[PATCH v6 05/11] arm64: rsi: Map unprotected MMIO as decrypted

[Gavin Shan]

On 10/11/24 11:19 PM, Catalin Marinas wrote:
> On Tue, Oct 08, 2024 at 10:31:06AM +1000, Gavin Shan wrote:
>> On 10/5/24 12:43 AM, Steven Price wrote:
>>> diff --git a/arch/arm64/kernel/rsi.c b/arch/arm64/kernel/rsi.c
>>> index d7bba4cee627..f1add76f89ce 100644
>>> --- a/arch/arm64/kernel/rsi.c
>>> +++ b/arch/arm64/kernel/rsi.c
>>> @@ -6,6 +6,8 @@
>>>    #include <linux/jump_label.h>
>>>    #include <linux/memblock.h>
>>>    #include <linux/psci.h>
>>> +
>>> +#include <asm/io.h>
>>>    #include <asm/rsi.h>
>>>    struct realm_config config;
>>> @@ -92,6 +94,16 @@ bool arm64_is_protected_mmio(phys_addr_t base, size_t size)
>>>    }
>>>    EXPORT_SYMBOL(arm64_is_protected_mmio);
>>> +static int realm_ioremap_hook(phys_addr_t phys, size_t size, pgprot_t *prot)
>>> +{
>>> +	if (arm64_is_protected_mmio(phys, size))
>>> +		*prot = pgprot_encrypted(*prot);
>>> +	else
>>> +		*prot = pgprot_decrypted(*prot);
>>> +
>>> +	return 0;
>>> +}
>>> +
>>
>> We probably need arm64_is_mmio_private() here, meaning arm64_is_protected_mmio() isn't
>> sufficient to avoid invoking SMCCC call SMC_RSI_IPA_STATE_GET in a regular guest where
>> realm capability isn't present.
> 
> I think we get away with this since the hook won't be registered in a
> normal guest (done from arm64_rsi_init()). So the additional check in
> arm64_is_mmio_private() is unnecessary.
> 

Indeed. I missed the point that the hook won't be registered for a normal
guest. So we're good and safe.

Thanks,
Gavin