From c4567a9188c5b4cd7378d643f3f2d6f2e62cfc8d Mon Sep 17 00:00:00 2001
From: "GunaShekar, Ajay" <AJAY.GunaShekar@amd.com>
Date: Mon, 28 Apr 2025 22:53:43 -0700
Subject: [PATCH] SWDEV-523028 - print PAL failure return values in logs (#81)

* print PAL failure return values in logs
* dump kernel info incase of PAL failure

[ROCm/clr commit: 99ef573399af2245d47b7089ff48e5480c697790]
---
 projects/clr/rocclr/device/pal/palvirtual.cpp | 37 ++++++++++++-------
 projects/clr/rocclr/device/pal/palvirtual.hpp |  2 +-
 2 files changed, 24 insertions(+), 15 deletions(-)

diff --git a/projects/clr/rocclr/device/pal/palvirtual.cpp b/projects/clr/rocclr/device/pal/palvirtual.cpp
index 4a6dc96bbc..15b7bf220a 100644
--- a/projects/clr/rocclr/device/pal/palvirtual.cpp
+++ b/projects/clr/rocclr/device/pal/palvirtual.cpp
@@ -369,24 +369,29 @@ bool VirtualGPU::Queue::flush() {
   const Settings& settings = gpu_.dev().settings();
 
   if (!settings.alwaysResident_ && palMemRefs_.size() != 0) {
-    if (Pal::Result::Success !=
-        iDev_->AddGpuMemoryReferences(palMemRefs_.size(), &palMemRefs_[0], iQueue_,
-                                      Pal::GpuMemoryRefCantTrim)) {
-      LogError("PAL failed to make resident resources!");
+    Pal::Result result = iDev_->AddGpuMemoryReferences(
+                                      palMemRefs_.size(),
+                                      &palMemRefs_[0], iQueue_,
+                                      Pal::GpuMemoryRefCantTrim);
+    if (Pal::Result::Success != result) {
+      LogPrintfError("PAL failed to make resident resources! result: %d", result);
       return false;
     }
     palMemRefs_.clear();
   }
 
   // Stop commands building
-  if (Pal::Result::Success != iCmdBuffs_[cmdBufIdSlot_]->End()) {
-    LogError("PAL failed to finalize a command buffer!");
+  Pal::Result result;
+  result = iCmdBuffs_[cmdBufIdSlot_]->End();
+  if (Pal::Result::Success != result) {
+    LogPrintfError("PAL failed to finalize a command buffer! result: %d", result);
     return false;
   }
 
   // Reset the fence. PAL will reset OS event
-  if (Pal::Result::Success != iDev_->ResetFences(1, &iCmdFences_[cmdBufIdSlot_])) {
-    LogError("PAL failed to reset a fence!");
+  result = iDev_->ResetFences(1, &iCmdFences_[cmdBufIdSlot_]);
+  if (Pal::Result::Success != result) {
+    LogPrintfError("PAL failed to reset a fence! result:%d", result);
     return false;
   }
 
@@ -420,14 +425,16 @@ bool VirtualGPU::Queue::flush() {
     }
   }
   // Submit command buffer to OS
-  Pal::Result result;
   if (gpu_.rgpCaptureEna()) {
     result = gpu_.dev().captureMgr()->TimedQueueSubmit(iQueue_, cmdBufIdCurrent_, submitInfo);
   } else {
     result = iQueue_->Submit(submitInfo);
   }
   if (Pal::Result::Success != result) {
-    LogError("PAL failed to submit CMD!");
+    LogPrintfError("PAL failed to submit CMD! result:%d", result);
+    if (GPU_ANALYZE_HANG) {
+      DumpMemoryReferences();
+    }
     return false;
   }
   // Make sure the slot isn't busy
@@ -461,15 +468,17 @@ bool VirtualGPU::Queue::flush() {
   }
 
   // Reset command buffer, so CB chunks could be reused
-  if (Pal::Result::Success != iCmdBuffs_[cmdBufIdSlot_]->Reset(nullptr, false)) {
-    LogError("PAL failed CB reset!");
+  result = iCmdBuffs_[cmdBufIdSlot_]->Reset(nullptr, false);
+  if (Pal::Result::Success != result) {
+    LogPrintfError("PAL failed CB reset! result:%d", result);
     return false;
   }
   // Start command buffer building
   Pal::CmdBufferBuildInfo cmdBuildInfo = {};
   cmdBuildInfo.pMemAllocator = &vlAlloc_;
-  if (Pal::Result::Success != iCmdBuffs_[cmdBufIdSlot_]->Begin(cmdBuildInfo)) {
-    LogError("PAL failed CB building initialization!");
+  result = iCmdBuffs_[cmdBufIdSlot_]->Begin(cmdBuildInfo);
+  if (Pal::Result::Success != result) {
+    LogPrintfError("PAL failed CB building initialization! result:%d", result);
     return false;
   }
 
diff --git a/projects/clr/rocclr/device/pal/palvirtual.hpp b/projects/clr/rocclr/device/pal/palvirtual.hpp
index 8a21da8367..2d4a123296 100644
--- a/projects/clr/rocclr/device/pal/palvirtual.hpp
+++ b/projects/clr/rocclr/device/pal/palvirtual.hpp
@@ -161,7 +161,7 @@ class VirtualGPU : public device::VirtualDevice {
         if (Pal::Result::Success == result) {
           break;
         } else if ((Pal::Result::NotReady == result) || (Pal::Result::Timeout == result)) {
-          LogWarning("PAL fence isn't ready!");
+          LogPrintfWarning("PAL fence isn't ready! result:%d", result);
           if (GPU_ANALYZE_HANG) {
             DumpMemoryReferences();
           }