Re: [PATCH v4 10/19] iommu: Add set_platform_dma_ops iommu ops
From: Baolu Lu
Date: Fri Jan 06 2023 - 21:48:59 EST
On 1/6/2023 10:26 PM, Jason Gunthorpe wrote:
On Fri, Jan 06, 2023 at 02:07:32PM +0800, Baolu Lu wrote:
Above looks good to me. Thanks! I have updated this part of code like
below:
@@ -2177,10 +2188,20 @@ static int __iommu_group_set_domain(struct
iommu_group *group,
* platform specific behavior.
*/
if (!new_domain) {
- if (WARN_ON(!group->domain->ops->detach_dev))
- return -EINVAL;
- __iommu_group_for_each_dev(group, group->domain,
- iommu_group_do_detach_device);
+ struct group_device *grp_dev;
+
+ grp_dev = list_first_entry(&group->devices,
+ struct group_device, list);
It seems OK - I hope we naturally can't ever get in a situation where
a group has disjoint iommu drivers.
The final code after cleanup looks like below. We will WARN_ON the lack
of callback in the iommu_group_do_set_platform_dma() helper.
2152 static int iommu_group_do_set_platform_dma(struct device *dev, void
*data)
2153 {
2154 const struct iommu_ops *ops = dev_iommu_ops(dev);
2155
2156 if (!WARN_ON(!ops->set_platform_dma_ops))
2157 ops->set_platform_dma_ops(dev);
2158
2159 return 0;
2160 }
2161
2162 static int __iommu_group_set_domain(struct iommu_group *group,
2163 struct iommu_domain *new_domain)
2164 {
2165 int ret;
2166
2167 if (group->domain == new_domain)
2168 return 0;
2169
2170 /*
2171 * New drivers should support default domains, so
set_platform_dma()
2172 * op will never be called. Otherwise the NULL domain
represents some
2173 * platform specific behavior.
2174 */
2175 if (!new_domain) {
2176 __iommu_group_for_each_dev(group, NULL,
2177
iommu_group_do_set_platform_dma);
2178 group->domain = NULL;
2179 return 0;
2180 }
How do you like this?
--
Best regards,
baolu