gru_instructions.h 21 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736
  1. /*
  2. * Copyright (c) 2008 Silicon Graphics, Inc. All Rights Reserved.
  3. *
  4. * This program is free software; you can redistribute it and/or modify
  5. * it under the terms of the GNU Lesser General Public License as published by
  6. * the Free Software Foundation; either version 2.1 of the License, or
  7. * (at your option) any later version.
  8. *
  9. * This program is distributed in the hope that it will be useful,
  10. * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11. * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
  12. * GNU Lesser General Public License for more details.
  13. *
  14. * You should have received a copy of the GNU Lesser General Public License
  15. * along with this program; if not, write to the Free Software
  16. * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
  17. */
  18. #ifndef __GRU_INSTRUCTIONS_H__
  19. #define __GRU_INSTRUCTIONS_H__
  20. extern int gru_check_status_proc(void *cb);
  21. extern int gru_wait_proc(void *cb);
  22. extern void gru_wait_abort_proc(void *cb);
  23. /*
  24. * Architecture dependent functions
  25. */
  26. #if defined(CONFIG_IA64)
  27. #include <linux/compiler.h>
  28. #include <asm/intrinsics.h>
  29. #define __flush_cache(p) ia64_fc((unsigned long)p)
  30. /* Use volatile on IA64 to ensure ordering via st4.rel */
  31. #define gru_ordered_store_ulong(p, v) \
  32. do { \
  33. barrier(); \
  34. *((volatile unsigned long *)(p)) = v; /* force st.rel */ \
  35. } while (0)
  36. #elif defined(CONFIG_X86_64)
  37. #define __flush_cache(p) clflush(p)
  38. #define gru_ordered_store_ulong(p, v) \
  39. do { \
  40. barrier(); \
  41. *(unsigned long *)p = v; \
  42. } while (0)
  43. #else
  44. #error "Unsupported architecture"
  45. #endif
  46. /*
  47. * Control block status and exception codes
  48. */
  49. #define CBS_IDLE 0
  50. #define CBS_EXCEPTION 1
  51. #define CBS_ACTIVE 2
  52. #define CBS_CALL_OS 3
  53. /* CB substatus bitmasks */
  54. #define CBSS_MSG_QUEUE_MASK 7
  55. #define CBSS_IMPLICIT_ABORT_ACTIVE_MASK 8
  56. /* CB substatus message queue values (low 3 bits of substatus) */
  57. #define CBSS_NO_ERROR 0
  58. #define CBSS_LB_OVERFLOWED 1
  59. #define CBSS_QLIMIT_REACHED 2
  60. #define CBSS_PAGE_OVERFLOW 3
  61. #define CBSS_AMO_NACKED 4
  62. #define CBSS_PUT_NACKED 5
  63. /*
  64. * Structure used to fetch exception detail for CBs that terminate with
  65. * CBS_EXCEPTION
  66. */
  67. struct control_block_extended_exc_detail {
  68. unsigned long cb;
  69. int opc;
  70. int ecause;
  71. int exopc;
  72. long exceptdet0;
  73. int exceptdet1;
  74. int cbrstate;
  75. int cbrexecstatus;
  76. };
  77. /*
  78. * Instruction formats
  79. */
  80. /*
  81. * Generic instruction format.
  82. * This definition has precise bit field definitions.
  83. */
  84. struct gru_instruction_bits {
  85. /* DW 0 - low */
  86. unsigned int icmd: 1;
  87. unsigned char ima: 3; /* CB_DelRep, unmapped mode */
  88. unsigned char reserved0: 4;
  89. unsigned int xtype: 3;
  90. unsigned int iaa0: 2;
  91. unsigned int iaa1: 2;
  92. unsigned char reserved1: 1;
  93. unsigned char opc: 8; /* opcode */
  94. unsigned char exopc: 8; /* extended opcode */
  95. /* DW 0 - high */
  96. unsigned int idef2: 22; /* TRi0 */
  97. unsigned char reserved2: 2;
  98. unsigned char istatus: 2;
  99. unsigned char isubstatus:4;
  100. unsigned char reserved3: 1;
  101. unsigned char tlb_fault_color: 1;
  102. /* DW 1 */
  103. unsigned long idef4; /* 42 bits: TRi1, BufSize */
  104. /* DW 2-6 */
  105. unsigned long idef1; /* BAddr0 */
  106. unsigned long idef5; /* Nelem */
  107. unsigned long idef6; /* Stride, Operand1 */
  108. unsigned long idef3; /* BAddr1, Value, Operand2 */
  109. unsigned long reserved4;
  110. /* DW 7 */
  111. unsigned long avalue; /* AValue */
  112. };
  113. /*
  114. * Generic instruction with friendlier names. This format is used
  115. * for inline instructions.
  116. */
  117. struct gru_instruction {
  118. /* DW 0 */
  119. union {
  120. unsigned long op64; /* icmd,xtype,iaa0,ima,opc,tri0 */
  121. struct {
  122. unsigned int op32;
  123. unsigned int tri0;
  124. };
  125. };
  126. unsigned long tri1_bufsize; /* DW 1 */
  127. unsigned long baddr0; /* DW 2 */
  128. unsigned long nelem; /* DW 3 */
  129. unsigned long op1_stride; /* DW 4 */
  130. unsigned long op2_value_baddr1; /* DW 5 */
  131. unsigned long reserved0; /* DW 6 */
  132. unsigned long avalue; /* DW 7 */
  133. };
  134. /* Some shifts and masks for the low 64 bits of a GRU command */
  135. #define GRU_CB_ICMD_SHFT 0
  136. #define GRU_CB_ICMD_MASK 0x1
  137. #define GRU_CB_XTYPE_SHFT 8
  138. #define GRU_CB_XTYPE_MASK 0x7
  139. #define GRU_CB_IAA0_SHFT 11
  140. #define GRU_CB_IAA0_MASK 0x3
  141. #define GRU_CB_IAA1_SHFT 13
  142. #define GRU_CB_IAA1_MASK 0x3
  143. #define GRU_CB_IMA_SHFT 1
  144. #define GRU_CB_IMA_MASK 0x3
  145. #define GRU_CB_OPC_SHFT 16
  146. #define GRU_CB_OPC_MASK 0xff
  147. #define GRU_CB_EXOPC_SHFT 24
  148. #define GRU_CB_EXOPC_MASK 0xff
  149. #define GRU_IDEF2_SHFT 32
  150. #define GRU_IDEF2_MASK 0x3ffff
  151. #define GRU_ISTATUS_SHFT 56
  152. #define GRU_ISTATUS_MASK 0x3
  153. /* GRU instruction opcodes (opc field) */
  154. #define OP_NOP 0x00
  155. #define OP_BCOPY 0x01
  156. #define OP_VLOAD 0x02
  157. #define OP_IVLOAD 0x03
  158. #define OP_VSTORE 0x04
  159. #define OP_IVSTORE 0x05
  160. #define OP_VSET 0x06
  161. #define OP_IVSET 0x07
  162. #define OP_MESQ 0x08
  163. #define OP_GAMXR 0x09
  164. #define OP_GAMIR 0x0a
  165. #define OP_GAMIRR 0x0b
  166. #define OP_GAMER 0x0c
  167. #define OP_GAMERR 0x0d
  168. #define OP_BSTORE 0x0e
  169. #define OP_VFLUSH 0x0f
  170. /* Extended opcodes values (exopc field) */
  171. /* GAMIR - AMOs with implicit operands */
  172. #define EOP_IR_FETCH 0x01 /* Plain fetch of memory */
  173. #define EOP_IR_CLR 0x02 /* Fetch and clear */
  174. #define EOP_IR_INC 0x05 /* Fetch and increment */
  175. #define EOP_IR_DEC 0x07 /* Fetch and decrement */
  176. #define EOP_IR_QCHK1 0x0d /* Queue check, 64 byte msg */
  177. #define EOP_IR_QCHK2 0x0e /* Queue check, 128 byte msg */
  178. /* GAMIRR - Registered AMOs with implicit operands */
  179. #define EOP_IRR_FETCH 0x01 /* Registered fetch of memory */
  180. #define EOP_IRR_CLR 0x02 /* Registered fetch and clear */
  181. #define EOP_IRR_INC 0x05 /* Registered fetch and increment */
  182. #define EOP_IRR_DEC 0x07 /* Registered fetch and decrement */
  183. #define EOP_IRR_DECZ 0x0f /* Registered fetch and decrement, update on zero*/
  184. /* GAMER - AMOs with explicit operands */
  185. #define EOP_ER_SWAP 0x00 /* Exchange argument and memory */
  186. #define EOP_ER_OR 0x01 /* Logical OR with memory */
  187. #define EOP_ER_AND 0x02 /* Logical AND with memory */
  188. #define EOP_ER_XOR 0x03 /* Logical XOR with memory */
  189. #define EOP_ER_ADD 0x04 /* Add value to memory */
  190. #define EOP_ER_CSWAP 0x08 /* Compare with operand2, write operand1 if match*/
  191. #define EOP_ER_CADD 0x0c /* Queue check, operand1*64 byte msg */
  192. /* GAMERR - Registered AMOs with explicit operands */
  193. #define EOP_ERR_SWAP 0x00 /* Exchange argument and memory */
  194. #define EOP_ERR_OR 0x01 /* Logical OR with memory */
  195. #define EOP_ERR_AND 0x02 /* Logical AND with memory */
  196. #define EOP_ERR_XOR 0x03 /* Logical XOR with memory */
  197. #define EOP_ERR_ADD 0x04 /* Add value to memory */
  198. #define EOP_ERR_CSWAP 0x08 /* Compare with operand2, write operand1 if match*/
  199. #define EOP_ERR_EPOLL 0x09 /* Poll for equality */
  200. #define EOP_ERR_NPOLL 0x0a /* Poll for inequality */
  201. /* GAMXR - SGI Arithmetic unit */
  202. #define EOP_XR_CSWAP 0x0b /* Masked compare exchange */
  203. /* Transfer types (xtype field) */
  204. #define XTYPE_B 0x0 /* byte */
  205. #define XTYPE_S 0x1 /* short (2-byte) */
  206. #define XTYPE_W 0x2 /* word (4-byte) */
  207. #define XTYPE_DW 0x3 /* doubleword (8-byte) */
  208. #define XTYPE_CL 0x6 /* cacheline (64-byte) */
  209. /* Instruction access attributes (iaa0, iaa1 fields) */
  210. #define IAA_RAM 0x0 /* normal cached RAM access */
  211. #define IAA_NCRAM 0x2 /* noncoherent RAM access */
  212. #define IAA_MMIO 0x1 /* noncoherent memory-mapped I/O space */
  213. #define IAA_REGISTER 0x3 /* memory-mapped registers, etc. */
  214. /* Instruction mode attributes (ima field) */
  215. #define IMA_MAPPED 0x0 /* Virtual mode */
  216. #define IMA_CB_DELAY 0x1 /* hold read responses until status changes */
  217. #define IMA_UNMAPPED 0x2 /* bypass the TLBs (OS only) */
  218. #define IMA_INTERRUPT 0x4 /* Interrupt when instruction completes */
  219. /* CBE ecause bits */
  220. #define CBE_CAUSE_RI (1 << 0)
  221. #define CBE_CAUSE_INVALID_INSTRUCTION (1 << 1)
  222. #define CBE_CAUSE_UNMAPPED_MODE_FORBIDDEN (1 << 2)
  223. #define CBE_CAUSE_PE_CHECK_DATA_ERROR (1 << 3)
  224. #define CBE_CAUSE_IAA_GAA_MISMATCH (1 << 4)
  225. #define CBE_CAUSE_DATA_SEGMENT_LIMIT_EXCEPTION (1 << 5)
  226. #define CBE_CAUSE_OS_FATAL_TLB_FAULT (1 << 6)
  227. #define CBE_CAUSE_EXECUTION_HW_ERROR (1 << 7)
  228. #define CBE_CAUSE_TLBHW_ERROR (1 << 8)
  229. #define CBE_CAUSE_RA_REQUEST_TIMEOUT (1 << 9)
  230. #define CBE_CAUSE_HA_REQUEST_TIMEOUT (1 << 10)
  231. #define CBE_CAUSE_RA_RESPONSE_FATAL (1 << 11)
  232. #define CBE_CAUSE_RA_RESPONSE_NON_FATAL (1 << 12)
  233. #define CBE_CAUSE_HA_RESPONSE_FATAL (1 << 13)
  234. #define CBE_CAUSE_HA_RESPONSE_NON_FATAL (1 << 14)
  235. #define CBE_CAUSE_ADDRESS_SPACE_DECODE_ERROR (1 << 15)
  236. #define CBE_CAUSE_PROTOCOL_STATE_DATA_ERROR (1 << 16)
  237. #define CBE_CAUSE_RA_RESPONSE_DATA_ERROR (1 << 17)
  238. #define CBE_CAUSE_HA_RESPONSE_DATA_ERROR (1 << 18)
  239. #define CBE_CAUSE_FORCED_ERROR (1 << 19)
  240. /* CBE cbrexecstatus bits */
  241. #define CBR_EXS_ABORT_OCC_BIT 0
  242. #define CBR_EXS_INT_OCC_BIT 1
  243. #define CBR_EXS_PENDING_BIT 2
  244. #define CBR_EXS_QUEUED_BIT 3
  245. #define CBR_EXS_TLB_INVAL_BIT 4
  246. #define CBR_EXS_EXCEPTION_BIT 5
  247. #define CBR_EXS_CB_INT_PENDING_BIT 6
  248. #define CBR_EXS_ABORT_OCC (1 << CBR_EXS_ABORT_OCC_BIT)
  249. #define CBR_EXS_INT_OCC (1 << CBR_EXS_INT_OCC_BIT)
  250. #define CBR_EXS_PENDING (1 << CBR_EXS_PENDING_BIT)
  251. #define CBR_EXS_QUEUED (1 << CBR_EXS_QUEUED_BIT)
  252. #define CBR_EXS_TLB_INVAL (1 << CBR_EXS_TLB_INVAL_BIT)
  253. #define CBR_EXS_EXCEPTION (1 << CBR_EXS_EXCEPTION_BIT)
  254. #define CBR_EXS_CB_INT_PENDING (1 << CBR_EXS_CB_INT_PENDING_BIT)
  255. /*
  256. * Exceptions are retried for the following cases. If any OTHER bits are set
  257. * in ecause, the exception is not retryable.
  258. */
  259. #define EXCEPTION_RETRY_BITS (CBE_CAUSE_EXECUTION_HW_ERROR | \
  260. CBE_CAUSE_TLBHW_ERROR | \
  261. CBE_CAUSE_RA_REQUEST_TIMEOUT | \
  262. CBE_CAUSE_RA_RESPONSE_NON_FATAL | \
  263. CBE_CAUSE_HA_RESPONSE_NON_FATAL | \
  264. CBE_CAUSE_RA_RESPONSE_DATA_ERROR | \
  265. CBE_CAUSE_HA_RESPONSE_DATA_ERROR \
  266. )
  267. /* Message queue head structure */
  268. union gru_mesqhead {
  269. unsigned long val;
  270. struct {
  271. unsigned int head;
  272. unsigned int limit;
  273. };
  274. };
  275. /* Generate the low word of a GRU instruction */
  276. static inline unsigned long
  277. __opdword(unsigned char opcode, unsigned char exopc, unsigned char xtype,
  278. unsigned char iaa0, unsigned char iaa1,
  279. unsigned long idef2, unsigned char ima)
  280. {
  281. return (1 << GRU_CB_ICMD_SHFT) |
  282. ((unsigned long)CBS_ACTIVE << GRU_ISTATUS_SHFT) |
  283. (idef2<< GRU_IDEF2_SHFT) |
  284. (iaa0 << GRU_CB_IAA0_SHFT) |
  285. (iaa1 << GRU_CB_IAA1_SHFT) |
  286. (ima << GRU_CB_IMA_SHFT) |
  287. (xtype << GRU_CB_XTYPE_SHFT) |
  288. (opcode << GRU_CB_OPC_SHFT) |
  289. (exopc << GRU_CB_EXOPC_SHFT);
  290. }
  291. /*
  292. * Architecture specific intrinsics
  293. */
  294. static inline void gru_flush_cache(void *p)
  295. {
  296. __flush_cache(p);
  297. }
  298. /*
  299. * Store the lower 64 bits of the command including the "start" bit. Then
  300. * start the instruction executing.
  301. */
  302. static inline void gru_start_instruction(struct gru_instruction *ins, unsigned long op64)
  303. {
  304. gru_ordered_store_ulong(ins, op64);
  305. mb();
  306. gru_flush_cache(ins);
  307. }
  308. /* Convert "hints" to IMA */
  309. #define CB_IMA(h) ((h) | IMA_UNMAPPED)
  310. /* Convert data segment cache line index into TRI0 / TRI1 value */
  311. #define GRU_DINDEX(i) ((i) * GRU_CACHE_LINE_BYTES)
  312. /* Inline functions for GRU instructions.
  313. * Note:
  314. * - nelem and stride are in elements
  315. * - tri0/tri1 is in bytes for the beginning of the data segment.
  316. */
  317. static inline void gru_vload_phys(void *cb, unsigned long gpa,
  318. unsigned int tri0, int iaa, unsigned long hints)
  319. {
  320. struct gru_instruction *ins = (struct gru_instruction *)cb;
  321. ins->baddr0 = (long)gpa | ((unsigned long)iaa << 62);
  322. ins->nelem = 1;
  323. ins->op1_stride = 1;
  324. gru_start_instruction(ins, __opdword(OP_VLOAD, 0, XTYPE_DW, iaa, 0,
  325. (unsigned long)tri0, CB_IMA(hints)));
  326. }
  327. static inline void gru_vstore_phys(void *cb, unsigned long gpa,
  328. unsigned int tri0, int iaa, unsigned long hints)
  329. {
  330. struct gru_instruction *ins = (struct gru_instruction *)cb;
  331. ins->baddr0 = (long)gpa | ((unsigned long)iaa << 62);
  332. ins->nelem = 1;
  333. ins->op1_stride = 1;
  334. gru_start_instruction(ins, __opdword(OP_VSTORE, 0, XTYPE_DW, iaa, 0,
  335. (unsigned long)tri0, CB_IMA(hints)));
  336. }
  337. static inline void gru_vload(void *cb, unsigned long mem_addr,
  338. unsigned int tri0, unsigned char xtype, unsigned long nelem,
  339. unsigned long stride, unsigned long hints)
  340. {
  341. struct gru_instruction *ins = (struct gru_instruction *)cb;
  342. ins->baddr0 = (long)mem_addr;
  343. ins->nelem = nelem;
  344. ins->op1_stride = stride;
  345. gru_start_instruction(ins, __opdword(OP_VLOAD, 0, xtype, IAA_RAM, 0,
  346. (unsigned long)tri0, CB_IMA(hints)));
  347. }
  348. static inline void gru_vstore(void *cb, unsigned long mem_addr,
  349. unsigned int tri0, unsigned char xtype, unsigned long nelem,
  350. unsigned long stride, unsigned long hints)
  351. {
  352. struct gru_instruction *ins = (void *)cb;
  353. ins->baddr0 = (long)mem_addr;
  354. ins->nelem = nelem;
  355. ins->op1_stride = stride;
  356. gru_start_instruction(ins, __opdword(OP_VSTORE, 0, xtype, IAA_RAM, 0,
  357. tri0, CB_IMA(hints)));
  358. }
  359. static inline void gru_ivload(void *cb, unsigned long mem_addr,
  360. unsigned int tri0, unsigned int tri1, unsigned char xtype,
  361. unsigned long nelem, unsigned long hints)
  362. {
  363. struct gru_instruction *ins = (void *)cb;
  364. ins->baddr0 = (long)mem_addr;
  365. ins->nelem = nelem;
  366. ins->tri1_bufsize = tri1;
  367. gru_start_instruction(ins, __opdword(OP_IVLOAD, 0, xtype, IAA_RAM, 0,
  368. tri0, CB_IMA(hints)));
  369. }
  370. static inline void gru_ivstore(void *cb, unsigned long mem_addr,
  371. unsigned int tri0, unsigned int tri1,
  372. unsigned char xtype, unsigned long nelem, unsigned long hints)
  373. {
  374. struct gru_instruction *ins = (void *)cb;
  375. ins->baddr0 = (long)mem_addr;
  376. ins->nelem = nelem;
  377. ins->tri1_bufsize = tri1;
  378. gru_start_instruction(ins, __opdword(OP_IVSTORE, 0, xtype, IAA_RAM, 0,
  379. tri0, CB_IMA(hints)));
  380. }
  381. static inline void gru_vset(void *cb, unsigned long mem_addr,
  382. unsigned long value, unsigned char xtype, unsigned long nelem,
  383. unsigned long stride, unsigned long hints)
  384. {
  385. struct gru_instruction *ins = (void *)cb;
  386. ins->baddr0 = (long)mem_addr;
  387. ins->op2_value_baddr1 = value;
  388. ins->nelem = nelem;
  389. ins->op1_stride = stride;
  390. gru_start_instruction(ins, __opdword(OP_VSET, 0, xtype, IAA_RAM, 0,
  391. 0, CB_IMA(hints)));
  392. }
  393. static inline void gru_ivset(void *cb, unsigned long mem_addr,
  394. unsigned int tri1, unsigned long value, unsigned char xtype,
  395. unsigned long nelem, unsigned long hints)
  396. {
  397. struct gru_instruction *ins = (void *)cb;
  398. ins->baddr0 = (long)mem_addr;
  399. ins->op2_value_baddr1 = value;
  400. ins->nelem = nelem;
  401. ins->tri1_bufsize = tri1;
  402. gru_start_instruction(ins, __opdword(OP_IVSET, 0, xtype, IAA_RAM, 0,
  403. 0, CB_IMA(hints)));
  404. }
  405. static inline void gru_vflush(void *cb, unsigned long mem_addr,
  406. unsigned long nelem, unsigned char xtype, unsigned long stride,
  407. unsigned long hints)
  408. {
  409. struct gru_instruction *ins = (void *)cb;
  410. ins->baddr0 = (long)mem_addr;
  411. ins->op1_stride = stride;
  412. ins->nelem = nelem;
  413. gru_start_instruction(ins, __opdword(OP_VFLUSH, 0, xtype, IAA_RAM, 0,
  414. 0, CB_IMA(hints)));
  415. }
  416. static inline void gru_nop(void *cb, int hints)
  417. {
  418. struct gru_instruction *ins = (void *)cb;
  419. gru_start_instruction(ins, __opdword(OP_NOP, 0, 0, 0, 0, 0, CB_IMA(hints)));
  420. }
  421. static inline void gru_bcopy(void *cb, const unsigned long src,
  422. unsigned long dest,
  423. unsigned int tri0, unsigned int xtype, unsigned long nelem,
  424. unsigned int bufsize, unsigned long hints)
  425. {
  426. struct gru_instruction *ins = (void *)cb;
  427. ins->baddr0 = (long)src;
  428. ins->op2_value_baddr1 = (long)dest;
  429. ins->nelem = nelem;
  430. ins->tri1_bufsize = bufsize;
  431. gru_start_instruction(ins, __opdword(OP_BCOPY, 0, xtype, IAA_RAM,
  432. IAA_RAM, tri0, CB_IMA(hints)));
  433. }
  434. static inline void gru_bstore(void *cb, const unsigned long src,
  435. unsigned long dest, unsigned int tri0, unsigned int xtype,
  436. unsigned long nelem, unsigned long hints)
  437. {
  438. struct gru_instruction *ins = (void *)cb;
  439. ins->baddr0 = (long)src;
  440. ins->op2_value_baddr1 = (long)dest;
  441. ins->nelem = nelem;
  442. gru_start_instruction(ins, __opdword(OP_BSTORE, 0, xtype, 0, IAA_RAM,
  443. tri0, CB_IMA(hints)));
  444. }
  445. static inline void gru_gamir(void *cb, int exopc, unsigned long src,
  446. unsigned int xtype, unsigned long hints)
  447. {
  448. struct gru_instruction *ins = (void *)cb;
  449. ins->baddr0 = (long)src;
  450. gru_start_instruction(ins, __opdword(OP_GAMIR, exopc, xtype, IAA_RAM, 0,
  451. 0, CB_IMA(hints)));
  452. }
  453. static inline void gru_gamirr(void *cb, int exopc, unsigned long src,
  454. unsigned int xtype, unsigned long hints)
  455. {
  456. struct gru_instruction *ins = (void *)cb;
  457. ins->baddr0 = (long)src;
  458. gru_start_instruction(ins, __opdword(OP_GAMIRR, exopc, xtype, IAA_RAM, 0,
  459. 0, CB_IMA(hints)));
  460. }
  461. static inline void gru_gamer(void *cb, int exopc, unsigned long src,
  462. unsigned int xtype,
  463. unsigned long operand1, unsigned long operand2,
  464. unsigned long hints)
  465. {
  466. struct gru_instruction *ins = (void *)cb;
  467. ins->baddr0 = (long)src;
  468. ins->op1_stride = operand1;
  469. ins->op2_value_baddr1 = operand2;
  470. gru_start_instruction(ins, __opdword(OP_GAMER, exopc, xtype, IAA_RAM, 0,
  471. 0, CB_IMA(hints)));
  472. }
  473. static inline void gru_gamerr(void *cb, int exopc, unsigned long src,
  474. unsigned int xtype, unsigned long operand1,
  475. unsigned long operand2, unsigned long hints)
  476. {
  477. struct gru_instruction *ins = (void *)cb;
  478. ins->baddr0 = (long)src;
  479. ins->op1_stride = operand1;
  480. ins->op2_value_baddr1 = operand2;
  481. gru_start_instruction(ins, __opdword(OP_GAMERR, exopc, xtype, IAA_RAM, 0,
  482. 0, CB_IMA(hints)));
  483. }
  484. static inline void gru_gamxr(void *cb, unsigned long src,
  485. unsigned int tri0, unsigned long hints)
  486. {
  487. struct gru_instruction *ins = (void *)cb;
  488. ins->baddr0 = (long)src;
  489. ins->nelem = 4;
  490. gru_start_instruction(ins, __opdword(OP_GAMXR, EOP_XR_CSWAP, XTYPE_DW,
  491. IAA_RAM, 0, 0, CB_IMA(hints)));
  492. }
  493. static inline void gru_mesq(void *cb, unsigned long queue,
  494. unsigned long tri0, unsigned long nelem,
  495. unsigned long hints)
  496. {
  497. struct gru_instruction *ins = (void *)cb;
  498. ins->baddr0 = (long)queue;
  499. ins->nelem = nelem;
  500. gru_start_instruction(ins, __opdword(OP_MESQ, 0, XTYPE_CL, IAA_RAM, 0,
  501. tri0, CB_IMA(hints)));
  502. }
  503. static inline unsigned long gru_get_amo_value(void *cb)
  504. {
  505. struct gru_instruction *ins = (void *)cb;
  506. return ins->avalue;
  507. }
  508. static inline int gru_get_amo_value_head(void *cb)
  509. {
  510. struct gru_instruction *ins = (void *)cb;
  511. return ins->avalue & 0xffffffff;
  512. }
  513. static inline int gru_get_amo_value_limit(void *cb)
  514. {
  515. struct gru_instruction *ins = (void *)cb;
  516. return ins->avalue >> 32;
  517. }
  518. static inline union gru_mesqhead gru_mesq_head(int head, int limit)
  519. {
  520. union gru_mesqhead mqh;
  521. mqh.head = head;
  522. mqh.limit = limit;
  523. return mqh;
  524. }
  525. /*
  526. * Get struct control_block_extended_exc_detail for CB.
  527. */
  528. extern int gru_get_cb_exception_detail(void *cb,
  529. struct control_block_extended_exc_detail *excdet);
  530. #define GRU_EXC_STR_SIZE 256
  531. /*
  532. * Control block definition for checking status
  533. */
  534. struct gru_control_block_status {
  535. unsigned int icmd :1;
  536. unsigned int ima :3;
  537. unsigned int reserved0 :4;
  538. unsigned int unused1 :24;
  539. unsigned int unused2 :24;
  540. unsigned int istatus :2;
  541. unsigned int isubstatus :4;
  542. unsigned int unused3 :2;
  543. };
  544. /* Get CB status */
  545. static inline int gru_get_cb_status(void *cb)
  546. {
  547. struct gru_control_block_status *cbs = (void *)cb;
  548. return cbs->istatus;
  549. }
  550. /* Get CB message queue substatus */
  551. static inline int gru_get_cb_message_queue_substatus(void *cb)
  552. {
  553. struct gru_control_block_status *cbs = (void *)cb;
  554. return cbs->isubstatus & CBSS_MSG_QUEUE_MASK;
  555. }
  556. /* Get CB substatus */
  557. static inline int gru_get_cb_substatus(void *cb)
  558. {
  559. struct gru_control_block_status *cbs = (void *)cb;
  560. return cbs->isubstatus;
  561. }
  562. /*
  563. * User interface to check an instruction status. UPM and exceptions
  564. * are handled automatically. However, this function does NOT wait
  565. * for an active instruction to complete.
  566. *
  567. */
  568. static inline int gru_check_status(void *cb)
  569. {
  570. struct gru_control_block_status *cbs = (void *)cb;
  571. int ret;
  572. ret = cbs->istatus;
  573. if (ret != CBS_ACTIVE)
  574. ret = gru_check_status_proc(cb);
  575. return ret;
  576. }
  577. /*
  578. * User interface (via inline function) to wait for an instruction
  579. * to complete. Completion status (IDLE or EXCEPTION is returned
  580. * to the user. Exception due to hardware errors are automatically
  581. * retried before returning an exception.
  582. *
  583. */
  584. static inline int gru_wait(void *cb)
  585. {
  586. return gru_wait_proc(cb);
  587. }
  588. /*
  589. * Wait for CB to complete. Aborts program if error. (Note: error does NOT
  590. * mean TLB mis - only fatal errors such as memory parity error or user
  591. * bugs will cause termination.
  592. */
  593. static inline void gru_wait_abort(void *cb)
  594. {
  595. gru_wait_abort_proc(cb);
  596. }
  597. /*
  598. * Get a pointer to the start of a gseg
  599. * p - Any valid pointer within the gseg
  600. */
  601. static inline void *gru_get_gseg_pointer (void *p)
  602. {
  603. return (void *)((unsigned long)p & ~(GRU_GSEG_PAGESIZE - 1));
  604. }
  605. /*
  606. * Get a pointer to a control block
  607. * gseg - GSeg address returned from gru_get_thread_gru_segment()
  608. * index - index of desired CB
  609. */
  610. static inline void *gru_get_cb_pointer(void *gseg,
  611. int index)
  612. {
  613. return gseg + GRU_CB_BASE + index * GRU_HANDLE_STRIDE;
  614. }
  615. /*
  616. * Get a pointer to a cacheline in the data segment portion of a GSeg
  617. * gseg - GSeg address returned from gru_get_thread_gru_segment()
  618. * index - index of desired cache line
  619. */
  620. static inline void *gru_get_data_pointer(void *gseg, int index)
  621. {
  622. return gseg + GRU_DS_BASE + index * GRU_CACHE_LINE_BYTES;
  623. }
  624. /*
  625. * Convert a vaddr into the tri index within the GSEG
  626. * vaddr - virtual address of within gseg
  627. */
  628. static inline int gru_get_tri(void *vaddr)
  629. {
  630. return ((unsigned long)vaddr & (GRU_GSEG_PAGESIZE - 1)) - GRU_DS_BASE;
  631. }
  632. #endif /* __GRU_INSTRUCTIONS_H__ */