Re: [GIT PULL] Driver core update for 4.7-rc1

[Linus Torvalds]

On Sat, May 21, 2016 at 9:31 AM, Linus Torvalds
<torvalds@xxxxxxxxxxxxxxxxxxxx> wrote:
>
> We're not suddenly enabling ISA on x86-64 after having successfully
> gotten rid of that insane crap long long ago.
>
> If you have *specific* dribvers that are actually relevant for some
> reason, make those ones available based on other options. For example,
> we've had the ISA_DMA_API config option to say "we support a subset of
> ISA, even when we don't actually want to ever see actual plug-in ISA
> cards".

I am planning on committing something like the attached.

Note that anything that added "depends on ISA" because of the other
patches will now be disabled on x86-64, even if that driver wasn't
disabled before.

Added Linus Walleij and Guenther Roech to the cc because of the
drivers that I'm aware that did this.

To repeat: I'm not at all interested in enabling random old drivers on
x86-64. Any driver enablement needs to be done on a one-by-one basis,
after having actually gotten TESTING, and after having had people make
sure that we don't get tons of new warnings like we did with the
"let's just enable ISA randomly" approach.

I'm ok with enabling some "ISA_BUS_API" support (like we have
ISA_DMA_API) that allows people to enable drivers one-by-one as they
are converted to a convenience function. The patch series seems to
have had that kind of approach initially.

I'd suggest doing that ISA_BUS_API config option as a *general* thing
(not arch-specific), and make it start out like

      config ISA_BUS_API
          def_bool ISA

which means that everybody gets it, and if ISA was enabled, it will be
enabled automatically. Then, architectures that do *not* enable ISA
itself (like x86-64), could choose to have a config option like

    config ISA_BUS
        bool "support ISA-style drivers on modern systems" if (x86 && EXPERT)
        default y
        select ISA_BUS_API

or something, which means that ISA_BUS_API would then get enabled (but
not ISA itself).

That's similar to what we do today with ISA_DMA_API, except we made
the mistake of making all the different architectures define the
ISA_DMA_API option.

It's the wholesale "enable random crap" that I will not accept. Not
even if there is then "hide the crap again when it causes warnings".
Even a driver that doesn't warn isn't necessarily useful, and I don't
want people to suddenly start seeing a lot of options for random ISA
drivers that simply aren't relevant, and never will be. Not even if
they compile cleanly.

               Linus
Author: Linus Torvalds <torvalds@xxxxxxxxxxxxxxxxxxxx>
Date:   Sat May 21 09:13:41 2016 -0700

    x86 isa: add back X86_32 dependency on CONFIG_ISA
    
    Commit b3c1be1b789c ("base: isa: Remove X86_32 dependency") made ISA
    support available on x86-64 too.  That's not right - while there are
    some LPC-style devices that might be useful still and be based on
    ISA-like IP blocks, that is *not* an excuse to try to enable any random
    legacy drivers.
    
    Such drivers should be individually enabled and made to perhaps depend
    on ISA_DMA_API instead (which we have continued to support on x86-64).
    Or we could add another "ISA_XYZ_API" that we support that doesn't
    enable random old drivers that aren't even 64-bit clean nor do we have
    any test coverage for.
    
    Turning off ISA will now also turn off some drivers that have been
    marked as depending on it as part of this series, and that used to work
    on modern platforms.
    
    See for example commits ad7afc38eab3..cc736607c86d, which may also need
    to be reverted.
    
    Cc: William Breathitt Gray <vilhelm.gray@xxxxxxxxx>
    Cc: Linus Walleij <linus.walleij@xxxxxxxxxx>
    Cc: Guenter Roeck <linux@xxxxxxxxxxxx>
    Cc: Greg Kroah-Hartman <gregkh@xxxxxxxxxxxxxxxxxxx>
    Signed-off-by: Linus Torvalds <torvalds@xxxxxxxxxxxxxxxxxxxx>

diff --git a/arch/x86/Kconfig b/arch/x86/Kconfig
index 48ac29034e1e..0a7b885964ba 100644
--- a/arch/x86/Kconfig
+++ b/arch/x86/Kconfig
@@ -2447,6 +2447,8 @@ config ISA_DMA_API
 	  Enables ISA-style DMA support for devices requiring such controllers.
 	  If unsure, say Y.
 
+if X86_32
+
 config ISA
 	bool "ISA support"
 	---help---
@@ -2456,8 +2458,6 @@ config ISA
 	  (MCA) or VESA.  ISA is an older system, now being displaced by PCI;
 	  newer boards don't support it.  If you have ISA, say Y, otherwise N.
 
-if X86_32
-
 config EISA
 	bool "EISA support"
 	depends on ISA