• Steven Rostedt's avatar
    init: scream bloody murder if interrupts are enabled too early · f91eb62f
    Steven Rostedt authored
    As I was testing a lot of my code recently, and having several
    "successes", I accidentally noticed in the dmesg this little line:
    
      start_kernel(): bug: interrupts were enabled *very* early, fixing it
    
    Sure enough, one of my patches two commits ago enabled interrupts early.
    The sad part here is that I never noticed it, and I ran several tests with
    ktest too, and ktest did not notice this line.
    
    What ktest looks for (and so does many other automated testing scripts) is
    a back trace produced by a WARN_ON() or BUG().  As a back trace was never
    produced, my buggy patch could have slipped into linux-next, or even
    worse, mainline.
    
    Adding a WARN(!irqs_disabled()) makes this bug a little more obvious:
    
      PID hash table entries: 4096 (order: 3, 32768 bytes)
      __ex_table already sorted, skipping sort
      Checking aperture...
      No AGP bridge found
      Calgary: detecting Calgary via BIOS EBDA area
      Calgary: Unable to locate Rio Grande table in EBDA - bailing!
      Memory: 2003252k/2054848k available (4857k kernel code, 460k absent, 51136k reserved, 6210k data, 1096k init)
      ------------[ cut here ]------------
      WARNING: at /home/rostedt/work/git/linux-trace.git/init/main.c:543 start_kernel+0x21e/0x415()
      Hardware name: To Be Filled By O.E.M.
      Interrupts were enabled *very* early, fixing it
      Modules linked in:
      Pid: 0, comm: swapper/0 Not tainted 3.8.0-test+ #286
      Call Trace:
        warn_slowpath_common+0x83/0x9b
        warn_slowpath_fmt+0x46/0x48
        start_kernel+0x21e/0x415
        x86_64_start_reservations+0x10e/0x112
        x86_64_start_kernel+0x102/0x111
      ---[ end trace 007d8b0491b4f5d8 ]---
      Preemptible hierarchical RCU implementation.
       RCU restricting CPUs from NR_CPUS=8 to nr_cpu_ids=4.
      NR_IRQS:4352 nr_irqs:712 16
      Console: colour VGA+ 80x25
      console [ttyS0] enabled, bootconsole disabled
    
    Do you see it?
    
    The original version of this patch just slapped a WARN_ON() in there and
    kept the printk().  Ard van Breemen suggested using the WARN() interface,
    which makes the code a bit cleaner.
    
    Also, while examining other warnings in init/main.c, I found two other
    locations that deserve a bloody murder scream if their conditions are hit,
    and updated them accordingly.
    Signed-off-by: default avatarSteven Rostedt <rostedt@goodmis.org>
    Cc: Ard van Breemen <ard@telegraafnet.nl>
    Signed-off-by: default avatarAndrew Morton <akpm@linux-foundation.org>
    Signed-off-by: default avatarLinus Torvalds <torvalds@linux-foundation.org>
    f91eb62f
main.c 22 KB