christoph/fail
1
Fork 0
Code Actions Packages Releases Activity

checksum-oostubs: single experiments work now

Browse Source 
next: campaign implementation

git-svn-id: https://www4.informatik.uni-erlangen.de/i4svn/danceos/trunk/devel/fail@1033 8c4709b5-6ec9-48aa-a5cd-a96041d1645a
This commit is contained in:
hsc
2012-04-05 15:37:08 +00:00
parent c0f3aac4b7
commit 08aa7ff3a3
4 changed files with 131 additions and 65 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
core/experiments/checksum-oostubs/campaign.cc
View File

@ -88,9 +88,11 @@ bool CoolChecksumCampaign::run()
for (int bit_offset = 0; bit_offset < (memoryMap[member][1])*8; ++bit_offset) { for (int bit_offset = 0; bit_offset < (memoryMap[member][1])*8; ++bit_offset) {
for (int instr_offset = 0; instr_offset < OOSTUBS_NUMINSTR; ++instr_offset) { for (int instr_offset = 0; instr_offset < OOSTUBS_NUMINSTR; ++instr_offset) {
CoolChecksumExperimentData *d = new CoolChecksumExperimentData; CoolChecksumExperimentData *d = new CoolChecksumExperimentData;
/*
d->msg.set_instr_offset(instr_offset); d->msg.set_instr_offset(instr_offset);
d->msg.set_mem_addr(memoryMap[member][0]); d->msg.set_mem_addr(memoryMap[member][0]);
d->msg.set_bit_offset(bit_offset); d->msg.set_bit_offset(bit_offset);
*/
fi::campaignmanager.addParam(d); fi::campaignmanager.addParam(d);
++count; ++count;
@ -122,6 +124,7 @@ bool CoolChecksumCampaign::run()
while ((res = static_cast<CoolChecksumExperimentData *>(fi::campaignmanager.getDone()))) { while ((res = static_cast<CoolChecksumExperimentData *>(fi::campaignmanager.getDone()))) {
rescount++; rescount++;
/*
results results
<< res->msg.injection_ip() << "\t" << res->msg.injection_ip() << "\t"
<< res->msg.instr_offset() << "\t" << res->msg.instr_offset() << "\t"
@ -131,6 +134,7 @@ bool CoolChecksumCampaign::run()
<< res->msg.resultdata() << "\t" << res->msg.resultdata() << "\t"
<< res->msg.error_corrected() << "\t" << res->msg.error_corrected() << "\t"
<< res->msg.details() << "\n"; << res->msg.details() << "\n";
*/
delete res; delete res;
} }
log << "done. sent " << count << " received " << rescount << endl; log << "done. sent " << count << " received " << rescount << endl;

56
core/experiments/checksum-oostubs/checksum-oostubs.proto
View File

@ -1,29 +1,41 @@
message OOStuBSProtoMsg { message OOStuBSProtoMsg {
// parameters // Input: experiment parameters
required int32 instr_offset = 1; required int32 instr_offset = 1;
required int32 mem_addr = 2; optional int32 instr_address = 2; // for sanity checks
required int32 bit_offset = 3; required int32 mem_addr = 3;
required int32 bit_offset = 4;
// ----------------------------------------------------
// Output: experiment results
// (make these optional to reduce overhead for server->client communication)
// results
// make these optional to reduce overhead for server->client communication
enum ResultType {
CALCDONE = 1;
TIMEOUT = 2;
TRAP = 3;
UNKNOWN = 4;
}
// instruction pointer where injection was done // instruction pointer where injection was done
optional uint32 injection_ip = 4; optional uint32 injection_ip = 5;
// result type, see above
optional ResultType resulttype = 5; // result type:
// result data, depending on resulttype: // FINISHED = planned number of instructions were executed
// CALCDONE: resultdata = calculated value // TRAP = premature guest "crash"
// TIMEOUT: resultdata = latest EIP enum ResultType {
// TRAP: resultdata = latest EIP FINISHED = 1;
// UNKNOWN: resultdata = latest EIP TRAP = 2;
optional uint32 resultdata = 6; UNKNOWN = 3;
}
optional ResultType resulttype = 6;
// result details:
// resultdata = result[0-2]
repeated uint32 resultdata = 7 [packed=true];
// was finish() ever reached?
optional bool finish_reached = 8;
// especially interesting for TRAP/ UNKNOWN: latest IP
optional uint32 latest_ip = 9;
// did ECC correct the fault? // did ECC correct the fault?
optional int32 error_corrected = 7; optional int32 error_corrected = 10;
// optional textual description of what happened // optional textual description of what happened
optional string details = 8; optional string details = 11;
} }

134
core/experiments/checksum-oostubs/experiment.cc
View File

@ -22,6 +22,7 @@ using std::endl;
bool CoolChecksumExperiment::run() bool CoolChecksumExperiment::run()
{ {
char const *statename = "checksum-oostubs.state";
Logger log("Checksum-OOStuBS", false); Logger log("Checksum-OOStuBS", false);
fi::BPEvent bp; fi::BPEvent bp;
@ -41,11 +42,11 @@ bool CoolChecksumExperiment::run()
log << "test function entry reached, saving state" << endl; log << "test function entry reached, saving state" << endl;
log << "EIP = " << std::hex << bp.getTriggerInstructionPointer() << " or " << sal::simulator.getRegisterManager().getInstructionPointer() << endl; log << "EIP = " << std::hex << bp.getTriggerInstructionPointer() << " or " << sal::simulator.getRegisterManager().getInstructionPointer() << endl;
log << "error_corrected = " << std::dec << ((int)sal::simulator.getMemoryManager().getByte(OOSTUBS_ERROR_CORRECTED)) << endl; log << "error_corrected = " << std::dec << ((int)sal::simulator.getMemoryManager().getByte(OOSTUBS_ERROR_CORRECTED)) << endl;
sal::simulator.save("checksum-oostubs.state"); sal::simulator.save(statename);
#elif 1 #elif 1
// STEP 2: record trace for fault-space pruning // STEP 2: record trace for fault-space pruning
log << "restoring state" << endl; log << "restoring state" << endl;
sal::simulator.restore("checksum-oostubs.state"); sal::simulator.restore(statename);
log << "EIP = " << std::hex << sal::simulator.getRegisterManager().getInstructionPointer() << endl; log << "EIP = " << std::hex << sal::simulator.getRegisterManager().getInstructionPointer() << endl;
log << "enabling tracing" << endl; log << "enabling tracing" << endl;
@ -67,7 +68,15 @@ bool CoolChecksumExperiment::run()
bp.setWatchInstructionPointer(fi::ANY_ADDR); bp.setWatchInstructionPointer(fi::ANY_ADDR);
bp.setCounter(OOSTUBS_NUMINSTR); bp.setCounter(OOSTUBS_NUMINSTR);
sal::simulator.addEventAndWait(&bp); sal::simulator.addEvent(&bp);
fi::BPEvent func_finish(OOSTUBS_FUNC_FINISH);
sal::simulator.addEvent(&func_finish);
if (sal::simulator.waitAny() == &func_finish) {
log << "experiment reached finish()" << endl;
// FIXME add instruction counter to SimulatorController
sal::simulator.waitAny();
}
log << "experiment finished after " << std::dec << OOSTUBS_NUMINSTR << " instructions" << endl; log << "experiment finished after " << std::dec << OOSTUBS_NUMINSTR << " instructions" << endl;
uint32_t results[OOSTUBS_RESULTS_BYTES / sizeof(uint32_t)]; uint32_t results[OOSTUBS_RESULTS_BYTES / sizeof(uint32_t)];
@ -93,83 +102,124 @@ bool CoolChecksumExperiment::run()
// FIXME consider moving experiment repetition into Fail* or even the // FIXME consider moving experiment repetition into Fail* or even the
// SAL -- whether and how this is possible with the chosen backend is // SAL -- whether and how this is possible with the chosen backend is
// backend specific // backend specific
for (int i = 0; i < 20; ++i) { //for (int i = 0; i < 2000; ++i) {
// STEP 3: The actual experiment. // STEP 3: The actual experiment.
log << "restoring state" << endl; log << "restoring state" << endl;
sal::simulator.restore("coolecc.state"); sal::simulator.restore(statename);
// get an experiment parameter set
log << "asking job server for experiment parameters" << endl; log << "asking job server for experiment parameters" << endl;
CoolChecksumExperimentData param; CoolChecksumExperimentData param;
/*
if (!m_jc.getParam(param)) { if (!m_jc.getParam(param)) {
log << "Dying." << endl; log << "Dying." << endl;
// communicate that we were told to die // communicate that we were told to die
sal::simulator.terminate(1); sal::simulator.terminate(1);
} }
*/
param.msg.set_instr_offset(1);
param.msg.set_mem_addr(1024*1024*8);
param.msg.set_bit_offset(0);
int id = param.getWorkloadID(); int id = param.getWorkloadID();
int instr_offset = param.msg.instr_offset(); int instr_offset = param.msg.instr_offset();
int mem_addr = param.msg.mem_addr();
int bit_offset = param.msg.bit_offset(); int bit_offset = param.msg.bit_offset();
log << "job " << id << " instr " << instr_offset << " bit " << bit_offset << endl; log << "job " << id << " instr " << instr_offset << " mem " << mem_addr << "+" << bit_offset << endl;
// FIXME could be improved (especially for backends supporting // reaching finish() could happen before OR after FI
// breakpoints natively) by utilizing a previously recorded instruction fi::BPEvent func_finish(OOSTUBS_FUNC_FINISH);
// trace sal::simulator.addEvent(&func_finish);
bool finish_reached = false;
// XXX test this with coolchecksum first (or reassure with sanity checks)
// XXX could be improved with intermediate states (reducing runtime until injection)
bp.setWatchInstructionPointer(fi::ANY_ADDR); bp.setWatchInstructionPointer(fi::ANY_ADDR);
for (int count = 0; count < instr_offset; ++count) { bp.setCounter(instr_offset);
sal::simulator.addEventAndWait(&bp); sal::simulator.addEvent(&bp);
// finish() before FI?
if (sal::simulator.waitAny() == &func_finish) {
finish_reached = true;
log << "experiment reached finish() before FI" << endl;
// wait for bp
sal::simulator.waitAny();
} }
// inject // --- fault injection ---
sal::guest_address_t inject_addr = COOL_ECC_OBJUNDERTEST + bit_offset / 8;
sal::MemoryManager& mm = sal::simulator.getMemoryManager(); sal::MemoryManager& mm = sal::simulator.getMemoryManager();
sal::byte_t data = mm.getByte(inject_addr); sal::byte_t data = mm.getByte(mem_addr);
sal::byte_t newdata = data ^ (1 << (bit_offset % 8)); sal::byte_t newdata = data ^ (1 << bit_offset);
mm.setByte(inject_addr, newdata); mm.setByte(mem_addr, newdata);
// note at what IP we did it // note at what IP we did it
int32_t injection_ip = sal::simulator.getRegisterManager().getInstructionPointer(); int32_t injection_ip = sal::simulator.getRegisterManager().getInstructionPointer();
param.msg.set_injection_ip(injection_ip); param.msg.set_injection_ip(injection_ip);
log << "inject @ ip " << injection_ip log << "fault injected @ ip " << injection_ip
<< " offset " << std::dec << (bit_offset / 8) << " 0x" << std::hex << ((int)data) << " -> 0x" << ((int)newdata) << endl;
<< " (bit " << (bit_offset % 8) << ") 0x" // XXX sanity check
<< std::hex << ((int)data) << " -> 0x" << ((int)newdata) << endl;
// aftermath // --- aftermath ---
fi::BPEvent ev_done(COOL_ECC_CALCDONE); // four possible outcomes:
sal::simulator.addEvent(&ev_done); // - guest causes a trap, "crashes"
fi::BPEvent ev_timeout(fi::ANY_ADDR); // - guest runs OOSTUBS_NUMINSTR+OOSTUBS_RECOVERYINSTR instructions but
ev_timeout.setCounter(COOL_ECC_NUMINSTR + 3000); // never reaches finish()
sal::simulator.addEvent(&ev_timeout); // - guest reaches finish() within OOSTUBS_NUMINSTR+OOSTUBS_RECOVERYINSTR
// instructions with
// * a wrong result[0-2]
// * a correct result[0-2]
// catch traps as "extraordinary" ending
fi::TrapEvent ev_trap(fi::ANY_TRAP); fi::TrapEvent ev_trap(fi::ANY_TRAP);
sal::simulator.addEvent(&ev_trap); sal::simulator.addEvent(&ev_trap);
// remaining instructions until "normal" ending
fi::BPEvent ev_done(fi::ANY_ADDR);
ev_done.setCounter(OOSTUBS_NUMINSTR + OOSTUBS_RECOVERYINSTR - instr_offset);
sal::simulator.addEvent(&ev_done);
fi::BaseEvent* ev = sal::simulator.waitAny(); fi::BaseEvent* ev = sal::simulator.waitAny();
// Do we reach finish() while waiting for ev_trap/ev_done?
if (ev == &func_finish) {
finish_reached = true;
log << "experiment reached finish()" << endl;
// wait for ev_trap/ev_done
ev = sal::simulator.waitAny();
}
// record resultdata, finish_reached and error_corrected regardless of result
uint32_t results[OOSTUBS_RESULTS_BYTES / sizeof(uint32_t)];
sal::simulator.getMemoryManager().getBytes(OOSTUBS_RESULTS_ADDR, sizeof(results), results);
for (unsigned i = 0; i < sizeof(results) / sizeof(*results); ++i) {
log << "results[" << i << "]: " << std::dec << results[i] << endl;
param.msg.add_resultdata(results[i]);
}
param.msg.set_finish_reached(finish_reached);
int32_t error_corrected = sal::simulator.getMemoryManager().getByte(OOSTUBS_ERROR_CORRECTED);
param.msg.set_error_corrected(error_corrected);
param.msg.set_latest_ip(sal::simulator.getRegisterManager().getInstructionPointer());
if (ev == &ev_done) { if (ev == &ev_done) {
int32_t data = sal::simulator.getRegisterManager().getRegister(targetreg)->getData(); log << std::dec << "Result FINISHED" << endl;
log << std::dec << "Result EDX = " << data << endl; param.msg.set_resulttype(param.msg.FINISHED);
param.msg.set_resulttype(CoolChecksumProtoMsg_ResultType_CALCDONE);
param.msg.set_resultdata(data);
} else if (ev == &ev_timeout) {
log << std::dec << "Result TIMEOUT" << endl;
param.msg.set_resulttype(CoolChecksumProtoMsg_ResultType_TIMEOUT);
param.msg.set_resultdata(sal::simulator.getRegisterManager().getInstructionPointer());
} else if (ev == &ev_trap) { } else if (ev == &ev_trap) {
log << std::dec << "Result TRAP #" << ev_trap.getTriggerNumber() << endl; log << std::dec << "Result TRAP #" << ev_trap.getTriggerNumber() << endl;
param.msg.set_resulttype(CoolChecksumProtoMsg_ResultType_TRAP); param.msg.set_resulttype(param.msg.TRAP);
param.msg.set_resultdata(sal::simulator.getRegisterManager().getInstructionPointer());
} else { } else {
log << std::dec << "Result WTF?" << endl; log << std::dec << "Result WTF?" << endl;
param.msg.set_resulttype(CoolChecksumProtoMsg_ResultType_UNKNOWN); param.msg.set_resulttype(param.msg.UNKNOWN);
param.msg.set_resultdata(sal::simulator.getRegisterManager().getInstructionPointer());
std::stringstream ss; std::stringstream ss;
ss << "eventid " << ev << " EIP " << sal::simulator.getRegisterManager().getInstructionPointer(); ss << "eventid " << ev->getId() << " EIP " << sal::simulator.getRegisterManager().getInstructionPointer();
param.msg.set_details(ss.str()); param.msg.set_details(ss.str());
} }
int32_t error_corrected = sal::simulator.getMemoryManager().getByte(COOL_ECC_ERROR_CORRECTED); /*
param.msg.set_error_corrected(error_corrected);
m_jc.sendResult(param); m_jc.sendResult(param);
*/
} //}
#endif #endif
// Explicitly terminate, or the simulator will continue to run. // Explicitly terminate, or the simulator will continue to run.
sal::simulator.terminate(); sal::simulator.terminate();

2
core/experiments/checksum-oostubs/experimentInfo.hpp
View File

@ -10,7 +10,7 @@
#define OOSTUBS_FUNC_ENTRY 0x00103f2c #define OOSTUBS_FUNC_ENTRY 0x00103f2c
// empty function that is called explicitly when the experiment finished // empty function that is called explicitly when the experiment finished
// nm -C ecc.elf|fgrep "finished()" // nm -C ecc.elf|fgrep "finished()"
#define OOSTUBS_FUNC_DONE 0x001093f0 #define OOSTUBS_FUNC_FINISH 0x001093f0
// number of instructions the target executes under non-error conditions from ENTRY to DONE: // number of instructions the target executes under non-error conditions from ENTRY to DONE:
// (result of experiment's step #2) // (result of experiment's step #2)
#define OOSTUBS_NUMINSTR 0x4a3401 #define OOSTUBS_NUMINSTR 0x4a3401