Skip to contentSkip to conversation

xen: Check NX state before setting up GDT mapping

GIT PULL Jeremy Fitzhardinge 2 participants 5 messages Sep 21, 2009 – Sep 23, 2009

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

Hi Ingo,

This is a little followup fix to the previous Xen stack-protector patches to deal with the case of a 64-bit processor without NX support (either old Intel or disabled in BIOS). Before creating the mapping for the GDT we need to make sure that NX is actually supported or the processor will fault on us.

This ends up making mm/init.o non-stack-protected. Given that this file has some fairly complex functions with local arrays, I've made an additional change to move the NX setup into its own file to limit the non-stack-protected code. I added it as a later delta so that the pure bugfix can be applied to stable, and the cleanup can go on top into mainline.

Thanks,
J

The following changes since commit 79b520e87e1214cfa107bdc8528b5d6c055a8b82:
Linus Torvalds (1):
Merge branch 'for-linus' of git://oss.sgi.com/xfs/xfs

are available in the git repository at:

git://git.kernel.org/pub/scm/linux/kernel/git/jeremy/xen.git bugfix

Jeremy Fitzhardinge (2):
xen: check EFER for NX before setting up GDT mapping
x86: split NX setup into separate file to limit unstack-protected code

 arch/x86/include/asm/pgtable_types.h |    1 +
 arch/x86/mm/Makefile                 |    3 +-
 arch/x86/mm/init.c                   |   63 -------------------------------
 arch/x86/mm/setup_nx.c               |   69 ++++++++++++++++++++++++++++++++++
 arch/x86/xen/enlighten.c             |   10 ++--
 5 files changed, 77 insertions(+), 69 deletions(-)
 create mode 100644 arch/x86/mm/setup_nx.c
arch/x86/include/asm/pgtable_types.h +1 -0
index 54cb697..e9918d9 100644
@@ -277,6 +277,7 @@ static inline pteval_t pte_flags(pte_t pte)
 typedef struct page *pgtable_t;
 
 extern pteval_t __supported_pte_mask;
+extern void set_nx(void);
 extern int nx_enabled;
 
 #define pgprot_writecombine	pgprot_writecombine
arch/x86/mm/Makefile +2 -1
index 9b5a9f5..06630d2 100644
@@ -1,9 +1,10 @@
 obj-y	:=  init.o init_$(BITS).o fault.o ioremap.o extable.o pageattr.o mmap.o \
-	    pat.o pgtable.o physaddr.o gup.o
+	    pat.o pgtable.o physaddr.o gup.o setup_nx.o
 
 # Make sure __phys_addr has no stackprotector
 nostackp := $(call cc-option, -fno-stack-protector)
 CFLAGS_physaddr.o		:= $(nostackp)
+CFLAGS_setup_nx.o		:= $(nostackp)
 
 obj-$(CONFIG_SMP)		+= tlb.o
 
arch/x86/mm/init.c +0 -63

69 lines not on this page · raw message

arch/x86/mm/setup_nx.c new file +69 -0

69 lines not on this page · raw message

arch/x86/xen/enlighten.c +5 -5
index 0dd0c2c..5d701bf 100644
@@ -1083,6 +1083,11 @@ asmlinkage void __init xen_start_kernel(void)
 
 	__supported_pte_mask |= _PAGE_IOMAP;
 
+#ifdef CONFIG_X86_64
+	/* Work out if we support NX */
+	check_efer();
+#endif
+
 	xen_setup_features();
 
 	/* Get mfn list */
@@ -1123,11 +1128,6 @@ asmlinkage void __init xen_start_kernel(void)
 
 	pgd = (pgd_t *)xen_start_info->pt_base;
 
-#ifdef CONFIG_X86_64
-	/* Work out if we support NX */
-	check_efer();
-#endif
-
 	/* Don't do the full vcpu_info placement stuff until we have a
 	   possible map and a non-dummy shared_info. */
 	per_cpu(xen_vcpu, 0) = &HYPERVISOR_shared_info->vcpu_info[0];
#1 lore raw
Quoted 30 lines · Jeremy Fitzhardinge

Hi Ingo,

This is a little followup fix to the previous Xen stack-protector patches to deal with the case of a 64-bit processor without NX support (either old Intel or disabled in BIOS). Before creating the mapping for the GDT we need to make sure that NX is actually supported or the processor will fault on us.

This ends up making mm/init.o non-stack-protected. Given that this file has some fairly complex functions with local arrays, I've made an additional change to move the NX setup into its own file to limit the non-stack-protected code. I added it as a later delta so that the pure bugfix can be applied to stable, and the cleanup can go on top into mainline.

Thanks,
J

The following changes since commit 79b520e87e1214cfa107bdc8528b5d6c055a8b82:
Linus Torvalds (1):
Merge branch 'for-linus' of git://oss.sgi.com/xfs/xfs

are available in the git repository at:

git://git.kernel.org/pub/scm/linux/kernel/git/jeremy/xen.git bugfix

Jeremy Fitzhardinge (2):
xen: check EFER for NX before setting up GDT mapping
x86: split NX setup into separate file to limit unstack-protected code

Quoted 7 lines
 arch/x86/include/asm/pgtable_types.h |    1 +
 arch/x86/mm/Makefile                 |    3 +-
 arch/x86/mm/init.c                   |   63 -------------------------------
 arch/x86/mm/setup_nx.c               |   69 ++++++++++++++++++++++++++++++++++
 arch/x86/xen/enlighten.c             |   10 ++--
 5 files changed, 77 insertions(+), 69 deletions(-)
 create mode 100644 arch/x86/mm/setup_nx.c

Pulled, thanks Jeremy!

Ingo

#2 lore raw
In reply to Jeremy Fitzhardinge

arch/x86/xen/enlighten.c asmlinkage void __init xen_start_kernel(void) · line 1090

Quoted 13 lines · Jeremy Fitzhardinge
arch/x86/xen/enlighten.c +5 -0
index 0dd0c2c..5d701bf 100644
@@ -1083,6 +1083,11 @@ asmlinkage void __init xen_start_kernel(void)
 
 	__supported_pte_mask |= _PAGE_IOMAP;
 
+#ifdef CONFIG_X86_64
+	/* Work out if we support NX */
+	check_efer();
+#endif
+

btw., this #ifdef could sure be eliminated, right?

Ingo

#3 lore raw
Quoted 8 lines · Ingo Molnar

arch/x86/xen/enlighten.c asmlinkage void __init xen_start_kernel(void) · line 1090

Quoted 6 lines
arch/x86/xen/enlighten.c +5 -0
+#ifdef CONFIG_X86_64
+	/* Work out if we support NX */
+	check_efer();
+#endif
+

btw., this #ifdef could sure be eliminated, right?

check_efer() is only defined on x86-64 and all its existing callsites have a similar #ifdef. NX setup is an area which could do with a bit of hitting with the unification stick.

J

#4 lore raw
Quoted 10 lines · Jeremy Fitzhardinge
Quoted 8 lines · Ingo Molnar

arch/x86/xen/enlighten.c asmlinkage void __init xen_start_kernel(void) · line 1090

Quoted 6 lines
arch/x86/xen/enlighten.c +5 -0
+#ifdef CONFIG_X86_64
+	/* Work out if we support NX */
+	check_efer();
+#endif
+

btw., this #ifdef could sure be eliminated, right?

Quoted 2 lines

check_efer() is only defined on x86-64 and all its existing callsites have a similar #ifdef.

Yes, a familar pattern that prompted my comment.

Quoted 2 lines

[...] NX setup is an area which could do with a bit of hitting with the unification stick.

Yes.

Ingo