On Tue, Aug 17, 2021 at 11:14:08AM +0200, Hannes Reinecke wrote:

The current handler does both, bus reset and host reset.
So split them off into two distinct functions.

Well, it splits out both the bus and the host reset from sym_eh_handler,
which also handles abort and device reset.

+	/*
+	 * Escalate to host reset if the PCI bus went down
 	 */

That's a pretty big change from the old code and warrants a little
explanation.  I suspect this is fine, but I'd like to read the
thoughts about it in the commit log.

+	spin_lock_irq(shost->host_lock);
+	sym_reset_scsi_bus(np, 1);
+	spin_unlock_irq(shost->host_lock);

This loses the cmd_queued handling.  Which I guess should be fine, but
again needs to be properly documented.  Same for the host reset.

+						(sym_data->io_reset,
+						WAIT_FOR_PCI_RECOVERY*HZ);
+		spin_lock_irq(shost->host_lock);
+		sym_data->io_reset = NULL;
+		spin_unlock_irq(shost->host_lock);
+	}
+
+	if (finished_reset) {
+		sym_reset_scsi_bus(np, 0);
+		sym_start_up(shost, 1);
+	}
+
+	shost_printk(KERN_WARNING, shost, "HOST RESET operation %s.\n",
+			finished_reset==1 ? "complete" : "failed");
+	return finished_reset ? SCSI_SUCCESS : SCSI_FAILED;

The old code just returned early for the !finished_reset case, why
does this change it?