On Saturday 30 January 2016 18:33:26 Vasily Averin wrote:
> On 26.01.2016 02:58, Jeff Kirsher wrote:
> > From: Arnd Bergmann <a...@arndb.de>
> >
> > intel/i40e/i40e_txrx.c: In function 'i40e_xmit_frame_ring':
> > intel/i40e/i40e_txrx.c:2367:20: error: 'oiph' may be used uninitialized in
> > this function [-Werror=maybe-uninitialized]
> > intel/i40e/i40e_txrx.c:2317:16: note: 'oiph' was declared here
> > intel/i40e/i40e_txrx.c:2367:17: error: 'oudph' may be used uninitialized in
> > this function [-Werror=maybe-uninitialized]
> > intel/i40e/i40e_txrx.c:2316:17: note: 'oudph' was declared here
>
> 2364 if ((tx_ring->flags & I40E_TXR_FLAGS_OUTER_UDP_CSUM) &&
> 2365 (l4_tunnel == I40E_TXD_CTX_UDP_TUNNELING) &&
> 2366 (*cd_tunneling & I40E_TXD_CTX_QW0_EXT_IP_MASK)) {
> 2367 oudph->check = ~csum_tcpudp_magic(oiph->saddr,
> 2368 oiph->daddr,
> 2369 (skb->len -
> skb_transport_offset(skb)),
> 2370 IPPROTO_UDP, 0);
>
> if compiler reports that oudph and oiph can be unitialized here,
> it's not enough just to set them to NULL.
>
> Do we need probably to check that variables was initialized before access
> here?
> i.e. add oudph && oiph into condition?
Sorry, I should not have mentioned my patch when it wasn't meant as
a serious submission. The patch I sent did not have a proper changelog
on it so it failed to explain it.
The reason why my patch works correctly is that the check on "l4_tunnel
== I40E_TXD_CTX_UDP_TUNNELING" means we can only get here if the
two variables have been initialized, and gcc fails to see this.
Jeff mentioned that he already had a patch for this, so I did not follow
up with a real patch.
Arnd
