drivers/misc/sgi-gru/gru_instructions.h

   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
  19 #ifndef __GRU_INSTRUCTIONS_H__
  20 #define __GRU_INSTRUCTIONS_H__
  21
  22 extern int gru_check_status_proc(void *cb);
  23 extern int gru_wait_proc(void *cb);
  24 extern void gru_wait_abort_proc(void *cb);
  25
  26
  27
  28 /*
  29  * Architecture dependent functions
  30  */
  31
  32 #if defined(CONFIG_IA64)
  33 #include <linux/compiler.h>
  34 #include <asm/intrinsics.h>
  35 #define __flush_cache(p)                ia64_fc((unsigned long)p)
  36 /* Use volatile on IA64 to ensure ordering via st4.rel */
  37 #define gru_ordered_store_ulong(p, v)                                   \
  38                 do {                                                    \
  39                         barrier();                                      \
  40                         *((volatile unsigned long *)(p)) = v; /* force st.rel */        \
  41                 } while (0)
  42 #elif defined(CONFIG_X86_64)
  43 #define __flush_cache(p)                clflush(p)
  44 #define gru_ordered_store_ulong(p, v)                                   \
  45                 do {                                                    \
  46                         barrier();                                      \
  47                         *(unsigned long *)p = v;                        \
  48                 } while (0)
  49 #else
  50 #error "Unsupported architecture"
  51 #endif
  52
  53 /*
  54  * Control block status and exception codes
  55  */
  56 #define CBS_IDLE                        0
  57 #define CBS_EXCEPTION                   1
  58 #define CBS_ACTIVE                      2
  59 #define CBS_CALL_OS                     3
  60
  61 /* CB substatus bitmasks */
  62 #define CBSS_MSG_QUEUE_MASK             7
  63 #define CBSS_IMPLICIT_ABORT_ACTIVE_MASK 8
  64
  65 /* CB substatus message queue values (low 3 bits of substatus) */
  66 #define CBSS_NO_ERROR                   0
  67 #define CBSS_LB_OVERFLOWED              1
  68 #define CBSS_QLIMIT_REACHED             2
  69 #define CBSS_PAGE_OVERFLOW              3
  70 #define CBSS_AMO_NACKED                 4
  71 #define CBSS_PUT_NACKED                 5
  72
  73 /*
  74  * Structure used to fetch exception detail for CBs that terminate with
  75  * CBS_EXCEPTION
  76  */
  77 struct control_block_extended_exc_detail {
  78         unsigned long   cb;
  79         int             opc;
  80         int             ecause;
  81         int             exopc;
  82         long            exceptdet0;
  83         int             exceptdet1;
  84         int             cbrstate;
  85         int             cbrexecstatus;
  86 };
  87
  88 /*
  89  * Instruction formats
  90  */
  91
  92 /*
  93  * Generic instruction format.
  94  * This definition has precise bit field definitions.
  95  */
  96 struct gru_instruction_bits {
  97     /* DW 0  - low */
  98     unsigned int                icmd:      1;
  99     unsigned char               ima:       3;   /* CB_DelRep, unmapped mode */
 100     unsigned char               reserved0: 4;
 101     unsigned int                xtype:     3;
 102     unsigned int                iaa0:      2;
 103     unsigned int                iaa1:      2;
 104     unsigned char               reserved1: 1;
 105     unsigned char               opc:       8;   /* opcode */
 106     unsigned char               exopc:     8;   /* extended opcode */
 107     /* DW 0  - high */
 108     unsigned int                idef2:    22;   /* TRi0 */
 109     unsigned char               reserved2: 2;
 110     unsigned char               istatus:   2;
 111     unsigned char               isubstatus:4;
 112     unsigned char               reserved3: 1;
 113     unsigned char               tlb_fault_color: 1;
 114     /* DW 1 */
 115     unsigned long               idef4;          /* 42 bits: TRi1, BufSize */
 116     /* DW 2-6 */
 117     unsigned long               idef1;          /* BAddr0 */
 118     unsigned long               idef5;          /* Nelem */
 119     unsigned long               idef6;          /* Stride, Operand1 */
 120     unsigned long               idef3;          /* BAddr1, Value, Operand2 */
 121     unsigned long               reserved4;
 122     /* DW 7 */
 123     unsigned long               avalue;          /* AValue */
 124 };
 125
 126 /*
 127  * Generic instruction with friendlier names. This format is used
 128  * for inline instructions.
 129  */
 130 struct gru_instruction {
 131     /* DW 0 */
 132     union {
 133         unsigned long           op64;    /* icmd,xtype,iaa0,ima,opc,tri0 */
 134         struct {
 135                 unsigned int    op32;
 136                 unsigned int    tri0;
 137         };
 138     };
 139     unsigned long               tri1_bufsize;           /* DW 1 */
 140     unsigned long               baddr0;                 /* DW 2 */
 141     unsigned long               nelem;                  /* DW 3 */
 142     unsigned long               op1_stride;             /* DW 4 */
 143     unsigned long               op2_value_baddr1;       /* DW 5 */
 144     unsigned long               reserved0;              /* DW 6 */
 145     unsigned long               avalue;                 /* DW 7 */
 146 };
 147
 148 /* Some shifts and masks for the low 64 bits of a GRU command */
 149 #define GRU_CB_ICMD_SHFT        0
 150 #define GRU_CB_ICMD_MASK        0x1
 151 #define GRU_CB_XTYPE_SHFT       8
 152 #define GRU_CB_XTYPE_MASK       0x7
 153 #define GRU_CB_IAA0_SHFT        11
 154 #define GRU_CB_IAA0_MASK        0x3
 155 #define GRU_CB_IAA1_SHFT        13
 156 #define GRU_CB_IAA1_MASK        0x3
 157 #define GRU_CB_IMA_SHFT         1
 158 #define GRU_CB_IMA_MASK         0x3
 159 #define GRU_CB_OPC_SHFT         16
 160 #define GRU_CB_OPC_MASK         0xff
 161 #define GRU_CB_EXOPC_SHFT       24
 162 #define GRU_CB_EXOPC_MASK       0xff
 163 #define GRU_IDEF2_SHFT          32
 164 #define GRU_IDEF2_MASK          0x3ffff
 165 #define GRU_ISTATUS_SHFT        56
 166 #define GRU_ISTATUS_MASK        0x3
 167
 168 /* GRU instruction opcodes (opc field) */
 169 #define OP_NOP          0x00
 170 #define OP_BCOPY        0x01
 171 #define OP_VLOAD        0x02
 172 #define OP_IVLOAD       0x03
 173 #define OP_VSTORE       0x04
 174 #define OP_IVSTORE      0x05
 175 #define OP_VSET         0x06
 176 #define OP_IVSET        0x07
 177 #define OP_MESQ         0x08
 178 #define OP_GAMXR        0x09
 179 #define OP_GAMIR        0x0a
 180 #define OP_GAMIRR       0x0b
 181 #define OP_GAMER        0x0c
 182 #define OP_GAMERR       0x0d
 183 #define OP_BSTORE       0x0e
 184 #define OP_VFLUSH       0x0f
 185
 186
 187 /* Extended opcodes values (exopc field) */
 188
 189 /* GAMIR - AMOs with implicit operands */
 190 #define EOP_IR_FETCH    0x01 /* Plain fetch of memory */
 191 #define EOP_IR_CLR      0x02 /* Fetch and clear */
 192 #define EOP_IR_INC      0x05 /* Fetch and increment */
 193 #define EOP_IR_DEC      0x07 /* Fetch and decrement */
 194 #define EOP_IR_QCHK1    0x0d /* Queue check, 64 byte msg */
 195 #define EOP_IR_QCHK2    0x0e /* Queue check, 128 byte msg */
 196
 197 /* GAMIRR - Registered AMOs with implicit operands */
 198 #define EOP_IRR_FETCH   0x01 /* Registered fetch of memory */
 199 #define EOP_IRR_CLR     0x02 /* Registered fetch and clear */
 200 #define EOP_IRR_INC     0x05 /* Registered fetch and increment */
 201 #define EOP_IRR_DEC     0x07 /* Registered fetch and decrement */
 202 #define EOP_IRR_DECZ    0x0f /* Registered fetch and decrement, update on zero*/
 203
 204 /* GAMER - AMOs with explicit operands */
 205 #define EOP_ER_SWAP     0x00 /* Exchange argument and memory */
 206 #define EOP_ER_OR       0x01 /* Logical OR with memory */
 207 #define EOP_ER_AND      0x02 /* Logical AND with memory */
 208 #define EOP_ER_XOR      0x03 /* Logical XOR with memory */
 209 #define EOP_ER_ADD      0x04 /* Add value to memory */
 210 #define EOP_ER_CSWAP    0x08 /* Compare with operand2, write operand1 if match*/
 211 #define EOP_ER_CADD     0x0c /* Queue check, operand1*64 byte msg */
 212
 213 /* GAMERR - Registered AMOs with explicit operands */
 214 #define EOP_ERR_SWAP    0x00 /* Exchange argument and memory */
 215 #define EOP_ERR_OR      0x01 /* Logical OR with memory */
 216 #define EOP_ERR_AND     0x02 /* Logical AND with memory */
 217 #define EOP_ERR_XOR     0x03 /* Logical XOR with memory */
 218 #define EOP_ERR_ADD     0x04 /* Add value to memory */
 219 #define EOP_ERR_CSWAP   0x08 /* Compare with operand2, write operand1 if match*/
 220 #define EOP_ERR_EPOLL   0x09 /* Poll for equality */
 221 #define EOP_ERR_NPOLL   0x0a /* Poll for inequality */
 222
 223 /* GAMXR - SGI Arithmetic unit */
 224 #define EOP_XR_CSWAP    0x0b /* Masked compare exchange */
 225
 226
 227 /* Transfer types (xtype field) */
 228 #define XTYPE_B         0x0     /* byte */
 229 #define XTYPE_S         0x1     /* short (2-byte) */
 230 #define XTYPE_W         0x2     /* word (4-byte) */
 231 #define XTYPE_DW        0x3     /* doubleword (8-byte) */
 232 #define XTYPE_CL        0x6     /* cacheline (64-byte) */
 233
 234
 235 /* Instruction access attributes (iaa0, iaa1 fields) */
 236 #define IAA_RAM         0x0     /* normal cached RAM access */
 237 #define IAA_NCRAM       0x2     /* noncoherent RAM access */
 238 #define IAA_MMIO        0x1     /* noncoherent memory-mapped I/O space */
 239 #define IAA_REGISTER    0x3     /* memory-mapped registers, etc. */
 240
 241
 242 /* Instruction mode attributes (ima field) */
 243 #define IMA_MAPPED      0x0     /* Virtual mode  */
 244 #define IMA_CB_DELAY    0x1     /* hold read responses until status changes */
 245 #define IMA_UNMAPPED    0x2     /* bypass the TLBs (OS only) */
 246 #define IMA_INTERRUPT   0x4     /* Interrupt when instruction completes */
 247
 248 /* CBE ecause bits */
 249 #define CBE_CAUSE_RI                            (1 << 0)
 250 #define CBE_CAUSE_INVALID_INSTRUCTION           (1 << 1)
 251 #define CBE_CAUSE_UNMAPPED_MODE_FORBIDDEN       (1 << 2)
 252 #define CBE_CAUSE_PE_CHECK_DATA_ERROR           (1 << 3)
 253 #define CBE_CAUSE_IAA_GAA_MISMATCH              (1 << 4)
 254 #define CBE_CAUSE_DATA_SEGMENT_LIMIT_EXCEPTION  (1 << 5)
 255 #define CBE_CAUSE_OS_FATAL_TLB_FAULT            (1 << 6)
 256 #define CBE_CAUSE_EXECUTION_HW_ERROR            (1 << 7)
 257 #define CBE_CAUSE_TLBHW_ERROR                   (1 << 8)
 258 #define CBE_CAUSE_RA_REQUEST_TIMEOUT            (1 << 9)
 259 #define CBE_CAUSE_HA_REQUEST_TIMEOUT            (1 << 10)
 260 #define CBE_CAUSE_RA_RESPONSE_FATAL             (1 << 11)
 261 #define CBE_CAUSE_RA_RESPONSE_NON_FATAL         (1 << 12)
 262 #define CBE_CAUSE_HA_RESPONSE_FATAL             (1 << 13)
 263 #define CBE_CAUSE_HA_RESPONSE_NON_FATAL         (1 << 14)
 264 #define CBE_CAUSE_ADDRESS_SPACE_DECODE_ERROR    (1 << 15)
 265 #define CBE_CAUSE_PROTOCOL_STATE_DATA_ERROR     (1 << 16)
 266 #define CBE_CAUSE_RA_RESPONSE_DATA_ERROR        (1 << 17)
 267 #define CBE_CAUSE_HA_RESPONSE_DATA_ERROR        (1 << 18)
 268
 269 /* CBE cbrexecstatus bits */
 270 #define CBR_EXS_ABORT_OCC_BIT                   0
 271 #define CBR_EXS_INT_OCC_BIT                     1
 272 #define CBR_EXS_PENDING_BIT                     2
 273 #define CBR_EXS_QUEUED_BIT                      3
 274 #define CBR_EXS_TLB_INVAL_BIT                   4
 275 #define CBR_EXS_EXCEPTION_BIT                   5
 276
 277 #define CBR_EXS_ABORT_OCC                       (1 << CBR_EXS_ABORT_OCC_BIT)
 278 #define CBR_EXS_INT_OCC                         (1 << CBR_EXS_INT_OCC_BIT)
 279 #define CBR_EXS_PENDING                         (1 << CBR_EXS_PENDING_BIT)
 280 #define CBR_EXS_QUEUED                          (1 << CBR_EXS_QUEUED_BIT)
 281 #define CBR_TLB_INVAL                           (1 << CBR_EXS_TLB_INVAL_BIT)
 282 #define CBR_EXS_EXCEPTION                       (1 << CBR_EXS_EXCEPTION_BIT)
 283
 284 /*
 285  * Exceptions are retried for the following cases. If any OTHER bits are set
 286  * in ecause, the exception is not retryable.
 287  */
 288 #define EXCEPTION_RETRY_BITS (CBE_CAUSE_EXECUTION_HW_ERROR |            \
 289                               CBE_CAUSE_TLBHW_ERROR |                   \
 290                               CBE_CAUSE_RA_REQUEST_TIMEOUT |            \
 291                               CBE_CAUSE_RA_RESPONSE_NON_FATAL |         \
 292                               CBE_CAUSE_HA_RESPONSE_NON_FATAL |         \
 293                               CBE_CAUSE_RA_RESPONSE_DATA_ERROR |        \
 294                               CBE_CAUSE_HA_RESPONSE_DATA_ERROR          \
 295                               )
 296
 297 /* Message queue head structure */
 298 union gru_mesqhead {
 299         unsigned long   val;
 300         struct {
 301                 unsigned int    head;
 302                 unsigned int    limit;
 303         };
 304 };
 305
 306
 307 /* Generate the low word of a GRU instruction */
 308 static inline unsigned long
 309 __opdword(unsigned char opcode, unsigned char exopc, unsigned char xtype,
 310        unsigned char iaa0, unsigned char iaa1,
 311        unsigned long idef2, unsigned char ima)
 312 {
 313     return (1 << GRU_CB_ICMD_SHFT) |
 314            ((unsigned long)CBS_ACTIVE << GRU_ISTATUS_SHFT) |
 315            (idef2<< GRU_IDEF2_SHFT) |
 316            (iaa0 << GRU_CB_IAA0_SHFT) |
 317            (iaa1 << GRU_CB_IAA1_SHFT) |
 318            (ima << GRU_CB_IMA_SHFT) |
 319            (xtype << GRU_CB_XTYPE_SHFT) |
 320            (opcode << GRU_CB_OPC_SHFT) |
 321            (exopc << GRU_CB_EXOPC_SHFT);
 322 }
 323
 324 /*
 325  * Architecture specific intrinsics
 326  */
 327 static inline void gru_flush_cache(void *p)
 328 {
 329         __flush_cache(p);
 330 }
 331
 332 /*
 333  * Store the lower 64 bits of the command including the "start" bit. Then
 334  * start the instruction executing.
 335  */
 336 static inline void gru_start_instruction(struct gru_instruction *ins, unsigned long op64)
 337 {
 338         gru_ordered_store_ulong(ins, op64);
 339         mb();
 340         gru_flush_cache(ins);
 341 }
 342
 343
 344 /* Convert "hints" to IMA */
 345 #define CB_IMA(h)               ((h) | IMA_UNMAPPED)
 346
 347 /* Convert data segment cache line index into TRI0 / TRI1 value */
 348 #define GRU_DINDEX(i)           ((i) * GRU_CACHE_LINE_BYTES)
 349
 350 /* Inline functions for GRU instructions.
 351  *     Note:
 352  *      - nelem and stride are in elements
 353  *      - tri0/tri1 is in bytes for the beginning of the data segment.
 354  */
 355 static inline void gru_vload_phys(void *cb, unsigned long gpa,
 356                 unsigned int tri0, int iaa, unsigned long hints)
 357 {
 358         struct gru_instruction *ins = (struct gru_instruction *)cb;
 359
 360         ins->baddr0 = (long)gpa | ((unsigned long)iaa << 62);
 361         ins->nelem = 1;
 362         ins->op1_stride = 1;
 363         gru_start_instruction(ins, __opdword(OP_VLOAD, 0, XTYPE_DW, iaa, 0,
 364                                         (unsigned long)tri0, CB_IMA(hints)));
 365 }
 366
 367 static inline void gru_vload(void *cb, unsigned long mem_addr,
 368                 unsigned int tri0, unsigned char xtype, unsigned long nelem,
 369                 unsigned long stride, unsigned long hints)
 370 {
 371         struct gru_instruction *ins = (struct gru_instruction *)cb;
 372
 373         ins->baddr0 = (long)mem_addr;
 374         ins->nelem = nelem;
 375         ins->op1_stride = stride;
 376         gru_start_instruction(ins, __opdword(OP_VLOAD, 0, xtype, IAA_RAM, 0,
 377                                         (unsigned long)tri0, CB_IMA(hints)));
 378 }
 379
 380 static inline void gru_vstore(void *cb, unsigned long mem_addr,
 381                 unsigned int tri0, unsigned char xtype, unsigned long nelem,
 382                 unsigned long stride, unsigned long hints)
 383 {
 384         struct gru_instruction *ins = (void *)cb;
 385
 386         ins->baddr0 = (long)mem_addr;
 387         ins->nelem = nelem;
 388         ins->op1_stride = stride;
 389         gru_start_instruction(ins, __opdword(OP_VSTORE, 0, xtype, IAA_RAM, 0,
 390                                         tri0, CB_IMA(hints)));
 391 }
 392
 393 static inline void gru_ivload(void *cb, unsigned long mem_addr,
 394                 unsigned int tri0, unsigned int tri1, unsigned char xtype,
 395                 unsigned long nelem, unsigned long hints)
 396 {
 397         struct gru_instruction *ins = (void *)cb;
 398
 399         ins->baddr0 = (long)mem_addr;
 400         ins->nelem = nelem;
 401         ins->tri1_bufsize = tri1;
 402         gru_start_instruction(ins, __opdword(OP_IVLOAD, 0, xtype, IAA_RAM, 0,
 403                                         tri0, CB_IMA(hints)));
 404 }
 405
 406 static inline void gru_ivstore(void *cb, unsigned long mem_addr,
 407                 unsigned int tri0, unsigned int tri1,
 408                 unsigned char xtype, unsigned long nelem, unsigned long hints)
 409 {
 410         struct gru_instruction *ins = (void *)cb;
 411
 412         ins->baddr0 = (long)mem_addr;
 413         ins->nelem = nelem;
 414         ins->tri1_bufsize = tri1;
 415         gru_start_instruction(ins, __opdword(OP_IVSTORE, 0, xtype, IAA_RAM, 0,
 416                                         tri0, CB_IMA(hints)));
 417 }
 418
 419 static inline void gru_vset(void *cb, unsigned long mem_addr,
 420                 unsigned long value, unsigned char xtype, unsigned long nelem,
 421                 unsigned long stride, unsigned long hints)
 422 {
 423         struct gru_instruction *ins = (void *)cb;
 424
 425         ins->baddr0 = (long)mem_addr;
 426         ins->op2_value_baddr1 = value;
 427         ins->nelem = nelem;
 428         ins->op1_stride = stride;
 429         gru_start_instruction(ins, __opdword(OP_VSET, 0, xtype, IAA_RAM, 0,
 430                                          0, CB_IMA(hints)));
 431 }
 432
 433 static inline void gru_ivset(void *cb, unsigned long mem_addr,
 434                 unsigned int tri1, unsigned long value, unsigned char xtype,
 435                 unsigned long nelem, unsigned long hints)
 436 {
 437         struct gru_instruction *ins = (void *)cb;
 438
 439         ins->baddr0 = (long)mem_addr;
 440         ins->op2_value_baddr1 = value;
 441         ins->nelem = nelem;
 442         ins->tri1_bufsize = tri1;
 443         gru_start_instruction(ins, __opdword(OP_IVSET, 0, xtype, IAA_RAM, 0,
 444                                         0, CB_IMA(hints)));
 445 }
 446
 447 static inline void gru_vflush(void *cb, unsigned long mem_addr,
 448                 unsigned long nelem, unsigned char xtype, unsigned long stride,
 449                 unsigned long hints)
 450 {
 451         struct gru_instruction *ins = (void *)cb;
 452
 453         ins->baddr0 = (long)mem_addr;
 454         ins->op1_stride = stride;
 455         ins->nelem = nelem;
 456         gru_start_instruction(ins, __opdword(OP_VFLUSH, 0, xtype, IAA_RAM, 0,
 457                                         0, CB_IMA(hints)));
 458 }
 459
 460 static inline void gru_nop(void *cb, int hints)
 461 {
 462         struct gru_instruction *ins = (void *)cb;
 463
 464         gru_start_instruction(ins, __opdword(OP_NOP, 0, 0, 0, 0, 0, CB_IMA(hints)));
 465 }
 466
 467
 468 static inline void gru_bcopy(void *cb, const unsigned long src,
 469                 unsigned long dest,
 470                 unsigned int tri0, unsigned int xtype, unsigned long nelem,
 471                 unsigned int bufsize, unsigned long hints)
 472 {
 473         struct gru_instruction *ins = (void *)cb;
 474
 475         ins->baddr0 = (long)src;
 476         ins->op2_value_baddr1 = (long)dest;
 477         ins->nelem = nelem;
 478         ins->tri1_bufsize = bufsize;
 479         gru_start_instruction(ins, __opdword(OP_BCOPY, 0, xtype, IAA_RAM,
 480                                         IAA_RAM, tri0, CB_IMA(hints)));
 481 }
 482
 483 static inline void gru_bstore(void *cb, const unsigned long src,
 484                 unsigned long dest, unsigned int tri0, unsigned int xtype,
 485                 unsigned long nelem, unsigned long hints)
 486 {
 487         struct gru_instruction *ins = (void *)cb;
 488
 489         ins->baddr0 = (long)src;
 490         ins->op2_value_baddr1 = (long)dest;
 491         ins->nelem = nelem;
 492         gru_start_instruction(ins, __opdword(OP_BSTORE, 0, xtype, 0, IAA_RAM,
 493                                         tri0, CB_IMA(hints)));
 494 }
 495
 496 static inline void gru_gamir(void *cb, int exopc, unsigned long src,
 497                 unsigned int xtype, unsigned long hints)
 498 {
 499         struct gru_instruction *ins = (void *)cb;
 500
 501         ins->baddr0 = (long)src;
 502         gru_start_instruction(ins, __opdword(OP_GAMIR, exopc, xtype, IAA_RAM, 0,
 503                                         0, CB_IMA(hints)));
 504 }
 505
 506 static inline void gru_gamirr(void *cb, int exopc, unsigned long src,
 507                 unsigned int xtype, unsigned long hints)
 508 {
 509         struct gru_instruction *ins = (void *)cb;
 510
 511         ins->baddr0 = (long)src;
 512         gru_start_instruction(ins, __opdword(OP_GAMIRR, exopc, xtype, IAA_RAM, 0,
 513                                         0, CB_IMA(hints)));
 514 }
 515
 516 static inline void gru_gamer(void *cb, int exopc, unsigned long src,
 517                 unsigned int xtype,
 518                 unsigned long operand1, unsigned long operand2,
 519                 unsigned long hints)
 520 {
 521         struct gru_instruction *ins = (void *)cb;
 522
 523         ins->baddr0 = (long)src;
 524         ins->op1_stride = operand1;
 525         ins->op2_value_baddr1 = operand2;
 526         gru_start_instruction(ins, __opdword(OP_GAMER, exopc, xtype, IAA_RAM, 0,
 527                                         0, CB_IMA(hints)));
 528 }
 529
 530 static inline void gru_gamerr(void *cb, int exopc, unsigned long src,
 531                 unsigned int xtype, unsigned long operand1,
 532                 unsigned long operand2, unsigned long hints)
 533 {
 534         struct gru_instruction *ins = (void *)cb;
 535
 536         ins->baddr0 = (long)src;
 537         ins->op1_stride = operand1;
 538         ins->op2_value_baddr1 = operand2;
 539         gru_start_instruction(ins, __opdword(OP_GAMERR, exopc, xtype, IAA_RAM, 0,
 540                                         0, CB_IMA(hints)));
 541 }
 542
 543 static inline void gru_gamxr(void *cb, unsigned long src,
 544                 unsigned int tri0, unsigned long hints)
 545 {
 546         struct gru_instruction *ins = (void *)cb;
 547
 548         ins->baddr0 = (long)src;
 549         ins->nelem = 4;
 550         gru_start_instruction(ins, __opdword(OP_GAMXR, EOP_XR_CSWAP, XTYPE_DW,
 551                                  IAA_RAM, 0, 0, CB_IMA(hints)));
 552 }
 553
 554 static inline void gru_mesq(void *cb, unsigned long queue,
 555                 unsigned long tri0, unsigned long nelem,
 556                 unsigned long hints)
 557 {
 558         struct gru_instruction *ins = (void *)cb;
 559
 560         ins->baddr0 = (long)queue;
 561         ins->nelem = nelem;
 562         gru_start_instruction(ins, __opdword(OP_MESQ, 0, XTYPE_CL, IAA_RAM, 0,
 563                                         tri0, CB_IMA(hints)));
 564 }
 565
 566 static inline unsigned long gru_get_amo_value(void *cb)
 567 {
 568         struct gru_instruction *ins = (void *)cb;
 569
 570         return ins->avalue;
 571 }
 572
 573 static inline int gru_get_amo_value_head(void *cb)
 574 {
 575         struct gru_instruction *ins = (void *)cb;
 576
 577         return ins->avalue & 0xffffffff;
 578 }
 579
 580 static inline int gru_get_amo_value_limit(void *cb)
 581 {
 582         struct gru_instruction *ins = (void *)cb;
 583
 584         return ins->avalue >> 32;
 585 }
 586
 587 static inline union gru_mesqhead  gru_mesq_head(int head, int limit)
 588 {
 589         union gru_mesqhead mqh;
 590
 591         mqh.head = head;
 592         mqh.limit = limit;
 593         return mqh;
 594 }
 595
 596 /*
 597  * Get struct control_block_extended_exc_detail for CB.
 598  */
 599 extern int gru_get_cb_exception_detail(void *cb,
 600                        struct control_block_extended_exc_detail *excdet);
 601
 602 #define GRU_EXC_STR_SIZE                256
 603
 604
 605 /*
 606  * Control block definition for checking status
 607  */
 608 struct gru_control_block_status {
 609         unsigned int    icmd            :1;
 610         unsigned int    ima             :3;
 611         unsigned int    reserved0       :4;
 612         unsigned int    unused1         :24;
 613         unsigned int    unused2         :24;
 614         unsigned int    istatus         :2;
 615         unsigned int    isubstatus      :4;
 616         unsigned int    unused3         :2;
 617 };
 618
 619 /* Get CB status */
 620 static inline int gru_get_cb_status(void *cb)
 621 {
 622         struct gru_control_block_status *cbs = (void *)cb;
 623
 624         return cbs->istatus;
 625 }
 626
 627 /* Get CB message queue substatus */
 628 static inline int gru_get_cb_message_queue_substatus(void *cb)
 629 {
 630         struct gru_control_block_status *cbs = (void *)cb;
 631
 632         return cbs->isubstatus & CBSS_MSG_QUEUE_MASK;
 633 }
 634
 635 /* Get CB substatus */
 636 static inline int gru_get_cb_substatus(void *cb)
 637 {
 638         struct gru_control_block_status *cbs = (void *)cb;
 639
 640         return cbs->isubstatus;
 641 }
 642
 643 /*
 644  * User interface to check an instruction status. UPM and exceptions
 645  * are handled automatically. However, this function does NOT wait
 646  * for an active instruction to complete.
 647  *
 648  */
 649 static inline int gru_check_status(void *cb)
 650 {
 651         struct gru_control_block_status *cbs = (void *)cb;
 652         int ret;
 653
 654         ret = cbs->istatus;
 655         if (ret != CBS_ACTIVE)
 656                 ret = gru_check_status_proc(cb);
 657         return ret;
 658 }
 659
 660 /*
 661  * User interface (via inline function) to wait for an instruction
 662  * to complete. Completion status (IDLE or EXCEPTION is returned
 663  * to the user. Exception due to hardware errors are automatically
 664  * retried before returning an exception.
 665  *
 666  */
 667 static inline int gru_wait(void *cb)
 668 {
 669         return gru_wait_proc(cb);
 670 }
 671
 672 /*
 673  * Wait for CB to complete. Aborts program if error. (Note: error does NOT
 674  * mean TLB mis - only fatal errors such as memory parity error or user
 675  * bugs will cause termination.
 676  */
 677 static inline void gru_wait_abort(void *cb)
 678 {
 679         gru_wait_abort_proc(cb);
 680 }
 681
 682 /*
 683  * Get a pointer to the start of a gseg
 684  *      p       - Any valid pointer within the gseg
 685  */
 686 static inline void *gru_get_gseg_pointer (void *p)
 687 {
 688         return (void *)((unsigned long)p & ~(GRU_GSEG_PAGESIZE - 1));
 689 }
 690
 691 /*
 692  * Get a pointer to a control block
 693  *      gseg    - GSeg address returned from gru_get_thread_gru_segment()
 694  *      index   - index of desired CB
 695  */
 696 static inline void *gru_get_cb_pointer(void *gseg,
 697                                                       int index)
 698 {
 699         return gseg + GRU_CB_BASE + index * GRU_HANDLE_STRIDE;
 700 }
 701
 702 /*
 703  * Get a pointer to a cacheline in the data segment portion of a GSeg
 704  *      gseg    - GSeg address returned from gru_get_thread_gru_segment()
 705  *      index   - index of desired cache line
 706  */
 707 static inline void *gru_get_data_pointer(void *gseg, int index)
 708 {
 709         return gseg + GRU_DS_BASE + index * GRU_CACHE_LINE_BYTES;
 710 }
 711
 712 /*
 713  * Convert a vaddr into the tri index within the GSEG
 714  *      vaddr           - virtual address of within gseg
 715  */
 716 static inline int gru_get_tri(void *vaddr)
 717 {
 718         return ((unsigned long)vaddr & (GRU_GSEG_PAGESIZE - 1)) - GRU_DS_BASE;
 719 }
 720 #endif          /* __GRU_INSTRUCTIONS_H__ */