gem5/src/arch/arm/tlb.cc

/*
 * Copyright (c) 2010-2013, 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.
 *
 * Copyright (c) 2001-2005 The Regents of The University of Michigan
 * 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
 *          Nathan Binkert
 *          Steve Reinhardt
 */

#include "arch/arm/tlb.hh"

#include <memory>
#include <string>
#include <vector>

#include "arch/arm/faults.hh"
#include "arch/arm/pagetable.hh"
#include "arch/arm/system.hh"
#include "arch/arm/table_walker.hh"
#include "arch/arm/stage2_lookup.hh"
#include "arch/arm/stage2_mmu.hh"
#include "arch/arm/utility.hh"
#include "base/inifile.hh"
#include "base/str.hh"
#include "base/trace.hh"
#include "cpu/base.hh"
#include "cpu/thread_context.hh"
#include "debug/Checkpoint.hh"
#include "debug/TLB.hh"
#include "debug/TLBVerbose.hh"
#include "mem/page_table.hh"
#include "params/ArmTLB.hh"
#include "sim/full_system.hh"
#include "sim/process.hh"

using namespace std;
using namespace ArmISA;

TLB::TLB(const ArmTLBParams *p)
    : BaseTLB(p), table(new TlbEntry[p->size]), size(p->size),
      isStage2(p->is_stage2), stage2Req(false), _attr(0),
      directToStage2(false), tableWalker(p->walker), stage2Tlb(NULL),
      stage2Mmu(NULL), test(nullptr), rangeMRU(1),
      aarch64(false), aarch64EL(EL0), isPriv(false), isSecure(false),
      isHyp(false), asid(0), vmid(0), dacr(0),
      miscRegValid(false), miscRegContext(0), curTranType(NormalTran)
{
    tableWalker->setTlb(this);

    // Cache system-level properties
    haveLPAE = tableWalker->haveLPAE();
    haveVirtualization = tableWalker->haveVirtualization();
    haveLargeAsid64 = tableWalker->haveLargeAsid64();
}

TLB::~TLB()
{
    delete[] table;
}

void
TLB::init()
{
    if (stage2Mmu && !isStage2)
        stage2Tlb = stage2Mmu->stage2Tlb();
}

void
TLB::setMMU(Stage2MMU *m, MasterID master_id)
{
    stage2Mmu = m;
    tableWalker->setMMU(m, master_id);
}

bool
TLB::translateFunctional(ThreadContext *tc, Addr va, Addr &pa)
{
    updateMiscReg(tc);

    if (directToStage2) {
        assert(stage2Tlb);
        return stage2Tlb->translateFunctional(tc, va, pa);
    }

    TlbEntry *e = lookup(va, asid, vmid, isHyp, isSecure, true, false,
                         aarch64 ? aarch64EL : EL1);
    if (!e)
        return false;
    pa = e->pAddr(va);
    return true;
}

Fault
TLB::finalizePhysical(RequestPtr req, ThreadContext *tc, Mode mode) const
{
    return NoFault;
}

TlbEntry*
TLB::lookup(Addr va, uint16_t asn, uint8_t vmid, bool hyp, bool secure,
            bool functional, bool ignore_asn, uint8_t target_el)
{

    TlbEntry *retval = NULL;

    // Maintaining LRU array
    int x = 0;
    while (retval == NULL && x < size) {
        if ((!ignore_asn && table[x].match(va, asn, vmid, hyp, secure, false,
             target_el)) ||
            (ignore_asn && table[x].match(va, vmid, hyp, secure, target_el))) {
            // We only move the hit entry ahead when the position is higher
            // than rangeMRU
            if (x > rangeMRU && !functional) {
                TlbEntry tmp_entry = table[x];
                for (int i = x; i > 0; i--)
                    table[i] = table[i - 1];
                table[0] = tmp_entry;
                retval = &table[0];
            } else {
                retval = &table[x];
            }
            break;
        }
        ++x;
    }

    DPRINTF(TLBVerbose, "Lookup %#x, asn %#x -> %s vmn 0x%x hyp %d secure %d "
            "ppn %#x size: %#x pa: %#x ap:%d ns:%d nstid:%d g:%d asid: %d "
            "el: %d\n",
            va, asn, retval ? "hit" : "miss", vmid, hyp, secure,
            retval ? retval->pfn       : 0, retval ? retval->size  : 0,
            retval ? retval->pAddr(va) : 0, retval ? retval->ap    : 0,
            retval ? retval->ns        : 0, retval ? retval->nstid : 0,
            retval ? retval->global    : 0, retval ? retval->asid  : 0,
            retval ? retval->el        : 0);

    return retval;
}

// insert a new TLB entry
void
TLB::insert(Addr addr, TlbEntry &entry)
{
    DPRINTF(TLB, "Inserting entry into TLB with pfn:%#x size:%#x vpn: %#x"
            " asid:%d vmid:%d N:%d global:%d valid:%d nc:%d xn:%d"
            " ap:%#x domain:%#x ns:%d nstid:%d isHyp:%d\n", entry.pfn,
            entry.size, entry.vpn, entry.asid, entry.vmid, entry.N,
            entry.global, entry.valid, entry.nonCacheable, entry.xn,
            entry.ap, static_cast<uint8_t>(entry.domain), entry.ns, entry.nstid,
            entry.isHyp);

    if (table[size - 1].valid)
        DPRINTF(TLB, " - Replacing Valid entry %#x, asn %d vmn %d ppn %#x "
                "size: %#x ap:%d ns:%d nstid:%d g:%d isHyp:%d el: %d\n",
                table[size-1].vpn << table[size-1].N, table[size-1].asid,
                table[size-1].vmid, table[size-1].pfn << table[size-1].N,
                table[size-1].size, table[size-1].ap, table[size-1].ns,
                table[size-1].nstid, table[size-1].global, table[size-1].isHyp,
                table[size-1].el);

    //inserting to MRU position and evicting the LRU one

    for (int i = size - 1; i > 0; --i)
        table[i] = table[i-1];
    table[0] = entry;

    inserts++;
    ppRefills->notify(1);
}

void
TLB::printTlb() const
{
    int x = 0;
    TlbEntry *te;
    DPRINTF(TLB, "Current TLB contents:\n");
    while (x < size) {
        te = &table[x];
        if (te->valid)
            DPRINTF(TLB, " *  %s\n", te->print());
        ++x;
    }
}

void
TLB::flushAllSecurity(bool secure_lookup, uint8_t target_el, bool ignore_el)
{
    DPRINTF(TLB, "Flushing all TLB entries (%s lookup)\n",
            (secure_lookup ? "secure" : "non-secure"));
    int x = 0;
    TlbEntry *te;
    while (x < size) {
        te = &table[x];
        if (te->valid && secure_lookup == !te->nstid &&
            (te->vmid == vmid || secure_lookup) &&
            checkELMatch(target_el, te->el, ignore_el)) {

            DPRINTF(TLB, " -  %s\n", te->print());
            te->valid = false;
            flushedEntries++;
        }
        ++x;
    }

    flushTlb++;

    // If there's a second stage TLB (and we're not it) then flush it as well
    // if we're currently in hyp mode
    if (!isStage2 && isHyp) {
        stage2Tlb->flushAllSecurity(secure_lookup, true);
    }
}

void
TLB::flushAllNs(bool hyp, uint8_t target_el, bool ignore_el)
{
    DPRINTF(TLB, "Flushing all NS TLB entries (%s lookup)\n",
            (hyp ? "hyp" : "non-hyp"));
    int x = 0;
    TlbEntry *te;
    while (x < size) {
        te = &table[x];
        if (te->valid && te->nstid && te->isHyp == hyp &&
            checkELMatch(target_el, te->el, ignore_el)) {

            DPRINTF(TLB, " -  %s\n", te->print());
            flushedEntries++;
            te->valid = false;
        }
        ++x;
    }

    flushTlb++;

    // If there's a second stage TLB (and we're not it) then flush it as well
    if (!isStage2 && !hyp) {
        stage2Tlb->flushAllNs(false, true);
    }
}

void
TLB::flushMvaAsid(Addr mva, uint64_t asn, bool secure_lookup, uint8_t target_el)
{
    DPRINTF(TLB, "Flushing TLB entries with mva: %#x, asid: %#x "
            "(%s lookup)\n", mva, asn, (secure_lookup ?
            "secure" : "non-secure"));
    _flushMva(mva, asn, secure_lookup, false, false, target_el);
    flushTlbMvaAsid++;
}

void
TLB::flushAsid(uint64_t asn, bool secure_lookup, uint8_t target_el)
{
    DPRINTF(TLB, "Flushing TLB entries with asid: %#x (%s lookup)\n", asn,
            (secure_lookup ? "secure" : "non-secure"));

    int x = 0 ;
    TlbEntry *te;

    while (x < size) {
        te = &table[x];
        if (te->valid && te->asid == asn && secure_lookup == !te->nstid &&
            (te->vmid == vmid || secure_lookup) &&
            checkELMatch(target_el, te->el, false)) {

            te->valid = false;
            DPRINTF(TLB, " -  %s\n", te->print());
            flushedEntries++;
        }
        ++x;
    }
    flushTlbAsid++;
}

void
TLB::flushMva(Addr mva, bool secure_lookup, bool hyp, uint8_t target_el)
{
    DPRINTF(TLB, "Flushing TLB entries with mva: %#x (%s lookup)\n", mva,
            (secure_lookup ? "secure" : "non-secure"));
    _flushMva(mva, 0xbeef, secure_lookup, hyp, true, target_el);
    flushTlbMva++;
}

void
TLB::_flushMva(Addr mva, uint64_t asn, bool secure_lookup, bool hyp,
               bool ignore_asn, uint8_t target_el)
{
    TlbEntry *te;
    // D5.7.2: Sign-extend address to 64 bits
    mva = sext<56>(mva);
    te = lookup(mva, asn, vmid, hyp, secure_lookup, false, ignore_asn,
                target_el);
    while (te != NULL) {
        if (secure_lookup == !te->nstid) {
            DPRINTF(TLB, " -  %s\n", te->print());
            te->valid = false;
            flushedEntries++;
        }
        te = lookup(mva, asn, vmid, hyp, secure_lookup, false, ignore_asn,
                    target_el);
    }
}

bool
TLB::checkELMatch(uint8_t target_el, uint8_t tentry_el, bool ignore_el)
{
    bool elMatch = true;
    if (!ignore_el) {
        if (target_el == 2 || target_el == 3) {
            elMatch = (tentry_el  == target_el);
        } else {
            elMatch = (tentry_el == 0) || (tentry_el  == 1);
        }
    }
    return elMatch;
}

void
TLB::drainResume()
{
    // We might have unserialized something or switched CPUs, so make
    // sure to re-read the misc regs.
    miscRegValid = false;
}

void
TLB::takeOverFrom(BaseTLB *_otlb)
{
    TLB *otlb = dynamic_cast<TLB*>(_otlb);
    /* Make sure we actually have a valid type */
    if (otlb) {
        _attr = otlb->_attr;
        haveLPAE = otlb->haveLPAE;
        directToStage2 = otlb->directToStage2;
        stage2Req = otlb->stage2Req;

        /* Sync the stage2 MMU if they exist in both
         * the old CPU and the new
         */
        if (!isStage2 &&
            stage2Tlb && otlb->stage2Tlb) {
            stage2Tlb->takeOverFrom(otlb->stage2Tlb);
        }
    } else {
        panic("Incompatible TLB type!");
    }
}

void
TLB::serialize(CheckpointOut &cp) const
{
    DPRINTF(Checkpoint, "Serializing Arm TLB\n");

    SERIALIZE_SCALAR(_attr);
    SERIALIZE_SCALAR(haveLPAE);
    SERIALIZE_SCALAR(directToStage2);
    SERIALIZE_SCALAR(stage2Req);

    int num_entries = size;
    SERIALIZE_SCALAR(num_entries);
    for (int i = 0; i < size; i++)
        table[i].serializeSection(cp, csprintf("TlbEntry%d", i));
}

void
TLB::unserialize(CheckpointIn &cp)
{
    DPRINTF(Checkpoint, "Unserializing Arm TLB\n");

    UNSERIALIZE_SCALAR(_attr);
    UNSERIALIZE_SCALAR(haveLPAE);
    UNSERIALIZE_SCALAR(directToStage2);
    UNSERIALIZE_SCALAR(stage2Req);

    int num_entries;
    UNSERIALIZE_SCALAR(num_entries);
    for (int i = 0; i < min(size, num_entries); i++)
        table[i].unserializeSection(cp, csprintf("TlbEntry%d", i));
}

void
TLB::regStats()
{
    instHits
        .name(name() + ".inst_hits")
        .desc("ITB inst hits")
        ;

    instMisses
        .name(name() + ".inst_misses")
        .desc("ITB inst misses")
        ;

    instAccesses
        .name(name() + ".inst_accesses")
        .desc("ITB inst accesses")
        ;

    readHits
        .name(name() + ".read_hits")
        .desc("DTB read hits")
        ;

    readMisses
        .name(name() + ".read_misses")
        .desc("DTB read misses")
        ;

    readAccesses
        .name(name() + ".read_accesses")
        .desc("DTB read accesses")
        ;

    writeHits
        .name(name() + ".write_hits")
        .desc("DTB write hits")
        ;

    writeMisses
        .name(name() + ".write_misses")
        .desc("DTB write misses")
        ;

    writeAccesses
        .name(name() + ".write_accesses")
        .desc("DTB write accesses")
        ;

    hits
        .name(name() + ".hits")
        .desc("DTB hits")
        ;

    misses
        .name(name() + ".misses")
        .desc("DTB misses")
        ;

    accesses
        .name(name() + ".accesses")
        .desc("DTB accesses")
        ;

    flushTlb
        .name(name() + ".flush_tlb")
        .desc("Number of times complete TLB was flushed")
        ;

    flushTlbMva
        .name(name() + ".flush_tlb_mva")
        .desc("Number of times TLB was flushed by MVA")
        ;

    flushTlbMvaAsid
        .name(name() + ".flush_tlb_mva_asid")
        .desc("Number of times TLB was flushed by MVA & ASID")
        ;

    flushTlbAsid
        .name(name() + ".flush_tlb_asid")
        .desc("Number of times TLB was flushed by ASID")
        ;

    flushedEntries
        .name(name() + ".flush_entries")
        .desc("Number of entries that have been flushed from TLB")
        ;

    alignFaults
        .name(name() + ".align_faults")
        .desc("Number of TLB faults due to alignment restrictions")
        ;

    prefetchFaults
        .name(name() + ".prefetch_faults")
        .desc("Number of TLB faults due to prefetch")
        ;

    domainFaults
        .name(name() + ".domain_faults")
        .desc("Number of TLB faults due to domain restrictions")
        ;

    permsFaults
        .name(name() + ".perms_faults")
        .desc("Number of TLB faults due to permissions restrictions")
        ;

    instAccesses = instHits + instMisses;
    readAccesses = readHits + readMisses;
    writeAccesses = writeHits + writeMisses;
    hits = readHits + writeHits + instHits;
    misses = readMisses + writeMisses + instMisses;
    accesses = readAccesses + writeAccesses + instAccesses;
}

void
TLB::regProbePoints()
{
    ppRefills.reset(new ProbePoints::PMU(getProbeManager(), "Refills"));
}

Fault
TLB::translateSe(RequestPtr req, ThreadContext *tc, Mode mode,
                 Translation *translation, bool &delay, bool timing)
{
    updateMiscReg(tc);
    Addr vaddr_tainted = req->getVaddr();
    Addr vaddr = 0;
    if (aarch64)
        vaddr = purifyTaggedAddr(vaddr_tainted, tc, aarch64EL, ttbcr);
    else
        vaddr = vaddr_tainted;
    uint32_t flags = req->getFlags();

    bool is_fetch = (mode == Execute);
    bool is_write = (mode == Write);

    if (!is_fetch) {
        assert(flags & MustBeOne);
        if (sctlr.a || !(flags & AllowUnaligned)) {
            if (vaddr & mask(flags & AlignmentMask)) {
                // LPAE is always disabled in SE mode
                return std::make_shared<DataAbort>(
                    vaddr_tainted,
                    TlbEntry::DomainType::NoAccess, is_write,
                    ArmFault::AlignmentFault, isStage2,
                    ArmFault::VmsaTran);
            }
        }
    }

    Addr paddr;
    Process *p = tc->getProcessPtr();

    if (!p->pTable->translate(vaddr, paddr))
        return std::make_shared<GenericPageTableFault>(vaddr_tainted);
    req->setPaddr(paddr);

    return NoFault;
}

Fault
TLB::checkPermissions(TlbEntry *te, RequestPtr req, Mode mode)
{
    Addr vaddr = req->getVaddr(); // 32-bit don't have to purify
    uint32_t flags = req->getFlags();
    bool is_fetch  = (mode == Execute);
    bool is_write  = (mode == Write);
    bool is_priv   = isPriv && !(flags & UserMode);

    // Get the translation type from the actuall table entry
    ArmFault::TranMethod tranMethod = te->longDescFormat ? ArmFault::LpaeTran
                                                         : ArmFault::VmsaTran;

    // If this is the second stage of translation and the request is for a
    // stage 1 page table walk then we need to check the HCR.PTW bit. This
    // allows us to generate a fault if the request targets an area marked
    // as a device or strongly ordered.
    if (isStage2 && req->isPTWalk() && hcr.ptw &&
        (te->mtype != TlbEntry::MemoryType::Normal)) {
        return std::make_shared<DataAbort>(
            vaddr, te->domain, is_write,
            ArmFault::PermissionLL + te->lookupLevel,
            isStage2, tranMethod);
    }

    // Generate an alignment fault for unaligned data accesses to device or
    // strongly ordered memory
    if (!is_fetch) {
        if (te->mtype != TlbEntry::MemoryType::Normal) {
            if (vaddr & mask(flags & AlignmentMask)) {
                alignFaults++;
                return std::make_shared<DataAbort>(
                    vaddr, TlbEntry::DomainType::NoAccess, is_write,
                    ArmFault::AlignmentFault, isStage2,
                    tranMethod);
            }
        }
    }

    if (te->nonCacheable) {
        // Prevent prefetching from I/O devices.
        if (req->isPrefetch()) {
            // Here we can safely use the fault status for the short
            // desc. format in all cases
            return std::make_shared<PrefetchAbort>(
                vaddr, ArmFault::PrefetchUncacheable,
                isStage2, tranMethod);
        }
    }

    if (!te->longDescFormat) {
        switch ((dacr >> (static_cast<uint8_t>(te->domain) * 2)) & 0x3) {
          case 0:
            domainFaults++;
            DPRINTF(TLB, "TLB Fault: Data abort on domain. DACR: %#x"
                    " domain: %#x write:%d\n", dacr,
                    static_cast<uint8_t>(te->domain), is_write);
            if (is_fetch)
                return std::make_shared<PrefetchAbort>(
                    vaddr,
                    ArmFault::DomainLL + te->lookupLevel,
                    isStage2, tranMethod);
            else
                return std::make_shared<DataAbort>(
                    vaddr, te->domain, is_write,
                    ArmFault::DomainLL + te->lookupLevel,
                    isStage2, tranMethod);
          case 1:
            // Continue with permissions check
            break;
          case 2:
            panic("UNPRED domain\n");
          case 3:
            return NoFault;
        }
    }

    // The 'ap' variable is AP[2:0] or {AP[2,1],1b'0}, i.e. always three bits
    uint8_t ap  = te->longDescFormat ? te->ap << 1 : te->ap;
    uint8_t hap = te->hap;

    if (sctlr.afe == 1 || te->longDescFormat)
        ap |= 1;

    bool abt;
    bool isWritable = true;
    // If this is a stage 2 access (eg for reading stage 1 page table entries)
    // then don't perform the AP permissions check, we stil do the HAP check
    // below.
    if (isStage2) {
        abt = false;
    } else {
        switch (ap) {
          case 0:
            DPRINTF(TLB, "Access permissions 0, checking rs:%#x\n",
                    (int)sctlr.rs);
            if (!sctlr.xp) {
                switch ((int)sctlr.rs) {
                  case 2:
                    abt = is_write;
                    break;
                  case 1:
                    abt = is_write || !is_priv;
                    break;
                  case 0:
                  case 3:
                  default:
                    abt = true;
                    break;
                }
            } else {
                abt = true;
            }
            break;
          case 1:
            abt = !is_priv;
            break;
          case 2:
            abt = !is_priv && is_write;
            isWritable = is_priv;
            break;
          case 3:
            abt = false;
            break;
          case 4:
            panic("UNPRED premissions\n");
          case 5:
            abt = !is_priv || is_write;
            isWritable = false;
            break;
          case 6:
          case 7:
            abt        = is_write;
            isWritable = false;
            break;
          default:
            panic("Unknown permissions %#x\n", ap);
        }
    }

    bool hapAbt = is_write ? !(hap & 2) : !(hap & 1);
    bool xn     = te->xn || (isWritable && sctlr.wxn) ||
                            (ap == 3    && sctlr.uwxn && is_priv);
    if (is_fetch && (abt || xn ||
                     (te->longDescFormat && te->pxn && is_priv) ||
                     (isSecure && te->ns && scr.sif))) {
        permsFaults++;
        DPRINTF(TLB, "TLB Fault: Prefetch abort on permission check. AP:%d "
                     "priv:%d write:%d ns:%d sif:%d sctlr.afe: %d \n",
                     ap, is_priv, is_write, te->ns, scr.sif,sctlr.afe);
        return std::make_shared<PrefetchAbort>(
            vaddr,
            ArmFault::PermissionLL + te->lookupLevel,
            isStage2, tranMethod);
    } else if (abt | hapAbt) {
        permsFaults++;
        DPRINTF(TLB, "TLB Fault: Data abort on permission check. AP:%d priv:%d"
               " write:%d\n", ap, is_priv, is_write);
        return std::make_shared<DataAbort>(
            vaddr, te->domain, is_write,
            ArmFault::PermissionLL + te->lookupLevel,
            isStage2 | !abt, tranMethod);
    }
    return NoFault;
}


Fault
TLB::checkPermissions64(TlbEntry *te, RequestPtr req, Mode mode,
                        ThreadContext *tc)
{
    assert(aarch64);

    Addr vaddr_tainted = req->getVaddr();
    Addr vaddr = purifyTaggedAddr(vaddr_tainted, tc, aarch64EL, ttbcr);

    uint32_t flags = req->getFlags();
    bool is_fetch  = (mode == Execute);
    bool is_write  = (mode == Write);
    bool is_priv M5_VAR_USED  = isPriv && !(flags & UserMode);

    updateMiscReg(tc, curTranType);

    // If this is the second stage of translation and the request is for a
    // stage 1 page table walk then we need to check the HCR.PTW bit. This
    // allows us to generate a fault if the request targets an area marked
    // as a device or strongly ordered.
    if (isStage2 && req->isPTWalk() && hcr.ptw &&
        (te->mtype != TlbEntry::MemoryType::Normal)) {
        return std::make_shared<DataAbort>(
            vaddr_tainted, te->domain, is_write,
            ArmFault::PermissionLL + te->lookupLevel,
            isStage2, ArmFault::LpaeTran);
    }

    // Generate an alignment fault for unaligned accesses to device or
    // strongly ordered memory
    if (!is_fetch) {
        if (te->mtype != TlbEntry::MemoryType::Normal) {
            if (vaddr & mask(flags & AlignmentMask)) {
                alignFaults++;
                return std::make_shared<DataAbort>(
                    vaddr_tainted,
                    TlbEntry::DomainType::NoAccess, is_write,
                    ArmFault::AlignmentFault, isStage2,
                    ArmFault::LpaeTran);
            }
        }
    }

    if (te->nonCacheable) {
        // Prevent prefetching from I/O devices.
        if (req->isPrefetch()) {
            // Here we can safely use the fault status for the short
            // desc. format in all cases
            return std::make_shared<PrefetchAbort>(
                vaddr_tainted,
                ArmFault::PrefetchUncacheable,
                isStage2, ArmFault::LpaeTran);
        }
    }

    uint8_t ap  = 0x3 & (te->ap);  // 2-bit access protection field
    bool grant = false;

    uint8_t xn =  te->xn;
    uint8_t pxn = te->pxn;
    bool r = !is_write && !is_fetch;
    bool w = is_write;
    bool x = is_fetch;
    DPRINTF(TLBVerbose, "Checking permissions: ap:%d, xn:%d, pxn:%d, r:%d, "
                        "w:%d, x:%d\n", ap, xn, pxn, r, w, x);

    if (isStage2) {
        panic("Virtualization in AArch64 state is not supported yet");
    } else {
        switch (aarch64EL) {
          case EL0:
            {
                uint8_t perm = (ap << 2)  | (xn << 1) | pxn;
                switch (perm) {
                  case 0:
                  case 1:
                  case 8:
                  case 9:
                    grant = x;
                    break;
                  case 4:
                  case 5:
                    grant = r || w || (x && !sctlr.wxn);
                    break;
                  case 6:
                  case 7:
                    grant = r || w;
                    break;
                  case 12:
                  case 13:
                    grant = r || x;
                    break;
                  case 14:
                  case 15:
                    grant = r;
                    break;
                  default:
                    grant = false;
                }
            }
            break;
          case EL1:
            {
                uint8_t perm = (ap << 2)  | (xn << 1) | pxn;
                switch (perm) {
                  case 0:
                  case 2:
                    grant = r || w || (x && !sctlr.wxn);
                    break;
                  case 1:
                  case 3:
                  case 4:
                  case 5:
                  case 6:
                  case 7:
                    // regions that are writeable at EL0 should not be
                    // executable at EL1
                    grant = r || w;
                    break;
                  case 8:
                  case 10:
                  case 12:
                  case 14:
                    grant = r || x;
                    break;
                  case 9:
                  case 11:
                  case 13:
                  case 15:
                    grant = r;
                    break;
                  default:
                    grant = false;
                }
            }
            break;
          case EL2:
          case EL3:
            {
                uint8_t perm = (ap & 0x2) | xn;
                switch (perm) {
                  case 0:
                    grant = r || w || (x && !sctlr.wxn) ;
                    break;
                  case 1:
                    grant = r || w;
                    break;
                  case 2:
                    grant = r || x;
                    break;
                  case 3:
                    grant = r;
                    break;
                  default:
                    grant = false;
                }
            }
            break;
        }
    }

    if (!grant) {
        if (is_fetch) {
            permsFaults++;
            DPRINTF(TLB, "TLB Fault: Prefetch abort on permission check. "
                    "AP:%d priv:%d write:%d ns:%d sif:%d "
                    "sctlr.afe: %d\n",
                    ap, is_priv, is_write, te->ns, scr.sif, sctlr.afe);
            // Use PC value instead of vaddr because vaddr might be aligned to
            // cache line and should not be the address reported in FAR
            return std::make_shared<PrefetchAbort>(
                req->getPC(),
                ArmFault::PermissionLL + te->lookupLevel,
                isStage2, ArmFault::LpaeTran);
        } else {
            permsFaults++;
            DPRINTF(TLB, "TLB Fault: Data abort on permission check. AP:%d "
                    "priv:%d write:%d\n", ap, is_priv, is_write);
            return std::make_shared<DataAbort>(
                vaddr_tainted, te->domain, is_write,
                ArmFault::PermissionLL + te->lookupLevel,
                isStage2, ArmFault::LpaeTran);
        }
    }

    return NoFault;
}

Fault
TLB::translateFs(RequestPtr req, ThreadContext *tc, Mode mode,
        Translation *translation, bool &delay, bool timing,
        TLB::ArmTranslationType tranType, bool functional)
{
    // No such thing as a functional timing access
    assert(!(timing && functional));

    updateMiscReg(tc, tranType);

    Addr vaddr_tainted = req->getVaddr();
    Addr vaddr = 0;
    if (aarch64)
        vaddr = purifyTaggedAddr(vaddr_tainted, tc, aarch64EL, ttbcr);
    else
        vaddr = vaddr_tainted;
    uint32_t flags = req->getFlags();

    bool is_fetch  = (mode == Execute);
    bool is_write  = (mode == Write);
    bool long_desc_format = aarch64 || (haveLPAE && ttbcr.eae);
    ArmFault::TranMethod tranMethod = long_desc_format ? ArmFault::LpaeTran
                                                       : ArmFault::VmsaTran;

    req->setAsid(asid);

    DPRINTF(TLBVerbose, "CPSR is priv:%d UserMode:%d secure:%d S1S2NsTran:%d\n",
            isPriv, flags & UserMode, isSecure, tranType & S1S2NsTran);

    DPRINTF(TLB, "translateFs addr %#x, mode %d, st2 %d, scr %#x sctlr %#x "
                 "flags %#x tranType 0x%x\n", vaddr_tainted, mode, isStage2,
                 scr, sctlr, flags, tranType);

    if ((req->isInstFetch() && (!sctlr.i)) ||
        ((!req->isInstFetch()) && (!sctlr.c))){
       req->setFlags(Request::UNCACHEABLE | Request::STRICT_ORDER);
    }
    if (!is_fetch) {
        assert(flags & MustBeOne);
        if (sctlr.a || !(flags & AllowUnaligned)) {
            if (vaddr & mask(flags & AlignmentMask)) {
                alignFaults++;
                return std::make_shared<DataAbort>(
                    vaddr_tainted,
                    TlbEntry::DomainType::NoAccess, is_write,
                    ArmFault::AlignmentFault, isStage2,
                    tranMethod);
            }
        }
    }

    // If guest MMU is off or hcr.vm=0 go straight to stage2
    if ((isStage2 && !hcr.vm) || (!isStage2 && !sctlr.m)) {

        req->setPaddr(vaddr);
        // When the MMU is off the security attribute corresponds to the
        // security state of the processor
        if (isSecure)
            req->setFlags(Request::SECURE);

        // @todo: double check this (ARM ARM issue C B3.2.1)
        if (long_desc_format || sctlr.tre == 0) {
            req->setFlags(Request::UNCACHEABLE | Request::STRICT_ORDER);
        } else {
            if (nmrr.ir0 == 0 || nmrr.or0 == 0 || prrr.tr0 != 0x2)
                req->setFlags(Request::UNCACHEABLE | Request::STRICT_ORDER);
        }

        // Set memory attributes
        TlbEntry temp_te;
        temp_te.ns = !isSecure;
        if (isStage2 || hcr.dc == 0 || isSecure ||
           (isHyp && !(tranType & S1CTran))) {

            temp_te.mtype      = is_fetch ? TlbEntry::MemoryType::Normal
                                          : TlbEntry::MemoryType::StronglyOrdered;
            temp_te.innerAttrs = 0x0;
            temp_te.outerAttrs = 0x0;
            temp_te.shareable  = true;
            temp_te.outerShareable = true;
        } else {
            temp_te.mtype      = TlbEntry::MemoryType::Normal;
            temp_te.innerAttrs = 0x3;
            temp_te.outerAttrs = 0x3;
            temp_te.shareable  = false;
            temp_te.outerShareable = false;
        }
        temp_te.setAttributes(long_desc_format);
        DPRINTF(TLBVerbose, "(No MMU) setting memory attributes: shareable: "
                "%d, innerAttrs: %d, outerAttrs: %d, isStage2: %d\n",
                temp_te.shareable, temp_te.innerAttrs, temp_te.outerAttrs,
                isStage2);
        setAttr(temp_te.attributes);

        return testTranslation(req, mode, TlbEntry::DomainType::NoAccess);
    }

    DPRINTF(TLBVerbose, "Translating %s=%#x context=%d\n",
            isStage2 ? "IPA" : "VA", vaddr_tainted, asid);
    // Translation enabled

    TlbEntry *te = NULL;
    TlbEntry mergeTe;
    Fault fault = getResultTe(&te, req, tc, mode, translation, timing,
                              functional, &mergeTe);
    // only proceed if we have a valid table entry
    if ((te == NULL) && (fault == NoFault)) delay = true;

    // If we have the table entry transfer some of the attributes to the
    // request that triggered the translation
    if (te != NULL) {
        // Set memory attributes
        DPRINTF(TLBVerbose,
                "Setting memory attributes: shareable: %d, innerAttrs: %d, "
                "outerAttrs: %d, mtype: %d, isStage2: %d\n",
                te->shareable, te->innerAttrs, te->outerAttrs,
                static_cast<uint8_t>(te->mtype), isStage2);
        setAttr(te->attributes);

        if (te->nonCacheable)
            req->setFlags(Request::UNCACHEABLE);

        // Require requests to be ordered if the request goes to
        // strongly ordered or device memory (i.e., anything other
        // than normal memory requires strict order).
        if (te->mtype != TlbEntry::MemoryType::Normal)
            req->setFlags(Request::STRICT_ORDER);

        Addr pa = te->pAddr(vaddr);
        req->setPaddr(pa);

        if (isSecure && !te->ns) {
            req->setFlags(Request::SECURE);
        }
        if ((!is_fetch) && (vaddr & mask(flags & AlignmentMask)) &&
            (te->mtype != TlbEntry::MemoryType::Normal)) {
                // Unaligned accesses to Device memory should always cause an
                // abort regardless of sctlr.a
                alignFaults++;
                return std::make_shared<DataAbort>(
                    vaddr_tainted,
                    TlbEntry::DomainType::NoAccess, is_write,
                    ArmFault::AlignmentFault, isStage2,
                    tranMethod);
        }

        // Check for a trickbox generated address fault
        if (fault == NoFault)
            fault = testTranslation(req, mode, te->domain);
    }

    // Generate Illegal Inst Set State fault if IL bit is set in CPSR
    if (fault == NoFault) {
        if (aarch64 && is_fetch && cpsr.il == 1) {
            return std::make_shared<IllegalInstSetStateFault>();
        }
    }

    return fault;
}

Fault
TLB::translateAtomic(RequestPtr req, ThreadContext *tc, Mode mode,
    TLB::ArmTranslationType tranType)
{
    updateMiscReg(tc, tranType);

    if (directToStage2) {
        assert(stage2Tlb);
        return stage2Tlb->translateAtomic(req, tc, mode, tranType);
    }

    bool delay = false;
    Fault fault;
    if (FullSystem)
        fault = translateFs(req, tc, mode, NULL, delay, false, tranType);
    else
        fault = translateSe(req, tc, mode, NULL, delay, false);
    assert(!delay);
    return fault;
}

Fault
TLB::translateFunctional(RequestPtr req, ThreadContext *tc, Mode mode,
    TLB::ArmTranslationType tranType)
{
    updateMiscReg(tc, tranType);

    if (directToStage2) {
        assert(stage2Tlb);
        return stage2Tlb->translateFunctional(req, tc, mode, tranType);
    }

    bool delay = false;
    Fault fault;
    if (FullSystem)
        fault = translateFs(req, tc, mode, NULL, delay, false, tranType, true);
   else
        fault = translateSe(req, tc, mode, NULL, delay, false);
    assert(!delay);
    return fault;
}

Fault
TLB::translateTiming(RequestPtr req, ThreadContext *tc,
    Translation *translation, Mode mode, TLB::ArmTranslationType tranType)
{
    updateMiscReg(tc, tranType);

    if (directToStage2) {
        assert(stage2Tlb);
        return stage2Tlb->translateTiming(req, tc, translation, mode, tranType);
    }

    assert(translation);

    return translateComplete(req, tc, translation, mode, tranType, isStage2);
}

Fault
TLB::translateComplete(RequestPtr req, ThreadContext *tc,
        Translation *translation, Mode mode, TLB::ArmTranslationType tranType,
        bool callFromS2)
{
    bool delay = false;
    Fault fault;
    if (FullSystem)
        fault = translateFs(req, tc, mode, translation, delay, true, tranType);
    else
        fault = translateSe(req, tc, mode, translation, delay, true);
    DPRINTF(TLBVerbose, "Translation returning delay=%d fault=%d\n", delay, fault !=
            NoFault);
    // If we have a translation, and we're not in the middle of doing a stage
    // 2 translation tell the translation that we've either finished or its
    // going to take a while. By not doing this when we're in the middle of a
    // stage 2 translation we prevent marking the translation as delayed twice,
    // one when the translation starts and again when the stage 1 translation
    // completes.
    if (translation && (callFromS2 || !stage2Req || req->hasPaddr() || fault != NoFault)) {
        if (!delay)
            translation->finish(fault, req, tc, mode);
        else
            translation->markDelayed();
    }
    return fault;
}

BaseMasterPort*
TLB::getMasterPort()
{
    return &stage2Mmu->getPort();
}

void
TLB::updateMiscReg(ThreadContext *tc, ArmTranslationType tranType)
{
    // check if the regs have changed, or the translation mode is different.
    // NOTE: the tran type doesn't affect stage 2 TLB's as they only handle
    // one type of translation anyway
    if (miscRegValid && miscRegContext == tc->contextId() &&
            ((tranType == curTranType) || isStage2)) {
        return;
    }

    DPRINTF(TLBVerbose, "TLB variables changed!\n");
    cpsr = tc->readMiscReg(MISCREG_CPSR);
    // Dependencies: SCR/SCR_EL3, CPSR
    isSecure  = inSecureState(tc);
    isSecure &= (tranType & HypMode)    == 0;
    isSecure &= (tranType & S1S2NsTran) == 0;
    aarch64 = !cpsr.width;
    if (aarch64) {  // AArch64
        aarch64EL = (ExceptionLevel) (uint8_t) cpsr.el;
        switch (aarch64EL) {
          case EL0:
          case EL1:
            {
                sctlr = tc->readMiscReg(MISCREG_SCTLR_EL1);
                ttbcr = tc->readMiscReg(MISCREG_TCR_EL1);
                uint64_t ttbr_asid = ttbcr.a1 ?
                    tc->readMiscReg(MISCREG_TTBR1_EL1) :
                    tc->readMiscReg(MISCREG_TTBR0_EL1);
                asid = bits(ttbr_asid,
                            (haveLargeAsid64 && ttbcr.as) ? 63 : 55, 48);
            }
            break;
          case EL2:
            sctlr = tc->readMiscReg(MISCREG_SCTLR_EL2);
            ttbcr = tc->readMiscReg(MISCREG_TCR_EL2);
            asid = -1;
            break;
          case EL3:
            sctlr = tc->readMiscReg(MISCREG_SCTLR_EL3);
            ttbcr = tc->readMiscReg(MISCREG_TCR_EL3);
            asid = -1;
            break;
        }
        scr = tc->readMiscReg(MISCREG_SCR_EL3);
        isPriv = aarch64EL != EL0;
        // @todo: modify this behaviour to support Virtualization in
        // AArch64
        vmid           = 0;
        isHyp          = false;
        directToStage2 = false;
        stage2Req      = false;
    } else {  // AArch32
        sctlr  = tc->readMiscReg(flattenMiscRegNsBanked(MISCREG_SCTLR, tc,
                                 !isSecure));
        ttbcr  = tc->readMiscReg(flattenMiscRegNsBanked(MISCREG_TTBCR, tc,
                                 !isSecure));
        scr    = tc->readMiscReg(MISCREG_SCR);
        isPriv = cpsr.mode != MODE_USER;
        if (haveLPAE && ttbcr.eae) {
            // Long-descriptor translation table format in use
            uint64_t ttbr_asid = tc->readMiscReg(
                flattenMiscRegNsBanked(ttbcr.a1 ? MISCREG_TTBR1
                                                : MISCREG_TTBR0,
                                       tc, !isSecure));
            asid = bits(ttbr_asid, 55, 48);
        } else {
            // Short-descriptor translation table format in use
            CONTEXTIDR context_id = tc->readMiscReg(flattenMiscRegNsBanked(
                MISCREG_CONTEXTIDR, tc,!isSecure));
            asid = context_id.asid;
        }
        prrr = tc->readMiscReg(flattenMiscRegNsBanked(MISCREG_PRRR, tc,
                               !isSecure));
        nmrr = tc->readMiscReg(flattenMiscRegNsBanked(MISCREG_NMRR, tc,
                               !isSecure));
        dacr = tc->readMiscReg(flattenMiscRegNsBanked(MISCREG_DACR, tc,
                               !isSecure));
        hcr  = tc->readMiscReg(MISCREG_HCR);

        if (haveVirtualization) {
            vmid   = bits(tc->readMiscReg(MISCREG_VTTBR), 55, 48);
            isHyp  = cpsr.mode == MODE_HYP;
            isHyp |=  tranType & HypMode;
            isHyp &= (tranType & S1S2NsTran) == 0;
            isHyp &= (tranType & S1CTran)    == 0;
            if (isHyp) {
                sctlr = tc->readMiscReg(MISCREG_HSCTLR);
            }
            // Work out if we should skip the first stage of translation and go
            // directly to stage 2. This value is cached so we don't have to
            // compute it for every translation.
            stage2Req      = hcr.vm && !isStage2 && !isHyp && !isSecure &&
                             !(tranType & S1CTran);
            directToStage2 = stage2Req && !sctlr.m;
        } else {
            vmid           = 0;
            stage2Req      = false;
            isHyp          = false;
            directToStage2 = false;
        }
    }
    miscRegValid = true;
    miscRegContext = tc->contextId();
    curTranType  = tranType;
}

Fault
TLB::getTE(TlbEntry **te, RequestPtr req, ThreadContext *tc, Mode mode,
        Translation *translation, bool timing, bool functional,
        bool is_secure, TLB::ArmTranslationType tranType)
{
    bool is_fetch = (mode == Execute);
    bool is_write = (mode == Write);

    Addr vaddr_tainted = req->getVaddr();
    Addr vaddr = 0;
    ExceptionLevel target_el = aarch64 ? aarch64EL : EL1;
    if (aarch64) {
        vaddr = purifyTaggedAddr(vaddr_tainted, tc, target_el, ttbcr);
    } else {
        vaddr = vaddr_tainted;
    }
    *te = lookup(vaddr, asid, vmid, isHyp, is_secure, false, false, target_el);
    if (*te == NULL) {
        if (req->isPrefetch()) {
            // if the request is a prefetch don't attempt to fill the TLB or go
            // any further with the memory access (here we can safely use the
            // fault status for the short desc. format in all cases)
           prefetchFaults++;
           return std::make_shared<PrefetchAbort>(
               vaddr_tainted, ArmFault::PrefetchTLBMiss, isStage2);
        }

        if (is_fetch)
            instMisses++;
        else if (is_write)
            writeMisses++;
        else
            readMisses++;

        // start translation table walk, pass variables rather than
        // re-retreaving in table walker for speed
        DPRINTF(TLB, "TLB Miss: Starting hardware table walker for %#x(%d:%d)\n",
                vaddr_tainted, asid, vmid);
        Fault fault;
        fault = tableWalker->walk(req, tc, asid, vmid, isHyp, mode,
                                  translation, timing, functional, is_secure,
                                  tranType);
        // for timing mode, return and wait for table walk,
        if (timing || fault != NoFault) {
            return fault;
        }

        *te = lookup(vaddr, asid, vmid, isHyp, is_secure, false, false, target_el);
        if (!*te)
            printTlb();
        assert(*te);
    } else {
        if (is_fetch)
            instHits++;
        else if (is_write)
            writeHits++;
        else
            readHits++;
    }
    return NoFault;
}

Fault
TLB::getResultTe(TlbEntry **te, RequestPtr req, ThreadContext *tc, Mode mode,
        Translation *translation, bool timing, bool functional,
        TlbEntry *mergeTe)
{
    Fault fault;
    TlbEntry *s1Te = NULL;

    Addr vaddr_tainted = req->getVaddr();

    // Get the stage 1 table entry
    fault = getTE(&s1Te, req, tc, mode, translation, timing, functional,
                  isSecure, curTranType);
    // only proceed if we have a valid table entry
    if ((s1Te != NULL) && (fault == NoFault)) {
        // Check stage 1 permissions before checking stage 2
        if (aarch64)
            fault = checkPermissions64(s1Te, req, mode, tc);
        else
            fault = checkPermissions(s1Te, req, mode);
        if (stage2Req & (fault == NoFault)) {
            Stage2LookUp *s2Lookup = new Stage2LookUp(this, stage2Tlb, *s1Te,
                req, translation, mode, timing, functional, curTranType);
            fault = s2Lookup->getTe(tc, mergeTe);
            if (s2Lookup->isComplete()) {
                *te = mergeTe;
                // We've finished with the lookup so delete it
                delete s2Lookup;
            } else {
                // The lookup hasn't completed, so we can't delete it now. We
                // get round this by asking the object to self delete when the
                // translation is complete.
                s2Lookup->setSelfDelete();
            }
        } else {
            // This case deals with an S1 hit (or bypass), followed by
            // an S2 hit-but-perms issue
            if (isStage2) {
                DPRINTF(TLBVerbose, "s2TLB: reqVa %#x, reqPa %#x, fault %p\n",
                        vaddr_tainted, req->hasPaddr() ? req->getPaddr() : ~0, fault);
                if (fault != NoFault) {
                    ArmFault *armFault = reinterpret_cast<ArmFault *>(fault.get());
                    armFault->annotate(ArmFault::S1PTW, false);
                    armFault->annotate(ArmFault::OVA, vaddr_tainted);
                }
            }
            *te = s1Te;
        }
    }
    return fault;
}

void
TLB::setTestInterface(SimObject *_ti)
{
    if (!_ti) {
        test = nullptr;
    } else {
        TlbTestInterface *ti(dynamic_cast<TlbTestInterface *>(_ti));
        fatal_if(!ti, "%s is not a valid ARM TLB tester\n", _ti->name());
        test = ti;
    }
}

Fault
TLB::testTranslation(RequestPtr req, Mode mode, TlbEntry::DomainType domain)
{
    if (!test) {
        return NoFault;
    } else {
        return test->translationCheck(req, isPriv, mode, domain);
    }
}

Fault
TLB::testWalk(Addr pa, Addr size, Addr va, bool is_secure, Mode mode,
              TlbEntry::DomainType domain, LookupLevel lookup_level)
{
    if (!test) {
        return NoFault;
    } else {
        return test->walkCheck(pa, size, va, is_secure, isPriv, mode,
                               domain, lookup_level);
    }
}


ArmISA::TLB *
ArmTLBParams::create()
{
    return new ArmISA::TLB(this);
}