Re: RT-Linux and SMP

Ingo Molnar (
Fri, 25 Apr 1997 03:06:11 +0200 (MET DST)

On Fri, 25 Apr 1997, Ingo Molnar wrote:

> On Fri, 25 Apr 1997, Ingo Molnar wrote:
> > [...] to build a per-driver 'EIP color map', which
> > works similarly to profiling, but does not build a histogram, just a "i've
> > been here" map. [...]
> ie. we can get this via append="profile=1" and by trivially converting
> /proc/profile, plus a mildly smart shellscript which builds the per-line
> colormap for _one_ C module.
> then we do a diff -u mydriver.out.uni mydriver.out.smp ?

a few finishing touches to the concept:

hm, we have to make the 'color' irq dependent. (16 bits deep with 16
irqs). And in the SMP case we have to broadcast every received interrupt
via a special broadcast IRQ to other IRQs (both when we enter do_IRQ/
do_fast_IRQ, and when we exit them). This way can we catch only those
dreaded 'sneaking interrupts' which are prevented by synronize_irqs().

we can resurrect the discontinued rescheduling interrupt for this, whee.

but then with these things we have a per-IRQ 'color map', which should be
the very same for uniprocessor and SMP kernels, and which is different if
broken in _any_ way?

-- mingo