Re: [PATCH] net: stmmac: guard FCS stripping against runt frames

On Thu, Sep 24, 2026 at 10:51:11AM -0300, Aldo Ariel wrote:

Hi Andrew, Lorenzo,

Fair point on the subject — the trigger is not limited to runt frames.
 
The FCS stripping in stmmac_rx() at lines 5810-5817 runs on every
last-segment descriptor regardless of total frame size:

    if (likely(!(status & rx_not_ls))) {
        if (buf2_len) {
            buf2_len -= ETH_FCS_LEN;
        } else if (buf1_len) {
            buf1_len -= ETH_FCS_LEN;
        }
    }      
       
For a multi-descriptor frame, the last descriptor's buf2_len or
buf1_len carries only the tail of the frame. With the default
dma_buf_sz = BUF_SIZE_2KiB for MTU 1500, a 4097-byte frame splits as:

Is the MTU programmed into the hardware? Will it even try to transfer
an 4097 octet frame when the MTU is configured to 1500?

   Andrew