Re: [PATCH v2 1/3] ARM: use arch_extension directive instead of arch argument

[Stefan Agner]

On 09.04.2019 14:25, MÃns RullgÃrd wrote:
> Stefan Agner <stefan@xxxxxxxx> writes:
> 
>> The LLVM Target parser currently does not allow to specify the security
>> extension as part of -march (see also LLVM Bug 40186 [0]). When trying
>> to use Clang with LLVM's integrated assembler, this leads to build
>> errors such as this:
>>   clang-8: error: the clang compiler does not support '-Wa,-march=armv7-a+sec'
>>
>> Use ".arch_extension sec" to enable the security extension in a more
>> portable fasion. Also make sure to use ".arch armv7-a" in case a v6/v7
>> multi-platform kernel is being built.
>>
>> Note that this is technically not exactly the same as the old code
>> checked for availabilty of the security extension by calling as-instr.
>> However, there are already other sites which use ".arch_extension sec"
>> unconditionally, hence de-facto we need an assembler capable of
>> ".arch_extension sec" already today (arch/arm/mm/proc-v7.S). The
>> arch extension "sec" is available since binutils 2.21 according to
>> its documentation [1].
>>
>> [0] https://bugs.llvm.org/show_bug.cgi?id=40186
>> [1] https://sourceware.org/binutils/docs-2.21/as/ARM-Options.html
>>
>> Signed-off-by: Stefan Agner <stefan@xxxxxxxx>
>> Acked-by: Mans Rullgard <mans@xxxxxxxxx>
>> Acked-by: Arnd Bergmann <arnd@xxxxxxxx>
>> Acked-by: Krzysztof Kozlowski <krzk@xxxxxxxxxx>
>> ---
>> Changes since v1:
>> - Explicitly specify assembler architecture as armv7-a to avoid
>>   build issues when bulding v6/v7 multi arch kernel.
>>
>>  arch/arm/mach-bcm/Makefile         | 3 ---
>>  arch/arm/mach-bcm/bcm_kona_smc.c   | 2 --
>>  arch/arm/mach-exynos/Makefile      | 4 ----
>>  arch/arm/mach-exynos/exynos-smc.S  | 3 ++-
>>  arch/arm/mach-exynos/sleep.S       | 3 ++-
>>  arch/arm/mach-highbank/Makefile    | 3 ---
>>  arch/arm/mach-highbank/smc.S       | 3 ++-
>>  arch/arm/mach-keystone/Makefile    | 3 ---
>>  arch/arm/mach-keystone/smc.S       | 1 +
>>  arch/arm/mach-omap2/Makefile       | 8 --------
>>  arch/arm/mach-omap2/omap-headsmp.S | 2 ++
>>  arch/arm/mach-omap2/omap-smc.S     | 3 ++-
>>  arch/arm/mach-omap2/sleep33xx.S    | 1 +
>>  arch/arm/mach-omap2/sleep34xx.S    | 2 ++
>>  arch/arm/mach-omap2/sleep43xx.S    | 2 ++
>>  arch/arm/mach-omap2/sleep44xx.S    | 2 ++
>>  arch/arm/mach-tango/Makefile       | 3 ---
>>  arch/arm/mach-tango/smc.S          | 1 +
>>  18 files changed, 19 insertions(+), 30 deletions(-)
> 
> [...]
> 
>> diff --git a/arch/arm/mach-bcm/bcm_kona_smc.c b/arch/arm/mach-bcm/bcm_kona_smc.c
>> index a55a7ecf146a..541e850a736c 100644
>> --- a/arch/arm/mach-bcm/bcm_kona_smc.c
>> +++ b/arch/arm/mach-bcm/bcm_kona_smc.c
>> @@ -125,9 +125,7 @@ static int bcm_kona_do_smc(u32 service_id, u32 buffer_phys)
>>  		__asmeq("%2", "r4")
>>  		__asmeq("%3", "r5")
>>  		__asmeq("%4", "r6")
>> -#ifdef REQUIRES_SEC
>>  		".arch_extension sec\n"
>> -#endif
>>  		"	smc    #0\n"
>>  		: "=r" (ip), "=r" (r0)
>>  		: "r" (r4), "r" (r5), "r" (r6)
> 
> [...]
> 
>> diff --git a/arch/arm/mach-keystone/smc.S b/arch/arm/mach-keystone/smc.S
>> index d15de8179fab..ec03dc499270 100644
>> --- a/arch/arm/mach-keystone/smc.S
>> +++ b/arch/arm/mach-keystone/smc.S
>> @@ -21,6 +21,7 @@
>>   *
>>   * Return: Non zero value on failure
>>   */
>> +	.arch_extension sec
>>  ENTRY(keystone_cpu_smc)
>>  	stmfd   sp!, {r4-r11, lr}
>>  	smc	#0
> 
> [...]
> 
>> diff --git a/arch/arm/mach-tango/smc.S b/arch/arm/mach-tango/smc.S
>> index 361a8dc89804..cf2d21e5226c 100644
>> --- a/arch/arm/mach-tango/smc.S
>> +++ b/arch/arm/mach-tango/smc.S
>> @@ -1,6 +1,7 @@
>>  /* SPDX-License-Identifier: GPL-2.0 */
>>  #include <linux/linkage.h>
>>
>> +	.arch_extension sec
>>  ENTRY(tango_smc)
>>  	push	{lr}
>>  	mov	ip, r1
> 
> Is there some reason these three don't need the .arch directive?

They all do not use a memory barrier instruction (e.g. dmb) which caused
issues on the other files.

--
Stefan