amd64_edac.h 13 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485
  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/msr.h>
  19. #include "edac_core.h"
  20. #include "mce_amd.h"
  21. #define amd64_debug(fmt, arg...) \
  22. edac_printk(KERN_DEBUG, "amd64", fmt, ##arg)
  23. #define amd64_info(fmt, arg...) \
  24. edac_printk(KERN_INFO, "amd64", fmt, ##arg)
  25. #define amd64_notice(fmt, arg...) \
  26. edac_printk(KERN_NOTICE, "amd64", fmt, ##arg)
  27. #define amd64_warn(fmt, arg...) \
  28. edac_printk(KERN_WARNING, "amd64", fmt, ##arg)
  29. #define amd64_err(fmt, arg...) \
  30. edac_printk(KERN_ERR, "amd64", fmt, ##arg)
  31. #define amd64_mc_warn(mci, fmt, arg...) \
  32. edac_mc_chipset_printk(mci, KERN_WARNING, "amd64", fmt, ##arg)
  33. #define amd64_mc_err(mci, fmt, arg...) \
  34. edac_mc_chipset_printk(mci, KERN_ERR, "amd64", fmt, ##arg)
  35. /*
  36. * Throughout the comments in this code, the following terms are used:
  37. *
  38. * SysAddr, DramAddr, and InputAddr
  39. *
  40. * These terms come directly from the amd64 documentation
  41. * (AMD publication #26094). They are defined as follows:
  42. *
  43. * SysAddr:
  44. * This is a physical address generated by a CPU core or a device
  45. * doing DMA. If generated by a CPU core, a SysAddr is the result of
  46. * a virtual to physical address translation by the CPU core's address
  47. * translation mechanism (MMU).
  48. *
  49. * DramAddr:
  50. * A DramAddr is derived from a SysAddr by subtracting an offset that
  51. * depends on which node the SysAddr maps to and whether the SysAddr
  52. * is within a range affected by memory hoisting. The DRAM Base
  53. * (section 3.4.4.1) and DRAM Limit (section 3.4.4.2) registers
  54. * determine which node a SysAddr maps to.
  55. *
  56. * If the DRAM Hole Address Register (DHAR) is enabled and the SysAddr
  57. * is within the range of addresses specified by this register, then
  58. * a value x from the DHAR is subtracted from the SysAddr to produce a
  59. * DramAddr. Here, x represents the base address for the node that
  60. * the SysAddr maps to plus an offset due to memory hoisting. See
  61. * section 3.4.8 and the comments in amd64_get_dram_hole_info() and
  62. * sys_addr_to_dram_addr() below for more information.
  63. *
  64. * If the SysAddr is not affected by the DHAR then a value y is
  65. * subtracted from the SysAddr to produce a DramAddr. Here, y is the
  66. * base address for the node that the SysAddr maps to. See section
  67. * 3.4.4 and the comments in sys_addr_to_dram_addr() below for more
  68. * information.
  69. *
  70. * InputAddr:
  71. * A DramAddr is translated to an InputAddr before being passed to the
  72. * memory controller for the node that the DramAddr is associated
  73. * with. The memory controller then maps the InputAddr to a csrow.
  74. * If node interleaving is not in use, then the InputAddr has the same
  75. * value as the DramAddr. Otherwise, the InputAddr is produced by
  76. * discarding the bits used for node interleaving from the DramAddr.
  77. * See section 3.4.4 for more information.
  78. *
  79. * The memory controller for a given node uses its DRAM CS Base and
  80. * DRAM CS Mask registers to map an InputAddr to a csrow. See
  81. * sections 3.5.4 and 3.5.5 for more information.
  82. */
  83. #define EDAC_AMD64_VERSION "3.4.0"
  84. #define EDAC_MOD_STR "amd64_edac"
  85. /* Extended Model from CPUID, for CPU Revision numbers */
  86. #define K8_REV_D 1
  87. #define K8_REV_E 2
  88. #define K8_REV_F 4
  89. /* Hardware limit on ChipSelect rows per MC and processors per system */
  90. #define NUM_CHIPSELECTS 8
  91. #define DRAM_RANGES 8
  92. #define ON true
  93. #define OFF false
  94. /*
  95. * PCI-defined configuration space registers
  96. */
  97. #define PCI_DEVICE_ID_AMD_15H_NB_F1 0x1601
  98. #define PCI_DEVICE_ID_AMD_15H_NB_F2 0x1602
  99. #define PCI_DEVICE_ID_AMD_15H_M30H_NB_F1 0x141b
  100. #define PCI_DEVICE_ID_AMD_15H_M30H_NB_F2 0x141c
  101. #define PCI_DEVICE_ID_AMD_15H_M60H_NB_F1 0x1571
  102. #define PCI_DEVICE_ID_AMD_15H_M60H_NB_F2 0x1572
  103. #define PCI_DEVICE_ID_AMD_16H_NB_F1 0x1531
  104. #define PCI_DEVICE_ID_AMD_16H_NB_F2 0x1532
  105. #define PCI_DEVICE_ID_AMD_16H_M30H_NB_F1 0x1581
  106. #define PCI_DEVICE_ID_AMD_16H_M30H_NB_F2 0x1582
  107. /*
  108. * Function 1 - Address Map
  109. */
  110. #define DRAM_BASE_LO 0x40
  111. #define DRAM_LIMIT_LO 0x44
  112. /*
  113. * F15 M30h D18F1x2[1C:00]
  114. */
  115. #define DRAM_CONT_BASE 0x200
  116. #define DRAM_CONT_LIMIT 0x204
  117. /*
  118. * F15 M30h D18F1x2[4C:40]
  119. */
  120. #define DRAM_CONT_HIGH_OFF 0x240
  121. #define dram_rw(pvt, i) ((u8)(pvt->ranges[i].base.lo & 0x3))
  122. #define dram_intlv_sel(pvt, i) ((u8)((pvt->ranges[i].lim.lo >> 8) & 0x7))
  123. #define dram_dst_node(pvt, i) ((u8)(pvt->ranges[i].lim.lo & 0x7))
  124. #define DHAR 0xf0
  125. #define dhar_mem_hoist_valid(pvt) ((pvt)->dhar & BIT(1))
  126. #define dhar_base(pvt) ((pvt)->dhar & 0xff000000)
  127. #define k8_dhar_offset(pvt) (((pvt)->dhar & 0x0000ff00) << 16)
  128. /* NOTE: Extra mask bit vs K8 */
  129. #define f10_dhar_offset(pvt) (((pvt)->dhar & 0x0000ff80) << 16)
  130. #define DCT_CFG_SEL 0x10C
  131. #define DRAM_LOCAL_NODE_BASE 0x120
  132. #define DRAM_LOCAL_NODE_LIM 0x124
  133. #define DRAM_BASE_HI 0x140
  134. #define DRAM_LIMIT_HI 0x144
  135. /*
  136. * Function 2 - DRAM controller
  137. */
  138. #define DCSB0 0x40
  139. #define DCSB1 0x140
  140. #define DCSB_CS_ENABLE BIT(0)
  141. #define DCSM0 0x60
  142. #define DCSM1 0x160
  143. #define csrow_enabled(i, dct, pvt) ((pvt)->csels[(dct)].csbases[(i)] & DCSB_CS_ENABLE)
  144. #define DRAM_CONTROL 0x78
  145. #define DBAM0 0x80
  146. #define DBAM1 0x180
  147. /* Extract the DIMM 'type' on the i'th DIMM from the DBAM reg value passed */
  148. #define DBAM_DIMM(i, reg) ((((reg) >> (4*(i)))) & 0xF)
  149. #define DBAM_MAX_VALUE 11
  150. #define DCLR0 0x90
  151. #define DCLR1 0x190
  152. #define REVE_WIDTH_128 BIT(16)
  153. #define WIDTH_128 BIT(11)
  154. #define DCHR0 0x94
  155. #define DCHR1 0x194
  156. #define DDR3_MODE BIT(8)
  157. #define DCT_SEL_LO 0x110
  158. #define dct_high_range_enabled(pvt) ((pvt)->dct_sel_lo & BIT(0))
  159. #define dct_interleave_enabled(pvt) ((pvt)->dct_sel_lo & BIT(2))
  160. #define dct_ganging_enabled(pvt) ((boot_cpu_data.x86 == 0x10) && ((pvt)->dct_sel_lo & BIT(4)))
  161. #define dct_data_intlv_enabled(pvt) ((pvt)->dct_sel_lo & BIT(5))
  162. #define dct_memory_cleared(pvt) ((pvt)->dct_sel_lo & BIT(10))
  163. #define SWAP_INTLV_REG 0x10c
  164. #define DCT_SEL_HI 0x114
  165. #define F15H_M60H_SCRCTRL 0x1C8
  166. /*
  167. * Function 3 - Misc Control
  168. */
  169. #define NBCTL 0x40
  170. #define NBCFG 0x44
  171. #define NBCFG_CHIPKILL BIT(23)
  172. #define NBCFG_ECC_ENABLE BIT(22)
  173. /* F3x48: NBSL */
  174. #define F10_NBSL_EXT_ERR_ECC 0x8
  175. #define NBSL_PP_OBS 0x2
  176. #define SCRCTRL 0x58
  177. #define F10_ONLINE_SPARE 0xB0
  178. #define online_spare_swap_done(pvt, c) (((pvt)->online_spare >> (1 + 2 * (c))) & 0x1)
  179. #define online_spare_bad_dramcs(pvt, c) (((pvt)->online_spare >> (4 + 4 * (c))) & 0x7)
  180. #define F10_NB_ARRAY_ADDR 0xB8
  181. #define F10_NB_ARRAY_DRAM BIT(31)
  182. /* Bits [2:1] are used to select 16-byte section within a 64-byte cacheline */
  183. #define SET_NB_ARRAY_ADDR(section) (((section) & 0x3) << 1)
  184. #define F10_NB_ARRAY_DATA 0xBC
  185. #define F10_NB_ARR_ECC_WR_REQ BIT(17)
  186. #define SET_NB_DRAM_INJECTION_WRITE(inj) \
  187. (BIT(((inj.word) & 0xF) + 20) | \
  188. F10_NB_ARR_ECC_WR_REQ | inj.bit_map)
  189. #define SET_NB_DRAM_INJECTION_READ(inj) \
  190. (BIT(((inj.word) & 0xF) + 20) | \
  191. BIT(16) | inj.bit_map)
  192. #define NBCAP 0xE8
  193. #define NBCAP_CHIPKILL BIT(4)
  194. #define NBCAP_SECDED BIT(3)
  195. #define NBCAP_DCT_DUAL BIT(0)
  196. #define EXT_NB_MCA_CFG 0x180
  197. /* MSRs */
  198. #define MSR_MCGCTL_NBE BIT(4)
  199. enum amd_families {
  200. K8_CPUS = 0,
  201. F10_CPUS,
  202. F15_CPUS,
  203. F15_M30H_CPUS,
  204. F15_M60H_CPUS,
  205. F16_CPUS,
  206. F16_M30H_CPUS,
  207. NUM_FAMILIES,
  208. };
  209. /* Error injection control structure */
  210. struct error_injection {
  211. u32 section;
  212. u32 word;
  213. u32 bit_map;
  214. };
  215. /* low and high part of PCI config space regs */
  216. struct reg_pair {
  217. u32 lo, hi;
  218. };
  219. /*
  220. * See F1x[1, 0][7C:40] DRAM Base/Limit Registers
  221. */
  222. struct dram_range {
  223. struct reg_pair base;
  224. struct reg_pair lim;
  225. };
  226. /* A DCT chip selects collection */
  227. struct chip_select {
  228. u32 csbases[NUM_CHIPSELECTS];
  229. u8 b_cnt;
  230. u32 csmasks[NUM_CHIPSELECTS];
  231. u8 m_cnt;
  232. };
  233. struct amd64_pvt {
  234. struct low_ops *ops;
  235. /* pci_device handles which we utilize */
  236. struct pci_dev *F1, *F2, *F3;
  237. u16 mc_node_id; /* MC index of this MC node */
  238. u8 fam; /* CPU family */
  239. u8 model; /* ... model */
  240. u8 stepping; /* ... stepping */
  241. int ext_model; /* extended model value of this node */
  242. int channel_count;
  243. /* Raw registers */
  244. u32 dclr0; /* DRAM Configuration Low DCT0 reg */
  245. u32 dclr1; /* DRAM Configuration Low DCT1 reg */
  246. u32 dchr0; /* DRAM Configuration High DCT0 reg */
  247. u32 dchr1; /* DRAM Configuration High DCT1 reg */
  248. u32 nbcap; /* North Bridge Capabilities */
  249. u32 nbcfg; /* F10 North Bridge Configuration */
  250. u32 ext_nbcfg; /* Extended F10 North Bridge Configuration */
  251. u32 dhar; /* DRAM Hoist reg */
  252. u32 dbam0; /* DRAM Base Address Mapping reg for DCT0 */
  253. u32 dbam1; /* DRAM Base Address Mapping reg for DCT1 */
  254. /* one for each DCT */
  255. struct chip_select csels[2];
  256. /* DRAM base and limit pairs F1x[78,70,68,60,58,50,48,40] */
  257. struct dram_range ranges[DRAM_RANGES];
  258. u64 top_mem; /* top of memory below 4GB */
  259. u64 top_mem2; /* top of memory above 4GB */
  260. u32 dct_sel_lo; /* DRAM Controller Select Low */
  261. u32 dct_sel_hi; /* DRAM Controller Select High */
  262. u32 online_spare; /* On-Line spare Reg */
  263. /* x4 or x8 syndromes in use */
  264. u8 ecc_sym_sz;
  265. /* place to store error injection parameters prior to issue */
  266. struct error_injection injection;
  267. /* cache the dram_type */
  268. enum mem_type dram_type;
  269. };
  270. enum err_codes {
  271. DECODE_OK = 0,
  272. ERR_NODE = -1,
  273. ERR_CSROW = -2,
  274. ERR_CHANNEL = -3,
  275. };
  276. struct err_info {
  277. int err_code;
  278. struct mem_ctl_info *src_mci;
  279. int csrow;
  280. int channel;
  281. u16 syndrome;
  282. u32 page;
  283. u32 offset;
  284. };
  285. static inline u64 get_dram_base(struct amd64_pvt *pvt, u8 i)
  286. {
  287. u64 addr = ((u64)pvt->ranges[i].base.lo & 0xffff0000) << 8;
  288. if (boot_cpu_data.x86 == 0xf)
  289. return addr;
  290. return (((u64)pvt->ranges[i].base.hi & 0x000000ff) << 40) | addr;
  291. }
  292. static inline u64 get_dram_limit(struct amd64_pvt *pvt, u8 i)
  293. {
  294. u64 lim = (((u64)pvt->ranges[i].lim.lo & 0xffff0000) << 8) | 0x00ffffff;
  295. if (boot_cpu_data.x86 == 0xf)
  296. return lim;
  297. return (((u64)pvt->ranges[i].lim.hi & 0x000000ff) << 40) | lim;
  298. }
  299. static inline u16 extract_syndrome(u64 status)
  300. {
  301. return ((status >> 47) & 0xff) | ((status >> 16) & 0xff00);
  302. }
  303. static inline u8 dct_sel_interleave_addr(struct amd64_pvt *pvt)
  304. {
  305. if (pvt->fam == 0x15 && pvt->model >= 0x30)
  306. return (((pvt->dct_sel_hi >> 9) & 0x1) << 2) |
  307. ((pvt->dct_sel_lo >> 6) & 0x3);
  308. return ((pvt)->dct_sel_lo >> 6) & 0x3;
  309. }
  310. /*
  311. * per-node ECC settings descriptor
  312. */
  313. struct ecc_settings {
  314. u32 old_nbctl;
  315. bool nbctl_valid;
  316. struct flags {
  317. unsigned long nb_mce_enable:1;
  318. unsigned long nb_ecc_prev:1;
  319. } flags;
  320. };
  321. #ifdef CONFIG_EDAC_DEBUG
  322. extern const struct attribute_group amd64_edac_dbg_group;
  323. #endif
  324. #ifdef CONFIG_EDAC_AMD64_ERROR_INJECTION
  325. extern const struct attribute_group amd64_edac_inj_group;
  326. #endif
  327. /*
  328. * Each of the PCI Device IDs types have their own set of hardware accessor
  329. * functions and per device encoding/decoding logic.
  330. */
  331. struct low_ops {
  332. int (*early_channel_count) (struct amd64_pvt *pvt);
  333. void (*map_sysaddr_to_csrow) (struct mem_ctl_info *mci, u64 sys_addr,
  334. struct err_info *);
  335. int (*dbam_to_cs) (struct amd64_pvt *pvt, u8 dct,
  336. unsigned cs_mode, int cs_mask_nr);
  337. };
  338. struct amd64_family_type {
  339. const char *ctl_name;
  340. u16 f1_id, f2_id;
  341. struct low_ops ops;
  342. };
  343. int __amd64_read_pci_cfg_dword(struct pci_dev *pdev, int offset,
  344. u32 *val, const char *func);
  345. int __amd64_write_pci_cfg_dword(struct pci_dev *pdev, int offset,
  346. u32 val, const char *func);
  347. #define amd64_read_pci_cfg(pdev, offset, val) \
  348. __amd64_read_pci_cfg_dword(pdev, offset, val, __func__)
  349. #define amd64_write_pci_cfg(pdev, offset, val) \
  350. __amd64_write_pci_cfg_dword(pdev, offset, val, __func__)
  351. int amd64_get_dram_hole_info(struct mem_ctl_info *mci, u64 *hole_base,
  352. u64 *hole_offset, u64 *hole_size);
  353. #define to_mci(k) container_of(k, struct mem_ctl_info, dev)
  354. /* Injection helpers */
  355. static inline void disable_caches(void *dummy)
  356. {
  357. write_cr0(read_cr0() | X86_CR0_CD);
  358. wbinvd();
  359. }
  360. static inline void enable_caches(void *dummy)
  361. {
  362. write_cr0(read_cr0() & ~X86_CR0_CD);
  363. }
  364. static inline u8 dram_intlv_en(struct amd64_pvt *pvt, unsigned int i)
  365. {
  366. if (pvt->fam == 0x15 && pvt->model >= 0x30) {
  367. u32 tmp;
  368. amd64_read_pci_cfg(pvt->F1, DRAM_CONT_LIMIT, &tmp);
  369. return (u8) tmp & 0xF;
  370. }
  371. return (u8) (pvt->ranges[i].base.lo >> 8) & 0x7;
  372. }
  373. static inline u8 dhar_valid(struct amd64_pvt *pvt)
  374. {
  375. if (pvt->fam == 0x15 && pvt->model >= 0x30) {
  376. u32 tmp;
  377. amd64_read_pci_cfg(pvt->F1, DRAM_CONT_BASE, &tmp);
  378. return (tmp >> 1) & BIT(0);
  379. }
  380. return (pvt)->dhar & BIT(0);
  381. }
  382. static inline u32 dct_sel_baseaddr(struct amd64_pvt *pvt)
  383. {
  384. if (pvt->fam == 0x15 && pvt->model >= 0x30) {
  385. u32 tmp;
  386. amd64_read_pci_cfg(pvt->F1, DRAM_CONT_BASE, &tmp);
  387. return (tmp >> 11) & 0x1FFF;
  388. }
  389. return (pvt)->dct_sel_lo & 0xFFFFF800;
  390. }