I am porting Linux to a new architecture (a massively multi-threaded network processor), which has a unique and bothersome characteristic: sizeof (void*) > sizeof (long). In most regards, it is a 32-bit machine, but pointers and all GPRs are 48 bit. Pointers occupy 64 bits in memory, with 16 bits ignored by load/store.
Unfortunately, the kernel is thick with pointer/int casts and universally assumes that sizeof (void*) == sizeof(long). The first phase of the port is s/long/intptr_t/ and s/unsigned long/uintptr_t/ for data that hold addresses and for casts to/from address data. I think it improves code comprehensibility to use (u)intptr_t for opaque types and for address arithmetic. I'm three weeks into the project and have almost 5000 changes. I probably have another week and a few thousand more to go before I have eliminated the pointer/int cast warnings and can work on more substantive phases.
For most (all?) existing ports, (u)intptr_t=(unsigned) long, so in theory there should be zero change to the compiled Linux image for conventional ILP=32 LP=64 architectures.
If the intent is really to extend or truncate with a cast between pointer and integer, it can be done like so:
void *ptr;
long l;
int i;
long long ll;
ptr = (void *)(intptr_t) l;
l = (long)(intptr_t) ptr;
ptr = (void *)(intptr_t) ll;
ll = (long long)(intptr_t) ptr;
ptr = (void *)(intptr_t) i;
i = (int)(intptr_t) ptr;
Are maintainers willing to accept patches to covert from (unsigned) long and (u)intptr_t ? I propose to do this in many small installments for digestibility and to moderate the downstream impact of merge conflicts.
G
--
To unsubscribe from this list: send the line "unsubscribe linux-arch" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
