kernel: pipe reads copy slices and zero only what is buffered

Pipe::read popped a byte at a time, and k_pipe_read zeroed the whole
request (128 KiB for wc) on every call. In the single-thread build's
profile of yes | head -c 20M | wc -c these were 14% of the run
(Pipe::read 182 ms, alloc_zeroed 240 ms of 2970 ms). Both builds gain.

Co-Authored-By: Claude Opus 5.5 (1M context) <[email protected]>
Claude-Session: https://claude.ai/code/session_781bad2ec05143db9067c86b4b2678e2

crates/gex-kernel/src/pipe.rs

@@ -103,9 +103,13 @@ impl Pipe {
         loop {
             if !s.buf.is_empty() {
                 let n = buf.len().min(s.buf.len());
-                for slot in buf.iter_mut().take(n) {
-                    *slot = s.buf.pop_front().unwrap();
-                }
+                // Two slice copies, not a pop per byte: a pipeline moves
+                // megabytes through here.
+                let (front, back) = s.buf.as_slices();
+                let k = n.min(front.len());
+                buf[..k].copy_from_slice(&front[..k]);
+                buf[k..n].copy_from_slice(&back[..n - k]);
+                s.buf.drain(..n);
                 self.ready.notify_all();
                 return Ok(n as u64);
             }

shims/kernel/src/lib.rs

@@ -1442,7 +1442,11 @@ pub extern "C" fn k_pipe_write(id: i32, at: gex_kmem::PAddr, len: u32) -> i32 {
 #[no_mangle]
 pub extern "C" fn k_pipe_read(id: i32, at: gex_kmem::PAddr, len: u32) -> i32 {
     let Some(p) = pipe_at(id) else { return -9 };
-    let mut buf = vec![0u8; len as usize];
+    // No bigger than what is buffered (at least one byte, so an empty pipe
+    // still answers EOF or EAGAIN): a reader asking for 128 KiB of a pipe
+    // holding a few bytes zeroed the whole request on every call.
+    let want = (len as usize).min(p.buffered().max(1));
+    let mut buf = vec![0u8; want];
     match p.read(&mut buf) {
         Ok(n) => {
             unsafe { gex_kmem::push_checked(at, buf.as_ptr(), n as u32) };