The previous patch "mm: wrap calls to set_pte_at_notify with
invalidate_range_start and invalidate_range_end" only called the
invalidate_range_end mmu notifier function in do_wp_page when the new_page
variable wasn't NULL. This was done in order to only call invalidate_range_end
after invalidate_range_start was called. Unfortunately, there are situations
where new_page is NULL and invalidate_range_start is called. This caused
invalidate_range_start to be called without a matching invalidate_range_end,
causing kvm to loop indefinitely on the first page fault.
This patch adds a flag variable to do_wp_page that marks whether the
invalidate_range_start notifier was called. invalidate_range_end is then
called if the flag is true.
Reported-by: Jiri Slaby <jslaby@xxxxxxx>
Cc: Avi Kivity <avi@xxxxxxxxxx>
Cc: Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx>
Signed-off-by: Haggai Eran <haggaie@xxxxxxxxxxxx>
---
I tested this patch against yesterday's linux-next (next-20120918), and it
seems to solve the problem with kvm. I used the same command line you reported:
qemu-kvm -k en-us -usbdevice tablet -balloon virtio -hda IMAGE -smp 2 \
-m 1000M -net user -net nic,model=e1000 -usb -serial pty
I was hoping you could also test it yourself, and see that it also works for
you, if you don't mind.
mm/memory.c | 9 +++++----
1 file changed, 5 insertions(+), 4 deletions(-)
diff --git a/mm/memory.c b/mm/memory.c
index 1a92d87..76ec199 100644
--- a/mm/memory.c
+++ b/mm/memory.c
@@ -2529,6 +2529,7 @@ static int do_wp_page(struct mm_struct *mm, struct vm_area_struct *vma,
struct page *dirty_page = NULL;
unsigned long mmun_start; /* For mmu_notifiers */
unsigned long mmun_end; /* For mmu_notifiers */
+ bool mmun_called = false; /* For mmu_notifiers */
old_page = vm_normal_page(vma, address, orig_pte);
if (!old_page) {
@@ -2706,8 +2707,9 @@ gotten:
if (mem_cgroup_newpage_charge(new_page, mm, GFP_KERNEL))
goto oom_free_new;
- mmun_start = address & PAGE_MASK;
- mmun_end = (address & PAGE_MASK) + PAGE_SIZE;
+ mmun_start = address & PAGE_MASK;
+ mmun_end = (address & PAGE_MASK) + PAGE_SIZE;
+ mmun_called = true;
mmu_notifier_invalidate_range_start(mm, mmun_start, mmun_end);
/*
@@ -2776,8 +2778,7 @@ gotten:
page_cache_release(new_page);
unlock:
pte_unmap_unlock(page_table, ptl);
- if (new_page)
- /* Only call the end notifier if the begin was called. */
+ if (mmun_called)
mmu_notifier_invalidate_range_end(mm, mmun_start, mmun_end);
if (old_page) {
/*
--
1.7.11.2
--
To unsubscribe from this list: send the line "unsubscribe kvm" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
