RandomX: try to recover after an error

src/log.h

@@ -288,7 +288,7 @@ template<> struct log::Stream::Entry<const_buf>
 
 struct hex_buf
 {
-	FORCEINLINE hex_buf(const uint8_t* data, size_t size) : m_data(data), m_size(size) {}
+	FORCEINLINE hex_buf(const void* data, size_t size) : m_data(reinterpret_cast<const uint8_t*>(data)), m_size(size) {}
 
 	template<typename T>
 	explicit FORCEINLINE hex_buf(const T* data) : m_data(reinterpret_cast<const uint8_t*>(data)), m_size(sizeof(T)) {}

src/miner.cpp

@@ -224,7 +224,16 @@ void Miner::run(WorkerData* data)
 		job[index].set_nonce(static_cast<uint32_t>(full_nonce), static_cast<uint32_t>(full_nonce >> 32));
 
 		hash h;
-		randomx_calculate_hash_next(vm, job[index].m_blob, job[index].m_blobSize, &h);
+		try {
+			randomx_calculate_hash_next(vm, job[index].m_blob, job[index].m_blobSize, &h);
+		}
+		catch (const std::exception& e) {
+			LOGERR(0, "Failed to calculate RandomX hash: exception \"" << e.what() << "\". Is your CPU/RAM unstable?" <<
+				"\nFailed RandomX hash input: " << log::hex_buf(j.m_blob, j.m_blobSize));
+
+			// Make the result hash all FF's to fail difficulty checks
+			memset(h.h, -1, HASH_SIZE);
+		}
 
 		if (j.m_diff.check_pow(h)) {
 			LOGINFO(0, log::Green() << "worker thread " << data->m_index << '/' << data->m_count << " found a mainchain block at height " << j.m_height << ", submitting it");

src/p2pool.cpp

@@ -1691,8 +1691,7 @@ int p2pool::run()
 		m_ZMQReader = nullptr;
 	}
 	catch (const std::exception& e) {
-		const char* s = e.what();
+		LOGERR(1, "exception " << e.what());
-		LOGERR(1, "exception " << s);
 		PANIC_STOP();
 	}
 

src/pool_block_parser.inl

@@ -393,8 +393,7 @@ int PoolBlock::deserialize(const uint8_t* data, size_t size, const SideChain& si
 		}
 	}
 	catch (std::exception& e) {
-		const char* msg = e.what();
+		LOGERR(0, "Exception in PoolBlock::deserialize(): " << e.what());
-		LOGERR(0, "Exception in PoolBlock::deserialize(): " << (msg ? msg : "unknown exception"));
 		return __LINE__;
 	}
 

src/pow_hash.cpp

@@ -322,6 +322,22 @@ void RandomX_Hasher::sync_wait()
 	ReadLock lock2(m_cacheLock);
 }
 
+static bool randomx_calculate_hash_safe(randomx_vm* machine, const void* input, size_t inputSize, void* output)
+{
+	// Try to calculate the hash again if something went wrong the first time (for example, because of an unstable CPU)
+	for (size_t i = 0; i < 2; ++i) {
+		try {
+			randomx_calculate_hash(machine, input, inputSize, output);
+			return true;
+		}
+		catch (const std::exception& e) {
+			LOGERR(0, "Failed to calculate RandomX hash: exception \"" << e.what() << "\". Is your CPU/RAM unstable?" <<
+				"\nFailed RandomX hash input: " << log::hex_buf(input, inputSize));
+		}
+	}
+	return false;
+}
+
 bool RandomX_Hasher::calculate(const void* data, size_t size, uint64_t /*height*/, const hash& seed, hash& result, bool force_light_mode)
 {
 	// First try to use the dataset if it's ready
@@ -335,8 +351,7 @@ bool RandomX_Hasher::calculate(const void* data, size_t size, uint64_t /*height*
 		MutexLock lock(m_vm[FULL_DATASET_VM].mutex);
 
 		if (m_vm[FULL_DATASET_VM].vm && (seed == m_seed[m_index])) {
-			randomx_calculate_hash(m_vm[FULL_DATASET_VM].vm, data, size, &result);
+			return randomx_calculate_hash_safe(m_vm[FULL_DATASET_VM].vm, data, size, &result);
-			return true;
 		}
 	}
 
@@ -350,8 +365,7 @@ bool RandomX_Hasher::calculate(const void* data, size_t size, uint64_t /*height*
 	{
 		MutexLock lock2(m_vm[m_index].mutex);
 		if (m_vm[m_index].vm && (seed == m_seed[m_index])) {
-			randomx_calculate_hash(m_vm[m_index].vm, data, size, &result);
+			return randomx_calculate_hash_safe(m_vm[m_index].vm, data, size, &result);
-			return true;
 		}
 	}
 
@@ -360,8 +374,7 @@ bool RandomX_Hasher::calculate(const void* data, size_t size, uint64_t /*height*
 	MutexLock lock2(m_vm[prev_index].mutex);
 
 	if (m_vm[prev_index].vm && (seed == m_seed[prev_index])) {
-		randomx_calculate_hash(m_vm[prev_index].vm, data, size, &result);
+		return randomx_calculate_hash_safe(m_vm[prev_index].vm, data, size, &result);
-		return true;
 	}
 
 	return false;