changeset ee6641d7c713 in /z/repo/m5 details: http://repo.m5sim.org/m5?cmd=changeset;node=ee6641d7c713 description: O3: Fix itstate prediction and recovery.
Any change of control flow now resets the itstate to 0 mask and 0 condition, except where the control flow alteration write into the cpsr register. These case, for example return from an iterrupt, require the predecoder to recover the itstate. As there is a window of opportunity between the return from an interrupt changing the control flow at the head of the pipe and the commit of the update to the CPSR, the predecoder needs to be able to grab the ITstate early. This is now handled by setting the forcedItState inside a PCstate for the control flow altering instruction. That instruction will have the correct mask/cond, but will not have a valid itstate until advancePC is called (note this happens to advance the execution). When the new PCstate is copy constructed it gets the itstate cond/mask, and upon advancing the PC the itstate becomes valid. Subsequent advancing invalidates the state and zeroes the cond/mask. This is handled in isolation for the ARM ISA and should have no impact on other ISAs. Refer arch/arm/types.hh and arch/arm/predecoder.cc for the details. diffstat: src/arch/arm/isa/insts/data.isa | 2 + src/arch/arm/isa/insts/ldr.isa | 6 +++- src/arch/arm/isa/insts/macromem.isa | 2 + src/arch/arm/isa/insts/misc.isa | 4 +++ src/arch/arm/isa/operands.isa | 1 + src/arch/arm/predecoder.cc | 14 ++++++++++-- src/arch/arm/predecoder.hh | 7 ++++++ src/arch/arm/types.hh | 39 ++++++++++++++++++++++++++++++++++-- 8 files changed, 67 insertions(+), 8 deletions(-) diffs (216 lines): diff -r b2c7e56572a4 -r ee6641d7c713 src/arch/arm/isa/insts/data.isa --- a/src/arch/arm/isa/insts/data.isa Tue Jan 18 16:30:05 2011 -0600 +++ b/src/arch/arm/isa/insts/data.isa Tue Jan 18 16:30:05 2011 -0600 @@ -241,6 +241,8 @@ CondCodes = CondCodesMask & newCpsr; NextThumb = ((CPSR)newCpsr).t; NextJazelle = ((CPSR)newCpsr).j; + ForcedItState = ((((CPSR)newCpsr).it2 << 2) & 0xFC) + | (((CPSR)newCpsr).it1 & 0x3); ''' buildImmDataInst(mnem + 's', code, flagType, suffix = "ImmPclr", buildCc = False, diff -r b2c7e56572a4 -r ee6641d7c713 src/arch/arm/isa/insts/ldr.isa --- a/src/arch/arm/isa/insts/ldr.isa Tue Jan 18 16:30:05 2011 -0600 +++ b/src/arch/arm/isa/insts/ldr.isa Tue Jan 18 16:30:05 2011 -0600 @@ -107,13 +107,15 @@ SCTLR sctlr = Sctlr; // Use the version of NPC that gets set before NextThumb pNPC = cSwap<uint32_t>(Mem.ud, cpsr.e); + uint32_t tempSpsr = cSwap<uint32_t>(Mem.ud >> 32, cpsr.e); uint32_t newCpsr = - cpsrWriteByInstr(cpsr | CondCodes, - cSwap<uint32_t>(Mem.ud >> 32, cpsr.e), + cpsrWriteByInstr(cpsr | CondCodes, tempSpsr, 0xF, true, sctlr.nmfi); Cpsr = ~CondCodesMask & newCpsr; NextThumb = ((CPSR)newCpsr).t; NextJazelle = ((CPSR)newCpsr).j; + ForcedItState = ((((CPSR)tempSpsr).it2 << 2) & 0xFC) + | (((CPSR)tempSpsr).it1 & 0x3); CondCodes = CondCodesMask & newCpsr; ''' self.codeBlobs["memacc_code"] = accCode diff -r b2c7e56572a4 -r ee6641d7c713 src/arch/arm/isa/insts/macromem.isa --- a/src/arch/arm/isa/insts/macromem.isa Tue Jan 18 16:30:05 2011 -0600 +++ b/src/arch/arm/isa/insts/macromem.isa Tue Jan 18 16:30:05 2011 -0600 @@ -94,6 +94,8 @@ Cpsr = ~CondCodesMask & newCpsr; CondCodes = CondCodesMask & newCpsr; IWNPC = cSwap(Mem.uw, cpsr.e) | ((Spsr & 0x20) ? 1 : 0); + ForcedItState = ((((CPSR)Spsr).it2 << 2) & 0xFC) + | (((CPSR)Spsr).it1 & 0x3); ''' microLdrRetUopIop = InstObjParams('ldr_ret_uop', 'MicroLdrRetUop', 'MicroMemOp', diff -r b2c7e56572a4 -r ee6641d7c713 src/arch/arm/isa/insts/misc.isa --- a/src/arch/arm/isa/insts/misc.isa Tue Jan 18 16:30:05 2011 -0600 +++ b/src/arch/arm/isa/insts/misc.isa Tue Jan 18 16:30:05 2011 -0600 @@ -85,6 +85,8 @@ Cpsr = ~CondCodesMask & newCpsr; NextThumb = ((CPSR)newCpsr).t; NextJazelle = ((CPSR)newCpsr).j; + ForcedItState = ((((CPSR)Op1).it2 << 2) & 0xFC) + | (((CPSR)Op1).it1 & 0x3); CondCodes = CondCodesMask & newCpsr; ''' msrCpsrRegIop = InstObjParams("msr", "MsrCpsrReg", "MsrRegOp", @@ -111,6 +113,8 @@ Cpsr = ~CondCodesMask & newCpsr; NextThumb = ((CPSR)newCpsr).t; NextJazelle = ((CPSR)newCpsr).j; + ForcedItState = ((((CPSR)imm).it2 << 2) & 0xFC) + | (((CPSR)imm).it1 & 0x3); CondCodes = CondCodesMask & newCpsr; ''' msrCpsrImmIop = InstObjParams("msr", "MsrCpsrImm", "MsrImmOp", diff -r b2c7e56572a4 -r ee6641d7c713 src/arch/arm/isa/operands.isa --- a/src/arch/arm/isa/operands.isa Tue Jan 18 16:30:05 2011 -0600 +++ b/src/arch/arm/isa/operands.isa Tue Jan 18 16:30:05 2011 -0600 @@ -245,6 +245,7 @@ 'Thumb': pcStateReg('thumb', srtPC), 'NextThumb': pcStateReg('nextThumb', srtMode), 'NextJazelle': pcStateReg('nextJazelle', srtMode), + 'ForcedItState': pcStateReg('forcedItState', srtMode), #Register operands depending on a field in the instruction encoding. These #should be avoided since they may not be portable across different diff -r b2c7e56572a4 -r ee6641d7c713 src/arch/arm/predecoder.cc --- a/src/arch/arm/predecoder.cc Tue Jan 18 16:30:05 2011 -0600 +++ b/src/arch/arm/predecoder.cc Tue Jan 18 16:30:05 2011 -0600 @@ -162,9 +162,17 @@ FPSCR fpscr = tc->readMiscReg(MISCREG_FPSCR); emi.fpscrLen = fpscr.len; emi.fpscrStride = fpscr.stride; - CPSR cpsr = tc->readMiscReg(MISCREG_CPSR); - itstate.top6 = cpsr.it2; - itstate.bottom2 = cpsr.it1; + + if (pc.forcedItStateIsValid()) { + // returns from exceptions/interrupts force the it state. + itstate = pc.forcedItState(); + DPRINTF(Predecoder, "Predecoder, itstate forced = %08x.\n", pc.forcedItState()); + } else if (predAddrValid && (pc.instAddr() != predAddr)) { + // Control flow changes necessitate a 0 itstate. + itstate.top6 = 0; + itstate.bottom2 = 0; + } + outOfBytes = false; process(); } diff -r b2c7e56572a4 -r ee6641d7c713 src/arch/arm/predecoder.hh --- a/src/arch/arm/predecoder.hh Tue Jan 18 16:30:05 2011 -0600 +++ b/src/arch/arm/predecoder.hh Tue Jan 18 16:30:05 2011 -0600 @@ -67,6 +67,8 @@ bool outOfBytes; int offset; ITSTATE itstate; + Addr predAddr; + bool predAddrValid; public: void reset() @@ -76,6 +78,9 @@ emi = 0; emiReady = false; outOfBytes = true; + itstate = 0; + predAddr = 0; + predAddrValid = false; } Predecoder(ThreadContext * _tc) : @@ -138,6 +143,8 @@ assert(emiReady); ExtMachInst thisEmi = emi; pc.npc(pc.pc() + getInstSize()); + predAddrValid = true; + predAddr = pc.pc() + getInstSize(); emi = 0; emiReady = false; return thisEmi; diff -r b2c7e56572a4 -r ee6641d7c713 src/arch/arm/types.hh --- a/src/arch/arm/types.hh Tue Jan 18 16:30:05 2011 -0600 +++ b/src/arch/arm/types.hh Tue Jan 18 16:30:05 2011 -0600 @@ -202,9 +202,10 @@ }; uint8_t flags; uint8_t nextFlags; - + uint8_t forcedItStateValue; + bool forcedItStateValid; public: - PCState() : flags(0), nextFlags(0) + PCState() : flags(0), nextFlags(0), forcedItStateValue(0), forcedItStateValid(false) {} void @@ -214,7 +215,7 @@ npc(val + (thumb() ? 2 : 4)); } - PCState(Addr val) : flags(0), nextFlags(0) + PCState(Addr val) : flags(0), nextFlags(0), forcedItStateValue(0), forcedItStateValid(false) { set(val); } bool @@ -277,12 +278,40 @@ nextFlags &= ~JazelleBit; } + uint8_t + forcedItState() const + { + return forcedItStateValue; + } + + void + forcedItState(uint8_t value) + { + forcedItStateValue = value; + // Not valid unless the advance is called. + forcedItStateValid = false; + } + + bool + forcedItStateIsValid() const + { + return forcedItStateValid; + } + void advance() { Base::advance(); npc(pc() + (thumb() ? 2 : 4)); flags = nextFlags; + + // Validate the itState + if (forcedItStateValue != 0 && !forcedItStateValid) { + forcedItStateValid = true; + } else { + forcedItStateValid = false; + forcedItStateValue = 0; + } } void @@ -366,6 +395,8 @@ Base::serialize(os); SERIALIZE_SCALAR(flags); SERIALIZE_SCALAR(nextFlags); + SERIALIZE_SCALAR(forcedItStateValue); + SERIALIZE_SCALAR(forcedItStateValid); } void @@ -374,6 +405,8 @@ Base::unserialize(cp, section); UNSERIALIZE_SCALAR(flags); UNSERIALIZE_SCALAR(nextFlags); + UNSERIALIZE_SCALAR(forcedItStateValue); + UNSERIALIZE_SCALAR(forcedItStateValid); } }; _______________________________________________ m5-dev mailing list m5-dev@m5sim.org http://m5sim.org/mailman/listinfo/m5-dev