Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
From:   Sven Peter <sven@svenpeter.dev>
To:     iommu@lists.linux-foundation.org
Cc:     Sven Peter <sven@svenpeter.dev>, Joerg Roedel <joro@8bytes.org>,
        Will Deacon <will@kernel.org>,
        Robin Murphy <robin.murphy@arm.com>,
        Arnd Bergmann <arnd@kernel.org>,
        Mohamed Mediouni <mohamed.mediouni@caramail.com>,
        Alexander Graf <graf@amazon.com>,
        Hector Martin <marcan@marcan.st>,
        Alyssa Rosenzweig <alyssa@rosenzweig.io>,
        linux-kernel@vger.kernel.org
Subject: [PATCH v2 0/8] Support IOMMU page sizes larger than the CPU page size
Date:   Sat, 28 Aug 2021 17:36:34 +0200
Message-Id: <20210828153642.19396-1-sven@svenpeter.dev>
MIME-Version: 1.0
Content-Transfer-Encoding: 8bit
Precedence: bulk

RFC Patch: https://lore.kernel.org/linux-iommu/20210806155523.50429-1-sven@svenpeter.dev/

Hi,

After a very helpful discussion with Robin Murphy on the RFC, here's v2 that is slowly
starting to look sane.
I've been running this code for two weeks now and mainly tested it with usb storage devices
connected to dwc3 and to xhci over pcie on the M1.

Some background: On the Apple M1 the IOMMUs are hardwired to only support 16 KB pages.
We'd still like to boot Linux with 4KB pages though because that's what most distros
ship these days. This patch series adds support for that setup to the IOMMU DMA API.

This is essentially done by always mapping the encapsulating IOMMU page and adjusting
the returned iova offset. There are also changes to only allow DMA domains to make use
of this and prevent UNMANAGED domains from encountering unexpected situations.

For untrusted devices the allocation size is simply aligned to iovad->granule if they
don't already go through the swiotlb path. I have not been able to test that part
so far though since there's no Thunderbolt support for the M1 yet.

The series is based on top of iommu/next (and without the last commit probably also on
iommu/core). It won't apply cleanly on apple/dart since it already takes Robin's DMA domain
cleanup series into account.


Best,

Sven
 
Sven Peter (8):
  iommu/dma: Align size for untrusted devs to IOVA granule
  iommu/dma: Fail unaligned map requests for untrusted devs
  iommu/dma: Disable get_sgtable for granule > PAGE_SIZE
  iommu/dma: Support granule > PAGE_SIZE in dma_map_sg
  iommu/dma: Support PAGE_SIZE < iovad->granule allocations
  iommu: Move IOMMU pagesize check to attach_device
  iommu: Introduce __IOMMU_DOMAIN_LP
  iommu/dart: Remove force_bypass logic

 drivers/iommu/apple-dart.c |  14 +--
 drivers/iommu/dma-iommu.c  | 172 ++++++++++++++++++++++++++++++++-----
 drivers/iommu/iommu.c      |  36 +++++++-
 drivers/iommu/iova.c       |   7 +-
 include/linux/iommu.h      |   8 +-
 5 files changed, 197 insertions(+), 40 deletions(-)

-- 
2.25.1