Problem with commit: x86, iosf: Make IOSF driver modular and usable by more drivers

Problem with commit: x86, iosf: Make IOSF driver modular and usable by more drivers

[Fejes József]

Hi,

I think there's a problem with commit
6b8f0c8780c71d78624f736d7849645b64cc88b7: config IOSF_MBI is
automatically a module and I cannot change that.

I've been using 3.15.* stable kernels. I have module support enabled,
but I build everything into the kernel, so I don't actually have any
modules built. I just upgraded to 3.16.1, and found out that I now
have this one module. I cannot find it in the menu, so I edited the
.config file by hand, but it changes back from =y to =m. Could you
please look into fixing it, and push it to 3.16.* stable branch?

Best regards,
Jozsef

Re: Problem with commit: x86, iosf: Make IOSF driver modular and usable by more drivers

[Randy Dunlap]

On 08/23/14 02:31, Fejes József wrote:
> Hi,
> 
> I think there's a problem with commit
> 6b8f0c8780c71d78624f736d7849645b64cc88b7: config IOSF_MBI is
> automatically a module and I cannot change that.
> 
> I've been using 3.15.* stable kernels. I have module support enabled,
> but I build everything into the kernel, so I don't actually have any
> modules built. I just upgraded to 3.16.1, and found out that I now
> have this one module. I cannot find it in the menu, so I edited the
> .config file by hand, but it changes back from =y to =m. Could you
> please look into fixing it, and push it to 3.16.* stable branch?

[adding David E. Box to email]

This is a mainline issue, not just a -stable issue.
Once fixed in mainline (if ever), then that fix can be added to -stable.

Fejes, you could just disable module support and then iosf_mbi would be built
into the kernel.  But as a loadable module, it won't waste memory if it's
not needed.

David, any other suggestions?
Why can't the users of IOSF_MBI just select it?  That's what many other
drivers do when they need to be sure that some functionality is present.

I'm surprised that someone else (e.g. Linus) has not complained about the
'default m' for this driver.


-- 
~Randy

Re: Problem with commit: x86, iosf: Make IOSF driver modular and usable by more drivers

[David E. Box]

On Sat, Aug 23, 2014 at 10:40:14AM -0700, Randy Dunlap wrote:
> On 08/23/14 02:31, Fejes József wrote:
> > Hi,
> > 
> > I think there's a problem with commit
> > 6b8f0c8780c71d78624f736d7849645b64cc88b7: config IOSF_MBI is
> > automatically a module and I cannot change that.
> > 
> > I've been using 3.15.* stable kernels. I have module support enabled,
> > but I build everything into the kernel, so I don't actually have any
> > modules built. I just upgraded to 3.16.1, and found out that I now
> > have this one module. I cannot find it in the menu, so I edited the
> > .config file by hand, but it changes back from =y to =m. Could you
> > please look into fixing it, and push it to 3.16.* stable branch?
> 
> [adding David E. Box to email]
> 
> This is a mainline issue, not just a -stable issue.
> Once fixed in mainline (if ever), then that fix can be added to -stable.
> 
> Fejes, you could just disable module support and then iosf_mbi would be built
> into the kernel.  But as a loadable module, it won't waste memory if it's
> not needed.
> 
> David, any other suggestions?
> Why can't the users of IOSF_MBI just select it? That's what many other
> drivers do when they need to be sure that some functionality is present.

They could, but it's only required on SoC's. Some registers, while available
through an MSR on x86 core systems, are only available through the sideband on
x86 SoC's. So these drivers would waste space on core platforms. There is no
Kconfig option that builds exclusively for x86 SoC's.

> I'm surprised that someone else (e.g. Linus) has not complained about the
> 'default m' for this driver.

I'll just move to prompt for selection. I was advised against this early on
since the driver doesn't have a hook to userspace. Unfortunantely I didn't
question this and locked myself into pursuing the 'default m' option as a way to
make sure the IOSF_MBI was available for the drivers that will use it. That was
obviosuly a mistake.

Dave

Re: Problem with commit: x86, iosf: Make IOSF driver modular and usable by more drivers

[One Thousand Gnomes]

> David, any other suggestions?
> Why can't the users of IOSF_MBI just select it?  That's what many other
> drivers do when they need to be sure that some functionality is present.

I think the obvious problem is that most users of IOSF_MBI are
non-obvious and there will be many of them over time.
> 
> I'm surprised that someone else (e.g. Linus) has not complained about the
> 'default m' for this driver.

That's certainly wrong, but drivers that use it should probably
auto-select it or we will get a billion bug reports asking why graphics
doesn't work etc 

Alan