Re: multiple scsi drives..

Dan Merillat (
Tue, 17 Oct 1995 15:38:08 -0400 (EDT)

On Mon, 16 Oct 1995, Eric Youngdale wrote:

> Date: Mon, 16 Oct 95 22:18 EDT
> From: Eric Youngdale <>
> To: Dan Merillat <>
> Cc: Jeremiah Jahn <>,
> Subject: Re: multiple scsi drives..
> >I take it then that there is still no way to grab the driver
> > by controller:deviceid? I'd rather have scsi/ have devices
> > of the type scsi/[a-d][1-7][1-40?] then devices that get changed
> The utility that I described generates entries like:

> crw------- 1 root root 9, 128 Aug 4 13:16 rsth4c0i5l0
I take it this is rewinding scsi-tape right? It's the only one that
dosn't follow the pattern of the rest.

[more examples snipped]

> Given entries like this, then it is possible to grab devices
> by controller:deviceid. The numbers for disks are
> sdh4 - disk, attached to host type 4 (i.e. aha1542).
> -334 - identifies which 1542, in this case the I/O port it is using.
> c0 - channel 0 (1542 only has one chan, others have more than 1)
> i0 - scsi id 0
> l0 - scsi lun 0
> p1 - partition 1

Very nice. This is what I was looking for, even if the device numbers

> If you are asking about a fixed mapping of minor numbers to
> devices, then this is not possible with a 16 bit dev_t (where you have
> 8 bits for the minor). With a 32 bit dev_t, it is still easily
> arguable that you do not have enough dynamic range to cover all of the
> possibilities without making assumptions about the ranges of some of
> the numbers, esp if you split it 16/16. I have even seen someone
> argue that you need about 48 bits for the minor number to allow for
> all of the possible permutations involving disk drives.

I know the argument. I had forgotten about multi-lun devices. I figured
8 channels + 8 devices/channel is 8 bits right there, plenty for the minor

> If and when a larger dev_t is implemented, we might recondsider
> and attempt to implement a fixed-mapping for scsi, but as of the moment
> there are no plans to change the mapping.

Well, I still have yet to see a machine running linux with more then 255
devices on it, so dynamic mapping should be ok. (at least untill we get
raid devices. I'd love to see a linux box serving for a terrabyte-tower.)

Thanks for the info, I'll check your package out.

Dan Merillat <>
Animaniac! Linux Activist and net.junkie
Also at: <>