svn commit: r276626 - head/sys/kern

[Gleb Smirnoff]

  Hi!

On Sat, Jan 03, 2015 at 05:21:20PM +0000, Hans Petter Selasky wrote:
H> Author: hselasky
H> Date: Sat Jan  3 17:21:19 2015
H> New Revision: 276626
H> URL: https://svnweb.freebsd.org/changeset/base/276626
H> 
H> Log:
H>   Rework r276532 a bit. Always avoid recursing into the console drivers
H>   clients, hence they might not handle it very well. This change allows
H>   debugging mutex problems with kernel console drivers when
H>   "debug.witness.skipspin=0" is set in the boot environment.
H>   
H>   MFC after:	1 week

I'd ask Andriy avg@ to review r276532 and this change. AFAIR, he attacked
this problem in 2011/2012 and it appeared much more complex than just adding
MTX_RECURSE flag.

H> Modified:
H>   head/sys/kern/kern_cons.c
H> 
H> Modified: head/sys/kern/kern_cons.c
H> ==============================================================================
H> --- head/sys/kern/kern_cons.c	Sat Jan  3 16:48:08 2015	(r276625)
H> +++ head/sys/kern/kern_cons.c	Sat Jan  3 17:21:19 2015	(r276626)
H> @@ -512,6 +512,13 @@ cnputs(char *p)
H>  	int unlock_reqd = 0;
H>  
H>  	if (use_cnputs_mtx) {
H> +	  	/*
H> +		 * NOTE: Debug prints and/or witness printouts in
H> +		 * console driver clients can cause the "cnputs_mtx"
H> +		 * mutex to recurse. Simply return if that happens.
H> +		 */
H> +		if (mtx_owned(&cnputs_mtx))
H> +			return;
H>  		mtx_lock_spin(&cnputs_mtx);
H>  		unlock_reqd = 1;
H>  	}
H> @@ -601,13 +608,7 @@ static void
H>  cn_drvinit(void *unused)
H>  {
H>  
H> -	/*
H> -	 * NOTE: Debug prints and/or witness printouts in console
H> -	 * driver clients can cause the "cnputs_mtx" mutex to
H> -	 * recurse. Make sure the "MTX_RECURSE" flags is set!
H> -	 */
H> -	mtx_init(&cnputs_mtx, "cnputs_mtx", NULL, MTX_SPIN |
H> -	    MTX_NOWITNESS | MTX_RECURSE);
H> +	mtx_init(&cnputs_mtx, "cnputs_mtx", NULL, MTX_SPIN | MTX_NOWITNESS);
H>  	use_cnputs_mtx = 1;
H>  }
H>  
H> 

-- 
Totus tuus, Glebius.