From adfa22dd89640042c0df2d8906781ea74da9166c Mon Sep 17 00:00:00 2001 From: Harry Yoo Date: Mon, 18 Aug 2025 11:02:04 +0900 Subject: mm: move page table sync declarations to linux/pgtable.h During our internal testing, we started observing intermittent boot failures when the machine uses 4-level paging and has a large amount of persistent memory: BUG: unable to handle page fault for address: ffffe70000000034 #PF: supervisor write access in kernel mode #PF: error_code(0x0002) - not-present page PGD 0 P4D 0 Oops: 0002 [#1] SMP NOPTI RIP: 0010:__init_single_page+0x9/0x6d Call Trace: __init_zone_device_page+0x17/0x5d memmap_init_zone_device+0x154/0x1bb pagemap_range+0x2e0/0x40f memremap_pages+0x10b/0x2f0 devm_memremap_pages+0x1e/0x60 dev_dax_probe+0xce/0x2ec [device_dax] dax_bus_probe+0x6d/0xc9 [... snip ...] It turns out that the kernel panics while initializing vmemmap (struct page array) when the vmemmap region spans two PGD entries, because the new PGD entry is only installed in init_mm.pgd, but not in the page tables of other tasks. And looking at __populate_section_memmap(): if (vmemmap_can_optimize(altmap, pgmap)) // does not sync top level page tables r = vmemmap_populate_compound_pages(pfn, start, end, nid, pgmap); else // sync top level page tables in x86 r = vmemmap_populate(start, end, nid, altmap); In the normal path, vmemmap_populate() in arch/x86/mm/init_64.c synchronizes the top level page table (See commit 9b861528a801 ("x86-64, mem: Update all PGDs for direct mapping and vmemmap mapping changes")) so that all tasks in the system can see the new vmemmap area. However, when vmemmap_can_optimize() returns true, the optimized path skips synchronization of top-level page tables. This is because vmemmap_populate_compound_pages() is implemented in core MM code, which does not handle synchronization of the top-level page tables. Instead, the core MM has historically relied on each architecture to perform this synchronization manually. We're not the first party to encounter a crash caused by not-sync'd top level page tables: earlier this year, Gwan-gyeong Mun attempted to address the issue [1] [2] after hitting a kernel panic when x86 code accessed the vmemmap area before the corresponding top-level entries were synced. At that time, the issue was believed to be triggered only when struct page was enlarged for debugging purposes, and the patch did not get further updates. It turns out that current approach of relying on each arch to handle the page table sync manually is fragile because 1) it's easy to forget to sync the top level page table, and 2) it's also easy to overlook that the kernel should not access the vmemmap and direct mapping areas before the sync. To address this, Dave Hansen suggested [3] [4] introducing {pgd,p4d}_populate_kernel() for updating kernel portion of the page tables and allow each architecture to explicitly perform synchronization when installing top-level entries. With this approach, we no longer need to worry about missing the sync step, reducing the risk of future regressions. The new interface reuses existing ARCH_PAGE_TABLE_SYNC_MASK, PGTBL_P*D_MODIFIED and arch_sync_kernel_mappings() facility used by vmalloc and ioremap to synchronize page tables. pgd_populate_kernel() looks like this: static inline void pgd_populate_kernel(unsigned long addr, pgd_t *pgd, p4d_t *p4d) { pgd_populate(&init_mm, pgd, p4d); if (ARCH_PAGE_TABLE_SYNC_MASK & PGTBL_PGD_MODIFIED) arch_sync_kernel_mappings(addr, addr); } It is worth noting that vmalloc() and apply_to_range() carefully synchronizes page tables by calling p*d_alloc_track() and arch_sync_kernel_mappings(), and thus they are not affected by this patch series. This series was hugely inspired by Dave Hansen's suggestion and hence added Suggested-by: Dave Hansen. Cc stable because lack of this series opens the door to intermittent boot failures. This patch (of 3): Move ARCH_PAGE_TABLE_SYNC_MASK and arch_sync_kernel_mappings() to linux/pgtable.h so that they can be used outside of vmalloc and ioremap. Link: https://lkml.kernel.org/r/20250818020206.4517-1-harry.yoo@oracle.com Link: https://lkml.kernel.org/r/20250818020206.4517-2-harry.yoo@oracle.com Link: https://lore.kernel.org/linux-mm/20250220064105.808339-1-gwan-gyeong.mun@intel.com [1] Link: https://lore.kernel.org/linux-mm/20250311114420.240341-1-gwan-gyeong.mun@intel.com [2] Link: https://lore.kernel.org/linux-mm/d1da214c-53d3-45ac-a8b6-51821c5416e4@intel.com [3] Link: https://lore.kernel.org/linux-mm/4d800744-7b88-41aa-9979-b245e8bf794b@intel.com [4] Fixes: 8d400913c231 ("x86/vmemmap: handle unpopulated sub-pmd ranges") Signed-off-by: Harry Yoo Acked-by: Kiryl Shutsemau Reviewed-by: Mike Rapoport (Microsoft) Reviewed-by: "Uladzislau Rezki (Sony)" Reviewed-by: Lorenzo Stoakes Acked-by: David Hildenbrand Cc: Alexander Potapenko Cc: Alistair Popple Cc: Andrey Konovalov Cc: Andrey Ryabinin Cc: Andy Lutomirski Cc: "Aneesh Kumar K.V" Cc: Anshuman Khandual Cc: Ard Biesheuvel Cc: Arnd Bergmann Cc: bibo mao Cc: Borislav Betkov Cc: Christoph Lameter (Ampere) Cc: Dennis Zhou Cc: Dev Jain Cc: Dmitriy Vyukov Cc: Gwan-gyeong Mun Cc: Ingo Molnar Cc: Jane Chu Cc: Joao Martins Cc: Joerg Roedel Cc: John Hubbard Cc: Kevin Brodsky Cc: Liam Howlett Cc: Michal Hocko Cc: Oscar Salvador Cc: Peter Xu Cc: Peter Zijlstra Cc: Qi Zheng Cc: Ryan Roberts Cc: Suren Baghdasaryan Cc: Tejun Heo Cc: Thomas Gleinxer Cc: Thomas Huth Cc: Vincenzo Frascino Cc: Vlastimil Babka Cc: Dave Hansen Cc: Signed-off-by: Andrew Morton --- include/linux/pgtable.h | 17 +++++++++++++++++ include/linux/vmalloc.h | 16 ---------------- 2 files changed, 17 insertions(+), 16 deletions(-) --- a/include/linux/pgtable.h +++ b/include/linux/pgtable.h @@ -1329,6 +1329,23 @@ static inline void ptep_modify_prot_comm __ptep_modify_prot_commit(vma, addr, ptep, pte); } #endif /* __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION */ + +/* + * Architectures can set this mask to a combination of PGTBL_P?D_MODIFIED values + * and let generic vmalloc and ioremap code know when arch_sync_kernel_mappings() + * needs to be called. + */ +#ifndef ARCH_PAGE_TABLE_SYNC_MASK +#define ARCH_PAGE_TABLE_SYNC_MASK 0 +#endif + +/* + * There is no default implementation for arch_sync_kernel_mappings(). It is + * relied upon the compiler to optimize calls out if ARCH_PAGE_TABLE_SYNC_MASK + * is 0. + */ +void arch_sync_kernel_mappings(unsigned long start, unsigned long end); + #endif /* CONFIG_MMU */ /* --- a/include/linux/vmalloc.h +++ b/include/linux/vmalloc.h @@ -220,22 +220,6 @@ int vmap_pages_range(unsigned long addr, struct page **pages, unsigned int page_shift); /* - * Architectures can set this mask to a combination of PGTBL_P?D_MODIFIED values - * and let generic vmalloc and ioremap code know when arch_sync_kernel_mappings() - * needs to be called. - */ -#ifndef ARCH_PAGE_TABLE_SYNC_MASK -#define ARCH_PAGE_TABLE_SYNC_MASK 0 -#endif - -/* - * There is no default implementation for arch_sync_kernel_mappings(). It is - * relied upon the compiler to optimize calls out if ARCH_PAGE_TABLE_SYNC_MASK - * is 0. - */ -void arch_sync_kernel_mappings(unsigned long start, unsigned long end); - -/* * Lowlevel-APIs (not for driver use!) */