[PATCH 0/2] x86/mm: vmalloc_fault fix for CONFIG_HUGETLBFS off

[PATCH 0/2] x86/mm: vmalloc_fault fix for CONFIG_HUGETLBFS off

[Toshi Kani]

Gratian Crisan reported that vmalloc_fault() crashes when
CONFIG_HUGETLBFS is not set since the function inadvertently
uses pXn_huge(), which always return 0 in this case. [1]
ioremap() does not depend on CONFIG_HUGETLBFS.

Patch 01 fixes the issue in vmalloc_fault().
Patch 02 is a clean-up for vmalloc_fault().

[1] https://lkml.org/lkml/2018/3/8/1281

---
Toshi Kani (2):
 1/2 x86/mm: fix vmalloc_fault to use pXd_large
 2/2 x86/mm: remove pointless checks in vmalloc_fault

---
 arch/x86/mm/fault.c | 62 +++++++++++++++++------------------------------------
 1 file changed, 20 insertions(+), 42 deletions(-)

[PATCH 1/2] x86/mm: fix vmalloc_fault to use pXd_large

[Toshi Kani]

Gratian Crisan reported that vmalloc_fault() crashes when
CONFIG_HUGETLBFS is not set since the function inadvertently
uses pXn_huge(), which always return 0 in this case.  ioremap()
does not depend on CONFIG_HUGETLBFS.

Fix vmalloc_fault() to call pXd_large() instead.

fixes: f4eafd8bcd52 ("x86/mm: Fix vmalloc_fault() to handle large pages properly")
Reported-by: Gratian Crisan <gratian.crisan@ni.com>
Signed-off-by: Toshi Kani <toshi.kani@hpe.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Gratian Crisan <gratian.crisan@ni.com>
Cc: stable@vger.kernel.org
---
 arch/x86/mm/fault.c |    6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)

diff --git a/arch/x86/mm/fault.c b/arch/x86/mm/fault.c
index c88573d90f3e..25a30b5d6582 100644
--- a/arch/x86/mm/fault.c
+++ b/arch/x86/mm/fault.c
@@ -330,7 +330,7 @@ static noinline int vmalloc_fault(unsigned long address)
 	if (!pmd_k)
 		return -1;
 
-	if (pmd_huge(*pmd_k))
+	if (pmd_large(*pmd_k))
 		return 0;
 
 	pte_k = pte_offset_kernel(pmd_k, address);
@@ -475,7 +475,7 @@ static noinline int vmalloc_fault(unsigned long address)
 	if (pud_none(*pud) || pud_pfn(*pud) != pud_pfn(*pud_ref))
 		BUG();
 
-	if (pud_huge(*pud))
+	if (pud_large(*pud))
 		return 0;
 
 	pmd = pmd_offset(pud, address);
@@ -486,7 +486,7 @@ static noinline int vmalloc_fault(unsigned long address)
 	if (pmd_none(*pmd) || pmd_pfn(*pmd) != pmd_pfn(*pmd_ref))
 		BUG();
 
-	if (pmd_huge(*pmd))
+	if (pmd_large(*pmd))
 		return 0;
 
 	pte_ref = pte_offset_kernel(pmd_ref, address);

[PATCH 2/2] x86/mm: remove pointless checks in vmalloc_fault

[Toshi Kani]

vmalloc_fault() sets user's pgd or p4d from the kernel page table.
Once it's set, all tables underneath are identical. There is no point
of following the same page table with two separate pointers and makes
sure they see the same with BUG().

Remove the pointless checks in vmalloc_fault(). Also rename the kernel
pgd/p4d pointers to pgd_k/p4d_k so that their names are consistent in
the file.

Suggested-by: Andy Lutomirski <luto@kernel.org>
Signed-off-by: Toshi Kani <toshi.kani@hpe.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Gratian Crisan <gratian.crisan@ni.com>
---
 arch/x86/mm/fault.c |   56 +++++++++++++++------------------------------------
 1 file changed, 17 insertions(+), 39 deletions(-)

diff --git a/arch/x86/mm/fault.c b/arch/x86/mm/fault.c
index 25a30b5d6582..e7bc79853538 100644
--- a/arch/x86/mm/fault.c
+++ b/arch/x86/mm/fault.c
@@ -417,11 +417,11 @@ void vmalloc_sync_all(void)
  */
 static noinline int vmalloc_fault(unsigned long address)
 {
-	pgd_t *pgd, *pgd_ref;
-	p4d_t *p4d, *p4d_ref;
-	pud_t *pud, *pud_ref;
-	pmd_t *pmd, *pmd_ref;
-	pte_t *pte, *pte_ref;
+	pgd_t *pgd, *pgd_k;
+	p4d_t *p4d, *p4d_k;
+	pud_t *pud;
+	pmd_t *pmd;
+	pte_t *pte;
 
 	/* Make sure we are in vmalloc area: */
 	if (!(address >= VMALLOC_START && address < VMALLOC_END))
@@ -435,73 +435,51 @@ static noinline int vmalloc_fault(unsigned long address)
 	 * case just flush:
 	 */
 	pgd = (pgd_t *)__va(read_cr3_pa()) + pgd_index(address);
-	pgd_ref = pgd_offset_k(address);
-	if (pgd_none(*pgd_ref))
+	pgd_k = pgd_offset_k(address);
+	if (pgd_none(*pgd_k))
 		return -1;
 
 	if (CONFIG_PGTABLE_LEVELS > 4) {
 		if (pgd_none(*pgd)) {
-			set_pgd(pgd, *pgd_ref);
+			set_pgd(pgd, *pgd_k);
 			arch_flush_lazy_mmu_mode();
 		} else {
-			BUG_ON(pgd_page_vaddr(*pgd) != pgd_page_vaddr(*pgd_ref));
+			BUG_ON(pgd_page_vaddr(*pgd) != pgd_page_vaddr(*pgd_k));
 		}
 	}
 
 	/* With 4-level paging, copying happens on the p4d level. */
 	p4d = p4d_offset(pgd, address);
-	p4d_ref = p4d_offset(pgd_ref, address);
-	if (p4d_none(*p4d_ref))
+	p4d_k = p4d_offset(pgd_k, address);
+	if (p4d_none(*p4d_k))
 		return -1;
 
 	if (p4d_none(*p4d) && CONFIG_PGTABLE_LEVELS == 4) {
-		set_p4d(p4d, *p4d_ref);
+		set_p4d(p4d, *p4d_k);
 		arch_flush_lazy_mmu_mode();
 	} else {
-		BUG_ON(p4d_pfn(*p4d) != p4d_pfn(*p4d_ref));
+		BUG_ON(p4d_pfn(*p4d) != p4d_pfn(*p4d_k));
 	}
 
-	/*
-	 * Below here mismatches are bugs because these lower tables
-	 * are shared:
-	 */
 	BUILD_BUG_ON(CONFIG_PGTABLE_LEVELS < 4);
 
 	pud = pud_offset(p4d, address);
-	pud_ref = pud_offset(p4d_ref, address);
-	if (pud_none(*pud_ref))
+	if (pud_none(*pud))
 		return -1;
 
-	if (pud_none(*pud) || pud_pfn(*pud) != pud_pfn(*pud_ref))
-		BUG();
-
 	if (pud_large(*pud))
 		return 0;
 
 	pmd = pmd_offset(pud, address);
-	pmd_ref = pmd_offset(pud_ref, address);
-	if (pmd_none(*pmd_ref))
+	if (pmd_none(*pmd))
 		return -1;
 
-	if (pmd_none(*pmd) || pmd_pfn(*pmd) != pmd_pfn(*pmd_ref))
-		BUG();
-
 	if (pmd_large(*pmd))
 		return 0;
 
-	pte_ref = pte_offset_kernel(pmd_ref, address);
-	if (!pte_present(*pte_ref))
-		return -1;
-
 	pte = pte_offset_kernel(pmd, address);
-
-	/*
-	 * Don't use pte_page here, because the mappings can point
-	 * outside mem_map, and the NUMA hash lookup cannot handle
-	 * that:
-	 */
-	if (!pte_present(*pte) || pte_pfn(*pte) != pte_pfn(*pte_ref))
-		BUG();
+	if (!pte_present(*pte))
+		return -1;
 
 	return 0;
 }

Re: [PATCH 2/2] x86/mm: remove pointless checks in vmalloc_fault

[Thomas Gleixner]

On Tue, 13 Mar 2018, Toshi Kani wrote:

> vmalloc_fault() sets user's pgd or p4d from the kernel page table.
> Once it's set, all tables underneath are identical. There is no point
> of following the same page table with two separate pointers and makes
> sure they see the same with BUG().
> 
> Remove the pointless checks in vmalloc_fault(). Also rename the kernel
> pgd/p4d pointers to pgd_k/p4d_k so that their names are consistent in
> the file.

I have no idea to which branch this might apply. The first patch applies
cleanly on linus head, but this one fails in hunk #2 on everything I
tried. Can you please check?

Thanks,

	tglx

[tip:x86/urgent] x86/mm: Fix vmalloc_fault to use pXd_large

[tip-bot for Toshi Kani]

Commit-ID:  18a955219bf7d9008ce480d4451b6b8bf4483a22
Gitweb:     https://git.kernel.org/tip/18a955219bf7d9008ce480d4451b6b8bf4483a22
Author:     Toshi Kani <toshi.kani@hpe.com>
AuthorDate: Tue, 13 Mar 2018 11:03:46 -0600
Committer:  Thomas Gleixner <tglx@linutronix.de>
CommitDate: Wed, 14 Mar 2018 20:22:42 +0100

x86/mm: Fix vmalloc_fault to use pXd_large

Gratian Crisan reported that vmalloc_fault() crashes when CONFIG_HUGETLBFS
is not set since the function inadvertently uses pXn_huge(), which always
return 0 in this case.  ioremap() does not depend on CONFIG_HUGETLBFS.

Fix vmalloc_fault() to call pXd_large() instead.

Fixes: f4eafd8bcd52 ("x86/mm: Fix vmalloc_fault() to handle large pages properly")
Reported-by: Gratian Crisan <gratian.crisan@ni.com>
Signed-off-by: Toshi Kani <toshi.kani@hpe.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Cc: stable@vger.kernel.org
Cc: linux-mm@kvack.org
Cc: Borislav Petkov <bp@alien8.de>
Cc: Andy Lutomirski <luto@kernel.org>
Link: https://lkml.kernel.org/r/20180313170347.3829-2-toshi.kani@hpe.com

---
 arch/x86/mm/fault.c | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)

diff --git a/arch/x86/mm/fault.c b/arch/x86/mm/fault.c
index c88573d90f3e..25a30b5d6582 100644
--- a/arch/x86/mm/fault.c
+++ b/arch/x86/mm/fault.c
@@ -330,7 +330,7 @@ static noinline int vmalloc_fault(unsigned long address)
 	if (!pmd_k)
 		return -1;
 
-	if (pmd_huge(*pmd_k))
+	if (pmd_large(*pmd_k))
 		return 0;
 
 	pte_k = pte_offset_kernel(pmd_k, address);
@@ -475,7 +475,7 @@ static noinline int vmalloc_fault(unsigned long address)
 	if (pud_none(*pud) || pud_pfn(*pud) != pud_pfn(*pud_ref))
 		BUG();
 
-	if (pud_huge(*pud))
+	if (pud_large(*pud))
 		return 0;
 
 	pmd = pmd_offset(pud, address);
@@ -486,7 +486,7 @@ static noinline int vmalloc_fault(unsigned long address)
 	if (pmd_none(*pmd) || pmd_pfn(*pmd) != pmd_pfn(*pmd_ref))
 		BUG();
 
-	if (pmd_huge(*pmd))
+	if (pmd_large(*pmd))
 		return 0;
 
 	pte_ref = pte_offset_kernel(pmd_ref, address);

Re: [PATCH 2/2] x86/mm: remove pointless checks in vmalloc_fault

[Kani, Toshi]

On Wed, 2018-03-14 at 20:27 +0100, Thomas Gleixner wrote:
> On Tue, 13 Mar 2018, Toshi Kani wrote:
> 
> > vmalloc_fault() sets user's pgd or p4d from the kernel page table.
> > Once it's set, all tables underneath are identical. There is no point
> > of following the same page table with two separate pointers and makes
> > sure they see the same with BUG().
> > 
> > Remove the pointless checks in vmalloc_fault(). Also rename the kernel
> > pgd/p4d pointers to pgd_k/p4d_k so that their names are consistent in
> > the file.
> 
> I have no idea to which branch this might apply. The first patch applies
> cleanly on linus head, but this one fails in hunk #2 on everything I
> tried. Can you please check?

Sorry for the trouble. The patches are based on linus head. I just tried
and they applied clean to me... 

Thanks,
-Toshi

Re: [PATCH 2/2] x86/mm: remove pointless checks in vmalloc_fault

[Thomas Gleixner]

On Wed, 14 Mar 2018, Kani, Toshi wrote:
> On Wed, 2018-03-14 at 20:27 +0100, Thomas Gleixner wrote:
> > On Tue, 13 Mar 2018, Toshi Kani wrote:
> > 
> > > vmalloc_fault() sets user's pgd or p4d from the kernel page table.
> > > Once it's set, all tables underneath are identical. There is no point
> > > of following the same page table with two separate pointers and makes
> > > sure they see the same with BUG().
> > > 
> > > Remove the pointless checks in vmalloc_fault(). Also rename the kernel
> > > pgd/p4d pointers to pgd_k/p4d_k so that their names are consistent in
> > > the file.
> > 
> > I have no idea to which branch this might apply. The first patch applies
> > cleanly on linus head, but this one fails in hunk #2 on everything I
> > tried. Can you please check?
> 
> Sorry for the trouble. The patches are based on linus head. I just tried
> and they applied clean to me... 

Hmm. Looks like I tried on the wrong branch. Nevertheless, 1/2 is queued in
urgent, but 2/2 will go through tip/x86/mm which already has changes in
that area causing the patch to fail. I just merged x86/urgent into x86/mm
and pushed it out. Can you please rebase 2/2 on top of that bracnh and
resend ?

Thanks,

	tglx

Re: [PATCH 2/2] x86/mm: remove pointless checks in vmalloc_fault

[Kani, Toshi]

On Wed, 2018-03-14 at 20:56 +0100, Thomas Gleixner wrote:
> On Wed, 14 Mar 2018, Kani, Toshi wrote:
> > On Wed, 2018-03-14 at 20:27 +0100, Thomas Gleixner wrote:
> > > On Tue, 13 Mar 2018, Toshi Kani wrote:
> > > 
> > > > vmalloc_fault() sets user's pgd or p4d from the kernel page table.
> > > > Once it's set, all tables underneath are identical. There is no point
> > > > of following the same page table with two separate pointers and makes
> > > > sure they see the same with BUG().
> > > > 
> > > > Remove the pointless checks in vmalloc_fault(). Also rename the kernel
> > > > pgd/p4d pointers to pgd_k/p4d_k so that their names are consistent in
> > > > the file.
> > > 
> > > I have no idea to which branch this might apply. The first patch applies
> > > cleanly on linus head, but this one fails in hunk #2 on everything I
> > > tried. Can you please check?
> > 
> > Sorry for the trouble. The patches are based on linus head. I just tried
> > and they applied clean to me... 
> 
> Hmm. Looks like I tried on the wrong branch. Nevertheless, 1/2 is queued in
> urgent, but 2/2 will go through tip/x86/mm which already has changes in
> that area causing the patch to fail. I just merged x86/urgent into x86/mm
> and pushed it out. Can you please rebase 2/2 on top of that bracnh and
> resend ?

Yes, I will merge up 2/2 and resend.

Thanks,
-Toshi