Re: [PATCH v2] arm64: add ptrace regsets for ptrauth key management

[Kristina Martsenko <kristina.martsenko@arm.com>]

On 19/01/2019 23:21, Will Deacon wrote:
> On Wed, Jan 16, 2019 at 03:13:03PM +0000, Dave Martin wrote:
>> On Tue, Jan 15, 2019 at 07:32:30PM +0000, Kristina Martsenko wrote:
>>> We don't currently have any other separate nested structs in the arm64
>>> ptrace userspace interface. Other architectures also seem to only have a
>>> few rare instances of this.
>>>
>>> It seems odd to complicate the userspace interface because of a kernel
>>> implementation detail, but if you think it's better I could change the
>>> structs to:
>>>
>>> struct user_pac_key {
>>> 	__u64		lo;
>>> 	__u64		hi;
>>> };
>>>
>>> struct user_pac_address_keys {
>>> 	struct user_pac_key	apiakey;
>>> 	struct user_pac_key	apibkey;
>>> 	struct user_pac_key	apdakey;
>>> 	struct user_pac_key	apdbkey;
>>> };
>>>
>>> struct user_pac_generic_keys {
>>> 	struct user_pac_key	apgakey;
>>> };
>>
>> I don't have a strong opinion really.
>>
>> There's another option that occurred to me just now, and that's simply
>> to represent each key as a single __uint128_t in the user regset view
>> (with the _hi half strictly in the high 64 bits).
>>
>> This avoids questions about how to represent the halves of the key,
>> and prevents userspace from accessing half-keys.  How the half-keys
>> map to system registers is an architectural quirk that userspace perhaps
>> need not care about.
>>
>> If you like that approach, you could keep the kernel code as-is for now
>> and just do the conversion explicitly in the ptrace accessors (as in
>> your original patch).
> 
> I think that would be worth persuing. We already have __uint128_t types in
> our uapi/ headers, so we're not introducing anything new by using them here.

Ok, I'll send v3 with __uint128_t keys.

Thanks,
Kristina

_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel