zephyr: lib: alloc: Use cached memory for L3 Heap

This patch implements recommended hardware flow for
Intel ACE platforms.
The L3 heap should be accessed via cached pointers
including management data.

Signed-off-by: Jaroslaw Stelter <Jaroslaw.Stelter@intel.com>

Author: jxstelter

zephyr/lib/alloc.c

@@ -116,8 +116,8 @@ static inline uintptr_t get_l3_heap_start(void)
 	 * - main_fw_load_offset
 	 * - main fw size in manifest
 	 */
-	return (uintptr_t)z_soc_uncached_ptr((__sparse_force void __sparse_cache *)
-					     ROUND_UP(IMR_L3_HEAP_BASE, L3_MEM_PAGE_SIZE));
+	return (uintptr_t)((__sparse_force void __sparse_cache *)
+			   ROUND_UP(IMR_L3_HEAP_BASE, L3_MEM_PAGE_SIZE));
 }
 
 /**
@@ -145,14 +145,89 @@ static bool is_l3_heap_pointer(void *ptr)
 	uintptr_t l3_heap_start = get_l3_heap_start();
 	uintptr_t l3_heap_end = l3_heap_start + get_l3_heap_size();
 
-	if (is_cached(ptr))
-		ptr = z_soc_uncached_ptr((__sparse_force void __sparse_cache *)ptr);
-
 	if ((POINTER_TO_UINT(ptr) >= l3_heap_start) && (POINTER_TO_UINT(ptr) < l3_heap_end))
 		return true;
 
 	return false;
 }
+
+/**
+ * Flush and invalidate cached L3 heap.
+ */
+static inline void l3_heap_flush_invalidate(struct k_heap *h)
+{
+	if (h == &l3_heap)
+		sys_cache_data_flush_and_invd_range((void *)get_l3_heap_start(),
+						    get_l3_heap_size());
+}
+
+/**
+ * Invalidate cached L3 heap management data.
+ */
+static inline void l3_heap_flush(struct k_heap *h)
+{
+	if (h == &l3_heap)
+		sys_cache_data_flush_range((void *)get_l3_heap_start(),
+					   get_l3_heap_size());
+}
+
+static void *l3_heap_alloc_aligned(struct k_heap *h, size_t min_align, size_t bytes)
+{
+	k_spinlock_key_t key;
+	void *ret;
+#if CONFIG_SYS_HEAP_RUNTIME_STATS && CONFIG_IPC_MAJOR_4
+	struct sys_memory_stats stats;
+#endif
+
+	key = k_spin_lock(&h->lock);
+	l3_heap_flush_invalidate(h);
+	ret = sys_heap_aligned_alloc(&h->heap, min_align, bytes);
+	l3_heap_flush(h);
+	k_spin_unlock(&h->lock, key);
+
+#if CONFIG_SYS_HEAP_RUNTIME_STATS && CONFIG_IPC_MAJOR_4
+	sys_heap_runtime_stats_get(&h->heap, &stats);
+	tr_info(&zephyr_tr, "heap allocated: %u free: %u max allocated: %u",
+		stats.allocated_bytes, stats.free_bytes, stats.max_allocated_bytes);
+#endif
+
+	return ret;
+}
+
+static void __sparse_cache *l3_heap_alloc_aligned_cached(struct k_heap *h,
+							 size_t min_align, size_t bytes)
+{
+	void __sparse_cache *ptr;
+
+	/*
+	 * Zephyr sys_heap stores metadata at start of each
+	 * heap allocation. To ensure no allocated cached buffer
+	 * overlaps the same cacheline with the metadata chunk,
+	 * align both allocation start and size of allocation
+	 * to cacheline. As cached and non-cached allocations are
+	 * mixed, same rules need to be followed for both type of
+	 * allocations.
+	 */
+	min_align = MAX(PLATFORM_DCACHE_ALIGN, min_align);
+	bytes = ALIGN_UP(bytes, min_align);
+
+	ptr = (__sparse_force void __sparse_cache *)l3_heap_alloc_aligned(h, min_align, bytes);
+
+	if (ptr)
+		ptr = z_soc_cached_ptr((__sparse_force void *)ptr);
+
+	return ptr;
+}
+
+static void l3_heap_free(struct k_heap *h, void *mem)
+{
+	k_spinlock_key_t key = k_spin_lock(&h->lock);
+
+	sys_heap_free(&h->heap, mem);
+	l3_heap_flush(h);
+	k_spin_unlock(&h->lock, key);
+}
+
 #endif
 
 static void *heap_alloc_aligned(struct k_heap *h, size_t min_align, size_t bytes)
@@ -251,6 +326,15 @@ void *rmalloc(enum mem_zone zone, uint32_t flags, uint32_t caps, size_t bytes)
 	if (caps & SOF_MEM_CAPS_L3) {
 #if CONFIG_L3_HEAP
 		heap = &l3_heap;
+		/* Uncached L3_HEAP should be not used */
+		if (!zone_is_cached(zone))
+			k_panic();
+		ptr = (__sparse_force void *)l3_heap_alloc_aligned_cached(heap, 0, bytes);
+
+		if (!ptr && zone == SOF_MEM_ZONE_SYS)
+			k_panic();
+
+		return ptr;
 #else
 		k_panic();
 #endif
@@ -352,7 +436,7 @@ void rfree(void *ptr)
 
 #if CONFIG_L3_HEAP
 	if (is_l3_heap_pointer(ptr)) {
-		heap_free(&l3_heap, ptr);
+		l3_heap_free(&l3_heap, ptr);
 		return;
 	}
 #endif
@@ -367,6 +451,7 @@ static int heap_init(void)
 
 #if CONFIG_L3_HEAP
 	sys_heap_init(&l3_heap.heap, UINT_TO_POINTER(get_l3_heap_start()), get_l3_heap_size());
+	l3_heap_flush_invalidate(&l3_heap);
 #endif
 
 	return 0;