2.4.x libata resync

2.4.x libata resync

[Tom Mortensen]

To Jeff Garzik & others,
I was wondering if there are any plans for another resync of the latest
2.6.x libata changes back into the 2.4.x kernel?
thanks,
-tom
ps, sorry for repeat of previous message with no subject line.

Re: 2.4.x libata resync

[Willy Tarreau]

On Fri, Sep 15, 2006 at 10:14:07PM -0700, Tom Mortensen wrote:
> To Jeff Garzik & others,
> I was wondering if there are any plans for another resync of the latest
> 2.6.x libata changes back into the 2.4.x kernel?

When Jeff posted his last version (which got merged), he said that it
would be his last work on this backport. I've been regularly checking
what has changed in 2.6, because often some bugs are fixed, but I see
that the code has considerably evolved since the last resync, and I'm
not even sure that those bugfixes are needed for 2.4.

A full resync of latest 2.6 would require a considerable effort it seems.
Do you encounter any problems right now ? I get very few feedback from
SATA users in general.

Regards,
Willy

Re: 2.4.x libata resync

[Tejun Heo]

Hello,

Willy Tarreau wrote:
> On Fri, Sep 15, 2006 at 10:14:07PM -0700, Tom Mortensen wrote:
>> To Jeff Garzik & others,
>> I was wondering if there are any plans for another resync of the latest
>> 2.6.x libata changes back into the 2.4.x kernel?
> 
> When Jeff posted his last version (which got merged), he said that it
> would be his last work on this backport. I've been regularly checking
> what has changed in 2.6, because often some bugs are fixed, but I see
> that the code has considerably evolved since the last resync, and I'm
> not even sure that those bugfixes are needed for 2.4.
> 
> A full resync of latest 2.6 would require a considerable effort it seems.
> Do you encounter any problems right now ? I get very few feedback from
> SATA users in general.

I don't think it's gonna happen.  Later libata changes depend on a 
number of SCSI updates, which in turn are deeply dependent upon new 
driver model used in 2.6.  So, apart from bug fixes, there won't be 2.4 
resync.

-- 
tejun

Re: 2.4.x libata resync

[Willy Tarreau]

On Sat, Sep 16, 2006 at 03:27:12PM +0900, Tejun Heo wrote:
> Hello,
> 
> Willy Tarreau wrote:
> >On Fri, Sep 15, 2006 at 10:14:07PM -0700, Tom Mortensen wrote:
> >>To Jeff Garzik & others,
> >>I was wondering if there are any plans for another resync of the latest
> >>2.6.x libata changes back into the 2.4.x kernel?
> >
> >When Jeff posted his last version (which got merged), he said that it
> >would be his last work on this backport. I've been regularly checking
> >what has changed in 2.6, because often some bugs are fixed, but I see
> >that the code has considerably evolved since the last resync, and I'm
> >not even sure that those bugfixes are needed for 2.4.
> >
> >A full resync of latest 2.6 would require a considerable effort it seems.
> >Do you encounter any problems right now ? I get very few feedback from
> >SATA users in general.
> 
> I don't think it's gonna happen.  Later libata changes depend on a 
> number of SCSI updates, which in turn are deeply dependent upon new 
> driver model used in 2.6.  So, apart from bug fixes, there won't be 2.4 
> resync.

There are a bunch of small patches in the early 2.6 version which look
like bugfixes, but with non-descriptive comments, so I'm not sure what
they fix. Several of them would apply to 2.4, but I don't want to touch
this area as long as nobody complains about problems.

Regards,
Willy

Re: 2.4.x libata resync

[Jeff Garzik]

Willy Tarreau wrote:
> There are a bunch of small patches in the early 2.6 version which look
> like bugfixes, but with non-descriptive comments, so I'm not sure what
> they fix. Several of them would apply to 2.4, but I don't want to touch
> this area as long as nobody complains about problems.

Oh there are tons of SATA bug fixes that 2.4.x is missing.  One of the 
biggest is the completely crappy exception handling.  If a SATA device 
is unplugged or spazzes out, the system may or may not recover.

	Jeff

Re: 2.4.x libata resync

[Willy Tarreau]

On Sat, Sep 16, 2006 at 11:49:06AM -0400, Jeff Garzik wrote:
> Willy Tarreau wrote:
> >There are a bunch of small patches in the early 2.6 version which look
> >like bugfixes, but with non-descriptive comments, so I'm not sure what
> >they fix. Several of them would apply to 2.4, but I don't want to touch
> >this area as long as nobody complains about problems.
> 
> Oh there are tons of SATA bug fixes that 2.4.x is missing.  One of the 
> biggest is the completely crappy exception handling.  If a SATA device 
> is unplugged or spazzes out, the system may or may not recover.

Already encountered on sata_nv in a sun x2100 :-)

Jeff, I did not want to blindly merge patches from 2.6 to 2.4, but if
you point me to a few ones you consider important, I'm willing to merge
them.

Regards,
Willy

Re: 2.4.x libata resync

[Jeff Garzik]

Willy Tarreau wrote:
> On Sat, Sep 16, 2006 at 11:49:06AM -0400, Jeff Garzik wrote:
>> Willy Tarreau wrote:
>>> There are a bunch of small patches in the early 2.6 version which look
>>> like bugfixes, but with non-descriptive comments, so I'm not sure what
>>> they fix. Several of them would apply to 2.4, but I don't want to touch
>>> this area as long as nobody complains about problems.
>> Oh there are tons of SATA bug fixes that 2.4.x is missing.  One of the 
>> biggest is the completely crappy exception handling.  If a SATA device 
>> is unplugged or spazzes out, the system may or may not recover.
> 
> Already encountered on sata_nv in a sun x2100 :-)
> 
> Jeff, I did not want to blindly merge patches from 2.6 to 2.4, but if
> you point me to a few ones you consider important, I'm willing to merge
> them.

As was hinted, it's not that easy, otherwise someone would have done it 
by now.  libata bug fixes require infrastructure that isn't present on 
2.4.  The overall codebase is just too different to easily pull out 
select bug fixes.

	Jeff

Re: 2.4.x libata resync

[Willy Tarreau]

On Sat, Sep 16, 2006 at 12:07:16PM -0400, Jeff Garzik wrote:
> Willy Tarreau wrote:
> >On Sat, Sep 16, 2006 at 11:49:06AM -0400, Jeff Garzik wrote:
> >>Willy Tarreau wrote:
> >>>There are a bunch of small patches in the early 2.6 version which look
> >>>like bugfixes, but with non-descriptive comments, so I'm not sure what
> >>>they fix. Several of them would apply to 2.4, but I don't want to touch
> >>>this area as long as nobody complains about problems.
> >>Oh there are tons of SATA bug fixes that 2.4.x is missing.  One of the 
> >>biggest is the completely crappy exception handling.  If a SATA device 
> >>is unplugged or spazzes out, the system may or may not recover.
> >
> >Already encountered on sata_nv in a sun x2100 :-)
> >
> >Jeff, I did not want to blindly merge patches from 2.6 to 2.4, but if
> >you point me to a few ones you consider important, I'm willing to merge
> >them.
> 
> As was hinted, it's not that easy, otherwise someone would have done it 
> by now.  libata bug fixes require infrastructure that isn't present on 
> 2.4.  The overall codebase is just too different to easily pull out 
> select bug fixes.

Of course for those. I was thinking about those which just change one
register or things like this that I cannot identify the expected effect.
If you agree, I'll enumerate the ones I've already noticed so that you
just have to say yes/no/unknown on them. Don't worry, I don't want to
spend lots of hours on this, since as I said, I do not receive any
feedback from SATA users on 2.4 (neither positive nor negative).

Willy

Re: 2.4.x libata resync

[Jeff Garzik]

Willy Tarreau wrote:
> Of course for those. I was thinking about those which just change one
> register or things like this that I cannot identify the expected effect.
> If you agree, I'll enumerate the ones I've already noticed so that you
> just have to say yes/no/unknown on them. Don't worry, I don't want to
> spend lots of hours on this, since as I said, I do not receive any
> feedback from SATA users on 2.4 (neither positive nor negative).


I'm more than happy to review any 2.4 libata patches people post.

	Jeff

Re: 2.4.x libata resync

[Willy Tarreau]

On Sat, Sep 16, 2006 at 12:27:33PM -0400, Jeff Garzik wrote:
> Willy Tarreau wrote:
> >Of course for those. I was thinking about those which just change one
> >register or things like this that I cannot identify the expected effect.
> >If you agree, I'll enumerate the ones I've already noticed so that you
> >just have to say yes/no/unknown on them. Don't worry, I don't want to
> >spend lots of hours on this, since as I said, I do not receive any
> >feedback from SATA users on 2.4 (neither positive nor negative).
> 
> 
> I'm more than happy to review any 2.4 libata patches people post.

OK, I will check what I found in 2.6 when I have time.

Thanks,
Willy

Re: 2.4.x libata resync

[Tom Mortensen]

Thank you for the info.  In studying the code that's what my
impression was as well (that it would require a considerable effort).

I haven't encountered any problems with SATA in 2.4.33, but I guess
what I was concerned about was when/if it might become necessary to
migrate to 2.6.x in order to support newer hardware.  I guess I'll
know when I encounter it :)

On 9/15/06, Willy Tarreau <w@1wt.eu> wrote:
> On Fri, Sep 15, 2006 at 10:14:07PM -0700, Tom Mortensen wrote:
> > To Jeff Garzik & others,
> > I was wondering if there are any plans for another resync of the latest
> > 2.6.x libata changes back into the 2.4.x kernel?
>
> When Jeff posted his last version (which got merged), he said that it
> would be his last work on this backport. I've been regularly checking
> what has changed in 2.6, because often some bugs are fixed, but I see
> that the code has considerably evolved since the last resync, and I'm
> not even sure that those bugfixes are needed for 2.4.
>
> A full resync of latest 2.6 would require a considerable effort it seems.
> Do you encounter any problems right now ? I get very few feedback from
> SATA users in general.
>
> Regards,
> Willy
>
>

Re: 2.4.x libata resync

[Jeff Garzik]

Tom Mortensen wrote:
> I was wondering if there are any plans for another resync of the latest
> 2.6.x libata changes back into the 2.4.x kernel?

No plans from me, anyway.

The 2.6.x SCSI API is quite different from the 2.4 API.  Dropping 2.4 
support removed a lot of limitations on the codebase, and allowed 
accelerated development.

That shouldn't stop any sufficiently motivated 3rd party, though...

	Jeff

Re: 2.4.x libata resync

[Mikael Pettersson]

On Sat, 16 Sep 2006 18:05:20 +0200, Willy Tarreau wrote:
>> >>Oh there are tons of SATA bug fixes that 2.4.x is missing.  One of the 
>> >>biggest is the completely crappy exception handling.  If a SATA device 
>> >>is unplugged or spazzes out, the system may or may not recover.
>> >
>> >Already encountered on sata_nv in a sun x2100 :-)
>> >
>> >Jeff, I did not want to blindly merge patches from 2.6 to 2.4, but if
>> >you point me to a few ones you consider important, I'm willing to merge
>> >them.
>> 
>> As was hinted, it's not that easy, otherwise someone would have done it 
>> by now.  libata bug fixes require infrastructure that isn't present on 
>> 2.4.  The overall codebase is just too different to easily pull out 
>> select bug fixes.
>
>Of course for those. I was thinking about those which just change one
>register or things like this that I cannot identify the expected effect.
>If you agree, I'll enumerate the ones I've already noticed so that you
>just have to say yes/no/unknown on them. Don't worry, I don't want to
>spend lots of hours on this, since as I said, I do not receive any
>feedback from SATA users on 2.4 (neither positive nor negative).

Here's some positive feedback: My leafnode News server
(a 440BX chipset mobo with a Tualatin PIII-S) has been using
Promise SATA PCI cards (currently a SATA300 TX2plus) for its
disk storage since the SATA update patch was made available
for 2.4.29. It's been fast and reliable the entire time.

/Mikael