FAILED: patch "[PATCH] mm: don't install PMD mappings when THPs are disabled by the" failed to apply to 6.6-stable tree

The patch below does not apply to the 6.6-stable tree.
If someone wants it applied there, or to any other stable or longterm tree, then please email the backport, including the original git commit id to <[email protected]>.

To reproduce the conflict and resubmit, you may use the following commands:

git fetch https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git/ linux-6.6.y
git checkout FETCH_HEAD
git cherry-pick -x 2b0f922323ccfa76219bcaacd35cd50aeaa13592
# <resolve conflicts, build, test, etc.>
git commit -s
git send-email --to '<[email protected]>' --in-reply-to '2024101842-empty-espresso-c8a3@gregkh' --subject-prefix 'PATCH 6.6.y' HEAD^..

Possible dependencies:

thanks,

greg k-h

------------------ original commit in Linus's tree ------------------

From 2b0f922323ccfa76219bcaacd35cd50aeaa13592 Mon Sep 17 00:00:00 2001
From: David Hildenbrand <[email protected]>
Date: Fri, 11 Oct 2024 12:24:45 +0200
Subject: [PATCH] mm: don't install PMD mappings when THPs are disabled by the
hw/process/vma

We (or rather, readahead logic :) ) might be allocating a THP in the pagecache and then try mapping it into a process that explicitly disabled
THP: we might end up installing PMD mappings.

This is a problem for s390x KVM, which explicitly remaps all PMD-mapped THPs to be PTE-mapped in s390_enable_sie()->thp_split_mm(), before starting the VM.

For example, starting a VM backed on a file system with large folios supported makes the VM crash when the VM tries accessing such a mapping using KVM.

Is it also a problem when the HW disabled THP using TRANSPARENT_HUGEPAGE_UNSUPPORTED? At least on x86 this would be the case without X86_FEATURE_PSE.

In the future, we might be able to do better on s390x and only disallow PMD mappings -- what s390x and likely TRANSPARENT_HUGEPAGE_UNSUPPORTED really wants. For now, fix it by essentially performing the same check as would be done in __thp_vma_allowable_orders() or in shmem code, where this works as expected, and disallow PMD mappings, making us fallback to PTE mappings.

index c0869a962ddd..30feedabc932 100644
@@ -4920,6 +4920,15 @@ vm_fault_t do_set_pmd(struct vm_fault *vmf, struct page *page)
 	pmd_t entry;
 	vm_fault_t ret = VM_FAULT_FALLBACK;
 
+	/*
+	 * It is too late to allocate a small folio, we already have a large
+	 * folio in the pagecache: especially s390 KVM cannot tolerate any
+	 * PMD mappings, but PTE-mapped THP are fine. So let's simply refuse any
+	 * PMD mappings if THPs are disabled.
+	 */
+	if (thp_disabled_by_hw() || vma_thp_disabled(vma, vma->vm_flags))
+		return ret;
+
 	if (!thp_vma_suitable_order(vma, haddr, PMD_ORDER))
 		return ret;
 

From: Kefeng Wang <[email protected]>

Patch series "mm: don't install PMD mappings when THPs are disabled by the hw/process/vma".

During testing, it was found that we can get PMD mappings in processes where THP (and more precisely, PMD mappings) are supposed to be disabled. While it works as expected for anon+shmem, the pagecache is the problematic bit.

For s390 KVM this currently means that a VM backed by a file located on filesystem with large folio support can crash when KVM tries accessing the problematic page, because the readahead logic might decide to use a PMD-sized THP and faulting it into the page tables will install a PMD mapping, something that s390 KVM cannot tolerate.

This might also be a problem with HW that does not support PMD mappings, but I did not try reproducing it.

Fix it by respecting the ways to disable THPs when deciding whether we can install a PMD mapping. khugepaged should already be taking care of not collapsing if THPs are effectively disabled for the hw/process/vma.

This patch (of 2):

Add vma_thp_disabled() and thp_disabled_by_hw() helpers to be shared by shmem_allowable_huge_orders() and __thp_vma_allowable_orders().

---

The change in mm/shmem.c does not exist yet.

This patch is required to backport the fix 2b0f922323ccfa76219bcaacd35cd50aeaa13592, for which a backport will be sent separately in reply to the "FAILED: ..." mail.

---
 include/linux/huge_mm.h | 18 ++++++++++++++++++
 mm/huge_memory.c        | 15 ++-------------
 2 files changed, 20 insertions(+), 13 deletions(-)

index fa0350b0812a..fc789c0ac85b 100644
@@ -137,6 +137,24 @@ bool hugepage_vma_check(struct vm_area_struct *vma, unsigned long vm_flags,
 	(transparent_hugepage_flags &					\
 	 (1<<TRANSPARENT_HUGEPAGE_USE_ZERO_PAGE_FLAG))
 
+static inline bool vma_thp_disabled(struct vm_area_struct *vma,
+		unsigned long vm_flags)
+{
+	/*
+	 * Explicitly disabled through madvise or prctl, or some
+	 * architectures may disable THP for some mappings, for
+	 * example, s390 kvm.
+	 */
+	return (vm_flags & VM_NOHUGEPAGE) ||
+	       test_bit(MMF_DISABLE_THP, &vma->vm_mm->flags);
+}
+
+static inline bool thp_disabled_by_hw(void)
+{
+	/* If the hardware/firmware marked hugepage support disabled. */
+	return transparent_hugepage_flags & (1 << TRANSPARENT_HUGEPAGE_UNSUPPORTED);
+}
+
 unsigned long thp_get_unmapped_area(struct file *filp, unsigned long addr,
 		unsigned long len, unsigned long pgoff, unsigned long flags);
 

index 9aea11b1477c..dfd6577225d8 100644
@@ -78,19 +78,8 @@ bool hugepage_vma_check(struct vm_area_struct *vma, unsigned long vm_flags,
 	if (!vma->vm_mm)		/* vdso */
 		return false;
 
-	/*
-	 * Explicitly disabled through madvise or prctl, or some
-	 * architectures may disable THP for some mappings, for
-	 * example, s390 kvm.
-	 * */
-	if ((vm_flags & VM_NOHUGEPAGE) ||
-	    test_bit(MMF_DISABLE_THP, &vma->vm_mm->flags))
-		return false;
-	/*
-	 * If the hardware/firmware marked hugepage support disabled.
-	 */
-	if (transparent_hugepage_flags & (1 << TRANSPARENT_HUGEPAGE_UNSUPPORTED))
-		return false;
+	if (thp_disabled_by_hw() || vma_thp_disabled(vma, vm_flags))
+		return 0;
 
 	/* khugepaged doesn't collapse DAX vma, but page fault is fine. */
 	if (vma_is_dax(vma))

2.46.1

We (or rather, readahead logic :) ) might be allocating a THP in the pagecache and then try mapping it into a process that explicitly disabled
THP: we might end up installing PMD mappings.

This is a problem for s390x KVM, which explicitly remaps all PMD-mapped THPs to be PTE-mapped in s390_enable_sie()->thp_split_mm(), before starting the VM.

For example, starting a VM backed on a file system with large folios supported makes the VM crash when the VM tries accessing such a mapping using KVM.

Is it also a problem when the HW disabled THP using TRANSPARENT_HUGEPAGE_UNSUPPORTED? At least on x86 this would be the case without X86_FEATURE_PSE.

In the future, we might be able to do better on s390x and only disallow PMD mappings -- what s390x and likely TRANSPARENT_HUGEPAGE_UNSUPPORTED really wants. For now, fix it by essentially performing the same check as would be done in __thp_vma_allowable_orders() or in shmem code, where this works as expected, and disallow PMD mappings, making us fallback to PTE mappings.

---

Minor contextual difference.

Note that the backport of 963756aac1f011d904ddd9548ae82286d3a91f96 is required (send separately as reply to the "FAILED:" mail).

---
 mm/memory.c | 9 +++++++++
 1 file changed, 9 insertions(+)

index b6ddfe22c5d5..742c2f65c2c8 100644
@@ -4293,6 +4293,15 @@ vm_fault_t do_set_pmd(struct vm_fault *vmf, struct page *page)
 	pmd_t entry;
 	vm_fault_t ret = VM_FAULT_FALLBACK;
 
+	/*
+	 * It is too late to allocate a small folio, we already have a large
+	 * folio in the pagecache: especially s390 KVM cannot tolerate any
+	 * PMD mappings, but PTE-mapped THP are fine. So let's simply refuse any
+	 * PMD mappings if THPs are disabled.
+	 */
+	if (thp_disabled_by_hw() || vma_thp_disabled(vma, vma->vm_flags))
+		return ret;
+
 	if (!transhuge_vma_suitable(vma, haddr))
 		return ret;
 

2.46.1

Gentle ping, XEN PV users reported an issue fixed by this fix upstream.

Cheers,

David / dhildenb

Hi David,

Petr

Hi Petr,

Yes, that's better. Same applies to the 6.1.y backport of this.

Cheers,

David / dhildenb

Ok, dropping this from the review queue, please resend the updated versions.

thansk,

greg k-h

Resending both patches in one series now, easier for everybody that way.

Conflicts:
* "mm: huge_memory: add vma_thp_disabled() and thp_disabled_by_hw()" -> The change in mm/shmem.c does not exist yet.
-> Small contextual conflict.
* "mm: don't install PMD mappings when THPs are disabled by the hw/process/vma" -> Small contextual conflict.

v1 -> v2:
* "mm: huge_memory: add vma_thp_disabled() and thp_disabled_by_hw()" -> Keep using "return false;" instead of "return 0;"

David Hildenbrand (1):
mm: don't install PMD mappings when THPs are disabled by the
hw/process/vma

Kefeng Wang (1):
mm: huge_memory: add vma_thp_disabled() and thp_disabled_by_hw()

 include/linux/huge_mm.h | 18 ++++++++++++++++++
 mm/huge_memory.c        | 13 +------------
 mm/memory.c             |  9 +++++++++
 3 files changed, 28 insertions(+), 12 deletions(-)

2.47.0

From: Kefeng Wang <[email protected]>

Patch series "mm: don't install PMD mappings when THPs are disabled by the hw/process/vma".

During testing, it was found that we can get PMD mappings in processes where THP (and more precisely, PMD mappings) are supposed to be disabled. While it works as expected for anon+shmem, the pagecache is the problematic bit.

For s390 KVM this currently means that a VM backed by a file located on filesystem with large folio support can crash when KVM tries accessing the problematic page, because the readahead logic might decide to use a PMD-sized THP and faulting it into the page tables will install a PMD mapping, something that s390 KVM cannot tolerate.

This might also be a problem with HW that does not support PMD mappings, but I did not try reproducing it.

Fix it by respecting the ways to disable THPs when deciding whether we can install a PMD mapping. khugepaged should already be taking care of not collapsing if THPs are effectively disabled for the hw/process/vma.

This patch (of 2):

Add vma_thp_disabled() and thp_disabled_by_hw() helpers to be shared by shmem_allowable_huge_orders() and __thp_vma_allowable_orders().

---
 include/linux/huge_mm.h | 18 ++++++++++++++++++
 mm/huge_memory.c        | 13 +------------
 2 files changed, 19 insertions(+), 12 deletions(-)

index fa0350b0812ab..fc789c0ac85b8 100644
@@ -137,6 +137,24 @@ bool hugepage_vma_check(struct vm_area_struct *vma, unsigned long vm_flags,
 	(transparent_hugepage_flags &					\
 	 (1<<TRANSPARENT_HUGEPAGE_USE_ZERO_PAGE_FLAG))
 
+static inline bool vma_thp_disabled(struct vm_area_struct *vma,
+		unsigned long vm_flags)
+{
+	/*
+	 * Explicitly disabled through madvise or prctl, or some
+	 * architectures may disable THP for some mappings, for
+	 * example, s390 kvm.
+	 */
+	return (vm_flags & VM_NOHUGEPAGE) ||
+	       test_bit(MMF_DISABLE_THP, &vma->vm_mm->flags);
+}
+
+static inline bool thp_disabled_by_hw(void)
+{
+	/* If the hardware/firmware marked hugepage support disabled. */
+	return transparent_hugepage_flags & (1 << TRANSPARENT_HUGEPAGE_UNSUPPORTED);
+}
+
 unsigned long thp_get_unmapped_area(struct file *filp, unsigned long addr,
 		unsigned long len, unsigned long pgoff, unsigned long flags);
 

index 9aea11b1477c8..7b4cb5c68b61b 100644
@@ -78,18 +78,7 @@ bool hugepage_vma_check(struct vm_area_struct *vma, unsigned long vm_flags,
 	if (!vma->vm_mm)		/* vdso */
 		return false;
 
-	/*
-	 * Explicitly disabled through madvise or prctl, or some
-	 * architectures may disable THP for some mappings, for
-	 * example, s390 kvm.
-	 * */
-	if ((vm_flags & VM_NOHUGEPAGE) ||
-	    test_bit(MMF_DISABLE_THP, &vma->vm_mm->flags))
-		return false;
-	/*
-	 * If the hardware/firmware marked hugepage support disabled.
-	 */
-	if (transparent_hugepage_flags & (1 << TRANSPARENT_HUGEPAGE_UNSUPPORTED))
+	if (thp_disabled_by_hw() || vma_thp_disabled(vma, vm_flags))
 		return false;
 
 	/* khugepaged doesn't collapse DAX vma, but page fault is fine. */

2.47.0

We (or rather, readahead logic :) ) might be allocating a THP in the pagecache and then try mapping it into a process that explicitly disabled
THP: we might end up installing PMD mappings.

This is a problem for s390x KVM, which explicitly remaps all PMD-mapped THPs to be PTE-mapped in s390_enable_sie()->thp_split_mm(), before starting the VM.

For example, starting a VM backed on a file system with large folios supported makes the VM crash when the VM tries accessing such a mapping using KVM.

Is it also a problem when the HW disabled THP using TRANSPARENT_HUGEPAGE_UNSUPPORTED? At least on x86 this would be the case without X86_FEATURE_PSE.

In the future, we might be able to do better on s390x and only disallow PMD mappings -- what s390x and likely TRANSPARENT_HUGEPAGE_UNSUPPORTED really wants. For now, fix it by essentially performing the same check as would be done in __thp_vma_allowable_orders() or in shmem code, where this works as expected, and disallow PMD mappings, making us fallback to PTE mappings.

---
 mm/memory.c | 9 +++++++++
 1 file changed, 9 insertions(+)

index b6ddfe22c5d5c..742c2f65c2c85 100644
@@ -4293,6 +4293,15 @@ vm_fault_t do_set_pmd(struct vm_fault *vmf, struct page *page)
 	pmd_t entry;
 	vm_fault_t ret = VM_FAULT_FALLBACK;
 
+	/*
+	 * It is too late to allocate a small folio, we already have a large
+	 * folio in the pagecache: especially s390 KVM cannot tolerate any
+	 * PMD mappings, but PTE-mapped THP are fine. So let's simply refuse any
+	 * PMD mappings if THPs are disabled.
+	 */
+	if (thp_disabled_by_hw() || vma_thp_disabled(vma, vma->vm_flags))
+		return ret;
+
 	if (!transhuge_vma_suitable(vma, haddr))
 		return ret;
 

2.47.0

Much better, now queued up, thanks!

greg k-h