gem5/src/arch/alpha/tlb.cc
Andreas Sandberg 48281375ee mem, cpu: Add a separate flag for strictly ordered memory
The Request::UNCACHEABLE flag currently has two different
functions. The first, and obvious, function is to prevent the memory
system from caching data in the request. The second function is to
prevent reordering and speculation in CPU models.

This changeset gives the order/speculation requirement a separate flag
(Request::STRICT_ORDER). This flag prevents CPU models from doing the
following optimizations:

    * Speculation: CPU models are not allowed to issue speculative
      loads.

    * Write combining: CPU models and caches are not allowed to merge
      writes to the same cache line.

Note: The memory system may still reorder accesses unless the
UNCACHEABLE flag is set. It is therefore expected that the
STRICT_ORDER flag is combined with the UNCACHEABLE flag to prevent
this behavior.
2015-05-05 03:22:33 -04:00

637 lines
19 KiB
C++

/*
* 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: Nathan Binkert
* Steve Reinhardt
* Andrew Schultz
*/
#include <memory>
#include <string>
#include <vector>
#include "arch/alpha/faults.hh"
#include "arch/alpha/pagetable.hh"
#include "arch/alpha/tlb.hh"
#include "arch/generic/debugfaults.hh"
#include "base/inifile.hh"
#include "base/str.hh"
#include "base/trace.hh"
#include "cpu/thread_context.hh"
#include "debug/TLB.hh"
#include "sim/full_system.hh"
using namespace std;
namespace AlphaISA {
///////////////////////////////////////////////////////////////////////
//
// Alpha TLB
//
#ifdef DEBUG
bool uncacheBit39 = false;
bool uncacheBit40 = false;
#endif
#define MODE2MASK(X) (1 << (X))
TLB::TLB(const Params *p)
: BaseTLB(p), size(p->size), nlu(0)
{
table = new TlbEntry[size];
memset(table, 0, sizeof(TlbEntry) * size);
flushCache();
}
TLB::~TLB()
{
if (table)
delete [] table;
}
void
TLB::regStats()
{
fetch_hits
.name(name() + ".fetch_hits")
.desc("ITB hits");
fetch_misses
.name(name() + ".fetch_misses")
.desc("ITB misses");
fetch_acv
.name(name() + ".fetch_acv")
.desc("ITB acv");
fetch_accesses
.name(name() + ".fetch_accesses")
.desc("ITB accesses");
fetch_accesses = fetch_hits + fetch_misses;
read_hits
.name(name() + ".read_hits")
.desc("DTB read hits")
;
read_misses
.name(name() + ".read_misses")
.desc("DTB read misses")
;
read_acv
.name(name() + ".read_acv")
.desc("DTB read access violations")
;
read_accesses
.name(name() + ".read_accesses")
.desc("DTB read accesses")
;
write_hits
.name(name() + ".write_hits")
.desc("DTB write hits")
;
write_misses
.name(name() + ".write_misses")
.desc("DTB write misses")
;
write_acv
.name(name() + ".write_acv")
.desc("DTB write access violations")
;
write_accesses
.name(name() + ".write_accesses")
.desc("DTB write accesses")
;
data_hits
.name(name() + ".data_hits")
.desc("DTB hits")
;
data_misses
.name(name() + ".data_misses")
.desc("DTB misses")
;
data_acv
.name(name() + ".data_acv")
.desc("DTB access violations")
;
data_accesses
.name(name() + ".data_accesses")
.desc("DTB accesses")
;
data_hits = read_hits + write_hits;
data_misses = read_misses + write_misses;
data_acv = read_acv + write_acv;
data_accesses = read_accesses + write_accesses;
}
// look up an entry in the TLB
TlbEntry *
TLB::lookup(Addr vpn, uint8_t asn)
{
// assume not found...
TlbEntry *retval = NULL;
if (EntryCache[0]) {
if (vpn == EntryCache[0]->tag &&
(EntryCache[0]->asma || EntryCache[0]->asn == asn))
retval = EntryCache[0];
else if (EntryCache[1]) {
if (vpn == EntryCache[1]->tag &&
(EntryCache[1]->asma || EntryCache[1]->asn == asn))
retval = EntryCache[1];
else if (EntryCache[2] && vpn == EntryCache[2]->tag &&
(EntryCache[2]->asma || EntryCache[2]->asn == asn))
retval = EntryCache[2];
}
}
if (retval == NULL) {
PageTable::const_iterator i = lookupTable.find(vpn);
if (i != lookupTable.end()) {
while (i->first == vpn) {
int index = i->second;
TlbEntry *entry = &table[index];
assert(entry->valid);
if (vpn == entry->tag && (entry->asma || entry->asn == asn)) {
retval = updateCache(entry);
break;
}
++i;
}
}
}
DPRINTF(TLB, "lookup %#x, asn %#x -> %s ppn %#x\n", vpn, (int)asn,
retval ? "hit" : "miss", retval ? retval->ppn : 0);
return retval;
}
Fault
TLB::checkCacheability(RequestPtr &req, bool itb)
{
// in Alpha, cacheability is controlled by upper-level bits of the
// physical address
/*
* We support having the uncacheable bit in either bit 39 or bit
* 40. The Turbolaser platform (and EV5) support having the bit
* in 39, but Tsunami (which Linux assumes uses an EV6) generates
* accesses with the bit in 40. So we must check for both, but we
* have debug flags to catch a weird case where both are used,
* which shouldn't happen.
*/
if (req->getPaddr() & PAddrUncachedBit43) {
// IPR memory space not implemented
if (PAddrIprSpace(req->getPaddr())) {
return std::make_shared<UnimpFault>(
"IPR memory space not implemented!");
} else {
// mark request as uncacheable
req->setFlags(Request::UNCACHEABLE | Request::STRICT_ORDER);
// Clear bits 42:35 of the physical address (10-2 in
// Tsunami manual)
req->setPaddr(req->getPaddr() & PAddrUncachedMask);
}
// We shouldn't be able to read from an uncachable address in Alpha as
// we don't have a ROM and we don't want to try to fetch from a device
// register as we destroy any data that is clear-on-read.
if (req->isUncacheable() && itb)
return std::make_shared<UnimpFault>(
"CPU trying to fetch from uncached I/O");
}
return NoFault;
}
// insert a new TLB entry
void
TLB::insert(Addr addr, TlbEntry &entry)
{
flushCache();
VAddr vaddr = addr;
if (table[nlu].valid) {
Addr oldvpn = table[nlu].tag;
PageTable::iterator i = lookupTable.find(oldvpn);
if (i == lookupTable.end())
panic("TLB entry not found in lookupTable");
int index;
while ((index = i->second) != nlu) {
if (table[index].tag != oldvpn)
panic("TLB entry not found in lookupTable");
++i;
}
DPRINTF(TLB, "remove @%d: %#x -> %#x\n", nlu, oldvpn, table[nlu].ppn);
lookupTable.erase(i);
}
DPRINTF(TLB, "insert @%d: %#x -> %#x\n", nlu, vaddr.vpn(), entry.ppn);
table[nlu] = entry;
table[nlu].tag = vaddr.vpn();
table[nlu].valid = true;
lookupTable.insert(make_pair(vaddr.vpn(), nlu));
nextnlu();
}
void
TLB::flushAll()
{
DPRINTF(TLB, "flushAll\n");
memset(table, 0, sizeof(TlbEntry) * size);
flushCache();
lookupTable.clear();
nlu = 0;
}
void
TLB::flushProcesses()
{
flushCache();
PageTable::iterator i = lookupTable.begin();
PageTable::iterator end = lookupTable.end();
while (i != end) {
int index = i->second;
TlbEntry *entry = &table[index];
assert(entry->valid);
// we can't increment i after we erase it, so save a copy and
// increment it to get the next entry now
PageTable::iterator cur = i;
++i;
if (!entry->asma) {
DPRINTF(TLB, "flush @%d: %#x -> %#x\n", index,
entry->tag, entry->ppn);
entry->valid = false;
lookupTable.erase(cur);
}
}
}
void
TLB::flushAddr(Addr addr, uint8_t asn)
{
flushCache();
VAddr vaddr = addr;
PageTable::iterator i = lookupTable.find(vaddr.vpn());
if (i == lookupTable.end())
return;
while (i != lookupTable.end() && i->first == vaddr.vpn()) {
int index = i->second;
TlbEntry *entry = &table[index];
assert(entry->valid);
if (vaddr.vpn() == entry->tag && (entry->asma || entry->asn == asn)) {
DPRINTF(TLB, "flushaddr @%d: %#x -> %#x\n", index, vaddr.vpn(),
entry->ppn);
// invalidate this entry
entry->valid = false;
lookupTable.erase(i++);
} else {
++i;
}
}
}
void
TLB::serialize(ostream &os)
{
SERIALIZE_SCALAR(size);
SERIALIZE_SCALAR(nlu);
for (int i = 0; i < size; i++) {
nameOut(os, csprintf("%s.Entry%d", name(), i));
table[i].serialize(os);
}
}
void
TLB::unserialize(Checkpoint *cp, const string &section)
{
UNSERIALIZE_SCALAR(size);
UNSERIALIZE_SCALAR(nlu);
for (int i = 0; i < size; i++) {
table[i].unserialize(cp, csprintf("%s.Entry%d", section, i));
if (table[i].valid) {
lookupTable.insert(make_pair(table[i].tag, i));
}
}
}
Fault
TLB::translateInst(RequestPtr req, ThreadContext *tc)
{
//If this is a pal pc, then set PHYSICAL
if (FullSystem && PcPAL(req->getPC()))
req->setFlags(Request::PHYSICAL);
if (PcPAL(req->getPC())) {
// strip off PAL PC marker (lsb is 1)
req->setPaddr((req->getVaddr() & ~3) & PAddrImplMask);
fetch_hits++;
return NoFault;
}
if (req->getFlags() & Request::PHYSICAL) {
req->setPaddr(req->getVaddr());
} else {
// verify that this is a good virtual address
if (!validVirtualAddress(req->getVaddr())) {
fetch_acv++;
return std::make_shared<ItbAcvFault>(req->getVaddr());
}
// VA<42:41> == 2, VA<39:13> maps directly to PA<39:13> for EV5
// VA<47:41> == 0x7e, VA<40:13> maps directly to PA<40:13> for EV6
if (VAddrSpaceEV6(req->getVaddr()) == 0x7e) {
// only valid in kernel mode
if (ICM_CM(tc->readMiscRegNoEffect(IPR_ICM)) !=
mode_kernel) {
fetch_acv++;
return std::make_shared<ItbAcvFault>(req->getVaddr());
}
req->setPaddr(req->getVaddr() & PAddrImplMask);
// sign extend the physical address properly
if (req->getPaddr() & PAddrUncachedBit40)
req->setPaddr(req->getPaddr() | ULL(0xf0000000000));
else
req->setPaddr(req->getPaddr() & ULL(0xffffffffff));
} else {
// not a physical address: need to look up pte
int asn = DTB_ASN_ASN(tc->readMiscRegNoEffect(IPR_DTB_ASN));
TlbEntry *entry = lookup(VAddr(req->getVaddr()).vpn(),
asn);
if (!entry) {
fetch_misses++;
return std::make_shared<ItbPageFault>(req->getVaddr());
}
req->setPaddr((entry->ppn << PageShift) +
(VAddr(req->getVaddr()).offset()
& ~3));
// check permissions for this access
if (!(entry->xre &
(1 << ICM_CM(tc->readMiscRegNoEffect(IPR_ICM))))) {
// instruction access fault
fetch_acv++;
return std::make_shared<ItbAcvFault>(req->getVaddr());
}
fetch_hits++;
}
}
// check that the physical address is ok (catch bad physical addresses)
if (req->getPaddr() & ~PAddrImplMask) {
return std::make_shared<MachineCheckFault>();
}
return checkCacheability(req, true);
}
Fault
TLB::translateData(RequestPtr req, ThreadContext *tc, bool write)
{
mode_type mode =
(mode_type)DTB_CM_CM(tc->readMiscRegNoEffect(IPR_DTB_CM));
/**
* Check for alignment faults
*/
if (req->getVaddr() & (req->getSize() - 1)) {
DPRINTF(TLB, "Alignment Fault on %#x, size = %d\n", req->getVaddr(),
req->getSize());
uint64_t flags = write ? MM_STAT_WR_MASK : 0;
return std::make_shared<DtbAlignmentFault>(req->getVaddr(),
req->getFlags(),
flags);
}
if (PcPAL(req->getPC())) {
mode = (req->getFlags() & AlphaRequestFlags::ALTMODE) ?
(mode_type)ALT_MODE_AM(
tc->readMiscRegNoEffect(IPR_ALT_MODE))
: mode_kernel;
}
if (req->getFlags() & Request::PHYSICAL) {
req->setPaddr(req->getVaddr());
} else {
// verify that this is a good virtual address
if (!validVirtualAddress(req->getVaddr())) {
if (write) { write_acv++; } else { read_acv++; }
uint64_t flags = (write ? MM_STAT_WR_MASK : 0) |
MM_STAT_BAD_VA_MASK |
MM_STAT_ACV_MASK;
return std::make_shared<DtbPageFault>(req->getVaddr(),
req->getFlags(),
flags);
}
// Check for "superpage" mapping
if (VAddrSpaceEV6(req->getVaddr()) == 0x7e) {
// only valid in kernel mode
if (DTB_CM_CM(tc->readMiscRegNoEffect(IPR_DTB_CM)) !=
mode_kernel) {
if (write) { write_acv++; } else { read_acv++; }
uint64_t flags = ((write ? MM_STAT_WR_MASK : 0) |
MM_STAT_ACV_MASK);
return std::make_shared<DtbAcvFault>(req->getVaddr(),
req->getFlags(),
flags);
}
req->setPaddr(req->getVaddr() & PAddrImplMask);
// sign extend the physical address properly
if (req->getPaddr() & PAddrUncachedBit40)
req->setPaddr(req->getPaddr() | ULL(0xf0000000000));
else
req->setPaddr(req->getPaddr() & ULL(0xffffffffff));
} else {
if (write)
write_accesses++;
else
read_accesses++;
int asn = DTB_ASN_ASN(tc->readMiscRegNoEffect(IPR_DTB_ASN));
// not a physical address: need to look up pte
TlbEntry *entry = lookup(VAddr(req->getVaddr()).vpn(), asn);
if (!entry) {
// page fault
if (write) { write_misses++; } else { read_misses++; }
uint64_t flags = (write ? MM_STAT_WR_MASK : 0) |
MM_STAT_DTB_MISS_MASK;
return (req->getFlags() & AlphaRequestFlags::VPTE) ?
(Fault)(std::make_shared<PDtbMissFault>(req->getVaddr(),
req->getFlags(),
flags)) :
(Fault)(std::make_shared<NDtbMissFault>(req->getVaddr(),
req->getFlags(),
flags));
}
req->setPaddr((entry->ppn << PageShift) +
VAddr(req->getVaddr()).offset());
if (write) {
if (!(entry->xwe & MODE2MASK(mode))) {
// declare the instruction access fault
write_acv++;
uint64_t flags = MM_STAT_WR_MASK |
MM_STAT_ACV_MASK |
(entry->fonw ? MM_STAT_FONW_MASK : 0);
return std::make_shared<DtbPageFault>(req->getVaddr(),
req->getFlags(),
flags);
}
if (entry->fonw) {
write_acv++;
uint64_t flags = MM_STAT_WR_MASK | MM_STAT_FONW_MASK;
return std::make_shared<DtbPageFault>(req->getVaddr(),
req->getFlags(),
flags);
}
} else {
if (!(entry->xre & MODE2MASK(mode))) {
read_acv++;
uint64_t flags = MM_STAT_ACV_MASK |
(entry->fonr ? MM_STAT_FONR_MASK : 0);
return std::make_shared<DtbAcvFault>(req->getVaddr(),
req->getFlags(),
flags);
}
if (entry->fonr) {
read_acv++;
uint64_t flags = MM_STAT_FONR_MASK;
return std::make_shared<DtbPageFault>(req->getVaddr(),
req->getFlags(),
flags);
}
}
}
if (write)
write_hits++;
else
read_hits++;
}
// check that the physical address is ok (catch bad physical addresses)
if (req->getPaddr() & ~PAddrImplMask) {
return std::make_shared<MachineCheckFault>();
}
return checkCacheability(req);
}
TlbEntry &
TLB::index(bool advance)
{
TlbEntry *entry = &table[nlu];
if (advance)
nextnlu();
return *entry;
}
Fault
TLB::translateAtomic(RequestPtr req, ThreadContext *tc, Mode mode)
{
if (mode == Execute)
return translateInst(req, tc);
else
return translateData(req, tc, mode == Write);
}
void
TLB::translateTiming(RequestPtr req, ThreadContext *tc,
Translation *translation, Mode mode)
{
assert(translation);
translation->finish(translateAtomic(req, tc, mode), req, tc, mode);
}
Fault
TLB::translateFunctional(RequestPtr req, ThreadContext *tc, Mode mode)
{
panic("Not implemented\n");
return NoFault;
}
Fault
TLB::finalizePhysical(RequestPtr req, ThreadContext *tc, Mode mode) const
{
return NoFault;
}
} // namespace AlphaISA
AlphaISA::TLB *
AlphaTLBParams::create()
{
return new AlphaISA::TLB(this);
}