如何判断iommu/smmu 是否能用

/sys/class/iommu   ??

/sys/kernel/iommu_groups  ??   

vfio API ??

转自  https://blog.csdn.net/tiantao2012/article/details/60756851        ？？？？？

https://stackoverflow.com/questions/44286683/check-for-iommu-support-on-linux

我认为不能单单从 /sys/kernel/iommu_groups 下面是否为空来判断，因为在没有iommu的时候，若使用vfio+noiommu模式(echo 1 > /sys/module/vfio/parameters/enable_unsafe_noiommu_mode),  vfio 也会在 iommu_groups下面添加设备，此时该目录下非空，但其实iommu不能用。

struct iommu_group *vfio_iommu_group_get(struct device *dev)
{
 struct iommu_group *group;
 int __maybe_unused ret;

 group = iommu_group_get(dev);

#ifdef CONFIG_VFIO_NOIOMMU
 /*
  * With noiommu enabled, an IOMMU group will be created for a device
  * that doesn't already have one and doesn't have an iommu_ops on their
  * bus.  We set iommudata simply to be able to identify these groups
  * as special use and for reclamation later.
  */
 if (group || !noiommu || iommu_present(dev->bus))
  return group;

 group = iommu_group_alloc();
 if (IS_ERR(group))
  return NULL;

 iommu_group_set_name(group, "vfio-noiommu");
 iommu_group_set_iommudata(group, &noiommu, NULL);
 ret = iommu_group_add_device(group, dev);
 if (ret) {
  iommu_group_put(group);
  return NULL;
 }

 /*
  * Where to taint?  At this point we've added an IOMMU group for a
  * device that is not backed by iommu_ops, therefore any iommu_
  * callback using iommu_ops can legitimately Oops.  So, while we may
  * be about to give a DMA capable device to a user without IOMMU
  * protection, which is clearly taint-worthy, let's go ahead and do
  * it here.
  */
 add_taint(TAINT_USER, LOCKDEP_STILL_OK);
 dev_warn(dev, "Adding kernel taint for vfio-noiommu group on device\n");
#endif

 return group;
}