[PATCH] arm64: Define only {pud/pmd}_{set/clear}_huge when usefull

When PUD and/or PMD are folded, those functions are useless
and we now have a stub in linux/pgtable.h

Reported-by: Naresh Kamboju <redacted>
Signed-off-by: Christophe Leroy <redacted>
---
 arch/arm64/mm/mmu.c | 20 ++++++++++++--------
 1 file changed, 12 insertions(+), 8 deletions(-)

diff --git a/arch/arm64/mm/mmu.c b/arch/arm64/mm/mmu.c
index 6dd9369e3ea0..98af085afe90 100644
--- a/arch/arm64/mm/mmu.c
+++ b/arch/arm64/mm/mmu.c

@@ -1337,6 +1337,7 @@ void *__init fixmap_remap_fdt(phys_addr_t dt_phys, int *size, pgprot_t prot)
 	return dt_virt;
 }
 
+#if CONFIG_PGTABLE_LEVELS > 3
 int pud_set_huge(pud_t *pudp, phys_addr_t phys, pgprot_t prot)
 {
 	pud_t new_pud = pfn_pud(__phys_to_pfn(phys), mk_pud_sect_prot(prot));

@@ -1351,6 +1352,16 @@ int pud_set_huge(pud_t *pudp, phys_addr_t phys, pgprot_t prot)
 	return 1;
 }
 
+int pud_clear_huge(pud_t *pudp)
+{
+	if (!pud_sect(READ_ONCE(*pudp)))
+		return 0;
+	pud_clear(pudp);
+	return 1;
+}
+#endif
+
+#if CONFIG_PGTABLE_LEVELS > 2
 int pmd_set_huge(pmd_t *pmdp, phys_addr_t phys, pgprot_t prot)
 {
 	pmd_t new_pmd = pfn_pmd(__phys_to_pfn(phys), mk_pmd_sect_prot(prot));

@@ -1365,14 +1376,6 @@ int pmd_set_huge(pmd_t *pmdp, phys_addr_t phys, pgprot_t prot)
 	return 1;
 }
 
-int pud_clear_huge(pud_t *pudp)
-{
-	if (!pud_sect(READ_ONCE(*pudp)))
-		return 0;
-	pud_clear(pudp);
-	return 1;
-}
-
 int pmd_clear_huge(pmd_t *pmdp)
 {
 	if (!pmd_sect(READ_ONCE(*pmdp)))

@@ -1380,6 +1383,7 @@ int pmd_clear_huge(pmd_t *pmdp)
 	pmd_clear(pmdp);
 	return 1;
 }
+#endif
 
 int pmd_free_pte_page(pmd_t *pmdp, unsigned long addr)
 {

-- 
2.25.0

On Fri, 14 May 2021 11:08:53 +0000 (UTC) Christophe Leroy [off-list ref] wrote:

When PUD and/or PMD are folded, those functions are useless
and we now have a stub in linux/pgtable.h

OK, help me out here please.  What patch does this fix?

On Fri, May 14, 2021 at 02:42:00PM -0700, Andrew Morton wrote:

On Fri, 14 May 2021 11:08:53 +0000 (UTC) Christophe Leroy [off-list ref] wrote:

quoted

When PUD and/or PMD are folded, those functions are useless
and we now have a stub in linux/pgtable.h

OK, help me out here please.  What patch does this fix?

Naresh's original report is here it seems:

https://lore.kernel.org/r/CA+G9fYv79t0+2W4Rt3wDkBShc4eY3M3utC5BHqUgGDwMYExYMw@mail.gmail.com/

I can reproduce the failure that he reported with
ARCH=arm64 allmodconfig and this patch resolves it for me.

Cheers,
Nathan

Le 14/05/2021 à 23:42, Andrew Morton a écrit :

On Fri, 14 May 2021 11:08:53 +0000 (UTC) Christophe Leroy [off-list ref] wrote:

quoted

When PUD and/or PMD are folded, those functions are useless
and we now have a stub in linux/pgtable.h

OK, help me out here please.  What patch does this fix?

Both this one and the x86 one from the day before fix 1cff41494b15cd82c1ec418bb5c ("mm/pgtable: add 
stubs for {pmd/pub}_{set/clear}_huge")

I think both the x86 fix and the arm64 fix should be squashed into that patch at the end.

I checked, the only other architecture involving pud_set_huge() and friends is powerpc, and powerpc 
doesn't have this problem as it only defined those for book3s/64 platforms which have 4 level page 
tables by definition.

Thanks
Christophe