Re: ybin on powerpc

Re: ybin on powerpc

[Andrew Morton]

(cc's added)

On Sun, 20 Aug 2006 07:57:24 +1000
Benjamin Herrenschmidt <benh@kernel.crashing.org> wrote:

> On Sat, 2006-08-19 at 11:19 -0700, Andrew Morton wrote:
> > I notice that 2.6.18-rc4-mm1 and 2.6.18-rc4-mm2-candidate do this:
> > 
> > g5:/home/akpm> 0 ybin
> > ofpath: /dev/sdb: Device not configured
> > ybin: Unable to find OpenFirmware path for boot=/dev/sdb2
> > ybin: Please add ofboot=<path> where <path> is the OpenFirmware path to /dev/sdb2 to /etc/yaboot.conf
> > ofpath: /dev/sdb: Device not configured
> > ybin: Unable to determine OpenFirmware path for macosx=/dev/sdb3
> > ybin: Try specifying the real OpenFirmware path for macosx=/dev/sdb3 in /etc/yaboot.conf
> 
> Did the layout of scsi stuff in either /proc or sysfs change ?

Bisection shows that this brokenness is introduced by git-scsi-misc.patch.

Someone has gone in and made some ill-advised "improvements" to
/proc/scsi/scsi.  The switch from "Direct-Access" to "Direct access" broke
ofpath.  I fixed that (in ofpath) but it remains broken for other reasons.

James, this is a non-compatible change to the kernel->userspace interface.  Please
drop it like a hot potato.

Re: ybin on powerpc

[Matthew Wilcox]

On Sat, Aug 19, 2006 at 06:28:18PM -0700, Andrew Morton wrote:
> Bisection shows that this brokenness is introduced by git-scsi-misc.patch.
> 
> Someone has gone in and made some ill-advised "improvements" to
> /proc/scsi/scsi.  The switch from "Direct-Access" to "Direct access" broke
> ofpath.  I fixed that (in ofpath) but it remains broken for other reasons.
> 
> James, this is a non-compatible change to the kernel->userspace interface.  Please
> drop it like a hot potato.

This one's my fault.  I didn't realise that bit got exposed to userspace
like that.  James, do you want a replacement patch for the bad one, or
an incremental one which restores the old spelling?

BTW, the SCSI specs changed the name from CD-ROM to CD/DVD.  I'd like to
reflect that change in /proc/scsi/scsi; is there any userspace currently
relying on that spelling?

Re: ybin on powerpc

[Andrew Morton]

On Sat, 19 Aug 2006 19:56:30 -0600
Matthew Wilcox <matthew@wil.cx> wrote:

> BTW, the SCSI specs changed the name from CD-ROM to CD/DVD.  I'd like to
> reflect that change in /proc/scsi/scsi; is there any userspace currently
> relying on that spelling?

I think that's basically unknowable.  How does one grep all userspace?

The only practical way to find out is to change it, see what breaks and
hope that people report it in time.  That's a terribly blunt weapon.

It's better to get the interfaces right first time ;)

Re: ybin on powerpc

[Matthew Wilcox]

On Sat, Aug 19, 2006 at 07:28:11PM -0700, Andrew Morton wrote:
> On Sat, 19 Aug 2006 19:56:30 -0600
> Matthew Wilcox <matthew@wil.cx> wrote:
> > BTW, the SCSI specs changed the name from CD-ROM to CD/DVD.  I'd like to
> > reflect that change in /proc/scsi/scsi; is there any userspace currently
> > relying on that spelling?
> 
> I think that's basically unknowable.  How does one grep all userspace?

apt-cache search?  ;-)

> The only practical way to find out is to change it, see what breaks and
> hope that people report it in time.  That's a terribly blunt weapon.
> 
> It's better to get the interfaces right first time ;)

Yeah; kind of hard to do when the SCSI committee changes the spec on you.

Re: ybin on powerpc

[Andrew Morton]

On Sat, 19 Aug 2006 21:05:59 -0600
Matthew Wilcox <matthew@wil.cx> wrote:

> On Sat, Aug 19, 2006 at 07:28:11PM -0700, Andrew Morton wrote:
> > On Sat, 19 Aug 2006 19:56:30 -0600
> > Matthew Wilcox <matthew@wil.cx> wrote:
> > > BTW, the SCSI specs changed the name from CD-ROM to CD/DVD.  I'd like to
> > > reflect that change in /proc/scsi/scsi; is there any userspace currently
> > > relying on that spelling?
> > 
> > I think that's basically unknowable.  How does one grep all userspace?
> 
> apt-cache search?  ;-)

Well yeah.  Smiley noted, but let us remember that "userspace" includes
non-published and non-open-source code.

> > The only practical way to find out is to change it, see what breaks and
> > hope that people report it in time.  That's a terribly blunt weapon.
> > 
> > It's better to get the interfaces right first time ;)
> 
> Yeah; kind of hard to do when the SCSI committee changes the spec on you.

I think people will continue to understand what we mean.

Re: ybin on powerpc

[James Bottomley]

On Sat, 2006-08-19 at 18:28 -0700, Andrew Morton wrote:
> Bisection shows that this brokenness is introduced by
> git-scsi-misc.patch.
> 
> Someone has gone in and made some ill-advised "improvements" to
> /proc/scsi/scsi.  The switch from "Direct-Access" to "Direct access"
> broke
> ofpath.  I fixed that (in ofpath) but it remains broken for other
> reasons.
> 
> James, this is a non-compatible change to the kernel->userspace
> interface.  Please
> drop it like a hot potato.

Well ... OK, we shouldn't break things gratuitously.  However,
the /proc/scsi interface is deprecated and has been for a while now.

Secondly, as Matthew pointed out, the SCSI standards actually keep
changing their minds about the names of these things ...

But, thirdly, in order to get around the issue, we provided the

/sys/class/scsi_disk/<x:y:z:i>/device/type

interface which supplies the numeric field value precisely so people
shouldn't parse the translated strings.

So, it sounds like yaboot is also in need of modification.

James

Re: ybin on powerpc

[Benjamin Herrenschmidt]

> Well ... OK, we shouldn't break things gratuitously.  However,
> the /proc/scsi interface is deprecated and has been for a while now.
> 
> Secondly, as Matthew pointed out, the SCSI standards actually keep
> changing their minds about the names of these things ...
> 
> But, thirdly, in order to get around the issue, we provided the
> 
> /sys/class/scsi_disk/<x:y:z:i>/device/type
> 
> interface which supplies the numeric field value precisely so people
> shouldn't parse the translated strings.
> 
> So, it sounds like yaboot is also in need of modification.

CC'ed Paul Nasrat who is the current yaboot maintainer

(Paul, the problem is about a /proc/scsi change breaking ofpath. The
change will probably be reverted, but still, ofpath should be changed to
not rely on that crap anymore)

Cheers,
Ben.

Re: ybin on powerpc

[Andrew Morton]

On Mon, 21 Aug 2006 07:54:17 +1000
Benjamin Herrenschmidt <benh@kernel.crashing.org> wrote:

> 
> > Well ... OK, we shouldn't break things gratuitously.  However,
> > the /proc/scsi interface is deprecated and has been for a while now.
> > 
> > Secondly, as Matthew pointed out, the SCSI standards actually keep
> > changing their minds about the names of these things ...
> > 
> > But, thirdly, in order to get around the issue, we provided the
> > 
> > /sys/class/scsi_disk/<x:y:z:i>/device/type
> > 
> > interface which supplies the numeric field value precisely so people
> > shouldn't parse the translated strings.
> > 
> > So, it sounds like yaboot is also in need of modification.
> 
> CC'ed Paul Nasrat who is the current yaboot maintainer
> 
> (Paul, the problem is about a /proc/scsi change breaking ofpath. The
> change will probably be reverted, but still, ofpath should be changed to
> not rely on that crap anymore)
> 

yup.

James, I think the best (only) way in which we can communicate with all
users of the kernel is via the kernel.  So we add a once-per-boot printk on
first access to /proc/scsi/scsi (it should include current->comm).

That should get people migrating off /proc/scsi/scsi as quickly as we can
reasonably expect.  Then in a year or four we can perhaps remove it.

Re: ybin on powerpc

[Paul Nasrat]

On Mon, 2006-08-21 at 07:54 +1000, Benjamin Herrenschmidt wrote:
> > Well ... OK, we shouldn't break things gratuitously.  However,
> > the /proc/scsi interface is deprecated and has been for a while now.
> > 

> > /sys/class/scsi_disk/<x:y:z:i>/device/type
> > 
> > interface which supplies the numeric field value precisely so people
> > shouldn't parse the translated strings.
> > 
> > So, it sounds like yaboot is also in need of modification.
> 
> CC'ed Paul Nasrat who is the current yaboot maintainer
> 
> (Paul, the problem is about a /proc/scsi change breaking ofpath. The
> change will probably be reverted, but still, ofpath should be changed to
> not rely on that crap anymore)

Thanks for the heads up.  I'll look into fixing up ofpath not to
use /proc/scsi.  I'm just in the process of moving countries, so I might
not get a chance to look at for a week or two.

Paul