If we change run_bottom_halves to test bh_mask_count against
zero and clear the bit itself if it finds it non-zero,
then everything will work fine without any spin_lock_irqsave.
To be rpecise, it looks like:
static inline void run_bottom_halves(void)
{
unsigned long active, mask=1;
void (**bh)(void);
atomic_t *countp = bh_mask_count;
active = get_active_bhs();
bh = bh_base;
do {
if (active & mask) {
if (*countp <= 0) {
clear_active_bhs(mask);
(*bh)();
} else {
bh_mask &= ~mask;
}
active &= ~mask;
}
bh++;
countp++;
mask <<= 1;
} while (active);
}
I'm assuming locking code in do_bottom_half which I don't undeerstand
prevents anyone from fiddling with the disable counts while this is
running. (If they could, we'd still have a race between the
get_active_bhs() call and the call to (*bh)().)
If it is possible, some retrying needs to be added to the
non-zero-counter case in case it is zeroed between the test of *countp
and the clearing of the bh_mask bit, and remove_bh() needs to be
fiddlied with considerably.
-- -Colin- 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/