Re: Race in kdb()

[Keith Owens <kaos@xxxxxxx>]

On Mon, 19 Apr 2004 19:50:08 +0500, 
Ananth N Mavinakayanahalli <ananth@xxxxxxxxxx> wrote:
>Hi,
>
>While working on KDB for ppc64, I hit a race in kdb(). Verified that the
>problem exists on x86 too..
>
>Problem occurs when a bp is set on a very frequently hit routine (eg., 
>sys_read() or schedule()), and multiple processors hit the bp 
>simultaneously. This is the problem I am seeing...
>
>        CPU0                            CPU1
>         |                               |
>        kdb(reason = break)              |
>         |                               |
>         |                               |
>        spin_lock()                      |
>         |                               |
>        kdb_initial_cpu = 0;             |
>         |                               |
>        spin_unlock()                    |
>         ..                       kdb(reason = break)
>         |                               |
>         |                      if (reason != KDB_REASON_SWITCH) {
>         |                      if (KDB_IS_RUNNING() && !KDB_STATE(REENTRY))
>         |                              <is true>
>        |                              ....
>         |                               
>         |
>        smp_kdb_stop();

I know.  Fixing this trace requires significant changes to the way that
breakpoints are handled and will break kdb for all architectures.  It
is on my list for kdb v5.0.

---------------------------
Use http://oss.sgi.com/ecartis to modify your settings or to unsubscribe.