Re: [PATCH v3 3/7] iommu: register the DMA channel bus for IOMMU notifications

[sashiko-bot@kernel.org]

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
- [High] iommu: iommu_device_register() fails before arch_initcall(dma_bus_init)
--

commit 9240bcc4b7e8d99a1e2af1ff943ce2458c2e21eb
Author: Peng Fan <peng.fan@nxp.com>

iommu: register the DMA channel bus for IOMMU notifications

This commit adds dma_chan_bus_type (guarded by CONFIG_DMA_ENGINE) to
iommu_buses[] so the IOMMU core installs its bus notifier for DMA channel
devices.

> diff --git a/drivers/iommu/iommu.c b/drivers/iommu/iommu.c
> index b486b8bbd1fcd..6fa871bbd63ad 100644
> --- a/drivers/iommu/iommu.c
> +++ b/drivers/iommu/iommu.c

[ ... ]

> @@ -191,6 +192,9 @@ static const struct bus_type * const iommu_buses[] = {
>  #ifdef CONFIG_ARM_AMBA
>  	&amba_bustype,
>  #endif
> +#ifdef CONFIG_DMA_ENGINE
> +	&dma_chan_bus_type,
> +#endif

[Severity: High]
Could this cause iommu_device_register() to fail when called before
arch_initcall(dma_bus_init)?

The commit message notes that dma_chan_bus_type is registered at
arch_initcall before iommu_subsys_init() runs at subsys_initcall, but
iommu_buses[] is also iterated by iommu_device_register():

drivers/iommu/iommu.c:iommu_device_register() {
	...
	for (int i = 0; i < ARRAY_SIZE(iommu_buses) && !err; i++)
		err = bus_iommu_probe(iommu_buses[i]);
	if (err)
		iommu_device_unregister(iommu);
	...
}

And bus_iommu_probe() calls bus_for_each_dev(), which returns -EINVAL if
the bus has not been registered via bus_register() yet:

drivers/base/bus.c:bus_for_each_dev() {
	struct subsys_private *sp = bus_to_subsys(bus);
	...
	if (!sp)
		return -EINVAL;
}

Every other bus in iommu_buses[] is registered at postcore_initcall or
earlier. On PowerPC, spapr_tce_setup_phb_iommus_initcall() registers PHB
IOMMU devices at postcore_initcall_sync, before arch_initcall(dma_bus_init):

arch/powerpc/kernel/iommu.c:spapr_tce_setup_phb_iommus_initcall() {
	struct pci_controller *hose;

	list_for_each_entry(hose, &hose_list, list_node) {
		ppc_iommu_register_device(hose);
	}
	return 0;
}
postcore_initcall_sync(spapr_tce_setup_phb_iommus_initcall);

When CONFIG_DMA_ENGINE=y, wouldn't bus_iommu_probe(&dma_chan_bus_type)
return -EINVAL during spapr_tce_setup_phb_iommus_initcall(), causing
iommu_device_register() to unregister all PHB IOMMU devices?

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20261102-edma-iommu-v3-v3-0-9d0e62a954eb@Example.Com?part=3