On Fri, 2018-09-14 at 17:01 +0800, kernel test robot wrote:
FYI, we noticed the following commit (built with gcc-4.9):
Thanks,
[..]
[ 245.080051] INFO: task lkp-setup-rootf:500 blocked for more than 120 seconds. [ 245.082850] Tainted: G W T 4.19.0-rc3-00014- gf26eb68 #1 [ 245.084673] "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message. [ 245.086455] lkp-setup-rootf D 6516 500 498 0x00000000 [ 245.087766] Call Trace: [ 245.088391] __schedule+0x4b1/0xe80 [ 245.089322] schedule+0x25/0x60 [ 245.090178] schedule_timeout+0x275/0x4e0 [ 245.091105] ? __ldsem_down_write_nested+0x1aa/0x300 [ 245.092217] __ldsem_down_write_nested+0x1b2/0x300 [ 245.093286] ? ldsem_down_write+0x2a/0x2e [ 245.094305] ldsem_down_write+0x2a/0x2e [ 245.095211] tty_ldisc_lock+0x16/0x40 [ 245.096104] tty_reopen+0x3f/0xc0 [ 245.096892] tty_open+0x3cf/0x450 [ 245.097718] chrdev_open+0x9c/0x1f0 [ 245.098511] ? security_file_open+0x99/0xa0 [ 245.099512] do_dentry_open+0x3d5/0x5d0 [ 245.100428] ? inode_permission+0x95/0x180 [ 245.101351] ? cdev_put+0x30/0x30 [ 245.102138] vfs_open+0x2d/0x30 [ 245.102898] path_openat+0x9e1/0x17d0 [ 245.103733] do_filp_open+0x6a/0x120 [ 245.104541] ? _raw_spin_unlock+0x22/0x30 [ 245.105439] ? __alloc_fd+0xa5/0x170 [ 245.106309] do_sys_open+0x13a/0x250 [ 245.107118] sys_open+0x22/0x30 [ 245.107980] do_int80_syscall_32+0x94/0x1c0 [ 245.108982] entry_INT80_32+0xf3/0xf3
So, unfortunately, I'll still have to set timeout to 5 sec for tty_reopen(). The better fix would be to release read lock whenever a writer is pending (say n_tty_receive_buf_common() can fallback), but that's probably better to put in another patches set, as this one is already packed. Annoyingly, it means that open(tty) will have now more chances to fail then before.. But still it's better than a panic with null-ptr deref.