static IRBB* CLG_(instrument)( VgCallbackClosure* closure, IRBB* bbIn, VexGuestLayout* layout, VexGuestExtents* vge, IRType gWordTy, IRType hWordTy ) { Int i; IRBB* bbOut; IRStmt* st, *stnext; Addr instrAddr, origAddr; UInt instrLen = 0, dataSize; UInt instrCount, costOffset; IRExpr *loadAddrExpr, *storeAddrExpr; BB* bb; IRDirty* di; IRExpr *arg1, **argv; Bool bb_seen_before = False; UInt cJumps = 0, cJumpsCorrected; Bool beforeIBoundary, instrIssued; if (gWordTy != hWordTy) { /* We don't currently support this case. */ VG_(tool_panic)("host/guest word size mismatch"); } // No instrumentation if it is switched off if (! CLG_(instrument_state)) { CLG_DEBUG(5, "instrument(BB %p) [Instrumentation OFF]\n", (Addr)closure->readdr); return bbIn; } CLG_DEBUG(3, "+ instrument(BB %p)\n", (Addr)closure->readdr); /* Set up BB for instrumented IR */ bbOut = emptyIRBB(); bbOut->tyenv = dopyIRTypeEnv(bbIn->tyenv); bbOut->next = dopyIRExpr(bbIn->next); bbOut->jumpkind = bbIn->jumpkind; // Copy verbatim any IR preamble preceding the first IMark i = 0; while (i < bbIn->stmts_used && bbIn->stmts[i]->tag != Ist_IMark) { addStmtToIRBB( bbOut, bbIn->stmts[i] ); i++; } // Get the first statement, and origAddr from it CLG_ASSERT(bbIn->stmts_used > 0); st = bbIn->stmts[i]; CLG_ASSERT(Ist_IMark == st->tag); instrAddr = origAddr = (Addr)st->Ist.IMark.addr; CLG_ASSERT(origAddr == st->Ist.IMark.addr); // XXX: check no overflow /* Get BB (creating if necessary). * JS: The hash table is keyed with orig_addr_noredir -- important! * JW: Why? If it is because of different chasing of the redirection, * this is not needed, as chasing is switched off in callgrind */ bb = CLG_(get_bb)(origAddr, bbIn, &bb_seen_before); //bb = CLG_(get_bb)(orig_addr_noredir, bbIn, &bb_seen_before); /* * Precondition: * - jmps_passed has number of cond.jumps passed in last executed BB * - current_bbcc has a pointer to the BBCC of the last executed BB * Thus, if bbcc_jmpkind is != -1 (JmpNone), * current_bbcc->bb->jmp_addr * gives the address of the jump source. * * The BBCC setup does 2 things: * - trace call: * * Unwind own call stack, i.e sync our ESP with real ESP * This is for ESP manipulation (longjmps, C++ exec handling) and RET * * For CALLs or JMPs crossing objects, record call arg + * push are on own call stack * * - prepare for cache log functions: * Set current_bbcc to BBCC that gets the costs for this BB execution * attached */ // helper call to setup_bbcc, with pointer to basic block info struct as argument arg1 = mkIRExpr_HWord( (HWord)bb ); argv = mkIRExprVec_1(arg1); di = unsafeIRDirty_0_N( 1, "setup_bbcc", VG_(fnptr_to_fnentry)( & CLG_(setup_bbcc) ), argv); addStmtToIRBB( bbOut, IRStmt_Dirty(di) ); instrCount = 0; costOffset = 0; // loop for each host instruction (starting from 'i') do { // We should be at an IMark statement CLG_ASSERT(Ist_IMark == st->tag); // Reset stuff for this original instruction loadAddrExpr = storeAddrExpr = NULL; instrIssued = False; dataSize = 0; // Process all the statements for this original instruction (ie. until // the next IMark statement, or the end of the block) do { i++; stnext = ( i < bbIn->stmts_used ? bbIn->stmts[i] : NULL ); beforeIBoundary = !stnext || (Ist_IMark == stnext->tag); collectStatementInfo(bbIn->tyenv, bbOut, st, &instrAddr, &instrLen, &loadAddrExpr, &storeAddrExpr, &dataSize, hWordTy); // instrument a simulator call before conditional jumps if (st->tag == Ist_Exit) { // Nb: instrLen will be zero if Vex failed to decode it. // Also Client requests can appear to be very large (eg. 18 // bytes on x86) because they are really multiple instructions. CLG_ASSERT( 0 == instrLen || bbIn->jumpkind == Ijk_ClientReq || (instrLen >= VG_MIN_INSTR_SZB && instrLen <= VG_MAX_INSTR_SZB) ); // Add instrumentation before this statement endOfInstr(bbOut, &(bb->instr[instrCount]), bb_seen_before, instrAddr - origAddr, instrLen, dataSize, &costOffset, instrIssued, loadAddrExpr, storeAddrExpr); // prepare for a possible further simcall in same host instr loadAddrExpr = storeAddrExpr = NULL; instrIssued = True; if (!bb_seen_before) { bb->jmp[cJumps].instr = instrCount; bb->jmp[cJumps].skip = False; } /* Update global variable jmps_passed (this is before the jump!) * A correction is needed if VEX inverted the last jump condition */ cJumpsCorrected = cJumps; if ((cJumps+1 == bb->cjmp_count) && bb->cjmp_inverted) cJumpsCorrected++; addConstMemStoreStmt( bbOut, (UWord) &CLG_(current_state).jmps_passed, cJumpsCorrected, hWordTy); cJumps++; } addStmtToIRBB( bbOut, st ); st = stnext; } while (!beforeIBoundary); // Add instrumentation for this original instruction. if (!instrIssued || (loadAddrExpr != 0) || (storeAddrExpr !=0)) endOfInstr(bbOut, &(bb->instr[instrCount]), bb_seen_before, instrAddr - origAddr, instrLen, dataSize, &costOffset, instrIssued, loadAddrExpr, storeAddrExpr); instrCount++; } while (st); /* Always update global variable jmps_passed (at end of BB) * A correction is needed if VEX inverted the last jump condition */ cJumpsCorrected = cJumps; if (bb->cjmp_inverted) cJumpsCorrected--; addConstMemStoreStmt( bbOut, (UWord) &CLG_(current_state).jmps_passed, cJumpsCorrected, hWordTy); /* This stores the instr of the call/ret at BB end */ bb->jmp[cJumps].instr = instrCount-1; CLG_ASSERT(bb->cjmp_count == cJumps); CLG_ASSERT(bb->instr_count == instrCount); instrAddr += instrLen; if (bb_seen_before) { CLG_ASSERT(bb->instr_len == instrAddr - origAddr); CLG_ASSERT(bb->cost_count == costOffset); CLG_ASSERT(bb->jmpkind == bbIn->jumpkind); } else { bb->instr_len = instrAddr - origAddr; bb->cost_count = costOffset; bb->jmpkind = bbIn->jumpkind; } CLG_DEBUG(3, "- instrument(BB %p): byteLen %u, CJumps %u, CostLen %u\n", origAddr, bb->instr_len, bb->cjmp_count, bb->cost_count); if (cJumps>0) { CLG_DEBUG(3, " [ "); for (i=0; i<cJumps; i++) CLG_DEBUG(3, "%d ", bb->jmp[i].instr); CLG_DEBUG(3, "], last inverted: %s \n", bb->cjmp_inverted ? "yes":"no"); } return bbOut; }
static IRBB* cg_instrument ( IRBB* bbIn, VexGuestLayout* layout, Addr64 orig_addr_noredir, VexGuestExtents* vge, IRType gWordTy, IRType hWordTy ) { Int i, isize; IRStmt* st; Addr64 cia; /* address of current insn */ CgState cgs; IRTypeEnv* tyenv = bbIn->tyenv; InstrInfo* curr_inode = NULL; if (gWordTy != hWordTy) { /* We don't currently support this case. */ VG_(tool_panic)("host/guest word size mismatch"); } /* Set up BB, including copying of the where-next stuff. */ cgs.bbOut = emptyIRBB(); cgs.bbOut->tyenv = dopyIRTypeEnv(tyenv); tl_assert( isIRAtom(bbIn->next) ); cgs.bbOut->next = dopyIRExpr(bbIn->next); cgs.bbOut->jumpkind = bbIn->jumpkind; // Copy verbatim any IR preamble preceding the first IMark i = 0; while (i < bbIn->stmts_used && bbIn->stmts[i]->tag != Ist_IMark) { addStmtToIRBB( cgs.bbOut, bbIn->stmts[i] ); i++; } // Get the first statement, and initial cia from it tl_assert(bbIn->stmts_used > 0); tl_assert(i < bbIn->stmts_used); st = bbIn->stmts[i]; tl_assert(Ist_IMark == st->tag); cia = st->Ist.IMark.addr; // Set up running state and get block info cgs.events_used = 0; cgs.bbInfo = get_BB_info(bbIn, (Addr)orig_addr_noredir); cgs.bbInfo_i = 0; if (DEBUG_CG) VG_(printf)("\n\n---------- cg_instrument ----------\n"); // Traverse the block, initialising inodes, adding events and flushing as // necessary. for (/*use current i*/; i < bbIn->stmts_used; i++) { st = bbIn->stmts[i]; tl_assert(isFlatIRStmt(st)); switch (st->tag) { case Ist_NoOp: case Ist_AbiHint: case Ist_Put: case Ist_PutI: case Ist_MFence: break; case Ist_IMark: cia = st->Ist.IMark.addr; isize = st->Ist.IMark.len; // If Vex fails to decode an instruction, the size will be zero. // Pretend otherwise. if (isize == 0) isize = VG_MIN_INSTR_SZB; // Sanity-check size. tl_assert( (VG_MIN_INSTR_SZB <= isize && isize <= VG_MAX_INSTR_SZB) || VG_CLREQ_SZB == isize ); // Get space for and init the inode, record it as the current one. // Subsequent Dr/Dw/Dm events from the same instruction will // also use it. curr_inode = setup_InstrInfo(&cgs, cia, isize); addEvent_Ir( &cgs, curr_inode ); break; case Ist_Tmp: { IRExpr* data = st->Ist.Tmp.data; if (data->tag == Iex_Load) { IRExpr* aexpr = data->Iex.Load.addr; // Note also, endianness info is ignored. I guess // that's not interesting. addEvent_Dr( &cgs, curr_inode, sizeofIRType(data->Iex.Load.ty), aexpr ); } break; } case Ist_Store: { IRExpr* data = st->Ist.Store.data; IRExpr* aexpr = st->Ist.Store.addr; addEvent_Dw( &cgs, curr_inode, sizeofIRType(typeOfIRExpr(tyenv, data)), aexpr ); break; } case Ist_Dirty: { Int dataSize; IRDirty* d = st->Ist.Dirty.details; if (d->mFx != Ifx_None) { /* This dirty helper accesses memory. Collect the details. */ tl_assert(d->mAddr != NULL); tl_assert(d->mSize != 0); dataSize = d->mSize; // Large (eg. 28B, 108B, 512B on x86) data-sized // instructions will be done inaccurately, but they're // very rare and this avoids errors from hitting more // than two cache lines in the simulation. if (dataSize > MIN_LINE_SIZE) dataSize = MIN_LINE_SIZE; if (d->mFx == Ifx_Read || d->mFx == Ifx_Modify) addEvent_Dr( &cgs, curr_inode, dataSize, d->mAddr ); if (d->mFx == Ifx_Write || d->mFx == Ifx_Modify) addEvent_Dw( &cgs, curr_inode, dataSize, d->mAddr ); } else { tl_assert(d->mAddr == NULL); tl_assert(d->mSize == 0); } break; } case Ist_Exit: /* We may never reach the next statement, so need to flush all outstanding transactions now. */ flushEvents( &cgs ); break; default: tl_assert(0); break; } /* Copy the original statement */ addStmtToIRBB( cgs.bbOut, st ); if (DEBUG_CG) { ppIRStmt(st); VG_(printf)("\n"); } } /* At the end of the bb. Flush outstandings. */ flushEvents( &cgs ); /* done. stay sane ... */ tl_assert(cgs.bbInfo_i == cgs.bbInfo->n_instrs); if (DEBUG_CG) { VG_(printf)( "goto {"); ppIRJumpKind(bbIn->jumpkind); VG_(printf)( "} "); ppIRExpr( bbIn->next ); VG_(printf)( "}\n"); } return cgs.bbOut; }
static IRBB* cg_instrument ( IRBB* bbIn, VexGuestLayout* layout, IRType gWordTy, IRType hWordTy ) { Int i, dataSize = 0, bbInfo_i; IRBB* bbOut; IRStmt* st; BB_info* bbInfo; Bool bbSeenBefore = False, addedInstrumentation, addInstNow; Addr instrAddr, origAddr; UInt instrLen; IRExpr *loadAddrExpr, *storeAddrExpr; if (gWordTy != hWordTy) { /* We don't currently support this case. */ VG_(tool_panic)("host/guest word size mismatch"); } /* Set up BB */ bbOut = emptyIRBB(); bbOut->tyenv = dopyIRTypeEnv(bbIn->tyenv); bbOut->next = dopyIRExpr(bbIn->next); bbOut->jumpkind = bbIn->jumpkind; // Get the first statement, and origAddr from it i = 0; tl_assert(bbIn->stmts_used > 0); st = bbIn->stmts[0]; tl_assert(Ist_IMark == st->tag); origAddr = (Addr)st->Ist.IMark.addr; tl_assert(origAddr == st->Ist.IMark.addr); // XXX: check no overflow // Get block info bbInfo = get_BB_info(bbIn, origAddr, &bbSeenBefore); bbInfo_i = 0; do { // We should be at an IMark statement tl_assert(Ist_IMark == st->tag); // Reset stuff for this original instruction loadAddrExpr = storeAddrExpr = NULL; dataSize = 0; addedInstrumentation = False; // Process all the statements for this original instruction (ie. until // the next IMark statement, or the end of the block) do { IRStmt* st2 = ( i+1 < bbIn->stmts_used ? bbIn->stmts[i+1] : NULL ); addInstNow = handleOneStatement(bbIn->tyenv, bbOut, st, st2, &instrAddr, &instrLen, &loadAddrExpr, &storeAddrExpr, &dataSize); if (addInstNow) { tl_assert(!addedInstrumentation); addedInstrumentation = True; // Add instrumentation before this statement. instrumentInstr(bbOut, &bbInfo->instrs[ bbInfo_i ], bbSeenBefore, instrAddr, instrLen, dataSize, loadAddrExpr, storeAddrExpr); } addStmtToIRBB( bbOut, st ); i++; st = st2; } while (st && Ist_IMark != st->tag); if (!addedInstrumentation) { // Add instrumentation now, after all the instruction's statements. instrumentInstr(bbOut, &bbInfo->instrs[ bbInfo_i ], bbSeenBefore, instrAddr, instrLen, dataSize, loadAddrExpr, storeAddrExpr); } bbInfo_i++; } while (st); return bbOut; }