diff --git a/sys/kern/kern_malloc.c b/sys/kern/kern_malloc.c
--- a/sys/kern/kern_malloc.c
+++ b/sys/kern/kern_malloc.c
@@ -763,6 +763,67 @@
 	return (malloc_large(&size, mtp, ds, flags DEBUG_REDZONE_ARG));
 }
 
+void *
+malloc_domainset_aligned(size_t size, size_t align, struct malloc_type *mtp,
+    struct domainset *ds, int flags)
+{
+	char *mem, *res;
+	size_t addend;
+	uintptr_t x;
+
+	MPASS(align > 0 && (align & (align - 1)) == 0);
+
+	if (align < sizeof(void *))
+		align = sizeof(void *);
+
+	addend = sizeof(void *) + align;
+#ifdef INVARIANTS
+	addend += sizeof(size_t);
+#endif
+	mem = malloc_domainset(size + addend, mtp, ds, flags);
+	if (mem == NULL)
+		return (NULL);
+
+	x = roundup((uintptr_t)mem + sizeof(void *), align);
+	res = (void *)x;
+	x -= sizeof(void *);
+	memcpy((void *)x, &mem, sizeof(mem));
+#ifdef INVARIANTS
+	x -= sizeof(size_t);
+	memcpy((void *)x, &align, sizeof(size_t));
+#endif
+	return (res);
+}
+
+void
+free_aligned(void *addr, struct malloc_type *mtp)
+{
+	void *mem;
+	uintptr_t x;
+
+	if (addr == NULL)
+		return;
+	x = (uintptr_t)addr;
+	x -= sizeof(void *);
+	memcpy(&mem, (void *)x, sizeof(mem));
+#ifdef INVARIANTS
+	{
+		size_t align;
+
+		x -= sizeof(size_t);
+		memcpy(&align, (void *)x, sizeof(size_t));
+		if (align == 0 || (align & (align - 1)) != 0)
+			panic("free_aligned: bad align "
+			    "addr %p alloc %p align %#zx", addr, mem, align);
+		if ((uintptr_t)addr - (uintptr_t)mem > sizeof(void *) +
+		    sizeof(size_t) + align)
+			panic("free_aligned: addresses too far "
+			    "addr %p alloc %p align %#zx", addr, mem, align);
+	}
+#endif
+	free(mem, mtp);
+}
+
 void *
 mallocarray(size_t nmemb, size_t size, struct malloc_type *type, int flags)
 {
diff --git a/sys/sys/malloc.h b/sys/sys/malloc.h
--- a/sys/sys/malloc.h
+++ b/sys/sys/malloc.h
@@ -261,6 +261,10 @@
 	    __result_use_check __alloc_size(2);
 void	*reallocf(void *addr, size_t size, struct malloc_type *type, int flags)
 	    __result_use_check __alloc_size(2);
+void	*malloc_domainset_aligned(size_t size, size_t align,
+	    struct malloc_type *mtp, struct domainset *ds, int flags)
+	    __malloc_like __result_use_check __alloc_size(1);
+void	free_aligned(void *addr, struct malloc_type *mtp);
 
 struct malloc_type *malloc_desc2type(const char *desc);
 
diff --git a/sys/x86/x86/busdma_bounce.c b/sys/x86/x86/busdma_bounce.c
--- a/sys/x86/x86/busdma_bounce.c
+++ b/sys/x86/x86/busdma_bounce.c
@@ -445,18 +445,13 @@
 	 * else allocate a block of contiguous pages because one or more of the
 	 * constraints is something that only the contig allocator can fulfill.
 	 *
-	 * NOTE: The (dmat->common.alignment <= dmat->maxsize) check
-	 * below is just a quick hack. The exact alignment guarantees
-	 * of malloc(9) need to be nailed down, and the code below
-	 * should be rewritten to take that into account.
-	 *
-	 * In the meantime warn the user if malloc gets it wrong.
+	 * Warn the user if we get it wrong.
 	 */
 	if (dmat->common.maxsize <= PAGE_SIZE &&
-	    dmat->common.alignment <= dmat->common.maxsize &&
 	    dmat->common.lowaddr >= ptoa((vm_paddr_t)Maxmem) &&
 	    attr == VM_MEMATTR_DEFAULT) {
-		*vaddr = malloc_domainset(dmat->common.maxsize, M_DEVBUF,
+		*vaddr = malloc_domainset_aligned(dmat->common.maxsize,
+		    dmat->common.alignment, M_DEVBUF,
 		    DOMAINSET_PREF(dmat->common.domain), mflags);
 	} else if (dmat->common.nsegments >=
 	    howmany(dmat->common.maxsize, MIN(dmat->common.maxsegsz,
@@ -503,7 +498,7 @@
 	if (map != NULL)
 		panic("bus_dmamem_free: Invalid map freed\n");
 	if ((dmat->bounce_flags & BUS_DMA_KMEM_ALLOC) == 0)
-		free(vaddr, M_DEVBUF);
+		free_aligned(vaddr, M_DEVBUF);
 	else
 		kmem_free((vm_offset_t)vaddr, dmat->common.maxsize);
 	CTR3(KTR_BUSDMA, "%s: tag %p flags 0x%x", __func__, dmat,