Re: [PATCH v2] iommu/of: Fix pci_request_acs() before enumerating PCI devices

[Xingang Wang <wangxingang5@huawei.com>]

On 2021/5/20 3:14, Rob Herring wrote:
> On Mon, May 17, 2021 at 01:17:05PM +0000, Wang Xingang wrote:
>> From: Xingang Wang <wangxingang5@huawei.com>
>>
>> When booting with devicetree, the pci_request_acs() is called after the
>> enumeration and initialization of PCI devices, thus the ACS is not
>> enabled. This patch add check for IOMMU in of_core_init(), and call
>> pci_request_acs() when iommu is detected, making sure that the ACS will
>> be enabled.
>>
>> Fixes: 6bf6c24720d33 ("iommu/of: Request ACS from the PCI core when
>> configuring IOMMU linkage")
>> Signed-off-by: Xingang Wang <wangxingang5@huawei.com>
>> ---
>>   drivers/iommu/of_iommu.c | 1 -
>>   drivers/of/base.c        | 9 ++++++++-
>>   2 files changed, 8 insertions(+), 2 deletions(-)
>>
>> Change log:
>> v1->v2:
>>   - remove pci_request_acs() in of_iommu_configure
>>   - check and call pci_request_acs() in of_core_init()
>>
>> diff --git a/drivers/iommu/of_iommu.c b/drivers/iommu/of_iommu.c
>> index a9d2df001149..54a14da242cc 100644
>> --- a/drivers/iommu/of_iommu.c
>> +++ b/drivers/iommu/of_iommu.c
>> @@ -205,7 +205,6 @@ const struct iommu_ops *of_iommu_configure(struct device *dev,
>>   			.np = master_np,
>>   		};
>>   
>> -		pci_request_acs();
>>   		err = pci_for_each_dma_alias(to_pci_dev(dev),
>>   					     of_pci_iommu_init, &info);
>>   	} else {
>> diff --git a/drivers/of/base.c b/drivers/of/base.c
>> index 48e941f99558..95cd8f0e5435 100644
>> --- a/drivers/of/base.c
>> +++ b/drivers/of/base.c
>> @@ -24,6 +24,7 @@
>>   #include <linux/of.h>
>>   #include <linux/of_device.h>
>>   #include <linux/of_graph.h>
>> +#include <linux/pci.h>
>>   #include <linux/spinlock.h>
>>   #include <linux/slab.h>
>>   #include <linux/string.h>
>> @@ -166,7 +167,7 @@ void __of_phandle_cache_inv_entry(phandle handle)
>>   void __init of_core_init(void)
>>   {
>>   	struct device_node *np;
>> -
>> +	bool of_iommu_detect = false;
>>   
>>   	/* Create the kset, and register existing nodes */
>>   	mutex_lock(&of_mutex);
>> @@ -180,6 +181,12 @@ void __init of_core_init(void)
>>   		__of_attach_node_sysfs(np);
>>   		if (np->phandle && !phandle_cache[of_phandle_cache_hash(np->phandle)])
>>   			phandle_cache[of_phandle_cache_hash(np->phandle)] = np;
>> +
>> +		/* Detect IOMMU and make sure ACS will be enabled */
>> +		if (!of_iommu_detect && of_get_property(np, "iommu-map", NULL)) {
>> +			of_iommu_detect = true;
>> +			pci_request_acs();
>> +		}
> 
> Private DT internal init code doesn't seem like the right place for
> this. If this needs to be ordered WRT PCI device enumeration, then
> somewhere in the PCI host bridge or bus init code seems like the right
> place to me.
> 
> Also, shouldn't this be conditional on 'iommu-map' being in the host
> bridge or a parent or ??? rather than just any iommu-map anywhere in the
> DT.
> 

Thanks for your suggestion, I will fix this and move the modification to 
procedure of pci_host_common_probe(), and before pci_host_probe().

>>   	}
>>   	mutex_unlock(&of_mutex);
>>   
>> -- 
>> 2.19.1
>>
> .
>