[RFC 3/9] arm64: handle 52-bit addresses in TTBR

[Robin Murphy]

On 07/12/17 12:29, Kristina Martsenko wrote:
> On 21/11/17 14:39, Robin Murphy wrote:
>> Hi Kristina,
>>
>> On 21/11/17 11:57, Kristina Martsenko wrote:
>>> The top 4 bits of a 52-bit physical address are positioned at bits 2..5
>>> in the TTBR registers. Introduce a couple of macros to move the bits
>>> there, and change all TTBR writers to use them.
>>>
>>> Leave TTBR0 PAN code unchanged, to avoid complicating it. A system with
>>> 52-bit PA will have PAN anyway (because it's ARMv8.1 or later), and a
>>> system without 52-bit PA can only use up to 48-bit PAs. Add a kconfig
>>> dependency to ensure PAN is configured.
>>>
>>> In addition, when using 52-bit PA there is a special alignment
>>> requirement on the top-level table. We don't currently have any VA_BITS
>>> configuration that would violate the requirement, but one could be added
>>> in the future, so add a compile-time BUG_ON to check for it.
>>>
>>> Signed-off-by: Kristina Martsenko <kristina.martsenko at arm.com>
>>> ---
> 
> [...]
> 
>>> diff --git a/arch/arm64/include/asm/assembler.h
>>> b/arch/arm64/include/asm/assembler.h
>>> index 04cf94766b78..ba3c796b9fe1 100644
>>> --- a/arch/arm64/include/asm/assembler.h
>>> +++ b/arch/arm64/include/asm/assembler.h
>>> @@ -512,4 +512,21 @@ alternative_else_nop_endif
>>>    #endif
>>>        .endm
>>>    +/*
>>> + * Arrange a physical address in a TTBR register, taking care of 52-bit
>>> + * addresses.
>>> + *
>>> + *     phys:    physical address, preserved
>>> + *     ttbr:    returns the TTBR value
>>> + */
>>> +    .macro    phys_to_ttbr, phys, ttbr
>>> +#ifdef CONFIG_ARM64_PA_BITS_52
>>> +    and    \ttbr, \phys, #(1 << 48) - 1
>>> +    orr    \ttbr, \ttbr, \phys, lsr #48 - 2
>>> +    bic    \ttbr, \ttbr, #(1 << 2) - 1
>>
>> Is there any reason for masking off each end of the result separately
>> like this, or could we just do it more straightforwardly?
> 
> I don't recall any reason, maybe just to keep it simple, to avoid having
> a separate mask macro.
> 
>> #define TTBR_BADDR_MASK ((1 << 46) - 1) << 2
>>
>>      orr    \ttbr, \phys, \phys, lsr #46
>>      and    \ttbr, \ttbr, #TTBR_BADDR_MASK
>>
>> (and equivalently for phys_to_pte in patch 4)
> 
> Ok, I'll rewrite it like this. (Note that mask is 52-bit-specific though.)

I don't see that it need be 52-bit specific - true the BADDR field 
itself is strictly bits 47:1, but AFAICS bit 1 is always going to be 
RES0: either explicitly in the 52-bit case, or from the (x-1):1 
definition otherwise, since the requirement that a table must be aligned 
to its size infers an absolute lower bound of x >= 3 (it may be larger 
still due to other reasons, but I'm finding this area of the ARM ARM 
obnoxiously difficult to read). Thus defining the mask as covering 47:2 
should still be reasonable in all cases.

>> Even better if there's a convenient place to define the mask such that
>> it can be shared with KVM's existing VTTBR stuff too.
> 
> Do you mean VTTBR_BADDR_MASK? I don't think this would be useful there,
> VTTBR_BADDR_MASK checks the alignment of the address that goes into
> VTTBR (not the VTTBR value itself), and takes into account specifically
> the 40-bit IPA and concatenated page tables.

Ah, I see - from skimming the code I managed to assume VTTBR_BADDR_MASK 
was a mask for the actual VTTBR.BADDR register field; I hadn't delved 
into all the VTTBR_X gubbins (yuck). Fair enough, scratch that idea. At 
least TTBR_ASID_MASK[1] gets to have a friend :)

Robin.

[1] 
https://www.mail-archive.com/linux-kernel@vger.kernel.org/msg1555176.html

> 
> Kristina
> 
>>> +#else
>>> +    mov    \ttbr, \phys
>>> +#endif
>>> +    .endm
>>> +
>>>    #endif    /* __ASM_ASSEMBLER_H */