feature: add main lock for both execute and flush

- both execute and flush have their own locks
- spliting these locks makes race condition
- obtain lock at higher level
- do not lock in queue when calling execute from immediate
- pass main lock to flush so can be unlocked before synchronize

Related-To: NEO-10356

Signed-off-by: Zbigniew Zdanowicz <zbigniew.zdanowicz@intel.com>

level_zero/core/source/cmdlist/cmdlist_hw_immediate.inl

@@ -424,7 +424,7 @@ inline ze_result_t CommandListCoreFamilyImmediate<gfxCoreFamily>::executeCommand
     size_t commandStreamStart = this->cmdListCurrentStartOffset;
 
     auto csr = static_cast<CommandQueueImp *>(cmdQ)->getCsr();
-    auto lockCSR = csr->obtainUniqueOwnership();
+    auto lockCSR = outerLock != nullptr ? std::move(*outerLock) : csr->obtainUniqueOwnership();
 
     if (NEO::ApiSpecificConfig::isSharedAllocPrefetchEnabled()) {
         auto svmAllocMgr = this->device->getDriverHandle()->getSvmAllocsManager();
@@ -1685,8 +1685,8 @@ ze_result_t CommandListCoreFamilyImmediate<gfxCoreFamily>::appendCommandLists(ui
         return ret;
     }
 
-    auto cmdQueue = this->cmdQImmediate;
-    ret = cmdQueue->executeCommandLists(numCommandLists, phCommandLists, nullptr, true, this->commandContainer.getCommandStream());
+    auto mainAppendLock = static_cast<CommandQueueImp *>(this->cmdQImmediate)->getCsr()->obtainUniqueOwnership();
+    ret = this->cmdQImmediate->executeCommandLists(numCommandLists, phCommandLists, nullptr, true, this->commandContainer.getCommandStream());
     if (ret != ZE_RESULT_SUCCESS) {
         return ret;
     }
@@ -1701,7 +1701,7 @@ ze_result_t CommandListCoreFamilyImmediate<gfxCoreFamily>::appendCommandLists(ui
     }
 
     bool hasStallingCmds = true;
-    return flushImmediate(ret, true, hasStallingCmds, relaxedOrderingDispatch, NEO::AppendOperations::kernel, false, hSignalEvent, true, nullptr);
+    return flushImmediate(ret, true, hasStallingCmds, relaxedOrderingDispatch, NEO::AppendOperations::kernel, false, hSignalEvent, true, &mainAppendLock);
 }
 
 } // namespace L0

level_zero/core/source/cmdqueue/cmdqueue_hw.inl

@@ -73,11 +73,12 @@ ze_result_t CommandQueueHw<gfxCoreFamily>::executeCommandLists(
         this->csr->ensurePrimaryCsrInitialized(*this->device->getNEODevice());
     }
 
-    auto lockCSR = this->csr->obtainUniqueOwnership();
+    std::unique_lock<NEO::CommandStreamReceiver::MutexType> lockCSR;
 
     if (parentImmediateCommandlistLinearStream != nullptr) {
         this->startingCmdBuffer = parentImmediateCommandlistLinearStream;
     } else {
+        lockCSR = this->csr->obtainUniqueOwnership();
         this->startingCmdBuffer = &this->commandStream;
     }
 

level_zero/core/test/unit_tests/sources/cmdlist/test_cmdlist_4.cpp

@@ -1567,12 +1567,19 @@ HWTEST2_F(ImmediateCommandListTest, givenImmediateCmdListWhenAppendingRegularThe
     commandList->close();
     auto cmdListHandle = commandList->toHandle();
 
+    auto &ultCsr = neoDevice->getUltCommandStreamReceiver<FamilyType>();
+    ultCsr.recursiveLockCounter = 0;
+
     // first append can carry preamble
     commandListImmediate->appendCommandLists(1, &cmdListHandle, nullptr, 0, nullptr);
 
+    EXPECT_EQ(1u, ultCsr.recursiveLockCounter);
+
     // regular append can dispatch bb_start to secondary regular or primary directly regular
     commandListImmediate->appendCommandLists(1, &cmdListHandle, nullptr, 0, nullptr);
 
+    EXPECT_EQ(2u, ultCsr.recursiveLockCounter);
+
     auto startStream = static_cast<L0::CommandQueueImp *>(commandListImmediate->cmdQImmediate)->getStartingCmdBuffer();
 
     if (commandListImmediate->getCmdListBatchBufferFlag()) {

level_zero/core/test/unit_tests/sources/cmdqueue/test_cmdqueue_1.cpp

@@ -1379,9 +1379,14 @@ HWTEST2_F(ExecuteCommandListTests, givenSuccessfulSubmitBatchBufferThenExecuteCo
     auto commandListHandle = commandList->toHandle();
     commandList->close();
 
+    auto &ultCsr = neoDevice->getUltCommandStreamReceiver<FamilyType>();
+    ultCsr.recursiveLockCounter = 0;
+
     auto res = commandQueue->executeCommandLists(1, &commandListHandle, nullptr, false, nullptr);
     EXPECT_EQ(ZE_RESULT_SUCCESS, res);
 
+    EXPECT_EQ(1u, ultCsr.recursiveLockCounter);
+
     commandQueue->destroy();
     commandList->destroy();
 }