linuxppc-dev.lists.ozlabs.org archive mirror
 help / color / mirror / Atom feed
* [PATCH] powerpc/hugetlb: Fix 512k hugepages on 8xx with 16k page size
@ 2020-02-06 13:50 Christophe Leroy
  2020-02-19 12:39 ` Michael Ellerman
  0 siblings, 1 reply; 2+ messages in thread
From: Christophe Leroy @ 2020-02-06 13:50 UTC (permalink / raw)
  To: Benjamin Herrenschmidt, Paul Mackerras, Michael Ellerman,
	aneesh.kumar
  Cc: linuxppc-dev, linux-kernel

Commit 55c8fc3f4930 ("powerpc/8xx: reintroduce 16K pages with HW
assistance") redefined pte_t as a struct of 4 pte_basic_t, because
in 16K pages mode there are four identical entries in the
page table. But the size of hugepage tables is calculated based
of the size of (void *). Therefore, we end up with page tables
of size 1k instead of 4k for 512k pages.

As 512k hugepage tables are the same size as standard page tables,
ie 4k, use the standard page tables instead of PGT_CACHE tables.

Fixes: 3fb69c6a1a13 ("powerpc/8xx: Enable 512k hugepage support with HW assistance")
Cc: stable@vger.kernel.org
Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
---
 arch/powerpc/mm/hugetlbpage.c | 29 ++++++++++++++++++-----------
 1 file changed, 18 insertions(+), 11 deletions(-)

diff --git a/arch/powerpc/mm/hugetlbpage.c b/arch/powerpc/mm/hugetlbpage.c
index c61032580185..edf511c2a30a 100644
--- a/arch/powerpc/mm/hugetlbpage.c
+++ b/arch/powerpc/mm/hugetlbpage.c
@@ -54,20 +54,24 @@ static int __hugepte_alloc(struct mm_struct *mm, hugepd_t *hpdp,
 	if (pshift >= pdshift) {
 		cachep = PGT_CACHE(PTE_T_ORDER);
 		num_hugepd = 1 << (pshift - pdshift);
+		new = NULL;
 	} else if (IS_ENABLED(CONFIG_PPC_8xx)) {
-		cachep = PGT_CACHE(PTE_INDEX_SIZE);
+		cachep = NULL;
 		num_hugepd = 1;
+		new = pte_alloc_one(mm);
 	} else {
 		cachep = PGT_CACHE(pdshift - pshift);
 		num_hugepd = 1;
+		new = NULL;
 	}
 
-	if (!cachep) {
+	if (!cachep && !new) {
 		WARN_ONCE(1, "No page table cache created for hugetlb tables");
 		return -ENOMEM;
 	}
 
-	new = kmem_cache_alloc(cachep, pgtable_gfp_flags(mm, GFP_KERNEL));
+	if (cachep)
+		new = kmem_cache_alloc(cachep, pgtable_gfp_flags(mm, GFP_KERNEL));
 
 	BUG_ON(pshift > HUGEPD_SHIFT_MASK);
 	BUG_ON((unsigned long)new & HUGEPD_SHIFT_MASK);
@@ -98,7 +102,10 @@ static int __hugepte_alloc(struct mm_struct *mm, hugepd_t *hpdp,
 	if (i < num_hugepd) {
 		for (i = i - 1 ; i >= 0; i--, hpdp--)
 			*hpdp = __hugepd(0);
-		kmem_cache_free(cachep, new);
+		if (cachep)
+			kmem_cache_free(cachep, new);
+		else
+			pte_free(mm, new);
 	} else {
 		kmemleak_ignore(new);
 	}
@@ -325,8 +332,7 @@ static void free_hugepd_range(struct mmu_gather *tlb, hugepd_t *hpdp, int pdshif
 	if (shift >= pdshift)
 		hugepd_free(tlb, hugepte);
 	else if (IS_ENABLED(CONFIG_PPC_8xx))
-		pgtable_free_tlb(tlb, hugepte,
-				 get_hugepd_cache_index(PTE_INDEX_SIZE));
+		pgtable_free_tlb(tlb, hugepte, 0);
 	else
 		pgtable_free_tlb(tlb, hugepte,
 				 get_hugepd_cache_index(pdshift - shift));
@@ -640,12 +646,13 @@ static int __init hugetlbpage_init(void)
 		 * if we have pdshift and shift value same, we don't
 		 * use pgt cache for hugepd.
 		 */
-		if (pdshift > shift && IS_ENABLED(CONFIG_PPC_8xx))
-			pgtable_cache_add(PTE_INDEX_SIZE);
-		else if (pdshift > shift)
-			pgtable_cache_add(pdshift - shift);
-		else if (IS_ENABLED(CONFIG_PPC_FSL_BOOK3E) || IS_ENABLED(CONFIG_PPC_8xx))
+		if (pdshift > shift) {
+			if (!IS_ENABLED(CONFIG_PPC_8xx))
+				pgtable_cache_add(pdshift - shift);
+		} else if (IS_ENABLED(CONFIG_PPC_FSL_BOOK3E) ||
+			   IS_ENABLED(CONFIG_PPC_8xx)) {
 			pgtable_cache_add(PTE_T_ORDER);
+		}
 
 		configured = true;
 	}
-- 
2.25.0


^ permalink raw reply related	[flat|nested] 2+ messages in thread

end of thread, other threads:[~2020-02-19 12:43 UTC | newest]

Thread overview: 2+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2020-02-06 13:50 [PATCH] powerpc/hugetlb: Fix 512k hugepages on 8xx with 16k page size Christophe Leroy
2020-02-19 12:39 ` Michael Ellerman

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).