> From: Jason Gunthorpe <jgg@xxxxxxxxxx>
> Sent: Wednesday, August 9, 2023 1:27 AM
>
> The two callers coming from bus_tye -> dma_configure() already have the
s/bus_tye/bus_type/
> device_lock held for the device being probed. Mark this in the API so that
> the core code doesn't attempt to re-acquire the same lock and deadlock.
>
> __iommu_probe_device from iommu_probe_device+0x10/0x40
> iommu_probe_device from of_iommu_configure+0xf8/0x1c8
> of_iommu_configure from of_dma_configure_id+0x188/0x450
> of_dma_configure_id from platform_dma_configure+0x24/0x60
> platform_dma_configure from really_probe+0xac/0x3d4
> really_probe from __driver_probe_device+0xa0/0x1e8
> __driver_probe_device from driver_probe_device+0x30/0xd0
> driver_probe_device from __driver_attach+0x10c/0x190
> __driver_attach from bus_for_each_dev+0x60/0xb4
> bus_for_each_dev from bus_add_driver+0xe0/0x208
> bus_add_driver from driver_register+0x7c/0x118
> driver_register from exynos_drm_init+0xe0/0x14c
> exynos_drm_init from do_one_initcall+0x6c/0x318
> do_one_initcall from kernel_init_freeable+0x1c4/0x214
> kernel_init_freeable from kernel_init+0x18/0x12c
> kernel_init from ret_from_fork+0x14/0x2c
>
> Internally make __iommu_probe_device() require the caller to get the
> device_lock().
>
> The three remaining callers of iommu_probe_device() don't seem to hold the
> device_lock():
>
> powerpc/kernel/iommu.c: iommu_add_device()
> iommu/iommu.c: iommu_bus_notifier()/BUS_NOTIFY_ADD_DEVICE/
> intel/iommu.c: probe_acpi_namespace_devices()
>
> Fixes: a2dde836050f ("iommu: Complete the locking for dev->iommu_group")
> Reported-by: Marek Szyprowski <m.szyprowski@xxxxxxxxxxx>
> Closes: https://lore.kernel.org/linux-iommu/bc5d6aa8-e8ca-c896-2f39-
> af074d55e436@xxxxxxxxxxx
> Tested-by: Marek Szyprowski <m.szyprowski@xxxxxxxxxxx>
> Signed-off-by: Jason Gunthorpe <jgg@xxxxxxxxxx>
Reviewed-by: Kevin Tian <kevin.tian@xxxxxxxxx>
