fanghuilin / rpms / kernel

Forked from rpms/kernel 3 years ago
Clone
e293be
centosplus kernel patch [bug#15216]
e293be
e293be
commit 89da619bc18d79bca5304724c11d4ba3b67ce2c6
e293be
Author: Jiang Biao <jiang.biao2@zte.com.cn>
e293be
Date:   Wed Jul 18 10:29:28 2018 +0800
e293be
e293be
    virtio_balloon: fix another race between migration and ballooning
e293be
    
e293be
    Kernel panic when with high memory pressure, calltrace looks like,
e293be
    
e293be
    PID: 21439 TASK: ffff881be3afedd0 CPU: 16 COMMAND: "java"
e293be
     #0 [ffff881ec7ed7630] machine_kexec at ffffffff81059beb
e293be
     #1 [ffff881ec7ed7690] __crash_kexec at ffffffff81105942
e293be
     #2 [ffff881ec7ed7760] crash_kexec at ffffffff81105a30
e293be
     #3 [ffff881ec7ed7778] oops_end at ffffffff816902c8
e293be
     #4 [ffff881ec7ed77a0] no_context at ffffffff8167ff46
e293be
     #5 [ffff881ec7ed77f0] __bad_area_nosemaphore at ffffffff8167ffdc
e293be
     #6 [ffff881ec7ed7838] __node_set at ffffffff81680300
e293be
     #7 [ffff881ec7ed7860] __do_page_fault at ffffffff8169320f
e293be
     #8 [ffff881ec7ed78c0] do_page_fault at ffffffff816932b5
e293be
     #9 [ffff881ec7ed78f0] page_fault at ffffffff8168f4c8
e293be
        [exception RIP: _raw_spin_lock_irqsave+47]
e293be
        RIP: ffffffff8168edef RSP: ffff881ec7ed79a8 RFLAGS: 00010046
e293be
        RAX: 0000000000000246 RBX: ffffea0019740d00 RCX: ffff881ec7ed7fd8
e293be
        RDX: 0000000000020000 RSI: 0000000000000016 RDI: 0000000000000008
e293be
        RBP: ffff881ec7ed79a8 R8: 0000000000000246 R9: 000000000001a098
e293be
        R10: ffff88107ffda000 R11: 0000000000000000 R12: 0000000000000000
e293be
        R13: 0000000000000008 R14: ffff881ec7ed7a80 R15: ffff881be3afedd0
e293be
        ORIG_RAX: ffffffffffffffff CS: 0010 SS: 0018
e293be
    
e293be
    It happens in the pagefault and results in double pagefault
e293be
    during compacting pages when memory allocation fails.
e293be
    
e293be
    Analysed the vmcore, the page leads to second pagefault is corrupted
e293be
    with _mapcount=-256, but private=0.
e293be
    
e293be
    It's caused by the race between migration and ballooning, and lock
e293be
    missing in virtballoon_migratepage() of virtio_balloon driver.
e293be
    This patch fix the bug.
e293be
    
e293be
    Fixes: e22504296d4f64f ("virtio_balloon: introduce migration primitives to balloon pages")
e293be
    Cc: stable@vger.kernel.org
e293be
    Signed-off-by: Jiang Biao <jiang.biao2@zte.com.cn>
e293be
    Signed-off-by: Huang Chong <huang.chong@zte.com.cn>
e293be
    Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
e293be
e293be
    Applied-by: Akemi Yagi <toracat@centos.org>
e293be
e293be
diff --git a/drivers/virtio/virtio_balloon.c b/drivers/virtio/virtio_balloon.c
e293be
index 6b237e3..3988c09 100644
e293be
--- a/drivers/virtio/virtio_balloon.c
e293be
+++ b/drivers/virtio/virtio_balloon.c
e293be
@@ -513,7 +513,9 @@ static int virtballoon_migratepage(struct balloon_dev_info *vb_dev_info,
e293be
 	tell_host(vb, vb->inflate_vq);
e293be
 
e293be
 	/* balloon's page migration 2nd step -- deflate "page" */
e293be
+	spin_lock_irqsave(&vb_dev_info->pages_lock, flags);
e293be
 	balloon_page_delete(page);
e293be
+	spin_unlock_irqrestore(&vb_dev_info->pages_lock, flags);
e293be
 	vb->num_pfns = VIRTIO_BALLOON_PAGES_PER_PAGE;
e293be
 	set_page_pfns(vb, vb->pfns, page);
e293be
 	tell_host(vb, vb->deflate_vq);