On Tue 2016-03-01 18:24:26, Sergey Senozhatsky wrote: > Hello, > > On (02/29/16 20:19), Sergey Senozhatsky wrote: > [..] > > > That is the problem. zap_locks() is not a solution. > > > > > > First, it handles only lockbuf_lock and console_sem. There are other > > > locks used by particular consoles that might cause a deadlock. > > > > yes, well, that's true for panic() in general. > > Petr, what do you think of this (added PRINTK_NMI_FLUSH_ON_PANIC)? > > 1) zap_locks() in console_flush_on_panic() > 2) add PRINTK_NMI_FLUSH_ON_PANIC symbols > 3) add printk_nmi_flush_on_panic()
This is definitely better than nothing. Well, it seems that the printk/NMI patches that motivated this patch will be removed for a while, see http://thread.gmane.org/gmane.linux.ports.arm.kernel/482845/focus=483002 I want to play with the panic handling a bit more and make it better working out of box. It might be enough to put the messages into the rind buffer when crashdump is going to be produced. Also there is still the idea about using the lock-less early console. I think that the solution from this patch might be the last fallback. Thanks a lot for proposals, Petr
