> No. There is hardware like Yukon-EC used in mac mini where the cost > of the CPU unaligned access is trivial and the cost of copying every > frame to shut up the stupid warning would be high. This is correct, there are architectures that do support unaligned access in trivial way. But there are others that will simply mis-behave in this case. Most important here is ARM that is found in many hand-helds. Impact is, drivers that don't align IP header on dword will simply cease to work on hand-helds. Among drivers, most important case is wifi ones, there we do have mix of 24/26 byte headers. Having wifi non functional on hand-held is not a good thing. To ensure valid operation, one can use get_unaligned/put_unaligned; but this will introduce overhead on architectures that don't have trivial implementation of get_unaligned like #define get_unaligned(ptr) (*(ptr)) #define put_unaligned(val, ptr) ((void)( *(ptr) = (val) )) Brief examination shows that among architectures in kernel tree only these do support trivial non-aligned access: m68k cris x86 s390 powerpc What is interesting, alpha (Yukon is alpha, yes?) is not in this list; it use generic implementation (from asm-generic) for get_unaligned; this generic implementation is not same as trivial; it includes intermediate structure dereferencing. Thus, I believe, for IP stack it is wise to have IP header aligned by drivers. I warn once, so it do not flood system log with messages. Intention it to help driver writers to identify this problem. Perhaps, this should be put under #ifdef CONFIG_DEBUG_KERNEL - To unsubscribe from this list: send the line "unsubscribe linux-net" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
