Re: [bug report] iommupt: Add map_pages op

[Jason Gunthorpe <jgg@ziepe.ca>]

On Tue, Nov 18, 2025 at 10:28:53AM +0300, Dan Carpenter wrote:
> Hello Jason Gunthorpe,
> 
> Commit dcd6a011a8d5 ("iommupt: Add map_pages op") from Nov 4, 2025
> (linux-next), leads to the following Smatch static checker warning:
> 
> 	drivers/iommu/generic_pt/fmt/../iommu_pt.h:752 increase_top()
> 	error: uninitialized symbol 'new_level'.
> 
> drivers/iommu/generic_pt/fmt/../iommu_pt.h
>     665 static int increase_top(struct pt_iommu *iommu_table, struct pt_range *range,
>     666                         struct pt_iommu_map_args *map)
>     667 {
>     668         struct iommu_pages_list free_list = IOMMU_PAGES_LIST_INIT(free_list);
>     669         struct pt_common *common = common_from_iommu(iommu_table);
>     670         uintptr_t top_of_table = READ_ONCE(common->top_of_table);
>     671         uintptr_t new_top_of_table = top_of_table;
>     672         struct pt_table_p *table_mem;
>     673         unsigned int new_level;
>     674         spinlock_t *domain_lock;
>     675         unsigned long flags;
>     676         int ret;
>     677 
>     678         while (true) {
>     679                 struct pt_range top_range =
>     680                         _pt_top_range(common, new_top_of_table);
>     681                 struct pt_state pts = pt_init_top(&top_range);
>     682 
>     683                 top_range.va = range->va;
>     684                 top_range.last_va = range->last_va;
>     685 
>     686                 if (!pt_check_range(&top_range) && map->leaf_level <= pts.level)
>     687                         break;
> 
> Smatch worries that we could break on the first iteration.  I don't
> know if that's possible.

Hmm, I wanted to say it can't because of the caller, but the locking
doesn't work like that, so yeah this is a bug. I will make a patch

Jason