From bmc Tue Sep  2 19:01:13 2003
Subject: VTRACE changes
To: dtrace-core@kiowa
Date: Tue, 2 Sep 2003 19:01:13 -0700 (PDT)
Cc: jim.litchfield@sun.com
X-Mailer: ELM [version 2.4ME+ PL31H (25)]
MIME-Version: 1.0
Content-Type: text/plain; charset=US-ASCII
Content-Transfer-Encoding: 7bit
Content-Length:  2271
Status: RO


Well, I can say without reservation that this has been the most stressful
day of my life.  It was reported this afternoon by Jim Litchfield that
UltraSPARC-Is no longer booted when running non-DEBUG with our bits.
At around 2p, Mike and I grabbed two UltraSPARC-I tip lines and went to
work.

Working in parallel, we ascertained that it was _not_ related to the 
krtld changes, address space layout, vmem arena changes and the other
modifications we made to common code.  We further ascertained that the
machines _would_ boot if running non-DEBUG bits with a DEBUG esp or fas
driver.

This data lead us to this code in esp:

  start_action:
+ #ifdef ESPDEBUG
        while (action != ACTION_RETURN) {
                ASSERT((action >= 0) && (action <= ACTION_SELECT));
                TRACE_3(TR_FAC_SCSI, TR_ESPSVC_ACTION_CALL,
!                       "espsvc call: esp %p, action %d (%d)",
!                       esp, action, i);
!               i++;
                action = (*evec[action])(esp);
        }
+ #endif
  exit:
        TRACE_0(TR_FAC_SCSI, TR_ESPSVC_END, "espsvc_end");
  }

Unfortunately, the ESPDEBUG (which was added as part of the VTRACE cleanup),
effectively cut out the "action = (*evec[action])(esp);" line in non-DEBUG
kernels.  As it turns out, that line's important.  And fas has a nearly 
identical problem.

With the drivers modified appropriately, both esp- and fas-based machines
boot.  We are _greatly_ relieved to have this problem resolved -- thank you
Jim for reporting it.

The plan from here:  Mike will putback shortly to rectify the esp and 
fas drivers.  He and I will print out all of the VTRACE diffs, and review
them carefully again in a conference room.  Adam:  please do the same in
parallel from home.  At ~9:30p, I will begin the sync and stage from the
gate.  Assuming we don't find anything else in our review, that will be
that.  If we do find other problems, they will obviously need to be
rectified before we build.

And our RTI is approved, so we're -- amazingly -- back on target.

	- Bryan

P.S.  I didn't even bother to mention the panic that Claudia saw today.
She BFU'd last night when the archives happened to be updating, and got
toxic bits.  Let's just say that that didn't lower the anxiety level at all...