On Thu, Jul 10, 2025, at 07:59, Nicolin Chen wrote:

 EXPORT_SYMBOL_NS_GPL(_iommufd_object_undepend, "IOMMUFD");

+/*
+ * Allocate an @offset to return to user space to use for an mmap() 
syscall
+ *
+ * Driver should use a per-structure helper in include/linux/iommufd.h
+ */
+int _iommufd_alloc_mmap(struct iommufd_ctx *ictx, struct 
iommufd_object *owner,
+			phys_addr_t mmio_addr, size_t length,
+			unsigned long *offset)
+{

...

+
+	/* Skip the first page to ease caller identifying the returned offset 
*/
+	rc = mtree_alloc_range(&ictx->mt_mmap, &startp, immap, immap->length,
+			       PAGE_SIZE, PHYS_ADDR_MAX, GFP_KERNEL);

This produces a warning on 32-bit targets with a 64-bit phys_addr_t,
in practice this would be ARM Cortex-A15 or -A17 systems:

In file included from include/linux/overflow.h:6,
                 ...
                 from drivers/iommu/iommufd/driver.c:4:
drivers/iommu/iommufd/driver.c: In function '_iommufd_alloc_mmap':
include/linux/limits.h:11:25: error: conversion from 'long long unsigned int' to 'long unsigned int' changes value from '18446744073709551615' to '4294967295' [-Werror=overflow]
   11 | #define PHYS_ADDR_MAX   (~(phys_addr_t)0)
      |                         ^~~~~~~~~~~~~~~~~
drivers/iommu/iommufd/driver.c:61:43: note: in expansion of macro 'PHYS_ADDR_MAX'
   61 |                                PAGE_SIZE, PHYS_ADDR_MAX, GFP_KERNEL);
      |                                           ^~~~~~~~~~~~~


The mtree_alloc_range() interface explicitly operates on 'unsigned long'
address values, so I don't see an immediate workaround for this that would
make it work on these machines. On the other hand, 

At the moment, the only drivers that select CONFIG_IOMMUFD_DRIVER
on 32-bit Arm systems are CONFIG_PDS_VFIO_PCI and CONFIG_MLX5_VFIO_PCI.
It's probably fine to make all three symbols "depends on 64BIT" for
now, but I don't know if there may be more drivers like this in the
future that actually could make sense on embedded systems.

     Arnd