reduce error spam for dead hwc resources

Author: erfrimod

vm/devices/net/mana_driver/src/gdma_driver.rs

@@ -1495,15 +1495,16 @@ impl<T: DeviceBacking> GdmaDriver<T> {
         )
         .await
     }
+
+    /// Returns true if the HWC has failed and no further requests will succeed.
+    pub fn hwc_failure(&self) -> bool {
+        self.hwc_failure
+    }
 }
 
 #[cfg(test)]
 impl<T: DeviceBacking> GdmaDriver<T> {
     pub(crate) fn hwc_timeout_in_ms(&self) -> u32 {
         self.hwc_timeout_in_ms
     }
-
-    pub(crate) fn hwc_failure(&self) -> bool {
-        self.hwc_failure
-    }
 }

vm/devices/net/mana_driver/src/resources.rs

@@ -63,12 +63,23 @@ impl ResourceArena {
     }
 
     pub(crate) async fn destroy<T: DeviceBacking>(mut self, gdma: &mut GdmaDriver<T>) {
+        // When HWC has already failed, skip sending teardown commands for HWC resources:
+        // DmaRegion, Eq, BnicQueue. HWC requests all fail: "Previous hardware failure".
+        // Device should reclaim resources on its own reset.
+        let skip_hwc = gdma.hwc_failure();
+        if skip_hwc {
+            tracing::info!(
+                count = self.resources.len(),
+                "skipping HWC resource teardown after hardware failure"
+            );
+        }
         for resource in self.resources.drain(..).rev() {
             let r = match resource {
                 Resource::MemoryBlock(mem) => {
                     drop(ManuallyDrop::into_inner(mem));
                     Ok(())
                 }
+                _ if skip_hwc => continue,
                 Resource::DmaRegion {
                     dev_id,
                     gdma_region,