Re: [PATCH v5 4/4] sparc64: Add support for ADI (Application Data Integrity)

[Khalid Aziz <khalid.aziz@oracle.com>]

On 01/25/2017 03:50 PM, Rob Gardner wrote:
> On 01/25/2017 03:20 PM, Khalid Aziz wrote:
>> On 01/25/2017 03:00 PM, Rob Gardner wrote:
>>> On 01/25/2017 12:57 PM, Khalid Aziz wrote:
>>>>
>>>> @@ -157,6 +158,24 @@ int __get_user_pages_fast(unsigned long start,
>>>> int nr_pages, int write,
>>>>       pgd_t *pgdp;
>>>>       int nr = 0;
>>>>   +#ifdef CONFIG_SPARC64
>>>> +    if (adi_capable()) {
>>>> +        long addr = start;
>>>> +
>>>> +        /* If userspace has passed a versioned address, kernel
>>>> +         * will not find it in the VMAs since it does not store
>>>> +         * the version tags in the list of VMAs. Storing version
>>>> +         * tags in list of VMAs is impractical since they can be
>>>> +         * changed any time from userspace without dropping into
>>>> +         * kernel. Any address search in VMAs will be done with
>>>> +         * non-versioned addresses. Ensure the ADI version bits
>>>> +         * are dropped here by sign extending the last bit before
>>>> +         * ADI bits. IOMMU does not implement version tags.
>>>> +         */
>>>> +        addr = (addr << (long)adi_nbits()) >> (long)adi_nbits();
>>>
>>>
>>> So you are depending on the sign extension to clear the ADI bits... but
>>> this only happens if there is a zero in that "last bit before ADI bits".
>>> If the last bit is a 1, then the ADI bits will be set instead of
>>> cleared.  That seems like an unintended consequence given the comment. I
>>> am aware of the value of adi_nbits() and of the number of valid bits in
>>> a virtual address on the M7 processor, but wouldn't using 'unsigned
>>> long' for everything here guarantee the ADI bits get cleared regardless
>>> of the state of the last non-adi bit?
>>
>> Sign extension is the right thing to do. MMU considers values of 0 and
>> 15 for bits 63-60 to be untagged addresses and expects bit 59 to be
>> sign-extended for untagged virtual addresses. The code I added is
>> explicitly meant to sign-extend, not zero out the top 4 bits.
>
> OK, that wasn't perfectly clear from the comment, which said "version
> bits are dropped".
>
> So sign extending will produce an address that the MMU can use, but will
> it produce an address that will allow a successful search in the page
> tables? ie, was this same sign extending done when first handing out
> that virtual address to the user?
>

Yes to both your questions. When virtual addresses are handed out, the 
last implemented virtual address bit is sign-extended. Sign-extending 
when dropping version bits preserves that original sign-extension. This 
is why MMU considers tag values of 0 as well as 15 to be invalid because 
they both represent sign-extension of the last implemented virtual address.

--
Khalid