Re: build issue #503 for v2.6.26-rc2-433-gf26a398 : undefinedreference to `request_firmware'

[James Bottomley]

On Fri, 2008-05-23 at 15:39 -0700, Randy Dunlap wrote:
> On Fri, 23 May 2008 17:00:30 -0500 James Bottomley wrote:
> 
> > On Fri, 2008-05-23 at 14:48 -0700, Randy Dunlap wrote:
> > > On Fri, 23 May 2008 15:33:00 -0500 James Bottomley wrote:
> > > 
> > > > On Fri, 2008-05-23 at 11:04 -0700, Randy Dunlap wrote:
> > > > > On Fri, 23 May 2008 12:59:11 -0500 James Bottomley wrote:
> > > > > 
> > > > > > On Fri, 2008-05-23 at 10:54 -0700, Randy Dunlap wrote:
> > > > > > > On Thu, 22 May 2008 11:52:24 +0100 James Bottomley wrote:
> > > > > > > 
> > > > > > > > On Sat, 2008-05-17 at 17:17 +0300, Adrian Bunk wrote:
> > > > > > > > > On Sat, May 17, 2008 at 11:04:31AM +0100, James Bottomley wrote:
> > > > > > > > > > On Fri, 2008-05-16 at 19:06 +0200, Toralf FÃrster wrote:
> > > > > > > > > > > Hello,
> > > > > > > > > > > 
> > > > > > > > > > > the build (.config attached) failed, make ends with :
> > > > > > > > > > > ...
> > > > > > > > > > >   UPD     include/linux/compile.h
> > > > > > > > > > >   CC      init/version.o
> > > > > > > > > > >   LD      init/built-in.o
> > > > > > > > > > >   LD      vmlinux
> > > > > > > > > > > drivers/built-in.o: In function `sas_request_addr':
> > > > > > > > > > > (.text+0x33bab): undefined reference to `request_firmware'
> > > > > > > > > > > drivers/built-in.o: In function `sas_request_addr':
> > > > > > > > > > > (.text+0x33c3f): undefined reference to `release_firmware'
> > > > > > > > > > > make: *** [vmlinux] Error 1
> > > > > > > > > > 
> > > > > > > > > > There's a slight fault in the stub logic.  It fails for FW_LOADER=m and
> > > > > > > > > > the user =y.
> > > > > > > > > > 
> > > > > > > > > > This should fix it.
> > > > > > > > > 
> > > > > > > > > An interesting question is whether we actually want to have the stub at 
> > > > > > > > > all - the compile errors catched cases where someone forgot to select 
> > > > > > > > > FW_LOADER at compile time instead of turning them into possible runtime 
> > > > > > > > > problems as will happen after your patch.
> > > > > > > > 
> > > > > > > > Well, the original intent of the stubs was to have the loader nop if it
> > > > > > > > wasn't available.  There was simply one case forgotten (the fact that a
> > > > > > > > modular loader isn't usable from built in components).
> > > > > > > > 
> > > > > > > > > Are there any serious use cases where SCSI_SAS_LIBSAS is used in very 
> > > > > > > > > space limited environments and without any other drivers that select 
> > > > > > > > > FW_LOADER in the kernel?
> > > > > > > > 
> > > > > > > > Probably in the future, certainly on embedded OSs used for storage
> > > > > > > > devices.  Given the convergence of SAS/SATA chips, it's possible for
> > > > > > > > other handhelds, but I'm not convinced of that.
> > > > > > > 
> > > > > > > Has there been any resolution to this build issue?
> > > > > > > 
> > > > > > > Do we know what we want?  then maybe a patch can be done...
> > > > > > 
> > > > > > As long as the patch works, that's what I propose.
> > > > > 
> > > > > Sorry, I am not understanding what you are saying.
> > > > > Please explain.
> > > > 
> > > > As long as the patch I previously posted on this thread
> > > > 
> > > > http://marc.info/?l=linux-scsi&m=121101871800303
> > > > 
> > > > Actually works.  This is the patch I propose to use to fix the issue.
> > > 
> > > 
> > > Thanks for the pointer to the patch.  That does actually build cleanly
> > > (but I would say that the patch is incorrect).
> > > 
> > > The case is (similar to that scsi_dh problem):
> > > CONFIG_SCSI_SAS_LIBSAS=y and CONFIG_FW_LOADER=m, so libsas's
> > > sas_request_addr() always returns -EINVAL as the value of
> > > request_firmware().  Is that what we want/expect for these CONFIGs?
> > 
> > Actually, I disagree.  There are two choices here
> 
> Disagree with which part(s)?

Disagree that the patch was incorrect.

> >      1. No stubs and simply select FW_LOADER for everything
> >      2. Stubs that actually work in all situations.
> > 
> > My patch does 2.  I can see an argument for 1. but since we have the
> > stubs, making them work in all cases seems like the smaller change to
> > functionality.
> 
> Oh, I'm for the stubs approach.
> 
> So with SCSI_SAS_LIBSAS=y and FW_LOADER=m, you would want/expect
> request_firmware() to return -EINVAL all of the time?

If you want stubs, yes

> Hm, I suppose that it has to either do that or select FW_LOADER. ?

Exactly.  I could imagine using Kconfig select trickery to force the
choice down to Y/N for firmware loader, but I bet it would be incredibly
fragile and difficult to understand.

James


--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/