1 /* -----------------------------------------------------------------------------
2 * $Id: StgCRun.c,v 1.28 2002/02/06 15:48:56 sewardj Exp $
4 * (c) The GHC Team, 1998-2000
8 * To run an STG function from C land, call
10 * rv = StgRun(f,BaseReg);
12 * where "f" is the STG function to call, and BaseReg is the address of the
13 * RegTable for this run (we might have separate RegTables if we're running
14 * multiple threads on an SMP machine).
16 * In the end, "f" must JMP to StgReturn (defined below),
17 * passing the return-value "rv" in R1,
18 * to return to the caller of StgRun returning "rv" in
19 * the whatever way C returns a value.
21 * NOTE: StgRun/StgReturn do *NOT* load or store Hp or any
22 * other registers (other than saving the C callee-saves
23 * registers). Instead, the called function "f" must do that
26 * GCC will have assumed that pushing/popping of C-stack frames is
27 * going on when it generated its code, and used stack space
28 * accordingly. However, we actually {\em post-process away} all
29 * such stack-framery (see \tr{ghc/driver/ghc-asm.lprl}). Things will
30 * be OK however, if we initially make sure there are
31 * @RESERVED_C_STACK_BYTES@ on the C-stack to begin with, for local
34 * -------------------------------------------------------------------------- */
36 #include "PosixSource.h"
40 * We define the following (unused) global register variables, because for
41 * some reason gcc generates sub-optimal code for StgRun() on the Alpha
42 * (unnecessarily saving extra registers on the stack) if we don't.
44 * Why do it at the top of this file, rather than near StgRun() below? Because
45 * gcc doesn't let us define global register variables after any function
46 * definition has been read. Any point after #include "Stg.h" would be too
49 * We define alpha_EXTRA_CAREFUL here to save $s6, $f8 and $f9 -- registers
50 * that we don't use but which are callee-save registers. The __divq() routine
51 * in libc.a clobbers $s6.
54 #ifdef alpha_TARGET_ARCH
55 #define alpha_EXTRA_CAREFUL
56 register long fake_ra __asm__("$26");
57 #ifdef alpha_EXTRA_CAREFUL
58 register long fake_s6 __asm__("$15");
59 register double fake_f8 __asm__("$f8");
60 register double fake_f9 __asm__("$f9");
64 /* include Stg.h first because we want real machine regs in here: we
65 * have to get the value of R1 back from Stg land to C land intact.
77 #ifdef USE_MINIINTERPRETER
79 /* -----------------------------------------------------------------------------
80 any architecture (using miniinterpreter)
81 -------------------------------------------------------------------------- */
83 extern StgThreadReturnCode StgRun(StgFunPtr f, StgRegTable *basereg)
87 fprintf(stderr,"Jumping to ");
91 f = (StgFunPtr) (f)();
93 return (StgThreadReturnCode)R1.i;
101 #else /* !USE_MINIINTERPRETER */
103 #ifdef LEADING_UNDERSCORE
104 #define STG_RETURN "_StgReturn"
106 #define STG_RETURN "StgReturn"
109 /* -----------------------------------------------------------------------------
111 -------------------------------------------------------------------------- */
113 #ifdef i386_TARGET_ARCH
116 StgRun(StgFunPtr f, StgRegTable *basereg) {
118 static volatile void* stg_esp_saved_before_stack_align;
120 unsigned char space[ RESERVED_C_STACK_BYTES + 4*sizeof(void *) ];
121 StgThreadReturnCode r;
125 * save callee-saves registers on behalf of the STG code.
127 "movl %%esp, %%eax\n\t"
129 "movl %%ebx,0(%%eax)\n\t"
130 "movl %%esi,4(%%eax)\n\t"
131 "movl %%edi,8(%%eax)\n\t"
132 "movl %%ebp,12(%%eax)\n\t"
138 * grab the function argument from the stack
142 * Get %%esp 8-aligned, first saving the current value. This
143 * moves it down another zero or four bytes, depending on
144 * whether it was already aligned or not. This is a bit
145 * subtle -- we must not have a swizzled %esp at any place
146 * where we might refer to one of the %digit parameters to
147 * this piece of assembly code, since gcc may generate
148 * %esp-relative stack offsets which will then be wrong.
152 "andl $0xFFFFFFF8, %%esp\n\t"
154 * And finally, jump to the function argument.
158 ".global " STG_RETURN "\n"
161 * Restore %%esp to pre-alignment state.
165 * Move return value from R1 to %%eax
167 "movl %%esi, %%eax\n\t"
169 * restore callee-saves registers. (Don't stomp on %%eax!)
171 "movl %%esp, %%edx\n\t"
173 "movl 0(%%edx),%%ebx\n\t" /* restore the registers saved above */
174 "movl 4(%%edx),%%esi\n\t"
175 "movl 8(%%edx),%%edi\n\t"
176 "movl 12(%%edx),%%ebp\n\t"
178 : "=&a" (r), "=m" (space),
179 "=m" (stg_esp_saved_before_stack_align)
180 : "m" (f), "m" (basereg), "i" (RESERVED_C_STACK_BYTES),
181 "m" (stg_esp_saved_before_stack_align)
183 : "edx" /* stomps on %edx */
191 /* -----------------------------------------------------------------------------
195 OLD COMMENT from GHC-3.02:
197 We want tailjumps to be calls, because `call xxx' is the only Sparc
198 branch that allows an arbitrary label as a target. (Gcc's ``goto
199 *target'' construct ends up loading the label into a register and
200 then jumping, at the cost of two extra instructions for the 32-bit
203 When entering the threaded world, we stash our return address in a
204 known location so that \tr{%i7} is available as an extra
205 callee-saves register. Of course, we have to restore this when
206 coming out of the threaded world.
208 I hate this god-forsaken architecture. Since the top of the
209 reserved stack space is used for globals and the bottom is reserved
210 for outgoing arguments, we have to stick our return address
211 somewhere in the middle. Currently, I'm allowing 100 extra
212 outgoing arguments beyond the first 6. --JSM
214 Updated info (GHC 4.06): we don't appear to use %i7 any more, so
215 I'm not sure whether we still need to save it. Incedentally, what
216 does the last paragraph above mean when it says "the top of the
217 stack is used for globals"? What globals? --SDM
219 Updated info (GHC 4.08.2): not saving %i7 any more (see below).
220 -------------------------------------------------------------------------- */
222 #ifdef sparc_TARGET_ARCH
225 StgRun(StgFunPtr f, StgRegTable *basereg) {
227 unsigned char space[RESERVED_C_STACK_BYTES];
229 register void *i7 __asm__("%i7");
230 ((void **)(space))[100] = i7;
235 ".global " STG_RETURN "\n"
237 : : : "l0","l1","l2","l3","l4","l5","l6","l7");
238 /* we tell the C compiler that l0-l7 are clobbered on return to
239 * StgReturn, otherwise it tries to use these to save eg. the
240 * address of space[100] across the call. The correct thing
241 * to do would be to save all the callee-saves regs, but we
242 * can't be bothered to do that.
244 * The code that gcc generates for this little fragment is now
245 * terrible. We could do much better by coding it directly in
249 /* updated 4.08.2: we don't save %i7 in the middle of the reserved
250 * space any more, since gcc tries to save its address across the
251 * call to f(), this gets clobbered in STG land and we end up
252 * dereferencing a bogus pointer in StgReturn.
254 __asm__ volatile ("ld %1,%0"
255 : "=r" (i7) : "m" (((void **)(space))[100]));
257 return (StgThreadReturnCode)R1.i;
262 /* -----------------------------------------------------------------------------
265 "The stack pointer (SP) must at all times denote an address that has octaword
266 alignment. (This restriction has the side effect that the in-memory portion
267 of the argument list, if any, will start on an octaword boundary.) Note that
268 the stack grows toward lower addresses. During a procedure invocation, SP
269 can never be set to a value that is higher than the value of SP at entry to
270 that procedure invocation.
272 "The contents of the stack, located above the portion of the argument list
273 (if any) that is passed in memory, belong to the calling procedure. Because
274 they are part of the calling procedure, they should not be read or written
275 by the called procedure, except as specified by indirect arguments or
276 language-controlled up-level references.
278 "The SP value might be used by the hardware when raising exceptions and
279 asynchronous interrupts. It must be assumed that the contents of the stack
280 below the current SP value and within the stack for the current thread are
281 continually and unpredictably modified, as specified in the _Alpha
282 Architecture Reference Manual_, and as a result of asynchronous software
285 -- Compaq Computer Corporation, Houston. Tru64 UNIX Calling Standard for
286 Alpha Systems, 5.1 edition, August 2000, section 3.2.1. http://www.
287 tru64unix.compaq.com/docs/base_doc/DOCUMENTATION/V51_PDF/ARH9MBTE.PDF
288 -------------------------------------------------------------------------- */
290 #ifdef alpha_TARGET_ARCH
293 StgRun(StgFunPtr f, StgRegTable *basereg)
295 register long real_ra __asm__("$26"); volatile long save_ra;
297 register long real_s0 __asm__("$9" ); volatile long save_s0;
298 register long real_s1 __asm__("$10"); volatile long save_s1;
299 register long real_s2 __asm__("$11"); volatile long save_s2;
300 register long real_s3 __asm__("$12"); volatile long save_s3;
301 register long real_s4 __asm__("$13"); volatile long save_s4;
302 register long real_s5 __asm__("$14"); volatile long save_s5;
303 #ifdef alpha_EXTRA_CAREFUL
304 register long real_s6 __asm__("$15"); volatile long save_s6;
307 register double real_f2 __asm__("$f2"); volatile double save_f2;
308 register double real_f3 __asm__("$f3"); volatile double save_f3;
309 register double real_f4 __asm__("$f4"); volatile double save_f4;
310 register double real_f5 __asm__("$f5"); volatile double save_f5;
311 register double real_f6 __asm__("$f6"); volatile double save_f6;
312 register double real_f7 __asm__("$f7"); volatile double save_f7;
313 #ifdef alpha_EXTRA_CAREFUL
314 register double real_f8 __asm__("$f8"); volatile double save_f8;
315 register double real_f9 __asm__("$f9"); volatile double save_f9;
318 register StgFunPtr real_pv __asm__("$27");
320 StgThreadReturnCode ret;
330 #ifdef alpha_EXTRA_CAREFUL
340 #ifdef alpha_EXTRA_CAREFUL
347 __asm__ volatile( "lda $30,-%0($30)" "\n"
348 "\t" "jmp ($27)" "\n"
350 ".globl " STG_RETURN "\n"
352 "\t" "lda $30,%0($30)" "\n"
353 : : "K" (RESERVED_C_STACK_BYTES));
363 #ifdef alpha_EXTRA_CAREFUL
373 #ifdef alpha_EXTRA_CAREFUL
383 #endif /* alpha_TARGET_ARCH */
385 /* -----------------------------------------------------------------------------
387 -------------------------------------------------------------------------- */
389 #ifdef hppa1_1_TARGET_ARCH
392 StgRun(StgFunPtr f, StgRegTable *basereg)
394 StgChar space[RESERVED_C_STACK_BYTES+16*sizeof(long)+10*sizeof(double)];
395 StgThreadReturnCode ret;
397 __asm__ volatile ("ldo %0(%%r30),%%r19\n"
398 "\tstw %%r3, 0(0,%%r19)\n"
399 "\tstw %%r4, 4(0,%%r19)\n"
400 "\tstw %%r5, 8(0,%%r19)\n"
401 "\tstw %%r6,12(0,%%r19)\n"
402 "\tstw %%r7,16(0,%%r19)\n"
403 "\tstw %%r8,20(0,%%r19)\n"
404 "\tstw %%r9,24(0,%%r19)\n"
405 "\tstw %%r10,28(0,%%r19)\n"
406 "\tstw %%r11,32(0,%%r19)\n"
407 "\tstw %%r12,36(0,%%r19)\n"
408 "\tstw %%r13,40(0,%%r19)\n"
409 "\tstw %%r14,44(0,%%r19)\n"
410 "\tstw %%r15,48(0,%%r19)\n"
411 "\tstw %%r16,52(0,%%r19)\n"
412 "\tstw %%r17,56(0,%%r19)\n"
413 "\tstw %%r18,60(0,%%r19)\n"
414 "\tldo 80(%%r19),%%r19\n"
415 "\tfstds %%fr12,-16(0,%%r19)\n"
416 "\tfstds %%fr13, -8(0,%%r19)\n"
417 "\tfstds %%fr14, 0(0,%%r19)\n"
418 "\tfstds %%fr15, 8(0,%%r19)\n"
419 "\tldo 32(%%r19),%%r19\n"
420 "\tfstds %%fr16,-16(0,%%r19)\n"
421 "\tfstds %%fr17, -8(0,%%r19)\n"
422 "\tfstds %%fr18, 0(0,%%r19)\n"
423 "\tfstds %%fr19, 8(0,%%r19)\n"
424 "\tldo 32(%%r19),%%r19\n"
425 "\tfstds %%fr20,-16(0,%%r19)\n"
426 "\tfstds %%fr21, -8(0,%%r19)\n" : :
427 "n" (-(116 * sizeof(long) + 10 * sizeof(double))) : "%r19"
432 __asm__ volatile (".align 4\n"
433 "\t.EXPORT " STG_RETURN ",CODE\n"
434 "\t.EXPORT " STG_RETURN ",ENTRY,PRIV_LEV=3\n"
436 /* "\tldo %0(%%r3),%%r19\n" */
437 "\tldo %1(%%r30),%%r19\n"
438 "\tcopy %%r11, %0\n" /* save R1 */
439 "\tldw 0(0,%%r19),%%r3\n"
440 "\tldw 4(0,%%r19),%%r4\n"
441 "\tldw 8(0,%%r19),%%r5\n"
442 "\tldw 12(0,%%r19),%%r6\n"
443 "\tldw 16(0,%%r19),%%r7\n"
444 "\tldw 20(0,%%r19),%%r8\n"
445 "\tldw 24(0,%%r19),%%r9\n"
446 "\tldw 28(0,%%r19),%%r10\n"
447 "\tldw 32(0,%%r19),%%r11\n"
448 "\tldw 36(0,%%r19),%%r12\n"
449 "\tldw 40(0,%%r19),%%r13\n"
450 "\tldw 44(0,%%r19),%%r14\n"
451 "\tldw 48(0,%%r19),%%r15\n"
452 "\tldw 52(0,%%r19),%%r16\n"
453 "\tldw 56(0,%%r19),%%r17\n"
454 "\tldw 60(0,%%r19),%%r18\n"
455 "\tldo 80(%%r19),%%r19\n"
456 "\tfldds -16(0,%%r19),%%fr12\n"
457 "\tfldds -8(0,%%r19),%%fr13\n"
458 "\tfldds 0(0,%%r19),%%fr14\n"
459 "\tfldds 8(0,%%r19),%%fr15\n"
460 "\tldo 32(%%r19),%%r19\n"
461 "\tfldds -16(0,%%r19),%%fr16\n"
462 "\tfldds -8(0,%%r19),%%fr17\n"
463 "\tfldds 0(0,%%r19),%%fr18\n"
464 "\tfldds 8(0,%%r19),%%fr19\n"
465 "\tldo 32(%%r19),%%r19\n"
466 "\tfldds -16(0,%%r19),%%fr20\n"
467 "\tfldds -8(0,%%r19),%%fr21\n"
469 : "n" (-(116 * sizeof(long) + 10 * sizeof(double)))
476 #endif /* hppa1_1_TARGET_ARCH */
478 #endif /* !USE_MINIINTERPRETER */