RT-V0.7.22 Bug with fbdev and e100

RT-V0.7.22 Bug with fbdev and e100

[Amit Shah]

Hi Ingo,

I got this:

ifconfig/4106: BUG in enable_irq at kernel/irq/manage.c:112
BUG: sleeping function called from invalid context ifconfig(4106) at 
kernel/rt.c:1320
in_atomic():1 [00000001], irqs_disabled():0
 [<c0108047>] dump_stack+0x1e/0x22 (20)
 [<c0120f35>] __might_sleep+0xba/0xc9 (36)
 [<c013d07a>] __spin_lock+0x33/0x52 (24)
 [<c013d0b1>] _spin_lock+0x18/0x1c (16)
 [<c01501ea>] kmem_cache_alloc+0x40/0xff (32)
 [<c01f48dc>] soft_cursor+0x50/0x23c (80)
 [<c01efa57>] bit_cursor+0x297/0x474 (160)
 [<c01eb73f>] fbcon_cursor+0x1af/0x273 (68)
 [<c022752b>] hide_cursor+0x2c/0x41 (20)
 [<c022abbb>] vt_console_print+0x307/0x316 (52)
 [<c0124ada>] __call_console_drivers+0x5e/0x60 (32)
 [<c0124c0b>] call_console_drivers+0xb3/0x154 (40)
 [<c0124fb3>] release_console_sem+0x52/0xe3 (32)
 [<c0124eb2>] vprintk+0x119/0x160 (36)
 [<c0124d95>] printk+0x18/0x1c (16)
 [<c014508a>] enable_irq+0xe6/0xf0 (52)
 [<f8c2b370>] e100_up+0x111/0x21f [e100] (48)
 [<f8c2c5d8>] e100_open+0x2c/0x71 [e100] (32)
 [<c027a4f5>] dev_open+0x78/0x86 (28)
 [<c027bc4a>] dev_change_flags+0x56/0x127 (36)
 [<c02b8c60>] devinet_ioctl+0x242/0x5bc (104)
 [<c02badc6>] inet_ioctl+0x5a/0x9a (28)
 [<c027131c>] sock_ioctl+0xbb/0x21f (32)
 [<c0179ef7>] sys_ioctl+0xdc/0x241 (44)
 [<c0107177>] syscall_call+0x7/0xb (-8124)
---------------------------
| preempt count: 00000002 ]
| 2-level deep critical section nesting:
----------------------------------------
.. [<c02cc920>] .... _raw_spin_lock_irqsave+0x1d/0x82
.....[<c0144fcc>] ..   ( <= enable_irq+0x28/0xf0)
.. [<c013ddd0>] .... print_traces+0x18/0x4c
.....[<c0108047>] ..   ( <= dump_stack+0x1e/0x22)

 [<c0108047>] dump_stack+0x1e/0x22 (20)
 [<c014508f>] enable_irq+0xeb/0xf0 (52)
 [<f8c2b370>] e100_up+0x111/0x21f [e100] (48)
 [<f8c2c5d8>] e100_open+0x2c/0x71 [e100] (32)
 [<c027a4f5>] dev_open+0x78/0x86 (28)
 [<c027bc4a>] dev_change_flags+0x56/0x127 (36)
 [<c02b8c60>] devinet_ioctl+0x242/0x5bc (104)
 [<c02badc6>] inet_ioctl+0x5a/0x9a (28)
 [<c027131c>] sock_ioctl+0xbb/0x21f (32)
 [<c0179ef7>] sys_ioctl+0xdc/0x241 (44)
 [<c0107177>] syscall_call+0x7/0xb (-8124)
---------------------------
| preempt count: 00000002 ]
| 2-level deep critical section nesting:
----------------------------------------
.. [<c02cc920>] .... _raw_spin_lock_irqsave+0x1d/0x82
.....[<c0144fcc>] ..   ( <= enable_irq+0x28/0xf0)
.. [<c013ddd0>] .... print_traces+0x18/0x4c
.....[<c0108047>] ..   ( <= dump_stack+0x1e/0x22)

IRQ#17 thread RT prio: 44.
e1000: eth0: e1000_watchdog: NIC Link is Up 100 Mbps Full Duplex
IRQ#3 thread RT prio: 43.
IRQ#8 thread RT prio: 42.


drivers/video/console/fbcon.c:fbcon_cursor needs additional locking?

Amit.
-- 
Amit Shah
Codito Technologies Pvt. Ltd.

Re: RT-V0.7.22 Bug with fbdev and e100

[Ingo Molnar]

* Amit Shah <amit.shah@codito.com> wrote:

> drivers/video/console/fbcon.c:fbcon_cursor needs additional locking?

this issue is a bit different. The problem is that the irq warning
message (which is harmless) is printed from an atomic context. In the
next release i'll try something different: simply dont print to the
fbcon if the kernel is in an atomic context. This will reduce the
utility of fbcon when debugging kernel crashes, but it should avoid this
assert (and potential problems resulting out of it). Other types of
consoles (serial, text, netconsole) will still try to output everything,
independently of atomicity. (of these, only netconsole wants to
reschedule.)

	Ingo

Re: RT-V0.7.22 Bug with fbdev and e100

[Ingo Molnar]

* Ingo Molnar <mingo@elte.hu> wrote:

> [...] This will reduce the utility of fbcon when debugging kernel
> crashes, but it should avoid this assert [...]

in fact the way i implemented it in my tree:

#define in_atomic_rt()  (!oops_in_progress && (in_atomic() || irqs_disabled()))

still enables crash messages to make it to fbcon. So the only thing
skipped will be non-fatal messages printed from 'raw' critical sections.
(which are very rare in PREEMPT_RT kernels).

	Ingo