pv_ops dom0 x86_32 PAE kernel buildlog of missing symbols...

pv_ops dom0 x86_32 PAE kernel buildlog of missing symbols...

[Marc - A. Dahlhaus]

[-- Attachment #1: Type: text/plain, Size: 266 bytes --]

Hello,

i did a quick compile of changeset 2225 today and got some missing 
symbol errors.
It is on a x86_32 PAE system with up to date stable glibc, gcc and 
binutils releases.
I attached the logs and config i used to build.

Maybe it could be of some use...

Marc

[-- Attachment #2: files.tbz --]
[-- Type: application/octet-stream, Size: 27236 bytes --]

[-- Attachment #3: Type: text/plain, Size: 138 bytes --]

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

Re: pv_ops dom0 x86_32 PAE kernel buildlog of missing symbols...

[Ian Campbell]

On Wed, 2008-12-10 at 22:07 +0100, Marc - A. Dahlhaus wrote:
> Hello,
> 
> i did a quick compile of changeset 2225 today and got some missing 
> symbol errors.
> It is on a x86_32 PAE system with up to date stable glibc, gcc and 
> binutils releases.
> I attached the logs and config i used to build.
> 
> Maybe it could be of some use...

Thanks, I think you are the first to try to build blkback modular.

ERROR: "alloc_empty_pages_and_pagevec" [drivers/xen/blkback/blkbk.ko]
undefined!
ERROR: "blk_put_queue" [drivers/xen/blkback/blkbk.ko] undefined!
ERROR: "free_empty_pages_and_pagevec" [drivers/xen/blkback/blkbk.ko] undefined!
ERROR: "blk_get_queue" [drivers/xen/blkback/blkbk.ko] undefined!
ERROR: "set_phys_to_machine" [drivers/xen/blkback/blkbk.ko] undefined!

The *_empty_pages_and_pagevec and set_phys_to_machine are probably fair
game for an EXPORT_SYMBOL_GPL. The two blk subsystem ones are
suspicious, the fact that they aren't already exported suggests there is
another API we should be using.

Ian.

Re: pv_ops dom0 x86_32 PAE kernel buildlog of missingsymbols...

[Jan Beulich]

>>> Ian Campbell <Ian.Campbell@citrix.com> 11.12.08 13:00 >>>
>ERROR: "alloc_empty_pages_and_pagevec" [drivers/xen/blkback/blkbk.ko]
>undefined!
>ERROR: "blk_put_queue" [drivers/xen/blkback/blkbk.ko] undefined!
>ERROR: "free_empty_pages_and_pagevec" [drivers/xen/blkback/blkbk.ko] undefined!
>ERROR: "blk_get_queue" [drivers/xen/blkback/blkbk.ko] undefined!
>ERROR: "set_phys_to_machine" [drivers/xen/blkback/blkbk.ko] undefined!
>
>The *_empty_pages_and_pagevec and set_phys_to_machine are probably fair
>game for an EXPORT_SYMBOL_GPL. The two blk subsystem ones are
>suspicious, the fact that they aren't already exported suggests there is
>another API we should be using.

I'd be curious if you can find one - in our 2.6.25 merge patch the only way
to address this was to open-code what they do in blkback (short of re-adding
the exports):

--- head-2008-12-08.orig/drivers/xen/blkback/blkback.c	2008-12-08 13:24:54.000000000 +0100
+++ head-2008-12-08/drivers/xen/blkback/blkback.c	2008-12-08 13:24:57.000000000 +0100
@@ -148,7 +148,7 @@ static void unplug_queue(blkif_t *blkif)
 		return;
 	if (blkif->plug->unplug_fn)
 		blkif->plug->unplug_fn(blkif->plug);
-	blk_put_queue(blkif->plug);
+	kobject_put(&blkif->plug->kobj);
 	blkif->plug = NULL;
 }
 
@@ -159,7 +159,8 @@ static void plug_queue(blkif_t *blkif, s
 	if (q == blkif->plug)
 		return;
 	unplug_queue(blkif);
-	blk_get_queue(q);
+	WARN_ON(test_bit(QUEUE_FLAG_DEAD, &q->queue_flags));
+	kobject_get(&q->kobj);
 	blkif->plug = q;
 }
 

Jan

Re: pv_ops dom0 x86_32 PAE kernel buildlog of missingsymbols...

[Marc - A. Dahlhaus [ Administration | Westermann GmbH ]]

Hello Jan,
hello Ian,


i did another build today (c/s 2239) and can confirm that only the
following two warnings are left:

ERROR: "blk_put_queue" [drivers/xen/blkback/blkbk.ko] undefined!
ERROR: "blk_get_queue" [drivers/xen/blkback/blkbk.ko] undefined!

Should i give Jan's patch a try or is another patch to address this
errors planed to be added to the patchset soon?

I ask this because i have some spare time and hardware over the weekend
i could use to do some boot tests. Maybe i can provide some valuable
logs then...


Marc

Am Donnerstag, den 11.12.2008, 13:19 +0000 schrieb Jan Beulich:
> >>> Ian Campbell <Ian.Campbell@citrix.com> 11.12.08 13:00 >>>
> >ERROR: "alloc_empty_pages_and_pagevec" [drivers/xen/blkback/blkbk.ko]
> >undefined!
> >ERROR: "blk_put_queue" [drivers/xen/blkback/blkbk.ko] undefined!
> >ERROR: "free_empty_pages_and_pagevec" [drivers/xen/blkback/blkbk.ko] undefined!
> >ERROR: "blk_get_queue" [drivers/xen/blkback/blkbk.ko] undefined!
> >ERROR: "set_phys_to_machine" [drivers/xen/blkback/blkbk.ko] undefined!
> >
> >The *_empty_pages_and_pagevec and set_phys_to_machine are probably fair
> >game for an EXPORT_SYMBOL_GPL. The two blk subsystem ones are
> >suspicious, the fact that they aren't already exported suggests there is
> >another API we should be using.
> 
> I'd be curious if you can find one - in our 2.6.25 merge patch the only way
> to address this was to open-code what they do in blkback (short of re-adding
> the exports):
> 
> --- head-2008-12-08.orig/drivers/xen/blkback/blkback.c	2008-12-08 13:24:54.000000000 +0100
> +++ head-2008-12-08/drivers/xen/blkback/blkback.c	2008-12-08 13:24:57.000000000 +0100
> @@ -148,7 +148,7 @@ static void unplug_queue(blkif_t *blkif)
>  		return;
>  	if (blkif->plug->unplug_fn)
>  		blkif->plug->unplug_fn(blkif->plug);
> -	blk_put_queue(blkif->plug);
> +	kobject_put(&blkif->plug->kobj);
>  	blkif->plug = NULL;
>  }
>  
> @@ -159,7 +159,8 @@ static void plug_queue(blkif_t *blkif, s
>  	if (q == blkif->plug)
>  		return;
>  	unplug_queue(blkif);
> -	blk_get_queue(q);
> +	WARN_ON(test_bit(QUEUE_FLAG_DEAD, &q->queue_flags));
> +	kobject_get(&q->kobj);
>  	blkif->plug = q;
>  }
>  
> 
> Jan

Re: pv_ops dom0 x86_32 PAE kernel buildlog of missingsymbols...

[Ian Campbell]

On Wed, 2008-12-17 at 16:49 +0100, Marc - A. Dahlhaus [ Administration |
Westermann GmbH ] wrote:
> Hello Jan,
> hello Ian,
> 
> 
> i did another build today (c/s 2239) and can confirm that only the
> following two warnings are left:
> 
> ERROR: "blk_put_queue" [drivers/xen/blkback/blkbk.ko] undefined!
> ERROR: "blk_get_queue" [drivers/xen/blkback/blkbk.ko] undefined!
> 
> Should i give Jan's patch a try or is another patch to address this
> errors planed to be added to the patchset soon?

I say just make these symbols EXPORT_SYMBOL_GPL or build blkback
statically for the time being.

The exports were removed upstream because there was no in-tree module
users but if we are adding one then reverting that could be acceptable.
If that turns out to be unacceptable upstream then we could consider
Jan's patch.

Ian.

> 
> I ask this because i have some spare time and hardware over the weekend
> i could use to do some boot tests. Maybe i can provide some valuable
> logs then...
> 
> 
> Marc
> 
> Am Donnerstag, den 11.12.2008, 13:19 +0000 schrieb Jan Beulich:
> > >>> Ian Campbell <Ian.Campbell@citrix.com> 11.12.08 13:00 >>>
> > >ERROR: "alloc_empty_pages_and_pagevec" [drivers/xen/blkback/blkbk.ko]
> > >undefined!
> > >ERROR: "blk_put_queue" [drivers/xen/blkback/blkbk.ko] undefined!
> > >ERROR: "free_empty_pages_and_pagevec" [drivers/xen/blkback/blkbk.ko] undefined!
> > >ERROR: "blk_get_queue" [drivers/xen/blkback/blkbk.ko] undefined!
> > >ERROR: "set_phys_to_machine" [drivers/xen/blkback/blkbk.ko] undefined!
> > >
> > >The *_empty_pages_and_pagevec and set_phys_to_machine are probably fair
> > >game for an EXPORT_SYMBOL_GPL. The two blk subsystem ones are
> > >suspicious, the fact that they aren't already exported suggests there is
> > >another API we should be using.
> > 
> > I'd be curious if you can find one - in our 2.6.25 merge patch the only way
> > to address this was to open-code what they do in blkback (short of re-adding
> > the exports):
> > 
> > --- head-2008-12-08.orig/drivers/xen/blkback/blkback.c	2008-12-08 13:24:54.000000000 +0100
> > +++ head-2008-12-08/drivers/xen/blkback/blkback.c	2008-12-08 13:24:57.000000000 +0100
> > @@ -148,7 +148,7 @@ static void unplug_queue(blkif_t *blkif)
> >  		return;
> >  	if (blkif->plug->unplug_fn)
> >  		blkif->plug->unplug_fn(blkif->plug);
> > -	blk_put_queue(blkif->plug);
> > +	kobject_put(&blkif->plug->kobj);
> >  	blkif->plug = NULL;
> >  }
> >  
> > @@ -159,7 +159,8 @@ static void plug_queue(blkif_t *blkif, s
> >  	if (q == blkif->plug)
> >  		return;
> >  	unplug_queue(blkif);
> > -	blk_get_queue(q);
> > +	WARN_ON(test_bit(QUEUE_FLAG_DEAD, &q->queue_flags));
> > +	kobject_get(&q->kobj);
> >  	blkif->plug = q;
> >  }
> >  
> > 
> > Jan
> 
> 

Re: pv_ops dom0 x86_32 PAE kernel buildlog of missingsymbols...

[Marc - A. Dahlhaus [ Administration | Westermann GmbH ]]

Am Mittwoch, den 17.12.2008, 15:55 +0000 schrieb Ian Campbell:
> On Wed, 2008-12-17 at 16:49 +0100, Marc - A. Dahlhaus [ Administration |
> Westermann GmbH ] wrote:
> > Hello Jan,
> > hello Ian,
> > 
> > 
> > i did another build today (c/s 2239) and can confirm that only the
> > following two warnings are left:
> > 
> > ERROR: "blk_put_queue" [drivers/xen/blkback/blkbk.ko] undefined!
> > ERROR: "blk_get_queue" [drivers/xen/blkback/blkbk.ko] undefined!
> > 
> > Should i give Jan's patch a try or is another patch to address this
> > errors planed to be added to the patchset soon?
> 
> I say just make these symbols EXPORT_SYMBOL_GPL or build blkback
> statically for the time being.

Used EXPORT_SYMBOL as remainder of exports in file used the same...
Build went silently of course.

> The exports were removed upstream because there was no in-tree module
> users but if we are adding one then reverting that could be acceptable.
> If that turns out to be unacceptable upstream then we could consider
> Jan's patch.
> 
> Ian.

Thanks for the hint.

Marc

Re: pv_ops dom0 x86_32 PAE kernel buildlog of missingsymbols...

[Ian Campbell]

On Wed, 2008-12-17 at 17:32 +0100, Marc - A. Dahlhaus [ Administration |
Westermann GmbH ] wrote:
> Am Mittwoch, den 17.12.2008, 15:55 +0000 schrieb Ian Campbell:
> > I say just make these symbols EXPORT_SYMBOL_GPL or build blkback
> > statically for the time being.
> 
> Used EXPORT_SYMBOL as remainder of exports in file used the same...

Ah yes, the ones which were removed were EXPORT_SYMBOL not _GPL, I was
thinking of something else.

Ian.