i82443bxgx_edac.c 12 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401
  1. /*
  2. * Intel 82443BX/GX (440BX/GX chipset) Memory Controller EDAC kernel
  3. * module (C) 2006 Tim Small
  4. *
  5. * This file may be distributed under the terms of the GNU General
  6. * Public License.
  7. *
  8. * Written by Tim Small <tim@buttersideup.com>, based on work by Linux
  9. * Networx, Thayne Harbaugh, Dan Hollis <goemon at anime dot net> and
  10. * others.
  11. *
  12. * 440GX fix by Jason Uhlenkott <juhlenko@akamai.com>.
  13. *
  14. * Written with reference to 82443BX Host Bridge Datasheet:
  15. * http://www.intel.com/design/chipsets/440/documentation.htm
  16. * references to this document given in [].
  17. *
  18. * This module doesn't support the 440LX, but it may be possible to
  19. * make it do so (the 440LX's register definitions are different, but
  20. * not completely so - I haven't studied them in enough detail to know
  21. * how easy this would be).
  22. */
  23. #include <linux/module.h>
  24. #include <linux/init.h>
  25. #include <linux/pci.h>
  26. #include <linux/pci_ids.h>
  27. #include <linux/slab.h>
  28. #include "edac_core.h"
  29. #define I82443_REVISION "0.1"
  30. #define EDAC_MOD_STR "i82443bxgx_edac"
  31. /* The 82443BX supports SDRAM, or EDO (EDO for mobile only), "Memory
  32. * Size: 8 MB to 512 MB (1GB with Registered DIMMs) with eight memory
  33. * rows" "The 82443BX supports multiple-bit error detection and
  34. * single-bit error correction when ECC mode is enabled and
  35. * single/multi-bit error detection when correction is disabled.
  36. * During writes to the DRAM, the 82443BX generates ECC for the data
  37. * on a QWord basis. Partial QWord writes require a read-modify-write
  38. * cycle when ECC is enabled."
  39. */
  40. /* "Additionally, the 82443BX ensures that the data is corrected in
  41. * main memory so that accumulation of errors is prevented. Another
  42. * error within the same QWord would result in a double-bit error
  43. * which is unrecoverable. This is known as hardware scrubbing since
  44. * it requires no software intervention to correct the data in memory."
  45. */
  46. /* [Also see page 100 (section 4.3), "DRAM Interface"]
  47. * [Also see page 112 (section 4.6.1.4), ECC]
  48. */
  49. #define I82443BXGX_NR_CSROWS 8
  50. #define I82443BXGX_NR_CHANS 1
  51. #define I82443BXGX_NR_DIMMS 4
  52. /* 82443 PCI Device 0 */
  53. #define I82443BXGX_NBXCFG 0x50 /* 32bit register starting at this PCI
  54. * config space offset */
  55. #define I82443BXGX_NBXCFG_OFFSET_NON_ECCROW 24 /* Array of bits, zero if
  56. * row is non-ECC */
  57. #define I82443BXGX_NBXCFG_OFFSET_DRAM_FREQ 12 /* 2 bits,00=100MHz,10=66 MHz */
  58. #define I82443BXGX_NBXCFG_OFFSET_DRAM_INTEGRITY 7 /* 2 bits: */
  59. #define I82443BXGX_NBXCFG_INTEGRITY_NONE 0x0 /* 00 = Non-ECC */
  60. #define I82443BXGX_NBXCFG_INTEGRITY_EC 0x1 /* 01 = EC (only) */
  61. #define I82443BXGX_NBXCFG_INTEGRITY_ECC 0x2 /* 10 = ECC */
  62. #define I82443BXGX_NBXCFG_INTEGRITY_SCRUB 0x3 /* 11 = ECC + HW Scrub */
  63. #define I82443BXGX_NBXCFG_OFFSET_ECC_DIAG_ENABLE 6
  64. /* 82443 PCI Device 0 */
  65. #define I82443BXGX_EAP 0x80 /* 32bit register starting at this PCI
  66. * config space offset, Error Address
  67. * Pointer Register */
  68. #define I82443BXGX_EAP_OFFSET_EAP 12 /* High 20 bits of error address */
  69. #define I82443BXGX_EAP_OFFSET_MBE BIT(1) /* Err at EAP was multi-bit (W1TC) */
  70. #define I82443BXGX_EAP_OFFSET_SBE BIT(0) /* Err at EAP was single-bit (W1TC) */
  71. #define I82443BXGX_ERRCMD 0x90 /* 8bit register starting at this PCI
  72. * config space offset. */
  73. #define I82443BXGX_ERRCMD_OFFSET_SERR_ON_MBE BIT(1) /* 1 = enable */
  74. #define I82443BXGX_ERRCMD_OFFSET_SERR_ON_SBE BIT(0) /* 1 = enable */
  75. #define I82443BXGX_ERRSTS 0x91 /* 16bit register starting at this PCI
  76. * config space offset. */
  77. #define I82443BXGX_ERRSTS_OFFSET_MBFRE 5 /* 3 bits - first err row multibit */
  78. #define I82443BXGX_ERRSTS_OFFSET_MEF BIT(4) /* 1 = MBE occurred */
  79. #define I82443BXGX_ERRSTS_OFFSET_SBFRE 1 /* 3 bits - first err row singlebit */
  80. #define I82443BXGX_ERRSTS_OFFSET_SEF BIT(0) /* 1 = SBE occurred */
  81. #define I82443BXGX_DRAMC 0x57 /* 8bit register starting at this PCI
  82. * config space offset. */
  83. #define I82443BXGX_DRAMC_OFFSET_DT 3 /* 2 bits, DRAM Type */
  84. #define I82443BXGX_DRAMC_DRAM_IS_EDO 0 /* 00 = EDO */
  85. #define I82443BXGX_DRAMC_DRAM_IS_SDRAM 1 /* 01 = SDRAM */
  86. #define I82443BXGX_DRAMC_DRAM_IS_RSDRAM 2 /* 10 = Registered SDRAM */
  87. #define I82443BXGX_DRB 0x60 /* 8x 8bit registers starting at this PCI
  88. * config space offset. */
  89. /* FIXME - don't poll when ECC disabled? */
  90. struct i82443bxgx_edacmc_error_info {
  91. u32 eap;
  92. };
  93. static struct edac_pci_ctl_info *i82443bxgx_pci;
  94. static void i82443bxgx_edacmc_get_error_info(struct mem_ctl_info *mci,
  95. struct i82443bxgx_edacmc_error_info
  96. *info)
  97. {
  98. struct pci_dev *pdev;
  99. pdev = to_pci_dev(mci->dev);
  100. pci_read_config_dword(pdev, I82443BXGX_EAP, &info->eap);
  101. if (info->eap & I82443BXGX_EAP_OFFSET_SBE)
  102. /* Clear error to allow next error to be reported [p.61] */
  103. pci_write_bits32(pdev, I82443BXGX_EAP,
  104. I82443BXGX_EAP_OFFSET_SBE,
  105. I82443BXGX_EAP_OFFSET_SBE);
  106. if (info->eap & I82443BXGX_EAP_OFFSET_MBE)
  107. /* Clear error to allow next error to be reported [p.61] */
  108. pci_write_bits32(pdev, I82443BXGX_EAP,
  109. I82443BXGX_EAP_OFFSET_MBE,
  110. I82443BXGX_EAP_OFFSET_MBE);
  111. }
  112. static int i82443bxgx_edacmc_process_error_info(struct mem_ctl_info *mci,
  113. struct
  114. i82443bxgx_edacmc_error_info
  115. *info, int handle_errors)
  116. {
  117. int error_found = 0;
  118. u32 eapaddr, page, pageoffset;
  119. /* bits 30:12 hold the 4kb block in which the error occurred
  120. * [p.61] */
  121. eapaddr = (info->eap & 0xfffff000);
  122. page = eapaddr >> PAGE_SHIFT;
  123. pageoffset = eapaddr - (page << PAGE_SHIFT);
  124. if (info->eap & I82443BXGX_EAP_OFFSET_SBE) {
  125. error_found = 1;
  126. if (handle_errors)
  127. edac_mc_handle_ce(mci, page, pageoffset,
  128. /* 440BX/GX don't make syndrome information available */
  129. 0, edac_mc_find_csrow_by_page(mci, page), 0, /* channel */
  130. mci->ctl_name);
  131. }
  132. if (info->eap & I82443BXGX_EAP_OFFSET_MBE) {
  133. error_found = 1;
  134. if (handle_errors)
  135. edac_mc_handle_ue(mci, page, pageoffset,
  136. edac_mc_find_csrow_by_page(mci, page),
  137. mci->ctl_name);
  138. }
  139. return error_found;
  140. }
  141. static void i82443bxgx_edacmc_check(struct mem_ctl_info *mci)
  142. {
  143. struct i82443bxgx_edacmc_error_info info;
  144. debugf1("MC%d: " __FILE__ ": %s()\n", mci->mc_idx, __func__);
  145. i82443bxgx_edacmc_get_error_info(mci, &info);
  146. i82443bxgx_edacmc_process_error_info(mci, &info, 1);
  147. }
  148. static void i82443bxgx_init_csrows(struct mem_ctl_info *mci,
  149. struct pci_dev *pdev,
  150. enum edac_type edac_mode,
  151. enum mem_type mtype)
  152. {
  153. struct csrow_info *csrow;
  154. int index;
  155. u8 drbar, dramc;
  156. u32 row_base, row_high_limit, row_high_limit_last;
  157. pci_read_config_byte(pdev, I82443BXGX_DRAMC, &dramc);
  158. row_high_limit_last = 0;
  159. for (index = 0; index < mci->nr_csrows; index++) {
  160. csrow = &mci->csrows[index];
  161. pci_read_config_byte(pdev, I82443BXGX_DRB + index, &drbar);
  162. debugf1("MC%d: " __FILE__ ": %s() Row=%d DRB = %#0x\n",
  163. mci->mc_idx, __func__, index, drbar);
  164. row_high_limit = ((u32) drbar << 23);
  165. /* find the DRAM Chip Select Base address and mask */
  166. debugf1("MC%d: " __FILE__ ": %s() Row=%d, "
  167. "Boundry Address=%#0x, Last = %#0x \n",
  168. mci->mc_idx, __func__, index, row_high_limit,
  169. row_high_limit_last);
  170. /* 440GX goes to 2GB, represented with a DRB of 0. */
  171. if (row_high_limit_last && !row_high_limit)
  172. row_high_limit = 1UL << 31;
  173. /* This row is empty [p.49] */
  174. if (row_high_limit == row_high_limit_last)
  175. continue;
  176. row_base = row_high_limit_last;
  177. csrow->first_page = row_base >> PAGE_SHIFT;
  178. csrow->last_page = (row_high_limit >> PAGE_SHIFT) - 1;
  179. csrow->nr_pages = csrow->last_page - csrow->first_page + 1;
  180. /* EAP reports in 4kilobyte granularity [61] */
  181. csrow->grain = 1 << 12;
  182. csrow->mtype = mtype;
  183. /* I don't think 440BX can tell you device type? FIXME? */
  184. csrow->dtype = DEV_UNKNOWN;
  185. /* Mode is global to all rows on 440BX */
  186. csrow->edac_mode = edac_mode;
  187. row_high_limit_last = row_high_limit;
  188. }
  189. }
  190. static int i82443bxgx_edacmc_probe1(struct pci_dev *pdev, int dev_idx)
  191. {
  192. struct mem_ctl_info *mci;
  193. u8 dramc;
  194. u32 nbxcfg, ecc_mode;
  195. enum mem_type mtype;
  196. enum edac_type edac_mode;
  197. debugf0("MC: " __FILE__ ": %s()\n", __func__);
  198. /* Something is really hosed if PCI config space reads from
  199. the MC aren't working. */
  200. if (pci_read_config_dword(pdev, I82443BXGX_NBXCFG, &nbxcfg))
  201. return -EIO;
  202. mci = edac_mc_alloc(0, I82443BXGX_NR_CSROWS, I82443BXGX_NR_CHANS);
  203. if (mci == NULL)
  204. return -ENOMEM;
  205. debugf0("MC: " __FILE__ ": %s(): mci = %p\n", __func__, mci);
  206. mci->dev = &pdev->dev;
  207. mci->mtype_cap = MEM_FLAG_EDO | MEM_FLAG_SDR | MEM_FLAG_RDR;
  208. mci->edac_ctl_cap = EDAC_FLAG_NONE | EDAC_FLAG_EC | EDAC_FLAG_SECDED;
  209. pci_read_config_byte(pdev, I82443BXGX_DRAMC, &dramc);
  210. switch ((dramc >> I82443BXGX_DRAMC_OFFSET_DT) & (BIT(0) | BIT(1))) {
  211. case I82443BXGX_DRAMC_DRAM_IS_EDO:
  212. mtype = MEM_EDO;
  213. break;
  214. case I82443BXGX_DRAMC_DRAM_IS_SDRAM:
  215. mtype = MEM_SDR;
  216. break;
  217. case I82443BXGX_DRAMC_DRAM_IS_RSDRAM:
  218. mtype = MEM_RDR;
  219. break;
  220. default:
  221. debugf0
  222. ("Unknown/reserved DRAM type value in DRAMC register!\n");
  223. mtype = -MEM_UNKNOWN;
  224. }
  225. if ((mtype == MEM_SDR) || (mtype == MEM_RDR))
  226. mci->edac_cap = mci->edac_ctl_cap;
  227. else
  228. mci->edac_cap = EDAC_FLAG_NONE;
  229. mci->scrub_cap = SCRUB_FLAG_HW_SRC;
  230. pci_read_config_dword(pdev, I82443BXGX_NBXCFG, &nbxcfg);
  231. ecc_mode = ((nbxcfg >> I82443BXGX_NBXCFG_OFFSET_DRAM_INTEGRITY) &
  232. (BIT(0) | BIT(1)));
  233. mci->scrub_mode = (ecc_mode == I82443BXGX_NBXCFG_INTEGRITY_SCRUB)
  234. ? SCRUB_HW_SRC : SCRUB_NONE;
  235. switch (ecc_mode) {
  236. case I82443BXGX_NBXCFG_INTEGRITY_NONE:
  237. edac_mode = EDAC_NONE;
  238. break;
  239. case I82443BXGX_NBXCFG_INTEGRITY_EC:
  240. edac_mode = EDAC_EC;
  241. break;
  242. case I82443BXGX_NBXCFG_INTEGRITY_ECC:
  243. case I82443BXGX_NBXCFG_INTEGRITY_SCRUB:
  244. edac_mode = EDAC_SECDED;
  245. break;
  246. default:
  247. debugf0
  248. ("%s(): Unknown/reserved ECC state in NBXCFG register!\n",
  249. __func__);
  250. edac_mode = EDAC_UNKNOWN;
  251. break;
  252. }
  253. i82443bxgx_init_csrows(mci, pdev, edac_mode, mtype);
  254. /* Many BIOSes don't clear error flags on boot, so do this
  255. * here, or we get "phantom" errors occuring at module-load
  256. * time. */
  257. pci_write_bits32(pdev, I82443BXGX_EAP,
  258. (I82443BXGX_EAP_OFFSET_SBE |
  259. I82443BXGX_EAP_OFFSET_MBE),
  260. (I82443BXGX_EAP_OFFSET_SBE |
  261. I82443BXGX_EAP_OFFSET_MBE));
  262. mci->mod_name = EDAC_MOD_STR;
  263. mci->mod_ver = I82443_REVISION;
  264. mci->ctl_name = "I82443BXGX";
  265. mci->dev_name = pci_name(pdev);
  266. mci->edac_check = i82443bxgx_edacmc_check;
  267. mci->ctl_page_to_phys = NULL;
  268. if (edac_mc_add_mc(mci, 0)) {
  269. debugf3("%s(): failed edac_mc_add_mc()\n", __func__);
  270. goto fail;
  271. }
  272. /* allocating generic PCI control info */
  273. i82443bxgx_pci = edac_pci_create_generic_ctl(&pdev->dev, EDAC_MOD_STR);
  274. if (!i82443bxgx_pci) {
  275. printk(KERN_WARNING
  276. "%s(): Unable to create PCI control\n",
  277. __func__);
  278. printk(KERN_WARNING
  279. "%s(): PCI error report via EDAC not setup\n",
  280. __func__);
  281. }
  282. debugf3("MC: " __FILE__ ": %s(): success\n", __func__);
  283. return 0;
  284. fail:
  285. edac_mc_free(mci);
  286. return -ENODEV;
  287. }
  288. EXPORT_SYMBOL_GPL(i82443bxgx_edacmc_probe1);
  289. /* returns count (>= 0), or negative on error */
  290. static int __devinit i82443bxgx_edacmc_init_one(struct pci_dev *pdev,
  291. const struct pci_device_id *ent)
  292. {
  293. debugf0("MC: " __FILE__ ": %s()\n", __func__);
  294. /* don't need to call pci_device_enable() */
  295. return i82443bxgx_edacmc_probe1(pdev, ent->driver_data);
  296. }
  297. static void __devexit i82443bxgx_edacmc_remove_one(struct pci_dev *pdev)
  298. {
  299. struct mem_ctl_info *mci;
  300. debugf0(__FILE__ ": %s()\n", __func__);
  301. if (i82443bxgx_pci)
  302. edac_pci_release_generic_ctl(i82443bxgx_pci);
  303. if ((mci = edac_mc_del_mc(&pdev->dev)) == NULL)
  304. return;
  305. edac_mc_free(mci);
  306. }
  307. EXPORT_SYMBOL_GPL(i82443bxgx_edacmc_remove_one);
  308. static const struct pci_device_id i82443bxgx_pci_tbl[] __devinitdata = {
  309. {PCI_DEVICE(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_82443BX_0)},
  310. {PCI_DEVICE(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_82443BX_2)},
  311. {PCI_DEVICE(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_82443GX_0)},
  312. {PCI_DEVICE(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_82443GX_2)},
  313. {0,} /* 0 terminated list. */
  314. };
  315. MODULE_DEVICE_TABLE(pci, i82443bxgx_pci_tbl);
  316. static struct pci_driver i82443bxgx_edacmc_driver = {
  317. .name = EDAC_MOD_STR,
  318. .probe = i82443bxgx_edacmc_init_one,
  319. .remove = __devexit_p(i82443bxgx_edacmc_remove_one),
  320. .id_table = i82443bxgx_pci_tbl,
  321. };
  322. static int __init i82443bxgx_edacmc_init(void)
  323. {
  324. return pci_register_driver(&i82443bxgx_edacmc_driver);
  325. }
  326. static void __exit i82443bxgx_edacmc_exit(void)
  327. {
  328. pci_unregister_driver(&i82443bxgx_edacmc_driver);
  329. }
  330. module_init(i82443bxgx_edacmc_init);
  331. module_exit(i82443bxgx_edacmc_exit);
  332. MODULE_LICENSE("GPL");
  333. MODULE_AUTHOR("Tim Small <tim@buttersideup.com> - WPAD");
  334. MODULE_DESCRIPTION("EDAC MC support for Intel 82443BX/GX memory controllers");