Scheduling while atomic in tty code with cdc-acm driver, kernel 2.6.29.1

Scheduling while atomic in tty code with cdc-acm driver, kernel 2.6.29.1

[Chuck Ebbert]

BUG: sleeping function called from invalid context at kernel/mutex.c:88
in_atomic(): 1, irqs_disabled(): 0, pid: 0, name: swapper
Pid: 0, comm: swapper Not tainted 2.6.29.1-54.fc11.x86_64 #1
Call Trace:
 <IRQ>  [<ffffffff8104255c>] __might_sleep+0x10b/0x110
 [<ffffffff813a8f2c>] mutex_lock+0x22/0x38
 [<ffffffff8103f597>] ? enqueue_task_fair+0x2a/0x6d
 [<ffffffff8122f794>] echo_char+0x25/0x81
 [<ffffffff81230e03>] n_tty_receive_buf+0x309/0xe50
 [<ffffffff8103be38>] ? task_rq_unlock+0x11/0x13
 [<ffffffff81042d5c>] ? default_wake_function+0x12/0x14
 [<ffffffff8103a1e0>] ? __wake_up_common+0x4e/0x84
 [<ffffffff8104fe2d>] ? __tasklet_schedule+0x77/0x8e
 [<ffffffff8102af47>] ? default_spin_lock_flags+0x9/0xe
 [<ffffffff81232a48>] ? tty_ldisc_try+0x3c/0x44
 [<ffffffff81233781>] flush_to_ldisc+0x10c/0x19f
 [<ffffffff81233864>] tty_flip_buffer_push+0x50/0x61
 [<ffffffffa03ab6ef>] acm_rx_tasklet+0x10c/0x28c [cdc_acm]
 [<ffffffff81050025>] tasklet_action+0x8e/0xf8
 [<ffffffff81050803>] __do_softirq+0x94/0x179
 [<ffffffff810127cc>] call_softirq+0x1c/0x30
 [<ffffffff8101394e>] do_softirq+0x52/0xb9
 [<ffffffff81050426>] irq_exit+0x53/0x90
 [<ffffffff81013c77>] do_IRQ+0x12c/0x151
 [<ffffffff81011f13>] ret_from_intr+0x0/0x2e
 <EOI>  [<ffffffff812187bd>] ? acpi_idle_enter_bm+0x28f/0x2de
 [<ffffffff812187b5>] ? acpi_idle_enter_bm+0x287/0x2de
 [<ffffffff812ee957>] ? cpuidle_idle_call+0x8d/0xc2
 [<ffffffff810102c7>] ? cpu_idle+0x68/0xb3
 [<ffffffff813a435c>] ? start_secondary+0x199/0x19e



Partial trace by kernel line numbers:

drivers/char/n_tty.c:828

drivers/char/n_tty.c:1160

drivers/char/tty_buffer.c:444

drivers/char/tty_buffer.c:486
  if (tty->low_latency)
    flush_to_ldisc(&tty->buf.work.work);  

Re: Scheduling while atomic in tty code with cdc-acm driver, kernel 2.6.29.1

[Oliver Neukum]

Am Dienstag 14 April 2009 01:36:33 schrieb Chuck Ebbert:
> BUG: sleeping function called from invalid context at kernel/mutex.c:88
> in_atomic(): 1, irqs_disabled(): 0, pid: 0, name: swapper
> Pid: 0, comm: swapper Not tainted 2.6.29.1-54.fc11.x86_64 #1

Is this specific to 2.6.29.1? It looks like setting low_latency, as cdc-acm,
does for legacy reasons, is harmful now. Please remove these lines:


	/* force low_latency on so that our tty_push actually forces the data 
through,
	   otherwise it is scheduled, and with high data rates data can get lost. */
	tty->low_latency = 1;

	HTH
		Oliver

Re: Scheduling while atomic in tty code with cdc-acm driver, kernel 2.6.29.1

[Alan Cox]

On Tue, 14 Apr 2009 01:50:10 +0200
Oliver Neukum <oliver@neukum.org> wrote:

> Am Dienstag 14 April 2009 01:36:33 schrieb Chuck Ebbert:
> > BUG: sleeping function called from invalid context at kernel/mutex.c:88
> > in_atomic(): 1, irqs_disabled(): 0, pid: 0, name: swapper
> > Pid: 0, comm: swapper Not tainted 2.6.29.1-54.fc11.x86_64 #1
> 
> Is this specific to 2.6.29.1? It looks like setting low_latency, as cdc-acm,
> does for legacy reasons, is harmful now. Please remove these lines:

The only thing that has changed is that this incorrect behaviour (setting
low_latency and doing a flip_buffer_push() in IRQ context) now causes a
warn rather than random failure.

The hack setting of it for high data rate devices was obsolete when the
tty buffer changes went in so it can definitely go in those cases.