Add victim cache for I-cache

Introduce a small victim cache to reduce conflict misses
in the direct-mapped instruction cache. On an I-cache miss,
probe the victim cache; on hit, swap the victim block with
the current I-cache block and return the data.

Also rename ic.block → ic.i_block to distinguish between
primary I-cache blocks and victim cache blocks.

Author: yy214123

riscv.c

@@ -321,7 +321,7 @@ static void mmu_fetch(hart_t *vm, uint32_t addr, uint32_t *value)
     /* cache hit */
     uint32_t idx = (addr >> IC_OFFSET_BITS) & IC_INDEX_MASK;
     uint32_t tag = addr >> (IC_OFFSET_BITS + IC_INDEX_BITS);
-    icache_block_t *blk = &vm->ic.block[idx];
+    icache_block_t *blk = &vm->ic.i_block[idx];
 
     if (likely(blk->valid && blk->tag == tag)) {
 #ifdef MMU_CACHE_STATS
@@ -332,6 +332,21 @@ static void mmu_fetch(hart_t *vm, uint32_t addr, uint32_t *value)
         return;
     }
 
+    /* search the victim cache */
+    for (int i = 0; i < VC_BLOCKS; i++) {
+        victim_cache_block_t *vblk = &vm->ic.v_block[i];
+        if (vblk->valid && vblk->tag == tag) {
+            /* victim cache hit, block swap*/
+            icache_block_t tmp = *blk;
+            *blk = *vblk;
+            *vblk = tmp;
+
+            uint32_t ofs = addr & IC_BLOCK_MASK;
+            *value = *(const uint32_t *) (blk->base + ofs);
+            return;
+        }
+    }
+
 #ifdef MMU_CACHE_STATS
     vm->cache_fetch.misses++;
 #endif

riscv.h

@@ -91,6 +91,16 @@ typedef struct __vm_internel vm_t;
 #define IC_OFFSET_BITS (__builtin_ctz((IC_BLOCKS_SIZE)))
 #define IC_INDEX_BITS (__builtin_ctz((IC_BLOCKS)))
 
+/* Define the victim cache.
+ *
+ * The block size of the victim cache is identical to that of the primary
+ * instruction cache (IC), ensuring full block compatibility.
+ * However, the number of blocks is smaller, allowing the VC to store
+ * a few recently evicted cache lines to reduce conflict misses.
+ */
+#define VC_BLOCK_SIZE IC_BLOCKS_SIZE
+#define VC_BLOCKS 16
+
 /* For power-of-two sizes, (size - 1) sets all low bits to 1,
  * allowing fast extraction of an address.
  */
@@ -104,8 +114,11 @@ typedef struct {
     bool valid;
 } icache_block_t;
 
+typedef icache_block_t victim_cache_block_t;
+
 typedef struct {
-    icache_block_t block[IC_BLOCKS];
+    icache_block_t i_block[IC_BLOCKS];
+    victim_cache_block_t v_block[VC_BLOCKS];
 } icache_t;
 
 struct __hart_internal {