Re: 3.7 RC1

Re: 3.7 RC1

[Greg KH]

On Mon, Oct 22, 2012 at 04:37:45PM -0700, K. Y. Srinivasan wrote:
> 
> While testing 3.7 RC1 I discovered that invoking the function orderly_poweroff()
> from an interrupt context will trigger an ASSERT(). This was not the case till
> recently. The comment preceding the orderly_poweroff() function claims that this
> function can be invoked from any context and in the current Hyper-V util driver,
> we support host-driven orderly shut down of the guest by invoking this
> orderly_poweroff() function in the context of the message callback. This code has
> been working for a very long time and it is broken now. Is my assumption that
> orderly_poweroff() could be invoked from the interrupt context a wrong assumption?

Can you check 3.7-rc2?  If that also fails, care to do a 'git bisect'
from 3.6 to 3.7-rc1 to find the problem patch?

thanks,

greg k-h

3.7 RC1

[K. Y. Srinivasan]

While testing 3.7 RC1 I discovered that invoking the function orderly_poweroff()
from an interrupt context will trigger an ASSERT(). This was not the case till
recently. The comment preceding the orderly_poweroff() function claims that this
function can be invoked from any context and in the current Hyper-V util driver,
we support host-driven orderly shut down of the guest by invoking this
orderly_poweroff() function in the context of the message callback. This code has
been working for a very long time and it is broken now. Is my assumption that
orderly_poweroff() could be invoked from the interrupt context a wrong assumption?

Regards,

K. Y

RE: 3.7 RC1

[KY Srinivasan]

Will do and I will report back.

K. Y

> -----Original Message-----
> From: Greg KH [mailto:gregkh@linuxfoundation.org]
> Sent: Monday, October 22, 2012 7:34 PM
> To: KY Srinivasan
> Cc: linux-kernel@vger.kernel.org; devel@linuxdriverproject.org; olaf@aepfle.de;
> apw@canonical.com; jasowang@redhat.com
> Subject: Re: 3.7 RC1
> 
> On Mon, Oct 22, 2012 at 04:37:45PM -0700, K. Y. Srinivasan wrote:
> >
> > While testing 3.7 RC1 I discovered that invoking the function
> orderly_poweroff()
> > from an interrupt context will trigger an ASSERT(). This was not the case till
> > recently. The comment preceding the orderly_poweroff() function claims that
> this
> > function can be invoked from any context and in the current Hyper-V util driver,
> > we support host-driven orderly shut down of the guest by invoking this
> > orderly_poweroff() function in the context of the message callback. This code
> has
> > been working for a very long time and it is broken now. Is my assumption that
> > orderly_poweroff() could be invoked from the interrupt context a wrong
> assumption?
> 
> Can you check 3.7-rc2?  If that also fails, care to do a 'git bisect'
> from 3.6 to 3.7-rc1 to find the problem patch?
> 
> thanks,
> 
> greg k-h
> 
> 

Re: 3.7 RC1

[Dan Carpenter]

On Mon, Oct 22, 2012 at 04:37:45PM -0700, K. Y. Srinivasan wrote:
> 
> While testing 3.7 RC1 I discovered that invoking the function orderly_poweroff()
> from an interrupt context will trigger an ASSERT(). This was not the case till
> recently. The comment preceding the orderly_poweroff() function claims that this
> function can be invoked from any context and in the current Hyper-V util driver,
> we support host-driven orderly shut down of the guest by invoking this
> orderly_poweroff() function in the context of the message callback. This code has
> been working for a very long time and it is broken now. Is my assumption that
> orderly_poweroff() could be invoked from the interrupt context a wrong assumption?

You can't call orderly_poweroff() from interrupt context.

regards,
dan carpenter

RE: 3.7 RC1

[KY Srinivasan]

> -----Original Message-----
> From: Dan Carpenter [mailto:dan.carpenter@oracle.com]
> Sent: Tuesday, October 23, 2012 1:47 AM
> To: KY Srinivasan
> Cc: gregkh@linuxfoundation.org; linux-kernel@vger.kernel.org;
> devel@linuxdriverproject.org; olaf@aepfle.de; apw@canonical.com;
> jasowang@redhat.com
> Subject: Re: 3.7 RC1
> 
> On Mon, Oct 22, 2012 at 04:37:45PM -0700, K. Y. Srinivasan wrote:
> >
> > While testing 3.7 RC1 I discovered that invoking the function
> orderly_poweroff()
> > from an interrupt context will trigger an ASSERT(). This was not the case till
> > recently. The comment preceding the orderly_poweroff() function claims that
> this
> > function can be invoked from any context and in the current Hyper-V util driver,
> > we support host-driven orderly shut down of the guest by invoking this
> > orderly_poweroff() function in the context of the message callback. This code
> has
> > been working for a very long time and it is broken now. Is my assumption that
> > orderly_poweroff() could be invoked from the interrupt context a wrong
> assumption?
> 
> You can't call orderly_poweroff() from interrupt context.

Thanks Dan; I am curious  to understand the basis for your assertion.
As I noted earlier  the documentation for this function clearly says it can
be called from any context. Furthermore, __orderly_poweroff(), the helper
function allocates memory with the GFP_ATOMIC flag set. Lastly, the behavior
of orderly_poweroff() has been such that this function could be called from interrupt context
for a very long time and something has changed now. For what it is worth, there are other users in
the kernel (in 3.7 RC1) that are invoking the orderly_poweroff() function from interrupt
context other than the Hyper-V shutdown handler:  fsl_hv_shutdown_isr() in drivers/virt/fsl_hypervisor.c.
I suspect there are other users as well that have made this similar assumption.


Regards,

K. Y

RE: 3.7 RC1

[KY Srinivasan]

> -----Original Message-----
> From: Greg KH [mailto:gregkh@linuxfoundation.org]
> Sent: Monday, October 22, 2012 7:34 PM
> To: KY Srinivasan
> Cc: linux-kernel@vger.kernel.org; devel@linuxdriverproject.org; olaf@aepfle.de;
> apw@canonical.com; jasowang@redhat.com
> Subject: Re: 3.7 RC1
> 
> On Mon, Oct 22, 2012 at 04:37:45PM -0700, K. Y. Srinivasan wrote:
> >
> > While testing 3.7 RC1 I discovered that invoking the function
> orderly_poweroff()
> > from an interrupt context will trigger an ASSERT(). This was not the case till
> > recently. The comment preceding the orderly_poweroff() function claims that
> this
> > function can be invoked from any context and in the current Hyper-V util driver,
> > we support host-driven orderly shut down of the guest by invoking this
> > orderly_poweroff() function in the context of the message callback. This code
> has
> > been working for a very long time and it is broken now. Is my assumption that
> > orderly_poweroff() could be invoked from the interrupt context a wrong
> assumption?
> 
> Can you check 3.7-rc2?  If that also fails, care to do a 'git bisect'
> from 3.6 to 3.7-rc1 to find the problem patch?

Looks like this problem is fixed in rc2.

Regards,

K. Y

Re: 3.7 RC1

[Dan Carpenter]

On Tue, Oct 23, 2012 at 02:24:58PM +0000, KY Srinivasan wrote:
> 
> 
> > -----Original Message-----
> > From: Dan Carpenter [mailto:dan.carpenter@oracle.com]
> > Sent: Tuesday, October 23, 2012 1:47 AM
> > To: KY Srinivasan
> > Cc: gregkh@linuxfoundation.org; linux-kernel@vger.kernel.org;
> > devel@linuxdriverproject.org; olaf@aepfle.de; apw@canonical.com;
> > jasowang@redhat.com
> > Subject: Re: 3.7 RC1
> > 
> > On Mon, Oct 22, 2012 at 04:37:45PM -0700, K. Y. Srinivasan wrote:
> > >
> > > While testing 3.7 RC1 I discovered that invoking the function
> > orderly_poweroff()
> > > from an interrupt context will trigger an ASSERT(). This was not the case till
> > > recently. The comment preceding the orderly_poweroff() function claims that
> > this
> > > function can be invoked from any context and in the current Hyper-V util driver,
> > > we support host-driven orderly shut down of the guest by invoking this
> > > orderly_poweroff() function in the context of the message callback. This code
> > has
> > > been working for a very long time and it is broken now. Is my assumption that
> > > orderly_poweroff() could be invoked from the interrupt context a wrong
> > assumption?
> > 
> > You can't call orderly_poweroff() from interrupt context.
> 
> Thanks Dan; I am curious  to understand the basis for your assertion.
> As I noted earlier  the documentation for this function clearly says it can
> be called from any context. Furthermore, __orderly_poweroff(), the helper
> function allocates memory with the GFP_ATOMIC flag set. Lastly, the behavior
> of orderly_poweroff() has been such that this function could be called from interrupt context
> for a very long time and something has changed now. For what it is worth, there are other users in
> the kernel (in 3.7 RC1) that are invoking the orderly_poweroff() function from interrupt
> context other than the Hyper-V shutdown handler:  fsl_hv_shutdown_isr() in drivers/virt/fsl_hypervisor.c.
> I suspect there are other users as well that have made this similar assumption.
> 

Aw crap.  I was wrong.  Sorry about that.

regards,
dan carpenter

RE: 3.7 RC1

[KY Srinivasan]

> -----Original Message-----
> From: Greg KH [mailto:gregkh@linuxfoundation.org]
> Sent: Monday, October 22, 2012 7:34 PM
> To: KY Srinivasan
> Cc: linux-kernel@vger.kernel.org; devel@linuxdriverproject.org; olaf@aepfle.de;
> apw@canonical.com; jasowang@redhat.com
> Subject: Re: 3.7 RC1
> 
> On Mon, Oct 22, 2012 at 04:37:45PM -0700, K. Y. Srinivasan wrote:
> >
> > While testing 3.7 RC1 I discovered that invoking the function
> orderly_poweroff()
> > from an interrupt context will trigger an ASSERT(). This was not the case till
> > recently. The comment preceding the orderly_poweroff() function claims that
> this
> > function can be invoked from any context and in the current Hyper-V util driver,
> > we support host-driven orderly shut down of the guest by invoking this
> > orderly_poweroff() function in the context of the message callback. This code
> has
> > been working for a very long time and it is broken now. Is my assumption that
> > orderly_poweroff() could be invoked from the interrupt context a wrong
> assumption?
> 
> Can you check 3.7-rc2?  If that also fails, care to do a 'git bisect'
> from 3.6 to 3.7-rc1 to find the problem patch?

The problem still persists in 3.7-rc2. Sorry for the false positive; we had a script problem
that indicated that the problem had gone away. Manual testing has shown that the problem
persists. We will investigate it further.

Regards,

K. Y
> 
> thanks,
> 
> greg k-h
> 
>