[RFC v1 1/2] drivers/vfio_pci_core: Change PXD_ORDER check from switch case to if/else block

Architectures like PowerPC uses runtime defined values for
PMD_ORDER/PUD_ORDER. This is because it can use either RADIX or HASH MMU
at runtime using kernel cmdline. So the pXd_index_size is not known at
compile time. Without this fix, when we add huge pfn support on powerpc
in the next patch, vfio_pci_core driver compilation can fail with the
following errors.

  CC [M]  drivers/vfio/vfio_main.o
  CC [M]  drivers/vfio/group.o
  CC [M]  drivers/vfio/container.o
  CC [M]  drivers/vfio/virqfd.o
  CC [M]  drivers/vfio/vfio_iommu_spapr_tce.o
  CC [M]  drivers/vfio/pci/vfio_pci_core.o
  CC [M]  drivers/vfio/pci/vfio_pci_intrs.o
  CC [M]  drivers/vfio/pci/vfio_pci_rdwr.o
  CC [M]  drivers/vfio/pci/vfio_pci_config.o
  CC [M]  drivers/vfio/pci/vfio_pci.o
  AR      kernel/built-in.a
../drivers/vfio/pci/vfio_pci_core.c: In function ‘vfio_pci_vmf_insert_pfn’:
../drivers/vfio/pci/vfio_pci_core.c:1678:9: error: case label does not reduce to an integer constant
 1678 |         case PMD_ORDER:
      |         ^~~~
../drivers/vfio/pci/vfio_pci_core.c:1682:9: error: case label does not reduce to an integer constant
 1682 |         case PUD_ORDER:
      |         ^~~~
make[6]: *** [../scripts/Makefile.build:289: drivers/vfio/pci/vfio_pci_core.o] Error 1
make[6]: *** Waiting for unfinished jobs....
make[5]: *** [../scripts/Makefile.build:546: drivers/vfio/pci] Error 2
make[5]: *** Waiting for unfinished jobs....
make[4]: *** [../scripts/Makefile.build:546: drivers/vfio] Error 2
make[3]: *** [../scripts/Makefile.build:546: drivers] Error 2

Fixes: f9e54c3a2f5b7 ("vfio/pci: implement huge_fault support")
Signed-off-by: Ritesh Harjani (IBM) <ritesh.list@gmail.com>
---
 drivers/vfio/pci/vfio_pci_core.c | 15 +++++++--------
 1 file changed, 7 insertions(+), 8 deletions(-)

diff --git a/drivers/vfio/pci/vfio_pci_core.c b/drivers/vfio/pci/vfio_pci_core.c
index d43745fe4c84..5395a6f30904 100644
--- a/drivers/vfio/pci/vfio_pci_core.c
+++ b/drivers/vfio/pci/vfio_pci_core.c

@@ -1670,21 +1670,20 @@ vm_fault_t vfio_pci_vmf_insert_pfn(struct vfio_pci_core_device *vdev,
 	if (vdev->pm_runtime_engaged || !__vfio_pci_memory_enabled(vdev))
 		return VM_FAULT_SIGBUS;

-	switch (order) {
-	case 0:
+	if (order == 0) {
 		return vmf_insert_pfn(vmf->vma, vmf->address, pfn);
+	}
 #ifdef CONFIG_ARCH_SUPPORTS_PMD_PFNMAP
-	case PMD_ORDER:
+	 else if (order == PMD_ORDER) {
 		return vmf_insert_pfn_pmd(vmf, pfn, false);
+	 }
 #endif
 #ifdef CONFIG_ARCH_SUPPORTS_PUD_PFNMAP
-	case PUD_ORDER:
+	 else if (order == PUD_ORDER) {
 		return vmf_insert_pfn_pud(vmf, pfn, false);
-		break;
+	 }
 #endif
-	default:
-		return VM_FAULT_FALLBACK;
-	}
+	return VM_FAULT_FALLBACK;
 }
 EXPORT_SYMBOL_GPL(vfio_pci_vmf_insert_pfn);

--

2.53.0

This uses _RPAGE_SW2 bit for the PMD and PUDs similar to PTEs.
This also adds support for {pte,pmd,pud}_pgprot helpers needed for
follow_pfnmap APIs.

This allows us to extend the PFN mappings, e.g. PCI MMIO bars where
it can grow as large as 8GB or even bigger, to map at PMD / PUD level.
VFIO PCI core driver already supports fault handling at PMD / PUD level
for more efficient BAR mappings.

Signed-off-by: Ritesh Harjani (IBM) <ritesh.list@gmail.com>
---

@linux-mm:
Is there any official test which I could use to verify this functionality.

For now I used basic ivshmem setup + vfio using Qemu and validated using some
basic test to see that we are seeing these prints.

[ 4351.435050] vfio_pci_mmap_huge_fault: 3 callbacks suppressed
[ 4351.435234] vfio-pci 0001:00:00.0: vfio_pci_mmap_huge_fault(,order = 5) BAR 2 page offset 0x0: 0x100
[ 4351.457005] vfio-pci 0001:00:00.0: vfio_pci_mmap_huge_fault(,order = 5) BAR 2 page offset 0x40: 0x100
[ 4351.463684] vfio-pci 0001:00:00.0: vfio_pci_mmap_huge_fault(,order = 5) BAR 2 page offset 0x20: 0x100

 arch/powerpc/Kconfig                         |  1 +
 arch/powerpc/include/asm/book3s/64/pgtable.h | 23 ++++++++++++++++++++
 arch/powerpc/include/asm/pgtable.h           | 12 ++++++++++
 3 files changed, 36 insertions(+)

diff --git a/arch/powerpc/Kconfig b/arch/powerpc/Kconfig
index ad7a2fe63a2a..cf9283757e5d 100644
--- a/arch/powerpc/Kconfig
+++ b/arch/powerpc/Kconfig

@@ -172,6 +172,7 @@ config PPC
 	select ARCH_STACKWALK
 	select ARCH_SUPPORTS_ATOMIC_RMW
 	select ARCH_SUPPORTS_DEBUG_PAGEALLOC	if PPC_BOOK3S || PPC_8xx
+	select ARCH_SUPPORTS_HUGE_PFNMAP	if PPC_BOOK3S_64 && TRANSPARENT_HUGEPAGE
 	select ARCH_SUPPORTS_PAGE_TABLE_CHECK	if !HUGETLB_PAGE
 	select ARCH_SUPPORTS_SCHED_MC		if SMP
 	select ARCH_SUPPORTS_SCHED_SMT		if PPC64 && SMP

diff --git a/arch/powerpc/include/asm/book3s/64/pgtable.h b/arch/powerpc/include/asm/book3s/64/pgtable.h
index 1a91762b455d..639cbf34f752 100644
--- a/arch/powerpc/include/asm/book3s/64/pgtable.h
+++ b/arch/powerpc/include/asm/book3s/64/pgtable.h

@@ -1289,6 +1289,29 @@ static inline pud_t pud_mkhuge(pud_t pud)
 	return pud;
 }

+#ifdef CONFIG_ARCH_SUPPORTS_PMD_PFNMAP
+static inline bool pmd_special(pmd_t pmd)
+{
+	return pte_special(pmd_pte(pmd));
+}
+
+static inline pmd_t pmd_mkspecial(pmd_t pmd)
+{
+	return pte_pmd(pte_mkspecial(pmd_pte(pmd)));
+}
+#endif
+
+#ifdef CONFIG_ARCH_SUPPORTS_PUD_PFNMAP
+static inline bool pud_special(pud_t pud)
+{
+	return pte_special(pud_pte(pud));
+}
+
+static inline pud_t pud_mkspecial(pud_t pud)
+{
+	return pte_pud(pte_mkspecial(pud_pte(pud)));
+}
+#endif

 #define __HAVE_ARCH_PMDP_SET_ACCESS_FLAGS
 extern int pmdp_set_access_flags(struct vm_area_struct *vma,

diff --git a/arch/powerpc/include/asm/pgtable.h b/arch/powerpc/include/asm/pgtable.h
index dcd3a88caaf6..2d27cb1c2334 100644
--- a/arch/powerpc/include/asm/pgtable.h
+++ b/arch/powerpc/include/asm/pgtable.h

@@ -63,6 +63,18 @@ static inline pgprot_t pte_pgprot(pte_t pte)
 	return __pgprot(pte_flags);
 }

+#define pmd_pgprot pmd_pgprot
+static inline pgprot_t pmd_pgprot(pmd_t pmd)
+{
+	return pte_pgprot(pmd_pte(pmd));
+}
+
+#define pud_pgprot pud_pgprot
+static inline pgprot_t pud_pgprot(pud_t pud)
+{
+	return pte_pgprot(pud_pte(pud));
+}
+
 static inline pgprot_t pgprot_nx(pgprot_t prot)
 {
 	return pte_pgprot(pte_exprotect(__pte(pgprot_val(prot))));
--

2.53.0

Le 27/02/2026 à 07:16, Ritesh Harjani (IBM) a écrit :

Those braces are unneeded as all legs of the if/else are single lines

  #ifdef CONFIG_ARCH_SUPPORTS_PMD_PFNMAP

ifdef could be replaced by IS_ENABLED() because PxD_ORDER and 
vmf_insert_pfn_xxx() are declared all the time

-	case PMD_ORDER:
+	 else if (order == PMD_ORDER) {

'else' is not needed because every 'if' leads to a return statement

  		return vmf_insert_pfn_pmd(vmf, pfn, false);
+	 }
  #endif
  #ifdef CONFIG_ARCH_SUPPORTS_PUD_PFNMAP
-	case PUD_ORDER:
+	 else if (order == PUD_ORDER) {
  		return vmf_insert_pfn_pud(vmf, pfn, false);
-		break;
+	 }
  #endif
-	default:
-		return VM_FAULT_FALLBACK;
-	}
+	return VM_FAULT_FALLBACK;

So at the end we should get something like:

	if (!order)
		return vmf_insert_pfn(vmf->vma, vmf->address, pfn);

	if (IS_ENABLED(CONFIG_ARCH_SUPPORTS_PMD_PFNMAP) && order == PMD_ORDER)
		return vmf_insert_pfn_pmd(vmf, pfn, false);

	if (IS_ENABLED(CONFIG_ARCH_SUPPORTS_PUD_PFNMAP) && order == PMD_ORDER)
		return vmf_insert_pfn_pud(vmf, pfn, false);

	return VM_FAULT_FALLBACK;

  }
  EXPORT_SYMBOL_GPL(vfio_pci_vmf_insert_pfn);

--
2.53.0

Le 27/02/2026 à 07:16, Ritesh Harjani (IBM) a écrit :

This uses _RPAGE_SW2 bit for the PMD and PUDs similar to PTEs.
This also adds support for {pte,pmd,pud}_pgprot helpers needed for
follow_pfnmap APIs.

This allows us to extend the PFN mappings, e.g. PCI MMIO bars where
it can grow as large as 8GB or even bigger, to map at PMD / PUD level.
VFIO PCI core driver already supports fault handling at PMD / PUD level
for more efficient BAR mappings.

Signed-off-by: Ritesh Harjani (IBM) <ritesh.list@gmail.com>

Reviewed-by: Christophe Leroy (CS GROUP) <chleroy@kernel.org>

"Christophe Leroy (CS GROUP)" [off-list ref] writes:

Le 27/02/2026 à 07:16, Ritesh Harjani (IBM) a écrit :

quoted

Architectures like PowerPC uses runtime defined values for
PMD_ORDER/PUD_ORDER. This is because it can use either RADIX or HASH MMU
at runtime using kernel cmdline. So the pXd_index_size is not known at
compile time. Without this fix, when we add huge pfn support on powerpc
in the next patch, vfio_pci_core driver compilation can fail with the
following errors.

   CC [M]  drivers/vfio/vfio_main.o
   CC [M]  drivers/vfio/group.o
   CC [M]  drivers/vfio/container.o
   CC [M]  drivers/vfio/virqfd.o
   CC [M]  drivers/vfio/vfio_iommu_spapr_tce.o
   CC [M]  drivers/vfio/pci/vfio_pci_core.o
   CC [M]  drivers/vfio/pci/vfio_pci_intrs.o
   CC [M]  drivers/vfio/pci/vfio_pci_rdwr.o
   CC [M]  drivers/vfio/pci/vfio_pci_config.o
   CC [M]  drivers/vfio/pci/vfio_pci.o
   AR      kernel/built-in.a
../drivers/vfio/pci/vfio_pci_core.c: In function ‘vfio_pci_vmf_insert_pfn’:
../drivers/vfio/pci/vfio_pci_core.c:1678:9: error: case label does not reduce to an integer constant
  1678 |         case PMD_ORDER:
       |         ^~~~
../drivers/vfio/pci/vfio_pci_core.c:1682:9: error: case label does not reduce to an integer constant
  1682 |         case PUD_ORDER:
       |         ^~~~
make[6]: *** [../scripts/Makefile.build:289: drivers/vfio/pci/vfio_pci_core.o] Error 1
make[6]: *** Waiting for unfinished jobs....
make[5]: *** [../scripts/Makefile.build:546: drivers/vfio/pci] Error 2
make[5]: *** Waiting for unfinished jobs....
make[4]: *** [../scripts/Makefile.build:546: drivers/vfio] Error 2
make[3]: *** [../scripts/Makefile.build:546: drivers] Error 2

Fixes: f9e54c3a2f5b7 ("vfio/pci: implement huge_fault support")
Signed-off-by: Ritesh Harjani (IBM) <ritesh.list@gmail.com>
---
  drivers/vfio/pci/vfio_pci_core.c | 15 +++++++--------
  1 file changed, 7 insertions(+), 8 deletions(-)

diff --git a/drivers/vfio/pci/vfio_pci_core.c b/drivers/vfio/pci/vfio_pci_core.c
index d43745fe4c84..5395a6f30904 100644
--- a/drivers/vfio/pci/vfio_pci_core.c
+++ b/drivers/vfio/pci/vfio_pci_core.c

@@ -1670,21 +1670,20 @@ vm_fault_t vfio_pci_vmf_insert_pfn(struct vfio_pci_core_device *vdev,
  	if (vdev->pm_runtime_engaged || !__vfio_pci_memory_enabled(vdev))
  		return VM_FAULT_SIGBUS;

-	switch (order) {
-	case 0:
+	if (order == 0) {
  		return vmf_insert_pfn(vmf->vma, vmf->address, pfn);
+	}

Those braces are unneeded as all legs of the if/else are single lines

quoted

  #ifdef CONFIG_ARCH_SUPPORTS_PMD_PFNMAP

ifdef could be replaced by IS_ENABLED() because PxD_ORDER and 
vmf_insert_pfn_xxx() are declared all the time

quoted

-	case PMD_ORDER:
+	 else if (order == PMD_ORDER) {

'else' is not needed because every 'if' leads to a return statement

quoted

  		return vmf_insert_pfn_pmd(vmf, pfn, false);
+	 }
  #endif
  #ifdef CONFIG_ARCH_SUPPORTS_PUD_PFNMAP
-	case PUD_ORDER:
+	 else if (order == PUD_ORDER) {
  		return vmf_insert_pfn_pud(vmf, pfn, false);
-		break;
+	 }
  #endif
-	default:
-		return VM_FAULT_FALLBACK;
-	}
+	return VM_FAULT_FALLBACK;

So at the end we should get something like:

	if (!order)
		return vmf_insert_pfn(vmf->vma, vmf->address, pfn);

	if (IS_ENABLED(CONFIG_ARCH_SUPPORTS_PMD_PFNMAP) && order == PMD_ORDER)
		return vmf_insert_pfn_pmd(vmf, pfn, false);

	if (IS_ENABLED(CONFIG_ARCH_SUPPORTS_PUD_PFNMAP) && order == PMD_ORDER)

                                                                ^^^ PUD_ORDER

		return vmf_insert_pfn_pud(vmf, pfn, false);

	return VM_FAULT_FALLBACK;

Looks a lot cleaner. Thanks!
I will make that change in v2.

-ritesh

"Christophe Leroy (CS GROUP)" [off-list ref] writes:

Le 27/02/2026 à 07:16, Ritesh Harjani (IBM) a écrit :

quoted

This uses _RPAGE_SW2 bit for the PMD and PUDs similar to PTEs.
This also adds support for {pte,pmd,pud}_pgprot helpers needed for
follow_pfnmap APIs.

This allows us to extend the PFN mappings, e.g. PCI MMIO bars where
it can grow as large as 8GB or even bigger, to map at PMD / PUD level.
VFIO PCI core driver already supports fault handling at PMD / PUD level
for more efficient BAR mappings.

Signed-off-by: Ritesh Harjani (IBM) <ritesh.list@gmail.com>

Reviewed-by: Christophe Leroy (CS GROUP) <chleroy@kernel.org>

Thanks for the review!

quoted

  #define __HAVE_ARCH_PMDP_SET_ACCESS_FLAGS
  extern int pmdp_set_access_flags(struct vm_area_struct *vma,

diff --git a/arch/powerpc/include/asm/pgtable.h b/arch/powerpc/include/asm/pgtable.h
index dcd3a88caaf6..2d27cb1c2334 100644
--- a/arch/powerpc/include/asm/pgtable.h
+++ b/arch/powerpc/include/asm/pgtable.h

@@ -63,6 +63,18 @@ static inline pgprot_t pte_pgprot(pte_t pte)
  	return __pgprot(pte_flags);
  }

+#define pmd_pgprot pmd_pgprot
+static inline pgprot_t pmd_pgprot(pmd_t pmd)
+{
+	return pte_pgprot(pmd_pte(pmd));
+}
+
+#define pud_pgprot pud_pgprot
+static inline pgprot_t pud_pgprot(pud_t pud)
+{
+	return pte_pgprot(pud_pte(pud));
+}
+

In v2 - I will add above under #ifdef CONFIG_PPC_BOOK3S_64 
to avoid build issues with 32-bit PPC.

-ritesh

Ritesh Harjani (IBM) [off-list ref] writes:

"Christophe Leroy (CS GROUP)" [off-list ref] writes:

quoted

Le 27/02/2026 à 07:16, Ritesh Harjani (IBM) a écrit :

quoted

This uses _RPAGE_SW2 bit for the PMD and PUDs similar to PTEs.
This also adds support for {pte,pmd,pud}_pgprot helpers needed for
follow_pfnmap APIs.

This allows us to extend the PFN mappings, e.g. PCI MMIO bars where
it can grow as large as 8GB or even bigger, to map at PMD / PUD level.
VFIO PCI core driver already supports fault handling at PMD / PUD level
for more efficient BAR mappings.

Signed-off-by: Ritesh Harjani (IBM) <ritesh.list@gmail.com>

Reviewed-by: Christophe Leroy (CS GROUP) <chleroy@kernel.org>

Thanks for the review!

quoted

quoted

  #define __HAVE_ARCH_PMDP_SET_ACCESS_FLAGS
  extern int pmdp_set_access_flags(struct vm_area_struct *vma,

diff --git a/arch/powerpc/include/asm/pgtable.h b/arch/powerpc/include/asm/pgtable.h
index dcd3a88caaf6..2d27cb1c2334 100644
--- a/arch/powerpc/include/asm/pgtable.h
+++ b/arch/powerpc/include/asm/pgtable.h

@@ -63,6 +63,18 @@ static inline pgprot_t pte_pgprot(pte_t pte)
  	return __pgprot(pte_flags);
  }

+#define pmd_pgprot pmd_pgprot
+static inline pgprot_t pmd_pgprot(pmd_t pmd)
+{
+	return pte_pgprot(pmd_pte(pmd));
+}
+
+#define pud_pgprot pud_pgprot
+static inline pgprot_t pud_pgprot(pud_t pud)
+{
+	return pte_pgprot(pud_pte(pud));
+}
+

In v2 - I will add above under #ifdef CONFIG_PPC_BOOK3S_64 
to avoid build issues with 32-bit PPC.

On second thoughts, I am thinking maybe we should guard it with CONFIG_PPC64.  
Currently the build fails on 32-bit since no definitions of pmd_pte()
and pud_pte().  Though, we could open-code that, but I think as of
today, this only gets excercised from follow_pfnmap_start() which gates
it with VM_PFNMAP | VM_IO, which I think could only happen for THP which
is only true for book3s/64. 
But to keep the generic definitions of pXd_pgprot() and since pmd_pte()
and pud_pte() are anyways available on book3s/64 & nohash/64, so let's
just guard this with PPC64.

I will amend this change in RFC-v2 and will keep the RB from Christophe.


+#ifdef CONFIG_PPC64
+#define pmd_pgprot pmd_pgprot
+static inline pgprot_t pmd_pgprot(pmd_t pmd)
+{
+       return pte_pgprot(pmd_pte(pmd));
+}
+
+#define pud_pgprot pud_pgprot
+static inline pgprot_t pud_pgprot(pud_t pud)
+{
+       return pte_pgprot(pud_pte(pud));
+}
+#endif /* CONFIG_PPC64 */
+


-ritesh