On Wed, 20 Feb 2013 15:37:14 -0800, Christoffer Dall
<cdall@xxxxxxxxxxxxxxx> wrote:
> On Wed, Feb 13, 2013 at 03:46:59PM +0000, Marc Zyngier wrote:
>> As we carry on deconstructing stage2_set_pte(), introduce a method
>> to write a pte at a given location.
>>
>> Make the prototype similar to the kernel set_pte_at(), so the
properties
>> are not completely surprising. Convert stage2_set_pte() to make use
>> of stage_set_pte_at().
>>
>> Signed-off-by: Marc Zyngier <marc.zyngier@xxxxxxx>
>> ---
>> arch/arm/kvm/mmu.c | 24 ++++++++++++++++--------
>> 1 file changed, 16 insertions(+), 8 deletions(-)
>>
>> diff --git a/arch/arm/kvm/mmu.c b/arch/arm/kvm/mmu.c
>> index cb2b42e..ab35906 100644
>> --- a/arch/arm/kvm/mmu.c
>> +++ b/arch/arm/kvm/mmu.c
>> @@ -444,10 +444,24 @@ static pte_t *stage2_get_pte(struct kvm *kvm,
>> struct kvm_mmu_memory_cache *cache
>> return pte_offset_kernel(pmd, addr);
>> }
>>
>> +static void stage2_set_pte_at(struct kvm *kvm, phys_addr_t addr,
>> + pte_t *pte, const pte_t pteval)
>> +{
>> + pte_t old_pte;
>> +
>> + /* Create 2nd stage page table mapping - Level 3 */
>> + old_pte = *pte;
>> + kvm_set_pte(pte, pteval);
>> + if (pte_present(old_pte))
>> + kvm_tlb_flush_vmid_ipa(kvm, addr);
>> + else
>> + get_page(virt_to_page(pte));
>> +}
>
> what's the reason between this level of indirection? We already have
> quite a few xxx_set_pte_xxx functions, ...
The idea is to be able to perform a read-modify write on an PTE. At the
moment, we can only write a full PTE.
M.
--
Fast, cheap, reliable. Pick two.
_______________________________________________
kvmarm mailing list
kvmarm@xxxxxxxxxxxxxxxxxxxxx
https://lists.cs.columbia.edu/cucslists/listinfo/kvmarm
