Re: [PATCH v3 13/13] mm/debug_vm_pgtable: populate a pte entry before fetching it

["Aneesh Kumar K.V" <aneesh.kumar@linux.ibm.com>]

On 9/2/20 9:19 AM, Anshuman Khandual wrote:
> 
> 
> On 09/01/2020 03:28 PM, Aneesh Kumar K.V wrote:
>> On 9/1/20 1:08 PM, Anshuman Khandual wrote:
>>>
>>>
>>> On 09/01/2020 12:07 PM, Aneesh Kumar K.V wrote:
>>>> On 9/1/20 8:55 AM, Anshuman Khandual wrote:
>>>>>
>>>>>
>>>>> On 08/27/2020 01:34 PM, Aneesh Kumar K.V wrote:
>>>>>> pte_clear_tests operate on an existing pte entry. Make sure that is not a none
>>>>>> pte entry.
>>>>>>
>>>>>> Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.ibm.com>
>>>>>> ---
>>>>>>     mm/debug_vm_pgtable.c | 6 ++++--
>>>>>>     1 file changed, 4 insertions(+), 2 deletions(-)
>>>>>>
>>>>>> diff --git a/mm/debug_vm_pgtable.c b/mm/debug_vm_pgtable.c
>>>>>> index 21329c7d672f..8527ebb75f2c 100644
>>>>>> --- a/mm/debug_vm_pgtable.c
>>>>>> +++ b/mm/debug_vm_pgtable.c
>>>>>> @@ -546,7 +546,7 @@ static void __init pgd_populate_tests(struct mm_struct *mm, pgd_t *pgdp,
>>>>>>     static void __init pte_clear_tests(struct mm_struct *mm, pte_t *ptep,
>>>>>>                        unsigned long vaddr)
>>>>>>     {
>>>>>> -    pte_t pte = ptep_get(ptep);
>>>>>> +    pte_t pte =  ptep_get_and_clear(mm, vaddr, ptep);
>>>>>
>>>>> Seems like ptep_get_and_clear() here just clears the entry in preparation
>>>>> for a following set_pte_at() which otherwise would have been a problem on
>>>>> ppc64 as you had pointed out earlier i.e set_pte_at() should not update an
>>>>> existing valid entry. So the commit message here is bit misleading.
>>>>>
>>>>
>>>> and also fetch the pte value which is used further.
>>>>
>>>>
>>>>>>           pr_debug("Validating PTE clear\n");
>>>>>>         pte = __pte(pte_val(pte) | RANDOM_ORVALUE);
>>>>>> @@ -944,7 +944,7 @@ static int __init debug_vm_pgtable(void)
>>>>>>         p4d_t *p4dp, *saved_p4dp;
>>>>>>         pud_t *pudp, *saved_pudp;
>>>>>>         pmd_t *pmdp, *saved_pmdp, pmd;
>>>>>> -    pte_t *ptep;
>>>>>> +    pte_t *ptep, pte;
>>>>>>         pgtable_t saved_ptep;
>>>>>>         pgprot_t prot, protnone;
>>>>>>         phys_addr_t paddr;
>>>>>> @@ -1049,6 +1049,8 @@ static int __init debug_vm_pgtable(void)
>>>>>>          */
>>>>>>           ptep = pte_alloc_map_lock(mm, pmdp, vaddr, &ptl);
>>>>>> +    pte = pfn_pte(pte_aligned, prot);
>>>>>> +    set_pte_at(mm, vaddr, ptep, pte);
>>>>>
>>>>> Not here, creating and populating an entry must be done in respective
>>>>> test functions itself. Besides, this seems bit redundant as well. The
>>>>> test pte_clear_tests() with the above change added, already
>>>>>
>>>>> - Clears the PTEP entry with ptep_get_and_clear()
>>>>
>>>> and fetch the old value set previously.
>>>
>>> In that case, please move above two lines i.e
>>>
>>> pte = pfn_pte(pte_aligned, prot);
>>> set_pte_at(mm, vaddr, ptep, pte);
>>>
>>> from debug_vm_pgtable() to pte_clear_tests() and update it's arguments
>>> as required.
>>>
>>
>> Frankly, I don't understand what these tests are testing. It all looks like some random clear and set.
> 
> The idea here is to have some value with some randomness preferably, in
> a given PTEP before attempting to clear the entry, in order to make sure
> that pte_clear() is indeed clearing something of non-zero value.
> 
>>
>> static void __init pte_clear_tests(struct mm_struct *mm, pte_t *ptep,
>>                     unsigned long vaddr, unsigned long pfn,
>>                     pgprot_t prot)
>> {
>>
>>      pte_t pte = pfn_pte(pfn, prot);
>>      set_pte_at(mm, vaddr, ptep, pte);
>>
>>      pte =  ptep_get_and_clear(mm, vaddr, ptep);
> 
> Looking at this again, this preceding pfn_pte() followed by set_pte_at()
> is not really required. Its reasonable to start with what ever was there
> in the PTEP as a seed value which anyway gets added with RANDOM_ORVALUE.
> s/ptep_get/ptep_get_and_clear is sufficient to take care of the powerpc
> set_pte_at() constraint.
> 

But the way test is written we had none pte before. That is why I added 
that set_pte_at to put something there. With none pte the below sequence 
fails.

	pte = __pte(pte_val(pte) | RANDOM_ORVALUE);
	set_pte_at(mm, vaddr, ptep, pte);


because nobody is marking a _PAGE_PTE there.

	pte_t pte = pfn_pte(pfn, prot);

	pr_debug("Validating PTE clear\n");
	pte = __pte(pte_val(pte) | RANDOM_ORVALUE);
	set_pte_at(mm, vaddr, ptep, pte);
	barrier();
	pte_clear(mm, vaddr, ptep);
	pte = ptep_get(ptep);
	WARN_ON(!pte_none(pte));

will that work for you?

-aneesh