Xen devel - May 2012 - [PATCH] xen: do not map the same GSI twice

PV on HVM guests map GSIs into event channels; at restore time the
event channels are resumed by restore_pirqs.
Device drivers might try to register the same GSI again through ACPI at
restore time, but the GSI has already been mapped and bound by
restore_pirqs.
This patch detects these situations and avoid mapping the same GSI
multiple times.

Signed-off-by: Stefano Stabellini <stefano.stabellini@eu.citrix.com>
---
 arch/x86/pci/xen.c   |    4 ++++
 drivers/xen/events.c |    5 +++--
 include/xen/events.h |    3 +++
 3 files changed, 10 insertions(+), 2 deletions(-)

diff --git a/arch/x86/pci/xen.c b/arch/x86/pci/xen.c
index 7415aa9..56ab749 100644
--- a/arch/x86/pci/xen.c
+++ b/arch/x86/pci/xen.c
@@ -64,6 +64,10 @@ static int xen_register_pirq(u32 gsi, int gsi_override, int
triggering,
 	int shareable = 0;
 	char *name;
 
+	irq = xen_irq_from_gsi(gsi);
+	if (irq > 0)
+		return irq;
+
 	if (set_pirq)
 		pirq = gsi;
 
diff --git a/drivers/xen/events.c b/drivers/xen/events.c
index 0a8a17c..6908e4c 100644
--- a/drivers/xen/events.c
+++ b/drivers/xen/events.c
@@ -611,7 +611,7 @@ static void disable_pirq(struct irq_data *data)
 	disable_dynirq(data);
 }
 
-static int find_irq_by_gsi(unsigned gsi)
+int xen_irq_from_gsi(unsigned gsi)
 {
 	struct irq_info *info;
 
@@ -625,6 +625,7 @@ static int find_irq_by_gsi(unsigned gsi)
 
 	return -1;
 }
+EXPORT_SYMBOL_GPL(xen_irq_from_gsi);
 
 /*
  * Do not make any assumptions regarding the relationship between the
@@ -644,7 +645,7 @@ int xen_bind_pirq_gsi_to_irq(unsigned gsi,
 
 	mutex_lock(&irq_mapping_update_lock);
 
-	irq = find_irq_by_gsi(gsi);
+	irq = xen_irq_from_gsi(gsi);
 	if (irq != -1) {
 		printk(KERN_INFO "xen_map_pirq_gsi: returning irq %d for gsi %u\n",
 		       irq, gsi);
diff --git a/include/xen/events.h b/include/xen/events.h
index 0f77370..b332bd7 100644
--- a/include/xen/events.h
+++ b/include/xen/events.h
@@ -103,6 +103,9 @@ int xen_irq_from_pirq(unsigned pirq);
 /* Return the pirq allocated to the irq. */
 int xen_pirq_from_irq(unsigned irq);
 
+/* Return the irq allocated to the gsi */
+int xen_irq_from_gsi (unsigned gsi);
+
 /* Determine whether to ignore this IRQ if it is passed to a guest. */
 int xen_test_irq_shared(int irq);
 
-- 
1.7.2.5

On Mon, May 21, 2012 at 04:54:10PM +0100, Stefano Stabellini
wrote:
> PV on HVM guests map GSIs into event channels; at restore time the
> event channels are resumed by restore_pirqs.
> Device drivers might try to register the same GSI again through ACPI at
> restore time, but the GSI has already been mapped and bound by
> restore_pirqs.
Which means... what kind of error do we get without this patch?
> This patch detects these situations and avoid mapping the same GSI
> multiple times.
> 
> Signed-off-by: Stefano Stabellini <stefano.stabellini@eu.citrix.com>
> ---
>  arch/x86/pci/xen.c   |    4 ++++
>  drivers/xen/events.c |    5 +++--
>  include/xen/events.h |    3 +++
>  3 files changed, 10 insertions(+), 2 deletions(-)
> 
> diff --git a/arch/x86/pci/xen.c b/arch/x86/pci/xen.c
> index 7415aa9..56ab749 100644
> --- a/arch/x86/pci/xen.c
> +++ b/arch/x86/pci/xen.c
> @@ -64,6 +64,10 @@ static int xen_register_pirq(u32 gsi, int gsi_override,
int triggering,
>  	int shareable = 0;
>  	char *name;
>  
> +	irq = xen_irq_from_gsi(gsi);
> +	if (irq > 0)
> +		return irq;
> +
>  	if (set_pirq)
>  		pirq = gsi;
>  
> diff --git a/drivers/xen/events.c b/drivers/xen/events.c
> index 0a8a17c..6908e4c 100644
> --- a/drivers/xen/events.c
> +++ b/drivers/xen/events.c
> @@ -611,7 +611,7 @@ static void disable_pirq(struct irq_data *data)
>  	disable_dynirq(data);
>  }
>  
> -static int find_irq_by_gsi(unsigned gsi)
> +int xen_irq_from_gsi(unsigned gsi)
>  {
>  	struct irq_info *info;
>  
> @@ -625,6 +625,7 @@ static int find_irq_by_gsi(unsigned gsi)
>  
>  	return -1;
>  }
> +EXPORT_SYMBOL_GPL(xen_irq_from_gsi);
>  
>  /*
>   * Do not make any assumptions regarding the relationship between the
> @@ -644,7 +645,7 @@ int xen_bind_pirq_gsi_to_irq(unsigned gsi,
>  
>  	mutex_lock(&irq_mapping_update_lock);
>  
> -	irq = find_irq_by_gsi(gsi);
> +	irq = xen_irq_from_gsi(gsi);
>  	if (irq != -1) {
>  		printk(KERN_INFO "xen_map_pirq_gsi: returning irq %d for gsi
%u\n",
>  		       irq, gsi);
> diff --git a/include/xen/events.h b/include/xen/events.h
> index 0f77370..b332bd7 100644
> --- a/include/xen/events.h
> +++ b/include/xen/events.h
> @@ -103,6 +103,9 @@ int xen_irq_from_pirq(unsigned pirq);
>  /* Return the pirq allocated to the irq. */
>  int xen_pirq_from_irq(unsigned irq);
>  
> +/* Return the irq allocated to the gsi */
> +int xen_irq_from_gsi (unsigned gsi);
> +
>  /* Determine whether to ignore this IRQ if it is passed to a guest. */
>  int xen_test_irq_shared(int irq);
>  
> -- 
> 1.7.2.5

On Mon, 21 May 2012, Konrad Rzeszutek Wilk wrote:
> On Mon, May 21, 2012 at 04:54:10PM +0100, Stefano Stabellini wrote:
> > PV on HVM guests map GSIs into event channels; at restore time the
> > event channels are resumed by restore_pirqs.
> > Device drivers might try to register the same GSI again through ACPI
at
> > restore time, but the GSI has already been mapped and bound by
> > restore_pirqs.
> 
> Which means... what kind of error do we get without this patch?
Xen would print:

(XEN) irq.c:2235: dom4: pirq 23 or emuirq 28 already mapped

and waste a pirq

On Mon, May 21, 2012 at 06:47:44PM +0100, Stefano Stabellini
wrote:
> On Mon, 21 May 2012, Konrad Rzeszutek Wilk wrote:
> > On Mon, May 21, 2012 at 04:54:10PM +0100, Stefano Stabellini wrote:
> > > PV on HVM guests map GSIs into event channels; at restore time
the
> > > event channels are resumed by restore_pirqs.
> > > Device drivers might try to register the same GSI again through
ACPI at
> > > restore time, but the GSI has already been mapped and bound by
> > > restore_pirqs.
> > 
> > Which means... what kind of error do we get without this patch?
> 
> Xen would print:
> 
> (XEN) irq.c:2235: dom4: pirq 23 or emuirq 28 already mapped
> 
> and waste a pirq
OK. This sounds like it has been a bug since ..oh forever then?

On Mon, 21 May 2012, Konrad Rzeszutek Wilk wrote:
> On Mon, May 21, 2012 at 06:47:44PM +0100, Stefano Stabellini wrote:
> > On Mon, 21 May 2012, Konrad Rzeszutek Wilk wrote:
> > > On Mon, May 21, 2012 at 04:54:10PM +0100, Stefano Stabellini
wrote:
> > > > PV on HVM guests map GSIs into event channels; at restore
time the
> > > > event channels are resumed by restore_pirqs.
> > > > Device drivers might try to register the same GSI again
through ACPI at
> > > > restore time, but the GSI has already been mapped and bound
by
> > > > restore_pirqs.
> > > 
> > > Which means... what kind of error do we get without this patch?
> > 
> > Xen would print:
> > 
> > (XEN) irq.c:2235: dom4: pirq 23 or emuirq 28 already mapped
> > 
> > and waste a pirq
> 
> OK. This sounds like it has been a bug since ..oh forever then?
> 
Yes. It only showed up since the ata_piix driver started registering the
GSI again on resume (I don''t know when this behavior started exactly).