ARM errata 430973 on multi platform kernels

[Tony Lindgren]

* Russell King - ARM Linux <linux at arm.linux.org.uk> [150423 03:26]:
> On Mon, Apr 20, 2015 at 04:40:32PM -0700, Tony Lindgren wrote:
> > * Sebastian Reichel <sre at kernel.org> [150417 11:43]:
> > > On Thu, Apr 16, 2015 at 09:08:58AM -0700, Tony Lindgren wrote:
> > > > * Sebastian Reichel <sre at kernel.org> [150415 09:32]:
> > > > > Hi,
> > > > > 
> > > > > On Thu, Apr 09, 2015 at 02:48:43PM +0100, Russell King - ARM Linux wrote:
> > > > > > On Thu, Apr 09, 2015 at 12:06:58AM +0100, Russell King - ARM Linux wrote:
> > > > > > > On Tue, Apr 07, 2015 at 08:22:08AM -0700, Tony Lindgren wrote:
> > > > > > > > Works for me. The above needs the following fix folded in to build:
> > > > > > > > 
> > > > > > > > --- a/arch/arm/mm/proc-v7.S
> > > > > > > > +++ b/arch/arm/mm/proc-v7.S
> > > > > > > > @@ -532,7 +532,7 @@ __v7_ca9mp_proc_info:
> > > > > > > >  __v7_ca8_proc_info:
> > > > > > > >  	.long	0x410fc080
> > > > > > > >  	.long	0xff0ffff0
> > > > > > > > -	__v7_proc __v7_ca8mp_proc_info, proc_fns = ca8_processor_functions
> > > > > > > > +	__v7_proc __v7_ca8_proc_info, __v7_setup, proc_fns = ca8_processor_functions
> > > > > > > >  	.size	__v7_ca8_proc_info, . - __v7_ca8_proc_info
> > > > > > > >  
> > > > > > > >  #endif	/* CONFIG_ARM_LPAE */
> > > > > > > 
> > > > > > > Thanks, merged into the original patch.
> > > > > > 
> > > > > > Do you want to give me an ack for this, thanks?
> > > > > 
> > > > > I tried to test this together with Tony's follow up patch, but I get
> > > > > this after applying the patch to v4.0:
> > > > > 
> > > > > sre at earth ~/src/linux [430973-fix] % make -j4
> > > > >   CHK     include/config/kernel.release
> > > > >   CHK     include/generated/uapi/linux/version.h
> > > > >   CHK     include/generated/utsrelease.h
> > > > > make[1]: 'include/generated/mach-types.h' is up to date.
> > > > >   CALL    scripts/checksyscalls.sh
> > > > >   CHK     include/generated/compile.h
> > > > >   AS      arch/arm/mm/proc-v7.o
> > > > > arch/arm/mm/proc-v7.S: Assembler messages:
> > > > > arch/arm/mm/proc-v7.S:535: Error: invalid operands (*ABS* and .text sections) for `|'
> > > > > arch/arm/mm/proc-v7.S:535: Error: invalid operands (*ABS* and .text sections) for `|'
> > > > > scripts/Makefile.build:294: recipe for target 'arch/arm/mm/proc-v7.o' failed
> > > > > make[1]: *** [arch/arm/mm/proc-v7.o] Error 1
> > > > > Makefile:947: recipe for target 'arch/arm/mm' failed
> > > > > make: *** [arch/arm/mm] Error 2
> > > > > make: *** Waiting for unfinished jobs....
> > > > 
> > > > Maybe test the version in Linux next:
> > > > 
> > > > a6d746789825 ("ARM: proc-v7: avoid errata 430973 workaround for non-Cortex A8 CPUs")
> > > 
> > > DONE with your your patch added on top:
> > > 
> > > Tested-By: Sebastian Reichel <sre at kernel.org>
> > > 
> > > (on N900)
> > 
> > OK thanks, patch now uploaded to Russell's patch system:
> > 
> > http://www.arm.linux.org.uk/developer/patches/viewpatch.php?id=8345/1
> 
> I have a concern with that patch.
> 
> The reason that it's disabled for multiplatform is because we can't
> guarantee that the auxctrl register will be writable.  The solution
> we came up with for multiplatform was to require firmware to be
> updated to enable this bit.
> 
> Enabling it on a platform where firmware has not been updated, but
> runs in the non-secure world will lead to the kernel hanging in the
> early assembly code.
> 
> I've discussed it with Catalin, and Catalin's position is that we
> should not remove the !multiplatform conditional.  That's something
> which I find that I'm agreeing with Catalin on - any other non-secure
> Cortex A8 user who is setting this bit in firmware will instantly
> break if this patch is applied.
> 
> However, I don't think anyone is willing to say that they have a
> solution to this problem - obviously, you can't build OMAP as a
> non-multiplatform kernel anymore, so in effect you can never have
> the kernel enable this errata.  And you can't detect whether you're
> running in secure mode or not.
> 
> We could do the "only write the bit if it was originally clear" but
> we still have the problem that doing so may cause other people
> regressions.

How about we keep the bit writing part !multiplatform conditional
(or even remove it) but always do the flush for ca-8?

Then we could also do a warning for a misconfigured ca-8 later on.

Regards,

Tony