amd76x_edac.c 9.1 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367
  1. /*
  2. * AMD 76x Memory Controller kernel module
  3. * (C) 2003 Linux Networx (http://lnxi.com)
  4. * This file may be distributed under the terms of the
  5. * GNU General Public License.
  6. *
  7. * Written by Thayne Harbaugh
  8. * Based on work by Dan Hollis <goemon at anime dot net> and others.
  9. * http://www.anime.net/~goemon/linux-ecc/
  10. *
  11. * $Id: edac_amd76x.c,v 1.4.2.5 2005/10/05 00:43:44 dsp_llnl Exp $
  12. *
  13. */
  14. #include <linux/module.h>
  15. #include <linux/init.h>
  16. #include <linux/pci.h>
  17. #include <linux/pci_ids.h>
  18. #include <linux/edac.h>
  19. #include "edac_core.h"
  20. #define AMD76X_REVISION " Ver: 2.0.2"
  21. #define EDAC_MOD_STR "amd76x_edac"
  22. #define amd76x_printk(level, fmt, arg...) \
  23. edac_printk(level, "amd76x", fmt, ##arg)
  24. #define amd76x_mc_printk(mci, level, fmt, arg...) \
  25. edac_mc_chipset_printk(mci, level, "amd76x", fmt, ##arg)
  26. #define AMD76X_NR_CSROWS 8
  27. #define AMD76X_NR_CHANS 1
  28. #define AMD76X_NR_DIMMS 4
  29. /* AMD 76x register addresses - device 0 function 0 - PCI bridge */
  30. #define AMD76X_ECC_MODE_STATUS 0x48 /* Mode and status of ECC (32b)
  31. *
  32. * 31:16 reserved
  33. * 15:14 SERR enabled: x1=ue 1x=ce
  34. * 13 reserved
  35. * 12 diag: disabled, enabled
  36. * 11:10 mode: dis, EC, ECC, ECC+scrub
  37. * 9:8 status: x1=ue 1x=ce
  38. * 7:4 UE cs row
  39. * 3:0 CE cs row
  40. */
  41. #define AMD76X_DRAM_MODE_STATUS 0x58 /* DRAM Mode and status (32b)
  42. *
  43. * 31:26 clock disable 5 - 0
  44. * 25 SDRAM init
  45. * 24 reserved
  46. * 23 mode register service
  47. * 22:21 suspend to RAM
  48. * 20 burst refresh enable
  49. * 19 refresh disable
  50. * 18 reserved
  51. * 17:16 cycles-per-refresh
  52. * 15:8 reserved
  53. * 7:0 x4 mode enable 7 - 0
  54. */
  55. #define AMD76X_MEM_BASE_ADDR 0xC0 /* Memory base address (8 x 32b)
  56. *
  57. * 31:23 chip-select base
  58. * 22:16 reserved
  59. * 15:7 chip-select mask
  60. * 6:3 reserved
  61. * 2:1 address mode
  62. * 0 chip-select enable
  63. */
  64. struct amd76x_error_info {
  65. u32 ecc_mode_status;
  66. };
  67. enum amd76x_chips {
  68. AMD761 = 0,
  69. AMD762
  70. };
  71. struct amd76x_dev_info {
  72. const char *ctl_name;
  73. };
  74. static const struct amd76x_dev_info amd76x_devs[] = {
  75. [AMD761] = {
  76. .ctl_name = "AMD761"},
  77. [AMD762] = {
  78. .ctl_name = "AMD762"},
  79. };
  80. static struct edac_pci_ctl_info *amd76x_pci;
  81. /**
  82. * amd76x_get_error_info - fetch error information
  83. * @mci: Memory controller
  84. * @info: Info to fill in
  85. *
  86. * Fetch and store the AMD76x ECC status. Clear pending status
  87. * on the chip so that further errors will be reported
  88. */
  89. static void amd76x_get_error_info(struct mem_ctl_info *mci,
  90. struct amd76x_error_info *info)
  91. {
  92. struct pci_dev *pdev;
  93. pdev = to_pci_dev(mci->dev);
  94. pci_read_config_dword(pdev, AMD76X_ECC_MODE_STATUS,
  95. &info->ecc_mode_status);
  96. if (info->ecc_mode_status & BIT(8))
  97. pci_write_bits32(pdev, AMD76X_ECC_MODE_STATUS,
  98. (u32) BIT(8), (u32) BIT(8));
  99. if (info->ecc_mode_status & BIT(9))
  100. pci_write_bits32(pdev, AMD76X_ECC_MODE_STATUS,
  101. (u32) BIT(9), (u32) BIT(9));
  102. }
  103. /**
  104. * amd76x_process_error_info - Error check
  105. * @mci: Memory controller
  106. * @info: Previously fetched information from chip
  107. * @handle_errors: 1 if we should do recovery
  108. *
  109. * Process the chip state and decide if an error has occurred.
  110. * A return of 1 indicates an error. Also if handle_errors is true
  111. * then attempt to handle and clean up after the error
  112. */
  113. static int amd76x_process_error_info(struct mem_ctl_info *mci,
  114. struct amd76x_error_info *info,
  115. int handle_errors)
  116. {
  117. int error_found;
  118. u32 row;
  119. error_found = 0;
  120. /*
  121. * Check for an uncorrectable error
  122. */
  123. if (info->ecc_mode_status & BIT(8)) {
  124. error_found = 1;
  125. if (handle_errors) {
  126. row = (info->ecc_mode_status >> 4) & 0xf;
  127. edac_mc_handle_ue(mci, mci->csrows[row].first_page, 0,
  128. row, mci->ctl_name);
  129. }
  130. }
  131. /*
  132. * Check for a correctable error
  133. */
  134. if (info->ecc_mode_status & BIT(9)) {
  135. error_found = 1;
  136. if (handle_errors) {
  137. row = info->ecc_mode_status & 0xf;
  138. edac_mc_handle_ce(mci, mci->csrows[row].first_page, 0,
  139. 0, row, 0, mci->ctl_name);
  140. }
  141. }
  142. return error_found;
  143. }
  144. /**
  145. * amd76x_check - Poll the controller
  146. * @mci: Memory controller
  147. *
  148. * Called by the poll handlers this function reads the status
  149. * from the controller and checks for errors.
  150. */
  151. static void amd76x_check(struct mem_ctl_info *mci)
  152. {
  153. struct amd76x_error_info info;
  154. debugf3("%s()\n", __func__);
  155. amd76x_get_error_info(mci, &info);
  156. amd76x_process_error_info(mci, &info, 1);
  157. }
  158. static void amd76x_init_csrows(struct mem_ctl_info *mci, struct pci_dev *pdev,
  159. enum edac_type edac_mode)
  160. {
  161. struct csrow_info *csrow;
  162. u32 mba, mba_base, mba_mask, dms;
  163. int index;
  164. for (index = 0; index < mci->nr_csrows; index++) {
  165. csrow = &mci->csrows[index];
  166. /* find the DRAM Chip Select Base address and mask */
  167. pci_read_config_dword(pdev,
  168. AMD76X_MEM_BASE_ADDR + (index * 4), &mba);
  169. if (!(mba & BIT(0)))
  170. continue;
  171. mba_base = mba & 0xff800000UL;
  172. mba_mask = ((mba & 0xff80) << 16) | 0x7fffffUL;
  173. pci_read_config_dword(pdev, AMD76X_DRAM_MODE_STATUS, &dms);
  174. csrow->first_page = mba_base >> PAGE_SHIFT;
  175. csrow->nr_pages = (mba_mask + 1) >> PAGE_SHIFT;
  176. csrow->last_page = csrow->first_page + csrow->nr_pages - 1;
  177. csrow->page_mask = mba_mask >> PAGE_SHIFT;
  178. csrow->grain = csrow->nr_pages << PAGE_SHIFT;
  179. csrow->mtype = MEM_RDDR;
  180. csrow->dtype = ((dms >> index) & 0x1) ? DEV_X4 : DEV_UNKNOWN;
  181. csrow->edac_mode = edac_mode;
  182. }
  183. }
  184. /**
  185. * amd76x_probe1 - Perform set up for detected device
  186. * @pdev; PCI device detected
  187. * @dev_idx: Device type index
  188. *
  189. * We have found an AMD76x and now need to set up the memory
  190. * controller status reporting. We configure and set up the
  191. * memory controller reporting and claim the device.
  192. */
  193. static int amd76x_probe1(struct pci_dev *pdev, int dev_idx)
  194. {
  195. static const enum edac_type ems_modes[] = {
  196. EDAC_NONE,
  197. EDAC_EC,
  198. EDAC_SECDED,
  199. EDAC_SECDED
  200. };
  201. struct mem_ctl_info *mci = NULL;
  202. u32 ems;
  203. u32 ems_mode;
  204. struct amd76x_error_info discard;
  205. debugf0("%s()\n", __func__);
  206. pci_read_config_dword(pdev, AMD76X_ECC_MODE_STATUS, &ems);
  207. ems_mode = (ems >> 10) & 0x3;
  208. mci = edac_mc_alloc(0, AMD76X_NR_CSROWS, AMD76X_NR_CHANS, 0);
  209. if (mci == NULL) {
  210. return -ENOMEM;
  211. }
  212. debugf0("%s(): mci = %p\n", __func__, mci);
  213. mci->dev = &pdev->dev;
  214. mci->mtype_cap = MEM_FLAG_RDDR;
  215. mci->edac_ctl_cap = EDAC_FLAG_NONE | EDAC_FLAG_EC | EDAC_FLAG_SECDED;
  216. mci->edac_cap = ems_mode ?
  217. (EDAC_FLAG_EC | EDAC_FLAG_SECDED) : EDAC_FLAG_NONE;
  218. mci->mod_name = EDAC_MOD_STR;
  219. mci->mod_ver = AMD76X_REVISION;
  220. mci->ctl_name = amd76x_devs[dev_idx].ctl_name;
  221. mci->dev_name = pci_name(pdev);
  222. mci->edac_check = amd76x_check;
  223. mci->ctl_page_to_phys = NULL;
  224. amd76x_init_csrows(mci, pdev, ems_modes[ems_mode]);
  225. amd76x_get_error_info(mci, &discard); /* clear counters */
  226. /* Here we assume that we will never see multiple instances of this
  227. * type of memory controller. The ID is therefore hardcoded to 0.
  228. */
  229. if (edac_mc_add_mc(mci)) {
  230. debugf3("%s(): failed edac_mc_add_mc()\n", __func__);
  231. goto fail;
  232. }
  233. /* allocating generic PCI control info */
  234. amd76x_pci = edac_pci_create_generic_ctl(&pdev->dev, EDAC_MOD_STR);
  235. if (!amd76x_pci) {
  236. printk(KERN_WARNING
  237. "%s(): Unable to create PCI control\n",
  238. __func__);
  239. printk(KERN_WARNING
  240. "%s(): PCI error report via EDAC not setup\n",
  241. __func__);
  242. }
  243. /* get this far and it's successful */
  244. debugf3("%s(): success\n", __func__);
  245. return 0;
  246. fail:
  247. edac_mc_free(mci);
  248. return -ENODEV;
  249. }
  250. /* returns count (>= 0), or negative on error */
  251. static int __devinit amd76x_init_one(struct pci_dev *pdev,
  252. const struct pci_device_id *ent)
  253. {
  254. debugf0("%s()\n", __func__);
  255. /* don't need to call pci_enable_device() */
  256. return amd76x_probe1(pdev, ent->driver_data);
  257. }
  258. /**
  259. * amd76x_remove_one - driver shutdown
  260. * @pdev: PCI device being handed back
  261. *
  262. * Called when the driver is unloaded. Find the matching mci
  263. * structure for the device then delete the mci and free the
  264. * resources.
  265. */
  266. static void __devexit amd76x_remove_one(struct pci_dev *pdev)
  267. {
  268. struct mem_ctl_info *mci;
  269. debugf0("%s()\n", __func__);
  270. if (amd76x_pci)
  271. edac_pci_release_generic_ctl(amd76x_pci);
  272. if ((mci = edac_mc_del_mc(&pdev->dev)) == NULL)
  273. return;
  274. edac_mc_free(mci);
  275. }
  276. static const struct pci_device_id amd76x_pci_tbl[] __devinitdata = {
  277. {
  278. PCI_VEND_DEV(AMD, FE_GATE_700C), PCI_ANY_ID, PCI_ANY_ID, 0, 0,
  279. AMD762},
  280. {
  281. PCI_VEND_DEV(AMD, FE_GATE_700E), PCI_ANY_ID, PCI_ANY_ID, 0, 0,
  282. AMD761},
  283. {
  284. 0,
  285. } /* 0 terminated list. */
  286. };
  287. MODULE_DEVICE_TABLE(pci, amd76x_pci_tbl);
  288. static struct pci_driver amd76x_driver = {
  289. .name = EDAC_MOD_STR,
  290. .probe = amd76x_init_one,
  291. .remove = __devexit_p(amd76x_remove_one),
  292. .id_table = amd76x_pci_tbl,
  293. };
  294. static int __init amd76x_init(void)
  295. {
  296. /* Ensure that the OPSTATE is set correctly for POLL or NMI */
  297. opstate_init();
  298. return pci_register_driver(&amd76x_driver);
  299. }
  300. static void __exit amd76x_exit(void)
  301. {
  302. pci_unregister_driver(&amd76x_driver);
  303. }
  304. module_init(amd76x_init);
  305. module_exit(amd76x_exit);
  306. MODULE_LICENSE("GPL");
  307. MODULE_AUTHOR("Linux Networx (http://lnxi.com) Thayne Harbaugh");
  308. MODULE_DESCRIPTION("MC support for AMD 76x memory controllers");
  309. module_param(edac_op_state, int, 0444);
  310. MODULE_PARM_DESC(edac_op_state, "EDAC Error Reporting state: 0=Poll,1=NMI");