On 23 April 2018 at 13:10, Auger Eric <eric.au...@redhat.com> wrote:
> Hi Peter,
>
> On 04/16/2018 02:59 PM, Peter Maydell wrote:
>> On 12 April 2018 at 08:37, Eric Auger <eric.au...@redhat.com> wrote:
>>> +/**
>>> + * TODO: handle the case where the level resolves less than
>>> + * granule_sz -3 IA bits.
>>> + */
>>> +static inline
>>> +uint64_t iova_level_offset(uint64_t iova, int level, int granule_sz)
>>> +{
>>> + return (iova >> level_shift(level, granule_sz)) &
>>> + MAKE_64BIT_MASK(0, granule_sz - 3);
>>> +}
>>> +
>>> +#endif
>>
>> When does the TODO case happen, and what goes wrong?
> Sorry for the delay [back to the office].
>
> This refers to
> D4.2 of ARM ARM , Effect of granule size on translation table addressing
> and indexing/ Reduced IA width.
>
> "
> Depending on the configuration and implementation choices, the required
> input address width for the initial level of lookup might be smaller
> than the number of address bits that can be resolved at that level"
>
> For instance with 4k granule and IA limited to 35 bit, we use a 9-bit
> mask here and initial level 1 offset is based on IA[30, 38] although IA
> is limited to 35.
Hmm. We do get this right for the table walk code in
target/arm/helper.c, so it would be kind of nice to get it
right here too.
thanks
-- PMM
