On 2/21/18 11:57 AM, Alexey Dobriyan wrote:
On Tue, Feb 20, 2018 at 03:38:24PM -0800, Yang Shi wrote:

On 2/20/18 2:38 PM, Alexey Dobriyan wrote:
On Wed, Feb 21, 2018 at 03:49:29AM +0800, Yang Shi wrote:
When running vm-scalability with large memory (> 300GB), the below hung
task issue happens occasionally.

INFO: task ps:14018 blocked for more than 120 seconds.
         Tainted: G            E 4.9.79-009.ali3000.alios7.x86_64 #1
   "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message.
   ps              D    0 14018      1 0x00000004
    ffff885582f84000 ffff885e8682f000 ffff880972943000 ffff885ebf499bc0
    ffff8828ee120000 ffffc900349bfca8 ffffffff817154d0 0000000000000040
    00ffffff812f872a ffff885ebf499bc0 024000d000948300 ffff880972943000
   Call Trace:
    [<ffffffff817154d0>] ? __schedule+0x250/0x730
    [<ffffffff817159e6>] schedule+0x36/0x80
    [<ffffffff81718560>] rwsem_down_read_failed+0xf0/0x150
    [<ffffffff81390a28>] call_rwsem_down_read_failed+0x18/0x30
    [<ffffffff81717db0>] down_read+0x20/0x40
    [<ffffffff812b9439>] proc_pid_cmdline_read+0xd9/0x4e0
    [<ffffffff81253c95>] ? do_filp_open+0xa5/0x100
    [<ffffffff81241d87>] __vfs_read+0x37/0x150
    [<ffffffff812f824b>] ? security_file_permission+0x9b/0xc0
    [<ffffffff81242266>] vfs_read+0x96/0x130
    [<ffffffff812437b5>] SyS_read+0x55/0xc0
    [<ffffffff8171a6da>] entry_SYSCALL_64_fastpath+0x1a/0xc5

When manipulating a large mapping, the process may hold the mmap_sem for
long time, so reading /proc/<pid>/cmdline may be blocked in
uninterruptible state for long time.

down_read_trylock() sounds too aggressive, and we already have killable
version APIs for semaphore, here use down_read_killable() to improve the
-       down_read(&mm->mmap_sem);
+       rv = down_read_killable(&mm->mmap_sem);
+       if (rv)
+               goto out_mmput;
        arg_start = mm->arg_start;
        arg_end = mm->arg_end;
        env_start = mm->env_start;
Fix is incomplete
Yes, it is. Since I just ran into the above splat, so I just did the
minimum change.

1) /proc/*/cmdline only wants to read 4 values atomically,
     those 4 values are basically random values and aren't
     related to VM part at all (after C/R went in, they are
     settable to arbitrary values)
Sorry, I don't get your point here. Could you please elaborate?
I hoped there is some random spinlock those 4 values could be moved to
but no.

2) access_remote_vm() et al will do the same ->mmap_sem, and
Yes, it does. But, __access_remote_vm() is called by access_process_vm()
too, which is used by much more places, i.e. ptrace, so I was not sure
if it is preferred to convert to killable version. So, I leave it untouched.
Yeah, but ->mmap_sem is taken 3 times per /proc/*/cmdline read
and your scalability tests should trigger next backtrace right away.

3) /proc/*/environ and get_cmdline() do the same.
They look suitable to use killable version.

Had a look at get_cmdline() further. It looks it is not necessary to convert to killable since it is called mainly by audit_exit(), which is called when process or syscall exit, so it might be preferred to wait until the mmap_sem is released. So, it sounds pointless to use down_read_killable().


BTW, I just realized the code should go to out_free_page instead of
out_mmput. Will rectify in newer version once we decide the extra places
to use killable version.

Reply via email to