1 // Copyright 2015 The Go Authors. All rights reserved.
2 // Use of this source code is governed by a BSD-style
3 // license that can be found in the LICENSE file.
11 TEXT runtime·rt0_go(SB),NOSPLIT|TOPFRAME,$0
12 // SP = stack; R0 = argc; R1 = argv
15 MOVW R0, 8(RSP) // argc
16 MOVD R1, 16(RSP) // argv
20 MOVD ZR, g // clear g, make sure it's not junk.
24 MOVD R0, 16(RSP) // arg2: TLS base
25 MOVD $runtime·tls_g(SB), R2
26 MOVD R2, 8(RSP) // arg1: &tlsg
31 // create istack out of the given (operating system) stack.
32 // _cgo_init may update stackguard.
33 MOVD $runtime·g0(SB), g
35 MOVD $(-64*1024)(R7), R0
36 MOVD R0, g_stackguard0(g)
37 MOVD R0, g_stackguard1(g)
38 MOVD R0, (g_stack+stack_lo)(g)
39 MOVD R7, (g_stack+stack_hi)(g)
41 // if there is a _cgo_init, call it using the gcc ABI.
42 MOVD _cgo_init(SB), R12
46 MRS_TPIDR_R0 // load TLS base pointer
47 MOVD R0, R3 // arg 3: TLS base pointer
48 MOVD $runtime·tls_g(SB), R2 // arg 2: &tls_g
50 MOVD $0, R2 // arg 2: not used when using platform's TLS
52 MOVD $setg_gcc<>(SB), R1 // arg 1: setg
53 MOVD g, R0 // arg 0: G
54 SUB $16, RSP // reserve 16 bytes for sp-8 where fp may be saved.
60 // update stackguard after _cgo_init
61 MOVD (g_stack+stack_lo)(g), R0
62 ADD $const__StackGuard, R0
63 MOVD R0, g_stackguard0(g)
64 MOVD R0, g_stackguard1(g)
66 // set the per-goroutine and per-mach "registers"
67 MOVD $runtime·m0(SB), R0
80 MOVW 8(RSP), R0 // copy argc
82 MOVD 16(RSP), R0 // copy argv
86 BL runtime·schedinit(SB)
88 // create a new goroutine to start program
89 MOVD $runtime·mainPC(SB), R0 // entry
91 MOVD R0, 8(RSP) // arg
92 MOVD $0, 0(RSP) // dummy LR
93 BL runtime·newproc(SB)
99 // Prevent dead-code elimination of debugCallV2, which is
100 // intended to be called by debuggers.
101 MOVD $runtime·debugCallV2<ABIInternal>(SB), R0
104 MOVD R0, (R0) // boom
107 DATA runtime·mainPC+0(SB)/8,$runtime·main<ABIInternal>(SB)
108 GLOBL runtime·mainPC(SB),RODATA,$8
110 // Windows ARM64 needs an immediate 0xf000 argument.
111 // See go.dev/issues/53837.
113 #ifdef GOOS_windows \
120 TEXT runtime·breakpoint(SB),NOSPLIT|NOFRAME,$0-0
124 TEXT runtime·asminit(SB),NOSPLIT|NOFRAME,$0-0
127 TEXT runtime·mstart(SB),NOSPLIT|TOPFRAME,$0
128 BL runtime·mstart0(SB)
136 // restore state from Gobuf; longjmp
137 TEXT runtime·gogo(SB), NOSPLIT|NOFRAME, $0-8
140 MOVD 0(R6), R4 // make sure g != nil
143 TEXT gogo<>(SB), NOSPLIT|NOFRAME, $0
145 BL runtime·save_g(SB)
147 MOVD gobuf_sp(R5), R0
149 MOVD gobuf_bp(R5), R29
150 MOVD gobuf_lr(R5), LR
151 MOVD gobuf_ret(R5), R0
152 MOVD gobuf_ctxt(R5), R26
153 MOVD $0, gobuf_sp(R5)
154 MOVD $0, gobuf_bp(R5)
155 MOVD $0, gobuf_ret(R5)
156 MOVD $0, gobuf_lr(R5)
157 MOVD $0, gobuf_ctxt(R5)
158 CMP ZR, ZR // set condition codes for == test, needed by stack split
159 MOVD gobuf_pc(R5), R6
162 // void mcall(fn func(*g))
163 // Switch to m->g0's stack, call fn(g).
164 // Fn must never return. It should gogo(&g->sched)
165 // to keep running g.
166 TEXT runtime·mcall<ABIInternal>(SB), NOSPLIT|NOFRAME, $0-8
167 MOVD R0, R26 // context
169 // Save caller state in g->sched
171 MOVD R0, (g_sched+gobuf_sp)(g)
172 MOVD R29, (g_sched+gobuf_bp)(g)
173 MOVD LR, (g_sched+gobuf_pc)(g)
174 MOVD $0, (g_sched+gobuf_lr)(g)
176 // Switch to m->g0 & its stack, call fn.
180 BL runtime·save_g(SB)
183 B runtime·badmcall(SB)
185 MOVD (g_sched+gobuf_sp)(g), R0
186 MOVD R0, RSP // sp = m->g0->sched.sp
187 MOVD (g_sched+gobuf_bp)(g), R29
188 MOVD R3, R0 // arg = g
189 MOVD $0, -16(RSP) // dummy LR
191 MOVD 0(R26), R4 // code pointer
193 B runtime·badmcall2(SB)
195 // systemstack_switch is a dummy routine that systemstack leaves at the bottom
196 // of the G stack. We need to distinguish the routine that
197 // lives at the bottom of the G stack from the one that lives
198 // at the top of the system stack because the one at the top of
199 // the system stack terminates the stack walk (see topofstack()).
200 TEXT runtime·systemstack_switch(SB), NOSPLIT, $0-0
202 BL (LR) // make sure this function is not leaf
205 // func systemstack(fn func())
206 TEXT runtime·systemstack(SB), NOSPLIT, $0-8
207 MOVD fn+0(FP), R3 // R3 = fn
208 MOVD R3, R26 // context
209 MOVD g_m(g), R4 // R4 = m
211 MOVD m_gsignal(R4), R5 // R5 = gsignal
215 MOVD m_g0(R4), R5 // R5 = g0
223 // Bad: g is not gsignal, not g0, not curg. What is it?
224 // Hide call from linker nosplit analysis.
225 MOVD $runtime·badsystemstack(SB), R3
230 // save our state in g->sched. Pretend to
231 // be systemstack_switch if the G stack is scanned.
232 BL gosave_systemstack_switch<>(SB)
236 BL runtime·save_g(SB)
237 MOVD (g_sched+gobuf_sp)(g), R3
239 MOVD (g_sched+gobuf_bp)(g), R29
241 // call target function
242 MOVD 0(R26), R3 // code pointer
248 BL runtime·save_g(SB)
249 MOVD (g_sched+gobuf_sp)(g), R0
251 MOVD (g_sched+gobuf_bp)(g), R29
252 MOVD $0, (g_sched+gobuf_sp)(g)
253 MOVD $0, (g_sched+gobuf_bp)(g)
257 // already on m stack, just call directly
258 // Using a tail call here cleans up tracebacks since we won't stop
259 // at an intermediate systemstack.
260 MOVD 0(R26), R3 // code pointer
261 MOVD.P 16(RSP), R30 // restore LR
262 SUB $8, RSP, R29 // restore FP
266 * support for morestack
269 // Called during function prolog when more stack is needed.
270 // Caller has already loaded:
271 // R3 prolog's LR (R30)
273 // The traceback routines see morestack on a g0 as being
274 // the top of a stack (for example, morestack calling newstack
275 // calling the scheduler calling newm calling gc), so we must
276 // record an argument size. For that purpose, it has no arguments.
277 TEXT runtime·morestack(SB),NOSPLIT|NOFRAME,$0-0
278 // Cannot grow scheduler stack (m->g0).
283 BL runtime·badmorestackg0(SB)
286 // Cannot grow signal stack (m->gsignal).
287 MOVD m_gsignal(R8), R4
290 BL runtime·badmorestackgsignal(SB)
294 // Set g->sched to context in f
296 MOVD R0, (g_sched+gobuf_sp)(g)
297 MOVD R29, (g_sched+gobuf_bp)(g)
298 MOVD LR, (g_sched+gobuf_pc)(g)
299 MOVD R3, (g_sched+gobuf_lr)(g)
300 MOVD R26, (g_sched+gobuf_ctxt)(g)
303 // Set m->morebuf to f's callers.
304 MOVD R3, (m_morebuf+gobuf_pc)(R8) // f's caller's PC
306 MOVD R0, (m_morebuf+gobuf_sp)(R8) // f's caller's RSP
307 MOVD g, (m_morebuf+gobuf_g)(R8)
309 // Call newstack on m->g0's stack.
311 BL runtime·save_g(SB)
312 MOVD (g_sched+gobuf_sp)(g), R0
314 MOVD (g_sched+gobuf_bp)(g), R29
315 MOVD.W $0, -16(RSP) // create a call frame on g0 (saved LR; keep 16-aligned)
316 BL runtime·newstack(SB)
318 // Not reached, but make sure the return PC from the call to newstack
319 // is still in this function, and not the beginning of the next.
322 TEXT runtime·morestack_noctxt(SB),NOSPLIT|NOFRAME,$0-0
323 // Force SPWRITE. This function doesn't actually write SP,
324 // but it is called with a special calling convention where
325 // the caller doesn't save LR on stack but passes it as a
326 // register (R3), and the unwinder currently doesn't understand.
327 // Make it SPWRITE to stop unwinding. (See issue 54332)
331 B runtime·morestack(SB)
333 // spillArgs stores return values from registers to a *internal/abi.RegArgs in R20.
334 TEXT ·spillArgs(SB),NOSPLIT,$0-0
335 STP (R0, R1), (0*8)(R20)
336 STP (R2, R3), (2*8)(R20)
337 STP (R4, R5), (4*8)(R20)
338 STP (R6, R7), (6*8)(R20)
339 STP (R8, R9), (8*8)(R20)
340 STP (R10, R11), (10*8)(R20)
341 STP (R12, R13), (12*8)(R20)
342 STP (R14, R15), (14*8)(R20)
343 FSTPD (F0, F1), (16*8)(R20)
344 FSTPD (F2, F3), (18*8)(R20)
345 FSTPD (F4, F5), (20*8)(R20)
346 FSTPD (F6, F7), (22*8)(R20)
347 FSTPD (F8, F9), (24*8)(R20)
348 FSTPD (F10, F11), (26*8)(R20)
349 FSTPD (F12, F13), (28*8)(R20)
350 FSTPD (F14, F15), (30*8)(R20)
353 // unspillArgs loads args into registers from a *internal/abi.RegArgs in R20.
354 TEXT ·unspillArgs(SB),NOSPLIT,$0-0
355 LDP (0*8)(R20), (R0, R1)
356 LDP (2*8)(R20), (R2, R3)
357 LDP (4*8)(R20), (R4, R5)
358 LDP (6*8)(R20), (R6, R7)
359 LDP (8*8)(R20), (R8, R9)
360 LDP (10*8)(R20), (R10, R11)
361 LDP (12*8)(R20), (R12, R13)
362 LDP (14*8)(R20), (R14, R15)
363 FLDPD (16*8)(R20), (F0, F1)
364 FLDPD (18*8)(R20), (F2, F3)
365 FLDPD (20*8)(R20), (F4, F5)
366 FLDPD (22*8)(R20), (F6, F7)
367 FLDPD (24*8)(R20), (F8, F9)
368 FLDPD (26*8)(R20), (F10, F11)
369 FLDPD (28*8)(R20), (F12, F13)
370 FLDPD (30*8)(R20), (F14, F15)
373 // reflectcall: call a function with the given argument list
374 // func call(stackArgsType *_type, f *FuncVal, stackArgs *byte, stackArgsSize, stackRetOffset, frameSize uint32, regArgs *abi.RegArgs).
375 // we don't have variable-sized frames, so we use a small number
376 // of constant-sized-frame functions to encode a few bits of size in the pc.
377 // Caution: ugly multiline assembly macros in your future!
379 #define DISPATCH(NAME,MAXSIZE) \
380 MOVD $MAXSIZE, R27; \
383 MOVD $NAME(SB), R27; \
385 // Note: can't just "B NAME(SB)" - bad inlining results.
387 TEXT ·reflectcall(SB), NOSPLIT|NOFRAME, $0-48
388 MOVWU frameSize+32(FP), R16
389 DISPATCH(runtime·call16, 16)
390 DISPATCH(runtime·call32, 32)
391 DISPATCH(runtime·call64, 64)
392 DISPATCH(runtime·call128, 128)
393 DISPATCH(runtime·call256, 256)
394 DISPATCH(runtime·call512, 512)
395 DISPATCH(runtime·call1024, 1024)
396 DISPATCH(runtime·call2048, 2048)
397 DISPATCH(runtime·call4096, 4096)
398 DISPATCH(runtime·call8192, 8192)
399 DISPATCH(runtime·call16384, 16384)
400 DISPATCH(runtime·call32768, 32768)
401 DISPATCH(runtime·call65536, 65536)
402 DISPATCH(runtime·call131072, 131072)
403 DISPATCH(runtime·call262144, 262144)
404 DISPATCH(runtime·call524288, 524288)
405 DISPATCH(runtime·call1048576, 1048576)
406 DISPATCH(runtime·call2097152, 2097152)
407 DISPATCH(runtime·call4194304, 4194304)
408 DISPATCH(runtime·call8388608, 8388608)
409 DISPATCH(runtime·call16777216, 16777216)
410 DISPATCH(runtime·call33554432, 33554432)
411 DISPATCH(runtime·call67108864, 67108864)
412 DISPATCH(runtime·call134217728, 134217728)
413 DISPATCH(runtime·call268435456, 268435456)
414 DISPATCH(runtime·call536870912, 536870912)
415 DISPATCH(runtime·call1073741824, 1073741824)
416 MOVD $runtime·badreflectcall(SB), R0
419 #define CALLFN(NAME,MAXSIZE) \
420 TEXT NAME(SB), WRAPPER, $MAXSIZE-48; \
422 /* copy arguments to stack */ \
423 MOVD stackArgs+16(FP), R3; \
424 MOVWU stackArgsSize+24(FP), R4; \
428 /* if R6=(argsize&~15) != 0 */ \
430 /* copy 16 bytes a time */ \
431 LDP.P 16(R3), (R7, R8); \
432 STP.P (R7, R8), 16(R5); \
437 /* if R6=(argsize&15) != 0 */ \
439 /* copy 1 byte a time for the rest */ \
444 /* set up argument registers */ \
445 MOVD regArgs+40(FP), R20; \
446 CALL ·unspillArgs(SB); \
447 /* call function */ \
450 PCDATA $PCDATA_StackMapIndex, $0; \
452 /* copy return values back */ \
453 MOVD regArgs+40(FP), R20; \
454 CALL ·spillArgs(SB); \
455 MOVD stackArgsType+0(FP), R7; \
456 MOVD stackArgs+16(FP), R3; \
457 MOVWU stackArgsSize+24(FP), R4; \
458 MOVWU stackRetOffset+28(FP), R6; \
466 // callRet copies return values back at the end of call*. This is a
467 // separate function so it can allocate stack space for the arguments
468 // to reflectcallmove. It does not follow the Go ABI; it expects its
469 // arguments in registers.
470 TEXT callRet<>(SB), NOSPLIT, $48-0
473 STP (R5, R4), 24(RSP)
475 BL runtime·reflectcallmove(SB)
481 CALLFN(·call128, 128)
482 CALLFN(·call256, 256)
483 CALLFN(·call512, 512)
484 CALLFN(·call1024, 1024)
485 CALLFN(·call2048, 2048)
486 CALLFN(·call4096, 4096)
487 CALLFN(·call8192, 8192)
488 CALLFN(·call16384, 16384)
489 CALLFN(·call32768, 32768)
490 CALLFN(·call65536, 65536)
491 CALLFN(·call131072, 131072)
492 CALLFN(·call262144, 262144)
493 CALLFN(·call524288, 524288)
494 CALLFN(·call1048576, 1048576)
495 CALLFN(·call2097152, 2097152)
496 CALLFN(·call4194304, 4194304)
497 CALLFN(·call8388608, 8388608)
498 CALLFN(·call16777216, 16777216)
499 CALLFN(·call33554432, 33554432)
500 CALLFN(·call67108864, 67108864)
501 CALLFN(·call134217728, 134217728)
502 CALLFN(·call268435456, 268435456)
503 CALLFN(·call536870912, 536870912)
504 CALLFN(·call1073741824, 1073741824)
506 // func memhash32(p unsafe.Pointer, h uintptr) uintptr
507 TEXT runtime·memhash32<ABIInternal>(SB),NOSPLIT|NOFRAME,$0-24
508 MOVB runtime·useAeshash(SB), R10
510 MOVD $runtime·aeskeysched+0(SB), R3
512 VEOR V0.B16, V0.B16, V0.B16
526 B runtime·memhash32Fallback<ABIInternal>(SB)
528 // func memhash64(p unsafe.Pointer, h uintptr) uintptr
529 TEXT runtime·memhash64<ABIInternal>(SB),NOSPLIT|NOFRAME,$0-24
530 MOVB runtime·useAeshash(SB), R10
532 MOVD $runtime·aeskeysched+0(SB), R3
534 VEOR V0.B16, V0.B16, V0.B16
548 B runtime·memhash64Fallback<ABIInternal>(SB)
550 // func memhash(p unsafe.Pointer, h, size uintptr) uintptr
551 TEXT runtime·memhash<ABIInternal>(SB),NOSPLIT|NOFRAME,$0-32
552 MOVB runtime·useAeshash(SB), R10
556 B runtime·memhashFallback<ABIInternal>(SB)
558 // func strhash(p unsafe.Pointer, h uintptr) uintptr
559 TEXT runtime·strhash<ABIInternal>(SB),NOSPLIT|NOFRAME,$0-24
560 MOVB runtime·useAeshash(SB), R10
562 LDP (R0), (R0, R2) // string data / length
565 B runtime·strhashFallback<ABIInternal>(SB)
570 // At return, R0 = return value
571 TEXT aeshashbody<>(SB),NOSPLIT|NOFRAME,$0
572 VEOR V30.B16, V30.B16, V30.B16
574 VMOV R2, V30.D[1] // load length into seed
576 MOVD $runtime·aeskeysched+0(SB), R4
577 VLD1.P 16(R4), [V0.B16]
593 VEOR V2.B16, V2.B16, V2.B16
594 TBZ $3, R2, less_than_8
595 VLD1.P 8(R0), V2.D[0]
598 TBZ $2, R2, less_than_4
599 VLD1.P 4(R0), V2.S[2]
602 TBZ $1, R2, less_than_2
603 VLD1.P 2(R0), V2.H[6]
632 VLD1.P (R0)(R10), [V2.B16]
648 VEOR V3.B16, V2.B16, V2.B16
654 VLD1 (R4), [V1.B16, V2.B16, V3.B16]
663 VLD1.P (R0)(R10), [V4.B16, V5.B16]
664 VLD1 (R0), [V6.B16, V7.B16]
689 VEOR V6.B16, V4.B16, V4.B16
690 VEOR V7.B16, V5.B16, V5.B16
691 VEOR V5.B16, V4.B16, V4.B16
697 VLD1.P 64(R4), [V1.B16, V2.B16, V3.B16, V4.B16]
698 VLD1 (R4), [V5.B16, V6.B16, V7.B16]
715 VLD1.P (R0)(R10), [V8.B16, V9.B16, V10.B16, V11.B16]
716 VLD1 (R0), [V12.B16, V13.B16, V14.B16, V15.B16]
722 AESMC V10.B16, V10.B16
724 AESMC V11.B16, V11.B16
726 AESMC V12.B16, V12.B16
728 AESMC V13.B16, V13.B16
730 AESMC V14.B16, V14.B16
732 AESMC V15.B16, V15.B16
739 AESMC V10.B16, V10.B16
741 AESMC V11.B16, V11.B16
743 AESMC V12.B16, V12.B16
745 AESMC V13.B16, V13.B16
747 AESMC V14.B16, V14.B16
749 AESMC V15.B16, V15.B16
760 VEOR V12.B16, V8.B16, V8.B16
761 VEOR V13.B16, V9.B16, V9.B16
762 VEOR V14.B16, V10.B16, V10.B16
763 VEOR V15.B16, V11.B16, V11.B16
764 VEOR V10.B16, V8.B16, V8.B16
765 VEOR V11.B16, V9.B16, V9.B16
766 VEOR V9.B16, V8.B16, V8.B16
773 VLD1.P 64(R4), [V1.B16, V2.B16, V3.B16, V4.B16]
774 VLD1 (R4), [V5.B16, V6.B16, V7.B16]
791 VLD1.P 64(R10), [V8.B16, V9.B16, V10.B16, V11.B16]
792 VLD1 (R10), [V12.B16, V13.B16, V14.B16, V15.B16]
814 VLD1.P 64(R0), [V8.B16, V9.B16, V10.B16, V11.B16]
824 VLD1.P 64(R0), [V12.B16, V13.B16, V14.B16, V15.B16]
879 VEOR V0.B16, V1.B16, V0.B16
880 VEOR V2.B16, V3.B16, V2.B16
881 VEOR V4.B16, V5.B16, V4.B16
882 VEOR V6.B16, V7.B16, V6.B16
883 VEOR V0.B16, V2.B16, V0.B16
884 VEOR V4.B16, V6.B16, V4.B16
885 VEOR V4.B16, V0.B16, V0.B16
890 TEXT runtime·procyield(SB),NOSPLIT,$0-0
891 MOVWU cycles+0(FP), R0
898 // Save state of caller into g->sched,
899 // but using fake PC from systemstack_switch.
900 // Must only be called from functions with no locals ($0)
901 // or else unwinding from systemstack_switch is incorrect.
903 TEXT gosave_systemstack_switch<>(SB),NOSPLIT|NOFRAME,$0
904 MOVD $runtime·systemstack_switch(SB), R0
905 ADD $8, R0 // get past prologue
906 MOVD R0, (g_sched+gobuf_pc)(g)
908 MOVD R0, (g_sched+gobuf_sp)(g)
909 MOVD R29, (g_sched+gobuf_bp)(g)
910 MOVD $0, (g_sched+gobuf_lr)(g)
911 MOVD $0, (g_sched+gobuf_ret)(g)
912 // Assert ctxt is zero. See func save.
913 MOVD (g_sched+gobuf_ctxt)(g), R0
915 CALL runtime·abort(SB)
918 // func asmcgocall_no_g(fn, arg unsafe.Pointer)
919 // Call fn(arg) aligned appropriately for the gcc ABI.
920 // Called on a system stack, and there may be no g yet (during needm).
921 TEXT ·asmcgocall_no_g(SB),NOSPLIT,$0-16
924 SUB $16, RSP // skip over saved frame pointer below RSP
926 ADD $16, RSP // skip over saved frame pointer below RSP
929 // func asmcgocall(fn, arg unsafe.Pointer) int32
930 // Call fn(arg) on the scheduler stack,
931 // aligned appropriately for the gcc ABI.
932 // See cgocall.go for more details.
933 TEXT ·asmcgocall(SB),NOSPLIT,$0-20
937 MOVD RSP, R2 // save original stack pointer
941 // Figure out if we need to switch to m->g0 stack.
942 // We get called to create new OS threads too, and those
943 // come in on the m->g0 stack already. Or we might already
944 // be on the m->gsignal stack.
946 MOVD m_gsignal(R8), R3
953 // Switch to system stack.
954 MOVD R0, R9 // gosave_systemstack_switch<> and save_g might clobber R0
955 BL gosave_systemstack_switch<>(SB)
957 BL runtime·save_g(SB)
958 MOVD (g_sched+gobuf_sp)(g), R0
960 MOVD (g_sched+gobuf_bp)(g), R29
963 // Now on a scheduling stack (a pthread-created stack).
964 // Save room for two of our pointers /*, plus 32 bytes of callee
965 // save area that lives on the caller stack. */
969 MOVD R4, 0(RSP) // save old g on stack
970 MOVD (g_stack+stack_hi)(R4), R4
972 MOVD R4, 8(RSP) // save depth in old g stack (can't just save SP, as stack might be copied during a callback)
976 // Restore g, stack pointer. R0 is errno, so don't touch it
978 BL runtime·save_g(SB)
979 MOVD (g_stack+stack_hi)(g), R5
989 // Running on a system stack, perhaps even without a g.
990 // Having no g can happen during thread creation or thread teardown
991 // (see needm/dropm on Solaris, for example).
992 // This code is like the above sequence but without saving/restoring g
993 // and without worrying about the stack moving out from under us
994 // (because we're on a system stack, not a goroutine stack).
995 // The above code could be used directly if already on a system stack,
996 // but then the only path through this code would be a rare case on Solaris.
997 // Using this code for all "already on system stack" calls exercises it more,
998 // which should help keep it correct.
1003 MOVD R4, 0(RSP) // Where above code stores g, in case someone looks during debugging.
1004 MOVD R2, 8(RSP) // Save original stack pointer.
1006 // Restore stack pointer.
1012 // cgocallback(fn, frame unsafe.Pointer, ctxt uintptr)
1013 // See cgocall.go for more details.
1014 TEXT ·cgocallback(SB),NOSPLIT,$24-24
1017 // Skip cgocallbackg, just dropm when fn is nil, and frame is the saved g.
1018 // It is used to dropm while thread is exiting.
1021 // Restore the g from frame.
1026 // Load g from thread-local storage.
1027 BL runtime·load_g(SB)
1029 // If g is nil, Go did not create the current thread,
1030 // or if this thread never called into Go on pthread platforms.
1031 // Call needm to obtain one for temporary use.
1032 // In this case, we're running on the thread stack, so there's
1033 // lots of space, but the linker doesn't know. Hide the call from
1034 // the linker analysis by using an indirect call.
1038 MOVD R8, savedm-8(SP)
1042 MOVD g, savedm-8(SP) // g is zero, so is m.
1043 MOVD $runtime·needAndBindM(SB), R0
1046 // Set m->g0->sched.sp = SP, so that if a panic happens
1047 // during the function we are about to execute, it will
1048 // have a valid SP to run on the g0 stack.
1049 // The next few lines (after the havem label)
1050 // will save this SP onto the stack and then write
1051 // the same SP back to m->sched.sp. That seems redundant,
1052 // but if an unrecovered panic happens, unwindm will
1053 // restore the g->sched.sp from the stack location
1054 // and then systemstack will try to use it. If we don't set it here,
1055 // that restored SP will be uninitialized (typically 0) and
1056 // will not be usable.
1060 MOVD R0, (g_sched+gobuf_sp)(R3)
1061 MOVD R29, (g_sched+gobuf_bp)(R3)
1064 // Now there's a valid m, and we're running on its m->g0.
1065 // Save current m->g0->sched.sp on stack and then set it to SP.
1066 // Save current sp in m->g0->sched.sp in preparation for
1067 // switch back to m->curg stack.
1068 // NOTE: unwindm knows that the saved g->sched.sp is at 16(RSP) aka savedsp-16(SP).
1069 // Beware that the frame size is actually 32+16.
1071 MOVD (g_sched+gobuf_sp)(R3), R4
1072 MOVD R4, savedsp-16(SP)
1074 MOVD R0, (g_sched+gobuf_sp)(R3)
1076 // Switch to m->curg stack and call runtime.cgocallbackg.
1077 // Because we are taking over the execution of m->curg
1078 // but *not* resuming what had been running, we need to
1079 // save that information (m->curg->sched) so we can restore it.
1080 // We can restore m->curg->sched.sp easily, because calling
1081 // runtime.cgocallbackg leaves SP unchanged upon return.
1082 // To save m->curg->sched.pc, we push it onto the curg stack and
1083 // open a frame the same size as cgocallback's g0 frame.
1084 // Once we switch to the curg stack, the pushed PC will appear
1085 // to be the return PC of cgocallback, so that the traceback
1086 // will seamlessly trace back into the earlier calls.
1088 BL runtime·save_g(SB)
1089 MOVD (g_sched+gobuf_sp)(g), R4 // prepare stack as R4
1090 MOVD (g_sched+gobuf_pc)(g), R5
1092 MOVD (g_sched+gobuf_bp)(g), R5
1094 // Gather our arguments into registers.
1096 MOVD frame+8(FP), R2
1097 MOVD ctxt+16(FP), R3
1098 MOVD $-48(R4), R0 // maintain 16-byte SP alignment
1099 MOVD R0, RSP // switch stack
1103 MOVD $runtime·cgocallbackg(SB), R0
1104 CALL (R0) // indirect call to bypass nosplit check. We're on a different stack now.
1106 // Restore g->sched (== m->curg->sched) from saved values.
1108 MOVD R5, (g_sched+gobuf_pc)(g)
1111 MOVD R4, (g_sched+gobuf_sp)(g)
1113 // Switch back to m->g0's stack and restore m->g0->sched.sp.
1114 // (Unlike m->curg, the g0 goroutine never uses sched.pc,
1115 // so we do not have to restore it.)
1118 BL runtime·save_g(SB)
1119 MOVD (g_sched+gobuf_sp)(g), R0
1121 MOVD savedsp-16(SP), R4
1122 MOVD R4, (g_sched+gobuf_sp)(g)
1124 // If the m on entry was nil, we called needm above to borrow an m,
1125 // 1. for the duration of the call on non-pthread platforms,
1126 // 2. or the duration of the C thread alive on pthread platforms.
1127 // If the m on entry wasn't nil,
1128 // 1. the thread might be a Go thread,
1129 // 2. or it's wasn't the first call from a C thread on pthread platforms,
1130 // since the we skip dropm to resue the m in the first call.
1131 MOVD savedm-8(SP), R6
1134 // Skip dropm to reuse it in the next call, when a pthread key has been created.
1135 MOVD _cgo_pthread_key_created(SB), R6
1136 // It means cgo is disabled when _cgo_pthread_key_created is a nil pointer, need dropm.
1142 MOVD $runtime·dropm(SB), R0
1149 // Called from cgo wrappers, this function returns g->m->curg.stack.hi.
1150 // Must obey the gcc calling convention.
1151 TEXT _cgo_topofstack(SB),NOSPLIT,$24
1152 // g (R28) and REGTMP (R27) might be clobbered by load_g. They
1153 // are callee-save in the gcc calling convention, so save them.
1154 MOVD R27, savedR27-8(SP)
1155 MOVD g, saveG-16(SP)
1157 BL runtime·load_g(SB)
1160 MOVD (g_stack+stack_hi)(R0), R0
1162 MOVD saveG-16(SP), g
1163 MOVD savedR28-8(SP), R27
1166 // void setg(G*); set g. for use by needm.
1167 TEXT runtime·setg(SB), NOSPLIT, $0-8
1169 // This only happens if iscgo, so jump straight to save_g
1170 BL runtime·save_g(SB)
1173 // void setg_gcc(G*); set g called from gcc
1174 TEXT setg_gcc<>(SB),NOSPLIT,$8
1176 MOVD R27, savedR27-8(SP)
1177 BL runtime·save_g(SB)
1178 MOVD savedR27-8(SP), R27
1181 TEXT runtime·emptyfunc(SB),0,$0-0
1184 TEXT runtime·abort(SB),NOSPLIT|NOFRAME,$0-0
1189 TEXT runtime·return0(SB), NOSPLIT, $0
1193 // The top-most function running on a goroutine
1194 // returns to goexit+PCQuantum.
1195 TEXT runtime·goexit(SB),NOSPLIT|NOFRAME|TOPFRAME,$0-0
1197 BL runtime·goexit1(SB) // does not return
1199 // This is called from .init_array and follows the platform, not Go, ABI.
1200 TEXT runtime·addmoduledata(SB),NOSPLIT,$0-0
1202 MOVD R27, 8(RSP) // The access to global variables below implicitly uses R27, which is callee-save
1203 MOVD runtime·lastmoduledatap(SB), R1
1204 MOVD R0, moduledata_next(R1)
1205 MOVD R0, runtime·lastmoduledatap(SB)
1210 TEXT ·checkASM(SB),NOSPLIT,$0-1
1215 // gcWriteBarrier informs the GC about heap pointer writes.
1217 // gcWriteBarrier does NOT follow the Go ABI. It accepts the
1218 // number of bytes of buffer needed in R25, and returns a pointer
1219 // to the buffer space in R25.
1220 // It clobbers condition codes.
1221 // It does not clobber any general-purpose registers except R27,
1222 // but may clobber others (e.g., floating point registers)
1223 // The act of CALLing gcWriteBarrier will clobber R30 (LR).
1224 TEXT gcWriteBarrier<>(SB),NOSPLIT,$200
1225 // Save the registers clobbered by the fast path.
1226 STP (R0, R1), 184(RSP)
1230 MOVD (p_wbBuf+wbBuf_next)(R0), R1
1231 MOVD (p_wbBuf+wbBuf_end)(R0), R27
1232 // Increment wbBuf.next position.
1234 // Is the buffer full?
1237 // Commit to the larger buffer.
1238 MOVD R1, (p_wbBuf+wbBuf_next)(R0)
1239 // Make return value (the original next position)
1241 // Restore registers.
1242 LDP 184(RSP), (R0, R1)
1246 // Save all general purpose registers since these could be
1247 // clobbered by wbBufFlush and were not saved by the caller.
1248 // R0 and R1 already saved
1249 STP (R2, R3), 1*8(RSP)
1250 STP (R4, R5), 3*8(RSP)
1251 STP (R6, R7), 5*8(RSP)
1252 STP (R8, R9), 7*8(RSP)
1253 STP (R10, R11), 9*8(RSP)
1254 STP (R12, R13), 11*8(RSP)
1255 STP (R14, R15), 13*8(RSP)
1256 // R16, R17 may be clobbered by linker trampoline
1258 STP (R19, R20), 15*8(RSP)
1259 STP (R21, R22), 17*8(RSP)
1260 STP (R23, R24), 19*8(RSP)
1261 STP (R25, R26), 21*8(RSP)
1262 // R27 is temp register.
1264 // R29 is frame pointer (unused).
1265 // R30 is LR, which was saved by the prologue.
1268 CALL runtime·wbBufFlush(SB)
1269 LDP 1*8(RSP), (R2, R3)
1270 LDP 3*8(RSP), (R4, R5)
1271 LDP 5*8(RSP), (R6, R7)
1272 LDP 7*8(RSP), (R8, R9)
1273 LDP 9*8(RSP), (R10, R11)
1274 LDP 11*8(RSP), (R12, R13)
1275 LDP 13*8(RSP), (R14, R15)
1276 LDP 15*8(RSP), (R19, R20)
1277 LDP 17*8(RSP), (R21, R22)
1278 LDP 19*8(RSP), (R23, R24)
1279 LDP 21*8(RSP), (R25, R26)
1282 TEXT runtime·gcWriteBarrier1<ABIInternal>(SB),NOSPLIT,$0
1284 JMP gcWriteBarrier<>(SB)
1285 TEXT runtime·gcWriteBarrier2<ABIInternal>(SB),NOSPLIT,$0
1287 JMP gcWriteBarrier<>(SB)
1288 TEXT runtime·gcWriteBarrier3<ABIInternal>(SB),NOSPLIT,$0
1290 JMP gcWriteBarrier<>(SB)
1291 TEXT runtime·gcWriteBarrier4<ABIInternal>(SB),NOSPLIT,$0
1293 JMP gcWriteBarrier<>(SB)
1294 TEXT runtime·gcWriteBarrier5<ABIInternal>(SB),NOSPLIT,$0
1296 JMP gcWriteBarrier<>(SB)
1297 TEXT runtime·gcWriteBarrier6<ABIInternal>(SB),NOSPLIT,$0
1299 JMP gcWriteBarrier<>(SB)
1300 TEXT runtime·gcWriteBarrier7<ABIInternal>(SB),NOSPLIT,$0
1302 JMP gcWriteBarrier<>(SB)
1303 TEXT runtime·gcWriteBarrier8<ABIInternal>(SB),NOSPLIT,$0
1305 JMP gcWriteBarrier<>(SB)
1307 DATA debugCallFrameTooLarge<>+0x00(SB)/20, $"call frame too large"
1308 GLOBL debugCallFrameTooLarge<>(SB), RODATA, $20 // Size duplicated below
1310 // debugCallV2 is the entry point for debugger-injected function
1311 // calls on running goroutines. It informs the runtime that a
1312 // debug call has been injected and creates a call frame for the
1313 // debugger to fill in.
1315 // To inject a function call, a debugger should:
1316 // 1. Check that the goroutine is in state _Grunning and that
1317 // there are at least 288 bytes free on the stack.
1318 // 2. Set SP as SP-16.
1319 // 3. Store the current LR in (SP) (using the SP after step 2).
1320 // 4. Store the current PC in the LR register.
1321 // 5. Write the desired argument frame size at SP-16
1322 // 6. Save all machine registers (including flags and fpsimd registers)
1323 // so they can be restored later by the debugger.
1324 // 7. Set the PC to debugCallV2 and resume execution.
1326 // If the goroutine is in state _Grunnable, then it's not generally
1327 // safe to inject a call because it may return out via other runtime
1328 // operations. Instead, the debugger should unwind the stack to find
1329 // the return to non-runtime code, add a temporary breakpoint there,
1330 // and inject the call once that breakpoint is hit.
1332 // If the goroutine is in any other state, it's not safe to inject a call.
1334 // This function communicates back to the debugger by setting R20 and
1335 // invoking BRK to raise a breakpoint signal. Note that the signal PC of
1336 // the signal triggered by the BRK instruction is the PC where the signal
1337 // is trapped, not the next PC, so to resume execution, the debugger needs
1338 // to set the signal PC to PC+4. See the comments in the implementation for
1339 // the protocol the debugger is expected to follow. InjectDebugCall in the
1340 // runtime tests demonstrates this protocol.
1342 // The debugger must ensure that any pointers passed to the function
1343 // obey escape analysis requirements. Specifically, it must not pass
1344 // a stack pointer to an escaping argument. debugCallV2 cannot check
1347 // This is ABIInternal because Go code injects its PC directly into new
1348 // goroutine stacks.
1349 TEXT runtime·debugCallV2<ABIInternal>(SB),NOSPLIT|NOFRAME,$0-0
1350 STP (R29, R30), -280(RSP)
1353 // Save all registers that may contain pointers so they can be
1354 // conservatively scanned.
1356 // We can't do anything that might clobber any of these
1357 // registers before this.
1358 STP (R27, g), (30*8)(RSP)
1359 STP (R25, R26), (28*8)(RSP)
1360 STP (R23, R24), (26*8)(RSP)
1361 STP (R21, R22), (24*8)(RSP)
1362 STP (R19, R20), (22*8)(RSP)
1363 STP (R16, R17), (20*8)(RSP)
1364 STP (R14, R15), (18*8)(RSP)
1365 STP (R12, R13), (16*8)(RSP)
1366 STP (R10, R11), (14*8)(RSP)
1367 STP (R8, R9), (12*8)(RSP)
1368 STP (R6, R7), (10*8)(RSP)
1369 STP (R4, R5), (8*8)(RSP)
1370 STP (R2, R3), (6*8)(RSP)
1371 STP (R0, R1), (4*8)(RSP)
1373 // Perform a safe-point check.
1374 MOVD R30, 8(RSP) // Caller's PC
1375 CALL runtime·debugCallCheck(SB)
1379 // The safety check failed. Put the reason string at the top
1385 // Set R20 to 8 and invoke BRK. The debugger should get the
1386 // reason a call can't be injected from SP+8 and resume execution.
1392 // Registers are saved and it's safe to make a call.
1393 // Open up a call frame, moving the stack if necessary.
1395 // Once the frame is allocated, this will set R20 to 0 and
1396 // invoke BRK. The debugger should write the argument
1397 // frame for the call at SP+8, set up argument registers,
1398 // set the LR as the signal PC + 4, set the PC to the function
1399 // to call, set R26 to point to the closure (if a closure call),
1400 // and resume execution.
1402 // If the function returns, this will set R20 to 1 and invoke
1403 // BRK. The debugger can then inspect any return value saved
1404 // on the stack at SP+8 and in registers. To resume execution,
1405 // the debugger should restore the LR from (SP).
1407 // If the function panics, this will set R20 to 2 and invoke BRK.
1408 // The interface{} value of the panic will be at SP+8. The debugger
1409 // can inspect the panic value and resume execution again.
1410 #define DEBUG_CALL_DISPATCH(NAME,MAXSIZE) \
1413 MOVD $NAME(SB), R0; \
1415 CALL runtime·debugCallWrap(SB); \
1418 MOVD 256(RSP), R0 // the argument frame size
1419 DEBUG_CALL_DISPATCH(debugCall32<>, 32)
1420 DEBUG_CALL_DISPATCH(debugCall64<>, 64)
1421 DEBUG_CALL_DISPATCH(debugCall128<>, 128)
1422 DEBUG_CALL_DISPATCH(debugCall256<>, 256)
1423 DEBUG_CALL_DISPATCH(debugCall512<>, 512)
1424 DEBUG_CALL_DISPATCH(debugCall1024<>, 1024)
1425 DEBUG_CALL_DISPATCH(debugCall2048<>, 2048)
1426 DEBUG_CALL_DISPATCH(debugCall4096<>, 4096)
1427 DEBUG_CALL_DISPATCH(debugCall8192<>, 8192)
1428 DEBUG_CALL_DISPATCH(debugCall16384<>, 16384)
1429 DEBUG_CALL_DISPATCH(debugCall32768<>, 32768)
1430 DEBUG_CALL_DISPATCH(debugCall65536<>, 65536)
1431 // The frame size is too large. Report the error.
1432 MOVD $debugCallFrameTooLarge<>(SB), R0
1435 MOVD R0, 16(RSP) // length of debugCallFrameTooLarge string
1441 // Calls and failures resume here.
1443 // Set R20 to 16 and invoke BRK. The debugger should restore
1444 // all registers except for PC and RSP and resume execution.
1447 // We must not modify flags after this point.
1449 // Restore pointer-containing registers, which may have been
1450 // modified from the debugger's copy by stack copying.
1451 LDP (30*8)(RSP), (R27, g)
1452 LDP (28*8)(RSP), (R25, R26)
1453 LDP (26*8)(RSP), (R23, R24)
1454 LDP (24*8)(RSP), (R21, R22)
1455 LDP (22*8)(RSP), (R19, R20)
1456 LDP (20*8)(RSP), (R16, R17)
1457 LDP (18*8)(RSP), (R14, R15)
1458 LDP (16*8)(RSP), (R12, R13)
1459 LDP (14*8)(RSP), (R10, R11)
1460 LDP (12*8)(RSP), (R8, R9)
1461 LDP (10*8)(RSP), (R6, R7)
1462 LDP (8*8)(RSP), (R4, R5)
1463 LDP (6*8)(RSP), (R2, R3)
1464 LDP (4*8)(RSP), (R0, R1)
1466 LDP -8(RSP), (R29, R27)
1467 ADD $288, RSP, RSP // Add 16 more bytes, see saveSigContext
1468 MOVD -16(RSP), R30 // restore old lr
1471 // runtime.debugCallCheck assumes that functions defined with the
1472 // DEBUG_CALL_FN macro are safe points to inject calls.
1473 #define DEBUG_CALL_FN(NAME,MAXSIZE) \
1474 TEXT NAME(SB),WRAPPER,$MAXSIZE-0; \
1475 NO_LOCAL_POINTERS; \
1481 DEBUG_CALL_FN(debugCall32<>, 32)
1482 DEBUG_CALL_FN(debugCall64<>, 64)
1483 DEBUG_CALL_FN(debugCall128<>, 128)
1484 DEBUG_CALL_FN(debugCall256<>, 256)
1485 DEBUG_CALL_FN(debugCall512<>, 512)
1486 DEBUG_CALL_FN(debugCall1024<>, 1024)
1487 DEBUG_CALL_FN(debugCall2048<>, 2048)
1488 DEBUG_CALL_FN(debugCall4096<>, 4096)
1489 DEBUG_CALL_FN(debugCall8192<>, 8192)
1490 DEBUG_CALL_FN(debugCall16384<>, 16384)
1491 DEBUG_CALL_FN(debugCall32768<>, 32768)
1492 DEBUG_CALL_FN(debugCall65536<>, 65536)
1494 // func debugCallPanicked(val interface{})
1495 TEXT runtime·debugCallPanicked(SB),NOSPLIT,$16-16
1496 // Copy the panic value to the top of stack at SP+8.
1497 MOVD val_type+0(FP), R0
1499 MOVD val_data+8(FP), R0
1505 // Note: these functions use a special calling convention to save generated code space.
1506 // Arguments are passed in registers, but the space for those arguments are allocated
1507 // in the caller's stack frame. These stubs write the args into that stack space and
1508 // then tail call to the corresponding runtime handler.
1509 // The tail call makes these stubs disappear in backtraces.
1511 // Defined as ABIInternal since the compiler generates ABIInternal
1512 // calls to it directly and it does not use the stack-based Go ABI.
1513 TEXT runtime·panicIndex<ABIInternal>(SB),NOSPLIT,$0-16
1514 JMP runtime·goPanicIndex<ABIInternal>(SB)
1515 TEXT runtime·panicIndexU<ABIInternal>(SB),NOSPLIT,$0-16
1516 JMP runtime·goPanicIndexU<ABIInternal>(SB)
1517 TEXT runtime·panicSliceAlen<ABIInternal>(SB),NOSPLIT,$0-16
1520 JMP runtime·goPanicSliceAlen<ABIInternal>(SB)
1521 TEXT runtime·panicSliceAlenU<ABIInternal>(SB),NOSPLIT,$0-16
1524 JMP runtime·goPanicSliceAlenU<ABIInternal>(SB)
1525 TEXT runtime·panicSliceAcap<ABIInternal>(SB),NOSPLIT,$0-16
1528 JMP runtime·goPanicSliceAcap<ABIInternal>(SB)
1529 TEXT runtime·panicSliceAcapU<ABIInternal>(SB),NOSPLIT,$0-16
1532 JMP runtime·goPanicSliceAcapU<ABIInternal>(SB)
1533 TEXT runtime·panicSliceB<ABIInternal>(SB),NOSPLIT,$0-16
1534 JMP runtime·goPanicSliceB<ABIInternal>(SB)
1535 TEXT runtime·panicSliceBU<ABIInternal>(SB),NOSPLIT,$0-16
1536 JMP runtime·goPanicSliceBU<ABIInternal>(SB)
1537 TEXT runtime·panicSlice3Alen<ABIInternal>(SB),NOSPLIT,$0-16
1540 JMP runtime·goPanicSlice3Alen<ABIInternal>(SB)
1541 TEXT runtime·panicSlice3AlenU<ABIInternal>(SB),NOSPLIT,$0-16
1544 JMP runtime·goPanicSlice3AlenU<ABIInternal>(SB)
1545 TEXT runtime·panicSlice3Acap<ABIInternal>(SB),NOSPLIT,$0-16
1548 JMP runtime·goPanicSlice3Acap<ABIInternal>(SB)
1549 TEXT runtime·panicSlice3AcapU<ABIInternal>(SB),NOSPLIT,$0-16
1552 JMP runtime·goPanicSlice3AcapU<ABIInternal>(SB)
1553 TEXT runtime·panicSlice3B<ABIInternal>(SB),NOSPLIT,$0-16
1556 JMP runtime·goPanicSlice3B<ABIInternal>(SB)
1557 TEXT runtime·panicSlice3BU<ABIInternal>(SB),NOSPLIT,$0-16
1560 JMP runtime·goPanicSlice3BU<ABIInternal>(SB)
1561 TEXT runtime·panicSlice3C<ABIInternal>(SB),NOSPLIT,$0-16
1562 JMP runtime·goPanicSlice3C<ABIInternal>(SB)
1563 TEXT runtime·panicSlice3CU<ABIInternal>(SB),NOSPLIT,$0-16
1564 JMP runtime·goPanicSlice3CU<ABIInternal>(SB)
1565 TEXT runtime·panicSliceConvert<ABIInternal>(SB),NOSPLIT,$0-16
1568 JMP runtime·goPanicSliceConvert<ABIInternal>(SB)
1570 TEXT ·getcallerfp<ABIInternal>(SB),NOSPLIT|NOFRAME,$0