1 /* 2 * AMD64 class Memory Controller kernel module 3 * 4 * Copyright (c) 2009 SoftwareBitMaker. 5 * Copyright (c) 2009 Advanced Micro Devices, Inc. 6 * 7 * This file may be distributed under the terms of the 8 * GNU General Public License. 9 * 10 * Originally Written by Thayne Harbaugh 11 * 12 * Changes by Douglas "norsk" Thompson <dougthompson@xmission.com>: 13 * - K8 CPU Revision D and greater support 14 * 15 * Changes by Dave Peterson <dsp@llnl.gov> <dave_peterson@pobox.com>: 16 * - Module largely rewritten, with new (and hopefully correct) 17 * code for dealing with node and chip select interleaving, 18 * various code cleanup, and bug fixes 19 * - Added support for memory hoisting using DRAM hole address 20 * register 21 * 22 * Changes by Douglas "norsk" Thompson <dougthompson@xmission.com>: 23 * -K8 Rev (1207) revision support added, required Revision 24 * specific mini-driver code to support Rev F as well as 25 * prior revisions 26 * 27 * Changes by Douglas "norsk" Thompson <dougthompson@xmission.com>: 28 * -Family 10h revision support added. New PCI Device IDs, 29 * indicating new changes. Actual registers modified 30 * were slight, less than the Rev E to Rev F transition 31 * but changing the PCI Device ID was the proper thing to 32 * do, as it provides for almost automactic family 33 * detection. The mods to Rev F required more family 34 * information detection. 35 * 36 * Changes/Fixes by Borislav Petkov <borislav.petkov@amd.com>: 37 * - misc fixes and code cleanups 38 * 39 * This module is based on the following documents 40 * (available from http://www.amd.com/): 41 * 42 * Title: BIOS and Kernel Developer's Guide for AMD Athlon 64 and AMD 43 * Opteron Processors 44 * AMD publication #: 26094 45 *` Revision: 3.26 46 * 47 * Title: BIOS and Kernel Developer's Guide for AMD NPT Family 0Fh 48 * Processors 49 * AMD publication #: 32559 50 * Revision: 3.00 51 * Issue Date: May 2006 52 * 53 * Title: BIOS and Kernel Developer's Guide (BKDG) For AMD Family 10h 54 * Processors 55 * AMD publication #: 31116 56 * Revision: 3.00 57 * Issue Date: September 07, 2007 58 * 59 * Sections in the first 2 documents are no longer in sync with each other. 60 * The Family 10h BKDG was totally re-written from scratch with a new 61 * presentation model. 62 * Therefore, comments that refer to a Document section might be off. 63 */ 64 65 #include <linux/module.h> 66 #include <linux/ctype.h> 67 #include <linux/init.h> 68 #include <linux/pci.h> 69 #include <linux/pci_ids.h> 70 #include <linux/slab.h> 71 #include <linux/mmzone.h> 72 #include <linux/edac.h> 73 #include <asm/msr.h> 74 #include "edac_core.h" 75 #include "edac_mce_amd.h" 76 77 #define amd64_printk(level, fmt, arg...) \ 78 edac_printk(level, "amd64", fmt, ##arg) 79 80 #define amd64_mc_printk(mci, level, fmt, arg...) \ 81 edac_mc_chipset_printk(mci, level, "amd64", fmt, ##arg) 82 83 /* 84 * Throughout the comments in this code, the following terms are used: 85 * 86 * SysAddr, DramAddr, and InputAddr 87 * 88 * These terms come directly from the amd64 documentation 89 * (AMD publication #26094). They are defined as follows: 90 * 91 * SysAddr: 92 * This is a physical address generated by a CPU core or a device 93 * doing DMA. If generated by a CPU core, a SysAddr is the result of 94 * a virtual to physical address translation by the CPU core's address 95 * translation mechanism (MMU). 96 * 97 * DramAddr: 98 * A DramAddr is derived from a SysAddr by subtracting an offset that 99 * depends on which node the SysAddr maps to and whether the SysAddr 100 * is within a range affected by memory hoisting. The DRAM Base 101 * (section 3.4.4.1) and DRAM Limit (section 3.4.4.2) registers 102 * determine which node a SysAddr maps to. 103 * 104 * If the DRAM Hole Address Register (DHAR) is enabled and the SysAddr 105 * is within the range of addresses specified by this register, then 106 * a value x from the DHAR is subtracted from the SysAddr to produce a 107 * DramAddr. Here, x represents the base address for the node that 108 * the SysAddr maps to plus an offset due to memory hoisting. See 109 * section 3.4.8 and the comments in amd64_get_dram_hole_info() and 110 * sys_addr_to_dram_addr() below for more information. 111 * 112 * If the SysAddr is not affected by the DHAR then a value y is 113 * subtracted from the SysAddr to produce a DramAddr. Here, y is the 114 * base address for the node that the SysAddr maps to. See section 115 * 3.4.4 and the comments in sys_addr_to_dram_addr() below for more 116 * information. 117 * 118 * InputAddr: 119 * A DramAddr is translated to an InputAddr before being passed to the 120 * memory controller for the node that the DramAddr is associated 121 * with. The memory controller then maps the InputAddr to a csrow. 122 * If node interleaving is not in use, then the InputAddr has the same 123 * value as the DramAddr. Otherwise, the InputAddr is produced by 124 * discarding the bits used for node interleaving from the DramAddr. 125 * See section 3.4.4 for more information. 126 * 127 * The memory controller for a given node uses its DRAM CS Base and 128 * DRAM CS Mask registers to map an InputAddr to a csrow. See 129 * sections 3.5.4 and 3.5.5 for more information. 130 */ 131 132 #define EDAC_AMD64_VERSION " Ver: 3.2.0 " __DATE__ 133 #define EDAC_MOD_STR "amd64_edac" 134 135 #define EDAC_MAX_NUMNODES 8 136 137 /* Extended Model from CPUID, for CPU Revision numbers */ 138 #define OPTERON_CPU_LE_REV_C 0 139 #define OPTERON_CPU_REV_D 1 140 #define OPTERON_CPU_REV_E 2 141 142 /* NPT processors have the following Extended Models */ 143 #define OPTERON_CPU_REV_F 4 144 #define OPTERON_CPU_REV_FA 5 145 146 /* Hardware limit on ChipSelect rows per MC and processors per system */ 147 #define MAX_CS_COUNT 8 148 #define DRAM_REG_COUNT 8 149 150 151 /* 152 * PCI-defined configuration space registers 153 */ 154 155 156 /* 157 * Function 1 - Address Map 158 */ 159 #define K8_DRAM_BASE_LOW 0x40 160 #define K8_DRAM_LIMIT_LOW 0x44 161 #define K8_DHAR 0xf0 162 163 #define DHAR_VALID BIT(0) 164 #define F10_DRAM_MEM_HOIST_VALID BIT(1) 165 166 #define DHAR_BASE_MASK 0xff000000 167 #define dhar_base(dhar) (dhar & DHAR_BASE_MASK) 168 169 #define K8_DHAR_OFFSET_MASK 0x0000ff00 170 #define k8_dhar_offset(dhar) ((dhar & K8_DHAR_OFFSET_MASK) << 16) 171 172 #define F10_DHAR_OFFSET_MASK 0x0000ff80 173 /* NOTE: Extra mask bit vs K8 */ 174 #define f10_dhar_offset(dhar) ((dhar & F10_DHAR_OFFSET_MASK) << 16) 175 176 177 /* F10 High BASE/LIMIT registers */ 178 #define F10_DRAM_BASE_HIGH 0x140 179 #define F10_DRAM_LIMIT_HIGH 0x144 180 181 182 /* 183 * Function 2 - DRAM controller 184 */ 185 #define K8_DCSB0 0x40 186 #define F10_DCSB1 0x140 187 188 #define K8_DCSB_CS_ENABLE BIT(0) 189 #define K8_DCSB_NPT_SPARE BIT(1) 190 #define K8_DCSB_NPT_TESTFAIL BIT(2) 191 192 /* 193 * REV E: select [31:21] and [15:9] from DCSB and the shift amount to form 194 * the address 195 */ 196 #define REV_E_DCSB_BASE_BITS (0xFFE0FE00ULL) 197 #define REV_E_DCS_SHIFT 4 198 199 #define REV_F_F1Xh_DCSB_BASE_BITS (0x1FF83FE0ULL) 200 #define REV_F_F1Xh_DCS_SHIFT 8 201 202 /* 203 * REV F and later: selects [28:19] and [13:5] from DCSB and the shift amount 204 * to form the address 205 */ 206 #define REV_F_DCSB_BASE_BITS (0x1FF83FE0ULL) 207 #define REV_F_DCS_SHIFT 8 208 209 /* DRAM CS Mask Registers */ 210 #define K8_DCSM0 0x60 211 #define F10_DCSM1 0x160 212 213 /* REV E: select [29:21] and [15:9] from DCSM */ 214 #define REV_E_DCSM_MASK_BITS 0x3FE0FE00 215 216 /* unused bits [24:20] and [12:0] */ 217 #define REV_E_DCS_NOTUSED_BITS 0x01F01FFF 218 219 /* REV F and later: select [28:19] and [13:5] from DCSM */ 220 #define REV_F_F1Xh_DCSM_MASK_BITS 0x1FF83FE0 221 222 /* unused bits [26:22] and [12:0] */ 223 #define REV_F_F1Xh_DCS_NOTUSED_BITS 0x07C01FFF 224 225 #define DBAM0 0x80 226 #define DBAM1 0x180 227 228 /* Extract the DIMM 'type' on the i'th DIMM from the DBAM reg value passed */ 229 #define DBAM_DIMM(i, reg) ((((reg) >> (4*i))) & 0xF) 230 231 #define DBAM_MAX_VALUE 11 232 233 234 #define F10_DCLR_0 0x90 235 #define F10_DCLR_1 0x190 236 #define REVE_WIDTH_128 BIT(16) 237 #define F10_WIDTH_128 BIT(11) 238 239 240 #define F10_DCHR_0 0x94 241 #define F10_DCHR_1 0x194 242 243 #define F10_DCHR_FOUR_RANK_DIMM BIT(18) 244 #define F10_DCHR_Ddr3Mode BIT(8) 245 #define F10_DCHR_MblMode BIT(6) 246 247 248 #define F10_DCTL_SEL_LOW 0x110 249 250 #define dct_sel_baseaddr(pvt) \ 251 ((pvt->dram_ctl_select_low) & 0xFFFFF800) 252 253 #define dct_sel_interleave_addr(pvt) \ 254 (((pvt->dram_ctl_select_low) >> 6) & 0x3) 255 256 enum { 257 F10_DCTL_SEL_LOW_DctSelHiRngEn = BIT(0), 258 F10_DCTL_SEL_LOW_DctSelIntLvEn = BIT(2), 259 F10_DCTL_SEL_LOW_DctGangEn = BIT(4), 260 F10_DCTL_SEL_LOW_DctDatIntLv = BIT(5), 261 F10_DCTL_SEL_LOW_DramEnable = BIT(8), 262 F10_DCTL_SEL_LOW_MemCleared = BIT(10), 263 }; 264 265 #define dct_high_range_enabled(pvt) \ 266 (pvt->dram_ctl_select_low & F10_DCTL_SEL_LOW_DctSelHiRngEn) 267 268 #define dct_interleave_enabled(pvt) \ 269 (pvt->dram_ctl_select_low & F10_DCTL_SEL_LOW_DctSelIntLvEn) 270 271 #define dct_ganging_enabled(pvt) \ 272 (pvt->dram_ctl_select_low & F10_DCTL_SEL_LOW_DctGangEn) 273 274 #define dct_data_intlv_enabled(pvt) \ 275 (pvt->dram_ctl_select_low & F10_DCTL_SEL_LOW_DctDatIntLv) 276 277 #define dct_dram_enabled(pvt) \ 278 (pvt->dram_ctl_select_low & F10_DCTL_SEL_LOW_DramEnable) 279 280 #define dct_memory_cleared(pvt) \ 281 (pvt->dram_ctl_select_low & F10_DCTL_SEL_LOW_MemCleared) 282 283 284 #define F10_DCTL_SEL_HIGH 0x114 285 286 287 /* 288 * Function 3 - Misc Control 289 */ 290 #define K8_NBCTL 0x40 291 292 /* Correctable ECC error reporting enable */ 293 #define K8_NBCTL_CECCEn BIT(0) 294 295 /* UnCorrectable ECC error reporting enable */ 296 #define K8_NBCTL_UECCEn BIT(1) 297 298 #define K8_NBCFG 0x44 299 #define K8_NBCFG_CHIPKILL BIT(23) 300 #define K8_NBCFG_ECC_ENABLE BIT(22) 301 302 #define K8_NBSL 0x48 303 304 305 /* Family F10h: Normalized Extended Error Codes */ 306 #define F10_NBSL_EXT_ERR_RES 0x0 307 #define F10_NBSL_EXT_ERR_ECC 0x8 308 309 /* Next two are overloaded values */ 310 #define F10_NBSL_EXT_ERR_LINK_PROTO 0xB 311 #define F10_NBSL_EXT_ERR_L3_PROTO 0xB 312 313 #define F10_NBSL_EXT_ERR_NB_ARRAY 0xC 314 #define F10_NBSL_EXT_ERR_DRAM_PARITY 0xD 315 #define F10_NBSL_EXT_ERR_LINK_RETRY 0xE 316 317 /* Next two are overloaded values */ 318 #define F10_NBSL_EXT_ERR_GART_WALK 0xF 319 #define F10_NBSL_EXT_ERR_DEV_WALK 0xF 320 321 /* 0x10 to 0x1B: Reserved */ 322 #define F10_NBSL_EXT_ERR_L3_DATA 0x1C 323 #define F10_NBSL_EXT_ERR_L3_TAG 0x1D 324 #define F10_NBSL_EXT_ERR_L3_LRU 0x1E 325 326 /* K8: Normalized Extended Error Codes */ 327 #define K8_NBSL_EXT_ERR_ECC 0x0 328 #define K8_NBSL_EXT_ERR_CRC 0x1 329 #define K8_NBSL_EXT_ERR_SYNC 0x2 330 #define K8_NBSL_EXT_ERR_MST 0x3 331 #define K8_NBSL_EXT_ERR_TGT 0x4 332 #define K8_NBSL_EXT_ERR_GART 0x5 333 #define K8_NBSL_EXT_ERR_RMW 0x6 334 #define K8_NBSL_EXT_ERR_WDT 0x7 335 #define K8_NBSL_EXT_ERR_CHIPKILL_ECC 0x8 336 #define K8_NBSL_EXT_ERR_DRAM_PARITY 0xD 337 338 /* 339 * The following are for BUS type errors AFTER values have been normalized by 340 * shifting right 341 */ 342 #define K8_NBSL_PP_SRC 0x0 343 #define K8_NBSL_PP_RES 0x1 344 #define K8_NBSL_PP_OBS 0x2 345 #define K8_NBSL_PP_GENERIC 0x3 346 347 #define EXTRACT_ERR_CPU_MAP(x) ((x) & 0xF) 348 349 #define K8_NBEAL 0x50 350 #define K8_NBEAH 0x54 351 #define K8_SCRCTRL 0x58 352 353 #define F10_NB_CFG_LOW 0x88 354 #define F10_NB_CFG_LOW_ENABLE_EXT_CFG BIT(14) 355 356 #define F10_NB_CFG_HIGH 0x8C 357 358 #define F10_ONLINE_SPARE 0xB0 359 #define F10_ONLINE_SPARE_SWAPDONE0(x) ((x) & BIT(1)) 360 #define F10_ONLINE_SPARE_SWAPDONE1(x) ((x) & BIT(3)) 361 #define F10_ONLINE_SPARE_BADDRAM_CS0(x) (((x) >> 4) & 0x00000007) 362 #define F10_ONLINE_SPARE_BADDRAM_CS1(x) (((x) >> 8) & 0x00000007) 363 364 #define F10_NB_ARRAY_ADDR 0xB8 365 366 #define F10_NB_ARRAY_DRAM_ECC 0x80000000 367 368 /* Bits [2:1] are used to select 16-byte section within a 64-byte cacheline */ 369 #define SET_NB_ARRAY_ADDRESS(section) (((section) & 0x3) << 1) 370 371 #define F10_NB_ARRAY_DATA 0xBC 372 373 #define SET_NB_DRAM_INJECTION_WRITE(word, bits) \ 374 (BIT(((word) & 0xF) + 20) | \ 375 BIT(17) | bits) 376 377 #define SET_NB_DRAM_INJECTION_READ(word, bits) \ 378 (BIT(((word) & 0xF) + 20) | \ 379 BIT(16) | bits) 380 381 #define K8_NBCAP 0xE8 382 #define K8_NBCAP_CORES (BIT(12)|BIT(13)) 383 #define K8_NBCAP_CHIPKILL BIT(4) 384 #define K8_NBCAP_SECDED BIT(3) 385 #define K8_NBCAP_8_NODE BIT(2) 386 #define K8_NBCAP_DUAL_NODE BIT(1) 387 #define K8_NBCAP_DCT_DUAL BIT(0) 388 389 /* 390 * MSR Regs 391 */ 392 #define K8_MSR_MCGCTL 0x017b 393 #define K8_MSR_MCGCTL_NBE BIT(4) 394 395 #define K8_MSR_MC4CTL 0x0410 396 #define K8_MSR_MC4STAT 0x0411 397 #define K8_MSR_MC4ADDR 0x0412 398 399 /* AMD sets the first MC device at device ID 0x18. */ 400 static inline int get_node_id(struct pci_dev *pdev) 401 { 402 return PCI_SLOT(pdev->devfn) - 0x18; 403 } 404 405 enum amd64_chipset_families { 406 K8_CPUS = 0, 407 F10_CPUS, 408 F11_CPUS, 409 }; 410 411 /* Error injection control structure */ 412 struct error_injection { 413 u32 section; 414 u32 word; 415 u32 bit_map; 416 }; 417 418 struct amd64_pvt { 419 /* pci_device handles which we utilize */ 420 struct pci_dev *addr_f1_ctl; 421 struct pci_dev *dram_f2_ctl; 422 struct pci_dev *misc_f3_ctl; 423 424 int mc_node_id; /* MC index of this MC node */ 425 int ext_model; /* extended model value of this node */ 426 427 struct low_ops *ops; /* pointer to per PCI Device ID func table */ 428 429 int channel_count; 430 431 /* Raw registers */ 432 u32 dclr0; /* DRAM Configuration Low DCT0 reg */ 433 u32 dclr1; /* DRAM Configuration Low DCT1 reg */ 434 u32 dchr0; /* DRAM Configuration High DCT0 reg */ 435 u32 dchr1; /* DRAM Configuration High DCT1 reg */ 436 u32 nbcap; /* North Bridge Capabilities */ 437 u32 nbcfg; /* F10 North Bridge Configuration */ 438 u32 ext_nbcfg; /* Extended F10 North Bridge Configuration */ 439 u32 dhar; /* DRAM Hoist reg */ 440 u32 dbam0; /* DRAM Base Address Mapping reg for DCT0 */ 441 u32 dbam1; /* DRAM Base Address Mapping reg for DCT1 */ 442 443 /* DRAM CS Base Address Registers F2x[1,0][5C:40] */ 444 u32 dcsb0[MAX_CS_COUNT]; 445 u32 dcsb1[MAX_CS_COUNT]; 446 447 /* DRAM CS Mask Registers F2x[1,0][6C:60] */ 448 u32 dcsm0[MAX_CS_COUNT]; 449 u32 dcsm1[MAX_CS_COUNT]; 450 451 /* 452 * Decoded parts of DRAM BASE and LIMIT Registers 453 * F1x[78,70,68,60,58,50,48,40] 454 */ 455 u64 dram_base[DRAM_REG_COUNT]; 456 u64 dram_limit[DRAM_REG_COUNT]; 457 u8 dram_IntlvSel[DRAM_REG_COUNT]; 458 u8 dram_IntlvEn[DRAM_REG_COUNT]; 459 u8 dram_DstNode[DRAM_REG_COUNT]; 460 u8 dram_rw_en[DRAM_REG_COUNT]; 461 462 /* 463 * The following fields are set at (load) run time, after CPU revision 464 * has been determined, since the dct_base and dct_mask registers vary 465 * based on revision 466 */ 467 u32 dcsb_base; /* DCSB base bits */ 468 u32 dcsm_mask; /* DCSM mask bits */ 469 u32 cs_count; /* num chip selects (== num DCSB registers) */ 470 u32 num_dcsm; /* Number of DCSM registers */ 471 u32 dcs_mask_notused; /* DCSM notused mask bits */ 472 u32 dcs_shift; /* DCSB and DCSM shift value */ 473 474 u64 top_mem; /* top of memory below 4GB */ 475 u64 top_mem2; /* top of memory above 4GB */ 476 477 u32 dram_ctl_select_low; /* DRAM Controller Select Low Reg */ 478 u32 dram_ctl_select_high; /* DRAM Controller Select High Reg */ 479 u32 online_spare; /* On-Line spare Reg */ 480 481 /* temp storage for when input is received from sysfs */ 482 struct err_regs ctl_error_info; 483 484 /* place to store error injection parameters prior to issue */ 485 struct error_injection injection; 486 487 /* Save old hw registers' values before we modified them */ 488 u32 nbctl_mcgctl_saved; /* When true, following 2 are valid */ 489 u32 old_nbctl; 490 unsigned long old_mcgctl; /* per core on this node */ 491 492 /* MC Type Index value: socket F vs Family 10h */ 493 u32 mc_type_index; 494 495 /* misc settings */ 496 struct flags { 497 unsigned long cf8_extcfg:1; 498 } flags; 499 }; 500 501 struct scrubrate { 502 u32 scrubval; /* bit pattern for scrub rate */ 503 u32 bandwidth; /* bandwidth consumed (bytes/sec) */ 504 }; 505 506 extern struct scrubrate scrubrates[23]; 507 extern u32 revf_quad_ddr2_shift[16]; 508 extern const char *tt_msgs[4]; 509 extern const char *ll_msgs[4]; 510 extern const char *rrrr_msgs[16]; 511 extern const char *to_msgs[2]; 512 extern const char *pp_msgs[4]; 513 extern const char *ii_msgs[4]; 514 extern const char *ext_msgs[32]; 515 extern const char *htlink_msgs[8]; 516 517 #ifdef CONFIG_EDAC_DEBUG 518 #define NUM_DBG_ATTRS 9 519 #else 520 #define NUM_DBG_ATTRS 0 521 #endif 522 523 #ifdef CONFIG_EDAC_AMD64_ERROR_INJECTION 524 #define NUM_INJ_ATTRS 5 525 #else 526 #define NUM_INJ_ATTRS 0 527 #endif 528 529 extern struct mcidev_sysfs_attribute amd64_dbg_attrs[NUM_DBG_ATTRS], 530 amd64_inj_attrs[NUM_INJ_ATTRS]; 531 532 /* 533 * Each of the PCI Device IDs types have their own set of hardware accessor 534 * functions and per device encoding/decoding logic. 535 */ 536 struct low_ops { 537 int (*probe_valid_hardware)(struct amd64_pvt *pvt); 538 int (*early_channel_count)(struct amd64_pvt *pvt); 539 540 u64 (*get_error_address)(struct mem_ctl_info *mci, 541 struct err_regs *info); 542 void (*read_dram_base_limit)(struct amd64_pvt *pvt, int dram); 543 void (*read_dram_ctl_register)(struct amd64_pvt *pvt); 544 void (*map_sysaddr_to_csrow)(struct mem_ctl_info *mci, 545 struct err_regs *info, 546 u64 SystemAddr); 547 int (*dbam_map_to_pages)(struct amd64_pvt *pvt, int dram_map); 548 }; 549 550 struct amd64_family_type { 551 const char *ctl_name; 552 u16 addr_f1_ctl; 553 u16 misc_f3_ctl; 554 struct low_ops ops; 555 }; 556 557 static struct amd64_family_type amd64_family_types[]; 558 559 static inline const char *get_amd_family_name(int index) 560 { 561 return amd64_family_types[index].ctl_name; 562 } 563 564 static inline struct low_ops *family_ops(int index) 565 { 566 return &amd64_family_types[index].ops; 567 } 568 569 /* 570 * For future CPU versions, verify the following as new 'slow' rates appear and 571 * modify the necessary skip values for the supported CPU. 572 */ 573 #define K8_MIN_SCRUB_RATE_BITS 0x0 574 #define F10_MIN_SCRUB_RATE_BITS 0x5 575 #define F11_MIN_SCRUB_RATE_BITS 0x6 576 577 int amd64_get_dram_hole_info(struct mem_ctl_info *mci, u64 *hole_base, 578 u64 *hole_offset, u64 *hole_size); 579