2 * Copyright (c) 2008 Silicon Graphics, Inc. All Rights Reserved.
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.
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.
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
19 #ifndef __GRU_INSTRUCTIONS_H__
20 #define __GRU_INSTRUCTIONS_H__
22 #define gru_flush_cache_hook(p)
23 #define gru_emulator_wait_hook(p, w)
26 * Architecture dependent functions
29 #if defined CONFIG_IA64
30 #include <linux/compiler.h>
31 #include <asm/intrinsics.h>
32 #define __flush_cache(p) ia64_fc(p)
33 /* Use volatile on IA64 to ensure ordering via st4.rel */
34 #define gru_ordered_store_int(p,v) \
37 *((volatile int *)(p)) = v; /* force st.rel */ \
39 #elif defined CONFIG_X86_64
40 #define __flush_cache(p) clflush(p)
41 #define gru_ordered_store_int(p,v) \
47 #error "Unsupported architecture"
51 * Control block status and exception codes
54 #define CBS_EXCEPTION 1
58 /* CB substatus bitmasks */
59 #define CBSS_MSG_QUEUE_MASK 7
60 #define CBSS_IMPLICIT_ABORT_ACTIVE_MASK 8
62 /* CB substatus message queue values (low 3 bits of substatus) */
63 #define CBSS_NO_ERROR 0
64 #define CBSS_LB_OVERFLOWED 1
65 #define CBSS_QLIMIT_REACHED 2
66 #define CBSS_PAGE_OVERFLOW 3
67 #define CBSS_AMO_NACKED 4
68 #define CBSS_PUT_NACKED 5
71 * Structure used to fetch exception detail for CBs that terminate with
74 struct control_block_extended_exc_detail
{
88 * Generic instruction format.
89 * This definition has precise bit field definitions.
91 struct gru_instruction_bits
{
94 unsigned char ima
: 3; /* CB_DelRep, unmapped mode */
95 unsigned char reserved0
: 4;
96 unsigned int xtype
: 3;
99 unsigned char reserved1
: 1;
100 unsigned char opc
: 8; /* opcode */
101 unsigned char exopc
: 8; /* extended opcode */
103 unsigned int idef2
: 22; /* TRi0 */
104 unsigned char reserved2
: 2;
105 unsigned char istatus
: 2;
106 unsigned char isubstatus
:4;
107 unsigned char reserved3
: 2;
109 unsigned long idef4
; /* 42 bits: TRi1, BufSize */
111 unsigned long idef1
; /* BAddr0 */
112 unsigned long idef5
; /* Nelem */
113 unsigned long idef6
; /* Stride, Operand1 */
114 unsigned long idef3
; /* BAddr1, Value, Operand2 */
115 unsigned long reserved4
;
117 unsigned long avalue
; /* AValue */
121 * Generic instruction with friendlier names. This format is used
122 * for inline instructions.
124 struct gru_instruction
{
126 unsigned int op32
; /* icmd,xtype,iaa0,ima,opc */
128 unsigned long tri1_bufsize
; /* DW 1 */
129 unsigned long baddr0
; /* DW 2 */
130 unsigned long nelem
; /* DW 3 */
131 unsigned long op1_stride
; /* DW 4 */
132 unsigned long op2_value_baddr1
; /* DW 5 */
133 unsigned long reserved0
; /* DW 6 */
134 unsigned long avalue
; /* DW 7 */
137 /* Some shifts and masks for the low 32 bits of a GRU command */
138 #define GRU_CB_ICMD_SHFT 0
139 #define GRU_CB_ICMD_MASK 0x1
140 #define GRU_CB_XTYPE_SHFT 8
141 #define GRU_CB_XTYPE_MASK 0x7
142 #define GRU_CB_IAA0_SHFT 11
143 #define GRU_CB_IAA0_MASK 0x3
144 #define GRU_CB_IAA1_SHFT 13
145 #define GRU_CB_IAA1_MASK 0x3
146 #define GRU_CB_IMA_SHFT 1
147 #define GRU_CB_IMA_MASK 0x3
148 #define GRU_CB_OPC_SHFT 16
149 #define GRU_CB_OPC_MASK 0xff
150 #define GRU_CB_EXOPC_SHFT 24
151 #define GRU_CB_EXOPC_MASK 0xff
153 /* GRU instruction opcodes (opc field) */
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
161 #define OP_IVSET 0x07
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
172 /* Extended opcodes values (exopc field) */
174 /* GAMIR - AMOs with implicit operands */
175 #define EOP_IR_FETCH 0x01 /* Plain fetch of memory */
176 #define EOP_IR_CLR 0x02 /* Fetch and clear */
177 #define EOP_IR_INC 0x05 /* Fetch and increment */
178 #define EOP_IR_DEC 0x07 /* Fetch and decrement */
179 #define EOP_IR_QCHK1 0x0d /* Queue check, 64 byte msg */
180 #define EOP_IR_QCHK2 0x0e /* Queue check, 128 byte msg */
182 /* GAMIRR - Registered AMOs with implicit operands */
183 #define EOP_IRR_FETCH 0x01 /* Registered fetch of memory */
184 #define EOP_IRR_CLR 0x02 /* Registered fetch and clear */
185 #define EOP_IRR_INC 0x05 /* Registered fetch and increment */
186 #define EOP_IRR_DEC 0x07 /* Registered fetch and decrement */
187 #define EOP_IRR_DECZ 0x0f /* Registered fetch and decrement, update on zero*/
189 /* GAMER - AMOs with explicit operands */
190 #define EOP_ER_SWAP 0x00 /* Exchange argument and memory */
191 #define EOP_ER_OR 0x01 /* Logical OR with memory */
192 #define EOP_ER_AND 0x02 /* Logical AND with memory */
193 #define EOP_ER_XOR 0x03 /* Logical XOR with memory */
194 #define EOP_ER_ADD 0x04 /* Add value to memory */
195 #define EOP_ER_CSWAP 0x08 /* Compare with operand2, write operand1 if match*/
196 #define EOP_ER_CADD 0x0c /* Queue check, operand1*64 byte msg */
198 /* GAMERR - Registered AMOs with explicit operands */
199 #define EOP_ERR_SWAP 0x00 /* Exchange argument and memory */
200 #define EOP_ERR_OR 0x01 /* Logical OR with memory */
201 #define EOP_ERR_AND 0x02 /* Logical AND with memory */
202 #define EOP_ERR_XOR 0x03 /* Logical XOR with memory */
203 #define EOP_ERR_ADD 0x04 /* Add value to memory */
204 #define EOP_ERR_CSWAP 0x08 /* Compare with operand2, write operand1 if match*/
205 #define EOP_ERR_EPOLL 0x09 /* Poll for equality */
206 #define EOP_ERR_NPOLL 0x0a /* Poll for inequality */
208 /* GAMXR - SGI Arithmetic unit */
209 #define EOP_XR_CSWAP 0x0b /* Masked compare exchange */
212 /* Transfer types (xtype field) */
213 #define XTYPE_B 0x0 /* byte */
214 #define XTYPE_S 0x1 /* short (2-byte) */
215 #define XTYPE_W 0x2 /* word (4-byte) */
216 #define XTYPE_DW 0x3 /* doubleword (8-byte) */
217 #define XTYPE_CL 0x6 /* cacheline (64-byte) */
220 /* Instruction access attributes (iaa0, iaa1 fields) */
221 #define IAA_RAM 0x0 /* normal cached RAM access */
222 #define IAA_NCRAM 0x2 /* noncoherent RAM access */
223 #define IAA_MMIO 0x1 /* noncoherent memory-mapped I/O space */
224 #define IAA_REGISTER 0x3 /* memory-mapped registers, etc. */
227 /* Instruction mode attributes (ima field) */
228 #define IMA_MAPPED 0x0 /* Virtual mode */
229 #define IMA_CB_DELAY 0x1 /* hold read responses until status changes */
230 #define IMA_UNMAPPED 0x2 /* bypass the TLBs (OS only) */
231 #define IMA_INTERRUPT 0x4 /* Interrupt when instruction completes */
233 /* CBE ecause bits */
234 #define CBE_CAUSE_RI (1 << 0)
235 #define CBE_CAUSE_INVALID_INSTRUCTION (1 << 1)
236 #define CBE_CAUSE_UNMAPPED_MODE_FORBIDDEN (1 << 2)
237 #define CBE_CAUSE_PE_CHECK_DATA_ERROR (1 << 3)
238 #define CBE_CAUSE_IAA_GAA_MISMATCH (1 << 4)
239 #define CBE_CAUSE_DATA_SEGMENT_LIMIT_EXCEPTION (1 << 5)
240 #define CBE_CAUSE_OS_FATAL_TLB_FAULT (1 << 6)
241 #define CBE_CAUSE_EXECUTION_HW_ERROR (1 << 7)
242 #define CBE_CAUSE_TLBHW_ERROR (1 << 8)
243 #define CBE_CAUSE_RA_REQUEST_TIMEOUT (1 << 9)
244 #define CBE_CAUSE_HA_REQUEST_TIMEOUT (1 << 10)
245 #define CBE_CAUSE_RA_RESPONSE_FATAL (1 << 11)
246 #define CBE_CAUSE_RA_RESPONSE_NON_FATAL (1 << 12)
247 #define CBE_CAUSE_HA_RESPONSE_FATAL (1 << 13)
248 #define CBE_CAUSE_HA_RESPONSE_NON_FATAL (1 << 14)
249 #define CBE_CAUSE_ADDRESS_SPACE_DECODE_ERROR (1 << 15)
250 #define CBE_CAUSE_RESPONSE_DATA_ERROR (1 << 16)
251 #define CBE_CAUSE_PROTOCOL_STATE_DATA_ERROR (1 << 17)
254 * Exceptions are retried for the following cases. If any OTHER bits are set
255 * in ecause, the exception is not retryable.
257 #define EXCEPTION_RETRY_BITS (CBE_CAUSE_RESPONSE_DATA_ERROR | \
258 CBE_CAUSE_RA_REQUEST_TIMEOUT | \
259 CBE_CAUSE_TLBHW_ERROR | \
260 CBE_CAUSE_HA_REQUEST_TIMEOUT)
262 /* Message queue head structure */
272 /* Generate the low word of a GRU instruction */
273 static inline unsigned int
274 __opword(unsigned char opcode
, unsigned char exopc
, unsigned char xtype
,
275 unsigned char iaa0
, unsigned char iaa1
,
278 return (1 << GRU_CB_ICMD_SHFT
) |
279 (iaa0
<< GRU_CB_IAA0_SHFT
) |
280 (iaa1
<< GRU_CB_IAA1_SHFT
) |
281 (ima
<< GRU_CB_IMA_SHFT
) |
282 (xtype
<< GRU_CB_XTYPE_SHFT
) |
283 (opcode
<< GRU_CB_OPC_SHFT
) |
284 (exopc
<< GRU_CB_EXOPC_SHFT
);
288 * Prefetch a cacheline. Fetch is unconditional. Must page fault if
289 * no valid TLB entry is found.
290 * ??? should I use actual "load" or hardware prefetch???
292 static inline void gru_prefetch(void *p
)
298 * Architecture specific intrinsics
300 static inline void gru_flush_cache(void *p
)
306 * Store the lower 32 bits of the command including the "start" bit. Then
307 * start the instruction executing.
309 static inline void gru_start_instruction(struct gru_instruction
*ins
, int op32
)
311 gru_ordered_store_int(ins
, op32
);
315 /* Convert "hints" to IMA */
316 #define CB_IMA(h) ((h) | IMA_UNMAPPED)
318 /* Convert data segment cache line index into TRI0 / TRI1 value */
319 #define GRU_DINDEX(i) ((i) * GRU_CACHE_LINE_BYTES)
321 /* Inline functions for GRU instructions.
323 * - nelem and stride are in elements
324 * - tri0/tri1 is in bytes for the beginning of the data segment.
326 static inline void gru_vload(void *cb
, unsigned long mem_addr
,
327 unsigned int tri0
, unsigned char xtype
, unsigned long nelem
,
328 unsigned long stride
, unsigned long hints
)
330 struct gru_instruction
*ins
= (struct gru_instruction
*)cb
;
332 ins
->baddr0
= (long)mem_addr
;
335 ins
->op1_stride
= stride
;
336 gru_start_instruction(ins
, __opword(OP_VLOAD
, 0, xtype
, IAA_RAM
, 0,
340 static inline void gru_vstore(void *cb
, unsigned long mem_addr
,
341 unsigned int tri0
, unsigned char xtype
, unsigned long nelem
,
342 unsigned long stride
, unsigned long hints
)
344 struct gru_instruction
*ins
= (void *)cb
;
346 ins
->baddr0
= (long)mem_addr
;
349 ins
->op1_stride
= stride
;
350 gru_start_instruction(ins
, __opword(OP_VSTORE
, 0, xtype
, IAA_RAM
, 0,
354 static inline void gru_ivload(void *cb
, unsigned long mem_addr
,
355 unsigned int tri0
, unsigned int tri1
, unsigned char xtype
,
356 unsigned long nelem
, unsigned long hints
)
358 struct gru_instruction
*ins
= (void *)cb
;
360 ins
->baddr0
= (long)mem_addr
;
363 ins
->tri1_bufsize
= tri1
;
364 gru_start_instruction(ins
, __opword(OP_IVLOAD
, 0, xtype
, IAA_RAM
, 0,
368 static inline void gru_ivstore(void *cb
, unsigned long mem_addr
,
369 unsigned int tri0
, unsigned int tri1
,
370 unsigned char xtype
, unsigned long nelem
, unsigned long hints
)
372 struct gru_instruction
*ins
= (void *)cb
;
374 ins
->baddr0
= (long)mem_addr
;
377 ins
->tri1_bufsize
= tri1
;
378 gru_start_instruction(ins
, __opword(OP_IVSTORE
, 0, xtype
, IAA_RAM
, 0,
382 static inline void gru_vset(void *cb
, unsigned long mem_addr
,
383 unsigned long value
, unsigned char xtype
, unsigned long nelem
,
384 unsigned long stride
, unsigned long hints
)
386 struct gru_instruction
*ins
= (void *)cb
;
388 ins
->baddr0
= (long)mem_addr
;
389 ins
->op2_value_baddr1
= value
;
391 ins
->op1_stride
= stride
;
392 gru_start_instruction(ins
, __opword(OP_VSET
, 0, xtype
, IAA_RAM
, 0,
396 static inline void gru_ivset(void *cb
, unsigned long mem_addr
,
397 unsigned int tri1
, unsigned long value
, unsigned char xtype
,
398 unsigned long nelem
, unsigned long hints
)
400 struct gru_instruction
*ins
= (void *)cb
;
402 ins
->baddr0
= (long)mem_addr
;
403 ins
->op2_value_baddr1
= value
;
405 ins
->tri1_bufsize
= tri1
;
406 gru_start_instruction(ins
, __opword(OP_IVSET
, 0, xtype
, IAA_RAM
, 0,
410 static inline void gru_vflush(void *cb
, unsigned long mem_addr
,
411 unsigned long nelem
, unsigned char xtype
, unsigned long stride
,
414 struct gru_instruction
*ins
= (void *)cb
;
416 ins
->baddr0
= (long)mem_addr
;
417 ins
->op1_stride
= stride
;
419 gru_start_instruction(ins
, __opword(OP_VFLUSH
, 0, xtype
, IAA_RAM
, 0,
423 static inline void gru_nop(void *cb
, int hints
)
425 struct gru_instruction
*ins
= (void *)cb
;
427 gru_start_instruction(ins
, __opword(OP_NOP
, 0, 0, 0, 0, CB_IMA(hints
)));
431 static inline void gru_bcopy(void *cb
, const unsigned long src
,
433 unsigned int tri0
, unsigned int xtype
, unsigned long nelem
,
434 unsigned int bufsize
, unsigned long hints
)
436 struct gru_instruction
*ins
= (void *)cb
;
438 ins
->baddr0
= (long)src
;
439 ins
->op2_value_baddr1
= (long)dest
;
442 ins
->tri1_bufsize
= bufsize
;
443 gru_start_instruction(ins
, __opword(OP_BCOPY
, 0, xtype
, IAA_RAM
,
444 IAA_RAM
, CB_IMA(hints
)));
447 static inline void gru_bstore(void *cb
, const unsigned long src
,
448 unsigned long dest
, unsigned int tri0
, unsigned int xtype
,
449 unsigned long nelem
, unsigned long hints
)
451 struct gru_instruction
*ins
= (void *)cb
;
453 ins
->baddr0
= (long)src
;
454 ins
->op2_value_baddr1
= (long)dest
;
457 gru_start_instruction(ins
, __opword(OP_BSTORE
, 0, xtype
, 0, IAA_RAM
,
461 static inline void gru_gamir(void *cb
, int exopc
, unsigned long src
,
462 unsigned int xtype
, unsigned long hints
)
464 struct gru_instruction
*ins
= (void *)cb
;
466 ins
->baddr0
= (long)src
;
467 gru_start_instruction(ins
, __opword(OP_GAMIR
, exopc
, xtype
, IAA_RAM
, 0,
471 static inline void gru_gamirr(void *cb
, int exopc
, unsigned long src
,
472 unsigned int xtype
, unsigned long hints
)
474 struct gru_instruction
*ins
= (void *)cb
;
476 ins
->baddr0
= (long)src
;
477 gru_start_instruction(ins
, __opword(OP_GAMIRR
, exopc
, xtype
, IAA_RAM
, 0,
481 static inline void gru_gamer(void *cb
, int exopc
, unsigned long src
,
483 unsigned long operand1
, unsigned long operand2
,
486 struct gru_instruction
*ins
= (void *)cb
;
488 ins
->baddr0
= (long)src
;
489 ins
->op1_stride
= operand1
;
490 ins
->op2_value_baddr1
= operand2
;
491 gru_start_instruction(ins
, __opword(OP_GAMER
, exopc
, xtype
, IAA_RAM
, 0,
495 static inline void gru_gamerr(void *cb
, int exopc
, unsigned long src
,
496 unsigned int xtype
, unsigned long operand1
,
497 unsigned long operand2
, unsigned long hints
)
499 struct gru_instruction
*ins
= (void *)cb
;
501 ins
->baddr0
= (long)src
;
502 ins
->op1_stride
= operand1
;
503 ins
->op2_value_baddr1
= operand2
;
504 gru_start_instruction(ins
, __opword(OP_GAMERR
, exopc
, xtype
, IAA_RAM
, 0,
508 static inline void gru_gamxr(void *cb
, unsigned long src
,
509 unsigned int tri0
, unsigned long hints
)
511 struct gru_instruction
*ins
= (void *)cb
;
513 ins
->baddr0
= (long)src
;
515 gru_start_instruction(ins
, __opword(OP_GAMXR
, EOP_XR_CSWAP
, XTYPE_DW
,
516 IAA_RAM
, 0, CB_IMA(hints
)));
519 static inline void gru_mesq(void *cb
, unsigned long queue
,
520 unsigned long tri0
, unsigned long nelem
,
523 struct gru_instruction
*ins
= (void *)cb
;
525 ins
->baddr0
= (long)queue
;
528 gru_start_instruction(ins
, __opword(OP_MESQ
, 0, XTYPE_CL
, IAA_RAM
, 0,
532 static inline unsigned long gru_get_amo_value(void *cb
)
534 struct gru_instruction
*ins
= (void *)cb
;
539 static inline int gru_get_amo_value_head(void *cb
)
541 struct gru_instruction
*ins
= (void *)cb
;
543 return ins
->avalue
& 0xffffffff;
546 static inline int gru_get_amo_value_limit(void *cb
)
548 struct gru_instruction
*ins
= (void *)cb
;
550 return ins
->avalue
>> 32;
553 static inline union gru_mesqhead
gru_mesq_head(int head
, int limit
)
555 union gru_mesqhead mqh
;
563 * Get struct control_block_extended_exc_detail for CB.
565 extern int gru_get_cb_exception_detail(void *cb
,
566 struct control_block_extended_exc_detail
*excdet
);
568 #define GRU_EXC_STR_SIZE 256
570 extern int gru_check_status_proc(void *cb
);
571 extern int gru_wait_proc(void *cb
);
572 extern void gru_wait_abort_proc(void *cb
);
575 * Control block definition for checking status
577 struct gru_control_block_status
{
578 unsigned int icmd
:1;
579 unsigned int unused1
:31;
580 unsigned int unused2
:24;
581 unsigned int istatus
:2;
582 unsigned int isubstatus
:4;
583 unsigned int inused3
:2;
587 static inline int gru_get_cb_status(void *cb
)
589 struct gru_control_block_status
*cbs
= (void *)cb
;
594 /* Get CB message queue substatus */
595 static inline int gru_get_cb_message_queue_substatus(void *cb
)
597 struct gru_control_block_status
*cbs
= (void *)cb
;
599 return cbs
->isubstatus
& CBSS_MSG_QUEUE_MASK
;
602 /* Get CB substatus */
603 static inline int gru_get_cb_substatus(void *cb
)
605 struct gru_control_block_status
*cbs
= (void *)cb
;
607 return cbs
->isubstatus
;
610 /* Check the status of a CB. If the CB is in UPM mode, call the
611 * OS to handle the UPM status.
612 * Returns the CB status field value (0 for normal completion)
614 static inline int gru_check_status(void *cb
)
616 struct gru_control_block_status
*cbs
= (void *)cb
;
617 int ret
= cbs
->istatus
;
619 if (ret
== CBS_CALL_OS
)
620 ret
= gru_check_status_proc(cb
);
624 /* Wait for CB to complete.
625 * Returns the CB status field value (0 for normal completion)
627 static inline int gru_wait(void *cb
)
629 struct gru_control_block_status
*cbs
= (void *)cb
;
630 int ret
= cbs
->istatus
;;
633 ret
= gru_wait_proc(cb
);
637 /* Wait for CB to complete. Aborts program if error. (Note: error does NOT
638 * mean TLB mis - only fatal errors such as memory parity error or user
639 * bugs will cause termination.
641 static inline void gru_wait_abort(void *cb
)
643 struct gru_control_block_status
*cbs
= (void *)cb
;
645 if (cbs
->istatus
!= CBS_IDLE
)
646 gru_wait_abort_proc(cb
);
651 * Get a pointer to a control block
652 * gseg - GSeg address returned from gru_get_thread_gru_segment()
653 * index - index of desired CB
655 static inline void *gru_get_cb_pointer(void *gseg
,
658 return gseg
+ GRU_CB_BASE
+ index
* GRU_HANDLE_STRIDE
;
662 * Get a pointer to a cacheline in the data segment portion of a GSeg
663 * gseg - GSeg address returned from gru_get_thread_gru_segment()
664 * index - index of desired cache line
666 static inline void *gru_get_data_pointer(void *gseg
, int index
)
668 return gseg
+ GRU_DS_BASE
+ index
* GRU_CACHE_LINE_BYTES
;
672 * Convert a vaddr into the tri index within the GSEG
673 * vaddr - virtual address of within gseg
675 static inline int gru_get_tri(void *vaddr
)
677 return ((unsigned long)vaddr
& (GRU_GSEG_PAGESIZE
- 1)) - GRU_DS_BASE
;
679 #endif /* __GRU_INSTRUCTIONS_H__ */