From 9910159f06590c17df4fbddedaabb4c0201cc4cb Mon Sep 17 00:00:00 2001 From: Rasmus Villemoes Date: Mon, 15 Dec 2025 14:17:23 +0100 Subject: iio: core: add separate lockdep class for info_exist_lock When one iio device is a consumer of another, it is possible that the ->info_exist_lock of both ends up being taken when reading the value of the consumer device. Since they currently belong to the same lockdep class (being initialized in a single location with mutex_init()), that results in a lockdep warning CPU0 ---- lock(&iio_dev_opaque->info_exist_lock); lock(&iio_dev_opaque->info_exist_lock); *** DEADLOCK *** May be due to missing lock nesting notation 4 locks held by sensors/414: #0: c31fd6dc (&p->lock){+.+.}-{3:3}, at: seq_read_iter+0x44/0x4e4 #1: c4f5a1c4 (&of->mutex){+.+.}-{3:3}, at: kernfs_seq_start+0x1c/0xac #2: c2827548 (kn->active#34){.+.+}-{0:0}, at: kernfs_seq_start+0x30/0xac #3: c1dd2b68 (&iio_dev_opaque->info_exist_lock){+.+.}-{3:3}, at: iio_read_channel_processed_scale+0x24/0xd8 stack backtrace: CPU: 0 UID: 0 PID: 414 Comm: sensors Not tainted 6.17.11 #5 NONE Hardware name: Generic AM33XX (Flattened Device Tree) Call trace: unwind_backtrace from show_stack+0x10/0x14 show_stack from dump_stack_lvl+0x44/0x60 dump_stack_lvl from print_deadlock_bug+0x2b8/0x334 print_deadlock_bug from __lock_acquire+0x13a4/0x2ab0 __lock_acquire from lock_acquire+0xd0/0x2c0 lock_acquire from __mutex_lock+0xa0/0xe8c __mutex_lock from mutex_lock_nested+0x1c/0x24 mutex_lock_nested from iio_read_channel_raw+0x20/0x6c iio_read_channel_raw from rescale_read_raw+0x128/0x1c4 rescale_read_raw from iio_channel_read+0xe4/0xf4 iio_channel_read from iio_read_channel_processed_scale+0x6c/0xd8 iio_read_channel_processed_scale from iio_hwmon_read_val+0x68/0xbc iio_hwmon_read_val from dev_attr_show+0x18/0x48 dev_attr_show from sysfs_kf_seq_show+0x80/0x110 sysfs_kf_seq_show from seq_read_iter+0xdc/0x4e4 seq_read_iter from vfs_read+0x238/0x2e4 vfs_read from ksys_read+0x6c/0xec ksys_read from ret_fast_syscall+0x0/0x1c Just as the mlock_key already has its own lockdep class, add a lock_class_key for the info_exist mutex. Note that this has in theory been a problem since before IIO first left staging, but it only occurs when a chain of consumers is in use and that is not often done. Fixes: ac917a81117c ("staging:iio:core set the iio_dev.info pointer to null on unregister under lock.") Signed-off-by: Rasmus Villemoes Reviewed-by: Peter Rosin Cc: Signed-off-by: Jonathan Cameron --- include/linux/iio/iio-opaque.h | 2 ++ 1 file changed, 2 insertions(+) (limited to 'include/linux') diff --git a/include/linux/iio/iio-opaque.h b/include/linux/iio/iio-opaque.h index 4247497f3f8b..b87841a355f8 100644 --- a/include/linux/iio/iio-opaque.h +++ b/include/linux/iio/iio-opaque.h @@ -14,6 +14,7 @@ * @mlock: lock used to prevent simultaneous device state changes * @mlock_key: lockdep class for iio_dev lock * @info_exist_lock: lock to prevent use during removal + * @info_exist_key: lockdep class for info_exist lock * @trig_readonly: mark the current trigger immutable * @event_interface: event chrdevs associated with interrupt lines * @attached_buffers: array of buffers statically attached by the driver @@ -47,6 +48,7 @@ struct iio_dev_opaque { struct mutex mlock; struct lock_class_key mlock_key; struct mutex info_exist_lock; + struct lock_class_key info_exist_key; bool trig_readonly; struct iio_event_interface *event_interface; struct iio_buffer **attached_buffers; -- cgit v1.2.3 From 6626734dd2b151753e134730e27d17e64784c345 Mon Sep 17 00:00:00 2001 From: Robin Murphy Date: Mon, 12 Jan 2026 15:46:37 +0000 Subject: mm_zone: Generalise has_managed_dma() It would be useful to be able to check for potential DMA pages beyond just ZONE_DMA - generalise the existing has_managed_dma() function to allow checking other zones too. Signed-off-by: Robin Murphy Acked-by: David Hildenbrand (Red Hat) Acked-by: Mike Rapoport (Microsoft) Tested-by: Vladimir Kondratiev Reviewed-by: Baoquan He Signed-off-by: Marek Szyprowski Link: https://lore.kernel.org/r/bd002d2351074e57be1ca08f03f333debac658fb.1768230104.git.robin.murphy@arm.com --- include/linux/mmzone.h | 9 +++++---- 1 file changed, 5 insertions(+), 4 deletions(-) (limited to 'include/linux') diff --git a/include/linux/mmzone.h b/include/linux/mmzone.h index 75ef7c9f9307..fc5d6c88d2f0 100644 --- a/include/linux/mmzone.h +++ b/include/linux/mmzone.h @@ -1648,14 +1648,15 @@ static inline int is_highmem(const struct zone *zone) return is_highmem_idx(zone_idx(zone)); } -#ifdef CONFIG_ZONE_DMA -bool has_managed_dma(void); -#else +bool has_managed_zone(enum zone_type zone); static inline bool has_managed_dma(void) { +#ifdef CONFIG_ZONE_DMA + return has_managed_zone(ZONE_DMA); +#else return false; -} #endif +} #ifndef CONFIG_NUMA -- cgit v1.2.3 From a995fe1a3aa78b7d06cc1cc7b6b8436c5e93b07f Mon Sep 17 00:00:00 2001 From: Danilo Krummrich Date: Wed, 7 Jan 2026 11:35:05 +0100 Subject: rust: driver: drop device private data post unbind Currently, the driver's device private data is allocated and initialized from driver core code called from bus abstractions after the driver's probe() callback returned the corresponding initializer. Similarly, the driver's device private data is dropped within the remove() callback of bus abstractions after calling the remove() callback of the corresponding driver. However, commit 6f61a2637abe ("rust: device: introduce Device::drvdata()") introduced an accessor for the driver's device private data for a Device, i.e. a device that is currently bound to a driver. Obviously, this is in conflict with dropping the driver's device private data in remove(), since a device can not be considered to be fully unbound after remove() has finished: We also have to consider registrations guarded by devres - such as IRQ or class device registrations - which are torn down after remove() in devres_release_all(). Thus, it can happen that, for instance, a class device or IRQ callback still calls Device::drvdata(), which then runs concurrently to remove() (which sets dev->driver_data to NULL and drops the driver's device private data), before devres_release_all() started to tear down the corresponding registration. This is because devres guarded registrations can, as expected, access the corresponding Device that defines their scope. In C it simply is the driver's responsibility to ensure that its device private data is freed after e.g. an IRQ registration is unregistered. Typically, C drivers achieve this by allocating their device private data with e.g. devm_kzalloc() before doing anything else, i.e. before e.g. registering an IRQ with devm_request_threaded_irq(), relying on the reverse order cleanup of devres. Technically, we could do something similar in Rust. However, the resulting code would be pretty messy: In Rust we have to differentiate between allocated but uninitialized memory and initialized memory in the type system. Thus, we would need to somehow keep track of whether the driver's device private data object has been initialized (i.e. probe() was successful and returned a valid initializer for this memory) and conditionally call the destructor of the corresponding object when it is freed. This is because we'd need to allocate and register the memory of the driver's device private data *before* it is initialized by the initializer returned by the driver's probe() callback, because the driver could already register devres guarded registrations within probe() outside of the driver's device private data initializer. Luckily there is a much simpler solution: Instead of dropping the driver's device private data at the end of remove(), we just drop it after the device has been fully unbound, i.e. after all devres callbacks have been processed. For this, we introduce a new post_unbind() callback private to the driver-core, i.e. the callback is neither exposed to drivers, nor to bus abstractions. This way, the driver-core code can simply continue to conditionally allocate the memory for the driver's device private data when the driver's initializer is returned from probe() - no change needed - and drop it when the driver-core code receives the post_unbind() callback. Closes: https://lore.kernel.org/all/DEZMS6Y4A7XE.XE7EUBT5SJFJ@kernel.org/ Fixes: 6f61a2637abe ("rust: device: introduce Device::drvdata()") Acked-by: Alice Ryhl Acked-by: Greg Kroah-Hartman Acked-by: Igor Korotin Link: https://patch.msgid.link/20260107103511.570525-7-dakr@kernel.org [ Remove #ifdef CONFIG_RUST, rename post_unbind() to post_unbind_rust(). - Danilo] Signed-off-by: Danilo Krummrich --- include/linux/device/driver.h | 9 +++++++++ 1 file changed, 9 insertions(+) (limited to 'include/linux') diff --git a/include/linux/device/driver.h b/include/linux/device/driver.h index cd8e0f0a634b..bbc67ec513ed 100644 --- a/include/linux/device/driver.h +++ b/include/linux/device/driver.h @@ -85,6 +85,8 @@ enum probe_type { * uevent. * @p: Driver core's private data, no one other than the driver * core can touch this. + * @p_cb: Callbacks private to the driver core; no one other than the + * driver core is allowed to touch this. * * The device driver-model tracks all of the drivers known to the system. * The main reason for this tracking is to enable the driver core to match @@ -119,6 +121,13 @@ struct device_driver { void (*coredump) (struct device *dev); struct driver_private *p; + struct { + /* + * Called after remove() and after all devres entries have been + * processed. This is a Rust only callback. + */ + void (*post_unbind_rust)(struct device *dev); + } p_cb; }; -- cgit v1.2.3 From 6ac433f8b2590b09ca00863d218665729ac985f7 Mon Sep 17 00:00:00 2001 From: Mathieu Desnoyers Date: Wed, 24 Dec 2025 12:33:57 -0500 Subject: mm: rename cpu_bitmap field to flexible_array MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit The cpu_bitmap flexible array now contains more than just the cpu_bitmap. In preparation for changing the static mm_struct definitions to cover for the additional space required, change the cpu_bitmap type from "unsigned long" to "char", require an unsigned long alignment of the flexible array, and rename the field from "cpu_bitmap" to "flexible_array". Introduce the MM_STRUCT_FLEXIBLE_ARRAY_INIT macro to statically initialize the flexible array. This covers the init_mm and efi_mm static definitions. This is a preparation step for fixing the missing mm_cid size for static mm_struct definitions. Link: https://lkml.kernel.org/r/20251224173358.647691-3-mathieu.desnoyers@efficios.com Fixes: af7f588d8f73 ("sched: Introduce per-memory-map concurrency ID") Signed-off-by: Mathieu Desnoyers Reviewed-by: Thomas Gleixner Cc: Mark Brown Cc: Aboorva Devarajan Cc: Al Viro Cc: Baolin Wang Cc: Christan König Cc: Christian Brauner Cc: Christoph Lameter Cc: David Hildenbrand Cc: David Rientjes Cc: Dennis Zhou Cc: Johannes Weiner Cc: "Liam R . Howlett" Cc: Lorenzo Stoakes Cc: Martin Liu Cc: Masami Hiramatsu Cc: Mateusz Guzik Cc: Matthew Wilcox Cc: Miaohe Lin Cc: Michal Hocko Cc: Mike Rapoport Cc: "Paul E. McKenney" Cc: Roman Gushchin Cc: SeongJae Park Cc: Shakeel Butt Cc: Steven Rostedt Cc: Suren Baghdasaryan Cc: Sweet Tea Dorminy Cc: Tejun Heo Cc: Vlastimil Babka Cc: Wei Yang Cc: Yu Zhao Cc: Peter Zijlstra (Intel) Cc: Signed-off-by: Andrew Morton --- include/linux/mm_types.h | 13 +++++++++---- 1 file changed, 9 insertions(+), 4 deletions(-) (limited to 'include/linux') diff --git a/include/linux/mm_types.h b/include/linux/mm_types.h index 42af2292951d..110b319a2ffb 100644 --- a/include/linux/mm_types.h +++ b/include/linux/mm_types.h @@ -1329,7 +1329,7 @@ struct mm_struct { * The mm_cpumask needs to be at the end of mm_struct, because it * is dynamically sized based on nr_cpu_ids. */ - unsigned long cpu_bitmap[]; + char flexible_array[] __aligned(__alignof__(unsigned long)); }; /* Copy value to the first system word of mm flags, non-atomically. */ @@ -1366,19 +1366,24 @@ static inline void __mm_flags_set_mask_bits_word(struct mm_struct *mm, MT_FLAGS_USE_RCU) extern struct mm_struct init_mm; +#define MM_STRUCT_FLEXIBLE_ARRAY_INIT \ +{ \ + [0 ... sizeof(cpumask_t)-1] = 0 \ +} + /* Pointer magic because the dynamic array size confuses some compilers. */ static inline void mm_init_cpumask(struct mm_struct *mm) { unsigned long cpu_bitmap = (unsigned long)mm; - cpu_bitmap += offsetof(struct mm_struct, cpu_bitmap); + cpu_bitmap += offsetof(struct mm_struct, flexible_array); cpumask_clear((struct cpumask *)cpu_bitmap); } /* Future-safe accessor for struct mm_struct's cpu_vm_mask. */ static inline cpumask_t *mm_cpumask(struct mm_struct *mm) { - return (struct cpumask *)&mm->cpu_bitmap; + return (struct cpumask *)&mm->flexible_array; } #ifdef CONFIG_LRU_GEN @@ -1469,7 +1474,7 @@ static inline cpumask_t *mm_cpus_allowed(struct mm_struct *mm) { unsigned long bitmap = (unsigned long)mm; - bitmap += offsetof(struct mm_struct, cpu_bitmap); + bitmap += offsetof(struct mm_struct, flexible_array); /* Skip cpu_bitmap */ bitmap += cpumask_size(); return (struct cpumask *)bitmap; -- cgit v1.2.3 From be31340a4cc259340044b7fc4f7e97f58c74ee8e Mon Sep 17 00:00:00 2001 From: Mathieu Desnoyers Date: Wed, 24 Dec 2025 12:33:58 -0500 Subject: mm: take into account mm_cid size for mm_struct static definitions MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Both init_mm and efi_mm static definitions need to make room for the 2 mm_cid cpumasks. This fixes possible out-of-bounds accesses to init_mm and efi_mm. Add a space between # and define for the mm_alloc_cid() definition to make it consistent with the coding style used in the rest of this header file. Link: https://lkml.kernel.org/r/20251224173358.647691-4-mathieu.desnoyers@efficios.com Fixes: af7f588d8f73 ("sched: Introduce per-memory-map concurrency ID") Signed-off-by: Mathieu Desnoyers Reviewed-by: Thomas Gleixner Cc: Mark Brown Cc: Aboorva Devarajan Cc: Al Viro Cc: Baolin Wang Cc: Christan König Cc: Christian Brauner Cc: Christoph Lameter Cc: David Hildenbrand Cc: David Rientjes Cc: Dennis Zhou Cc: Johannes Weiner Cc: "Liam R . Howlett" Cc: Lorenzo Stoakes Cc: Martin Liu Cc: Masami Hiramatsu Cc: Mateusz Guzik Cc: Matthew Wilcox Cc: Miaohe Lin Cc: Michal Hocko Cc: Mike Rapoport Cc: "Paul E. McKenney" Cc: Roman Gushchin Cc: SeongJae Park Cc: Shakeel Butt Cc: Steven Rostedt Cc: Suren Baghdasaryan Cc: Sweet Tea Dorminy Cc: Tejun Heo Cc: Vlastimil Babka Cc: Wei Yang Cc: Yu Zhao Cc: Peter Zijlstra (Intel) Cc: Signed-off-by: Andrew Morton --- include/linux/mm_types.h | 7 +++++-- 1 file changed, 5 insertions(+), 2 deletions(-) (limited to 'include/linux') diff --git a/include/linux/mm_types.h b/include/linux/mm_types.h index 110b319a2ffb..aa4639888f89 100644 --- a/include/linux/mm_types.h +++ b/include/linux/mm_types.h @@ -1368,7 +1368,7 @@ extern struct mm_struct init_mm; #define MM_STRUCT_FLEXIBLE_ARRAY_INIT \ { \ - [0 ... sizeof(cpumask_t)-1] = 0 \ + [0 ... sizeof(cpumask_t) + MM_CID_STATIC_SIZE - 1] = 0 \ } /* Pointer magic because the dynamic array size confuses some compilers. */ @@ -1500,7 +1500,7 @@ static inline int mm_alloc_cid_noprof(struct mm_struct *mm, struct task_struct * mm_init_cid(mm, p); return 0; } -#define mm_alloc_cid(...) alloc_hooks(mm_alloc_cid_noprof(__VA_ARGS__)) +# define mm_alloc_cid(...) alloc_hooks(mm_alloc_cid_noprof(__VA_ARGS__)) static inline void mm_destroy_cid(struct mm_struct *mm) { @@ -1514,6 +1514,8 @@ static inline unsigned int mm_cid_size(void) return cpumask_size() + bitmap_size(num_possible_cpus()); } +/* Use 2 * NR_CPUS as worse case for static allocation. */ +# define MM_CID_STATIC_SIZE (2 * sizeof(cpumask_t)) #else /* CONFIG_SCHED_MM_CID */ static inline void mm_init_cid(struct mm_struct *mm, struct task_struct *p) { } static inline int mm_alloc_cid(struct mm_struct *mm, struct task_struct *p) { return 0; } @@ -1522,6 +1524,7 @@ static inline unsigned int mm_cid_size(void) { return 0; } +# define MM_CID_STATIC_SIZE 0 #endif /* CONFIG_SCHED_MM_CID */ struct mmu_gather; -- cgit v1.2.3 From f9a49aa302a05e91ca01f69031cb79a0ea33031f Mon Sep 17 00:00:00 2001 From: Joanne Koong Date: Mon, 5 Jan 2026 13:17:27 -0800 Subject: fs/writeback: skip AS_NO_DATA_INTEGRITY mappings in wait_sb_inodes() MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Above the while() loop in wait_sb_inodes(), we document that we must wait for all pages under writeback for data integrity. Consequently, if a mapping, like fuse, traditionally does not have data integrity semantics, there is no need to wait at all; we can simply skip these inodes. This restores fuse back to prior behavior where syncs are no-ops. This fixes a user regression where if a system is running a faulty fuse server that does not reply to issued write requests, this causes wait_sb_inodes() to wait forever. Link: https://lkml.kernel.org/r/20260105211737.4105620-2-joannelkoong@gmail.com Fixes: 0c58a97f919c ("fuse: remove tmp folio for writebacks and internal rb tree") Signed-off-by: Joanne Koong Reported-by: Athul Krishna Reported-by: J. Neuschäfer Reviewed-by: Bernd Schubert Tested-by: J. Neuschäfer Cc: Alexander Viro Cc: Bernd Schubert Cc: Bonaccorso Salvatore Cc: Christian Brauner Cc: David Hildenbrand Cc: Jan Kara Cc: "Liam R. Howlett" Cc: Lorenzo Stoakes Cc: "Matthew Wilcox (Oracle)" Cc: Michal Hocko Cc: Mike Rapoport Cc: Miklos Szeredi Cc: Suren Baghdasaryan Cc: Vlastimil Babka Cc: Signed-off-by: Andrew Morton --- include/linux/pagemap.h | 11 +++++++++++ 1 file changed, 11 insertions(+) (limited to 'include/linux') diff --git a/include/linux/pagemap.h b/include/linux/pagemap.h index 31a848485ad9..ec442af3f886 100644 --- a/include/linux/pagemap.h +++ b/include/linux/pagemap.h @@ -210,6 +210,7 @@ enum mapping_flags { AS_WRITEBACK_MAY_DEADLOCK_ON_RECLAIM = 9, AS_KERNEL_FILE = 10, /* mapping for a fake kernel file that shouldn't account usage to user cgroups */ + AS_NO_DATA_INTEGRITY = 11, /* no data integrity guarantees */ /* Bits 16-25 are used for FOLIO_ORDER */ AS_FOLIO_ORDER_BITS = 5, AS_FOLIO_ORDER_MIN = 16, @@ -345,6 +346,16 @@ static inline bool mapping_writeback_may_deadlock_on_reclaim(const struct addres return test_bit(AS_WRITEBACK_MAY_DEADLOCK_ON_RECLAIM, &mapping->flags); } +static inline void mapping_set_no_data_integrity(struct address_space *mapping) +{ + set_bit(AS_NO_DATA_INTEGRITY, &mapping->flags); +} + +static inline bool mapping_no_data_integrity(const struct address_space *mapping) +{ + return test_bit(AS_NO_DATA_INTEGRITY, &mapping->flags); +} + static inline gfp_t mapping_gfp_mask(const struct address_space *mapping) { return mapping->gfp_mask; -- cgit v1.2.3 From ca1a47cd3f5f4c46ca188b1c9a27af87d1ab2216 Mon Sep 17 00:00:00 2001 From: "David Hildenbrand (Red Hat)" Date: Tue, 23 Dec 2025 22:40:34 +0100 Subject: mm/hugetlb: fix hugetlb_pmd_shared() Patch series "mm/hugetlb: fixes for PMD table sharing (incl. using mmu_gather)", v3. One functional fix, one performance regression fix, and two related comment fixes. I cleaned up my prototype I recently shared [1] for the performance fix, deferring most of the cleanups I had in the prototype to a later point. While doing that I identified the other things. The goal of this patch set is to be backported to stable trees "fairly" easily. At least patch #1 and #4. Patch #1 fixes hugetlb_pmd_shared() not detecting any sharing Patch #2 + #3 are simple comment fixes that patch #4 interacts with. Patch #4 is a fix for the reported performance regression due to excessive IPI broadcasts during fork()+exit(). The last patch is all about TLB flushes, IPIs and mmu_gather. Read: complicated There are plenty of cleanups in the future to be had + one reasonable optimization on x86. But that's all out of scope for this series. Runtime tested, with a focus on fixing the performance regression using the original reproducer [2] on x86. This patch (of 4): We switched from (wrongly) using the page count to an independent shared count. Now, shared page tables have a refcount of 1 (excluding speculative references) and instead use ptdesc->pt_share_count to identify sharing. We didn't convert hugetlb_pmd_shared(), so right now, we would never detect a shared PMD table as such, because sharing/unsharing no longer touches the refcount of a PMD table. Page migration, like mbind() or migrate_pages() would allow for migrating folios mapped into such shared PMD tables, even though the folios are not exclusive. In smaps we would account them as "private" although they are "shared", and we would be wrongly setting the PM_MMAP_EXCLUSIVE in the pagemap interface. Fix it by properly using ptdesc_pmd_is_shared() in hugetlb_pmd_shared(). Link: https://lkml.kernel.org/r/20251223214037.580860-1-david@kernel.org Link: https://lkml.kernel.org/r/20251223214037.580860-2-david@kernel.org Link: https://lore.kernel.org/all/8cab934d-4a56-44aa-b641-bfd7e23bd673@kernel.org/ [1] Link: https://lore.kernel.org/all/8cab934d-4a56-44aa-b641-bfd7e23bd673@kernel.org/ [2] Fixes: 59d9094df3d7 ("mm: hugetlb: independent PMD page table shared count") Signed-off-by: David Hildenbrand (Red Hat) Reviewed-by: Rik van Riel Reviewed-by: Lance Yang Tested-by: Lance Yang Reviewed-by: Harry Yoo Tested-by: Laurence Oberman Reviewed-by: Lorenzo Stoakes Acked-by: Oscar Salvador Cc: Liu Shixin Cc: Uschakow, Stanislav" Cc: Signed-off-by: Andrew Morton --- include/linux/hugetlb.h | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'include/linux') diff --git a/include/linux/hugetlb.h b/include/linux/hugetlb.h index 019a1c5281e4..03c8725efa28 100644 --- a/include/linux/hugetlb.h +++ b/include/linux/hugetlb.h @@ -1326,7 +1326,7 @@ static inline __init void hugetlb_cma_reserve(int order) #ifdef CONFIG_HUGETLB_PMD_PAGE_TABLE_SHARING static inline bool hugetlb_pmd_shared(pte_t *pte) { - return page_count(virt_to_page(pte)) > 1; + return ptdesc_pmd_is_shared(virt_to_ptdesc(pte)); } #else static inline bool hugetlb_pmd_shared(pte_t *pte) -- cgit v1.2.3 From 8ce720d5bd91e9dc16db3604aa4b1bf76770a9a1 Mon Sep 17 00:00:00 2001 From: "David Hildenbrand (Red Hat)" Date: Tue, 23 Dec 2025 22:40:37 +0100 Subject: mm/hugetlb: fix excessive IPI broadcasts when unsharing PMD tables using mmu_gather As reported, ever since commit 1013af4f585f ("mm/hugetlb: fix huge_pmd_unshare() vs GUP-fast race") we can end up in some situations where we perform so many IPI broadcasts when unsharing hugetlb PMD page tables that it severely regresses some workloads. In particular, when we fork()+exit(), or when we munmap() a large area backed by many shared PMD tables, we perform one IPI broadcast per unshared PMD table. There are two optimizations to be had: (1) When we process (unshare) multiple such PMD tables, such as during exit(), it is sufficient to send a single IPI broadcast (as long as we respect locking rules) instead of one per PMD table. Locking prevents that any of these PMD tables could get reused before we drop the lock. (2) When we are not the last sharer (> 2 users including us), there is no need to send the IPI broadcast. The shared PMD tables cannot become exclusive (fully unshared) before an IPI will be broadcasted by the last sharer. Concurrent GUP-fast could walk into a PMD table just before we unshared it. It could then succeed in grabbing a page from the shared page table even after munmap() etc succeeded (and supressed an IPI). But there is not difference compared to GUP-fast just sleeping for a while after grabbing the page and re-enabling IRQs. Most importantly, GUP-fast will never walk into page tables that are no-longer shared, because the last sharer will issue an IPI broadcast. (if ever required, checking whether the PUD changed in GUP-fast after grabbing the page like we do in the PTE case could handle this) So let's rework PMD sharing TLB flushing + IPI sync to use the mmu_gather infrastructure so we can implement these optimizations and demystify the code at least a bit. Extend the mmu_gather infrastructure to be able to deal with our special hugetlb PMD table sharing implementation. To make initialization of the mmu_gather easier when working on a single VMA (in particular, when dealing with hugetlb), provide tlb_gather_mmu_vma(). We'll consolidate the handling for (full) unsharing of PMD tables in tlb_unshare_pmd_ptdesc() and tlb_flush_unshared_tables(), and track in "struct mmu_gather" whether we had (full) unsharing of PMD tables. Because locking is very special (concurrent unsharing+reuse must be prevented), we disallow deferring flushing to tlb_finish_mmu() and instead require an explicit earlier call to tlb_flush_unshared_tables(). From hugetlb code, we call huge_pmd_unshare_flush() where we make sure that the expected lock protecting us from concurrent unsharing+reuse is still held. Check with a VM_WARN_ON_ONCE() in tlb_finish_mmu() that tlb_flush_unshared_tables() was properly called earlier. Document it all properly. Notes about tlb_remove_table_sync_one() interaction with unsharing: There are two fairly tricky things: (1) tlb_remove_table_sync_one() is a NOP on architectures without CONFIG_MMU_GATHER_RCU_TABLE_FREE. Here, the assumption is that the previous TLB flush would send an IPI to all relevant CPUs. Careful: some architectures like x86 only send IPIs to all relevant CPUs when tlb->freed_tables is set. The relevant architectures should be selecting MMU_GATHER_RCU_TABLE_FREE, but x86 might not do that in stable kernels and it might have been problematic before this patch. Also, the arch flushing behavior (independent of IPIs) is different when tlb->freed_tables is set. Do we have to enlighten them to also take care of tlb->unshared_tables? So far we didn't care, so hopefully we are fine. Of course, we could be setting tlb->freed_tables as well, but that might then unnecessarily flush too much, because the semantics of tlb->freed_tables are a bit fuzzy. This patch changes nothing in this regard. (2) tlb_remove_table_sync_one() is not a NOP on architectures with CONFIG_MMU_GATHER_RCU_TABLE_FREE that actually don't need a sync. Take x86 as an example: in the common case (!pv, !X86_FEATURE_INVLPGB) we still issue IPIs during TLB flushes and don't actually need the second tlb_remove_table_sync_one(). This optimized can be implemented on top of this, by checking e.g., in tlb_remove_table_sync_one() whether we really need IPIs. But as described in (1), it really must honor tlb->freed_tables then to send IPIs to all relevant CPUs. Notes on TLB flushing changes: (1) Flushing for non-shared PMD tables We're converting from flush_hugetlb_tlb_range() to tlb_remove_huge_tlb_entry(). Given that we properly initialize the MMU gather in tlb_gather_mmu_vma() to be hugetlb aware, similar to __unmap_hugepage_range(), that should be fine. (2) Flushing for shared PMD tables We're converting from various things (flush_hugetlb_tlb_range(), tlb_flush_pmd_range(), flush_tlb_range()) to tlb_flush_pmd_range(). tlb_flush_pmd_range() achieves the same that tlb_remove_huge_tlb_entry() would achieve in these scenarios. Note that tlb_remove_huge_tlb_entry() also calls __tlb_remove_tlb_entry(), however that is only implemented on powerpc, which does not support PMD table sharing. Similar to (1), tlb_gather_mmu_vma() should make sure that TLB flushing keeps on working as expected. Further, note that the ptdesc_pmd_pts_dec() in huge_pmd_share() is not a concern, as we are holding the i_mmap_lock the whole time, preventing concurrent unsharing. That ptdesc_pmd_pts_dec() usage will be removed separately as a cleanup later. There are plenty more cleanups to be had, but they have to wait until this is fixed. [david@kernel.org: fix kerneldoc] Link: https://lkml.kernel.org/r/f223dd74-331c-412d-93fc-69e360a5006c@kernel.org Link: https://lkml.kernel.org/r/20251223214037.580860-5-david@kernel.org Fixes: 1013af4f585f ("mm/hugetlb: fix huge_pmd_unshare() vs GUP-fast race") Signed-off-by: David Hildenbrand (Red Hat) Reported-by: Uschakow, Stanislav" Closes: https://lore.kernel.org/all/4d3878531c76479d9f8ca9789dc6485d@amazon.de/ Tested-by: Laurence Oberman Acked-by: Harry Yoo Reviewed-by: Lorenzo Stoakes Cc: Lance Yang Cc: Liu Shixin Cc: Oscar Salvador Cc: Rik van Riel Cc: Signed-off-by: Andrew Morton --- include/linux/hugetlb.h | 15 ++++++++++----- include/linux/mm_types.h | 1 + 2 files changed, 11 insertions(+), 5 deletions(-) (limited to 'include/linux') diff --git a/include/linux/hugetlb.h b/include/linux/hugetlb.h index 03c8725efa28..e51b8ef0cebd 100644 --- a/include/linux/hugetlb.h +++ b/include/linux/hugetlb.h @@ -240,8 +240,9 @@ pte_t *huge_pte_alloc(struct mm_struct *mm, struct vm_area_struct *vma, pte_t *huge_pte_offset(struct mm_struct *mm, unsigned long addr, unsigned long sz); unsigned long hugetlb_mask_last_page(struct hstate *h); -int huge_pmd_unshare(struct mm_struct *mm, struct vm_area_struct *vma, - unsigned long addr, pte_t *ptep); +int huge_pmd_unshare(struct mmu_gather *tlb, struct vm_area_struct *vma, + unsigned long addr, pte_t *ptep); +void huge_pmd_unshare_flush(struct mmu_gather *tlb, struct vm_area_struct *vma); void adjust_range_if_pmd_sharing_possible(struct vm_area_struct *vma, unsigned long *start, unsigned long *end); @@ -300,13 +301,17 @@ static inline struct address_space *hugetlb_folio_mapping_lock_write( return NULL; } -static inline int huge_pmd_unshare(struct mm_struct *mm, - struct vm_area_struct *vma, - unsigned long addr, pte_t *ptep) +static inline int huge_pmd_unshare(struct mmu_gather *tlb, + struct vm_area_struct *vma, unsigned long addr, pte_t *ptep) { return 0; } +static inline void huge_pmd_unshare_flush(struct mmu_gather *tlb, + struct vm_area_struct *vma) +{ +} + static inline void adjust_range_if_pmd_sharing_possible( struct vm_area_struct *vma, unsigned long *start, unsigned long *end) diff --git a/include/linux/mm_types.h b/include/linux/mm_types.h index aa4639888f89..78950eb8926d 100644 --- a/include/linux/mm_types.h +++ b/include/linux/mm_types.h @@ -1530,6 +1530,7 @@ static inline unsigned int mm_cid_size(void) struct mmu_gather; extern void tlb_gather_mmu(struct mmu_gather *tlb, struct mm_struct *mm); extern void tlb_gather_mmu_fullmm(struct mmu_gather *tlb, struct mm_struct *mm); +void tlb_gather_mmu_vma(struct mmu_gather *tlb, struct vm_area_struct *vma); extern void tlb_finish_mmu(struct mmu_gather *tlb); struct vm_fault; -- cgit v1.2.3 From 35e247032606f06c2f19d90a6562bc315206b7a7 Mon Sep 17 00:00:00 2001 From: Lorenzo Stoakes Date: Wed, 14 Jan 2026 11:00:06 +0000 Subject: mm: do not copy page tables unnecessarily for VM_UFFD_WP Commit ab04b530e7e8 ("mm: introduce copy-on-fork VMAs and make VM_MAYBE_GUARD one") aggregates flags checks in vma_needs_copy(), including VM_UFFD_WP. However in doing so, it incorrectly performed this check against src_vma. This check was done on the assumption that all relevant flags are copied upon fork. However the userfaultfd logic is very innovative in that it implements custom logic on fork in dup_userfaultfd(), including a rather well hidden case where lacking UFFD_FEATURE_EVENT_FORK causes VM_UFFD_WP to not be propagated to the destination VMA. And indeed, vma_needs_copy(), prior to this patch, did check this property on dst_vma, not src_vma. Since all the other relevant flags are copied on fork, we can simply fix this by checking against dst_vma. While we're here, we fix a comment against VM_COPY_ON_FORK (noting that it did indeed already reference dst_vma) to make it abundantly clear that we must check against the destination VMA. Link: https://lkml.kernel.org/r/20260114110006.1047071-1-lorenzo.stoakes@oracle.com Fixes: ab04b530e7e8 ("mm: introduce copy-on-fork VMAs and make VM_MAYBE_GUARD one") Signed-off-by: Lorenzo Stoakes Reported-by: Chris Mason Closes: https://lore.kernel.org/all/20260113231257.3002271-1-clm@meta.com/ Acked-by: David Hildenbrand (Red Hat) Acked-by: Pedro Falcato Cc: Liam Howlett Cc: Michal Hocko Cc: Mike Rapoport Cc: Suren Baghdasaryan Cc: Vlastimil Babka Signed-off-by: Andrew Morton --- include/linux/mm.h | 6 +++++- 1 file changed, 5 insertions(+), 1 deletion(-) (limited to 'include/linux') diff --git a/include/linux/mm.h b/include/linux/mm.h index 6f959d8ca4b4..f0d5be9dc736 100644 --- a/include/linux/mm.h +++ b/include/linux/mm.h @@ -608,7 +608,11 @@ enum { /* * Flags which should result in page tables being copied on fork. These are * flags which indicate that the VMA maps page tables which cannot be - * reconsistuted upon page fault, so necessitate page table copying upon + * reconsistuted upon page fault, so necessitate page table copying upon fork. + * + * Note that these flags should be compared with the DESTINATION VMA not the + * source, as VM_UFFD_WP may not be propagated to destination, while all other + * flags will be. * * VM_PFNMAP / VM_MIXEDMAP - These contain kernel-mapped data which cannot be * reasonably reconstructed on page fault. -- cgit v1.2.3