The old code would += the total errors every time
stats were gathered.  Instead, keep a count of short-pkt
and long-pkt counters and then simply add them together
for the rx-over-length stat.
Signed-off-by: Ben Greear <greearb@candelatech.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
        u32 rx_fc_pause;
        u32 rx_fc_unsupported;
        u32 rx_tco_frames;
+       u32 rx_short_frame_errors;
        u32 rx_over_length_errors;
 
        u16 eeprom_wc;
                ns->collisions += nic->tx_collisions;
                ns->tx_errors += le32_to_cpu(s->tx_max_collisions) +
                        le32_to_cpu(s->tx_lost_crs);
-               ns->rx_length_errors += le32_to_cpu(s->rx_short_frame_errors) +
+               nic->rx_short_frame_errors +=
+                       le32_to_cpu(s->rx_short_frame_errors);
+               ns->rx_length_errors = nic->rx_short_frame_errors +
                        nic->rx_over_length_errors;
                ns->rx_crc_errors += le32_to_cpu(s->rx_crc_errors);
                ns->rx_frame_errors += le32_to_cpu(s->rx_alignment_errors);