On 01.07.2019 21:56, Al Viro wrote: > On Mon, Jul 01, 2019 at 09:35:28PM +0200, Heiner Kallweit wrote: >> Sparse complains about casting to/from restricted __be16. Fix this. > > Fix what, exactly? Force-cast is not a fix - it's "STFU, I know > better, it's really correct" to sparse. Which may or may not > match the reality, but it definitely requires more in way of > commit message than "sparse says it's wrong; shut it up". > >> static void rtl8169_rx_vlan_tag(struct RxDesc *desc, struct sk_buff *skb) >> @@ -1537,7 +1537,7 @@ static void rtl8169_rx_vlan_tag(struct RxDesc *desc, >> struct sk_buff *skb) >> >> if (opts2 & RxVlanTag) >> __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), >> - ntohs(opts2 & 0xffff)); >> + ntohs((__force __be16)(opts2 & 0xffff))); >> } > > Should that be ntohs at all? What behaviour is correct on big-endian host? > > AFAICS, in that code opts2 comes from little-endian 32bit. It's converted to > host-endian, lower 16 bits (i.e. the first two octets in memory) are then > fed to ntohs. Suppose we had in-core value stored as A0, A1, A2, A3. > On little-endian that code will yield A0 * 256 + A1, treated as host-endian. > On big-endian the same will yield A1 * 256 + A0. Is that actually correct? > I think you're right and the original patch should be reverted.
> The code dealing with the value passed to __vlan_hwaccel_put_tag() as the > third argument treats it as a host-endian integer. So... Has anyone > tested that code on b-e host? Should that ntohs() actually be swab16(), > yielding (on any host) the same value we currently get for l-e hosts only? > I haven't seen any b-e host with a Realtek network chip yet.
