>ugh, some time ago i added a 'sti' to lock_kernel() for exactly this
>reason. The bug got reintroduced when we switched to a spin_lock()-ed
>kernel lock.
_Silenty_ forcing a sti() you was not only hiding a bug that may
cause an IPI deadlock,
but you was also pontentially opening a window for deadlocking in
a spinlock or to race with an irq handler.
>as i said in the previous mail, _dont_ do this Andrea. At least not in
>2.2. This way we might trigger CPU bugs never seen before.
The NMI should be the most robust and bug-free part of the CPU since it's
the way the hardware delivers urgent things to do to the CPU (it's been
almost unused for doing useful things in the PC so far of course).
But think if somebody would use
an Intel CPU for an embedded application where he want to use the NMI. Do
you suggest him to not use an Intel CPU because the NMI may trigger CPU
bugs never seen before? If so which CPU does have a relialable NMI
mechanism?
Andrea Arcangeli
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/