amd64_edac.h 14 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463
  1. /*
  2. * AMD64 class Memory Controller kernel module
  3. *
  4. * Copyright (c) 2009 SoftwareBitMaker.
  5. * Copyright (c) 2009 Advanced Micro Devices, Inc.
  6. *
  7. * This file may be distributed under the terms of the
  8. * GNU General Public License.
  9. *
  10. * Originally Written by Thayne Harbaugh
  11. *
  12. * Changes by Douglas "norsk" Thompson <dougthompson@xmission.com>:
  13. * - K8 CPU Revision D and greater support
  14. *
  15. * Changes by Dave Peterson <dsp@llnl.gov> <dave_peterson@pobox.com>:
  16. * - Module largely rewritten, with new (and hopefully correct)
  17. * code for dealing with node and chip select interleaving,
  18. * various code cleanup, and bug fixes
  19. * - Added support for memory hoisting using DRAM hole address
  20. * register
  21. *
  22. * Changes by Douglas "norsk" Thompson <dougthompson@xmission.com>:
  23. * -K8 Rev (1207) revision support added, required Revision
  24. * specific mini-driver code to support Rev F as well as
  25. * prior revisions
  26. *
  27. * Changes by Douglas "norsk" Thompson <dougthompson@xmission.com>:
  28. * -Family 10h revision support added. New PCI Device IDs,
  29. * indicating new changes. Actual registers modified
  30. * were slight, less than the Rev E to Rev F transition
  31. * but changing the PCI Device ID was the proper thing to
  32. * do, as it provides for almost automactic family
  33. * detection. The mods to Rev F required more family
  34. * information detection.
  35. *
  36. * Changes/Fixes by Borislav Petkov <borislav.petkov@amd.com>:
  37. * - misc fixes and code cleanups
  38. *
  39. * This module is based on the following documents
  40. * (available from http://www.amd.com/):
  41. *
  42. * Title: BIOS and Kernel Developer's Guide for AMD Athlon 64 and AMD
  43. * Opteron Processors
  44. * AMD publication #: 26094
  45. *` Revision: 3.26
  46. *
  47. * Title: BIOS and Kernel Developer's Guide for AMD NPT Family 0Fh
  48. * Processors
  49. * AMD publication #: 32559
  50. * Revision: 3.00
  51. * Issue Date: May 2006
  52. *
  53. * Title: BIOS and Kernel Developer's Guide (BKDG) For AMD Family 10h
  54. * Processors
  55. * AMD publication #: 31116
  56. * Revision: 3.00
  57. * Issue Date: September 07, 2007
  58. *
  59. * Sections in the first 2 documents are no longer in sync with each other.
  60. * The Family 10h BKDG was totally re-written from scratch with a new
  61. * presentation model.
  62. * Therefore, comments that refer to a Document section might be off.
  63. */
  64. #include <linux/module.h>
  65. #include <linux/ctype.h>
  66. #include <linux/init.h>
  67. #include <linux/pci.h>
  68. #include <linux/pci_ids.h>
  69. #include <linux/slab.h>
  70. #include <linux/mmzone.h>
  71. #include <linux/edac.h>
  72. #include <asm/msr.h>
  73. #include "edac_core.h"
  74. #include "mce_amd.h"
  75. #define amd64_debug(fmt, arg...) \
  76. edac_printk(KERN_DEBUG, "amd64", fmt, ##arg)
  77. #define amd64_info(fmt, arg...) \
  78. edac_printk(KERN_INFO, "amd64", fmt, ##arg)
  79. #define amd64_notice(fmt, arg...) \
  80. edac_printk(KERN_NOTICE, "amd64", fmt, ##arg)
  81. #define amd64_warn(fmt, arg...) \
  82. edac_printk(KERN_WARNING, "amd64", fmt, ##arg)
  83. #define amd64_err(fmt, arg...) \
  84. edac_printk(KERN_ERR, "amd64", fmt, ##arg)
  85. #define amd64_mc_warn(mci, fmt, arg...) \
  86. edac_mc_chipset_printk(mci, KERN_WARNING, "amd64", fmt, ##arg)
  87. #define amd64_mc_err(mci, fmt, arg...) \
  88. edac_mc_chipset_printk(mci, KERN_ERR, "amd64", fmt, ##arg)
  89. /*
  90. * Throughout the comments in this code, the following terms are used:
  91. *
  92. * SysAddr, DramAddr, and InputAddr
  93. *
  94. * These terms come directly from the amd64 documentation
  95. * (AMD publication #26094). They are defined as follows:
  96. *
  97. * SysAddr:
  98. * This is a physical address generated by a CPU core or a device
  99. * doing DMA. If generated by a CPU core, a SysAddr is the result of
  100. * a virtual to physical address translation by the CPU core's address
  101. * translation mechanism (MMU).
  102. *
  103. * DramAddr:
  104. * A DramAddr is derived from a SysAddr by subtracting an offset that
  105. * depends on which node the SysAddr maps to and whether the SysAddr
  106. * is within a range affected by memory hoisting. The DRAM Base
  107. * (section 3.4.4.1) and DRAM Limit (section 3.4.4.2) registers
  108. * determine which node a SysAddr maps to.
  109. *
  110. * If the DRAM Hole Address Register (DHAR) is enabled and the SysAddr
  111. * is within the range of addresses specified by this register, then
  112. * a value x from the DHAR is subtracted from the SysAddr to produce a
  113. * DramAddr. Here, x represents the base address for the node that
  114. * the SysAddr maps to plus an offset due to memory hoisting. See
  115. * section 3.4.8 and the comments in amd64_get_dram_hole_info() and
  116. * sys_addr_to_dram_addr() below for more information.
  117. *
  118. * If the SysAddr is not affected by the DHAR then a value y is
  119. * subtracted from the SysAddr to produce a DramAddr. Here, y is the
  120. * base address for the node that the SysAddr maps to. See section
  121. * 3.4.4 and the comments in sys_addr_to_dram_addr() below for more
  122. * information.
  123. *
  124. * InputAddr:
  125. * A DramAddr is translated to an InputAddr before being passed to the
  126. * memory controller for the node that the DramAddr is associated
  127. * with. The memory controller then maps the InputAddr to a csrow.
  128. * If node interleaving is not in use, then the InputAddr has the same
  129. * value as the DramAddr. Otherwise, the InputAddr is produced by
  130. * discarding the bits used for node interleaving from the DramAddr.
  131. * See section 3.4.4 for more information.
  132. *
  133. * The memory controller for a given node uses its DRAM CS Base and
  134. * DRAM CS Mask registers to map an InputAddr to a csrow. See
  135. * sections 3.5.4 and 3.5.5 for more information.
  136. */
  137. #define EDAC_AMD64_VERSION "3.4.0"
  138. #define EDAC_MOD_STR "amd64_edac"
  139. /* Extended Model from CPUID, for CPU Revision numbers */
  140. #define K8_REV_D 1
  141. #define K8_REV_E 2
  142. #define K8_REV_F 4
  143. /* Hardware limit on ChipSelect rows per MC and processors per system */
  144. #define NUM_CHIPSELECTS 8
  145. #define DRAM_RANGES 8
  146. #define ON true
  147. #define OFF false
  148. /*
  149. * Create a contiguous bitmask starting at bit position @lo and ending at
  150. * position @hi. For example
  151. *
  152. * GENMASK(21, 39) gives us the 64bit vector 0x000000ffffe00000.
  153. */
  154. #define GENMASK(lo, hi) (((1ULL << ((hi) - (lo) + 1)) - 1) << (lo))
  155. /*
  156. * PCI-defined configuration space registers
  157. */
  158. #define PCI_DEVICE_ID_AMD_15H_NB_F1 0x1601
  159. #define PCI_DEVICE_ID_AMD_15H_NB_F2 0x1602
  160. /*
  161. * Function 1 - Address Map
  162. */
  163. #define DRAM_BASE_LO 0x40
  164. #define DRAM_LIMIT_LO 0x44
  165. #define dram_intlv_en(pvt, i) ((u8)((pvt->ranges[i].base.lo >> 8) & 0x7))
  166. #define dram_rw(pvt, i) ((u8)(pvt->ranges[i].base.lo & 0x3))
  167. #define dram_intlv_sel(pvt, i) ((u8)((pvt->ranges[i].lim.lo >> 8) & 0x7))
  168. #define dram_dst_node(pvt, i) ((u8)(pvt->ranges[i].lim.lo & 0x7))
  169. #define DHAR 0xf0
  170. #define dhar_valid(pvt) ((pvt)->dhar & BIT(0))
  171. #define dhar_mem_hoist_valid(pvt) ((pvt)->dhar & BIT(1))
  172. #define dhar_base(pvt) ((pvt)->dhar & 0xff000000)
  173. #define k8_dhar_offset(pvt) (((pvt)->dhar & 0x0000ff00) << 16)
  174. /* NOTE: Extra mask bit vs K8 */
  175. #define f10_dhar_offset(pvt) (((pvt)->dhar & 0x0000ff80) << 16)
  176. #define DCT_CFG_SEL 0x10C
  177. #define DRAM_LOCAL_NODE_BASE 0x120
  178. #define DRAM_LOCAL_NODE_LIM 0x124
  179. #define DRAM_BASE_HI 0x140
  180. #define DRAM_LIMIT_HI 0x144
  181. /*
  182. * Function 2 - DRAM controller
  183. */
  184. #define DCSB0 0x40
  185. #define DCSB1 0x140
  186. #define DCSB_CS_ENABLE BIT(0)
  187. #define DCSM0 0x60
  188. #define DCSM1 0x160
  189. #define csrow_enabled(i, dct, pvt) ((pvt)->csels[(dct)].csbases[(i)] & DCSB_CS_ENABLE)
  190. #define DBAM0 0x80
  191. #define DBAM1 0x180
  192. /* Extract the DIMM 'type' on the i'th DIMM from the DBAM reg value passed */
  193. #define DBAM_DIMM(i, reg) ((((reg) >> (4*i))) & 0xF)
  194. #define DBAM_MAX_VALUE 11
  195. #define DCLR0 0x90
  196. #define DCLR1 0x190
  197. #define REVE_WIDTH_128 BIT(16)
  198. #define WIDTH_128 BIT(11)
  199. #define DCHR0 0x94
  200. #define DCHR1 0x194
  201. #define DDR3_MODE BIT(8)
  202. #define DCT_SEL_LO 0x110
  203. #define dct_sel_baseaddr(pvt) ((pvt)->dct_sel_lo & 0xFFFFF800)
  204. #define dct_sel_interleave_addr(pvt) (((pvt)->dct_sel_lo >> 6) & 0x3)
  205. #define dct_high_range_enabled(pvt) ((pvt)->dct_sel_lo & BIT(0))
  206. #define dct_interleave_enabled(pvt) ((pvt)->dct_sel_lo & BIT(2))
  207. #define dct_ganging_enabled(pvt) ((boot_cpu_data.x86 == 0x10) && ((pvt)->dct_sel_lo & BIT(4)))
  208. #define dct_data_intlv_enabled(pvt) ((pvt)->dct_sel_lo & BIT(5))
  209. #define dct_memory_cleared(pvt) ((pvt)->dct_sel_lo & BIT(10))
  210. #define SWAP_INTLV_REG 0x10c
  211. #define DCT_SEL_HI 0x114
  212. /*
  213. * Function 3 - Misc Control
  214. */
  215. #define NBCTL 0x40
  216. #define NBCFG 0x44
  217. #define NBCFG_CHIPKILL BIT(23)
  218. #define NBCFG_ECC_ENABLE BIT(22)
  219. /* F3x48: NBSL */
  220. #define F10_NBSL_EXT_ERR_ECC 0x8
  221. #define NBSL_PP_OBS 0x2
  222. #define SCRCTRL 0x58
  223. #define F10_ONLINE_SPARE 0xB0
  224. #define online_spare_swap_done(pvt, c) (((pvt)->online_spare >> (1 + 2 * (c))) & 0x1)
  225. #define online_spare_bad_dramcs(pvt, c) (((pvt)->online_spare >> (4 + 4 * (c))) & 0x7)
  226. #define F10_NB_ARRAY_ADDR 0xB8
  227. #define F10_NB_ARRAY_DRAM_ECC BIT(31)
  228. /* Bits [2:1] are used to select 16-byte section within a 64-byte cacheline */
  229. #define SET_NB_ARRAY_ADDRESS(section) (((section) & 0x3) << 1)
  230. #define F10_NB_ARRAY_DATA 0xBC
  231. #define SET_NB_DRAM_INJECTION_WRITE(word, bits) \
  232. (BIT(((word) & 0xF) + 20) | \
  233. BIT(17) | bits)
  234. #define SET_NB_DRAM_INJECTION_READ(word, bits) \
  235. (BIT(((word) & 0xF) + 20) | \
  236. BIT(16) | bits)
  237. #define NBCAP 0xE8
  238. #define NBCAP_CHIPKILL BIT(4)
  239. #define NBCAP_SECDED BIT(3)
  240. #define NBCAP_DCT_DUAL BIT(0)
  241. #define EXT_NB_MCA_CFG 0x180
  242. /* MSRs */
  243. #define MSR_MCGCTL_NBE BIT(4)
  244. /* AMD sets the first MC device at device ID 0x18. */
  245. static inline u8 get_node_id(struct pci_dev *pdev)
  246. {
  247. return PCI_SLOT(pdev->devfn) - 0x18;
  248. }
  249. enum amd_families {
  250. K8_CPUS = 0,
  251. F10_CPUS,
  252. F15_CPUS,
  253. NUM_FAMILIES,
  254. };
  255. /* Error injection control structure */
  256. struct error_injection {
  257. u32 section;
  258. u32 word;
  259. u32 bit_map;
  260. };
  261. /* low and high part of PCI config space regs */
  262. struct reg_pair {
  263. u32 lo, hi;
  264. };
  265. /*
  266. * See F1x[1, 0][7C:40] DRAM Base/Limit Registers
  267. */
  268. struct dram_range {
  269. struct reg_pair base;
  270. struct reg_pair lim;
  271. };
  272. /* A DCT chip selects collection */
  273. struct chip_select {
  274. u32 csbases[NUM_CHIPSELECTS];
  275. u8 b_cnt;
  276. u32 csmasks[NUM_CHIPSELECTS];
  277. u8 m_cnt;
  278. };
  279. struct amd64_pvt {
  280. struct low_ops *ops;
  281. /* pci_device handles which we utilize */
  282. struct pci_dev *F1, *F2, *F3;
  283. unsigned mc_node_id; /* MC index of this MC node */
  284. int ext_model; /* extended model value of this node */
  285. int channel_count;
  286. /* Raw registers */
  287. u32 dclr0; /* DRAM Configuration Low DCT0 reg */
  288. u32 dclr1; /* DRAM Configuration Low DCT1 reg */
  289. u32 dchr0; /* DRAM Configuration High DCT0 reg */
  290. u32 dchr1; /* DRAM Configuration High DCT1 reg */
  291. u32 nbcap; /* North Bridge Capabilities */
  292. u32 nbcfg; /* F10 North Bridge Configuration */
  293. u32 ext_nbcfg; /* Extended F10 North Bridge Configuration */
  294. u32 dhar; /* DRAM Hoist reg */
  295. u32 dbam0; /* DRAM Base Address Mapping reg for DCT0 */
  296. u32 dbam1; /* DRAM Base Address Mapping reg for DCT1 */
  297. /* one for each DCT */
  298. struct chip_select csels[2];
  299. /* DRAM base and limit pairs F1x[78,70,68,60,58,50,48,40] */
  300. struct dram_range ranges[DRAM_RANGES];
  301. u64 top_mem; /* top of memory below 4GB */
  302. u64 top_mem2; /* top of memory above 4GB */
  303. u32 dct_sel_lo; /* DRAM Controller Select Low */
  304. u32 dct_sel_hi; /* DRAM Controller Select High */
  305. u32 online_spare; /* On-Line spare Reg */
  306. /* x4 or x8 syndromes in use */
  307. u8 ecc_sym_sz;
  308. /* place to store error injection parameters prior to issue */
  309. struct error_injection injection;
  310. };
  311. static inline u64 get_dram_base(struct amd64_pvt *pvt, unsigned i)
  312. {
  313. u64 addr = ((u64)pvt->ranges[i].base.lo & 0xffff0000) << 8;
  314. if (boot_cpu_data.x86 == 0xf)
  315. return addr;
  316. return (((u64)pvt->ranges[i].base.hi & 0x000000ff) << 40) | addr;
  317. }
  318. static inline u64 get_dram_limit(struct amd64_pvt *pvt, unsigned i)
  319. {
  320. u64 lim = (((u64)pvt->ranges[i].lim.lo & 0xffff0000) << 8) | 0x00ffffff;
  321. if (boot_cpu_data.x86 == 0xf)
  322. return lim;
  323. return (((u64)pvt->ranges[i].lim.hi & 0x000000ff) << 40) | lim;
  324. }
  325. static inline u16 extract_syndrome(u64 status)
  326. {
  327. return ((status >> 47) & 0xff) | ((status >> 16) & 0xff00);
  328. }
  329. /*
  330. * per-node ECC settings descriptor
  331. */
  332. struct ecc_settings {
  333. u32 old_nbctl;
  334. bool nbctl_valid;
  335. struct flags {
  336. unsigned long nb_mce_enable:1;
  337. unsigned long nb_ecc_prev:1;
  338. } flags;
  339. };
  340. #ifdef CONFIG_EDAC_DEBUG
  341. #define NUM_DBG_ATTRS 5
  342. #else
  343. #define NUM_DBG_ATTRS 0
  344. #endif
  345. #ifdef CONFIG_EDAC_AMD64_ERROR_INJECTION
  346. #define NUM_INJ_ATTRS 5
  347. #else
  348. #define NUM_INJ_ATTRS 0
  349. #endif
  350. extern struct mcidev_sysfs_attribute amd64_dbg_attrs[NUM_DBG_ATTRS],
  351. amd64_inj_attrs[NUM_INJ_ATTRS];
  352. /*
  353. * Each of the PCI Device IDs types have their own set of hardware accessor
  354. * functions and per device encoding/decoding logic.
  355. */
  356. struct low_ops {
  357. int (*early_channel_count) (struct amd64_pvt *pvt);
  358. void (*map_sysaddr_to_csrow) (struct mem_ctl_info *mci, u64 sys_addr,
  359. u16 syndrome);
  360. int (*dbam_to_cs) (struct amd64_pvt *pvt, u8 dct, unsigned cs_mode);
  361. int (*read_dct_pci_cfg) (struct amd64_pvt *pvt, int offset,
  362. u32 *val, const char *func);
  363. };
  364. struct amd64_family_type {
  365. const char *ctl_name;
  366. u16 f1_id, f3_id;
  367. struct low_ops ops;
  368. };
  369. int __amd64_write_pci_cfg_dword(struct pci_dev *pdev, int offset,
  370. u32 val, const char *func);
  371. #define amd64_read_pci_cfg(pdev, offset, val) \
  372. __amd64_read_pci_cfg_dword(pdev, offset, val, __func__)
  373. #define amd64_write_pci_cfg(pdev, offset, val) \
  374. __amd64_write_pci_cfg_dword(pdev, offset, val, __func__)
  375. #define amd64_read_dct_pci_cfg(pvt, offset, val) \
  376. pvt->ops->read_dct_pci_cfg(pvt, offset, val, __func__)
  377. int amd64_get_dram_hole_info(struct mem_ctl_info *mci, u64 *hole_base,
  378. u64 *hole_offset, u64 *hole_size);