Re: [RFC PATCH v2 05/11] ACPI: platform: setup MSI domain for ACPI based platform device

[Marc Zyngier <marc.zyngier@arm.com>]

On 15/09/16 15:05, Hanjun Guo wrote:
> Hi Marc,
> 
> Thanks for your review, reply inline.
> 
> On 09/14/2016 11:45 PM, Marc Zyngier wrote:
>> On 14/09/16 15:21, Hanjun Guo wrote:
>>> From: Hanjun Guo <hanjun.guo@linaro.org>
>>>
>>> With the platform msi domain created, we can set up the msi domain
>>> for a platform device when it's probed.
>>>
>>> This patch introduces acpi_configure_msi_domain(), which retrieves
>>> the domain from iort and set it to platform device.
>>>
>>> As some platform devices such as an irqchip needs the msi irqdomain
>>> to be the interrupt parent domain, we need to get irqdomain before
>>> platform device is probed.
>>>
>>> Cc: Marc Zyngier <marc.zyngier@arm.com>
>>> Cc: Greg KH <gregkh@linuxfoundation.org>
>>> Cc: Thomas Gleixner <tglx@linutronix.de>
>>> Cc: Bjorn Helgaas <bhelgaas@google.com>
>>> Cc: Lorenzo Pieralisi <lorenzo.pieralisi@arm.com>
>>> Cc: Tomasz Nowicki <tn@semihalf.com>
>>> Signed-off-by: Hanjun Guo <hanjun.guo@linaro.org>
>>> ---
>>>   drivers/acpi/arm64/iort.c   |  5 ++++-
>>>   drivers/base/platform-msi.c | 15 ++++++++++++++-
>>>   drivers/base/platform.c     |  2 ++
>>>   include/linux/msi.h         |  1 +
>>>   4 files changed, 21 insertions(+), 2 deletions(-)
>>>
>>> diff --git a/drivers/acpi/arm64/iort.c b/drivers/acpi/arm64/iort.c
>>> index 13a1905..bccd3cc 100644
>>> --- a/drivers/acpi/arm64/iort.c
>>> +++ b/drivers/acpi/arm64/iort.c
>>> @@ -478,6 +478,7 @@ struct irq_domain *iort_get_device_domain(struct device *dev, u32 req_id)
>>>   {
>>>   	struct fwnode_handle *handle;
>>>   	int its_id;
>>> +	enum irq_domain_bus_token bus_token;
>>>
>>>   	if (iort_dev_find_its_id(dev, req_id, 0, &its_id))
>>>   		return NULL;
>>> @@ -486,7 +487,9 @@ struct irq_domain *iort_get_device_domain(struct device *dev, u32 req_id)
>>>   	if (!handle)
>>>   		return NULL;
>>>
>>> -	return irq_find_matching_fwnode(handle, DOMAIN_BUS_PCI_MSI);
>>> +	bus_token = dev_is_pci(dev) ?
>>> +			DOMAIN_BUS_PCI_MSI : DOMAIN_BUS_PLATFORM_MSI;
>>> +	return irq_find_matching_fwnode(handle, bus_token);
>>>   }
>>>
>>>   static int __get_pci_rid(struct pci_dev *pdev, u16 alias, void *data)
>>> diff --git a/drivers/base/platform-msi.c b/drivers/base/platform-msi.c
>>> index 279e539..f6eae18 100644
>>> --- a/drivers/base/platform-msi.c
>>> +++ b/drivers/base/platform-msi.c
>>> @@ -17,8 +17,8 @@
>>>    * along with this program.  If not, see <http://www.gnu.org/licenses/>.
>>>    */
>>>
>>> +#include <linux/acpi_iort.h>
>>>   #include <linux/device.h>
>>> -#include <linux/idr.h>
>>>   #include <linux/irq.h>
>>>   #include <linux/irqdomain.h>
>>>   #include <linux/msi.h>
>>> @@ -416,3 +416,16 @@ int platform_msi_domain_alloc(struct irq_domain *domain, unsigned int virq,
>>>
>>>   	return err;
>>>   }
>>> +
>>> +int acpi_configure_msi_domain(struct device *dev)
>>> +{
>>> +	struct irq_domain *d = NULL;
>>> +
>>> +	d = iort_get_device_domain(dev, 0);
>>
>> This looks completely wrong. Why RID 0? As far as I can see, 0 is not a
>> special value, and could be something else.
> 
> You are right. I tried to reuse the API of get irqdomain in IORT for
> PCI devices, but for platform device, we don't have req id in named
> component, so I just pass 0 here, I think I need to prepare another
> API for platform devices.
> 
>>
>>> +	if (d) {
>>> +		dev_set_msi_domain(dev, d);
>>> +		return 0;
>>> +	}
>>> +
>>> +	return -EINVAL;
>>> +}
>>
>> I really hate this, as the platform MSI code is intentionally free of
>> any firmware reference. This should live in the ACPI code.
> 
> Will do, I think locate it in iort.c is better.
> 
>>
>>> diff --git a/drivers/base/platform.c b/drivers/base/platform.c
>>> index 6482d47..ea01a37 100644
>>> --- a/drivers/base/platform.c
>>> +++ b/drivers/base/platform.c
>>> @@ -24,6 +24,7 @@
>>>   #include <linux/pm_domain.h>
>>>   #include <linux/idr.h>
>>>   #include <linux/acpi.h>
>>> +#include <linux/msi.h>
>>>   #include <linux/clk/clk-conf.h>
>>>   #include <linux/limits.h>
>>>   #include <linux/property.h>
>>> @@ -500,6 +501,7 @@ struct platform_device *platform_device_register_full(
>>>   	pdev->dev.parent = pdevinfo->parent;
>>>   	pdev->dev.fwnode = pdevinfo->fwnode;
>>>
>>> +	acpi_configure_msi_domain(&pdev->dev);
>>
>> It feels odd to put this in the generic code, while you could perfectly
>> put the call into acpi_platform.c and keep the firmware stuff away from
>> the generic code.
> 
> My feeling is the same, I'm still trying to find a new way to do it,
> but I can't simply put that in acpi_platform.c, because
> 
> acpi_create_platform_device()
>     platform_device_register_full()
> 	platform_device_alloc()  --> dev is alloced
>          ...
>          dev.fwnode  is set
> 	(I get the msi domain by the fwnode in acpi_configure_msi_domain)
>          ...
>          platform_device_add()  --> which the device is probed.
> 
> For devices like irqchip which needs the dev->msi_domain to be
> set before it's really probed, because it needs the msi domain
> to be the parent domain.
> 
> If I call the function in acpi_create_platform_device() before
> platform_device_register_full(), we just can't set dev's msi
> domain, but if call it after platform_device_register_full(),
> the irqchip like mbigen will not get its parent domain...
> 
> DT is using another API for platform device probe, so has no
> problems like I said above, any suggestions to do it right in
> ACPI?

How about having something that's completely generic and solves
the problem once and for all? Something like this:

diff --git a/drivers/base/platform.c b/drivers/base/platform.c
index 6482d47..6f0f90b 100644
--- a/drivers/base/platform.c
+++ b/drivers/base/platform.c
@@ -533,6 +533,9 @@ struct platform_device *platform_device_register_full(
 			goto err;
 	}
 
+	if (pdevinfo->pre_add_cb)
+		pdevinfo->pre_add_cb(&pdev->dev);
+
 	ret = platform_device_add(pdev);
 	if (ret) {
 err:
diff --git a/include/linux/platform_device.h b/include/linux/platform_device.h
index 98c2a7c..44ea133 100644
--- a/include/linux/platform_device.h
+++ b/include/linux/platform_device.h
@@ -74,6 +74,7 @@ struct platform_device_info {
 		u64 dma_mask;
 
 		struct property_entry *properties;
+		void (*pre_add_cb)(struct device *);
 };
 extern struct platform_device *platform_device_register_full(
 		const struct platform_device_info *pdevinfo);

Plug pre_add_cb with your ACPI callback where you can do all the
processing you want before the device is actually added.

Thanks,

	M.
-- 
Jazz is not dead. It just smells funny...