[PATCH] iommu/of: Only do IOMMU lookup for available ones

[PATCH] iommu/of: Only do IOMMU lookup for available ones

[Jeffy Chen]

The for_each_matching_node_and_match() would return every matching
nodes including unavailable ones.

It's pointless to init unavailable IOMMUs, so add a sanity check to
avoid that.

Signed-off-by: Jeffy Chen <jeffy.chen@rock-chips.com>
---

 drivers/iommu/of_iommu.c | 3 +++
 1 file changed, 3 insertions(+)

diff --git a/drivers/iommu/of_iommu.c b/drivers/iommu/of_iommu.c
index 50947ebb6d17..6f7456caa30d 100644
--- a/drivers/iommu/of_iommu.c
+++ b/drivers/iommu/of_iommu.c
@@ -240,6 +240,9 @@ static int __init of_iommu_init(void)
 	for_each_matching_node_and_match(np, matches, &match) {
 		const of_iommu_init_fn init_fn = match->data;
 
+		if (!of_device_is_available(np))
+			continue;
+
 		if (init_fn && init_fn(np))
 			pr_err("Failed to initialise IOMMU %pOF\n", np);
 	}
-- 
2.11.0

Re: [PATCH] iommu/of: Only do IOMMU lookup for available ones

[Robin Murphy]

Hi Jeffy,

On 03/01/18 06:09, Jeffy Chen wrote:
> The for_each_matching_node_and_match() would return every matching
> nodes including unavailable ones.
> 
> It's pointless to init unavailable IOMMUs, so add a sanity check to
> avoid that.

Even better would be to clean up the last remaining init_fn user and get 
rid of the whole business. With the probe-deferral mechanism, early 
initialisation hooks are no longer needed, and the IOMMU_OF_DECLARE 
section really only remains as a way of detecting builtin drivers long 
before their registration initcalls run.

Robin.

> Signed-off-by: Jeffy Chen <jeffy.chen-TNX95d0MmH7DzftRWevZcw@public.gmane.org>
> ---
> 
>   drivers/iommu/of_iommu.c | 3 +++
>   1 file changed, 3 insertions(+)
> 
> diff --git a/drivers/iommu/of_iommu.c b/drivers/iommu/of_iommu.c
> index 50947ebb6d17..6f7456caa30d 100644
> --- a/drivers/iommu/of_iommu.c
> +++ b/drivers/iommu/of_iommu.c
> @@ -240,6 +240,9 @@ static int __init of_iommu_init(void)
>   	for_each_matching_node_and_match(np, matches, &match) {
>   		const of_iommu_init_fn init_fn = match->data;
>   
> +		if (!of_device_is_available(np))
> +			continue;
> +
>   		if (init_fn && init_fn(np))
>   			pr_err("Failed to initialise IOMMU %pOF\n", np);
>   	}
> 

Re: [PATCH] iommu/of: Only do IOMMU lookup for available ones

[Joerg Roedel]

On Wed, Jan 03, 2018 at 02:09:20PM +0800, Jeffy Chen wrote:
> The for_each_matching_node_and_match() would return every matching
> nodes including unavailable ones.
> 
> It's pointless to init unavailable IOMMUs, so add a sanity check to
> avoid that.
> 
> Signed-off-by: Jeffy Chen <jeffy.chen@rock-chips.com>
> ---
> 
>  drivers/iommu/of_iommu.c | 3 +++
>  1 file changed, 3 insertions(+)
> 
> diff --git a/drivers/iommu/of_iommu.c b/drivers/iommu/of_iommu.c
> index 50947ebb6d17..6f7456caa30d 100644
> --- a/drivers/iommu/of_iommu.c
> +++ b/drivers/iommu/of_iommu.c
> @@ -240,6 +240,9 @@ static int __init of_iommu_init(void)
>  	for_each_matching_node_and_match(np, matches, &match) {
>  		const of_iommu_init_fn init_fn = match->data;
>  
> +		if (!of_device_is_available(np))
> +			continue;
> +

Makes sense to me, but I'd like to have an OK from Robin or Will (added
to Cc) before applying this.



	Joerg

Re: [PATCH] iommu/of: Only do IOMMU lookup for available ones

[Will Deacon]

On Wed, Jan 17, 2018 at 02:28:08PM +0100, Joerg Roedel wrote:
> On Wed, Jan 03, 2018 at 02:09:20PM +0800, Jeffy Chen wrote:
> > The for_each_matching_node_and_match() would return every matching
> > nodes including unavailable ones.
> > 
> > It's pointless to init unavailable IOMMUs, so add a sanity check to
> > avoid that.
> > 
> > Signed-off-by: Jeffy Chen <jeffy.chen-TNX95d0MmH7DzftRWevZcw@public.gmane.org>
> > ---
> > 
> >  drivers/iommu/of_iommu.c | 3 +++
> >  1 file changed, 3 insertions(+)
> > 
> > diff --git a/drivers/iommu/of_iommu.c b/drivers/iommu/of_iommu.c
> > index 50947ebb6d17..6f7456caa30d 100644
> > --- a/drivers/iommu/of_iommu.c
> > +++ b/drivers/iommu/of_iommu.c
> > @@ -240,6 +240,9 @@ static int __init of_iommu_init(void)
> >  	for_each_matching_node_and_match(np, matches, &match) {
> >  		const of_iommu_init_fn init_fn = match->data;
> >  
> > +		if (!of_device_is_available(np))
> > +			continue;
> > +
> 
> Makes sense to me, but I'd like to have an OK from Robin or Will (added
> to Cc) before applying this.

I don't think this patch makes a lot of sense in isolation: the SMMU drivers
themselves will likely still probe, and it's unclear what we should about
DMA when an IOMMU is not deemed to be available. See:

https://patchwork.kernel.org/patch/9681211/

Jeffy -- are you solving a real issue here, or is this just an attempt at
some cleanup?

Will

Re: [PATCH] iommu/of: Only do IOMMU lookup for available ones

[JeffyChen]

Hi Will,

Thanks for your reply.

On 01/18/2018 10:41 PM, Will Deacon wrote:
>> >
>> >Makes sense to me, but I'd like to have an OK from Robin or Will (added
>> >to Cc) before applying this.
> I don't think this patch makes a lot of sense in isolation: the SMMU drivers
> themselves will likely still probe, and it's unclear what we should about
> DMA when an IOMMU is not deemed to be available. See:
>
> https://patchwork.kernel.org/patch/9681211/
>
> Jeffy -- are you solving a real issue here, or is this just an attempt at
> some cleanup?

hmmm, it's just a cleanup, but it seems like we already removed this 
of_iommu_init_fn in:
b0c560f7d8a4 iommu: Clean up of_iommu_init_fn

so this patch can be disregarded too :)
>
> Will
>
>
>