broken reverse_scan in aic7xxx

broken reverse_scan in aic7xxx

[Stephan Seidl]

Hallo,
between the Linux versions 2.6.8 and 2.6.14, anybody cleaned up
the code of `linux/drivers/scsi/aic7xxx/aic7xxx_osm.c'.
The Kernel 2.6.8 I refer to came with Debian Sarge.

$ ls -al linux-2.6.8-16s/drivers/scsi/aic7xxx/aic7xxx_osm.c
-rw-r--r--  1 seidl seidl 143292 May 19  2005 linux-2.6.8-16s/drivers/scsi/aic7xxx/aic7xxx_osm.c
$ ls -al linux-2.6.14.3/drivers/scsi/aic7xxx/aic7xxx_osm.c
-rw-r--r--  1 seidl seidl 74505 Nov 24  2005 linux-2.6.14.3/drivers/scsi/aic7xxx/aic7xxx_osm.c

A victim of the cleaning operation is the option `aic7xxx=reverse_scan'
which is still documented, but, which does no longer work. Here is some
output.

$ (cd linux-2.6.8-16s/drivers/scsi/aic7xxx && grep reverse_scan *)
aic79xx_osm.c:static uint32_t aic79xx_reverse_scan;
aic79xx_osm.c:" reverse_scan            Sort PCI devices highest Bus/Slot
to lowest\n"
aic79xx_osm.c:  if (aic79xx_reverse_scan != 0)
aic79xx_osm.c:  if (aic79xx_reverse_scan != 0)
aic79xx_osm.c:          { "reverse_scan", &aic79xx_reverse_scan },
aic7xxx_osm.c:static uint32_t aic7xxx_reverse_scan;
aic7xxx_osm.c:" reverse_scan            Sort PCI devices highest Bus/Slot
to lowest\n"
aic7xxx_osm.c:          if (aic7xxx_reverse_scan != 0)
aic7xxx_osm.c:          if (aic7xxx_reverse_scan != 0)
aic7xxx_osm.c:          { "reverse_scan", &aic7xxx_reverse_scan },

$ (cd linux-2.6.14.3/drivers/scsi/aic7xxx && grep reverse_scan *)
aic79xx_osm.c:static uint32_t aic79xx_reverse_scan;
aic79xx_osm.c:" reverse_scan            Sort PCI devices highest Bus/Slot
to lowest\n"
aic79xx_osm.c:  if (aic79xx_reverse_scan != 0)
aic79xx_osm.c:  if (aic79xx_reverse_scan != 0)
aic79xx_osm.c:          { "reverse_scan", &aic79xx_reverse_scan },
aic7xxx_osm.c:" reverse_scan            Sort PCI devices highest Bus/Slot
to lowest\n"

Since `aic7xxx=reverse_scan' is part of my disaster recovery concept,
I would like to ask whether anybody has the resources to reactivate
this driver option as it was before ?

Kind regards,
Stephan Seidl

Re: broken reverse_scan in aic7xxx

[Matthew Wilcox]

On Sat, Jul 08, 2006 at 05:21:59PM +0200, Stephan Seidl wrote:
> A victim of the cleaning operation is the option `aic7xxx=reverse_scan'
> which is still documented, but, which does no longer work. Here is some
> output.
> 
> Since `aic7xxx=reverse_scan' is part of my disaster recovery concept,
> I would like to ask whether anybody has the resources to reactivate
> this driver option as it was before ?

The documentation should be removed ;-)

It's a simple choice between reverse_scan and supporting PCI hotplug.
You can't have both, and we chose the latter.  The same choice has been
made in virtually every other device driver.

I'm afraid you're going to have to change your concept.

Re: broken reverse_scan in aic7xxx

[Christoph Hellwig]

On Sat, Jul 08, 2006 at 09:44:49AM -0600, Matthew Wilcox wrote:
> On Sat, Jul 08, 2006 at 05:21:59PM +0200, Stephan Seidl wrote:
> > A victim of the cleaning operation is the option `aic7xxx=reverse_scan'
> > which is still documented, but, which does no longer work. Here is some
> > output.
> > 
> > Since `aic7xxx=reverse_scan' is part of my disaster recovery concept,
> > I would like to ask whether anybody has the resources to reactivate
> > this driver option as it was before ?
> 
> The documentation should be removed ;-)
> 
> It's a simple choice between reverse_scan and supporting PCI hotplug.
> You can't have both, and we chose the latter.  The same choice has been
> made in virtually every other device driver.
> 
> I'm afraid you're going to have to change your concept.

Or you could submit a patch to implement a reverse scanning option in
the core PCI code..

Re: broken reverse_scan in aic7xxx

[Matthew Wilcox]

On Sat, Jul 08, 2006 at 06:48:44PM +0100, Christoph Hellwig wrote:
> On Sat, Jul 08, 2006 at 09:44:49AM -0600, Matthew Wilcox wrote:
> > It's a simple choice between reverse_scan and supporting PCI hotplug.
> > You can't have both, and we chose the latter.  The same choice has been
> > made in virtually every other device driver.
> 
> Or you could submit a patch to implement a reverse scanning option in
> the core PCI code..

You'd want to specify that on a per-driver basis, though, surely.  Or
all your ethernet cards would get reversed too.  Oh, but it wouldn't be
enough for sym2 (iterate through card types instead of through pci ids),
so you'd want another option for that.

In the end it becomes a hideous mess and it's just easier to adapt
userspace.

Hey, anyone want to implement SIOCSIFNAME for scsi devices?  ;-)

Re: broken reverse_scan in aic7xxx

[Matthew Wilcox]

On Sat, Jul 08, 2006 at 12:37:52PM -0600, Matthew Wilcox wrote:
> On Sat, Jul 08, 2006 at 06:48:44PM +0100, Christoph Hellwig wrote:
> > Or you could submit a patch to implement a reverse scanning option in
> > the core PCI code..
> 
> You'd want to specify that on a per-driver basis, though, surely.  Or
> all your ethernet cards would get reversed too.  Oh, but it wouldn't be
> enough for sym2 (iterate through card types instead of through pci ids),
> so you'd want another option for that.

Actually, it's even worse.  PCI doesn't do its own matching any more;
it's now part of the driver core ... so the flag/alternate
implementation has to go all through the callchain:

pci_register_driver
  __pci_register_driver
    driver_register
      bus_add_driver
        driver_attach

At this point, you'd want to change/replace the bus_for_each_dev()
macro.  Wouldn't be too hard to have a bus_for_each_dev_rev() to give
aic7xxx backward compatibility, but sym2 is completely SOL.  The real
pain is passing the information that we want to do this in reverse
all the way down the call chain.

Re: broken reverse_scan in aic7xxx

[Arjan van de Ven]

On Sat, 2006-07-08 at 12:37 -0600, Matthew Wilcox wrote:
> On Sat, Jul 08, 2006 at 06:48:44PM +0100, Christoph Hellwig wrote:
> > On Sat, Jul 08, 2006 at 09:44:49AM -0600, Matthew Wilcox wrote:
> > > It's a simple choice between reverse_scan and supporting PCI hotplug.
> > > You can't have both, and we chose the latter.  The same choice has been
> > > made in virtually every other device driver.
> > 
> > Or you could submit a patch to implement a reverse scanning option in
> > the core PCI code..
> 
> You'd want to specify that on a per-driver basis, though, surely.  Or
> all your ethernet cards would get reversed too.  Oh, but it wouldn't be
> enough for sym2 (iterate through card types instead of through pci ids),
> so you'd want another option for that.
> 
> In the end it becomes a hideous mess and it's just easier to adapt
> userspace.
> 
> Hey, anyone want to implement SIOCSIFNAME for scsi devices?  ;-)

it's called udev, and it sets device names by UUID already ;-)

Re: broken reverse_scan in aic7xxx

[Matthew Wilcox]

On Sun, Jul 09, 2006 at 10:44:43AM +0200, Arjan van de Ven wrote:
> On Sat, 2006-07-08 at 12:37 -0600, Matthew Wilcox wrote:
> > Hey, anyone want to implement SIOCSIFNAME for scsi devices?  ;-)
> 
> it's called udev, and it sets device names by UUID already ;-)

That's not true.  The kernel has a name for the scsi disc with major 8,
minor 0 and it's 'sda'.  It uses it in all the printks from sd.c.  The
poor user now has the job of mapping UUID <-> sdN <-> bus address <->
physical location, for we use the first three and often what they really
need to know is the fourth.