Re: linux-next: triage for March 18, 2012

[David Howells <dhowells@redhat.com>]

Valdis.Kletnieks@vt.edu wrote:

> Umm.. it's not clear to *me* that it's intended to be a negative 16 bit?  Or
> am I just missing context not present in the patch?
> 
> (I have no idea if the rest of the patch is OK or not, but that comment
> didn't give me warm fuzzies....)

Sorry, I haven't explained it well.

The patch permits a 64-bit hosted assembler to represent a large 32-bit
unsigned integer (such as 0xfffffff1) as a negative integer where the
instruction being assembled has a signed immediate operand.

For instance, the ANDI instruction on FRV takes a small signed integer (10
bits IIRC) that it sign-extends to 32-bits before using - so to clear a single
low-order bit, I pass in, say, ~0x2 and the assembler represents this as a
negative value.

The problem was that in a 64-bit hosted assembler, the unsigned 32-bit integer
gets converted to an unsigned 64-bit integer - which doesn't then appear
negative (whereas in a 32-bit hosted assembler it does).

David