[PATCH v14 07/44] arm64: RMI: Configure the RMM with the host's page size

[Steven Price]

On 21/05/2026 14:30, Marc Zyngier wrote:
> On Wed, 13 May 2026 14:17:15 +0100,
> Steven Price <steven.price at arm.com> wrote:
>>
>> RMM v2.0 brings the ability to set the RMM's granule size. Check the
>> feature registers and configure the RMM so that it matches the host's
>> page size. This means that operations can be done with a granulatity
>> equal to PAGE_SIZE.
>>
>> Signed-off-by: Steven Price <steven.price at arm.com>
>> ---
>> Changes since v13:
>>  * Moved out of KVM.
>> ---
>>  arch/arm64/kernel/rmi.c | 42 +++++++++++++++++++++++++++++++++++++++++
>>  1 file changed, 42 insertions(+)
>>
>> diff --git a/arch/arm64/kernel/rmi.c b/arch/arm64/kernel/rmi.c
>> index 99c1ccc35c11..a14ead5dedda 100644
>> --- a/arch/arm64/kernel/rmi.c
>> +++ b/arch/arm64/kernel/rmi.c
>> @@ -49,6 +49,45 @@ static int rmi_check_version(void)
>>  	return 0;
>>  }
>>  
>> +static int rmi_configure(void)
>> +{
>> +	struct rmm_config *config __free(free_page) = NULL;
>> +	unsigned long ret;
>> +
>> +	config = (struct rmm_config *)get_zeroed_page(GFP_KERNEL);
>> +	if (!config)
>> +		return -ENOMEM;
> 
> This is the sort of buggy construct that is highlighted in
> include/linux/cleanup.h: initialising the object for cleanup with
> NULL, and only later assigning the expected value.
> 
> It may not matter here, but it will catch you (or more probably me) in
> the future.

Good spot. I have to admit I'm still getting the hang of these cleanup
handlers.

>> +
>> +	switch (PAGE_SIZE) {
>> +	case SZ_4K:
>> +		config->rmi_granule_size = RMI_GRANULE_SIZE_4KB;
>> +		break;
>> +	case SZ_16K:
>> +		config->rmi_granule_size = RMI_GRANULE_SIZE_16KB;
>> +		break;
>> +	case SZ_64K:
>> +		config->rmi_granule_size = RMI_GRANULE_SIZE_64KB;
>> +		break;
>> +	default:
>> +		pr_err("Unsupported PAGE_SIZE for RMM\n");
> 
> Do you really anticipate PAGE_SIZE being any other value? This is 100%
> dead code. If you want to be extra cautious, have a BUILD_BUg_ON().

No, but falling through is clearly wrong (and likely to trigger AI
review comments if nothing else) - BUILD_BUG() sounds like a good solution.

>> +		return -EINVAL;
>> +	}
>> +
>> +	ret = rmi_rmm_config_set(virt_to_phys(config));
>> +	if (ret) {
>> +		pr_err("RMM config set failed\n");
>> +		return -EINVAL;
>> +	}
> 
> What is the live cycle of the page when the call succeeds? Is it
> switched back to the NS PAS and allowed to be freed?

Yes, as Suzuki answered - it never leaves the NS PAS. The RMM just reads it.

Thanks,
Steve

>> +
>> +	ret = rmi_rmm_activate();
>> +	if (ret) {
>> +		pr_err("RMM activate failed\n");
>> +		return -ENXIO;
>> +	}
>> +
>> +	return 0;
>> +}
>> +
>>  static int __init arm64_init_rmi(void)
>>  {
>>  	/* Continue without realm support if we can't agree on a version */
>> @@ -60,6 +99,9 @@ static int __init arm64_init_rmi(void)
>>  	if (WARN_ON(rmi_features(1, &rmm_feat_reg1)))
>>  		return 0;
>>  
>> +	if (rmi_configure())
>> +		return 0;
>> +
>>  	return 0;
>>  }
>>  subsys_initcall(arm64_init_rmi);
> 
> Thanks,
> 
> 	M.
>