From: Luming Yu <luyu@redhat.com> Date: Fri, 9 Nov 2007 14:02:44 +0800 Subject: [ia64] mm: register backing store bug Message-id: 4733F804.1060906@redhat.com O-Subject: [RHEL 5.2 PATCH] bz310801: Register backing store bug Bugzilla: 310801 bz310801 Description of problem: A bug in the register backing store code can interact badly with the address space randomisation feature to cause applications to seg-fault. This is fixed by the attached patch, which has already been submitted back to the mainline kernel and is currently included in the 2.6.23-rc8 prepatch. Version-Release number of selected component (if applicable): Any IA64 Linux RH version could be effected, though it's unlikely that this issue will be noticed until the address space randomisation is turned on. Upstream status: It is upstream. http://git.kernel.org/?p=linux/kernel/git/torvalds/linux-2.6.git;a=commit;h=e8c59c0cf9c91dccfb6367c306d753500d5a0150 [IA64] Failure to grow RBS There is a bug in the ia64_do_page_fault code that can cause a failure to grow the register backing store, or any mapping that is marked as VM_GROWSUP if the mapping is the highest mapped area of memory. When the address accessed is below the first mapping the previous mapping is returned as NULL, and this case is handled. However, when the address accessed is above the highest mapping the vma returned is NULL, this case is not handled correctly, and it fails to spot that this access might require an existing mapping to grow upwards. Signed-off-by: Andrew Burgess <andrew@transitive.com> Signed-off-by: Tony Luck <tony.luck@intel.com> The following back port has been tested by me. I don't see any problem with this patch so far. Management status: The pm_ack, devel_ack, and qa_ack are __not__ set on the bugzilla. Please review, test and AC. Thanks, Luming Acked-by: Prarit Bhargava <prarit@redhat.com> Acked-by: Jon Masters <jcm@redhat.com> Acked-by: Jarod Wilson <jwilson@redhat.com> diff --git a/arch/ia64/mm/fault.c b/arch/ia64/mm/fault.c index 14ef7cc..42bc87c 100644 --- a/arch/ia64/mm/fault.c +++ b/arch/ia64/mm/fault.c @@ -125,11 +125,17 @@ ia64_do_page_fault (unsigned long address, unsigned long isr, struct pt_regs *re down_read(&mm->mmap_sem); vma = find_vma_prev(mm, address, &prev_vma); - if (!vma) + if (!vma && !prev_vma ) goto bad_area; - /* find_vma_prev() returns vma such that address < vma->vm_end or NULL */ - if (address < vma->vm_start) + /* + * find_vma_prev() returns vma such that address < vma->vm_end or NULL + * + * May find no vma, but could be that the last vm area is the + * register backing store that needs to expand upwards, in + * this case vma will be null, but prev_vma will ne non-null + */ + if (( !vma && prev_vma ) || (address < vma->vm_start) ) goto check_expansion; good_area: @@ -184,6 +190,8 @@ ia64_do_page_fault (unsigned long address, unsigned long isr, struct pt_regs *re check_expansion: if (!(prev_vma && (prev_vma->vm_flags & VM_GROWSUP) && (address == prev_vma->vm_end))) { + if (!vma) + goto bad_area; if (!(vma->vm_flags & VM_GROWSDOWN)) goto bad_area; if (REGION_NUMBER(address) != REGION_NUMBER(vma->vm_start)