drivers/edac/amd64_edac.h

   1 /*
   2  * AMD64 class Memory Controller kernel module
   3  *
   4  * Copyright (c) 2009 SoftwareBitMaker.
   5  * Copyright (c) 2009-15 Advanced Micro Devices, Inc.
   6  *
   7  * This file may be distributed under the terms of the
   8  * GNU General Public License.
   9  */
  10
  11 #include <linux/module.h>
  12 #include <linux/ctype.h>
  13 #include <linux/init.h>
  14 #include <linux/pci.h>
  15 #include <linux/pci_ids.h>
  16 #include <linux/slab.h>
  17 #include <linux/mmzone.h>
  18 #include <linux/edac.h>
  19 #include <asm/cpu_device_id.h>
  20 #include <asm/msr.h>
  21 #include "edac_module.h"
  22 #include "mce_amd.h"
  23
  24 #define amd64_info(fmt, arg...) \
  25         edac_printk(KERN_INFO, "amd64", fmt, ##arg)
  26
  27 #define amd64_warn(fmt, arg...) \
  28         edac_printk(KERN_WARNING, "amd64", "Warning: " fmt, ##arg)
  29
  30 #define amd64_err(fmt, arg...) \
  31         edac_printk(KERN_ERR, "amd64", "Error: " fmt, ##arg)
  32
  33 #define amd64_mc_warn(mci, fmt, arg...) \
  34         edac_mc_chipset_printk(mci, KERN_WARNING, "amd64", fmt, ##arg)
  35
  36 #define amd64_mc_err(mci, fmt, arg...) \
  37         edac_mc_chipset_printk(mci, KERN_ERR, "amd64", fmt, ##arg)
  38
  39 /*
  40  * Throughout the comments in this code, the following terms are used:
  41  *
  42  *      SysAddr, DramAddr, and InputAddr
  43  *
  44  *  These terms come directly from the amd64 documentation
  45  * (AMD publication #26094).  They are defined as follows:
  46  *
  47  *     SysAddr:
  48  *         This is a physical address generated by a CPU core or a device
  49  *         doing DMA.  If generated by a CPU core, a SysAddr is the result of
  50  *         a virtual to physical address translation by the CPU core's address
  51  *         translation mechanism (MMU).
  52  *
  53  *     DramAddr:
  54  *         A DramAddr is derived from a SysAddr by subtracting an offset that
  55  *         depends on which node the SysAddr maps to and whether the SysAddr
  56  *         is within a range affected by memory hoisting.  The DRAM Base
  57  *         (section 3.4.4.1) and DRAM Limit (section 3.4.4.2) registers
  58  *         determine which node a SysAddr maps to.
  59  *
  60  *         If the DRAM Hole Address Register (DHAR) is enabled and the SysAddr
  61  *         is within the range of addresses specified by this register, then
  62  *         a value x from the DHAR is subtracted from the SysAddr to produce a
  63  *         DramAddr.  Here, x represents the base address for the node that
  64  *         the SysAddr maps to plus an offset due to memory hoisting.  See
  65  *         section 3.4.8 and the comments in amd64_get_dram_hole_info() and
  66  *         sys_addr_to_dram_addr() below for more information.
  67  *
  68  *         If the SysAddr is not affected by the DHAR then a value y is
  69  *         subtracted from the SysAddr to produce a DramAddr.  Here, y is the
  70  *         base address for the node that the SysAddr maps to.  See section
  71  *         3.4.4 and the comments in sys_addr_to_dram_addr() below for more
  72  *         information.
  73  *
  74  *     InputAddr:
  75  *         A DramAddr is translated to an InputAddr before being passed to the
  76  *         memory controller for the node that the DramAddr is associated
  77  *         with.  The memory controller then maps the InputAddr to a csrow.
  78  *         If node interleaving is not in use, then the InputAddr has the same
  79  *         value as the DramAddr.  Otherwise, the InputAddr is produced by
  80  *         discarding the bits used for node interleaving from the DramAddr.
  81  *         See section 3.4.4 for more information.
  82  *
  83  *         The memory controller for a given node uses its DRAM CS Base and
  84  *         DRAM CS Mask registers to map an InputAddr to a csrow.  See
  85  *         sections 3.5.4 and 3.5.5 for more information.
  86  */
  87
  88 #define EDAC_AMD64_VERSION              "3.5.0"
  89 #define EDAC_MOD_STR                    "amd64_edac"
  90
  91 /* Extended Model from CPUID, for CPU Revision numbers */
  92 #define K8_REV_D                        1
  93 #define K8_REV_E                        2
  94 #define K8_REV_F                        4
  95
  96 /* Hardware limit on ChipSelect rows per MC and processors per system */
  97 #define NUM_CHIPSELECTS                 8
  98 #define DRAM_RANGES                     8
  99
 100 #define ON true
 101 #define OFF false
 102
 103 /*
 104  * PCI-defined configuration space registers
 105  */
 106 #define PCI_DEVICE_ID_AMD_15H_NB_F1     0x1601
 107 #define PCI_DEVICE_ID_AMD_15H_NB_F2     0x1602
 108 #define PCI_DEVICE_ID_AMD_15H_M30H_NB_F1 0x141b
 109 #define PCI_DEVICE_ID_AMD_15H_M30H_NB_F2 0x141c
 110 #define PCI_DEVICE_ID_AMD_15H_M60H_NB_F1 0x1571
 111 #define PCI_DEVICE_ID_AMD_15H_M60H_NB_F2 0x1572
 112 #define PCI_DEVICE_ID_AMD_16H_NB_F1     0x1531
 113 #define PCI_DEVICE_ID_AMD_16H_NB_F2     0x1532
 114 #define PCI_DEVICE_ID_AMD_16H_M30H_NB_F1 0x1581
 115 #define PCI_DEVICE_ID_AMD_16H_M30H_NB_F2 0x1582
 116 #define PCI_DEVICE_ID_AMD_17H_DF_F0     0x1460
 117 #define PCI_DEVICE_ID_AMD_17H_DF_F6     0x1466
 118
 119 /*
 120  * Function 1 - Address Map
 121  */
 122 #define DRAM_BASE_LO                    0x40
 123 #define DRAM_LIMIT_LO                   0x44
 124
 125 /*
 126  * F15 M30h D18F1x2[1C:00]
 127  */
 128 #define DRAM_CONT_BASE                  0x200
 129 #define DRAM_CONT_LIMIT                 0x204
 130
 131 /*
 132  * F15 M30h D18F1x2[4C:40]
 133  */
 134 #define DRAM_CONT_HIGH_OFF              0x240
 135
 136 #define dram_rw(pvt, i)                 ((u8)(pvt->ranges[i].base.lo & 0x3))
 137 #define dram_intlv_sel(pvt, i)          ((u8)((pvt->ranges[i].lim.lo >> 8) & 0x7))
 138 #define dram_dst_node(pvt, i)           ((u8)(pvt->ranges[i].lim.lo & 0x7))
 139
 140 #define DHAR                            0xf0
 141 #define dhar_mem_hoist_valid(pvt)       ((pvt)->dhar & BIT(1))
 142 #define dhar_base(pvt)                  ((pvt)->dhar & 0xff000000)
 143 #define k8_dhar_offset(pvt)             (((pvt)->dhar & 0x0000ff00) << 16)
 144
 145                                         /* NOTE: Extra mask bit vs K8 */
 146 #define f10_dhar_offset(pvt)            (((pvt)->dhar & 0x0000ff80) << 16)
 147
 148 #define DCT_CFG_SEL                     0x10C
 149
 150 #define DRAM_LOCAL_NODE_BASE            0x120
 151 #define DRAM_LOCAL_NODE_LIM             0x124
 152
 153 #define DRAM_BASE_HI                    0x140
 154 #define DRAM_LIMIT_HI                   0x144
 155
 156
 157 /*
 158  * Function 2 - DRAM controller
 159  */
 160 #define DCSB0                           0x40
 161 #define DCSB1                           0x140
 162 #define DCSB_CS_ENABLE                  BIT(0)
 163
 164 #define DCSM0                           0x60
 165 #define DCSM1                           0x160
 166
 167 #define csrow_enabled(i, dct, pvt)      ((pvt)->csels[(dct)].csbases[(i)] & DCSB_CS_ENABLE)
 168
 169 #define DRAM_CONTROL                    0x78
 170
 171 #define DBAM0                           0x80
 172 #define DBAM1                           0x180
 173
 174 /* Extract the DIMM 'type' on the i'th DIMM from the DBAM reg value passed */
 175 #define DBAM_DIMM(i, reg)               ((((reg) >> (4*(i)))) & 0xF)
 176
 177 #define DBAM_MAX_VALUE                  11
 178
 179 #define DCLR0                           0x90
 180 #define DCLR1                           0x190
 181 #define REVE_WIDTH_128                  BIT(16)
 182 #define WIDTH_128                       BIT(11)
 183
 184 #define DCHR0                           0x94
 185 #define DCHR1                           0x194
 186 #define DDR3_MODE                       BIT(8)
 187
 188 #define DCT_SEL_LO                      0x110
 189 #define dct_high_range_enabled(pvt)     ((pvt)->dct_sel_lo & BIT(0))
 190 #define dct_interleave_enabled(pvt)     ((pvt)->dct_sel_lo & BIT(2))
 191
 192 #define dct_ganging_enabled(pvt)        ((boot_cpu_data.x86 == 0x10) && ((pvt)->dct_sel_lo & BIT(4)))
 193
 194 #define dct_data_intlv_enabled(pvt)     ((pvt)->dct_sel_lo & BIT(5))
 195 #define dct_memory_cleared(pvt)         ((pvt)->dct_sel_lo & BIT(10))
 196
 197 #define SWAP_INTLV_REG                  0x10c
 198
 199 #define DCT_SEL_HI                      0x114
 200
 201 #define F15H_M60H_SCRCTRL               0x1C8
 202 #define F17H_SCR_BASE_ADDR              0x48
 203 #define F17H_SCR_LIMIT_ADDR             0x4C
 204
 205 /*
 206  * Function 3 - Misc Control
 207  */
 208 #define NBCTL                           0x40
 209
 210 #define NBCFG                           0x44
 211 #define NBCFG_CHIPKILL                  BIT(23)
 212 #define NBCFG_ECC_ENABLE                BIT(22)
 213
 214 /* F3x48: NBSL */
 215 #define F10_NBSL_EXT_ERR_ECC            0x8
 216 #define NBSL_PP_OBS                     0x2
 217
 218 #define SCRCTRL                         0x58
 219
 220 #define F10_ONLINE_SPARE                0xB0
 221 #define online_spare_swap_done(pvt, c)  (((pvt)->online_spare >> (1 + 2 * (c))) & 0x1)
 222 #define online_spare_bad_dramcs(pvt, c) (((pvt)->online_spare >> (4 + 4 * (c))) & 0x7)
 223
 224 #define F10_NB_ARRAY_ADDR               0xB8
 225 #define F10_NB_ARRAY_DRAM               BIT(31)
 226
 227 /* Bits [2:1] are used to select 16-byte section within a 64-byte cacheline  */
 228 #define SET_NB_ARRAY_ADDR(section)      (((section) & 0x3) << 1)
 229
 230 #define F10_NB_ARRAY_DATA               0xBC
 231 #define F10_NB_ARR_ECC_WR_REQ           BIT(17)
 232 #define SET_NB_DRAM_INJECTION_WRITE(inj)  \
 233                                         (BIT(((inj.word) & 0xF) + 20) | \
 234                                         F10_NB_ARR_ECC_WR_REQ | inj.bit_map)
 235 #define SET_NB_DRAM_INJECTION_READ(inj)  \
 236                                         (BIT(((inj.word) & 0xF) + 20) | \
 237                                         BIT(16) |  inj.bit_map)
 238
 239
 240 #define NBCAP                           0xE8
 241 #define NBCAP_CHIPKILL                  BIT(4)
 242 #define NBCAP_SECDED                    BIT(3)
 243 #define NBCAP_DCT_DUAL                  BIT(0)
 244
 245 #define EXT_NB_MCA_CFG                  0x180
 246
 247 /* MSRs */
 248 #define MSR_MCGCTL_NBE                  BIT(4)
 249
 250 /* F17h */
 251
 252 /* F0: */
 253 #define DF_DHAR                         0x104
 254
 255 /* UMC CH register offsets */
 256 #define UMCCH_BASE_ADDR                 0x0
 257 #define UMCCH_ADDR_MASK                 0x20
 258 #define UMCCH_ADDR_CFG                  0x30
 259 #define UMCCH_DIMM_CFG                  0x80
 260 #define UMCCH_UMC_CFG                   0x100
 261 #define UMCCH_SDP_CTRL                  0x104
 262 #define UMCCH_ECC_CTRL                  0x14C
 263 #define UMCCH_ECC_BAD_SYMBOL            0xD90
 264 #define UMCCH_UMC_CAP                   0xDF0
 265 #define UMCCH_UMC_CAP_HI                0xDF4
 266
 267 /* UMC CH bitfields */
 268 #define UMC_ECC_CHIPKILL_CAP            BIT(31)
 269 #define UMC_ECC_ENABLED                 BIT(30)
 270
 271 #define UMC_SDP_INIT                    BIT(31)
 272
 273 #define NUM_UMCS                        2
 274
 275 enum amd_families {
 276         K8_CPUS = 0,
 277         F10_CPUS,
 278         F15_CPUS,
 279         F15_M30H_CPUS,
 280         F15_M60H_CPUS,
 281         F16_CPUS,
 282         F16_M30H_CPUS,
 283         F17_CPUS,
 284         NUM_FAMILIES,
 285 };
 286
 287 /* Error injection control structure */
 288 struct error_injection {
 289         u32      section;
 290         u32      word;
 291         u32      bit_map;
 292 };
 293
 294 /* low and high part of PCI config space regs */
 295 struct reg_pair {
 296         u32 lo, hi;
 297 };
 298
 299 /*
 300  * See F1x[1, 0][7C:40] DRAM Base/Limit Registers
 301  */
 302 struct dram_range {
 303         struct reg_pair base;
 304         struct reg_pair lim;
 305 };
 306
 307 /* A DCT chip selects collection */
 308 struct chip_select {
 309         u32 csbases[NUM_CHIPSELECTS];
 310         u8 b_cnt;
 311
 312         u32 csmasks[NUM_CHIPSELECTS];
 313         u8 m_cnt;
 314 };
 315
 316 struct amd64_umc {
 317         u32 dimm_cfg;           /* DIMM Configuration reg */
 318         u32 umc_cfg;            /* Configuration reg */
 319         u32 sdp_ctrl;           /* SDP Control reg */
 320         u32 ecc_ctrl;           /* DRAM ECC Control reg */
 321         u32 umc_cap_hi;         /* Capabilities High reg */
 322 };
 323
 324 struct amd64_pvt {
 325         struct low_ops *ops;
 326
 327         /* pci_device handles which we utilize */
 328         struct pci_dev *F0, *F1, *F2, *F3, *F6;
 329
 330         u16 mc_node_id;         /* MC index of this MC node */
 331         u8 fam;                 /* CPU family */
 332         u8 model;               /* ... model */
 333         u8 stepping;            /* ... stepping */
 334
 335         int ext_model;          /* extended model value of this node */
 336         int channel_count;
 337
 338         /* Raw registers */
 339         u32 dclr0;              /* DRAM Configuration Low DCT0 reg */
 340         u32 dclr1;              /* DRAM Configuration Low DCT1 reg */
 341         u32 dchr0;              /* DRAM Configuration High DCT0 reg */
 342         u32 dchr1;              /* DRAM Configuration High DCT1 reg */
 343         u32 nbcap;              /* North Bridge Capabilities */
 344         u32 nbcfg;              /* F10 North Bridge Configuration */
 345         u32 ext_nbcfg;          /* Extended F10 North Bridge Configuration */
 346         u32 dhar;               /* DRAM Hoist reg */
 347         u32 dbam0;              /* DRAM Base Address Mapping reg for DCT0 */
 348         u32 dbam1;              /* DRAM Base Address Mapping reg for DCT1 */
 349
 350         /* one for each DCT */
 351         struct chip_select csels[2];
 352
 353         /* DRAM base and limit pairs F1x[78,70,68,60,58,50,48,40] */
 354         struct dram_range ranges[DRAM_RANGES];
 355
 356         u64 top_mem;            /* top of memory below 4GB */
 357         u64 top_mem2;           /* top of memory above 4GB */
 358
 359         u32 dct_sel_lo;         /* DRAM Controller Select Low */
 360         u32 dct_sel_hi;         /* DRAM Controller Select High */
 361         u32 online_spare;       /* On-Line spare Reg */
 362
 363         /* x4 or x8 syndromes in use */
 364         u8 ecc_sym_sz;
 365
 366         /* place to store error injection parameters prior to issue */
 367         struct error_injection injection;
 368
 369         /* cache the dram_type */
 370         enum mem_type dram_type;
 371
 372         struct amd64_umc *umc;  /* UMC registers */
 373 };
 374
 375 enum err_codes {
 376         DECODE_OK       =  0,
 377         ERR_NODE        = -1,
 378         ERR_CSROW       = -2,
 379         ERR_CHANNEL     = -3,
 380         ERR_SYND        = -4,
 381         ERR_NORM_ADDR   = -5,
 382 };
 383
 384 struct err_info {
 385         int err_code;
 386         struct mem_ctl_info *src_mci;
 387         int csrow;
 388         int channel;
 389         u16 syndrome;
 390         u32 page;
 391         u32 offset;
 392 };
 393
 394 static inline u32 get_umc_base(u8 channel)
 395 {
 396         /* ch0: 0x50000, ch1: 0x150000 */
 397         return 0x50000 + (!!channel << 20);
 398 }
 399
 400 static inline u64 get_dram_base(struct amd64_pvt *pvt, u8 i)
 401 {
 402         u64 addr = ((u64)pvt->ranges[i].base.lo & 0xffff0000) << 8;
 403
 404         if (boot_cpu_data.x86 == 0xf)
 405                 return addr;
 406
 407         return (((u64)pvt->ranges[i].base.hi & 0x000000ff) << 40) | addr;
 408 }
 409
 410 static inline u64 get_dram_limit(struct amd64_pvt *pvt, u8 i)
 411 {
 412         u64 lim = (((u64)pvt->ranges[i].lim.lo & 0xffff0000) << 8) | 0x00ffffff;
 413
 414         if (boot_cpu_data.x86 == 0xf)
 415                 return lim;
 416
 417         return (((u64)pvt->ranges[i].lim.hi & 0x000000ff) << 40) | lim;
 418 }
 419
 420 static inline u16 extract_syndrome(u64 status)
 421 {
 422         return ((status >> 47) & 0xff) | ((status >> 16) & 0xff00);
 423 }
 424
 425 static inline u8 dct_sel_interleave_addr(struct amd64_pvt *pvt)
 426 {
 427         if (pvt->fam == 0x15 && pvt->model >= 0x30)
 428                 return (((pvt->dct_sel_hi >> 9) & 0x1) << 2) |
 429                         ((pvt->dct_sel_lo >> 6) & 0x3);
 430
 431         return  ((pvt)->dct_sel_lo >> 6) & 0x3;
 432 }
 433 /*
 434  * per-node ECC settings descriptor
 435  */
 436 struct ecc_settings {
 437         u32 old_nbctl;
 438         bool nbctl_valid;
 439
 440         struct flags {
 441                 unsigned long nb_mce_enable:1;
 442                 unsigned long nb_ecc_prev:1;
 443         } flags;
 444 };
 445
 446 #ifdef CONFIG_EDAC_DEBUG
 447 extern const struct attribute_group amd64_edac_dbg_group;
 448 #endif
 449
 450 #ifdef CONFIG_EDAC_AMD64_ERROR_INJECTION
 451 extern const struct attribute_group amd64_edac_inj_group;
 452 #endif
 453
 454 /*
 455  * Each of the PCI Device IDs types have their own set of hardware accessor
 456  * functions and per device encoding/decoding logic.
 457  */
 458 struct low_ops {
 459         int (*early_channel_count)      (struct amd64_pvt *pvt);
 460         void (*map_sysaddr_to_csrow)    (struct mem_ctl_info *mci, u64 sys_addr,
 461                                          struct err_info *);
 462         int (*dbam_to_cs)               (struct amd64_pvt *pvt, u8 dct,
 463                                          unsigned cs_mode, int cs_mask_nr);
 464 };
 465
 466 struct amd64_family_type {
 467         const char *ctl_name;
 468         u16 f0_id, f1_id, f2_id, f6_id;
 469         struct low_ops ops;
 470 };
 471
 472 int __amd64_read_pci_cfg_dword(struct pci_dev *pdev, int offset,
 473                                u32 *val, const char *func);
 474 int __amd64_write_pci_cfg_dword(struct pci_dev *pdev, int offset,
 475                                 u32 val, const char *func);
 476
 477 #define amd64_read_pci_cfg(pdev, offset, val)   \
 478         __amd64_read_pci_cfg_dword(pdev, offset, val, __func__)
 479
 480 #define amd64_write_pci_cfg(pdev, offset, val)  \
 481         __amd64_write_pci_cfg_dword(pdev, offset, val, __func__)
 482
 483 int amd64_get_dram_hole_info(struct mem_ctl_info *mci, u64 *hole_base,
 484                              u64 *hole_offset, u64 *hole_size);
 485
 486 #define to_mci(k) container_of(k, struct mem_ctl_info, dev)
 487
 488 /* Injection helpers */
 489 static inline void disable_caches(void *dummy)
 490 {
 491         write_cr0(read_cr0() | X86_CR0_CD);
 492         wbinvd();
 493 }
 494
 495 static inline void enable_caches(void *dummy)
 496 {
 497         write_cr0(read_cr0() & ~X86_CR0_CD);
 498 }
 499
 500 static inline u8 dram_intlv_en(struct amd64_pvt *pvt, unsigned int i)
 501 {
 502         if (pvt->fam == 0x15 && pvt->model >= 0x30) {
 503                 u32 tmp;
 504                 amd64_read_pci_cfg(pvt->F1, DRAM_CONT_LIMIT, &tmp);
 505                 return (u8) tmp & 0xF;
 506         }
 507         return (u8) (pvt->ranges[i].base.lo >> 8) & 0x7;
 508 }
 509
 510 static inline u8 dhar_valid(struct amd64_pvt *pvt)
 511 {
 512         if (pvt->fam == 0x15 && pvt->model >= 0x30) {
 513                 u32 tmp;
 514                 amd64_read_pci_cfg(pvt->F1, DRAM_CONT_BASE, &tmp);
 515                 return (tmp >> 1) & BIT(0);
 516         }
 517         return (pvt)->dhar & BIT(0);
 518 }
 519
 520 static inline u32 dct_sel_baseaddr(struct amd64_pvt *pvt)
 521 {
 522         if (pvt->fam == 0x15 && pvt->model >= 0x30) {
 523                 u32 tmp;
 524                 amd64_read_pci_cfg(pvt->F1, DRAM_CONT_BASE, &tmp);
 525                 return (tmp >> 11) & 0x1FFF;
 526         }
 527         return (pvt)->dct_sel_lo & 0xFFFFF800;
 528 }