Andrew Morton wrote:I must say that having an unreliable early-current is going to be quite a
pita for evermore. Things like mcount-based tricks and
basic-block-profiling-based tricks, for example.
Is it really going to be too messy to fake up some statically-defined gdt
which points at init_task, install that before we call any C at all?
That's on my TODO list - make %gs set correctly before hitting C code, and get rid of all the early_* stuff. I had already encountered a PDA-related oops with lockdep enabled, and addressed it.
It's pretty easy to solve in general for the boot CPU, but its a bit more tricky to handle for secondary CPUs.
Laurent, could you resend your original oops? It doesn't seem to have appeared on lkml.
In the meantime, I'll work on a proper fix for this.
J