On 13 March 2012 03:15, Christoffer Dall <c.dall at virtualopensystems.com> wrote: > On Mon, Mar 12, 2012 at 10:43 PM, Rusty Russell <rusty at rustcorp.com.au> wrote: >> But I suspect it wouldn't be so explicit in the manual if there wasn't >> some silicon coming which actually needs this... > not sure what the process is here, maybe they just don't want to > dictate that architecture, if that for some reason can clutter some > hardware implementation logic magic. Who knows. As a general rule of thumb you can assume that impdef leeway in the architecture manual is there for a reason, yes. If you have access to the AEM model it has a take_ccfail_undef flag you can use to enable (some of) this behaviour. Since we've dropped all that asm now, I don't think the 'cc failed?' checks are all that much code and I'd prefer to have them in rather than out. This would be in line with existing code in the kernel for eg swp emulation which is handling this case already. -- PMM
