[PATCH] ahci: don't use MSI when PCI is disabled

[PATCH] ahci: don't use MSI when PCI is disabled

[Arnd Bergmann]

The AHCI driver is used for some on-chip devices that do not
use PCI for probing, and it can be built even when CONFIG_PCI
is disabled, but that now results in a build failure:

ata/libahci.c: In function 'ahci_host_activate_multi_irqs':
ata/libahci.c:2475:4: error: invalid use of undefined type 'struct msix_entry'
ata/libahci.c:2475:21: error: dereferencing pointer to incomplete type 'struct msix_entry'

This adds an #ifdef so we use that code only if actual PCI MSI
is available.

Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Fixes: d684a90d38e2 ("ahci: per-port msix support")

diff --git a/drivers/ata/libahci.c b/drivers/ata/libahci.c
index 1b6c7cc415bf..54fe01d8145b 100644
--- a/drivers/ata/libahci.c
+++ b/drivers/ata/libahci.c
@@ -1796,7 +1796,8 @@ static void ahci_port_intr(struct ata_port *ap)
 	ahci_handle_port_interrupt(ap, port_mmio, status);
 }
 
-static irqreturn_t ahci_multi_irqs_intr_hard(int irq, void *dev_instance)
+static irqreturn_t __maybe_unused ahci_multi_irqs_intr_hard(int irq,
+							    void *dev_instance)
 {
 	struct ata_port *ap = dev_instance;
 	void __iomem *port_mmio = ahci_port_base(ap);
@@ -2454,6 +2455,7 @@ void ahci_set_em_messages(struct ahci_host_priv *hpriv,
 }
 EXPORT_SYMBOL_GPL(ahci_set_em_messages);
 
+#ifdef CONFIG_PCI_MSI
 static int ahci_host_activate_multi_irqs(struct ata_host *host,
 					 struct scsi_host_template *sht)
 {
@@ -2492,6 +2494,7 @@ static int ahci_host_activate_multi_irqs(struct ata_host *host,
 
 	return ata_host_register(host, sht);
 }
+#endif
 
 /**
  *	ahci_host_activate - start AHCI host, request IRQs and register it
@@ -2510,9 +2513,12 @@ int ahci_host_activate(struct ata_host *host, struct scsi_host_template *sht)
 	int irq = hpriv->irq;
 	int rc;
 
+#ifdef CONFIG_PCI_MSI
 	if (hpriv->flags & (AHCI_HFLAG_MULTI_MSI | AHCI_HFLAG_MULTI_MSIX))
 		rc = ahci_host_activate_multi_irqs(host, sht);
-	else if (hpriv->flags & AHCI_HFLAG_EDGE_IRQ)
+	else
+#endif
+	if (hpriv->flags & AHCI_HFLAG_EDGE_IRQ)
 		rc = ata_host_activate(host, irq, ahci_single_edge_irq_intr,
 				       IRQF_SHARED, sht);
 	else

[PATCH] ahci: don't use MSI when PCI is disabled

[Dan Williams]

On Mon, Nov 23, 2015 at 1:47 AM, Arnd Bergmann <arnd@arndb.de> wrote:
> The AHCI driver is used for some on-chip devices that do not
> use PCI for probing, and it can be built even when CONFIG_PCI
> is disabled, but that now results in a build failure:
>
> ata/libahci.c: In function 'ahci_host_activate_multi_irqs':
> ata/libahci.c:2475:4: error: invalid use of undefined type 'struct msix_entry'
> ata/libahci.c:2475:21: error: dereferencing pointer to incomplete type 'struct msix_entry'
>
> This adds an #ifdef so we use that code only if actual PCI MSI
> is available.
>
> Signed-off-by: Arnd Bergmann <arnd@arndb.de>
> Fixes: d684a90d38e2 ("ahci: per-port msix support")
>
> diff --git a/drivers/ata/libahci.c b/drivers/ata/libahci.c
> index 1b6c7cc415bf..54fe01d8145b 100644
> --- a/drivers/ata/libahci.c
> +++ b/drivers/ata/libahci.c
> @@ -1796,7 +1796,8 @@ static void ahci_port_intr(struct ata_port *ap)
>         ahci_handle_port_interrupt(ap, port_mmio, status);
>  }
>
> -static irqreturn_t ahci_multi_irqs_intr_hard(int irq, void *dev_instance)
> +static irqreturn_t __maybe_unused ahci_multi_irqs_intr_hard(int irq,
> +                                                           void *dev_instance)
>  {
>         struct ata_port *ap = dev_instance;
>         void __iomem *port_mmio = ahci_port_base(ap);
> @@ -2454,6 +2455,7 @@ void ahci_set_em_messages(struct ahci_host_priv *hpriv,
>  }
>  EXPORT_SYMBOL_GPL(ahci_set_em_messages);
>
> +#ifdef CONFIG_PCI_MSI
>  static int ahci_host_activate_multi_irqs(struct ata_host *host,
>                                          struct scsi_host_template *sht)
>  {
> @@ -2492,6 +2494,7 @@ static int ahci_host_activate_multi_irqs(struct ata_host *host,
>
>         return ata_host_register(host, sht);
>  }
> +#endif
>
>  /**
>   *     ahci_host_activate - start AHCI host, request IRQs and register it
> @@ -2510,9 +2513,12 @@ int ahci_host_activate(struct ata_host *host, struct scsi_host_template *sht)
>         int irq = hpriv->irq;
>         int rc;
>
> +#ifdef CONFIG_PCI_MSI
>         if (hpriv->flags & (AHCI_HFLAG_MULTI_MSI | AHCI_HFLAG_MULTI_MSIX))
>                 rc = ahci_host_activate_multi_irqs(host, sht);
> -       else if (hpriv->flags & AHCI_HFLAG_EDGE_IRQ)
> +       else
> +#endif
> +       if (hpriv->flags & AHCI_HFLAG_EDGE_IRQ)
>                 rc = ata_host_activate(host, irq, ahci_single_edge_irq_intr,
>                                        IRQF_SHARED, sht);
>         else
>

If we had an "static inline ahci_irq_vector(int port)"  helper to
compile out the struct msix_entry de-reference would that be
sufficient?

[PATCH] ahci: don't use MSI when PCI is disabled

[Arnd Bergmann]

On Monday 23 November 2015 09:25:38 Dan Williams wrote:
> 
> If we had an "static inline ahci_irq_vector(int port)"  helper to
> compile out the struct msix_entry de-reference would that be
> sufficient?

Yes, that is probably a nicer way to do it. We should then also do
something like this:

@@ -2510,7 +2513,8 @@ int ahci_host_activate(struct ata_host *host, struct scsi_host_template *sht)
 	int irq = hpriv->irq;
 	int rc;
 
-	if (hpriv->flags & (AHCI_HFLAG_MULTI_MSI | AHCI_HFLAG_MULTI_MSIX))
+	if (IS_ENABLED(CONFIG_PCI_MSI) && \
+	    hpriv->flags & (AHCI_HFLAG_MULTI_MSI | AHCI_HFLAG_MULTI_MSIX))
 		rc = ahci_host_activate_multi_irqs(host, sht);
 	else if (hpriv->flags & AHCI_HFLAG_EDGE_IRQ)
 		rc = ata_host_activate(host, irq, ahci_single_edge_irq_intr,

which will let gcc leave out the entire ahci_host_activate_multi_irqs()
function but still flag compile errors in it even if CONFIG_PCI is
disabled.

	Arnd

[PATCH] ahci: don't use MSI when PCI is disabled

[Arnd Bergmann]

On Monday 23 November 2015 20:34:30 Arnd Bergmann wrote:
> On Monday 23 November 2015 09:25:38 Dan Williams wrote:
> > 
> > If we had an "static inline ahci_irq_vector(int port)"  helper to
> > compile out the struct msix_entry de-reference would that be
> > sufficient?
> 
> Yes, that is probably a nicer way to do it. We should then also do
> something like this:
> 
> @@ -2510,7 +2513,8 @@ int ahci_host_activate(struct ata_host *host, struct scsi_host_template *sht)
>         int irq = hpriv->irq;
>         int rc;
>  
> -       if (hpriv->flags & (AHCI_HFLAG_MULTI_MSI | AHCI_HFLAG_MULTI_MSIX))
> +       if (IS_ENABLED(CONFIG_PCI_MSI) && \
> +           hpriv->flags & (AHCI_HFLAG_MULTI_MSI | AHCI_HFLAG_MULTI_MSIX))
>                 rc = ahci_host_activate_multi_irqs(host, sht);
>         else if (hpriv->flags & AHCI_HFLAG_EDGE_IRQ)
>                 rc = ata_host_activate(host, irq, ahci_single_edge_irq_intr,
> 
> which will let gcc leave out the entire ahci_host_activate_multi_irqs()
> function but still flag compile errors in it even if CONFIG_PCI is
> disabled.
> 

Dan, are you going to do the ahci_irq_vector patch, or should we use my
original patch to fix up the build error?

Chen-Yu Tsai just stumbled over the same problem that I noticed, and I'm
sure others have run into it too.

	Arnd

[PATCH] ahci: don't use MSI when PCI is disabled

[Dan Williams]

On Fri, Dec 4, 2015 at 8:34 AM, Arnd Bergmann <arnd@arndb.de> wrote:
> On Monday 23 November 2015 20:34:30 Arnd Bergmann wrote:
>> On Monday 23 November 2015 09:25:38 Dan Williams wrote:
>> >
>> > If we had an "static inline ahci_irq_vector(int port)"  helper to
>> > compile out the struct msix_entry de-reference would that be
>> > sufficient?
>>
>> Yes, that is probably a nicer way to do it. We should then also do
>> something like this:
>>
>> @@ -2510,7 +2513,8 @@ int ahci_host_activate(struct ata_host *host, struct scsi_host_template *sht)
>>         int irq = hpriv->irq;
>>         int rc;
>>
>> -       if (hpriv->flags & (AHCI_HFLAG_MULTI_MSI | AHCI_HFLAG_MULTI_MSIX))
>> +       if (IS_ENABLED(CONFIG_PCI_MSI) && \
>> +           hpriv->flags & (AHCI_HFLAG_MULTI_MSI | AHCI_HFLAG_MULTI_MSIX))
>>                 rc = ahci_host_activate_multi_irqs(host, sht);
>>         else if (hpriv->flags & AHCI_HFLAG_EDGE_IRQ)
>>                 rc = ata_host_activate(host, irq, ahci_single_edge_irq_intr,
>>
>> which will let gcc leave out the entire ahci_host_activate_multi_irqs()
>> function but still flag compile errors in it even if CONFIG_PCI is
>> disabled.
>>
>
> Dan, are you going to do the ahci_irq_vector patch, or should we use my
> original patch to fix up the build error?

Ah, I was about to ask you the same question.  No worries, I'll send
out a patch with the above approach today.