linux/drivers/edac/amd64_edac.h
<<
>>
Prefs
   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#define PCI_DEVICE_ID_AMD_17H_M10H_DF_F0 0x15e8
 119#define PCI_DEVICE_ID_AMD_17H_M10H_DF_F6 0x15ee
 120
 121/*
 122 * Function 1 - Address Map
 123 */
 124#define DRAM_BASE_LO                    0x40
 125#define DRAM_LIMIT_LO                   0x44
 126
 127/*
 128 * F15 M30h D18F1x2[1C:00]
 129 */
 130#define DRAM_CONT_BASE                  0x200
 131#define DRAM_CONT_LIMIT                 0x204
 132
 133/*
 134 * F15 M30h D18F1x2[4C:40]
 135 */
 136#define DRAM_CONT_HIGH_OFF              0x240
 137
 138#define dram_rw(pvt, i)                 ((u8)(pvt->ranges[i].base.lo & 0x3))
 139#define dram_intlv_sel(pvt, i)          ((u8)((pvt->ranges[i].lim.lo >> 8) & 0x7))
 140#define dram_dst_node(pvt, i)           ((u8)(pvt->ranges[i].lim.lo & 0x7))
 141
 142#define DHAR                            0xf0
 143#define dhar_mem_hoist_valid(pvt)       ((pvt)->dhar & BIT(1))
 144#define dhar_base(pvt)                  ((pvt)->dhar & 0xff000000)
 145#define k8_dhar_offset(pvt)             (((pvt)->dhar & 0x0000ff00) << 16)
 146
 147                                        /* NOTE: Extra mask bit vs K8 */
 148#define f10_dhar_offset(pvt)            (((pvt)->dhar & 0x0000ff80) << 16)
 149
 150#define DCT_CFG_SEL                     0x10C
 151
 152#define DRAM_LOCAL_NODE_BASE            0x120
 153#define DRAM_LOCAL_NODE_LIM             0x124
 154
 155#define DRAM_BASE_HI                    0x140
 156#define DRAM_LIMIT_HI                   0x144
 157
 158
 159/*
 160 * Function 2 - DRAM controller
 161 */
 162#define DCSB0                           0x40
 163#define DCSB1                           0x140
 164#define DCSB_CS_ENABLE                  BIT(0)
 165
 166#define DCSM0                           0x60
 167#define DCSM1                           0x160
 168
 169#define csrow_enabled(i, dct, pvt)      ((pvt)->csels[(dct)].csbases[(i)] & DCSB_CS_ENABLE)
 170
 171#define DRAM_CONTROL                    0x78
 172
 173#define DBAM0                           0x80
 174#define DBAM1                           0x180
 175
 176/* Extract the DIMM 'type' on the i'th DIMM from the DBAM reg value passed */
 177#define DBAM_DIMM(i, reg)               ((((reg) >> (4*(i)))) & 0xF)
 178
 179#define DBAM_MAX_VALUE                  11
 180
 181#define DCLR0                           0x90
 182#define DCLR1                           0x190
 183#define REVE_WIDTH_128                  BIT(16)
 184#define WIDTH_128                       BIT(11)
 185
 186#define DCHR0                           0x94
 187#define DCHR1                           0x194
 188#define DDR3_MODE                       BIT(8)
 189
 190#define DCT_SEL_LO                      0x110
 191#define dct_high_range_enabled(pvt)     ((pvt)->dct_sel_lo & BIT(0))
 192#define dct_interleave_enabled(pvt)     ((pvt)->dct_sel_lo & BIT(2))
 193
 194#define dct_ganging_enabled(pvt)        ((boot_cpu_data.x86 == 0x10) && ((pvt)->dct_sel_lo & BIT(4)))
 195
 196#define dct_data_intlv_enabled(pvt)     ((pvt)->dct_sel_lo & BIT(5))
 197#define dct_memory_cleared(pvt)         ((pvt)->dct_sel_lo & BIT(10))
 198
 199#define SWAP_INTLV_REG                  0x10c
 200
 201#define DCT_SEL_HI                      0x114
 202
 203#define F15H_M60H_SCRCTRL               0x1C8
 204#define F17H_SCR_BASE_ADDR              0x48
 205#define F17H_SCR_LIMIT_ADDR             0x4C
 206
 207/*
 208 * Function 3 - Misc Control
 209 */
 210#define NBCTL                           0x40
 211
 212#define NBCFG                           0x44
 213#define NBCFG_CHIPKILL                  BIT(23)
 214#define NBCFG_ECC_ENABLE                BIT(22)
 215
 216/* F3x48: NBSL */
 217#define F10_NBSL_EXT_ERR_ECC            0x8
 218#define NBSL_PP_OBS                     0x2
 219
 220#define SCRCTRL                         0x58
 221
 222#define F10_ONLINE_SPARE                0xB0
 223#define online_spare_swap_done(pvt, c)  (((pvt)->online_spare >> (1 + 2 * (c))) & 0x1)
 224#define online_spare_bad_dramcs(pvt, c) (((pvt)->online_spare >> (4 + 4 * (c))) & 0x7)
 225
 226#define F10_NB_ARRAY_ADDR               0xB8
 227#define F10_NB_ARRAY_DRAM               BIT(31)
 228
 229/* Bits [2:1] are used to select 16-byte section within a 64-byte cacheline  */
 230#define SET_NB_ARRAY_ADDR(section)      (((section) & 0x3) << 1)
 231
 232#define F10_NB_ARRAY_DATA               0xBC
 233#define F10_NB_ARR_ECC_WR_REQ           BIT(17)
 234#define SET_NB_DRAM_INJECTION_WRITE(inj)  \
 235                                        (BIT(((inj.word) & 0xF) + 20) | \
 236                                        F10_NB_ARR_ECC_WR_REQ | inj.bit_map)
 237#define SET_NB_DRAM_INJECTION_READ(inj)  \
 238                                        (BIT(((inj.word) & 0xF) + 20) | \
 239                                        BIT(16) |  inj.bit_map)
 240
 241
 242#define NBCAP                           0xE8
 243#define NBCAP_CHIPKILL                  BIT(4)
 244#define NBCAP_SECDED                    BIT(3)
 245#define NBCAP_DCT_DUAL                  BIT(0)
 246
 247#define EXT_NB_MCA_CFG                  0x180
 248
 249/* MSRs */
 250#define MSR_MCGCTL_NBE                  BIT(4)
 251
 252/* F17h */
 253
 254/* F0: */
 255#define DF_DHAR                         0x104
 256
 257/* UMC CH register offsets */
 258#define UMCCH_BASE_ADDR                 0x0
 259#define UMCCH_ADDR_MASK                 0x20
 260#define UMCCH_ADDR_CFG                  0x30
 261#define UMCCH_DIMM_CFG                  0x80
 262#define UMCCH_UMC_CFG                   0x100
 263#define UMCCH_SDP_CTRL                  0x104
 264#define UMCCH_ECC_CTRL                  0x14C
 265#define UMCCH_ECC_BAD_SYMBOL            0xD90
 266#define UMCCH_UMC_CAP                   0xDF0
 267#define UMCCH_UMC_CAP_HI                0xDF4
 268
 269/* UMC CH bitfields */
 270#define UMC_ECC_CHIPKILL_CAP            BIT(31)
 271#define UMC_ECC_ENABLED                 BIT(30)
 272
 273#define UMC_SDP_INIT                    BIT(31)
 274
 275#define NUM_UMCS                        2
 276
 277enum amd_families {
 278        K8_CPUS = 0,
 279        F10_CPUS,
 280        F15_CPUS,
 281        F15_M30H_CPUS,
 282        F15_M60H_CPUS,
 283        F16_CPUS,
 284        F16_M30H_CPUS,
 285        F17_CPUS,
 286        F17_M10H_CPUS,
 287        NUM_FAMILIES,
 288};
 289
 290/* Error injection control structure */
 291struct error_injection {
 292        u32      section;
 293        u32      word;
 294        u32      bit_map;
 295};
 296
 297/* low and high part of PCI config space regs */
 298struct reg_pair {
 299        u32 lo, hi;
 300};
 301
 302/*
 303 * See F1x[1, 0][7C:40] DRAM Base/Limit Registers
 304 */
 305struct dram_range {
 306        struct reg_pair base;
 307        struct reg_pair lim;
 308};
 309
 310/* A DCT chip selects collection */
 311struct chip_select {
 312        u32 csbases[NUM_CHIPSELECTS];
 313        u8 b_cnt;
 314
 315        u32 csmasks[NUM_CHIPSELECTS];
 316        u8 m_cnt;
 317};
 318
 319struct amd64_umc {
 320        u32 dimm_cfg;           /* DIMM Configuration reg */
 321        u32 umc_cfg;            /* Configuration reg */
 322        u32 sdp_ctrl;           /* SDP Control reg */
 323        u32 ecc_ctrl;           /* DRAM ECC Control reg */
 324        u32 umc_cap_hi;         /* Capabilities High reg */
 325};
 326
 327struct amd64_pvt {
 328        struct low_ops *ops;
 329
 330        /* pci_device handles which we utilize */
 331        struct pci_dev *F0, *F1, *F2, *F3, *F6;
 332
 333        u16 mc_node_id;         /* MC index of this MC node */
 334        u8 fam;                 /* CPU family */
 335        u8 model;               /* ... model */
 336        u8 stepping;            /* ... stepping */
 337
 338        int ext_model;          /* extended model value of this node */
 339        int channel_count;
 340
 341        /* Raw registers */
 342        u32 dclr0;              /* DRAM Configuration Low DCT0 reg */
 343        u32 dclr1;              /* DRAM Configuration Low DCT1 reg */
 344        u32 dchr0;              /* DRAM Configuration High DCT0 reg */
 345        u32 dchr1;              /* DRAM Configuration High DCT1 reg */
 346        u32 nbcap;              /* North Bridge Capabilities */
 347        u32 nbcfg;              /* F10 North Bridge Configuration */
 348        u32 ext_nbcfg;          /* Extended F10 North Bridge Configuration */
 349        u32 dhar;               /* DRAM Hoist reg */
 350        u32 dbam0;              /* DRAM Base Address Mapping reg for DCT0 */
 351        u32 dbam1;              /* DRAM Base Address Mapping reg for DCT1 */
 352
 353        /* one for each DCT */
 354        struct chip_select csels[2];
 355
 356        /* DRAM base and limit pairs F1x[78,70,68,60,58,50,48,40] */
 357        struct dram_range ranges[DRAM_RANGES];
 358
 359        u64 top_mem;            /* top of memory below 4GB */
 360        u64 top_mem2;           /* top of memory above 4GB */
 361
 362        u32 dct_sel_lo;         /* DRAM Controller Select Low */
 363        u32 dct_sel_hi;         /* DRAM Controller Select High */
 364        u32 online_spare;       /* On-Line spare Reg */
 365
 366        /* x4 or x8 syndromes in use */
 367        u8 ecc_sym_sz;
 368
 369        /* place to store error injection parameters prior to issue */
 370        struct error_injection injection;
 371
 372        /* cache the dram_type */
 373        enum mem_type dram_type;
 374
 375        struct amd64_umc *umc;  /* UMC registers */
 376};
 377
 378enum err_codes {
 379        DECODE_OK       =  0,
 380        ERR_NODE        = -1,
 381        ERR_CSROW       = -2,
 382        ERR_CHANNEL     = -3,
 383        ERR_SYND        = -4,
 384        ERR_NORM_ADDR   = -5,
 385};
 386
 387struct err_info {
 388        int err_code;
 389        struct mem_ctl_info *src_mci;
 390        int csrow;
 391        int channel;
 392        u16 syndrome;
 393        u32 page;
 394        u32 offset;
 395};
 396
 397static inline u32 get_umc_base(u8 channel)
 398{
 399        /* ch0: 0x50000, ch1: 0x150000 */
 400        return 0x50000 + (!!channel << 20);
 401}
 402
 403static inline u64 get_dram_base(struct amd64_pvt *pvt, u8 i)
 404{
 405        u64 addr = ((u64)pvt->ranges[i].base.lo & 0xffff0000) << 8;
 406
 407        if (boot_cpu_data.x86 == 0xf)
 408                return addr;
 409
 410        return (((u64)pvt->ranges[i].base.hi & 0x000000ff) << 40) | addr;
 411}
 412
 413static inline u64 get_dram_limit(struct amd64_pvt *pvt, u8 i)
 414{
 415        u64 lim = (((u64)pvt->ranges[i].lim.lo & 0xffff0000) << 8) | 0x00ffffff;
 416
 417        if (boot_cpu_data.x86 == 0xf)
 418                return lim;
 419
 420        return (((u64)pvt->ranges[i].lim.hi & 0x000000ff) << 40) | lim;
 421}
 422
 423static inline u16 extract_syndrome(u64 status)
 424{
 425        return ((status >> 47) & 0xff) | ((status >> 16) & 0xff00);
 426}
 427
 428static inline u8 dct_sel_interleave_addr(struct amd64_pvt *pvt)
 429{
 430        if (pvt->fam == 0x15 && pvt->model >= 0x30)
 431                return (((pvt->dct_sel_hi >> 9) & 0x1) << 2) |
 432                        ((pvt->dct_sel_lo >> 6) & 0x3);
 433
 434        return  ((pvt)->dct_sel_lo >> 6) & 0x3;
 435}
 436/*
 437 * per-node ECC settings descriptor
 438 */
 439struct ecc_settings {
 440        u32 old_nbctl;
 441        bool nbctl_valid;
 442
 443        struct flags {
 444                unsigned long nb_mce_enable:1;
 445                unsigned long nb_ecc_prev:1;
 446        } flags;
 447};
 448
 449#ifdef CONFIG_EDAC_DEBUG
 450extern const struct attribute_group amd64_edac_dbg_group;
 451#endif
 452
 453#ifdef CONFIG_EDAC_AMD64_ERROR_INJECTION
 454extern const struct attribute_group amd64_edac_inj_group;
 455#endif
 456
 457/*
 458 * Each of the PCI Device IDs types have their own set of hardware accessor
 459 * functions and per device encoding/decoding logic.
 460 */
 461struct low_ops {
 462        int (*early_channel_count)      (struct amd64_pvt *pvt);
 463        void (*map_sysaddr_to_csrow)    (struct mem_ctl_info *mci, u64 sys_addr,
 464                                         struct err_info *);
 465        int (*dbam_to_cs)               (struct amd64_pvt *pvt, u8 dct,
 466                                         unsigned cs_mode, int cs_mask_nr);
 467};
 468
 469struct amd64_family_type {
 470        const char *ctl_name;
 471        u16 f0_id, f1_id, f2_id, f6_id;
 472        struct low_ops ops;
 473};
 474
 475int __amd64_read_pci_cfg_dword(struct pci_dev *pdev, int offset,
 476                               u32 *val, const char *func);
 477int __amd64_write_pci_cfg_dword(struct pci_dev *pdev, int offset,
 478                                u32 val, const char *func);
 479
 480#define amd64_read_pci_cfg(pdev, offset, val)   \
 481        __amd64_read_pci_cfg_dword(pdev, offset, val, __func__)
 482
 483#define amd64_write_pci_cfg(pdev, offset, val)  \
 484        __amd64_write_pci_cfg_dword(pdev, offset, val, __func__)
 485
 486int amd64_get_dram_hole_info(struct mem_ctl_info *mci, u64 *hole_base,
 487                             u64 *hole_offset, u64 *hole_size);
 488
 489#define to_mci(k) container_of(k, struct mem_ctl_info, dev)
 490
 491/* Injection helpers */
 492static inline void disable_caches(void *dummy)
 493{
 494        write_cr0(read_cr0() | X86_CR0_CD);
 495        wbinvd();
 496}
 497
 498static inline void enable_caches(void *dummy)
 499{
 500        write_cr0(read_cr0() & ~X86_CR0_CD);
 501}
 502
 503static inline u8 dram_intlv_en(struct amd64_pvt *pvt, unsigned int i)
 504{
 505        if (pvt->fam == 0x15 && pvt->model >= 0x30) {
 506                u32 tmp;
 507                amd64_read_pci_cfg(pvt->F1, DRAM_CONT_LIMIT, &tmp);
 508                return (u8) tmp & 0xF;
 509        }
 510        return (u8) (pvt->ranges[i].base.lo >> 8) & 0x7;
 511}
 512
 513static inline u8 dhar_valid(struct amd64_pvt *pvt)
 514{
 515        if (pvt->fam == 0x15 && pvt->model >= 0x30) {
 516                u32 tmp;
 517                amd64_read_pci_cfg(pvt->F1, DRAM_CONT_BASE, &tmp);
 518                return (tmp >> 1) & BIT(0);
 519        }
 520        return (pvt)->dhar & BIT(0);
 521}
 522
 523static inline u32 dct_sel_baseaddr(struct amd64_pvt *pvt)
 524{
 525        if (pvt->fam == 0x15 && pvt->model >= 0x30) {
 526                u32 tmp;
 527                amd64_read_pci_cfg(pvt->F1, DRAM_CONT_BASE, &tmp);
 528                return (tmp >> 11) & 0x1FFF;
 529        }
 530        return (pvt)->dct_sel_lo & 0xFFFFF800;
 531}
 532