[Linux-parport] Problem with ECP+EPP

[vdb128 at picaros.org]

> > > /* Check for Intel bug. */
> > > if (priv->ecr) {
> > > 	unsigned char i;
> > > 	for (i = 0x00; i < 0x80; i += 0x20) {
> > > 		ECR_WRITE (pb, i);
> > > 		if (clear_epp_timeout (pb)) {
> > > 			/* Phony EPP in ECP. */
> > > 			return 0;
> > > 		}
> > > 	}
> > > }

This code often disables the EPP mode where it shouldn't.  Moreover,
it enables interrupts and usually initiates a spurious interrupt.  If
another handler is already installed (shared IRQ) then the 
kernel will not disable the IRQ in the PIC, which leads to a screaming
interrupt for PCI cards ... locking up the system.  The PCI intx# lines
are level sensitive.

So I support removal of the Intel test and propose to combine 
parport_EPP_supported() and parport_ECPEPP_supported() into one 
routine.  This gives something like 

static int __devinit parport_EPP_supported(struct parport *pb) {
	struct parport_pc_private *priv = pb->private_data;
	int result=0;
	unsigned char oecr=0x34;

	if (priv->ecrok) {
	  oecr = inb (ECONTROL (pb));
	  /* Search for SMC style EPP+ECP mode */
	  ECR_WRITE (pb, ECR_EPP<<5 | 0x14);   /* EPP no interrupts for PCI */
	  outb (0x04, CONTROL (pb));
	}

	/*
	 * Theory:
	 *	Bit 0 of STR is the EPP timeout bit, this bit is 0
	 *	when EPP is possible and is set high when an EPP timeout
	 *	occurs (EPP uses the HALT line to stop the CPU while it does
	 *	the byte transfer, an EPP timeout occurs if the attached
	 *	device fails to respond after 10 micro seconds).
	 *
	 *	This bit is cleared by either reading it (National Semi, 
	 *	Netmos, Oxford Semi)
	 *	or writing a 1 to the bit (SMC, UMC, WinBond), others ???
	 *	This bit is always high in non EPP modes.  (except Netmos)
	 */

	/* If EPP timeout bit clear then EPP available */
	if (!clear_epp_timeout(pb)) 
		goto end;  /* No way to clear timeout */

	pb->modes |= PARPORT_MODE_EPP;

	/* Set up access functions to use EPP hardware. */
	if(!priv->ecrok) {
	  pb->ops->epp_read_data = parport_pc_epp_read_data;
	  pb->ops->epp_write_data = parport_pc_epp_write_data;
	  pb->ops->epp_read_addr = parport_pc_epp_read_addr;
	  pb->ops->epp_write_addr = parport_pc_epp_write_addr;
	} else {
	  /* Set up access functions to use ECP+EPP hardware. */
	  pb->ops->epp_read_data = parport_pc_ecpepp_read_data;
	  pb->ops->epp_write_data = parport_pc_ecpepp_write_data;
	  pb->ops->epp_read_addr = parport_pc_ecpepp_read_addr;
	  pb->ops->epp_write_addr = parport_pc_ecpepp_write_addr;
	}

	result=1;
 end:
	if (priv->ecrok) 
	  ECR_WRITE (pb, oecr);

	if (verbose_probing) {
	  printk (KERN_DEBUG "parport_EPP_supported(%s): result=%d\n", 
		  pb->name, result);
	}
	return(result);
}

The EPP detection in parport_pc_probe_port() then reduces to

  if (base != 0x3bc && !check_region(base+0x3, 5))
    parport_EPP_supported(p);

A further description is at  picaros.org/ftp/key/parport-2.4.23.diff .
This diff modifies too much but might get ported to the 2.6 kernel in
parts if there is interest.  (the cli sti should be replaced by semaphores)