/*
- * Copyright (c) 2012-2013, 2015 ARM Limited
+ * Copyright (c) 2012-2013, 2015-2018 ARM Limited
* All rights reserved.
*
* The license below extends only to copyright in the software and shall
* Miss Status and Handling Register (MSHR) definitions.
*/
-#include <algorithm>
+#include "mem/cache/mshr.hh"
+
#include <cassert>
#include <string>
-#include <vector>
-#include "base/misc.hh"
+#include "base/logging.hh"
+#include "base/trace.hh"
#include "base/types.hh"
#include "debug/Cache.hh"
-#include "mem/cache/cache.hh"
-#include "mem/cache/mshr.hh"
+#include "mem/cache/base.hh"
+#include "mem/request.hh"
#include "sim/core.hh"
-using namespace std;
-
-MSHR::MSHR() : readyTime(0), _isUncacheable(false), downstreamPending(false),
- pendingDirty(false),
+MSHR::MSHR() : downstreamPending(false),
+ pendingModified(false),
postInvalidate(false), postDowngrade(false),
- queue(NULL), order(0), blkAddr(0),
- blkSize(0), isSecure(false), inService(false),
- isForward(false), threadNum(InvalidThreadID), data(NULL)
+ isForward(false)
{
}
-
MSHR::TargetList::TargetList()
- : needsExclusive(false), hasUpgrade(false)
+ : needsWritable(false), hasUpgrade(false), allocOnFill(false),
+ hasFromCache(false)
{}
-inline void
-MSHR::TargetList::add(PacketPtr pkt, Tick readyTime,
- Counter order, Target::Source source, bool markPending)
+void
+MSHR::TargetList::updateFlags(PacketPtr pkt, Target::Source source,
+ bool alloc_on_fill)
{
if (source != Target::FromSnoop) {
- if (pkt->needsExclusive()) {
- needsExclusive = true;
+ if (pkt->needsWritable()) {
+ needsWritable = true;
}
// StoreCondReq is effectively an upgrade if it's in an MSHR
if (pkt->isUpgrade() || pkt->cmd == MemCmd::StoreCondReq) {
hasUpgrade = true;
}
+
+ // potentially re-evaluate whether we should allocate on a fill or
+ // not
+ allocOnFill = allocOnFill || alloc_on_fill;
+
+ if (source != Target::FromPrefetcher) {
+ hasFromCache = hasFromCache || pkt->fromCache();
+ }
}
+}
+void
+MSHR::TargetList::populateFlags()
+{
+ resetFlags();
+ for (auto& t: *this) {
+ updateFlags(t.pkt, t.source, t.allocOnFill);
+ }
+}
+
+inline void
+MSHR::TargetList::add(PacketPtr pkt, Tick readyTime,
+ Counter order, Target::Source source, bool markPending,
+ bool alloc_on_fill)
+{
+ updateFlags(pkt, source, alloc_on_fill);
if (markPending) {
// Iterate over the SenderState stack and see if we find
// an MSHR entry. If we do, set the downstreamPending
// flag. Otherwise, do nothing.
MSHR *mshr = pkt->findNextSenderState<MSHR>();
- if (mshr != NULL) {
+ if (mshr != nullptr) {
assert(!mshr->downstreamPending);
mshr->downstreamPending = true;
+ } else {
+ // No need to clear downstreamPending later
+ markPending = false;
}
}
- push_back(Target(pkt, readyTime, order, source, markPending));
+ emplace_back(pkt, readyTime, order, source, markPending, alloc_on_fill);
}
static void
replaceUpgrade(PacketPtr pkt)
{
+ // remember if the current packet has data allocated
+ bool has_data = pkt->hasData() || pkt->hasRespData();
+
if (pkt->cmd == MemCmd::UpgradeReq) {
pkt->cmd = MemCmd::ReadExReq;
DPRINTF(Cache, "Replacing UpgradeReq with ReadExReq\n");
pkt->cmd = MemCmd::StoreCondFailReq;
DPRINTF(Cache, "Replacing StoreCondReq with StoreCondFailReq\n");
}
+
+ if (!has_data) {
+ // there is no sensible way of setting the data field if the
+ // new command actually would carry data
+ assert(!pkt->hasData());
+
+ if (pkt->hasRespData()) {
+ // we went from a packet that had no data (neither request,
+ // nor response), to one that does, and therefore we need to
+ // actually allocate space for the data payload
+ pkt->allocate();
+ }
+ }
}
if (!hasUpgrade)
return;
- Iterator end_i = end();
- for (Iterator i = begin(); i != end_i; ++i) {
- replaceUpgrade(i->pkt);
+ for (auto& t : *this) {
+ replaceUpgrade(t.pkt);
}
hasUpgrade = false;
void
-MSHR::TargetList::clearDownstreamPending()
+MSHR::TargetList::clearDownstreamPending(MSHR::TargetList::iterator begin,
+ MSHR::TargetList::iterator end)
{
- Iterator end_i = end();
- for (Iterator i = begin(); i != end_i; ++i) {
- if (i->markedPending) {
+ for (auto t = begin; t != end; t++) {
+ if (t->markedPending) {
// Iterate over the SenderState stack and see if we find
// an MSHR entry. If we find one, clear the
// downstreamPending flag by calling
// clearDownstreamPending(). This recursively clears the
// downstreamPending flag in all caches this packet has
// passed through.
- MSHR *mshr = i->pkt->findNextSenderState<MSHR>();
- if (mshr != NULL) {
+ MSHR *mshr = t->pkt->findNextSenderState<MSHR>();
+ if (mshr != nullptr) {
mshr->clearDownstreamPending();
}
+ t->markedPending = false;
}
}
}
+void
+MSHR::TargetList::clearDownstreamPending()
+{
+ clearDownstreamPending(begin(), end());
+}
+
bool
-MSHR::TargetList::checkFunctional(PacketPtr pkt)
+MSHR::TargetList::trySatisfyFunctional(PacketPtr pkt)
{
- Iterator end_i = end();
- for (Iterator i = begin(); i != end_i; ++i) {
- if (pkt->checkFunctional(i->pkt)) {
+ for (auto& t : *this) {
+ if (pkt->trySatisfyFunctional(t.pkt)) {
return true;
}
}
void
-MSHR::TargetList::
-print(std::ostream &os, int verbosity, const std::string &prefix) const
+MSHR::TargetList::print(std::ostream &os, int verbosity,
+ const std::string &prefix) const
{
- ConstIterator end_i = end();
- for (ConstIterator i = begin(); i != end_i; ++i) {
+ for (auto& t : *this) {
const char *s;
- switch (i->source) {
+ switch (t.source) {
case Target::FromCPU:
s = "FromCPU";
break;
break;
}
ccprintf(os, "%s%s: ", prefix, s);
- i->pkt->print(os, verbosity, "");
+ t.pkt->print(os, verbosity, "");
+ ccprintf(os, "\n");
}
}
void
MSHR::allocate(Addr blk_addr, unsigned blk_size, PacketPtr target,
- Tick when_ready, Counter _order)
+ Tick when_ready, Counter _order, bool alloc_on_fill)
{
blkAddr = blk_addr;
blkSize = blk_size;
_isUncacheable = target->req->isUncacheable();
inService = false;
downstreamPending = false;
- threadNum = 0;
assert(targets.isReset());
// Don't know of a case where we would allocate a new MSHR for a
// snoop (mem-side request), so set source according to request here
Target::Source source = (target->cmd == MemCmd::HardPFReq) ?
Target::FromPrefetcher : Target::FromCPU;
- targets.add(target, when_ready, _order, source, true);
+ targets.add(target, when_ready, _order, source, true, alloc_on_fill);
assert(deferredTargets.isReset());
- data = NULL;
}
targets.clearDownstreamPending();
}
-bool
-MSHR::markInService(bool pending_dirty_resp)
+void
+MSHR::markInService(bool pending_modified_resp)
{
assert(!inService);
- if (isForwardNoResponse()) {
- // we just forwarded the request packet & don't expect a
- // response, so get rid of it
- assert(getNumTargets() == 1);
- popTarget();
- return true;
- }
inService = true;
- pendingDirty = targets.needsExclusive || pending_dirty_resp;
+ pendingModified = targets.needsWritable || pending_modified_resp;
postInvalidate = postDowngrade = false;
if (!downstreamPending) {
// level where it's going to get a response
targets.clearDownstreamPending();
}
- return false;
}
* Adds a target to an MSHR
*/
void
-MSHR::allocateTarget(PacketPtr pkt, Tick whenReady, Counter _order)
+MSHR::allocateTarget(PacketPtr pkt, Tick whenReady, Counter _order,
+ bool alloc_on_fill)
{
+ // assume we'd never issue a prefetch when we've got an
+ // outstanding miss
+ assert(pkt->cmd != MemCmd::HardPFReq);
+
// if there's a request already in service for this MSHR, we will
// have to defer the new target until after the response if any of
// the following are true:
// - there are other targets already deferred
// - there's a pending invalidate to be applied after the response
// comes back (but before this target is processed)
- // - this target requires an exclusive block and either we're not
- // getting an exclusive block back or we have already snooped
- // another read request that will downgrade our exclusive block
- // to shared
-
- // assume we'd never issue a prefetch when we've got an
- // outstanding miss
- assert(pkt->cmd != MemCmd::HardPFReq);
-
- if (inService &&
- (!deferredTargets.empty() || hasPostInvalidate() ||
- (pkt->needsExclusive() &&
- (!isPendingDirty() || hasPostDowngrade() || isForward)))) {
+ // - the MSHR's first (and only) non-deferred target is a cache
+ // maintenance packet
+ // - the new target is a cache maintenance packet (this is probably
+ // overly conservative but certainly safe)
+ // - this target requires a writable block and either we're not
+ // getting a writable block back or we have already snooped
+ // another read request that will downgrade our writable block
+ // to non-writable (Shared or Owned)
+ PacketPtr tgt_pkt = targets.front().pkt;
+ if (pkt->req->isCacheMaintenance() ||
+ tgt_pkt->req->isCacheMaintenance() ||
+ !deferredTargets.empty() ||
+ (inService &&
+ (hasPostInvalidate() ||
+ (pkt->needsWritable() &&
+ (!isPendingModified() || hasPostDowngrade() || isForward))))) {
// need to put on deferred list
- if (hasPostInvalidate())
+ if (inService && hasPostInvalidate())
replaceUpgrade(pkt);
- deferredTargets.add(pkt, whenReady, _order, Target::FromCPU, true);
+ deferredTargets.add(pkt, whenReady, _order, Target::FromCPU, true,
+ alloc_on_fill);
} else {
// No request outstanding, or still OK to append to
// outstanding request: append to regular target list. Only
// mark pending if current request hasn't been issued yet
// (isn't in service).
- targets.add(pkt, whenReady, _order, Target::FromCPU, !inService);
+ targets.add(pkt, whenReady, _order, Target::FromCPU, !inService,
+ alloc_on_fill);
}
}
bool
MSHR::handleSnoop(PacketPtr pkt, Counter _order)
{
- DPRINTF(Cache, "%s for %s addr %#llx size %d\n", __func__,
- pkt->cmdString(), pkt->getAddr(), pkt->getSize());
+ DPRINTF(Cache, "%s for %s\n", __func__, pkt->print());
+
+ // when we snoop packets the needsWritable and isInvalidate flags
+ // should always be the same, however, this assumes that we never
+ // snoop writes as they are currently not marked as invalidations
+ panic_if((pkt->needsWritable() != pkt->isInvalidate()) &&
+ !pkt->req->isCacheMaintenance(),
+ "%s got snoop %s where needsWritable, "
+ "does not match isInvalidate", name(), pkt->print());
+
if (!inService || (pkt->isExpressSnoop() && downstreamPending)) {
// Request has not been issued yet, or it's been issued
// locally but is buffered unissued at some downstream cache
// That is, even though the upper-level cache got out on its
// local bus first, some other invalidating transaction
// reached the global bus before the upgrade did.
- if (pkt->needsExclusive()) {
+ if (pkt->needsWritable() || pkt->req->isCacheInvalidate()) {
targets.replaceUpgrades();
deferredTargets.replaceUpgrades();
}
// From here on down, the request issued by this MSHR logically
// precedes the request we're snooping.
- if (pkt->needsExclusive()) {
+ if (pkt->needsWritable() || pkt->req->isCacheInvalidate()) {
// snooped request still precedes the re-request we'll have to
// issue for deferred targets, if any...
deferredTargets.replaceUpgrades();
}
- if (hasPostInvalidate()) {
- // a prior snoop has already appended an invalidation, so
- // logically we don't have the block anymore; no need for
- // further snooping.
+ PacketPtr tgt_pkt = targets.front().pkt;
+ if (hasPostInvalidate() || tgt_pkt->req->isCacheInvalidate()) {
+ // a prior snoop has already appended an invalidation or a
+ // cache invalidation operation is in progress, so logically
+ // we don't have the block anymore; no need for further
+ // snooping.
return true;
}
- if (isPendingDirty() || pkt->isInvalidate()) {
+ if (isPendingModified() || pkt->isInvalidate()) {
// We need to save and replay the packet in two cases:
- // 1. We're awaiting an exclusive copy, so ownership is pending,
- // and we need to respond after we receive data.
+ // 1. We're awaiting a writable copy (Modified or Exclusive),
+ // so this MSHR is the orgering point, and we need to respond
+ // after we receive data.
// 2. It's an invalidation (e.g., UpgradeReq), and we need
// to forward the snoop up the hierarchy after the current
// transaction completes.
-
- // Actual target device (typ. a memory) will delete the
- // packet on reception, so we need to save a copy here.
-
- // Clear flags and also allocate new data as the original
- // packet data storage may have been deleted by the time we
- // get to send this packet.
- PacketPtr cp_pkt = new Packet(pkt, true, true);
- targets.add(cp_pkt, curTick(), _order, Target::FromSnoop,
- downstreamPending && targets.needsExclusive);
- if (isPendingDirty()) {
- pkt->assertMemInhibit();
- pkt->setSupplyExclusive();
+ // Start by determining if we will eventually respond or not,
+ // matching the conditions checked in Cache::handleSnoop
+ bool will_respond = isPendingModified() && pkt->needsResponse() &&
+ !pkt->isClean();
+
+ // The packet we are snooping may be deleted by the time we
+ // actually process the target, and we consequently need to
+ // save a copy here. Clear flags and also allocate new data as
+ // the original packet data storage may have been deleted by
+ // the time we get to process this packet. In the cases where
+ // we are not responding after handling the snoop we also need
+ // to create a copy of the request to be on the safe side. In
+ // the latter case the cache is responsible for deleting both
+ // the packet and the request as part of handling the deferred
+ // snoop.
+ PacketPtr cp_pkt = will_respond ? new Packet(pkt, true, true) :
+ new Packet(std::make_shared<Request>(*pkt->req), pkt->cmd,
+ blkSize, pkt->id);
+
+ if (will_respond) {
+ // we are the ordering point, and will consequently
+ // respond, and depending on whether the packet
+ // needsWritable or not we either pass a Shared line or a
+ // Modified line
+ pkt->setCacheResponding();
+
+ // inform the cache hierarchy that this cache had the line
+ // in the Modified state, even if the response is passed
+ // as Shared (and thus non-writable)
+ pkt->setResponderHadWritable();
+
+ // in the case of an uncacheable request there is no need
+ // to set the responderHadWritable flag, but since the
+ // recipient does not care there is no harm in doing so
}
+ targets.add(cp_pkt, curTick(), _order, Target::FromSnoop,
+ downstreamPending && targets.needsWritable, false);
- if (pkt->needsExclusive()) {
+ if (pkt->needsWritable() || pkt->isInvalidate()) {
// This transaction will take away our pending copy
postInvalidate = true;
}
+
+ if (isPendingModified() && pkt->isClean()) {
+ pkt->setSatisfied();
+ }
}
- if (!pkt->needsExclusive()) {
+ if (!pkt->needsWritable() && !pkt->req->isUncacheable()) {
// This transaction will get a read-shared copy, downgrading
- // our copy if we had an exclusive one
+ // our copy if we had a writable one
postDowngrade = true;
- pkt->assertShared();
+ // make sure that any downstream cache does not respond with a
+ // writable (and dirty) copy even if it has one, unless it was
+ // explicitly asked for one
+ pkt->setHasSharers();
}
return true;
}
+MSHR::TargetList
+MSHR::extractServiceableTargets(PacketPtr pkt)
+{
+ TargetList ready_targets;
+ // If the downstream MSHR got an invalidation request then we only
+ // service the first of the FromCPU targets and any other
+ // non-FromCPU target. This way the remaining FromCPU targets
+ // issue a new request and get a fresh copy of the block and we
+ // avoid memory consistency violations.
+ if (pkt->cmd == MemCmd::ReadRespWithInvalidate) {
+ auto it = targets.begin();
+ assert((it->source == Target::FromCPU) ||
+ (it->source == Target::FromPrefetcher));
+ ready_targets.push_back(*it);
+ it = targets.erase(it);
+ while (it != targets.end()) {
+ if (it->source == Target::FromCPU) {
+ it++;
+ } else {
+ assert(it->source == Target::FromSnoop);
+ ready_targets.push_back(*it);
+ it = targets.erase(it);
+ }
+ }
+ ready_targets.populateFlags();
+ } else {
+ std::swap(ready_targets, targets);
+ }
+ targets.populateFlags();
+
+ return ready_targets;
+}
bool
MSHR::promoteDeferredTargets()
{
- assert(targets.empty());
- if (deferredTargets.empty()) {
+ if (targets.empty() && deferredTargets.empty()) {
+ // nothing to promote
return false;
}
- // swap targets & deferredTargets lists
- std::swap(targets, deferredTargets);
-
- // clear deferredTargets flags
- deferredTargets.resetFlags();
+ // the deferred targets can be generally promoted unless they
+ // contain a cache maintenance request
+
+ // find the first target that is a cache maintenance request
+ auto it = std::find_if(deferredTargets.begin(), deferredTargets.end(),
+ [](MSHR::Target &t) {
+ return t.pkt->req->isCacheMaintenance();
+ });
+ if (it == deferredTargets.begin()) {
+ // if the first deferred target is a cache maintenance packet
+ // then we can promote provided the targets list is empty and
+ // we can service it on its own
+ if (targets.empty()) {
+ targets.splice(targets.end(), deferredTargets, it);
+ }
+ } else {
+ // if a cache maintenance operation exists, we promote all the
+ // deferred targets that precede it, or all deferred targets
+ // otherwise
+ targets.splice(targets.end(), deferredTargets,
+ deferredTargets.begin(), it);
+ }
+ deferredTargets.populateFlags();
+ targets.populateFlags();
order = targets.front().order;
readyTime = std::max(curTick(), targets.front().readyTime);
return true;
}
+void
+MSHR::promoteIf(const std::function<bool (Target &)>& pred)
+{
+ // if any of the deferred targets were upper-level cache
+ // requests marked downstreamPending, need to clear that
+ assert(!downstreamPending); // not pending here anymore
+
+ // find the first target does not satisfy the condition
+ auto last_it = std::find_if_not(deferredTargets.begin(),
+ deferredTargets.end(),
+ pred);
+
+ // for the prefix of the deferredTargets [begin(), last_it) clear
+ // the downstreamPending flag and move them to the target list
+ deferredTargets.clearDownstreamPending(deferredTargets.begin(),
+ last_it);
+ targets.splice(targets.end(), deferredTargets,
+ deferredTargets.begin(), last_it);
+ // We need to update the flags for the target lists after the
+ // modifications
+ deferredTargets.populateFlags();
+}
+
+void
+MSHR::promoteReadable()
+{
+ if (!deferredTargets.empty() && !hasPostInvalidate()) {
+ // We got a non invalidating response, and we have the block
+ // but we have deferred targets which are waiting and they do
+ // not need writable. This can happen if the original request
+ // was for a cache clean operation and we had a copy of the
+ // block. Since we serviced the cache clean operation and we
+ // have the block, there's no need to defer the targets, so
+ // move them up to the regular target list.
+
+ auto pred = [](Target &t) {
+ assert(t.source == Target::FromCPU);
+ return !t.pkt->req->isCacheInvalidate() &&
+ !t.pkt->needsWritable();
+ };
+ promoteIf(pred);
+ }
+}
void
-MSHR::handleFill(Packet *pkt, CacheBlk *blk)
+MSHR::promoteWritable()
{
- if (!pkt->sharedAsserted()
- && !(hasPostInvalidate() || hasPostDowngrade())
- && deferredTargets.needsExclusive) {
- // We got an exclusive response, but we have deferred targets
- // which are waiting to request an exclusive copy (not because
+ if (deferredTargets.needsWritable &&
+ !(hasPostInvalidate() || hasPostDowngrade())) {
+ // We got a writable response, but we have deferred targets
+ // which are waiting to request a writable copy (not because
// of a pending invalidate). This can happen if the original
- // request was for a read-only (non-exclusive) block, but we
- // got an exclusive copy anyway because of the E part of the
- // MOESI/MESI protocol. Since we got the exclusive copy
- // there's no need to defer the targets, so move them up to
- // the regular target list.
- assert(!targets.needsExclusive);
- targets.needsExclusive = true;
- // if any of the deferred targets were upper-level cache
- // requests marked downstreamPending, need to clear that
- assert(!downstreamPending); // not pending here anymore
- deferredTargets.clearDownstreamPending();
- // this clears out deferredTargets too
- targets.splice(targets.end(), deferredTargets);
- deferredTargets.resetFlags();
+ // request was for a read-only block, but we got a writable
+ // response anyway. Since we got the writable copy there's no
+ // need to defer the targets, so move them up to the regular
+ // target list.
+ assert(!targets.needsWritable);
+ targets.needsWritable = true;
+
+ auto pred = [](Target &t) {
+ assert(t.source == Target::FromCPU);
+ return !t.pkt->req->isCacheInvalidate();
+ };
+
+ promoteIf(pred);
}
}
bool
-MSHR::checkFunctional(PacketPtr pkt)
+MSHR::trySatisfyFunctional(PacketPtr pkt)
{
// For printing, we treat the MSHR as a whole as single entity.
// For other requests, we iterate over the individual targets
// since that's where the actual data lies.
if (pkt->isPrint()) {
- pkt->checkFunctional(this, blkAddr, isSecure, blkSize, NULL);
+ pkt->trySatisfyFunctional(this, blkAddr, isSecure, blkSize, nullptr);
return false;
} else {
- return (targets.checkFunctional(pkt) ||
- deferredTargets.checkFunctional(pkt));
+ return (targets.trySatisfyFunctional(pkt) ||
+ deferredTargets.trySatisfyFunctional(pkt));
}
}
+bool
+MSHR::sendPacket(BaseCache &cache)
+{
+ return cache.sendMSHRQueuePacket(this);
+}
void
MSHR::print(std::ostream &os, int verbosity, const std::string &prefix) const
{
- ccprintf(os, "%s[%#llx:%#llx](%s) %s %s %s state: %s %s %s %s %s\n",
+ ccprintf(os, "%s[%#llx:%#llx](%s) %s %s %s state: %s %s %s %s %s %s\n",
prefix, blkAddr, blkAddr + blkSize - 1,
isSecure ? "s" : "ns",
isForward ? "Forward" : "",
- isForwardNoResponse() ? "ForwNoResp" : "",
- needsExclusive() ? "Excl" : "",
+ allocOnFill() ? "AllocOnFill" : "",
+ needsWritable() ? "Wrtbl" : "",
_isUncacheable ? "Unc" : "",
inService ? "InSvc" : "",
downstreamPending ? "DwnPend" : "",
- hasPostInvalidate() ? "PostInv" : "",
- hasPostDowngrade() ? "PostDowngr" : "");
+ postInvalidate ? "PostInv" : "",
+ postDowngrade ? "PostDowngr" : "",
+ hasFromCache() ? "HasFromCache" : "");
- ccprintf(os, "%s Targets:\n", prefix);
- targets.print(os, verbosity, prefix + " ");
+ if (!targets.empty()) {
+ ccprintf(os, "%s Targets:\n", prefix);
+ targets.print(os, verbosity, prefix + " ");
+ }
if (!deferredTargets.empty()) {
ccprintf(os, "%s Deferred Targets:\n", prefix);
deferredTargets.print(os, verbosity, prefix + " ");
std::string
MSHR::print() const
{
- ostringstream str;
+ std::ostringstream str;
print(str);
return str.str();
}