linux - 内核挂tty子系统

Question

我在 RHEL 机器上遇到 tty 子系统的一些问题。从我在日志中看到的，每次生成新控制台(无论是 pts 还是 tty)时都会生成一些内核 oops。在我看来，那里似乎存在某种竞争条件。这是堆栈跟踪:

kernel:  INFO: task sshd:6338 blocked for more than 120 seconds.
kernel:       Tainted: P           ---------------    2.6.32-504.el6.x86_64 #1
kernel:  "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message.
kernel:  sshd          D 0000000000000000     0  6338   6195 0x00000080
kernel:  ffff88035be8d728 0000000000000082 0000000000000000 0000000000000000
kernel:  ffff88035be8d7f8 ffffffff8105ca34 00009488ef033e83 ffff88035be8d708
kernel:  ffff88035be8d880 0000000109b91c98 ffff881eea341098 ffff88035be8dfd8
kernel:  Call Trace:
kernel:  [<ffffffff8105ca34>] ? find_busiest_group+0x244/0x9e0
kernel:  [<ffffffff8152a8c5>] schedule_timeout+0x215/0x2e0
kernel:  [<ffffffff8152a543>] wait_for_common+0x123/0x180
kernel:  [<ffffffff81064b90>] ? default_wake_function+0x0/0x20
kernel:  [<ffffffff8152a65d>] wait_for_completion+0x1d/0x20
kernel:  [<ffffffff81098bf7>] flush_work+0x77/0xc0
kernel:  [<ffffffff81098460>] ? wq_barrier_func+0x0/0x20
kernel:  [<ffffffff81098e14>] flush_delayed_work+0x54/0x70
kernel:  [<ffffffff813392f5>] tty_flush_to_ldisc+0x15/0x20
kernel:  [<ffffffff81333cc7>] n_tty_poll+0x67/0x1d0
kernel:  [<ffffffff8132f80a>] tty_poll+0x8a/0xa0
kernel:  [<ffffffff811a6895>] do_select+0x3c5/0x7c0
kernel:  [<ffffffff8149cf18>] ? ip_finish_output+0x148/0x310
kernel:  [<ffffffff811a59f0>] ? __pollwait+0x0/0xf0
kernel:  [<ffffffff811a5ae0>] ? pollwake+0x0/0x60
kernel:  [<ffffffff811a5ae0>] ? pollwake+0x0/0x60
kernel:  [<ffffffff811a5ae0>] ? pollwake+0x0/0x60
kernel:  [<ffffffff811a5ae0>] ? pollwake+0x0/0x60
kernel:  [<ffffffff8152d04b>] ? _spin_unlock_bh+0x1b/0x20
kernel:  [<ffffffff8144b835>] ? release_sock+0xe5/0x110
kernel:  [<ffffffff814a52cc>] ? tcp_sendmsg+0x73c/0xa20
kernel:  [<ffffffff8144a72b>] ? sock_aio_write+0x19b/0x1c0
kernel:  [<ffffffff8133158d>] ? tty_wakeup+0x3d/0x80
kernel:  [<ffffffff811a6e1a>] core_sys_select+0x18a/0x2c0
kernel:  [<ffffffff8109eb00>] ? autoremove_wake_function+0x0/0x40
kernel:  [<ffffffff811a71a7>] sys_select+0x47/0x110
kernel:  [<ffffffff810e5c87>] ? audit_syscall_entry+0x1d7/0x200
kernel:  [<ffffffff810e5a7e>] ? __audit_syscall_exit+0x25e/0x290
kernel:  [<ffffffff8100b072>] system_call_fastpath+0x16/0x1b

因此，查看最后 2 个函数调用，似乎任务通过 schedule_timeout() 安排休眠一段时间，之后 find_busiest_group 尝试平衡该任务产生的负载。它是正确的还是我在这里遗漏了什么？

谢谢。

Answer 1

如果有人感兴趣，我向 RedHat 提交了一个案例，问题似乎与阵列 Controller (hpsa) 中的 HP 固件错误有关。更多详情:https://access.redhat.com/solutions/1179703

**> 我查看了 bugzilla，发现该错误可能与

Smart Array firmware revision.

The bug was reproduced on system with below combination of controller version and hpsa module version. (New hpsa module version and old firmware version)

• kmod-hpsa: 3.4.4-1-RH1
• SA Firmware: 3.22
• Controller: P220i (103c:323b 103c:3355 rev 01)

The system running new hpsa module version and new firmware version did not reproduced this bug.

• kmod-hpsa: 3.4.4-1-RH1
• SA Firmware: 3.42
• Controller: P220i (103c:323b 103c:3355 rev 01)

The system running old hpsa module version and old firmware version also did not reproduced this bug.

• kmod-hpsa: 3.4.0-1-RH1
• SA Firmware: 3.22
• Controller: P220i (103c:323b 103c:3355 rev 01)

In our case the controller firmware version is 3.22 and we were using new hpsa module version 3.4.4-1-RH2.

$ cat proc/scsi/scsi | grep -A 5 P220i Vendor: HP Model: P220i Rev: 3.22 Type: RAID
ANSI SCSI revision: 05

Now I see that with old kernel we are using old version of hpsa module (3.4.0-1-RH1). With this the system should not encounter this bug.

modinfo hpsa filename: /lib/modules/2.6.32-431.23.3.el6.x86_64/kernel/drivers/scsi/hpsa.ko

license: GPL version: 3.4.0-1-RH1 description: Driver for HP Smart Array Controller version 3.4.0-1-RH1 author:
Hewlett-Packard Company**

这是 RedHat 工程师的说法。