amd64_edac.c 73 KB

1234567891011121314151617181920212223242526272829303132333435363738394041424344454647484950515253545556575859606162636465666768697071727374757677787980818283848586878889909192939495969798991001011021031041051061071081091101111121131141151161171181191201211221231241251261271281291301311321331341351361371381391401411421431441451461471481491501511521531541551561571581591601611621631641651661671681691701711721731741751761771781791801811821831841851861871881891901911921931941951961971981992002012022032042052062072082092102112122132142152162172182192202212222232242252262272282292302312322332342352362372382392402412422432442452462472482492502512522532542552562572582592602612622632642652662672682692702712722732742752762772782792802812822832842852862872882892902912922932942952962972982993003013023033043053063073083093103113123133143153163173183193203213223233243253263273283293303313323333343353363373383393403413423433443453463473483493503513523533543553563573583593603613623633643653663673683693703713723733743753763773783793803813823833843853863873883893903913923933943953963973983994004014024034044054064074084094104114124134144154164174184194204214224234244254264274284294304314324334344354364374384394404414424434444454464474484494504514524534544554564574584594604614624634644654664674684694704714724734744754764774784794804814824834844854864874884894904914924934944954964974984995005015025035045055065075085095105115125135145155165175185195205215225235245255265275285295305315325335345355365375385395405415425435445455465475485495505515525535545555565575585595605615625635645655665675685695705715725735745755765775785795805815825835845855865875885895905915925935945955965975985996006016026036046056066076086096106116126136146156166176186196206216226236246256266276286296306316326336346356366376386396406416426436446456466476486496506516526536546556566576586596606616626636646656666676686696706716726736746756766776786796806816826836846856866876886896906916926936946956966976986997007017027037047057067077087097107117127137147157167177187197207217227237247257267277287297307317327337347357367377387397407417427437447457467477487497507517527537547557567577587597607617627637647657667677687697707717727737747757767777787797807817827837847857867877887897907917927937947957967977987998008018028038048058068078088098108118128138148158168178188198208218228238248258268278288298308318328338348358368378388398408418428438448458468478488498508518528538548558568578588598608618628638648658668678688698708718728738748758768778788798808818828838848858868878888898908918928938948958968978988999009019029039049059069079089099109119129139149159169179189199209219229239249259269279289299309319329339349359369379389399409419429439449459469479489499509519529539549559569579589599609619629639649659669679689699709719729739749759769779789799809819829839849859869879889899909919929939949959969979989991000100110021003100410051006100710081009101010111012101310141015101610171018101910201021102210231024102510261027102810291030103110321033103410351036103710381039104010411042104310441045104610471048104910501051105210531054105510561057105810591060106110621063106410651066106710681069107010711072107310741075107610771078107910801081108210831084108510861087108810891090109110921093109410951096109710981099110011011102110311041105110611071108110911101111111211131114111511161117111811191120112111221123112411251126112711281129113011311132113311341135113611371138113911401141114211431144114511461147114811491150115111521153115411551156115711581159116011611162116311641165116611671168116911701171117211731174117511761177117811791180118111821183118411851186118711881189119011911192119311941195119611971198119912001201120212031204120512061207120812091210121112121213121412151216121712181219122012211222122312241225122612271228122912301231123212331234123512361237123812391240124112421243124412451246124712481249125012511252125312541255125612571258125912601261126212631264126512661267126812691270127112721273127412751276127712781279128012811282128312841285128612871288128912901291129212931294129512961297129812991300130113021303130413051306130713081309131013111312131313141315131613171318131913201321132213231324132513261327132813291330133113321333133413351336133713381339134013411342134313441345134613471348134913501351135213531354135513561357135813591360136113621363136413651366136713681369137013711372137313741375137613771378137913801381138213831384138513861387138813891390139113921393139413951396139713981399140014011402140314041405140614071408140914101411141214131414141514161417141814191420142114221423142414251426142714281429143014311432143314341435143614371438143914401441144214431444144514461447144814491450145114521453145414551456145714581459146014611462146314641465146614671468146914701471147214731474147514761477147814791480148114821483148414851486148714881489149014911492149314941495149614971498149915001501150215031504150515061507150815091510151115121513151415151516151715181519152015211522152315241525152615271528152915301531153215331534153515361537153815391540154115421543154415451546154715481549155015511552155315541555155615571558155915601561156215631564156515661567156815691570157115721573157415751576157715781579158015811582158315841585158615871588158915901591159215931594159515961597159815991600160116021603160416051606160716081609161016111612161316141615161616171618161916201621162216231624162516261627162816291630163116321633163416351636163716381639164016411642164316441645164616471648164916501651165216531654165516561657165816591660166116621663166416651666166716681669167016711672167316741675167616771678167916801681168216831684168516861687168816891690169116921693169416951696169716981699170017011702170317041705170617071708170917101711171217131714171517161717171817191720172117221723172417251726172717281729173017311732173317341735173617371738173917401741174217431744174517461747174817491750175117521753175417551756175717581759176017611762176317641765176617671768176917701771177217731774177517761777177817791780178117821783178417851786178717881789179017911792179317941795179617971798179918001801180218031804180518061807180818091810181118121813181418151816181718181819182018211822182318241825182618271828182918301831183218331834183518361837183818391840184118421843184418451846184718481849185018511852185318541855185618571858185918601861186218631864186518661867186818691870187118721873187418751876187718781879188018811882188318841885188618871888188918901891189218931894189518961897189818991900190119021903190419051906190719081909191019111912191319141915191619171918191919201921192219231924192519261927192819291930193119321933193419351936193719381939194019411942194319441945194619471948194919501951195219531954195519561957195819591960196119621963196419651966196719681969197019711972197319741975197619771978197919801981198219831984198519861987198819891990199119921993199419951996199719981999200020012002200320042005200620072008200920102011201220132014201520162017201820192020202120222023202420252026202720282029203020312032203320342035203620372038203920402041204220432044204520462047204820492050205120522053205420552056205720582059206020612062206320642065206620672068206920702071207220732074207520762077207820792080208120822083208420852086208720882089209020912092209320942095209620972098209921002101210221032104210521062107210821092110211121122113211421152116211721182119212021212122212321242125212621272128212921302131213221332134213521362137213821392140214121422143214421452146214721482149215021512152215321542155215621572158215921602161216221632164216521662167216821692170217121722173217421752176217721782179218021812182218321842185218621872188218921902191219221932194219521962197219821992200220122022203220422052206220722082209221022112212221322142215221622172218221922202221222222232224222522262227222822292230223122322233223422352236223722382239224022412242224322442245224622472248224922502251225222532254225522562257225822592260226122622263226422652266226722682269227022712272227322742275227622772278227922802281228222832284228522862287228822892290229122922293229422952296229722982299230023012302230323042305230623072308230923102311231223132314231523162317231823192320232123222323232423252326232723282329233023312332233323342335233623372338233923402341234223432344234523462347234823492350235123522353235423552356235723582359236023612362236323642365236623672368236923702371237223732374237523762377237823792380238123822383238423852386238723882389239023912392239323942395239623972398239924002401240224032404240524062407240824092410241124122413241424152416241724182419242024212422242324242425242624272428242924302431243224332434243524362437243824392440244124422443244424452446244724482449245024512452245324542455245624572458
  1. #include "amd64_edac.h"
  2. static struct edac_pci_ctl_info *amd64_ctl_pci;
  3. static int report_gart_errors;
  4. module_param(report_gart_errors, int, 0644);
  5. /*
  6. * Set by command line parameter. If BIOS has enabled the ECC, this override is
  7. * cleared to prevent re-enabling the hardware by this driver.
  8. */
  9. static int ecc_enable_override;
  10. module_param(ecc_enable_override, int, 0644);
  11. /* Lookup table for all possible MC control instances */
  12. struct amd64_pvt;
  13. static struct mem_ctl_info *mci_lookup[MAX_NUMNODES];
  14. static struct amd64_pvt *pvt_lookup[MAX_NUMNODES];
  15. /*
  16. * Memory scrubber control interface. For K8, memory scrubbing is handled by
  17. * hardware and can involve L2 cache, dcache as well as the main memory. With
  18. * F10, this is extended to L3 cache scrubbing on CPU models sporting that
  19. * functionality.
  20. *
  21. * This causes the "units" for the scrubbing speed to vary from 64 byte blocks
  22. * (dram) over to cache lines. This is nasty, so we will use bandwidth in
  23. * bytes/sec for the setting.
  24. *
  25. * Currently, we only do dram scrubbing. If the scrubbing is done in software on
  26. * other archs, we might not have access to the caches directly.
  27. */
  28. /*
  29. * scan the scrub rate mapping table for a close or matching bandwidth value to
  30. * issue. If requested is too big, then use last maximum value found.
  31. */
  32. static int amd64_search_set_scrub_rate(struct pci_dev *ctl, u32 new_bw,
  33. u32 min_scrubrate)
  34. {
  35. u32 scrubval;
  36. int i;
  37. /*
  38. * map the configured rate (new_bw) to a value specific to the AMD64
  39. * memory controller and apply to register. Search for the first
  40. * bandwidth entry that is greater or equal than the setting requested
  41. * and program that. If at last entry, turn off DRAM scrubbing.
  42. */
  43. for (i = 0; i < ARRAY_SIZE(scrubrates); i++) {
  44. /*
  45. * skip scrub rates which aren't recommended
  46. * (see F10 BKDG, F3x58)
  47. */
  48. if (scrubrates[i].scrubval < min_scrubrate)
  49. continue;
  50. if (scrubrates[i].bandwidth <= new_bw)
  51. break;
  52. /*
  53. * if no suitable bandwidth found, turn off DRAM scrubbing
  54. * entirely by falling back to the last element in the
  55. * scrubrates array.
  56. */
  57. }
  58. scrubval = scrubrates[i].scrubval;
  59. if (scrubval)
  60. edac_printk(KERN_DEBUG, EDAC_MC,
  61. "Setting scrub rate bandwidth: %u\n",
  62. scrubrates[i].bandwidth);
  63. else
  64. edac_printk(KERN_DEBUG, EDAC_MC, "Turning scrubbing off.\n");
  65. pci_write_bits32(ctl, K8_SCRCTRL, scrubval, 0x001F);
  66. return 0;
  67. }
  68. static int amd64_set_scrub_rate(struct mem_ctl_info *mci, u32 *bandwidth)
  69. {
  70. struct amd64_pvt *pvt = mci->pvt_info;
  71. u32 min_scrubrate = 0x0;
  72. switch (boot_cpu_data.x86) {
  73. case 0xf:
  74. min_scrubrate = K8_MIN_SCRUB_RATE_BITS;
  75. break;
  76. case 0x10:
  77. min_scrubrate = F10_MIN_SCRUB_RATE_BITS;
  78. break;
  79. case 0x11:
  80. min_scrubrate = F11_MIN_SCRUB_RATE_BITS;
  81. break;
  82. default:
  83. amd64_printk(KERN_ERR, "Unsupported family!\n");
  84. break;
  85. }
  86. return amd64_search_set_scrub_rate(pvt->misc_f3_ctl, *bandwidth,
  87. min_scrubrate);
  88. }
  89. static int amd64_get_scrub_rate(struct mem_ctl_info *mci, u32 *bw)
  90. {
  91. struct amd64_pvt *pvt = mci->pvt_info;
  92. u32 scrubval = 0;
  93. int status = -1, i, ret = 0;
  94. ret = pci_read_config_dword(pvt->misc_f3_ctl, K8_SCRCTRL, &scrubval);
  95. if (ret)
  96. debugf0("Reading K8_SCRCTRL failed\n");
  97. scrubval = scrubval & 0x001F;
  98. edac_printk(KERN_DEBUG, EDAC_MC,
  99. "pci-read, sdram scrub control value: %d \n", scrubval);
  100. for (i = 0; ARRAY_SIZE(scrubrates); i++) {
  101. if (scrubrates[i].scrubval == scrubval) {
  102. *bw = scrubrates[i].bandwidth;
  103. status = 0;
  104. break;
  105. }
  106. }
  107. return status;
  108. }
  109. /* Map from a CSROW entry to the mask entry that operates on it */
  110. static inline u32 amd64_map_to_dcs_mask(struct amd64_pvt *pvt, int csrow)
  111. {
  112. return csrow >> (pvt->num_dcsm >> 3);
  113. }
  114. /* return the 'base' address the i'th CS entry of the 'dct' DRAM controller */
  115. static u32 amd64_get_dct_base(struct amd64_pvt *pvt, int dct, int csrow)
  116. {
  117. if (dct == 0)
  118. return pvt->dcsb0[csrow];
  119. else
  120. return pvt->dcsb1[csrow];
  121. }
  122. /*
  123. * Return the 'mask' address the i'th CS entry. This function is needed because
  124. * there number of DCSM registers on Rev E and prior vs Rev F and later is
  125. * different.
  126. */
  127. static u32 amd64_get_dct_mask(struct amd64_pvt *pvt, int dct, int csrow)
  128. {
  129. if (dct == 0)
  130. return pvt->dcsm0[amd64_map_to_dcs_mask(pvt, csrow)];
  131. else
  132. return pvt->dcsm1[amd64_map_to_dcs_mask(pvt, csrow)];
  133. }
  134. /*
  135. * In *base and *limit, pass back the full 40-bit base and limit physical
  136. * addresses for the node given by node_id. This information is obtained from
  137. * DRAM Base (section 3.4.4.1) and DRAM Limit (section 3.4.4.2) registers. The
  138. * base and limit addresses are of type SysAddr, as defined at the start of
  139. * section 3.4.4 (p. 70). They are the lowest and highest physical addresses
  140. * in the address range they represent.
  141. */
  142. static void amd64_get_base_and_limit(struct amd64_pvt *pvt, int node_id,
  143. u64 *base, u64 *limit)
  144. {
  145. *base = pvt->dram_base[node_id];
  146. *limit = pvt->dram_limit[node_id];
  147. }
  148. /*
  149. * Return 1 if the SysAddr given by sys_addr matches the base/limit associated
  150. * with node_id
  151. */
  152. static int amd64_base_limit_match(struct amd64_pvt *pvt,
  153. u64 sys_addr, int node_id)
  154. {
  155. u64 base, limit, addr;
  156. amd64_get_base_and_limit(pvt, node_id, &base, &limit);
  157. /* The K8 treats this as a 40-bit value. However, bits 63-40 will be
  158. * all ones if the most significant implemented address bit is 1.
  159. * Here we discard bits 63-40. See section 3.4.2 of AMD publication
  160. * 24592: AMD x86-64 Architecture Programmer's Manual Volume 1
  161. * Application Programming.
  162. */
  163. addr = sys_addr & 0x000000ffffffffffull;
  164. return (addr >= base) && (addr <= limit);
  165. }
  166. /*
  167. * Attempt to map a SysAddr to a node. On success, return a pointer to the
  168. * mem_ctl_info structure for the node that the SysAddr maps to.
  169. *
  170. * On failure, return NULL.
  171. */
  172. static struct mem_ctl_info *find_mc_by_sys_addr(struct mem_ctl_info *mci,
  173. u64 sys_addr)
  174. {
  175. struct amd64_pvt *pvt;
  176. int node_id;
  177. u32 intlv_en, bits;
  178. /*
  179. * Here we use the DRAM Base (section 3.4.4.1) and DRAM Limit (section
  180. * 3.4.4.2) registers to map the SysAddr to a node ID.
  181. */
  182. pvt = mci->pvt_info;
  183. /*
  184. * The value of this field should be the same for all DRAM Base
  185. * registers. Therefore we arbitrarily choose to read it from the
  186. * register for node 0.
  187. */
  188. intlv_en = pvt->dram_IntlvEn[0];
  189. if (intlv_en == 0) {
  190. for (node_id = 0; ; ) {
  191. if (amd64_base_limit_match(pvt, sys_addr, node_id))
  192. break;
  193. if (++node_id >= DRAM_REG_COUNT)
  194. goto err_no_match;
  195. }
  196. goto found;
  197. }
  198. if (unlikely((intlv_en != (0x01 << 8)) &&
  199. (intlv_en != (0x03 << 8)) &&
  200. (intlv_en != (0x07 << 8)))) {
  201. amd64_printk(KERN_WARNING, "junk value of 0x%x extracted from "
  202. "IntlvEn field of DRAM Base Register for node 0: "
  203. "This probably indicates a BIOS bug.\n", intlv_en);
  204. return NULL;
  205. }
  206. bits = (((u32) sys_addr) >> 12) & intlv_en;
  207. for (node_id = 0; ; ) {
  208. if ((pvt->dram_limit[node_id] & intlv_en) == bits)
  209. break; /* intlv_sel field matches */
  210. if (++node_id >= DRAM_REG_COUNT)
  211. goto err_no_match;
  212. }
  213. /* sanity test for sys_addr */
  214. if (unlikely(!amd64_base_limit_match(pvt, sys_addr, node_id))) {
  215. amd64_printk(KERN_WARNING,
  216. "%s(): sys_addr 0x%lx falls outside base/limit "
  217. "address range for node %d with node interleaving "
  218. "enabled.\n", __func__, (unsigned long)sys_addr,
  219. node_id);
  220. return NULL;
  221. }
  222. found:
  223. return edac_mc_find(node_id);
  224. err_no_match:
  225. debugf2("sys_addr 0x%lx doesn't match any node\n",
  226. (unsigned long)sys_addr);
  227. return NULL;
  228. }
  229. /*
  230. * Extract the DRAM CS base address from selected csrow register.
  231. */
  232. static u64 base_from_dct_base(struct amd64_pvt *pvt, int csrow)
  233. {
  234. return ((u64) (amd64_get_dct_base(pvt, 0, csrow) & pvt->dcsb_base)) <<
  235. pvt->dcs_shift;
  236. }
  237. /*
  238. * Extract the mask from the dcsb0[csrow] entry in a CPU revision-specific way.
  239. */
  240. static u64 mask_from_dct_mask(struct amd64_pvt *pvt, int csrow)
  241. {
  242. u64 dcsm_bits, other_bits;
  243. u64 mask;
  244. /* Extract bits from DRAM CS Mask. */
  245. dcsm_bits = amd64_get_dct_mask(pvt, 0, csrow) & pvt->dcsm_mask;
  246. other_bits = pvt->dcsm_mask;
  247. other_bits = ~(other_bits << pvt->dcs_shift);
  248. /*
  249. * The extracted bits from DCSM belong in the spaces represented by
  250. * the cleared bits in other_bits.
  251. */
  252. mask = (dcsm_bits << pvt->dcs_shift) | other_bits;
  253. return mask;
  254. }
  255. /*
  256. * @input_addr is an InputAddr associated with the node given by mci. Return the
  257. * csrow that input_addr maps to, or -1 on failure (no csrow claims input_addr).
  258. */
  259. static int input_addr_to_csrow(struct mem_ctl_info *mci, u64 input_addr)
  260. {
  261. struct amd64_pvt *pvt;
  262. int csrow;
  263. u64 base, mask;
  264. pvt = mci->pvt_info;
  265. /*
  266. * Here we use the DRAM CS Base and DRAM CS Mask registers. For each CS
  267. * base/mask register pair, test the condition shown near the start of
  268. * section 3.5.4 (p. 84, BKDG #26094, K8, revA-E).
  269. */
  270. for (csrow = 0; csrow < CHIPSELECT_COUNT; csrow++) {
  271. /* This DRAM chip select is disabled on this node */
  272. if ((pvt->dcsb0[csrow] & K8_DCSB_CS_ENABLE) == 0)
  273. continue;
  274. base = base_from_dct_base(pvt, csrow);
  275. mask = ~mask_from_dct_mask(pvt, csrow);
  276. if ((input_addr & mask) == (base & mask)) {
  277. debugf2("InputAddr 0x%lx matches csrow %d (node %d)\n",
  278. (unsigned long)input_addr, csrow,
  279. pvt->mc_node_id);
  280. return csrow;
  281. }
  282. }
  283. debugf2("no matching csrow for InputAddr 0x%lx (MC node %d)\n",
  284. (unsigned long)input_addr, pvt->mc_node_id);
  285. return -1;
  286. }
  287. /*
  288. * Return the base value defined by the DRAM Base register for the node
  289. * represented by mci. This function returns the full 40-bit value despite the
  290. * fact that the register only stores bits 39-24 of the value. See section
  291. * 3.4.4.1 (BKDG #26094, K8, revA-E)
  292. */
  293. static inline u64 get_dram_base(struct mem_ctl_info *mci)
  294. {
  295. struct amd64_pvt *pvt = mci->pvt_info;
  296. return pvt->dram_base[pvt->mc_node_id];
  297. }
  298. /*
  299. * Obtain info from the DRAM Hole Address Register (section 3.4.8, pub #26094)
  300. * for the node represented by mci. Info is passed back in *hole_base,
  301. * *hole_offset, and *hole_size. Function returns 0 if info is valid or 1 if
  302. * info is invalid. Info may be invalid for either of the following reasons:
  303. *
  304. * - The revision of the node is not E or greater. In this case, the DRAM Hole
  305. * Address Register does not exist.
  306. *
  307. * - The DramHoleValid bit is cleared in the DRAM Hole Address Register,
  308. * indicating that its contents are not valid.
  309. *
  310. * The values passed back in *hole_base, *hole_offset, and *hole_size are
  311. * complete 32-bit values despite the fact that the bitfields in the DHAR
  312. * only represent bits 31-24 of the base and offset values.
  313. */
  314. int amd64_get_dram_hole_info(struct mem_ctl_info *mci, u64 *hole_base,
  315. u64 *hole_offset, u64 *hole_size)
  316. {
  317. struct amd64_pvt *pvt = mci->pvt_info;
  318. u64 base;
  319. /* only revE and later have the DRAM Hole Address Register */
  320. if (boot_cpu_data.x86 == 0xf && pvt->ext_model < OPTERON_CPU_REV_E) {
  321. debugf1(" revision %d for node %d does not support DHAR\n",
  322. pvt->ext_model, pvt->mc_node_id);
  323. return 1;
  324. }
  325. /* only valid for Fam10h */
  326. if (boot_cpu_data.x86 == 0x10 &&
  327. (pvt->dhar & F10_DRAM_MEM_HOIST_VALID) == 0) {
  328. debugf1(" Dram Memory Hoisting is DISABLED on this system\n");
  329. return 1;
  330. }
  331. if ((pvt->dhar & DHAR_VALID) == 0) {
  332. debugf1(" Dram Memory Hoisting is DISABLED on this node %d\n",
  333. pvt->mc_node_id);
  334. return 1;
  335. }
  336. /* This node has Memory Hoisting */
  337. /* +------------------+--------------------+--------------------+-----
  338. * | memory | DRAM hole | relocated |
  339. * | [0, (x - 1)] | [x, 0xffffffff] | addresses from |
  340. * | | | DRAM hole |
  341. * | | | [0x100000000, |
  342. * | | | (0x100000000+ |
  343. * | | | (0xffffffff-x))] |
  344. * +------------------+--------------------+--------------------+-----
  345. *
  346. * Above is a diagram of physical memory showing the DRAM hole and the
  347. * relocated addresses from the DRAM hole. As shown, the DRAM hole
  348. * starts at address x (the base address) and extends through address
  349. * 0xffffffff. The DRAM Hole Address Register (DHAR) relocates the
  350. * addresses in the hole so that they start at 0x100000000.
  351. */
  352. base = dhar_base(pvt->dhar);
  353. *hole_base = base;
  354. *hole_size = (0x1ull << 32) - base;
  355. if (boot_cpu_data.x86 > 0xf)
  356. *hole_offset = f10_dhar_offset(pvt->dhar);
  357. else
  358. *hole_offset = k8_dhar_offset(pvt->dhar);
  359. debugf1(" DHAR info for node %d base 0x%lx offset 0x%lx size 0x%lx\n",
  360. pvt->mc_node_id, (unsigned long)*hole_base,
  361. (unsigned long)*hole_offset, (unsigned long)*hole_size);
  362. return 0;
  363. }
  364. EXPORT_SYMBOL_GPL(amd64_get_dram_hole_info);
  365. /*
  366. * Return the DramAddr that the SysAddr given by @sys_addr maps to. It is
  367. * assumed that sys_addr maps to the node given by mci.
  368. *
  369. * The first part of section 3.4.4 (p. 70) shows how the DRAM Base (section
  370. * 3.4.4.1) and DRAM Limit (section 3.4.4.2) registers are used to translate a
  371. * SysAddr to a DramAddr. If the DRAM Hole Address Register (DHAR) is enabled,
  372. * then it is also involved in translating a SysAddr to a DramAddr. Sections
  373. * 3.4.8 and 3.5.8.2 describe the DHAR and how it is used for memory hoisting.
  374. * These parts of the documentation are unclear. I interpret them as follows:
  375. *
  376. * When node n receives a SysAddr, it processes the SysAddr as follows:
  377. *
  378. * 1. It extracts the DRAMBase and DRAMLimit values from the DRAM Base and DRAM
  379. * Limit registers for node n. If the SysAddr is not within the range
  380. * specified by the base and limit values, then node n ignores the Sysaddr
  381. * (since it does not map to node n). Otherwise continue to step 2 below.
  382. *
  383. * 2. If the DramHoleValid bit of the DHAR for node n is clear, the DHAR is
  384. * disabled so skip to step 3 below. Otherwise see if the SysAddr is within
  385. * the range of relocated addresses (starting at 0x100000000) from the DRAM
  386. * hole. If not, skip to step 3 below. Else get the value of the
  387. * DramHoleOffset field from the DHAR. To obtain the DramAddr, subtract the
  388. * offset defined by this value from the SysAddr.
  389. *
  390. * 3. Obtain the base address for node n from the DRAMBase field of the DRAM
  391. * Base register for node n. To obtain the DramAddr, subtract the base
  392. * address from the SysAddr, as shown near the start of section 3.4.4 (p.70).
  393. */
  394. static u64 sys_addr_to_dram_addr(struct mem_ctl_info *mci, u64 sys_addr)
  395. {
  396. u64 dram_base, hole_base, hole_offset, hole_size, dram_addr;
  397. int ret = 0;
  398. dram_base = get_dram_base(mci);
  399. ret = amd64_get_dram_hole_info(mci, &hole_base, &hole_offset,
  400. &hole_size);
  401. if (!ret) {
  402. if ((sys_addr >= (1ull << 32)) &&
  403. (sys_addr < ((1ull << 32) + hole_size))) {
  404. /* use DHAR to translate SysAddr to DramAddr */
  405. dram_addr = sys_addr - hole_offset;
  406. debugf2("using DHAR to translate SysAddr 0x%lx to "
  407. "DramAddr 0x%lx\n",
  408. (unsigned long)sys_addr,
  409. (unsigned long)dram_addr);
  410. return dram_addr;
  411. }
  412. }
  413. /*
  414. * Translate the SysAddr to a DramAddr as shown near the start of
  415. * section 3.4.4 (p. 70). Although sys_addr is a 64-bit value, the k8
  416. * only deals with 40-bit values. Therefore we discard bits 63-40 of
  417. * sys_addr below. If bit 39 of sys_addr is 1 then the bits we
  418. * discard are all 1s. Otherwise the bits we discard are all 0s. See
  419. * section 3.4.2 of AMD publication 24592: AMD x86-64 Architecture
  420. * Programmer's Manual Volume 1 Application Programming.
  421. */
  422. dram_addr = (sys_addr & 0xffffffffffull) - dram_base;
  423. debugf2("using DRAM Base register to translate SysAddr 0x%lx to "
  424. "DramAddr 0x%lx\n", (unsigned long)sys_addr,
  425. (unsigned long)dram_addr);
  426. return dram_addr;
  427. }
  428. /*
  429. * @intlv_en is the value of the IntlvEn field from a DRAM Base register
  430. * (section 3.4.4.1). Return the number of bits from a SysAddr that are used
  431. * for node interleaving.
  432. */
  433. static int num_node_interleave_bits(unsigned intlv_en)
  434. {
  435. static const int intlv_shift_table[] = { 0, 1, 0, 2, 0, 0, 0, 3 };
  436. int n;
  437. BUG_ON(intlv_en > 7);
  438. n = intlv_shift_table[intlv_en];
  439. return n;
  440. }
  441. /* Translate the DramAddr given by @dram_addr to an InputAddr. */
  442. static u64 dram_addr_to_input_addr(struct mem_ctl_info *mci, u64 dram_addr)
  443. {
  444. struct amd64_pvt *pvt;
  445. int intlv_shift;
  446. u64 input_addr;
  447. pvt = mci->pvt_info;
  448. /*
  449. * See the start of section 3.4.4 (p. 70, BKDG #26094, K8, revA-E)
  450. * concerning translating a DramAddr to an InputAddr.
  451. */
  452. intlv_shift = num_node_interleave_bits(pvt->dram_IntlvEn[0]);
  453. input_addr = ((dram_addr >> intlv_shift) & 0xffffff000ull) +
  454. (dram_addr & 0xfff);
  455. debugf2(" Intlv Shift=%d DramAddr=0x%lx maps to InputAddr=0x%lx\n",
  456. intlv_shift, (unsigned long)dram_addr,
  457. (unsigned long)input_addr);
  458. return input_addr;
  459. }
  460. /*
  461. * Translate the SysAddr represented by @sys_addr to an InputAddr. It is
  462. * assumed that @sys_addr maps to the node given by mci.
  463. */
  464. static u64 sys_addr_to_input_addr(struct mem_ctl_info *mci, u64 sys_addr)
  465. {
  466. u64 input_addr;
  467. input_addr =
  468. dram_addr_to_input_addr(mci, sys_addr_to_dram_addr(mci, sys_addr));
  469. debugf2("SysAdddr 0x%lx translates to InputAddr 0x%lx\n",
  470. (unsigned long)sys_addr, (unsigned long)input_addr);
  471. return input_addr;
  472. }
  473. /*
  474. * @input_addr is an InputAddr associated with the node represented by mci.
  475. * Translate @input_addr to a DramAddr and return the result.
  476. */
  477. static u64 input_addr_to_dram_addr(struct mem_ctl_info *mci, u64 input_addr)
  478. {
  479. struct amd64_pvt *pvt;
  480. int node_id, intlv_shift;
  481. u64 bits, dram_addr;
  482. u32 intlv_sel;
  483. /*
  484. * Near the start of section 3.4.4 (p. 70, BKDG #26094, K8, revA-E)
  485. * shows how to translate a DramAddr to an InputAddr. Here we reverse
  486. * this procedure. When translating from a DramAddr to an InputAddr, the
  487. * bits used for node interleaving are discarded. Here we recover these
  488. * bits from the IntlvSel field of the DRAM Limit register (section
  489. * 3.4.4.2) for the node that input_addr is associated with.
  490. */
  491. pvt = mci->pvt_info;
  492. node_id = pvt->mc_node_id;
  493. BUG_ON((node_id < 0) || (node_id > 7));
  494. intlv_shift = num_node_interleave_bits(pvt->dram_IntlvEn[0]);
  495. if (intlv_shift == 0) {
  496. debugf1(" InputAddr 0x%lx translates to DramAddr of "
  497. "same value\n", (unsigned long)input_addr);
  498. return input_addr;
  499. }
  500. bits = ((input_addr & 0xffffff000ull) << intlv_shift) +
  501. (input_addr & 0xfff);
  502. intlv_sel = pvt->dram_IntlvSel[node_id] & ((1 << intlv_shift) - 1);
  503. dram_addr = bits + (intlv_sel << 12);
  504. debugf1("InputAddr 0x%lx translates to DramAddr 0x%lx "
  505. "(%d node interleave bits)\n", (unsigned long)input_addr,
  506. (unsigned long)dram_addr, intlv_shift);
  507. return dram_addr;
  508. }
  509. /*
  510. * @dram_addr is a DramAddr that maps to the node represented by mci. Convert
  511. * @dram_addr to a SysAddr.
  512. */
  513. static u64 dram_addr_to_sys_addr(struct mem_ctl_info *mci, u64 dram_addr)
  514. {
  515. struct amd64_pvt *pvt = mci->pvt_info;
  516. u64 hole_base, hole_offset, hole_size, base, limit, sys_addr;
  517. int ret = 0;
  518. ret = amd64_get_dram_hole_info(mci, &hole_base, &hole_offset,
  519. &hole_size);
  520. if (!ret) {
  521. if ((dram_addr >= hole_base) &&
  522. (dram_addr < (hole_base + hole_size))) {
  523. sys_addr = dram_addr + hole_offset;
  524. debugf1("using DHAR to translate DramAddr 0x%lx to "
  525. "SysAddr 0x%lx\n", (unsigned long)dram_addr,
  526. (unsigned long)sys_addr);
  527. return sys_addr;
  528. }
  529. }
  530. amd64_get_base_and_limit(pvt, pvt->mc_node_id, &base, &limit);
  531. sys_addr = dram_addr + base;
  532. /*
  533. * The sys_addr we have computed up to this point is a 40-bit value
  534. * because the k8 deals with 40-bit values. However, the value we are
  535. * supposed to return is a full 64-bit physical address. The AMD
  536. * x86-64 architecture specifies that the most significant implemented
  537. * address bit through bit 63 of a physical address must be either all
  538. * 0s or all 1s. Therefore we sign-extend the 40-bit sys_addr to a
  539. * 64-bit value below. See section 3.4.2 of AMD publication 24592:
  540. * AMD x86-64 Architecture Programmer's Manual Volume 1 Application
  541. * Programming.
  542. */
  543. sys_addr |= ~((sys_addr & (1ull << 39)) - 1);
  544. debugf1(" Node %d, DramAddr 0x%lx to SysAddr 0x%lx\n",
  545. pvt->mc_node_id, (unsigned long)dram_addr,
  546. (unsigned long)sys_addr);
  547. return sys_addr;
  548. }
  549. /*
  550. * @input_addr is an InputAddr associated with the node given by mci. Translate
  551. * @input_addr to a SysAddr.
  552. */
  553. static inline u64 input_addr_to_sys_addr(struct mem_ctl_info *mci,
  554. u64 input_addr)
  555. {
  556. return dram_addr_to_sys_addr(mci,
  557. input_addr_to_dram_addr(mci, input_addr));
  558. }
  559. /*
  560. * Find the minimum and maximum InputAddr values that map to the given @csrow.
  561. * Pass back these values in *input_addr_min and *input_addr_max.
  562. */
  563. static void find_csrow_limits(struct mem_ctl_info *mci, int csrow,
  564. u64 *input_addr_min, u64 *input_addr_max)
  565. {
  566. struct amd64_pvt *pvt;
  567. u64 base, mask;
  568. pvt = mci->pvt_info;
  569. BUG_ON((csrow < 0) || (csrow >= CHIPSELECT_COUNT));
  570. base = base_from_dct_base(pvt, csrow);
  571. mask = mask_from_dct_mask(pvt, csrow);
  572. *input_addr_min = base & ~mask;
  573. *input_addr_max = base | mask | pvt->dcs_mask_notused;
  574. }
  575. /*
  576. * Extract error address from MCA NB Address Low (section 3.6.4.5) and MCA NB
  577. * Address High (section 3.6.4.6) register values and return the result. Address
  578. * is located in the info structure (nbeah and nbeal), the encoding is device
  579. * specific.
  580. */
  581. static u64 extract_error_address(struct mem_ctl_info *mci,
  582. struct amd64_error_info_regs *info)
  583. {
  584. struct amd64_pvt *pvt = mci->pvt_info;
  585. return pvt->ops->get_error_address(mci, info);
  586. }
  587. /* Map the Error address to a PAGE and PAGE OFFSET. */
  588. static inline void error_address_to_page_and_offset(u64 error_address,
  589. u32 *page, u32 *offset)
  590. {
  591. *page = (u32) (error_address >> PAGE_SHIFT);
  592. *offset = ((u32) error_address) & ~PAGE_MASK;
  593. }
  594. /*
  595. * @sys_addr is an error address (a SysAddr) extracted from the MCA NB Address
  596. * Low (section 3.6.4.5) and MCA NB Address High (section 3.6.4.6) registers
  597. * of a node that detected an ECC memory error. mci represents the node that
  598. * the error address maps to (possibly different from the node that detected
  599. * the error). Return the number of the csrow that sys_addr maps to, or -1 on
  600. * error.
  601. */
  602. static int sys_addr_to_csrow(struct mem_ctl_info *mci, u64 sys_addr)
  603. {
  604. int csrow;
  605. csrow = input_addr_to_csrow(mci, sys_addr_to_input_addr(mci, sys_addr));
  606. if (csrow == -1)
  607. amd64_mc_printk(mci, KERN_ERR,
  608. "Failed to translate InputAddr to csrow for "
  609. "address 0x%lx\n", (unsigned long)sys_addr);
  610. return csrow;
  611. }
  612. static int get_channel_from_ecc_syndrome(unsigned short syndrome);
  613. static void amd64_cpu_display_info(struct amd64_pvt *pvt)
  614. {
  615. if (boot_cpu_data.x86 == 0x11)
  616. edac_printk(KERN_DEBUG, EDAC_MC, "F11h CPU detected\n");
  617. else if (boot_cpu_data.x86 == 0x10)
  618. edac_printk(KERN_DEBUG, EDAC_MC, "F10h CPU detected\n");
  619. else if (boot_cpu_data.x86 == 0xf)
  620. edac_printk(KERN_DEBUG, EDAC_MC, "%s detected\n",
  621. (pvt->ext_model >= OPTERON_CPU_REV_F) ?
  622. "Rev F or later" : "Rev E or earlier");
  623. else
  624. /* we'll hardly ever ever get here */
  625. edac_printk(KERN_ERR, EDAC_MC, "Unknown cpu!\n");
  626. }
  627. /*
  628. * Determine if the DIMMs have ECC enabled. ECC is enabled ONLY if all the DIMMs
  629. * are ECC capable.
  630. */
  631. static enum edac_type amd64_determine_edac_cap(struct amd64_pvt *pvt)
  632. {
  633. int bit;
  634. enum dev_type edac_cap = EDAC_NONE;
  635. bit = (boot_cpu_data.x86 > 0xf || pvt->ext_model >= OPTERON_CPU_REV_F)
  636. ? 19
  637. : 17;
  638. if (pvt->dclr0 >> BIT(bit))
  639. edac_cap = EDAC_FLAG_SECDED;
  640. return edac_cap;
  641. }
  642. static void f10_debug_display_dimm_sizes(int ctrl, struct amd64_pvt *pvt,
  643. int ganged);
  644. /* Display and decode various NB registers for debug purposes. */
  645. static void amd64_dump_misc_regs(struct amd64_pvt *pvt)
  646. {
  647. int ganged;
  648. debugf1(" nbcap:0x%8.08x DctDualCap=%s DualNode=%s 8-Node=%s\n",
  649. pvt->nbcap,
  650. (pvt->nbcap & K8_NBCAP_DCT_DUAL) ? "True" : "False",
  651. (pvt->nbcap & K8_NBCAP_DUAL_NODE) ? "True" : "False",
  652. (pvt->nbcap & K8_NBCAP_8_NODE) ? "True" : "False");
  653. debugf1(" ECC Capable=%s ChipKill Capable=%s\n",
  654. (pvt->nbcap & K8_NBCAP_SECDED) ? "True" : "False",
  655. (pvt->nbcap & K8_NBCAP_CHIPKILL) ? "True" : "False");
  656. debugf1(" DramCfg0-low=0x%08x DIMM-ECC=%s Parity=%s Width=%s\n",
  657. pvt->dclr0,
  658. (pvt->dclr0 & BIT(19)) ? "Enabled" : "Disabled",
  659. (pvt->dclr0 & BIT(8)) ? "Enabled" : "Disabled",
  660. (pvt->dclr0 & BIT(11)) ? "128b" : "64b");
  661. debugf1(" DIMM x4 Present: L0=%s L1=%s L2=%s L3=%s DIMM Type=%s\n",
  662. (pvt->dclr0 & BIT(12)) ? "Y" : "N",
  663. (pvt->dclr0 & BIT(13)) ? "Y" : "N",
  664. (pvt->dclr0 & BIT(14)) ? "Y" : "N",
  665. (pvt->dclr0 & BIT(15)) ? "Y" : "N",
  666. (pvt->dclr0 & BIT(16)) ? "UN-Buffered" : "Buffered");
  667. debugf1(" online-spare: 0x%8.08x\n", pvt->online_spare);
  668. if (boot_cpu_data.x86 == 0xf) {
  669. debugf1(" dhar: 0x%8.08x Base=0x%08x Offset=0x%08x\n",
  670. pvt->dhar, dhar_base(pvt->dhar),
  671. k8_dhar_offset(pvt->dhar));
  672. debugf1(" DramHoleValid=%s\n",
  673. (pvt->dhar & DHAR_VALID) ? "True" : "False");
  674. debugf1(" dbam-dkt: 0x%8.08x\n", pvt->dbam0);
  675. /* everything below this point is Fam10h and above */
  676. return;
  677. } else {
  678. debugf1(" dhar: 0x%8.08x Base=0x%08x Offset=0x%08x\n",
  679. pvt->dhar, dhar_base(pvt->dhar),
  680. f10_dhar_offset(pvt->dhar));
  681. debugf1(" DramMemHoistValid=%s DramHoleValid=%s\n",
  682. (pvt->dhar & F10_DRAM_MEM_HOIST_VALID) ?
  683. "True" : "False",
  684. (pvt->dhar & DHAR_VALID) ?
  685. "True" : "False");
  686. }
  687. /* Only if NOT ganged does dcl1 have valid info */
  688. if (!dct_ganging_enabled(pvt)) {
  689. debugf1(" DramCfg1-low=0x%08x DIMM-ECC=%s Parity=%s "
  690. "Width=%s\n", pvt->dclr1,
  691. (pvt->dclr1 & BIT(19)) ? "Enabled" : "Disabled",
  692. (pvt->dclr1 & BIT(8)) ? "Enabled" : "Disabled",
  693. (pvt->dclr1 & BIT(11)) ? "128b" : "64b");
  694. debugf1(" DIMM x4 Present: L0=%s L1=%s L2=%s L3=%s "
  695. "DIMM Type=%s\n",
  696. (pvt->dclr1 & BIT(12)) ? "Y" : "N",
  697. (pvt->dclr1 & BIT(13)) ? "Y" : "N",
  698. (pvt->dclr1 & BIT(14)) ? "Y" : "N",
  699. (pvt->dclr1 & BIT(15)) ? "Y" : "N",
  700. (pvt->dclr1 & BIT(16)) ? "UN-Buffered" : "Buffered");
  701. }
  702. /*
  703. * Determine if ganged and then dump memory sizes for first controller,
  704. * and if NOT ganged dump info for 2nd controller.
  705. */
  706. ganged = dct_ganging_enabled(pvt);
  707. f10_debug_display_dimm_sizes(0, pvt, ganged);
  708. if (!ganged)
  709. f10_debug_display_dimm_sizes(1, pvt, ganged);
  710. }
  711. /* Read in both of DBAM registers */
  712. static void amd64_read_dbam_reg(struct amd64_pvt *pvt)
  713. {
  714. int err = 0;
  715. unsigned int reg;
  716. reg = DBAM0;
  717. err = pci_read_config_dword(pvt->dram_f2_ctl, reg, &pvt->dbam0);
  718. if (err)
  719. goto err_reg;
  720. if (boot_cpu_data.x86 >= 0x10) {
  721. reg = DBAM1;
  722. err = pci_read_config_dword(pvt->dram_f2_ctl, reg, &pvt->dbam1);
  723. if (err)
  724. goto err_reg;
  725. }
  726. err_reg:
  727. debugf0("Error reading F2x%03x.\n", reg);
  728. }
  729. /*
  730. * NOTE: CPU Revision Dependent code: Rev E and Rev F
  731. *
  732. * Set the DCSB and DCSM mask values depending on the CPU revision value. Also
  733. * set the shift factor for the DCSB and DCSM values.
  734. *
  735. * ->dcs_mask_notused, RevE:
  736. *
  737. * To find the max InputAddr for the csrow, start with the base address and set
  738. * all bits that are "don't care" bits in the test at the start of section
  739. * 3.5.4 (p. 84).
  740. *
  741. * The "don't care" bits are all set bits in the mask and all bits in the gaps
  742. * between bit ranges [35:25] and [19:13]. The value REV_E_DCS_NOTUSED_BITS
  743. * represents bits [24:20] and [12:0], which are all bits in the above-mentioned
  744. * gaps.
  745. *
  746. * ->dcs_mask_notused, RevF and later:
  747. *
  748. * To find the max InputAddr for the csrow, start with the base address and set
  749. * all bits that are "don't care" bits in the test at the start of NPT section
  750. * 4.5.4 (p. 87).
  751. *
  752. * The "don't care" bits are all set bits in the mask and all bits in the gaps
  753. * between bit ranges [36:27] and [21:13].
  754. *
  755. * The value REV_F_F1Xh_DCS_NOTUSED_BITS represents bits [26:22] and [12:0],
  756. * which are all bits in the above-mentioned gaps.
  757. */
  758. static void amd64_set_dct_base_and_mask(struct amd64_pvt *pvt)
  759. {
  760. if (pvt->ext_model >= OPTERON_CPU_REV_F) {
  761. pvt->dcsb_base = REV_F_F1Xh_DCSB_BASE_BITS;
  762. pvt->dcsm_mask = REV_F_F1Xh_DCSM_MASK_BITS;
  763. pvt->dcs_mask_notused = REV_F_F1Xh_DCS_NOTUSED_BITS;
  764. pvt->dcs_shift = REV_F_F1Xh_DCS_SHIFT;
  765. switch (boot_cpu_data.x86) {
  766. case 0xf:
  767. pvt->num_dcsm = REV_F_DCSM_COUNT;
  768. break;
  769. case 0x10:
  770. pvt->num_dcsm = F10_DCSM_COUNT;
  771. break;
  772. case 0x11:
  773. pvt->num_dcsm = F11_DCSM_COUNT;
  774. break;
  775. default:
  776. amd64_printk(KERN_ERR, "Unsupported family!\n");
  777. break;
  778. }
  779. } else {
  780. pvt->dcsb_base = REV_E_DCSB_BASE_BITS;
  781. pvt->dcsm_mask = REV_E_DCSM_MASK_BITS;
  782. pvt->dcs_mask_notused = REV_E_DCS_NOTUSED_BITS;
  783. pvt->dcs_shift = REV_E_DCS_SHIFT;
  784. pvt->num_dcsm = REV_E_DCSM_COUNT;
  785. }
  786. }
  787. /*
  788. * Function 2 Offset F10_DCSB0; read in the DCS Base and DCS Mask hw registers
  789. */
  790. static void amd64_read_dct_base_mask(struct amd64_pvt *pvt)
  791. {
  792. int cs, reg, err = 0;
  793. amd64_set_dct_base_and_mask(pvt);
  794. for (cs = 0; cs < CHIPSELECT_COUNT; cs++) {
  795. reg = K8_DCSB0 + (cs * 4);
  796. err = pci_read_config_dword(pvt->dram_f2_ctl, reg,
  797. &pvt->dcsb0[cs]);
  798. if (unlikely(err))
  799. debugf0("Reading K8_DCSB0[%d] failed\n", cs);
  800. else
  801. debugf0(" DCSB0[%d]=0x%08x reg: F2x%x\n",
  802. cs, pvt->dcsb0[cs], reg);
  803. /* If DCT are NOT ganged, then read in DCT1's base */
  804. if (boot_cpu_data.x86 >= 0x10 && !dct_ganging_enabled(pvt)) {
  805. reg = F10_DCSB1 + (cs * 4);
  806. err = pci_read_config_dword(pvt->dram_f2_ctl, reg,
  807. &pvt->dcsb1[cs]);
  808. if (unlikely(err))
  809. debugf0("Reading F10_DCSB1[%d] failed\n", cs);
  810. else
  811. debugf0(" DCSB1[%d]=0x%08x reg: F2x%x\n",
  812. cs, pvt->dcsb1[cs], reg);
  813. } else {
  814. pvt->dcsb1[cs] = 0;
  815. }
  816. }
  817. for (cs = 0; cs < pvt->num_dcsm; cs++) {
  818. reg = K8_DCSB0 + (cs * 4);
  819. err = pci_read_config_dword(pvt->dram_f2_ctl, reg,
  820. &pvt->dcsm0[cs]);
  821. if (unlikely(err))
  822. debugf0("Reading K8_DCSM0 failed\n");
  823. else
  824. debugf0(" DCSM0[%d]=0x%08x reg: F2x%x\n",
  825. cs, pvt->dcsm0[cs], reg);
  826. /* If DCT are NOT ganged, then read in DCT1's mask */
  827. if (boot_cpu_data.x86 >= 0x10 && !dct_ganging_enabled(pvt)) {
  828. reg = F10_DCSM1 + (cs * 4);
  829. err = pci_read_config_dword(pvt->dram_f2_ctl, reg,
  830. &pvt->dcsm1[cs]);
  831. if (unlikely(err))
  832. debugf0("Reading F10_DCSM1[%d] failed\n", cs);
  833. else
  834. debugf0(" DCSM1[%d]=0x%08x reg: F2x%x\n",
  835. cs, pvt->dcsm1[cs], reg);
  836. } else
  837. pvt->dcsm1[cs] = 0;
  838. }
  839. }
  840. static enum mem_type amd64_determine_memory_type(struct amd64_pvt *pvt)
  841. {
  842. enum mem_type type;
  843. if (boot_cpu_data.x86 >= 0x10 || pvt->ext_model >= OPTERON_CPU_REV_F) {
  844. /* Rev F and later */
  845. type = (pvt->dclr0 & BIT(16)) ? MEM_DDR2 : MEM_RDDR2;
  846. } else {
  847. /* Rev E and earlier */
  848. type = (pvt->dclr0 & BIT(18)) ? MEM_DDR : MEM_RDDR;
  849. }
  850. debugf1(" Memory type is: %s\n",
  851. (type == MEM_DDR2) ? "MEM_DDR2" :
  852. (type == MEM_RDDR2) ? "MEM_RDDR2" :
  853. (type == MEM_DDR) ? "MEM_DDR" : "MEM_RDDR");
  854. return type;
  855. }
  856. /*
  857. * Read the DRAM Configuration Low register. It differs between CG, D & E revs
  858. * and the later RevF memory controllers (DDR vs DDR2)
  859. *
  860. * Return:
  861. * number of memory channels in operation
  862. * Pass back:
  863. * contents of the DCL0_LOW register
  864. */
  865. static int k8_early_channel_count(struct amd64_pvt *pvt)
  866. {
  867. int flag, err = 0;
  868. err = pci_read_config_dword(pvt->dram_f2_ctl, F10_DCLR_0, &pvt->dclr0);
  869. if (err)
  870. return err;
  871. if ((boot_cpu_data.x86_model >> 4) >= OPTERON_CPU_REV_F) {
  872. /* RevF (NPT) and later */
  873. flag = pvt->dclr0 & F10_WIDTH_128;
  874. } else {
  875. /* RevE and earlier */
  876. flag = pvt->dclr0 & REVE_WIDTH_128;
  877. }
  878. /* not used */
  879. pvt->dclr1 = 0;
  880. return (flag) ? 2 : 1;
  881. }
  882. /* extract the ERROR ADDRESS for the K8 CPUs */
  883. static u64 k8_get_error_address(struct mem_ctl_info *mci,
  884. struct amd64_error_info_regs *info)
  885. {
  886. return (((u64) (info->nbeah & 0xff)) << 32) +
  887. (info->nbeal & ~0x03);
  888. }
  889. /*
  890. * Read the Base and Limit registers for K8 based Memory controllers; extract
  891. * fields from the 'raw' reg into separate data fields
  892. *
  893. * Isolates: BASE, LIMIT, IntlvEn, IntlvSel, RW_EN
  894. */
  895. static void k8_read_dram_base_limit(struct amd64_pvt *pvt, int dram)
  896. {
  897. u32 low;
  898. u32 off = dram << 3; /* 8 bytes between DRAM entries */
  899. int err;
  900. err = pci_read_config_dword(pvt->addr_f1_ctl,
  901. K8_DRAM_BASE_LOW + off, &low);
  902. if (err)
  903. debugf0("Reading K8_DRAM_BASE_LOW failed\n");
  904. /* Extract parts into separate data entries */
  905. pvt->dram_base[dram] = ((u64) low & 0xFFFF0000) << 8;
  906. pvt->dram_IntlvEn[dram] = (low >> 8) & 0x7;
  907. pvt->dram_rw_en[dram] = (low & 0x3);
  908. err = pci_read_config_dword(pvt->addr_f1_ctl,
  909. K8_DRAM_LIMIT_LOW + off, &low);
  910. if (err)
  911. debugf0("Reading K8_DRAM_LIMIT_LOW failed\n");
  912. /*
  913. * Extract parts into separate data entries. Limit is the HIGHEST memory
  914. * location of the region, so lower 24 bits need to be all ones
  915. */
  916. pvt->dram_limit[dram] = (((u64) low & 0xFFFF0000) << 8) | 0x00FFFFFF;
  917. pvt->dram_IntlvSel[dram] = (low >> 8) & 0x7;
  918. pvt->dram_DstNode[dram] = (low & 0x7);
  919. }
  920. static void k8_map_sysaddr_to_csrow(struct mem_ctl_info *mci,
  921. struct amd64_error_info_regs *info,
  922. u64 SystemAddress)
  923. {
  924. struct mem_ctl_info *src_mci;
  925. unsigned short syndrome;
  926. int channel, csrow;
  927. u32 page, offset;
  928. /* Extract the syndrome parts and form a 16-bit syndrome */
  929. syndrome = EXTRACT_HIGH_SYNDROME(info->nbsl) << 8;
  930. syndrome |= EXTRACT_LOW_SYNDROME(info->nbsh);
  931. /* CHIPKILL enabled */
  932. if (info->nbcfg & K8_NBCFG_CHIPKILL) {
  933. channel = get_channel_from_ecc_syndrome(syndrome);
  934. if (channel < 0) {
  935. /*
  936. * Syndrome didn't map, so we don't know which of the
  937. * 2 DIMMs is in error. So we need to ID 'both' of them
  938. * as suspect.
  939. */
  940. amd64_mc_printk(mci, KERN_WARNING,
  941. "unknown syndrome 0x%x - possible error "
  942. "reporting race\n", syndrome);
  943. edac_mc_handle_ce_no_info(mci, EDAC_MOD_STR);
  944. return;
  945. }
  946. } else {
  947. /*
  948. * non-chipkill ecc mode
  949. *
  950. * The k8 documentation is unclear about how to determine the
  951. * channel number when using non-chipkill memory. This method
  952. * was obtained from email communication with someone at AMD.
  953. * (Wish the email was placed in this comment - norsk)
  954. */
  955. channel = ((SystemAddress & BIT(3)) != 0);
  956. }
  957. /*
  958. * Find out which node the error address belongs to. This may be
  959. * different from the node that detected the error.
  960. */
  961. src_mci = find_mc_by_sys_addr(mci, SystemAddress);
  962. if (src_mci) {
  963. amd64_mc_printk(mci, KERN_ERR,
  964. "failed to map error address 0x%lx to a node\n",
  965. (unsigned long)SystemAddress);
  966. edac_mc_handle_ce_no_info(mci, EDAC_MOD_STR);
  967. return;
  968. }
  969. /* Now map the SystemAddress to a CSROW */
  970. csrow = sys_addr_to_csrow(src_mci, SystemAddress);
  971. if (csrow < 0) {
  972. edac_mc_handle_ce_no_info(src_mci, EDAC_MOD_STR);
  973. } else {
  974. error_address_to_page_and_offset(SystemAddress, &page, &offset);
  975. edac_mc_handle_ce(src_mci, page, offset, syndrome, csrow,
  976. channel, EDAC_MOD_STR);
  977. }
  978. }
  979. /*
  980. * determrine the number of PAGES in for this DIMM's size based on its DRAM
  981. * Address Mapping.
  982. *
  983. * First step is to calc the number of bits to shift a value of 1 left to
  984. * indicate show many pages. Start with the DBAM value as the starting bits,
  985. * then proceed to adjust those shift bits, based on CPU rev and the table.
  986. * See BKDG on the DBAM
  987. */
  988. static int k8_dbam_map_to_pages(struct amd64_pvt *pvt, int dram_map)
  989. {
  990. int nr_pages;
  991. if (pvt->ext_model >= OPTERON_CPU_REV_F) {
  992. nr_pages = 1 << (revf_quad_ddr2_shift[dram_map] - PAGE_SHIFT);
  993. } else {
  994. /*
  995. * RevE and less section; this line is tricky. It collapses the
  996. * table used by RevD and later to one that matches revisions CG
  997. * and earlier.
  998. */
  999. dram_map -= (pvt->ext_model >= OPTERON_CPU_REV_D) ?
  1000. (dram_map > 8 ? 4 : (dram_map > 5 ?
  1001. 3 : (dram_map > 2 ? 1 : 0))) : 0;
  1002. /* 25 shift is 32MiB minimum DIMM size in RevE and prior */
  1003. nr_pages = 1 << (dram_map + 25 - PAGE_SHIFT);
  1004. }
  1005. return nr_pages;
  1006. }
  1007. /*
  1008. * Get the number of DCT channels in use.
  1009. *
  1010. * Return:
  1011. * number of Memory Channels in operation
  1012. * Pass back:
  1013. * contents of the DCL0_LOW register
  1014. */
  1015. static int f10_early_channel_count(struct amd64_pvt *pvt)
  1016. {
  1017. int err = 0, channels = 0;
  1018. u32 dbam;
  1019. err = pci_read_config_dword(pvt->dram_f2_ctl, F10_DCLR_0, &pvt->dclr0);
  1020. if (err)
  1021. goto err_reg;
  1022. err = pci_read_config_dword(pvt->dram_f2_ctl, F10_DCLR_1, &pvt->dclr1);
  1023. if (err)
  1024. goto err_reg;
  1025. /* If we are in 128 bit mode, then we are using 2 channels */
  1026. if (pvt->dclr0 & F10_WIDTH_128) {
  1027. debugf0("Data WIDTH is 128 bits - 2 channels\n");
  1028. channels = 2;
  1029. return channels;
  1030. }
  1031. /*
  1032. * Need to check if in UN-ganged mode: In such, there are 2 channels,
  1033. * but they are NOT in 128 bit mode and thus the above 'dcl0' status bit
  1034. * will be OFF.
  1035. *
  1036. * Need to check DCT0[0] and DCT1[0] to see if only one of them has
  1037. * their CSEnable bit on. If so, then SINGLE DIMM case.
  1038. */
  1039. debugf0("Data WIDTH is NOT 128 bits - need more decoding\n");
  1040. /*
  1041. * Check DRAM Bank Address Mapping values for each DIMM to see if there
  1042. * is more than just one DIMM present in unganged mode. Need to check
  1043. * both controllers since DIMMs can be placed in either one.
  1044. */
  1045. channels = 0;
  1046. err = pci_read_config_dword(pvt->dram_f2_ctl, DBAM0, &dbam);
  1047. if (err)
  1048. goto err_reg;
  1049. if (DBAM_DIMM(0, dbam) > 0)
  1050. channels++;
  1051. if (DBAM_DIMM(1, dbam) > 0)
  1052. channels++;
  1053. if (DBAM_DIMM(2, dbam) > 0)
  1054. channels++;
  1055. if (DBAM_DIMM(3, dbam) > 0)
  1056. channels++;
  1057. /* If more than 2 DIMMs are present, then we have 2 channels */
  1058. if (channels > 2)
  1059. channels = 2;
  1060. else if (channels == 0) {
  1061. /* No DIMMs on DCT0, so look at DCT1 */
  1062. err = pci_read_config_dword(pvt->dram_f2_ctl, DBAM1, &dbam);
  1063. if (err)
  1064. goto err_reg;
  1065. if (DBAM_DIMM(0, dbam) > 0)
  1066. channels++;
  1067. if (DBAM_DIMM(1, dbam) > 0)
  1068. channels++;
  1069. if (DBAM_DIMM(2, dbam) > 0)
  1070. channels++;
  1071. if (DBAM_DIMM(3, dbam) > 0)
  1072. channels++;
  1073. if (channels > 2)
  1074. channels = 2;
  1075. }
  1076. /* If we found ALL 0 values, then assume just ONE DIMM-ONE Channel */
  1077. if (channels == 0)
  1078. channels = 1;
  1079. debugf0("DIMM count= %d\n", channels);
  1080. return channels;
  1081. err_reg:
  1082. return -1;
  1083. }
  1084. static int f10_dbam_map_to_pages(struct amd64_pvt *pvt, int dram_map)
  1085. {
  1086. return 1 << (revf_quad_ddr2_shift[dram_map] - PAGE_SHIFT);
  1087. }
  1088. /* Enable extended configuration access via 0xCF8 feature */
  1089. static void amd64_setup(struct amd64_pvt *pvt)
  1090. {
  1091. u32 reg;
  1092. pci_read_config_dword(pvt->misc_f3_ctl, F10_NB_CFG_HIGH, &reg);
  1093. pvt->flags.cf8_extcfg = !!(reg & F10_NB_CFG_LOW_ENABLE_EXT_CFG);
  1094. reg |= F10_NB_CFG_LOW_ENABLE_EXT_CFG;
  1095. pci_write_config_dword(pvt->misc_f3_ctl, F10_NB_CFG_HIGH, reg);
  1096. }
  1097. /* Restore the extended configuration access via 0xCF8 feature */
  1098. static void amd64_teardown(struct amd64_pvt *pvt)
  1099. {
  1100. u32 reg;
  1101. pci_read_config_dword(pvt->misc_f3_ctl, F10_NB_CFG_HIGH, &reg);
  1102. reg &= ~F10_NB_CFG_LOW_ENABLE_EXT_CFG;
  1103. if (pvt->flags.cf8_extcfg)
  1104. reg |= F10_NB_CFG_LOW_ENABLE_EXT_CFG;
  1105. pci_write_config_dword(pvt->misc_f3_ctl, F10_NB_CFG_HIGH, reg);
  1106. }
  1107. static u64 f10_get_error_address(struct mem_ctl_info *mci,
  1108. struct amd64_error_info_regs *info)
  1109. {
  1110. return (((u64) (info->nbeah & 0xffff)) << 32) +
  1111. (info->nbeal & ~0x01);
  1112. }
  1113. /*
  1114. * Read the Base and Limit registers for F10 based Memory controllers. Extract
  1115. * fields from the 'raw' reg into separate data fields.
  1116. *
  1117. * Isolates: BASE, LIMIT, IntlvEn, IntlvSel, RW_EN.
  1118. */
  1119. static void f10_read_dram_base_limit(struct amd64_pvt *pvt, int dram)
  1120. {
  1121. u32 high_offset, low_offset, high_base, low_base, high_limit, low_limit;
  1122. low_offset = K8_DRAM_BASE_LOW + (dram << 3);
  1123. high_offset = F10_DRAM_BASE_HIGH + (dram << 3);
  1124. /* read the 'raw' DRAM BASE Address register */
  1125. pci_read_config_dword(pvt->addr_f1_ctl, low_offset, &low_base);
  1126. /* Read from the ECS data register */
  1127. pci_read_config_dword(pvt->addr_f1_ctl, high_offset, &high_base);
  1128. /* Extract parts into separate data entries */
  1129. pvt->dram_rw_en[dram] = (low_base & 0x3);
  1130. if (pvt->dram_rw_en[dram] == 0)
  1131. return;
  1132. pvt->dram_IntlvEn[dram] = (low_base >> 8) & 0x7;
  1133. pvt->dram_base[dram] = (((((u64) high_base & 0x000000FF) << 32) |
  1134. ((u64) low_base & 0xFFFF0000))) << 8;
  1135. low_offset = K8_DRAM_LIMIT_LOW + (dram << 3);
  1136. high_offset = F10_DRAM_LIMIT_HIGH + (dram << 3);
  1137. /* read the 'raw' LIMIT registers */
  1138. pci_read_config_dword(pvt->addr_f1_ctl, low_offset, &low_limit);
  1139. /* Read from the ECS data register for the HIGH portion */
  1140. pci_read_config_dword(pvt->addr_f1_ctl, high_offset, &high_limit);
  1141. debugf0(" HW Regs: BASE=0x%08x-%08x LIMIT= 0x%08x-%08x\n",
  1142. high_base, low_base, high_limit, low_limit);
  1143. pvt->dram_DstNode[dram] = (low_limit & 0x7);
  1144. pvt->dram_IntlvSel[dram] = (low_limit >> 8) & 0x7;
  1145. /*
  1146. * Extract address values and form a LIMIT address. Limit is the HIGHEST
  1147. * memory location of the region, so low 24 bits need to be all ones.
  1148. */
  1149. low_limit |= 0x0000FFFF;
  1150. pvt->dram_limit[dram] =
  1151. ((((u64) high_limit << 32) + (u64) low_limit) << 8) | (0xFF);
  1152. }
  1153. static void f10_read_dram_ctl_register(struct amd64_pvt *pvt)
  1154. {
  1155. int err = 0;
  1156. err = pci_read_config_dword(pvt->dram_f2_ctl, F10_DCTL_SEL_LOW,
  1157. &pvt->dram_ctl_select_low);
  1158. if (err) {
  1159. debugf0("Reading F10_DCTL_SEL_LOW failed\n");
  1160. } else {
  1161. debugf0("DRAM_DCTL_SEL_LOW=0x%x DctSelBaseAddr=0x%x\n",
  1162. pvt->dram_ctl_select_low, dct_sel_baseaddr(pvt));
  1163. debugf0(" DRAM DCTs are=%s DRAM Is=%s DRAM-Ctl-"
  1164. "sel-hi-range=%s\n",
  1165. (dct_ganging_enabled(pvt) ? "GANGED" : "NOT GANGED"),
  1166. (dct_dram_enabled(pvt) ? "Enabled" : "Disabled"),
  1167. (dct_high_range_enabled(pvt) ? "Enabled" : "Disabled"));
  1168. debugf0(" DctDatIntLv=%s MemCleared=%s DctSelIntLvAddr=0x%x\n",
  1169. (dct_data_intlv_enabled(pvt) ? "Enabled" : "Disabled"),
  1170. (dct_memory_cleared(pvt) ? "True " : "False "),
  1171. dct_sel_interleave_addr(pvt));
  1172. }
  1173. err = pci_read_config_dword(pvt->dram_f2_ctl, F10_DCTL_SEL_HIGH,
  1174. &pvt->dram_ctl_select_high);
  1175. if (err)
  1176. debugf0("Reading F10_DCTL_SEL_HIGH failed\n");
  1177. }
  1178. /*
  1179. * determine channel based on the interleaving mode: F10h BKDG, 2.8.9 Memory
  1180. * Interleaving Modes.
  1181. */
  1182. static u32 f10_determine_channel(struct amd64_pvt *pvt, u64 sys_addr,
  1183. int hi_range_sel, u32 intlv_en)
  1184. {
  1185. u32 cs, temp, dct_sel_high = (pvt->dram_ctl_select_low >> 1) & 1;
  1186. if (dct_ganging_enabled(pvt))
  1187. cs = 0;
  1188. else if (hi_range_sel)
  1189. cs = dct_sel_high;
  1190. else if (dct_interleave_enabled(pvt)) {
  1191. /*
  1192. * see F2x110[DctSelIntLvAddr] - channel interleave mode
  1193. */
  1194. if (dct_sel_interleave_addr(pvt) == 0)
  1195. cs = sys_addr >> 6 & 1;
  1196. else if ((dct_sel_interleave_addr(pvt) >> 1) & 1) {
  1197. temp = hweight_long((u32) ((sys_addr >> 16) & 0x1F)) % 2;
  1198. if (dct_sel_interleave_addr(pvt) & 1)
  1199. cs = (sys_addr >> 9 & 1) ^ temp;
  1200. else
  1201. cs = (sys_addr >> 6 & 1) ^ temp;
  1202. } else if (intlv_en & 4)
  1203. cs = sys_addr >> 15 & 1;
  1204. else if (intlv_en & 2)
  1205. cs = sys_addr >> 14 & 1;
  1206. else if (intlv_en & 1)
  1207. cs = sys_addr >> 13 & 1;
  1208. else
  1209. cs = sys_addr >> 12 & 1;
  1210. } else if (dct_high_range_enabled(pvt) && !dct_ganging_enabled(pvt))
  1211. cs = ~dct_sel_high & 1;
  1212. else
  1213. cs = 0;
  1214. return cs;
  1215. }
  1216. static inline u32 f10_map_intlv_en_to_shift(u32 intlv_en)
  1217. {
  1218. if (intlv_en == 1)
  1219. return 1;
  1220. else if (intlv_en == 3)
  1221. return 2;
  1222. else if (intlv_en == 7)
  1223. return 3;
  1224. return 0;
  1225. }
  1226. /* See F10h BKDG, 2.8.10.2 DctSelBaseOffset Programming */
  1227. static inline u64 f10_get_base_addr_offset(u64 sys_addr, int hi_range_sel,
  1228. u32 dct_sel_base_addr,
  1229. u64 dct_sel_base_off,
  1230. u32 hole_valid, u32 hole_off,
  1231. u64 dram_base)
  1232. {
  1233. u64 chan_off;
  1234. if (hi_range_sel) {
  1235. if (!(dct_sel_base_addr & 0xFFFFF800) &&
  1236. hole_valid && (sys_addr >= 0x100000000ULL))
  1237. chan_off = hole_off << 16;
  1238. else
  1239. chan_off = dct_sel_base_off;
  1240. } else {
  1241. if (hole_valid && (sys_addr >= 0x100000000ULL))
  1242. chan_off = hole_off << 16;
  1243. else
  1244. chan_off = dram_base & 0xFFFFF8000000ULL;
  1245. }
  1246. return (sys_addr & 0x0000FFFFFFFFFFC0ULL) -
  1247. (chan_off & 0x0000FFFFFF800000ULL);
  1248. }
  1249. /* Hack for the time being - Can we get this from BIOS?? */
  1250. #define CH0SPARE_RANK 0
  1251. #define CH1SPARE_RANK 1
  1252. /*
  1253. * checks if the csrow passed in is marked as SPARED, if so returns the new
  1254. * spare row
  1255. */
  1256. static inline int f10_process_possible_spare(int csrow,
  1257. u32 cs, struct amd64_pvt *pvt)
  1258. {
  1259. u32 swap_done;
  1260. u32 bad_dram_cs;
  1261. /* Depending on channel, isolate respective SPARING info */
  1262. if (cs) {
  1263. swap_done = F10_ONLINE_SPARE_SWAPDONE1(pvt->online_spare);
  1264. bad_dram_cs = F10_ONLINE_SPARE_BADDRAM_CS1(pvt->online_spare);
  1265. if (swap_done && (csrow == bad_dram_cs))
  1266. csrow = CH1SPARE_RANK;
  1267. } else {
  1268. swap_done = F10_ONLINE_SPARE_SWAPDONE0(pvt->online_spare);
  1269. bad_dram_cs = F10_ONLINE_SPARE_BADDRAM_CS0(pvt->online_spare);
  1270. if (swap_done && (csrow == bad_dram_cs))
  1271. csrow = CH0SPARE_RANK;
  1272. }
  1273. return csrow;
  1274. }
  1275. /*
  1276. * Iterate over the DRAM DCT "base" and "mask" registers looking for a
  1277. * SystemAddr match on the specified 'ChannelSelect' and 'NodeID'
  1278. *
  1279. * Return:
  1280. * -EINVAL: NOT FOUND
  1281. * 0..csrow = Chip-Select Row
  1282. */
  1283. static int f10_lookup_addr_in_dct(u32 in_addr, u32 nid, u32 cs)
  1284. {
  1285. struct mem_ctl_info *mci;
  1286. struct amd64_pvt *pvt;
  1287. u32 cs_base, cs_mask;
  1288. int cs_found = -EINVAL;
  1289. int csrow;
  1290. mci = mci_lookup[nid];
  1291. if (!mci)
  1292. return cs_found;
  1293. pvt = mci->pvt_info;
  1294. debugf1("InputAddr=0x%x channelselect=%d\n", in_addr, cs);
  1295. for (csrow = 0; csrow < CHIPSELECT_COUNT; csrow++) {
  1296. cs_base = amd64_get_dct_base(pvt, cs, csrow);
  1297. if (!(cs_base & K8_DCSB_CS_ENABLE))
  1298. continue;
  1299. /*
  1300. * We have an ENABLED CSROW, Isolate just the MASK bits of the
  1301. * target: [28:19] and [13:5], which map to [36:27] and [21:13]
  1302. * of the actual address.
  1303. */
  1304. cs_base &= REV_F_F1Xh_DCSB_BASE_BITS;
  1305. /*
  1306. * Get the DCT Mask, and ENABLE the reserved bits: [18:16] and
  1307. * [4:0] to become ON. Then mask off bits [28:0] ([36:8])
  1308. */
  1309. cs_mask = amd64_get_dct_mask(pvt, cs, csrow);
  1310. debugf1(" CSROW=%d CSBase=0x%x RAW CSMask=0x%x\n",
  1311. csrow, cs_base, cs_mask);
  1312. cs_mask = (cs_mask | 0x0007C01F) & 0x1FFFFFFF;
  1313. debugf1(" Final CSMask=0x%x\n", cs_mask);
  1314. debugf1(" (InputAddr & ~CSMask)=0x%x "
  1315. "(CSBase & ~CSMask)=0x%x\n",
  1316. (in_addr & ~cs_mask), (cs_base & ~cs_mask));
  1317. if ((in_addr & ~cs_mask) == (cs_base & ~cs_mask)) {
  1318. cs_found = f10_process_possible_spare(csrow, cs, pvt);
  1319. debugf1(" MATCH csrow=%d\n", cs_found);
  1320. break;
  1321. }
  1322. }
  1323. return cs_found;
  1324. }
  1325. /* For a given @dram_range, check if @sys_addr falls within it. */
  1326. static int f10_match_to_this_node(struct amd64_pvt *pvt, int dram_range,
  1327. u64 sys_addr, int *nid, int *chan_sel)
  1328. {
  1329. int node_id, cs_found = -EINVAL, high_range = 0;
  1330. u32 intlv_en, intlv_sel, intlv_shift, hole_off;
  1331. u32 hole_valid, tmp, dct_sel_base, channel;
  1332. u64 dram_base, chan_addr, dct_sel_base_off;
  1333. dram_base = pvt->dram_base[dram_range];
  1334. intlv_en = pvt->dram_IntlvEn[dram_range];
  1335. node_id = pvt->dram_DstNode[dram_range];
  1336. intlv_sel = pvt->dram_IntlvSel[dram_range];
  1337. debugf1("(dram=%d) Base=0x%llx SystemAddr= 0x%llx Limit=0x%llx\n",
  1338. dram_range, dram_base, sys_addr, pvt->dram_limit[dram_range]);
  1339. /*
  1340. * This assumes that one node's DHAR is the same as all the other
  1341. * nodes' DHAR.
  1342. */
  1343. hole_off = (pvt->dhar & 0x0000FF80);
  1344. hole_valid = (pvt->dhar & 0x1);
  1345. dct_sel_base_off = (pvt->dram_ctl_select_high & 0xFFFFFC00) << 16;
  1346. debugf1(" HoleOffset=0x%x HoleValid=0x%x IntlvSel=0x%x\n",
  1347. hole_off, hole_valid, intlv_sel);
  1348. if (intlv_en ||
  1349. (intlv_sel != ((sys_addr >> 12) & intlv_en)))
  1350. return -EINVAL;
  1351. dct_sel_base = dct_sel_baseaddr(pvt);
  1352. /*
  1353. * check whether addresses >= DctSelBaseAddr[47:27] are to be used to
  1354. * select between DCT0 and DCT1.
  1355. */
  1356. if (dct_high_range_enabled(pvt) &&
  1357. !dct_ganging_enabled(pvt) &&
  1358. ((sys_addr >> 27) >= (dct_sel_base >> 11)))
  1359. high_range = 1;
  1360. channel = f10_determine_channel(pvt, sys_addr, high_range, intlv_en);
  1361. chan_addr = f10_get_base_addr_offset(sys_addr, high_range, dct_sel_base,
  1362. dct_sel_base_off, hole_valid,
  1363. hole_off, dram_base);
  1364. intlv_shift = f10_map_intlv_en_to_shift(intlv_en);
  1365. /* remove Node ID (in case of memory interleaving) */
  1366. tmp = chan_addr & 0xFC0;
  1367. chan_addr = ((chan_addr >> intlv_shift) & 0xFFFFFFFFF000ULL) | tmp;
  1368. /* remove channel interleave and hash */
  1369. if (dct_interleave_enabled(pvt) &&
  1370. !dct_high_range_enabled(pvt) &&
  1371. !dct_ganging_enabled(pvt)) {
  1372. if (dct_sel_interleave_addr(pvt) != 1)
  1373. chan_addr = (chan_addr >> 1) & 0xFFFFFFFFFFFFFFC0ULL;
  1374. else {
  1375. tmp = chan_addr & 0xFC0;
  1376. chan_addr = ((chan_addr & 0xFFFFFFFFFFFFC000ULL) >> 1)
  1377. | tmp;
  1378. }
  1379. }
  1380. debugf1(" (ChannelAddrLong=0x%llx) >> 8 becomes InputAddr=0x%x\n",
  1381. chan_addr, (u32)(chan_addr >> 8));
  1382. cs_found = f10_lookup_addr_in_dct(chan_addr >> 8, node_id, channel);
  1383. if (cs_found >= 0) {
  1384. *nid = node_id;
  1385. *chan_sel = channel;
  1386. }
  1387. return cs_found;
  1388. }
  1389. static int f10_translate_sysaddr_to_cs(struct amd64_pvt *pvt, u64 sys_addr,
  1390. int *node, int *chan_sel)
  1391. {
  1392. int dram_range, cs_found = -EINVAL;
  1393. u64 dram_base, dram_limit;
  1394. for (dram_range = 0; dram_range < DRAM_REG_COUNT; dram_range++) {
  1395. if (!pvt->dram_rw_en[dram_range])
  1396. continue;
  1397. dram_base = pvt->dram_base[dram_range];
  1398. dram_limit = pvt->dram_limit[dram_range];
  1399. if ((dram_base <= sys_addr) && (sys_addr <= dram_limit)) {
  1400. cs_found = f10_match_to_this_node(pvt, dram_range,
  1401. sys_addr, node,
  1402. chan_sel);
  1403. if (cs_found >= 0)
  1404. break;
  1405. }
  1406. }
  1407. return cs_found;
  1408. }
  1409. /*
  1410. * This the F10h reference code from AMD to map a @sys_addr to NodeID,
  1411. * CSROW, Channel.
  1412. *
  1413. * The @sys_addr is usually an error address received from the hardware.
  1414. */
  1415. static void f10_map_sysaddr_to_csrow(struct mem_ctl_info *mci,
  1416. struct amd64_error_info_regs *info,
  1417. u64 sys_addr)
  1418. {
  1419. struct amd64_pvt *pvt = mci->pvt_info;
  1420. u32 page, offset;
  1421. unsigned short syndrome;
  1422. int nid, csrow, chan = 0;
  1423. csrow = f10_translate_sysaddr_to_cs(pvt, sys_addr, &nid, &chan);
  1424. if (csrow >= 0) {
  1425. error_address_to_page_and_offset(sys_addr, &page, &offset);
  1426. syndrome = EXTRACT_HIGH_SYNDROME(info->nbsl) << 8;
  1427. syndrome |= EXTRACT_LOW_SYNDROME(info->nbsh);
  1428. /*
  1429. * Is CHIPKILL on? If so, then we can attempt to use the
  1430. * syndrome to isolate which channel the error was on.
  1431. */
  1432. if (pvt->nbcfg & K8_NBCFG_CHIPKILL)
  1433. chan = get_channel_from_ecc_syndrome(syndrome);
  1434. if (chan >= 0) {
  1435. edac_mc_handle_ce(mci, page, offset, syndrome,
  1436. csrow, chan, EDAC_MOD_STR);
  1437. } else {
  1438. /*
  1439. * Channel unknown, report all channels on this
  1440. * CSROW as failed.
  1441. */
  1442. for (chan = 0; chan < mci->csrows[csrow].nr_channels;
  1443. chan++) {
  1444. edac_mc_handle_ce(mci, page, offset,
  1445. syndrome,
  1446. csrow, chan,
  1447. EDAC_MOD_STR);
  1448. }
  1449. }
  1450. } else {
  1451. edac_mc_handle_ce_no_info(mci, EDAC_MOD_STR);
  1452. }
  1453. }
  1454. /*
  1455. * Input (@index) is the DBAM DIMM value (1 of 4) used as an index into a shift
  1456. * table (revf_quad_ddr2_shift) which starts at 128MB DIMM size. Index of 0
  1457. * indicates an empty DIMM slot, as reported by Hardware on empty slots.
  1458. *
  1459. * Normalize to 128MB by subracting 27 bit shift.
  1460. */
  1461. static int map_dbam_to_csrow_size(int index)
  1462. {
  1463. int mega_bytes = 0;
  1464. if (index > 0 && index <= DBAM_MAX_VALUE)
  1465. mega_bytes = ((128 << (revf_quad_ddr2_shift[index]-27)));
  1466. return mega_bytes;
  1467. }
  1468. /*
  1469. * debug routine to display the memory sizes of a DIMM (ganged or not) and it
  1470. * CSROWs as well
  1471. */
  1472. static void f10_debug_display_dimm_sizes(int ctrl, struct amd64_pvt *pvt,
  1473. int ganged)
  1474. {
  1475. int dimm, size0, size1;
  1476. u32 dbam;
  1477. u32 *dcsb;
  1478. debugf1(" dbam%d: 0x%8.08x CSROW is %s\n", ctrl,
  1479. ctrl ? pvt->dbam1 : pvt->dbam0,
  1480. ganged ? "GANGED - dbam1 not used" : "NON-GANGED");
  1481. dbam = ctrl ? pvt->dbam1 : pvt->dbam0;
  1482. dcsb = ctrl ? pvt->dcsb1 : pvt->dcsb0;
  1483. /* Dump memory sizes for DIMM and its CSROWs */
  1484. for (dimm = 0; dimm < 4; dimm++) {
  1485. size0 = 0;
  1486. if (dcsb[dimm*2] & K8_DCSB_CS_ENABLE)
  1487. size0 = map_dbam_to_csrow_size(DBAM_DIMM(dimm, dbam));
  1488. size1 = 0;
  1489. if (dcsb[dimm*2 + 1] & K8_DCSB_CS_ENABLE)
  1490. size1 = map_dbam_to_csrow_size(DBAM_DIMM(dimm, dbam));
  1491. debugf1(" CTRL-%d DIMM-%d=%5dMB CSROW-%d=%5dMB "
  1492. "CSROW-%d=%5dMB\n",
  1493. ctrl,
  1494. dimm,
  1495. size0 + size1,
  1496. dimm * 2,
  1497. size0,
  1498. dimm * 2 + 1,
  1499. size1);
  1500. }
  1501. }
  1502. /*
  1503. * Very early hardware probe on pci_probe thread to determine if this module
  1504. * supports the hardware.
  1505. *
  1506. * Return:
  1507. * 0 for OK
  1508. * 1 for error
  1509. */
  1510. static int f10_probe_valid_hardware(struct amd64_pvt *pvt)
  1511. {
  1512. int ret = 0;
  1513. /*
  1514. * If we are on a DDR3 machine, we don't know yet if
  1515. * we support that properly at this time
  1516. */
  1517. if ((pvt->dchr0 & F10_DCHR_Ddr3Mode) ||
  1518. (pvt->dchr1 & F10_DCHR_Ddr3Mode)) {
  1519. amd64_printk(KERN_WARNING,
  1520. "%s() This machine is running with DDR3 memory. "
  1521. "This is not currently supported. "
  1522. "DCHR0=0x%x DCHR1=0x%x\n",
  1523. __func__, pvt->dchr0, pvt->dchr1);
  1524. amd64_printk(KERN_WARNING,
  1525. " Contact '%s' module MAINTAINER to help add"
  1526. " support.\n",
  1527. EDAC_MOD_STR);
  1528. ret = 1;
  1529. }
  1530. return ret;
  1531. }
  1532. /*
  1533. * There currently are 3 types type of MC devices for AMD Athlon/Opterons
  1534. * (as per PCI DEVICE_IDs):
  1535. *
  1536. * Family K8: That is the Athlon64 and Opteron CPUs. They all have the same PCI
  1537. * DEVICE ID, even though there is differences between the different Revisions
  1538. * (CG,D,E,F).
  1539. *
  1540. * Family F10h and F11h.
  1541. *
  1542. */
  1543. static struct amd64_family_type amd64_family_types[] = {
  1544. [K8_CPUS] = {
  1545. .ctl_name = "RevF",
  1546. .addr_f1_ctl = PCI_DEVICE_ID_AMD_K8_NB_ADDRMAP,
  1547. .misc_f3_ctl = PCI_DEVICE_ID_AMD_K8_NB_MISC,
  1548. .ops = {
  1549. .early_channel_count = k8_early_channel_count,
  1550. .get_error_address = k8_get_error_address,
  1551. .read_dram_base_limit = k8_read_dram_base_limit,
  1552. .map_sysaddr_to_csrow = k8_map_sysaddr_to_csrow,
  1553. .dbam_map_to_pages = k8_dbam_map_to_pages,
  1554. }
  1555. },
  1556. [F10_CPUS] = {
  1557. .ctl_name = "Family 10h",
  1558. .addr_f1_ctl = PCI_DEVICE_ID_AMD_10H_NB_MAP,
  1559. .misc_f3_ctl = PCI_DEVICE_ID_AMD_10H_NB_MISC,
  1560. .ops = {
  1561. .probe_valid_hardware = f10_probe_valid_hardware,
  1562. .early_channel_count = f10_early_channel_count,
  1563. .get_error_address = f10_get_error_address,
  1564. .read_dram_base_limit = f10_read_dram_base_limit,
  1565. .read_dram_ctl_register = f10_read_dram_ctl_register,
  1566. .map_sysaddr_to_csrow = f10_map_sysaddr_to_csrow,
  1567. .dbam_map_to_pages = f10_dbam_map_to_pages,
  1568. }
  1569. },
  1570. [F11_CPUS] = {
  1571. .ctl_name = "Family 11h",
  1572. .addr_f1_ctl = PCI_DEVICE_ID_AMD_11H_NB_MAP,
  1573. .misc_f3_ctl = PCI_DEVICE_ID_AMD_11H_NB_MISC,
  1574. .ops = {
  1575. .probe_valid_hardware = f10_probe_valid_hardware,
  1576. .early_channel_count = f10_early_channel_count,
  1577. .get_error_address = f10_get_error_address,
  1578. .read_dram_base_limit = f10_read_dram_base_limit,
  1579. .read_dram_ctl_register = f10_read_dram_ctl_register,
  1580. .map_sysaddr_to_csrow = f10_map_sysaddr_to_csrow,
  1581. .dbam_map_to_pages = f10_dbam_map_to_pages,
  1582. }
  1583. },
  1584. };
  1585. static struct pci_dev *pci_get_related_function(unsigned int vendor,
  1586. unsigned int device,
  1587. struct pci_dev *related)
  1588. {
  1589. struct pci_dev *dev = NULL;
  1590. dev = pci_get_device(vendor, device, dev);
  1591. while (dev) {
  1592. if ((dev->bus->number == related->bus->number) &&
  1593. (PCI_SLOT(dev->devfn) == PCI_SLOT(related->devfn)))
  1594. break;
  1595. dev = pci_get_device(vendor, device, dev);
  1596. }
  1597. return dev;
  1598. }
  1599. /*
  1600. * syndrome mapping table for ECC ChipKill devices
  1601. *
  1602. * The comment in each row is the token (nibble) number that is in error.
  1603. * The least significant nibble of the syndrome is the mask for the bits
  1604. * that are in error (need to be toggled) for the particular nibble.
  1605. *
  1606. * Each row contains 16 entries.
  1607. * The first entry (0th) is the channel number for that row of syndromes.
  1608. * The remaining 15 entries are the syndromes for the respective Error
  1609. * bit mask index.
  1610. *
  1611. * 1st index entry is 0x0001 mask, indicating that the rightmost bit is the
  1612. * bit in error.
  1613. * The 2nd index entry is 0x0010 that the second bit is damaged.
  1614. * The 3rd index entry is 0x0011 indicating that the rightmost 2 bits
  1615. * are damaged.
  1616. * Thus so on until index 15, 0x1111, whose entry has the syndrome
  1617. * indicating that all 4 bits are damaged.
  1618. *
  1619. * A search is performed on this table looking for a given syndrome.
  1620. *
  1621. * See the AMD documentation for ECC syndromes. This ECC table is valid
  1622. * across all the versions of the AMD64 processors.
  1623. *
  1624. * A fast lookup is to use the LAST four bits of the 16-bit syndrome as a
  1625. * COLUMN index, then search all ROWS of that column, looking for a match
  1626. * with the input syndrome. The ROW value will be the token number.
  1627. *
  1628. * The 0'th entry on that row, can be returned as the CHANNEL (0 or 1) of this
  1629. * error.
  1630. */
  1631. #define NUMBER_ECC_ROWS 36
  1632. static const unsigned short ecc_chipkill_syndromes[NUMBER_ECC_ROWS][16] = {
  1633. /* Channel 0 syndromes */
  1634. {/*0*/ 0, 0xe821, 0x7c32, 0x9413, 0xbb44, 0x5365, 0xc776, 0x2f57,
  1635. 0xdd88, 0x35a9, 0xa1ba, 0x499b, 0x66cc, 0x8eed, 0x1afe, 0xf2df },
  1636. {/*1*/ 0, 0x5d31, 0xa612, 0xfb23, 0x9584, 0xc8b5, 0x3396, 0x6ea7,
  1637. 0xeac8, 0xb7f9, 0x4cda, 0x11eb, 0x7f4c, 0x227d, 0xd95e, 0x846f },
  1638. {/*2*/ 0, 0x0001, 0x0002, 0x0003, 0x0004, 0x0005, 0x0006, 0x0007,
  1639. 0x0008, 0x0009, 0x000a, 0x000b, 0x000c, 0x000d, 0x000e, 0x000f },
  1640. {/*3*/ 0, 0x2021, 0x3032, 0x1013, 0x4044, 0x6065, 0x7076, 0x5057,
  1641. 0x8088, 0xa0a9, 0xb0ba, 0x909b, 0xc0cc, 0xe0ed, 0xf0fe, 0xd0df },
  1642. {/*4*/ 0, 0x5041, 0xa082, 0xf0c3, 0x9054, 0xc015, 0x30d6, 0x6097,
  1643. 0xe0a8, 0xb0e9, 0x402a, 0x106b, 0x70fc, 0x20bd, 0xd07e, 0x803f },
  1644. {/*5*/ 0, 0xbe21, 0xd732, 0x6913, 0x2144, 0x9f65, 0xf676, 0x4857,
  1645. 0x3288, 0x8ca9, 0xe5ba, 0x5b9b, 0x13cc, 0xaded, 0xc4fe, 0x7adf },
  1646. {/*6*/ 0, 0x4951, 0x8ea2, 0xc7f3, 0x5394, 0x1ac5, 0xdd36, 0x9467,
  1647. 0xa1e8, 0xe8b9, 0x2f4a, 0x661b, 0xf27c, 0xbb2d, 0x7cde, 0x358f },
  1648. {/*7*/ 0, 0x74e1, 0x9872, 0xec93, 0xd6b4, 0xa255, 0x4ec6, 0x3a27,
  1649. 0x6bd8, 0x1f39, 0xf3aa, 0x874b, 0xbd6c, 0xc98d, 0x251e, 0x51ff },
  1650. {/*8*/ 0, 0x15c1, 0x2a42, 0x3f83, 0xcef4, 0xdb35, 0xe4b6, 0xf177,
  1651. 0x4758, 0x5299, 0x6d1a, 0x78db, 0x89ac, 0x9c6d, 0xa3ee, 0xb62f },
  1652. {/*9*/ 0, 0x3d01, 0x1602, 0x2b03, 0x8504, 0xb805, 0x9306, 0xae07,
  1653. 0xca08, 0xf709, 0xdc0a, 0xe10b, 0x4f0c, 0x720d, 0x590e, 0x640f },
  1654. {/*a*/ 0, 0x9801, 0xec02, 0x7403, 0x6b04, 0xf305, 0x8706, 0x1f07,
  1655. 0xbd08, 0x2509, 0x510a, 0xc90b, 0xd60c, 0x4e0d, 0x3a0e, 0xa20f },
  1656. {/*b*/ 0, 0xd131, 0x6212, 0xb323, 0x3884, 0xe9b5, 0x5a96, 0x8ba7,
  1657. 0x1cc8, 0xcdf9, 0x7eda, 0xafeb, 0x244c, 0xf57d, 0x465e, 0x976f },
  1658. {/*c*/ 0, 0xe1d1, 0x7262, 0x93b3, 0xb834, 0x59e5, 0xca56, 0x2b87,
  1659. 0xdc18, 0x3dc9, 0xae7a, 0x4fab, 0x542c, 0x85fd, 0x164e, 0xf79f },
  1660. {/*d*/ 0, 0x6051, 0xb0a2, 0xd0f3, 0x1094, 0x70c5, 0xa036, 0xc067,
  1661. 0x20e8, 0x40b9, 0x904a, 0x601b, 0x307c, 0x502d, 0x80de, 0xe08f },
  1662. {/*e*/ 0, 0xa4c1, 0xf842, 0x5c83, 0xe6f4, 0x4235, 0x1eb6, 0xba77,
  1663. 0x7b58, 0xdf99, 0x831a, 0x27db, 0x9dac, 0x396d, 0x65ee, 0xc12f },
  1664. {/*f*/ 0, 0x11c1, 0x2242, 0x3383, 0xc8f4, 0xd935, 0xeab6, 0xfb77,
  1665. 0x4c58, 0x5d99, 0x6e1a, 0x7fdb, 0x84ac, 0x956d, 0xa6ee, 0xb72f },
  1666. /* Channel 1 syndromes */
  1667. {/*10*/ 1, 0x45d1, 0x8a62, 0xcfb3, 0x5e34, 0x1be5, 0xd456, 0x9187,
  1668. 0xa718, 0xe2c9, 0x2d7a, 0x68ab, 0xf92c, 0xbcfd, 0x734e, 0x369f },
  1669. {/*11*/ 1, 0x63e1, 0xb172, 0xd293, 0x14b4, 0x7755, 0xa5c6, 0xc627,
  1670. 0x28d8, 0x4b39, 0x99aa, 0xfa4b, 0x3c6c, 0x5f8d, 0x8d1e, 0xeeff },
  1671. {/*12*/ 1, 0xb741, 0xd982, 0x6ec3, 0x2254, 0x9515, 0xfbd6, 0x4c97,
  1672. 0x33a8, 0x84e9, 0xea2a, 0x5d6b, 0x11fc, 0xa6bd, 0xc87e, 0x7f3f },
  1673. {/*13*/ 1, 0xdd41, 0x6682, 0xbbc3, 0x3554, 0xe815, 0x53d6, 0xce97,
  1674. 0x1aa8, 0xc7e9, 0x7c2a, 0xa1fb, 0x2ffc, 0xf2bd, 0x497e, 0x943f },
  1675. {/*14*/ 1, 0x2bd1, 0x3d62, 0x16b3, 0x4f34, 0x64e5, 0x7256, 0x5987,
  1676. 0x8518, 0xaec9, 0xb87a, 0x93ab, 0xca2c, 0xe1fd, 0xf74e, 0xdc9f },
  1677. {/*15*/ 1, 0x83c1, 0xc142, 0x4283, 0xa4f4, 0x2735, 0x65b6, 0xe677,
  1678. 0xf858, 0x7b99, 0x391a, 0xbadb, 0x5cac, 0xdf6d, 0x9dee, 0x1e2f },
  1679. {/*16*/ 1, 0x8fd1, 0xc562, 0x4ab3, 0xa934, 0x26e5, 0x6c56, 0xe387,
  1680. 0xfe18, 0x71c9, 0x3b7a, 0xb4ab, 0x572c, 0xd8fd, 0x924e, 0x1d9f },
  1681. {/*17*/ 1, 0x4791, 0x89e2, 0xce73, 0x5264, 0x15f5, 0xdb86, 0x9c17,
  1682. 0xa3b8, 0xe429, 0x2a5a, 0x6dcb, 0xf1dc, 0xb64d, 0x783e, 0x3faf },
  1683. {/*18*/ 1, 0x5781, 0xa9c2, 0xfe43, 0x92a4, 0xc525, 0x3b66, 0x6ce7,
  1684. 0xe3f8, 0xb479, 0x4a3a, 0x1dbb, 0x715c, 0x26dd, 0xd89e, 0x8f1f },
  1685. {/*19*/ 1, 0xbf41, 0xd582, 0x6ac3, 0x2954, 0x9615, 0xfcd6, 0x4397,
  1686. 0x3ea8, 0x81e9, 0xeb2a, 0x546b, 0x17fc, 0xa8bd, 0xc27e, 0x7d3f },
  1687. {/*1a*/ 1, 0x9891, 0xe1e2, 0x7273, 0x6464, 0xf7f5, 0x8586, 0x1617,
  1688. 0xb8b8, 0x2b29, 0x595a, 0xcacb, 0xdcdc, 0x4f4d, 0x3d3e, 0xaeaf },
  1689. {/*1b*/ 1, 0xcce1, 0x4472, 0x8893, 0xfdb4, 0x3f55, 0xb9c6, 0x7527,
  1690. 0x56d8, 0x9a39, 0x12aa, 0xde4b, 0xab6c, 0x678d, 0xef1e, 0x23ff },
  1691. {/*1c*/ 1, 0xa761, 0xf9b2, 0x5ed3, 0xe214, 0x4575, 0x1ba6, 0xbcc7,
  1692. 0x7328, 0xd449, 0x8a9a, 0x2dfb, 0x913c, 0x365d, 0x688e, 0xcfef },
  1693. {/*1d*/ 1, 0xff61, 0x55b2, 0xaad3, 0x7914, 0x8675, 0x2ca6, 0xd3c7,
  1694. 0x9e28, 0x6149, 0xcb9a, 0x34fb, 0xe73c, 0x185d, 0xb28e, 0x4def },
  1695. {/*1e*/ 1, 0x5451, 0xa8a2, 0xfcf3, 0x9694, 0xc2c5, 0x3e36, 0x6a67,
  1696. 0xebe8, 0xbfb9, 0x434a, 0x171b, 0x7d7c, 0x292d, 0xd5de, 0x818f },
  1697. {/*1f*/ 1, 0x6fc1, 0xb542, 0xda83, 0x19f4, 0x7635, 0xacb6, 0xc377,
  1698. 0x2e58, 0x4199, 0x9b1a, 0xf4db, 0x37ac, 0x586d, 0x82ee, 0xed2f },
  1699. /* ECC bits are also in the set of tokens and they too can go bad
  1700. * first 2 cover channel 0, while the second 2 cover channel 1
  1701. */
  1702. {/*20*/ 0, 0xbe01, 0xd702, 0x6903, 0x2104, 0x9f05, 0xf606, 0x4807,
  1703. 0x3208, 0x8c09, 0xe50a, 0x5b0b, 0x130c, 0xad0d, 0xc40e, 0x7a0f },
  1704. {/*21*/ 0, 0x4101, 0x8202, 0xc303, 0x5804, 0x1905, 0xda06, 0x9b07,
  1705. 0xac08, 0xed09, 0x2e0a, 0x6f0b, 0x640c, 0xb50d, 0x760e, 0x370f },
  1706. {/*22*/ 1, 0xc441, 0x4882, 0x8cc3, 0xf654, 0x3215, 0xbed6, 0x7a97,
  1707. 0x5ba8, 0x9fe9, 0x132a, 0xd76b, 0xadfc, 0x69bd, 0xe57e, 0x213f },
  1708. {/*23*/ 1, 0x7621, 0x9b32, 0xed13, 0xda44, 0xac65, 0x4176, 0x3757,
  1709. 0x6f88, 0x19a9, 0xf4ba, 0x829b, 0xb5cc, 0xc3ed, 0x2efe, 0x58df }
  1710. };
  1711. /*
  1712. * Given the syndrome argument, scan each of the channel tables for a syndrome
  1713. * match. Depending on which table it is found, return the channel number.
  1714. */
  1715. static int get_channel_from_ecc_syndrome(unsigned short syndrome)
  1716. {
  1717. int row;
  1718. int column;
  1719. /* Determine column to scan */
  1720. column = syndrome & 0xF;
  1721. /* Scan all rows, looking for syndrome, or end of table */
  1722. for (row = 0; row < NUMBER_ECC_ROWS; row++) {
  1723. if (ecc_chipkill_syndromes[row][column] == syndrome)
  1724. return ecc_chipkill_syndromes[row][0];
  1725. }
  1726. debugf0("syndrome(%x) not found\n", syndrome);
  1727. return -1;
  1728. }
  1729. /*
  1730. * Check for valid error in the NB Status High register. If so, proceed to read
  1731. * NB Status Low, NB Address Low and NB Address High registers and store data
  1732. * into error structure.
  1733. *
  1734. * Returns:
  1735. * - 1: if hardware regs contains valid error info
  1736. * - 0: if no valid error is indicated
  1737. */
  1738. static int amd64_get_error_info_regs(struct mem_ctl_info *mci,
  1739. struct amd64_error_info_regs *regs)
  1740. {
  1741. struct amd64_pvt *pvt;
  1742. struct pci_dev *misc_f3_ctl;
  1743. int err = 0;
  1744. pvt = mci->pvt_info;
  1745. misc_f3_ctl = pvt->misc_f3_ctl;
  1746. err = pci_read_config_dword(misc_f3_ctl, K8_NBSH, &regs->nbsh);
  1747. if (err)
  1748. goto err_reg;
  1749. if (!(regs->nbsh & K8_NBSH_VALID_BIT))
  1750. return 0;
  1751. /* valid error, read remaining error information registers */
  1752. err = pci_read_config_dword(misc_f3_ctl, K8_NBSL, &regs->nbsl);
  1753. if (err)
  1754. goto err_reg;
  1755. err = pci_read_config_dword(misc_f3_ctl, K8_NBEAL, &regs->nbeal);
  1756. if (err)
  1757. goto err_reg;
  1758. err = pci_read_config_dword(misc_f3_ctl, K8_NBEAH, &regs->nbeah);
  1759. if (err)
  1760. goto err_reg;
  1761. err = pci_read_config_dword(misc_f3_ctl, K8_NBCFG, &regs->nbcfg);
  1762. if (err)
  1763. goto err_reg;
  1764. return 1;
  1765. err_reg:
  1766. debugf0("Reading error info register failed\n");
  1767. return 0;
  1768. }
  1769. /*
  1770. * This function is called to retrieve the error data from hardware and store it
  1771. * in the info structure.
  1772. *
  1773. * Returns:
  1774. * - 1: if a valid error is found
  1775. * - 0: if no error is found
  1776. */
  1777. static int amd64_get_error_info(struct mem_ctl_info *mci,
  1778. struct amd64_error_info_regs *info)
  1779. {
  1780. struct amd64_pvt *pvt;
  1781. struct amd64_error_info_regs regs;
  1782. pvt = mci->pvt_info;
  1783. if (!amd64_get_error_info_regs(mci, info))
  1784. return 0;
  1785. /*
  1786. * Here's the problem with the K8's EDAC reporting: There are four
  1787. * registers which report pieces of error information. They are shared
  1788. * between CEs and UEs. Furthermore, contrary to what is stated in the
  1789. * BKDG, the overflow bit is never used! Every error always updates the
  1790. * reporting registers.
  1791. *
  1792. * Can you see the race condition? All four error reporting registers
  1793. * must be read before a new error updates them! There is no way to read
  1794. * all four registers atomically. The best than can be done is to detect
  1795. * that a race has occured and then report the error without any kind of
  1796. * precision.
  1797. *
  1798. * What is still positive is that errors are still reported and thus
  1799. * problems can still be detected - just not localized because the
  1800. * syndrome and address are spread out across registers.
  1801. *
  1802. * Grrrrr!!!!! Here's hoping that AMD fixes this in some future K8 rev.
  1803. * UEs and CEs should have separate register sets with proper overflow
  1804. * bits that are used! At very least the problem can be fixed by
  1805. * honoring the ErrValid bit in 'nbsh' and not updating registers - just
  1806. * set the overflow bit - unless the current error is CE and the new
  1807. * error is UE which would be the only situation for overwriting the
  1808. * current values.
  1809. */
  1810. regs = *info;
  1811. /* Use info from the second read - most current */
  1812. if (unlikely(!amd64_get_error_info_regs(mci, info)))
  1813. return 0;
  1814. /* clear the error bits in hardware */
  1815. pci_write_bits32(pvt->misc_f3_ctl, K8_NBSH, 0, K8_NBSH_VALID_BIT);
  1816. /* Check for the possible race condition */
  1817. if ((regs.nbsh != info->nbsh) ||
  1818. (regs.nbsl != info->nbsl) ||
  1819. (regs.nbeah != info->nbeah) ||
  1820. (regs.nbeal != info->nbeal)) {
  1821. amd64_mc_printk(mci, KERN_WARNING,
  1822. "hardware STATUS read access race condition "
  1823. "detected!\n");
  1824. return 0;
  1825. }
  1826. return 1;
  1827. }
  1828. static inline void amd64_decode_gart_tlb_error(struct mem_ctl_info *mci,
  1829. struct amd64_error_info_regs *info)
  1830. {
  1831. u32 err_code;
  1832. u32 ec_tt; /* error code transaction type (2b) */
  1833. u32 ec_ll; /* error code cache level (2b) */
  1834. err_code = EXTRACT_ERROR_CODE(info->nbsl);
  1835. ec_ll = EXTRACT_LL_CODE(err_code);
  1836. ec_tt = EXTRACT_TT_CODE(err_code);
  1837. amd64_mc_printk(mci, KERN_ERR,
  1838. "GART TLB event: transaction type(%s), "
  1839. "cache level(%s)\n", tt_msgs[ec_tt], ll_msgs[ec_ll]);
  1840. }
  1841. static inline void amd64_decode_mem_cache_error(struct mem_ctl_info *mci,
  1842. struct amd64_error_info_regs *info)
  1843. {
  1844. u32 err_code;
  1845. u32 ec_rrrr; /* error code memory transaction (4b) */
  1846. u32 ec_tt; /* error code transaction type (2b) */
  1847. u32 ec_ll; /* error code cache level (2b) */
  1848. err_code = EXTRACT_ERROR_CODE(info->nbsl);
  1849. ec_ll = EXTRACT_LL_CODE(err_code);
  1850. ec_tt = EXTRACT_TT_CODE(err_code);
  1851. ec_rrrr = EXTRACT_RRRR_CODE(err_code);
  1852. amd64_mc_printk(mci, KERN_ERR,
  1853. "cache hierarchy error: memory transaction type(%s), "
  1854. "transaction type(%s), cache level(%s)\n",
  1855. rrrr_msgs[ec_rrrr], tt_msgs[ec_tt], ll_msgs[ec_ll]);
  1856. }
  1857. /*
  1858. * Handle any Correctable Errors (CEs) that have occurred. Check for valid ERROR
  1859. * ADDRESS and process.
  1860. */
  1861. static void amd64_handle_ce(struct mem_ctl_info *mci,
  1862. struct amd64_error_info_regs *info)
  1863. {
  1864. struct amd64_pvt *pvt = mci->pvt_info;
  1865. u64 SystemAddress;
  1866. /* Ensure that the Error Address is VALID */
  1867. if ((info->nbsh & K8_NBSH_VALID_ERROR_ADDR) == 0) {
  1868. amd64_mc_printk(mci, KERN_ERR,
  1869. "HW has no ERROR_ADDRESS available\n");
  1870. edac_mc_handle_ce_no_info(mci, EDAC_MOD_STR);
  1871. return;
  1872. }
  1873. SystemAddress = extract_error_address(mci, info);
  1874. amd64_mc_printk(mci, KERN_ERR,
  1875. "CE ERROR_ADDRESS= 0x%llx\n", SystemAddress);
  1876. pvt->ops->map_sysaddr_to_csrow(mci, info, SystemAddress);
  1877. }
  1878. /* Handle any Un-correctable Errors (UEs) */
  1879. static void amd64_handle_ue(struct mem_ctl_info *mci,
  1880. struct amd64_error_info_regs *info)
  1881. {
  1882. int csrow;
  1883. u64 SystemAddress;
  1884. u32 page, offset;
  1885. struct mem_ctl_info *log_mci, *src_mci = NULL;
  1886. log_mci = mci;
  1887. if ((info->nbsh & K8_NBSH_VALID_ERROR_ADDR) == 0) {
  1888. amd64_mc_printk(mci, KERN_CRIT,
  1889. "HW has no ERROR_ADDRESS available\n");
  1890. edac_mc_handle_ue_no_info(log_mci, EDAC_MOD_STR);
  1891. return;
  1892. }
  1893. SystemAddress = extract_error_address(mci, info);
  1894. /*
  1895. * Find out which node the error address belongs to. This may be
  1896. * different from the node that detected the error.
  1897. */
  1898. src_mci = find_mc_by_sys_addr(mci, SystemAddress);
  1899. if (!src_mci) {
  1900. amd64_mc_printk(mci, KERN_CRIT,
  1901. "ERROR ADDRESS (0x%lx) value NOT mapped to a MC\n",
  1902. (unsigned long)SystemAddress);
  1903. edac_mc_handle_ue_no_info(log_mci, EDAC_MOD_STR);
  1904. return;
  1905. }
  1906. log_mci = src_mci;
  1907. csrow = sys_addr_to_csrow(log_mci, SystemAddress);
  1908. if (csrow < 0) {
  1909. amd64_mc_printk(mci, KERN_CRIT,
  1910. "ERROR_ADDRESS (0x%lx) value NOT mapped to 'csrow'\n",
  1911. (unsigned long)SystemAddress);
  1912. edac_mc_handle_ue_no_info(log_mci, EDAC_MOD_STR);
  1913. } else {
  1914. error_address_to_page_and_offset(SystemAddress, &page, &offset);
  1915. edac_mc_handle_ue(log_mci, page, offset, csrow, EDAC_MOD_STR);
  1916. }
  1917. }
  1918. static void amd64_decode_bus_error(struct mem_ctl_info *mci,
  1919. struct amd64_error_info_regs *info)
  1920. {
  1921. u32 err_code, ext_ec;
  1922. u32 ec_pp; /* error code participating processor (2p) */
  1923. u32 ec_to; /* error code timed out (1b) */
  1924. u32 ec_rrrr; /* error code memory transaction (4b) */
  1925. u32 ec_ii; /* error code memory or I/O (2b) */
  1926. u32 ec_ll; /* error code cache level (2b) */
  1927. ext_ec = EXTRACT_EXT_ERROR_CODE(info->nbsl);
  1928. err_code = EXTRACT_ERROR_CODE(info->nbsl);
  1929. ec_ll = EXTRACT_LL_CODE(err_code);
  1930. ec_ii = EXTRACT_II_CODE(err_code);
  1931. ec_rrrr = EXTRACT_RRRR_CODE(err_code);
  1932. ec_to = EXTRACT_TO_CODE(err_code);
  1933. ec_pp = EXTRACT_PP_CODE(err_code);
  1934. amd64_mc_printk(mci, KERN_ERR,
  1935. "BUS ERROR:\n"
  1936. " time-out(%s) mem or i/o(%s)\n"
  1937. " participating processor(%s)\n"
  1938. " memory transaction type(%s)\n"
  1939. " cache level(%s) Error Found by: %s\n",
  1940. to_msgs[ec_to],
  1941. ii_msgs[ec_ii],
  1942. pp_msgs[ec_pp],
  1943. rrrr_msgs[ec_rrrr],
  1944. ll_msgs[ec_ll],
  1945. (info->nbsh & K8_NBSH_ERR_SCRUBER) ?
  1946. "Scrubber" : "Normal Operation");
  1947. /* If this was an 'observed' error, early out */
  1948. if (ec_pp == K8_NBSL_PP_OBS)
  1949. return; /* We aren't the node involved */
  1950. /* Parse out the extended error code for ECC events */
  1951. switch (ext_ec) {
  1952. /* F10 changed to one Extended ECC error code */
  1953. case F10_NBSL_EXT_ERR_RES: /* Reserved field */
  1954. case F10_NBSL_EXT_ERR_ECC: /* F10 ECC ext err code */
  1955. break;
  1956. default:
  1957. amd64_mc_printk(mci, KERN_ERR, "NOT ECC: no special error "
  1958. "handling for this error\n");
  1959. return;
  1960. }
  1961. if (info->nbsh & K8_NBSH_CECC)
  1962. amd64_handle_ce(mci, info);
  1963. else if (info->nbsh & K8_NBSH_UECC)
  1964. amd64_handle_ue(mci, info);
  1965. /*
  1966. * If main error is CE then overflow must be CE. If main error is UE
  1967. * then overflow is unknown. We'll call the overflow a CE - if
  1968. * panic_on_ue is set then we're already panic'ed and won't arrive
  1969. * here. Else, then apparently someone doesn't think that UE's are
  1970. * catastrophic.
  1971. */
  1972. if (info->nbsh & K8_NBSH_OVERFLOW)
  1973. edac_mc_handle_ce_no_info(mci, EDAC_MOD_STR
  1974. "Error Overflow set");
  1975. }
  1976. int amd64_process_error_info(struct mem_ctl_info *mci,
  1977. struct amd64_error_info_regs *info,
  1978. int handle_errors)
  1979. {
  1980. struct amd64_pvt *pvt;
  1981. struct amd64_error_info_regs *regs;
  1982. u32 err_code, ext_ec;
  1983. int gart_tlb_error = 0;
  1984. pvt = mci->pvt_info;
  1985. /* If caller doesn't want us to process the error, return */
  1986. if (!handle_errors)
  1987. return 1;
  1988. regs = info;
  1989. debugf1("NorthBridge ERROR: mci(0x%p)\n", mci);
  1990. debugf1(" MC node(%d) Error-Address(0x%.8x-%.8x)\n",
  1991. pvt->mc_node_id, regs->nbeah, regs->nbeal);
  1992. debugf1(" nbsh(0x%.8x) nbsl(0x%.8x)\n",
  1993. regs->nbsh, regs->nbsl);
  1994. debugf1(" Valid Error=%s Overflow=%s\n",
  1995. (regs->nbsh & K8_NBSH_VALID_BIT) ? "True" : "False",
  1996. (regs->nbsh & K8_NBSH_OVERFLOW) ? "True" : "False");
  1997. debugf1(" Err Uncorrected=%s MCA Error Reporting=%s\n",
  1998. (regs->nbsh & K8_NBSH_UNCORRECTED_ERR) ?
  1999. "True" : "False",
  2000. (regs->nbsh & K8_NBSH_ERR_ENABLE) ?
  2001. "True" : "False");
  2002. debugf1(" MiscErr Valid=%s ErrAddr Valid=%s PCC=%s\n",
  2003. (regs->nbsh & K8_NBSH_MISC_ERR_VALID) ?
  2004. "True" : "False",
  2005. (regs->nbsh & K8_NBSH_VALID_ERROR_ADDR) ?
  2006. "True" : "False",
  2007. (regs->nbsh & K8_NBSH_PCC) ?
  2008. "True" : "False");
  2009. debugf1(" CECC=%s UECC=%s Found by Scruber=%s\n",
  2010. (regs->nbsh & K8_NBSH_CECC) ?
  2011. "True" : "False",
  2012. (regs->nbsh & K8_NBSH_UECC) ?
  2013. "True" : "False",
  2014. (regs->nbsh & K8_NBSH_ERR_SCRUBER) ?
  2015. "True" : "False");
  2016. debugf1(" CORE0=%s CORE1=%s CORE2=%s CORE3=%s\n",
  2017. (regs->nbsh & K8_NBSH_CORE0) ? "True" : "False",
  2018. (regs->nbsh & K8_NBSH_CORE1) ? "True" : "False",
  2019. (regs->nbsh & K8_NBSH_CORE2) ? "True" : "False",
  2020. (regs->nbsh & K8_NBSH_CORE3) ? "True" : "False");
  2021. err_code = EXTRACT_ERROR_CODE(regs->nbsl);
  2022. /* Determine which error type:
  2023. * 1) GART errors - non-fatal, developmental events
  2024. * 2) MEMORY errors
  2025. * 3) BUS errors
  2026. * 4) Unknown error
  2027. */
  2028. if (TEST_TLB_ERROR(err_code)) {
  2029. /*
  2030. * GART errors are intended to help graphics driver developers
  2031. * to detect bad GART PTEs. It is recommended by AMD to disable
  2032. * GART table walk error reporting by default[1] (currently
  2033. * being disabled in mce_cpu_quirks()) and according to the
  2034. * comment in mce_cpu_quirks(), such GART errors can be
  2035. * incorrectly triggered. We may see these errors anyway and
  2036. * unless requested by the user, they won't be reported.
  2037. *
  2038. * [1] section 13.10.1 on BIOS and Kernel Developers Guide for
  2039. * AMD NPT family 0Fh processors
  2040. */
  2041. if (report_gart_errors == 0)
  2042. return 1;
  2043. /*
  2044. * Only if GART error reporting is requested should we generate
  2045. * any logs.
  2046. */
  2047. gart_tlb_error = 1;
  2048. debugf1("GART TLB error\n");
  2049. amd64_decode_gart_tlb_error(mci, info);
  2050. } else if (TEST_MEM_ERROR(err_code)) {
  2051. debugf1("Memory/Cache error\n");
  2052. amd64_decode_mem_cache_error(mci, info);
  2053. } else if (TEST_BUS_ERROR(err_code)) {
  2054. debugf1("Bus (Link/DRAM) error\n");
  2055. amd64_decode_bus_error(mci, info);
  2056. } else {
  2057. /* shouldn't reach here! */
  2058. amd64_mc_printk(mci, KERN_WARNING,
  2059. "%s(): unknown MCE error 0x%x\n", __func__,
  2060. err_code);
  2061. }
  2062. ext_ec = EXTRACT_EXT_ERROR_CODE(regs->nbsl);
  2063. amd64_mc_printk(mci, KERN_ERR,
  2064. "ExtErr=(0x%x) %s\n", ext_ec, ext_msgs[ext_ec]);
  2065. if (((ext_ec >= F10_NBSL_EXT_ERR_CRC &&
  2066. ext_ec <= F10_NBSL_EXT_ERR_TGT) ||
  2067. (ext_ec == F10_NBSL_EXT_ERR_RMW)) &&
  2068. EXTRACT_LDT_LINK(info->nbsh)) {
  2069. amd64_mc_printk(mci, KERN_ERR,
  2070. "Error on hypertransport link: %s\n",
  2071. htlink_msgs[
  2072. EXTRACT_LDT_LINK(info->nbsh)]);
  2073. }
  2074. /*
  2075. * Check the UE bit of the NB status high register, if set generate some
  2076. * logs. If NOT a GART error, then process the event as a NO-INFO event.
  2077. * If it was a GART error, skip that process.
  2078. */
  2079. if (regs->nbsh & K8_NBSH_UNCORRECTED_ERR) {
  2080. amd64_mc_printk(mci, KERN_CRIT, "uncorrected error\n");
  2081. if (!gart_tlb_error)
  2082. edac_mc_handle_ue_no_info(mci, "UE bit is set\n");
  2083. }
  2084. if (regs->nbsh & K8_NBSH_PCC)
  2085. amd64_mc_printk(mci, KERN_CRIT,
  2086. "PCC (processor context corrupt) set\n");
  2087. return 1;
  2088. }
  2089. EXPORT_SYMBOL_GPL(amd64_process_error_info);