port-i386/47432: ddb stack trace from interrupt context randomly fails

[gson%gson.org@localhost (Andreas Gustafsson)]

>Number:         47432
>Category:       port-i386
>Synopsis:       ddb stack trace from interrupt context randomly fails
>Confidential:   no
>Severity:       serious
>Priority:       high
>Responsible:    port-i386-maintainer
>State:          open
>Class:          sw-bug
>Submitter-Id:   net
>Arrival-Date:   Fri Jan 11 09:35:00 +0000 2013
>Originator:     Andreas Gustafsson
>Release:        NetBSD-current, source date 2013.01.10.06.52.35
>Organization:
>Environment:
System: NetBSD
Architecture: i386
Machine: i386
>Description:

On NetBSD/i386, when entering ddb via an interrupt, for example by
sending a break sequence to a serial console, the stack trace
displayed by the ddb "trace" command often ends prematurely with
an internal DDB page fault.

Here's an example of such an incomplete stack trace:

  db{0}> bt
  breakpoint(c0c90ee0,3f8,5,c05949d2,c10ea5cc,0,5f5e100,0,c109e9ec,c109e974) at 
netbsd:breakpoint+0x4
  comintr(c109e8c0,c3564f30,7,c0c90010,30,c10c0010,c0c90010,d5646c,1,c36bacd8) 
at netbsd:comintr+0x597
  Xintr_ioapic_edge7() at netbsd:Xintr_ioapic_edge7+0xb5
  --- interrupt ---
  fatal page fault in supervisor mode
  trap type 6 code 0 eip c028b2c6 cs 8 eflags 246 cr2 64 ilevel 8 esp c356475c
  curlwp 0xc10c1d20 pid 0 lid 2 lowest kstack 0xc36b8000
  kernel: supervisor trap page fault, code=0
  Faulted in DDB; continuing...
  db{0}> 

For comparison, here's a correct-looking stack trace:

  db{0}> bt
  breakpoint(c0c90ee0,3f8,5,c0c91740,0,0,c0c91740,c0595838,c109e9ec,c109e974) 
at netbsd:breakpoint+0x4
  comintr(c109e8c0,c36bac78,0,0,0,0,0,0,0,0) at netbsd:comintr+0x597
  --- switch to interrupt stack ---
  Xintr_ioapic_edge7() at netbsd:Xintr_ioapic_edge7+0xb5
  --- interrupt ---
  
acpicpu_cstate_idle_enter(c117ada8,c0c91740,25,0,e58d8ee3,0,c36bad04,c10c1d20,c0
  c91740,c36bad2c) at netbsd:acpicpu_cstate_idle_enter+0x42
  acpicpu_cstate_idle(c10c1d20,0,0,0,0,c10c1d20,c056d7d0,c10c1d20,0,c0100321) 
at n
  etbsd:acpicpu_cstate_idle+0xaa
  idle_loop(c10c1d20,e99000,ea2000,0,c0100307,0,0,0,0,0) at 
netbsd:idle_loop+0x17c

Repeatedly taking stack traces from an idle system sometimes results
in incomplete traces and sometimes complete ones, seemingly at random.
This is contrast to PR 44260, where the stack trace was always incompete.

>How-To-Repeat:

pkg_add py-anita
anita --sets=kern-GENERIC,modules,base,etc interact 
http://nyftp.netbsd.org/pub/NetBSD-daily/HEAD/201301091830Z/i386/
[wait for a login prompt]
[press "control-a b" to send a break sequence the virtual serial console]
trace
cont
[press "control-a b" to send a break sequence the virtual serial console]
trace
cont
(etc)

>Fix: