[PATCH] Intel IOMMU: RMRRs do not necessarily have to be present on all VT-d capable platforms

[PATCH] Intel IOMMU: RMRRs do not necessarily have to be present on all VT-d capable platforms

[Yong Wang]

RMRRs do not necessarily have to be present on all VT-d capable platforms.
The printk is just informational and does not need to be followed by an
error return.

Signed-off-by: Yong Y Wang <yong.y.wang@intel.com>
---
 dmar.c |    4 +---
 1 file changed, 1 insertion(+), 3 deletions(-)

diff --git a/drivers/pci/dmar.c b/drivers/pci/dmar.c
index f941f60..8bf86ae 100644
--- a/drivers/pci/dmar.c
+++ b/drivers/pci/dmar.c
@@ -317,10 +317,8 @@ int __init dmar_table_init(void)
 		return -ENODEV;
 	}
 
-	if (list_empty(&dmar_rmrr_units)) {
+	if (list_empty(&dmar_rmrr_units))
 		printk(KERN_INFO PREFIX "No RMRR found\n");
-		return -ENODEV;
-	}
 
 	return 0;
 }

Re: [PATCH] Intel IOMMU: RMRRs do not necessarily have to be present on all VT-d capable platforms

[Andrew Morton]

On Tue, 1 Jul 2008 05:13:54 -0400 Yong Wang <yong.y.wang@linux.intel.com> wrote:

> RMRRs do not necessarily have to be present on all VT-d capable platforms.
> The printk is just informational and does not need to be followed by an
> error return.
> 
> Signed-off-by: Yong Y Wang <yong.y.wang@intel.com>
> ---
>  dmar.c |    4 +---
>  1 file changed, 1 insertion(+), 3 deletions(-)
> 
> diff --git a/drivers/pci/dmar.c b/drivers/pci/dmar.c
> index f941f60..8bf86ae 100644
> --- a/drivers/pci/dmar.c
> +++ b/drivers/pci/dmar.c
> @@ -317,10 +317,8 @@ int __init dmar_table_init(void)
>  		return -ENODEV;
>  	}
>  
> -	if (list_empty(&dmar_rmrr_units)) {
> +	if (list_empty(&dmar_rmrr_units))
>  		printk(KERN_INFO PREFIX "No RMRR found\n");
> -		return -ENODEV;
> -	}
>  
>  	return 0;
>  }

>From your description I am unable to work out how important this fix is.

Is it needed in 2.6.25.x?  In 2.6.26?  In 2.6.27?  There's no way for me to
know :(

Re: [PATCH] Intel IOMMU: RMRRs do not necessarily have to be present on all VT-d capable platforms

[Yong Wang]

On Thu, Jul 10, 2008 at 01:01:49AM -0700, Andrew Morton wrote:
> On Tue, 1 Jul 2008 05:13:54 -0400 Yong Wang <yong.y.wang@linux.intel.com> wrote:
> 
> > RMRRs do not necessarily have to be present on all VT-d capable platforms.
> > The printk is just informational and does not need to be followed by an
> > error return.
> > 
> > Signed-off-by: Yong Y Wang <yong.y.wang@intel.com>
> > ---
> >  dmar.c |    4 +---
> >  1 file changed, 1 insertion(+), 3 deletions(-)
> > 
> > diff --git a/drivers/pci/dmar.c b/drivers/pci/dmar.c
> > index f941f60..8bf86ae 100644
> > --- a/drivers/pci/dmar.c
> > +++ b/drivers/pci/dmar.c
> > @@ -317,10 +317,8 @@ int __init dmar_table_init(void)
> >  		return -ENODEV;
> >  	}
> >  
> > -	if (list_empty(&dmar_rmrr_units)) {
> > +	if (list_empty(&dmar_rmrr_units))
> >  		printk(KERN_INFO PREFIX "No RMRR found\n");
> > -		return -ENODEV;
> > -	}
> >  
> >  	return 0;
> >  }
> 
> >From your description I am unable to work out how important this fix is.
> 
> Is it needed in 2.6.25.x?  In 2.6.26?  In 2.6.27?  There's no way for me to
> know :(

Actually this is the same as Suresh' x2apic patch 05/26. The absence of RMRRs
will prevent kernel from parsing IOxAPICs without this patch and thus leaves
IOxAPICs in an uninitialized state.

Re: [PATCH] Intel IOMMU: RMRRs do not necessarily have to be present on all VT-d capable platforms

[Andrew Morton]

On Thu, 10 Jul 2008 19:53:26 -0400 Yong Wang <yong.y.wang@linux.intel.com> wrote:

> On Thu, Jul 10, 2008 at 01:01:49AM -0700, Andrew Morton wrote:
> > On Tue, 1 Jul 2008 05:13:54 -0400 Yong Wang <yong.y.wang@linux.intel.com> wrote:
> > 
> > > RMRRs do not necessarily have to be present on all VT-d capable platforms.
> > > The printk is just informational and does not need to be followed by an
> > > error return.
> > > 
> > > Signed-off-by: Yong Y Wang <yong.y.wang@intel.com>
> > > ---
> > >  dmar.c |    4 +---
> > >  1 file changed, 1 insertion(+), 3 deletions(-)
> > > 
> > > diff --git a/drivers/pci/dmar.c b/drivers/pci/dmar.c
> > > index f941f60..8bf86ae 100644
> > > --- a/drivers/pci/dmar.c
> > > +++ b/drivers/pci/dmar.c
> > > @@ -317,10 +317,8 @@ int __init dmar_table_init(void)
> > >  		return -ENODEV;
> > >  	}
> > >  
> > > -	if (list_empty(&dmar_rmrr_units)) {
> > > +	if (list_empty(&dmar_rmrr_units))
> > >  		printk(KERN_INFO PREFIX "No RMRR found\n");
> > > -		return -ENODEV;
> > > -	}
> > >  
> > >  	return 0;
> > >  }
> > 
> > >From your description I am unable to work out how important this fix is.
> > 
> > Is it needed in 2.6.25.x?  In 2.6.26?  In 2.6.27?  There's no way for me to
> > know :(
> 
> Actually this is the same as Suresh' x2apic patch 05/26. The absence of RMRRs
> will prevent kernel from parsing IOxAPICs without this patch and thus leaves
> IOxAPICs in an uninitialized state.

And what runtime effect does that have?

Re: [PATCH] Intel IOMMU: RMRRs do not necessarily have to be present on all VT-d capable platforms

[Suresh Siddha]

On Thu, Jul 10, 2008 at 05:32:43PM -0700, Andrew Morton wrote:
> On Thu, 10 Jul 2008 19:53:26 -0400 Yong Wang <yong.y.wang@linux.intel.com> wrote:
> 
> > On Thu, Jul 10, 2008 at 01:01:49AM -0700, Andrew Morton wrote:
> > > On Tue, 1 Jul 2008 05:13:54 -0400 Yong Wang <yong.y.wang@linux.intel.com> wrote:
> > >
> > > > RMRRs do not necessarily have to be present on all VT-d capable platforms.
> > > > The printk is just informational and does not need to be followed by an
> > > > error return.
> > > >
> > > > Signed-off-by: Yong Y Wang <yong.y.wang@intel.com>
> > > > ---
> > > >  dmar.c |    4 +---
> > > >  1 file changed, 1 insertion(+), 3 deletions(-)
> > > >
> > > > diff --git a/drivers/pci/dmar.c b/drivers/pci/dmar.c
> > > > index f941f60..8bf86ae 100644
> > > > --- a/drivers/pci/dmar.c
> > > > +++ b/drivers/pci/dmar.c
> > > > @@ -317,10 +317,8 @@ int __init dmar_table_init(void)
> > > >           return -ENODEV;
> > > >   }
> > > >
> > > > - if (list_empty(&dmar_rmrr_units)) {
> > > > + if (list_empty(&dmar_rmrr_units))
> > > >           printk(KERN_INFO PREFIX "No RMRR found\n");
> > > > -         return -ENODEV;
> > > > - }
> > > >
> > > >   return 0;
> > > >  }
> > >
> > > >From your description I am unable to work out how important this fix is.
> > >
> > > Is it needed in 2.6.25.x?  In 2.6.26?  In 2.6.27?  There's no way for me to
> > > know :(
> >
> > Actually this is the same as Suresh' x2apic patch 05/26. The absence of RMRRs
> > will prevent kernel from parsing IOxAPICs without this patch and thus leaves
> > IOxAPICs in an uninitialized state.
> 
> And what runtime effect does that have?

Andrew, this is not urgent. It is now getting tracked as part of x2apic
patchset. And also, this doesn't need to be backported to any kernel.

thanks
suresh