On Wed, Sep 27, 2023 at 11:59:21AM +0800, Liu Shixin wrote:
> Since kmemleak_alloc_phys() rather than kmemleak_alloc() was called from
> memblock_alloc_range_nid(), kmemleak_free_part_phys() should be used to
> delete kmemleak object in free_bootmem_page() and put_page_bootmem().
>
> Fixes: 028725e73375 ("bootmem: remove the vmemmap pages from kmemleak in free_bootmem_page")
> Fixes: dd0ff4d12dd2 ("bootmem: remove the vmemmap pages from kmemleak in put_page_bootmem")
> Signed-off-by: Liu Shixin <liushixin2@xxxxxxxxxx>
Acked-by: Catalin Marinas <catalin.marinas@xxxxxxx>
> include/linux/bootmem_info.h | 2 +-
> mm/bootmem_info.c | 2 +-
> 2 files changed, 2 insertions(+), 2 deletions(-)
>
> diff --git a/include/linux/bootmem_info.h b/include/linux/bootmem_info.h
> index e1a3c9c9754c..cffa38a73618 100644
> --- a/include/linux/bootmem_info.h
> +++ b/include/linux/bootmem_info.h
> @@ -60,7 +60,7 @@ static inline void get_page_bootmem(unsigned long info, struct page *page,
>
> static inline void free_bootmem_page(struct page *page)
> {
> - kmemleak_free_part(page_to_virt(page), PAGE_SIZE);
> + kmemleak_free_part_phys(PFN_PHYS(page_to_pfn(page)), PAGE_SIZE);
We have a page_to_phys(), though it looks like the generic
implementation (wrongly) assumes a dma_addr_t type rather than
phys_addr_t. We have phys_do_dma() for such conversion as it needs to
take the device into account, so I guess most page_to_phys() uses under
drivers/ are wrong.
Maybe as a separate series clean-up those page_to_phys() uses in
drivers.
> free_reserved_page(page);
> }
> #endif
> diff --git a/mm/bootmem_info.c b/mm/bootmem_info.c
> index b1efebfcf94b..fa7cb0c87c03 100644
> --- a/mm/bootmem_info.c
> +++ b/mm/bootmem_info.c
> @@ -34,7 +34,7 @@ void put_page_bootmem(struct page *page)
> ClearPagePrivate(page);
> set_page_private(page, 0);
> INIT_LIST_HEAD(&page->lru);
> - kmemleak_free_part(page_to_virt(page), PAGE_SIZE);
> + kmemleak_free_part_phys(PFN_PHYS(page_to_pfn(page)), PAGE_SIZE);
> free_reserved_page(page);
> }
> }
> --
> 2.25.1
--
Catalin
