dev-hsa: Update QCntxt readIndex in HW scheduler write

The QCntxt is reused when a queue is unmapped and mapped again. This is
fairly common in GPU full system. If this is not done the readIndex on
the queue context is reset to 1, causing getCommandsFromHost to read
from the wrong slot which is typically an old dispatch packet or an
invalid packet. This causes simulation to stall as the incorrect
completion signal is eventually written.

Change-Id: I65541e559fe04f5eb44b936ca37e3f802262fe6a
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/57670
Reviewed-by: Matt Sinclair <mattdsinclair@gmail.com>
Maintainer: Matt Sinclair <mattdsinclair@gmail.com>
Tested-by: kokoro <noreply+kokoro@google.com>

src/dev/hsa/hw_scheduler.cc

@@ -335,6 +335,11 @@ HWScheduler::write(Addr db_addr, uint64_t doorbell_reg)
     uint32_t al_idx = dbMap[db_addr];
     // Modify the write pointer
     activeList[al_idx].qDesc->writeIndex = doorbell_reg;
+    // If a queue is unmapped and remapped (common in full system) the qDesc
+    // gets reused. Keep the readIndex up to date so that when the HSA packet
+    // processor gets commands from host, the correct entry is read after
+    // remapping.
+    activeList[al_idx].qDesc->readIndex = doorbell_reg - 1;
     DPRINTF(HSAPacketProcessor, "queue %d qDesc->writeIndex %d\n",
             al_idx, activeList[al_idx].qDesc->writeIndex);
     // If this queue is mapped, then start DMA to fetch the