Re: Problems with USB

Re: Problems with USB

[Petr.Titera]

Hello,

     yes it look that this patch can work. But who send that signal. I
tried to hack kernel to print signal number and it looks, that khubd gets
SIGHUP.

Petr Titera
petr.titera@whitesoft.cz



                                                                                            
                    Greg KH                                                                 
                    <greg@kroah.co       To:     Petr.Titera@whitesoft.cz                   
                    m>                   cc:     linux-kernel@vger.kernel.org               
                                         Fax to:                                            
                    13.01.2003           Subject:     Re: Problems with USB                 
                    06:55                                                                   
                                                                                            
                                                                                            




On Sun, Jan 12, 2003 at 09:44:42PM +0100, Petr.Titera@whitesoft.cz wrote:
> Hello all,
>
>      I have problems with USB in recent kernels (tested on 2.5.56) and
> RedHat 8.0. Right after end of script  '/etc/rc.d/rc.sysinit' and before
> script '/etc/rc.d/rc' which runs after USB  daemon khubd gets some signal
> and ends. From this point USB does not work as as system does not get any
> plug events. If I disable USB at startup and load modules later,
everything
> works.

Can you try this patch that I just added to the USB tree?  It seems to
work for me.

thanks,

greg k-h


# USB: Fix from Jeff and Pete to keep khubd from being able to be killed
#      by a signal

diff -Nru a/drivers/usb/core/hub.c b/drivers/usb/core/hub.c
--- a/drivers/usb/core/hub.c  Sun Jan 12 22:03:13 2003
+++ b/drivers/usb/core/hub.c  Sun Jan 12 22:03:13 2003
@@ -1085,6 +1085,12 @@

     daemonize();

+    /* keep others from killing us */
+    spin_lock_irq(&current->sig->siglock);
+    sigemptyset(&current->blocked);
+    recalc_sigpending();
+    spin_unlock_irq(&current->sig->siglock);
+
     /* Setup a nice name */
     strcpy(current->comm, "khubd");

Re: Problems with USB

[Greg KH]

On Mon, Jan 13, 2003 at 09:13:44AM +0100, Petr.Titera@whitesoft.cz wrote:
> 
> Hello,
> 
>      yes it look that this patch can work.

Great, thanks for testing this.

> But who send that signal. I tried to hack kernel to print signal
> number and it looks, that khubd gets SIGHUP.

I think it's something in the NGTL code, but I'm not really sure.

thanks,

greg k-h

Problems with USB

[Petr.Titera]

Hello all,

     I have problems with USB in recent kernels (tested on 2.5.56) and
RedHat 8.0. Right after end of script  '/etc/rc.d/rc.sysinit' and before
script '/etc/rc.d/rc' which runs after USB  daemon khubd gets some signal
and ends. From this point USB does not work as as system does not get any
plug events. If I disable USB at startup and load modules later, everything
works.
     Under 2.4.x kernel everithing works as expected.


Petr Titera
petr.titera@whitesoft.cz

Re: Problems with USB

[Greg KH]

On Sun, Jan 12, 2003 at 09:44:42PM +0100, Petr.Titera@whitesoft.cz wrote:
> Hello all,
> 
>      I have problems with USB in recent kernels (tested on 2.5.56) and
> RedHat 8.0. Right after end of script  '/etc/rc.d/rc.sysinit' and before
> script '/etc/rc.d/rc' which runs after USB  daemon khubd gets some signal
> and ends. From this point USB does not work as as system does not get any
> plug events. If I disable USB at startup and load modules later, everything
> works.

Can you try this patch that I just added to the USB tree?  It seems to
work for me.

thanks,

greg k-h


# USB: Fix from Jeff and Pete to keep khubd from being able to be killed
#      by a signal

diff -Nru a/drivers/usb/core/hub.c b/drivers/usb/core/hub.c
--- a/drivers/usb/core/hub.c	Sun Jan 12 22:03:13 2003
+++ b/drivers/usb/core/hub.c	Sun Jan 12 22:03:13 2003
@@ -1085,6 +1085,12 @@
 
 	daemonize();
 
+	/* keep others from killing us */
+	spin_lock_irq(&current->sig->siglock);
+	sigemptyset(&current->blocked);
+	recalc_sigpending();
+	spin_unlock_irq(&current->sig->siglock);
+
 	/* Setup a nice name */
 	strcpy(current->comm, "khubd");
 

Re: Problems with USB

[Pete Zaitcev]

> On Sun, Jan 12, 2003 at 09:44:42PM +0100, Petr.Titera@whitesoft.cz wrote:
>>      I have problems with USB in recent kernels (tested on 2.5.56) and
>> RedHat 8.0. Right after end of script  '/etc/rc.d/rc.sysinit' and before
>> script '/etc/rc.d/rc' which runs after USB  daemon khubd gets some signal
>> and ends.

> greg k-h
> 
> # USB: Fix from Jeff and Pete to keep khubd from being able to be killed
> #      by a signal
> 
> diff -Nru a/drivers/usb/core/hub.c b/drivers/usb/core/hub.c
> --- a/drivers/usb/core/hub.c	Sun Jan 12 22:03:13 2003
> +++ b/drivers/usb/core/hub.c	Sun Jan 12 22:03:13 2003
> @@ -1085,6 +1085,12 @@
>  
>  	daemonize();
>  
> +	/* keep others from killing us */
> +	spin_lock_irq(&current->sig->siglock);
> +	sigemptyset(&current->blocked);
> +	recalc_sigpending();
> +	spin_unlock_irq(&current->sig->siglock);
> +
>  	/* Setup a nice name */
>  	strcpy(current->comm, "khubd");
>  

For the record, I disagree with this strongly.

In khubd case, the existing code did it righ. It ran
daemonize(), which should have divorced it from the session
and process group. If daemonize is buggy, it is the place
to fix it. Ingo has the fix for 2.5, in fact, I think he
may have sent it to Linus already (it's __set_special_pids()).

My version of the patch above never was intended as anything
but a stop-gap solution which allowed us to ship a beta on
schedule, while I was investigating the cause.

Jeff told me on IRC that "every bit in 8139too.c thread was
added as a response to a particular problem", but he did not
remember what particular problem this kludge fixed there.

I stole the stop-gap from Stephen's kjournald. Note, that
I do not have any idea why he put it in there. Very likely,
for entirely different reason than working around bugs in
daemonize().

-- Pete

[P.S. Greg, if Jeff's P.O.V. prevails in the court of Linus,
change that sigemptyset() to siginitsetinv(...., sigmask(SIGKILL)).
Otherwise the whole signal checking path in khubd becomes utterly
meaningless.]

[P.P.S And take my name from the bk message. Yes, I wrote
the patch, but I do not want to endorse it, however indirectly]

Re: Problems with USB

[Greg KH]

On Mon, Jan 13, 2003 at 01:59:09PM -0500, Pete Zaitcev wrote:
> > On Sun, Jan 12, 2003 at 09:44:42PM +0100, Petr.Titera@whitesoft.cz wrote:
> >>      I have problems with USB in recent kernels (tested on 2.5.56) and
> >> RedHat 8.0. Right after end of script  '/etc/rc.d/rc.sysinit' and before
> >> script '/etc/rc.d/rc' which runs after USB  daemon khubd gets some signal
> >> and ends.
> 
> > greg k-h
> > 
> > # USB: Fix from Jeff and Pete to keep khubd from being able to be killed
> > #      by a signal
> > 
> > diff -Nru a/drivers/usb/core/hub.c b/drivers/usb/core/hub.c
> > --- a/drivers/usb/core/hub.c	Sun Jan 12 22:03:13 2003
> > +++ b/drivers/usb/core/hub.c	Sun Jan 12 22:03:13 2003
> > @@ -1085,6 +1085,12 @@
> >  
> >  	daemonize();
> >  
> > +	/* keep others from killing us */
> > +	spin_lock_irq(&current->sig->siglock);
> > +	sigemptyset(&current->blocked);
> > +	recalc_sigpending();
> > +	spin_unlock_irq(&current->sig->siglock);
> > +
> >  	/* Setup a nice name */
> >  	strcpy(current->comm, "khubd");
> >  
> 
> For the record, I disagree with this strongly.

Ok, Arjan also just convinced me to revert this patch, as Ingo's patch
is the correct fix.

thanks,

greg k-h