X-Git-Url: http://git.megacz.com/?a=blobdiff_plain;f=ghc%2Frts%2FStgCRun.c;h=7597ff228048f6e4124028ad9f36d0d650104c79;hb=bfb041829d38b711d7f5f783b1ed944f01c23a9b;hp=b835a49caaafe619ec6ffb49c5d7323c298c32c5;hpb=2f6e73fc393bb526e03f391bfa40c1912560e378;p=ghc-hetmet.git diff --git a/ghc/rts/StgCRun.c b/ghc/rts/StgCRun.c index b835a49..7597ff2 100644 --- a/ghc/rts/StgCRun.c +++ b/ghc/rts/StgCRun.c @@ -50,7 +50,7 @@ * in libc.a clobbers $s6. */ #include "ghcconfig.h" -#ifdef alpha_TARGET_ARCH +#ifdef alpha_HOST_ARCH #define alpha_EXTRA_CAREFUL register long fake_ra __asm__("$26"); register long fake_gp __asm__("$29"); @@ -82,17 +82,17 @@ register double fake_f9 __asm__("$f9"); any architecture (using miniinterpreter) -------------------------------------------------------------------------- */ -StgThreadReturnCode StgRun(StgFunPtr f, StgRegTable *basereg STG_UNUSED) +StgRegTable * StgRun(StgFunPtr f, StgRegTable *basereg STG_UNUSED) { while (f) { - if (RtsFlags[0].DebugFlags.interpreter) { + IF_DEBUG(interpreter, debugBelch("Jumping to "); printPtr((P_)f); fflush(stdout); debugBelch("\n"); - } + ); f = (StgFunPtr) (f)(); } - return (StgThreadReturnCode)R1.i; + return (StgRegTable *)R1.p; } StgFunPtr StgReturn(void) @@ -112,13 +112,19 @@ StgFunPtr StgReturn(void) x86 architecture -------------------------------------------------------------------------- */ -#ifdef i386_TARGET_ARCH +#ifdef i386_HOST_ARCH -StgThreadReturnCode +#ifdef darwin_TARGET_OS +#define STG_GLOBAL ".globl " +#else +#define STG_GLOBAL ".global " +#endif + +StgRegTable * StgRun(StgFunPtr f, StgRegTable *basereg) { unsigned char space[ RESERVED_C_STACK_BYTES + 4*sizeof(void *) ]; - StgThreadReturnCode r; + StgRegTable * r; __asm__ volatile ( /* @@ -135,14 +141,32 @@ StgRun(StgFunPtr f, StgRegTable *basereg) { */ "movl %3,%%ebx\n\t" /* - * grab the function argument from the stack, and jump to it. + * grab the function argument from the stack */ "movl %2,%%eax\n\t" + +#if darwin_TARGET_OS + /* + * Darwin: keep the stack aligned + */ + "subl $12,%%esp\n\t" +#endif + + /* + * jump to it + */ "jmp *%%eax\n\t" - ".global " STG_RETURN "\n" + STG_GLOBAL STG_RETURN "\n" STG_RETURN ":\n\t" +#if darwin_TARGET_OS + /* + * Darwin: keep the stack aligned + */ + "addl $12,%%esp\n\t" +#endif + "movl %%esi, %%eax\n\t" /* Return value in R1 */ /* @@ -175,11 +199,12 @@ StgRun(StgFunPtr f, StgRegTable *basereg) { ------------------------------------------------------------------------- */ -#ifdef x86_64_TARGET_ARCH +#ifdef x86_64_HOST_ARCH -extern StgThreadReturnCode StgRun(StgFunPtr f, StgRegTable *basereg); +extern StgRegTable * StgRun(StgFunPtr f, StgRegTable *basereg); -static void StgRunIsImplementedInAssembler(void) +void StgRunIsImplementedInAssembler(void); +void StgRunIsImplementedInAssembler(void) { __asm__ volatile ( /* @@ -225,7 +250,41 @@ static void StgRunIsImplementedInAssembler(void) "addq %0, %%rsp\n\t" "retq" - : : "i"(RESERVED_C_STACK_BYTES+48 /*stack frame size*/)); + : : "i"(RESERVED_C_STACK_BYTES+48+8 /*stack frame size*/)); + /* + HACK alert! + + The x86_64 ABI specifies that on a procedure call, %rsp is + aligned on a 16-byte boundary + 8. That is, the first + argument on the stack after the return address will be + 16-byte aligned. + + Which should be fine: RESERVED_C_STACK_BYTES+48 is a multiple + of 16 bytes. + + BUT... when we do a C-call from STG land, gcc likes to put the + stack alignment adjustment in the prolog. eg. if we're calling + a function with arguments in regs, gcc will insert 'subq $8,%rsp' + in the prolog, to keep %rsp aligned (the return address is 8 + bytes, remember). The mangler throws away the prolog, so we + lose the stack alignment. + + The hack is to add this extra 8 bytes to our %rsp adjustment + here, so that throughout STG code, %rsp is 16-byte aligned, + ready for a C-call. + + A quick way to see if this is wrong is to compile this code: + + main = System.Exit.exitWith ExitSuccess + + And run it with +RTS -sstderr. The stats code in the RTS, in + particular statsPrintf(), relies on the stack alignment because + it saves the %xmm regs on the stack, so it'll fall over if the + stack isn't aligned, and calling exitWith from Haskell invokes + shutdownHaskellAndExit using a C call. + + Future gcc releases will almost certainly break this hack... + */ } #endif /* x86-64 */ @@ -261,9 +320,9 @@ static void StgRunIsImplementedInAssembler(void) Updated info (GHC 4.08.2): not saving %i7 any more (see below). -------------------------------------------------------------------------- */ -#ifdef sparc_TARGET_ARCH +#ifdef sparc_HOST_ARCH -StgThreadReturnCode +StgRegTable * StgRun(StgFunPtr f, StgRegTable *basereg) { unsigned char space[RESERVED_C_STACK_BYTES]; @@ -296,7 +355,7 @@ StgRun(StgFunPtr f, StgRegTable *basereg) { __asm__ volatile ("ld %1,%0" : "=r" (i7) : "m" (((void **)(space))[100])); #endif - return (StgThreadReturnCode)R1.i; + return (StgRegTable *)R1.i; } #endif @@ -329,9 +388,9 @@ StgRun(StgFunPtr f, StgRegTable *basereg) { tru64unix.compaq.com/docs/base_doc/DOCUMENTATION/V51_PDF/ARH9MBTE.PDF -------------------------------------------------------------------------- */ -#ifdef alpha_TARGET_ARCH +#ifdef alpha_HOST_ARCH -StgThreadReturnCode +StgRegTable * StgRun(StgFunPtr f, StgRegTable *basereg) { register long real_ra __asm__("$26"); volatile long save_ra; @@ -360,7 +419,7 @@ StgRun(StgFunPtr f, StgRegTable *basereg) register StgFunPtr real_pv __asm__("$27"); - StgThreadReturnCode ret; + StgRegTable * ret; save_ra = real_ra; save_gp = real_gp; @@ -425,19 +484,19 @@ StgRun(StgFunPtr f, StgRegTable *basereg) return ret; } -#endif /* alpha_TARGET_ARCH */ +#endif /* alpha_HOST_ARCH */ /* ----------------------------------------------------------------------------- HP-PA architecture -------------------------------------------------------------------------- */ -#ifdef hppa1_1_TARGET_ARCH +#ifdef hppa1_1_HOST_ARCH -StgThreadReturnCode +StgRegTable * StgRun(StgFunPtr f, StgRegTable *basereg) { StgChar space[RESERVED_C_STACK_BYTES+16*sizeof(long)+10*sizeof(double)]; - StgThreadReturnCode ret; + StgRegTable * ret; __asm__ volatile ("ldo %0(%%r30),%%r19\n" "\tstw %%r3, 0(0,%%r19)\n" @@ -518,7 +577,7 @@ StgRun(StgFunPtr f, StgRegTable *basereg) return ret; } -#endif /* hppa1_1_TARGET_ARCH */ +#endif /* hppa1_1_HOST_ARCH */ /* ----------------------------------------------------------------------------- PowerPC architecture @@ -527,13 +586,18 @@ StgRun(StgFunPtr f, StgRegTable *basereg) -------------------------------------------------------------------------- */ -#ifdef powerpc_TARGET_ARCH +#ifdef powerpc_HOST_ARCH -extern StgThreadReturnCode StgRun(StgFunPtr f, StgRegTable *basereg); +extern StgRegTable * StgRun(StgFunPtr f, StgRegTable *basereg); -#ifdef darwin_TARGET_OS -static void StgRunIsImplementedInAssembler(void) +#ifdef darwin_HOST_OS +void StgRunIsImplementedInAssembler(void) { +#if HAVE_SUBSECTIONS_VIA_SYMBOLS + // if the toolchain supports deadstripping, we have to + // prevent it here (it tends to get confused here). + __asm__ volatile (".no_dead_strip _StgRunIsImplementedInAssembler"); +#endif __asm__ volatile ( "\n.globl _StgRun\n" "_StgRun:\n" @@ -639,10 +703,10 @@ static void StgRunIsImplementedInAssembler(void) -------------------------------------------------------------------------- */ -#ifdef powerpc64_TARGET_ARCH +#ifdef powerpc64_HOST_ARCH -#ifdef linux_TARGET_OS -extern StgThreadReturnCode StgRun(StgFunPtr f, StgRegTable *basereg); +#ifdef linux_HOST_OS +extern StgRegTable * StgRun(StgFunPtr f, StgRegTable *basereg); static void StgRunIsImplementedInAssembler(void) { @@ -762,7 +826,7 @@ static void StgRunIsImplementedInAssembler(void) "\tblr\n" : : "i"(RESERVED_C_STACK_BYTES+304 /*stack frame size*/)); } -#else // linux_TARGET_OS +#else // linux_HOST_OS #error Only linux support for power64 right now. #endif @@ -781,7 +845,7 @@ static void StgRunIsImplementedInAssembler(void) loc31: saved gp (gcc 3.3 uses this slot) -------------------------------------------------------------------------- */ -#ifdef ia64_TARGET_ARCH +#ifdef ia64_HOST_ARCH /* the memory stack is rarely used, so 16K is excessive */ #undef RESERVED_C_STACK_BYTES