7.3-STABLE and Linux version of SIMetrix

[Andre Albsmeier]

On Thu, 06-Jan-2011 at 08:57:53 +0100, Alexander Leidinger wrote:
> Quoting Andre Albsmeier <Andre.Albsmeier at siemens.com> (from Wed, 5 Jan  
> 2011 20:19:15 +0100):
> 
> > On Sun, 02-Jan-2011 at 11:50:21 +0100, Alexander Leidinger wrote:
> >> On Sat, 1 Jan 2011 23:46:29 +0100 Andre Albsmeier
> >> <Andre.Albsmeier at siemens.com> wrote:
> >>
> >> > On Fri, 31-Dec-2010 at 14:48:00 +0100, Alexander Leidinger wrote:
> >> > > On Thu, 30 Dec 2010 08:51:24 +0100 Andre Albsmeier
> >> > > <Andre.Albsmeier at siemens.com> wrote:
> >> > >
> >> > > > I try to get the Linux version of SIMetrix (a very nice circuit
> >> > > > simulator) running. Everything looks fine: It starts, the GUI
> >> > > > comes up, you can draw schematics and so on. But when it comes
> >> > > > to simulation, the (SIMetrix-)console says:
> >> > > >
> >> > > > *** Fatal error, out of memory ***
> >> > > > Could not allocate shared heap
> >> > > > Exception occurred while executing script command Run
> >> > >
> >> > > Is there something in the dmesg output? In case it tries to execute
> >> > > an unsupported ioctl/syscall it should show up there. If not I
> >> > > suggest to give 8.x a try, it has an improved linuxulator.
> >> >
> >> > Bad luck. I just started the PC-BSD 8.1 live system and
> >> > the error there is exactly the same...
> >>
> >> Then there is only ktrace + linux_kdump (use the package) or dtrace
> >> left.
> >
> > Got it running... A short explanation:
> >
> > Linux' shm_open() fails because it wants to find some funky shmfs
> > to construct the full pathname. It starts to search at the default
> > mountpoint which is /dev/shm. If this fails it runs through fstab
> > and searches for shmfs and tmpfs. Whatever it finds will be
> > statfs()'ed to be checked for Linux' fs magic for shmfs (0x01021994).
> >
> > My solution is eerie but it works. This is what I did:
> >
> > 1. cd /compat/linux
> > 2. mkdir dev
> > 3. cd dev
> > 4. ln -s /tmp shm
> 
> As long as you do not need to access something from dev (FreeBSD one),  
> this works. As soon as you need to access something from there it  
> depends upon the software. If the software is doing a readdir, the  
> user experience will be worse. And if it does not allow to enter a  

Let me see if I got this: Now I have a /compat/linux/dev directory
which means that if some Linux programme wants to look into /dev it
will see just the shm entry. If there wasn't my dev directory it
would have continued to FreeBSD's /dev...

> name by hand, you are lost. So this is a solution for your situation,  
> but not a generic solution which we can use in the ports.

Of course not. I was just happy that it works and that that's all
what has been missing. But you are right, it is not very generic
now that I understand the impact.

> 
> Can you please try a solution with devfs.conf (I never tried if it  
> allows to link outside of dev)?

I added the following line to devfs.conf

link /tmp shm

and removed my dev stuff under /compat/linux. And it still works ;-).

> 
> > This is not enough since statfs() won't return the proper magic
> > because it's no shmfs. So let's fix this ;-):
> >
> > --- sys/compat/linux/linux_stats.c.ORI	2008-01-10 20:11:51.000000000 +0100
> > +++ sys/compat/linux/linux_stats.c	2011-01-05 20:04:18.000000000 +0100
> > @@ -379,6 +379,8 @@
> >  	struct statfs bsd_statfs;
> >  	char *path;
> >  	int error;
> > +	char* cp;
> > +	unsigned char shm = 0;
> >
> >  	LCONVPATHEXIST(td, args->path, &path);
> >
> > @@ -387,10 +389,15 @@
> >  		printf(ARGS(statfs, "%s, *"), path);
> >  #endif
> >  	error = kern_statfs(td, path, UIO_SYSSPACE, &bsd_statfs);
> > +	if( (cp = strstr( path, "/dev/shm" )) != NULL )
> > +	  shm = ( cp[8] == '\0' || (cp[8] == '/' && cp[9] == '\0') );
> > +
> >  	LFREEPATH(path);
> >  	if (error)
> >  		return (error);
> >  	bsd_to_linux_statfs(&bsd_statfs, &linux_statfs);
> > +	if( shm )
> > +	  linux_statfs.f_type = 0x01021994;
> >  	return copyout(&linux_statfs, args->buf, sizeof(linux_statfs));
> >  }
> 
> This is something we could do (with a little bit of cleanup), when we  
> find a nice solution for the dev-part.

The devfs.conf thing makes it possible to simplify the (ugly)
patch a bit:

--- /sys/compat/linux/linux_stats.c.ORI	2008-01-10 20:11:51.000000000 +0100
+++ /sys/compat/linux/linux_stats.c	2011-01-06 11:26:53.000000000 +0100
@@ -379,6 +379,7 @@
 	struct statfs bsd_statfs;
 	char *path;
 	int error;
+	unsigned char shm = 0;
 
 	LCONVPATHEXIST(td, args->path, &path);
 
@@ -387,10 +388,14 @@
 		printf(ARGS(statfs, "%s, *"), path);
 #endif
 	error = kern_statfs(td, path, UIO_SYSSPACE, &bsd_statfs);
+	if( strncmp( path, "/dev/shm", 8 ) == 0 )
+	  shm = ( path[8] == '\0' || (path[8] == '/' && path[9] == '\0') );
 	LFREEPATH(path);
 	if (error)
 		return (error);
 	bsd_to_linux_statfs(&bsd_statfs, &linux_statfs);
+	if( shm )
+	  linux_statfs.f_type = 0x01021994;
 	return copyout(&linux_statfs, args->buf, sizeof(linux_statfs));
 }

Thanks for the hint with devfs.conf, that's a lot better!

	-Andre
 
> 
> Bye,
> Alexander.
> 
> -- 
> And do you think (fop that I am) that I could be the Scarlet Pumpernickel?
> 
> http://www.Leidinger.net    Alexander @ Leidinger.net: PGP ID = B0063FE7
> http://www.FreeBSD.org       netchild @ FreeBSD.org  : PGP ID = 72077137
> 

-- 
Windows NT Multitasking: Messing up several things at once.