arm64: mm: reserve hugetlb CMA after numa_init

hugetlb_cma_reserve() is called at the wrong place. numa_init has not been
done yet. so all reserved memory will be located at node0.

Fixes: cf11e85 ("mm: hugetlb: optionally allocate gigantic hugepages using cma")
Signed-off-by: Barry Song <[email protected]>
Reviewed-by: Anshuman Khandual <[email protected]>
Acked-by: Roman Gushchin <[email protected]>
Cc: Matthias Brugger <[email protected]>
Cc: Will Deacon <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Will Deacon <[email protected]>

Author: Barry Song

arch/arm64/mm/init.c

@@ -404,11 +404,6 @@ void __init arm64_memblock_init(void)
 	high_memory = __va(memblock_end_of_DRAM() - 1) + 1;
 
 	dma_contiguous_reserve(arm64_dma32_phys_limit);
-
-#ifdef CONFIG_ARM64_4K_PAGES
-	hugetlb_cma_reserve(PUD_SHIFT - PAGE_SHIFT);
-#endif
-
 }
 
 void __init bootmem_init(void)
@@ -424,6 +419,16 @@ void __init bootmem_init(void)
 	min_low_pfn = min;
 
 	arm64_numa_init();
+
+	/*
+	 * must be done after arm64_numa_init() which calls numa_init() to
+	 * initialize node_online_map that gets used in hugetlb_cma_reserve()
+	 * while allocating required CMA size across online nodes.
+	 */
+#ifdef CONFIG_ARM64_4K_PAGES
+	hugetlb_cma_reserve(PUD_SHIFT - PAGE_SHIFT);
+#endif
+
 	/*
 	 * Sparsemem tries to allocate bootmem in memory_present(), so must be
 	 * done after the fixed reservations.