1 // SPDX-License-Identifier: GPL-2.0-only 2 /* Intel Sandy Bridge -EN/-EP/-EX Memory Controller kernel module 3 * 4 * This driver supports the memory controllers found on the Intel 5 * processor family Sandy Bridge. 6 * 7 * Copyright (c) 2011 by: 8 * Mauro Carvalho Chehab 9 */ 10 11 #include <linux/module.h> 12 #include <linux/init.h> 13 #include <linux/pci.h> 14 #include <linux/pci_ids.h> 15 #include <linux/slab.h> 16 #include <linux/delay.h> 17 #include <linux/edac.h> 18 #include <linux/mmzone.h> 19 #include <linux/smp.h> 20 #include <linux/bitmap.h> 21 #include <linux/math64.h> 22 #include <linux/mod_devicetable.h> 23 #include <asm/cpu_device_id.h> 24 #include <asm/intel-family.h> 25 #include <asm/processor.h> 26 #include <asm/mce.h> 27 28 #include "edac_module.h" 29 30 /* Static vars */ 31 static LIST_HEAD(sbridge_edac_list); 32 33 /* 34 * Alter this version for the module when modifications are made 35 */ 36 #define SBRIDGE_REVISION " Ver: 1.1.2 " 37 #define EDAC_MOD_STR "sb_edac" 38 39 /* 40 * Debug macros 41 */ 42 #define sbridge_printk(level, fmt, arg...) \ 43 edac_printk(level, "sbridge", fmt, ##arg) 44 45 #define sbridge_mc_printk(mci, level, fmt, arg...) \ 46 edac_mc_chipset_printk(mci, level, "sbridge", fmt, ##arg) 47 48 /* 49 * Get a bit field at register value <v>, from bit <lo> to bit <hi> 50 */ 51 #define GET_BITFIELD(v, lo, hi) \ 52 (((v) & GENMASK_ULL(hi, lo)) >> (lo)) 53 54 /* Devices 12 Function 6, Offsets 0x80 to 0xcc */ 55 static const u32 sbridge_dram_rule[] = { 56 0x80, 0x88, 0x90, 0x98, 0xa0, 57 0xa8, 0xb0, 0xb8, 0xc0, 0xc8, 58 }; 59 60 static const u32 ibridge_dram_rule[] = { 61 0x60, 0x68, 0x70, 0x78, 0x80, 62 0x88, 0x90, 0x98, 0xa0, 0xa8, 63 0xb0, 0xb8, 0xc0, 0xc8, 0xd0, 64 0xd8, 0xe0, 0xe8, 0xf0, 0xf8, 65 }; 66 67 static const u32 knl_dram_rule[] = { 68 0x60, 0x68, 0x70, 0x78, 0x80, /* 0-4 */ 69 0x88, 0x90, 0x98, 0xa0, 0xa8, /* 5-9 */ 70 0xb0, 0xb8, 0xc0, 0xc8, 0xd0, /* 10-14 */ 71 0xd8, 0xe0, 0xe8, 0xf0, 0xf8, /* 15-19 */ 72 0x100, 0x108, 0x110, 0x118, /* 20-23 */ 73 }; 74 75 #define DRAM_RULE_ENABLE(reg) GET_BITFIELD(reg, 0, 0) 76 #define A7MODE(reg) GET_BITFIELD(reg, 26, 26) 77 78 static char *show_dram_attr(u32 attr) 79 { 80 switch (attr) { 81 case 0: 82 return "DRAM"; 83 case 1: 84 return "MMCFG"; 85 case 2: 86 return "NXM"; 87 default: 88 return "unknown"; 89 } 90 } 91 92 static const u32 sbridge_interleave_list[] = { 93 0x84, 0x8c, 0x94, 0x9c, 0xa4, 94 0xac, 0xb4, 0xbc, 0xc4, 0xcc, 95 }; 96 97 static const u32 ibridge_interleave_list[] = { 98 0x64, 0x6c, 0x74, 0x7c, 0x84, 99 0x8c, 0x94, 0x9c, 0xa4, 0xac, 100 0xb4, 0xbc, 0xc4, 0xcc, 0xd4, 101 0xdc, 0xe4, 0xec, 0xf4, 0xfc, 102 }; 103 104 static const u32 knl_interleave_list[] = { 105 0x64, 0x6c, 0x74, 0x7c, 0x84, /* 0-4 */ 106 0x8c, 0x94, 0x9c, 0xa4, 0xac, /* 5-9 */ 107 0xb4, 0xbc, 0xc4, 0xcc, 0xd4, /* 10-14 */ 108 0xdc, 0xe4, 0xec, 0xf4, 0xfc, /* 15-19 */ 109 0x104, 0x10c, 0x114, 0x11c, /* 20-23 */ 110 }; 111 #define MAX_INTERLEAVE \ 112 (max_t(unsigned int, ARRAY_SIZE(sbridge_interleave_list), \ 113 max_t(unsigned int, ARRAY_SIZE(ibridge_interleave_list), \ 114 ARRAY_SIZE(knl_interleave_list)))) 115 116 struct interleave_pkg { 117 unsigned char start; 118 unsigned char end; 119 }; 120 121 static const struct interleave_pkg sbridge_interleave_pkg[] = { 122 { 0, 2 }, 123 { 3, 5 }, 124 { 8, 10 }, 125 { 11, 13 }, 126 { 16, 18 }, 127 { 19, 21 }, 128 { 24, 26 }, 129 { 27, 29 }, 130 }; 131 132 static const struct interleave_pkg ibridge_interleave_pkg[] = { 133 { 0, 3 }, 134 { 4, 7 }, 135 { 8, 11 }, 136 { 12, 15 }, 137 { 16, 19 }, 138 { 20, 23 }, 139 { 24, 27 }, 140 { 28, 31 }, 141 }; 142 143 static inline int sad_pkg(const struct interleave_pkg *table, u32 reg, 144 int interleave) 145 { 146 return GET_BITFIELD(reg, table[interleave].start, 147 table[interleave].end); 148 } 149 150 /* Devices 12 Function 7 */ 151 152 #define TOLM 0x80 153 #define TOHM 0x84 154 #define HASWELL_TOLM 0xd0 155 #define HASWELL_TOHM_0 0xd4 156 #define HASWELL_TOHM_1 0xd8 157 #define KNL_TOLM 0xd0 158 #define KNL_TOHM_0 0xd4 159 #define KNL_TOHM_1 0xd8 160 161 #define GET_TOLM(reg) ((GET_BITFIELD(reg, 0, 3) << 28) | 0x3ffffff) 162 #define GET_TOHM(reg) ((GET_BITFIELD(reg, 0, 20) << 25) | 0x3ffffff) 163 164 /* Device 13 Function 6 */ 165 166 #define SAD_TARGET 0xf0 167 168 #define SOURCE_ID(reg) GET_BITFIELD(reg, 9, 11) 169 170 #define SOURCE_ID_KNL(reg) GET_BITFIELD(reg, 12, 14) 171 172 #define SAD_CONTROL 0xf4 173 174 /* Device 14 function 0 */ 175 176 static const u32 tad_dram_rule[] = { 177 0x40, 0x44, 0x48, 0x4c, 178 0x50, 0x54, 0x58, 0x5c, 179 0x60, 0x64, 0x68, 0x6c, 180 }; 181 #define MAX_TAD ARRAY_SIZE(tad_dram_rule) 182 183 #define TAD_LIMIT(reg) ((GET_BITFIELD(reg, 12, 31) << 26) | 0x3ffffff) 184 #define TAD_SOCK(reg) GET_BITFIELD(reg, 10, 11) 185 #define TAD_CH(reg) GET_BITFIELD(reg, 8, 9) 186 #define TAD_TGT3(reg) GET_BITFIELD(reg, 6, 7) 187 #define TAD_TGT2(reg) GET_BITFIELD(reg, 4, 5) 188 #define TAD_TGT1(reg) GET_BITFIELD(reg, 2, 3) 189 #define TAD_TGT0(reg) GET_BITFIELD(reg, 0, 1) 190 191 /* Device 15, function 0 */ 192 193 #define MCMTR 0x7c 194 #define KNL_MCMTR 0x624 195 196 #define IS_ECC_ENABLED(mcmtr) GET_BITFIELD(mcmtr, 2, 2) 197 #define IS_LOCKSTEP_ENABLED(mcmtr) GET_BITFIELD(mcmtr, 1, 1) 198 #define IS_CLOSE_PG(mcmtr) GET_BITFIELD(mcmtr, 0, 0) 199 200 /* Device 15, function 1 */ 201 202 #define RASENABLES 0xac 203 #define IS_MIRROR_ENABLED(reg) GET_BITFIELD(reg, 0, 0) 204 205 /* Device 15, functions 2-5 */ 206 207 static const int mtr_regs[] = { 208 0x80, 0x84, 0x88, 209 }; 210 211 static const int knl_mtr_reg = 0xb60; 212 213 #define RANK_DISABLE(mtr) GET_BITFIELD(mtr, 16, 19) 214 #define IS_DIMM_PRESENT(mtr) GET_BITFIELD(mtr, 14, 14) 215 #define RANK_CNT_BITS(mtr) GET_BITFIELD(mtr, 12, 13) 216 #define RANK_WIDTH_BITS(mtr) GET_BITFIELD(mtr, 2, 4) 217 #define COL_WIDTH_BITS(mtr) GET_BITFIELD(mtr, 0, 1) 218 219 static const u32 tad_ch_nilv_offset[] = { 220 0x90, 0x94, 0x98, 0x9c, 221 0xa0, 0xa4, 0xa8, 0xac, 222 0xb0, 0xb4, 0xb8, 0xbc, 223 }; 224 #define CHN_IDX_OFFSET(reg) GET_BITFIELD(reg, 28, 29) 225 #define TAD_OFFSET(reg) (GET_BITFIELD(reg, 6, 25) << 26) 226 227 static const u32 rir_way_limit[] = { 228 0x108, 0x10c, 0x110, 0x114, 0x118, 229 }; 230 #define MAX_RIR_RANGES ARRAY_SIZE(rir_way_limit) 231 232 #define IS_RIR_VALID(reg) GET_BITFIELD(reg, 31, 31) 233 #define RIR_WAY(reg) GET_BITFIELD(reg, 28, 29) 234 235 #define MAX_RIR_WAY 8 236 237 static const u32 rir_offset[MAX_RIR_RANGES][MAX_RIR_WAY] = { 238 { 0x120, 0x124, 0x128, 0x12c, 0x130, 0x134, 0x138, 0x13c }, 239 { 0x140, 0x144, 0x148, 0x14c, 0x150, 0x154, 0x158, 0x15c }, 240 { 0x160, 0x164, 0x168, 0x16c, 0x170, 0x174, 0x178, 0x17c }, 241 { 0x180, 0x184, 0x188, 0x18c, 0x190, 0x194, 0x198, 0x19c }, 242 { 0x1a0, 0x1a4, 0x1a8, 0x1ac, 0x1b0, 0x1b4, 0x1b8, 0x1bc }, 243 }; 244 245 #define RIR_RNK_TGT(type, reg) (((type) == BROADWELL) ? \ 246 GET_BITFIELD(reg, 20, 23) : GET_BITFIELD(reg, 16, 19)) 247 248 #define RIR_OFFSET(type, reg) (((type) == HASWELL || (type) == BROADWELL) ? \ 249 GET_BITFIELD(reg, 2, 15) : GET_BITFIELD(reg, 2, 14)) 250 251 /* Device 16, functions 2-7 */ 252 253 /* 254 * FIXME: Implement the error count reads directly 255 */ 256 257 static const u32 correrrcnt[] = { 258 0x104, 0x108, 0x10c, 0x110, 259 }; 260 261 #define RANK_ODD_OV(reg) GET_BITFIELD(reg, 31, 31) 262 #define RANK_ODD_ERR_CNT(reg) GET_BITFIELD(reg, 16, 30) 263 #define RANK_EVEN_OV(reg) GET_BITFIELD(reg, 15, 15) 264 #define RANK_EVEN_ERR_CNT(reg) GET_BITFIELD(reg, 0, 14) 265 266 static const u32 correrrthrsld[] = { 267 0x11c, 0x120, 0x124, 0x128, 268 }; 269 270 #define RANK_ODD_ERR_THRSLD(reg) GET_BITFIELD(reg, 16, 30) 271 #define RANK_EVEN_ERR_THRSLD(reg) GET_BITFIELD(reg, 0, 14) 272 273 274 /* Device 17, function 0 */ 275 276 #define SB_RANK_CFG_A 0x0328 277 278 #define IB_RANK_CFG_A 0x0320 279 280 /* 281 * sbridge structs 282 */ 283 284 #define NUM_CHANNELS 6 /* Max channels per MC */ 285 #define MAX_DIMMS 3 /* Max DIMMS per channel */ 286 #define KNL_MAX_CHAS 38 /* KNL max num. of Cache Home Agents */ 287 #define KNL_MAX_CHANNELS 6 /* KNL max num. of PCI channels */ 288 #define KNL_MAX_EDCS 8 /* Embedded DRAM controllers */ 289 #define CHANNEL_UNSPECIFIED 0xf /* Intel IA32 SDM 15-14 */ 290 291 enum type { 292 SANDY_BRIDGE, 293 IVY_BRIDGE, 294 HASWELL, 295 BROADWELL, 296 KNIGHTS_LANDING, 297 }; 298 299 enum domain { 300 IMC0 = 0, 301 IMC1, 302 SOCK, 303 }; 304 305 enum mirroring_mode { 306 NON_MIRRORING, 307 ADDR_RANGE_MIRRORING, 308 FULL_MIRRORING, 309 }; 310 311 struct sbridge_pvt; 312 struct sbridge_info { 313 enum type type; 314 u32 mcmtr; 315 u32 rankcfgr; 316 u64 (*get_tolm)(struct sbridge_pvt *pvt); 317 u64 (*get_tohm)(struct sbridge_pvt *pvt); 318 u64 (*rir_limit)(u32 reg); 319 u64 (*sad_limit)(u32 reg); 320 u32 (*interleave_mode)(u32 reg); 321 u32 (*dram_attr)(u32 reg); 322 const u32 *dram_rule; 323 const u32 *interleave_list; 324 const struct interleave_pkg *interleave_pkg; 325 u8 max_sad; 326 u8 (*get_node_id)(struct sbridge_pvt *pvt); 327 u8 (*get_ha)(u8 bank); 328 enum mem_type (*get_memory_type)(struct sbridge_pvt *pvt); 329 enum dev_type (*get_width)(struct sbridge_pvt *pvt, u32 mtr); 330 struct pci_dev *pci_vtd; 331 }; 332 333 struct sbridge_channel { 334 u32 ranks; 335 u32 dimms; 336 }; 337 338 struct pci_id_descr { 339 int dev_id; 340 int optional; 341 enum domain dom; 342 }; 343 344 struct pci_id_table { 345 const struct pci_id_descr *descr; 346 int n_devs_per_imc; 347 int n_devs_per_sock; 348 int n_imcs_per_sock; 349 enum type type; 350 }; 351 352 struct sbridge_dev { 353 struct list_head list; 354 int seg; 355 u8 bus, mc; 356 u8 node_id, source_id; 357 struct pci_dev **pdev; 358 enum domain dom; 359 int n_devs; 360 int i_devs; 361 struct mem_ctl_info *mci; 362 }; 363 364 struct knl_pvt { 365 struct pci_dev *pci_cha[KNL_MAX_CHAS]; 366 struct pci_dev *pci_channel[KNL_MAX_CHANNELS]; 367 struct pci_dev *pci_mc0; 368 struct pci_dev *pci_mc1; 369 struct pci_dev *pci_mc0_misc; 370 struct pci_dev *pci_mc1_misc; 371 struct pci_dev *pci_mc_info; /* tolm, tohm */ 372 }; 373 374 struct sbridge_pvt { 375 /* Devices per socket */ 376 struct pci_dev *pci_ddrio; 377 struct pci_dev *pci_sad0, *pci_sad1; 378 struct pci_dev *pci_br0, *pci_br1; 379 /* Devices per memory controller */ 380 struct pci_dev *pci_ha, *pci_ta, *pci_ras; 381 struct pci_dev *pci_tad[NUM_CHANNELS]; 382 383 struct sbridge_dev *sbridge_dev; 384 385 struct sbridge_info info; 386 struct sbridge_channel channel[NUM_CHANNELS]; 387 388 /* Memory type detection */ 389 bool is_cur_addr_mirrored, is_lockstep, is_close_pg; 390 bool is_chan_hash; 391 enum mirroring_mode mirror_mode; 392 393 /* Memory description */ 394 u64 tolm, tohm; 395 struct knl_pvt knl; 396 }; 397 398 #define PCI_DESCR(device_id, opt, domain) \ 399 .dev_id = (device_id), \ 400 .optional = opt, \ 401 .dom = domain 402 403 static const struct pci_id_descr pci_dev_descr_sbridge[] = { 404 /* Processor Home Agent */ 405 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_HA0, 0, IMC0) }, 406 407 /* Memory controller */ 408 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TA, 0, IMC0) }, 409 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_RAS, 0, IMC0) }, 410 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD0, 0, IMC0) }, 411 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD1, 0, IMC0) }, 412 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD2, 0, IMC0) }, 413 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD3, 0, IMC0) }, 414 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_DDRIO, 1, SOCK) }, 415 416 /* System Address Decoder */ 417 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_SAD0, 0, SOCK) }, 418 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_SAD1, 0, SOCK) }, 419 420 /* Broadcast Registers */ 421 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_BR, 0, SOCK) }, 422 }; 423 424 #define PCI_ID_TABLE_ENTRY(A, N, M, T) { \ 425 .descr = A, \ 426 .n_devs_per_imc = N, \ 427 .n_devs_per_sock = ARRAY_SIZE(A), \ 428 .n_imcs_per_sock = M, \ 429 .type = T \ 430 } 431 432 static const struct pci_id_table pci_dev_descr_sbridge_table[] = { 433 PCI_ID_TABLE_ENTRY(pci_dev_descr_sbridge, ARRAY_SIZE(pci_dev_descr_sbridge), 1, SANDY_BRIDGE), 434 {0,} /* 0 terminated list. */ 435 }; 436 437 /* This changes depending if 1HA or 2HA: 438 * 1HA: 439 * 0x0eb8 (17.0) is DDRIO0 440 * 2HA: 441 * 0x0ebc (17.4) is DDRIO0 442 */ 443 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_1HA_DDRIO0 0x0eb8 444 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_2HA_DDRIO0 0x0ebc 445 446 /* pci ids */ 447 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0 0x0ea0 448 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TA 0x0ea8 449 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_RAS 0x0e71 450 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD0 0x0eaa 451 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD1 0x0eab 452 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD2 0x0eac 453 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD3 0x0ead 454 #define PCI_DEVICE_ID_INTEL_IBRIDGE_SAD 0x0ec8 455 #define PCI_DEVICE_ID_INTEL_IBRIDGE_BR0 0x0ec9 456 #define PCI_DEVICE_ID_INTEL_IBRIDGE_BR1 0x0eca 457 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1 0x0e60 458 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TA 0x0e68 459 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_RAS 0x0e79 460 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD0 0x0e6a 461 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD1 0x0e6b 462 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD2 0x0e6c 463 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD3 0x0e6d 464 465 static const struct pci_id_descr pci_dev_descr_ibridge[] = { 466 /* Processor Home Agent */ 467 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0, 0, IMC0) }, 468 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1, 1, IMC1) }, 469 470 /* Memory controller */ 471 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TA, 0, IMC0) }, 472 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_RAS, 0, IMC0) }, 473 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD0, 0, IMC0) }, 474 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD1, 0, IMC0) }, 475 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD2, 0, IMC0) }, 476 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD3, 0, IMC0) }, 477 478 /* Optional, mode 2HA */ 479 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TA, 1, IMC1) }, 480 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_RAS, 1, IMC1) }, 481 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD0, 1, IMC1) }, 482 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD1, 1, IMC1) }, 483 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD2, 1, IMC1) }, 484 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD3, 1, IMC1) }, 485 486 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_1HA_DDRIO0, 1, SOCK) }, 487 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_2HA_DDRIO0, 1, SOCK) }, 488 489 /* System Address Decoder */ 490 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_SAD, 0, SOCK) }, 491 492 /* Broadcast Registers */ 493 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_BR0, 1, SOCK) }, 494 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_BR1, 0, SOCK) }, 495 496 }; 497 498 static const struct pci_id_table pci_dev_descr_ibridge_table[] = { 499 PCI_ID_TABLE_ENTRY(pci_dev_descr_ibridge, 12, 2, IVY_BRIDGE), 500 {0,} /* 0 terminated list. */ 501 }; 502 503 /* Haswell support */ 504 /* EN processor: 505 * - 1 IMC 506 * - 3 DDR3 channels, 2 DPC per channel 507 * EP processor: 508 * - 1 or 2 IMC 509 * - 4 DDR4 channels, 3 DPC per channel 510 * EP 4S processor: 511 * - 2 IMC 512 * - 4 DDR4 channels, 3 DPC per channel 513 * EX processor: 514 * - 2 IMC 515 * - each IMC interfaces with a SMI 2 channel 516 * - each SMI channel interfaces with a scalable memory buffer 517 * - each scalable memory buffer supports 4 DDR3/DDR4 channels, 3 DPC 518 */ 519 #define HASWELL_DDRCRCLKCONTROLS 0xa10 /* Ditto on Broadwell */ 520 #define HASWELL_HASYSDEFEATURE2 0x84 521 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_VTD_MISC 0x2f28 522 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0 0x2fa0 523 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1 0x2f60 524 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TA 0x2fa8 525 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TM 0x2f71 526 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TA 0x2f68 527 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TM 0x2f79 528 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_CBO_SAD0 0x2ffc 529 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_CBO_SAD1 0x2ffd 530 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD0 0x2faa 531 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD1 0x2fab 532 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD2 0x2fac 533 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD3 0x2fad 534 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD0 0x2f6a 535 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD1 0x2f6b 536 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD2 0x2f6c 537 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD3 0x2f6d 538 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO0 0x2fbd 539 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO1 0x2fbf 540 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO2 0x2fb9 541 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO3 0x2fbb 542 static const struct pci_id_descr pci_dev_descr_haswell[] = { 543 /* first item must be the HA */ 544 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0, 0, IMC0) }, 545 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1, 1, IMC1) }, 546 547 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TA, 0, IMC0) }, 548 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TM, 0, IMC0) }, 549 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD0, 0, IMC0) }, 550 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD1, 0, IMC0) }, 551 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD2, 1, IMC0) }, 552 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD3, 1, IMC0) }, 553 554 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TA, 1, IMC1) }, 555 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TM, 1, IMC1) }, 556 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD0, 1, IMC1) }, 557 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD1, 1, IMC1) }, 558 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD2, 1, IMC1) }, 559 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD3, 1, IMC1) }, 560 561 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_CBO_SAD0, 0, SOCK) }, 562 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_CBO_SAD1, 0, SOCK) }, 563 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO0, 1, SOCK) }, 564 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO1, 1, SOCK) }, 565 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO2, 1, SOCK) }, 566 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO3, 1, SOCK) }, 567 }; 568 569 static const struct pci_id_table pci_dev_descr_haswell_table[] = { 570 PCI_ID_TABLE_ENTRY(pci_dev_descr_haswell, 13, 2, HASWELL), 571 {0,} /* 0 terminated list. */ 572 }; 573 574 /* Knight's Landing Support */ 575 /* 576 * KNL's memory channels are swizzled between memory controllers. 577 * MC0 is mapped to CH3,4,5 and MC1 is mapped to CH0,1,2 578 */ 579 #define knl_channel_remap(mc, chan) ((mc) ? (chan) : (chan) + 3) 580 581 /* Memory controller, TAD tables, error injection - 2-8-0, 2-9-0 (2 of these) */ 582 #define PCI_DEVICE_ID_INTEL_KNL_IMC_MC 0x7840 583 /* DRAM channel stuff; bank addrs, dimmmtr, etc.. 2-8-2 - 2-9-4 (6 of these) */ 584 #define PCI_DEVICE_ID_INTEL_KNL_IMC_CHAN 0x7843 585 /* kdrwdbu TAD limits/offsets, MCMTR - 2-10-1, 2-11-1 (2 of these) */ 586 #define PCI_DEVICE_ID_INTEL_KNL_IMC_TA 0x7844 587 /* CHA broadcast registers, dram rules - 1-29-0 (1 of these) */ 588 #define PCI_DEVICE_ID_INTEL_KNL_IMC_SAD0 0x782a 589 /* SAD target - 1-29-1 (1 of these) */ 590 #define PCI_DEVICE_ID_INTEL_KNL_IMC_SAD1 0x782b 591 /* Caching / Home Agent */ 592 #define PCI_DEVICE_ID_INTEL_KNL_IMC_CHA 0x782c 593 /* Device with TOLM and TOHM, 0-5-0 (1 of these) */ 594 #define PCI_DEVICE_ID_INTEL_KNL_IMC_TOLHM 0x7810 595 596 /* 597 * KNL differs from SB, IB, and Haswell in that it has multiple 598 * instances of the same device with the same device ID, so we handle that 599 * by creating as many copies in the table as we expect to find. 600 * (Like device ID must be grouped together.) 601 */ 602 603 static const struct pci_id_descr pci_dev_descr_knl[] = { 604 [0 ... 1] = { PCI_DESCR(PCI_DEVICE_ID_INTEL_KNL_IMC_MC, 0, IMC0)}, 605 [2 ... 7] = { PCI_DESCR(PCI_DEVICE_ID_INTEL_KNL_IMC_CHAN, 0, IMC0) }, 606 [8] = { PCI_DESCR(PCI_DEVICE_ID_INTEL_KNL_IMC_TA, 0, IMC0) }, 607 [9] = { PCI_DESCR(PCI_DEVICE_ID_INTEL_KNL_IMC_TOLHM, 0, IMC0) }, 608 [10] = { PCI_DESCR(PCI_DEVICE_ID_INTEL_KNL_IMC_SAD0, 0, SOCK) }, 609 [11] = { PCI_DESCR(PCI_DEVICE_ID_INTEL_KNL_IMC_SAD1, 0, SOCK) }, 610 [12 ... 49] = { PCI_DESCR(PCI_DEVICE_ID_INTEL_KNL_IMC_CHA, 0, SOCK) }, 611 }; 612 613 static const struct pci_id_table pci_dev_descr_knl_table[] = { 614 PCI_ID_TABLE_ENTRY(pci_dev_descr_knl, ARRAY_SIZE(pci_dev_descr_knl), 1, KNIGHTS_LANDING), 615 {0,} 616 }; 617 618 /* 619 * Broadwell support 620 * 621 * DE processor: 622 * - 1 IMC 623 * - 2 DDR3 channels, 2 DPC per channel 624 * EP processor: 625 * - 1 or 2 IMC 626 * - 4 DDR4 channels, 3 DPC per channel 627 * EP 4S processor: 628 * - 2 IMC 629 * - 4 DDR4 channels, 3 DPC per channel 630 * EX processor: 631 * - 2 IMC 632 * - each IMC interfaces with a SMI 2 channel 633 * - each SMI channel interfaces with a scalable memory buffer 634 * - each scalable memory buffer supports 4 DDR3/DDR4 channels, 3 DPC 635 */ 636 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_VTD_MISC 0x6f28 637 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0 0x6fa0 638 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1 0x6f60 639 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TA 0x6fa8 640 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TM 0x6f71 641 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TA 0x6f68 642 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TM 0x6f79 643 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_CBO_SAD0 0x6ffc 644 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_CBO_SAD1 0x6ffd 645 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD0 0x6faa 646 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD1 0x6fab 647 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD2 0x6fac 648 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD3 0x6fad 649 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD0 0x6f6a 650 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD1 0x6f6b 651 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD2 0x6f6c 652 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD3 0x6f6d 653 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_DDRIO0 0x6faf 654 655 static const struct pci_id_descr pci_dev_descr_broadwell[] = { 656 /* first item must be the HA */ 657 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0, 0, IMC0) }, 658 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1, 1, IMC1) }, 659 660 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TA, 0, IMC0) }, 661 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TM, 0, IMC0) }, 662 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD0, 0, IMC0) }, 663 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD1, 0, IMC0) }, 664 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD2, 1, IMC0) }, 665 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD3, 1, IMC0) }, 666 667 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TA, 1, IMC1) }, 668 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TM, 1, IMC1) }, 669 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD0, 1, IMC1) }, 670 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD1, 1, IMC1) }, 671 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD2, 1, IMC1) }, 672 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD3, 1, IMC1) }, 673 674 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_CBO_SAD0, 0, SOCK) }, 675 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_CBO_SAD1, 0, SOCK) }, 676 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_DDRIO0, 1, SOCK) }, 677 }; 678 679 static const struct pci_id_table pci_dev_descr_broadwell_table[] = { 680 PCI_ID_TABLE_ENTRY(pci_dev_descr_broadwell, 10, 2, BROADWELL), 681 {0,} /* 0 terminated list. */ 682 }; 683 684 685 /**************************************************************************** 686 Ancillary status routines 687 ****************************************************************************/ 688 689 static inline int numrank(enum type type, u32 mtr) 690 { 691 int ranks = (1 << RANK_CNT_BITS(mtr)); 692 int max = 4; 693 694 if (type == HASWELL || type == BROADWELL || type == KNIGHTS_LANDING) 695 max = 8; 696 697 if (ranks > max) { 698 edac_dbg(0, "Invalid number of ranks: %d (max = %i) raw value = %x (%04x)\n", 699 ranks, max, (unsigned int)RANK_CNT_BITS(mtr), mtr); 700 return -EINVAL; 701 } 702 703 return ranks; 704 } 705 706 static inline int numrow(u32 mtr) 707 { 708 int rows = (RANK_WIDTH_BITS(mtr) + 12); 709 710 if (rows < 13 || rows > 18) { 711 edac_dbg(0, "Invalid number of rows: %d (should be between 14 and 17) raw value = %x (%04x)\n", 712 rows, (unsigned int)RANK_WIDTH_BITS(mtr), mtr); 713 return -EINVAL; 714 } 715 716 return 1 << rows; 717 } 718 719 static inline int numcol(u32 mtr) 720 { 721 int cols = (COL_WIDTH_BITS(mtr) + 10); 722 723 if (cols > 12) { 724 edac_dbg(0, "Invalid number of cols: %d (max = 4) raw value = %x (%04x)\n", 725 cols, (unsigned int)COL_WIDTH_BITS(mtr), mtr); 726 return -EINVAL; 727 } 728 729 return 1 << cols; 730 } 731 732 static struct sbridge_dev *get_sbridge_dev(int seg, u8 bus, enum domain dom, 733 int multi_bus, 734 struct sbridge_dev *prev) 735 { 736 struct sbridge_dev *sbridge_dev; 737 738 /* 739 * If we have devices scattered across several busses that pertain 740 * to the same memory controller, we'll lump them all together. 741 */ 742 if (multi_bus) { 743 return list_first_entry_or_null(&sbridge_edac_list, 744 struct sbridge_dev, list); 745 } 746 747 sbridge_dev = list_entry(prev ? prev->list.next 748 : sbridge_edac_list.next, struct sbridge_dev, list); 749 750 list_for_each_entry_from(sbridge_dev, &sbridge_edac_list, list) { 751 if ((sbridge_dev->seg == seg) && (sbridge_dev->bus == bus) && 752 (dom == SOCK || dom == sbridge_dev->dom)) 753 return sbridge_dev; 754 } 755 756 return NULL; 757 } 758 759 static struct sbridge_dev *alloc_sbridge_dev(int seg, u8 bus, enum domain dom, 760 const struct pci_id_table *table) 761 { 762 struct sbridge_dev *sbridge_dev; 763 764 sbridge_dev = kzalloc(sizeof(*sbridge_dev), GFP_KERNEL); 765 if (!sbridge_dev) 766 return NULL; 767 768 sbridge_dev->pdev = kcalloc(table->n_devs_per_imc, 769 sizeof(*sbridge_dev->pdev), 770 GFP_KERNEL); 771 if (!sbridge_dev->pdev) { 772 kfree(sbridge_dev); 773 return NULL; 774 } 775 776 sbridge_dev->seg = seg; 777 sbridge_dev->bus = bus; 778 sbridge_dev->dom = dom; 779 sbridge_dev->n_devs = table->n_devs_per_imc; 780 list_add_tail(&sbridge_dev->list, &sbridge_edac_list); 781 782 return sbridge_dev; 783 } 784 785 static void free_sbridge_dev(struct sbridge_dev *sbridge_dev) 786 { 787 list_del(&sbridge_dev->list); 788 kfree(sbridge_dev->pdev); 789 kfree(sbridge_dev); 790 } 791 792 static u64 sbridge_get_tolm(struct sbridge_pvt *pvt) 793 { 794 u32 reg; 795 796 /* Address range is 32:28 */ 797 pci_read_config_dword(pvt->pci_sad1, TOLM, ®); 798 return GET_TOLM(reg); 799 } 800 801 static u64 sbridge_get_tohm(struct sbridge_pvt *pvt) 802 { 803 u32 reg; 804 805 pci_read_config_dword(pvt->pci_sad1, TOHM, ®); 806 return GET_TOHM(reg); 807 } 808 809 static u64 ibridge_get_tolm(struct sbridge_pvt *pvt) 810 { 811 u32 reg; 812 813 pci_read_config_dword(pvt->pci_br1, TOLM, ®); 814 815 return GET_TOLM(reg); 816 } 817 818 static u64 ibridge_get_tohm(struct sbridge_pvt *pvt) 819 { 820 u32 reg; 821 822 pci_read_config_dword(pvt->pci_br1, TOHM, ®); 823 824 return GET_TOHM(reg); 825 } 826 827 static u64 rir_limit(u32 reg) 828 { 829 return ((u64)GET_BITFIELD(reg, 1, 10) << 29) | 0x1fffffff; 830 } 831 832 static u64 sad_limit(u32 reg) 833 { 834 return (GET_BITFIELD(reg, 6, 25) << 26) | 0x3ffffff; 835 } 836 837 static u32 interleave_mode(u32 reg) 838 { 839 return GET_BITFIELD(reg, 1, 1); 840 } 841 842 static u32 dram_attr(u32 reg) 843 { 844 return GET_BITFIELD(reg, 2, 3); 845 } 846 847 static u64 knl_sad_limit(u32 reg) 848 { 849 return (GET_BITFIELD(reg, 7, 26) << 26) | 0x3ffffff; 850 } 851 852 static u32 knl_interleave_mode(u32 reg) 853 { 854 return GET_BITFIELD(reg, 1, 2); 855 } 856 857 static const char * const knl_intlv_mode[] = { 858 "[8:6]", "[10:8]", "[14:12]", "[32:30]" 859 }; 860 861 static const char *get_intlv_mode_str(u32 reg, enum type t) 862 { 863 if (t == KNIGHTS_LANDING) 864 return knl_intlv_mode[knl_interleave_mode(reg)]; 865 else 866 return interleave_mode(reg) ? "[8:6]" : "[8:6]XOR[18:16]"; 867 } 868 869 static u32 dram_attr_knl(u32 reg) 870 { 871 return GET_BITFIELD(reg, 3, 4); 872 } 873 874 875 static enum mem_type get_memory_type(struct sbridge_pvt *pvt) 876 { 877 u32 reg; 878 enum mem_type mtype; 879 880 if (pvt->pci_ddrio) { 881 pci_read_config_dword(pvt->pci_ddrio, pvt->info.rankcfgr, 882 ®); 883 if (GET_BITFIELD(reg, 11, 11)) 884 /* FIXME: Can also be LRDIMM */ 885 mtype = MEM_RDDR3; 886 else 887 mtype = MEM_DDR3; 888 } else 889 mtype = MEM_UNKNOWN; 890 891 return mtype; 892 } 893 894 static enum mem_type haswell_get_memory_type(struct sbridge_pvt *pvt) 895 { 896 u32 reg; 897 bool registered = false; 898 enum mem_type mtype = MEM_UNKNOWN; 899 900 if (!pvt->pci_ddrio) 901 goto out; 902 903 pci_read_config_dword(pvt->pci_ddrio, 904 HASWELL_DDRCRCLKCONTROLS, ®); 905 /* Is_Rdimm */ 906 if (GET_BITFIELD(reg, 16, 16)) 907 registered = true; 908 909 pci_read_config_dword(pvt->pci_ta, MCMTR, ®); 910 if (GET_BITFIELD(reg, 14, 14)) { 911 if (registered) 912 mtype = MEM_RDDR4; 913 else 914 mtype = MEM_DDR4; 915 } else { 916 if (registered) 917 mtype = MEM_RDDR3; 918 else 919 mtype = MEM_DDR3; 920 } 921 922 out: 923 return mtype; 924 } 925 926 static enum dev_type knl_get_width(struct sbridge_pvt *pvt, u32 mtr) 927 { 928 /* for KNL value is fixed */ 929 return DEV_X16; 930 } 931 932 static enum dev_type sbridge_get_width(struct sbridge_pvt *pvt, u32 mtr) 933 { 934 /* there's no way to figure out */ 935 return DEV_UNKNOWN; 936 } 937 938 static enum dev_type __ibridge_get_width(u32 mtr) 939 { 940 enum dev_type type; 941 942 switch (mtr) { 943 case 3: 944 type = DEV_UNKNOWN; 945 break; 946 case 2: 947 type = DEV_X16; 948 break; 949 case 1: 950 type = DEV_X8; 951 break; 952 case 0: 953 type = DEV_X4; 954 break; 955 } 956 957 return type; 958 } 959 960 static enum dev_type ibridge_get_width(struct sbridge_pvt *pvt, u32 mtr) 961 { 962 /* 963 * ddr3_width on the documentation but also valid for DDR4 on 964 * Haswell 965 */ 966 return __ibridge_get_width(GET_BITFIELD(mtr, 7, 8)); 967 } 968 969 static enum dev_type broadwell_get_width(struct sbridge_pvt *pvt, u32 mtr) 970 { 971 /* ddr3_width on the documentation but also valid for DDR4 */ 972 return __ibridge_get_width(GET_BITFIELD(mtr, 8, 9)); 973 } 974 975 static enum mem_type knl_get_memory_type(struct sbridge_pvt *pvt) 976 { 977 /* DDR4 RDIMMS and LRDIMMS are supported */ 978 return MEM_RDDR4; 979 } 980 981 static u8 get_node_id(struct sbridge_pvt *pvt) 982 { 983 u32 reg; 984 pci_read_config_dword(pvt->pci_br0, SAD_CONTROL, ®); 985 return GET_BITFIELD(reg, 0, 2); 986 } 987 988 static u8 haswell_get_node_id(struct sbridge_pvt *pvt) 989 { 990 u32 reg; 991 992 pci_read_config_dword(pvt->pci_sad1, SAD_CONTROL, ®); 993 return GET_BITFIELD(reg, 0, 3); 994 } 995 996 static u8 knl_get_node_id(struct sbridge_pvt *pvt) 997 { 998 u32 reg; 999 1000 pci_read_config_dword(pvt->pci_sad1, SAD_CONTROL, ®); 1001 return GET_BITFIELD(reg, 0, 2); 1002 } 1003 1004 /* 1005 * Use the reporting bank number to determine which memory 1006 * controller (also known as "ha" for "home agent"). Sandy 1007 * Bridge only has one memory controller per socket, so the 1008 * answer is always zero. 1009 */ 1010 static u8 sbridge_get_ha(u8 bank) 1011 { 1012 return 0; 1013 } 1014 1015 /* 1016 * On Ivy Bridge, Haswell and Broadwell the error may be in a 1017 * home agent bank (7, 8), or one of the per-channel memory 1018 * controller banks (9 .. 16). 1019 */ 1020 static u8 ibridge_get_ha(u8 bank) 1021 { 1022 switch (bank) { 1023 case 7 ... 8: 1024 return bank - 7; 1025 case 9 ... 16: 1026 return (bank - 9) / 4; 1027 default: 1028 return 0xff; 1029 } 1030 } 1031 1032 /* Not used, but included for safety/symmetry */ 1033 static u8 knl_get_ha(u8 bank) 1034 { 1035 return 0xff; 1036 } 1037 1038 static u64 haswell_get_tolm(struct sbridge_pvt *pvt) 1039 { 1040 u32 reg; 1041 1042 pci_read_config_dword(pvt->info.pci_vtd, HASWELL_TOLM, ®); 1043 return (GET_BITFIELD(reg, 26, 31) << 26) | 0x3ffffff; 1044 } 1045 1046 static u64 haswell_get_tohm(struct sbridge_pvt *pvt) 1047 { 1048 u64 rc; 1049 u32 reg; 1050 1051 pci_read_config_dword(pvt->info.pci_vtd, HASWELL_TOHM_0, ®); 1052 rc = GET_BITFIELD(reg, 26, 31); 1053 pci_read_config_dword(pvt->info.pci_vtd, HASWELL_TOHM_1, ®); 1054 rc = ((reg << 6) | rc) << 26; 1055 1056 return rc | 0x1ffffff; 1057 } 1058 1059 static u64 knl_get_tolm(struct sbridge_pvt *pvt) 1060 { 1061 u32 reg; 1062 1063 pci_read_config_dword(pvt->knl.pci_mc_info, KNL_TOLM, ®); 1064 return (GET_BITFIELD(reg, 26, 31) << 26) | 0x3ffffff; 1065 } 1066 1067 static u64 knl_get_tohm(struct sbridge_pvt *pvt) 1068 { 1069 u64 rc; 1070 u32 reg_lo, reg_hi; 1071 1072 pci_read_config_dword(pvt->knl.pci_mc_info, KNL_TOHM_0, ®_lo); 1073 pci_read_config_dword(pvt->knl.pci_mc_info, KNL_TOHM_1, ®_hi); 1074 rc = ((u64)reg_hi << 32) | reg_lo; 1075 return rc | 0x3ffffff; 1076 } 1077 1078 1079 static u64 haswell_rir_limit(u32 reg) 1080 { 1081 return (((u64)GET_BITFIELD(reg, 1, 11) + 1) << 29) - 1; 1082 } 1083 1084 static inline u8 sad_pkg_socket(u8 pkg) 1085 { 1086 /* on Ivy Bridge, nodeID is SASS, where A is HA and S is node id */ 1087 return ((pkg >> 3) << 2) | (pkg & 0x3); 1088 } 1089 1090 static inline u8 sad_pkg_ha(u8 pkg) 1091 { 1092 return (pkg >> 2) & 0x1; 1093 } 1094 1095 static int haswell_chan_hash(int idx, u64 addr) 1096 { 1097 int i; 1098 1099 /* 1100 * XOR even bits from 12:26 to bit0 of idx, 1101 * odd bits from 13:27 to bit1 1102 */ 1103 for (i = 12; i < 28; i += 2) 1104 idx ^= (addr >> i) & 3; 1105 1106 return idx; 1107 } 1108 1109 /* Low bits of TAD limit, and some metadata. */ 1110 static const u32 knl_tad_dram_limit_lo[] = { 1111 0x400, 0x500, 0x600, 0x700, 1112 0x800, 0x900, 0xa00, 0xb00, 1113 }; 1114 1115 /* Low bits of TAD offset. */ 1116 static const u32 knl_tad_dram_offset_lo[] = { 1117 0x404, 0x504, 0x604, 0x704, 1118 0x804, 0x904, 0xa04, 0xb04, 1119 }; 1120 1121 /* High 16 bits of TAD limit and offset. */ 1122 static const u32 knl_tad_dram_hi[] = { 1123 0x408, 0x508, 0x608, 0x708, 1124 0x808, 0x908, 0xa08, 0xb08, 1125 }; 1126 1127 /* Number of ways a tad entry is interleaved. */ 1128 static const u32 knl_tad_ways[] = { 1129 8, 6, 4, 3, 2, 1, 1130 }; 1131 1132 /* 1133 * Retrieve the n'th Target Address Decode table entry 1134 * from the memory controller's TAD table. 1135 * 1136 * @pvt: driver private data 1137 * @entry: which entry you want to retrieve 1138 * @mc: which memory controller (0 or 1) 1139 * @offset: output tad range offset 1140 * @limit: output address of first byte above tad range 1141 * @ways: output number of interleave ways 1142 * 1143 * The offset value has curious semantics. It's a sort of running total 1144 * of the sizes of all the memory regions that aren't mapped in this 1145 * tad table. 1146 */ 1147 static int knl_get_tad(const struct sbridge_pvt *pvt, 1148 const int entry, 1149 const int mc, 1150 u64 *offset, 1151 u64 *limit, 1152 int *ways) 1153 { 1154 u32 reg_limit_lo, reg_offset_lo, reg_hi; 1155 struct pci_dev *pci_mc; 1156 int way_id; 1157 1158 switch (mc) { 1159 case 0: 1160 pci_mc = pvt->knl.pci_mc0; 1161 break; 1162 case 1: 1163 pci_mc = pvt->knl.pci_mc1; 1164 break; 1165 default: 1166 WARN_ON(1); 1167 return -EINVAL; 1168 } 1169 1170 pci_read_config_dword(pci_mc, 1171 knl_tad_dram_limit_lo[entry], ®_limit_lo); 1172 pci_read_config_dword(pci_mc, 1173 knl_tad_dram_offset_lo[entry], ®_offset_lo); 1174 pci_read_config_dword(pci_mc, 1175 knl_tad_dram_hi[entry], ®_hi); 1176 1177 /* Is this TAD entry enabled? */ 1178 if (!GET_BITFIELD(reg_limit_lo, 0, 0)) 1179 return -ENODEV; 1180 1181 way_id = GET_BITFIELD(reg_limit_lo, 3, 5); 1182 1183 if (way_id < ARRAY_SIZE(knl_tad_ways)) { 1184 *ways = knl_tad_ways[way_id]; 1185 } else { 1186 *ways = 0; 1187 sbridge_printk(KERN_ERR, 1188 "Unexpected value %d in mc_tad_limit_lo wayness field\n", 1189 way_id); 1190 return -ENODEV; 1191 } 1192 1193 /* 1194 * The least significant 6 bits of base and limit are truncated. 1195 * For limit, we fill the missing bits with 1s. 1196 */ 1197 *offset = ((u64) GET_BITFIELD(reg_offset_lo, 6, 31) << 6) | 1198 ((u64) GET_BITFIELD(reg_hi, 0, 15) << 32); 1199 *limit = ((u64) GET_BITFIELD(reg_limit_lo, 6, 31) << 6) | 63 | 1200 ((u64) GET_BITFIELD(reg_hi, 16, 31) << 32); 1201 1202 return 0; 1203 } 1204 1205 /* Determine which memory controller is responsible for a given channel. */ 1206 static int knl_channel_mc(int channel) 1207 { 1208 WARN_ON(channel < 0 || channel >= 6); 1209 1210 return channel < 3 ? 1 : 0; 1211 } 1212 1213 /* 1214 * Get the Nth entry from EDC_ROUTE_TABLE register. 1215 * (This is the per-tile mapping of logical interleave targets to 1216 * physical EDC modules.) 1217 * 1218 * entry 0: 0:2 1219 * 1: 3:5 1220 * 2: 6:8 1221 * 3: 9:11 1222 * 4: 12:14 1223 * 5: 15:17 1224 * 6: 18:20 1225 * 7: 21:23 1226 * reserved: 24:31 1227 */ 1228 static u32 knl_get_edc_route(int entry, u32 reg) 1229 { 1230 WARN_ON(entry >= KNL_MAX_EDCS); 1231 return GET_BITFIELD(reg, entry*3, (entry*3)+2); 1232 } 1233 1234 /* 1235 * Get the Nth entry from MC_ROUTE_TABLE register. 1236 * (This is the per-tile mapping of logical interleave targets to 1237 * physical DRAM channels modules.) 1238 * 1239 * entry 0: mc 0:2 channel 18:19 1240 * 1: mc 3:5 channel 20:21 1241 * 2: mc 6:8 channel 22:23 1242 * 3: mc 9:11 channel 24:25 1243 * 4: mc 12:14 channel 26:27 1244 * 5: mc 15:17 channel 28:29 1245 * reserved: 30:31 1246 * 1247 * Though we have 3 bits to identify the MC, we should only see 1248 * the values 0 or 1. 1249 */ 1250 1251 static u32 knl_get_mc_route(int entry, u32 reg) 1252 { 1253 int mc, chan; 1254 1255 WARN_ON(entry >= KNL_MAX_CHANNELS); 1256 1257 mc = GET_BITFIELD(reg, entry*3, (entry*3)+2); 1258 chan = GET_BITFIELD(reg, (entry*2) + 18, (entry*2) + 18 + 1); 1259 1260 return knl_channel_remap(mc, chan); 1261 } 1262 1263 /* 1264 * Render the EDC_ROUTE register in human-readable form. 1265 * Output string s should be at least KNL_MAX_EDCS*2 bytes. 1266 */ 1267 static void knl_show_edc_route(u32 reg, char *s) 1268 { 1269 int i; 1270 1271 for (i = 0; i < KNL_MAX_EDCS; i++) { 1272 s[i*2] = knl_get_edc_route(i, reg) + '0'; 1273 s[i*2+1] = '-'; 1274 } 1275 1276 s[KNL_MAX_EDCS*2 - 1] = '\0'; 1277 } 1278 1279 /* 1280 * Render the MC_ROUTE register in human-readable form. 1281 * Output string s should be at least KNL_MAX_CHANNELS*2 bytes. 1282 */ 1283 static void knl_show_mc_route(u32 reg, char *s) 1284 { 1285 int i; 1286 1287 for (i = 0; i < KNL_MAX_CHANNELS; i++) { 1288 s[i*2] = knl_get_mc_route(i, reg) + '0'; 1289 s[i*2+1] = '-'; 1290 } 1291 1292 s[KNL_MAX_CHANNELS*2 - 1] = '\0'; 1293 } 1294 1295 #define KNL_EDC_ROUTE 0xb8 1296 #define KNL_MC_ROUTE 0xb4 1297 1298 /* Is this dram rule backed by regular DRAM in flat mode? */ 1299 #define KNL_EDRAM(reg) GET_BITFIELD(reg, 29, 29) 1300 1301 /* Is this dram rule cached? */ 1302 #define KNL_CACHEABLE(reg) GET_BITFIELD(reg, 28, 28) 1303 1304 /* Is this rule backed by edc ? */ 1305 #define KNL_EDRAM_ONLY(reg) GET_BITFIELD(reg, 29, 29) 1306 1307 /* Is this rule backed by DRAM, cacheable in EDRAM? */ 1308 #define KNL_CACHEABLE(reg) GET_BITFIELD(reg, 28, 28) 1309 1310 /* Is this rule mod3? */ 1311 #define KNL_MOD3(reg) GET_BITFIELD(reg, 27, 27) 1312 1313 /* 1314 * Figure out how big our RAM modules are. 1315 * 1316 * The DIMMMTR register in KNL doesn't tell us the size of the DIMMs, so we 1317 * have to figure this out from the SAD rules, interleave lists, route tables, 1318 * and TAD rules. 1319 * 1320 * SAD rules can have holes in them (e.g. the 3G-4G hole), so we have to 1321 * inspect the TAD rules to figure out how large the SAD regions really are. 1322 * 1323 * When we know the real size of a SAD region and how many ways it's 1324 * interleaved, we know the individual contribution of each channel to 1325 * TAD is size/ways. 1326 * 1327 * Finally, we have to check whether each channel participates in each SAD 1328 * region. 1329 * 1330 * Fortunately, KNL only supports one DIMM per channel, so once we know how 1331 * much memory the channel uses, we know the DIMM is at least that large. 1332 * (The BIOS might possibly choose not to map all available memory, in which 1333 * case we will underreport the size of the DIMM.) 1334 * 1335 * In theory, we could try to determine the EDC sizes as well, but that would 1336 * only work in flat mode, not in cache mode. 1337 * 1338 * @mc_sizes: Output sizes of channels (must have space for KNL_MAX_CHANNELS 1339 * elements) 1340 */ 1341 static int knl_get_dimm_capacity(struct sbridge_pvt *pvt, u64 *mc_sizes) 1342 { 1343 u64 sad_base, sad_size, sad_limit = 0; 1344 u64 tad_base, tad_size, tad_limit, tad_deadspace, tad_livespace; 1345 int sad_rule = 0; 1346 int tad_rule = 0; 1347 int intrlv_ways, tad_ways; 1348 u32 first_pkg, pkg; 1349 int i; 1350 u64 sad_actual_size[2]; /* sad size accounting for holes, per mc */ 1351 u32 dram_rule, interleave_reg; 1352 u32 mc_route_reg[KNL_MAX_CHAS]; 1353 u32 edc_route_reg[KNL_MAX_CHAS]; 1354 int edram_only; 1355 char edc_route_string[KNL_MAX_EDCS*2]; 1356 char mc_route_string[KNL_MAX_CHANNELS*2]; 1357 int cur_reg_start; 1358 int mc; 1359 int channel; 1360 int participants[KNL_MAX_CHANNELS]; 1361 1362 for (i = 0; i < KNL_MAX_CHANNELS; i++) 1363 mc_sizes[i] = 0; 1364 1365 /* Read the EDC route table in each CHA. */ 1366 cur_reg_start = 0; 1367 for (i = 0; i < KNL_MAX_CHAS; i++) { 1368 pci_read_config_dword(pvt->knl.pci_cha[i], 1369 KNL_EDC_ROUTE, &edc_route_reg[i]); 1370 1371 if (i > 0 && edc_route_reg[i] != edc_route_reg[i-1]) { 1372 knl_show_edc_route(edc_route_reg[i-1], 1373 edc_route_string); 1374 if (cur_reg_start == i-1) 1375 edac_dbg(0, "edc route table for CHA %d: %s\n", 1376 cur_reg_start, edc_route_string); 1377 else 1378 edac_dbg(0, "edc route table for CHA %d-%d: %s\n", 1379 cur_reg_start, i-1, edc_route_string); 1380 cur_reg_start = i; 1381 } 1382 } 1383 knl_show_edc_route(edc_route_reg[i-1], edc_route_string); 1384 if (cur_reg_start == i-1) 1385 edac_dbg(0, "edc route table for CHA %d: %s\n", 1386 cur_reg_start, edc_route_string); 1387 else 1388 edac_dbg(0, "edc route table for CHA %d-%d: %s\n", 1389 cur_reg_start, i-1, edc_route_string); 1390 1391 /* Read the MC route table in each CHA. */ 1392 cur_reg_start = 0; 1393 for (i = 0; i < KNL_MAX_CHAS; i++) { 1394 pci_read_config_dword(pvt->knl.pci_cha[i], 1395 KNL_MC_ROUTE, &mc_route_reg[i]); 1396 1397 if (i > 0 && mc_route_reg[i] != mc_route_reg[i-1]) { 1398 knl_show_mc_route(mc_route_reg[i-1], mc_route_string); 1399 if (cur_reg_start == i-1) 1400 edac_dbg(0, "mc route table for CHA %d: %s\n", 1401 cur_reg_start, mc_route_string); 1402 else 1403 edac_dbg(0, "mc route table for CHA %d-%d: %s\n", 1404 cur_reg_start, i-1, mc_route_string); 1405 cur_reg_start = i; 1406 } 1407 } 1408 knl_show_mc_route(mc_route_reg[i-1], mc_route_string); 1409 if (cur_reg_start == i-1) 1410 edac_dbg(0, "mc route table for CHA %d: %s\n", 1411 cur_reg_start, mc_route_string); 1412 else 1413 edac_dbg(0, "mc route table for CHA %d-%d: %s\n", 1414 cur_reg_start, i-1, mc_route_string); 1415 1416 /* Process DRAM rules */ 1417 for (sad_rule = 0; sad_rule < pvt->info.max_sad; sad_rule++) { 1418 /* previous limit becomes the new base */ 1419 sad_base = sad_limit; 1420 1421 pci_read_config_dword(pvt->pci_sad0, 1422 pvt->info.dram_rule[sad_rule], &dram_rule); 1423 1424 if (!DRAM_RULE_ENABLE(dram_rule)) 1425 break; 1426 1427 edram_only = KNL_EDRAM_ONLY(dram_rule); 1428 1429 sad_limit = pvt->info.sad_limit(dram_rule)+1; 1430 sad_size = sad_limit - sad_base; 1431 1432 pci_read_config_dword(pvt->pci_sad0, 1433 pvt->info.interleave_list[sad_rule], &interleave_reg); 1434 1435 /* 1436 * Find out how many ways this dram rule is interleaved. 1437 * We stop when we see the first channel again. 1438 */ 1439 first_pkg = sad_pkg(pvt->info.interleave_pkg, 1440 interleave_reg, 0); 1441 for (intrlv_ways = 1; intrlv_ways < 8; intrlv_ways++) { 1442 pkg = sad_pkg(pvt->info.interleave_pkg, 1443 interleave_reg, intrlv_ways); 1444 1445 if ((pkg & 0x8) == 0) { 1446 /* 1447 * 0 bit means memory is non-local, 1448 * which KNL doesn't support 1449 */ 1450 edac_dbg(0, "Unexpected interleave target %d\n", 1451 pkg); 1452 return -1; 1453 } 1454 1455 if (pkg == first_pkg) 1456 break; 1457 } 1458 if (KNL_MOD3(dram_rule)) 1459 intrlv_ways *= 3; 1460 1461 edac_dbg(3, "dram rule %d (base 0x%llx, limit 0x%llx), %d way interleave%s\n", 1462 sad_rule, 1463 sad_base, 1464 sad_limit, 1465 intrlv_ways, 1466 edram_only ? ", EDRAM" : ""); 1467 1468 /* 1469 * Find out how big the SAD region really is by iterating 1470 * over TAD tables (SAD regions may contain holes). 1471 * Each memory controller might have a different TAD table, so 1472 * we have to look at both. 1473 * 1474 * Livespace is the memory that's mapped in this TAD table, 1475 * deadspace is the holes (this could be the MMIO hole, or it 1476 * could be memory that's mapped by the other TAD table but 1477 * not this one). 1478 */ 1479 for (mc = 0; mc < 2; mc++) { 1480 sad_actual_size[mc] = 0; 1481 tad_livespace = 0; 1482 for (tad_rule = 0; 1483 tad_rule < ARRAY_SIZE( 1484 knl_tad_dram_limit_lo); 1485 tad_rule++) { 1486 if (knl_get_tad(pvt, 1487 tad_rule, 1488 mc, 1489 &tad_deadspace, 1490 &tad_limit, 1491 &tad_ways)) 1492 break; 1493 1494 tad_size = (tad_limit+1) - 1495 (tad_livespace + tad_deadspace); 1496 tad_livespace += tad_size; 1497 tad_base = (tad_limit+1) - tad_size; 1498 1499 if (tad_base < sad_base) { 1500 if (tad_limit > sad_base) 1501 edac_dbg(0, "TAD region overlaps lower SAD boundary -- TAD tables may be configured incorrectly.\n"); 1502 } else if (tad_base < sad_limit) { 1503 if (tad_limit+1 > sad_limit) { 1504 edac_dbg(0, "TAD region overlaps upper SAD boundary -- TAD tables may be configured incorrectly.\n"); 1505 } else { 1506 /* TAD region is completely inside SAD region */ 1507 edac_dbg(3, "TAD region %d 0x%llx - 0x%llx (%lld bytes) table%d\n", 1508 tad_rule, tad_base, 1509 tad_limit, tad_size, 1510 mc); 1511 sad_actual_size[mc] += tad_size; 1512 } 1513 } 1514 } 1515 } 1516 1517 for (mc = 0; mc < 2; mc++) { 1518 edac_dbg(3, " total TAD DRAM footprint in table%d : 0x%llx (%lld bytes)\n", 1519 mc, sad_actual_size[mc], sad_actual_size[mc]); 1520 } 1521 1522 /* Ignore EDRAM rule */ 1523 if (edram_only) 1524 continue; 1525 1526 /* Figure out which channels participate in interleave. */ 1527 for (channel = 0; channel < KNL_MAX_CHANNELS; channel++) 1528 participants[channel] = 0; 1529 1530 /* For each channel, does at least one CHA have 1531 * this channel mapped to the given target? 1532 */ 1533 for (channel = 0; channel < KNL_MAX_CHANNELS; channel++) { 1534 int target; 1535 int cha; 1536 1537 for (target = 0; target < KNL_MAX_CHANNELS; target++) { 1538 for (cha = 0; cha < KNL_MAX_CHAS; cha++) { 1539 if (knl_get_mc_route(target, 1540 mc_route_reg[cha]) == channel 1541 && !participants[channel]) { 1542 participants[channel] = 1; 1543 break; 1544 } 1545 } 1546 } 1547 } 1548 1549 for (channel = 0; channel < KNL_MAX_CHANNELS; channel++) { 1550 mc = knl_channel_mc(channel); 1551 if (participants[channel]) { 1552 edac_dbg(4, "mc channel %d contributes %lld bytes via sad entry %d\n", 1553 channel, 1554 sad_actual_size[mc]/intrlv_ways, 1555 sad_rule); 1556 mc_sizes[channel] += 1557 sad_actual_size[mc]/intrlv_ways; 1558 } 1559 } 1560 } 1561 1562 return 0; 1563 } 1564 1565 static void get_source_id(struct mem_ctl_info *mci) 1566 { 1567 struct sbridge_pvt *pvt = mci->pvt_info; 1568 u32 reg; 1569 1570 if (pvt->info.type == HASWELL || pvt->info.type == BROADWELL || 1571 pvt->info.type == KNIGHTS_LANDING) 1572 pci_read_config_dword(pvt->pci_sad1, SAD_TARGET, ®); 1573 else 1574 pci_read_config_dword(pvt->pci_br0, SAD_TARGET, ®); 1575 1576 if (pvt->info.type == KNIGHTS_LANDING) 1577 pvt->sbridge_dev->source_id = SOURCE_ID_KNL(reg); 1578 else 1579 pvt->sbridge_dev->source_id = SOURCE_ID(reg); 1580 } 1581 1582 static int __populate_dimms(struct mem_ctl_info *mci, 1583 u64 knl_mc_sizes[KNL_MAX_CHANNELS], 1584 enum edac_type mode) 1585 { 1586 struct sbridge_pvt *pvt = mci->pvt_info; 1587 int channels = pvt->info.type == KNIGHTS_LANDING ? KNL_MAX_CHANNELS 1588 : NUM_CHANNELS; 1589 unsigned int i, j, banks, ranks, rows, cols, npages; 1590 struct dimm_info *dimm; 1591 enum mem_type mtype; 1592 u64 size; 1593 1594 mtype = pvt->info.get_memory_type(pvt); 1595 if (mtype == MEM_RDDR3 || mtype == MEM_RDDR4) 1596 edac_dbg(0, "Memory is registered\n"); 1597 else if (mtype == MEM_UNKNOWN) 1598 edac_dbg(0, "Cannot determine memory type\n"); 1599 else 1600 edac_dbg(0, "Memory is unregistered\n"); 1601 1602 if (mtype == MEM_DDR4 || mtype == MEM_RDDR4) 1603 banks = 16; 1604 else 1605 banks = 8; 1606 1607 for (i = 0; i < channels; i++) { 1608 u32 mtr; 1609 1610 int max_dimms_per_channel; 1611 1612 if (pvt->info.type == KNIGHTS_LANDING) { 1613 max_dimms_per_channel = 1; 1614 if (!pvt->knl.pci_channel[i]) 1615 continue; 1616 } else { 1617 max_dimms_per_channel = ARRAY_SIZE(mtr_regs); 1618 if (!pvt->pci_tad[i]) 1619 continue; 1620 } 1621 1622 for (j = 0; j < max_dimms_per_channel; j++) { 1623 dimm = EDAC_DIMM_PTR(mci->layers, mci->dimms, mci->n_layers, i, j, 0); 1624 if (pvt->info.type == KNIGHTS_LANDING) { 1625 pci_read_config_dword(pvt->knl.pci_channel[i], 1626 knl_mtr_reg, &mtr); 1627 } else { 1628 pci_read_config_dword(pvt->pci_tad[i], 1629 mtr_regs[j], &mtr); 1630 } 1631 edac_dbg(4, "Channel #%d MTR%d = %x\n", i, j, mtr); 1632 if (IS_DIMM_PRESENT(mtr)) { 1633 if (!IS_ECC_ENABLED(pvt->info.mcmtr)) { 1634 sbridge_printk(KERN_ERR, "CPU SrcID #%d, Ha #%d, Channel #%d has DIMMs, but ECC is disabled\n", 1635 pvt->sbridge_dev->source_id, 1636 pvt->sbridge_dev->dom, i); 1637 return -ENODEV; 1638 } 1639 pvt->channel[i].dimms++; 1640 1641 ranks = numrank(pvt->info.type, mtr); 1642 1643 if (pvt->info.type == KNIGHTS_LANDING) { 1644 /* For DDR4, this is fixed. */ 1645 cols = 1 << 10; 1646 rows = knl_mc_sizes[i] / 1647 ((u64) cols * ranks * banks * 8); 1648 } else { 1649 rows = numrow(mtr); 1650 cols = numcol(mtr); 1651 } 1652 1653 size = ((u64)rows * cols * banks * ranks) >> (20 - 3); 1654 npages = MiB_TO_PAGES(size); 1655 1656 edac_dbg(0, "mc#%d: ha %d channel %d, dimm %d, %lld MiB (%d pages) bank: %d, rank: %d, row: %#x, col: %#x\n", 1657 pvt->sbridge_dev->mc, pvt->sbridge_dev->dom, i, j, 1658 size, npages, 1659 banks, ranks, rows, cols); 1660 1661 dimm->nr_pages = npages; 1662 dimm->grain = 32; 1663 dimm->dtype = pvt->info.get_width(pvt, mtr); 1664 dimm->mtype = mtype; 1665 dimm->edac_mode = mode; 1666 snprintf(dimm->label, sizeof(dimm->label), 1667 "CPU_SrcID#%u_Ha#%u_Chan#%u_DIMM#%u", 1668 pvt->sbridge_dev->source_id, pvt->sbridge_dev->dom, i, j); 1669 } 1670 } 1671 } 1672 1673 return 0; 1674 } 1675 1676 static int get_dimm_config(struct mem_ctl_info *mci) 1677 { 1678 struct sbridge_pvt *pvt = mci->pvt_info; 1679 u64 knl_mc_sizes[KNL_MAX_CHANNELS]; 1680 enum edac_type mode; 1681 u32 reg; 1682 1683 pvt->sbridge_dev->node_id = pvt->info.get_node_id(pvt); 1684 edac_dbg(0, "mc#%d: Node ID: %d, source ID: %d\n", 1685 pvt->sbridge_dev->mc, 1686 pvt->sbridge_dev->node_id, 1687 pvt->sbridge_dev->source_id); 1688 1689 /* KNL doesn't support mirroring or lockstep, 1690 * and is always closed page 1691 */ 1692 if (pvt->info.type == KNIGHTS_LANDING) { 1693 mode = EDAC_S4ECD4ED; 1694 pvt->mirror_mode = NON_MIRRORING; 1695 pvt->is_cur_addr_mirrored = false; 1696 1697 if (knl_get_dimm_capacity(pvt, knl_mc_sizes) != 0) 1698 return -1; 1699 if (pci_read_config_dword(pvt->pci_ta, KNL_MCMTR, &pvt->info.mcmtr)) { 1700 edac_dbg(0, "Failed to read KNL_MCMTR register\n"); 1701 return -ENODEV; 1702 } 1703 } else { 1704 if (pvt->info.type == HASWELL || pvt->info.type == BROADWELL) { 1705 if (pci_read_config_dword(pvt->pci_ha, HASWELL_HASYSDEFEATURE2, ®)) { 1706 edac_dbg(0, "Failed to read HASWELL_HASYSDEFEATURE2 register\n"); 1707 return -ENODEV; 1708 } 1709 pvt->is_chan_hash = GET_BITFIELD(reg, 21, 21); 1710 if (GET_BITFIELD(reg, 28, 28)) { 1711 pvt->mirror_mode = ADDR_RANGE_MIRRORING; 1712 edac_dbg(0, "Address range partial memory mirroring is enabled\n"); 1713 goto next; 1714 } 1715 } 1716 if (pci_read_config_dword(pvt->pci_ras, RASENABLES, ®)) { 1717 edac_dbg(0, "Failed to read RASENABLES register\n"); 1718 return -ENODEV; 1719 } 1720 if (IS_MIRROR_ENABLED(reg)) { 1721 pvt->mirror_mode = FULL_MIRRORING; 1722 edac_dbg(0, "Full memory mirroring is enabled\n"); 1723 } else { 1724 pvt->mirror_mode = NON_MIRRORING; 1725 edac_dbg(0, "Memory mirroring is disabled\n"); 1726 } 1727 1728 next: 1729 if (pci_read_config_dword(pvt->pci_ta, MCMTR, &pvt->info.mcmtr)) { 1730 edac_dbg(0, "Failed to read MCMTR register\n"); 1731 return -ENODEV; 1732 } 1733 if (IS_LOCKSTEP_ENABLED(pvt->info.mcmtr)) { 1734 edac_dbg(0, "Lockstep is enabled\n"); 1735 mode = EDAC_S8ECD8ED; 1736 pvt->is_lockstep = true; 1737 } else { 1738 edac_dbg(0, "Lockstep is disabled\n"); 1739 mode = EDAC_S4ECD4ED; 1740 pvt->is_lockstep = false; 1741 } 1742 if (IS_CLOSE_PG(pvt->info.mcmtr)) { 1743 edac_dbg(0, "address map is on closed page mode\n"); 1744 pvt->is_close_pg = true; 1745 } else { 1746 edac_dbg(0, "address map is on open page mode\n"); 1747 pvt->is_close_pg = false; 1748 } 1749 } 1750 1751 return __populate_dimms(mci, knl_mc_sizes, mode); 1752 } 1753 1754 static void get_memory_layout(const struct mem_ctl_info *mci) 1755 { 1756 struct sbridge_pvt *pvt = mci->pvt_info; 1757 int i, j, k, n_sads, n_tads, sad_interl; 1758 u32 reg; 1759 u64 limit, prv = 0; 1760 u64 tmp_mb; 1761 u32 gb, mb; 1762 u32 rir_way; 1763 1764 /* 1765 * Step 1) Get TOLM/TOHM ranges 1766 */ 1767 1768 pvt->tolm = pvt->info.get_tolm(pvt); 1769 tmp_mb = (1 + pvt->tolm) >> 20; 1770 1771 gb = div_u64_rem(tmp_mb, 1024, &mb); 1772 edac_dbg(0, "TOLM: %u.%03u GB (0x%016Lx)\n", 1773 gb, (mb*1000)/1024, (u64)pvt->tolm); 1774 1775 /* Address range is already 45:25 */ 1776 pvt->tohm = pvt->info.get_tohm(pvt); 1777 tmp_mb = (1 + pvt->tohm) >> 20; 1778 1779 gb = div_u64_rem(tmp_mb, 1024, &mb); 1780 edac_dbg(0, "TOHM: %u.%03u GB (0x%016Lx)\n", 1781 gb, (mb*1000)/1024, (u64)pvt->tohm); 1782 1783 /* 1784 * Step 2) Get SAD range and SAD Interleave list 1785 * TAD registers contain the interleave wayness. However, it 1786 * seems simpler to just discover it indirectly, with the 1787 * algorithm bellow. 1788 */ 1789 prv = 0; 1790 for (n_sads = 0; n_sads < pvt->info.max_sad; n_sads++) { 1791 /* SAD_LIMIT Address range is 45:26 */ 1792 pci_read_config_dword(pvt->pci_sad0, pvt->info.dram_rule[n_sads], 1793 ®); 1794 limit = pvt->info.sad_limit(reg); 1795 1796 if (!DRAM_RULE_ENABLE(reg)) 1797 continue; 1798 1799 if (limit <= prv) 1800 break; 1801 1802 tmp_mb = (limit + 1) >> 20; 1803 gb = div_u64_rem(tmp_mb, 1024, &mb); 1804 edac_dbg(0, "SAD#%d %s up to %u.%03u GB (0x%016Lx) Interleave: %s reg=0x%08x\n", 1805 n_sads, 1806 show_dram_attr(pvt->info.dram_attr(reg)), 1807 gb, (mb*1000)/1024, 1808 ((u64)tmp_mb) << 20L, 1809 get_intlv_mode_str(reg, pvt->info.type), 1810 reg); 1811 prv = limit; 1812 1813 pci_read_config_dword(pvt->pci_sad0, pvt->info.interleave_list[n_sads], 1814 ®); 1815 sad_interl = sad_pkg(pvt->info.interleave_pkg, reg, 0); 1816 for (j = 0; j < 8; j++) { 1817 u32 pkg = sad_pkg(pvt->info.interleave_pkg, reg, j); 1818 if (j > 0 && sad_interl == pkg) 1819 break; 1820 1821 edac_dbg(0, "SAD#%d, interleave #%d: %d\n", 1822 n_sads, j, pkg); 1823 } 1824 } 1825 1826 if (pvt->info.type == KNIGHTS_LANDING) 1827 return; 1828 1829 /* 1830 * Step 3) Get TAD range 1831 */ 1832 prv = 0; 1833 for (n_tads = 0; n_tads < MAX_TAD; n_tads++) { 1834 pci_read_config_dword(pvt->pci_ha, tad_dram_rule[n_tads], ®); 1835 limit = TAD_LIMIT(reg); 1836 if (limit <= prv) 1837 break; 1838 tmp_mb = (limit + 1) >> 20; 1839 1840 gb = div_u64_rem(tmp_mb, 1024, &mb); 1841 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", 1842 n_tads, gb, (mb*1000)/1024, 1843 ((u64)tmp_mb) << 20L, 1844 (u32)(1 << TAD_SOCK(reg)), 1845 (u32)TAD_CH(reg) + 1, 1846 (u32)TAD_TGT0(reg), 1847 (u32)TAD_TGT1(reg), 1848 (u32)TAD_TGT2(reg), 1849 (u32)TAD_TGT3(reg), 1850 reg); 1851 prv = limit; 1852 } 1853 1854 /* 1855 * Step 4) Get TAD offsets, per each channel 1856 */ 1857 for (i = 0; i < NUM_CHANNELS; i++) { 1858 if (!pvt->channel[i].dimms) 1859 continue; 1860 for (j = 0; j < n_tads; j++) { 1861 pci_read_config_dword(pvt->pci_tad[i], 1862 tad_ch_nilv_offset[j], 1863 ®); 1864 tmp_mb = TAD_OFFSET(reg) >> 20; 1865 gb = div_u64_rem(tmp_mb, 1024, &mb); 1866 edac_dbg(0, "TAD CH#%d, offset #%d: %u.%03u GB (0x%016Lx), reg=0x%08x\n", 1867 i, j, 1868 gb, (mb*1000)/1024, 1869 ((u64)tmp_mb) << 20L, 1870 reg); 1871 } 1872 } 1873 1874 /* 1875 * Step 6) Get RIR Wayness/Limit, per each channel 1876 */ 1877 for (i = 0; i < NUM_CHANNELS; i++) { 1878 if (!pvt->channel[i].dimms) 1879 continue; 1880 for (j = 0; j < MAX_RIR_RANGES; j++) { 1881 pci_read_config_dword(pvt->pci_tad[i], 1882 rir_way_limit[j], 1883 ®); 1884 1885 if (!IS_RIR_VALID(reg)) 1886 continue; 1887 1888 tmp_mb = pvt->info.rir_limit(reg) >> 20; 1889 rir_way = 1 << RIR_WAY(reg); 1890 gb = div_u64_rem(tmp_mb, 1024, &mb); 1891 edac_dbg(0, "CH#%d RIR#%d, limit: %u.%03u GB (0x%016Lx), way: %d, reg=0x%08x\n", 1892 i, j, 1893 gb, (mb*1000)/1024, 1894 ((u64)tmp_mb) << 20L, 1895 rir_way, 1896 reg); 1897 1898 for (k = 0; k < rir_way; k++) { 1899 pci_read_config_dword(pvt->pci_tad[i], 1900 rir_offset[j][k], 1901 ®); 1902 tmp_mb = RIR_OFFSET(pvt->info.type, reg) << 6; 1903 1904 gb = div_u64_rem(tmp_mb, 1024, &mb); 1905 edac_dbg(0, "CH#%d RIR#%d INTL#%d, offset %u.%03u GB (0x%016Lx), tgt: %d, reg=0x%08x\n", 1906 i, j, k, 1907 gb, (mb*1000)/1024, 1908 ((u64)tmp_mb) << 20L, 1909 (u32)RIR_RNK_TGT(pvt->info.type, reg), 1910 reg); 1911 } 1912 } 1913 } 1914 } 1915 1916 static struct mem_ctl_info *get_mci_for_node_id(u8 node_id, u8 ha) 1917 { 1918 struct sbridge_dev *sbridge_dev; 1919 1920 list_for_each_entry(sbridge_dev, &sbridge_edac_list, list) { 1921 if (sbridge_dev->node_id == node_id && sbridge_dev->dom == ha) 1922 return sbridge_dev->mci; 1923 } 1924 return NULL; 1925 } 1926 1927 static int get_memory_error_data(struct mem_ctl_info *mci, 1928 u64 addr, 1929 u8 *socket, u8 *ha, 1930 long *channel_mask, 1931 u8 *rank, 1932 char **area_type, char *msg) 1933 { 1934 struct mem_ctl_info *new_mci; 1935 struct sbridge_pvt *pvt = mci->pvt_info; 1936 struct pci_dev *pci_ha; 1937 int n_rir, n_sads, n_tads, sad_way, sck_xch; 1938 int sad_interl, idx, base_ch; 1939 int interleave_mode, shiftup = 0; 1940 unsigned int sad_interleave[MAX_INTERLEAVE]; 1941 u32 reg, dram_rule; 1942 u8 ch_way, sck_way, pkg, sad_ha = 0; 1943 u32 tad_offset; 1944 u32 rir_way; 1945 u32 mb, gb; 1946 u64 ch_addr, offset, limit = 0, prv = 0; 1947 1948 1949 /* 1950 * Step 0) Check if the address is at special memory ranges 1951 * The check bellow is probably enough to fill all cases where 1952 * the error is not inside a memory, except for the legacy 1953 * range (e. g. VGA addresses). It is unlikely, however, that the 1954 * memory controller would generate an error on that range. 1955 */ 1956 if ((addr > (u64) pvt->tolm) && (addr < (1LL << 32))) { 1957 sprintf(msg, "Error at TOLM area, on addr 0x%08Lx", addr); 1958 return -EINVAL; 1959 } 1960 if (addr >= (u64)pvt->tohm) { 1961 sprintf(msg, "Error at MMIOH area, on addr 0x%016Lx", addr); 1962 return -EINVAL; 1963 } 1964 1965 /* 1966 * Step 1) Get socket 1967 */ 1968 for (n_sads = 0; n_sads < pvt->info.max_sad; n_sads++) { 1969 pci_read_config_dword(pvt->pci_sad0, pvt->info.dram_rule[n_sads], 1970 ®); 1971 1972 if (!DRAM_RULE_ENABLE(reg)) 1973 continue; 1974 1975 limit = pvt->info.sad_limit(reg); 1976 if (limit <= prv) { 1977 sprintf(msg, "Can't discover the memory socket"); 1978 return -EINVAL; 1979 } 1980 if (addr <= limit) 1981 break; 1982 prv = limit; 1983 } 1984 if (n_sads == pvt->info.max_sad) { 1985 sprintf(msg, "Can't discover the memory socket"); 1986 return -EINVAL; 1987 } 1988 dram_rule = reg; 1989 *area_type = show_dram_attr(pvt->info.dram_attr(dram_rule)); 1990 interleave_mode = pvt->info.interleave_mode(dram_rule); 1991 1992 pci_read_config_dword(pvt->pci_sad0, pvt->info.interleave_list[n_sads], 1993 ®); 1994 1995 if (pvt->info.type == SANDY_BRIDGE) { 1996 sad_interl = sad_pkg(pvt->info.interleave_pkg, reg, 0); 1997 for (sad_way = 0; sad_way < 8; sad_way++) { 1998 u32 pkg = sad_pkg(pvt->info.interleave_pkg, reg, sad_way); 1999 if (sad_way > 0 && sad_interl == pkg) 2000 break; 2001 sad_interleave[sad_way] = pkg; 2002 edac_dbg(0, "SAD interleave #%d: %d\n", 2003 sad_way, sad_interleave[sad_way]); 2004 } 2005 edac_dbg(0, "mc#%d: Error detected on SAD#%d: address 0x%016Lx < 0x%016Lx, Interleave [%d:6]%s\n", 2006 pvt->sbridge_dev->mc, 2007 n_sads, 2008 addr, 2009 limit, 2010 sad_way + 7, 2011 !interleave_mode ? "" : "XOR[18:16]"); 2012 if (interleave_mode) 2013 idx = ((addr >> 6) ^ (addr >> 16)) & 7; 2014 else 2015 idx = (addr >> 6) & 7; 2016 switch (sad_way) { 2017 case 1: 2018 idx = 0; 2019 break; 2020 case 2: 2021 idx = idx & 1; 2022 break; 2023 case 4: 2024 idx = idx & 3; 2025 break; 2026 case 8: 2027 break; 2028 default: 2029 sprintf(msg, "Can't discover socket interleave"); 2030 return -EINVAL; 2031 } 2032 *socket = sad_interleave[idx]; 2033 edac_dbg(0, "SAD interleave index: %d (wayness %d) = CPU socket %d\n", 2034 idx, sad_way, *socket); 2035 } else if (pvt->info.type == HASWELL || pvt->info.type == BROADWELL) { 2036 int bits, a7mode = A7MODE(dram_rule); 2037 2038 if (a7mode) { 2039 /* A7 mode swaps P9 with P6 */ 2040 bits = GET_BITFIELD(addr, 7, 8) << 1; 2041 bits |= GET_BITFIELD(addr, 9, 9); 2042 } else 2043 bits = GET_BITFIELD(addr, 6, 8); 2044 2045 if (interleave_mode == 0) { 2046 /* interleave mode will XOR {8,7,6} with {18,17,16} */ 2047 idx = GET_BITFIELD(addr, 16, 18); 2048 idx ^= bits; 2049 } else 2050 idx = bits; 2051 2052 pkg = sad_pkg(pvt->info.interleave_pkg, reg, idx); 2053 *socket = sad_pkg_socket(pkg); 2054 sad_ha = sad_pkg_ha(pkg); 2055 2056 if (a7mode) { 2057 /* MCChanShiftUpEnable */ 2058 pci_read_config_dword(pvt->pci_ha, HASWELL_HASYSDEFEATURE2, ®); 2059 shiftup = GET_BITFIELD(reg, 22, 22); 2060 } 2061 2062 edac_dbg(0, "SAD interleave package: %d = CPU socket %d, HA %i, shiftup: %i\n", 2063 idx, *socket, sad_ha, shiftup); 2064 } else { 2065 /* Ivy Bridge's SAD mode doesn't support XOR interleave mode */ 2066 idx = (addr >> 6) & 7; 2067 pkg = sad_pkg(pvt->info.interleave_pkg, reg, idx); 2068 *socket = sad_pkg_socket(pkg); 2069 sad_ha = sad_pkg_ha(pkg); 2070 edac_dbg(0, "SAD interleave package: %d = CPU socket %d, HA %d\n", 2071 idx, *socket, sad_ha); 2072 } 2073 2074 *ha = sad_ha; 2075 2076 /* 2077 * Move to the proper node structure, in order to access the 2078 * right PCI registers 2079 */ 2080 new_mci = get_mci_for_node_id(*socket, sad_ha); 2081 if (!new_mci) { 2082 sprintf(msg, "Struct for socket #%u wasn't initialized", 2083 *socket); 2084 return -EINVAL; 2085 } 2086 mci = new_mci; 2087 pvt = mci->pvt_info; 2088 2089 /* 2090 * Step 2) Get memory channel 2091 */ 2092 prv = 0; 2093 pci_ha = pvt->pci_ha; 2094 for (n_tads = 0; n_tads < MAX_TAD; n_tads++) { 2095 pci_read_config_dword(pci_ha, tad_dram_rule[n_tads], ®); 2096 limit = TAD_LIMIT(reg); 2097 if (limit <= prv) { 2098 sprintf(msg, "Can't discover the memory channel"); 2099 return -EINVAL; 2100 } 2101 if (addr <= limit) 2102 break; 2103 prv = limit; 2104 } 2105 if (n_tads == MAX_TAD) { 2106 sprintf(msg, "Can't discover the memory channel"); 2107 return -EINVAL; 2108 } 2109 2110 ch_way = TAD_CH(reg) + 1; 2111 sck_way = TAD_SOCK(reg); 2112 2113 if (ch_way == 3) 2114 idx = addr >> 6; 2115 else { 2116 idx = (addr >> (6 + sck_way + shiftup)) & 0x3; 2117 if (pvt->is_chan_hash) 2118 idx = haswell_chan_hash(idx, addr); 2119 } 2120 idx = idx % ch_way; 2121 2122 /* 2123 * FIXME: Shouldn't we use CHN_IDX_OFFSET() here, when ch_way == 3 ??? 2124 */ 2125 switch (idx) { 2126 case 0: 2127 base_ch = TAD_TGT0(reg); 2128 break; 2129 case 1: 2130 base_ch = TAD_TGT1(reg); 2131 break; 2132 case 2: 2133 base_ch = TAD_TGT2(reg); 2134 break; 2135 case 3: 2136 base_ch = TAD_TGT3(reg); 2137 break; 2138 default: 2139 sprintf(msg, "Can't discover the TAD target"); 2140 return -EINVAL; 2141 } 2142 *channel_mask = 1 << base_ch; 2143 2144 pci_read_config_dword(pvt->pci_tad[base_ch], tad_ch_nilv_offset[n_tads], &tad_offset); 2145 2146 if (pvt->mirror_mode == FULL_MIRRORING || 2147 (pvt->mirror_mode == ADDR_RANGE_MIRRORING && n_tads == 0)) { 2148 *channel_mask |= 1 << ((base_ch + 2) % 4); 2149 switch(ch_way) { 2150 case 2: 2151 case 4: 2152 sck_xch = (1 << sck_way) * (ch_way >> 1); 2153 break; 2154 default: 2155 sprintf(msg, "Invalid mirror set. Can't decode addr"); 2156 return -EINVAL; 2157 } 2158 2159 pvt->is_cur_addr_mirrored = true; 2160 } else { 2161 sck_xch = (1 << sck_way) * ch_way; 2162 pvt->is_cur_addr_mirrored = false; 2163 } 2164 2165 if (pvt->is_lockstep) 2166 *channel_mask |= 1 << ((base_ch + 1) % 4); 2167 2168 offset = TAD_OFFSET(tad_offset); 2169 2170 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", 2171 n_tads, 2172 addr, 2173 limit, 2174 sck_way, 2175 ch_way, 2176 offset, 2177 idx, 2178 base_ch, 2179 *channel_mask); 2180 2181 /* Calculate channel address */ 2182 /* Remove the TAD offset */ 2183 2184 if (offset > addr) { 2185 sprintf(msg, "Can't calculate ch addr: TAD offset 0x%08Lx is too high for addr 0x%08Lx!", 2186 offset, addr); 2187 return -EINVAL; 2188 } 2189 2190 ch_addr = addr - offset; 2191 ch_addr >>= (6 + shiftup); 2192 ch_addr /= sck_xch; 2193 ch_addr <<= (6 + shiftup); 2194 ch_addr |= addr & ((1 << (6 + shiftup)) - 1); 2195 2196 /* 2197 * Step 3) Decode rank 2198 */ 2199 for (n_rir = 0; n_rir < MAX_RIR_RANGES; n_rir++) { 2200 pci_read_config_dword(pvt->pci_tad[base_ch], rir_way_limit[n_rir], ®); 2201 2202 if (!IS_RIR_VALID(reg)) 2203 continue; 2204 2205 limit = pvt->info.rir_limit(reg); 2206 gb = div_u64_rem(limit >> 20, 1024, &mb); 2207 edac_dbg(0, "RIR#%d, limit: %u.%03u GB (0x%016Lx), way: %d\n", 2208 n_rir, 2209 gb, (mb*1000)/1024, 2210 limit, 2211 1 << RIR_WAY(reg)); 2212 if (ch_addr <= limit) 2213 break; 2214 } 2215 if (n_rir == MAX_RIR_RANGES) { 2216 sprintf(msg, "Can't discover the memory rank for ch addr 0x%08Lx", 2217 ch_addr); 2218 return -EINVAL; 2219 } 2220 rir_way = RIR_WAY(reg); 2221 2222 if (pvt->is_close_pg) 2223 idx = (ch_addr >> 6); 2224 else 2225 idx = (ch_addr >> 13); /* FIXME: Datasheet says to shift by 15 */ 2226 idx %= 1 << rir_way; 2227 2228 pci_read_config_dword(pvt->pci_tad[base_ch], rir_offset[n_rir][idx], ®); 2229 *rank = RIR_RNK_TGT(pvt->info.type, reg); 2230 2231 edac_dbg(0, "RIR#%d: channel address 0x%08Lx < 0x%08Lx, RIR interleave %d, index %d\n", 2232 n_rir, 2233 ch_addr, 2234 limit, 2235 rir_way, 2236 idx); 2237 2238 return 0; 2239 } 2240 2241 static int get_memory_error_data_from_mce(struct mem_ctl_info *mci, 2242 const struct mce *m, u8 *socket, 2243 u8 *ha, long *channel_mask, 2244 char *msg) 2245 { 2246 u32 reg, channel = GET_BITFIELD(m->status, 0, 3); 2247 struct mem_ctl_info *new_mci; 2248 struct sbridge_pvt *pvt; 2249 struct pci_dev *pci_ha; 2250 bool tad0; 2251 2252 if (channel >= NUM_CHANNELS) { 2253 sprintf(msg, "Invalid channel 0x%x", channel); 2254 return -EINVAL; 2255 } 2256 2257 pvt = mci->pvt_info; 2258 if (!pvt->info.get_ha) { 2259 sprintf(msg, "No get_ha()"); 2260 return -EINVAL; 2261 } 2262 *ha = pvt->info.get_ha(m->bank); 2263 if (*ha != 0 && *ha != 1) { 2264 sprintf(msg, "Impossible bank %d", m->bank); 2265 return -EINVAL; 2266 } 2267 2268 *socket = m->socketid; 2269 new_mci = get_mci_for_node_id(*socket, *ha); 2270 if (!new_mci) { 2271 strcpy(msg, "mci socket got corrupted!"); 2272 return -EINVAL; 2273 } 2274 2275 pvt = new_mci->pvt_info; 2276 pci_ha = pvt->pci_ha; 2277 pci_read_config_dword(pci_ha, tad_dram_rule[0], ®); 2278 tad0 = m->addr <= TAD_LIMIT(reg); 2279 2280 *channel_mask = 1 << channel; 2281 if (pvt->mirror_mode == FULL_MIRRORING || 2282 (pvt->mirror_mode == ADDR_RANGE_MIRRORING && tad0)) { 2283 *channel_mask |= 1 << ((channel + 2) % 4); 2284 pvt->is_cur_addr_mirrored = true; 2285 } else { 2286 pvt->is_cur_addr_mirrored = false; 2287 } 2288 2289 if (pvt->is_lockstep) 2290 *channel_mask |= 1 << ((channel + 1) % 4); 2291 2292 return 0; 2293 } 2294 2295 /**************************************************************************** 2296 Device initialization routines: put/get, init/exit 2297 ****************************************************************************/ 2298 2299 /* 2300 * sbridge_put_all_devices 'put' all the devices that we have 2301 * reserved via 'get' 2302 */ 2303 static void sbridge_put_devices(struct sbridge_dev *sbridge_dev) 2304 { 2305 int i; 2306 2307 edac_dbg(0, "\n"); 2308 for (i = 0; i < sbridge_dev->n_devs; i++) { 2309 struct pci_dev *pdev = sbridge_dev->pdev[i]; 2310 if (!pdev) 2311 continue; 2312 edac_dbg(0, "Removing dev %02x:%02x.%d\n", 2313 pdev->bus->number, 2314 PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn)); 2315 pci_dev_put(pdev); 2316 } 2317 } 2318 2319 static void sbridge_put_all_devices(void) 2320 { 2321 struct sbridge_dev *sbridge_dev, *tmp; 2322 2323 list_for_each_entry_safe(sbridge_dev, tmp, &sbridge_edac_list, list) { 2324 sbridge_put_devices(sbridge_dev); 2325 free_sbridge_dev(sbridge_dev); 2326 } 2327 } 2328 2329 static int sbridge_get_onedevice(struct pci_dev **prev, 2330 u8 *num_mc, 2331 const struct pci_id_table *table, 2332 const unsigned devno, 2333 const int multi_bus) 2334 { 2335 struct sbridge_dev *sbridge_dev = NULL; 2336 const struct pci_id_descr *dev_descr = &table->descr[devno]; 2337 struct pci_dev *pdev = NULL; 2338 int seg = 0; 2339 u8 bus = 0; 2340 int i = 0; 2341 2342 sbridge_printk(KERN_DEBUG, 2343 "Seeking for: PCI ID %04x:%04x\n", 2344 PCI_VENDOR_ID_INTEL, dev_descr->dev_id); 2345 2346 pdev = pci_get_device(PCI_VENDOR_ID_INTEL, 2347 dev_descr->dev_id, *prev); 2348 2349 if (!pdev) { 2350 if (*prev) { 2351 *prev = pdev; 2352 return 0; 2353 } 2354 2355 if (dev_descr->optional) 2356 return 0; 2357 2358 /* if the HA wasn't found */ 2359 if (devno == 0) 2360 return -ENODEV; 2361 2362 sbridge_printk(KERN_INFO, 2363 "Device not found: %04x:%04x\n", 2364 PCI_VENDOR_ID_INTEL, dev_descr->dev_id); 2365 2366 /* End of list, leave */ 2367 return -ENODEV; 2368 } 2369 seg = pci_domain_nr(pdev->bus); 2370 bus = pdev->bus->number; 2371 2372 next_imc: 2373 sbridge_dev = get_sbridge_dev(seg, bus, dev_descr->dom, 2374 multi_bus, sbridge_dev); 2375 if (!sbridge_dev) { 2376 /* If the HA1 wasn't found, don't create EDAC second memory controller */ 2377 if (dev_descr->dom == IMC1 && devno != 1) { 2378 edac_dbg(0, "Skip IMC1: %04x:%04x (since HA1 was absent)\n", 2379 PCI_VENDOR_ID_INTEL, dev_descr->dev_id); 2380 pci_dev_put(pdev); 2381 return 0; 2382 } 2383 2384 if (dev_descr->dom == SOCK) 2385 goto out_imc; 2386 2387 sbridge_dev = alloc_sbridge_dev(seg, bus, dev_descr->dom, table); 2388 if (!sbridge_dev) { 2389 pci_dev_put(pdev); 2390 return -ENOMEM; 2391 } 2392 (*num_mc)++; 2393 } 2394 2395 if (sbridge_dev->pdev[sbridge_dev->i_devs]) { 2396 sbridge_printk(KERN_ERR, 2397 "Duplicated device for %04x:%04x\n", 2398 PCI_VENDOR_ID_INTEL, dev_descr->dev_id); 2399 pci_dev_put(pdev); 2400 return -ENODEV; 2401 } 2402 2403 sbridge_dev->pdev[sbridge_dev->i_devs++] = pdev; 2404 2405 /* pdev belongs to more than one IMC, do extra gets */ 2406 if (++i > 1) 2407 pci_dev_get(pdev); 2408 2409 if (dev_descr->dom == SOCK && i < table->n_imcs_per_sock) 2410 goto next_imc; 2411 2412 out_imc: 2413 /* Be sure that the device is enabled */ 2414 if (unlikely(pci_enable_device(pdev) < 0)) { 2415 sbridge_printk(KERN_ERR, 2416 "Couldn't enable %04x:%04x\n", 2417 PCI_VENDOR_ID_INTEL, dev_descr->dev_id); 2418 return -ENODEV; 2419 } 2420 2421 edac_dbg(0, "Detected %04x:%04x\n", 2422 PCI_VENDOR_ID_INTEL, dev_descr->dev_id); 2423 2424 /* 2425 * As stated on drivers/pci/search.c, the reference count for 2426 * @from is always decremented if it is not %NULL. So, as we need 2427 * to get all devices up to null, we need to do a get for the device 2428 */ 2429 pci_dev_get(pdev); 2430 2431 *prev = pdev; 2432 2433 return 0; 2434 } 2435 2436 /* 2437 * sbridge_get_all_devices - Find and perform 'get' operation on the MCH's 2438 * devices we want to reference for this driver. 2439 * @num_mc: pointer to the memory controllers count, to be incremented in case 2440 * of success. 2441 * @table: model specific table 2442 * 2443 * returns 0 in case of success or error code 2444 */ 2445 static int sbridge_get_all_devices(u8 *num_mc, 2446 const struct pci_id_table *table) 2447 { 2448 int i, rc; 2449 struct pci_dev *pdev = NULL; 2450 int allow_dups = 0; 2451 int multi_bus = 0; 2452 2453 if (table->type == KNIGHTS_LANDING) 2454 allow_dups = multi_bus = 1; 2455 while (table && table->descr) { 2456 for (i = 0; i < table->n_devs_per_sock; i++) { 2457 if (!allow_dups || i == 0 || 2458 table->descr[i].dev_id != 2459 table->descr[i-1].dev_id) { 2460 pdev = NULL; 2461 } 2462 do { 2463 rc = sbridge_get_onedevice(&pdev, num_mc, 2464 table, i, multi_bus); 2465 if (rc < 0) { 2466 if (i == 0) { 2467 i = table->n_devs_per_sock; 2468 break; 2469 } 2470 sbridge_put_all_devices(); 2471 return -ENODEV; 2472 } 2473 } while (pdev && !allow_dups); 2474 } 2475 table++; 2476 } 2477 2478 return 0; 2479 } 2480 2481 /* 2482 * Device IDs for {SBRIDGE,IBRIDGE,HASWELL,BROADWELL}_IMC_HA0_TAD0 are in 2483 * the format: XXXa. So we can convert from a device to the corresponding 2484 * channel like this 2485 */ 2486 #define TAD_DEV_TO_CHAN(dev) (((dev) & 0xf) - 0xa) 2487 2488 static int sbridge_mci_bind_devs(struct mem_ctl_info *mci, 2489 struct sbridge_dev *sbridge_dev) 2490 { 2491 struct sbridge_pvt *pvt = mci->pvt_info; 2492 struct pci_dev *pdev; 2493 u8 saw_chan_mask = 0; 2494 int i; 2495 2496 for (i = 0; i < sbridge_dev->n_devs; i++) { 2497 pdev = sbridge_dev->pdev[i]; 2498 if (!pdev) 2499 continue; 2500 2501 switch (pdev->device) { 2502 case PCI_DEVICE_ID_INTEL_SBRIDGE_SAD0: 2503 pvt->pci_sad0 = pdev; 2504 break; 2505 case PCI_DEVICE_ID_INTEL_SBRIDGE_SAD1: 2506 pvt->pci_sad1 = pdev; 2507 break; 2508 case PCI_DEVICE_ID_INTEL_SBRIDGE_BR: 2509 pvt->pci_br0 = pdev; 2510 break; 2511 case PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_HA0: 2512 pvt->pci_ha = pdev; 2513 break; 2514 case PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TA: 2515 pvt->pci_ta = pdev; 2516 break; 2517 case PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_RAS: 2518 pvt->pci_ras = pdev; 2519 break; 2520 case PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD0: 2521 case PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD1: 2522 case PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD2: 2523 case PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD3: 2524 { 2525 int id = TAD_DEV_TO_CHAN(pdev->device); 2526 pvt->pci_tad[id] = pdev; 2527 saw_chan_mask |= 1 << id; 2528 } 2529 break; 2530 case PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_DDRIO: 2531 pvt->pci_ddrio = pdev; 2532 break; 2533 default: 2534 goto error; 2535 } 2536 2537 edac_dbg(0, "Associated PCI %02x:%02x, bus %d with dev = %p\n", 2538 pdev->vendor, pdev->device, 2539 sbridge_dev->bus, 2540 pdev); 2541 } 2542 2543 /* Check if everything were registered */ 2544 if (!pvt->pci_sad0 || !pvt->pci_sad1 || !pvt->pci_ha || 2545 !pvt->pci_ras || !pvt->pci_ta) 2546 goto enodev; 2547 2548 if (saw_chan_mask != 0x0f) 2549 goto enodev; 2550 return 0; 2551 2552 enodev: 2553 sbridge_printk(KERN_ERR, "Some needed devices are missing\n"); 2554 return -ENODEV; 2555 2556 error: 2557 sbridge_printk(KERN_ERR, "Unexpected device %02x:%02x\n", 2558 PCI_VENDOR_ID_INTEL, pdev->device); 2559 return -EINVAL; 2560 } 2561 2562 static int ibridge_mci_bind_devs(struct mem_ctl_info *mci, 2563 struct sbridge_dev *sbridge_dev) 2564 { 2565 struct sbridge_pvt *pvt = mci->pvt_info; 2566 struct pci_dev *pdev; 2567 u8 saw_chan_mask = 0; 2568 int i; 2569 2570 for (i = 0; i < sbridge_dev->n_devs; i++) { 2571 pdev = sbridge_dev->pdev[i]; 2572 if (!pdev) 2573 continue; 2574 2575 switch (pdev->device) { 2576 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0: 2577 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1: 2578 pvt->pci_ha = pdev; 2579 break; 2580 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TA: 2581 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TA: 2582 pvt->pci_ta = pdev; 2583 break; 2584 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_RAS: 2585 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_RAS: 2586 pvt->pci_ras = pdev; 2587 break; 2588 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD0: 2589 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD1: 2590 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD2: 2591 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD3: 2592 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD0: 2593 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD1: 2594 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD2: 2595 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD3: 2596 { 2597 int id = TAD_DEV_TO_CHAN(pdev->device); 2598 pvt->pci_tad[id] = pdev; 2599 saw_chan_mask |= 1 << id; 2600 } 2601 break; 2602 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_2HA_DDRIO0: 2603 pvt->pci_ddrio = pdev; 2604 break; 2605 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_1HA_DDRIO0: 2606 pvt->pci_ddrio = pdev; 2607 break; 2608 case PCI_DEVICE_ID_INTEL_IBRIDGE_SAD: 2609 pvt->pci_sad0 = pdev; 2610 break; 2611 case PCI_DEVICE_ID_INTEL_IBRIDGE_BR0: 2612 pvt->pci_br0 = pdev; 2613 break; 2614 case PCI_DEVICE_ID_INTEL_IBRIDGE_BR1: 2615 pvt->pci_br1 = pdev; 2616 break; 2617 default: 2618 goto error; 2619 } 2620 2621 edac_dbg(0, "Associated PCI %02x.%02d.%d with dev = %p\n", 2622 sbridge_dev->bus, 2623 PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn), 2624 pdev); 2625 } 2626 2627 /* Check if everything were registered */ 2628 if (!pvt->pci_sad0 || !pvt->pci_ha || !pvt->pci_br0 || 2629 !pvt->pci_br1 || !pvt->pci_ras || !pvt->pci_ta) 2630 goto enodev; 2631 2632 if (saw_chan_mask != 0x0f && /* -EN/-EX */ 2633 saw_chan_mask != 0x03) /* -EP */ 2634 goto enodev; 2635 return 0; 2636 2637 enodev: 2638 sbridge_printk(KERN_ERR, "Some needed devices are missing\n"); 2639 return -ENODEV; 2640 2641 error: 2642 sbridge_printk(KERN_ERR, 2643 "Unexpected device %02x:%02x\n", PCI_VENDOR_ID_INTEL, 2644 pdev->device); 2645 return -EINVAL; 2646 } 2647 2648 static int haswell_mci_bind_devs(struct mem_ctl_info *mci, 2649 struct sbridge_dev *sbridge_dev) 2650 { 2651 struct sbridge_pvt *pvt = mci->pvt_info; 2652 struct pci_dev *pdev; 2653 u8 saw_chan_mask = 0; 2654 int i; 2655 2656 /* there's only one device per system; not tied to any bus */ 2657 if (pvt->info.pci_vtd == NULL) 2658 /* result will be checked later */ 2659 pvt->info.pci_vtd = pci_get_device(PCI_VENDOR_ID_INTEL, 2660 PCI_DEVICE_ID_INTEL_HASWELL_IMC_VTD_MISC, 2661 NULL); 2662 2663 for (i = 0; i < sbridge_dev->n_devs; i++) { 2664 pdev = sbridge_dev->pdev[i]; 2665 if (!pdev) 2666 continue; 2667 2668 switch (pdev->device) { 2669 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_CBO_SAD0: 2670 pvt->pci_sad0 = pdev; 2671 break; 2672 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_CBO_SAD1: 2673 pvt->pci_sad1 = pdev; 2674 break; 2675 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0: 2676 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1: 2677 pvt->pci_ha = pdev; 2678 break; 2679 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TA: 2680 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TA: 2681 pvt->pci_ta = pdev; 2682 break; 2683 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TM: 2684 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TM: 2685 pvt->pci_ras = pdev; 2686 break; 2687 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD0: 2688 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD1: 2689 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD2: 2690 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD3: 2691 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD0: 2692 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD1: 2693 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD2: 2694 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD3: 2695 { 2696 int id = TAD_DEV_TO_CHAN(pdev->device); 2697 pvt->pci_tad[id] = pdev; 2698 saw_chan_mask |= 1 << id; 2699 } 2700 break; 2701 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO0: 2702 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO1: 2703 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO2: 2704 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO3: 2705 if (!pvt->pci_ddrio) 2706 pvt->pci_ddrio = pdev; 2707 break; 2708 default: 2709 break; 2710 } 2711 2712 edac_dbg(0, "Associated PCI %02x.%02d.%d with dev = %p\n", 2713 sbridge_dev->bus, 2714 PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn), 2715 pdev); 2716 } 2717 2718 /* Check if everything were registered */ 2719 if (!pvt->pci_sad0 || !pvt->pci_ha || !pvt->pci_sad1 || 2720 !pvt->pci_ras || !pvt->pci_ta || !pvt->info.pci_vtd) 2721 goto enodev; 2722 2723 if (saw_chan_mask != 0x0f && /* -EN/-EX */ 2724 saw_chan_mask != 0x03) /* -EP */ 2725 goto enodev; 2726 return 0; 2727 2728 enodev: 2729 sbridge_printk(KERN_ERR, "Some needed devices are missing\n"); 2730 return -ENODEV; 2731 } 2732 2733 static int broadwell_mci_bind_devs(struct mem_ctl_info *mci, 2734 struct sbridge_dev *sbridge_dev) 2735 { 2736 struct sbridge_pvt *pvt = mci->pvt_info; 2737 struct pci_dev *pdev; 2738 u8 saw_chan_mask = 0; 2739 int i; 2740 2741 /* there's only one device per system; not tied to any bus */ 2742 if (pvt->info.pci_vtd == NULL) 2743 /* result will be checked later */ 2744 pvt->info.pci_vtd = pci_get_device(PCI_VENDOR_ID_INTEL, 2745 PCI_DEVICE_ID_INTEL_BROADWELL_IMC_VTD_MISC, 2746 NULL); 2747 2748 for (i = 0; i < sbridge_dev->n_devs; i++) { 2749 pdev = sbridge_dev->pdev[i]; 2750 if (!pdev) 2751 continue; 2752 2753 switch (pdev->device) { 2754 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_CBO_SAD0: 2755 pvt->pci_sad0 = pdev; 2756 break; 2757 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_CBO_SAD1: 2758 pvt->pci_sad1 = pdev; 2759 break; 2760 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0: 2761 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1: 2762 pvt->pci_ha = pdev; 2763 break; 2764 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TA: 2765 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TA: 2766 pvt->pci_ta = pdev; 2767 break; 2768 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TM: 2769 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TM: 2770 pvt->pci_ras = pdev; 2771 break; 2772 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD0: 2773 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD1: 2774 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD2: 2775 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD3: 2776 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD0: 2777 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD1: 2778 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD2: 2779 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD3: 2780 { 2781 int id = TAD_DEV_TO_CHAN(pdev->device); 2782 pvt->pci_tad[id] = pdev; 2783 saw_chan_mask |= 1 << id; 2784 } 2785 break; 2786 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_DDRIO0: 2787 pvt->pci_ddrio = pdev; 2788 break; 2789 default: 2790 break; 2791 } 2792 2793 edac_dbg(0, "Associated PCI %02x.%02d.%d with dev = %p\n", 2794 sbridge_dev->bus, 2795 PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn), 2796 pdev); 2797 } 2798 2799 /* Check if everything were registered */ 2800 if (!pvt->pci_sad0 || !pvt->pci_ha || !pvt->pci_sad1 || 2801 !pvt->pci_ras || !pvt->pci_ta || !pvt->info.pci_vtd) 2802 goto enodev; 2803 2804 if (saw_chan_mask != 0x0f && /* -EN/-EX */ 2805 saw_chan_mask != 0x03) /* -EP */ 2806 goto enodev; 2807 return 0; 2808 2809 enodev: 2810 sbridge_printk(KERN_ERR, "Some needed devices are missing\n"); 2811 return -ENODEV; 2812 } 2813 2814 static int knl_mci_bind_devs(struct mem_ctl_info *mci, 2815 struct sbridge_dev *sbridge_dev) 2816 { 2817 struct sbridge_pvt *pvt = mci->pvt_info; 2818 struct pci_dev *pdev; 2819 int dev, func; 2820 2821 int i; 2822 int devidx; 2823 2824 for (i = 0; i < sbridge_dev->n_devs; i++) { 2825 pdev = sbridge_dev->pdev[i]; 2826 if (!pdev) 2827 continue; 2828 2829 /* Extract PCI device and function. */ 2830 dev = (pdev->devfn >> 3) & 0x1f; 2831 func = pdev->devfn & 0x7; 2832 2833 switch (pdev->device) { 2834 case PCI_DEVICE_ID_INTEL_KNL_IMC_MC: 2835 if (dev == 8) 2836 pvt->knl.pci_mc0 = pdev; 2837 else if (dev == 9) 2838 pvt->knl.pci_mc1 = pdev; 2839 else { 2840 sbridge_printk(KERN_ERR, 2841 "Memory controller in unexpected place! (dev %d, fn %d)\n", 2842 dev, func); 2843 continue; 2844 } 2845 break; 2846 2847 case PCI_DEVICE_ID_INTEL_KNL_IMC_SAD0: 2848 pvt->pci_sad0 = pdev; 2849 break; 2850 2851 case PCI_DEVICE_ID_INTEL_KNL_IMC_SAD1: 2852 pvt->pci_sad1 = pdev; 2853 break; 2854 2855 case PCI_DEVICE_ID_INTEL_KNL_IMC_CHA: 2856 /* There are one of these per tile, and range from 2857 * 1.14.0 to 1.18.5. 2858 */ 2859 devidx = ((dev-14)*8)+func; 2860 2861 if (devidx < 0 || devidx >= KNL_MAX_CHAS) { 2862 sbridge_printk(KERN_ERR, 2863 "Caching and Home Agent in unexpected place! (dev %d, fn %d)\n", 2864 dev, func); 2865 continue; 2866 } 2867 2868 WARN_ON(pvt->knl.pci_cha[devidx] != NULL); 2869 2870 pvt->knl.pci_cha[devidx] = pdev; 2871 break; 2872 2873 case PCI_DEVICE_ID_INTEL_KNL_IMC_CHAN: 2874 devidx = -1; 2875 2876 /* 2877 * MC0 channels 0-2 are device 9 function 2-4, 2878 * MC1 channels 3-5 are device 8 function 2-4. 2879 */ 2880 2881 if (dev == 9) 2882 devidx = func-2; 2883 else if (dev == 8) 2884 devidx = 3 + (func-2); 2885 2886 if (devidx < 0 || devidx >= KNL_MAX_CHANNELS) { 2887 sbridge_printk(KERN_ERR, 2888 "DRAM Channel Registers in unexpected place! (dev %d, fn %d)\n", 2889 dev, func); 2890 continue; 2891 } 2892 2893 WARN_ON(pvt->knl.pci_channel[devidx] != NULL); 2894 pvt->knl.pci_channel[devidx] = pdev; 2895 break; 2896 2897 case PCI_DEVICE_ID_INTEL_KNL_IMC_TOLHM: 2898 pvt->knl.pci_mc_info = pdev; 2899 break; 2900 2901 case PCI_DEVICE_ID_INTEL_KNL_IMC_TA: 2902 pvt->pci_ta = pdev; 2903 break; 2904 2905 default: 2906 sbridge_printk(KERN_ERR, "Unexpected device %d\n", 2907 pdev->device); 2908 break; 2909 } 2910 } 2911 2912 if (!pvt->knl.pci_mc0 || !pvt->knl.pci_mc1 || 2913 !pvt->pci_sad0 || !pvt->pci_sad1 || 2914 !pvt->pci_ta) { 2915 goto enodev; 2916 } 2917 2918 for (i = 0; i < KNL_MAX_CHANNELS; i++) { 2919 if (!pvt->knl.pci_channel[i]) { 2920 sbridge_printk(KERN_ERR, "Missing channel %d\n", i); 2921 goto enodev; 2922 } 2923 } 2924 2925 for (i = 0; i < KNL_MAX_CHAS; i++) { 2926 if (!pvt->knl.pci_cha[i]) { 2927 sbridge_printk(KERN_ERR, "Missing CHA %d\n", i); 2928 goto enodev; 2929 } 2930 } 2931 2932 return 0; 2933 2934 enodev: 2935 sbridge_printk(KERN_ERR, "Some needed devices are missing\n"); 2936 return -ENODEV; 2937 } 2938 2939 /**************************************************************************** 2940 Error check routines 2941 ****************************************************************************/ 2942 2943 /* 2944 * While Sandy Bridge has error count registers, SMI BIOS read values from 2945 * and resets the counters. So, they are not reliable for the OS to read 2946 * from them. So, we have no option but to just trust on whatever MCE is 2947 * telling us about the errors. 2948 */ 2949 static void sbridge_mce_output_error(struct mem_ctl_info *mci, 2950 const struct mce *m) 2951 { 2952 struct mem_ctl_info *new_mci; 2953 struct sbridge_pvt *pvt = mci->pvt_info; 2954 enum hw_event_mc_err_type tp_event; 2955 char *type, *optype, msg[256]; 2956 bool ripv = GET_BITFIELD(m->mcgstatus, 0, 0); 2957 bool overflow = GET_BITFIELD(m->status, 62, 62); 2958 bool uncorrected_error = GET_BITFIELD(m->status, 61, 61); 2959 bool recoverable; 2960 u32 core_err_cnt = GET_BITFIELD(m->status, 38, 52); 2961 u32 mscod = GET_BITFIELD(m->status, 16, 31); 2962 u32 errcode = GET_BITFIELD(m->status, 0, 15); 2963 u32 channel = GET_BITFIELD(m->status, 0, 3); 2964 u32 optypenum = GET_BITFIELD(m->status, 4, 6); 2965 /* 2966 * Bits 5-0 of MCi_MISC give the least significant bit that is valid. 2967 * A value 6 is for cache line aligned address, a value 12 is for page 2968 * aligned address reported by patrol scrubber. 2969 */ 2970 u32 lsb = GET_BITFIELD(m->misc, 0, 5); 2971 long channel_mask, first_channel; 2972 u8 rank = 0xff, socket, ha; 2973 int rc, dimm; 2974 char *area_type = "DRAM"; 2975 2976 if (pvt->info.type != SANDY_BRIDGE) 2977 recoverable = true; 2978 else 2979 recoverable = GET_BITFIELD(m->status, 56, 56); 2980 2981 if (uncorrected_error) { 2982 core_err_cnt = 1; 2983 if (ripv) { 2984 type = "FATAL"; 2985 tp_event = HW_EVENT_ERR_FATAL; 2986 } else { 2987 type = "NON_FATAL"; 2988 tp_event = HW_EVENT_ERR_UNCORRECTED; 2989 } 2990 } else { 2991 type = "CORRECTED"; 2992 tp_event = HW_EVENT_ERR_CORRECTED; 2993 } 2994 2995 /* 2996 * According with Table 15-9 of the Intel Architecture spec vol 3A, 2997 * memory errors should fit in this mask: 2998 * 000f 0000 1mmm cccc (binary) 2999 * where: 3000 * f = Correction Report Filtering Bit. If 1, subsequent errors 3001 * won't be shown 3002 * mmm = error type 3003 * cccc = channel 3004 * If the mask doesn't match, report an error to the parsing logic 3005 */ 3006 switch (optypenum) { 3007 case 0: 3008 optype = "generic undef request error"; 3009 break; 3010 case 1: 3011 optype = "memory read error"; 3012 break; 3013 case 2: 3014 optype = "memory write error"; 3015 break; 3016 case 3: 3017 optype = "addr/cmd error"; 3018 break; 3019 case 4: 3020 optype = "memory scrubbing error"; 3021 break; 3022 default: 3023 optype = "reserved"; 3024 break; 3025 } 3026 3027 if (pvt->info.type == KNIGHTS_LANDING) { 3028 if (channel == 14) { 3029 edac_dbg(0, "%s%s err_code:%04x:%04x EDRAM bank %d\n", 3030 overflow ? " OVERFLOW" : "", 3031 (uncorrected_error && recoverable) 3032 ? " recoverable" : "", 3033 mscod, errcode, 3034 m->bank); 3035 } else { 3036 char A = *("A"); 3037 3038 /* 3039 * Reported channel is in range 0-2, so we can't map it 3040 * back to mc. To figure out mc we check machine check 3041 * bank register that reported this error. 3042 * bank15 means mc0 and bank16 means mc1. 3043 */ 3044 channel = knl_channel_remap(m->bank == 16, channel); 3045 channel_mask = 1 << channel; 3046 3047 snprintf(msg, sizeof(msg), 3048 "%s%s err_code:%04x:%04x channel:%d (DIMM_%c)", 3049 overflow ? " OVERFLOW" : "", 3050 (uncorrected_error && recoverable) 3051 ? " recoverable" : " ", 3052 mscod, errcode, channel, A + channel); 3053 edac_mc_handle_error(tp_event, mci, core_err_cnt, 3054 m->addr >> PAGE_SHIFT, m->addr & ~PAGE_MASK, 0, 3055 channel, 0, -1, 3056 optype, msg); 3057 } 3058 return; 3059 } else if (lsb < 12) { 3060 rc = get_memory_error_data(mci, m->addr, &socket, &ha, 3061 &channel_mask, &rank, 3062 &area_type, msg); 3063 } else { 3064 rc = get_memory_error_data_from_mce(mci, m, &socket, &ha, 3065 &channel_mask, msg); 3066 } 3067 3068 if (rc < 0) 3069 goto err_parsing; 3070 new_mci = get_mci_for_node_id(socket, ha); 3071 if (!new_mci) { 3072 strcpy(msg, "Error: socket got corrupted!"); 3073 goto err_parsing; 3074 } 3075 mci = new_mci; 3076 pvt = mci->pvt_info; 3077 3078 first_channel = find_first_bit(&channel_mask, NUM_CHANNELS); 3079 3080 if (rank == 0xff) 3081 dimm = -1; 3082 else if (rank < 4) 3083 dimm = 0; 3084 else if (rank < 8) 3085 dimm = 1; 3086 else 3087 dimm = 2; 3088 3089 /* 3090 * FIXME: On some memory configurations (mirror, lockstep), the 3091 * Memory Controller can't point the error to a single DIMM. The 3092 * EDAC core should be handling the channel mask, in order to point 3093 * to the group of dimm's where the error may be happening. 3094 */ 3095 if (!pvt->is_lockstep && !pvt->is_cur_addr_mirrored && !pvt->is_close_pg) 3096 channel = first_channel; 3097 3098 snprintf(msg, sizeof(msg), 3099 "%s%s area:%s err_code:%04x:%04x socket:%d ha:%d channel_mask:%ld rank:%d", 3100 overflow ? " OVERFLOW" : "", 3101 (uncorrected_error && recoverable) ? " recoverable" : "", 3102 area_type, 3103 mscod, errcode, 3104 socket, ha, 3105 channel_mask, 3106 rank); 3107 3108 edac_dbg(0, "%s\n", msg); 3109 3110 /* FIXME: need support for channel mask */ 3111 3112 if (channel == CHANNEL_UNSPECIFIED) 3113 channel = -1; 3114 3115 /* Call the helper to output message */ 3116 edac_mc_handle_error(tp_event, mci, core_err_cnt, 3117 m->addr >> PAGE_SHIFT, m->addr & ~PAGE_MASK, 0, 3118 channel, dimm, -1, 3119 optype, msg); 3120 return; 3121 err_parsing: 3122 edac_mc_handle_error(tp_event, mci, core_err_cnt, 0, 0, 0, 3123 -1, -1, -1, 3124 msg, ""); 3125 3126 } 3127 3128 /* 3129 * Check that logging is enabled and that this is the right type 3130 * of error for us to handle. 3131 */ 3132 static int sbridge_mce_check_error(struct notifier_block *nb, unsigned long val, 3133 void *data) 3134 { 3135 struct mce *mce = (struct mce *)data; 3136 struct mem_ctl_info *mci; 3137 char *type; 3138 3139 if (edac_get_report_status() == EDAC_REPORTING_DISABLED) 3140 return NOTIFY_DONE; 3141 3142 /* 3143 * Just let mcelog handle it if the error is 3144 * outside the memory controller. A memory error 3145 * is indicated by bit 7 = 1 and bits = 8-11,13-15 = 0. 3146 * bit 12 has an special meaning. 3147 */ 3148 if ((mce->status & 0xefff) >> 7 != 1) 3149 return NOTIFY_DONE; 3150 3151 /* Check ADDRV bit in STATUS */ 3152 if (!GET_BITFIELD(mce->status, 58, 58)) 3153 return NOTIFY_DONE; 3154 3155 /* Check MISCV bit in STATUS */ 3156 if (!GET_BITFIELD(mce->status, 59, 59)) 3157 return NOTIFY_DONE; 3158 3159 /* Check address type in MISC (physical address only) */ 3160 if (GET_BITFIELD(mce->misc, 6, 8) != 2) 3161 return NOTIFY_DONE; 3162 3163 mci = get_mci_for_node_id(mce->socketid, IMC0); 3164 if (!mci) 3165 return NOTIFY_DONE; 3166 3167 if (mce->mcgstatus & MCG_STATUS_MCIP) 3168 type = "Exception"; 3169 else 3170 type = "Event"; 3171 3172 sbridge_mc_printk(mci, KERN_DEBUG, "HANDLING MCE MEMORY ERROR\n"); 3173 3174 sbridge_mc_printk(mci, KERN_DEBUG, "CPU %d: Machine Check %s: %Lx " 3175 "Bank %d: %016Lx\n", mce->extcpu, type, 3176 mce->mcgstatus, mce->bank, mce->status); 3177 sbridge_mc_printk(mci, KERN_DEBUG, "TSC %llx ", mce->tsc); 3178 sbridge_mc_printk(mci, KERN_DEBUG, "ADDR %llx ", mce->addr); 3179 sbridge_mc_printk(mci, KERN_DEBUG, "MISC %llx ", mce->misc); 3180 3181 sbridge_mc_printk(mci, KERN_DEBUG, "PROCESSOR %u:%x TIME %llu SOCKET " 3182 "%u APIC %x\n", mce->cpuvendor, mce->cpuid, 3183 mce->time, mce->socketid, mce->apicid); 3184 3185 sbridge_mce_output_error(mci, mce); 3186 3187 /* Advice mcelog that the error were handled */ 3188 return NOTIFY_STOP; 3189 } 3190 3191 static struct notifier_block sbridge_mce_dec = { 3192 .notifier_call = sbridge_mce_check_error, 3193 .priority = MCE_PRIO_EDAC, 3194 }; 3195 3196 /**************************************************************************** 3197 EDAC register/unregister logic 3198 ****************************************************************************/ 3199 3200 static void sbridge_unregister_mci(struct sbridge_dev *sbridge_dev) 3201 { 3202 struct mem_ctl_info *mci = sbridge_dev->mci; 3203 struct sbridge_pvt *pvt; 3204 3205 if (unlikely(!mci || !mci->pvt_info)) { 3206 edac_dbg(0, "MC: dev = %p\n", &sbridge_dev->pdev[0]->dev); 3207 3208 sbridge_printk(KERN_ERR, "Couldn't find mci handler\n"); 3209 return; 3210 } 3211 3212 pvt = mci->pvt_info; 3213 3214 edac_dbg(0, "MC: mci = %p, dev = %p\n", 3215 mci, &sbridge_dev->pdev[0]->dev); 3216 3217 /* Remove MC sysfs nodes */ 3218 edac_mc_del_mc(mci->pdev); 3219 3220 edac_dbg(1, "%s: free mci struct\n", mci->ctl_name); 3221 kfree(mci->ctl_name); 3222 edac_mc_free(mci); 3223 sbridge_dev->mci = NULL; 3224 } 3225 3226 static int sbridge_register_mci(struct sbridge_dev *sbridge_dev, enum type type) 3227 { 3228 struct mem_ctl_info *mci; 3229 struct edac_mc_layer layers[2]; 3230 struct sbridge_pvt *pvt; 3231 struct pci_dev *pdev = sbridge_dev->pdev[0]; 3232 int rc; 3233 3234 /* allocate a new MC control structure */ 3235 layers[0].type = EDAC_MC_LAYER_CHANNEL; 3236 layers[0].size = type == KNIGHTS_LANDING ? 3237 KNL_MAX_CHANNELS : NUM_CHANNELS; 3238 layers[0].is_virt_csrow = false; 3239 layers[1].type = EDAC_MC_LAYER_SLOT; 3240 layers[1].size = type == KNIGHTS_LANDING ? 1 : MAX_DIMMS; 3241 layers[1].is_virt_csrow = true; 3242 mci = edac_mc_alloc(sbridge_dev->mc, ARRAY_SIZE(layers), layers, 3243 sizeof(*pvt)); 3244 3245 if (unlikely(!mci)) 3246 return -ENOMEM; 3247 3248 edac_dbg(0, "MC: mci = %p, dev = %p\n", 3249 mci, &pdev->dev); 3250 3251 pvt = mci->pvt_info; 3252 memset(pvt, 0, sizeof(*pvt)); 3253 3254 /* Associate sbridge_dev and mci for future usage */ 3255 pvt->sbridge_dev = sbridge_dev; 3256 sbridge_dev->mci = mci; 3257 3258 mci->mtype_cap = type == KNIGHTS_LANDING ? 3259 MEM_FLAG_DDR4 : MEM_FLAG_DDR3; 3260 mci->edac_ctl_cap = EDAC_FLAG_NONE; 3261 mci->edac_cap = EDAC_FLAG_NONE; 3262 mci->mod_name = EDAC_MOD_STR; 3263 mci->dev_name = pci_name(pdev); 3264 mci->ctl_page_to_phys = NULL; 3265 3266 pvt->info.type = type; 3267 switch (type) { 3268 case IVY_BRIDGE: 3269 pvt->info.rankcfgr = IB_RANK_CFG_A; 3270 pvt->info.get_tolm = ibridge_get_tolm; 3271 pvt->info.get_tohm = ibridge_get_tohm; 3272 pvt->info.dram_rule = ibridge_dram_rule; 3273 pvt->info.get_memory_type = get_memory_type; 3274 pvt->info.get_node_id = get_node_id; 3275 pvt->info.get_ha = ibridge_get_ha; 3276 pvt->info.rir_limit = rir_limit; 3277 pvt->info.sad_limit = sad_limit; 3278 pvt->info.interleave_mode = interleave_mode; 3279 pvt->info.dram_attr = dram_attr; 3280 pvt->info.max_sad = ARRAY_SIZE(ibridge_dram_rule); 3281 pvt->info.interleave_list = ibridge_interleave_list; 3282 pvt->info.interleave_pkg = ibridge_interleave_pkg; 3283 pvt->info.get_width = ibridge_get_width; 3284 3285 /* Store pci devices at mci for faster access */ 3286 rc = ibridge_mci_bind_devs(mci, sbridge_dev); 3287 if (unlikely(rc < 0)) 3288 goto fail0; 3289 get_source_id(mci); 3290 mci->ctl_name = kasprintf(GFP_KERNEL, "Ivy Bridge SrcID#%d_Ha#%d", 3291 pvt->sbridge_dev->source_id, pvt->sbridge_dev->dom); 3292 break; 3293 case SANDY_BRIDGE: 3294 pvt->info.rankcfgr = SB_RANK_CFG_A; 3295 pvt->info.get_tolm = sbridge_get_tolm; 3296 pvt->info.get_tohm = sbridge_get_tohm; 3297 pvt->info.dram_rule = sbridge_dram_rule; 3298 pvt->info.get_memory_type = get_memory_type; 3299 pvt->info.get_node_id = get_node_id; 3300 pvt->info.get_ha = sbridge_get_ha; 3301 pvt->info.rir_limit = rir_limit; 3302 pvt->info.sad_limit = sad_limit; 3303 pvt->info.interleave_mode = interleave_mode; 3304 pvt->info.dram_attr = dram_attr; 3305 pvt->info.max_sad = ARRAY_SIZE(sbridge_dram_rule); 3306 pvt->info.interleave_list = sbridge_interleave_list; 3307 pvt->info.interleave_pkg = sbridge_interleave_pkg; 3308 pvt->info.get_width = sbridge_get_width; 3309 3310 /* Store pci devices at mci for faster access */ 3311 rc = sbridge_mci_bind_devs(mci, sbridge_dev); 3312 if (unlikely(rc < 0)) 3313 goto fail0; 3314 get_source_id(mci); 3315 mci->ctl_name = kasprintf(GFP_KERNEL, "Sandy Bridge SrcID#%d_Ha#%d", 3316 pvt->sbridge_dev->source_id, pvt->sbridge_dev->dom); 3317 break; 3318 case HASWELL: 3319 /* rankcfgr isn't used */ 3320 pvt->info.get_tolm = haswell_get_tolm; 3321 pvt->info.get_tohm = haswell_get_tohm; 3322 pvt->info.dram_rule = ibridge_dram_rule; 3323 pvt->info.get_memory_type = haswell_get_memory_type; 3324 pvt->info.get_node_id = haswell_get_node_id; 3325 pvt->info.get_ha = ibridge_get_ha; 3326 pvt->info.rir_limit = haswell_rir_limit; 3327 pvt->info.sad_limit = sad_limit; 3328 pvt->info.interleave_mode = interleave_mode; 3329 pvt->info.dram_attr = dram_attr; 3330 pvt->info.max_sad = ARRAY_SIZE(ibridge_dram_rule); 3331 pvt->info.interleave_list = ibridge_interleave_list; 3332 pvt->info.interleave_pkg = ibridge_interleave_pkg; 3333 pvt->info.get_width = ibridge_get_width; 3334 3335 /* Store pci devices at mci for faster access */ 3336 rc = haswell_mci_bind_devs(mci, sbridge_dev); 3337 if (unlikely(rc < 0)) 3338 goto fail0; 3339 get_source_id(mci); 3340 mci->ctl_name = kasprintf(GFP_KERNEL, "Haswell SrcID#%d_Ha#%d", 3341 pvt->sbridge_dev->source_id, pvt->sbridge_dev->dom); 3342 break; 3343 case BROADWELL: 3344 /* rankcfgr isn't used */ 3345 pvt->info.get_tolm = haswell_get_tolm; 3346 pvt->info.get_tohm = haswell_get_tohm; 3347 pvt->info.dram_rule = ibridge_dram_rule; 3348 pvt->info.get_memory_type = haswell_get_memory_type; 3349 pvt->info.get_node_id = haswell_get_node_id; 3350 pvt->info.get_ha = ibridge_get_ha; 3351 pvt->info.rir_limit = haswell_rir_limit; 3352 pvt->info.sad_limit = sad_limit; 3353 pvt->info.interleave_mode = interleave_mode; 3354 pvt->info.dram_attr = dram_attr; 3355 pvt->info.max_sad = ARRAY_SIZE(ibridge_dram_rule); 3356 pvt->info.interleave_list = ibridge_interleave_list; 3357 pvt->info.interleave_pkg = ibridge_interleave_pkg; 3358 pvt->info.get_width = broadwell_get_width; 3359 3360 /* Store pci devices at mci for faster access */ 3361 rc = broadwell_mci_bind_devs(mci, sbridge_dev); 3362 if (unlikely(rc < 0)) 3363 goto fail0; 3364 get_source_id(mci); 3365 mci->ctl_name = kasprintf(GFP_KERNEL, "Broadwell SrcID#%d_Ha#%d", 3366 pvt->sbridge_dev->source_id, pvt->sbridge_dev->dom); 3367 break; 3368 case KNIGHTS_LANDING: 3369 /* pvt->info.rankcfgr == ??? */ 3370 pvt->info.get_tolm = knl_get_tolm; 3371 pvt->info.get_tohm = knl_get_tohm; 3372 pvt->info.dram_rule = knl_dram_rule; 3373 pvt->info.get_memory_type = knl_get_memory_type; 3374 pvt->info.get_node_id = knl_get_node_id; 3375 pvt->info.get_ha = knl_get_ha; 3376 pvt->info.rir_limit = NULL; 3377 pvt->info.sad_limit = knl_sad_limit; 3378 pvt->info.interleave_mode = knl_interleave_mode; 3379 pvt->info.dram_attr = dram_attr_knl; 3380 pvt->info.max_sad = ARRAY_SIZE(knl_dram_rule); 3381 pvt->info.interleave_list = knl_interleave_list; 3382 pvt->info.interleave_pkg = ibridge_interleave_pkg; 3383 pvt->info.get_width = knl_get_width; 3384 3385 rc = knl_mci_bind_devs(mci, sbridge_dev); 3386 if (unlikely(rc < 0)) 3387 goto fail0; 3388 get_source_id(mci); 3389 mci->ctl_name = kasprintf(GFP_KERNEL, "Knights Landing SrcID#%d_Ha#%d", 3390 pvt->sbridge_dev->source_id, pvt->sbridge_dev->dom); 3391 break; 3392 } 3393 3394 if (!mci->ctl_name) { 3395 rc = -ENOMEM; 3396 goto fail0; 3397 } 3398 3399 /* Get dimm basic config and the memory layout */ 3400 rc = get_dimm_config(mci); 3401 if (rc < 0) { 3402 edac_dbg(0, "MC: failed to get_dimm_config()\n"); 3403 goto fail; 3404 } 3405 get_memory_layout(mci); 3406 3407 /* record ptr to the generic device */ 3408 mci->pdev = &pdev->dev; 3409 3410 /* add this new MC control structure to EDAC's list of MCs */ 3411 if (unlikely(edac_mc_add_mc(mci))) { 3412 edac_dbg(0, "MC: failed edac_mc_add_mc()\n"); 3413 rc = -EINVAL; 3414 goto fail; 3415 } 3416 3417 return 0; 3418 3419 fail: 3420 kfree(mci->ctl_name); 3421 fail0: 3422 edac_mc_free(mci); 3423 sbridge_dev->mci = NULL; 3424 return rc; 3425 } 3426 3427 static const struct x86_cpu_id sbridge_cpuids[] = { 3428 INTEL_CPU_FAM6(SANDYBRIDGE_X, pci_dev_descr_sbridge_table), 3429 INTEL_CPU_FAM6(IVYBRIDGE_X, pci_dev_descr_ibridge_table), 3430 INTEL_CPU_FAM6(HASWELL_X, pci_dev_descr_haswell_table), 3431 INTEL_CPU_FAM6(BROADWELL_X, pci_dev_descr_broadwell_table), 3432 INTEL_CPU_FAM6(BROADWELL_D, pci_dev_descr_broadwell_table), 3433 INTEL_CPU_FAM6(XEON_PHI_KNL, pci_dev_descr_knl_table), 3434 INTEL_CPU_FAM6(XEON_PHI_KNM, pci_dev_descr_knl_table), 3435 { } 3436 }; 3437 MODULE_DEVICE_TABLE(x86cpu, sbridge_cpuids); 3438 3439 /* 3440 * sbridge_probe Get all devices and register memory controllers 3441 * present. 3442 * return: 3443 * 0 for FOUND a device 3444 * < 0 for error code 3445 */ 3446 3447 static int sbridge_probe(const struct x86_cpu_id *id) 3448 { 3449 int rc = -ENODEV; 3450 u8 mc, num_mc = 0; 3451 struct sbridge_dev *sbridge_dev; 3452 struct pci_id_table *ptable = (struct pci_id_table *)id->driver_data; 3453 3454 /* get the pci devices we want to reserve for our use */ 3455 rc = sbridge_get_all_devices(&num_mc, ptable); 3456 3457 if (unlikely(rc < 0)) { 3458 edac_dbg(0, "couldn't get all devices\n"); 3459 goto fail0; 3460 } 3461 3462 mc = 0; 3463 3464 list_for_each_entry(sbridge_dev, &sbridge_edac_list, list) { 3465 edac_dbg(0, "Registering MC#%d (%d of %d)\n", 3466 mc, mc + 1, num_mc); 3467 3468 sbridge_dev->mc = mc++; 3469 rc = sbridge_register_mci(sbridge_dev, ptable->type); 3470 if (unlikely(rc < 0)) 3471 goto fail1; 3472 } 3473 3474 sbridge_printk(KERN_INFO, "%s\n", SBRIDGE_REVISION); 3475 3476 return 0; 3477 3478 fail1: 3479 list_for_each_entry(sbridge_dev, &sbridge_edac_list, list) 3480 sbridge_unregister_mci(sbridge_dev); 3481 3482 sbridge_put_all_devices(); 3483 fail0: 3484 return rc; 3485 } 3486 3487 /* 3488 * sbridge_remove cleanup 3489 * 3490 */ 3491 static void sbridge_remove(void) 3492 { 3493 struct sbridge_dev *sbridge_dev; 3494 3495 edac_dbg(0, "\n"); 3496 3497 list_for_each_entry(sbridge_dev, &sbridge_edac_list, list) 3498 sbridge_unregister_mci(sbridge_dev); 3499 3500 /* Release PCI resources */ 3501 sbridge_put_all_devices(); 3502 } 3503 3504 /* 3505 * sbridge_init Module entry function 3506 * Try to initialize this module for its devices 3507 */ 3508 static int __init sbridge_init(void) 3509 { 3510 const struct x86_cpu_id *id; 3511 const char *owner; 3512 int rc; 3513 3514 edac_dbg(2, "\n"); 3515 3516 owner = edac_get_owner(); 3517 if (owner && strncmp(owner, EDAC_MOD_STR, sizeof(EDAC_MOD_STR))) 3518 return -EBUSY; 3519 3520 id = x86_match_cpu(sbridge_cpuids); 3521 if (!id) 3522 return -ENODEV; 3523 3524 /* Ensure that the OPSTATE is set correctly for POLL or NMI */ 3525 opstate_init(); 3526 3527 rc = sbridge_probe(id); 3528 3529 if (rc >= 0) { 3530 mce_register_decode_chain(&sbridge_mce_dec); 3531 if (edac_get_report_status() == EDAC_REPORTING_DISABLED) 3532 sbridge_printk(KERN_WARNING, "Loading driver, error reporting disabled.\n"); 3533 return 0; 3534 } 3535 3536 sbridge_printk(KERN_ERR, "Failed to register device with error %d.\n", 3537 rc); 3538 3539 return rc; 3540 } 3541 3542 /* 3543 * sbridge_exit() Module exit function 3544 * Unregister the driver 3545 */ 3546 static void __exit sbridge_exit(void) 3547 { 3548 edac_dbg(2, "\n"); 3549 sbridge_remove(); 3550 mce_unregister_decode_chain(&sbridge_mce_dec); 3551 } 3552 3553 module_init(sbridge_init); 3554 module_exit(sbridge_exit); 3555 3556 module_param(edac_op_state, int, 0444); 3557 MODULE_PARM_DESC(edac_op_state, "EDAC Error Reporting state: 0=Poll,1=NMI"); 3558 3559 MODULE_LICENSE("GPL"); 3560 MODULE_AUTHOR("Mauro Carvalho Chehab"); 3561 MODULE_AUTHOR("Red Hat Inc. (http://www.redhat.com)"); 3562 MODULE_DESCRIPTION("MC Driver for Intel Sandy Bridge and Ivy Bridge memory controllers - " 3563 SBRIDGE_REVISION); 3564