FreeBSD 10-STABLE/sparc64 panic

[Edward Tomasz Napierała]

Wiadomość napisana przez Nathaniel W Filardo w dniu 20 maj 2014, o godz. 17:37:
> Well, I am having a bad day; I tried to roll back to 9 but ended up building
> a 10 kernel again (too many source checkouts, sigh).  In any case, I kicked
> WITNESS and INVARIANTS on this time and, while I cannot say for sure that
> it is related, early in boot I am told
> 
> lock order reversal:
> 1st 0xc0756998 entropy harvest mutex (entropy harvest mutex) @ /systank/src-git/sys/dev/random/random_harvestq.c:198
> 2nd 0xfffff800055c7c38 uart_hwmtx (uart_hwmtx) @ /systank/src-git/sys/dev/uart/uart_cpu.h:94
> KDB: stack backtrace:
> _witness_debugger() at _witness_debugger+0x38
> witness_checkorder() at witness_checkorder+0xea0
> __mtx_lock_spin_flags() at __mtx_lock_spin_flags+0x134
> uart_cnputc() at uart_cnputc+0x60
> cnputc() at cnputc+0xac
> putchar() at putchar+0xe8
> kvprintf() at kvprintf+0x88
> _vprintf() at _vprintf+0x38
> vprintf() at vprintf+0x10
> printf() at printf+0x20
> witness_checkorder() at witness_checkorder+0xbf8
> __mtx_lock_spin_flags() at __mtx_lock_spin_flags+0x134

I think I've seen this on 10.0/amd64.  This was caused by not having this:

options         WITNESS_SKIPSPIN        # Don't run witness on spinlocks for speed

in the kernel config.  Basically, WITNESS without WITNESS_SKIPSPIN
was unbootable.  I didn't investigate this further; just added that option.