search

riscv-non-isa / riscv-iommu

RISC-V IOMMU Specification
https://jira.riscv.org/browse/RVG-55
Creative Commons Attribution 4.0 International
98 stars 17 forks source link

ATS Control Register Field STU and IOMMU #267

Closed just-for-fun-too closed 1 year ago

just-for-fun-too commented 1 year ago

STU is programmed in the ATS Control register of IO devices (Functions) which support ATS requests. It can have a value from 0 to 31 indicating 1 to 2^31 4096-byte blocks.

The Untranslated Address Field in an ATS Translation Request has to be checked for validity against the implied (address) range to see if it is valid or not. This range is computed using the STU for that IO device using the formula 2^(STU+12) + (Length/2). Length is available in the Translation Request Header, and as such available to the IOMMU, but STU is not. Questions:

1- How does the IOMMU know the STU associated with an ATS request since, it is device specific, but no such field exists in the ATS request or the Device Context? 2- IS STU a fixed constant in the system? But there are no registers defined in the IOMMU to program it. 3- Can we use the custom field in the Translation control (tc) field of the Device Context to store a device specific STU.

Or am I missing something basic on how this check works.

Thanks

Thanks

ved-rivos commented 1 year ago

The STU is the smallest page size that the first or second stage page based virtual memory system mode selected for that device. The virtual memory system modes supported by the IOMMU are provided in the capabilities register. In case of RISC-V this is 4 KiB for all virtual memory system modes. Hence, as part of enabling ATS in a device connected to a host with a RISC-V IOMMU, the ATS control register should be programmed with a value of 0.

Whether an IOMMU supports returning multiple translations is implementation specific and the rules are specified by PCIe. The implementation may do one of the following for such requests:

  1. Perform translation only for the address in the Untranslated address field and provide an appropriate response. In such case the completion has a length and byte count set to 8 bytes and the lower byte address in the completion set to RCB-8 indicating that it is the final completion. This is typical.
  2. Perform multiple translations starting at the untranslated address - in one or more completions. When using multiple completions then each completion must carry a translation for an address adjacent to that in the previous completion. The number of translations may be less than or equal to number of requested translations.
just-for-fun-too commented 1 year ago

Thank you very much for the answer regarding STU and also including the case of sending back multiple translations!