fix duplicated nonce usage

- avoid that a nonce which not fits to the current job is used (check jobId after start nonce is consumed)
- move jobId check into the if condition to get a new bunch of nonces
- CPU: add jobId validation after the start nonce is consumed

4 files changed, 20 insertions, 12 deletions

diff --git a/xmrstak/backend/amd/minethd.cpp b/xmrstak/backend/amd/minethd.cpp
index 189576c..88431cc 100644
--- a/xmrstak/backend/amd/minethd.cpp
+++ b/xmrstak/backend/amd/minethd.cpp
@@ -231,10 +231,11 @@ void minethd::work_main()
 			if((round_ctr++ & 0xF) == 0)
 			{
 				globalStates::inst().calc_start_nonce(pGpuCtx->Nonce, oWork.bNiceHash, h_per_round * 16);
+				// check if the job is still valid, there is a small possibility that the job is switched
+				if(globalStates::inst().iGlobalJobNo.load(std::memory_order_relaxed) != iJobNo)
+					break;
 			}
-			// check if the job is still valid, there is a small posibility that the job is switched
-			if(globalStates::inst().iGlobalJobNo.load(std::memory_order_relaxed) != iJobNo)
-				break;
+			
 
 			cl_uint results[0x100];
 			memset(results,0,sizeof(cl_uint)*(0x100));
diff --git a/xmrstak/backend/cpu/minethd.cpp b/xmrstak/backend/cpu/minethd.cpp
index 8f79667..482c085 100644
--- a/xmrstak/backend/cpu/minethd.cpp
+++ b/xmrstak/backend/cpu/minethd.cpp
@@ -503,6 +503,9 @@ void minethd::work_main()
 			if((nonce_ctr++ & (nonce_chunk-1)) == 0)
 			{
 				globalStates::inst().calc_start_nonce(result.iNonce, oWork.bNiceHash, nonce_chunk);
+				// check if the job is still valid, there is a small posibility that the job is switched
+				if(globalStates::inst().iGlobalJobNo.load(std::memory_order_relaxed) != iJobNo)
+					break;
 			}
 
 			*piNonce = result.iNonce;
@@ -810,6 +813,9 @@ void minethd::multiway_work_main()
 			{
 				globalStates::inst().calc_start_nonce(iNonce, oWork.bNiceHash, nonce_chunk);
 				nonce_ctr = nonce_chunk;
+				// check if the job is still valid, there is a small posibility that the job is switched
+				if(globalStates::inst().iGlobalJobNo.load(std::memory_order_relaxed) != iJobNo)
+					break;
 			}
 
 			for (size_t i = 0; i < N; i++)
diff --git a/xmrstak/backend/globalStates.cpp b/xmrstak/backend/globalStates.cpp
index 9f41be2..3bd7d0e 100644
--- a/xmrstak/backend/globalStates.cpp
+++ b/xmrstak/backend/globalStates.cpp
@@ -47,20 +47,21 @@ void globalStates::switch_work(miner_work& pWork, pool_data& dat)
 {
 	jobLock.WriteLock();
 
+	/* This notifies all threads that the job has changed.
+	 * To avoid duplicated shared this must be done before the nonce is exchanged.
+	 */
+	iGlobalJobNo++;
+	
 	size_t xid = dat.pool_id;
 	dat.pool_id = pool_id;
 	pool_id = xid;
 
 	/* Maybe a worker thread is updating the nonce while we read it.
-	 * In that case GPUs check the job ID after a nonce update and in the
-	 * case that it is a CPU thread we have a small chance (max 6 nonces per CPU thread)
-	 * that we recalculate a nonce after we reconnect to the current pool
+	 * To avoid duplicated share calculations the job ID is checked in the worker thread
+	 * after the nonce is read.
 	 */
 	dat.iSavedNonce = iGlobalNonce.exchange(dat.iSavedNonce, std::memory_order_relaxed);
 	oGlobalWork = pWork;
-
-	// this notifies all threads that the job has changed
-	iGlobalJobNo++;
 	
 	jobLock.UnLock();
 }
diff --git a/xmrstak/backend/nvidia/minethd.cpp b/xmrstak/backend/nvidia/minethd.cpp
index a21e525..16171e1 100644
--- a/xmrstak/backend/nvidia/minethd.cpp
+++ b/xmrstak/backend/nvidia/minethd.cpp
@@ -275,10 +275,10 @@ void minethd::work_main()
 			if((round_ctr++ & 0xF) == 0)
 			{
 				globalStates::inst().calc_start_nonce(iNonce, oWork.bNiceHash, h_per_round * 16);
+				// check if the job is still valid, there is a small posibility that the job is switched
+				if(globalStates::inst().iGlobalJobNo.load(std::memory_order_relaxed) != iJobNo)
+					break;
 			}
-			// check if the job is still valid, there is a small posibility that the job is switched
-			if(globalStates::inst().iGlobalJobNo.load(std::memory_order_relaxed) != iJobNo)
-				break;
 
 			uint32_t foundNonce[10];
 			uint32_t foundCount;