| Message ID | 20230817234047.195194-1-baolu.lu@linux.intel.com (mailing list archive) |
|---|---|
| Headers | show |
| Series | iommu: Prepare to deliver page faults to user space | expand |
On Fri, Aug 18, 2023 at 07:40:36AM +0800, Lu Baolu wrote: > When a user-managed page table is attached to an IOMMU, it is necessary > to deliver IO page faults to user space so that they can be handled > appropriately. One use case for this is nested translation, which is > currently being discussed in the mailing list. > > I have posted a RFC series [1] that describes the implementation of > delivering page faults to user space through IOMMUFD. This series has > received several comments on the IOMMU refactoring, which I am trying to > address in this series. Looking at this after all the patches are applied.. iommu_report_device_fault() and iommu_queue_iopf() should be put in the same file. iommu_queue_iopf() seems misnamed since it isn't queuing anything. It is delivering the fault to the domain. It is weird that iommu_sva_domain_alloc is not in the sva file iopf_queue_work() wrappers a work queue, but it should trampoline through another function before invoking the driver's callback and not invoke it with a weird work_struct - decode the group and get back the domain. Every single handler will require the group and domain. Same for domain->iopf_handler, the domain should be an argument if we are invoking the function on a domain. Perhaps group->domain is a simple answer. Jason
On 2023/8/22 2:31, Jason Gunthorpe wrote: > On Fri, Aug 18, 2023 at 07:40:36AM +0800, Lu Baolu wrote: >> When a user-managed page table is attached to an IOMMU, it is necessary >> to deliver IO page faults to user space so that they can be handled >> appropriately. One use case for this is nested translation, which is >> currently being discussed in the mailing list. >> >> I have posted a RFC series [1] that describes the implementation of >> delivering page faults to user space through IOMMUFD. This series has >> received several comments on the IOMMU refactoring, which I am trying to >> address in this series. > > Looking at this after all the patches are applied.. Thank you very much for reviewing my patches. > > iommu_report_device_fault() and iommu_queue_iopf() should be put in > the same file. Yes. I will move both into io-pgfault.c. After that, iommu_queue_iopf() becomes static. > > iommu_queue_iopf() seems misnamed since it isn't queuing anything. It > is delivering the fault to the domain. Yeah, perhaps we can rename it to iommu_handle_iopf(). /** * iommu_handle_iopf - IO Page Fault handler * @fault: fault event * @dev: struct device. > > It is weird that iommu_sva_domain_alloc is not in the sva file Agreed. I will move it to iommu-sva.c. > iopf_queue_work() wrappers a work queue, but it should trampoline > through another function before invoking the driver's callback and not > invoke it with a weird work_struct - decode the group and get back the > domain. Every single handler will require the group and domain. The work queue wrapper is duplicate. I will remove it and let the driver to call queue_work() directly. > > Same for domain->iopf_handler, the domain should be an argument if we > are invoking the function on a domain. > > Perhaps group->domain is a simple answer. Yes. I will add domain in fault group and make it part of the parameters of the callback. Best regards, baolu