diff --git a/Cargo.toml b/Cargo.toml
index 2c5504421ec..708bed282b6 100644
--- a/Cargo.toml
+++ b/Cargo.toml
@@ -39,7 +39,7 @@ hex = "0.3"
 home = "0.3"
 ignore = "0.4"
 lazy_static = "1.2.0"
-jobserver = "0.1.11"
+jobserver = "0.1.13"
 lazycell = "1.2.0"
 libc = "0.2"
 log = "0.4.6"
diff --git a/src/cargo/util/flock.rs b/src/cargo/util/flock.rs
index bc87be6c9e6..e49a4daaa38 100644
--- a/src/cargo/util/flock.rs
+++ b/src/cargo/util/flock.rs
@@ -312,8 +312,27 @@ fn acquire(
     let msg = format!("waiting for file lock on {}", msg);
     config.shell().status_with_color("Blocking", &msg, Cyan)?;
 
-    block().chain_err(|| format!("failed to lock file: {}", path.display()))?;
-    return Ok(());
+    // We're about to block the current process and not really do anything
+    // productive for what could possibly be a very long time. We could be
+    // waiting, for example, on another Cargo to finish a download, finish an
+    // entire build, etc. Since we're not doing anything productive we're not
+    // making good use of our jobserver token, if we have one.
+    //
+    // This can typically come about if `cargo` is invoked from `make` (or some
+    // other jobserver-providing system). In this situation it's actually best
+    // if we release the token back to the original jobserver to let some other
+    // cpu-hungry work continue to make progress. After we're done blocking
+    // we'll block waiting to reacquire a token as we'll probably be doing cpu
+    // hungry work ourselves.
+    let jobserver = config.jobserver_from_env();
+    if let Some(server) = jobserver {
+        server.release_raw()?;
+    }
+    let result = block().chain_err(|| format!("failed to lock file: {}", path.display()));
+    if let Some(server) = jobserver {
+        server.acquire_raw()?;
+    }
+    return Ok(result?);
 
     #[cfg(all(target_os = "linux", not(target_env = "musl")))]
     fn is_on_nfs_mount(path: &Path) -> bool {