9fans archive / 1998 / 10 / 64 /    prev next

From: jmk@pla... jmk@pla...
Subject: [9fans] NCR SCSI drivers
Date: Wed, 21 Oct 1998 09:42:37 -0400

	>And then 
	>
	>	scsi0: ncr53c8xx: port 0 irq 0 addr 0 size 0
	>	
	>and a config prompt. So it looks like it is scsi.c is calling
	>the ncr driver's reset, and the ncr driver thinks it succeeded.
	>However, I'm wondering about all the '0' values -- I don't know
	>if that is more of the same 'everything hidden by PCI' or not.

the plan9 fs driver doesn't fill in any of the ISAConf structure in
the reset routine which the calling scsireset() routine prints out. there's
no reason it doesn't, just pass the isa pointer to the init() routine too
and fill it in there. this is essentially what brazil does because
brazil handles multiple controllers and sometimes the port value has to
be used to differentiate them

	Scsiio
	ncr53c8xxreset(int ctlrno, ISAConf* isa)
	{
		int differential, o;
		Controller *ctlr;
		static int scandone;
	
		if(scandone == 0){
			scanpci();
			scandone = 1;
		}
	
		differential = 0;
		for (o = 0; o < isa->nopt; o++) {
			if (strcmp(isa->opt[o], "diff") == 0)
				differential = 1;
		}
	
		if((ctlr = xalloc(sizeof(Controller))) == 0){
			print("scsi#%d: %s: controller allocation failed\n",
				ctlrno, isa->type);
			return 0;
		}
		ctlrxx[ctlrno] = ctlr;
		ctlr->ctlrno = ctlrno;
	
		if (init(ctlr, isa, differential))
			return exec;
	
		return 0;
	}

and init() does stuff like

	isa->port = (ulong)KADDR(regpa);
	isa->irq = pcidev->intl;

before returning, leading to a message like

	scsi#0: ncr53c8xx: port 0xE4000000 irq 9

at boot.

it's also worth mentioning that in brazil the reset() and exec() routines
are the same for both the fileserver and normal kernel, a small victory for
maintainability.

--jim