[PATCH] ARM: keystone: remove hand-coded smc instruction

[dave.martin@linaro.org (Dave Martin)]

On Tue, Jun 25, 2013 at 10:40:57AM -0400, Santosh Shilimkar wrote:
> On Tuesday 25 June 2013 10:32 AM, Dave Martin wrote:
> > On Tue, Jun 25, 2013 at 10:27:11AM -0400, Santosh Shilimkar wrote:
> >> On Tuesday 25 June 2013 10:13 AM, Dave Martin wrote:
> >>> On Fri, Jun 21, 2013 at 06:54:47PM -0400, Santosh Shilimkar wrote:
> >>>> On Friday 21 June 2013 05:41 PM, Arnd Bergmann wrote:
> >>>>> On Friday 21 June 2013, Santosh Shilimkar wrote:
> >>>>>>>
> >>>>>> I was curious how you will fix that for a c file.
> >>>>>> Just to be clear, I was planning to do that in 3.11-rcx/3.12
> >>>>>> time. Let me know if it needs to be done earlier than that.
> >>>>>
> >>>>> It breaks randconfig builds on arm-soc at the moment, so I'd
> >>>>> like the fix as early as possible for 3.11.
> >>>>>
> >>>> Ok, fix is at end of the email. Let me know if it makes
> >>>> to pass both the builds now. I have build and boot tested
> >>>> both ARM and THUMB2 builds on Keystone board.
> >>>>
> >>>> Regards,
> >>>> Santosh
> >>>>
> >>>> From 05d6a5b6cad624fb3791e8c1f8eb7c774f0790d9 Mon Sep 17 00:00:00 2001
> >>>> From: Santosh Shilimkar <santosh.shilimkar@ti.com>
> >>>> Date: Fri, 21 Jun 2013 18:35:32 -0400
> >>>> Subject: [PATCH] ARM: keystone: Move CPU bringup code to dedicated asm file
> >>>>
> >>>> Because of inline asm usage in platsmp.c, smc instruction
> >>>> creates build failure for ARM V6+V7 build where as using instruction
> >>>> encoding for smc breaks the thumb2 build.
> >>>>
> >>>> So move the code snippet to separate asm file and mark
> >>>> it with 'armv7-a$(plus_sec)' to avoid any build issues.
> >>>>
> >>>> Cc: Arnd Bergmann <arnd@arndb.de>
> >>>>
> >>>> Signed-off-by: Santosh Shilimkar <santosh.shilimkar@ti.com>
> >>>> ---
> >>
> >> [..]
> >>
> >>>> diff --git a/arch/arm/mach-keystone/smc.S b/arch/arm/mach-keystone/smc.S
> >>>> new file mode 100644
> >>>> index 0000000..9b9e4f7
> >>>> --- /dev/null
> >>>> +++ b/arch/arm/mach-keystone/smc.S
> >>>> @@ -0,0 +1,29 @@
> >>>> +/*
> >>>> + * Keystone Secure APIs
> >>>> + *
> >>>> + * Copyright (C) 2013 Texas Instruments, Inc.
> >>>> + * 	Santosh Shilimkar <santosh.shilimkar@ti.com>
> >>>> + *
> >>>> + * This program is free software,you can redistribute it and/or modify
> >>>> + * it under the terms of the GNU General Public License version 2 as
> >>>> + * published by the Free Software Foundation.
> >>>> + */
> >>>> +
> >>>> +#include <linux/linkage.h>
> >>>> +
> >>>> +/**
> >>>> + * u32 keystone_cpu_smc(u32 command, u32 cpu, u32 addr)
> >>>> + *
> >>>> + * Low level CPU monitor API
> >>>> + * @command:	Monitor command.
> >>>> + * @cpu:	CPU Number
> >>>> + * @addr:	Kernel jump address for boot CPU
> >>>> + *
> >>>> + * Return: Non zero value on failure
> >>>> + */
> >>>
> >>> Oops, looks like I missed the final mail on this thread.  Ignore my
> >>> previous mail.
> >>>
> >>> I still think it would be a good idea to try to consolidate all these
> >>> trivial SMC wrappers, but this remains debatable.
> >>>
> >>>
> >>>
> >>> Anyway, this looks like it should work, except:
> >>>
> >>>> +ENTRY(keystone_cpu_smc)
> >>>> +	stmfd   sp!, {r4-r12, lr}
> >>>> +	smc	#0
> >>>> +	dsb
> >>>
> >>> What's this DSB for?  (You didn't have it in the inline asm version)
> >>>
> >> Just to drain the write buffer before resuming on non-secure side.
> > 
> > Why do you need to do that?
> > 
> To commit any secure side pending writes. I don't remember exactly the
> issues but I remember facing issues in power management sequencing with
> SMC calls in between. That time a dsb did the trick. In fact I use to
> keep 1 before SMC and 1 after. 

Can we at least have a comment (as recommended in
Documentation/development-process/4.Coding)?

We don't want other people reading this and concluding that SMC always
needs barriers -- in general, this shouldn't be necessary and could
indicate bugs lurking somewhere.

Cheers
---Dave