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 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> 21 #include <linux/smp.h> 22 #include <linux/bitmap.h> 23 #include <linux/math64.h> 24 #include <asm/processor.h> 25 #include <asm/mce.h> 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: 1.1.1 " 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) & GENMASK_ULL(hi, lo)) >> (lo)) 54 55 /* Devices 12 Function 6, Offsets 0x80 to 0xcc */ 56 static const u32 sbridge_dram_rule[] = { 57 0x80, 0x88, 0x90, 0x98, 0xa0, 58 0xa8, 0xb0, 0xb8, 0xc0, 0xc8, 59 }; 60 61 static const u32 ibridge_dram_rule[] = { 62 0x60, 0x68, 0x70, 0x78, 0x80, 63 0x88, 0x90, 0x98, 0xa0, 0xa8, 64 0xb0, 0xb8, 0xc0, 0xc8, 0xd0, 65 0xd8, 0xe0, 0xe8, 0xf0, 0xf8, 66 }; 67 68 static const u32 knl_dram_rule[] = { 69 0x60, 0x68, 0x70, 0x78, 0x80, /* 0-4 */ 70 0x88, 0x90, 0x98, 0xa0, 0xa8, /* 5-9 */ 71 0xb0, 0xb8, 0xc0, 0xc8, 0xd0, /* 10-14 */ 72 0xd8, 0xe0, 0xe8, 0xf0, 0xf8, /* 15-19 */ 73 0x100, 0x108, 0x110, 0x118, /* 20-23 */ 74 }; 75 76 #define DRAM_RULE_ENABLE(reg) GET_BITFIELD(reg, 0, 0) 77 #define A7MODE(reg) GET_BITFIELD(reg, 26, 26) 78 79 static char *show_dram_attr(u32 attr) 80 { 81 switch (attr) { 82 case 0: 83 return "DRAM"; 84 case 1: 85 return "MMCFG"; 86 case 2: 87 return "NXM"; 88 default: 89 return "unknown"; 90 } 91 } 92 93 static const u32 sbridge_interleave_list[] = { 94 0x84, 0x8c, 0x94, 0x9c, 0xa4, 95 0xac, 0xb4, 0xbc, 0xc4, 0xcc, 96 }; 97 98 static const u32 ibridge_interleave_list[] = { 99 0x64, 0x6c, 0x74, 0x7c, 0x84, 100 0x8c, 0x94, 0x9c, 0xa4, 0xac, 101 0xb4, 0xbc, 0xc4, 0xcc, 0xd4, 102 0xdc, 0xe4, 0xec, 0xf4, 0xfc, 103 }; 104 105 static const u32 knl_interleave_list[] = { 106 0x64, 0x6c, 0x74, 0x7c, 0x84, /* 0-4 */ 107 0x8c, 0x94, 0x9c, 0xa4, 0xac, /* 5-9 */ 108 0xb4, 0xbc, 0xc4, 0xcc, 0xd4, /* 10-14 */ 109 0xdc, 0xe4, 0xec, 0xf4, 0xfc, /* 15-19 */ 110 0x104, 0x10c, 0x114, 0x11c, /* 20-23 */ 111 }; 112 113 struct interleave_pkg { 114 unsigned char start; 115 unsigned char end; 116 }; 117 118 static const struct interleave_pkg sbridge_interleave_pkg[] = { 119 { 0, 2 }, 120 { 3, 5 }, 121 { 8, 10 }, 122 { 11, 13 }, 123 { 16, 18 }, 124 { 19, 21 }, 125 { 24, 26 }, 126 { 27, 29 }, 127 }; 128 129 static const struct interleave_pkg ibridge_interleave_pkg[] = { 130 { 0, 3 }, 131 { 4, 7 }, 132 { 8, 11 }, 133 { 12, 15 }, 134 { 16, 19 }, 135 { 20, 23 }, 136 { 24, 27 }, 137 { 28, 31 }, 138 }; 139 140 static inline int sad_pkg(const struct interleave_pkg *table, u32 reg, 141 int interleave) 142 { 143 return GET_BITFIELD(reg, table[interleave].start, 144 table[interleave].end); 145 } 146 147 /* Devices 12 Function 7 */ 148 149 #define TOLM 0x80 150 #define TOHM 0x84 151 #define HASWELL_TOLM 0xd0 152 #define HASWELL_TOHM_0 0xd4 153 #define HASWELL_TOHM_1 0xd8 154 #define KNL_TOLM 0xd0 155 #define KNL_TOHM_0 0xd4 156 #define KNL_TOHM_1 0xd8 157 158 #define GET_TOLM(reg) ((GET_BITFIELD(reg, 0, 3) << 28) | 0x3ffffff) 159 #define GET_TOHM(reg) ((GET_BITFIELD(reg, 0, 20) << 25) | 0x3ffffff) 160 161 /* Device 13 Function 6 */ 162 163 #define SAD_TARGET 0xf0 164 165 #define SOURCE_ID(reg) GET_BITFIELD(reg, 9, 11) 166 167 #define SOURCE_ID_KNL(reg) GET_BITFIELD(reg, 12, 14) 168 169 #define SAD_CONTROL 0xf4 170 171 /* Device 14 function 0 */ 172 173 static const u32 tad_dram_rule[] = { 174 0x40, 0x44, 0x48, 0x4c, 175 0x50, 0x54, 0x58, 0x5c, 176 0x60, 0x64, 0x68, 0x6c, 177 }; 178 #define MAX_TAD ARRAY_SIZE(tad_dram_rule) 179 180 #define TAD_LIMIT(reg) ((GET_BITFIELD(reg, 12, 31) << 26) | 0x3ffffff) 181 #define TAD_SOCK(reg) GET_BITFIELD(reg, 10, 11) 182 #define TAD_CH(reg) GET_BITFIELD(reg, 8, 9) 183 #define TAD_TGT3(reg) GET_BITFIELD(reg, 6, 7) 184 #define TAD_TGT2(reg) GET_BITFIELD(reg, 4, 5) 185 #define TAD_TGT1(reg) GET_BITFIELD(reg, 2, 3) 186 #define TAD_TGT0(reg) GET_BITFIELD(reg, 0, 1) 187 188 /* Device 15, function 0 */ 189 190 #define MCMTR 0x7c 191 #define KNL_MCMTR 0x624 192 193 #define IS_ECC_ENABLED(mcmtr) GET_BITFIELD(mcmtr, 2, 2) 194 #define IS_LOCKSTEP_ENABLED(mcmtr) GET_BITFIELD(mcmtr, 1, 1) 195 #define IS_CLOSE_PG(mcmtr) GET_BITFIELD(mcmtr, 0, 0) 196 197 /* Device 15, function 1 */ 198 199 #define RASENABLES 0xac 200 #define IS_MIRROR_ENABLED(reg) GET_BITFIELD(reg, 0, 0) 201 202 /* Device 15, functions 2-5 */ 203 204 static const int mtr_regs[] = { 205 0x80, 0x84, 0x88, 206 }; 207 208 static const int knl_mtr_reg = 0xb60; 209 210 #define RANK_DISABLE(mtr) GET_BITFIELD(mtr, 16, 19) 211 #define IS_DIMM_PRESENT(mtr) GET_BITFIELD(mtr, 14, 14) 212 #define RANK_CNT_BITS(mtr) GET_BITFIELD(mtr, 12, 13) 213 #define RANK_WIDTH_BITS(mtr) GET_BITFIELD(mtr, 2, 4) 214 #define COL_WIDTH_BITS(mtr) GET_BITFIELD(mtr, 0, 1) 215 216 static const u32 tad_ch_nilv_offset[] = { 217 0x90, 0x94, 0x98, 0x9c, 218 0xa0, 0xa4, 0xa8, 0xac, 219 0xb0, 0xb4, 0xb8, 0xbc, 220 }; 221 #define CHN_IDX_OFFSET(reg) GET_BITFIELD(reg, 28, 29) 222 #define TAD_OFFSET(reg) (GET_BITFIELD(reg, 6, 25) << 26) 223 224 static const u32 rir_way_limit[] = { 225 0x108, 0x10c, 0x110, 0x114, 0x118, 226 }; 227 #define MAX_RIR_RANGES ARRAY_SIZE(rir_way_limit) 228 229 #define IS_RIR_VALID(reg) GET_BITFIELD(reg, 31, 31) 230 #define RIR_WAY(reg) GET_BITFIELD(reg, 28, 29) 231 232 #define MAX_RIR_WAY 8 233 234 static const u32 rir_offset[MAX_RIR_RANGES][MAX_RIR_WAY] = { 235 { 0x120, 0x124, 0x128, 0x12c, 0x130, 0x134, 0x138, 0x13c }, 236 { 0x140, 0x144, 0x148, 0x14c, 0x150, 0x154, 0x158, 0x15c }, 237 { 0x160, 0x164, 0x168, 0x16c, 0x170, 0x174, 0x178, 0x17c }, 238 { 0x180, 0x184, 0x188, 0x18c, 0x190, 0x194, 0x198, 0x19c }, 239 { 0x1a0, 0x1a4, 0x1a8, 0x1ac, 0x1b0, 0x1b4, 0x1b8, 0x1bc }, 240 }; 241 242 #define RIR_RNK_TGT(reg) GET_BITFIELD(reg, 16, 19) 243 #define RIR_OFFSET(reg) GET_BITFIELD(reg, 2, 14) 244 245 /* Device 16, functions 2-7 */ 246 247 /* 248 * FIXME: Implement the error count reads directly 249 */ 250 251 static const u32 correrrcnt[] = { 252 0x104, 0x108, 0x10c, 0x110, 253 }; 254 255 #define RANK_ODD_OV(reg) GET_BITFIELD(reg, 31, 31) 256 #define RANK_ODD_ERR_CNT(reg) GET_BITFIELD(reg, 16, 30) 257 #define RANK_EVEN_OV(reg) GET_BITFIELD(reg, 15, 15) 258 #define RANK_EVEN_ERR_CNT(reg) GET_BITFIELD(reg, 0, 14) 259 260 static const u32 correrrthrsld[] = { 261 0x11c, 0x120, 0x124, 0x128, 262 }; 263 264 #define RANK_ODD_ERR_THRSLD(reg) GET_BITFIELD(reg, 16, 30) 265 #define RANK_EVEN_ERR_THRSLD(reg) GET_BITFIELD(reg, 0, 14) 266 267 268 /* Device 17, function 0 */ 269 270 #define SB_RANK_CFG_A 0x0328 271 272 #define IB_RANK_CFG_A 0x0320 273 274 /* 275 * sbridge structs 276 */ 277 278 #define NUM_CHANNELS 8 /* 2MC per socket, four chan per MC */ 279 #define MAX_DIMMS 3 /* Max DIMMS per channel */ 280 #define KNL_MAX_CHAS 38 /* KNL max num. of Cache Home Agents */ 281 #define KNL_MAX_CHANNELS 6 /* KNL max num. of PCI channels */ 282 #define KNL_MAX_EDCS 8 /* Embedded DRAM controllers */ 283 #define CHANNEL_UNSPECIFIED 0xf /* Intel IA32 SDM 15-14 */ 284 285 enum type { 286 SANDY_BRIDGE, 287 IVY_BRIDGE, 288 HASWELL, 289 BROADWELL, 290 KNIGHTS_LANDING, 291 }; 292 293 struct sbridge_pvt; 294 struct sbridge_info { 295 enum type type; 296 u32 mcmtr; 297 u32 rankcfgr; 298 u64 (*get_tolm)(struct sbridge_pvt *pvt); 299 u64 (*get_tohm)(struct sbridge_pvt *pvt); 300 u64 (*rir_limit)(u32 reg); 301 u64 (*sad_limit)(u32 reg); 302 u32 (*interleave_mode)(u32 reg); 303 char* (*show_interleave_mode)(u32 reg); 304 u32 (*dram_attr)(u32 reg); 305 const u32 *dram_rule; 306 const u32 *interleave_list; 307 const struct interleave_pkg *interleave_pkg; 308 u8 max_sad; 309 u8 max_interleave; 310 u8 (*get_node_id)(struct sbridge_pvt *pvt); 311 enum mem_type (*get_memory_type)(struct sbridge_pvt *pvt); 312 enum dev_type (*get_width)(struct sbridge_pvt *pvt, u32 mtr); 313 struct pci_dev *pci_vtd; 314 }; 315 316 struct sbridge_channel { 317 u32 ranks; 318 u32 dimms; 319 }; 320 321 struct pci_id_descr { 322 int dev_id; 323 int optional; 324 }; 325 326 struct pci_id_table { 327 const struct pci_id_descr *descr; 328 int n_devs; 329 }; 330 331 struct sbridge_dev { 332 struct list_head list; 333 u8 bus, mc; 334 u8 node_id, source_id; 335 struct pci_dev **pdev; 336 int n_devs; 337 struct mem_ctl_info *mci; 338 }; 339 340 struct knl_pvt { 341 struct pci_dev *pci_cha[KNL_MAX_CHAS]; 342 struct pci_dev *pci_channel[KNL_MAX_CHANNELS]; 343 struct pci_dev *pci_mc0; 344 struct pci_dev *pci_mc1; 345 struct pci_dev *pci_mc0_misc; 346 struct pci_dev *pci_mc1_misc; 347 struct pci_dev *pci_mc_info; /* tolm, tohm */ 348 }; 349 350 struct sbridge_pvt { 351 struct pci_dev *pci_ta, *pci_ddrio, *pci_ras; 352 struct pci_dev *pci_sad0, *pci_sad1; 353 struct pci_dev *pci_ha0, *pci_ha1; 354 struct pci_dev *pci_br0, *pci_br1; 355 struct pci_dev *pci_ha1_ta; 356 struct pci_dev *pci_tad[NUM_CHANNELS]; 357 358 struct sbridge_dev *sbridge_dev; 359 360 struct sbridge_info info; 361 struct sbridge_channel channel[NUM_CHANNELS]; 362 363 /* Memory type detection */ 364 bool is_mirrored, is_lockstep, is_close_pg; 365 366 /* Fifo double buffers */ 367 struct mce mce_entry[MCE_LOG_LEN]; 368 struct mce mce_outentry[MCE_LOG_LEN]; 369 370 /* Fifo in/out counters */ 371 unsigned mce_in, mce_out; 372 373 /* Count indicator to show errors not got */ 374 unsigned mce_overrun; 375 376 /* Memory description */ 377 u64 tolm, tohm; 378 struct knl_pvt knl; 379 }; 380 381 #define PCI_DESCR(device_id, opt) \ 382 .dev_id = (device_id), \ 383 .optional = opt 384 385 static const struct pci_id_descr pci_dev_descr_sbridge[] = { 386 /* Processor Home Agent */ 387 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_HA0, 0) }, 388 389 /* Memory controller */ 390 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TA, 0) }, 391 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_RAS, 0) }, 392 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD0, 0) }, 393 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD1, 0) }, 394 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD2, 0) }, 395 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD3, 0) }, 396 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_DDRIO, 1) }, 397 398 /* System Address Decoder */ 399 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_SAD0, 0) }, 400 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_SAD1, 0) }, 401 402 /* Broadcast Registers */ 403 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_BR, 0) }, 404 }; 405 406 #define PCI_ID_TABLE_ENTRY(A) { .descr=A, .n_devs = ARRAY_SIZE(A) } 407 static const struct pci_id_table pci_dev_descr_sbridge_table[] = { 408 PCI_ID_TABLE_ENTRY(pci_dev_descr_sbridge), 409 {0,} /* 0 terminated list. */ 410 }; 411 412 /* This changes depending if 1HA or 2HA: 413 * 1HA: 414 * 0x0eb8 (17.0) is DDRIO0 415 * 2HA: 416 * 0x0ebc (17.4) is DDRIO0 417 */ 418 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_1HA_DDRIO0 0x0eb8 419 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_2HA_DDRIO0 0x0ebc 420 421 /* pci ids */ 422 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0 0x0ea0 423 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TA 0x0ea8 424 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_RAS 0x0e71 425 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD0 0x0eaa 426 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD1 0x0eab 427 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD2 0x0eac 428 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD3 0x0ead 429 #define PCI_DEVICE_ID_INTEL_IBRIDGE_SAD 0x0ec8 430 #define PCI_DEVICE_ID_INTEL_IBRIDGE_BR0 0x0ec9 431 #define PCI_DEVICE_ID_INTEL_IBRIDGE_BR1 0x0eca 432 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1 0x0e60 433 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TA 0x0e68 434 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_RAS 0x0e79 435 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD0 0x0e6a 436 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD1 0x0e6b 437 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD2 0x0e6c 438 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD3 0x0e6d 439 440 static const struct pci_id_descr pci_dev_descr_ibridge[] = { 441 /* Processor Home Agent */ 442 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0, 0) }, 443 444 /* Memory controller */ 445 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TA, 0) }, 446 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_RAS, 0) }, 447 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD0, 0) }, 448 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD1, 0) }, 449 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD2, 0) }, 450 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD3, 0) }, 451 452 /* System Address Decoder */ 453 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_SAD, 0) }, 454 455 /* Broadcast Registers */ 456 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_BR0, 1) }, 457 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_BR1, 0) }, 458 459 /* Optional, mode 2HA */ 460 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1, 1) }, 461 #if 0 462 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TA, 1) }, 463 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_RAS, 1) }, 464 #endif 465 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD0, 1) }, 466 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD1, 1) }, 467 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD2, 1) }, 468 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD3, 1) }, 469 470 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_1HA_DDRIO0, 1) }, 471 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_2HA_DDRIO0, 1) }, 472 }; 473 474 static const struct pci_id_table pci_dev_descr_ibridge_table[] = { 475 PCI_ID_TABLE_ENTRY(pci_dev_descr_ibridge), 476 {0,} /* 0 terminated list. */ 477 }; 478 479 /* Haswell support */ 480 /* EN processor: 481 * - 1 IMC 482 * - 3 DDR3 channels, 2 DPC per channel 483 * EP processor: 484 * - 1 or 2 IMC 485 * - 4 DDR4 channels, 3 DPC per channel 486 * EP 4S processor: 487 * - 2 IMC 488 * - 4 DDR4 channels, 3 DPC per channel 489 * EX processor: 490 * - 2 IMC 491 * - each IMC interfaces with a SMI 2 channel 492 * - each SMI channel interfaces with a scalable memory buffer 493 * - each scalable memory buffer supports 4 DDR3/DDR4 channels, 3 DPC 494 */ 495 #define HASWELL_DDRCRCLKCONTROLS 0xa10 /* Ditto on Broadwell */ 496 #define HASWELL_HASYSDEFEATURE2 0x84 497 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_VTD_MISC 0x2f28 498 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0 0x2fa0 499 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1 0x2f60 500 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TA 0x2fa8 501 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_THERMAL 0x2f71 502 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TA 0x2f68 503 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_THERMAL 0x2f79 504 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_CBO_SAD0 0x2ffc 505 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_CBO_SAD1 0x2ffd 506 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD0 0x2faa 507 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD1 0x2fab 508 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD2 0x2fac 509 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD3 0x2fad 510 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD0 0x2f6a 511 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD1 0x2f6b 512 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD2 0x2f6c 513 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD3 0x2f6d 514 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO0 0x2fbd 515 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO1 0x2fbf 516 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO2 0x2fb9 517 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO3 0x2fbb 518 static const struct pci_id_descr pci_dev_descr_haswell[] = { 519 /* first item must be the HA */ 520 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0, 0) }, 521 522 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_CBO_SAD0, 0) }, 523 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_CBO_SAD1, 0) }, 524 525 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1, 1) }, 526 527 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TA, 0) }, 528 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_THERMAL, 0) }, 529 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD0, 0) }, 530 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD1, 0) }, 531 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD2, 1) }, 532 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD3, 1) }, 533 534 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO0, 1) }, 535 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO1, 1) }, 536 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO2, 1) }, 537 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO3, 1) }, 538 539 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TA, 1) }, 540 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_THERMAL, 1) }, 541 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD0, 1) }, 542 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD1, 1) }, 543 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD2, 1) }, 544 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD3, 1) }, 545 }; 546 547 static const struct pci_id_table pci_dev_descr_haswell_table[] = { 548 PCI_ID_TABLE_ENTRY(pci_dev_descr_haswell), 549 {0,} /* 0 terminated list. */ 550 }; 551 552 /* Knight's Landing Support */ 553 /* 554 * KNL's memory channels are swizzled between memory controllers. 555 * MC0 is mapped to CH3,5,6 and MC1 is mapped to CH0,1,2 556 */ 557 #define knl_channel_remap(channel) ((channel + 3) % 6) 558 559 /* Memory controller, TAD tables, error injection - 2-8-0, 2-9-0 (2 of these) */ 560 #define PCI_DEVICE_ID_INTEL_KNL_IMC_MC 0x7840 561 /* DRAM channel stuff; bank addrs, dimmmtr, etc.. 2-8-2 - 2-9-4 (6 of these) */ 562 #define PCI_DEVICE_ID_INTEL_KNL_IMC_CHANNEL 0x7843 563 /* kdrwdbu TAD limits/offsets, MCMTR - 2-10-1, 2-11-1 (2 of these) */ 564 #define PCI_DEVICE_ID_INTEL_KNL_IMC_TA 0x7844 565 /* CHA broadcast registers, dram rules - 1-29-0 (1 of these) */ 566 #define PCI_DEVICE_ID_INTEL_KNL_IMC_SAD0 0x782a 567 /* SAD target - 1-29-1 (1 of these) */ 568 #define PCI_DEVICE_ID_INTEL_KNL_IMC_SAD1 0x782b 569 /* Caching / Home Agent */ 570 #define PCI_DEVICE_ID_INTEL_KNL_IMC_CHA 0x782c 571 /* Device with TOLM and TOHM, 0-5-0 (1 of these) */ 572 #define PCI_DEVICE_ID_INTEL_KNL_IMC_TOLHM 0x7810 573 574 /* 575 * KNL differs from SB, IB, and Haswell in that it has multiple 576 * instances of the same device with the same device ID, so we handle that 577 * by creating as many copies in the table as we expect to find. 578 * (Like device ID must be grouped together.) 579 */ 580 581 static const struct pci_id_descr pci_dev_descr_knl[] = { 582 [0] = { PCI_DESCR(PCI_DEVICE_ID_INTEL_KNL_IMC_SAD0, 0) }, 583 [1] = { PCI_DESCR(PCI_DEVICE_ID_INTEL_KNL_IMC_SAD1, 0) }, 584 [2 ... 3] = { PCI_DESCR(PCI_DEVICE_ID_INTEL_KNL_IMC_MC, 0)}, 585 [4 ... 41] = { PCI_DESCR(PCI_DEVICE_ID_INTEL_KNL_IMC_CHA, 0) }, 586 [42 ... 47] = { PCI_DESCR(PCI_DEVICE_ID_INTEL_KNL_IMC_CHANNEL, 0) }, 587 [48] = { PCI_DESCR(PCI_DEVICE_ID_INTEL_KNL_IMC_TA, 0) }, 588 [49] = { PCI_DESCR(PCI_DEVICE_ID_INTEL_KNL_IMC_TOLHM, 0) }, 589 }; 590 591 static const struct pci_id_table pci_dev_descr_knl_table[] = { 592 PCI_ID_TABLE_ENTRY(pci_dev_descr_knl), 593 {0,} 594 }; 595 596 /* 597 * Broadwell support 598 * 599 * DE processor: 600 * - 1 IMC 601 * - 2 DDR3 channels, 2 DPC per channel 602 * EP processor: 603 * - 1 or 2 IMC 604 * - 4 DDR4 channels, 3 DPC per channel 605 * EP 4S processor: 606 * - 2 IMC 607 * - 4 DDR4 channels, 3 DPC per channel 608 * EX processor: 609 * - 2 IMC 610 * - each IMC interfaces with a SMI 2 channel 611 * - each SMI channel interfaces with a scalable memory buffer 612 * - each scalable memory buffer supports 4 DDR3/DDR4 channels, 3 DPC 613 */ 614 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_VTD_MISC 0x6f28 615 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0 0x6fa0 616 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1 0x6f60 617 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TA 0x6fa8 618 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_THERMAL 0x6f71 619 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TA 0x6f68 620 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_THERMAL 0x6f79 621 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_CBO_SAD0 0x6ffc 622 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_CBO_SAD1 0x6ffd 623 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD0 0x6faa 624 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD1 0x6fab 625 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD2 0x6fac 626 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD3 0x6fad 627 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD0 0x6f6a 628 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD1 0x6f6b 629 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD2 0x6f6c 630 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD3 0x6f6d 631 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_DDRIO0 0x6faf 632 633 static const struct pci_id_descr pci_dev_descr_broadwell[] = { 634 /* first item must be the HA */ 635 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0, 0) }, 636 637 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_CBO_SAD0, 0) }, 638 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_CBO_SAD1, 0) }, 639 640 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1, 1) }, 641 642 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TA, 0) }, 643 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_THERMAL, 0) }, 644 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD0, 0) }, 645 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD1, 0) }, 646 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD2, 1) }, 647 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD3, 1) }, 648 649 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_DDRIO0, 1) }, 650 651 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TA, 1) }, 652 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_THERMAL, 1) }, 653 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD0, 1) }, 654 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD1, 1) }, 655 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD2, 1) }, 656 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD3, 1) }, 657 }; 658 659 static const struct pci_id_table pci_dev_descr_broadwell_table[] = { 660 PCI_ID_TABLE_ENTRY(pci_dev_descr_broadwell), 661 {0,} /* 0 terminated list. */ 662 }; 663 664 /* 665 * pci_device_id table for which devices we are looking for 666 */ 667 static const struct pci_device_id sbridge_pci_tbl[] = { 668 {PCI_DEVICE(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_HA0)}, 669 {PCI_DEVICE(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TA)}, 670 {PCI_DEVICE(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0)}, 671 {PCI_DEVICE(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0)}, 672 {PCI_DEVICE(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_KNL_IMC_SAD0)}, 673 {0,} /* 0 terminated list. */ 674 }; 675 676 677 /**************************************************************************** 678 Ancillary status routines 679 ****************************************************************************/ 680 681 static inline int numrank(enum type type, u32 mtr) 682 { 683 int ranks = (1 << RANK_CNT_BITS(mtr)); 684 int max = 4; 685 686 if (type == HASWELL || type == BROADWELL || type == KNIGHTS_LANDING) 687 max = 8; 688 689 if (ranks > max) { 690 edac_dbg(0, "Invalid number of ranks: %d (max = %i) raw value = %x (%04x)\n", 691 ranks, max, (unsigned int)RANK_CNT_BITS(mtr), mtr); 692 return -EINVAL; 693 } 694 695 return ranks; 696 } 697 698 static inline int numrow(u32 mtr) 699 { 700 int rows = (RANK_WIDTH_BITS(mtr) + 12); 701 702 if (rows < 13 || rows > 18) { 703 edac_dbg(0, "Invalid number of rows: %d (should be between 14 and 17) raw value = %x (%04x)\n", 704 rows, (unsigned int)RANK_WIDTH_BITS(mtr), mtr); 705 return -EINVAL; 706 } 707 708 return 1 << rows; 709 } 710 711 static inline int numcol(u32 mtr) 712 { 713 int cols = (COL_WIDTH_BITS(mtr) + 10); 714 715 if (cols > 12) { 716 edac_dbg(0, "Invalid number of cols: %d (max = 4) raw value = %x (%04x)\n", 717 cols, (unsigned int)COL_WIDTH_BITS(mtr), mtr); 718 return -EINVAL; 719 } 720 721 return 1 << cols; 722 } 723 724 static struct sbridge_dev *get_sbridge_dev(u8 bus, int multi_bus) 725 { 726 struct sbridge_dev *sbridge_dev; 727 728 /* 729 * If we have devices scattered across several busses that pertain 730 * to the same memory controller, we'll lump them all together. 731 */ 732 if (multi_bus) { 733 return list_first_entry_or_null(&sbridge_edac_list, 734 struct sbridge_dev, list); 735 } 736 737 list_for_each_entry(sbridge_dev, &sbridge_edac_list, list) { 738 if (sbridge_dev->bus == bus) 739 return sbridge_dev; 740 } 741 742 return NULL; 743 } 744 745 static struct sbridge_dev *alloc_sbridge_dev(u8 bus, 746 const struct pci_id_table *table) 747 { 748 struct sbridge_dev *sbridge_dev; 749 750 sbridge_dev = kzalloc(sizeof(*sbridge_dev), GFP_KERNEL); 751 if (!sbridge_dev) 752 return NULL; 753 754 sbridge_dev->pdev = kzalloc(sizeof(*sbridge_dev->pdev) * table->n_devs, 755 GFP_KERNEL); 756 if (!sbridge_dev->pdev) { 757 kfree(sbridge_dev); 758 return NULL; 759 } 760 761 sbridge_dev->bus = bus; 762 sbridge_dev->n_devs = table->n_devs; 763 list_add_tail(&sbridge_dev->list, &sbridge_edac_list); 764 765 return sbridge_dev; 766 } 767 768 static void free_sbridge_dev(struct sbridge_dev *sbridge_dev) 769 { 770 list_del(&sbridge_dev->list); 771 kfree(sbridge_dev->pdev); 772 kfree(sbridge_dev); 773 } 774 775 static u64 sbridge_get_tolm(struct sbridge_pvt *pvt) 776 { 777 u32 reg; 778 779 /* Address range is 32:28 */ 780 pci_read_config_dword(pvt->pci_sad1, TOLM, ®); 781 return GET_TOLM(reg); 782 } 783 784 static u64 sbridge_get_tohm(struct sbridge_pvt *pvt) 785 { 786 u32 reg; 787 788 pci_read_config_dword(pvt->pci_sad1, TOHM, ®); 789 return GET_TOHM(reg); 790 } 791 792 static u64 ibridge_get_tolm(struct sbridge_pvt *pvt) 793 { 794 u32 reg; 795 796 pci_read_config_dword(pvt->pci_br1, TOLM, ®); 797 798 return GET_TOLM(reg); 799 } 800 801 static u64 ibridge_get_tohm(struct sbridge_pvt *pvt) 802 { 803 u32 reg; 804 805 pci_read_config_dword(pvt->pci_br1, TOHM, ®); 806 807 return GET_TOHM(reg); 808 } 809 810 static u64 rir_limit(u32 reg) 811 { 812 return ((u64)GET_BITFIELD(reg, 1, 10) << 29) | 0x1fffffff; 813 } 814 815 static u64 sad_limit(u32 reg) 816 { 817 return (GET_BITFIELD(reg, 6, 25) << 26) | 0x3ffffff; 818 } 819 820 static u32 interleave_mode(u32 reg) 821 { 822 return GET_BITFIELD(reg, 1, 1); 823 } 824 825 char *show_interleave_mode(u32 reg) 826 { 827 return interleave_mode(reg) ? "8:6" : "[8:6]XOR[18:16]"; 828 } 829 830 static u32 dram_attr(u32 reg) 831 { 832 return GET_BITFIELD(reg, 2, 3); 833 } 834 835 static u64 knl_sad_limit(u32 reg) 836 { 837 return (GET_BITFIELD(reg, 7, 26) << 26) | 0x3ffffff; 838 } 839 840 static u32 knl_interleave_mode(u32 reg) 841 { 842 return GET_BITFIELD(reg, 1, 2); 843 } 844 845 static char *knl_show_interleave_mode(u32 reg) 846 { 847 char *s; 848 849 switch (knl_interleave_mode(reg)) { 850 case 0: 851 s = "use address bits [8:6]"; 852 break; 853 case 1: 854 s = "use address bits [10:8]"; 855 break; 856 case 2: 857 s = "use address bits [14:12]"; 858 break; 859 case 3: 860 s = "use address bits [32:30]"; 861 break; 862 default: 863 WARN_ON(1); 864 break; 865 } 866 867 return s; 868 } 869 870 static u32 dram_attr_knl(u32 reg) 871 { 872 return GET_BITFIELD(reg, 3, 4); 873 } 874 875 876 static enum mem_type get_memory_type(struct sbridge_pvt *pvt) 877 { 878 u32 reg; 879 enum mem_type mtype; 880 881 if (pvt->pci_ddrio) { 882 pci_read_config_dword(pvt->pci_ddrio, pvt->info.rankcfgr, 883 ®); 884 if (GET_BITFIELD(reg, 11, 11)) 885 /* FIXME: Can also be LRDIMM */ 886 mtype = MEM_RDDR3; 887 else 888 mtype = MEM_DDR3; 889 } else 890 mtype = MEM_UNKNOWN; 891 892 return mtype; 893 } 894 895 static enum mem_type haswell_get_memory_type(struct sbridge_pvt *pvt) 896 { 897 u32 reg; 898 bool registered = false; 899 enum mem_type mtype = MEM_UNKNOWN; 900 901 if (!pvt->pci_ddrio) 902 goto out; 903 904 pci_read_config_dword(pvt->pci_ddrio, 905 HASWELL_DDRCRCLKCONTROLS, ®); 906 /* Is_Rdimm */ 907 if (GET_BITFIELD(reg, 16, 16)) 908 registered = true; 909 910 pci_read_config_dword(pvt->pci_ta, MCMTR, ®); 911 if (GET_BITFIELD(reg, 14, 14)) { 912 if (registered) 913 mtype = MEM_RDDR4; 914 else 915 mtype = MEM_DDR4; 916 } else { 917 if (registered) 918 mtype = MEM_RDDR3; 919 else 920 mtype = MEM_DDR3; 921 } 922 923 out: 924 return mtype; 925 } 926 927 static enum dev_type knl_get_width(struct sbridge_pvt *pvt, u32 mtr) 928 { 929 /* for KNL value is fixed */ 930 return DEV_X16; 931 } 932 933 static enum dev_type sbridge_get_width(struct sbridge_pvt *pvt, u32 mtr) 934 { 935 /* there's no way to figure out */ 936 return DEV_UNKNOWN; 937 } 938 939 static enum dev_type __ibridge_get_width(u32 mtr) 940 { 941 enum dev_type type; 942 943 switch (mtr) { 944 case 3: 945 type = DEV_UNKNOWN; 946 break; 947 case 2: 948 type = DEV_X16; 949 break; 950 case 1: 951 type = DEV_X8; 952 break; 953 case 0: 954 type = DEV_X4; 955 break; 956 } 957 958 return type; 959 } 960 961 static enum dev_type ibridge_get_width(struct sbridge_pvt *pvt, u32 mtr) 962 { 963 /* 964 * ddr3_width on the documentation but also valid for DDR4 on 965 * Haswell 966 */ 967 return __ibridge_get_width(GET_BITFIELD(mtr, 7, 8)); 968 } 969 970 static enum dev_type broadwell_get_width(struct sbridge_pvt *pvt, u32 mtr) 971 { 972 /* ddr3_width on the documentation but also valid for DDR4 */ 973 return __ibridge_get_width(GET_BITFIELD(mtr, 8, 9)); 974 } 975 976 static enum mem_type knl_get_memory_type(struct sbridge_pvt *pvt) 977 { 978 /* DDR4 RDIMMS and LRDIMMS are supported */ 979 return MEM_RDDR4; 980 } 981 982 static u8 get_node_id(struct sbridge_pvt *pvt) 983 { 984 u32 reg; 985 pci_read_config_dword(pvt->pci_br0, SAD_CONTROL, ®); 986 return GET_BITFIELD(reg, 0, 2); 987 } 988 989 static u8 haswell_get_node_id(struct sbridge_pvt *pvt) 990 { 991 u32 reg; 992 993 pci_read_config_dword(pvt->pci_sad1, SAD_CONTROL, ®); 994 return GET_BITFIELD(reg, 0, 3); 995 } 996 997 static u8 knl_get_node_id(struct sbridge_pvt *pvt) 998 { 999 u32 reg; 1000 1001 pci_read_config_dword(pvt->pci_sad1, SAD_CONTROL, ®); 1002 return GET_BITFIELD(reg, 0, 2); 1003 } 1004 1005 1006 static u64 haswell_get_tolm(struct sbridge_pvt *pvt) 1007 { 1008 u32 reg; 1009 1010 pci_read_config_dword(pvt->info.pci_vtd, HASWELL_TOLM, ®); 1011 return (GET_BITFIELD(reg, 26, 31) << 26) | 0x3ffffff; 1012 } 1013 1014 static u64 haswell_get_tohm(struct sbridge_pvt *pvt) 1015 { 1016 u64 rc; 1017 u32 reg; 1018 1019 pci_read_config_dword(pvt->info.pci_vtd, HASWELL_TOHM_0, ®); 1020 rc = GET_BITFIELD(reg, 26, 31); 1021 pci_read_config_dword(pvt->info.pci_vtd, HASWELL_TOHM_1, ®); 1022 rc = ((reg << 6) | rc) << 26; 1023 1024 return rc | 0x1ffffff; 1025 } 1026 1027 static u64 knl_get_tolm(struct sbridge_pvt *pvt) 1028 { 1029 u32 reg; 1030 1031 pci_read_config_dword(pvt->knl.pci_mc_info, KNL_TOLM, ®); 1032 return (GET_BITFIELD(reg, 26, 31) << 26) | 0x3ffffff; 1033 } 1034 1035 static u64 knl_get_tohm(struct sbridge_pvt *pvt) 1036 { 1037 u64 rc; 1038 u32 reg_lo, reg_hi; 1039 1040 pci_read_config_dword(pvt->knl.pci_mc_info, KNL_TOHM_0, ®_lo); 1041 pci_read_config_dword(pvt->knl.pci_mc_info, KNL_TOHM_1, ®_hi); 1042 rc = ((u64)reg_hi << 32) | reg_lo; 1043 return rc | 0x3ffffff; 1044 } 1045 1046 1047 static u64 haswell_rir_limit(u32 reg) 1048 { 1049 return (((u64)GET_BITFIELD(reg, 1, 11) + 1) << 29) - 1; 1050 } 1051 1052 static inline u8 sad_pkg_socket(u8 pkg) 1053 { 1054 /* on Ivy Bridge, nodeID is SASS, where A is HA and S is node id */ 1055 return ((pkg >> 3) << 2) | (pkg & 0x3); 1056 } 1057 1058 static inline u8 sad_pkg_ha(u8 pkg) 1059 { 1060 return (pkg >> 2) & 0x1; 1061 } 1062 1063 /**************************************************************************** 1064 Memory check routines 1065 ****************************************************************************/ 1066 static struct pci_dev *get_pdev_same_bus(u8 bus, u32 id) 1067 { 1068 struct pci_dev *pdev = NULL; 1069 1070 do { 1071 pdev = pci_get_device(PCI_VENDOR_ID_INTEL, id, pdev); 1072 if (pdev && pdev->bus->number == bus) 1073 break; 1074 } while (pdev); 1075 1076 return pdev; 1077 } 1078 1079 /** 1080 * check_if_ecc_is_active() - Checks if ECC is active 1081 * @bus: Device bus 1082 * @type: Memory controller type 1083 * returns: 0 in case ECC is active, -ENODEV if it can't be determined or 1084 * disabled 1085 */ 1086 static int check_if_ecc_is_active(const u8 bus, enum type type) 1087 { 1088 struct pci_dev *pdev = NULL; 1089 u32 mcmtr, id; 1090 1091 switch (type) { 1092 case IVY_BRIDGE: 1093 id = PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TA; 1094 break; 1095 case HASWELL: 1096 id = PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TA; 1097 break; 1098 case SANDY_BRIDGE: 1099 id = PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TA; 1100 break; 1101 case BROADWELL: 1102 id = PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TA; 1103 break; 1104 case KNIGHTS_LANDING: 1105 /* 1106 * KNL doesn't group things by bus the same way 1107 * SB/IB/Haswell does. 1108 */ 1109 id = PCI_DEVICE_ID_INTEL_KNL_IMC_TA; 1110 break; 1111 default: 1112 return -ENODEV; 1113 } 1114 1115 if (type != KNIGHTS_LANDING) 1116 pdev = get_pdev_same_bus(bus, id); 1117 else 1118 pdev = pci_get_device(PCI_VENDOR_ID_INTEL, id, 0); 1119 1120 if (!pdev) { 1121 sbridge_printk(KERN_ERR, "Couldn't find PCI device " 1122 "%04x:%04x! on bus %02d\n", 1123 PCI_VENDOR_ID_INTEL, id, bus); 1124 return -ENODEV; 1125 } 1126 1127 pci_read_config_dword(pdev, 1128 type == KNIGHTS_LANDING ? KNL_MCMTR : MCMTR, &mcmtr); 1129 if (!IS_ECC_ENABLED(mcmtr)) { 1130 sbridge_printk(KERN_ERR, "ECC is disabled. Aborting\n"); 1131 return -ENODEV; 1132 } 1133 return 0; 1134 } 1135 1136 /* Low bits of TAD limit, and some metadata. */ 1137 static const u32 knl_tad_dram_limit_lo[] = { 1138 0x400, 0x500, 0x600, 0x700, 1139 0x800, 0x900, 0xa00, 0xb00, 1140 }; 1141 1142 /* Low bits of TAD offset. */ 1143 static const u32 knl_tad_dram_offset_lo[] = { 1144 0x404, 0x504, 0x604, 0x704, 1145 0x804, 0x904, 0xa04, 0xb04, 1146 }; 1147 1148 /* High 16 bits of TAD limit and offset. */ 1149 static const u32 knl_tad_dram_hi[] = { 1150 0x408, 0x508, 0x608, 0x708, 1151 0x808, 0x908, 0xa08, 0xb08, 1152 }; 1153 1154 /* Number of ways a tad entry is interleaved. */ 1155 static const u32 knl_tad_ways[] = { 1156 8, 6, 4, 3, 2, 1, 1157 }; 1158 1159 /* 1160 * Retrieve the n'th Target Address Decode table entry 1161 * from the memory controller's TAD table. 1162 * 1163 * @pvt: driver private data 1164 * @entry: which entry you want to retrieve 1165 * @mc: which memory controller (0 or 1) 1166 * @offset: output tad range offset 1167 * @limit: output address of first byte above tad range 1168 * @ways: output number of interleave ways 1169 * 1170 * The offset value has curious semantics. It's a sort of running total 1171 * of the sizes of all the memory regions that aren't mapped in this 1172 * tad table. 1173 */ 1174 static int knl_get_tad(const struct sbridge_pvt *pvt, 1175 const int entry, 1176 const int mc, 1177 u64 *offset, 1178 u64 *limit, 1179 int *ways) 1180 { 1181 u32 reg_limit_lo, reg_offset_lo, reg_hi; 1182 struct pci_dev *pci_mc; 1183 int way_id; 1184 1185 switch (mc) { 1186 case 0: 1187 pci_mc = pvt->knl.pci_mc0; 1188 break; 1189 case 1: 1190 pci_mc = pvt->knl.pci_mc1; 1191 break; 1192 default: 1193 WARN_ON(1); 1194 return -EINVAL; 1195 } 1196 1197 pci_read_config_dword(pci_mc, 1198 knl_tad_dram_limit_lo[entry], ®_limit_lo); 1199 pci_read_config_dword(pci_mc, 1200 knl_tad_dram_offset_lo[entry], ®_offset_lo); 1201 pci_read_config_dword(pci_mc, 1202 knl_tad_dram_hi[entry], ®_hi); 1203 1204 /* Is this TAD entry enabled? */ 1205 if (!GET_BITFIELD(reg_limit_lo, 0, 0)) 1206 return -ENODEV; 1207 1208 way_id = GET_BITFIELD(reg_limit_lo, 3, 5); 1209 1210 if (way_id < ARRAY_SIZE(knl_tad_ways)) { 1211 *ways = knl_tad_ways[way_id]; 1212 } else { 1213 *ways = 0; 1214 sbridge_printk(KERN_ERR, 1215 "Unexpected value %d in mc_tad_limit_lo wayness field\n", 1216 way_id); 1217 return -ENODEV; 1218 } 1219 1220 /* 1221 * The least significant 6 bits of base and limit are truncated. 1222 * For limit, we fill the missing bits with 1s. 1223 */ 1224 *offset = ((u64) GET_BITFIELD(reg_offset_lo, 6, 31) << 6) | 1225 ((u64) GET_BITFIELD(reg_hi, 0, 15) << 32); 1226 *limit = ((u64) GET_BITFIELD(reg_limit_lo, 6, 31) << 6) | 63 | 1227 ((u64) GET_BITFIELD(reg_hi, 16, 31) << 32); 1228 1229 return 0; 1230 } 1231 1232 /* Determine which memory controller is responsible for a given channel. */ 1233 static int knl_channel_mc(int channel) 1234 { 1235 WARN_ON(channel < 0 || channel >= 6); 1236 1237 return channel < 3 ? 1 : 0; 1238 } 1239 1240 /* 1241 * Get the Nth entry from EDC_ROUTE_TABLE register. 1242 * (This is the per-tile mapping of logical interleave targets to 1243 * physical EDC modules.) 1244 * 1245 * entry 0: 0:2 1246 * 1: 3:5 1247 * 2: 6:8 1248 * 3: 9:11 1249 * 4: 12:14 1250 * 5: 15:17 1251 * 6: 18:20 1252 * 7: 21:23 1253 * reserved: 24:31 1254 */ 1255 static u32 knl_get_edc_route(int entry, u32 reg) 1256 { 1257 WARN_ON(entry >= KNL_MAX_EDCS); 1258 return GET_BITFIELD(reg, entry*3, (entry*3)+2); 1259 } 1260 1261 /* 1262 * Get the Nth entry from MC_ROUTE_TABLE register. 1263 * (This is the per-tile mapping of logical interleave targets to 1264 * physical DRAM channels modules.) 1265 * 1266 * entry 0: mc 0:2 channel 18:19 1267 * 1: mc 3:5 channel 20:21 1268 * 2: mc 6:8 channel 22:23 1269 * 3: mc 9:11 channel 24:25 1270 * 4: mc 12:14 channel 26:27 1271 * 5: mc 15:17 channel 28:29 1272 * reserved: 30:31 1273 * 1274 * Though we have 3 bits to identify the MC, we should only see 1275 * the values 0 or 1. 1276 */ 1277 1278 static u32 knl_get_mc_route(int entry, u32 reg) 1279 { 1280 int mc, chan; 1281 1282 WARN_ON(entry >= KNL_MAX_CHANNELS); 1283 1284 mc = GET_BITFIELD(reg, entry*3, (entry*3)+2); 1285 chan = GET_BITFIELD(reg, (entry*2) + 18, (entry*2) + 18 + 1); 1286 1287 return knl_channel_remap(mc*3 + chan); 1288 } 1289 1290 /* 1291 * Render the EDC_ROUTE register in human-readable form. 1292 * Output string s should be at least KNL_MAX_EDCS*2 bytes. 1293 */ 1294 static void knl_show_edc_route(u32 reg, char *s) 1295 { 1296 int i; 1297 1298 for (i = 0; i < KNL_MAX_EDCS; i++) { 1299 s[i*2] = knl_get_edc_route(i, reg) + '0'; 1300 s[i*2+1] = '-'; 1301 } 1302 1303 s[KNL_MAX_EDCS*2 - 1] = '\0'; 1304 } 1305 1306 /* 1307 * Render the MC_ROUTE register in human-readable form. 1308 * Output string s should be at least KNL_MAX_CHANNELS*2 bytes. 1309 */ 1310 static void knl_show_mc_route(u32 reg, char *s) 1311 { 1312 int i; 1313 1314 for (i = 0; i < KNL_MAX_CHANNELS; i++) { 1315 s[i*2] = knl_get_mc_route(i, reg) + '0'; 1316 s[i*2+1] = '-'; 1317 } 1318 1319 s[KNL_MAX_CHANNELS*2 - 1] = '\0'; 1320 } 1321 1322 #define KNL_EDC_ROUTE 0xb8 1323 #define KNL_MC_ROUTE 0xb4 1324 1325 /* Is this dram rule backed by regular DRAM in flat mode? */ 1326 #define KNL_EDRAM(reg) GET_BITFIELD(reg, 29, 29) 1327 1328 /* Is this dram rule cached? */ 1329 #define KNL_CACHEABLE(reg) GET_BITFIELD(reg, 28, 28) 1330 1331 /* Is this rule backed by edc ? */ 1332 #define KNL_EDRAM_ONLY(reg) GET_BITFIELD(reg, 29, 29) 1333 1334 /* Is this rule backed by DRAM, cacheable in EDRAM? */ 1335 #define KNL_CACHEABLE(reg) GET_BITFIELD(reg, 28, 28) 1336 1337 /* Is this rule mod3? */ 1338 #define KNL_MOD3(reg) GET_BITFIELD(reg, 27, 27) 1339 1340 /* 1341 * Figure out how big our RAM modules are. 1342 * 1343 * The DIMMMTR register in KNL doesn't tell us the size of the DIMMs, so we 1344 * have to figure this out from the SAD rules, interleave lists, route tables, 1345 * and TAD rules. 1346 * 1347 * SAD rules can have holes in them (e.g. the 3G-4G hole), so we have to 1348 * inspect the TAD rules to figure out how large the SAD regions really are. 1349 * 1350 * When we know the real size of a SAD region and how many ways it's 1351 * interleaved, we know the individual contribution of each channel to 1352 * TAD is size/ways. 1353 * 1354 * Finally, we have to check whether each channel participates in each SAD 1355 * region. 1356 * 1357 * Fortunately, KNL only supports one DIMM per channel, so once we know how 1358 * much memory the channel uses, we know the DIMM is at least that large. 1359 * (The BIOS might possibly choose not to map all available memory, in which 1360 * case we will underreport the size of the DIMM.) 1361 * 1362 * In theory, we could try to determine the EDC sizes as well, but that would 1363 * only work in flat mode, not in cache mode. 1364 * 1365 * @mc_sizes: Output sizes of channels (must have space for KNL_MAX_CHANNELS 1366 * elements) 1367 */ 1368 static int knl_get_dimm_capacity(struct sbridge_pvt *pvt, u64 *mc_sizes) 1369 { 1370 u64 sad_base, sad_size, sad_limit = 0; 1371 u64 tad_base, tad_size, tad_limit, tad_deadspace, tad_livespace; 1372 int sad_rule = 0; 1373 int tad_rule = 0; 1374 int intrlv_ways, tad_ways; 1375 u32 first_pkg, pkg; 1376 int i; 1377 u64 sad_actual_size[2]; /* sad size accounting for holes, per mc */ 1378 u32 dram_rule, interleave_reg; 1379 u32 mc_route_reg[KNL_MAX_CHAS]; 1380 u32 edc_route_reg[KNL_MAX_CHAS]; 1381 int edram_only; 1382 char edc_route_string[KNL_MAX_EDCS*2]; 1383 char mc_route_string[KNL_MAX_CHANNELS*2]; 1384 int cur_reg_start; 1385 int mc; 1386 int channel; 1387 int way; 1388 int participants[KNL_MAX_CHANNELS]; 1389 int participant_count = 0; 1390 1391 for (i = 0; i < KNL_MAX_CHANNELS; i++) 1392 mc_sizes[i] = 0; 1393 1394 /* Read the EDC route table in each CHA. */ 1395 cur_reg_start = 0; 1396 for (i = 0; i < KNL_MAX_CHAS; i++) { 1397 pci_read_config_dword(pvt->knl.pci_cha[i], 1398 KNL_EDC_ROUTE, &edc_route_reg[i]); 1399 1400 if (i > 0 && edc_route_reg[i] != edc_route_reg[i-1]) { 1401 knl_show_edc_route(edc_route_reg[i-1], 1402 edc_route_string); 1403 if (cur_reg_start == i-1) 1404 edac_dbg(0, "edc route table for CHA %d: %s\n", 1405 cur_reg_start, edc_route_string); 1406 else 1407 edac_dbg(0, "edc route table for CHA %d-%d: %s\n", 1408 cur_reg_start, i-1, edc_route_string); 1409 cur_reg_start = i; 1410 } 1411 } 1412 knl_show_edc_route(edc_route_reg[i-1], edc_route_string); 1413 if (cur_reg_start == i-1) 1414 edac_dbg(0, "edc route table for CHA %d: %s\n", 1415 cur_reg_start, edc_route_string); 1416 else 1417 edac_dbg(0, "edc route table for CHA %d-%d: %s\n", 1418 cur_reg_start, i-1, edc_route_string); 1419 1420 /* Read the MC route table in each CHA. */ 1421 cur_reg_start = 0; 1422 for (i = 0; i < KNL_MAX_CHAS; i++) { 1423 pci_read_config_dword(pvt->knl.pci_cha[i], 1424 KNL_MC_ROUTE, &mc_route_reg[i]); 1425 1426 if (i > 0 && mc_route_reg[i] != mc_route_reg[i-1]) { 1427 knl_show_mc_route(mc_route_reg[i-1], mc_route_string); 1428 if (cur_reg_start == i-1) 1429 edac_dbg(0, "mc route table for CHA %d: %s\n", 1430 cur_reg_start, mc_route_string); 1431 else 1432 edac_dbg(0, "mc route table for CHA %d-%d: %s\n", 1433 cur_reg_start, i-1, mc_route_string); 1434 cur_reg_start = i; 1435 } 1436 } 1437 knl_show_mc_route(mc_route_reg[i-1], mc_route_string); 1438 if (cur_reg_start == i-1) 1439 edac_dbg(0, "mc route table for CHA %d: %s\n", 1440 cur_reg_start, mc_route_string); 1441 else 1442 edac_dbg(0, "mc route table for CHA %d-%d: %s\n", 1443 cur_reg_start, i-1, mc_route_string); 1444 1445 /* Process DRAM rules */ 1446 for (sad_rule = 0; sad_rule < pvt->info.max_sad; sad_rule++) { 1447 /* previous limit becomes the new base */ 1448 sad_base = sad_limit; 1449 1450 pci_read_config_dword(pvt->pci_sad0, 1451 pvt->info.dram_rule[sad_rule], &dram_rule); 1452 1453 if (!DRAM_RULE_ENABLE(dram_rule)) 1454 break; 1455 1456 edram_only = KNL_EDRAM_ONLY(dram_rule); 1457 1458 sad_limit = pvt->info.sad_limit(dram_rule)+1; 1459 sad_size = sad_limit - sad_base; 1460 1461 pci_read_config_dword(pvt->pci_sad0, 1462 pvt->info.interleave_list[sad_rule], &interleave_reg); 1463 1464 /* 1465 * Find out how many ways this dram rule is interleaved. 1466 * We stop when we see the first channel again. 1467 */ 1468 first_pkg = sad_pkg(pvt->info.interleave_pkg, 1469 interleave_reg, 0); 1470 for (intrlv_ways = 1; intrlv_ways < 8; intrlv_ways++) { 1471 pkg = sad_pkg(pvt->info.interleave_pkg, 1472 interleave_reg, intrlv_ways); 1473 1474 if ((pkg & 0x8) == 0) { 1475 /* 1476 * 0 bit means memory is non-local, 1477 * which KNL doesn't support 1478 */ 1479 edac_dbg(0, "Unexpected interleave target %d\n", 1480 pkg); 1481 return -1; 1482 } 1483 1484 if (pkg == first_pkg) 1485 break; 1486 } 1487 if (KNL_MOD3(dram_rule)) 1488 intrlv_ways *= 3; 1489 1490 edac_dbg(3, "dram rule %d (base 0x%llx, limit 0x%llx), %d way interleave%s\n", 1491 sad_rule, 1492 sad_base, 1493 sad_limit, 1494 intrlv_ways, 1495 edram_only ? ", EDRAM" : ""); 1496 1497 /* 1498 * Find out how big the SAD region really is by iterating 1499 * over TAD tables (SAD regions may contain holes). 1500 * Each memory controller might have a different TAD table, so 1501 * we have to look at both. 1502 * 1503 * Livespace is the memory that's mapped in this TAD table, 1504 * deadspace is the holes (this could be the MMIO hole, or it 1505 * could be memory that's mapped by the other TAD table but 1506 * not this one). 1507 */ 1508 for (mc = 0; mc < 2; mc++) { 1509 sad_actual_size[mc] = 0; 1510 tad_livespace = 0; 1511 for (tad_rule = 0; 1512 tad_rule < ARRAY_SIZE( 1513 knl_tad_dram_limit_lo); 1514 tad_rule++) { 1515 if (knl_get_tad(pvt, 1516 tad_rule, 1517 mc, 1518 &tad_deadspace, 1519 &tad_limit, 1520 &tad_ways)) 1521 break; 1522 1523 tad_size = (tad_limit+1) - 1524 (tad_livespace + tad_deadspace); 1525 tad_livespace += tad_size; 1526 tad_base = (tad_limit+1) - tad_size; 1527 1528 if (tad_base < sad_base) { 1529 if (tad_limit > sad_base) 1530 edac_dbg(0, "TAD region overlaps lower SAD boundary -- TAD tables may be configured incorrectly.\n"); 1531 } else if (tad_base < sad_limit) { 1532 if (tad_limit+1 > sad_limit) { 1533 edac_dbg(0, "TAD region overlaps upper SAD boundary -- TAD tables may be configured incorrectly.\n"); 1534 } else { 1535 /* TAD region is completely inside SAD region */ 1536 edac_dbg(3, "TAD region %d 0x%llx - 0x%llx (%lld bytes) table%d\n", 1537 tad_rule, tad_base, 1538 tad_limit, tad_size, 1539 mc); 1540 sad_actual_size[mc] += tad_size; 1541 } 1542 } 1543 tad_base = tad_limit+1; 1544 } 1545 } 1546 1547 for (mc = 0; mc < 2; mc++) { 1548 edac_dbg(3, " total TAD DRAM footprint in table%d : 0x%llx (%lld bytes)\n", 1549 mc, sad_actual_size[mc], sad_actual_size[mc]); 1550 } 1551 1552 /* Ignore EDRAM rule */ 1553 if (edram_only) 1554 continue; 1555 1556 /* Figure out which channels participate in interleave. */ 1557 for (channel = 0; channel < KNL_MAX_CHANNELS; channel++) 1558 participants[channel] = 0; 1559 1560 /* For each channel, does at least one CHA have 1561 * this channel mapped to the given target? 1562 */ 1563 for (channel = 0; channel < KNL_MAX_CHANNELS; channel++) { 1564 for (way = 0; way < intrlv_ways; way++) { 1565 int target; 1566 int cha; 1567 1568 if (KNL_MOD3(dram_rule)) 1569 target = way; 1570 else 1571 target = 0x7 & sad_pkg( 1572 pvt->info.interleave_pkg, interleave_reg, way); 1573 1574 for (cha = 0; cha < KNL_MAX_CHAS; cha++) { 1575 if (knl_get_mc_route(target, 1576 mc_route_reg[cha]) == channel 1577 && !participants[channel]) { 1578 participant_count++; 1579 participants[channel] = 1; 1580 break; 1581 } 1582 } 1583 } 1584 } 1585 1586 if (participant_count != intrlv_ways) 1587 edac_dbg(0, "participant_count (%d) != interleave_ways (%d): DIMM size may be incorrect\n", 1588 participant_count, intrlv_ways); 1589 1590 for (channel = 0; channel < KNL_MAX_CHANNELS; channel++) { 1591 mc = knl_channel_mc(channel); 1592 if (participants[channel]) { 1593 edac_dbg(4, "mc channel %d contributes %lld bytes via sad entry %d\n", 1594 channel, 1595 sad_actual_size[mc]/intrlv_ways, 1596 sad_rule); 1597 mc_sizes[channel] += 1598 sad_actual_size[mc]/intrlv_ways; 1599 } 1600 } 1601 } 1602 1603 return 0; 1604 } 1605 1606 static int get_dimm_config(struct mem_ctl_info *mci) 1607 { 1608 struct sbridge_pvt *pvt = mci->pvt_info; 1609 struct dimm_info *dimm; 1610 unsigned i, j, banks, ranks, rows, cols, npages; 1611 u64 size; 1612 u32 reg; 1613 enum edac_type mode; 1614 enum mem_type mtype; 1615 int channels = pvt->info.type == KNIGHTS_LANDING ? 1616 KNL_MAX_CHANNELS : NUM_CHANNELS; 1617 u64 knl_mc_sizes[KNL_MAX_CHANNELS]; 1618 1619 if (pvt->info.type == HASWELL || pvt->info.type == BROADWELL || 1620 pvt->info.type == KNIGHTS_LANDING) 1621 pci_read_config_dword(pvt->pci_sad1, SAD_TARGET, ®); 1622 else 1623 pci_read_config_dword(pvt->pci_br0, SAD_TARGET, ®); 1624 1625 if (pvt->info.type == KNIGHTS_LANDING) 1626 pvt->sbridge_dev->source_id = SOURCE_ID_KNL(reg); 1627 else 1628 pvt->sbridge_dev->source_id = SOURCE_ID(reg); 1629 1630 pvt->sbridge_dev->node_id = pvt->info.get_node_id(pvt); 1631 edac_dbg(0, "mc#%d: Node ID: %d, source ID: %d\n", 1632 pvt->sbridge_dev->mc, 1633 pvt->sbridge_dev->node_id, 1634 pvt->sbridge_dev->source_id); 1635 1636 /* KNL doesn't support mirroring or lockstep, 1637 * and is always closed page 1638 */ 1639 if (pvt->info.type == KNIGHTS_LANDING) { 1640 mode = EDAC_S4ECD4ED; 1641 pvt->is_mirrored = false; 1642 1643 if (knl_get_dimm_capacity(pvt, knl_mc_sizes) != 0) 1644 return -1; 1645 } else { 1646 pci_read_config_dword(pvt->pci_ras, RASENABLES, ®); 1647 if (IS_MIRROR_ENABLED(reg)) { 1648 edac_dbg(0, "Memory mirror is enabled\n"); 1649 pvt->is_mirrored = true; 1650 } else { 1651 edac_dbg(0, "Memory mirror is disabled\n"); 1652 pvt->is_mirrored = false; 1653 } 1654 1655 pci_read_config_dword(pvt->pci_ta, MCMTR, &pvt->info.mcmtr); 1656 if (IS_LOCKSTEP_ENABLED(pvt->info.mcmtr)) { 1657 edac_dbg(0, "Lockstep is enabled\n"); 1658 mode = EDAC_S8ECD8ED; 1659 pvt->is_lockstep = true; 1660 } else { 1661 edac_dbg(0, "Lockstep is disabled\n"); 1662 mode = EDAC_S4ECD4ED; 1663 pvt->is_lockstep = false; 1664 } 1665 if (IS_CLOSE_PG(pvt->info.mcmtr)) { 1666 edac_dbg(0, "address map is on closed page mode\n"); 1667 pvt->is_close_pg = true; 1668 } else { 1669 edac_dbg(0, "address map is on open page mode\n"); 1670 pvt->is_close_pg = false; 1671 } 1672 } 1673 1674 mtype = pvt->info.get_memory_type(pvt); 1675 if (mtype == MEM_RDDR3 || mtype == MEM_RDDR4) 1676 edac_dbg(0, "Memory is registered\n"); 1677 else if (mtype == MEM_UNKNOWN) 1678 edac_dbg(0, "Cannot determine memory type\n"); 1679 else 1680 edac_dbg(0, "Memory is unregistered\n"); 1681 1682 if (mtype == MEM_DDR4 || mtype == MEM_RDDR4) 1683 banks = 16; 1684 else 1685 banks = 8; 1686 1687 for (i = 0; i < channels; i++) { 1688 u32 mtr; 1689 1690 int max_dimms_per_channel; 1691 1692 if (pvt->info.type == KNIGHTS_LANDING) { 1693 max_dimms_per_channel = 1; 1694 if (!pvt->knl.pci_channel[i]) 1695 continue; 1696 } else { 1697 max_dimms_per_channel = ARRAY_SIZE(mtr_regs); 1698 if (!pvt->pci_tad[i]) 1699 continue; 1700 } 1701 1702 for (j = 0; j < max_dimms_per_channel; j++) { 1703 dimm = EDAC_DIMM_PTR(mci->layers, mci->dimms, mci->n_layers, 1704 i, j, 0); 1705 if (pvt->info.type == KNIGHTS_LANDING) { 1706 pci_read_config_dword(pvt->knl.pci_channel[i], 1707 knl_mtr_reg, &mtr); 1708 } else { 1709 pci_read_config_dword(pvt->pci_tad[i], 1710 mtr_regs[j], &mtr); 1711 } 1712 edac_dbg(4, "Channel #%d MTR%d = %x\n", i, j, mtr); 1713 if (IS_DIMM_PRESENT(mtr)) { 1714 pvt->channel[i].dimms++; 1715 1716 ranks = numrank(pvt->info.type, mtr); 1717 1718 if (pvt->info.type == KNIGHTS_LANDING) { 1719 /* For DDR4, this is fixed. */ 1720 cols = 1 << 10; 1721 rows = knl_mc_sizes[i] / 1722 ((u64) cols * ranks * banks * 8); 1723 } else { 1724 rows = numrow(mtr); 1725 cols = numcol(mtr); 1726 } 1727 1728 size = ((u64)rows * cols * banks * ranks) >> (20 - 3); 1729 npages = MiB_TO_PAGES(size); 1730 1731 edac_dbg(0, "mc#%d: ha %d channel %d, dimm %d, %lld Mb (%d pages) bank: %d, rank: %d, row: %#x, col: %#x\n", 1732 pvt->sbridge_dev->mc, i/4, i%4, j, 1733 size, npages, 1734 banks, ranks, rows, cols); 1735 1736 dimm->nr_pages = npages; 1737 dimm->grain = 32; 1738 dimm->dtype = pvt->info.get_width(pvt, mtr); 1739 dimm->mtype = mtype; 1740 dimm->edac_mode = mode; 1741 snprintf(dimm->label, sizeof(dimm->label), 1742 "CPU_SrcID#%u_Ha#%u_Chan#%u_DIMM#%u", 1743 pvt->sbridge_dev->source_id, i/4, i%4, j); 1744 } 1745 } 1746 } 1747 1748 return 0; 1749 } 1750 1751 static void get_memory_layout(const struct mem_ctl_info *mci) 1752 { 1753 struct sbridge_pvt *pvt = mci->pvt_info; 1754 int i, j, k, n_sads, n_tads, sad_interl; 1755 u32 reg; 1756 u64 limit, prv = 0; 1757 u64 tmp_mb; 1758 u32 gb, mb; 1759 u32 rir_way; 1760 1761 /* 1762 * Step 1) Get TOLM/TOHM ranges 1763 */ 1764 1765 pvt->tolm = pvt->info.get_tolm(pvt); 1766 tmp_mb = (1 + pvt->tolm) >> 20; 1767 1768 gb = div_u64_rem(tmp_mb, 1024, &mb); 1769 edac_dbg(0, "TOLM: %u.%03u GB (0x%016Lx)\n", 1770 gb, (mb*1000)/1024, (u64)pvt->tolm); 1771 1772 /* Address range is already 45:25 */ 1773 pvt->tohm = pvt->info.get_tohm(pvt); 1774 tmp_mb = (1 + pvt->tohm) >> 20; 1775 1776 gb = div_u64_rem(tmp_mb, 1024, &mb); 1777 edac_dbg(0, "TOHM: %u.%03u GB (0x%016Lx)\n", 1778 gb, (mb*1000)/1024, (u64)pvt->tohm); 1779 1780 /* 1781 * Step 2) Get SAD range and SAD Interleave list 1782 * TAD registers contain the interleave wayness. However, it 1783 * seems simpler to just discover it indirectly, with the 1784 * algorithm bellow. 1785 */ 1786 prv = 0; 1787 for (n_sads = 0; n_sads < pvt->info.max_sad; n_sads++) { 1788 /* SAD_LIMIT Address range is 45:26 */ 1789 pci_read_config_dword(pvt->pci_sad0, pvt->info.dram_rule[n_sads], 1790 ®); 1791 limit = pvt->info.sad_limit(reg); 1792 1793 if (!DRAM_RULE_ENABLE(reg)) 1794 continue; 1795 1796 if (limit <= prv) 1797 break; 1798 1799 tmp_mb = (limit + 1) >> 20; 1800 gb = div_u64_rem(tmp_mb, 1024, &mb); 1801 edac_dbg(0, "SAD#%d %s up to %u.%03u GB (0x%016Lx) Interleave: %s reg=0x%08x\n", 1802 n_sads, 1803 show_dram_attr(pvt->info.dram_attr(reg)), 1804 gb, (mb*1000)/1024, 1805 ((u64)tmp_mb) << 20L, 1806 pvt->info.show_interleave_mode(reg), 1807 reg); 1808 prv = limit; 1809 1810 pci_read_config_dword(pvt->pci_sad0, pvt->info.interleave_list[n_sads], 1811 ®); 1812 sad_interl = sad_pkg(pvt->info.interleave_pkg, reg, 0); 1813 for (j = 0; j < 8; j++) { 1814 u32 pkg = sad_pkg(pvt->info.interleave_pkg, reg, j); 1815 if (j > 0 && sad_interl == pkg) 1816 break; 1817 1818 edac_dbg(0, "SAD#%d, interleave #%d: %d\n", 1819 n_sads, j, pkg); 1820 } 1821 } 1822 1823 if (pvt->info.type == KNIGHTS_LANDING) 1824 return; 1825 1826 /* 1827 * Step 3) Get TAD range 1828 */ 1829 prv = 0; 1830 for (n_tads = 0; n_tads < MAX_TAD; n_tads++) { 1831 pci_read_config_dword(pvt->pci_ha0, tad_dram_rule[n_tads], 1832 ®); 1833 limit = TAD_LIMIT(reg); 1834 if (limit <= prv) 1835 break; 1836 tmp_mb = (limit + 1) >> 20; 1837 1838 gb = div_u64_rem(tmp_mb, 1024, &mb); 1839 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", 1840 n_tads, gb, (mb*1000)/1024, 1841 ((u64)tmp_mb) << 20L, 1842 (u32)(1 << TAD_SOCK(reg)), 1843 (u32)TAD_CH(reg) + 1, 1844 (u32)TAD_TGT0(reg), 1845 (u32)TAD_TGT1(reg), 1846 (u32)TAD_TGT2(reg), 1847 (u32)TAD_TGT3(reg), 1848 reg); 1849 prv = limit; 1850 } 1851 1852 /* 1853 * Step 4) Get TAD offsets, per each channel 1854 */ 1855 for (i = 0; i < NUM_CHANNELS; i++) { 1856 if (!pvt->channel[i].dimms) 1857 continue; 1858 for (j = 0; j < n_tads; j++) { 1859 pci_read_config_dword(pvt->pci_tad[i], 1860 tad_ch_nilv_offset[j], 1861 ®); 1862 tmp_mb = TAD_OFFSET(reg) >> 20; 1863 gb = div_u64_rem(tmp_mb, 1024, &mb); 1864 edac_dbg(0, "TAD CH#%d, offset #%d: %u.%03u GB (0x%016Lx), reg=0x%08x\n", 1865 i, j, 1866 gb, (mb*1000)/1024, 1867 ((u64)tmp_mb) << 20L, 1868 reg); 1869 } 1870 } 1871 1872 /* 1873 * Step 6) Get RIR Wayness/Limit, per each channel 1874 */ 1875 for (i = 0; i < NUM_CHANNELS; i++) { 1876 if (!pvt->channel[i].dimms) 1877 continue; 1878 for (j = 0; j < MAX_RIR_RANGES; j++) { 1879 pci_read_config_dword(pvt->pci_tad[i], 1880 rir_way_limit[j], 1881 ®); 1882 1883 if (!IS_RIR_VALID(reg)) 1884 continue; 1885 1886 tmp_mb = pvt->info.rir_limit(reg) >> 20; 1887 rir_way = 1 << RIR_WAY(reg); 1888 gb = div_u64_rem(tmp_mb, 1024, &mb); 1889 edac_dbg(0, "CH#%d RIR#%d, limit: %u.%03u GB (0x%016Lx), way: %d, reg=0x%08x\n", 1890 i, j, 1891 gb, (mb*1000)/1024, 1892 ((u64)tmp_mb) << 20L, 1893 rir_way, 1894 reg); 1895 1896 for (k = 0; k < rir_way; k++) { 1897 pci_read_config_dword(pvt->pci_tad[i], 1898 rir_offset[j][k], 1899 ®); 1900 tmp_mb = RIR_OFFSET(reg) << 6; 1901 1902 gb = div_u64_rem(tmp_mb, 1024, &mb); 1903 edac_dbg(0, "CH#%d RIR#%d INTL#%d, offset %u.%03u GB (0x%016Lx), tgt: %d, reg=0x%08x\n", 1904 i, j, k, 1905 gb, (mb*1000)/1024, 1906 ((u64)tmp_mb) << 20L, 1907 (u32)RIR_RNK_TGT(reg), 1908 reg); 1909 } 1910 } 1911 } 1912 } 1913 1914 static struct mem_ctl_info *get_mci_for_node_id(u8 node_id) 1915 { 1916 struct sbridge_dev *sbridge_dev; 1917 1918 list_for_each_entry(sbridge_dev, &sbridge_edac_list, list) { 1919 if (sbridge_dev->node_id == node_id) 1920 return sbridge_dev->mci; 1921 } 1922 return NULL; 1923 } 1924 1925 static int get_memory_error_data(struct mem_ctl_info *mci, 1926 u64 addr, 1927 u8 *socket, u8 *ha, 1928 long *channel_mask, 1929 u8 *rank, 1930 char **area_type, char *msg) 1931 { 1932 struct mem_ctl_info *new_mci; 1933 struct sbridge_pvt *pvt = mci->pvt_info; 1934 struct pci_dev *pci_ha; 1935 int n_rir, n_sads, n_tads, sad_way, sck_xch; 1936 int sad_interl, idx, base_ch; 1937 int interleave_mode, shiftup = 0; 1938 unsigned sad_interleave[pvt->info.max_interleave]; 1939 u32 reg, dram_rule; 1940 u8 ch_way, sck_way, pkg, sad_ha = 0, ch_add = 0; 1941 u32 tad_offset; 1942 u32 rir_way; 1943 u32 mb, gb; 1944 u64 ch_addr, offset, limit = 0, prv = 0; 1945 1946 1947 /* 1948 * Step 0) Check if the address is at special memory ranges 1949 * The check bellow is probably enough to fill all cases where 1950 * the error is not inside a memory, except for the legacy 1951 * range (e. g. VGA addresses). It is unlikely, however, that the 1952 * memory controller would generate an error on that range. 1953 */ 1954 if ((addr > (u64) pvt->tolm) && (addr < (1LL << 32))) { 1955 sprintf(msg, "Error at TOLM area, on addr 0x%08Lx", addr); 1956 return -EINVAL; 1957 } 1958 if (addr >= (u64)pvt->tohm) { 1959 sprintf(msg, "Error at MMIOH area, on addr 0x%016Lx", addr); 1960 return -EINVAL; 1961 } 1962 1963 /* 1964 * Step 1) Get socket 1965 */ 1966 for (n_sads = 0; n_sads < pvt->info.max_sad; n_sads++) { 1967 pci_read_config_dword(pvt->pci_sad0, pvt->info.dram_rule[n_sads], 1968 ®); 1969 1970 if (!DRAM_RULE_ENABLE(reg)) 1971 continue; 1972 1973 limit = pvt->info.sad_limit(reg); 1974 if (limit <= prv) { 1975 sprintf(msg, "Can't discover the memory socket"); 1976 return -EINVAL; 1977 } 1978 if (addr <= limit) 1979 break; 1980 prv = limit; 1981 } 1982 if (n_sads == pvt->info.max_sad) { 1983 sprintf(msg, "Can't discover the memory socket"); 1984 return -EINVAL; 1985 } 1986 dram_rule = reg; 1987 *area_type = show_dram_attr(pvt->info.dram_attr(dram_rule)); 1988 interleave_mode = pvt->info.interleave_mode(dram_rule); 1989 1990 pci_read_config_dword(pvt->pci_sad0, pvt->info.interleave_list[n_sads], 1991 ®); 1992 1993 if (pvt->info.type == SANDY_BRIDGE) { 1994 sad_interl = sad_pkg(pvt->info.interleave_pkg, reg, 0); 1995 for (sad_way = 0; sad_way < 8; sad_way++) { 1996 u32 pkg = sad_pkg(pvt->info.interleave_pkg, reg, sad_way); 1997 if (sad_way > 0 && sad_interl == pkg) 1998 break; 1999 sad_interleave[sad_way] = pkg; 2000 edac_dbg(0, "SAD interleave #%d: %d\n", 2001 sad_way, sad_interleave[sad_way]); 2002 } 2003 edac_dbg(0, "mc#%d: Error detected on SAD#%d: address 0x%016Lx < 0x%016Lx, Interleave [%d:6]%s\n", 2004 pvt->sbridge_dev->mc, 2005 n_sads, 2006 addr, 2007 limit, 2008 sad_way + 7, 2009 !interleave_mode ? "" : "XOR[18:16]"); 2010 if (interleave_mode) 2011 idx = ((addr >> 6) ^ (addr >> 16)) & 7; 2012 else 2013 idx = (addr >> 6) & 7; 2014 switch (sad_way) { 2015 case 1: 2016 idx = 0; 2017 break; 2018 case 2: 2019 idx = idx & 1; 2020 break; 2021 case 4: 2022 idx = idx & 3; 2023 break; 2024 case 8: 2025 break; 2026 default: 2027 sprintf(msg, "Can't discover socket interleave"); 2028 return -EINVAL; 2029 } 2030 *socket = sad_interleave[idx]; 2031 edac_dbg(0, "SAD interleave index: %d (wayness %d) = CPU socket %d\n", 2032 idx, sad_way, *socket); 2033 } else if (pvt->info.type == HASWELL || pvt->info.type == BROADWELL) { 2034 int bits, a7mode = A7MODE(dram_rule); 2035 2036 if (a7mode) { 2037 /* A7 mode swaps P9 with P6 */ 2038 bits = GET_BITFIELD(addr, 7, 8) << 1; 2039 bits |= GET_BITFIELD(addr, 9, 9); 2040 } else 2041 bits = GET_BITFIELD(addr, 6, 8); 2042 2043 if (interleave_mode == 0) { 2044 /* interleave mode will XOR {8,7,6} with {18,17,16} */ 2045 idx = GET_BITFIELD(addr, 16, 18); 2046 idx ^= bits; 2047 } else 2048 idx = bits; 2049 2050 pkg = sad_pkg(pvt->info.interleave_pkg, reg, idx); 2051 *socket = sad_pkg_socket(pkg); 2052 sad_ha = sad_pkg_ha(pkg); 2053 if (sad_ha) 2054 ch_add = 4; 2055 2056 if (a7mode) { 2057 /* MCChanShiftUpEnable */ 2058 pci_read_config_dword(pvt->pci_ha0, 2059 HASWELL_HASYSDEFEATURE2, ®); 2060 shiftup = GET_BITFIELD(reg, 22, 22); 2061 } 2062 2063 edac_dbg(0, "SAD interleave package: %d = CPU socket %d, HA %i, shiftup: %i\n", 2064 idx, *socket, sad_ha, shiftup); 2065 } else { 2066 /* Ivy Bridge's SAD mode doesn't support XOR interleave mode */ 2067 idx = (addr >> 6) & 7; 2068 pkg = sad_pkg(pvt->info.interleave_pkg, reg, idx); 2069 *socket = sad_pkg_socket(pkg); 2070 sad_ha = sad_pkg_ha(pkg); 2071 if (sad_ha) 2072 ch_add = 4; 2073 edac_dbg(0, "SAD interleave package: %d = CPU socket %d, HA %d\n", 2074 idx, *socket, sad_ha); 2075 } 2076 2077 *ha = sad_ha; 2078 2079 /* 2080 * Move to the proper node structure, in order to access the 2081 * right PCI registers 2082 */ 2083 new_mci = get_mci_for_node_id(*socket); 2084 if (!new_mci) { 2085 sprintf(msg, "Struct for socket #%u wasn't initialized", 2086 *socket); 2087 return -EINVAL; 2088 } 2089 mci = new_mci; 2090 pvt = mci->pvt_info; 2091 2092 /* 2093 * Step 2) Get memory channel 2094 */ 2095 prv = 0; 2096 if (pvt->info.type == SANDY_BRIDGE) 2097 pci_ha = pvt->pci_ha0; 2098 else { 2099 if (sad_ha) 2100 pci_ha = pvt->pci_ha1; 2101 else 2102 pci_ha = pvt->pci_ha0; 2103 } 2104 for (n_tads = 0; n_tads < MAX_TAD; n_tads++) { 2105 pci_read_config_dword(pci_ha, tad_dram_rule[n_tads], ®); 2106 limit = TAD_LIMIT(reg); 2107 if (limit <= prv) { 2108 sprintf(msg, "Can't discover the memory channel"); 2109 return -EINVAL; 2110 } 2111 if (addr <= limit) 2112 break; 2113 prv = limit; 2114 } 2115 if (n_tads == MAX_TAD) { 2116 sprintf(msg, "Can't discover the memory channel"); 2117 return -EINVAL; 2118 } 2119 2120 ch_way = TAD_CH(reg) + 1; 2121 sck_way = 1 << TAD_SOCK(reg); 2122 2123 if (ch_way == 3) 2124 idx = addr >> 6; 2125 else 2126 idx = (addr >> (6 + sck_way + shiftup)) & 0x3; 2127 idx = idx % ch_way; 2128 2129 /* 2130 * FIXME: Shouldn't we use CHN_IDX_OFFSET() here, when ch_way == 3 ??? 2131 */ 2132 switch (idx) { 2133 case 0: 2134 base_ch = TAD_TGT0(reg); 2135 break; 2136 case 1: 2137 base_ch = TAD_TGT1(reg); 2138 break; 2139 case 2: 2140 base_ch = TAD_TGT2(reg); 2141 break; 2142 case 3: 2143 base_ch = TAD_TGT3(reg); 2144 break; 2145 default: 2146 sprintf(msg, "Can't discover the TAD target"); 2147 return -EINVAL; 2148 } 2149 *channel_mask = 1 << base_ch; 2150 2151 pci_read_config_dword(pvt->pci_tad[ch_add + base_ch], 2152 tad_ch_nilv_offset[n_tads], 2153 &tad_offset); 2154 2155 if (pvt->is_mirrored) { 2156 *channel_mask |= 1 << ((base_ch + 2) % 4); 2157 switch(ch_way) { 2158 case 2: 2159 case 4: 2160 sck_xch = 1 << sck_way * (ch_way >> 1); 2161 break; 2162 default: 2163 sprintf(msg, "Invalid mirror set. Can't decode addr"); 2164 return -EINVAL; 2165 } 2166 } else 2167 sck_xch = (1 << sck_way) * ch_way; 2168 2169 if (pvt->is_lockstep) 2170 *channel_mask |= 1 << ((base_ch + 1) % 4); 2171 2172 offset = TAD_OFFSET(tad_offset); 2173 2174 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", 2175 n_tads, 2176 addr, 2177 limit, 2178 sck_way, 2179 ch_way, 2180 offset, 2181 idx, 2182 base_ch, 2183 *channel_mask); 2184 2185 /* Calculate channel address */ 2186 /* Remove the TAD offset */ 2187 2188 if (offset > addr) { 2189 sprintf(msg, "Can't calculate ch addr: TAD offset 0x%08Lx is too high for addr 0x%08Lx!", 2190 offset, addr); 2191 return -EINVAL; 2192 } 2193 2194 ch_addr = addr - offset; 2195 ch_addr >>= (6 + shiftup); 2196 ch_addr /= ch_way * sck_way; 2197 ch_addr <<= (6 + shiftup); 2198 ch_addr |= addr & ((1 << (6 + shiftup)) - 1); 2199 2200 /* 2201 * Step 3) Decode rank 2202 */ 2203 for (n_rir = 0; n_rir < MAX_RIR_RANGES; n_rir++) { 2204 pci_read_config_dword(pvt->pci_tad[ch_add + base_ch], 2205 rir_way_limit[n_rir], 2206 ®); 2207 2208 if (!IS_RIR_VALID(reg)) 2209 continue; 2210 2211 limit = pvt->info.rir_limit(reg); 2212 gb = div_u64_rem(limit >> 20, 1024, &mb); 2213 edac_dbg(0, "RIR#%d, limit: %u.%03u GB (0x%016Lx), way: %d\n", 2214 n_rir, 2215 gb, (mb*1000)/1024, 2216 limit, 2217 1 << RIR_WAY(reg)); 2218 if (ch_addr <= limit) 2219 break; 2220 } 2221 if (n_rir == MAX_RIR_RANGES) { 2222 sprintf(msg, "Can't discover the memory rank for ch addr 0x%08Lx", 2223 ch_addr); 2224 return -EINVAL; 2225 } 2226 rir_way = RIR_WAY(reg); 2227 2228 if (pvt->is_close_pg) 2229 idx = (ch_addr >> 6); 2230 else 2231 idx = (ch_addr >> 13); /* FIXME: Datasheet says to shift by 15 */ 2232 idx %= 1 << rir_way; 2233 2234 pci_read_config_dword(pvt->pci_tad[ch_add + base_ch], 2235 rir_offset[n_rir][idx], 2236 ®); 2237 *rank = RIR_RNK_TGT(reg); 2238 2239 edac_dbg(0, "RIR#%d: channel address 0x%08Lx < 0x%08Lx, RIR interleave %d, index %d\n", 2240 n_rir, 2241 ch_addr, 2242 limit, 2243 rir_way, 2244 idx); 2245 2246 return 0; 2247 } 2248 2249 /**************************************************************************** 2250 Device initialization routines: put/get, init/exit 2251 ****************************************************************************/ 2252 2253 /* 2254 * sbridge_put_all_devices 'put' all the devices that we have 2255 * reserved via 'get' 2256 */ 2257 static void sbridge_put_devices(struct sbridge_dev *sbridge_dev) 2258 { 2259 int i; 2260 2261 edac_dbg(0, "\n"); 2262 for (i = 0; i < sbridge_dev->n_devs; i++) { 2263 struct pci_dev *pdev = sbridge_dev->pdev[i]; 2264 if (!pdev) 2265 continue; 2266 edac_dbg(0, "Removing dev %02x:%02x.%d\n", 2267 pdev->bus->number, 2268 PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn)); 2269 pci_dev_put(pdev); 2270 } 2271 } 2272 2273 static void sbridge_put_all_devices(void) 2274 { 2275 struct sbridge_dev *sbridge_dev, *tmp; 2276 2277 list_for_each_entry_safe(sbridge_dev, tmp, &sbridge_edac_list, list) { 2278 sbridge_put_devices(sbridge_dev); 2279 free_sbridge_dev(sbridge_dev); 2280 } 2281 } 2282 2283 static int sbridge_get_onedevice(struct pci_dev **prev, 2284 u8 *num_mc, 2285 const struct pci_id_table *table, 2286 const unsigned devno, 2287 const int multi_bus) 2288 { 2289 struct sbridge_dev *sbridge_dev; 2290 const struct pci_id_descr *dev_descr = &table->descr[devno]; 2291 struct pci_dev *pdev = NULL; 2292 u8 bus = 0; 2293 2294 sbridge_printk(KERN_DEBUG, 2295 "Seeking for: PCI ID %04x:%04x\n", 2296 PCI_VENDOR_ID_INTEL, dev_descr->dev_id); 2297 2298 pdev = pci_get_device(PCI_VENDOR_ID_INTEL, 2299 dev_descr->dev_id, *prev); 2300 2301 if (!pdev) { 2302 if (*prev) { 2303 *prev = pdev; 2304 return 0; 2305 } 2306 2307 if (dev_descr->optional) 2308 return 0; 2309 2310 /* if the HA wasn't found */ 2311 if (devno == 0) 2312 return -ENODEV; 2313 2314 sbridge_printk(KERN_INFO, 2315 "Device not found: %04x:%04x\n", 2316 PCI_VENDOR_ID_INTEL, dev_descr->dev_id); 2317 2318 /* End of list, leave */ 2319 return -ENODEV; 2320 } 2321 bus = pdev->bus->number; 2322 2323 sbridge_dev = get_sbridge_dev(bus, multi_bus); 2324 if (!sbridge_dev) { 2325 sbridge_dev = alloc_sbridge_dev(bus, table); 2326 if (!sbridge_dev) { 2327 pci_dev_put(pdev); 2328 return -ENOMEM; 2329 } 2330 (*num_mc)++; 2331 } 2332 2333 if (sbridge_dev->pdev[devno]) { 2334 sbridge_printk(KERN_ERR, 2335 "Duplicated device for %04x:%04x\n", 2336 PCI_VENDOR_ID_INTEL, dev_descr->dev_id); 2337 pci_dev_put(pdev); 2338 return -ENODEV; 2339 } 2340 2341 sbridge_dev->pdev[devno] = pdev; 2342 2343 /* Be sure that the device is enabled */ 2344 if (unlikely(pci_enable_device(pdev) < 0)) { 2345 sbridge_printk(KERN_ERR, 2346 "Couldn't enable %04x:%04x\n", 2347 PCI_VENDOR_ID_INTEL, dev_descr->dev_id); 2348 return -ENODEV; 2349 } 2350 2351 edac_dbg(0, "Detected %04x:%04x\n", 2352 PCI_VENDOR_ID_INTEL, dev_descr->dev_id); 2353 2354 /* 2355 * As stated on drivers/pci/search.c, the reference count for 2356 * @from is always decremented if it is not %NULL. So, as we need 2357 * to get all devices up to null, we need to do a get for the device 2358 */ 2359 pci_dev_get(pdev); 2360 2361 *prev = pdev; 2362 2363 return 0; 2364 } 2365 2366 /* 2367 * sbridge_get_all_devices - Find and perform 'get' operation on the MCH's 2368 * devices we want to reference for this driver. 2369 * @num_mc: pointer to the memory controllers count, to be incremented in case 2370 * of success. 2371 * @table: model specific table 2372 * @allow_dups: allow for multiple devices to exist with the same device id 2373 * (as implemented, this isn't expected to work correctly in the 2374 * multi-socket case). 2375 * @multi_bus: don't assume devices on different buses belong to different 2376 * memory controllers. 2377 * 2378 * returns 0 in case of success or error code 2379 */ 2380 static int sbridge_get_all_devices_full(u8 *num_mc, 2381 const struct pci_id_table *table, 2382 int allow_dups, 2383 int multi_bus) 2384 { 2385 int i, rc; 2386 struct pci_dev *pdev = NULL; 2387 2388 while (table && table->descr) { 2389 for (i = 0; i < table->n_devs; i++) { 2390 if (!allow_dups || i == 0 || 2391 table->descr[i].dev_id != 2392 table->descr[i-1].dev_id) { 2393 pdev = NULL; 2394 } 2395 do { 2396 rc = sbridge_get_onedevice(&pdev, num_mc, 2397 table, i, multi_bus); 2398 if (rc < 0) { 2399 if (i == 0) { 2400 i = table->n_devs; 2401 break; 2402 } 2403 sbridge_put_all_devices(); 2404 return -ENODEV; 2405 } 2406 } while (pdev && !allow_dups); 2407 } 2408 table++; 2409 } 2410 2411 return 0; 2412 } 2413 2414 #define sbridge_get_all_devices(num_mc, table) \ 2415 sbridge_get_all_devices_full(num_mc, table, 0, 0) 2416 #define sbridge_get_all_devices_knl(num_mc, table) \ 2417 sbridge_get_all_devices_full(num_mc, table, 1, 1) 2418 2419 static int sbridge_mci_bind_devs(struct mem_ctl_info *mci, 2420 struct sbridge_dev *sbridge_dev) 2421 { 2422 struct sbridge_pvt *pvt = mci->pvt_info; 2423 struct pci_dev *pdev; 2424 u8 saw_chan_mask = 0; 2425 int i; 2426 2427 for (i = 0; i < sbridge_dev->n_devs; i++) { 2428 pdev = sbridge_dev->pdev[i]; 2429 if (!pdev) 2430 continue; 2431 2432 switch (pdev->device) { 2433 case PCI_DEVICE_ID_INTEL_SBRIDGE_SAD0: 2434 pvt->pci_sad0 = pdev; 2435 break; 2436 case PCI_DEVICE_ID_INTEL_SBRIDGE_SAD1: 2437 pvt->pci_sad1 = pdev; 2438 break; 2439 case PCI_DEVICE_ID_INTEL_SBRIDGE_BR: 2440 pvt->pci_br0 = pdev; 2441 break; 2442 case PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_HA0: 2443 pvt->pci_ha0 = pdev; 2444 break; 2445 case PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TA: 2446 pvt->pci_ta = pdev; 2447 break; 2448 case PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_RAS: 2449 pvt->pci_ras = pdev; 2450 break; 2451 case PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD0: 2452 case PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD1: 2453 case PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD2: 2454 case PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD3: 2455 { 2456 int id = pdev->device - PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD0; 2457 pvt->pci_tad[id] = pdev; 2458 saw_chan_mask |= 1 << id; 2459 } 2460 break; 2461 case PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_DDRIO: 2462 pvt->pci_ddrio = pdev; 2463 break; 2464 default: 2465 goto error; 2466 } 2467 2468 edac_dbg(0, "Associated PCI %02x:%02x, bus %d with dev = %p\n", 2469 pdev->vendor, pdev->device, 2470 sbridge_dev->bus, 2471 pdev); 2472 } 2473 2474 /* Check if everything were registered */ 2475 if (!pvt->pci_sad0 || !pvt->pci_sad1 || !pvt->pci_ha0 || 2476 !pvt-> pci_tad || !pvt->pci_ras || !pvt->pci_ta) 2477 goto enodev; 2478 2479 if (saw_chan_mask != 0x0f) 2480 goto enodev; 2481 return 0; 2482 2483 enodev: 2484 sbridge_printk(KERN_ERR, "Some needed devices are missing\n"); 2485 return -ENODEV; 2486 2487 error: 2488 sbridge_printk(KERN_ERR, "Unexpected device %02x:%02x\n", 2489 PCI_VENDOR_ID_INTEL, pdev->device); 2490 return -EINVAL; 2491 } 2492 2493 static int ibridge_mci_bind_devs(struct mem_ctl_info *mci, 2494 struct sbridge_dev *sbridge_dev) 2495 { 2496 struct sbridge_pvt *pvt = mci->pvt_info; 2497 struct pci_dev *pdev; 2498 u8 saw_chan_mask = 0; 2499 int i; 2500 2501 for (i = 0; i < sbridge_dev->n_devs; i++) { 2502 pdev = sbridge_dev->pdev[i]; 2503 if (!pdev) 2504 continue; 2505 2506 switch (pdev->device) { 2507 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0: 2508 pvt->pci_ha0 = pdev; 2509 break; 2510 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TA: 2511 pvt->pci_ta = pdev; 2512 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_RAS: 2513 pvt->pci_ras = pdev; 2514 break; 2515 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD0: 2516 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD1: 2517 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD2: 2518 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD3: 2519 { 2520 int id = pdev->device - PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD0; 2521 pvt->pci_tad[id] = pdev; 2522 saw_chan_mask |= 1 << id; 2523 } 2524 break; 2525 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_2HA_DDRIO0: 2526 pvt->pci_ddrio = pdev; 2527 break; 2528 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_1HA_DDRIO0: 2529 pvt->pci_ddrio = pdev; 2530 break; 2531 case PCI_DEVICE_ID_INTEL_IBRIDGE_SAD: 2532 pvt->pci_sad0 = pdev; 2533 break; 2534 case PCI_DEVICE_ID_INTEL_IBRIDGE_BR0: 2535 pvt->pci_br0 = pdev; 2536 break; 2537 case PCI_DEVICE_ID_INTEL_IBRIDGE_BR1: 2538 pvt->pci_br1 = pdev; 2539 break; 2540 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1: 2541 pvt->pci_ha1 = pdev; 2542 break; 2543 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD0: 2544 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD1: 2545 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD2: 2546 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD3: 2547 { 2548 int id = pdev->device - PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD0 + 4; 2549 pvt->pci_tad[id] = pdev; 2550 saw_chan_mask |= 1 << id; 2551 } 2552 break; 2553 default: 2554 goto error; 2555 } 2556 2557 edac_dbg(0, "Associated PCI %02x.%02d.%d with dev = %p\n", 2558 sbridge_dev->bus, 2559 PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn), 2560 pdev); 2561 } 2562 2563 /* Check if everything were registered */ 2564 if (!pvt->pci_sad0 || !pvt->pci_ha0 || !pvt->pci_br0 || 2565 !pvt->pci_br1 || !pvt->pci_tad || !pvt->pci_ras || 2566 !pvt->pci_ta) 2567 goto enodev; 2568 2569 if (saw_chan_mask != 0x0f && /* -EN */ 2570 saw_chan_mask != 0x33 && /* -EP */ 2571 saw_chan_mask != 0xff) /* -EX */ 2572 goto enodev; 2573 return 0; 2574 2575 enodev: 2576 sbridge_printk(KERN_ERR, "Some needed devices are missing\n"); 2577 return -ENODEV; 2578 2579 error: 2580 sbridge_printk(KERN_ERR, 2581 "Unexpected device %02x:%02x\n", PCI_VENDOR_ID_INTEL, 2582 pdev->device); 2583 return -EINVAL; 2584 } 2585 2586 static int haswell_mci_bind_devs(struct mem_ctl_info *mci, 2587 struct sbridge_dev *sbridge_dev) 2588 { 2589 struct sbridge_pvt *pvt = mci->pvt_info; 2590 struct pci_dev *pdev; 2591 u8 saw_chan_mask = 0; 2592 int i; 2593 2594 /* there's only one device per system; not tied to any bus */ 2595 if (pvt->info.pci_vtd == NULL) 2596 /* result will be checked later */ 2597 pvt->info.pci_vtd = pci_get_device(PCI_VENDOR_ID_INTEL, 2598 PCI_DEVICE_ID_INTEL_HASWELL_IMC_VTD_MISC, 2599 NULL); 2600 2601 for (i = 0; i < sbridge_dev->n_devs; i++) { 2602 pdev = sbridge_dev->pdev[i]; 2603 if (!pdev) 2604 continue; 2605 2606 switch (pdev->device) { 2607 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_CBO_SAD0: 2608 pvt->pci_sad0 = pdev; 2609 break; 2610 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_CBO_SAD1: 2611 pvt->pci_sad1 = pdev; 2612 break; 2613 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0: 2614 pvt->pci_ha0 = pdev; 2615 break; 2616 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TA: 2617 pvt->pci_ta = pdev; 2618 break; 2619 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_THERMAL: 2620 pvt->pci_ras = pdev; 2621 break; 2622 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD0: 2623 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD1: 2624 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD2: 2625 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD3: 2626 { 2627 int id = pdev->device - PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD0; 2628 2629 pvt->pci_tad[id] = pdev; 2630 saw_chan_mask |= 1 << id; 2631 } 2632 break; 2633 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD0: 2634 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD1: 2635 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD2: 2636 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD3: 2637 { 2638 int id = pdev->device - PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD0 + 4; 2639 2640 pvt->pci_tad[id] = pdev; 2641 saw_chan_mask |= 1 << id; 2642 } 2643 break; 2644 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO0: 2645 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO1: 2646 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO2: 2647 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO3: 2648 if (!pvt->pci_ddrio) 2649 pvt->pci_ddrio = pdev; 2650 break; 2651 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1: 2652 pvt->pci_ha1 = pdev; 2653 break; 2654 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TA: 2655 pvt->pci_ha1_ta = pdev; 2656 break; 2657 default: 2658 break; 2659 } 2660 2661 edac_dbg(0, "Associated PCI %02x.%02d.%d with dev = %p\n", 2662 sbridge_dev->bus, 2663 PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn), 2664 pdev); 2665 } 2666 2667 /* Check if everything were registered */ 2668 if (!pvt->pci_sad0 || !pvt->pci_ha0 || !pvt->pci_sad1 || 2669 !pvt->pci_ras || !pvt->pci_ta || !pvt->info.pci_vtd) 2670 goto enodev; 2671 2672 if (saw_chan_mask != 0x0f && /* -EN */ 2673 saw_chan_mask != 0x33 && /* -EP */ 2674 saw_chan_mask != 0xff) /* -EX */ 2675 goto enodev; 2676 return 0; 2677 2678 enodev: 2679 sbridge_printk(KERN_ERR, "Some needed devices are missing\n"); 2680 return -ENODEV; 2681 } 2682 2683 static int broadwell_mci_bind_devs(struct mem_ctl_info *mci, 2684 struct sbridge_dev *sbridge_dev) 2685 { 2686 struct sbridge_pvt *pvt = mci->pvt_info; 2687 struct pci_dev *pdev; 2688 u8 saw_chan_mask = 0; 2689 int i; 2690 2691 /* there's only one device per system; not tied to any bus */ 2692 if (pvt->info.pci_vtd == NULL) 2693 /* result will be checked later */ 2694 pvt->info.pci_vtd = pci_get_device(PCI_VENDOR_ID_INTEL, 2695 PCI_DEVICE_ID_INTEL_BROADWELL_IMC_VTD_MISC, 2696 NULL); 2697 2698 for (i = 0; i < sbridge_dev->n_devs; i++) { 2699 pdev = sbridge_dev->pdev[i]; 2700 if (!pdev) 2701 continue; 2702 2703 switch (pdev->device) { 2704 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_CBO_SAD0: 2705 pvt->pci_sad0 = pdev; 2706 break; 2707 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_CBO_SAD1: 2708 pvt->pci_sad1 = pdev; 2709 break; 2710 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0: 2711 pvt->pci_ha0 = pdev; 2712 break; 2713 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TA: 2714 pvt->pci_ta = pdev; 2715 break; 2716 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_THERMAL: 2717 pvt->pci_ras = pdev; 2718 break; 2719 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD0: 2720 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD1: 2721 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD2: 2722 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD3: 2723 { 2724 int id = pdev->device - PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD0; 2725 pvt->pci_tad[id] = pdev; 2726 saw_chan_mask |= 1 << id; 2727 } 2728 break; 2729 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD0: 2730 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD1: 2731 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD2: 2732 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD3: 2733 { 2734 int id = pdev->device - PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD0 + 4; 2735 pvt->pci_tad[id] = pdev; 2736 saw_chan_mask |= 1 << id; 2737 } 2738 break; 2739 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_DDRIO0: 2740 pvt->pci_ddrio = pdev; 2741 break; 2742 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1: 2743 pvt->pci_ha1 = pdev; 2744 break; 2745 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TA: 2746 pvt->pci_ha1_ta = pdev; 2747 break; 2748 default: 2749 break; 2750 } 2751 2752 edac_dbg(0, "Associated PCI %02x.%02d.%d with dev = %p\n", 2753 sbridge_dev->bus, 2754 PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn), 2755 pdev); 2756 } 2757 2758 /* Check if everything were registered */ 2759 if (!pvt->pci_sad0 || !pvt->pci_ha0 || !pvt->pci_sad1 || 2760 !pvt->pci_ras || !pvt->pci_ta || !pvt->info.pci_vtd) 2761 goto enodev; 2762 2763 if (saw_chan_mask != 0x0f && /* -EN */ 2764 saw_chan_mask != 0x33 && /* -EP */ 2765 saw_chan_mask != 0xff) /* -EX */ 2766 goto enodev; 2767 return 0; 2768 2769 enodev: 2770 sbridge_printk(KERN_ERR, "Some needed devices are missing\n"); 2771 return -ENODEV; 2772 } 2773 2774 static int knl_mci_bind_devs(struct mem_ctl_info *mci, 2775 struct sbridge_dev *sbridge_dev) 2776 { 2777 struct sbridge_pvt *pvt = mci->pvt_info; 2778 struct pci_dev *pdev; 2779 int dev, func; 2780 2781 int i; 2782 int devidx; 2783 2784 for (i = 0; i < sbridge_dev->n_devs; i++) { 2785 pdev = sbridge_dev->pdev[i]; 2786 if (!pdev) 2787 continue; 2788 2789 /* Extract PCI device and function. */ 2790 dev = (pdev->devfn >> 3) & 0x1f; 2791 func = pdev->devfn & 0x7; 2792 2793 switch (pdev->device) { 2794 case PCI_DEVICE_ID_INTEL_KNL_IMC_MC: 2795 if (dev == 8) 2796 pvt->knl.pci_mc0 = pdev; 2797 else if (dev == 9) 2798 pvt->knl.pci_mc1 = pdev; 2799 else { 2800 sbridge_printk(KERN_ERR, 2801 "Memory controller in unexpected place! (dev %d, fn %d)\n", 2802 dev, func); 2803 continue; 2804 } 2805 break; 2806 2807 case PCI_DEVICE_ID_INTEL_KNL_IMC_SAD0: 2808 pvt->pci_sad0 = pdev; 2809 break; 2810 2811 case PCI_DEVICE_ID_INTEL_KNL_IMC_SAD1: 2812 pvt->pci_sad1 = pdev; 2813 break; 2814 2815 case PCI_DEVICE_ID_INTEL_KNL_IMC_CHA: 2816 /* There are one of these per tile, and range from 2817 * 1.14.0 to 1.18.5. 2818 */ 2819 devidx = ((dev-14)*8)+func; 2820 2821 if (devidx < 0 || devidx >= KNL_MAX_CHAS) { 2822 sbridge_printk(KERN_ERR, 2823 "Caching and Home Agent in unexpected place! (dev %d, fn %d)\n", 2824 dev, func); 2825 continue; 2826 } 2827 2828 WARN_ON(pvt->knl.pci_cha[devidx] != NULL); 2829 2830 pvt->knl.pci_cha[devidx] = pdev; 2831 break; 2832 2833 case PCI_DEVICE_ID_INTEL_KNL_IMC_CHANNEL: 2834 devidx = -1; 2835 2836 /* 2837 * MC0 channels 0-2 are device 9 function 2-4, 2838 * MC1 channels 3-5 are device 8 function 2-4. 2839 */ 2840 2841 if (dev == 9) 2842 devidx = func-2; 2843 else if (dev == 8) 2844 devidx = 3 + (func-2); 2845 2846 if (devidx < 0 || devidx >= KNL_MAX_CHANNELS) { 2847 sbridge_printk(KERN_ERR, 2848 "DRAM Channel Registers in unexpected place! (dev %d, fn %d)\n", 2849 dev, func); 2850 continue; 2851 } 2852 2853 WARN_ON(pvt->knl.pci_channel[devidx] != NULL); 2854 pvt->knl.pci_channel[devidx] = pdev; 2855 break; 2856 2857 case PCI_DEVICE_ID_INTEL_KNL_IMC_TOLHM: 2858 pvt->knl.pci_mc_info = pdev; 2859 break; 2860 2861 case PCI_DEVICE_ID_INTEL_KNL_IMC_TA: 2862 pvt->pci_ta = pdev; 2863 break; 2864 2865 default: 2866 sbridge_printk(KERN_ERR, "Unexpected device %d\n", 2867 pdev->device); 2868 break; 2869 } 2870 } 2871 2872 if (!pvt->knl.pci_mc0 || !pvt->knl.pci_mc1 || 2873 !pvt->pci_sad0 || !pvt->pci_sad1 || 2874 !pvt->pci_ta) { 2875 goto enodev; 2876 } 2877 2878 for (i = 0; i < KNL_MAX_CHANNELS; i++) { 2879 if (!pvt->knl.pci_channel[i]) { 2880 sbridge_printk(KERN_ERR, "Missing channel %d\n", i); 2881 goto enodev; 2882 } 2883 } 2884 2885 for (i = 0; i < KNL_MAX_CHAS; i++) { 2886 if (!pvt->knl.pci_cha[i]) { 2887 sbridge_printk(KERN_ERR, "Missing CHA %d\n", i); 2888 goto enodev; 2889 } 2890 } 2891 2892 return 0; 2893 2894 enodev: 2895 sbridge_printk(KERN_ERR, "Some needed devices are missing\n"); 2896 return -ENODEV; 2897 } 2898 2899 /**************************************************************************** 2900 Error check routines 2901 ****************************************************************************/ 2902 2903 /* 2904 * While Sandy Bridge has error count registers, SMI BIOS read values from 2905 * and resets the counters. So, they are not reliable for the OS to read 2906 * from them. So, we have no option but to just trust on whatever MCE is 2907 * telling us about the errors. 2908 */ 2909 static void sbridge_mce_output_error(struct mem_ctl_info *mci, 2910 const struct mce *m) 2911 { 2912 struct mem_ctl_info *new_mci; 2913 struct sbridge_pvt *pvt = mci->pvt_info; 2914 enum hw_event_mc_err_type tp_event; 2915 char *type, *optype, msg[256]; 2916 bool ripv = GET_BITFIELD(m->mcgstatus, 0, 0); 2917 bool overflow = GET_BITFIELD(m->status, 62, 62); 2918 bool uncorrected_error = GET_BITFIELD(m->status, 61, 61); 2919 bool recoverable; 2920 u32 core_err_cnt = GET_BITFIELD(m->status, 38, 52); 2921 u32 mscod = GET_BITFIELD(m->status, 16, 31); 2922 u32 errcode = GET_BITFIELD(m->status, 0, 15); 2923 u32 channel = GET_BITFIELD(m->status, 0, 3); 2924 u32 optypenum = GET_BITFIELD(m->status, 4, 6); 2925 long channel_mask, first_channel; 2926 u8 rank, socket, ha; 2927 int rc, dimm; 2928 char *area_type = NULL; 2929 2930 if (pvt->info.type != SANDY_BRIDGE) 2931 recoverable = true; 2932 else 2933 recoverable = GET_BITFIELD(m->status, 56, 56); 2934 2935 if (uncorrected_error) { 2936 if (ripv) { 2937 type = "FATAL"; 2938 tp_event = HW_EVENT_ERR_FATAL; 2939 } else { 2940 type = "NON_FATAL"; 2941 tp_event = HW_EVENT_ERR_UNCORRECTED; 2942 } 2943 } else { 2944 type = "CORRECTED"; 2945 tp_event = HW_EVENT_ERR_CORRECTED; 2946 } 2947 2948 /* 2949 * According with Table 15-9 of the Intel Architecture spec vol 3A, 2950 * memory errors should fit in this mask: 2951 * 000f 0000 1mmm cccc (binary) 2952 * where: 2953 * f = Correction Report Filtering Bit. If 1, subsequent errors 2954 * won't be shown 2955 * mmm = error type 2956 * cccc = channel 2957 * If the mask doesn't match, report an error to the parsing logic 2958 */ 2959 if (! ((errcode & 0xef80) == 0x80)) { 2960 optype = "Can't parse: it is not a mem"; 2961 } else { 2962 switch (optypenum) { 2963 case 0: 2964 optype = "generic undef request error"; 2965 break; 2966 case 1: 2967 optype = "memory read error"; 2968 break; 2969 case 2: 2970 optype = "memory write error"; 2971 break; 2972 case 3: 2973 optype = "addr/cmd error"; 2974 break; 2975 case 4: 2976 optype = "memory scrubbing error"; 2977 break; 2978 default: 2979 optype = "reserved"; 2980 break; 2981 } 2982 } 2983 2984 /* Only decode errors with an valid address (ADDRV) */ 2985 if (!GET_BITFIELD(m->status, 58, 58)) 2986 return; 2987 2988 if (pvt->info.type == KNIGHTS_LANDING) { 2989 if (channel == 14) { 2990 edac_dbg(0, "%s%s err_code:%04x:%04x EDRAM bank %d\n", 2991 overflow ? " OVERFLOW" : "", 2992 (uncorrected_error && recoverable) 2993 ? " recoverable" : "", 2994 mscod, errcode, 2995 m->bank); 2996 } else { 2997 char A = *("A"); 2998 2999 channel = knl_channel_remap(channel); 3000 channel_mask = 1 << channel; 3001 snprintf(msg, sizeof(msg), 3002 "%s%s err_code:%04x:%04x channel:%d (DIMM_%c)", 3003 overflow ? " OVERFLOW" : "", 3004 (uncorrected_error && recoverable) 3005 ? " recoverable" : " ", 3006 mscod, errcode, channel, A + channel); 3007 edac_mc_handle_error(tp_event, mci, core_err_cnt, 3008 m->addr >> PAGE_SHIFT, m->addr & ~PAGE_MASK, 0, 3009 channel, 0, -1, 3010 optype, msg); 3011 } 3012 return; 3013 } else { 3014 rc = get_memory_error_data(mci, m->addr, &socket, &ha, 3015 &channel_mask, &rank, &area_type, msg); 3016 } 3017 3018 if (rc < 0) 3019 goto err_parsing; 3020 new_mci = get_mci_for_node_id(socket); 3021 if (!new_mci) { 3022 strcpy(msg, "Error: socket got corrupted!"); 3023 goto err_parsing; 3024 } 3025 mci = new_mci; 3026 pvt = mci->pvt_info; 3027 3028 first_channel = find_first_bit(&channel_mask, NUM_CHANNELS); 3029 3030 if (rank < 4) 3031 dimm = 0; 3032 else if (rank < 8) 3033 dimm = 1; 3034 else 3035 dimm = 2; 3036 3037 3038 /* 3039 * FIXME: On some memory configurations (mirror, lockstep), the 3040 * Memory Controller can't point the error to a single DIMM. The 3041 * EDAC core should be handling the channel mask, in order to point 3042 * to the group of dimm's where the error may be happening. 3043 */ 3044 if (!pvt->is_lockstep && !pvt->is_mirrored && !pvt->is_close_pg) 3045 channel = first_channel; 3046 3047 snprintf(msg, sizeof(msg), 3048 "%s%s area:%s err_code:%04x:%04x socket:%d ha:%d channel_mask:%ld rank:%d", 3049 overflow ? " OVERFLOW" : "", 3050 (uncorrected_error && recoverable) ? " recoverable" : "", 3051 area_type, 3052 mscod, errcode, 3053 socket, ha, 3054 channel_mask, 3055 rank); 3056 3057 edac_dbg(0, "%s\n", msg); 3058 3059 /* FIXME: need support for channel mask */ 3060 3061 if (channel == CHANNEL_UNSPECIFIED) 3062 channel = -1; 3063 3064 /* Call the helper to output message */ 3065 edac_mc_handle_error(tp_event, mci, core_err_cnt, 3066 m->addr >> PAGE_SHIFT, m->addr & ~PAGE_MASK, 0, 3067 4*ha+channel, dimm, -1, 3068 optype, msg); 3069 return; 3070 err_parsing: 3071 edac_mc_handle_error(tp_event, mci, core_err_cnt, 0, 0, 0, 3072 -1, -1, -1, 3073 msg, ""); 3074 3075 } 3076 3077 /* 3078 * sbridge_check_error Retrieve and process errors reported by the 3079 * hardware. Called by the Core module. 3080 */ 3081 static void sbridge_check_error(struct mem_ctl_info *mci) 3082 { 3083 struct sbridge_pvt *pvt = mci->pvt_info; 3084 int i; 3085 unsigned count = 0; 3086 struct mce *m; 3087 3088 /* 3089 * MCE first step: Copy all mce errors into a temporary buffer 3090 * We use a double buffering here, to reduce the risk of 3091 * loosing an error. 3092 */ 3093 smp_rmb(); 3094 count = (pvt->mce_out + MCE_LOG_LEN - pvt->mce_in) 3095 % MCE_LOG_LEN; 3096 if (!count) 3097 return; 3098 3099 m = pvt->mce_outentry; 3100 if (pvt->mce_in + count > MCE_LOG_LEN) { 3101 unsigned l = MCE_LOG_LEN - pvt->mce_in; 3102 3103 memcpy(m, &pvt->mce_entry[pvt->mce_in], sizeof(*m) * l); 3104 smp_wmb(); 3105 pvt->mce_in = 0; 3106 count -= l; 3107 m += l; 3108 } 3109 memcpy(m, &pvt->mce_entry[pvt->mce_in], sizeof(*m) * count); 3110 smp_wmb(); 3111 pvt->mce_in += count; 3112 3113 smp_rmb(); 3114 if (pvt->mce_overrun) { 3115 sbridge_printk(KERN_ERR, "Lost %d memory errors\n", 3116 pvt->mce_overrun); 3117 smp_wmb(); 3118 pvt->mce_overrun = 0; 3119 } 3120 3121 /* 3122 * MCE second step: parse errors and display 3123 */ 3124 for (i = 0; i < count; i++) 3125 sbridge_mce_output_error(mci, &pvt->mce_outentry[i]); 3126 } 3127 3128 /* 3129 * sbridge_mce_check_error Replicates mcelog routine to get errors 3130 * This routine simply queues mcelog errors, and 3131 * return. The error itself should be handled later 3132 * by sbridge_check_error. 3133 * WARNING: As this routine should be called at NMI time, extra care should 3134 * be taken to avoid deadlocks, and to be as fast as possible. 3135 */ 3136 static int sbridge_mce_check_error(struct notifier_block *nb, unsigned long val, 3137 void *data) 3138 { 3139 struct mce *mce = (struct mce *)data; 3140 struct mem_ctl_info *mci; 3141 struct sbridge_pvt *pvt; 3142 char *type; 3143 3144 if (get_edac_report_status() == EDAC_REPORTING_DISABLED) 3145 return NOTIFY_DONE; 3146 3147 mci = get_mci_for_node_id(mce->socketid); 3148 if (!mci) 3149 return NOTIFY_BAD; 3150 pvt = mci->pvt_info; 3151 3152 /* 3153 * Just let mcelog handle it if the error is 3154 * outside the memory controller. A memory error 3155 * is indicated by bit 7 = 1 and bits = 8-11,13-15 = 0. 3156 * bit 12 has an special meaning. 3157 */ 3158 if ((mce->status & 0xefff) >> 7 != 1) 3159 return NOTIFY_DONE; 3160 3161 if (mce->mcgstatus & MCG_STATUS_MCIP) 3162 type = "Exception"; 3163 else 3164 type = "Event"; 3165 3166 sbridge_mc_printk(mci, KERN_DEBUG, "HANDLING MCE MEMORY ERROR\n"); 3167 3168 sbridge_mc_printk(mci, KERN_DEBUG, "CPU %d: Machine Check %s: %Lx " 3169 "Bank %d: %016Lx\n", mce->extcpu, type, 3170 mce->mcgstatus, mce->bank, mce->status); 3171 sbridge_mc_printk(mci, KERN_DEBUG, "TSC %llx ", mce->tsc); 3172 sbridge_mc_printk(mci, KERN_DEBUG, "ADDR %llx ", mce->addr); 3173 sbridge_mc_printk(mci, KERN_DEBUG, "MISC %llx ", mce->misc); 3174 3175 sbridge_mc_printk(mci, KERN_DEBUG, "PROCESSOR %u:%x TIME %llu SOCKET " 3176 "%u APIC %x\n", mce->cpuvendor, mce->cpuid, 3177 mce->time, mce->socketid, mce->apicid); 3178 3179 smp_rmb(); 3180 if ((pvt->mce_out + 1) % MCE_LOG_LEN == pvt->mce_in) { 3181 smp_wmb(); 3182 pvt->mce_overrun++; 3183 return NOTIFY_DONE; 3184 } 3185 3186 /* Copy memory error at the ringbuffer */ 3187 memcpy(&pvt->mce_entry[pvt->mce_out], mce, sizeof(*mce)); 3188 smp_wmb(); 3189 pvt->mce_out = (pvt->mce_out + 1) % MCE_LOG_LEN; 3190 3191 /* Handle fatal errors immediately */ 3192 if (mce->mcgstatus & 1) 3193 sbridge_check_error(mci); 3194 3195 /* Advice mcelog that the error were handled */ 3196 return NOTIFY_STOP; 3197 } 3198 3199 static struct notifier_block sbridge_mce_dec = { 3200 .notifier_call = sbridge_mce_check_error, 3201 }; 3202 3203 /**************************************************************************** 3204 EDAC register/unregister logic 3205 ****************************************************************************/ 3206 3207 static void sbridge_unregister_mci(struct sbridge_dev *sbridge_dev) 3208 { 3209 struct mem_ctl_info *mci = sbridge_dev->mci; 3210 struct sbridge_pvt *pvt; 3211 3212 if (unlikely(!mci || !mci->pvt_info)) { 3213 edac_dbg(0, "MC: dev = %p\n", &sbridge_dev->pdev[0]->dev); 3214 3215 sbridge_printk(KERN_ERR, "Couldn't find mci handler\n"); 3216 return; 3217 } 3218 3219 pvt = mci->pvt_info; 3220 3221 edac_dbg(0, "MC: mci = %p, dev = %p\n", 3222 mci, &sbridge_dev->pdev[0]->dev); 3223 3224 /* Remove MC sysfs nodes */ 3225 edac_mc_del_mc(mci->pdev); 3226 3227 edac_dbg(1, "%s: free mci struct\n", mci->ctl_name); 3228 kfree(mci->ctl_name); 3229 edac_mc_free(mci); 3230 sbridge_dev->mci = NULL; 3231 } 3232 3233 static int sbridge_register_mci(struct sbridge_dev *sbridge_dev, enum type type) 3234 { 3235 struct mem_ctl_info *mci; 3236 struct edac_mc_layer layers[2]; 3237 struct sbridge_pvt *pvt; 3238 struct pci_dev *pdev = sbridge_dev->pdev[0]; 3239 int rc; 3240 3241 /* Check the number of active and not disabled channels */ 3242 rc = check_if_ecc_is_active(sbridge_dev->bus, type); 3243 if (unlikely(rc < 0)) 3244 return rc; 3245 3246 /* allocate a new MC control structure */ 3247 layers[0].type = EDAC_MC_LAYER_CHANNEL; 3248 layers[0].size = type == KNIGHTS_LANDING ? 3249 KNL_MAX_CHANNELS : NUM_CHANNELS; 3250 layers[0].is_virt_csrow = false; 3251 layers[1].type = EDAC_MC_LAYER_SLOT; 3252 layers[1].size = type == KNIGHTS_LANDING ? 1 : MAX_DIMMS; 3253 layers[1].is_virt_csrow = true; 3254 mci = edac_mc_alloc(sbridge_dev->mc, ARRAY_SIZE(layers), layers, 3255 sizeof(*pvt)); 3256 3257 if (unlikely(!mci)) 3258 return -ENOMEM; 3259 3260 edac_dbg(0, "MC: mci = %p, dev = %p\n", 3261 mci, &pdev->dev); 3262 3263 pvt = mci->pvt_info; 3264 memset(pvt, 0, sizeof(*pvt)); 3265 3266 /* Associate sbridge_dev and mci for future usage */ 3267 pvt->sbridge_dev = sbridge_dev; 3268 sbridge_dev->mci = mci; 3269 3270 mci->mtype_cap = type == KNIGHTS_LANDING ? 3271 MEM_FLAG_DDR4 : MEM_FLAG_DDR3; 3272 mci->edac_ctl_cap = EDAC_FLAG_NONE; 3273 mci->edac_cap = EDAC_FLAG_NONE; 3274 mci->mod_name = "sbridge_edac.c"; 3275 mci->mod_ver = SBRIDGE_REVISION; 3276 mci->dev_name = pci_name(pdev); 3277 mci->ctl_page_to_phys = NULL; 3278 3279 /* Set the function pointer to an actual operation function */ 3280 mci->edac_check = sbridge_check_error; 3281 3282 pvt->info.type = type; 3283 switch (type) { 3284 case IVY_BRIDGE: 3285 pvt->info.rankcfgr = IB_RANK_CFG_A; 3286 pvt->info.get_tolm = ibridge_get_tolm; 3287 pvt->info.get_tohm = ibridge_get_tohm; 3288 pvt->info.dram_rule = ibridge_dram_rule; 3289 pvt->info.get_memory_type = get_memory_type; 3290 pvt->info.get_node_id = get_node_id; 3291 pvt->info.rir_limit = rir_limit; 3292 pvt->info.sad_limit = sad_limit; 3293 pvt->info.interleave_mode = interleave_mode; 3294 pvt->info.show_interleave_mode = show_interleave_mode; 3295 pvt->info.dram_attr = dram_attr; 3296 pvt->info.max_sad = ARRAY_SIZE(ibridge_dram_rule); 3297 pvt->info.interleave_list = ibridge_interleave_list; 3298 pvt->info.max_interleave = ARRAY_SIZE(ibridge_interleave_list); 3299 pvt->info.interleave_pkg = ibridge_interleave_pkg; 3300 pvt->info.get_width = ibridge_get_width; 3301 mci->ctl_name = kasprintf(GFP_KERNEL, "Ivy Bridge Socket#%d", mci->mc_idx); 3302 3303 /* Store pci devices at mci for faster access */ 3304 rc = ibridge_mci_bind_devs(mci, sbridge_dev); 3305 if (unlikely(rc < 0)) 3306 goto fail0; 3307 break; 3308 case SANDY_BRIDGE: 3309 pvt->info.rankcfgr = SB_RANK_CFG_A; 3310 pvt->info.get_tolm = sbridge_get_tolm; 3311 pvt->info.get_tohm = sbridge_get_tohm; 3312 pvt->info.dram_rule = sbridge_dram_rule; 3313 pvt->info.get_memory_type = get_memory_type; 3314 pvt->info.get_node_id = get_node_id; 3315 pvt->info.rir_limit = rir_limit; 3316 pvt->info.sad_limit = sad_limit; 3317 pvt->info.interleave_mode = interleave_mode; 3318 pvt->info.show_interleave_mode = show_interleave_mode; 3319 pvt->info.dram_attr = dram_attr; 3320 pvt->info.max_sad = ARRAY_SIZE(sbridge_dram_rule); 3321 pvt->info.interleave_list = sbridge_interleave_list; 3322 pvt->info.max_interleave = ARRAY_SIZE(sbridge_interleave_list); 3323 pvt->info.interleave_pkg = sbridge_interleave_pkg; 3324 pvt->info.get_width = sbridge_get_width; 3325 mci->ctl_name = kasprintf(GFP_KERNEL, "Sandy Bridge Socket#%d", mci->mc_idx); 3326 3327 /* Store pci devices at mci for faster access */ 3328 rc = sbridge_mci_bind_devs(mci, sbridge_dev); 3329 if (unlikely(rc < 0)) 3330 goto fail0; 3331 break; 3332 case HASWELL: 3333 /* rankcfgr isn't used */ 3334 pvt->info.get_tolm = haswell_get_tolm; 3335 pvt->info.get_tohm = haswell_get_tohm; 3336 pvt->info.dram_rule = ibridge_dram_rule; 3337 pvt->info.get_memory_type = haswell_get_memory_type; 3338 pvt->info.get_node_id = haswell_get_node_id; 3339 pvt->info.rir_limit = haswell_rir_limit; 3340 pvt->info.sad_limit = sad_limit; 3341 pvt->info.interleave_mode = interleave_mode; 3342 pvt->info.show_interleave_mode = show_interleave_mode; 3343 pvt->info.dram_attr = dram_attr; 3344 pvt->info.max_sad = ARRAY_SIZE(ibridge_dram_rule); 3345 pvt->info.interleave_list = ibridge_interleave_list; 3346 pvt->info.max_interleave = ARRAY_SIZE(ibridge_interleave_list); 3347 pvt->info.interleave_pkg = ibridge_interleave_pkg; 3348 pvt->info.get_width = ibridge_get_width; 3349 mci->ctl_name = kasprintf(GFP_KERNEL, "Haswell Socket#%d", mci->mc_idx); 3350 3351 /* Store pci devices at mci for faster access */ 3352 rc = haswell_mci_bind_devs(mci, sbridge_dev); 3353 if (unlikely(rc < 0)) 3354 goto fail0; 3355 break; 3356 case BROADWELL: 3357 /* rankcfgr isn't used */ 3358 pvt->info.get_tolm = haswell_get_tolm; 3359 pvt->info.get_tohm = haswell_get_tohm; 3360 pvt->info.dram_rule = ibridge_dram_rule; 3361 pvt->info.get_memory_type = haswell_get_memory_type; 3362 pvt->info.get_node_id = haswell_get_node_id; 3363 pvt->info.rir_limit = haswell_rir_limit; 3364 pvt->info.sad_limit = sad_limit; 3365 pvt->info.interleave_mode = interleave_mode; 3366 pvt->info.show_interleave_mode = show_interleave_mode; 3367 pvt->info.dram_attr = dram_attr; 3368 pvt->info.max_sad = ARRAY_SIZE(ibridge_dram_rule); 3369 pvt->info.interleave_list = ibridge_interleave_list; 3370 pvt->info.max_interleave = ARRAY_SIZE(ibridge_interleave_list); 3371 pvt->info.interleave_pkg = ibridge_interleave_pkg; 3372 pvt->info.get_width = broadwell_get_width; 3373 mci->ctl_name = kasprintf(GFP_KERNEL, "Broadwell Socket#%d", mci->mc_idx); 3374 3375 /* Store pci devices at mci for faster access */ 3376 rc = broadwell_mci_bind_devs(mci, sbridge_dev); 3377 if (unlikely(rc < 0)) 3378 goto fail0; 3379 break; 3380 case KNIGHTS_LANDING: 3381 /* pvt->info.rankcfgr == ??? */ 3382 pvt->info.get_tolm = knl_get_tolm; 3383 pvt->info.get_tohm = knl_get_tohm; 3384 pvt->info.dram_rule = knl_dram_rule; 3385 pvt->info.get_memory_type = knl_get_memory_type; 3386 pvt->info.get_node_id = knl_get_node_id; 3387 pvt->info.rir_limit = NULL; 3388 pvt->info.sad_limit = knl_sad_limit; 3389 pvt->info.interleave_mode = knl_interleave_mode; 3390 pvt->info.show_interleave_mode = knl_show_interleave_mode; 3391 pvt->info.dram_attr = dram_attr_knl; 3392 pvt->info.max_sad = ARRAY_SIZE(knl_dram_rule); 3393 pvt->info.interleave_list = knl_interleave_list; 3394 pvt->info.max_interleave = ARRAY_SIZE(knl_interleave_list); 3395 pvt->info.interleave_pkg = ibridge_interleave_pkg; 3396 pvt->info.get_width = knl_get_width; 3397 mci->ctl_name = kasprintf(GFP_KERNEL, 3398 "Knights Landing Socket#%d", mci->mc_idx); 3399 3400 rc = knl_mci_bind_devs(mci, sbridge_dev); 3401 if (unlikely(rc < 0)) 3402 goto fail0; 3403 break; 3404 } 3405 3406 /* Get dimm basic config and the memory layout */ 3407 get_dimm_config(mci); 3408 get_memory_layout(mci); 3409 3410 /* record ptr to the generic device */ 3411 mci->pdev = &pdev->dev; 3412 3413 /* add this new MC control structure to EDAC's list of MCs */ 3414 if (unlikely(edac_mc_add_mc(mci))) { 3415 edac_dbg(0, "MC: failed edac_mc_add_mc()\n"); 3416 rc = -EINVAL; 3417 goto fail0; 3418 } 3419 3420 return 0; 3421 3422 fail0: 3423 kfree(mci->ctl_name); 3424 edac_mc_free(mci); 3425 sbridge_dev->mci = NULL; 3426 return rc; 3427 } 3428 3429 /* 3430 * sbridge_probe Probe for ONE instance of device to see if it is 3431 * present. 3432 * return: 3433 * 0 for FOUND a device 3434 * < 0 for error code 3435 */ 3436 3437 static int sbridge_probe(struct pci_dev *pdev, const struct pci_device_id *id) 3438 { 3439 int rc = -ENODEV; 3440 u8 mc, num_mc = 0; 3441 struct sbridge_dev *sbridge_dev; 3442 enum type type = SANDY_BRIDGE; 3443 3444 /* get the pci devices we want to reserve for our use */ 3445 mutex_lock(&sbridge_edac_lock); 3446 3447 /* 3448 * All memory controllers are allocated at the first pass. 3449 */ 3450 if (unlikely(probed >= 1)) { 3451 mutex_unlock(&sbridge_edac_lock); 3452 return -ENODEV; 3453 } 3454 probed++; 3455 3456 switch (pdev->device) { 3457 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TA: 3458 rc = sbridge_get_all_devices(&num_mc, 3459 pci_dev_descr_ibridge_table); 3460 type = IVY_BRIDGE; 3461 break; 3462 case PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_HA0: 3463 rc = sbridge_get_all_devices(&num_mc, 3464 pci_dev_descr_sbridge_table); 3465 type = SANDY_BRIDGE; 3466 break; 3467 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0: 3468 rc = sbridge_get_all_devices(&num_mc, 3469 pci_dev_descr_haswell_table); 3470 type = HASWELL; 3471 break; 3472 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0: 3473 rc = sbridge_get_all_devices(&num_mc, 3474 pci_dev_descr_broadwell_table); 3475 type = BROADWELL; 3476 break; 3477 case PCI_DEVICE_ID_INTEL_KNL_IMC_SAD0: 3478 rc = sbridge_get_all_devices_knl(&num_mc, 3479 pci_dev_descr_knl_table); 3480 type = KNIGHTS_LANDING; 3481 break; 3482 } 3483 if (unlikely(rc < 0)) { 3484 edac_dbg(0, "couldn't get all devices for 0x%x\n", pdev->device); 3485 goto fail0; 3486 } 3487 3488 mc = 0; 3489 3490 list_for_each_entry(sbridge_dev, &sbridge_edac_list, list) { 3491 edac_dbg(0, "Registering MC#%d (%d of %d)\n", 3492 mc, mc + 1, num_mc); 3493 3494 sbridge_dev->mc = mc++; 3495 rc = sbridge_register_mci(sbridge_dev, type); 3496 if (unlikely(rc < 0)) 3497 goto fail1; 3498 } 3499 3500 sbridge_printk(KERN_INFO, "%s\n", SBRIDGE_REVISION); 3501 3502 mutex_unlock(&sbridge_edac_lock); 3503 return 0; 3504 3505 fail1: 3506 list_for_each_entry(sbridge_dev, &sbridge_edac_list, list) 3507 sbridge_unregister_mci(sbridge_dev); 3508 3509 sbridge_put_all_devices(); 3510 fail0: 3511 mutex_unlock(&sbridge_edac_lock); 3512 return rc; 3513 } 3514 3515 /* 3516 * sbridge_remove destructor for one instance of device 3517 * 3518 */ 3519 static void sbridge_remove(struct pci_dev *pdev) 3520 { 3521 struct sbridge_dev *sbridge_dev; 3522 3523 edac_dbg(0, "\n"); 3524 3525 /* 3526 * we have a trouble here: pdev value for removal will be wrong, since 3527 * it will point to the X58 register used to detect that the machine 3528 * is a Nehalem or upper design. However, due to the way several PCI 3529 * devices are grouped together to provide MC functionality, we need 3530 * to use a different method for releasing the devices 3531 */ 3532 3533 mutex_lock(&sbridge_edac_lock); 3534 3535 if (unlikely(!probed)) { 3536 mutex_unlock(&sbridge_edac_lock); 3537 return; 3538 } 3539 3540 list_for_each_entry(sbridge_dev, &sbridge_edac_list, list) 3541 sbridge_unregister_mci(sbridge_dev); 3542 3543 /* Release PCI resources */ 3544 sbridge_put_all_devices(); 3545 3546 probed--; 3547 3548 mutex_unlock(&sbridge_edac_lock); 3549 } 3550 3551 MODULE_DEVICE_TABLE(pci, sbridge_pci_tbl); 3552 3553 /* 3554 * sbridge_driver pci_driver structure for this module 3555 * 3556 */ 3557 static struct pci_driver sbridge_driver = { 3558 .name = "sbridge_edac", 3559 .probe = sbridge_probe, 3560 .remove = sbridge_remove, 3561 .id_table = sbridge_pci_tbl, 3562 }; 3563 3564 /* 3565 * sbridge_init Module entry function 3566 * Try to initialize this module for its devices 3567 */ 3568 static int __init sbridge_init(void) 3569 { 3570 int pci_rc; 3571 3572 edac_dbg(2, "\n"); 3573 3574 /* Ensure that the OPSTATE is set correctly for POLL or NMI */ 3575 opstate_init(); 3576 3577 pci_rc = pci_register_driver(&sbridge_driver); 3578 if (pci_rc >= 0) { 3579 mce_register_decode_chain(&sbridge_mce_dec); 3580 if (get_edac_report_status() == EDAC_REPORTING_DISABLED) 3581 sbridge_printk(KERN_WARNING, "Loading driver, error reporting disabled.\n"); 3582 return 0; 3583 } 3584 3585 sbridge_printk(KERN_ERR, "Failed to register device with error %d.\n", 3586 pci_rc); 3587 3588 return pci_rc; 3589 } 3590 3591 /* 3592 * sbridge_exit() Module exit function 3593 * Unregister the driver 3594 */ 3595 static void __exit sbridge_exit(void) 3596 { 3597 edac_dbg(2, "\n"); 3598 pci_unregister_driver(&sbridge_driver); 3599 mce_unregister_decode_chain(&sbridge_mce_dec); 3600 } 3601 3602 module_init(sbridge_init); 3603 module_exit(sbridge_exit); 3604 3605 module_param(edac_op_state, int, 0444); 3606 MODULE_PARM_DESC(edac_op_state, "EDAC Error Reporting state: 0=Poll,1=NMI"); 3607 3608 MODULE_LICENSE("GPL"); 3609 MODULE_AUTHOR("Mauro Carvalho Chehab"); 3610 MODULE_AUTHOR("Red Hat Inc. (http://www.redhat.com)"); 3611 MODULE_DESCRIPTION("MC Driver for Intel Sandy Bridge and Ivy Bridge memory controllers - " 3612 SBRIDGE_REVISION); 3613