Re: [linux-next:master 8264/14724] drivers/firmware/smccc/smccc.c:141:1: error: r7 cannot be used in asm here

[Aneesh Kumar K.V <aneesh.kumar@kernel.org>]

Catalin Marinas <catalin.marinas@arm.com> writes:

> On Wed, Oct 07, 2026 at 09:44:29AM +0100, Sudeep Holla wrote:
>> On Wed, Oct 07, 2026 at 12:28:54PM +0530, Aneesh Kumar K.V wrote:
>> > kernel test robot <lkp@intel.com> writes:
>> > > tree:   https://git.kernel.org/pub/scm/linux/kernel/git/next/linux-next.git master
>> > > head:   eea3fef32a9cf36abcb5975a5a594e4135a6b026
>> > > commit: d8263a131bb05398faa72ac32be54a03e48fa197 [8264/14724] firmware: hwrng: arm_smccc_trng: Register as an SMCCC device
>> > > config: arm-randconfig-1001-20261007 (https://download.01.org/0day-ci/archive/20261007/202610070641.B8hxBJcf-lkp@intel.com/config)
>> > > compiler: arm-linux-gnueabi-gcc (GCC) 8.5.0
>> > > reproduce (this is a W=1 build): (https://download.01.org/0day-ci/archive/20261007/202610070641.B8hxBJcf-lkp@intel.com/reproduce)
>> > >
>> > > If you fix the issue in a separate patch/commit (i.e. not just a new version of
>> > > the same patch/commit), kindly add following tags
>> > > | Reported-by: kernel test robot <lkp@intel.com>
>> > > | Closes: https://lore.kernel.org/oe-kbuild-all/202610070641.B8hxBJcf-lkp@intel.com/
>> > >
>> > > All errors (new ones prefixed by >>):
>> > >
>> > >    drivers/firmware/smccc/smccc.c: In function 'smccc_devices_init':
>> > >>> drivers/firmware/smccc/smccc.c:141:1: error: r7 cannot be used in asm here
>> > >     }
>> > >     ^
>> > >>> drivers/firmware/smccc/smccc.c:141:1: error: r7 cannot be used in asm here
>> > >
>> > >
>> > > vim +141 drivers/firmware/smccc/smccc.c
>> > >
>> > > d8263a131bb0539 Aneesh Kumar K.V (Arm  2026-09-23  122) 
>> > > b83c2d92be719bf Andre Przywara         2021-07-31  123  static int __init smccc_devices_init(void)
>> > > b83c2d92be719bf Andre Przywara         2021-07-31  124  {
>> > > d8263a131bb0539 Aneesh Kumar K.V (Arm  2026-09-23  125) 	struct arm_smccc_device *sdev;
>> > > d8263a131bb0539 Aneesh Kumar K.V (Arm  2026-09-23  126) 	const struct smccc_device_info *smccc_dev;
>> > > d8263a131bb0539 Aneesh Kumar K.V (Arm  2026-09-23  127) 
>> > > d8263a131bb0539 Aneesh Kumar K.V (Arm  2026-09-23  128) 	for (int i = 0; i < ARRAY_SIZE(smccc_devices); i++) {
>> > > d8263a131bb0539 Aneesh Kumar K.V (Arm  2026-09-23  129) 		smccc_dev = &smccc_devices[i];
>> > > d8263a131bb0539 Aneesh Kumar K.V (Arm  2026-09-23  130) 
>> > > d8263a131bb0539 Aneesh Kumar K.V (Arm  2026-09-23  131) 		if (!smccc_probe_smccc_device(smccc_dev))
>> > > d8263a131bb0539 Aneesh Kumar K.V (Arm  2026-09-23  132) 			continue;
>> > > d8263a131bb0539 Aneesh Kumar K.V (Arm  2026-09-23  133) 
>> > > d8263a131bb0539 Aneesh Kumar K.V (Arm  2026-09-23  134) 		sdev = arm_smccc_device_register(smccc_dev->device_name, smccc_dev->func_id);
>> > > d8263a131bb0539 Aneesh Kumar K.V (Arm  2026-09-23  135) 		if (IS_ERR(sdev))
>> > > d8263a131bb0539 Aneesh Kumar K.V (Arm  2026-09-23  136) 			pr_err("%s: could not register device: %pe\n",
>> > > d8263a131bb0539 Aneesh Kumar K.V (Arm  2026-09-23  137) 			       smccc_dev->device_name, sdev);
>> > > b83c2d92be719bf Andre Przywara         2021-07-31  138  	}
>> > > b83c2d92be719bf Andre Przywara         2021-07-31  139  
>> > > b83c2d92be719bf Andre Przywara         2021-07-31  140  	return 0;
>> > > b83c2d92be719bf Andre Przywara         2021-07-31 @141  }
>> > 
>> > This is with:
>> > 
>> > CONFIG_ARM_THUMB=y
>> > CONFIG_ARM_THUMBEE=y
>> > 
>> > where there are restrictions on the use of r7. IMHO, this needs to be
>> > addressed in the implementation of arm_smccc_1_1_invoke(), rather than
>> 
>> We can even drop the extra arguments set 0 as it was original used in
>> TRNG or just use 4 arguments. That shouldn't impact <FEAT>_VERSION ABIs.
>> 
>> I think there is a precedent here with `SMCCC_ARCH_WORKAROUND_*` calls.
>> `__arm_smccc_1_1()` is deliberately variadic: it sets up only the input
>> registers supplied by the caller. Not sure if it is trivial to meddle with
>> the macro to handle this as workaround.
>> 
>> Regards,
>> Sudeep
>> 
>> -->8
>> 
>> diff --git i/drivers/firmware/smccc/smccc.c w/drivers/firmware/smccc/smccc.c
>> index afc2e4d602ba..dfbd04c8eeac 100644
>> --- i/drivers/firmware/smccc/smccc.c
>> +++ w/drivers/firmware/smccc/smccc.c
>> @@ -116,8 +116,7 @@ static bool __init smccc_probe_smccc_device(const struct smccc_device_info *smcc
>>         if (IS_ENABLED(CONFIG_ARM) && ARM_SMCCC_IS_64(smccc_dev->func_id))
>>                 return false;
>> 
>> -       arm_smccc_1_1_invoke(smccc_dev->func_id,
>> -                            0, 0, 0, 0, 0, 0, 0, &res);
>> +       arm_smccc_1_1_invoke(smccc_dev->func_id, &res);
>
> I think that's going in the right direction and works for TRNG. However,
> the RSI_ABI_VERSION func_id requires another argument (I think requested
> interface revision). We could pass (func_id, 0, &res) but I wonder
> whether this probe function is generic enough to always assume all
> arguments are 0.
>

We are only checking whether the result is SMCCC_RET_NOT_SUPPORTED. The
assumption was that the argument register values therefore did not
matter. If a function expects a value within a specific range, it may
return an error for an invalid value, but not SMCCC_RET_NOT_SUPPORTED.

We passed zeroes to avoid exposing stale register contents to components
that are not supposed to read them. In particular, if an SMC is
forwarded to a non-secure host, we do not want that host to see those
register contents. However, this concern does not apply to the two
version-query function IDs used here.

We could therefore use:

	arm_smccc_1_1_invoke(smccc_dev->func_id, &res);

and note that this table must not contain function IDs that are
forwarded to another component. Alternatively, we can retain the
seven-argument form and fix the macro as proposed here?:

https://lore.kernel.org/all/yq5a8q49522f.fsf@kernel.org/

-aneesh