RE: e100 badness (2.6.10-rc1-mm2)

Randy:

Thanks for the information. We are aware of this (from a previous
conversation between Scott Feldman, Andrew Morton and Jeremy
Fitzhardinge). We are working on a fix for this.

Thanks,
Ganesh.

-----Original Message-----
From: Randy.Dunlap [mailto:rddunlap@osdl.org]
Sent: Monday, November 01, 2004 8:24 PM
To: Venkatesan, Ganesh; Brandeburg, Jesse; netdev@oss.sgi.com
Subject: e100 badness (2.6.10-rc1-mm2)

Hi Ganesh, Jesse --

I see this in 2.6.10-rc1-mm2:

Badness in enable_irq at kernel/irq/manage.c:112
  [<c0106e09>] dump_stack+0x1e/0x20
  [<c013f7ab>] enable_irq+0xaa/0x114
  [<c02b0498>] e100_up+0x130/0x23f
  [<c02b164a>] e100_open+0x26/0x6e
  [<c0396896>] dev_open+0x6e/0x7c
  [<c0397e05>] dev_change_flags+0x56/0x126
  [<c03d2426>] devinet_ioctl+0x60b/0x6cd
  [<c03d4261>] inet_ioctl+0x81/0xae
  [<c038ddbc>] sock_ioctl+0x1d3/0x2d6
  [<c0172970>] sys_ioctl+0x179/0x21d
  [<c0105f5d>] sysenter_past_esp+0x52/0x71


I think that it happens because (in e100_up()),
disable_irq() is called before request_irq() is called.
If there are no other interrupt handlers on the same irq,
then desc->depth is cleared to 0 by setup_irq() [called
from request_irq()].
Then the following enable_irq() is confused by
desc->depth == 0.

--
~Randy