- 05 May, 2021 40 commits
-
-
Charan Teja Reddy authored
By definition, COMPACT[STALL|FAIL] events needs to be counted when there is 'At least in one zone compaction wasn't deferred or skipped from the direct compaction'. And when compaction is skipped or deferred, COMPACT_SKIPPED will be returned but it will still go and update these compaction events which is wrong in the sense that COMPACT[STALL|FAIL] is counted without even trying the compaction. Correct this by skipping the counting of these events when COMPACT_SKIPPED is returned for compaction. This indirectly also avoid the unnecessary try into the get_page_from_freelist() when compaction is not even tried. There is a corner case where compaction is skipped but still count COMPACTSTALL event, which is that IRQ came and freed the page and the same is captured in capture_control. Link: https://lkml.kernel.org/r/1613151184-21213-1-git-send-email-charante@codeaurora.orgSigned-off-by:
Charan Teja Reddy <charante@codeaurora.org> Acked-by:
Vlastimil Babka <vbabka@suse.cz> Acked-by:
David Rientjes <rientjes@google.com> Signed-off-by:
Andrew Morton <akpm@linux-foundation.org> Signed-off-by:
Linus Torvalds <torvalds@linux-foundation.org>
-
Pintu Kumar authored
The sysctl_compact_memory is mostly unused in mm/compaction.c It just acts as a place holder for sysctl to store .data. But the .data itself is not needed here. So we can get ride of this variable completely and make .data as NULL. This will also eliminate the extern declaration from header file. No functionality is broken or changed this way. Link: https://lkml.kernel.org/r/1614852224-14671-1-git-send-email-pintu@codeaurora.orgSigned-off-by:
Pintu Kumar <pintu@codeaurora.org> Signed-off-by:
Pintu Agarwal <pintu.ping@gmail.com> Reviewed-by:
-
Yang Shi authored
The number of deferred objects might get windup to an absurd number, and it results in clamp of slab objects. It is undesirable for sustaining workingset. So shrink deferred objects proportional to priority and cap nr_deferred to twice of cache items. The idea is borrowed from Dave Chinner's patch: https://lore.kernel.org/linux-xfs/20191031234618.15403-13-david@fromorbit.com/ Tested with kernel build and vfs metadata heavy workload in our production environment, no regression is spotted so far. Link: https://lkml.kernel.org/r/20210311190845.9708-14-shy828301@gmail.comSigned-off-by:
Yang Shi <shy828301@gmail.com> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: Kirill Tkhai <ktkhai@virtuozzo.com> Cc: Michal Hocko <mhocko@suse.com> Cc: Roman Gushchin <guro@fb.com> Cc: Shakeel Butt <shakeelb@google.com> Cc: Vlastimil Babka <vbabka@suse.cz> Signed-off-by:
-
Yang Shi authored
Now shrinker's nr_deferred is per memcg for memcg aware shrinkers, add to parent's corresponding nr_deferred when memcg offline. Link: https://lkml.kernel.org/r/20210311190845.9708-13-shy828301@gmail.comSigned-off-by:
Kirill Tkhai <ktkhai@virtuozzo.com> Acked-by:
Roman Gushchin <guro@fb.com> Reviewed-by:
Shakeel Butt <shakeelb@google.com> Cc: Dave Chinner <david@fromorbit.com> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: Michal Hocko <mhocko@suse.com> Signed-off-by:
-
Yang Shi authored
Now nr_deferred is available on per memcg level for memcg aware shrinkers, so don't need allocate shrinker->nr_deferred for such shrinkers anymore. The prealloc_memcg_shrinker() would return -ENOSYS if !CONFIG_MEMCG or memcg is disabled by kernel command line, then shrinker's SHRINKER_MEMCG_AWARE flag would be cleared. This makes the implementation of this patch simpler. Link: https://lkml.kernel.org/r/20210311190845.9708-12-shy828301@gmail.comSigned-off-by:
-
Yang Shi authored
Use per memcg's nr_deferred for memcg aware shrinkers. The shrinker's nr_deferred will be used in the following cases: 1. Non memcg aware shrinkers 2. !CONFIG_MEMCG 3. memcg is disabled by boot parameter Link: https://lkml.kernel.org/r/20210311190845.9708-11-shy828301@gmail.comSigned-off-by: -
Yang Shi authored
Currently the number of deferred objects are per shrinker, but some slabs, for example, vfs inode/dentry cache are per memcg, this would result in poor isolation among memcgs. The deferred objects typically are generated by __GFP_NOFS allocations, one memcg with excessive __GFP_NOFS allocations may blow up deferred objects, then other innocent memcgs may suffer from over shrink, excessive reclaim latency, etc. For example, two workloads run in memcgA and memcgB respectively, workload in B is vfs heavy workload. Workload in A generates excessive deferred objects, then B's vfs cache might be hit heavily (drop half of caches) by B's limit reclaim or global reclaim. We observed this hit in our production environment which was running vfs heavy workload shown as the below tracing log: <...>-409454 [016] .... 28286961.747146: mm_shrink_slab_start: super_cache_scan+0x0/0x1a0 ffff9a83046f3458: nid: 1 objects to shrink 3641681686040 gfp_flags GFP_HIGHUSER_MOVABLE|__GFP_ZERO pgs_scanned 1 lru_pgs 15721 cache items 246404277 delta 31345 total_scan 123202138 <...>-409454 [022] .... 28287105.928018: mm_shrink_slab_end: super_cache_scan+0x0/0x1a0 ffff9a83046f3458: nid: 1 unused scan count 3641681686040 new scan count 3641798379189 total_scan 602 last shrinker return val 123186855 The vfs cache and page cache ratio was 10:1 on this machine, and half of caches were dropped. This also resulted in significant amount of page caches were dropped due to inodes eviction. Make nr_deferred per memcg for memcg aware shrinkers would solve the unfairness and bring better isolation. The following patch will add nr_deferred to parent memcg when memcg offline. To preserve nr_deferred when reparenting memcgs to root, root memcg needs shrinker_info allocated too. When memcg is not enabled (!CONFIG_MEMCG or memcg disabled), the shrinker's nr_deferred would be used. And non memcg aware shrinkers use shrinker's nr_deferred all the time. Link: https://lkml.kernel.org/r/20210311190845.9708-10-shy828301@gmail.comSigned-off-by:
-
Yang Shi authored
Currently registered shrinker is indicated by non-NULL shrinker->nr_deferred. This approach is fine with nr_deferred at the shrinker level, but the following patches will move MEMCG_AWARE shrinkers' nr_deferred to memcg level, so their shrinker->nr_deferred would always be NULL. This would prevent the shrinkers from unregistering correctly. Remove SHRINKER_REGISTERING since we could check if shrinker is registered successfully by the new flag. Link: https://lkml.kernel.org/r/20210311190845.9708-9-shy828301@gmail.comSigned-off-by:
-
Yang Shi authored
The shrinker_info is dereferenced in a couple of places via rcu_dereference_protected with different calling conventions, for example, using mem_cgroup_nodeinfo helper or dereferencing memcg->nodeinfo[nid]->shrinker_info. And the later patch will add more dereference places. So extract the dereference into a helper to make the code more readable. No functional change. [akpm@linux-foundation.org: retain rcu_dereference_protected() in free_shrinker_info(), per Hugh] Link: https://lkml.kernel.org/r/20210311190845.9708-8-shy828301@gmail.comSigned-off-by:
-
Yang Shi authored
The following patch is going to add nr_deferred into shrinker_map, the change will make shrinker_map not only include map anymore, so rename it to "memcg_shrinker_info". And this should make the patch adding nr_deferred cleaner and readable and make review easier. Also remove the "memcg_" prefix. Link: https://lkml.kernel.org/r/20210311190845.9708-7-shy828301@gmail.comSigned-off-by:
-
Yang Shi authored
Using kvfree_rcu() to free the old shrinker_maps instead of call_rcu(). We don't have to define a dedicated callback for call_rcu() anymore. Link: https://lkml.kernel.org/r/20210311190845.9708-6-shy828301@gmail.comSigned-off-by:
-
Yang Shi authored
Both memcg_shrinker_map_size and shrinker_nr_max is maintained, but actually the map size can be calculated via shrinker_nr_max, so it seems unnecessary to keep both. Remove memcg_shrinker_map_size since shrinker_nr_max is also used by iterating the bit map. Link: https://lkml.kernel.org/r/20210311190845.9708-5-shy828301@gmail.comSigned-off-by:
-
Yang Shi authored
Since memcg_shrinker_map_size just can be changed under holding shrinker_rwsem exclusively, the read side can be protected by holding read lock, so it sounds superfluous to have a dedicated mutex. Kirill Tkhai suggested use write lock since: * We want the assignment to shrinker_maps is visible for shrink_slab_memcg(). * The rcu_dereference_protected() dereferrencing in shrink_slab_memcg(), but in case of we use READ lock in alloc_shrinker_maps(), the dereferrencing is not actually protected. * READ lock makes alloc_shrinker_info() racy against memory allocation fail. alloc_shrinker_info()->free_shrinker_info() may free memory right after shrink_slab_memcg() dereferenced it. You may say shrink_slab_memcg()->mem_cgroup_online() protects us from it? Yes, sure, but this is not the thing we want to remember in the future, since this spreads modularity. And a test with heavy paging workload didn't show write lock makes things worse. Link: https://lkml.kernel.org/r/20210311190845.9708-4-shy828301@gmail.comSigned-off-by: -
Yang Shi authored
The shrinker map management is not purely memcg specific, it is at the intersection between memory cgroup and shrinkers. It's allocation and assignment of a structure, and the only memcg bit is the map is being stored in a memcg structure. So move the shrinker_maps handling code into vmscan.c for tighter integration with shrinker code, and remove the "memcg_" prefix. There is no functional change. Link: https://lkml.kernel.org/r/20210311190845.9708-3-shy828301@gmail.comSigned-off-by:
-
Yang Shi authored
Patch series "Make shrinker's nr_deferred memcg aware", v10. Recently huge amount one-off slab drop was seen on some vfs metadata heavy workloads, it turned out there were huge amount accumulated nr_deferred objects seen by the shrinker. On our production machine, I saw absurd number of nr_deferred shown as the below tracing result: <...>-48776 [032] .... 27970562.458916: mm_shrink_slab_start: super_cache_scan+0x0/0x1a0 ffff9a83046f3458: nid: 0 objects to shrink 2531805877005 gfp_flags GFP_HIGHUSER_MOVABLE pgs_scanned 32 lru_pgs 9300 cache items 1667 delta 11 total_scan 833 There are 2.5 trillion deferred objects on one node, assuming all of them are dentry (192 bytes per object), so the total size of deferred on one node is ~480TB. It is definitely ridiculous. I managed to reproduce this problem with kernel build workload plus negative dentry generator. First step, run the below kernel build test script: NR_CPUS=`cat /proc/cpuinfo | grep -e processor | wc -l` cd /root/Buildarea/linux-stable for i in `seq 1500`; do cgcreate -g memory:kern_build echo 4G > /sys/fs/cgroup/memory/kern_build/memory.limit_in_bytes echo 3 > /proc/sys/vm/drop_caches cgexec -g memory:kern_build make clean > /dev/null 2>&1 cgexec -g memory:kern_build make -j$NR_CPUS > /dev/null 2>&1 cgdelete -g memory:kern_build done Then run the below negative dentry generator script: NR_CPUS=`cat /proc/cpuinfo | grep -e processor | wc -l` mkdir /sys/fs/cgroup/memory/test echo $$ > /sys/fs/cgroup/memory/test/tasks for i in `seq $NR_CPUS`; do while true; do FILE=`head /dev/urandom | tr -dc A-Za-z0-9 | head -c 64` cat $FILE 2>/dev/null done & done Then kswapd will shrink half of dentry cache in just one loop as the below tracing result showed: kswapd0-475 [028] .... 305968.252561: mm_shrink_slab_start: super_cache_scan+0x0/0x190 0000000024acf00c: nid: 0 objects to shrink 4994376020 gfp_flags GFP_KERNEL cache items 93689873 delta 45746 total_scan 46844936 priority 12 kswapd0-475 [021] .... 306013.099399: mm_shrink_slab_end: super_cache_scan+0x0/0x190 0000000024acf00c: nid: 0 unused scan count 4994376020 new scan count 4947576838 total_scan 8 last shrinker return val 46844928 There were huge number of deferred objects before the shrinker was called, the behavior does match the code but it might be not desirable from the user's stand of point. The excessive amount of nr_deferred might be accumulated due to various reasons, for example: * GFP_NOFS allocation * Significant times of small amount scan (< scan_batch, 1024 for vfs metadata) However the LRUs of slabs are per memcg (memcg-aware shrinkers) but the deferred objects is per shrinker, this may have some bad effects: * Poor isolation among memcgs. Some memcgs which happen to have frequent limit reclaim may get nr_deferred accumulated to a huge number, then other innocent memcgs may take the fall. In our case the main workload was hit. * Unbounded deferred objects. There is no cap for deferred objects, it can outgrow ridiculously as the tracing result showed. * Easy to get out of control. Although shrinkers take into account deferred objects, but it can go out of control easily. One misconfigured memcg could incur absurd amount of deferred objects in a period of time. * Sort of reclaim problems, i.e. over reclaim, long reclaim latency, etc. There may be hundred GB slab caches for vfe metadata heavy workload, shrink half of them may take minutes. We observed latency spike due to the prolonged reclaim. These issues also have been discussed in https://lore.kernel.org/linux-mm/20200916185823.5347-1-shy828301@gmail.com/. The patchset is the outcome of that discussion. So this patchset makes nr_deferred per-memcg to tackle the problem. It does: * Have memcg_shrinker_deferred per memcg per node, just like what shrinker_map does. Instead it is an atomic_long_t array, each element represent one shrinker even though the shrinker is not memcg aware, this simplifies the implementation. For memcg aware shrinkers, the deferred objects are just accumulated to its own memcg. The shrinkers just see nr_deferred from its own memcg. Non memcg aware shrinkers still use global nr_deferred from struct shrinker. * Once the memcg is offlined, its nr_deferred will be reparented to its parent along with LRUs. * The root memcg has memcg_shrinker_deferred array too. It simplifies the handling of reparenting to root memcg. * Cap nr_deferred to 2x of the length of lru. The idea is borrowed from Dave Chinner's series (https://lore.kernel.org/linux-xfs/20191031234618.15403-1-david@fromorbit.com/) The downside is each memcg has to allocate extra memory to store the nr_deferred array. On our production environment, there are typically around 40 shrinkers, so each memcg needs ~320 bytes. 10K memcgs would need ~3.2MB memory. It seems fine. We have been running the patched kernel on some hosts of our fleet (test and production) for months, it works very well. The monitor data shows the working set is sustained as expected. This patch (of 13): The tracepoint's nid should show what node the shrink happens on, the start tracepoint uses nid from shrinkctl, but the nid might be set to 0 before end tracepoint if the shrinker is not NUMA aware, so the tracing log may show the shrink happens on one node but end up on the other node. It seems confusing. And the following patch will remove using nid directly in do_shrink_slab(), this patch also helps cleanup the code. Link: https://lkml.kernel.org/r/20210311190845.9708-1-shy828301@gmail.com Link: https://lkml.kernel.org/r/20210311190845.9708-2-shy828301@gmail.comSigned-off-by: -
Dave Hansen authored
RECLAIM_ZONE was assumed to be unused because it was never explicitly used in the kernel. However, there were a number of places where it was checked implicitly by checking 'node_reclaim_mode' for a zero value. These zero checks are not great because it is not obvious what a zero mode *means* in the code. Replace them with a helper which makes it more obvious: node_reclaim_enabled(). This helper also provides a handy place to explicitly check the RECLAIM_ZONE bit itself. Check it explicitly there to make it more obvious where the bit can affect behavior. This should have no functional impact. Link: https://lkml.kernel.org/r/20210219172559.BF589C44@viggo.jf.intel.comSigned-off-by:
Dave Hansen <dave.hansen@linux.intel.com> Reviewed-by:
Ben Widawsky <ben.widawsky@intel.com> Reviewed-by:
Oscar Salvador <osalvador@suse.de> Acked-by:
Christoph Lameter <cl@linux.com> Acked-by:
-
Dave Hansen authored
It is currently not obvious that the RECLAIM_* bits are part of the uapi since they are defined in vmscan.c. Move them to a uapi header to make it obvious. This should have no functional impact. Link: https://lkml.kernel.org/r/20210219172557.08074910@viggo.jf.intel.comSigned-off-by:
-
Axel Rasmussen authored
Fix a dormant bug in userfaultfd_events_test(), where we did `return faulting_process(0)` instead of `exit(faulting_process(0))`. This caused the forked process to keep running, trying to execute any further test cases after the events test in parallel with the "real" process. Add a simple test case which exercises minor faults. In short, it does the following: 1. "Sets up" an area (area_dst) and a second shared mapping to the same underlying pages (area_dst_alias). 2. Register one of these areas with userfaultfd, in minor fault mode. 3. Start a second thread to handle any minor faults. 4. Populate the underlying pages with the non-UFFD-registered side of the mapping. Basically, memset() each page with some arbitrary contents. 5. Then, using the UFFD-registered mapping, read all of the page contents, asserting that the contents match expectations (we expect the minor fault handling thread can modify the page contents before resolving the fault). The minor fault handling thread, upon receiving an event, flips all the bits (~) in that page, just to prove that it can modify it in some arbitrary way. Then it issues a UFFDIO_CONTINUE ioctl, to setup the mapping and resolve the fault. The reading thread should wake up and see this modification. Currently the minor fault test is only enabled in hugetlb_shared mode, as this is the only configuration the kernel feature supports. Link: https://lkml.kernel.org/r/20210301222728.176417-7-axelrasmussen@google.comSigned-off-by:
Axel Rasmussen <axelrasmussen@google.com> Reviewed-by:
Peter Xu <peterx@redhat.com> Cc: Adam Ruprecht <ruprecht@google.com> Cc: Alexander Viro <viro@zeniv.linux.org.uk> Cc: Alexey Dobriyan <adobriyan@gmail.com> Cc: Andrea Arcangeli <aarcange@redhat.com> Cc: Anshuman Khandual <anshuman.khandual@arm.com> Cc: Cannon Matthews <cannonmatthews@google.com> Cc: Catalin Marinas <catalin.marinas@arm.com> Cc: Chinwen Chang <chinwen.chang@mediatek.com> Cc: David Rientjes <rientjes@google.com> Cc: "Dr . David Alan Gilbert" <dgilbert@redhat.com> Cc: Huang Ying <ying.huang@intel.com> Cc: Ingo Molnar <mingo@redhat.com> Cc: Jann Horn <jannh@google.com> Cc: Jerome Glisse <jglisse@redhat.com> Cc: Kirill A. Shutemov <kirill@shutemov.name> Cc: Lokesh Gidra <lokeshgidra@google.com> Cc: "Matthew Wilcox (Oracle)" <willy@infradead.org> Cc: Michael Ellerman <mpe@ellerman.id.au> Cc: "Michal Koutn" <mkoutny@suse.com> Cc: Michel Lespinasse <walken@google.com> Cc: Mike Kravetz <mike.kravetz@oracle.com> Cc: Mike Rapoport <rppt@linux.vnet.ibm.com> Cc: Mina Almasry <almasrymina@google.com> Cc: Nicholas Piggin <npiggin@gmail.com> Cc: Oliver Upton <oupton@google.com> Cc: Shaohua Li <shli@fb.com> Cc: Shawn Anastasio <shawn@anastas.io> Cc: Steven Price <steven.price@arm.com> Cc: Steven Rostedt <rostedt@goodmis.org> Cc: Vlastimil Babka <vbabka@suse.cz> Signed-off-by:
-
Axel Rasmussen authored
Reword / reorganize things a little bit into "lists", so new features / modes / ioctls can sort of just be appended. Describe how UFFDIO_REGISTER_MODE_MINOR and UFFDIO_CONTINUE can be used to intercept and resolve minor faults. Make it clear that COPY and ZEROPAGE are used for MISSING faults, whereas CONTINUE is used for MINOR faults. Link: https://lkml.kernel.org/r/20210301222728.176417-6-axelrasmussen@google.comSigned-off-by:
-
Axel Rasmussen authored
This ioctl is how userspace ought to resolve "minor" userfaults. The idea is, userspace is notified that a minor fault has occurred. It might change the contents of the page using its second non-UFFD mapping, or not. Then, it calls UFFDIO_CONTINUE to tell the kernel "I have ensured the page contents are correct, carry on setting up the mapping". Note that it doesn't make much sense to use UFFDIO_{COPY,ZEROPAGE} for MINOR registered VMAs. ZEROPAGE maps the VMA to the zero page; but in the minor fault case, we already have some pre-existing underlying page. Likewise, UFFDIO_COPY isn't useful if we have a second non-UFFD mapping. We'd just use memcpy() or similar instead. It turns out hugetlb_mcopy_atomic_pte() already does very close to what we want, if an existing page is provided via `struct page **pagep`. We already special-case the behavior a bit for the UFFDIO_ZEROPAGE case, so just extend that design: add an enum for the three modes of operation, and make the small adjustments needed for the MCOPY_ATOMIC_CONTINUE case. (Basically, look up the existing page, and avoid adding the existing page to the page cache or calling set_page_huge_active() on it.) Link: https://lkml.kernel.org/r/20210301222728.176417-5-axelrasmussen@google.comSigned-off-by: -
Axel Rasmussen authored
For background, mm/userfaultfd.c provides a general mcopy_atomic implementation. But some types of memory (i.e., hugetlb and shmem) need a slightly different implementation, so they provide their own helpers for this. In other words, userfaultfd is the only caller of these functions. This patch achieves two things: 1. Don't spend time compiling code which will end up never being referenced anyway (a small build time optimization). 2. In patches later in this series, we extend the signature of these helpers with UFFD-specific state (a mode enumeration). Once this happens, we *have to* either not compile the helpers, or unconditionally define the UFFD-only state (which seems messier to me). This includes the declarations in the headers, as otherwise they'd yield warnings about implicitly defining the type of those arguments. Link: https://lkml.kernel.org/r/20210301222728.176417-4-axelrasmussen@google.comSigned-off-by:
Mike Kravetz <mike.kravetz@oracle.com> Reviewed-by:
-
Axel Rasmussen authored
As the comment says: for the MINOR fault use case, although the page might be present and populated in the other (non-UFFD-registered) half of the mapping, it may be out of date, and we explicitly want userspace to get a minor fault so it can check and potentially update the page's contents. Huge PMD sharing would prevent these faults from occurring for suitably aligned areas, so disable it upon UFFD registration. Link: https://lkml.kernel.org/r/20210301222728.176417-3-axelrasmussen@google.comSigned-off-by:
-
Axel Rasmussen authored
Patch series "userfaultfd: add minor fault handling", v9. Overview ======== This series adds a new userfaultfd feature, UFFD_FEATURE_MINOR_HUGETLBFS. When enabled (via the UFFDIO_API ioctl), this feature means that any hugetlbfs VMAs registered with UFFDIO_REGISTER_MODE_MISSING will *also* get events for "minor" faults. By "minor" fault, I mean the following situation: Let there exist two mappings (i.e., VMAs) to the same page(s) (shared memory). One of the mappings is registered with userfaultfd (in minor mode), and the other is not. Via the non-UFFD mapping, the underlying pages have already been allocated & filled with some contents. The UFFD mapping has not yet been faulted in; when it is touched for the first time, this results in what I'm calling a "minor" fault. As a concrete example, when working with hugetlbfs, we have huge_pte_none(), but find_lock_page() finds an existing page. We also add a new ioctl to resolve such faults: UFFDIO_CONTINUE. The idea is, userspace resolves the fault by either a) doing nothing if the contents are already correct, or b) updating the underlying contents using the second, non-UFFD mapping (via memcpy/memset or similar, or something fancier like RDMA, or etc...). In either case, userspace issues UFFDIO_CONTINUE to tell the kernel "I have ensured the page contents are correct, carry on setting up the mapping". Use Case ======== Consider the use case of VM live migration (e.g. under QEMU/KVM): 1. While a VM is still running, we copy the contents of its memory to a target machine. The pages are populated on the target by writing to the non-UFFD mapping, using the setup described above. The VM is still running (and therefore its memory is likely changing), so this may be repeated several times, until we decide the target is "up to date enough". 2. We pause the VM on the source, and start executing on the target machine. During this gap, the VM's user(s) will *see* a pause, so it is desirable to minimize this window. 3. Between the last time any page was copied from the source to the target, and when the VM was paused, the contents of that page may have changed - and therefore the copy we have on the target machine is out of date. Although we can keep track of which pages are out of date, for VMs with large amounts of memory, it is "slow" to transfer this information to the target machine. We want to resume execution before such a transfer would complete. 4. So, the guest begins executing on the target machine. The first time it touches its memory (via the UFFD-registered mapping), userspace wants to intercept this fault. Userspace checks whether or not the page is up to date, and if not, copies the updated page from the source machine, via the non-UFFD mapping. Finally, whether a copy was performed or not, userspace issues a UFFDIO_CONTINUE ioctl to tell the kernel "I have ensured the page contents are correct, carry on setting up the mapping". We don't have to do all of the final updates on-demand. The userfaultfd manager can, in the background, also copy over updated pages once it receives the map of which pages are up-to-date or not. Interaction with Existing APIs ============================== Because this is a feature, a registered VMA could potentially receive both missing and minor faults. I spent some time thinking through how the existing API interacts with the new feature: UFFDIO_CONTINUE cannot be used to resolve non-minor faults, as it does not allocate a new page. If UFFDIO_CONTINUE is used on a non-minor fault: - For non-shared memory or shmem, -EINVAL is returned. - For hugetlb, -EFAULT is returned. UFFDIO_COPY and UFFDIO_ZEROPAGE cannot be used to resolve minor faults. Without modifications, the existing codepath assumes a new page needs to be allocated. This is okay, since userspace must have a second non-UFFD-registered mapping anyway, thus there isn't much reason to want to use these in any case (just memcpy or memset or similar). - If UFFDIO_COPY is used on a minor fault, -EEXIST is returned. - If UFFDIO_ZEROPAGE is used on a minor fault, -EEXIST is returned (or -EINVAL in the case of hugetlb, as UFFDIO_ZEROPAGE is unsupported in any case). - UFFDIO_WRITEPROTECT simply doesn't work with shared memory, and returns -ENOENT in that case (regardless of the kind of fault). Future Work =========== This series only supports hugetlbfs. I have a second series in flight to support shmem as well, extending the functionality. This series is more mature than the shmem support at this point, and the functionality works fully on hugetlbfs, so this series can be merged first and then shmem support will follow. This patch (of 6): This feature allows userspace to intercept "minor" faults. By "minor" faults, I mean the following situation: Let there exist two mappings (i.e., VMAs) to the same page(s). One of the mappings is registered with userfaultfd (in minor mode), and the other is not. Via the non-UFFD mapping, the underlying pages have already been allocated & filled with some contents. The UFFD mapping has not yet been faulted in; when it is touched for the first time, this results in what I'm calling a "minor" fault. As a concrete example, when working with hugetlbfs, we have huge_pte_none(), but find_lock_page() finds an existing page. This commit adds the new registration mode, and sets the relevant flag on the VMAs being registered. In the hugetlb fault path, if we find that we have huge_pte_none(), but find_lock_page() does indeed find an existing page, then we have a "minor" fault, and if the VMA has the userfaultfd registration flag, we call into userfaultfd to handle it. This is implemented as a new registration mode, instead of an API feature. This is because the alternative implementation has significant drawbacks [1]. However, doing it this was requires we allocate a VM_* flag for the new registration mode. On 32-bit systems, there are no unused bits, so this feature is only supported on architectures with CONFIG_ARCH_USES_HIGH_VMA_FLAGS. When attempting to register a VMA in MINOR mode on 32-bit architectures, we return -EINVAL. [1] https://lore.kernel.org/patchwork/patch/1380226/ [peterx@redhat.com: fix minor fault page leak] Link: https://lkml.kernel.org/r/20210322175132.36659-1-peterx@redhat.com Link: https://lkml.kernel.org/r/20210301222728.176417-1-axelrasmussen@google.com Link: https://lkml.kernel.org/r/20210301222728.176417-2-axelrasmussen@google.comSigned-off-by:
-
Oscar Salvador authored
pfn_range_valid_contig() bails out when it finds an in-use page or a hugetlb page, among other things. We can drop the in-use page check since __alloc_contig_pages can migrate away those pages, and the hugetlb page check can go too since isolate_migratepages_range is now capable of dealing with hugetlb pages. Either way, those checks are racy so let the end function handle it when the time comes. Link: https://lkml.kernel.org/r/20210419075413.1064-8-osalvador@suse.deSigned-off-by:
David Hildenbrand <david@redhat.com> Reviewed-by:
Michal Hocko <mhocko@suse.com> Cc: Muchun Song <songmuchun@bytedance.com> Cc: Vlastimil Babka <vbabka@suse.cz> Signed-off-by:
-
Oscar Salvador authored
alloc_contig_range() will fail if it finds a HugeTLB page within the range, without a chance to handle them. Since HugeTLB pages can be migrated as any LRU or Movable page, it does not make sense to bail out without trying. Enable the interface to recognize in-use HugeTLB pages so we can migrate them, and have much better chances to succeed the call. Link: https://lkml.kernel.org/r/20210419075413.1064-7-osalvador@suse.deSigned-off-by:
-
Oscar Salvador authored
alloc_contig_range will fail if it ever sees a HugeTLB page within the range we are trying to allocate, even when that page is free and can be easily reallocated. This has proved to be problematic for some users of alloc_contic_range, e.g: CMA and virtio-mem, where those would fail the call even when those pages lay in ZONE_MOVABLE and are free. We can do better by trying to replace such page. Free hugepages are tricky to handle so as to no userspace application notices disruption, we need to replace the current free hugepage with a new one. In order to do that, a new function called alloc_and_dissolve_huge_page is introduced. This function will first try to get a new fresh hugepage, and if it succeeds, it will replace the old one in the free hugepage pool. The free page replacement is done under hugetlb_lock, so no external users of hugetlb will notice the change. To allocate the new huge page, we use alloc_buddy_huge_page(), so we do not have to deal with any counters, and prep_new_huge_page() is not called. This is valulable because in case we need to free the new page, we only need to call __free_pages(). Once we know that the page to be replaced is a genuine 0-refcounted huge page, we remove the old page from the freelist by remove_hugetlb_page(). Then, we can call __prep_new_huge_page() and __prep_account_new_huge_page() for the new huge page to properly initialize it and increment the hstate->nr_huge_pages counter (previously decremented by remove_hugetlb_page()). Once done, the page is enqueued by enqueue_huge_page() and it is ready to be used. There is one tricky case when page's refcount is 0 because it is in the process of being released. A missing PageHugeFreed bit will tell us that freeing is in flight so we retry after dropping the hugetlb_lock. The race window should be small and the next retry should make a forward progress. E.g: CPU0 CPU1 free_huge_page() isolate_or_dissolve_huge_page PageHuge() == T alloc_and_dissolve_huge_page alloc_buddy_huge_page() spin_lock_irq(hugetlb_lock) // PageHuge() && !PageHugeFreed && // !PageCount() spin_unlock_irq(hugetlb_lock) spin_lock_irq(hugetlb_lock) 1) update_and_free_page PageHuge() == F __free_pages() 2) enqueue_huge_page SetPageHugeFreed() spin_unlock_irq(&hugetlb_lock) spin_lock_irq(hugetlb_lock) 1) PageHuge() == F (freed by case#1 from CPU0) 2) PageHuge() == T PageHugeFreed() == T - proceed with replacing the page In the case above we retry as the window race is quite small and we have high chances to succeed next time. With regard to the allocation, we restrict it to the node the page belongs to with __GFP_THISNODE, meaning we do not fallback on other node's zones. Note that gigantic hugetlb pages are fenced off since there is a cyclic dependency between them and alloc_contig_range. Link: https://lkml.kernel.org/r/20210419075413.1064-6-osalvador@suse.deSigned-off-by: -
Oscar Salvador authored
Currently, prep_new_huge_page() performs two functions. It sets the right state for a new hugetlb, and increases the hstate's counters to account for the new page. Let us split its functionality into two separate functions, decoupling the handling of the counters from initializing a hugepage. The outcome is having __prep_new_huge_page(), which only initializes the page , and __prep_account_new_huge_page(), which adds the new page to the hstate's counters. This allows us to be able to set a hugetlb without having to worry about the counter/locking. It will prove useful in the next patch. prep_new_huge_page() still calls both functions. Link: https://lkml.kernel.org/r/20210419075413.1064-5-osalvador@suse.deSigned-off-by:
-
Oscar Salvador authored
Pages allocated via the page allocator or CMA get its private field cleared by means of post_alloc_hook(). Pages allocated during boot, that is directly from the memblock allocator, get cleared by paging_init()-> .. ->memmap_init_zone-> .. ->__init_single_page() before any memblock allocation. Based on this ground, let us remove the clearing of the flag from prep_new_huge_page() as it is not needed. This was a leftover from commit 6c037149 ("hugetlb: convert PageHugeFreed to HPageFreed flag"). Previously the explicit clearing was necessary because compound allocations do not get this initialization (see prep_compound_page). Link: https://lkml.kernel.org/r/20210419075413.1064-4-osalvador@suse.deSigned-off-by:
-
Oscar Salvador authored
Currently, isolate_migratepages_{range,block} and their callers use a pfn == 0 vs pfn != 0 scheme to let the caller know whether there was any error during isolation. This does not work as soon as we need to start reporting different error codes and make sure we pass them down the chain, so they are properly interpreted by functions like e.g: alloc_contig_range. Let us rework isolate_migratepages_{range,block} so we can report error codes. Since isolate_migratepages_block will stop returning the next pfn to be scanned, we reuse the cc->migrate_pfn field to keep track of that. Link: https://lkml.kernel.org/r/20210419075413.1064-3-osalvador@suse.deSigned-off-by: -
Oscar Salvador authored
Patch series "Make alloc_contig_range handle Hugetlb pages", v10. alloc_contig_range lacks the ability to handle HugeTLB pages. This can be problematic for some users, e.g: CMA and virtio-mem, where those users will fail the call if alloc_contig_range ever sees a HugeTLB page, even when those pages lay in ZONE_MOVABLE and are free. That problem can be easily solved by replacing the page in the free hugepage pool. In-use HugeTLB are no exception though, as those can be isolated and migrated as any other LRU or Movable page. This aims to improve alloc_contig_range->isolate_migratepages_block, so that HugeTLB pages can be recognized and handled. Since we also need to start reporting errors down the chain (e.g: -ENOMEM due to not be able to allocate a new hugetlb page), isolate_migratepages_{range,block} interfaces need to change to start reporting error codes instead of the pfn == 0 vs pfn != 0 scheme it is using right now. From now on, isolate_migratepages_block will not return the next pfn to be scanned anymore, but -EINTR, -ENOMEM or 0, so we the next pfn to be scanned will be recorded in cc->migrate_pfn field (as it is already done in isolate_migratepages_range()). Below is an insight from David (thanks), where the problem can clearly be seen: "Start a VM with 4G. Hotplug 1G via virtio-mem and online it to ZONE_MOVABLE. Allocate 512 huge pages. [root@localhost ~]# cat /proc/meminfo MemTotal: 5061512 kB MemFree: 3319396 kB MemAvailable: 3457144 kB ... HugePages_Total: 512 HugePages_Free: 512 HugePages_Rsvd: 0 HugePages_Surp: 0 Hugepagesize: 2048 kB The huge pages get partially allocate from ZONE_MOVABLE. Try unplugging 1G via virtio-mem (remember, all ZONE_MOVABLE). Inside the guest: [ 180.058992] alloc_contig_range: [1b8000, 1c0000) PFNs busy [ 180.060531] alloc_contig_range: [1b8000, 1c0000) PFNs busy [ 180.061972] alloc_contig_range: [1b8000, 1c0000) PFNs busy [ 180.063413] alloc_contig_range: [1b8000, 1c0000) PFNs busy [ 180.064838] alloc_contig_range: [1b8000, 1c0000) PFNs busy [ 180.065848] alloc_contig_range: [1bfc00, 1c0000) PFNs busy [ 180.066794] alloc_contig_range: [1bfc00, 1c0000) PFNs busy [ 180.067738] alloc_contig_range: [1bfc00, 1c0000) PFNs busy [ 180.068669] alloc_contig_range: [1bfc00, 1c0000) PFNs busy [ 180.069598] alloc_contig_range: [1bfc00, 1c0000) PFNs busy" And then with this patchset running: "Same experiment with ZONE_MOVABLE: a) Free huge pages: all memory can get unplugged again. b) Allocated/populated but idle huge pages: all memory can get unplugged again. c) Allocated/populated but all 512 huge pages are read/written in a loop: all memory can get unplugged again, but I get a single [ 121.192345] alloc_contig_range: [180000, 188000) PFNs busy Most probably because it happened to try migrating a huge page while it was busy. As virtio-mem retries on ZONE_MOVABLE a couple of times, it can deal with this temporary failure. Last but not least, I did something extreme: # cat /proc/meminfo MemTotal: 5061568 kB MemFree: 186560 kB MemAvailable: 354524 kB ... HugePages_Total: 2048 HugePages_Free: 2048 HugePages_Rsvd: 0 HugePages_Surp: 0 Triggering unplug would require to dissolve+alloc - which now fails when trying to allocate an additional ~512 huge pages (1G). As expected, I can properly see memory unplug not fully succeeding. + I get a fairly continuous stream of [ 226.611584] alloc_contig_range: [19f400, 19f800) PFNs busy ... But more importantly, the hugepage count remains stable, as configured by the admin (me): HugePages_Total: 2048 HugePages_Free: 2048 HugePages_Rsvd: 0 HugePages_Surp: 0" This patch (of 7): Currently, __alloc_contig_migrate_range can generate -EINTR, -ENOMEM or -EBUSY, and report them down the chain. The problem is that when migrate_pages() reports -ENOMEM, we keep going till we exhaust all the try-attempts (5 at the moment) instead of bailing out. migrate_pages() bails out right away on -ENOMEM because it is considered a fatal error. Do the same here instead of keep going and retrying. Note that this is not fixing a real issue, just a cosmetic change. Although we can save some cycles by backing off ealier Link: https://lkml.kernel.org/r/20210419075413.1064-1-osalvador@suse.de Link: https://lkml.kernel.org/r/20210419075413.1064-2-osalvador@suse.deSigned-off-by: -
Mike Kravetz authored
After making hugetlb lock irq safe and separating some functionality done under the lock, add some lockdep_assert_held to help verify locking. Link: https://lkml.kernel.org/r/20210409205254.242291-9-mike.kravetz@oracle.comSigned-off-by:
Miaohe Lin <linmiaohe@huawei.com> Reviewed-by:
Muchun Song <songmuchun@bytedance.com> Reviewed-by:
-
Mike Kravetz authored
Commit c77c0a8a ("mm/hugetlb: defer freeing of huge pages if in non-task context") was added to address the issue of free_huge_page being called from irq context. That commit hands off free_huge_page processing to a workqueue if !in_task. However, this doesn't cover all the cases as pointed out by 0day bot lockdep report [1]. : Possible interrupt unsafe locking scenario: : : CPU0 CPU1 : ---- ---- : lock(hugetlb_lock); : local_irq_disable(); : lock(slock-AF_INET); : lock(hugetlb_lock); : <Interrupt> : lock(slock-AF_INET); Shakeel has later explained that this is very likely TCP TX zerocopy from hugetlb pages scenario when the networking code drops a last reference to hugetlb page while having IRQ disabled. Hugetlb freeing path doesn't disable IRQ while holding hugetlb_lock so a lock dependency chain can lead to a deadlock. This commit addresses the issue by doing the following: - Make hugetlb_lock irq safe. This is mostly a simple process of changing spin_*lock calls to spin_*lock_irq* calls. - Make subpool lock irq safe in a similar manner. - Revert the !in_task check and workqueue handoff. [1] https://lore.kernel.org/linux-mm/000000000000f1c03b05bc43aadc@google.com/ Link: https://lkml.kernel.org/r/20210409205254.242291-8-mike.kravetz@oracle.comSigned-off-by:
-
Mike Kravetz authored
free_pool_huge_page was called with hugetlb_lock held. It would remove a hugetlb page, and then free the corresponding pages to the lower level allocators such as buddy. free_pool_huge_page was called in a loop to remove hugetlb pages and these loops could hold the hugetlb_lock for a considerable time. Create new routine remove_pool_huge_page to replace free_pool_huge_page. remove_pool_huge_page will remove the hugetlb page, and it must be called with the hugetlb_lock held. It will return the removed page and it is the responsibility of the caller to free the page to the lower level allocators. The hugetlb_lock is dropped before freeing to these allocators which results in shorter lock hold times. Add new helper routine to call update_and_free_page for a list of pages. Note: Some changes to the routine return_unused_surplus_pages are in need of explanation. Commit e5bbc8a6 ("mm/hugetlb.c: fix reservation race when freeing surplus pages") modified this routine to address a race which could occur when dropping the hugetlb_lock in the loop that removes pool pages. Accounting changes introduced in that commit were subtle and took some thought to understand. This commit removes the cond_resched_lock() and the potential race. Therefore, remove the subtle code and restore the more straight forward accounting effectively reverting the commit. Link: https://lkml.kernel.org/r/20210409205254.242291-7-mike.kravetz@oracle.comSigned-off-by:
-
Mike Kravetz authored
With the introduction of remove_hugetlb_page(), there is no need for update_and_free_page to hold the hugetlb lock. Change all callers to drop the lock before calling. With additional code modifications, this will allow loops which decrease the huge page pool to drop the hugetlb_lock with each page to reduce long hold times. The ugly unlock/lock cycle in free_pool_huge_page will be removed in a subsequent patch which restructures free_pool_huge_page. Link: https://lkml.kernel.org/r/20210409205254.242291-6-mike.kravetz@oracle.comSigned-off-by:
-
Mike Kravetz authored
The new remove_hugetlb_page() routine is designed to remove a hugetlb page from hugetlbfs processing. It will remove the page from the active or free list, update global counters and set the compound page destructor to NULL so that PageHuge() will return false for the 'page'. After this call, the 'page' can be treated as a normal compound page or a collection of base size pages. update_and_free_page no longer decrements h->nr_huge_pages{_node} as this is performed in remove_hugetlb_page. The only functionality performed by update_and_free_page is to free the base pages to the lower level allocators. update_and_free_page is typically called after remove_hugetlb_page. remove_hugetlb_page is to be called with the hugetlb_lock held. Creating this routine and separating functionality is in preparation for restructuring code to reduce lock hold times. This commit should not introduce any changes to functionality. Link: https://lkml.kernel.org/r/20210409205254.242291-5-mike.kravetz@oracle.comSigned-off-by: -
Mike Kravetz authored
The helper routine hstate_next_node_to_alloc accesses and modifies the hstate variable next_nid_to_alloc. The helper is used by the routines alloc_pool_huge_page and adjust_pool_surplus. adjust_pool_surplus is called with hugetlb_lock held. However, alloc_pool_huge_page can not be called with the hugetlb lock held as it will call the page allocator. Two instances of alloc_pool_huge_page could be run in parallel or alloc_pool_huge_page could run in parallel with adjust_pool_surplus which may result in the variable next_nid_to_alloc becoming invalid for the caller and pages being allocated on the wrong node. Both alloc_pool_huge_page and adjust_pool_surplus are only called from the routine set_max_huge_pages after boot. set_max_huge_pages is only called as the reusult of a user writing to the proc/sysfs nr_hugepages, or nr_hugepages_mempolicy file to adjust the number of hugetlb pages. It makes little sense to allow multiple adjustment to the number of hugetlb pages in parallel. Add a mutex to the hstate and use it to only allow one hugetlb page adjustment at a time. This will synchronize modifications to the next_nid_to_alloc variable. Link: https://lkml.kernel.org/r/20210409205254.242291-4-mike.kravetz@oracle.comSigned-off-by:
-
Mike Kravetz authored
Now that cma_release is non-blocking and irq safe, there is no need to drop hugetlb_lock before calling. Link: https://lkml.kernel.org/r/20210409205254.242291-3-mike.kravetz@oracle.comSigned-off-by:
-
Mike Kravetz authored
Patch series "make hugetlb put_page safe for all calling contexts", v5. This effort is the result a recent bug report [1]. Syzbot found a potential deadlock in the hugetlb put_page/free_huge_page_path. WARNING: SOFTIRQ-safe -> SOFTIRQ-unsafe lock order detected Since the free_huge_page_path already has code to 'hand off' page free requests to a workqueue, a suggestion was proposed to make the in_irq() detection accurate by always enabling PREEMPT_COUNT [2]. The outcome of that discussion was that the hugetlb put_page path (free_huge_page) path should be properly fixed and safe for all calling contexts. [1] https://lore.kernel.org/linux-mm/000000000000f1c03b05bc43aadc@google.com/ [2] http://lkml.kernel.org/r/20210311021321.127500-1-mike.kravetz@oracle.com This patch (of 8): cma_release is currently a sleepable operatation because the bitmap manipulation is protected by cma->lock mutex. Hugetlb code which relies on cma_release for CMA backed (giga) hugetlb pages, however, needs to be irq safe. The lock doesn't protect any sleepable operation so it can be changed to a (irq aware) spin lock. The bitmap processing should be quite fast in typical case but if cma sizes grow to TB then we will likely need to replace the lock by a more optimized bitmap implementation. Link: https://lkml.kernel.org/r/20210409205254.242291-1-mike.kravetz@oracle.com Link: https://lkml.kernel.org/r/20210409205254.242291-2-mike.kravetz@oracle.comSigned-off-by:
-
Miaohe Lin authored
The local variable pseudo_vma is not used anymore. Link: https://lkml.kernel.org/r/20210410072348.20437-6-linmiaohe@huawei.comSigned-off-by:
-
Miaohe Lin authored
A rare out of memory error would prevent removal of the reserve map region for a page. hugetlb_fix_reserve_counts() handles this rare case to avoid dangling with incorrect counts. Unfortunately, hugepage_subpool_get_pages and hugetlb_acct_memory could possibly fail too. We should correctly handle these cases. Link: https://lkml.kernel.org/r/20210410072348.20437-5-linmiaohe@huawei.com Fixes: b5cec28d ("hugetlbfs: truncate_hugepages() takes a range of pages") Signed-off-by:
-
