changeset b7bc5b1084a4 in /z/repo/gem5
details: http://repo.gem5.org/gem5?cmd=changeset;node=b7bc5b1084a4
description:
arm: Add stats to table walker
This patch adds table walker stats for:
- Walk events
- Instruction vs Data
- Page size histogram
- Wait time and service time histograms
- Pending requests histogram (per cycle) - measures dist. of L
(p(1..) = how often busy, p(0) = how often idle)
- Squashes, before starting and after completion
diffstat:
src/arch/arm/table_walker.cc | 186 ++++++++++++++++++++++++++++++++++++++++++-
src/arch/arm/table_walker.hh | 31 +++++++
src/dev/dma_device.cc | 12 ++-
src/dev/dma_device.hh | 4 +-
4 files changed, 225 insertions(+), 8 deletions(-)
diffs (truncated from 476 to 300 lines):
diff -r 74834c49fbbe -r b7bc5b1084a4 src/arch/arm/table_walker.cc
--- a/src/arch/arm/table_walker.cc Tue Dec 23 09:31:18 2014 -0500
+++ b/src/arch/arm/table_walker.cc Tue Dec 23 09:31:18 2014 -0500
@@ -60,6 +60,8 @@
stage2Mmu(NULL), isStage2(p->is_stage2), tlb(NULL),
currState(NULL), pending(false), masterId(p->sys->getMasterId(name())),
numSquashable(p->num_squash_per_cycle),
+ pendingReqs(0),
+ pendingChangeTick(curTick()),
doL1DescEvent(this), doL2DescEvent(this),
doL0LongDescEvent(this), doL1LongDescEvent(this),
doL2LongDescEvent(this),
doL3LongDescEvent(this),
@@ -151,6 +153,7 @@
if (params()->sys->isTimingMode() && currState) {
delete currState;
currState = NULL;
+ pendingChange();
}
}
@@ -170,6 +173,8 @@
bool secure, TLB::ArmTranslationType tranType)
{
assert(!(_functional && _timing));
+ ++statWalks;
+
WalkerState *savedCurrState = NULL;
if (!currState && !_functional) {
@@ -196,10 +201,13 @@
// this fault to re-execute the faulting instruction which should clean
// up everything.
if (currState->vaddr_tainted == _req->getVaddr()) {
+ ++statSquashedBefore;
return std::make_shared<ReExec>();
}
}
+ pendingChange();
+ currState->startTime = curTick();
currState->tc = _tc;
currState->aarch64 = opModeIs64(currOpMode(_tc));
currState->el = currEL(_tc);
@@ -261,6 +269,8 @@
currState->isFetch = (currState->mode == TLB::Execute);
currState->isWrite = (currState->mode == TLB::Write);
+ statRequestOrigin[REQUESTED][currState->isFetch]++;
+
// We only do a second stage of translation if we're not secure, or in
// hyp mode, the second stage MMU is enabled, and this table walker
// instance is the first stage.
@@ -280,6 +290,10 @@
currState->userTable = true;
currState->xnTable = false;
currState->pxnTable = false;
+
+ ++statWalksLongDescriptor;
+ } else {
+ ++statWalksShortDescriptor;
}
if (!currState->timing) {
@@ -303,8 +317,10 @@
if (pending || pendingQueue.size()) {
pendingQueue.push_back(currState);
currState = NULL;
+ pendingChange();
} else {
pending = true;
+ pendingChange();
if (currState->aarch64)
return processWalkAArch64();
else if (long_desc_format)
@@ -321,6 +337,7 @@
{
assert(!currState);
assert(pendingQueue.size());
+ pendingChange();
currState = pendingQueue.front();
ExceptionLevel target_el = EL0;
@@ -372,6 +389,7 @@
(currState->transState->squashed() || te)) {
pendingQueue.pop_front();
num_squashed++;
+ statSquashedBefore++;
DPRINTF(TLB, "Squashing table walk for address %#x\n",
currState->vaddr_tainted);
@@ -383,6 +401,7 @@
currState->req, currState->tc, currState->mode);
} else {
// translate the request now that we know it will work
+ statWalkServiceTime.sample(curTick() - currState->startTime);
tlb->translateTiming(currState->req, currState->tc,
currState->transState, currState->mode);
@@ -402,8 +421,9 @@
currState = NULL;
}
}
+ pendingChange();
- // if we've still got pending translations schedule more work
+ // if we still have pending translations, schedule more work
nextWalk(tc);
currState = NULL;
}
@@ -420,6 +440,8 @@
currState->vaddr_tainted, currState->ttbcr,
mbits(currState->vaddr, 31,
32 -
currState->ttbcr.n));
+ statWalkWaitTime.sample(curTick() - currState->startTime);
+
if (currState->ttbcr.n == 0 || !mbits(currState->vaddr, 31,
32 - currState->ttbcr.n)) {
DPRINTF(TLB, " - Selecting TTBR0\n");
@@ -511,6 +533,8 @@
DPRINTF(TLB, "Beginning table walk for address %#x, TTBCR: %#x\n",
currState->vaddr_tainted, currState->ttbcr);
+ statWalkWaitTime.sample(curTick() - currState->startTime);
+
Request::Flags flag = 0;
if (currState->isSecure)
flag.set(Request::SECURE);
@@ -700,6 +724,8 @@
static const GrainSize GrainMap_EL1_tg1[] =
{ ReservedGrain, Grain16KB, Grain4KB, Grain64KB };
+ statWalkWaitTime.sample(curTick() - currState->startTime);
+
// Determine TTBR, table size, granule size and phys. address range
Addr ttbr = 0;
int tsz = 0, ps = 0;
@@ -920,8 +946,8 @@
panic("Invalid table lookup level");
break;
}
- port.dmaAction(MemCmd::ReadReq, desc_addr, sizeof(uint64_t), event,
- (uint8_t*) &currState->longDesc.data,
+ port.dmaAction(MemCmd::ReadReq, desc_addr, sizeof(uint64_t),
+ event, (uint8_t*) &currState->longDesc.data,
currState->tc->getCpuPtr()->clockPeriod(), flag);
DPRINTF(TLBVerbose,
"Adding to walker fifo: queue size before adding: %d\n",
@@ -1698,6 +1724,7 @@
if (currState->fault != NoFault) {
currState->transState->finish(currState->fault, currState->req,
currState->tc, currState->mode);
+ statWalksShortTerminatedAtLevel[0]++;
pending = false;
nextWalk(currState->tc);
@@ -1711,9 +1738,11 @@
// delay is not set so there is no L2 to do
// Don't finish the translation if a stage 2 look up is underway
if (!currState->doingStage2) {
+ statWalkServiceTime.sample(curTick() - currState->startTime);
DPRINTF(TLBVerbose, "calling translateTiming again\n");
currState->fault = tlb->translateTiming(currState->req,
currState->tc,
currState->transState, currState->mode);
+ statWalksShortTerminatedAtLevel[0]++;
}
pending = false;
@@ -1749,13 +1778,16 @@
if (currState->fault != NoFault) {
currState->transState->finish(currState->fault, currState->req,
currState->tc, currState->mode);
+ statWalksShortTerminatedAtLevel[1]++;
}
else {
// Don't finish the translation if a stage 2 look up is underway
if (!currState->doingStage2) {
+ statWalkServiceTime.sample(curTick() - currState->startTime);
DPRINTF(TLBVerbose, "calling translateTiming again\n");
currState->fault = tlb->translateTiming(currState->req,
currState->tc, currState->transState, currState->mode);
+ statWalksShortTerminatedAtLevel[1]++;
}
}
@@ -1832,9 +1864,11 @@
// Don't finish the translation if a stage 2 look up is underway
if (!currState->doingStage2) {
DPRINTF(TLBVerbose, "calling translateTiming again\n");
+ statWalkServiceTime.sample(curTick() - currState->startTime);
currState->fault = tlb->translateTiming(currState->req,
currState->tc,
currState->transState,
currState->mode);
+ statWalksLongTerminatedAtLevel[(unsigned) curr_lookup_level]++;
}
pending = false;
@@ -1906,7 +1940,7 @@
} else {
if (isTiming) {
port.dmaAction(MemCmd::ReadReq, descAddr, numBytes, event, data,
- currState->tc->getCpuPtr()->clockPeriod(), flags);
+ currState->tc->getCpuPtr()->clockPeriod(),flags);
if (queueIndex >= 0) {
DPRINTF(TLBVerbose, "Adding to walker fifo: queue size before
adding: %d\n",
stateQueues[queueIndex].size());
@@ -1956,6 +1990,9 @@
else
te.el = 1;
+ statPageSizes[pageSizeNtoStatBin(te.N)]++;
+ statRequestOrigin[COMPLETED][currState->isFetch]++;
+
// ASID has no meaning for stage 2 TLB entries, so mark all stage 2 entries
// as global
te.global = descriptor.global(currState) || isStage2;
@@ -2024,3 +2061,144 @@
panic("Invalid lookup level conversion");
}
}
+
+/* this method keeps track of the table walker queue's residency, so
+ * needs to be called whenever requests start and complete. */
+void
+TableWalker::pendingChange()
+{
+ unsigned n = pendingQueue.size();
+ if ((currState != NULL) && (currState != pendingQueue.front())) {
+ ++n;
+ }
+
+ if (n != pendingReqs) {
+ Tick now = curTick();
+ statPendingWalks.sample(pendingReqs, now - pendingChangeTick);
+ pendingReqs = n;
+ pendingChangeTick = now;
+ }
+}
+
+uint8_t
+TableWalker::pageSizeNtoStatBin(uint8_t N)
+{
+ /* for statPageSizes */
+ switch(N) {
+ case 12: return 0; // 4K
+ case 14: return 1; // 16K (using 16K granule in v8-64)
+ case 16: return 2; // 64K
+ case 20: return 3; // 1M
+ case 21: return 4; // 2M-LPAE
+ case 24: return 5; // 16M
+ case 25: return 6; // 32M (using 16K granule in v8-64)
+ case 29: return 7; // 512M (using 64K granule in v8-64)
+ case 30: return 8; // 1G-LPAE
+ default:
+ panic("unknown page size");
+ return 255;
+ }
+}
+
+void
+TableWalker::regStats()
+{
+ statWalks
+ .name(name() + ".walks")
+ .desc("Table walker walks requested")
+ ;
+
+ statWalksShortDescriptor
+ .name(name() + ".walksShort")
+ .desc("Table walker walks initiated with short descriptors")
+ .flags(Stats::nozero)
+ ;
+
+ statWalksLongDescriptor
+ .name(name() + ".walksLong")
+ .desc("Table walker walks initiated with long descriptors")
+ .flags(Stats::nozero)
+ ;
+
+ statWalksShortTerminatedAtLevel
+ .init(2)
+ .name(name() + ".walksShortTerminationLevel")
+ .desc("Level at which table walker walks "
+ "with short descriptors terminate")
+ .flags(Stats::nozero)
+ ;
+ statWalksShortTerminatedAtLevel.subname(0, "Level1");
+ statWalksShortTerminatedAtLevel.subname(1, "Level2");
+
+ statWalksLongTerminatedAtLevel
+ .init(4)
+ .name(name() + ".walksLongTerminationLevel")
+ .desc("Level at which table walker walks "
+ "with long descriptors terminate")
+ .flags(Stats::nozero)
+ ;
+ statWalksLongTerminatedAtLevel.subname(0, "Level0");
+ statWalksLongTerminatedAtLevel.subname(1, "Level1");
+ statWalksLongTerminatedAtLevel.subname(2, "Level2");
+ statWalksLongTerminatedAtLevel.subname(3, "Level3");
+
_______________________________________________
gem5-dev mailing list
gem5-dev@gem5.org
http://m5sim.org/mailman/listinfo/gem5-dev
