gem5/src/arch/arm/faults.cc

1458 lines
45 KiB
C++
Raw Normal View History

2009-04-06 03:53:15 +02:00
/*
* Copyright (c) 2010, 2012-2014, 2016 ARM Limited
* All rights reserved
*
* The license below extends only to copyright in the software and shall
* not be construed as granting a license to any other intellectual
* property including but not limited to intellectual property relating
* to a hardware implementation of the functionality of the software
* licensed hereunder. You may use the software subject to the license
* terms below provided that you ensure that this notice is replicated
* unmodified and in its entirety in all distributions of the software,
* modified or unmodified, in source code or in binary form.
*
2009-04-06 03:53:15 +02:00
* Copyright (c) 2003-2005 The Regents of The University of Michigan
* Copyright (c) 2007-2008 The Florida State University
* All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions are
* met: redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer;
* redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution;
* neither the name of the copyright holders nor the names of its
* contributors may be used to endorse or promote products derived from
* this software without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
* "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
* LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
* A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
* OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
* SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
* LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
* DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
* THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
* (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
* OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
*
* Authors: Ali Saidi
* Gabe Black
* Giacomo Gabrielli
* Thomas Grocutt
2009-04-06 03:53:15 +02:00
*/
#include "arch/arm/faults.hh"
#include "arch/arm/insts/static_inst.hh"
#include "arch/arm/system.hh"
#include "arch/arm/utility.hh"
#include "base/compiler.hh"
2009-04-06 03:53:15 +02:00
#include "base/trace.hh"
2011-04-15 19:44:06 +02:00
#include "cpu/base.hh"
#include "cpu/thread_context.hh"
#include "debug/Faults.hh"
#include "sim/full_system.hh"
2009-04-06 03:53:15 +02:00
namespace ArmISA
{
uint8_t ArmFault::shortDescFaultSources[] = {
0x01, // AlignmentFault
0x04, // InstructionCacheMaintenance
0xff, // SynchExtAbtOnTranslTableWalkL0 (INVALID)
0x0c, // SynchExtAbtOnTranslTableWalkL1
0x0e, // SynchExtAbtOnTranslTableWalkL2
0xff, // SynchExtAbtOnTranslTableWalkL3 (INVALID)
0xff, // SynchPtyErrOnTranslTableWalkL0 (INVALID)
0x1c, // SynchPtyErrOnTranslTableWalkL1
0x1e, // SynchPtyErrOnTranslTableWalkL2
0xff, // SynchPtyErrOnTranslTableWalkL3 (INVALID)
0xff, // TranslationL0 (INVALID)
0x05, // TranslationL1
0x07, // TranslationL2
0xff, // TranslationL3 (INVALID)
0xff, // AccessFlagL0 (INVALID)
0x03, // AccessFlagL1
0x06, // AccessFlagL2
0xff, // AccessFlagL3 (INVALID)
0xff, // DomainL0 (INVALID)
0x09, // DomainL1
0x0b, // DomainL2
0xff, // DomainL3 (INVALID)
0xff, // PermissionL0 (INVALID)
0x0d, // PermissionL1
0x0f, // PermissionL2
0xff, // PermissionL3 (INVALID)
0x02, // DebugEvent
0x08, // SynchronousExternalAbort
0x10, // TLBConflictAbort
0x19, // SynchPtyErrOnMemoryAccess
0x16, // AsynchronousExternalAbort
0x18, // AsynchPtyErrOnMemoryAccess
0xff, // AddressSizeL0 (INVALID)
0xff, // AddressSizeL1 (INVALID)
0xff, // AddressSizeL2 (INVALID)
0xff, // AddressSizeL3 (INVALID)
0x40, // PrefetchTLBMiss
0x80 // PrefetchUncacheable
};
2009-04-06 03:53:15 +02:00
static_assert(sizeof(ArmFault::shortDescFaultSources) ==
ArmFault::NumFaultSources,
"Invalid size of ArmFault::shortDescFaultSources[]");
2009-04-06 03:53:15 +02:00
uint8_t ArmFault::longDescFaultSources[] = {
0x21, // AlignmentFault
0xff, // InstructionCacheMaintenance (INVALID)
0xff, // SynchExtAbtOnTranslTableWalkL0 (INVALID)
0x15, // SynchExtAbtOnTranslTableWalkL1
0x16, // SynchExtAbtOnTranslTableWalkL2
0x17, // SynchExtAbtOnTranslTableWalkL3
0xff, // SynchPtyErrOnTranslTableWalkL0 (INVALID)
0x1d, // SynchPtyErrOnTranslTableWalkL1
0x1e, // SynchPtyErrOnTranslTableWalkL2
0x1f, // SynchPtyErrOnTranslTableWalkL3
0xff, // TranslationL0 (INVALID)
0x05, // TranslationL1
0x06, // TranslationL2
0x07, // TranslationL3
0xff, // AccessFlagL0 (INVALID)
0x09, // AccessFlagL1
0x0a, // AccessFlagL2
0x0b, // AccessFlagL3
0xff, // DomainL0 (INVALID)
0x3d, // DomainL1
0x3e, // DomainL2
0xff, // DomainL3 (RESERVED)
0xff, // PermissionL0 (INVALID)
0x0d, // PermissionL1
0x0e, // PermissionL2
0x0f, // PermissionL3
0x22, // DebugEvent
0x10, // SynchronousExternalAbort
0x30, // TLBConflictAbort
0x18, // SynchPtyErrOnMemoryAccess
0x11, // AsynchronousExternalAbort
0x19, // AsynchPtyErrOnMemoryAccess
0xff, // AddressSizeL0 (INVALID)
0xff, // AddressSizeL1 (INVALID)
0xff, // AddressSizeL2 (INVALID)
0xff, // AddressSizeL3 (INVALID)
0x40, // PrefetchTLBMiss
0x80 // PrefetchUncacheable
};
2009-04-06 03:53:15 +02:00
static_assert(sizeof(ArmFault::longDescFaultSources) ==
ArmFault::NumFaultSources,
"Invalid size of ArmFault::longDescFaultSources[]");
2009-04-06 03:53:15 +02:00
uint8_t ArmFault::aarch64FaultSources[] = {
0x21, // AlignmentFault
0xff, // InstructionCacheMaintenance (INVALID)
0x14, // SynchExtAbtOnTranslTableWalkL0
0x15, // SynchExtAbtOnTranslTableWalkL1
0x16, // SynchExtAbtOnTranslTableWalkL2
0x17, // SynchExtAbtOnTranslTableWalkL3
0x1c, // SynchPtyErrOnTranslTableWalkL0
0x1d, // SynchPtyErrOnTranslTableWalkL1
0x1e, // SynchPtyErrOnTranslTableWalkL2
0x1f, // SynchPtyErrOnTranslTableWalkL3
0x04, // TranslationL0
0x05, // TranslationL1
0x06, // TranslationL2
0x07, // TranslationL3
0x08, // AccessFlagL0
0x09, // AccessFlagL1
0x0a, // AccessFlagL2
0x0b, // AccessFlagL3
// @todo: Section & Page Domain Fault in AArch64?
0xff, // DomainL0 (INVALID)
0xff, // DomainL1 (INVALID)
0xff, // DomainL2 (INVALID)
0xff, // DomainL3 (INVALID)
0x0c, // PermissionL0
0x0d, // PermissionL1
0x0e, // PermissionL2
0x0f, // PermissionL3
0xff, // DebugEvent (INVALID)
0x10, // SynchronousExternalAbort
0x30, // TLBConflictAbort
0x18, // SynchPtyErrOnMemoryAccess
0xff, // AsynchronousExternalAbort (INVALID)
0xff, // AsynchPtyErrOnMemoryAccess (INVALID)
0x00, // AddressSizeL0
0x01, // AddressSizeL1
0x02, // AddressSizeL2
0x03, // AddressSizeL3
0x40, // PrefetchTLBMiss
0x80 // PrefetchUncacheable
};
2009-04-06 03:53:15 +02:00
static_assert(sizeof(ArmFault::aarch64FaultSources) ==
ArmFault::NumFaultSources,
"Invalid size of ArmFault::aarch64FaultSources[]");
2009-04-06 03:53:15 +02:00
// Fields: name, offset, cur{ELT,ELH}Offset, lowerEL{64,32}Offset, next mode,
// {ARM, Thumb, ARM_ELR, Thumb_ELR} PC offset, hyp trap,
// {A, F} disable, class, stat
template<> ArmFault::FaultVals ArmFaultVals<Reset>::vals = {
// Some dummy values (the reset vector has an IMPLEMENTATION DEFINED
// location in AArch64)
"Reset", 0x000, 0x000, 0x000, 0x000, 0x000, MODE_SVC,
0, 0, 0, 0, false, true, true, EC_UNKNOWN, FaultStat()
};
template<> ArmFault::FaultVals ArmFaultVals<UndefinedInstruction>::vals = {
"Undefined Instruction", 0x004, 0x000, 0x200, 0x400, 0x600, MODE_UNDEFINED,
4, 2, 0, 0, true, false, false, EC_UNKNOWN, FaultStat()
};
template<> ArmFault::FaultVals ArmFaultVals<SupervisorCall>::vals = {
"Supervisor Call", 0x008, 0x000, 0x200, 0x400, 0x600, MODE_SVC,
4, 2, 4, 2, true, false, false, EC_SVC_TO_HYP, FaultStat()
};
template<> ArmFault::FaultVals ArmFaultVals<SecureMonitorCall>::vals = {
"Secure Monitor Call", 0x008, 0x000, 0x200, 0x400, 0x600, MODE_MON,
4, 4, 4, 4, false, true, true, EC_SMC_TO_HYP, FaultStat()
};
template<> ArmFault::FaultVals ArmFaultVals<HypervisorCall>::vals = {
"Hypervisor Call", 0x008, 0x000, 0x200, 0x400, 0x600, MODE_HYP,
4, 4, 4, 4, true, false, false, EC_HVC, FaultStat()
};
template<> ArmFault::FaultVals ArmFaultVals<PrefetchAbort>::vals = {
"Prefetch Abort", 0x00C, 0x000, 0x200, 0x400, 0x600, MODE_ABORT,
4, 4, 0, 0, true, true, false, EC_PREFETCH_ABORT_TO_HYP, FaultStat()
};
template<> ArmFault::FaultVals ArmFaultVals<DataAbort>::vals = {
"Data Abort", 0x010, 0x000, 0x200, 0x400, 0x600, MODE_ABORT,
8, 8, 0, 0, true, true, false, EC_DATA_ABORT_TO_HYP, FaultStat()
};
template<> ArmFault::FaultVals ArmFaultVals<VirtualDataAbort>::vals = {
"Virtual Data Abort", 0x010, 0x000, 0x200, 0x400, 0x600, MODE_ABORT,
8, 8, 0, 0, true, true, false, EC_INVALID, FaultStat()
};
template<> ArmFault::FaultVals ArmFaultVals<HypervisorTrap>::vals = {
// @todo: double check these values
"Hypervisor Trap", 0x014, 0x000, 0x200, 0x400, 0x600, MODE_HYP,
0, 0, 0, 0, false, false, false, EC_UNKNOWN, FaultStat()
};
template<> ArmFault::FaultVals ArmFaultVals<Interrupt>::vals = {
"IRQ", 0x018, 0x080, 0x280, 0x480, 0x680, MODE_IRQ,
4, 4, 0, 0, false, true, false, EC_UNKNOWN, FaultStat()
};
template<> ArmFault::FaultVals ArmFaultVals<VirtualInterrupt>::vals = {
"Virtual IRQ", 0x018, 0x080, 0x280, 0x480, 0x680, MODE_IRQ,
4, 4, 0, 0, false, true, false, EC_INVALID, FaultStat()
};
template<> ArmFault::FaultVals ArmFaultVals<FastInterrupt>::vals = {
"FIQ", 0x01C, 0x100, 0x300, 0x500, 0x700, MODE_FIQ,
4, 4, 0, 0, false, true, true, EC_UNKNOWN, FaultStat()
};
template<> ArmFault::FaultVals ArmFaultVals<VirtualFastInterrupt>::vals = {
"Virtual FIQ", 0x01C, 0x100, 0x300, 0x500, 0x700, MODE_FIQ,
4, 4, 0, 0, false, true, true, EC_INVALID, FaultStat()
};
template<> ArmFault::FaultVals ArmFaultVals<SupervisorTrap>::vals = {
// Some dummy values (SupervisorTrap is AArch64-only)
"Supervisor Trap", 0x014, 0x000, 0x200, 0x400, 0x600, MODE_SVC,
0, 0, 0, 0, false, false, false, EC_UNKNOWN, FaultStat()
};
template<> ArmFault::FaultVals ArmFaultVals<SecureMonitorTrap>::vals = {
// Some dummy values (SecureMonitorTrap is AArch64-only)
"Secure Monitor Trap", 0x014, 0x000, 0x200, 0x400, 0x600, MODE_MON,
0, 0, 0, 0, false, false, false, EC_UNKNOWN, FaultStat()
};
template<> ArmFault::FaultVals ArmFaultVals<PCAlignmentFault>::vals = {
// Some dummy values (PCAlignmentFault is AArch64-only)
"PC Alignment Fault", 0x000, 0x000, 0x200, 0x400, 0x600, MODE_SVC,
0, 0, 0, 0, true, false, false, EC_PC_ALIGNMENT, FaultStat()
};
template<> ArmFault::FaultVals ArmFaultVals<SPAlignmentFault>::vals = {
// Some dummy values (SPAlignmentFault is AArch64-only)
"SP Alignment Fault", 0x000, 0x000, 0x200, 0x400, 0x600, MODE_SVC,
0, 0, 0, 0, true, false, false, EC_STACK_PTR_ALIGNMENT, FaultStat()
};
template<> ArmFault::FaultVals ArmFaultVals<SystemError>::vals = {
// Some dummy values (SError is AArch64-only)
"SError", 0x000, 0x180, 0x380, 0x580, 0x780, MODE_SVC,
0, 0, 0, 0, false, true, true, EC_SERROR, FaultStat()
};
template<> ArmFault::FaultVals ArmFaultVals<FlushPipe>::vals = {
// Some dummy values
"Pipe Flush", 0x000, 0x000, 0x000, 0x000, 0x000, MODE_SVC,
0, 0, 0, 0, false, true, true, EC_UNKNOWN, FaultStat()
};
template<> ArmFault::FaultVals ArmFaultVals<ArmSev>::vals = {
// Some dummy values
"ArmSev Flush", 0x000, 0x000, 0x000, 0x000, 0x000, MODE_SVC,
0, 0, 0, 0, false, true, true, EC_UNKNOWN, FaultStat()
};
template<> ArmFault::FaultVals ArmFaultVals<IllegalInstSetStateFault>::vals = {
// Some dummy values (SPAlignmentFault is AArch64-only)
"Illegal Inst Set State Fault", 0x000, 0x000, 0x200, 0x400, 0x600, MODE_SVC,
0, 0, 0, 0, true, false, false, EC_ILLEGAL_INST, FaultStat()
};
2009-04-06 03:53:15 +02:00
Addr
ArmFault::getVector(ThreadContext *tc)
2009-04-06 03:53:15 +02:00
{
Addr base;
2009-04-06 03:53:15 +02:00
// ARM ARM issue C B1.8.1
bool haveSecurity = ArmSystem::haveSecurity(tc);
2010-06-02 19:58:16 +02:00
// panic if SCTLR.VE because I have no idea what to do with vectored
// interrupts
SCTLR sctlr = tc->readMiscReg(MISCREG_SCTLR);
assert(!sctlr.ve);
// Check for invalid modes
CPSR cpsr = tc->readMiscRegNoEffect(MISCREG_CPSR);
assert(haveSecurity || cpsr.mode != MODE_MON);
assert(ArmSystem::haveVirtualization(tc) || cpsr.mode != MODE_HYP);
2010-06-02 19:58:16 +02:00
switch (cpsr.mode)
{
case MODE_MON:
base = tc->readMiscReg(MISCREG_MVBAR);
break;
case MODE_HYP:
base = tc->readMiscReg(MISCREG_HVBAR);
break;
default:
if (sctlr.v) {
base = HighVecs;
} else {
base = haveSecurity ? tc->readMiscReg(MISCREG_VBAR) : 0;
}
break;
}
return base + offset(tc);
}
2009-04-06 03:53:15 +02:00
Addr
ArmFault::getVector64(ThreadContext *tc)
{
Addr vbar;
switch (toEL) {
case EL3:
assert(ArmSystem::haveSecurity(tc));
vbar = tc->readMiscReg(MISCREG_VBAR_EL3);
break;
case EL2:
assert(ArmSystem::haveVirtualization(tc));
vbar = tc->readMiscReg(MISCREG_VBAR_EL2);
break;
case EL1:
vbar = tc->readMiscReg(MISCREG_VBAR_EL1);
break;
default:
panic("Invalid target exception level");
break;
}
return vbar + offset64();
2009-04-06 03:53:15 +02:00
}
MiscRegIndex
ArmFault::getSyndromeReg64() const
{
switch (toEL) {
case EL1:
return MISCREG_ESR_EL1;
case EL2:
return MISCREG_ESR_EL2;
case EL3:
return MISCREG_ESR_EL3;
default:
panic("Invalid exception level");
break;
}
}
MiscRegIndex
ArmFault::getFaultAddrReg64() const
{
switch (toEL) {
case EL1:
return MISCREG_FAR_EL1;
case EL2:
return MISCREG_FAR_EL2;
case EL3:
return MISCREG_FAR_EL3;
default:
panic("Invalid exception level");
break;
}
}
void
ArmFault::setSyndrome(ThreadContext *tc, MiscRegIndex syndrome_reg)
{
uint32_t value;
uint32_t exc_class = (uint32_t) ec(tc);
uint32_t issVal = iss();
assert(!from64 || ArmSystem::highestELIs64(tc));
value = exc_class << 26;
// HSR.IL not valid for Prefetch Aborts (0x20, 0x21) and Data Aborts (0x24,
// 0x25) for which the ISS information is not valid (ARMv7).
// @todo: ARMv8 revises AArch32 functionality: when HSR.IL is not
// valid it is treated as RES1.
if (to64) {
value |= 1 << 25;
} else if ((bits(exc_class, 5, 3) != 4) ||
(bits(exc_class, 2) && bits(issVal, 24))) {
if (!machInst.thumb || machInst.bigThumb)
value |= 1 << 25;
}
// Condition code valid for EC[5:4] nonzero
if (!from64 && ((bits(exc_class, 5, 4) == 0) &&
(bits(exc_class, 3, 0) != 0))) {
if (!machInst.thumb) {
uint32_t cond;
ConditionCode condCode = (ConditionCode) (uint32_t) machInst.condCode;
// If its on unconditional instruction report with a cond code of
// 0xE, ie the unconditional code
cond = (condCode == COND_UC) ? COND_AL : condCode;
value |= cond << 20;
value |= 1 << 24;
}
value |= bits(issVal, 19, 0);
} else {
value |= issVal;
}
tc->setMiscReg(syndrome_reg, value);
}
void
ArmFault::invoke(ThreadContext *tc, const StaticInstPtr &inst)
{
CPSR cpsr = tc->readMiscReg(MISCREG_CPSR);
if (ArmSystem::highestELIs64(tc)) { // ARMv8
// Determine source exception level and mode
fromMode = (OperatingMode) (uint8_t) cpsr.mode;
fromEL = opModeToEL(fromMode);
if (opModeIs64(fromMode))
from64 = true;
// Determine target exception level
if (ArmSystem::haveSecurity(tc) && routeToMonitor(tc))
toEL = EL3;
else if (ArmSystem::haveVirtualization(tc) && routeToHyp(tc))
toEL = EL2;
else
toEL = opModeToEL(nextMode());
if (fromEL > toEL)
toEL = fromEL;
if (toEL == ArmSystem::highestEL(tc) || ELIs64(tc, toEL)) {
// Invoke exception handler in AArch64 state
to64 = true;
invoke64(tc, inst);
return;
}
}
// ARMv7 (ARM ARM issue C B1.9)
bool have_security = ArmSystem::haveSecurity(tc);
bool have_virtualization = ArmSystem::haveVirtualization(tc);
FaultBase::invoke(tc);
if (!FullSystem)
return;
countStat()++;
SCTLR sctlr = tc->readMiscReg(MISCREG_SCTLR);
SCR scr = tc->readMiscReg(MISCREG_SCR);
CPSR saved_cpsr = tc->readMiscReg(MISCREG_CPSR);
saved_cpsr.nz = tc->readCCReg(CCREG_NZ);
saved_cpsr.c = tc->readCCReg(CCREG_C);
saved_cpsr.v = tc->readCCReg(CCREG_V);
saved_cpsr.ge = tc->readCCReg(CCREG_GE);
ISA,CPU,etc: Create an ISA defined PC type that abstracts out ISA behaviors. This change is a low level and pervasive reorganization of how PCs are managed in M5. Back when Alpha was the only ISA, there were only 2 PCs to worry about, the PC and the NPC, and the lsb of the PC signaled whether or not you were in PAL mode. As other ISAs were added, we had to add an NNPC, micro PC and next micropc, x86 and ARM introduced variable length instruction sets, and ARM started to keep track of mode bits in the PC. Each CPU model handled PCs in its own custom way that needed to be updated individually to handle the new dimensions of variability, or, in the case of ARMs mode-bit-in-the-pc hack, the complexity could be hidden in the ISA at the ISA implementation's expense. Areas like the branch predictor hadn't been updated to handle branch delay slots or micropcs, and it turns out that had introduced a significant (10s of percent) performance bug in SPARC and to a lesser extend MIPS. Rather than perpetuate the problem by reworking O3 again to handle the PC features needed by x86, this change was introduced to rework PC handling in a more modular, transparent, and hopefully efficient way. PC type: Rather than having the superset of all possible elements of PC state declared in each of the CPU models, each ISA defines its own PCState type which has exactly the elements it needs. A cross product of canned PCState classes are defined in the new "generic" ISA directory for ISAs with/without delay slots and microcode. These are either typedef-ed or subclassed by each ISA. To read or write this structure through a *Context, you use the new pcState() accessor which reads or writes depending on whether it has an argument. If you just want the address of the current or next instruction or the current micro PC, you can get those through read-only accessors on either the PCState type or the *Contexts. These are instAddr(), nextInstAddr(), and microPC(). Note the move away from readPC. That name is ambiguous since it's not clear whether or not it should be the actual address to fetch from, or if it should have extra bits in it like the PAL mode bit. Each class is free to define its own functions to get at whatever values it needs however it needs to to be used in ISA specific code. Eventually Alpha's PAL mode bit could be moved out of the PC and into a separate field like ARM. These types can be reset to a particular pc (where npc = pc + sizeof(MachInst), nnpc = npc + sizeof(MachInst), upc = 0, nupc = 1 as appropriate), printed, serialized, and compared. There is a branching() function which encapsulates code in the CPU models that checked if an instruction branched or not. Exactly what that means in the context of branch delay slots which can skip an instruction when not taken is ambiguous, and ideally this function and its uses can be eliminated. PCStates also generally know how to advance themselves in various ways depending on if they point at an instruction, a microop, or the last microop of a macroop. More on that later. Ideally, accessing all the PCs at once when setting them will improve performance of M5 even though more data needs to be moved around. This is because often all the PCs need to be manipulated together, and by getting them all at once you avoid multiple function calls. Also, the PCs of a particular thread will have spatial locality in the cache. Previously they were grouped by element in arrays which spread out accesses. Advancing the PC: The PCs were previously managed entirely by the CPU which had to know about PC semantics, try to figure out which dimension to increment the PC in, what to set NPC/NNPC, etc. These decisions are best left to the ISA in conjunction with the PC type itself. Because most of the information about how to increment the PC (mainly what type of instruction it refers to) is contained in the instruction object, a new advancePC virtual function was added to the StaticInst class. Subclasses provide an implementation that moves around the right element of the PC with a minimal amount of decision making. In ISAs like Alpha, the instructions always simply assign NPC to PC without having to worry about micropcs, nnpcs, etc. The added cost of a virtual function call should be outweighed by not having to figure out as much about what to do with the PCs and mucking around with the extra elements. One drawback of making the StaticInsts advance the PC is that you have to actually have one to advance the PC. This would, superficially, seem to require decoding an instruction before fetch could advance. This is, as far as I can tell, realistic. fetch would advance through memory addresses, not PCs, perhaps predicting new memory addresses using existing ones. More sophisticated decisions about control flow would be made later on, after the instruction was decoded, and handed back to fetch. If branching needs to happen, some amount of decoding needs to happen to see that it's a branch, what the target is, etc. This could get a little more complicated if that gets done by the predecoder, but I'm choosing to ignore that for now. Variable length instructions: To handle variable length instructions in x86 and ARM, the predecoder now takes in the current PC by reference to the getExtMachInst function. It can modify the PC however it needs to (by setting NPC to be the PC + instruction length, for instance). This could be improved since the CPU doesn't know if the PC was modified and always has to write it back. ISA parser: To support the new API, all PC related operand types were removed from the parser and replaced with a PCState type. There are two warts on this implementation. First, as with all the other operand types, the PCState still has to have a valid operand type even though it doesn't use it. Second, using syntax like PCS.npc(target) doesn't work for two reasons, this looks like the syntax for operand type overriding, and the parser can't figure out if you're reading or writing. Instructions that use the PCS operand (which I've consistently called it) need to first read it into a local variable, manipulate it, and then write it back out. Return address stack: The return address stack needed a little extra help because, in the presence of branch delay slots, it has to merge together elements of the return PC and the call PC. To handle that, a buildRetPC utility function was added. There are basically only two versions in all the ISAs, but it didn't seem short enough to put into the generic ISA directory. Also, the branch predictor code in O3 and InOrder were adjusted so that they always store the PC of the actual call instruction in the RAS, not the next PC. If the call instruction is a microop, the next PC refers to the next microop in the same macroop which is probably not desirable. The buildRetPC function advances the PC intelligently to the next macroop (in an ISA specific way) so that that case works. Change in stats: There were no change in stats except in MIPS and SPARC in the O3 model. MIPS runs in about 9% fewer ticks. SPARC runs with 30%-50% fewer ticks, which could likely be improved further by setting call/return instruction flags and taking advantage of the RAS. TODO: Add != operators to the PCState classes, defined trivially to be !(a==b). Smooth out places where PCs are split apart, passed around, and put back together later. I think this might happen in SPARC's fault code. Add ISA specific constructors that allow setting PC elements without calling a bunch of accessors. Try to eliminate the need for the branching() function. Factor out Alpha's PAL mode pc bit into a separate flag field, and eliminate places where it's blindly masked out or tested in the PC.
2010-10-31 08:07:20 +01:00
Addr curPc M5_VAR_USED = tc->pcState().pc();
ITSTATE it = tc->pcState().itstate();
saved_cpsr.it2 = it.top6;
saved_cpsr.it1 = it.bottom2;
// if we have a valid instruction then use it to annotate this fault with
// extra information. This is used to generate the correct fault syndrome
// information
if (inst) {
ArmStaticInst *armInst = reinterpret_cast<ArmStaticInst *>(inst.get());
armInst->annotateFault(this);
}
if (have_security && routeToMonitor(tc))
cpsr.mode = MODE_MON;
else if (have_virtualization && routeToHyp(tc))
cpsr.mode = MODE_HYP;
else
cpsr.mode = nextMode();
// Ensure Secure state if initially in Monitor mode
if (have_security && saved_cpsr.mode == MODE_MON) {
SCR scr = tc->readMiscRegNoEffect(MISCREG_SCR);
if (scr.ns) {
scr.ns = 0;
tc->setMiscRegNoEffect(MISCREG_SCR, scr);
}
}
// some bits are set differently if we have been routed to hyp mode
if (cpsr.mode == MODE_HYP) {
SCTLR hsctlr = tc->readMiscReg(MISCREG_HSCTLR);
cpsr.t = hsctlr.te;
cpsr.e = hsctlr.ee;
if (!scr.ea) {cpsr.a = 1;}
if (!scr.fiq) {cpsr.f = 1;}
if (!scr.irq) {cpsr.i = 1;}
} else if (cpsr.mode == MODE_MON) {
// Special case handling when entering monitor mode
cpsr.t = sctlr.te;
cpsr.e = sctlr.ee;
cpsr.a = 1;
cpsr.f = 1;
cpsr.i = 1;
} else {
cpsr.t = sctlr.te;
cpsr.e = sctlr.ee;
// The *Disable functions are virtual and different per fault
cpsr.a = cpsr.a | abortDisable(tc);
cpsr.f = cpsr.f | fiqDisable(tc);
cpsr.i = 1;
}
cpsr.it1 = cpsr.it2 = 0;
cpsr.j = 0;
tc->setMiscReg(MISCREG_CPSR, cpsr);
// Make sure mailbox sets to one always
tc->setMiscReg(MISCREG_SEV_MAILBOX, 1);
// Clear the exclusive monitor
tc->setMiscReg(MISCREG_LOCKFLAG, 0);
if (cpsr.mode == MODE_HYP) {
tc->setMiscReg(MISCREG_ELR_HYP, curPc +
(saved_cpsr.t ? thumbPcOffset(true) : armPcOffset(true)));
} else {
tc->setIntReg(INTREG_LR, curPc +
(saved_cpsr.t ? thumbPcOffset(false) : armPcOffset(false)));
}
switch (cpsr.mode) {
case MODE_FIQ:
tc->setMiscReg(MISCREG_SPSR_FIQ, saved_cpsr);
break;
case MODE_IRQ:
tc->setMiscReg(MISCREG_SPSR_IRQ, saved_cpsr);
break;
case MODE_SVC:
tc->setMiscReg(MISCREG_SPSR_SVC, saved_cpsr);
break;
case MODE_MON:
assert(have_security);
tc->setMiscReg(MISCREG_SPSR_MON, saved_cpsr);
break;
case MODE_ABORT:
tc->setMiscReg(MISCREG_SPSR_ABT, saved_cpsr);
break;
case MODE_UNDEFINED:
tc->setMiscReg(MISCREG_SPSR_UND, saved_cpsr);
if (ec(tc) != EC_UNKNOWN)
setSyndrome(tc, MISCREG_HSR);
break;
case MODE_HYP:
assert(have_virtualization);
tc->setMiscReg(MISCREG_SPSR_HYP, saved_cpsr);
setSyndrome(tc, MISCREG_HSR);
break;
default:
panic("unknown Mode\n");
}
ISA,CPU,etc: Create an ISA defined PC type that abstracts out ISA behaviors. This change is a low level and pervasive reorganization of how PCs are managed in M5. Back when Alpha was the only ISA, there were only 2 PCs to worry about, the PC and the NPC, and the lsb of the PC signaled whether or not you were in PAL mode. As other ISAs were added, we had to add an NNPC, micro PC and next micropc, x86 and ARM introduced variable length instruction sets, and ARM started to keep track of mode bits in the PC. Each CPU model handled PCs in its own custom way that needed to be updated individually to handle the new dimensions of variability, or, in the case of ARMs mode-bit-in-the-pc hack, the complexity could be hidden in the ISA at the ISA implementation's expense. Areas like the branch predictor hadn't been updated to handle branch delay slots or micropcs, and it turns out that had introduced a significant (10s of percent) performance bug in SPARC and to a lesser extend MIPS. Rather than perpetuate the problem by reworking O3 again to handle the PC features needed by x86, this change was introduced to rework PC handling in a more modular, transparent, and hopefully efficient way. PC type: Rather than having the superset of all possible elements of PC state declared in each of the CPU models, each ISA defines its own PCState type which has exactly the elements it needs. A cross product of canned PCState classes are defined in the new "generic" ISA directory for ISAs with/without delay slots and microcode. These are either typedef-ed or subclassed by each ISA. To read or write this structure through a *Context, you use the new pcState() accessor which reads or writes depending on whether it has an argument. If you just want the address of the current or next instruction or the current micro PC, you can get those through read-only accessors on either the PCState type or the *Contexts. These are instAddr(), nextInstAddr(), and microPC(). Note the move away from readPC. That name is ambiguous since it's not clear whether or not it should be the actual address to fetch from, or if it should have extra bits in it like the PAL mode bit. Each class is free to define its own functions to get at whatever values it needs however it needs to to be used in ISA specific code. Eventually Alpha's PAL mode bit could be moved out of the PC and into a separate field like ARM. These types can be reset to a particular pc (where npc = pc + sizeof(MachInst), nnpc = npc + sizeof(MachInst), upc = 0, nupc = 1 as appropriate), printed, serialized, and compared. There is a branching() function which encapsulates code in the CPU models that checked if an instruction branched or not. Exactly what that means in the context of branch delay slots which can skip an instruction when not taken is ambiguous, and ideally this function and its uses can be eliminated. PCStates also generally know how to advance themselves in various ways depending on if they point at an instruction, a microop, or the last microop of a macroop. More on that later. Ideally, accessing all the PCs at once when setting them will improve performance of M5 even though more data needs to be moved around. This is because often all the PCs need to be manipulated together, and by getting them all at once you avoid multiple function calls. Also, the PCs of a particular thread will have spatial locality in the cache. Previously they were grouped by element in arrays which spread out accesses. Advancing the PC: The PCs were previously managed entirely by the CPU which had to know about PC semantics, try to figure out which dimension to increment the PC in, what to set NPC/NNPC, etc. These decisions are best left to the ISA in conjunction with the PC type itself. Because most of the information about how to increment the PC (mainly what type of instruction it refers to) is contained in the instruction object, a new advancePC virtual function was added to the StaticInst class. Subclasses provide an implementation that moves around the right element of the PC with a minimal amount of decision making. In ISAs like Alpha, the instructions always simply assign NPC to PC without having to worry about micropcs, nnpcs, etc. The added cost of a virtual function call should be outweighed by not having to figure out as much about what to do with the PCs and mucking around with the extra elements. One drawback of making the StaticInsts advance the PC is that you have to actually have one to advance the PC. This would, superficially, seem to require decoding an instruction before fetch could advance. This is, as far as I can tell, realistic. fetch would advance through memory addresses, not PCs, perhaps predicting new memory addresses using existing ones. More sophisticated decisions about control flow would be made later on, after the instruction was decoded, and handed back to fetch. If branching needs to happen, some amount of decoding needs to happen to see that it's a branch, what the target is, etc. This could get a little more complicated if that gets done by the predecoder, but I'm choosing to ignore that for now. Variable length instructions: To handle variable length instructions in x86 and ARM, the predecoder now takes in the current PC by reference to the getExtMachInst function. It can modify the PC however it needs to (by setting NPC to be the PC + instruction length, for instance). This could be improved since the CPU doesn't know if the PC was modified and always has to write it back. ISA parser: To support the new API, all PC related operand types were removed from the parser and replaced with a PCState type. There are two warts on this implementation. First, as with all the other operand types, the PCState still has to have a valid operand type even though it doesn't use it. Second, using syntax like PCS.npc(target) doesn't work for two reasons, this looks like the syntax for operand type overriding, and the parser can't figure out if you're reading or writing. Instructions that use the PCS operand (which I've consistently called it) need to first read it into a local variable, manipulate it, and then write it back out. Return address stack: The return address stack needed a little extra help because, in the presence of branch delay slots, it has to merge together elements of the return PC and the call PC. To handle that, a buildRetPC utility function was added. There are basically only two versions in all the ISAs, but it didn't seem short enough to put into the generic ISA directory. Also, the branch predictor code in O3 and InOrder were adjusted so that they always store the PC of the actual call instruction in the RAS, not the next PC. If the call instruction is a microop, the next PC refers to the next microop in the same macroop which is probably not desirable. The buildRetPC function advances the PC intelligently to the next macroop (in an ISA specific way) so that that case works. Change in stats: There were no change in stats except in MIPS and SPARC in the O3 model. MIPS runs in about 9% fewer ticks. SPARC runs with 30%-50% fewer ticks, which could likely be improved further by setting call/return instruction flags and taking advantage of the RAS. TODO: Add != operators to the PCState classes, defined trivially to be !(a==b). Smooth out places where PCs are split apart, passed around, and put back together later. I think this might happen in SPARC's fault code. Add ISA specific constructors that allow setting PC elements without calling a bunch of accessors. Try to eliminate the need for the branching() function. Factor out Alpha's PAL mode pc bit into a separate flag field, and eliminate places where it's blindly masked out or tested in the PC.
2010-10-31 08:07:20 +01:00
Addr newPc = getVector(tc);
DPRINTF(Faults, "Invoking Fault:%s cpsr:%#x PC:%#x lr:%#x newVec: %#x\n",
ISA,CPU,etc: Create an ISA defined PC type that abstracts out ISA behaviors. This change is a low level and pervasive reorganization of how PCs are managed in M5. Back when Alpha was the only ISA, there were only 2 PCs to worry about, the PC and the NPC, and the lsb of the PC signaled whether or not you were in PAL mode. As other ISAs were added, we had to add an NNPC, micro PC and next micropc, x86 and ARM introduced variable length instruction sets, and ARM started to keep track of mode bits in the PC. Each CPU model handled PCs in its own custom way that needed to be updated individually to handle the new dimensions of variability, or, in the case of ARMs mode-bit-in-the-pc hack, the complexity could be hidden in the ISA at the ISA implementation's expense. Areas like the branch predictor hadn't been updated to handle branch delay slots or micropcs, and it turns out that had introduced a significant (10s of percent) performance bug in SPARC and to a lesser extend MIPS. Rather than perpetuate the problem by reworking O3 again to handle the PC features needed by x86, this change was introduced to rework PC handling in a more modular, transparent, and hopefully efficient way. PC type: Rather than having the superset of all possible elements of PC state declared in each of the CPU models, each ISA defines its own PCState type which has exactly the elements it needs. A cross product of canned PCState classes are defined in the new "generic" ISA directory for ISAs with/without delay slots and microcode. These are either typedef-ed or subclassed by each ISA. To read or write this structure through a *Context, you use the new pcState() accessor which reads or writes depending on whether it has an argument. If you just want the address of the current or next instruction or the current micro PC, you can get those through read-only accessors on either the PCState type or the *Contexts. These are instAddr(), nextInstAddr(), and microPC(). Note the move away from readPC. That name is ambiguous since it's not clear whether or not it should be the actual address to fetch from, or if it should have extra bits in it like the PAL mode bit. Each class is free to define its own functions to get at whatever values it needs however it needs to to be used in ISA specific code. Eventually Alpha's PAL mode bit could be moved out of the PC and into a separate field like ARM. These types can be reset to a particular pc (where npc = pc + sizeof(MachInst), nnpc = npc + sizeof(MachInst), upc = 0, nupc = 1 as appropriate), printed, serialized, and compared. There is a branching() function which encapsulates code in the CPU models that checked if an instruction branched or not. Exactly what that means in the context of branch delay slots which can skip an instruction when not taken is ambiguous, and ideally this function and its uses can be eliminated. PCStates also generally know how to advance themselves in various ways depending on if they point at an instruction, a microop, or the last microop of a macroop. More on that later. Ideally, accessing all the PCs at once when setting them will improve performance of M5 even though more data needs to be moved around. This is because often all the PCs need to be manipulated together, and by getting them all at once you avoid multiple function calls. Also, the PCs of a particular thread will have spatial locality in the cache. Previously they were grouped by element in arrays which spread out accesses. Advancing the PC: The PCs were previously managed entirely by the CPU which had to know about PC semantics, try to figure out which dimension to increment the PC in, what to set NPC/NNPC, etc. These decisions are best left to the ISA in conjunction with the PC type itself. Because most of the information about how to increment the PC (mainly what type of instruction it refers to) is contained in the instruction object, a new advancePC virtual function was added to the StaticInst class. Subclasses provide an implementation that moves around the right element of the PC with a minimal amount of decision making. In ISAs like Alpha, the instructions always simply assign NPC to PC without having to worry about micropcs, nnpcs, etc. The added cost of a virtual function call should be outweighed by not having to figure out as much about what to do with the PCs and mucking around with the extra elements. One drawback of making the StaticInsts advance the PC is that you have to actually have one to advance the PC. This would, superficially, seem to require decoding an instruction before fetch could advance. This is, as far as I can tell, realistic. fetch would advance through memory addresses, not PCs, perhaps predicting new memory addresses using existing ones. More sophisticated decisions about control flow would be made later on, after the instruction was decoded, and handed back to fetch. If branching needs to happen, some amount of decoding needs to happen to see that it's a branch, what the target is, etc. This could get a little more complicated if that gets done by the predecoder, but I'm choosing to ignore that for now. Variable length instructions: To handle variable length instructions in x86 and ARM, the predecoder now takes in the current PC by reference to the getExtMachInst function. It can modify the PC however it needs to (by setting NPC to be the PC + instruction length, for instance). This could be improved since the CPU doesn't know if the PC was modified and always has to write it back. ISA parser: To support the new API, all PC related operand types were removed from the parser and replaced with a PCState type. There are two warts on this implementation. First, as with all the other operand types, the PCState still has to have a valid operand type even though it doesn't use it. Second, using syntax like PCS.npc(target) doesn't work for two reasons, this looks like the syntax for operand type overriding, and the parser can't figure out if you're reading or writing. Instructions that use the PCS operand (which I've consistently called it) need to first read it into a local variable, manipulate it, and then write it back out. Return address stack: The return address stack needed a little extra help because, in the presence of branch delay slots, it has to merge together elements of the return PC and the call PC. To handle that, a buildRetPC utility function was added. There are basically only two versions in all the ISAs, but it didn't seem short enough to put into the generic ISA directory. Also, the branch predictor code in O3 and InOrder were adjusted so that they always store the PC of the actual call instruction in the RAS, not the next PC. If the call instruction is a microop, the next PC refers to the next microop in the same macroop which is probably not desirable. The buildRetPC function advances the PC intelligently to the next macroop (in an ISA specific way) so that that case works. Change in stats: There were no change in stats except in MIPS and SPARC in the O3 model. MIPS runs in about 9% fewer ticks. SPARC runs with 30%-50% fewer ticks, which could likely be improved further by setting call/return instruction flags and taking advantage of the RAS. TODO: Add != operators to the PCState classes, defined trivially to be !(a==b). Smooth out places where PCs are split apart, passed around, and put back together later. I think this might happen in SPARC's fault code. Add ISA specific constructors that allow setting PC elements without calling a bunch of accessors. Try to eliminate the need for the branching() function. Factor out Alpha's PAL mode pc bit into a separate flag field, and eliminate places where it's blindly masked out or tested in the PC.
2010-10-31 08:07:20 +01:00
name(), cpsr, curPc, tc->readIntReg(INTREG_LR), newPc);
PCState pc(newPc);
pc.thumb(cpsr.t);
pc.nextThumb(pc.thumb());
pc.jazelle(cpsr.j);
pc.nextJazelle(pc.jazelle());
pc.aarch64(!cpsr.width);
pc.nextAArch64(!cpsr.width);
tc->pcState(pc);
}
void
ArmFault::invoke64(ThreadContext *tc, const StaticInstPtr &inst)
{
// Determine actual misc. register indices for ELR_ELx and SPSR_ELx
MiscRegIndex elr_idx, spsr_idx;
switch (toEL) {
case EL1:
elr_idx = MISCREG_ELR_EL1;
spsr_idx = MISCREG_SPSR_EL1;
break;
case EL2:
assert(ArmSystem::haveVirtualization(tc));
elr_idx = MISCREG_ELR_EL2;
spsr_idx = MISCREG_SPSR_EL2;
break;
case EL3:
assert(ArmSystem::haveSecurity(tc));
elr_idx = MISCREG_ELR_EL3;
spsr_idx = MISCREG_SPSR_EL3;
break;
default:
panic("Invalid target exception level");
break;
}
// Save process state into SPSR_ELx
CPSR cpsr = tc->readMiscReg(MISCREG_CPSR);
CPSR spsr = cpsr;
spsr.nz = tc->readCCReg(CCREG_NZ);
spsr.c = tc->readCCReg(CCREG_C);
spsr.v = tc->readCCReg(CCREG_V);
if (from64) {
// Force some bitfields to 0
spsr.q = 0;
spsr.it1 = 0;
spsr.j = 0;
spsr.res0_23_22 = 0;
spsr.ge = 0;
spsr.it2 = 0;
spsr.t = 0;
} else {
spsr.ge = tc->readCCReg(CCREG_GE);
ITSTATE it = tc->pcState().itstate();
spsr.it2 = it.top6;
spsr.it1 = it.bottom2;
// Force some bitfields to 0
spsr.res0_23_22 = 0;
spsr.ss = 0;
}
tc->setMiscReg(spsr_idx, spsr);
// Save preferred return address into ELR_ELx
Addr curr_pc = tc->pcState().pc();
Addr ret_addr = curr_pc;
if (from64)
ret_addr += armPcElrOffset();
else
ret_addr += spsr.t ? thumbPcElrOffset() : armPcElrOffset();
tc->setMiscReg(elr_idx, ret_addr);
// Update process state
OperatingMode64 mode = 0;
mode.spX = 1;
mode.el = toEL;
mode.width = 0;
cpsr.mode = mode;
cpsr.daif = 0xf;
cpsr.il = 0;
cpsr.ss = 0;
tc->setMiscReg(MISCREG_CPSR, cpsr);
// Set PC to start of exception handler
Addr new_pc = purifyTaggedAddr(getVector64(tc), tc, toEL);
DPRINTF(Faults, "Invoking Fault (AArch64 target EL):%s cpsr:%#x PC:%#x "
"elr:%#x newVec: %#x\n", name(), cpsr, curr_pc, ret_addr, new_pc);
PCState pc(new_pc);
pc.aarch64(!cpsr.width);
pc.nextAArch64(!cpsr.width);
ISA,CPU,etc: Create an ISA defined PC type that abstracts out ISA behaviors. This change is a low level and pervasive reorganization of how PCs are managed in M5. Back when Alpha was the only ISA, there were only 2 PCs to worry about, the PC and the NPC, and the lsb of the PC signaled whether or not you were in PAL mode. As other ISAs were added, we had to add an NNPC, micro PC and next micropc, x86 and ARM introduced variable length instruction sets, and ARM started to keep track of mode bits in the PC. Each CPU model handled PCs in its own custom way that needed to be updated individually to handle the new dimensions of variability, or, in the case of ARMs mode-bit-in-the-pc hack, the complexity could be hidden in the ISA at the ISA implementation's expense. Areas like the branch predictor hadn't been updated to handle branch delay slots or micropcs, and it turns out that had introduced a significant (10s of percent) performance bug in SPARC and to a lesser extend MIPS. Rather than perpetuate the problem by reworking O3 again to handle the PC features needed by x86, this change was introduced to rework PC handling in a more modular, transparent, and hopefully efficient way. PC type: Rather than having the superset of all possible elements of PC state declared in each of the CPU models, each ISA defines its own PCState type which has exactly the elements it needs. A cross product of canned PCState classes are defined in the new "generic" ISA directory for ISAs with/without delay slots and microcode. These are either typedef-ed or subclassed by each ISA. To read or write this structure through a *Context, you use the new pcState() accessor which reads or writes depending on whether it has an argument. If you just want the address of the current or next instruction or the current micro PC, you can get those through read-only accessors on either the PCState type or the *Contexts. These are instAddr(), nextInstAddr(), and microPC(). Note the move away from readPC. That name is ambiguous since it's not clear whether or not it should be the actual address to fetch from, or if it should have extra bits in it like the PAL mode bit. Each class is free to define its own functions to get at whatever values it needs however it needs to to be used in ISA specific code. Eventually Alpha's PAL mode bit could be moved out of the PC and into a separate field like ARM. These types can be reset to a particular pc (where npc = pc + sizeof(MachInst), nnpc = npc + sizeof(MachInst), upc = 0, nupc = 1 as appropriate), printed, serialized, and compared. There is a branching() function which encapsulates code in the CPU models that checked if an instruction branched or not. Exactly what that means in the context of branch delay slots which can skip an instruction when not taken is ambiguous, and ideally this function and its uses can be eliminated. PCStates also generally know how to advance themselves in various ways depending on if they point at an instruction, a microop, or the last microop of a macroop. More on that later. Ideally, accessing all the PCs at once when setting them will improve performance of M5 even though more data needs to be moved around. This is because often all the PCs need to be manipulated together, and by getting them all at once you avoid multiple function calls. Also, the PCs of a particular thread will have spatial locality in the cache. Previously they were grouped by element in arrays which spread out accesses. Advancing the PC: The PCs were previously managed entirely by the CPU which had to know about PC semantics, try to figure out which dimension to increment the PC in, what to set NPC/NNPC, etc. These decisions are best left to the ISA in conjunction with the PC type itself. Because most of the information about how to increment the PC (mainly what type of instruction it refers to) is contained in the instruction object, a new advancePC virtual function was added to the StaticInst class. Subclasses provide an implementation that moves around the right element of the PC with a minimal amount of decision making. In ISAs like Alpha, the instructions always simply assign NPC to PC without having to worry about micropcs, nnpcs, etc. The added cost of a virtual function call should be outweighed by not having to figure out as much about what to do with the PCs and mucking around with the extra elements. One drawback of making the StaticInsts advance the PC is that you have to actually have one to advance the PC. This would, superficially, seem to require decoding an instruction before fetch could advance. This is, as far as I can tell, realistic. fetch would advance through memory addresses, not PCs, perhaps predicting new memory addresses using existing ones. More sophisticated decisions about control flow would be made later on, after the instruction was decoded, and handed back to fetch. If branching needs to happen, some amount of decoding needs to happen to see that it's a branch, what the target is, etc. This could get a little more complicated if that gets done by the predecoder, but I'm choosing to ignore that for now. Variable length instructions: To handle variable length instructions in x86 and ARM, the predecoder now takes in the current PC by reference to the getExtMachInst function. It can modify the PC however it needs to (by setting NPC to be the PC + instruction length, for instance). This could be improved since the CPU doesn't know if the PC was modified and always has to write it back. ISA parser: To support the new API, all PC related operand types were removed from the parser and replaced with a PCState type. There are two warts on this implementation. First, as with all the other operand types, the PCState still has to have a valid operand type even though it doesn't use it. Second, using syntax like PCS.npc(target) doesn't work for two reasons, this looks like the syntax for operand type overriding, and the parser can't figure out if you're reading or writing. Instructions that use the PCS operand (which I've consistently called it) need to first read it into a local variable, manipulate it, and then write it back out. Return address stack: The return address stack needed a little extra help because, in the presence of branch delay slots, it has to merge together elements of the return PC and the call PC. To handle that, a buildRetPC utility function was added. There are basically only two versions in all the ISAs, but it didn't seem short enough to put into the generic ISA directory. Also, the branch predictor code in O3 and InOrder were adjusted so that they always store the PC of the actual call instruction in the RAS, not the next PC. If the call instruction is a microop, the next PC refers to the next microop in the same macroop which is probably not desirable. The buildRetPC function advances the PC intelligently to the next macroop (in an ISA specific way) so that that case works. Change in stats: There were no change in stats except in MIPS and SPARC in the O3 model. MIPS runs in about 9% fewer ticks. SPARC runs with 30%-50% fewer ticks, which could likely be improved further by setting call/return instruction flags and taking advantage of the RAS. TODO: Add != operators to the PCState classes, defined trivially to be !(a==b). Smooth out places where PCs are split apart, passed around, and put back together later. I think this might happen in SPARC's fault code. Add ISA specific constructors that allow setting PC elements without calling a bunch of accessors. Try to eliminate the need for the branching() function. Factor out Alpha's PAL mode pc bit into a separate flag field, and eliminate places where it's blindly masked out or tested in the PC.
2010-10-31 08:07:20 +01:00
tc->pcState(pc);
// If we have a valid instruction then use it to annotate this fault with
// extra information. This is used to generate the correct fault syndrome
// information
if (inst)
reinterpret_cast<ArmStaticInst *>(inst.get())->annotateFault(this);
// Save exception syndrome
if ((nextMode() != MODE_IRQ) && (nextMode() != MODE_FIQ))
setSyndrome(tc, getSyndromeReg64());
2009-04-06 03:53:15 +02:00
}
2010-06-02 19:58:16 +02:00
void
Reset::invoke(ThreadContext *tc, const StaticInstPtr &inst)
2010-06-02 19:58:16 +02:00
{
if (FullSystem) {
tc->getCpuPtr()->clearInterrupts(tc->threadId());
tc->clearArchRegs();
}
if (!ArmSystem::highestELIs64(tc)) {
ArmFault::invoke(tc, inst);
tc->setMiscReg(MISCREG_VMPIDR,
getMPIDR(dynamic_cast<ArmSystem*>(tc->getSystemPtr()), tc));
// Unless we have SMC code to get us there, boot in HYP!
if (ArmSystem::haveVirtualization(tc) &&
!ArmSystem::haveSecurity(tc)) {
CPSR cpsr = tc->readMiscReg(MISCREG_CPSR);
cpsr.mode = MODE_HYP;
tc->setMiscReg(MISCREG_CPSR, cpsr);
}
} else {
// Advance the PC to the IMPLEMENTATION DEFINED reset value
PCState pc = ArmSystem::resetAddr64(tc);
pc.aarch64(true);
pc.nextAArch64(true);
tc->pcState(pc);
}
2010-06-02 19:58:16 +02:00
}
void
UndefinedInstruction::invoke(ThreadContext *tc, const StaticInstPtr &inst)
{
if (FullSystem) {
ArmFault::invoke(tc, inst);
return;
}
// If the mnemonic isn't defined this has to be an unknown instruction.
assert(unknown || mnemonic != NULL);
if (disabled) {
panic("Attempted to execute disabled instruction "
"'%s' (inst 0x%08x)", mnemonic, machInst);
} else if (unknown) {
panic("Attempted to execute unknown instruction (inst 0x%08x)",
machInst);
} else {
panic("Attempted to execute unimplemented instruction "
"'%s' (inst 0x%08x)", mnemonic, machInst);
}
}
bool
UndefinedInstruction::routeToHyp(ThreadContext *tc) const
{
bool toHyp;
SCR scr = tc->readMiscRegNoEffect(MISCREG_SCR);
HCR hcr = tc->readMiscRegNoEffect(MISCREG_HCR);
CPSR cpsr = tc->readMiscRegNoEffect(MISCREG_CPSR);
// if in Hyp mode then stay in Hyp mode
toHyp = scr.ns && (cpsr.mode == MODE_HYP);
// if HCR.TGE is set to 1, take to Hyp mode through Hyp Trap vector
toHyp |= !inSecureState(scr, cpsr) && hcr.tge && (cpsr.mode == MODE_USER);
return toHyp;
}
uint32_t
UndefinedInstruction::iss() const
{
if (overrideEc == EC_INVALID)
return issRaw;
uint32_t new_iss = 0;
uint32_t op0, op1, op2, CRn, CRm, Rt, dir;
dir = bits(machInst, 21, 21);
op0 = bits(machInst, 20, 19);
op1 = bits(machInst, 18, 16);
CRn = bits(machInst, 15, 12);
CRm = bits(machInst, 11, 8);
op2 = bits(machInst, 7, 5);
Rt = bits(machInst, 4, 0);
new_iss = op0 << 20 | op2 << 17 | op1 << 14 | CRn << 10 |
Rt << 5 | CRm << 1 | dir;
return new_iss;
}
void
SupervisorCall::invoke(ThreadContext *tc, const StaticInstPtr &inst)
{
if (FullSystem) {
ArmFault::invoke(tc, inst);
return;
}
// As of now, there isn't a 32 bit thumb version of this instruction.
assert(!machInst.bigThumb);
uint32_t callNum;
CPSR cpsr = tc->readMiscReg(MISCREG_CPSR);
OperatingMode mode = (OperatingMode)(uint8_t)cpsr.mode;
if (opModeIs64(mode))
callNum = tc->readIntReg(INTREG_X8);
else
callNum = tc->readIntReg(INTREG_R7);
syscall_emul: [patch 13/22] add system call retry capability This changeset adds functionality that allows system calls to retry without affecting thread context state such as the program counter or register values for the associated thread context (when system calls return with a retry fault). This functionality is needed to solve problems with blocking system calls in multi-process or multi-threaded simulations where information is passed between processes/threads. Blocking system calls can cause deadlock because the simulator itself is single threaded. There is only a single thread servicing the event queue which can cause deadlock if the thread hits a blocking system call instruction. To illustrate the problem, consider two processes using the producer/consumer sharing model. The processes can use file descriptors and the read and write calls to pass information to one another. If the consumer calls the blocking read system call before the producer has produced anything, the call will block the event queue (while executing the system call instruction) and deadlock the simulation. The solution implemented in this changeset is to recognize that the system calls will block and then generate a special retry fault. The fault will be sent back up through the function call chain until it is exposed to the cpu model's pipeline where the fault becomes visible. The fault will trigger the cpu model to replay the instruction at a future tick where the call has a chance to succeed without actually going into a blocking state. In subsequent patches, we recognize that a syscall will block by calling a non-blocking poll (from inside the system call implementation) and checking for events. When events show up during the poll, it signifies that the call would not have blocked and the syscall is allowed to proceed (calling an underlying host system call if necessary). If no events are returned from the poll, we generate the fault and try the instruction for the thread context at a distant tick. Note that retrying every tick is not efficient. As an aside, the simulator has some multi-threading support for the event queue, but it is not used by default and needs work. Even if the event queue was completely multi-threaded, meaning that there is a hardware thread on the host servicing a single simulator thread contexts with a 1:1 mapping between them, it's still possible to run into deadlock due to the event queue barriers on quantum boundaries. The solution of replaying at a later tick is the simplest solution and solves the problem generally.
2015-07-20 16:15:21 +02:00
Fault fault;
tc->syscall(callNum, &fault);
// Advance the PC since that won't happen automatically.
PCState pc = tc->pcState();
assert(inst);
inst->advancePC(pc);
tc->pcState(pc);
}
bool
SupervisorCall::routeToHyp(ThreadContext *tc) const
{
bool toHyp;
SCR scr = tc->readMiscRegNoEffect(MISCREG_SCR);
HCR hcr = tc->readMiscRegNoEffect(MISCREG_HCR);
CPSR cpsr = tc->readMiscRegNoEffect(MISCREG_CPSR);
// if in Hyp mode then stay in Hyp mode
toHyp = scr.ns && (cpsr.mode == MODE_HYP);
// if HCR.TGE is set to 1, take to Hyp mode through Hyp Trap vector
toHyp |= !inSecureState(scr, cpsr) && hcr.tge && (cpsr.mode == MODE_USER);
return toHyp;
}
ExceptionClass
SupervisorCall::ec(ThreadContext *tc) const
{
return (overrideEc != EC_INVALID) ? overrideEc :
(from64 ? EC_SVC_64 : vals.ec);
}
uint32_t
SupervisorCall::iss() const
{
// Even if we have a 24 bit imm from an arm32 instruction then we only use
// the bottom 16 bits for the ISS value (it doesn't hurt for AArch64 SVC).
return issRaw & 0xFFFF;
}
uint32_t
SecureMonitorCall::iss() const
{
if (from64)
return bits(machInst, 20, 5);
return 0;
}
ExceptionClass
UndefinedInstruction::ec(ThreadContext *tc) const
{
return (overrideEc != EC_INVALID) ? overrideEc : vals.ec;
}
HypervisorCall::HypervisorCall(ExtMachInst _machInst, uint32_t _imm) :
ArmFaultVals<HypervisorCall>(_machInst, _imm)
{}
ExceptionClass
HypervisorCall::ec(ThreadContext *tc) const
{
return from64 ? EC_HVC_64 : vals.ec;
}
ExceptionClass
HypervisorTrap::ec(ThreadContext *tc) const
{
return (overrideEc != EC_INVALID) ? overrideEc : vals.ec;
}
template<class T>
FaultOffset
ArmFaultVals<T>::offset(ThreadContext *tc)
{
bool isHypTrap = false;
// Normally we just use the exception vector from the table at the top if
// this file, however if this exception has caused a transition to hype
// mode, and its an exception type that would only do this if it has been
// trapped then we use the hyp trap vector instead of the normal vector
if (vals.hypTrappable) {
CPSR cpsr = tc->readMiscReg(MISCREG_CPSR);
if (cpsr.mode == MODE_HYP) {
CPSR spsr = tc->readMiscReg(MISCREG_SPSR_HYP);
isHypTrap = spsr.mode != MODE_HYP;
}
}
return isHypTrap ? 0x14 : vals.offset;
}
// void
// SupervisorCall::setSyndrome64(ThreadContext *tc, MiscRegIndex esr_idx)
// {
// ESR esr = 0;
// esr.ec = machInst.aarch64 ? SvcAArch64 : SvcAArch32;
// esr.il = !machInst.thumb;
// if (machInst.aarch64)
// esr.imm16 = bits(machInst.instBits, 20, 5);
// else if (machInst.thumb)
// esr.imm16 = bits(machInst.instBits, 7, 0);
// else
// esr.imm16 = bits(machInst.instBits, 15, 0);
// tc->setMiscReg(esr_idx, esr);
// }
void
SecureMonitorCall::invoke(ThreadContext *tc, const StaticInstPtr &inst)
{
if (FullSystem) {
ArmFault::invoke(tc, inst);
return;
}
}
ExceptionClass
SecureMonitorCall::ec(ThreadContext *tc) const
{
return (from64 ? EC_SMC_64 : vals.ec);
}
ExceptionClass
SupervisorTrap::ec(ThreadContext *tc) const
{
return (overrideEc != EC_INVALID) ? overrideEc : vals.ec;
}
ExceptionClass
SecureMonitorTrap::ec(ThreadContext *tc) const
{
return (overrideEc != EC_INVALID) ? overrideEc :
(from64 ? EC_SMC_64 : vals.ec);
}
template<class T>
void
AbortFault<T>::invoke(ThreadContext *tc, const StaticInstPtr &inst)
{
if (tranMethod == ArmFault::UnknownTran) {
tranMethod = longDescFormatInUse(tc) ? ArmFault::LpaeTran
: ArmFault::VmsaTran;
if ((tranMethod == ArmFault::VmsaTran) && this->routeToMonitor(tc)) {
// See ARM ARM B3-1416
bool override_LPAE = false;
TTBCR ttbcr_s = tc->readMiscReg(MISCREG_TTBCR_S);
TTBCR M5_VAR_USED ttbcr_ns = tc->readMiscReg(MISCREG_TTBCR_NS);
if (ttbcr_s.eae) {
override_LPAE = true;
} else {
// Unimplemented code option, not seen in testing. May need
// extension according to the manual exceprt above.
DPRINTF(Faults, "Warning: Incomplete translation method "
"override detected.\n");
}
if (override_LPAE)
tranMethod = ArmFault::LpaeTran;
}
}
if (source == ArmFault::AsynchronousExternalAbort) {
tc->getCpuPtr()->clearInterrupt(tc->threadId(), INT_ABT, 0);
}
// Get effective fault source encoding
CPSR cpsr = tc->readMiscReg(MISCREG_CPSR);
FSR fsr = getFsr(tc);
// source must be determined BEFORE invoking generic routines which will
// try to set hsr etc. and are based upon source!
ArmFaultVals<T>::invoke(tc, inst);
if (!this->to64) { // AArch32
if (cpsr.mode == MODE_HYP) {
tc->setMiscReg(T::HFarIndex, faultAddr);
} else if (stage2) {
tc->setMiscReg(MISCREG_HPFAR, (faultAddr >> 8) & ~0xf);
tc->setMiscReg(T::HFarIndex, OVAddr);
} else {
tc->setMiscReg(T::FsrIndex, fsr);
tc->setMiscReg(T::FarIndex, faultAddr);
}
DPRINTF(Faults, "Abort Fault source=%#x fsr=%#x faultAddr=%#x "\
"tranMethod=%#x\n", source, fsr, faultAddr, tranMethod);
} else { // AArch64
// Set the FAR register. Nothing else to do if we are in AArch64 state
// because the syndrome register has already been set inside invoke64()
if (stage2) {
// stage 2 fault, set HPFAR_EL2 to the faulting IPA
// and FAR_EL2 to the Original VA
tc->setMiscReg(AbortFault<T>::getFaultAddrReg64(), OVAddr);
tc->setMiscReg(MISCREG_HPFAR_EL2, bits(faultAddr, 47, 12) << 4);
DPRINTF(Faults, "Abort Fault (Stage 2) VA: 0x%x IPA: 0x%x\n",
OVAddr, faultAddr);
} else {
tc->setMiscReg(AbortFault<T>::getFaultAddrReg64(), faultAddr);
}
}
}
template<class T>
FSR
AbortFault<T>::getFsr(ThreadContext *tc)
{
FSR fsr = 0;
if (((CPSR) tc->readMiscRegNoEffect(MISCREG_CPSR)).width) {
// AArch32
assert(tranMethod != ArmFault::UnknownTran);
if (tranMethod == ArmFault::LpaeTran) {
srcEncoded = ArmFault::longDescFaultSources[source];
fsr.status = srcEncoded;
fsr.lpae = 1;
} else {
srcEncoded = ArmFault::shortDescFaultSources[source];
fsr.fsLow = bits(srcEncoded, 3, 0);
fsr.fsHigh = bits(srcEncoded, 4);
fsr.domain = static_cast<uint8_t>(domain);
}
fsr.wnr = (write ? 1 : 0);
fsr.ext = 0;
} else {
// AArch64
srcEncoded = ArmFault::aarch64FaultSources[source];
}
if (srcEncoded == ArmFault::FaultSourceInvalid) {
panic("Invalid fault source\n");
}
return fsr;
}
template<class T>
bool
AbortFault<T>::abortDisable(ThreadContext *tc)
{
if (ArmSystem::haveSecurity(tc)) {
SCR scr = tc->readMiscRegNoEffect(MISCREG_SCR);
return (!scr.ns || scr.aw);
}
return true;
}
template<class T>
void
AbortFault<T>::annotate(ArmFault::AnnotationIDs id, uint64_t val)
{
switch (id)
{
case ArmFault::S1PTW:
s1ptw = val;
break;
case ArmFault::OVA:
OVAddr = val;
break;
// Just ignore unknown ID's
default:
break;
}
}
template<class T>
uint32_t
AbortFault<T>::iss() const
{
uint32_t val;
val = srcEncoded & 0x3F;
val |= write << 6;
val |= s1ptw << 7;
return (val);
}
template<class T>
bool
AbortFault<T>::isMMUFault() const
{
// NOTE: Not relying on LL information being aligned to lowest bits here
return
(source == ArmFault::AlignmentFault) ||
((source >= ArmFault::TranslationLL) &&
(source < ArmFault::TranslationLL + 4)) ||
((source >= ArmFault::AccessFlagLL) &&
(source < ArmFault::AccessFlagLL + 4)) ||
((source >= ArmFault::DomainLL) &&
(source < ArmFault::DomainLL + 4)) ||
((source >= ArmFault::PermissionLL) &&
(source < ArmFault::PermissionLL + 4));
}
ExceptionClass
PrefetchAbort::ec(ThreadContext *tc) const
{
if (to64) {
// AArch64
if (toEL == fromEL)
return EC_PREFETCH_ABORT_CURR_EL;
else
return EC_PREFETCH_ABORT_LOWER_EL;
} else {
// AArch32
// Abort faults have different EC codes depending on whether
// the fault originated within HYP mode, or not. So override
// the method and add the extra adjustment of the EC value.
ExceptionClass ec = ArmFaultVals<PrefetchAbort>::vals.ec;
CPSR spsr = tc->readMiscReg(MISCREG_SPSR_HYP);
if (spsr.mode == MODE_HYP) {
ec = ((ExceptionClass) (((uint32_t) ec) + 1));
}
return ec;
}
}
bool
PrefetchAbort::routeToMonitor(ThreadContext *tc) const
{
SCR scr = 0;
if (from64)
scr = tc->readMiscRegNoEffect(MISCREG_SCR_EL3);
else
scr = tc->readMiscRegNoEffect(MISCREG_SCR);
return scr.ea && !isMMUFault();
}
bool
PrefetchAbort::routeToHyp(ThreadContext *tc) const
{
bool toHyp;
SCR scr = tc->readMiscRegNoEffect(MISCREG_SCR);
HCR hcr = tc->readMiscRegNoEffect(MISCREG_HCR);
CPSR cpsr = tc->readMiscRegNoEffect(MISCREG_CPSR);
HDCR hdcr = tc->readMiscRegNoEffect(MISCREG_HDCR);
// if in Hyp mode then stay in Hyp mode
toHyp = scr.ns && (cpsr.mode == MODE_HYP);
// otherwise, check whether to take to Hyp mode through Hyp Trap vector
toHyp |= (stage2 ||
( (source == DebugEvent) && hdcr.tde && (cpsr.mode != MODE_HYP)) ||
( (source == SynchronousExternalAbort) && hcr.tge && (cpsr.mode == MODE_USER))
) && !inSecureState(tc);
return toHyp;
}
ExceptionClass
DataAbort::ec(ThreadContext *tc) const
{
if (to64) {
// AArch64
if (source == ArmFault::AsynchronousExternalAbort) {
panic("Asynchronous External Abort should be handled with "
"SystemErrors (SErrors)!");
}
if (toEL == fromEL)
return EC_DATA_ABORT_CURR_EL;
else
return EC_DATA_ABORT_LOWER_EL;
} else {
// AArch32
// Abort faults have different EC codes depending on whether
// the fault originated within HYP mode, or not. So override
// the method and add the extra adjustment of the EC value.
ExceptionClass ec = ArmFaultVals<DataAbort>::vals.ec;
CPSR spsr = tc->readMiscReg(MISCREG_SPSR_HYP);
if (spsr.mode == MODE_HYP) {
ec = ((ExceptionClass) (((uint32_t) ec) + 1));
}
return ec;
}
}
bool
DataAbort::routeToMonitor(ThreadContext *tc) const
{
SCR scr = 0;
if (from64)
scr = tc->readMiscRegNoEffect(MISCREG_SCR_EL3);
else
scr = tc->readMiscRegNoEffect(MISCREG_SCR);
return scr.ea && !isMMUFault();
}
bool
DataAbort::routeToHyp(ThreadContext *tc) const
{
bool toHyp;
SCR scr = tc->readMiscRegNoEffect(MISCREG_SCR);
HCR hcr = tc->readMiscRegNoEffect(MISCREG_HCR);
CPSR cpsr = tc->readMiscRegNoEffect(MISCREG_CPSR);
HDCR hdcr = tc->readMiscRegNoEffect(MISCREG_HDCR);
// if in Hyp mode then stay in Hyp mode
toHyp = scr.ns && (cpsr.mode == MODE_HYP);
// otherwise, check whether to take to Hyp mode through Hyp Trap vector
toHyp |= (stage2 ||
( (cpsr.mode != MODE_HYP) && ( ((source == AsynchronousExternalAbort) && hcr.amo) ||
((source == DebugEvent) && hdcr.tde) )
) ||
( (cpsr.mode == MODE_USER) && hcr.tge &&
((source == AlignmentFault) ||
(source == SynchronousExternalAbort))
)
) && !inSecureState(tc);
return toHyp;
}
uint32_t
DataAbort::iss() const
{
uint32_t val;
// Add on the data abort specific fields to the generic abort ISS value
val = AbortFault<DataAbort>::iss();
// ISS is valid if not caused by a stage 1 page table walk, and when taken
// to AArch64 only when directed to EL2
if (!s1ptw && (!to64 || toEL == EL2)) {
val |= isv << 24;
if (isv) {
val |= sas << 22;
val |= sse << 21;
val |= srt << 16;
// AArch64 only. These assignments are safe on AArch32 as well
// because these vars are initialized to false
val |= sf << 15;
val |= ar << 14;
}
}
return (val);
}
void
DataAbort::annotate(AnnotationIDs id, uint64_t val)
{
AbortFault<DataAbort>::annotate(id, val);
switch (id)
{
case SAS:
isv = true;
sas = val;
break;
case SSE:
isv = true;
sse = val;
break;
case SRT:
isv = true;
srt = val;
break;
case SF:
isv = true;
sf = val;
break;
case AR:
isv = true;
ar = val;
break;
// Just ignore unknown ID's
default:
break;
}
}
void
VirtualDataAbort::invoke(ThreadContext *tc, const StaticInstPtr &inst)
{
AbortFault<VirtualDataAbort>::invoke(tc, inst);
HCR hcr = tc->readMiscRegNoEffect(MISCREG_HCR);
hcr.va = 0;
tc->setMiscRegNoEffect(MISCREG_HCR, hcr);
}
bool
Interrupt::routeToMonitor(ThreadContext *tc) const
{
assert(ArmSystem::haveSecurity(tc));
SCR scr = 0;
if (from64)
scr = tc->readMiscRegNoEffect(MISCREG_SCR_EL3);
else
scr = tc->readMiscRegNoEffect(MISCREG_SCR);
return scr.irq;
}
bool
Interrupt::routeToHyp(ThreadContext *tc) const
{
bool toHyp;
SCR scr = tc->readMiscRegNoEffect(MISCREG_SCR);
HCR hcr = tc->readMiscRegNoEffect(MISCREG_HCR);
CPSR cpsr = tc->readMiscRegNoEffect(MISCREG_CPSR);
// Determine whether IRQs are routed to Hyp mode.
toHyp = (!scr.irq && hcr.imo && !inSecureState(tc)) ||
(cpsr.mode == MODE_HYP);
return toHyp;
}
bool
Interrupt::abortDisable(ThreadContext *tc)
{
if (ArmSystem::haveSecurity(tc)) {
SCR scr = tc->readMiscRegNoEffect(MISCREG_SCR);
return (!scr.ns || scr.aw);
}
return true;
}
VirtualInterrupt::VirtualInterrupt()
{}
bool
FastInterrupt::routeToMonitor(ThreadContext *tc) const
{
assert(ArmSystem::haveSecurity(tc));
SCR scr = 0;
if (from64)
scr = tc->readMiscRegNoEffect(MISCREG_SCR_EL3);
else
scr = tc->readMiscRegNoEffect(MISCREG_SCR);
return scr.fiq;
}
bool
FastInterrupt::routeToHyp(ThreadContext *tc) const
{
bool toHyp;
SCR scr = tc->readMiscRegNoEffect(MISCREG_SCR);
HCR hcr = tc->readMiscRegNoEffect(MISCREG_HCR);
CPSR cpsr = tc->readMiscRegNoEffect(MISCREG_CPSR);
// Determine whether IRQs are routed to Hyp mode.
toHyp = (!scr.fiq && hcr.fmo && !inSecureState(tc)) ||
(cpsr.mode == MODE_HYP);
return toHyp;
}
bool
FastInterrupt::abortDisable(ThreadContext *tc)
{
if (ArmSystem::haveSecurity(tc)) {
SCR scr = tc->readMiscRegNoEffect(MISCREG_SCR);
return (!scr.ns || scr.aw);
}
return true;
}
bool
FastInterrupt::fiqDisable(ThreadContext *tc)
{
if (ArmSystem::haveVirtualization(tc)) {
return true;
} else if (ArmSystem::haveSecurity(tc)) {
SCR scr = tc->readMiscRegNoEffect(MISCREG_SCR);
return (!scr.ns || scr.fw);
}
return true;
}
VirtualFastInterrupt::VirtualFastInterrupt()
{}
void
PCAlignmentFault::invoke(ThreadContext *tc, const StaticInstPtr &inst)
{
ArmFaultVals<PCAlignmentFault>::invoke(tc, inst);
assert(from64);
// Set the FAR
tc->setMiscReg(getFaultAddrReg64(), faultPC);
}
SPAlignmentFault::SPAlignmentFault()
{}
SystemError::SystemError()
{}
void
SystemError::invoke(ThreadContext *tc, const StaticInstPtr &inst)
{
tc->getCpuPtr()->clearInterrupt(tc->threadId(), INT_ABT, 0);
ArmFault::invoke(tc, inst);
}
bool
SystemError::routeToMonitor(ThreadContext *tc) const
{
assert(ArmSystem::haveSecurity(tc));
assert(from64);
SCR scr = tc->readMiscRegNoEffect(MISCREG_SCR_EL3);
return scr.ea;
}
bool
SystemError::routeToHyp(ThreadContext *tc) const
{
bool toHyp;
assert(from64);
SCR scr = tc->readMiscRegNoEffect(MISCREG_SCR_EL3);
HCR hcr = tc->readMiscRegNoEffect(MISCREG_HCR);
toHyp = (!scr.ea && hcr.amo && !inSecureState(tc)) ||
(!scr.ea && !scr.rw && !hcr.amo && !inSecureState(tc));
return toHyp;
}
void
FlushPipe::invoke(ThreadContext *tc, const StaticInstPtr &inst) {
DPRINTF(Faults, "Invoking FlushPipe Fault\n");
// Set the PC to the next instruction of the faulting instruction.
// Net effect is simply squashing all instructions behind and
// start refetching from the next instruction.
ISA,CPU,etc: Create an ISA defined PC type that abstracts out ISA behaviors. This change is a low level and pervasive reorganization of how PCs are managed in M5. Back when Alpha was the only ISA, there were only 2 PCs to worry about, the PC and the NPC, and the lsb of the PC signaled whether or not you were in PAL mode. As other ISAs were added, we had to add an NNPC, micro PC and next micropc, x86 and ARM introduced variable length instruction sets, and ARM started to keep track of mode bits in the PC. Each CPU model handled PCs in its own custom way that needed to be updated individually to handle the new dimensions of variability, or, in the case of ARMs mode-bit-in-the-pc hack, the complexity could be hidden in the ISA at the ISA implementation's expense. Areas like the branch predictor hadn't been updated to handle branch delay slots or micropcs, and it turns out that had introduced a significant (10s of percent) performance bug in SPARC and to a lesser extend MIPS. Rather than perpetuate the problem by reworking O3 again to handle the PC features needed by x86, this change was introduced to rework PC handling in a more modular, transparent, and hopefully efficient way. PC type: Rather than having the superset of all possible elements of PC state declared in each of the CPU models, each ISA defines its own PCState type which has exactly the elements it needs. A cross product of canned PCState classes are defined in the new "generic" ISA directory for ISAs with/without delay slots and microcode. These are either typedef-ed or subclassed by each ISA. To read or write this structure through a *Context, you use the new pcState() accessor which reads or writes depending on whether it has an argument. If you just want the address of the current or next instruction or the current micro PC, you can get those through read-only accessors on either the PCState type or the *Contexts. These are instAddr(), nextInstAddr(), and microPC(). Note the move away from readPC. That name is ambiguous since it's not clear whether or not it should be the actual address to fetch from, or if it should have extra bits in it like the PAL mode bit. Each class is free to define its own functions to get at whatever values it needs however it needs to to be used in ISA specific code. Eventually Alpha's PAL mode bit could be moved out of the PC and into a separate field like ARM. These types can be reset to a particular pc (where npc = pc + sizeof(MachInst), nnpc = npc + sizeof(MachInst), upc = 0, nupc = 1 as appropriate), printed, serialized, and compared. There is a branching() function which encapsulates code in the CPU models that checked if an instruction branched or not. Exactly what that means in the context of branch delay slots which can skip an instruction when not taken is ambiguous, and ideally this function and its uses can be eliminated. PCStates also generally know how to advance themselves in various ways depending on if they point at an instruction, a microop, or the last microop of a macroop. More on that later. Ideally, accessing all the PCs at once when setting them will improve performance of M5 even though more data needs to be moved around. This is because often all the PCs need to be manipulated together, and by getting them all at once you avoid multiple function calls. Also, the PCs of a particular thread will have spatial locality in the cache. Previously they were grouped by element in arrays which spread out accesses. Advancing the PC: The PCs were previously managed entirely by the CPU which had to know about PC semantics, try to figure out which dimension to increment the PC in, what to set NPC/NNPC, etc. These decisions are best left to the ISA in conjunction with the PC type itself. Because most of the information about how to increment the PC (mainly what type of instruction it refers to) is contained in the instruction object, a new advancePC virtual function was added to the StaticInst class. Subclasses provide an implementation that moves around the right element of the PC with a minimal amount of decision making. In ISAs like Alpha, the instructions always simply assign NPC to PC without having to worry about micropcs, nnpcs, etc. The added cost of a virtual function call should be outweighed by not having to figure out as much about what to do with the PCs and mucking around with the extra elements. One drawback of making the StaticInsts advance the PC is that you have to actually have one to advance the PC. This would, superficially, seem to require decoding an instruction before fetch could advance. This is, as far as I can tell, realistic. fetch would advance through memory addresses, not PCs, perhaps predicting new memory addresses using existing ones. More sophisticated decisions about control flow would be made later on, after the instruction was decoded, and handed back to fetch. If branching needs to happen, some amount of decoding needs to happen to see that it's a branch, what the target is, etc. This could get a little more complicated if that gets done by the predecoder, but I'm choosing to ignore that for now. Variable length instructions: To handle variable length instructions in x86 and ARM, the predecoder now takes in the current PC by reference to the getExtMachInst function. It can modify the PC however it needs to (by setting NPC to be the PC + instruction length, for instance). This could be improved since the CPU doesn't know if the PC was modified and always has to write it back. ISA parser: To support the new API, all PC related operand types were removed from the parser and replaced with a PCState type. There are two warts on this implementation. First, as with all the other operand types, the PCState still has to have a valid operand type even though it doesn't use it. Second, using syntax like PCS.npc(target) doesn't work for two reasons, this looks like the syntax for operand type overriding, and the parser can't figure out if you're reading or writing. Instructions that use the PCS operand (which I've consistently called it) need to first read it into a local variable, manipulate it, and then write it back out. Return address stack: The return address stack needed a little extra help because, in the presence of branch delay slots, it has to merge together elements of the return PC and the call PC. To handle that, a buildRetPC utility function was added. There are basically only two versions in all the ISAs, but it didn't seem short enough to put into the generic ISA directory. Also, the branch predictor code in O3 and InOrder were adjusted so that they always store the PC of the actual call instruction in the RAS, not the next PC. If the call instruction is a microop, the next PC refers to the next microop in the same macroop which is probably not desirable. The buildRetPC function advances the PC intelligently to the next macroop (in an ISA specific way) so that that case works. Change in stats: There were no change in stats except in MIPS and SPARC in the O3 model. MIPS runs in about 9% fewer ticks. SPARC runs with 30%-50% fewer ticks, which could likely be improved further by setting call/return instruction flags and taking advantage of the RAS. TODO: Add != operators to the PCState classes, defined trivially to be !(a==b). Smooth out places where PCs are split apart, passed around, and put back together later. I think this might happen in SPARC's fault code. Add ISA specific constructors that allow setting PC elements without calling a bunch of accessors. Try to eliminate the need for the branching() function. Factor out Alpha's PAL mode pc bit into a separate flag field, and eliminate places where it's blindly masked out or tested in the PC.
2010-10-31 08:07:20 +01:00
PCState pc = tc->pcState();
assert(inst);
inst->advancePC(pc);
tc->pcState(pc);
}
void
ArmSev::invoke(ThreadContext *tc, const StaticInstPtr &inst) {
DPRINTF(Faults, "Invoking ArmSev Fault\n");
if (!FullSystem)
return;
// Set sev_mailbox to 1, clear the pending interrupt from remote
// SEV execution and let pipeline continue as pcState is still
// valid.
tc->setMiscReg(MISCREG_SEV_MAILBOX, 1);
tc->getCpuPtr()->clearInterrupt(tc->threadId(), INT_SEV, 0);
}
// Instantiate all the templates to make the linker happy
template class ArmFaultVals<Reset>;
template class ArmFaultVals<UndefinedInstruction>;
template class ArmFaultVals<SupervisorCall>;
template class ArmFaultVals<SecureMonitorCall>;
template class ArmFaultVals<HypervisorCall>;
template class ArmFaultVals<PrefetchAbort>;
template class ArmFaultVals<DataAbort>;
template class ArmFaultVals<VirtualDataAbort>;
template class ArmFaultVals<HypervisorTrap>;
template class ArmFaultVals<Interrupt>;
template class ArmFaultVals<VirtualInterrupt>;
template class ArmFaultVals<FastInterrupt>;
template class ArmFaultVals<VirtualFastInterrupt>;
template class ArmFaultVals<SupervisorTrap>;
template class ArmFaultVals<SecureMonitorTrap>;
template class ArmFaultVals<PCAlignmentFault>;
template class ArmFaultVals<SPAlignmentFault>;
template class ArmFaultVals<SystemError>;
template class ArmFaultVals<FlushPipe>;
template class ArmFaultVals<ArmSev>;
template class AbortFault<PrefetchAbort>;
template class AbortFault<DataAbort>;
template class AbortFault<VirtualDataAbort>;
IllegalInstSetStateFault::IllegalInstSetStateFault()
{}
2009-04-06 03:53:15 +02:00
} // namespace ArmISA