[PATCH] kdb on SMP (breakpoint hits multiple cpus simultaneously)

[Vamsi Krishna S .]

On Thu, Apr 03, 2003 at 11:55:37PM +1000, Keith Owens wrote:
> On Thu, 3 Apr 2003 14:16:07 +0530, 
> "Vamsi Krishna S ." <vamsi at in.ibm.com> wrote:
> >Yes, that is what I thought. My earlier patch fixed only the int3
> >points. I will do similar fix for debugreg breakpoints too
> >and update the man pages and code comments accordingly.
> 
> I suggest you wait a bit.  A lot of the breakpoint problems come from
> the fact that kdb tries to control who can enter kdb, in particular the
> way kdb tries to avoid two cpus entering kdb at the same time.  I have
> concluded that this design was wrong, there are some situations where
> kdb cannot prevent concurrent entry, e.g. external debugging interrupts
> that are sent to all cpus and cannot be blocked.  As usual, ia64 is a
> special case :(.
> 
> The answer is not to try to prevent concurrent entry to kdb but to
> accept that it occurs and work with it.  That means the breakpoint
> state must become per-cpu, as in the smphdr* patches.  Concurrent
> software breakpoints will be allowed.
> 
Ok. For software breakpoints, when user types "bc *", this means 
you have to remove the INT3 instruction from the last CPU that
is leaving KDB. I will take a look at the next release.

Thanks,
Vamsi.
-- 
Vamsi Krishna S.
Linux Technology Center,
IBM Software Lab, Bangalore.
Ph: +91 80 5044959
Internet: vamsi at in.ibm.com