Add QutexAcquisitionHistoryTracker; integrate plumbing

We add the new Qutex acquisision history tracker that allows us
to dynamically detect qutex gridlocks. We've integrated it into
LockerAndInvoker::operator() in a preliminary way.

We also moved all of the trace*ForGridlockOn() methods into the
new QutexAcquisitionHistoryTracker singleton class. They're
more appropriately located there. They're still unimplemented
though.
This commit is contained in:
2025-09-29 19:27:02 -04:00
parent 8123ec1227
commit 71564b4d83
6 changed files with 325 additions and 106 deletions
@@ -13,8 +13,12 @@ namespace smo {
*
* The chain walking logic can use dynamic_cast to determine the most
* derived type and perform appropriate operations.
*
* Inherits from enable_shared_from_this to allow objects to obtain a
* shared_ptr to themselves, which is useful for gridlock detection tracking.
*/
class AsynchronousContinuationChainLink
: public std::enable_shared_from_this<AsynchronousContinuationChainLink>
{
public:
virtual ~AsynchronousContinuationChainLink() = default;
+126
View File
@@ -0,0 +1,126 @@
#ifndef QUTEX_ACQUISITION_HISTORY_TRACKER_H
#define QUTEX_ACQUISITION_HISTORY_TRACKER_H
#include <unordered_map>
#include <memory>
#include <forward_list>
#include <functional>
namespace smo {
// Forward declarations
class Qutex;
class AsynchronousContinuationChainLink;
/**
* @brief QutexAcquisitionHistoryTracker - Tracks acquisition history for
* gridlock detection
*
* This class maintains a central acquisition history to track all lockvokers
* suspected of being gridlocked. It stores information about what locks each
* timed-out lockvoker wants and what locks they hold in their continuation
* history.
*/
class QutexAcquisitionHistoryTracker
{
public:
/**
* @brief Type definition for the acquisition history entry
*
* pair.first: The firstFailedQutex that this lockvoker WANTS but can't
* acquire
* pair.second: A unique_ptr to a list of all acquired Qutexes in this
* lockvoker's continuation history
*/
typedef std::pair<
std::reference_wrapper<Qutex>,
std::unique_ptr<std::forward_list<std::reference_wrapper<Qutex>>>
> AcquisitionHistoryEntry;
/**
* @brief Type definition for the acquisition history map
*
* Key: std::shared_ptr<AsynchronousContinuationChainLink>
* (the continuation that contains the timed-out lockvoker)
* Value: AcquisitionHistoryEntry
* (its wanted lock (aka: firstFailedQutex/pair.first) + held locks)
*/
typedef std::unordered_map<
std::shared_ptr<AsynchronousContinuationChainLink>,
AcquisitionHistoryEntry
> AcquisitionHistoryMap;
public:
static QutexAcquisitionHistoryTracker& getInstance()
{
static QutexAcquisitionHistoryTracker instance;
return instance;
}
/**
* @brief Add a continuation to the acquisition history if it doesn't
* already exist
* @param continuation Shared pointer to the
* AsynchronousContinuationChainLink
* @param wantedLock The lock that this continuation wants but can't
* acquire
* @param heldLocks Unique pointer to list of locks held in this
* continuation's history (will be moved)
*/
void addIfNotExists(
std::shared_ptr<AsynchronousContinuationChainLink> &continuation,
Qutex& wantedLock,
std::unique_ptr<std::forward_list<std::reference_wrapper<Qutex>>>
heldLocks
)
{
auto it = acquisitionHistory.find(continuation);
// If a continuation already exists, don't add it again
if (it != acquisitionHistory.end()) {
return;
}
acquisitionHistory.emplace(continuation, std::make_pair(
std::ref(wantedLock), std::move(heldLocks)));
}
/**
* @brief Remove a continuation from the acquisition history
* @param continuation Shared pointer to the AsynchronousContinuationChainLink
* to remove
* @return true if the continuation was found and removed, false if not found
*/
bool remove(std::shared_ptr<AsynchronousContinuationChainLink> &continuation)
{
auto it = acquisitionHistory.find(continuation);
if (it != acquisitionHistory.end()) {
acquisitionHistory.erase(it);
return true;
}
return false;
}
bool heuristicallyTraceContinuationHistoryForGridlockOn(
Qutex &firstFailedQutex) const;
bool completelyTraceContinuationHistoryForGridlockOn(
Qutex &firstFailedQutex) const;
// Disable copy constructor and assignment operator
QutexAcquisitionHistoryTracker(
const QutexAcquisitionHistoryTracker&) = delete;
QutexAcquisitionHistoryTracker& operator=(
const QutexAcquisitionHistoryTracker&) = delete;
private:
QutexAcquisitionHistoryTracker() = default;
~QutexAcquisitionHistoryTracker() = default;
private:
AcquisitionHistoryMap acquisitionHistory;
};
} // namespace smo
#endif // QUTEX_ACQUISITION_HISTORY_TRACKER_H
+54 -6
View File
@@ -12,6 +12,7 @@
#include <asynchronousContinuation.h>
#include <lockerAndInvokerBase.h>
#include <callback.h>
#include <qutexAcquisitionHistoryTracker.h>
namespace smo {
@@ -36,6 +37,10 @@ public:
std::forward<Args>(args)...);
}
// Return list of all qutexes in predecessors' LockSets; excludes self.
std::unique_ptr<std::forward_list<std::reference_wrapper<Qutex>>>
getAcquiredQutexHistory() const;
public:
LockSet<OriginalCbFnT> requiredLocks;
std::atomic<bool> isAwakeOrBeingAwakened{false};
@@ -116,8 +121,33 @@ public:
bool isDeadlock = traceContinuationHistoryForDeadlockOn(
firstFailedQutex);
bool isGridlock = heuristicallyTraceContinuationHistoryForGridlockOn(
firstFailedQutex);
bool gridlockIsHeuristicallyLikely = false;
bool gridlockIsAlgorithmicallyLikely = false;
if (gridlockLikely)
{
auto tracker = QutexAcquisitionHistoryTracker
::getInstance();
auto heldLocks = serializedContinuation
.getAcquiredQutexHistory();
// Add this continuation to the tracker
tracker.addIfNotExists(
serializedContinuation.shared_from_this(),
firstFailedQutex, std::move(heldLocks));
gridlockIsHeuristicallyLikely = tracker
.heuristicallyTraceContinuationHistoryForGridlockOn(
firstFailedQutex);
gridlockIsAlgorithmicallyLikely = tracker
.completelyTraceContinuationHistoryForGridlockOn(
firstFailedQutex);
}
bool isGridlock = (gridlockIsHeuristicallyLikely
|| gridlockIsAlgorithmicallyLikely);
if (!isDeadlock && !isGridlock)
{ return; }
@@ -145,6 +175,28 @@ public:
* can't acquire the locks anyway.
*/
serializedContinuation.requiredLocks.unregisterFromQutexQueues();
#ifdef CONFIG_ENABLE_DEBUG_LOCKS
/** EXPLANATION:
* If we were being tracked for gridlock detection but successfully
* acquired all locks, it was a false positive due to timed delay,
* long-running operation, or I/O delay
*/
if (gridlockLikely)
{
bool removed = QutexAcquisitionHistoryTracker::getInstance()
.remove(serializedContinuation.shared_from_this());
if (removed)
{
std::cerr << "LockerAndInvoker::operator(): False positive gridlock "
"detection - continuation was being tracked but successfully "
"acquired all locks. This was likely due to timed delay, "
"long-running operation, or I/O delay." << std::endl;
}
}
#endif
invocationTarget();
}
@@ -227,10 +279,6 @@ public:
};
bool traceContinuationHistoryForDeadlockOn(Qutex &firstFailedQutex);
bool heuristicallyTraceContinuationHistoryForGridlockOn(
Qutex &firstFailedQutex);
bool completelyTraceContinuationHistoryForGridlockOn(
Qutex &firstFailedQutex);
bool traceContinuationHistoryForDeadlock(void)
{
for (auto& lockUsageDesc