797 | .id_table = sbridge_pci_tbl, |
1798 | }; | |
1799 | ||
1800 | /* | |
1801 | * sbridge_init Module entry function | |
1802 | * Try to initialize this module for its devices | |
1803 | */ | |
1804 | static int __init sbridge_init(void) | |
1805 | { | |
1806 | int pci_rc; | |
1807 | ||
956b9ba1 | 1808 | edac_dbg(2, "\n"); |
eebf11a0 MCC |
1809 | |
1810 | /* Ensure that the OPSTATE is set correctly for POLL or NMI */ | |
1811 | opstate_init(); | |
1812 | ||
1813 | pci_rc = pci_register_driver(&sbridge_driver); | |
1814 | ||
e35fca47 CG |
1815 | if (pci_rc >= 0) { |
1816 | mce_register_decode_chain(&sbridge_mce_dec); | |
eebf11a0 | 1817 | return 0; |
e35fca47 | 1818 | } |
eebf11a0 MCC |
1819 | |
1820 | sbridge_printk(KERN_ERR, "Failed to register device with error %d.\n", | |
1821 | pci_rc); | |
1822 | ||
1823 | return pci_rc; | |
1824 | } | |
1825 | ||
1826 | /* | |
1827 | * sbridge_exit() Module exit function | |
1828 | * Unregister the driver | |
1829 | */ | |
1830 | static void __exit sbridge_exit(void) | |
1831 | { | |
956b9ba1 | 1832 | edac_dbg(2, "\n"); |
eebf11a0 | 1833 | pci_unregister_driver(&sbridge_driver); |
e35fca47 | 1834 | mce_unregister_decode_chain(&sbridge_mce_dec); |
eebf11a0 MCC |
1835 | } |
1836 | ||
1837 | module_init(sbridge_init); | |
1838 | module_exit(sbridge_exit); | |
1839 | ||
1840 | module_param(edac_op_state, int, 0444); | |
1841 | MODULE_PARM_DESC(edac_op_state, "EDAC Error Reporting state: 0=Poll,1=NMI"); | |
1842 | ||
1843 | MODULE_LICENSE("GPL"); | |
1844 | MODULE_AUTHOR("Mauro Carvalho Chehab <mchehab@redhat.com>"); | |
1845 | MODULE_AUTHOR("Red Hat Inc. (http://www.redhat.com)"); | |
1846 | MODULE_DESCRIPTION("MC Driver for Intel Sandy Bridge memory controllers - " | |
1847 | SBRIDGE_REVISION); |