The vzalloc_node(), already rounds the total size to whole pages, and
sizeof(u64) is smaller than sizeof(struct recv_comp_data). So
round_up of recv_completion_cnt is not necessary, and may cause extra
memory allocation.
To save memory, remove this unnecessary round_up for recv_completion_cnt.
Signed-off-by: Haiyang Zhang <haiyangz@microsoft.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
        net_device->recv_section_size = resp->sections[0].sub_alloc_size;
        net_device->recv_section_cnt = resp->sections[0].num_sub_allocs;
 
-       /* Setup receive completion ring */
-       net_device->recv_completion_cnt
-               = round_up(net_device->recv_section_cnt + 1,
-                          PAGE_SIZE / sizeof(u64));
+       /* Setup receive completion ring.
+        * Add 1 to the recv_section_cnt because at least one entry in a
+        * ring buffer has to be empty.
+        */
+       net_device->recv_completion_cnt = net_device->recv_section_cnt + 1;
        ret = netvsc_alloc_recv_comp_ring(net_device, 0);
        if (ret)
                goto cleanup;