From f6143a54899626fd373a37ead43fd7bf138691a2 Mon Sep 17 00:00:00 2001
From: Changyuan Lyu <changyuanl@google.com>
Date: Sun, 12 Jan 2025 22:24:03 -0800
Subject: [PATCH] fix(vm): only report the root failure

Signed-off-by: Changyuan Lyu <changyuanl@google.com>
---
 alioth-cli/src/main.rs    |  4 +---
 alioth/src/board/board.rs | 19 +++++++++++--------
 alioth/src/vm.rs          | 24 ++++++++++++------------
 3 files changed, 24 insertions(+), 23 deletions(-)

diff --git a/alioth-cli/src/main.rs b/alioth-cli/src/main.rs
index a14e589..76e36d0 100644
--- a/alioth-cli/src/main.rs
+++ b/alioth-cli/src/main.rs
@@ -488,9 +488,7 @@ fn main_run(args: RunArgs) -> Result<(), Error> {
     }
 
     vm.boot().context(error::BootVm)?;
-    for result in vm.wait() {
-        result.context(error::WaitVm)?;
-    }
+    vm.wait().context(error::WaitVm)?;
     Ok(())
 }
 
diff --git a/alioth/src/board/board.rs b/alioth/src/board/board.rs
index 6941979..44e7e6f 100644
--- a/alioth/src/board/board.rs
+++ b/alioth/src/board/board.rs
@@ -401,15 +401,18 @@ where
         let mut vcpu = self.create_vcpu(id, &event_tx)?;
 
         let ret = self.run_vcpu_inner(id, &mut vcpu, &boot_rx);
-        if ret.is_err() && !matches!(ret, Err(Error::PeerFailure { .. })) {
-            log::warn!("VCPU-{id} reported error, unblocking other VCPUs...");
-            let mut mp_sync = self.mp_sync.lock();
-            mp_sync.fatal = true;
-            if mp_sync.count > 0 {
-                self.cond_var.notify_all();
-            }
-        }
         event_tx.send(id).unwrap();
+
+        if matches!(ret, Ok(_) | Err(Error::PeerFailure { .. })) {
+            return Ok(());
+        }
+
+        log::warn!("VCPU-{id} reported error, unblocking other VCPUs...");
+        let mut mp_sync = self.mp_sync.lock();
+        mp_sync.fatal = true;
+        if mp_sync.count > 0 {
+            self.cond_var.notify_all();
+        }
         ret
     }
 
diff --git a/alioth/src/vm.rs b/alioth/src/vm.rs
index cff20c8..d2d6df7 100644
--- a/alioth/src/vm.rs
+++ b/alioth/src/vm.rs
@@ -273,7 +273,7 @@ where
         Ok(())
     }
 
-    pub fn wait(&self) -> Vec<Result<()>> {
+    pub fn wait(&self) -> Result<()> {
         self.event_rx.recv().unwrap();
         let vcpus = self.board.vcpus.read();
         for _ in 1..vcpus.len() {
@@ -281,17 +281,17 @@ where
         }
         drop(vcpus);
         let mut vcpus = self.board.vcpus.write();
-        vcpus
-            .drain(..)
-            .enumerate()
-            .map(|(id, (handle, _))| match handle.join() {
-                Err(e) => {
-                    log::error!("cannot join vcpu {}: {:?}", id, e);
-                    Ok(())
-                }
-                Ok(r) => r.context(error::Vcpu { id: id as u32 }),
-            })
-            .collect()
+        let mut ret = Ok(());
+        for (id, (handle, _)) in vcpus.drain(..).enumerate() {
+            let Ok(r) = handle.join() else {
+                log::error!("Cannot join VCPU-{id}");
+                continue;
+            };
+            if r.is_err() && ret.is_ok() {
+                ret = r.context(error::Vcpu { id: id as u32 });
+            }
+        }
+        ret
     }
 }