Hi,
five minutes starring at the wd33c93 driver solved the problem. When
applying Linus' patches a superfluous disable_irq() was left over in
wd33c93.c:wd33c93_queuecommand(). Removing the line makes the box feel
like a computer again. This should also explain the ultralong times
measured by Mark during which interrupt were disabled. Patch appended below.
Ralf
--- drivers/scsi/wd33c93.c.orig Sun Nov 30 07:12:08 1997
+++ drivers/scsi/wd33c93.c Sun Nov 30 07:53:27 1997
@@ -292,7 +292,6 @@
Scsi_Cmnd *tmp;
unsigned long flags;
- disable_irq(cmd->host->irq);
DB(DB_QCMD,printk("Q-%d-%02x-%ld( ",cmd->target,cmd->cmnd[0],cmd->pid));
/* Set up a few fields in the Scsi_Cmnd structure for our own use:
|