apple_sep: unbinding warns twice and leaks the firmware IOMMU mapping

[brentkearney]

Unbinding apple_sep triggers two IOMMU WARNs and leaves the SEP firmware mapped in the device's DART. remove() unmaps the firmware region with its physical address instead of the IOVA that dma_map_resource() returned, so the unmap tears down page table entries that were never present and the real mapping survives.

drivers/soc/apple/sep.rs keeps only the shifted IOVA that MSG_BOOT_IMG4 needs and discards the IOVA itself:

let res = bindings::dma_map_resource(
    self.dev.as_raw(),
    self.region_params.addr,
    ...
);
self.fw_mapped.store(true, Relaxed);
res >> IOVA_SHIFT            // res is never stored

remove() then passes the physical address of the reserved region:

bindings::dma_unmap_resource(
    self.dev.as_raw(),
    self.region_params.addr, // should be the IOVA from dma_map_resource()
    self.region_params.size,
    ...
);

Impact

• Two WARNs per unbind, which taint the kernel.
• The firmware region stays mapped in the SEP's DART until reboot.
• Reachable on shipping configurations: the sep node is enabled by default on t8103-j293, t8103-j313, t8103-j456, and t8103-j457.
• Anyone debugging the SEP hits this immediately, because unbind and rebind is the natural way to re-run probe().

Steps to reproduce

1. Boot a machine whose sep node is enabled and confirm the driver is bound:

   ls -l /sys/bus/platform/devices/396400000.sep/driver
   

2. Unbind the driver:

   echo 396400000.sep | sudo tee /sys/bus/platform/drivers/apple_sep/unbind
   

3. Read dmesg.

Expected: the firmware mapping is torn down and dmesg stays quiet.

Actual: two WARNs, and the mapping is still present.

WARNING: drivers/iommu/io-pgtable-dart.c:308 at dart_unmap_pages+0x11c/0x140, CPU#1
 pc : dart_unmap_pages+0x11c/0x140
 lr : dart_unmap_pages+0x44/0x140
  dart_unmap_pages+0x11c/0x140 (P)
  apple_dart_unmap_pages+0x24/0x30
  __iommu_dma_unmap+0x94/0x140
  iommu_dma_unmap_phys+0xf8/0x118
  _RNvMs0_...3sep9SepDriverEE20post_unbind_callback+0x34/0x70

WARNING: drivers/iommu/dma-iommu.c:844 at __iommu_dma_unmap+0x128/0x140, CPU#1
 pc : __iommu_dma_unmap+0x128/0x140
 lr : __iommu_dma_unmap+0x94/0x140
  __iommu_dma_unmap+0x128/0x140 (P)
  iommu_dma_unmap_phys+0xf8/0x118
  _RNvMs0_...3sep9SepDriverEE20post_unbind_callback+0x34/0x70

The reserved region on this machine is at 0x10006340000, outside the device's DART address space, which is why the unmap finds nothing:

OF: reserved mem: 0x0000010006340000..0x00000100068b7fff (5600 KiB) map non-reusable sep-firmware

A rebind afterwards sends MSG_BOOT_TZ0 and never receives an ack, because SEPOS is already running. That is expected, but it makes unbind one-way.

Environment

Machine	Apple MacBook Pro (16-inch, 2021), apple,j316s / apple,t6000
Kernel	7.1.6-1-1-ARCH (linux-asahi 7.1.6.asahi1-1), aarch64
Distribution	Arch Linux ARM (Omarchy)
Config	CONFIG_APPLE_SEP=y

t600x does not enable the sep node, so reproducing here required adding aliases { sep = &sep; } and status = "okay" to t6000-j316s.dtb. The bug is in the unmap path and does not depend on that change.

Fix

Store the IOVA that dma_map_resource() returns and unmap that. Fix in #592.