Suppress AIO kmem warnings

The new zpl_aio_write() and zpl_aio_read() functions use kmem_alloc()
to allocate enough memory to hold the vectorized IO.  While this
allocation will be small it's been observed in practice to sometimes
slightly exceed the 8K warning threshold by a few kilobytes.
Therefore, the KM_NODEBUG flag has been added to suppress warning.

Signed-off-by: Brian Behlendorf <[email protected]>
Signed-off-by: Richard Yao <[email protected]>
Closes #2774

module/zfs/zpl_file.c

@@ -248,7 +248,7 @@ zpl_aio_read(struct kiocb *kiocb, const struct iovec *iovp,
 	size_t count = kiocb->ki_nbytes;
 	ssize_t read;
 	size_t alloc_size = sizeof (struct iovec) * nr_segs;
-	struct iovec *iov_tmp = kmem_alloc(alloc_size, KM_SLEEP);
+	struct iovec *iov_tmp = kmem_alloc(alloc_size, KM_SLEEP | KM_NODEBUG);
 	bcopy(iovp, iov_tmp, alloc_size);
 
 	ASSERT(iovp);
@@ -325,7 +325,7 @@ zpl_aio_write(struct kiocb *kiocb, const struct iovec *iovp,
 	size_t count = kiocb->ki_nbytes;
 	ssize_t wrote;
 	size_t alloc_size = sizeof (struct iovec) * nr_segs;
-	struct iovec *iov_tmp = kmem_alloc(alloc_size, KM_SLEEP);
+	struct iovec *iov_tmp = kmem_alloc(alloc_size, KM_SLEEP | KM_NODEBUG);
 	bcopy(iovp, iov_tmp, alloc_size);
 
 	ASSERT(iovp);