tree: 9438363fa51e4a199115101f2340e3384fe178ed [path history] [tgz]
  1. test/
  2. Makefile.am.include
  3. mboxd_flash.cpp
  4. mboxd_lpc_reset.cpp
  5. mboxd_msg.cpp
  6. mboxd_msg.hpp
  7. mboxd_pnor_partition_table.cpp
  8. mboxd_pnor_partition_table.h
  9. pnor_partition.cpp
  10. pnor_partition.hpp
  11. pnor_partition_defs.h
  12. pnor_partition_table.cpp
  13. pnor_partition_table.hpp
  14. README.md
vpnor/README.md

Virtual PNOR Functionality

In the abstract, the virtual PNOR function shifts mboxd away from accessing raw flash to dynamically presenting a raw flash image to the host from a set of backing files.

Enabling the feature virtualises both the host's access to the flash (the mbox protocol), and the BMC's access to the flash (via some filesystem on top of the flash device).

Do I want to use this feature in my platform?

Maybe. It depends on how the image construction is managed, particularly the behaviour around writes from the host. It is likely the scheme will prevent firmware updates from being correctly applied when using flash tools on the host.

Use-case Requirements

Currently, the virtual PNOR implementation requires that:

  • The host expect an FFS layout (OpenPOWER systems)
  • The BMC provide a directory tree presenting the backing files in a hierarchy that reflects the partition properties in the FFS table of contents.

Implementation Behavioural Properties

  1. The FFS ToC defines the set of valid access ranges in terms of partitions
  2. The read-only property of partitions is enforced
  3. The ToC is considered read-only
  4. Read access to valid ranges must be granted
  5. Write access to valid ranges may be granted
  6. Access ranges that are valid may map into a backing file associated with the partition
  7. A read of a valid access range that maps into the backing file will render the data held in the backing file at the appropriate offset
  8. A read of a valid access range that does not map into the backing file will appear erased
  9. A read of an invalid access range will appear erased
  10. A write to a valid access range that maps into the backing file will update the data in the file at the appropriate offset
  11. A write to a valid access range that does not map into the backing file will expand the backing file to accommodate the write.
  12. A write to a valid access range may fail if the range is not marked as writeable. The error should be returned in response to the request to open the write window intersecting the read-only range.
  13. A write of an invalid access range will return an error. The error should be returned in response to the request to open the write window covering the invalid range.

The clarification on when the failure occurs in points 11 and 12 is useful for host-side error handling. Opening a write window gives the indication that future writes are expected to succeed, but in both cases we define them as always failing. Therefore we should not give the impression to the host that what it is asking for can be satisfied.