amd64_edac.h 14 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525
  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. #include <linux/module.h>
  11. #include <linux/ctype.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/mmzone.h>
  17. #include <linux/edac.h>
  18. #include <linux/bitfield.h>
  19. #include <asm/cpu_device_id.h>
  20. #include <asm/msr.h>
  21. #include "edac_module.h"
  22. #include "mce_amd.h"
  23. #define amd64_info(fmt, arg...) \
  24. edac_printk(KERN_INFO, "amd64", fmt, ##arg)
  25. #define amd64_warn(fmt, arg...) \
  26. edac_printk(KERN_WARNING, "amd64", "Warning: " fmt, ##arg)
  27. #define amd64_err(fmt, arg...) \
  28. edac_printk(KERN_ERR, "amd64", "Error: " fmt, ##arg)
  29. #define amd64_mc_warn(mci, fmt, arg...) \
  30. edac_mc_chipset_printk(mci, KERN_WARNING, "amd64", fmt, ##arg)
  31. #define amd64_mc_err(mci, fmt, arg...) \
  32. edac_mc_chipset_printk(mci, KERN_ERR, "amd64", fmt, ##arg)
  33. /*
  34. * Throughout the comments in this code, the following terms are used:
  35. *
  36. * SysAddr, DramAddr, and InputAddr
  37. *
  38. * These terms come directly from the amd64 documentation
  39. * (AMD publication #26094). They are defined as follows:
  40. *
  41. * SysAddr:
  42. * This is a physical address generated by a CPU core or a device
  43. * doing DMA. If generated by a CPU core, a SysAddr is the result of
  44. * a virtual to physical address translation by the CPU core's address
  45. * translation mechanism (MMU).
  46. *
  47. * DramAddr:
  48. * A DramAddr is derived from a SysAddr by subtracting an offset that
  49. * depends on which node the SysAddr maps to and whether the SysAddr
  50. * is within a range affected by memory hoisting. The DRAM Base
  51. * (section 3.4.4.1) and DRAM Limit (section 3.4.4.2) registers
  52. * determine which node a SysAddr maps to.
  53. *
  54. * If the DRAM Hole Address Register (DHAR) is enabled and the SysAddr
  55. * is within the range of addresses specified by this register, then
  56. * a value x from the DHAR is subtracted from the SysAddr to produce a
  57. * DramAddr. Here, x represents the base address for the node that
  58. * the SysAddr maps to plus an offset due to memory hoisting. See
  59. * section 3.4.8 and the comments in amd64_get_dram_hole_info() and
  60. * sys_addr_to_dram_addr() below for more information.
  61. *
  62. * If the SysAddr is not affected by the DHAR then a value y is
  63. * subtracted from the SysAddr to produce a DramAddr. Here, y is the
  64. * base address for the node that the SysAddr maps to. See section
  65. * 3.4.4 and the comments in sys_addr_to_dram_addr() below for more
  66. * information.
  67. *
  68. * InputAddr:
  69. * A DramAddr is translated to an InputAddr before being passed to the
  70. * memory controller for the node that the DramAddr is associated
  71. * with. The memory controller then maps the InputAddr to a csrow.
  72. * If node interleaving is not in use, then the InputAddr has the same
  73. * value as the DramAddr. Otherwise, the InputAddr is produced by
  74. * discarding the bits used for node interleaving from the DramAddr.
  75. * See section 3.4.4 for more information.
  76. *
  77. * The memory controller for a given node uses its DRAM CS Base and
  78. * DRAM CS Mask registers to map an InputAddr to a csrow. See
  79. * sections 3.5.4 and 3.5.5 for more information.
  80. */
  81. #define EDAC_MOD_STR "amd64_edac"
  82. /* Extended Model from CPUID, for CPU Revision numbers */
  83. #define K8_REV_D 1
  84. #define K8_REV_E 2
  85. #define K8_REV_F 4
  86. /* Hardware limit on ChipSelect rows per MC and processors per system */
  87. #define NUM_CHIPSELECTS 8
  88. #define DRAM_RANGES 8
  89. #define NUM_CONTROLLERS 12
  90. #define ON true
  91. #define OFF false
  92. /*
  93. * PCI-defined configuration space registers
  94. */
  95. #define PCI_DEVICE_ID_AMD_15H_NB_F1 0x1601
  96. #define PCI_DEVICE_ID_AMD_15H_NB_F2 0x1602
  97. #define PCI_DEVICE_ID_AMD_15H_M30H_NB_F1 0x141b
  98. #define PCI_DEVICE_ID_AMD_15H_M30H_NB_F2 0x141c
  99. #define PCI_DEVICE_ID_AMD_15H_M60H_NB_F1 0x1571
  100. #define PCI_DEVICE_ID_AMD_15H_M60H_NB_F2 0x1572
  101. #define PCI_DEVICE_ID_AMD_16H_NB_F1 0x1531
  102. #define PCI_DEVICE_ID_AMD_16H_NB_F2 0x1532
  103. #define PCI_DEVICE_ID_AMD_16H_M30H_NB_F1 0x1581
  104. #define PCI_DEVICE_ID_AMD_16H_M30H_NB_F2 0x1582
  105. /*
  106. * Function 1 - Address Map
  107. */
  108. #define DRAM_BASE_LO 0x40
  109. #define DRAM_LIMIT_LO 0x44
  110. /*
  111. * F15 M30h D18F1x2[1C:00]
  112. */
  113. #define DRAM_CONT_BASE 0x200
  114. #define DRAM_CONT_LIMIT 0x204
  115. /*
  116. * F15 M30h D18F1x2[4C:40]
  117. */
  118. #define DRAM_CONT_HIGH_OFF 0x240
  119. #define dram_rw(pvt, i) ((u8)(pvt->ranges[i].base.lo & 0x3))
  120. #define dram_intlv_sel(pvt, i) ((u8)((pvt->ranges[i].lim.lo >> 8) & 0x7))
  121. #define dram_dst_node(pvt, i) ((u8)(pvt->ranges[i].lim.lo & 0x7))
  122. #define DHAR 0xf0
  123. #define dhar_mem_hoist_valid(pvt) ((pvt)->dhar & BIT(1))
  124. #define dhar_base(pvt) ((pvt)->dhar & 0xff000000)
  125. #define k8_dhar_offset(pvt) (((pvt)->dhar & 0x0000ff00) << 16)
  126. /* NOTE: Extra mask bit vs K8 */
  127. #define f10_dhar_offset(pvt) (((pvt)->dhar & 0x0000ff80) << 16)
  128. #define DCT_CFG_SEL 0x10C
  129. #define DRAM_LOCAL_NODE_BASE 0x120
  130. #define DRAM_LOCAL_NODE_LIM 0x124
  131. #define DRAM_BASE_HI 0x140
  132. #define DRAM_LIMIT_HI 0x144
  133. /*
  134. * Function 2 - DRAM controller
  135. */
  136. #define DCSB0 0x40
  137. #define DCSB1 0x140
  138. #define DCSB_CS_ENABLE BIT(0)
  139. #define DCSM0 0x60
  140. #define DCSM1 0x160
  141. #define csrow_enabled(i, dct, pvt) ((pvt)->csels[(dct)].csbases[(i)] & DCSB_CS_ENABLE)
  142. #define csrow_sec_enabled(i, dct, pvt) ((pvt)->csels[(dct)].csbases_sec[(i)] & DCSB_CS_ENABLE)
  143. #define DRAM_CONTROL 0x78
  144. #define DBAM0 0x80
  145. #define DBAM1 0x180
  146. /* Extract the DIMM 'type' on the i'th DIMM from the DBAM reg value passed */
  147. #define DBAM_DIMM(i, reg) ((((reg) >> (4*(i)))) & 0xF)
  148. #define DBAM_MAX_VALUE 11
  149. #define DCLR0 0x90
  150. #define DCLR1 0x190
  151. #define REVE_WIDTH_128 BIT(16)
  152. #define WIDTH_128 BIT(11)
  153. #define DCHR0 0x94
  154. #define DCHR1 0x194
  155. #define DDR3_MODE BIT(8)
  156. #define DCT_SEL_LO 0x110
  157. #define dct_high_range_enabled(pvt) ((pvt)->dct_sel_lo & BIT(0))
  158. #define dct_interleave_enabled(pvt) ((pvt)->dct_sel_lo & BIT(2))
  159. #define dct_ganging_enabled(pvt) ((boot_cpu_data.x86 == 0x10) && ((pvt)->dct_sel_lo & BIT(4)))
  160. #define dct_data_intlv_enabled(pvt) ((pvt)->dct_sel_lo & BIT(5))
  161. #define dct_memory_cleared(pvt) ((pvt)->dct_sel_lo & BIT(10))
  162. #define SWAP_INTLV_REG 0x10c
  163. #define DCT_SEL_HI 0x114
  164. #define F15H_M60H_SCRCTRL 0x1C8
  165. /*
  166. * Function 3 - Misc Control
  167. */
  168. #define NBCTL 0x40
  169. #define NBCFG 0x44
  170. #define NBCFG_CHIPKILL BIT(23)
  171. #define NBCFG_ECC_ENABLE BIT(22)
  172. /* F3x48: NBSL */
  173. #define F10_NBSL_EXT_ERR_ECC 0x8
  174. #define NBSL_PP_OBS 0x2
  175. #define SCRCTRL 0x58
  176. #define F10_ONLINE_SPARE 0xB0
  177. #define online_spare_swap_done(pvt, c) (((pvt)->online_spare >> (1 + 2 * (c))) & 0x1)
  178. #define online_spare_bad_dramcs(pvt, c) (((pvt)->online_spare >> (4 + 4 * (c))) & 0x7)
  179. #define F10_NB_ARRAY_ADDR 0xB8
  180. #define F10_NB_ARRAY_DRAM BIT(31)
  181. /* Bits [2:1] are used to select 16-byte section within a 64-byte cacheline */
  182. #define SET_NB_ARRAY_ADDR(section) (((section) & 0x3) << 1)
  183. #define F10_NB_ARRAY_DATA 0xBC
  184. #define F10_NB_ARR_ECC_WR_REQ BIT(17)
  185. #define SET_NB_DRAM_INJECTION_WRITE(inj) \
  186. (BIT(((inj.word) & 0xF) + 20) | \
  187. F10_NB_ARR_ECC_WR_REQ | inj.bit_map)
  188. #define SET_NB_DRAM_INJECTION_READ(inj) \
  189. (BIT(((inj.word) & 0xF) + 20) | \
  190. BIT(16) | inj.bit_map)
  191. #define NBCAP 0xE8
  192. #define NBCAP_CHIPKILL BIT(4)
  193. #define NBCAP_SECDED BIT(3)
  194. #define NBCAP_DCT_DUAL BIT(0)
  195. #define EXT_NB_MCA_CFG 0x180
  196. /* MSRs */
  197. #define MSR_MCGCTL_NBE BIT(4)
  198. /* F17h */
  199. /* F0: */
  200. #define DF_DHAR 0x104
  201. /* UMC CH register offsets */
  202. #define UMCCH_BASE_ADDR 0x0
  203. #define UMCCH_BASE_ADDR_SEC 0x10
  204. #define UMCCH_ADDR_MASK 0x20
  205. #define UMCCH_ADDR_MASK_SEC 0x28
  206. #define UMCCH_ADDR_MASK_SEC_DDR5 0x30
  207. #define UMCCH_DIMM_CFG 0x80
  208. #define UMCCH_DIMM_CFG_DDR5 0x90
  209. #define UMCCH_UMC_CFG 0x100
  210. #define UMCCH_SDP_CTRL 0x104
  211. #define UMCCH_ECC_CTRL 0x14C
  212. #define UMCCH_UMC_CAP_HI 0xDF4
  213. /* UMC CH bitfields */
  214. #define UMC_ECC_CHIPKILL_CAP BIT(31)
  215. #define UMC_ECC_ENABLED BIT(30)
  216. #define UMC_SDP_INIT BIT(31)
  217. /* Error injection control structure */
  218. struct error_injection {
  219. u32 section;
  220. u32 word;
  221. u32 bit_map;
  222. };
  223. /* low and high part of PCI config space regs */
  224. struct reg_pair {
  225. u32 lo, hi;
  226. };
  227. /*
  228. * See F1x[1, 0][7C:40] DRAM Base/Limit Registers
  229. */
  230. struct dram_range {
  231. struct reg_pair base;
  232. struct reg_pair lim;
  233. };
  234. /* A DCT chip selects collection */
  235. struct chip_select {
  236. u32 csbases[NUM_CHIPSELECTS];
  237. u32 csbases_sec[NUM_CHIPSELECTS];
  238. u8 b_cnt;
  239. u32 csmasks[NUM_CHIPSELECTS];
  240. u32 csmasks_sec[NUM_CHIPSELECTS];
  241. u8 m_cnt;
  242. };
  243. struct amd64_umc {
  244. u32 dimm_cfg; /* DIMM Configuration reg */
  245. u32 umc_cfg; /* Configuration reg */
  246. u32 sdp_ctrl; /* SDP Control reg */
  247. u32 ecc_ctrl; /* DRAM ECC Control reg */
  248. u32 umc_cap_hi; /* Capabilities High reg */
  249. /* cache the dram_type */
  250. enum mem_type dram_type;
  251. };
  252. struct amd64_family_flags {
  253. /*
  254. * Indicates that the system supports the new register offsets, etc.
  255. * first introduced with Family 19h Model 10h.
  256. */
  257. __u64 zn_regs_v2 : 1,
  258. __reserved : 63;
  259. };
  260. struct amd64_pvt {
  261. struct low_ops *ops;
  262. /* pci_device handles which we utilize */
  263. struct pci_dev *F1, *F2, *F3;
  264. u16 mc_node_id; /* MC index of this MC node */
  265. u8 fam; /* CPU family */
  266. u8 model; /* ... model */
  267. u8 stepping; /* ... stepping */
  268. int ext_model; /* extended model value of this node */
  269. /* Raw registers */
  270. u32 dclr0; /* DRAM Configuration Low DCT0 reg */
  271. u32 dclr1; /* DRAM Configuration Low DCT1 reg */
  272. u32 dchr0; /* DRAM Configuration High DCT0 reg */
  273. u32 dchr1; /* DRAM Configuration High DCT1 reg */
  274. u32 nbcap; /* North Bridge Capabilities */
  275. u32 nbcfg; /* F10 North Bridge Configuration */
  276. u32 dhar; /* DRAM Hoist reg */
  277. u32 dbam0; /* DRAM Base Address Mapping reg for DCT0 */
  278. u32 dbam1; /* DRAM Base Address Mapping reg for DCT1 */
  279. /* one for each DCT/UMC */
  280. struct chip_select csels[NUM_CONTROLLERS];
  281. /* DRAM base and limit pairs F1x[78,70,68,60,58,50,48,40] */
  282. struct dram_range ranges[DRAM_RANGES];
  283. u64 top_mem; /* top of memory below 4GB */
  284. u64 top_mem2; /* top of memory above 4GB */
  285. u32 dct_sel_lo; /* DRAM Controller Select Low */
  286. u32 dct_sel_hi; /* DRAM Controller Select High */
  287. u32 online_spare; /* On-Line spare Reg */
  288. u32 gpu_umc_base; /* Base address used for channel selection on GPUs */
  289. /* x4, x8, or x16 syndromes in use */
  290. u8 ecc_sym_sz;
  291. const char *ctl_name;
  292. u16 f1_id, f2_id;
  293. /* Maximum number of memory controllers per die/node. */
  294. u8 max_mcs;
  295. struct amd64_family_flags flags;
  296. /* place to store error injection parameters prior to issue */
  297. struct error_injection injection;
  298. /*
  299. * cache the dram_type
  300. *
  301. * NOTE: Don't use this for Family 17h and later.
  302. * Use dram_type in struct amd64_umc instead.
  303. */
  304. enum mem_type dram_type;
  305. struct amd64_umc *umc; /* UMC registers */
  306. };
  307. enum err_codes {
  308. DECODE_OK = 0,
  309. ERR_NODE = -1,
  310. ERR_CSROW = -2,
  311. ERR_CHANNEL = -3,
  312. ERR_SYND = -4,
  313. ERR_NORM_ADDR = -5,
  314. };
  315. struct err_info {
  316. int err_code;
  317. struct mem_ctl_info *src_mci;
  318. int csrow;
  319. int channel;
  320. u16 syndrome;
  321. u32 page;
  322. u32 offset;
  323. };
  324. static inline u32 get_umc_base(u8 channel)
  325. {
  326. /* chY: 0xY50000 */
  327. return 0x50000 + (channel << 20);
  328. }
  329. static inline u64 get_dram_base(struct amd64_pvt *pvt, u8 i)
  330. {
  331. u64 addr = ((u64)pvt->ranges[i].base.lo & 0xffff0000) << 8;
  332. if (boot_cpu_data.x86 == 0xf)
  333. return addr;
  334. return (((u64)pvt->ranges[i].base.hi & 0x000000ff) << 40) | addr;
  335. }
  336. static inline u64 get_dram_limit(struct amd64_pvt *pvt, u8 i)
  337. {
  338. u64 lim = (((u64)pvt->ranges[i].lim.lo & 0xffff0000) << 8) | 0x00ffffff;
  339. if (boot_cpu_data.x86 == 0xf)
  340. return lim;
  341. return (((u64)pvt->ranges[i].lim.hi & 0x000000ff) << 40) | lim;
  342. }
  343. static inline u16 extract_syndrome(u64 status)
  344. {
  345. return ((status >> 47) & 0xff) | ((status >> 16) & 0xff00);
  346. }
  347. static inline u8 dct_sel_interleave_addr(struct amd64_pvt *pvt)
  348. {
  349. if (pvt->fam == 0x15 && pvt->model >= 0x30)
  350. return (((pvt->dct_sel_hi >> 9) & 0x1) << 2) |
  351. ((pvt->dct_sel_lo >> 6) & 0x3);
  352. return ((pvt)->dct_sel_lo >> 6) & 0x3;
  353. }
  354. /*
  355. * per-node ECC settings descriptor
  356. */
  357. struct ecc_settings {
  358. u32 old_nbctl;
  359. bool nbctl_valid;
  360. struct flags {
  361. unsigned long nb_mce_enable:1;
  362. unsigned long nb_ecc_prev:1;
  363. } flags;
  364. };
  365. /*
  366. * Each of the PCI Device IDs types have their own set of hardware accessor
  367. * functions and per device encoding/decoding logic.
  368. */
  369. struct low_ops {
  370. void (*map_sysaddr_to_csrow)(struct mem_ctl_info *mci, u64 sys_addr,
  371. struct err_info *err);
  372. int (*dbam_to_cs)(struct amd64_pvt *pvt, u8 dct,
  373. unsigned int cs_mode, int cs_mask_nr);
  374. int (*hw_info_get)(struct amd64_pvt *pvt);
  375. bool (*ecc_enabled)(struct amd64_pvt *pvt);
  376. void (*setup_mci_misc_attrs)(struct mem_ctl_info *mci);
  377. void (*dump_misc_regs)(struct amd64_pvt *pvt);
  378. void (*get_err_info)(struct mce *m, struct err_info *err);
  379. };
  380. int __amd64_read_pci_cfg_dword(struct pci_dev *pdev, int offset,
  381. u32 *val, const char *func);
  382. int __amd64_write_pci_cfg_dword(struct pci_dev *pdev, int offset,
  383. u32 val, const char *func);
  384. #define amd64_read_pci_cfg(pdev, offset, val) \
  385. __amd64_read_pci_cfg_dword(pdev, offset, val, __func__)
  386. #define amd64_write_pci_cfg(pdev, offset, val) \
  387. __amd64_write_pci_cfg_dword(pdev, offset, val, __func__)
  388. #define to_mci(k) container_of(k, struct mem_ctl_info, dev)
  389. /* Injection helpers */
  390. static inline void disable_caches(void *dummy)
  391. {
  392. write_cr0(read_cr0() | X86_CR0_CD);
  393. wbinvd();
  394. }
  395. static inline void enable_caches(void *dummy)
  396. {
  397. write_cr0(read_cr0() & ~X86_CR0_CD);
  398. }
  399. static inline u8 dram_intlv_en(struct amd64_pvt *pvt, unsigned int i)
  400. {
  401. if (pvt->fam == 0x15 && pvt->model >= 0x30) {
  402. u32 tmp;
  403. amd64_read_pci_cfg(pvt->F1, DRAM_CONT_LIMIT, &tmp);
  404. return (u8) tmp & 0xF;
  405. }
  406. return (u8) (pvt->ranges[i].base.lo >> 8) & 0x7;
  407. }
  408. static inline u8 dhar_valid(struct amd64_pvt *pvt)
  409. {
  410. if (pvt->fam == 0x15 && pvt->model >= 0x30) {
  411. u32 tmp;
  412. amd64_read_pci_cfg(pvt->F1, DRAM_CONT_BASE, &tmp);
  413. return (tmp >> 1) & BIT(0);
  414. }
  415. return (pvt)->dhar & BIT(0);
  416. }
  417. static inline u32 dct_sel_baseaddr(struct amd64_pvt *pvt)
  418. {
  419. if (pvt->fam == 0x15 && pvt->model >= 0x30) {
  420. u32 tmp;
  421. amd64_read_pci_cfg(pvt->F1, DRAM_CONT_BASE, &tmp);
  422. return (tmp >> 11) & 0x1FFF;
  423. }
  424. return (pvt)->dct_sel_lo & 0xFFFFF800;
  425. }