Re: [ARM32] Possible PMAP_KMPAGE = 0 produces lock collision

[Matt Thomas <matt%3am-software.com@localhost>]

On Jul 18, 2008, at 4:46 AM, Andy Shevchenko wrote:

> Hello.
>
> Possible the patch "Default PMAP_KMPAGE to 0 unless it's been
> previously defined by <machine/pmap.h>"
> produces lock collision when LOCKDEBUG option is set.
>
> With this patch (means the behaviour is changed w/o it) the kernel
> panics at very start (and I can't start even debugger):
>
> init subsystems: stacks vectors undefined page pmap done.
> initarm done.
> Reader / writer lock error: lockdebug_wantlock: locking against myself
>
> lock address : 0x0000000080542b8c type     :     sleep/adaptive
> initialized  : 0x00000000803afb78
> shared holds :                  0 exclusive:                  1
> shares wanted:                  0 exclusive:                  1
> current cpu  :                  0 last held:                  0
> current lwp  : 0x000000008051da80 last held: 0x000000008051da80
> last locked  : 0x00000000803b1264 unlocked : 0x00000000803b2a48
> owner/count  : 0x000000008051da80 flags    : 0x000000000000000c
>
> Turnstile chain at 0x80551540.
> => No active turnstile for this lock.
>
> panic: LOCKDEBUG

you have the updated arm/include/arm32/pmap.h which defines PMAP_KMPAGE?
If it's not defined, the kernel will definitely panic.  Does the kernel
work if LOCKDEBUG isn't set?

If you have a JTAG debugger, use a breakpoint on __kernassert to catch
these.