Subject: Re: [PATCH] misc/GenWQE: fix pci_enable_msi usage
From: Frank Haverkamp <haver@linux.vnet.ibm.com>
Reply-To: haver@linux.vnet.ibm.com
To: Bjorn Helgaas <bhelgaas@google.com>
Cc: Sebastian Ott <sebott@linux.vnet.ibm.com>,
        Greg Kroah-Hartman <gregkh@linuxfoundation.org>,
        Alexander Gordeev <agordeev@redhat.com>, linux-kernel@vger.kernel.org,
        linux-pci@vger.kernel.org
In-Reply-To: <20140716211052.GC14366@google.com>
References: <alpine.LFD.2.11.1407091237030.1590@denkbrett>
	 <20140716211052.GC14366@google.com>
Content-Type: text/plain; charset="UTF-8"
Organization: IBM
Date: Thu, 17 Jul 2014 17:48:59 +0200
Message-ID: <1405612139.30550.13.camel@oc7383187364.ibm.com>
Mime-Version: 1.0
Content-Transfer-Encoding: 7bit
Sender: linux-kernel-owner@vger.kernel.org

Hi Bjorn and Sebastian,

Am Mittwoch, den 16.07.2014, 15:10 -0600 schrieb Bjorn Helgaas:
> [+cc linux-pci]
> 
> On Wed, Jul 09, 2014 at 12:46:39PM +0200, Sebastian Ott wrote:
> > GenWQE used to call pci_enable_msi_block to allocate a desired number
> > of MSI's. If that was not possible pci_enable_msi_block returned with a
> > smaller number which might be possible to allocate. GenWQE then called
> > pci_enable_msi_block with that number.
> > 
> > Since commit a30d0108b
> > "GenWQE: Use pci_enable_msi_exact() instead of pci_enable_msi_block()"
> > pci_enable_msi_exact is used which fails if the desired number of MSI's
> > was not possible to allocate. Change GenWQE to use pci_enable_msi_range
> > to restore the old behavior.
> > 
> > Signed-off-by: Sebastian Ott <sebott@linux.vnet.ibm.com>
> 
> Since this fixes a30d0108b09a, which we merged via my tree in v3.16-rc1, I
> applied this with Alexander and Frank's acks to for-linus for v3.16,
> thanks!
> 
> > ---
> >  drivers/misc/genwqe/card_ddcb.c  |  4 +---
> >  drivers/misc/genwqe/card_utils.c | 10 ++++++----
> >  2 files changed, 7 insertions(+), 7 deletions(-)
> > 
> > diff --git a/drivers/misc/genwqe/card_ddcb.c b/drivers/misc/genwqe/card_ddcb.c
> > index c8046db..f66d43d 100644
> > --- a/drivers/misc/genwqe/card_ddcb.c
> > +++ b/drivers/misc/genwqe/card_ddcb.c
> > @@ -1237,9 +1237,7 @@ int genwqe_setup_service_layer(struct genwqe_dev *cd)
> >  	}
> >  
> >  	rc = genwqe_set_interrupt_capability(cd, GENWQE_MSI_IRQS);
> > -	if (rc > 0)
> > -		rc = genwqe_set_interrupt_capability(cd, rc);
> > -	if (rc != 0) {
> > +	if (rc) {
> >  		rc = -ENODEV;
> >  		goto stop_kthread;
> >  	}
> > diff --git a/drivers/misc/genwqe/card_utils.c b/drivers/misc/genwqe/card_utils.c
> > index 62cc6bb..6abc437 100644
> > --- a/drivers/misc/genwqe/card_utils.c
> > +++ b/drivers/misc/genwqe/card_utils.c
> > @@ -718,10 +718,12 @@ int genwqe_set_interrupt_capability(struct genwqe_dev *cd, int count)
> >  	int rc;
> >  	struct pci_dev *pci_dev = cd->pci_dev;
> >  
> > -	rc = pci_enable_msi_exact(pci_dev, count);
> > -	if (rc == 0)
> > -		cd->flags |= GENWQE_FLAG_MSI_ENABLED;
> > -	return rc;
> > +	rc = pci_enable_msi_range(pci_dev, 1, count);
> > +	if (rc < 0)
> > +		return rc;
> > +
> > +	cd->flags |= GENWQE_FLAG_MSI_ENABLED;
> > +	return 0;
> >  }

The original code tried to register for 4 interrupts. On my system the
following scenario was executed:

We tried to register for 4 MSI interrupts, which was the original plan
of interrupts for the card. Linux returned that it could not do that and
that just 1 would work. We tried to register for 1 MSI interrupt and it
worked. GENWQE_MSI_IRQS is still 4. That is how my original version
looked like:

        rc = genwqe_set_interrupt_capability(cd, GENWQE_MSI_IRQS);
        if (rc > 0)
                rc = genwqe_set_interrupt_capability(cd, rc);
        if (rc != 0) {
                rc = -ENODEV;
                goto stop_kthread;
        }

So genwqe_set_interrupt_capability() was indeed called two times.
Calling it with 4 returned with rc=1 and no interrupts enabled. So
calling it a 2nd time with rc=1 was leading to the proper enablement.

I think the simplest way is to set GENWQE_MSI_IRQS to 1. The rest of the
code at least assumes just one interrupt, because 4 never worked.

Regards

Frank

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/