On 11/12/22 04:06, Catalin Marinas wrote:
> On Thu, Nov 10, 2022 at 08:45:07AM +0530, Anshuman Khandual wrote:
>> On 11/10/22 00:48, Catalin Marinas wrote:
>>> On Thu, Oct 27, 2022 at 08:09:15AM +0530, Anshuman Khandual wrote:
>>>> +#define __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION
>>>> +static inline pte_t ptep_modify_prot_start(struct vm_area_struct *vma,
>>>> + unsigned long addr,
>>>> + pte_t *ptep)
>>>> +{
>>>> + pte_t pte = ptep_get_and_clear(vma->vm_mm, addr, ptep);
>>>>
>>>> + if (IS_ENABLED(CONFIG_ARM64_WORKAROUND_2645198)) {
>>>> + /*
>>>> + * Break-before-make (BBM) is required for all user space mappings
>>>> + * when the permission changes from executable to non-executable
>>>> + * in cases where cpu is affected with errata #2645198.
>>>> + */
>>>> + if (pte_user_exec(pte) && cpus_have_const_cap(ARM64_WORKAROUND_2645198))
>>>> + __flush_tlb_range(vma, addr, addr + PAGE_SIZE, PAGE_SIZE, false, 3);
>>>
>>> Why not flush_tlb_page() here?
>>>
>>> But more importantly, can we not use ptep_clear_flush() instead (and
>>
>> Something like ...
>>
>> ptep_modify_prot_start -
>>
>> if (IS_ENABLED(CONFIG_ARM64_WORKAROUND_2645198)) {
>> if (pte_user_exec(READ_ONCE(*ptep)) && cpus_have_const_cap(ARM64_WORKAROUND_2645198))
>> return ptep_clear_flush(vma, addr, ptep);
>> } else {
>> return ptep_get_and_clear(vma->vm_mm, addr, ptep);
>> }
>
> Yes, this should work but avoid the 'else' when you have a return, so
> something like:
>
> if (IS_ENABLED(CONFIG_ARM64_WORKAROUND_2645198) &&
> cpus_have_const_cap(ARM64_WORKAROUND_2645198) &&
> pte_user_exec(READ_ONCE(*ptep)))
> return ptep_clear_flush(vma, addr, ptep);
>
> return ptep_get_and_clear(vma->vm_mm, addr, ptep);
Right, realized that later.
>
>
>>> huge_ptep_clear_flush())? They return the pte and do the TLBI.
>>
>> huge_ptep_modify_prot_start -
>>
>> if (IS_ENABLED(CONFIG_ARM64_WORKAROUND_2645198)) {
>> if (pte_user_exec(READ_ONCE(*ptep)) && cpus_have_const_cap(ARM64_WORKAROUND_2645198))
>> return huge_ptep_clear_flush(vma, addr, ptep);
>> } else {
>> return huge_ptep_get_and_clear(vma->vm_mm, addr, ptep);
>> }
>>
>> pte_user_exec(READ_ONCE(*ptep) should identify an user exec mapping even though
>> ptep represents a cont PTE/PMD huge page ? OR should huge_ptep_get() helper be
>> used instead ?
>
> This should work as a shortcut. The contiguous ptes should all be the
> same, so it's sufficient to check one of them.
Sure, will read the first one.
