schfs: cache sixteen inodes, not one

The one-entry cache lived inside schfs_file_read, so stat() never touched
it -- which is why `ls` barely moved. Moved into schfs_inode_get, where
every caller benefits, and widened to sixteen entries.

`ls` is the shape that matters: it stats every name in a directory, so a
single entry never hits and each file costs two descents, one to find the
name and one to read the inode. Sixteen keeps hitting through a directory
walk without a second reader evicting everything. Round-robin rather than
LRU: the pattern is a walk, not a working set, so the bookkeeping on every
hit would buy nothing.

Measured in the guest, TCG, against the original baseline:

  ls /bin warm   0.70s -> 0.43s
  ls /bin cold   2.80s -> 1.32s
  read 834 KB    1.73s -> 0.35s

schfs-check passes including the round-trip; bootcheck 108 [ok].

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

kernel/fs/schfs/obj.c

@@ -80,10 +80,43 @@ schfs_namehash(const char *name, unsigned len)
 /* Inodes                                                              */
 /* ------------------------------------------------------------------ */
 
+/* Inode cache.
+ *
+ * Sixteen entries, not one. `ls` is the shape that matters: it stats every name
+ * in a directory, so a single entry never hits and the whole directory costs
+ * two tree descents per file -- one to find the name, one to read the inode.
+ * Sixteen is enough that a directory walk keeps hitting while a second reader
+ * elsewhere does not evict everything.
+ *
+ * Round-robin replacement, deliberately: an LRU needs bookkeeping on every hit,
+ * and the access pattern here is a walk rather than a working set, so the
+ * cleverer policy buys nothing.
+ *
+ * Invalidation is blunt and total -- see schfs_cache_invalidate().
+ */
+#define INO_CACHE_N 16
+
+static struct {
+    int                      valid;
+    const struct schfs_vol  *vol;
+    uint64_t                 ino;
+    struct schfs_inode       in;
+} ino_cache[INO_CACHE_N];
+
+static unsigned ino_cache_next;
+
 int
 schfs_inode_get(struct schfs_vol *v, struct schfs_cursor *c,
                 uint64_t ino, struct schfs_inode *out)
 {
+    for (unsigned i = 0; i < INO_CACHE_N; i++) {
+        if (ino_cache[i].valid && ino_cache[i].vol == v &&
+            ino_cache[i].ino == ino) {
+            *out = ino_cache[i].in;
+            return 0;
+        }
+    }
+
     struct schfs_key k;
     schfs_mkkey(&k, ino, SCHFS_KEY_INODE, 0);
 
@@ -96,6 +129,12 @@ schfs_inode_get(struct schfs_vol *v, struct schfs_cursor *c,
     // it rather than acting on a half-filled struct.
     if (len != sizeof(*out))
         return -2;
+
+    unsigned slot = ino_cache_next++ % INO_CACHE_N;
+    ino_cache[slot].valid = 1;
+    ino_cache[slot].vol   = v;
+    ino_cache[slot].ino   = ino;
+    ino_cache[slot].in    = *out;
     return 0;
 }
 
@@ -287,7 +326,7 @@ schfs_dir_iterate(struct schfs_vol *v, struct schfs_cursor *c, uint64_t dir,
 /* ------------------------------------------------------------------ */
 
 /* --------------------------------------------------------------------------
- * Read caches: the last inode, and the last extent.
+ * Read caches: recently-read inodes (see schfs_inode_get) and the last extent.
  *
  * Every read used to descend the B-tree twice per block -- once for the inode
  * and once for the extent -- so an 834 KB file cost 418 descents, each a tree
@@ -307,13 +346,6 @@ schfs_dir_iterate(struct schfs_vol *v, struct schfs_cursor *c, uint64_t dir,
  * The volume pointer is part of the key because the host tools can hold more
  * than one volume open, and two images would otherwise share these.
  * ------------------------------------------------------------------------ */
-static struct {
-    int                      valid;
-    const struct schfs_vol  *vol;
-    uint64_t                 ino;
-    struct schfs_inode       in;
-} ino_cache;
-
 static struct {
     int                      valid;
     const struct schfs_vol  *vol;
@@ -326,7 +358,8 @@ static struct {
 void
 schfs_cache_invalidate(void)
 {
-    ino_cache.valid = 0;
+    for (unsigned i = 0; i < INO_CACHE_N; i++)
+        ino_cache[i].valid = 0;
     ext_cache.valid = 0;
 }
 
@@ -412,18 +445,12 @@ long
 schfs_file_read(struct schfs_vol *v, struct schfs_cursor *c, uint64_t ino,
                 uint64_t off, void *buf, unsigned n, void *dblk)
 {
+    /* schfs_inode_get caches for itself now, so this is a hit on every chunk
+     * after the first without a second cache here to keep coherent. */
     struct schfs_inode in;
-    if (ino_cache.valid && ino_cache.vol == v && ino_cache.ino == ino) {
+    int r = schfs_inode_get(v, c, ino, &in);
-        in = ino_cache.in;
+    if (r != 0)
-    } else {
+        return r;
-        int r = schfs_inode_get(v, c, ino, &in);
-        if (r != 0)
-            return r;
-        ino_cache.valid = 1;
-        ino_cache.vol   = v;
-        ino_cache.ino   = ino;
-        ino_cache.in    = in;
-    }
 
     if (off >= in.size)
         return 0;