arm64: Align less than PAGE_SIZE pgds naturally

When the pgd size is smaller than PAGE_SIZE, pgd_alloc() uses kzalloc()
to save space. However, this is not always naturally aligned as required
by the architecture. This patch creates a kmem_cache for pgd allocations
with the correct alignment.

The current kernel configurations with 4K pages + 39-bit VA and 64K
pages + 42-bit VA use a full page for the pgd and are not affected. The
patch is required for 48-bit VA with 64K pages where the pgd is 512
bytes.

Reported-by: Christoffer Dall <christoffer.dall@linaro.org>
Reviewed-by: Christoffer Dall <christoffer.dall@linaro.org>
Signed-off-by: Catalin Marinas <catalin.marinas@arm.com>

arch/arm64/mm/pgd.c

@@ -30,18 +30,32 @@
 
 #define PGD_SIZE	(PTRS_PER_PGD * sizeof(pgd_t))
 
+static struct kmem_cache *pgd_cache;
+
 pgd_t *pgd_alloc(struct mm_struct *mm)
 {
 	if (PGD_SIZE == PAGE_SIZE)
 		return (pgd_t *)get_zeroed_page(GFP_KERNEL);
 	else
-		return kzalloc(PGD_SIZE, GFP_KERNEL);
+		return kmem_cache_zalloc(pgd_cache, GFP_KERNEL);
 }
 
 void pgd_free(struct mm_struct *mm, pgd_t *pgd)
 {
 	if (PGD_SIZE == PAGE_SIZE)
 		free_page((unsigned long)pgd);
 	else
-		kfree(pgd);
+		kmem_cache_free(pgd_cache, pgd);
+}
+
+static int __init pgd_cache_init(void)
+{
+	/*
+	 * Naturally aligned pgds required by the architecture.
+	 */
+	if (PGD_SIZE != PAGE_SIZE)
+		pgd_cache = kmem_cache_create("pgd_cache", PGD_SIZE, PGD_SIZE,
+					      SLAB_PANIC, NULL);
+	return 0;
 }
+core_initcall(pgd_cache_init);