On 6 November 2013 13:05, Ard Biesheuvel ard.biesheuvel@linaro.org wrote:
Well, to be honest, my primary motivation for posting it to the list was so I could refer to it from the JIRA issue you told me about yesterday, but I would be happy to fix these minor issues and do a proper submission as well. However, considering that this is just a minor slice of the crypto extensions, and that I should probably also implement disassemble support (?) for a complete implementation, I did not think the coverage of the patch would be sufficient for a proper upstream submission. (I mainly wrote it to unblock my own work)
I don't have convenient access to the spec at the moment, so I couldn't check how much of the instructions this covered, but my general view is that we're adding the new-in-v8 instructions piecemeal, so provided the implementation of the instructions that the patch adds is reasonably complete I don't see the need to hold them up waiting for full coverage of every insn in the crypto extensions.
Thanks, I was wondering about that. Should there be another check? Not all v8 cores will implement these extensions, so we might also want to emulate one that doesn't, I suppose?
Yes, that would probably be better, actually: add a new ARM_FEATURE_ bit and then make ARM_FEATURE_V8 imply that new bit in arm_cpu_realizefn().
thanks -- PMM