On Jan 30,  4:15pm, pcah8322@artax.karlin.mff.cuni.cz (Pavel Cahyna) wrote:
-- Subject: Re: kern/32644: gsip(4) sends byte-swapped vlan tags

| >  This does not seem right because in the first part of the patch you do:
| >  
| >  +			    htole32(EXTSTS_VPKT | 
| >  +					(bswap16(VLAN_TAG_VALUE(mtag)) &
| >  +					 EXTSTS_VTCI));
| >  
| >  and in the second:
| >  
| >  | +			VLAN_INPUT_TAG(ifp, m, bswap16(extsts & EXTSTS_VTCI),
| >  
| >  because in the first patch you are not bswapping16 EXTSTS_VTCI and in the
| >  second you do?
| 
| I think it can be justified algebraically:
| 
| The firts patched part does:
| 
| mtag(host byte order) -> bswap16 -> &= EXTSTS_VTCI -> |= EXTSTS_VPKT ->
| -> htole32 -> tx descriptor
| 
| while the second does:
| 
| rx descriptor -> le32toh -> &= EXTSTS_VTCI -> bswap16 -> mtag(host byte
| order)
| 
| So, the second sequence is inverse to the first.

Thanks, I see it now.

christos