Re: destroy_dev_drain stuck (devscd) after kldload, ls -l cloned device, kldunload (trying to fix kqemu...)

On Fri, Apr 18, 2008 at 09:51:50PM +0200, Juergen Lock wrote:
> Using this... after what has been done to /sys/netsmb/smb_dev.c:
> 
> Index: Makefile
> ===================================================================
> RCS file: /home/pcvs/ports/emulators/kqemu-kmod/Makefile,v
> retrieving revision 1.21
> diff -u -p -r1.21 Makefile
> --- Makefile	26 Jul 2007 21:46:30 -0000	1.21
> +++ Makefile	18 Apr 2008 18:58:29 -0000
> _at__at_ -7,7 +7,7 _at__at_
>  
>  PORTNAME=	kqemu
>  PORTVERSION=	1.3.0.p11
> -PORTREVISION=	2
> +PORTREVISION=	3
>  CATEGORIES=	emulators kld
>  MASTER_SITES=	http://fabrice.bellard.free.fr/qemu/ \
>  		http://qemu.org/ \
> _at__at_ -57,17 +57,6 _at__at_ IGNORE=		kqemu requires kernel source to
>  CFLAGS+=	-DKSE
>  .endif
>  
> -# XXX the following is wrong if you run a custom SMP kernel on a
> -# singleprocessor machine (but it probably doesn't really matter)
> -KERNCONF!=	${UNAME} -v | ${SED} 's-.*/--' | ${TR} -d ' '
> -NCPU!=		${SYSCTL} -n hw.ncpu 2>/dev/null
> -# we know 7.x GENERIC has SMP
> -.if ${KERNCONF} == "GENERIC" && ${OSVERSION} >= 700000
> -CFLAGS+=	-DSMP
> -.elif ${NCPU} > 1
> -CFLAGS+=	-DSMP
> -.endif
> -
>  post-extract:
>  .if ${OSVERSION} >= 700024 && !defined(NOKSE)
>  	_at_${ECHO_MSG} "Compiling with -DKSE."
> Index: files/patch-kqemu-freebsd.c
> ===================================================================
> RCS file: /home/pcvs/ports/emulators/kqemu-kmod/files/patch-kqemu-freebsd.c,v
> retrieving revision 1.6
> diff -u -p -r1.6 patch-kqemu-freebsd.c
> --- files/patch-kqemu-freebsd.c	21 Jul 2007 19:38:16 -0000	1.6
> +++ files/patch-kqemu-freebsd.c	18 Apr 2008 18:37:00 -0000
> _at__at_ -17,13 +17,30 _at__at_ Index: kqemu-freebsd.c
>       return SIGPENDING(curthread);
>   }
>   #endif
> -_at__at_ -320,6 +328,9 _at__at_
> +_at__at_ -320,8 +320,15 _at__at_
>   #if __FreeBSD_version >= 500000
>       dev->si_drv1 = NULL;
>       TAILQ_REMOVE(&kqemuhead, ks, kqemu_ent);
> ++#if __FreeBSD_version >= 700051
> ++    destroy_dev_sched(dev);
> ++#else
>  +#if __FreeBSD_version >= 700024
>  +    dev_relthread(dev); 
>  +#endif
>       destroy_dev(dev);
>   #endif
> ++#endif
>       free(ks, M_KQEMU);
> +     --kqemu_ref_count;
> + }
> +_at__at_ -500,6 +507,10 _at__at_
> + 	while ((ks = TAILQ_FIRST(&kqemuhead)) != NULL) {
> + 	    kqemu_destroy(ks);
> + 	}
> ++#if __FreeBSD_version >= 700051
> ++	drain_dev_clone_events();
> ++	/* destroy_dev_drain(&kqemu_cdevsw); */
> ++#endif
> + 	clone_cleanup(&kqemuclones);
> + #endif
> +         kqemu_global_delete(kqemu_gs);

The destroy_dev_drain() function waits for the cdevs having the
kqemu_cdevsw cdevsw to be destroyed. I did not looked into the actual
kqemu sources, but assumed that kqemuclones have kqemu_cdevsw as
cdevsw. If this is the case, you shall call clone_cleanup() before
destroy_dev_drain().

> 
>  (and uncommenting the destroy_dev_drain call) I get a hang when I do:
> 	kldload kqemu
> 	ls -l /dev/kqemu
> 	kldunload kqemu
> 
>  Am I doing something wrong or is the bug elsewhere?
> 
>  Oh and does anyone still see qemu hanging using the commented version?
> (Only you shouldn't kldunload too quickly after running qemu I guess... :)
> 
>  Thanx,
> 	Juergen
> _______________________________________________
> freebsd-emulation_at_freebsd.org mailing list
> http://lists.freebsd.org/mailman/listinfo/freebsd-emulation
> To unsubscribe, send any mail to "freebsd-emulation-unsubscribe_at_freebsd.org"