pci.c 42 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783784785786787788789790791792793794795796797798799800801802803804805806807808809810811812813814815816817818819820821822823824825826827828829830831832833834835836837838839840841842843844845846847848849850851852853854855856857858859860861862863864865866867868869870871872873874875876877878879880881882883884885886887888889890891892893894895896897898899900901902903904905906907908909910911912913914915916917918919920921922923924925926927928929930931932933934935936937938939940941942943944945946947948949950951952953954955956957958959960961962963964965966967968969970971972973974975976977978979980981982983984985986987988989990991992993994995996997998999100010011002100310041005100610071008100910101011101210131014101510161017101810191020102110221023102410251026102710281029103010311032103310341035103610371038103910401041104210431044104510461047104810491050105110521053105410551056105710581059106010611062106310641065106610671068106910701071107210731074107510761077107810791080108110821083108410851086108710881089109010911092109310941095109610971098109911001101110211031104110511061107110811091110111111121113111411151116111711181119112011211122112311241125112611271128112911301131113211331134113511361137113811391140114111421143114411451146114711481149115011511152115311541155115611571158115911601161116211631164116511661167116811691170117111721173117411751176117711781179118011811182118311841185118611871188118911901191119211931194119511961197119811991200120112021203120412051206120712081209121012111212121312141215121612171218121912201221122212231224122512261227122812291230123112321233123412351236123712381239124012411242124312441245124612471248124912501251125212531254125512561257125812591260126112621263126412651266126712681269127012711272127312741275127612771278127912801281128212831284128512861287128812891290129112921293129412951296129712981299130013011302130313041305130613071308130913101311131213131314131513161317131813191320132113221323132413251326132713281329133013311332133313341335133613371338133913401341134213431344134513461347134813491350135113521353135413551356135713581359136013611362136313641365136613671368136913701371137213731374137513761377137813791380138113821383138413851386138713881389139013911392139313941395139613971398139914001401140214031404140514061407140814091410141114121413141414151416141714181419142014211422142314241425142614271428142914301431143214331434143514361437143814391440144114421443144414451446144714481449145014511452145314541455145614571458145914601461146214631464146514661467146814691470147114721473147414751476147714781479148014811482148314841485148614871488148914901491149214931494149514961497149814991500150115021503150415051506150715081509151015111512151315141515151615171518151915201521152215231524152515261527152815291530153115321533153415351536153715381539154015411542154315441545154615471548154915501551
  1. /*
  2. * Copyright 2014 IBM Corp.
  3. *
  4. * This program is free software; you can redistribute it and/or
  5. * modify it under the terms of the GNU General Public License
  6. * as published by the Free Software Foundation; either version
  7. * 2 of the License, or (at your option) any later version.
  8. */
  9. #include <linux/pci_regs.h>
  10. #include <linux/pci_ids.h>
  11. #include <linux/device.h>
  12. #include <linux/module.h>
  13. #include <linux/kernel.h>
  14. #include <linux/slab.h>
  15. #include <linux/sort.h>
  16. #include <linux/pci.h>
  17. #include <linux/of.h>
  18. #include <linux/delay.h>
  19. #include <asm/opal.h>
  20. #include <asm/msi_bitmap.h>
  21. #include <asm/pci-bridge.h> /* for struct pci_controller */
  22. #include <asm/pnv-pci.h>
  23. #include <asm/io.h>
  24. #include "cxl.h"
  25. #include <misc/cxl.h>
  26. #define CXL_PCI_VSEC_ID 0x1280
  27. #define CXL_VSEC_MIN_SIZE 0x80
  28. #define CXL_READ_VSEC_LENGTH(dev, vsec, dest) \
  29. { \
  30. pci_read_config_word(dev, vsec + 0x6, dest); \
  31. *dest >>= 4; \
  32. }
  33. #define CXL_READ_VSEC_NAFUS(dev, vsec, dest) \
  34. pci_read_config_byte(dev, vsec + 0x8, dest)
  35. #define CXL_READ_VSEC_STATUS(dev, vsec, dest) \
  36. pci_read_config_byte(dev, vsec + 0x9, dest)
  37. #define CXL_STATUS_SECOND_PORT 0x80
  38. #define CXL_STATUS_MSI_X_FULL 0x40
  39. #define CXL_STATUS_MSI_X_SINGLE 0x20
  40. #define CXL_STATUS_FLASH_RW 0x08
  41. #define CXL_STATUS_FLASH_RO 0x04
  42. #define CXL_STATUS_LOADABLE_AFU 0x02
  43. #define CXL_STATUS_LOADABLE_PSL 0x01
  44. /* If we see these features we won't try to use the card */
  45. #define CXL_UNSUPPORTED_FEATURES \
  46. (CXL_STATUS_MSI_X_FULL | CXL_STATUS_MSI_X_SINGLE)
  47. #define CXL_READ_VSEC_MODE_CONTROL(dev, vsec, dest) \
  48. pci_read_config_byte(dev, vsec + 0xa, dest)
  49. #define CXL_WRITE_VSEC_MODE_CONTROL(dev, vsec, val) \
  50. pci_write_config_byte(dev, vsec + 0xa, val)
  51. #define CXL_VSEC_PROTOCOL_MASK 0xe0
  52. #define CXL_VSEC_PROTOCOL_1024TB 0x80
  53. #define CXL_VSEC_PROTOCOL_512TB 0x40
  54. #define CXL_VSEC_PROTOCOL_256TB 0x20 /* Power 8 uses this */
  55. #define CXL_VSEC_PROTOCOL_ENABLE 0x01
  56. #define CXL_READ_VSEC_PSL_REVISION(dev, vsec, dest) \
  57. pci_read_config_word(dev, vsec + 0xc, dest)
  58. #define CXL_READ_VSEC_CAIA_MINOR(dev, vsec, dest) \
  59. pci_read_config_byte(dev, vsec + 0xe, dest)
  60. #define CXL_READ_VSEC_CAIA_MAJOR(dev, vsec, dest) \
  61. pci_read_config_byte(dev, vsec + 0xf, dest)
  62. #define CXL_READ_VSEC_BASE_IMAGE(dev, vsec, dest) \
  63. pci_read_config_word(dev, vsec + 0x10, dest)
  64. #define CXL_READ_VSEC_IMAGE_STATE(dev, vsec, dest) \
  65. pci_read_config_byte(dev, vsec + 0x13, dest)
  66. #define CXL_WRITE_VSEC_IMAGE_STATE(dev, vsec, val) \
  67. pci_write_config_byte(dev, vsec + 0x13, val)
  68. #define CXL_VSEC_USER_IMAGE_LOADED 0x80 /* RO */
  69. #define CXL_VSEC_PERST_LOADS_IMAGE 0x20 /* RW */
  70. #define CXL_VSEC_PERST_SELECT_USER 0x10 /* RW */
  71. #define CXL_READ_VSEC_AFU_DESC_OFF(dev, vsec, dest) \
  72. pci_read_config_dword(dev, vsec + 0x20, dest)
  73. #define CXL_READ_VSEC_AFU_DESC_SIZE(dev, vsec, dest) \
  74. pci_read_config_dword(dev, vsec + 0x24, dest)
  75. #define CXL_READ_VSEC_PS_OFF(dev, vsec, dest) \
  76. pci_read_config_dword(dev, vsec + 0x28, dest)
  77. #define CXL_READ_VSEC_PS_SIZE(dev, vsec, dest) \
  78. pci_read_config_dword(dev, vsec + 0x2c, dest)
  79. /* This works a little different than the p1/p2 register accesses to make it
  80. * easier to pull out individual fields */
  81. #define AFUD_READ(afu, off) in_be64(afu->afu_desc_mmio + off)
  82. #define AFUD_READ_LE(afu, off) in_le64(afu->afu_desc_mmio + off)
  83. #define EXTRACT_PPC_BIT(val, bit) (!!(val & PPC_BIT(bit)))
  84. #define EXTRACT_PPC_BITS(val, bs, be) ((val & PPC_BITMASK(bs, be)) >> PPC_BITLSHIFT(be))
  85. #define AFUD_READ_INFO(afu) AFUD_READ(afu, 0x0)
  86. #define AFUD_NUM_INTS_PER_PROC(val) EXTRACT_PPC_BITS(val, 0, 15)
  87. #define AFUD_NUM_PROCS(val) EXTRACT_PPC_BITS(val, 16, 31)
  88. #define AFUD_NUM_CRS(val) EXTRACT_PPC_BITS(val, 32, 47)
  89. #define AFUD_MULTIMODE(val) EXTRACT_PPC_BIT(val, 48)
  90. #define AFUD_PUSH_BLOCK_TRANSFER(val) EXTRACT_PPC_BIT(val, 55)
  91. #define AFUD_DEDICATED_PROCESS(val) EXTRACT_PPC_BIT(val, 59)
  92. #define AFUD_AFU_DIRECTED(val) EXTRACT_PPC_BIT(val, 61)
  93. #define AFUD_TIME_SLICED(val) EXTRACT_PPC_BIT(val, 63)
  94. #define AFUD_READ_CR(afu) AFUD_READ(afu, 0x20)
  95. #define AFUD_CR_LEN(val) EXTRACT_PPC_BITS(val, 8, 63)
  96. #define AFUD_READ_CR_OFF(afu) AFUD_READ(afu, 0x28)
  97. #define AFUD_READ_PPPSA(afu) AFUD_READ(afu, 0x30)
  98. #define AFUD_PPPSA_PP(val) EXTRACT_PPC_BIT(val, 6)
  99. #define AFUD_PPPSA_PSA(val) EXTRACT_PPC_BIT(val, 7)
  100. #define AFUD_PPPSA_LEN(val) EXTRACT_PPC_BITS(val, 8, 63)
  101. #define AFUD_READ_PPPSA_OFF(afu) AFUD_READ(afu, 0x38)
  102. #define AFUD_READ_EB(afu) AFUD_READ(afu, 0x40)
  103. #define AFUD_EB_LEN(val) EXTRACT_PPC_BITS(val, 8, 63)
  104. #define AFUD_READ_EB_OFF(afu) AFUD_READ(afu, 0x48)
  105. u16 cxl_afu_cr_read16(struct cxl_afu *afu, int cr, u64 off)
  106. {
  107. u64 aligned_off = off & ~0x3L;
  108. u32 val;
  109. val = cxl_afu_cr_read32(afu, cr, aligned_off);
  110. return (val >> ((off & 0x2) * 8)) & 0xffff;
  111. }
  112. u8 cxl_afu_cr_read8(struct cxl_afu *afu, int cr, u64 off)
  113. {
  114. u64 aligned_off = off & ~0x3L;
  115. u32 val;
  116. val = cxl_afu_cr_read32(afu, cr, aligned_off);
  117. return (val >> ((off & 0x3) * 8)) & 0xff;
  118. }
  119. static const struct pci_device_id cxl_pci_tbl[] = {
  120. { PCI_DEVICE(PCI_VENDOR_ID_IBM, 0x0477), },
  121. { PCI_DEVICE(PCI_VENDOR_ID_IBM, 0x044b), },
  122. { PCI_DEVICE(PCI_VENDOR_ID_IBM, 0x04cf), },
  123. { PCI_DEVICE_CLASS(0x120000, ~0), },
  124. { }
  125. };
  126. MODULE_DEVICE_TABLE(pci, cxl_pci_tbl);
  127. /*
  128. * Mostly using these wrappers to avoid confusion:
  129. * priv 1 is BAR2, while priv 2 is BAR0
  130. */
  131. static inline resource_size_t p1_base(struct pci_dev *dev)
  132. {
  133. return pci_resource_start(dev, 2);
  134. }
  135. static inline resource_size_t p1_size(struct pci_dev *dev)
  136. {
  137. return pci_resource_len(dev, 2);
  138. }
  139. static inline resource_size_t p2_base(struct pci_dev *dev)
  140. {
  141. return pci_resource_start(dev, 0);
  142. }
  143. static inline resource_size_t p2_size(struct pci_dev *dev)
  144. {
  145. return pci_resource_len(dev, 0);
  146. }
  147. static int find_cxl_vsec(struct pci_dev *dev)
  148. {
  149. int vsec = 0;
  150. u16 val;
  151. while ((vsec = pci_find_next_ext_capability(dev, vsec, PCI_EXT_CAP_ID_VNDR))) {
  152. pci_read_config_word(dev, vsec + 0x4, &val);
  153. if (val == CXL_PCI_VSEC_ID)
  154. return vsec;
  155. }
  156. return 0;
  157. }
  158. static void dump_cxl_config_space(struct pci_dev *dev)
  159. {
  160. int vsec;
  161. u32 val;
  162. dev_info(&dev->dev, "dump_cxl_config_space\n");
  163. pci_read_config_dword(dev, PCI_BASE_ADDRESS_0, &val);
  164. dev_info(&dev->dev, "BAR0: %#.8x\n", val);
  165. pci_read_config_dword(dev, PCI_BASE_ADDRESS_1, &val);
  166. dev_info(&dev->dev, "BAR1: %#.8x\n", val);
  167. pci_read_config_dword(dev, PCI_BASE_ADDRESS_2, &val);
  168. dev_info(&dev->dev, "BAR2: %#.8x\n", val);
  169. pci_read_config_dword(dev, PCI_BASE_ADDRESS_3, &val);
  170. dev_info(&dev->dev, "BAR3: %#.8x\n", val);
  171. pci_read_config_dword(dev, PCI_BASE_ADDRESS_4, &val);
  172. dev_info(&dev->dev, "BAR4: %#.8x\n", val);
  173. pci_read_config_dword(dev, PCI_BASE_ADDRESS_5, &val);
  174. dev_info(&dev->dev, "BAR5: %#.8x\n", val);
  175. dev_info(&dev->dev, "p1 regs: %#llx, len: %#llx\n",
  176. p1_base(dev), p1_size(dev));
  177. dev_info(&dev->dev, "p2 regs: %#llx, len: %#llx\n",
  178. p2_base(dev), p2_size(dev));
  179. dev_info(&dev->dev, "BAR 4/5: %#llx, len: %#llx\n",
  180. pci_resource_start(dev, 4), pci_resource_len(dev, 4));
  181. if (!(vsec = find_cxl_vsec(dev)))
  182. return;
  183. #define show_reg(name, what) \
  184. dev_info(&dev->dev, "cxl vsec: %30s: %#x\n", name, what)
  185. pci_read_config_dword(dev, vsec + 0x0, &val);
  186. show_reg("Cap ID", (val >> 0) & 0xffff);
  187. show_reg("Cap Ver", (val >> 16) & 0xf);
  188. show_reg("Next Cap Ptr", (val >> 20) & 0xfff);
  189. pci_read_config_dword(dev, vsec + 0x4, &val);
  190. show_reg("VSEC ID", (val >> 0) & 0xffff);
  191. show_reg("VSEC Rev", (val >> 16) & 0xf);
  192. show_reg("VSEC Length", (val >> 20) & 0xfff);
  193. pci_read_config_dword(dev, vsec + 0x8, &val);
  194. show_reg("Num AFUs", (val >> 0) & 0xff);
  195. show_reg("Status", (val >> 8) & 0xff);
  196. show_reg("Mode Control", (val >> 16) & 0xff);
  197. show_reg("Reserved", (val >> 24) & 0xff);
  198. pci_read_config_dword(dev, vsec + 0xc, &val);
  199. show_reg("PSL Rev", (val >> 0) & 0xffff);
  200. show_reg("CAIA Ver", (val >> 16) & 0xffff);
  201. pci_read_config_dword(dev, vsec + 0x10, &val);
  202. show_reg("Base Image Rev", (val >> 0) & 0xffff);
  203. show_reg("Reserved", (val >> 16) & 0x0fff);
  204. show_reg("Image Control", (val >> 28) & 0x3);
  205. show_reg("Reserved", (val >> 30) & 0x1);
  206. show_reg("Image Loaded", (val >> 31) & 0x1);
  207. pci_read_config_dword(dev, vsec + 0x14, &val);
  208. show_reg("Reserved", val);
  209. pci_read_config_dword(dev, vsec + 0x18, &val);
  210. show_reg("Reserved", val);
  211. pci_read_config_dword(dev, vsec + 0x1c, &val);
  212. show_reg("Reserved", val);
  213. pci_read_config_dword(dev, vsec + 0x20, &val);
  214. show_reg("AFU Descriptor Offset", val);
  215. pci_read_config_dword(dev, vsec + 0x24, &val);
  216. show_reg("AFU Descriptor Size", val);
  217. pci_read_config_dword(dev, vsec + 0x28, &val);
  218. show_reg("Problem State Offset", val);
  219. pci_read_config_dword(dev, vsec + 0x2c, &val);
  220. show_reg("Problem State Size", val);
  221. pci_read_config_dword(dev, vsec + 0x30, &val);
  222. show_reg("Reserved", val);
  223. pci_read_config_dword(dev, vsec + 0x34, &val);
  224. show_reg("Reserved", val);
  225. pci_read_config_dword(dev, vsec + 0x38, &val);
  226. show_reg("Reserved", val);
  227. pci_read_config_dword(dev, vsec + 0x3c, &val);
  228. show_reg("Reserved", val);
  229. pci_read_config_dword(dev, vsec + 0x40, &val);
  230. show_reg("PSL Programming Port", val);
  231. pci_read_config_dword(dev, vsec + 0x44, &val);
  232. show_reg("PSL Programming Control", val);
  233. pci_read_config_dword(dev, vsec + 0x48, &val);
  234. show_reg("Reserved", val);
  235. pci_read_config_dword(dev, vsec + 0x4c, &val);
  236. show_reg("Reserved", val);
  237. pci_read_config_dword(dev, vsec + 0x50, &val);
  238. show_reg("Flash Address Register", val);
  239. pci_read_config_dword(dev, vsec + 0x54, &val);
  240. show_reg("Flash Size Register", val);
  241. pci_read_config_dword(dev, vsec + 0x58, &val);
  242. show_reg("Flash Status/Control Register", val);
  243. pci_read_config_dword(dev, vsec + 0x58, &val);
  244. show_reg("Flash Data Port", val);
  245. #undef show_reg
  246. }
  247. static void dump_afu_descriptor(struct cxl_afu *afu)
  248. {
  249. u64 val, afu_cr_num, afu_cr_off, afu_cr_len;
  250. int i;
  251. #define show_reg(name, what) \
  252. dev_info(&afu->dev, "afu desc: %30s: %#llx\n", name, what)
  253. val = AFUD_READ_INFO(afu);
  254. show_reg("num_ints_per_process", AFUD_NUM_INTS_PER_PROC(val));
  255. show_reg("num_of_processes", AFUD_NUM_PROCS(val));
  256. show_reg("num_of_afu_CRs", AFUD_NUM_CRS(val));
  257. show_reg("req_prog_mode", val & 0xffffULL);
  258. afu_cr_num = AFUD_NUM_CRS(val);
  259. val = AFUD_READ(afu, 0x8);
  260. show_reg("Reserved", val);
  261. val = AFUD_READ(afu, 0x10);
  262. show_reg("Reserved", val);
  263. val = AFUD_READ(afu, 0x18);
  264. show_reg("Reserved", val);
  265. val = AFUD_READ_CR(afu);
  266. show_reg("Reserved", (val >> (63-7)) & 0xff);
  267. show_reg("AFU_CR_len", AFUD_CR_LEN(val));
  268. afu_cr_len = AFUD_CR_LEN(val) * 256;
  269. val = AFUD_READ_CR_OFF(afu);
  270. afu_cr_off = val;
  271. show_reg("AFU_CR_offset", val);
  272. val = AFUD_READ_PPPSA(afu);
  273. show_reg("PerProcessPSA_control", (val >> (63-7)) & 0xff);
  274. show_reg("PerProcessPSA Length", AFUD_PPPSA_LEN(val));
  275. val = AFUD_READ_PPPSA_OFF(afu);
  276. show_reg("PerProcessPSA_offset", val);
  277. val = AFUD_READ_EB(afu);
  278. show_reg("Reserved", (val >> (63-7)) & 0xff);
  279. show_reg("AFU_EB_len", AFUD_EB_LEN(val));
  280. val = AFUD_READ_EB_OFF(afu);
  281. show_reg("AFU_EB_offset", val);
  282. for (i = 0; i < afu_cr_num; i++) {
  283. val = AFUD_READ_LE(afu, afu_cr_off + i * afu_cr_len);
  284. show_reg("CR Vendor", val & 0xffff);
  285. show_reg("CR Device", (val >> 16) & 0xffff);
  286. }
  287. #undef show_reg
  288. }
  289. static int init_implementation_adapter_regs(struct cxl *adapter, struct pci_dev *dev)
  290. {
  291. struct device_node *np;
  292. const __be32 *prop;
  293. u64 psl_dsnctl;
  294. u64 chipid;
  295. if (!(np = pnv_pci_get_phb_node(dev)))
  296. return -ENODEV;
  297. while (np && !(prop = of_get_property(np, "ibm,chip-id", NULL)))
  298. np = of_get_next_parent(np);
  299. if (!np)
  300. return -ENODEV;
  301. chipid = be32_to_cpup(prop);
  302. of_node_put(np);
  303. /* Tell PSL where to route data to */
  304. psl_dsnctl = 0x02E8900002000000ULL | (chipid << (63-5));
  305. cxl_p1_write(adapter, CXL_PSL_DSNDCTL, psl_dsnctl);
  306. cxl_p1_write(adapter, CXL_PSL_RESLCKTO, 0x20000000200ULL);
  307. /* snoop write mask */
  308. cxl_p1_write(adapter, CXL_PSL_SNWRALLOC, 0x00000000FFFFFFFFULL);
  309. /* set fir_accum */
  310. cxl_p1_write(adapter, CXL_PSL_FIR_CNTL, 0x0800000000000000ULL);
  311. /* for debugging with trace arrays */
  312. cxl_p1_write(adapter, CXL_PSL_TRACE, 0x0000FF7C00000000ULL);
  313. return 0;
  314. }
  315. #define TBSYNC_CNT(n) (((u64)n & 0x7) << (63-6))
  316. #define _2048_250MHZ_CYCLES 1
  317. static int cxl_setup_psl_timebase(struct cxl *adapter, struct pci_dev *dev)
  318. {
  319. u64 psl_tb;
  320. int delta;
  321. unsigned int retry = 0;
  322. struct device_node *np;
  323. if (!(np = pnv_pci_get_phb_node(dev)))
  324. return -ENODEV;
  325. /* Do not fail when CAPP timebase sync is not supported by OPAL */
  326. of_node_get(np);
  327. if (! of_get_property(np, "ibm,capp-timebase-sync", NULL)) {
  328. of_node_put(np);
  329. pr_err("PSL: Timebase sync: OPAL support missing\n");
  330. return 0;
  331. }
  332. of_node_put(np);
  333. /*
  334. * Setup PSL Timebase Control and Status register
  335. * with the recommended Timebase Sync Count value
  336. */
  337. cxl_p1_write(adapter, CXL_PSL_TB_CTLSTAT,
  338. TBSYNC_CNT(2 * _2048_250MHZ_CYCLES));
  339. /* Enable PSL Timebase */
  340. cxl_p1_write(adapter, CXL_PSL_Control, 0x0000000000000000);
  341. cxl_p1_write(adapter, CXL_PSL_Control, CXL_PSL_Control_tb);
  342. /* Wait until CORE TB and PSL TB difference <= 16usecs */
  343. do {
  344. msleep(1);
  345. if (retry++ > 5) {
  346. pr_err("PSL: Timebase sync: giving up!\n");
  347. return -EIO;
  348. }
  349. psl_tb = cxl_p1_read(adapter, CXL_PSL_Timebase);
  350. delta = mftb() - psl_tb;
  351. if (delta < 0)
  352. delta = -delta;
  353. } while (cputime_to_usecs(delta) > 16);
  354. return 0;
  355. }
  356. static int init_implementation_afu_regs(struct cxl_afu *afu)
  357. {
  358. /* read/write masks for this slice */
  359. cxl_p1n_write(afu, CXL_PSL_APCALLOC_A, 0xFFFFFFFEFEFEFEFEULL);
  360. /* APC read/write masks for this slice */
  361. cxl_p1n_write(afu, CXL_PSL_COALLOC_A, 0xFF000000FEFEFEFEULL);
  362. /* for debugging with trace arrays */
  363. cxl_p1n_write(afu, CXL_PSL_SLICE_TRACE, 0x0000FFFF00000000ULL);
  364. cxl_p1n_write(afu, CXL_PSL_RXCTL_A, CXL_PSL_RXCTL_AFUHP_4S);
  365. return 0;
  366. }
  367. int cxl_setup_irq(struct cxl *adapter, unsigned int hwirq,
  368. unsigned int virq)
  369. {
  370. struct pci_dev *dev = to_pci_dev(adapter->dev.parent);
  371. return pnv_cxl_ioda_msi_setup(dev, hwirq, virq);
  372. }
  373. int cxl_update_image_control(struct cxl *adapter)
  374. {
  375. struct pci_dev *dev = to_pci_dev(adapter->dev.parent);
  376. int rc;
  377. int vsec;
  378. u8 image_state;
  379. if (!(vsec = find_cxl_vsec(dev))) {
  380. dev_err(&dev->dev, "ABORTING: CXL VSEC not found!\n");
  381. return -ENODEV;
  382. }
  383. if ((rc = CXL_READ_VSEC_IMAGE_STATE(dev, vsec, &image_state))) {
  384. dev_err(&dev->dev, "failed to read image state: %i\n", rc);
  385. return rc;
  386. }
  387. if (adapter->perst_loads_image)
  388. image_state |= CXL_VSEC_PERST_LOADS_IMAGE;
  389. else
  390. image_state &= ~CXL_VSEC_PERST_LOADS_IMAGE;
  391. if (adapter->perst_select_user)
  392. image_state |= CXL_VSEC_PERST_SELECT_USER;
  393. else
  394. image_state &= ~CXL_VSEC_PERST_SELECT_USER;
  395. if ((rc = CXL_WRITE_VSEC_IMAGE_STATE(dev, vsec, image_state))) {
  396. dev_err(&dev->dev, "failed to update image control: %i\n", rc);
  397. return rc;
  398. }
  399. return 0;
  400. }
  401. int cxl_alloc_one_irq(struct cxl *adapter)
  402. {
  403. struct pci_dev *dev = to_pci_dev(adapter->dev.parent);
  404. return pnv_cxl_alloc_hwirqs(dev, 1);
  405. }
  406. void cxl_release_one_irq(struct cxl *adapter, int hwirq)
  407. {
  408. struct pci_dev *dev = to_pci_dev(adapter->dev.parent);
  409. return pnv_cxl_release_hwirqs(dev, hwirq, 1);
  410. }
  411. int cxl_alloc_irq_ranges(struct cxl_irq_ranges *irqs, struct cxl *adapter, unsigned int num)
  412. {
  413. struct pci_dev *dev = to_pci_dev(adapter->dev.parent);
  414. return pnv_cxl_alloc_hwirq_ranges(irqs, dev, num);
  415. }
  416. void cxl_release_irq_ranges(struct cxl_irq_ranges *irqs, struct cxl *adapter)
  417. {
  418. struct pci_dev *dev = to_pci_dev(adapter->dev.parent);
  419. pnv_cxl_release_hwirq_ranges(irqs, dev);
  420. }
  421. static int setup_cxl_bars(struct pci_dev *dev)
  422. {
  423. /* Safety check in case we get backported to < 3.17 without M64 */
  424. if ((p1_base(dev) < 0x100000000ULL) ||
  425. (p2_base(dev) < 0x100000000ULL)) {
  426. dev_err(&dev->dev, "ABORTING: M32 BAR assignment incompatible with CXL\n");
  427. return -ENODEV;
  428. }
  429. /*
  430. * BAR 4/5 has a special meaning for CXL and must be programmed with a
  431. * special value corresponding to the CXL protocol address range.
  432. * For POWER 8 that means bits 48:49 must be set to 10
  433. */
  434. pci_write_config_dword(dev, PCI_BASE_ADDRESS_4, 0x00000000);
  435. pci_write_config_dword(dev, PCI_BASE_ADDRESS_5, 0x00020000);
  436. return 0;
  437. }
  438. /* pciex node: ibm,opal-m64-window = <0x3d058 0x0 0x3d058 0x0 0x8 0x0>; */
  439. static int switch_card_to_cxl(struct pci_dev *dev)
  440. {
  441. int vsec;
  442. u8 val;
  443. int rc;
  444. dev_info(&dev->dev, "switch card to CXL\n");
  445. if (!(vsec = find_cxl_vsec(dev))) {
  446. dev_err(&dev->dev, "ABORTING: CXL VSEC not found!\n");
  447. return -ENODEV;
  448. }
  449. if ((rc = CXL_READ_VSEC_MODE_CONTROL(dev, vsec, &val))) {
  450. dev_err(&dev->dev, "failed to read current mode control: %i", rc);
  451. return rc;
  452. }
  453. val &= ~CXL_VSEC_PROTOCOL_MASK;
  454. val |= CXL_VSEC_PROTOCOL_256TB | CXL_VSEC_PROTOCOL_ENABLE;
  455. if ((rc = CXL_WRITE_VSEC_MODE_CONTROL(dev, vsec, val))) {
  456. dev_err(&dev->dev, "failed to enable CXL protocol: %i", rc);
  457. return rc;
  458. }
  459. /*
  460. * The CAIA spec (v0.12 11.6 Bi-modal Device Support) states
  461. * we must wait 100ms after this mode switch before touching
  462. * PCIe config space.
  463. */
  464. msleep(100);
  465. return 0;
  466. }
  467. static int cxl_map_slice_regs(struct cxl_afu *afu, struct cxl *adapter, struct pci_dev *dev)
  468. {
  469. u64 p1n_base, p2n_base, afu_desc;
  470. const u64 p1n_size = 0x100;
  471. const u64 p2n_size = 0x1000;
  472. p1n_base = p1_base(dev) + 0x10000 + (afu->slice * p1n_size);
  473. p2n_base = p2_base(dev) + (afu->slice * p2n_size);
  474. afu->psn_phys = p2_base(dev) + (adapter->ps_off + (afu->slice * adapter->ps_size));
  475. afu_desc = p2_base(dev) + adapter->afu_desc_off + (afu->slice * adapter->afu_desc_size);
  476. if (!(afu->p1n_mmio = ioremap(p1n_base, p1n_size)))
  477. goto err;
  478. if (!(afu->p2n_mmio = ioremap(p2n_base, p2n_size)))
  479. goto err1;
  480. if (afu_desc) {
  481. if (!(afu->afu_desc_mmio = ioremap(afu_desc, adapter->afu_desc_size)))
  482. goto err2;
  483. }
  484. return 0;
  485. err2:
  486. iounmap(afu->p2n_mmio);
  487. err1:
  488. iounmap(afu->p1n_mmio);
  489. err:
  490. dev_err(&afu->dev, "Error mapping AFU MMIO regions\n");
  491. return -ENOMEM;
  492. }
  493. static void cxl_unmap_slice_regs(struct cxl_afu *afu)
  494. {
  495. if (afu->p2n_mmio) {
  496. iounmap(afu->p2n_mmio);
  497. afu->p2n_mmio = NULL;
  498. }
  499. if (afu->p1n_mmio) {
  500. iounmap(afu->p1n_mmio);
  501. afu->p1n_mmio = NULL;
  502. }
  503. if (afu->afu_desc_mmio) {
  504. iounmap(afu->afu_desc_mmio);
  505. afu->afu_desc_mmio = NULL;
  506. }
  507. }
  508. static void cxl_release_afu(struct device *dev)
  509. {
  510. struct cxl_afu *afu = to_cxl_afu(dev);
  511. pr_devel("cxl_release_afu\n");
  512. idr_destroy(&afu->contexts_idr);
  513. cxl_release_spa(afu);
  514. kfree(afu);
  515. }
  516. static struct cxl_afu *cxl_alloc_afu(struct cxl *adapter, int slice)
  517. {
  518. struct cxl_afu *afu;
  519. if (!(afu = kzalloc(sizeof(struct cxl_afu), GFP_KERNEL)))
  520. return NULL;
  521. afu->adapter = adapter;
  522. afu->dev.parent = &adapter->dev;
  523. afu->dev.release = cxl_release_afu;
  524. afu->slice = slice;
  525. idr_init(&afu->contexts_idr);
  526. mutex_init(&afu->contexts_lock);
  527. spin_lock_init(&afu->afu_cntl_lock);
  528. mutex_init(&afu->spa_mutex);
  529. afu->prefault_mode = CXL_PREFAULT_NONE;
  530. afu->irqs_max = afu->adapter->user_irqs;
  531. return afu;
  532. }
  533. /* Expects AFU struct to have recently been zeroed out */
  534. static int cxl_read_afu_descriptor(struct cxl_afu *afu)
  535. {
  536. u64 val;
  537. val = AFUD_READ_INFO(afu);
  538. afu->pp_irqs = AFUD_NUM_INTS_PER_PROC(val);
  539. afu->max_procs_virtualised = AFUD_NUM_PROCS(val);
  540. afu->crs_num = AFUD_NUM_CRS(val);
  541. if (AFUD_AFU_DIRECTED(val))
  542. afu->modes_supported |= CXL_MODE_DIRECTED;
  543. if (AFUD_DEDICATED_PROCESS(val))
  544. afu->modes_supported |= CXL_MODE_DEDICATED;
  545. if (AFUD_TIME_SLICED(val))
  546. afu->modes_supported |= CXL_MODE_TIME_SLICED;
  547. val = AFUD_READ_PPPSA(afu);
  548. afu->pp_size = AFUD_PPPSA_LEN(val) * 4096;
  549. afu->psa = AFUD_PPPSA_PSA(val);
  550. if ((afu->pp_psa = AFUD_PPPSA_PP(val)))
  551. afu->pp_offset = AFUD_READ_PPPSA_OFF(afu);
  552. val = AFUD_READ_CR(afu);
  553. afu->crs_len = AFUD_CR_LEN(val) * 256;
  554. afu->crs_offset = AFUD_READ_CR_OFF(afu);
  555. /* eb_len is in multiple of 4K */
  556. afu->eb_len = AFUD_EB_LEN(AFUD_READ_EB(afu)) * 4096;
  557. afu->eb_offset = AFUD_READ_EB_OFF(afu);
  558. /* eb_off is 4K aligned so lower 12 bits are always zero */
  559. if (EXTRACT_PPC_BITS(afu->eb_offset, 0, 11) != 0) {
  560. dev_warn(&afu->dev,
  561. "Invalid AFU error buffer offset %Lx\n",
  562. afu->eb_offset);
  563. dev_info(&afu->dev,
  564. "Ignoring AFU error buffer in the descriptor\n");
  565. /* indicate that no afu buffer exists */
  566. afu->eb_len = 0;
  567. }
  568. return 0;
  569. }
  570. static int cxl_afu_descriptor_looks_ok(struct cxl_afu *afu)
  571. {
  572. int i;
  573. if (afu->psa && afu->adapter->ps_size <
  574. (afu->pp_offset + afu->pp_size*afu->max_procs_virtualised)) {
  575. dev_err(&afu->dev, "per-process PSA can't fit inside the PSA!\n");
  576. return -ENODEV;
  577. }
  578. if (afu->pp_psa && (afu->pp_size < PAGE_SIZE))
  579. dev_warn(&afu->dev, "AFU uses < PAGE_SIZE per-process PSA!");
  580. for (i = 0; i < afu->crs_num; i++) {
  581. if ((cxl_afu_cr_read32(afu, i, 0) == 0)) {
  582. dev_err(&afu->dev, "ABORTING: AFU configuration record %i is invalid\n", i);
  583. return -EINVAL;
  584. }
  585. }
  586. return 0;
  587. }
  588. static int sanitise_afu_regs(struct cxl_afu *afu)
  589. {
  590. u64 reg;
  591. /*
  592. * Clear out any regs that contain either an IVTE or address or may be
  593. * waiting on an acknowledgement to try to be a bit safer as we bring
  594. * it online
  595. */
  596. reg = cxl_p2n_read(afu, CXL_AFU_Cntl_An);
  597. if ((reg & CXL_AFU_Cntl_An_ES_MASK) != CXL_AFU_Cntl_An_ES_Disabled) {
  598. dev_warn(&afu->dev, "WARNING: AFU was not disabled: %#016llx\n", reg);
  599. if (__cxl_afu_reset(afu))
  600. return -EIO;
  601. if (cxl_afu_disable(afu))
  602. return -EIO;
  603. if (cxl_psl_purge(afu))
  604. return -EIO;
  605. }
  606. cxl_p1n_write(afu, CXL_PSL_SPAP_An, 0x0000000000000000);
  607. cxl_p1n_write(afu, CXL_PSL_IVTE_Limit_An, 0x0000000000000000);
  608. cxl_p1n_write(afu, CXL_PSL_IVTE_Offset_An, 0x0000000000000000);
  609. cxl_p1n_write(afu, CXL_PSL_AMBAR_An, 0x0000000000000000);
  610. cxl_p1n_write(afu, CXL_PSL_SPOffset_An, 0x0000000000000000);
  611. cxl_p1n_write(afu, CXL_HAURP_An, 0x0000000000000000);
  612. cxl_p2n_write(afu, CXL_CSRP_An, 0x0000000000000000);
  613. cxl_p2n_write(afu, CXL_AURP1_An, 0x0000000000000000);
  614. cxl_p2n_write(afu, CXL_AURP0_An, 0x0000000000000000);
  615. cxl_p2n_write(afu, CXL_SSTP1_An, 0x0000000000000000);
  616. cxl_p2n_write(afu, CXL_SSTP0_An, 0x0000000000000000);
  617. reg = cxl_p2n_read(afu, CXL_PSL_DSISR_An);
  618. if (reg) {
  619. dev_warn(&afu->dev, "AFU had pending DSISR: %#016llx\n", reg);
  620. if (reg & CXL_PSL_DSISR_TRANS)
  621. cxl_p2n_write(afu, CXL_PSL_TFC_An, CXL_PSL_TFC_An_AE);
  622. else
  623. cxl_p2n_write(afu, CXL_PSL_TFC_An, CXL_PSL_TFC_An_A);
  624. }
  625. reg = cxl_p1n_read(afu, CXL_PSL_SERR_An);
  626. if (reg) {
  627. if (reg & ~0xffff)
  628. dev_warn(&afu->dev, "AFU had pending SERR: %#016llx\n", reg);
  629. cxl_p1n_write(afu, CXL_PSL_SERR_An, reg & ~0xffff);
  630. }
  631. reg = cxl_p2n_read(afu, CXL_PSL_ErrStat_An);
  632. if (reg) {
  633. dev_warn(&afu->dev, "AFU had pending error status: %#016llx\n", reg);
  634. cxl_p2n_write(afu, CXL_PSL_ErrStat_An, reg);
  635. }
  636. return 0;
  637. }
  638. #define ERR_BUFF_MAX_COPY_SIZE PAGE_SIZE
  639. /*
  640. * afu_eb_read:
  641. * Called from sysfs and reads the afu error info buffer. The h/w only supports
  642. * 4/8 bytes aligned access. So in case the requested offset/count arent 8 byte
  643. * aligned the function uses a bounce buffer which can be max PAGE_SIZE.
  644. */
  645. ssize_t cxl_afu_read_err_buffer(struct cxl_afu *afu, char *buf,
  646. loff_t off, size_t count)
  647. {
  648. loff_t aligned_start, aligned_end;
  649. size_t aligned_length;
  650. void *tbuf;
  651. const void __iomem *ebuf = afu->afu_desc_mmio + afu->eb_offset;
  652. if (count == 0 || off < 0 || (size_t)off >= afu->eb_len)
  653. return 0;
  654. /* calculate aligned read window */
  655. count = min((size_t)(afu->eb_len - off), count);
  656. aligned_start = round_down(off, 8);
  657. aligned_end = round_up(off + count, 8);
  658. aligned_length = aligned_end - aligned_start;
  659. /* max we can copy in one read is PAGE_SIZE */
  660. if (aligned_length > ERR_BUFF_MAX_COPY_SIZE) {
  661. aligned_length = ERR_BUFF_MAX_COPY_SIZE;
  662. count = ERR_BUFF_MAX_COPY_SIZE - (off & 0x7);
  663. }
  664. /* use bounce buffer for copy */
  665. tbuf = (void *)__get_free_page(GFP_TEMPORARY);
  666. if (!tbuf)
  667. return -ENOMEM;
  668. /* perform aligned read from the mmio region */
  669. memcpy_fromio(tbuf, ebuf + aligned_start, aligned_length);
  670. memcpy(buf, tbuf + (off & 0x7), count);
  671. free_page((unsigned long)tbuf);
  672. return count;
  673. }
  674. static int cxl_configure_afu(struct cxl_afu *afu, struct cxl *adapter, struct pci_dev *dev)
  675. {
  676. int rc;
  677. if ((rc = cxl_map_slice_regs(afu, adapter, dev)))
  678. return rc;
  679. if ((rc = sanitise_afu_regs(afu)))
  680. goto err1;
  681. /* We need to reset the AFU before we can read the AFU descriptor */
  682. if ((rc = __cxl_afu_reset(afu)))
  683. goto err1;
  684. if (cxl_verbose)
  685. dump_afu_descriptor(afu);
  686. if ((rc = cxl_read_afu_descriptor(afu)))
  687. goto err1;
  688. if ((rc = cxl_afu_descriptor_looks_ok(afu)))
  689. goto err1;
  690. if ((rc = init_implementation_afu_regs(afu)))
  691. goto err1;
  692. if ((rc = cxl_register_serr_irq(afu)))
  693. goto err1;
  694. if ((rc = cxl_register_psl_irq(afu)))
  695. goto err2;
  696. return 0;
  697. err2:
  698. cxl_release_serr_irq(afu);
  699. err1:
  700. cxl_unmap_slice_regs(afu);
  701. return rc;
  702. }
  703. static void cxl_deconfigure_afu(struct cxl_afu *afu)
  704. {
  705. cxl_release_psl_irq(afu);
  706. cxl_release_serr_irq(afu);
  707. cxl_unmap_slice_regs(afu);
  708. }
  709. static int cxl_init_afu(struct cxl *adapter, int slice, struct pci_dev *dev)
  710. {
  711. struct cxl_afu *afu;
  712. int rc;
  713. afu = cxl_alloc_afu(adapter, slice);
  714. if (!afu)
  715. return -ENOMEM;
  716. rc = dev_set_name(&afu->dev, "afu%i.%i", adapter->adapter_num, slice);
  717. if (rc)
  718. goto err_free;
  719. rc = cxl_configure_afu(afu, adapter, dev);
  720. if (rc)
  721. goto err_free;
  722. /* Don't care if this fails */
  723. cxl_debugfs_afu_add(afu);
  724. /*
  725. * After we call this function we must not free the afu directly, even
  726. * if it returns an error!
  727. */
  728. if ((rc = cxl_register_afu(afu)))
  729. goto err_put1;
  730. if ((rc = cxl_sysfs_afu_add(afu)))
  731. goto err_put1;
  732. adapter->afu[afu->slice] = afu;
  733. if ((rc = cxl_pci_vphb_add(afu)))
  734. dev_info(&afu->dev, "Can't register vPHB\n");
  735. return 0;
  736. err_put1:
  737. cxl_deconfigure_afu(afu);
  738. cxl_debugfs_afu_remove(afu);
  739. device_unregister(&afu->dev);
  740. return rc;
  741. err_free:
  742. kfree(afu);
  743. return rc;
  744. }
  745. static void cxl_remove_afu(struct cxl_afu *afu)
  746. {
  747. pr_devel("cxl_remove_afu\n");
  748. if (!afu)
  749. return;
  750. cxl_sysfs_afu_remove(afu);
  751. cxl_debugfs_afu_remove(afu);
  752. spin_lock(&afu->adapter->afu_list_lock);
  753. afu->adapter->afu[afu->slice] = NULL;
  754. spin_unlock(&afu->adapter->afu_list_lock);
  755. cxl_context_detach_all(afu);
  756. cxl_afu_deactivate_mode(afu);
  757. cxl_deconfigure_afu(afu);
  758. device_unregister(&afu->dev);
  759. }
  760. int cxl_reset(struct cxl *adapter)
  761. {
  762. struct pci_dev *dev = to_pci_dev(adapter->dev.parent);
  763. int rc;
  764. if (adapter->perst_same_image) {
  765. dev_warn(&dev->dev,
  766. "cxl: refusing to reset/reflash when perst_reloads_same_image is set.\n");
  767. return -EINVAL;
  768. }
  769. dev_info(&dev->dev, "CXL reset\n");
  770. /* pcie_warm_reset requests a fundamental pci reset which includes a
  771. * PERST assert/deassert. PERST triggers a loading of the image
  772. * if "user" or "factory" is selected in sysfs */
  773. if ((rc = pci_set_pcie_reset_state(dev, pcie_warm_reset))) {
  774. dev_err(&dev->dev, "cxl: pcie_warm_reset failed\n");
  775. return rc;
  776. }
  777. return rc;
  778. }
  779. static int cxl_map_adapter_regs(struct cxl *adapter, struct pci_dev *dev)
  780. {
  781. if (pci_request_region(dev, 2, "priv 2 regs"))
  782. goto err1;
  783. if (pci_request_region(dev, 0, "priv 1 regs"))
  784. goto err2;
  785. pr_devel("cxl_map_adapter_regs: p1: %#016llx %#llx, p2: %#016llx %#llx",
  786. p1_base(dev), p1_size(dev), p2_base(dev), p2_size(dev));
  787. if (!(adapter->p1_mmio = ioremap(p1_base(dev), p1_size(dev))))
  788. goto err3;
  789. if (!(adapter->p2_mmio = ioremap(p2_base(dev), p2_size(dev))))
  790. goto err4;
  791. return 0;
  792. err4:
  793. iounmap(adapter->p1_mmio);
  794. adapter->p1_mmio = NULL;
  795. err3:
  796. pci_release_region(dev, 0);
  797. err2:
  798. pci_release_region(dev, 2);
  799. err1:
  800. return -ENOMEM;
  801. }
  802. static void cxl_unmap_adapter_regs(struct cxl *adapter)
  803. {
  804. if (adapter->p1_mmio) {
  805. iounmap(adapter->p1_mmio);
  806. adapter->p1_mmio = NULL;
  807. pci_release_region(to_pci_dev(adapter->dev.parent), 2);
  808. }
  809. if (adapter->p2_mmio) {
  810. iounmap(adapter->p2_mmio);
  811. adapter->p2_mmio = NULL;
  812. pci_release_region(to_pci_dev(adapter->dev.parent), 0);
  813. }
  814. }
  815. static int cxl_read_vsec(struct cxl *adapter, struct pci_dev *dev)
  816. {
  817. int vsec;
  818. u32 afu_desc_off, afu_desc_size;
  819. u32 ps_off, ps_size;
  820. u16 vseclen;
  821. u8 image_state;
  822. if (!(vsec = find_cxl_vsec(dev))) {
  823. dev_err(&dev->dev, "ABORTING: CXL VSEC not found!\n");
  824. return -ENODEV;
  825. }
  826. CXL_READ_VSEC_LENGTH(dev, vsec, &vseclen);
  827. if (vseclen < CXL_VSEC_MIN_SIZE) {
  828. dev_err(&dev->dev, "ABORTING: CXL VSEC too short\n");
  829. return -EINVAL;
  830. }
  831. CXL_READ_VSEC_STATUS(dev, vsec, &adapter->vsec_status);
  832. CXL_READ_VSEC_PSL_REVISION(dev, vsec, &adapter->psl_rev);
  833. CXL_READ_VSEC_CAIA_MAJOR(dev, vsec, &adapter->caia_major);
  834. CXL_READ_VSEC_CAIA_MINOR(dev, vsec, &adapter->caia_minor);
  835. CXL_READ_VSEC_BASE_IMAGE(dev, vsec, &adapter->base_image);
  836. CXL_READ_VSEC_IMAGE_STATE(dev, vsec, &image_state);
  837. adapter->user_image_loaded = !!(image_state & CXL_VSEC_USER_IMAGE_LOADED);
  838. adapter->perst_select_user = !!(image_state & CXL_VSEC_USER_IMAGE_LOADED);
  839. CXL_READ_VSEC_NAFUS(dev, vsec, &adapter->slices);
  840. CXL_READ_VSEC_AFU_DESC_OFF(dev, vsec, &afu_desc_off);
  841. CXL_READ_VSEC_AFU_DESC_SIZE(dev, vsec, &afu_desc_size);
  842. CXL_READ_VSEC_PS_OFF(dev, vsec, &ps_off);
  843. CXL_READ_VSEC_PS_SIZE(dev, vsec, &ps_size);
  844. /* Convert everything to bytes, because there is NO WAY I'd look at the
  845. * code a month later and forget what units these are in ;-) */
  846. adapter->ps_off = ps_off * 64 * 1024;
  847. adapter->ps_size = ps_size * 64 * 1024;
  848. adapter->afu_desc_off = afu_desc_off * 64 * 1024;
  849. adapter->afu_desc_size = afu_desc_size *64 * 1024;
  850. /* Total IRQs - 1 PSL ERROR - #AFU*(1 slice error + 1 DSI) */
  851. adapter->user_irqs = pnv_cxl_get_irq_count(dev) - 1 - 2*adapter->slices;
  852. return 0;
  853. }
  854. static int cxl_vsec_looks_ok(struct cxl *adapter, struct pci_dev *dev)
  855. {
  856. if (adapter->vsec_status & CXL_STATUS_SECOND_PORT)
  857. return -EBUSY;
  858. if (adapter->vsec_status & CXL_UNSUPPORTED_FEATURES) {
  859. dev_err(&dev->dev, "ABORTING: CXL requires unsupported features\n");
  860. return -EINVAL;
  861. }
  862. if (!adapter->slices) {
  863. /* Once we support dynamic reprogramming we can use the card if
  864. * it supports loadable AFUs */
  865. dev_err(&dev->dev, "ABORTING: Device has no AFUs\n");
  866. return -EINVAL;
  867. }
  868. if (!adapter->afu_desc_off || !adapter->afu_desc_size) {
  869. dev_err(&dev->dev, "ABORTING: VSEC shows no AFU descriptors\n");
  870. return -EINVAL;
  871. }
  872. if (adapter->ps_size > p2_size(dev) - adapter->ps_off) {
  873. dev_err(&dev->dev, "ABORTING: Problem state size larger than "
  874. "available in BAR2: 0x%llx > 0x%llx\n",
  875. adapter->ps_size, p2_size(dev) - adapter->ps_off);
  876. return -EINVAL;
  877. }
  878. return 0;
  879. }
  880. static void cxl_release_adapter(struct device *dev)
  881. {
  882. struct cxl *adapter = to_cxl_adapter(dev);
  883. pr_devel("cxl_release_adapter\n");
  884. cxl_remove_adapter_nr(adapter);
  885. kfree(adapter);
  886. }
  887. static struct cxl *cxl_alloc_adapter(void)
  888. {
  889. struct cxl *adapter;
  890. if (!(adapter = kzalloc(sizeof(struct cxl), GFP_KERNEL)))
  891. return NULL;
  892. spin_lock_init(&adapter->afu_list_lock);
  893. if (cxl_alloc_adapter_nr(adapter))
  894. goto err1;
  895. if (dev_set_name(&adapter->dev, "card%i", adapter->adapter_num))
  896. goto err2;
  897. return adapter;
  898. err2:
  899. cxl_remove_adapter_nr(adapter);
  900. err1:
  901. kfree(adapter);
  902. return NULL;
  903. }
  904. #define CXL_PSL_ErrIVTE_tberror (0x1ull << (63-31))
  905. static int sanitise_adapter_regs(struct cxl *adapter)
  906. {
  907. /* Clear PSL tberror bit by writing 1 to it */
  908. cxl_p1_write(adapter, CXL_PSL_ErrIVTE, CXL_PSL_ErrIVTE_tberror);
  909. return cxl_tlb_slb_invalidate(adapter);
  910. }
  911. /* This should contain *only* operations that can safely be done in
  912. * both creation and recovery.
  913. */
  914. static int cxl_configure_adapter(struct cxl *adapter, struct pci_dev *dev)
  915. {
  916. int rc;
  917. adapter->dev.parent = &dev->dev;
  918. adapter->dev.release = cxl_release_adapter;
  919. pci_set_drvdata(dev, adapter);
  920. rc = pci_enable_device(dev);
  921. if (rc) {
  922. dev_err(&dev->dev, "pci_enable_device failed: %i\n", rc);
  923. return rc;
  924. }
  925. if ((rc = cxl_read_vsec(adapter, dev)))
  926. return rc;
  927. if ((rc = cxl_vsec_looks_ok(adapter, dev)))
  928. return rc;
  929. if ((rc = setup_cxl_bars(dev)))
  930. return rc;
  931. if ((rc = switch_card_to_cxl(dev)))
  932. return rc;
  933. if ((rc = cxl_update_image_control(adapter)))
  934. return rc;
  935. if ((rc = cxl_map_adapter_regs(adapter, dev)))
  936. return rc;
  937. if ((rc = sanitise_adapter_regs(adapter)))
  938. goto err;
  939. if ((rc = init_implementation_adapter_regs(adapter, dev)))
  940. goto err;
  941. if ((rc = pnv_phb_to_cxl_mode(dev, OPAL_PHB_CAPI_MODE_CAPI)))
  942. goto err;
  943. /* If recovery happened, the last step is to turn on snooping.
  944. * In the non-recovery case this has no effect */
  945. if ((rc = pnv_phb_to_cxl_mode(dev, OPAL_PHB_CAPI_MODE_SNOOP_ON)))
  946. goto err;
  947. if ((rc = cxl_setup_psl_timebase(adapter, dev)))
  948. goto err;
  949. if ((rc = cxl_register_psl_err_irq(adapter)))
  950. goto err;
  951. return 0;
  952. err:
  953. cxl_unmap_adapter_regs(adapter);
  954. return rc;
  955. }
  956. static void cxl_deconfigure_adapter(struct cxl *adapter)
  957. {
  958. struct pci_dev *pdev = to_pci_dev(adapter->dev.parent);
  959. cxl_release_psl_err_irq(adapter);
  960. cxl_unmap_adapter_regs(adapter);
  961. pci_disable_device(pdev);
  962. }
  963. static struct cxl *cxl_init_adapter(struct pci_dev *dev)
  964. {
  965. struct cxl *adapter;
  966. int rc;
  967. adapter = cxl_alloc_adapter();
  968. if (!adapter)
  969. return ERR_PTR(-ENOMEM);
  970. /* Set defaults for parameters which need to persist over
  971. * configure/reconfigure
  972. */
  973. adapter->perst_loads_image = true;
  974. adapter->perst_same_image = false;
  975. rc = cxl_configure_adapter(adapter, dev);
  976. if (rc) {
  977. pci_disable_device(dev);
  978. cxl_release_adapter(&adapter->dev);
  979. return ERR_PTR(rc);
  980. }
  981. /* Don't care if this one fails: */
  982. cxl_debugfs_adapter_add(adapter);
  983. /*
  984. * After we call this function we must not free the adapter directly,
  985. * even if it returns an error!
  986. */
  987. if ((rc = cxl_register_adapter(adapter)))
  988. goto err_put1;
  989. if ((rc = cxl_sysfs_adapter_add(adapter)))
  990. goto err_put1;
  991. return adapter;
  992. err_put1:
  993. /* This should mirror cxl_remove_adapter, except without the
  994. * sysfs parts
  995. */
  996. cxl_debugfs_adapter_remove(adapter);
  997. cxl_deconfigure_adapter(adapter);
  998. device_unregister(&adapter->dev);
  999. return ERR_PTR(rc);
  1000. }
  1001. static void cxl_remove_adapter(struct cxl *adapter)
  1002. {
  1003. pr_devel("cxl_remove_adapter\n");
  1004. cxl_sysfs_adapter_remove(adapter);
  1005. cxl_debugfs_adapter_remove(adapter);
  1006. cxl_deconfigure_adapter(adapter);
  1007. device_unregister(&adapter->dev);
  1008. }
  1009. static int cxl_probe(struct pci_dev *dev, const struct pci_device_id *id)
  1010. {
  1011. struct cxl *adapter;
  1012. int slice;
  1013. int rc;
  1014. if (cxl_verbose)
  1015. dump_cxl_config_space(dev);
  1016. adapter = cxl_init_adapter(dev);
  1017. if (IS_ERR(adapter)) {
  1018. dev_err(&dev->dev, "cxl_init_adapter failed: %li\n", PTR_ERR(adapter));
  1019. return PTR_ERR(adapter);
  1020. }
  1021. for (slice = 0; slice < adapter->slices; slice++) {
  1022. if ((rc = cxl_init_afu(adapter, slice, dev))) {
  1023. dev_err(&dev->dev, "AFU %i failed to initialise: %i\n", slice, rc);
  1024. continue;
  1025. }
  1026. rc = cxl_afu_select_best_mode(adapter->afu[slice]);
  1027. if (rc)
  1028. dev_err(&dev->dev, "AFU %i failed to start: %i\n", slice, rc);
  1029. }
  1030. return 0;
  1031. }
  1032. static void cxl_remove(struct pci_dev *dev)
  1033. {
  1034. struct cxl *adapter = pci_get_drvdata(dev);
  1035. struct cxl_afu *afu;
  1036. int i;
  1037. /*
  1038. * Lock to prevent someone grabbing a ref through the adapter list as
  1039. * we are removing it
  1040. */
  1041. for (i = 0; i < adapter->slices; i++) {
  1042. afu = adapter->afu[i];
  1043. cxl_pci_vphb_remove(afu);
  1044. cxl_remove_afu(afu);
  1045. }
  1046. cxl_remove_adapter(adapter);
  1047. }
  1048. static pci_ers_result_t cxl_vphb_error_detected(struct cxl_afu *afu,
  1049. pci_channel_state_t state)
  1050. {
  1051. struct pci_dev *afu_dev;
  1052. pci_ers_result_t result = PCI_ERS_RESULT_NEED_RESET;
  1053. pci_ers_result_t afu_result = PCI_ERS_RESULT_NEED_RESET;
  1054. /* There should only be one entry, but go through the list
  1055. * anyway
  1056. */
  1057. list_for_each_entry(afu_dev, &afu->phb->bus->devices, bus_list) {
  1058. if (!afu_dev->driver)
  1059. continue;
  1060. afu_dev->error_state = state;
  1061. if (afu_dev->driver->err_handler)
  1062. afu_result = afu_dev->driver->err_handler->error_detected(afu_dev,
  1063. state);
  1064. /* Disconnect trumps all, NONE trumps NEED_RESET */
  1065. if (afu_result == PCI_ERS_RESULT_DISCONNECT)
  1066. result = PCI_ERS_RESULT_DISCONNECT;
  1067. else if ((afu_result == PCI_ERS_RESULT_NONE) &&
  1068. (result == PCI_ERS_RESULT_NEED_RESET))
  1069. result = PCI_ERS_RESULT_NONE;
  1070. }
  1071. return result;
  1072. }
  1073. static pci_ers_result_t cxl_pci_error_detected(struct pci_dev *pdev,
  1074. pci_channel_state_t state)
  1075. {
  1076. struct cxl *adapter = pci_get_drvdata(pdev);
  1077. struct cxl_afu *afu;
  1078. pci_ers_result_t result = PCI_ERS_RESULT_NEED_RESET;
  1079. int i;
  1080. /* At this point, we could still have an interrupt pending.
  1081. * Let's try to get them out of the way before they do
  1082. * anything we don't like.
  1083. */
  1084. schedule();
  1085. /* If we're permanently dead, give up. */
  1086. if (state == pci_channel_io_perm_failure) {
  1087. /* Tell the AFU drivers; but we don't care what they
  1088. * say, we're going away.
  1089. */
  1090. for (i = 0; i < adapter->slices; i++) {
  1091. afu = adapter->afu[i];
  1092. cxl_vphb_error_detected(afu, state);
  1093. }
  1094. return PCI_ERS_RESULT_DISCONNECT;
  1095. }
  1096. /* Are we reflashing?
  1097. *
  1098. * If we reflash, we could come back as something entirely
  1099. * different, including a non-CAPI card. As such, by default
  1100. * we don't participate in the process. We'll be unbound and
  1101. * the slot re-probed. (TODO: check EEH doesn't blindly rebind
  1102. * us!)
  1103. *
  1104. * However, this isn't the entire story: for reliablity
  1105. * reasons, we usually want to reflash the FPGA on PERST in
  1106. * order to get back to a more reliable known-good state.
  1107. *
  1108. * This causes us a bit of a problem: if we reflash we can't
  1109. * trust that we'll come back the same - we could have a new
  1110. * image and been PERSTed in order to load that
  1111. * image. However, most of the time we actually *will* come
  1112. * back the same - for example a regular EEH event.
  1113. *
  1114. * Therefore, we allow the user to assert that the image is
  1115. * indeed the same and that we should continue on into EEH
  1116. * anyway.
  1117. */
  1118. if (adapter->perst_loads_image && !adapter->perst_same_image) {
  1119. /* TODO take the PHB out of CXL mode */
  1120. dev_info(&pdev->dev, "reflashing, so opting out of EEH!\n");
  1121. return PCI_ERS_RESULT_NONE;
  1122. }
  1123. /*
  1124. * At this point, we want to try to recover. We'll always
  1125. * need a complete slot reset: we don't trust any other reset.
  1126. *
  1127. * Now, we go through each AFU:
  1128. * - We send the driver, if bound, an error_detected callback.
  1129. * We expect it to clean up, but it can also tell us to give
  1130. * up and permanently detach the card. To simplify things, if
  1131. * any bound AFU driver doesn't support EEH, we give up on EEH.
  1132. *
  1133. * - We detach all contexts associated with the AFU. This
  1134. * does not free them, but puts them into a CLOSED state
  1135. * which causes any the associated files to return useful
  1136. * errors to userland. It also unmaps, but does not free,
  1137. * any IRQs.
  1138. *
  1139. * - We clean up our side: releasing and unmapping resources we hold
  1140. * so we can wire them up again when the hardware comes back up.
  1141. *
  1142. * Driver authors should note:
  1143. *
  1144. * - Any contexts you create in your kernel driver (except
  1145. * those associated with anonymous file descriptors) are
  1146. * your responsibility to free and recreate. Likewise with
  1147. * any attached resources.
  1148. *
  1149. * - We will take responsibility for re-initialising the
  1150. * device context (the one set up for you in
  1151. * cxl_pci_enable_device_hook and accessed through
  1152. * cxl_get_context). If you've attached IRQs or other
  1153. * resources to it, they remains yours to free.
  1154. *
  1155. * You can call the same functions to release resources as you
  1156. * normally would: we make sure that these functions continue
  1157. * to work when the hardware is down.
  1158. *
  1159. * Two examples:
  1160. *
  1161. * 1) If you normally free all your resources at the end of
  1162. * each request, or if you use anonymous FDs, your
  1163. * error_detected callback can simply set a flag to tell
  1164. * your driver not to start any new calls. You can then
  1165. * clear the flag in the resume callback.
  1166. *
  1167. * 2) If you normally allocate your resources on startup:
  1168. * * Set a flag in error_detected as above.
  1169. * * Let CXL detach your contexts.
  1170. * * In slot_reset, free the old resources and allocate new ones.
  1171. * * In resume, clear the flag to allow things to start.
  1172. */
  1173. for (i = 0; i < adapter->slices; i++) {
  1174. afu = adapter->afu[i];
  1175. result = cxl_vphb_error_detected(afu, state);
  1176. /* Only continue if everyone agrees on NEED_RESET */
  1177. if (result != PCI_ERS_RESULT_NEED_RESET)
  1178. return result;
  1179. cxl_context_detach_all(afu);
  1180. cxl_afu_deactivate_mode(afu);
  1181. cxl_deconfigure_afu(afu);
  1182. }
  1183. cxl_deconfigure_adapter(adapter);
  1184. return result;
  1185. }
  1186. static pci_ers_result_t cxl_pci_slot_reset(struct pci_dev *pdev)
  1187. {
  1188. struct cxl *adapter = pci_get_drvdata(pdev);
  1189. struct cxl_afu *afu;
  1190. struct cxl_context *ctx;
  1191. struct pci_dev *afu_dev;
  1192. pci_ers_result_t afu_result = PCI_ERS_RESULT_RECOVERED;
  1193. pci_ers_result_t result = PCI_ERS_RESULT_RECOVERED;
  1194. int i;
  1195. if (cxl_configure_adapter(adapter, pdev))
  1196. goto err;
  1197. for (i = 0; i < adapter->slices; i++) {
  1198. afu = adapter->afu[i];
  1199. if (cxl_configure_afu(afu, adapter, pdev))
  1200. goto err;
  1201. if (cxl_afu_select_best_mode(afu))
  1202. goto err;
  1203. cxl_pci_vphb_reconfigure(afu);
  1204. list_for_each_entry(afu_dev, &afu->phb->bus->devices, bus_list) {
  1205. /* Reset the device context.
  1206. * TODO: make this less disruptive
  1207. */
  1208. ctx = cxl_get_context(afu_dev);
  1209. if (ctx && cxl_release_context(ctx))
  1210. goto err;
  1211. ctx = cxl_dev_context_init(afu_dev);
  1212. if (!ctx)
  1213. goto err;
  1214. afu_dev->dev.archdata.cxl_ctx = ctx;
  1215. if (cxl_afu_check_and_enable(afu))
  1216. goto err;
  1217. afu_dev->error_state = pci_channel_io_normal;
  1218. /* If there's a driver attached, allow it to
  1219. * chime in on recovery. Drivers should check
  1220. * if everything has come back OK, but
  1221. * shouldn't start new work until we call
  1222. * their resume function.
  1223. */
  1224. if (!afu_dev->driver)
  1225. continue;
  1226. if (afu_dev->driver->err_handler &&
  1227. afu_dev->driver->err_handler->slot_reset)
  1228. afu_result = afu_dev->driver->err_handler->slot_reset(afu_dev);
  1229. if (afu_result == PCI_ERS_RESULT_DISCONNECT)
  1230. result = PCI_ERS_RESULT_DISCONNECT;
  1231. }
  1232. }
  1233. return result;
  1234. err:
  1235. /* All the bits that happen in both error_detected and cxl_remove
  1236. * should be idempotent, so we don't need to worry about leaving a mix
  1237. * of unconfigured and reconfigured resources.
  1238. */
  1239. dev_err(&pdev->dev, "EEH recovery failed. Asking to be disconnected.\n");
  1240. return PCI_ERS_RESULT_DISCONNECT;
  1241. }
  1242. static void cxl_pci_resume(struct pci_dev *pdev)
  1243. {
  1244. struct cxl *adapter = pci_get_drvdata(pdev);
  1245. struct cxl_afu *afu;
  1246. struct pci_dev *afu_dev;
  1247. int i;
  1248. /* Everything is back now. Drivers should restart work now.
  1249. * This is not the place to be checking if everything came back up
  1250. * properly, because there's no return value: do that in slot_reset.
  1251. */
  1252. for (i = 0; i < adapter->slices; i++) {
  1253. afu = adapter->afu[i];
  1254. list_for_each_entry(afu_dev, &afu->phb->bus->devices, bus_list) {
  1255. if (afu_dev->driver && afu_dev->driver->err_handler &&
  1256. afu_dev->driver->err_handler->resume)
  1257. afu_dev->driver->err_handler->resume(afu_dev);
  1258. }
  1259. }
  1260. }
  1261. static const struct pci_error_handlers cxl_err_handler = {
  1262. .error_detected = cxl_pci_error_detected,
  1263. .slot_reset = cxl_pci_slot_reset,
  1264. .resume = cxl_pci_resume,
  1265. };
  1266. struct pci_driver cxl_pci_driver = {
  1267. .name = "cxl-pci",
  1268. .id_table = cxl_pci_tbl,
  1269. .probe = cxl_probe,
  1270. .remove = cxl_remove,
  1271. .shutdown = cxl_remove,
  1272. .err_handler = &cxl_err_handler,
  1273. };