X-Git-Url: http://git.megacz.com/?p=ghc-hetmet.git;a=blobdiff_plain;f=rts%2FThreadPaused.c;h=7aee59dcd5f55af964275b9df09b1a997c0e4386;hp=3aa66edc191a858aef32056bb59860bada32b5b4;hb=5d52d9b64c21dcf77849866584744722f8121389;hpb=53a442f10d80cd85b33620a023c4a8749a7c0b20 diff --git a/rts/ThreadPaused.c b/rts/ThreadPaused.c index 3aa66ed..7aee59d 100644 --- a/rts/ThreadPaused.c +++ b/rts/ThreadPaused.c @@ -6,13 +6,15 @@ * * ---------------------------------------------------------------------------*/ +// #include "PosixSource.h" #include "Rts.h" -#include "Storage.h" -#include "LdvProfile.h" + +#include "ThreadPaused.h" +#include "sm/Storage.h" #include "Updates.h" #include "RaiseAsync.h" #include "Trace.h" -#include "RtsFlags.h" +#include "Threads.h" #include // for memmove() @@ -27,7 +29,7 @@ struct stack_gap { StgWord gap_size; struct stack_gap *next_gap; }; static void -stackSqueeze(StgTSO *tso, StgPtr bottom) +stackSqueeze(Capability *cap, StgTSO *tso, StgPtr bottom) { StgPtr frame; rtsBool prev_was_update_frame; @@ -74,7 +76,7 @@ stackSqueeze(StgTSO *tso, StgPtr bottom) * screw us up if we don't check. */ if (upd->updatee != updatee && !closure_IND(upd->updatee)) { - UPD_IND_NOLOCK(upd->updatee, updatee); + updateThunk(cap, tso, upd->updatee, updatee); } // now mark this update frame as a stack gap. The gap @@ -141,23 +143,23 @@ stackSqueeze(StgTSO *tso, StgPtr bottom) // indicates unused // { - void *sp; - void *gap_start, *next_gap_start, *gap_end; + StgWord8 *sp; + StgWord8 *gap_start, *next_gap_start, *gap_end; nat chunk_size; - next_gap_start = (void *)((unsigned char*)gap + sizeof(StgUpdateFrame)); + next_gap_start = (StgWord8*)gap + sizeof(StgUpdateFrame); sp = next_gap_start; while ((StgPtr)gap > tso->sp) { // we're working in *bytes* now... gap_start = next_gap_start; - gap_end = (void*) ((unsigned char*)gap_start - gap->gap_size * sizeof(W_)); + gap_end = gap_start - gap->gap_size * sizeof(W_); gap = gap->next_gap; - next_gap_start = (void *)((unsigned char*)gap + sizeof(StgUpdateFrame)); + next_gap_start = (StgWord8*)gap + sizeof(StgUpdateFrame); - chunk_size = (unsigned char*)gap_end - (unsigned char*)next_gap_start; + chunk_size = gap_end - next_gap_start; sp -= chunk_size; memmove(sp, next_gap_start, chunk_size); } @@ -195,6 +197,10 @@ threadPaused(Capability *cap, StgTSO *tso) maybePerformBlockedException (cap, tso); if (tso->what_next == ThreadKilled) { return; } + // NB. Blackholing is *compulsory*, we must either do lazy + // blackholing, or eager blackholing consistently. See Note + // [upd-black-hole] in sm/Scav.c. + stack_end = &tso->stack[tso->stack_size]; frame = (StgClosure *)tso->sp; @@ -224,58 +230,67 @@ threadPaused(Capability *cap, StgTSO *tso) #ifdef THREADED_RTS retry: #endif - if (closure_flags[INFO_PTR_TO_STRUCT(bh_info)->type] & _IND - || bh_info == &stg_BLACKHOLE_info) { + if (bh_info == &stg_BLACKHOLE_info || + bh_info == &stg_WHITEHOLE_info) + { debugTrace(DEBUG_squeeze, "suspending duplicate work: %ld words of stack", (long)((StgPtr)frame - tso->sp)); // If this closure is already an indirection, then - // suspend the computation up to this point: - suspendComputation(cap,tso,(StgPtr)frame); + // suspend the computation up to this point. + // NB. check raiseAsync() to see what happens when + // we're in a loop (#2783). + suspendComputation(cap,tso,(StgUpdateFrame*)frame); // Now drop the update frame, and arrange to return // the value to the frame underneath: tso->sp = (StgPtr)frame + sizeofW(StgUpdateFrame) - 2; tso->sp[1] = (StgWord)bh; + ASSERT(bh->header.info != &stg_TSO_info); tso->sp[0] = (W_)&stg_enter_info; // And continue with threadPaused; there might be // yet more computation to suspend. - threadPaused(cap,tso); - return; + frame = (StgClosure *)tso->sp + 2; + prev_was_update_frame = rtsFalse; + continue; } - if (bh->header.info != &stg_CAF_BLACKHOLE_info) { -#if (!defined(LAZY_BLACKHOLING)) && defined(DEBUG) - debugBelch("Unexpected lazy BHing required at 0x%04lx\n",(long)bh); -#endif - // zero out the slop so that the sanity checker can tell - // where the next closure is. - DEBUG_FILL_SLOP(bh); -#ifdef PROFILING - // @LDV profiling - // We pretend that bh is now dead. - LDV_recordDead_FILL_SLOP_DYNAMIC((StgClosure *)bh); -#endif + // zero out the slop so that the sanity checker can tell + // where the next closure is. + DEBUG_FILL_SLOP(bh); + // @LDV profiling + // We pretend that bh is now dead. + LDV_RECORD_DEAD_FILL_SLOP_DYNAMIC((StgClosure *)bh); + + // an EAGER_BLACKHOLE or CAF_BLACKHOLE gets turned into a + // BLACKHOLE here. #ifdef THREADED_RTS - cur_bh_info = (const StgInfoTable *) - cas((StgVolatilePtr)&bh->header.info, - (StgWord)bh_info, - (StgWord)&stg_BLACKHOLE_info); - - if (cur_bh_info != bh_info) { - bh_info = cur_bh_info; - goto retry; - } -#else - SET_INFO(bh,&stg_BLACKHOLE_info); + // first we turn it into a WHITEHOLE to claim it, and if + // successful we write our TSO and then the BLACKHOLE info pointer. + cur_bh_info = (const StgInfoTable *) + cas((StgVolatilePtr)&bh->header.info, + (StgWord)bh_info, + (StgWord)&stg_WHITEHOLE_info); + + if (cur_bh_info != bh_info) { + bh_info = cur_bh_info; + goto retry; + } #endif - // We pretend that bh has just been created. - LDV_RECORD_CREATE(bh); - } + // The payload of the BLACKHOLE points to the TSO + ((StgInd *)bh)->indirectee = (StgClosure *)tso; + write_barrier(); + SET_INFO(bh,&stg_BLACKHOLE_info); + + // .. and we need a write barrier, since we just mutated the closure: + recordClosureMutated(cap,bh); + + // We pretend that bh has just been created. + LDV_RECORD_CREATE(bh); frame = (StgClosure *) ((StgUpdateFrame *)frame + 1); if (prev_was_update_frame) { @@ -310,7 +325,13 @@ end: // the number of words we have to shift down is less than the // number of stack words we squeeze away by doing so. if (RtsFlags.GcFlags.squeezeUpdFrames == rtsTrue && - ((weight <= 4 && words_to_squeeze > 0) || weight < words_to_squeeze)) { - stackSqueeze(tso, (StgPtr)frame); + ((weight <= 8 && words_to_squeeze > 0) || weight < words_to_squeeze)) { + // threshold above bumped from 5 to 8 as a result of #2797 + stackSqueeze(cap, tso, (StgPtr)frame); + tso->flags |= TSO_SQUEEZED; + // This flag tells threadStackOverflow() that the stack was + // squeezed, because it may not need to be expanded. + } else { + tso->flags &= ~TSO_SQUEEZED; } }