[PATCH v6 00/14] iommu/amd: SVA Support (Part 1) - cleanup/refactoring

[Vasant Hegde <vasant.hegde@amd.com>]

This is part 1 of the 4-part series to introduce Share Virtual Address
(SVA) support, which focuses on cleaning up and refactoring the existing
code in preparation for subsequent series.

It contains the following enhancements:

* Patch 1 - 7:
  Clean up, refactoring and miscellaneous improvements.

* Patch 8 - 9:
  Use global functions to check iommu features

* Patch 10 - 14:
  Modify logic to independently enable PCI capabilities (ATS/PASID/PRI)
  for devices. This allows more flexibility in preparation for SVA and
  IOPF supports.

This patch series is based on top of linux v6.6-rc1.
Base commit : 0bb80ecc33a (Linux 6.6-rc1)

This is also available at github :
  https://github.com/AMDESE/linux/tree/iommu_sva_part1_v6_v6.6_rc1

Thanks Jason, Jerry, Robin for reviewing previous versions and providing valuable
feedbacks.


Changes from v5 -> v6:
  - Rebased on top v6.6-rc1
  - Added Review-by tags

v5: https://lore.kernel.org/linux-iommu/20230821104227.706997-1-vasant.hegde@amd.com/T/#t

Changes from v4 -> v5:
  - Changed dev_data.ppr from bool to bit field
  - Reverted sysfs_emit that was changing existing behaviour

v4 : https://lore.kernel.org/linux-iommu/20230815102202.565012-1-vasant.hegde@amd.com/T/#t

Changes from v3 -> v4:
  - Dropped patches that were touching iommu_v2.c as we will be deprecating
    iommu_v2 module.
  - Dropped Review-by tags for the patches which got modified (mostly the
    patches that were touching iommu_v2.c code).
  - Introduced new patch to use global EFR/EFR2 to check the iommu features
  - Pass enum to set_dte_entry() instead of bool
  - Used DIV_ROUND_UP instead of loop to calculate gcr3 levels

v3 : https://lore.kernel.org/linux-iommu/20230804064216.835544-1-vasant.hegde@amd.com/T/#t

Changes from v2 -> v3:
  - Removed fallthrough from switch-case
  - Added lockdep_assert_held() instead of comment on top of the function
  - Moved macro defination (PPR_HANDLER_IOPF) to the patch where its
    actually used
  - Use helper function to allocate memory instead of get_zeroed_page()
  - Converted boolean to bit fields

v2 : https://lore.kernel.org/linux-iommu/20230728053609.165183-1-vasant.hegde@amd.com/T/#t

Changes from v1 -> v2:
  - Dropped GCR3 related changes from Part1. We are reworking
    GCR3 management based on Jason's comment. We will post them
    as separate part.
  - Addressed review comment from Jason
  - Added iommu_dev_data.ppr to track PPR status

v1 : https://lore.kernel.org/linux-iommu/20230712141516.154144-1-vasant.hegde@amd.com/


Thank you,
Vasant / Suravee


Suravee Suthikulpanit (8):
  iommu/amd: Remove unused amd_io_pgtable.pt_root variable
  iommu/amd: Consolidate timeout pre-define to amd_iommu_type.h
  iommu/amd: Consolidate logic to allocate protection domain
  iommu/amd: Introduce helper functions for managing GCR3 table
  iommu/amd: Miscellaneous clean up when free domain
  iommu/amd: Consolidate feature detection and reporting logic
  iommu/amd: Modify logic for checking GT and PPR features
  iommu/amd: Introduce iommu_dev_data.ppr

Vasant Hegde (6):
  iommu/amd: Refactor protection domain allocation code
  iommu/amd: Do not set amd_iommu_pgtable in pass-through mode
  iommu/amd: Rename ats related variables
  iommu/amd: Introduce iommu_dev_data.flags to track device capabilities
  iommu/amd: Enable device ATS/PASID/PRI capabilities independently
  iommu/amd: Initialize iommu_device->max_pasids

 drivers/iommu/amd/amd_iommu.h       |  28 +-
 drivers/iommu/amd/amd_iommu_types.h |  31 +-
 drivers/iommu/amd/init.c            | 116 +++-----
 drivers/iommu/amd/io_pgtable_v2.c   |   8 +-
 drivers/iommu/amd/iommu.c           | 442 +++++++++++++++-------------
 5 files changed, 322 insertions(+), 303 deletions(-)

-- 
2.31.1