On 8/21/20 1:31 PM, Anshuman Khandual wrote:

On 08/21/2020 12:23 PM, Aneesh Kumar K.V wrote:

quoted

On 8/21/20 9:03 AM, Anshuman Khandual wrote:

quoted

On 08/19/2020 07:15 PM, Aneesh Kumar K.V wrote:

quoted

"Aneesh Kumar K.V" [off-list ref] writes:

quoted

This patch series includes fixes for debug_vm_pgtable test code so that
they follow page table updates rules correctly. The first two patches introduce
changes w.r.t ppc64. The patches are included in this series for completeness. We can
merge them via ppc64 tree if required.

Hugetlb test is disabled on ppc64 because that needs larger change to satisfy
page table update rules.

Changes from V1:
* Address review feedback
* drop test specific pfn_pte and pfn_pmd.
* Update ppc64 page table helper to add _PAGE_PTE

Aneesh Kumar K.V (13):
    powerpc/mm: Add DEBUG_VM WARN for pmd_clear
    powerpc/mm: Move setting pte specific flags to pfn_pte
    mm/debug_vm_pgtable/ppc64: Avoid setting top bits in radom value
    mm/debug_vm_pgtables/hugevmap: Use the arch helper to identify huge
      vmap support.
    mm/debug_vm_pgtable/savedwrite: Enable savedwrite test with
      CONFIG_NUMA_BALANCING
    mm/debug_vm_pgtable/THP: Mark the pte entry huge before using
      set_pmd/pud_at
    mm/debug_vm_pgtable/set_pte/pmd/pud: Don't use set_*_at to update an
      existing pte entry
    mm/debug_vm_pgtable/thp: Use page table depost/withdraw with THP
    mm/debug_vm_pgtable/locks: Move non page table modifying test together
    mm/debug_vm_pgtable/locks: Take correct page table lock
    mm/debug_vm_pgtable/pmd_clear: Don't use pmd/pud_clear on pte entries
    mm/debug_vm_pgtable/hugetlb: Disable hugetlb test on ppc64
    mm/debug_vm_pgtable: populate a pte entry before fetching it

   arch/powerpc/include/asm/book3s/64/pgtable.h |  29 +++-
   arch/powerpc/include/asm/nohash/pgtable.h    |   5 -
   arch/powerpc/mm/book3s64/pgtable.c           |   2 +-
   arch/powerpc/mm/pgtable.c                    |   5 -
   include/linux/io.h                           |  12 ++
   mm/debug_vm_pgtable.c                        | 151 +++++++++++--------
   6 files changed, 127 insertions(+), 77 deletions(-)

BTW I picked a wrong branch when sending this. Attaching the diff
against what I want to send.  pfn_pmd() no more updates _PAGE_PTE
because that is handled by pmd_mkhuge().

diff --git a/arch/powerpc/mm/book3s64/pgtable.c b/arch/powerpc/mm/book3s64/pgtable.c
index 3b4da7c63e28..e18ae50a275c 100644
--- a/arch/powerpc/mm/book3s64/pgtable.c
+++ b/arch/powerpc/mm/book3s64/pgtable.c

@@ -141,7 +141,7 @@ pmd_t pfn_pmd(unsigned long pfn, pgprot_t pgprot)
       unsigned long pmdv;
         pmdv = (pfn << PAGE_SHIFT) & PTE_RPN_MASK;
-    return __pmd(pmdv | pgprot_val(pgprot) | _PAGE_PTE);
+    return pmd_set_protbits(__pmd(pmdv), pgprot);
   }
     pmd_t mk_pmd(struct page *page, pgprot_t pgprot)

diff --git a/mm/debug_vm_pgtable.c b/mm/debug_vm_pgtable.c
index 7d9f8e1d790f..cad61d22f33a 100644
--- a/mm/debug_vm_pgtable.c
+++ b/mm/debug_vm_pgtable.c

@@ -229,7 +229,7 @@ static void __init pmd_huge_tests(pmd_t *pmdp, unsigned long pfn, pgprot_t prot)
     static void __init pmd_savedwrite_tests(unsigned long pfn, pgprot_t prot)
   {
-    pmd_t pmd = pfn_pmd(pfn, prot);
+    pmd_t pmd = pmd_mkhuge(pfn_pmd(pfn, prot));
         if (!IS_ENABLED(CONFIG_NUMA_BALANCING))
           return;

Cover letter does not mention which branch or tag this series applies on.
Just assumed it to be 5.9-rc1. Should the above changes be captured as a
pre-requisite patch ?

Anyways, the series fails to be build on arm64.

A) Without CONFIG_TRANSPARENT_HUGEPAGE

mm/debug_vm_pgtable.c: In function ‘debug_vm_pgtable’:
mm/debug_vm_pgtable.c:1045:2: error: too many arguments to function ‘pmd_advanced_tests’
    pmd_advanced_tests(mm, vma, pmdp, pmd_aligned, vaddr, prot, saved_ptep);
    ^~~~~~~~~~~~~~~~~~
mm/debug_vm_pgtable.c:366:20: note: declared here
   static void __init pmd_advanced_tests(struct mm_struct *mm,
                      ^~~~~~~~~~~~~~~~~~

B) As mentioned previously, this should be solved by including <linux/io.h>

mm/debug_vm_pgtable.c: In function ‘pmd_huge_tests’:
mm/debug_vm_pgtable.c:215:7: error: implicit declaration of function ‘arch_ioremap_pmd_supported’; did you mean ‘arch_disable_smp_support’? [-Werror=implicit-function-declaration]
    if (!arch_ioremap_pmd_supported())
         ^~~~~~~~~~~~~~~~~~~~~~~~~~

Please make sure that the series builds on all enabled platforms i.e x86,
arm64, ppc32, ppc64, arc, s390 along with selectively enabling/disabling
all the features that make various #ifdefs in the test.

I was hoping to get kernel test robot build report to verify that. But if you can help with that i have pushed a branch to github with reported build failure fixes.

https://github.com/kvaneesh/linux/tree/debug_vm_pgtable

I still haven't looked at the PMD_FOLDED feedback from Christophe because I am not sure i follow why we are checking for PMD folded there.

If this series does not build on existing enabled platforms, wondering
how effective the review could be, assuming that things would need to
change again to fix those build failures on various platforms. Getting
this to build here is essential, as not all page table constructs are
available across these platforms. Hence wondering, it might be better
if you could resend the series after fixing build issues.

Sure. I am hoping kernel test robot will pick this up. I did an x86 and 
about 19 different ppc config build with the series. The git tree above 
was pushed with that. Considering you authored the change i am wondering 
if you could help with checking other architecture (may be atleast arm 
variant)

-aneesh