Question about the description of t2gpa usage scenario
Closed this issue · 2 comments
From my understanding, for a PCIe ATS Translation Request, the usage scenario of t2gpa when enabled would typically be like:
- firstly, untranslated request from device triggers IOMMU to perform the two-stage address translation, and GPA returns and gets cached in device ATC,
- then translated request from device accesses IOMMU with GPA to continue the second-stage translation, and SPA returns to device.
It seems like t2gpa means translating to gpa. Is the pattern above a fixed step-by-step usage flow? In other words, could we say it is expected a untranslated request would be immediately followed by a related translated request(with same gpa)? If this is the most common case, then I guess the translation cache is important for the second stage when t2gpa is enabled.
firstly, untranslated request from device triggers IOMMU to perform the two-stage address translation, and GPA returns and gets cached in device ATC,
Its only applicable to ATS Translation requests - not untranslated request. On a ATS Translation request, the IOMMU does an IOVA -> GPA -> SPA translation. However, it only returns the GPA and not the SPA in the completion to the device. The translation from GPA->SPA is done to ensure there are no faults and to determine accurate permissions.
A translation request in this usage would be followed by a translated requested using the translated address - a GPA in this configuration. The GPA in the translated request is then again translated to an SPA and the permissions verified.
That's clear. Thanks a lot.