amd64_edac.h 15 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549
  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 <asm/cpu_device_id.h>
  19. #include <asm/msr.h>
  20. #include "edac_module.h"
  21. #include "mce_amd.h"
  22. #define amd64_info(fmt, arg...) \
  23. edac_printk(KERN_INFO, "amd64", fmt, ##arg)
  24. #define amd64_warn(fmt, arg...) \
  25. edac_printk(KERN_WARNING, "amd64", "Warning: " fmt, ##arg)
  26. #define amd64_err(fmt, arg...) \
  27. edac_printk(KERN_ERR, "amd64", "Error: " fmt, ##arg)
  28. #define amd64_mc_warn(mci, fmt, arg...) \
  29. edac_mc_chipset_printk(mci, KERN_WARNING, "amd64", fmt, ##arg)
  30. #define amd64_mc_err(mci, fmt, arg...) \
  31. edac_mc_chipset_printk(mci, KERN_ERR, "amd64", fmt, ##arg)
  32. /*
  33. * Throughout the comments in this code, the following terms are used:
  34. *
  35. * SysAddr, DramAddr, and InputAddr
  36. *
  37. * These terms come directly from the amd64 documentation
  38. * (AMD publication #26094). They are defined as follows:
  39. *
  40. * SysAddr:
  41. * This is a physical address generated by a CPU core or a device
  42. * doing DMA. If generated by a CPU core, a SysAddr is the result of
  43. * a virtual to physical address translation by the CPU core's address
  44. * translation mechanism (MMU).
  45. *
  46. * DramAddr:
  47. * A DramAddr is derived from a SysAddr by subtracting an offset that
  48. * depends on which node the SysAddr maps to and whether the SysAddr
  49. * is within a range affected by memory hoisting. The DRAM Base
  50. * (section 3.4.4.1) and DRAM Limit (section 3.4.4.2) registers
  51. * determine which node a SysAddr maps to.
  52. *
  53. * If the DRAM Hole Address Register (DHAR) is enabled and the SysAddr
  54. * is within the range of addresses specified by this register, then
  55. * a value x from the DHAR is subtracted from the SysAddr to produce a
  56. * DramAddr. Here, x represents the base address for the node that
  57. * the SysAddr maps to plus an offset due to memory hoisting. See
  58. * section 3.4.8 and the comments in amd64_get_dram_hole_info() and
  59. * sys_addr_to_dram_addr() below for more information.
  60. *
  61. * If the SysAddr is not affected by the DHAR then a value y is
  62. * subtracted from the SysAddr to produce a DramAddr. Here, y is the
  63. * base address for the node that the SysAddr maps to. See section
  64. * 3.4.4 and the comments in sys_addr_to_dram_addr() below for more
  65. * information.
  66. *
  67. * InputAddr:
  68. * A DramAddr is translated to an InputAddr before being passed to the
  69. * memory controller for the node that the DramAddr is associated
  70. * with. The memory controller then maps the InputAddr to a csrow.
  71. * If node interleaving is not in use, then the InputAddr has the same
  72. * value as the DramAddr. Otherwise, the InputAddr is produced by
  73. * discarding the bits used for node interleaving from the DramAddr.
  74. * See section 3.4.4 for more information.
  75. *
  76. * The memory controller for a given node uses its DRAM CS Base and
  77. * DRAM CS Mask registers to map an InputAddr to a csrow. See
  78. * sections 3.5.4 and 3.5.5 for more information.
  79. */
  80. #define EDAC_AMD64_VERSION "3.5.0"
  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 8
  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. #define PCI_DEVICE_ID_AMD_17H_DF_F0 0x1460
  106. #define PCI_DEVICE_ID_AMD_17H_DF_F6 0x1466
  107. #define PCI_DEVICE_ID_AMD_17H_M10H_DF_F0 0x15e8
  108. #define PCI_DEVICE_ID_AMD_17H_M10H_DF_F6 0x15ee
  109. #define PCI_DEVICE_ID_AMD_17H_M30H_DF_F0 0x1490
  110. #define PCI_DEVICE_ID_AMD_17H_M30H_DF_F6 0x1496
  111. #define PCI_DEVICE_ID_AMD_17H_M60H_DF_F0 0x1448
  112. #define PCI_DEVICE_ID_AMD_17H_M60H_DF_F6 0x144e
  113. #define PCI_DEVICE_ID_AMD_17H_M70H_DF_F0 0x1440
  114. #define PCI_DEVICE_ID_AMD_17H_M70H_DF_F6 0x1446
  115. #define PCI_DEVICE_ID_AMD_19H_DF_F0 0x1650
  116. #define PCI_DEVICE_ID_AMD_19H_DF_F6 0x1656
  117. /*
  118. * Function 1 - Address Map
  119. */
  120. #define DRAM_BASE_LO 0x40
  121. #define DRAM_LIMIT_LO 0x44
  122. /*
  123. * F15 M30h D18F1x2[1C:00]
  124. */
  125. #define DRAM_CONT_BASE 0x200
  126. #define DRAM_CONT_LIMIT 0x204
  127. /*
  128. * F15 M30h D18F1x2[4C:40]
  129. */
  130. #define DRAM_CONT_HIGH_OFF 0x240
  131. #define dram_rw(pvt, i) ((u8)(pvt->ranges[i].base.lo & 0x3))
  132. #define dram_intlv_sel(pvt, i) ((u8)((pvt->ranges[i].lim.lo >> 8) & 0x7))
  133. #define dram_dst_node(pvt, i) ((u8)(pvt->ranges[i].lim.lo & 0x7))
  134. #define DHAR 0xf0
  135. #define dhar_mem_hoist_valid(pvt) ((pvt)->dhar & BIT(1))
  136. #define dhar_base(pvt) ((pvt)->dhar & 0xff000000)
  137. #define k8_dhar_offset(pvt) (((pvt)->dhar & 0x0000ff00) << 16)
  138. /* NOTE: Extra mask bit vs K8 */
  139. #define f10_dhar_offset(pvt) (((pvt)->dhar & 0x0000ff80) << 16)
  140. #define DCT_CFG_SEL 0x10C
  141. #define DRAM_LOCAL_NODE_BASE 0x120
  142. #define DRAM_LOCAL_NODE_LIM 0x124
  143. #define DRAM_BASE_HI 0x140
  144. #define DRAM_LIMIT_HI 0x144
  145. /*
  146. * Function 2 - DRAM controller
  147. */
  148. #define DCSB0 0x40
  149. #define DCSB1 0x140
  150. #define DCSB_CS_ENABLE BIT(0)
  151. #define DCSM0 0x60
  152. #define DCSM1 0x160
  153. #define csrow_enabled(i, dct, pvt) ((pvt)->csels[(dct)].csbases[(i)] & DCSB_CS_ENABLE)
  154. #define csrow_sec_enabled(i, dct, pvt) ((pvt)->csels[(dct)].csbases_sec[(i)] & DCSB_CS_ENABLE)
  155. #define DRAM_CONTROL 0x78
  156. #define DBAM0 0x80
  157. #define DBAM1 0x180
  158. /* Extract the DIMM 'type' on the i'th DIMM from the DBAM reg value passed */
  159. #define DBAM_DIMM(i, reg) ((((reg) >> (4*(i)))) & 0xF)
  160. #define DBAM_MAX_VALUE 11
  161. #define DCLR0 0x90
  162. #define DCLR1 0x190
  163. #define REVE_WIDTH_128 BIT(16)
  164. #define WIDTH_128 BIT(11)
  165. #define DCHR0 0x94
  166. #define DCHR1 0x194
  167. #define DDR3_MODE BIT(8)
  168. #define DCT_SEL_LO 0x110
  169. #define dct_high_range_enabled(pvt) ((pvt)->dct_sel_lo & BIT(0))
  170. #define dct_interleave_enabled(pvt) ((pvt)->dct_sel_lo & BIT(2))
  171. #define dct_ganging_enabled(pvt) ((boot_cpu_data.x86 == 0x10) && ((pvt)->dct_sel_lo & BIT(4)))
  172. #define dct_data_intlv_enabled(pvt) ((pvt)->dct_sel_lo & BIT(5))
  173. #define dct_memory_cleared(pvt) ((pvt)->dct_sel_lo & BIT(10))
  174. #define SWAP_INTLV_REG 0x10c
  175. #define DCT_SEL_HI 0x114
  176. #define F15H_M60H_SCRCTRL 0x1C8
  177. #define F17H_SCR_BASE_ADDR 0x48
  178. #define F17H_SCR_LIMIT_ADDR 0x4C
  179. /*
  180. * Function 3 - Misc Control
  181. */
  182. #define NBCTL 0x40
  183. #define NBCFG 0x44
  184. #define NBCFG_CHIPKILL BIT(23)
  185. #define NBCFG_ECC_ENABLE BIT(22)
  186. /* F3x48: NBSL */
  187. #define F10_NBSL_EXT_ERR_ECC 0x8
  188. #define NBSL_PP_OBS 0x2
  189. #define SCRCTRL 0x58
  190. #define F10_ONLINE_SPARE 0xB0
  191. #define online_spare_swap_done(pvt, c) (((pvt)->online_spare >> (1 + 2 * (c))) & 0x1)
  192. #define online_spare_bad_dramcs(pvt, c) (((pvt)->online_spare >> (4 + 4 * (c))) & 0x7)
  193. #define F10_NB_ARRAY_ADDR 0xB8
  194. #define F10_NB_ARRAY_DRAM BIT(31)
  195. /* Bits [2:1] are used to select 16-byte section within a 64-byte cacheline */
  196. #define SET_NB_ARRAY_ADDR(section) (((section) & 0x3) << 1)
  197. #define F10_NB_ARRAY_DATA 0xBC
  198. #define F10_NB_ARR_ECC_WR_REQ BIT(17)
  199. #define SET_NB_DRAM_INJECTION_WRITE(inj) \
  200. (BIT(((inj.word) & 0xF) + 20) | \
  201. F10_NB_ARR_ECC_WR_REQ | inj.bit_map)
  202. #define SET_NB_DRAM_INJECTION_READ(inj) \
  203. (BIT(((inj.word) & 0xF) + 20) | \
  204. BIT(16) | inj.bit_map)
  205. #define NBCAP 0xE8
  206. #define NBCAP_CHIPKILL BIT(4)
  207. #define NBCAP_SECDED BIT(3)
  208. #define NBCAP_DCT_DUAL BIT(0)
  209. #define EXT_NB_MCA_CFG 0x180
  210. /* MSRs */
  211. #define MSR_MCGCTL_NBE BIT(4)
  212. /* F17h */
  213. /* F0: */
  214. #define DF_DHAR 0x104
  215. /* UMC CH register offsets */
  216. #define UMCCH_BASE_ADDR 0x0
  217. #define UMCCH_BASE_ADDR_SEC 0x10
  218. #define UMCCH_ADDR_MASK 0x20
  219. #define UMCCH_ADDR_MASK_SEC 0x28
  220. #define UMCCH_ADDR_CFG 0x30
  221. #define UMCCH_DIMM_CFG 0x80
  222. #define UMCCH_UMC_CFG 0x100
  223. #define UMCCH_SDP_CTRL 0x104
  224. #define UMCCH_ECC_CTRL 0x14C
  225. #define UMCCH_ECC_BAD_SYMBOL 0xD90
  226. #define UMCCH_UMC_CAP 0xDF0
  227. #define UMCCH_UMC_CAP_HI 0xDF4
  228. /* UMC CH bitfields */
  229. #define UMC_ECC_CHIPKILL_CAP BIT(31)
  230. #define UMC_ECC_ENABLED BIT(30)
  231. #define UMC_SDP_INIT BIT(31)
  232. enum amd_families {
  233. K8_CPUS = 0,
  234. F10_CPUS,
  235. F15_CPUS,
  236. F15_M30H_CPUS,
  237. F15_M60H_CPUS,
  238. F16_CPUS,
  239. F16_M30H_CPUS,
  240. F17_CPUS,
  241. F17_M10H_CPUS,
  242. F17_M30H_CPUS,
  243. F17_M60H_CPUS,
  244. F17_M70H_CPUS,
  245. F19_CPUS,
  246. NUM_FAMILIES,
  247. };
  248. /* Error injection control structure */
  249. struct error_injection {
  250. u32 section;
  251. u32 word;
  252. u32 bit_map;
  253. };
  254. /* low and high part of PCI config space regs */
  255. struct reg_pair {
  256. u32 lo, hi;
  257. };
  258. /*
  259. * See F1x[1, 0][7C:40] DRAM Base/Limit Registers
  260. */
  261. struct dram_range {
  262. struct reg_pair base;
  263. struct reg_pair lim;
  264. };
  265. /* A DCT chip selects collection */
  266. struct chip_select {
  267. u32 csbases[NUM_CHIPSELECTS];
  268. u32 csbases_sec[NUM_CHIPSELECTS];
  269. u8 b_cnt;
  270. u32 csmasks[NUM_CHIPSELECTS];
  271. u32 csmasks_sec[NUM_CHIPSELECTS];
  272. u8 m_cnt;
  273. };
  274. struct amd64_umc {
  275. u32 dimm_cfg; /* DIMM Configuration reg */
  276. u32 umc_cfg; /* Configuration reg */
  277. u32 sdp_ctrl; /* SDP Control reg */
  278. u32 ecc_ctrl; /* DRAM ECC Control reg */
  279. u32 umc_cap_hi; /* Capabilities High reg */
  280. };
  281. struct amd64_pvt {
  282. struct low_ops *ops;
  283. /* pci_device handles which we utilize */
  284. struct pci_dev *F0, *F1, *F2, *F3, *F6;
  285. u16 mc_node_id; /* MC index of this MC node */
  286. u8 fam; /* CPU family */
  287. u8 model; /* ... model */
  288. u8 stepping; /* ... stepping */
  289. int ext_model; /* extended model value of this node */
  290. int channel_count;
  291. /* Raw registers */
  292. u32 dclr0; /* DRAM Configuration Low DCT0 reg */
  293. u32 dclr1; /* DRAM Configuration Low DCT1 reg */
  294. u32 dchr0; /* DRAM Configuration High DCT0 reg */
  295. u32 dchr1; /* DRAM Configuration High DCT1 reg */
  296. u32 nbcap; /* North Bridge Capabilities */
  297. u32 nbcfg; /* F10 North Bridge Configuration */
  298. u32 ext_nbcfg; /* Extended F10 North Bridge Configuration */
  299. u32 dhar; /* DRAM Hoist reg */
  300. u32 dbam0; /* DRAM Base Address Mapping reg for DCT0 */
  301. u32 dbam1; /* DRAM Base Address Mapping reg for DCT1 */
  302. /* one for each DCT/UMC */
  303. struct chip_select csels[NUM_CONTROLLERS];
  304. /* DRAM base and limit pairs F1x[78,70,68,60,58,50,48,40] */
  305. struct dram_range ranges[DRAM_RANGES];
  306. u64 top_mem; /* top of memory below 4GB */
  307. u64 top_mem2; /* top of memory above 4GB */
  308. u32 dct_sel_lo; /* DRAM Controller Select Low */
  309. u32 dct_sel_hi; /* DRAM Controller Select High */
  310. u32 online_spare; /* On-Line spare Reg */
  311. /* x4, x8, or x16 syndromes in use */
  312. u8 ecc_sym_sz;
  313. /* place to store error injection parameters prior to issue */
  314. struct error_injection injection;
  315. /* cache the dram_type */
  316. enum mem_type dram_type;
  317. struct amd64_umc *umc; /* UMC registers */
  318. };
  319. enum err_codes {
  320. DECODE_OK = 0,
  321. ERR_NODE = -1,
  322. ERR_CSROW = -2,
  323. ERR_CHANNEL = -3,
  324. ERR_SYND = -4,
  325. ERR_NORM_ADDR = -5,
  326. };
  327. struct err_info {
  328. int err_code;
  329. struct mem_ctl_info *src_mci;
  330. int csrow;
  331. int channel;
  332. u16 syndrome;
  333. u32 page;
  334. u32 offset;
  335. };
  336. static inline u32 get_umc_base(u8 channel)
  337. {
  338. /* chY: 0xY50000 */
  339. return 0x50000 + (channel << 20);
  340. }
  341. static inline u64 get_dram_base(struct amd64_pvt *pvt, u8 i)
  342. {
  343. u64 addr = ((u64)pvt->ranges[i].base.lo & 0xffff0000) << 8;
  344. if (boot_cpu_data.x86 == 0xf)
  345. return addr;
  346. return (((u64)pvt->ranges[i].base.hi & 0x000000ff) << 40) | addr;
  347. }
  348. static inline u64 get_dram_limit(struct amd64_pvt *pvt, u8 i)
  349. {
  350. u64 lim = (((u64)pvt->ranges[i].lim.lo & 0xffff0000) << 8) | 0x00ffffff;
  351. if (boot_cpu_data.x86 == 0xf)
  352. return lim;
  353. return (((u64)pvt->ranges[i].lim.hi & 0x000000ff) << 40) | lim;
  354. }
  355. static inline u16 extract_syndrome(u64 status)
  356. {
  357. return ((status >> 47) & 0xff) | ((status >> 16) & 0xff00);
  358. }
  359. static inline u8 dct_sel_interleave_addr(struct amd64_pvt *pvt)
  360. {
  361. if (pvt->fam == 0x15 && pvt->model >= 0x30)
  362. return (((pvt->dct_sel_hi >> 9) & 0x1) << 2) |
  363. ((pvt->dct_sel_lo >> 6) & 0x3);
  364. return ((pvt)->dct_sel_lo >> 6) & 0x3;
  365. }
  366. /*
  367. * per-node ECC settings descriptor
  368. */
  369. struct ecc_settings {
  370. u32 old_nbctl;
  371. bool nbctl_valid;
  372. struct flags {
  373. unsigned long nb_mce_enable:1;
  374. unsigned long nb_ecc_prev:1;
  375. } flags;
  376. };
  377. #ifdef CONFIG_EDAC_DEBUG
  378. extern const struct attribute_group amd64_edac_dbg_group;
  379. #endif
  380. #ifdef CONFIG_EDAC_AMD64_ERROR_INJECTION
  381. extern const struct attribute_group amd64_edac_inj_group;
  382. #endif
  383. /*
  384. * Each of the PCI Device IDs types have their own set of hardware accessor
  385. * functions and per device encoding/decoding logic.
  386. */
  387. struct low_ops {
  388. int (*early_channel_count) (struct amd64_pvt *pvt);
  389. void (*map_sysaddr_to_csrow) (struct mem_ctl_info *mci, u64 sys_addr,
  390. struct err_info *);
  391. int (*dbam_to_cs) (struct amd64_pvt *pvt, u8 dct,
  392. unsigned cs_mode, int cs_mask_nr);
  393. };
  394. struct amd64_family_type {
  395. const char *ctl_name;
  396. u16 f0_id, f1_id, f2_id, f6_id;
  397. /* Maximum number of memory controllers per die/node. */
  398. u8 max_mcs;
  399. struct low_ops ops;
  400. };
  401. int __amd64_read_pci_cfg_dword(struct pci_dev *pdev, int offset,
  402. u32 *val, const char *func);
  403. int __amd64_write_pci_cfg_dword(struct pci_dev *pdev, int offset,
  404. u32 val, const char *func);
  405. #define amd64_read_pci_cfg(pdev, offset, val) \
  406. __amd64_read_pci_cfg_dword(pdev, offset, val, __func__)
  407. #define amd64_write_pci_cfg(pdev, offset, val) \
  408. __amd64_write_pci_cfg_dword(pdev, offset, val, __func__)
  409. int amd64_get_dram_hole_info(struct mem_ctl_info *mci, u64 *hole_base,
  410. u64 *hole_offset, u64 *hole_size);
  411. #define to_mci(k) container_of(k, struct mem_ctl_info, dev)
  412. /* Injection helpers */
  413. static inline void disable_caches(void *dummy)
  414. {
  415. write_cr0(read_cr0() | X86_CR0_CD);
  416. wbinvd();
  417. }
  418. static inline void enable_caches(void *dummy)
  419. {
  420. write_cr0(read_cr0() & ~X86_CR0_CD);
  421. }
  422. static inline u8 dram_intlv_en(struct amd64_pvt *pvt, unsigned int i)
  423. {
  424. if (pvt->fam == 0x15 && pvt->model >= 0x30) {
  425. u32 tmp;
  426. amd64_read_pci_cfg(pvt->F1, DRAM_CONT_LIMIT, &tmp);
  427. return (u8) tmp & 0xF;
  428. }
  429. return (u8) (pvt->ranges[i].base.lo >> 8) & 0x7;
  430. }
  431. static inline u8 dhar_valid(struct amd64_pvt *pvt)
  432. {
  433. if (pvt->fam == 0x15 && pvt->model >= 0x30) {
  434. u32 tmp;
  435. amd64_read_pci_cfg(pvt->F1, DRAM_CONT_BASE, &tmp);
  436. return (tmp >> 1) & BIT(0);
  437. }
  438. return (pvt)->dhar & BIT(0);
  439. }
  440. static inline u32 dct_sel_baseaddr(struct amd64_pvt *pvt)
  441. {
  442. if (pvt->fam == 0x15 && pvt->model >= 0x30) {
  443. u32 tmp;
  444. amd64_read_pci_cfg(pvt->F1, DRAM_CONT_BASE, &tmp);
  445. return (tmp >> 11) & 0x1FFF;
  446. }
  447. return (pvt)->dct_sel_lo & 0xFFFFF800;
  448. }