| Message ID | 20231024134637.3120277-1-surenb@google.com (mailing list archive) |
|---|---|
| Headers | show |
| Series | Memory allocation profiling | expand |
On Tue, Oct 24, 2023 at 06:45:57AM -0700, Suren Baghdasaryan wrote: > Updates since the last version [1] > - Simplified allocation tagging macros; > - Runtime enable/disable sysctl switch (/proc/sys/vm/mem_profiling) > instead of kernel command-line option; > - CONFIG_MEM_ALLOC_PROFILING_BY_DEFAULT to select default enable state; > - Changed the user-facing API from debugfs to procfs (/proc/allocinfo); > - Removed context capture support to make patch incremental; > - Renamed uninstrumented allocation functions to use _noprof suffix; > - Added __GFP_LAST_BIT to make the code cleaner; > - Removed lazy per-cpu counters; it turned out the memory savings was > minimal and not worth the performance impact; Hello Suren, > Performance overhead: > To evaluate performance we implemented an in-kernel test executing > multiple get_free_page/free_page and kmalloc/kfree calls with allocation > sizes growing from 8 to 240 bytes with CPU frequency set to max and CPU > affinity set to a specific CPU to minimize the noise. Below is performance > comparison between the baseline kernel, profiling when enabled, profiling > when disabled and (for comparison purposes) baseline with > CONFIG_MEMCG_KMEM enabled and allocations using __GFP_ACCOUNT: > > kmalloc pgalloc > (1 baseline) 12.041s 49.190s > (2 default disabled) 14.970s (+24.33%) 49.684s (+1.00%) > (3 default enabled) 16.859s (+40.01%) 56.287s (+14.43%) > (4 runtime enabled) 16.983s (+41.04%) 55.760s (+13.36%) > (5 memcg) 33.831s (+180.96%) 51.433s (+4.56%) some recent changes [1] to the kmem accounting should have made it quite a bit faster. Would be great if you can provide new numbers for the comparison. Maybe with the next revision? And btw thank you (and Kent): your numbers inspired me to do this kmemcg performance work. I expect it still to be ~twice more expensive than your stuff because on the memcg side we handle separately charge and statistics, but hopefully the difference will be lower. Thank you! [1]: patches from next tree, so no stable hashes: mm: kmem: reimplement get_obj_cgroup_from_current() percpu: scoped objcg protection mm: kmem: scoped objcg protection mm: kmem: make memcg keep a reference to the original objcg mm: kmem: add direct objcg pointer to task_struct mm: kmem: optimize get_obj_cgroup_from_current()
On Tue, Oct 24, 2023 at 11:29 AM Roman Gushchin <roman.gushchin@linux.dev> wrote: > > On Tue, Oct 24, 2023 at 06:45:57AM -0700, Suren Baghdasaryan wrote: > > Updates since the last version [1] > > - Simplified allocation tagging macros; > > - Runtime enable/disable sysctl switch (/proc/sys/vm/mem_profiling) > > instead of kernel command-line option; > > - CONFIG_MEM_ALLOC_PROFILING_BY_DEFAULT to select default enable state; > > - Changed the user-facing API from debugfs to procfs (/proc/allocinfo); > > - Removed context capture support to make patch incremental; > > - Renamed uninstrumented allocation functions to use _noprof suffix; > > - Added __GFP_LAST_BIT to make the code cleaner; > > - Removed lazy per-cpu counters; it turned out the memory savings was > > minimal and not worth the performance impact; > > Hello Suren, > > > Performance overhead: > > To evaluate performance we implemented an in-kernel test executing > > multiple get_free_page/free_page and kmalloc/kfree calls with allocation > > sizes growing from 8 to 240 bytes with CPU frequency set to max and CPU > > affinity set to a specific CPU to minimize the noise. Below is performance > > comparison between the baseline kernel, profiling when enabled, profiling > > when disabled and (for comparison purposes) baseline with > > CONFIG_MEMCG_KMEM enabled and allocations using __GFP_ACCOUNT: > > > > kmalloc pgalloc > > (1 baseline) 12.041s 49.190s > > (2 default disabled) 14.970s (+24.33%) 49.684s (+1.00%) > > (3 default enabled) 16.859s (+40.01%) 56.287s (+14.43%) > > (4 runtime enabled) 16.983s (+41.04%) 55.760s (+13.36%) > > (5 memcg) 33.831s (+180.96%) 51.433s (+4.56%) > > some recent changes [1] to the kmem accounting should have made it quite a bit > faster. Would be great if you can provide new numbers for the comparison. > Maybe with the next revision? > > And btw thank you (and Kent): your numbers inspired me to do this kmemcg > performance work. I expect it still to be ~twice more expensive than your > stuff because on the memcg side we handle separately charge and statistics, > but hopefully the difference will be lower. Yes, I saw them! Well done! I'll definitely update my numbers once the patches land in their final form. > > Thank you! Thank you for the optimizations! > > [1]: > patches from next tree, so no stable hashes: > mm: kmem: reimplement get_obj_cgroup_from_current() > percpu: scoped objcg protection > mm: kmem: scoped objcg protection > mm: kmem: make memcg keep a reference to the original objcg > mm: kmem: add direct objcg pointer to task_struct > mm: kmem: optimize get_obj_cgroup_from_current()
Updates since the last version [1] - Simplified allocation tagging macros; - Runtime enable/disable sysctl switch (/proc/sys/vm/mem_profiling) instead of kernel command-line option; - CONFIG_MEM_ALLOC_PROFILING_BY_DEFAULT to select default enable state; - Changed the user-facing API from debugfs to procfs (/proc/allocinfo); - Removed context capture support to make patch incremental; - Renamed uninstrumented allocation functions to use _noprof suffix; - Added __GFP_LAST_BIT to make the code cleaner; - Removed lazy per-cpu counters; it turned out the memory savings was minimal and not worth the performance impact; Things we could not address: - Alternative way of instrument allocation functions. We discussed an alternative way of instrumenting the allocators and Steven Rostedt wrote a proposal [2] to provide compiler support for Callsite Trampolines - a special attribution of functions to be instrumented. So far we spoke to representatives of GNU and CLANG communities, will be presenting a proposal at LPC 2023 and we are working on a proof of concept for CLANG (see [example 1]). While we will keep working with compiler community on adding this support, posting the latest version of the patchset as an immediately available solution until compiler support is implemented; - Reclaim memory used by pageexts, slabexts, pcpuexts when profiling is disabled. The main obstacles of reclaiming the memory used for profiling: 1. References from already allocated objects still point to the allocation tags we are trying to reclaim; We would have to scan and kill all references before reclaiming; 2. pageext memory is allocated during early boot when fragmentation is low. Once reclaimed we might not be able to get it back; 3. pageext and slabext objects used for profiling can be interleaved with other data. Reformatting these vectors of objects at runtime is a complex and racy task. Overview Memory allocation profiling infrastructure provides a low overhead mechanism to make all kernel allocations in the system visible. It can be used to monitor memory usage, track memory hotspots, detect memory leaks, identify memory regressions. To keep the overhead to the minimum, we record only allocation sizes for every allocation in the codebase. The data is exposed to the user space via /proc/allocinfo interface. Usage example: $ sort -hr /proc/allocinfo | head 153MiB 8599 mm/slub.c:1826 module:slub func:alloc_slab_page 6.08MiB 49 mm/slab_common.c:950 module:slab_common func:_kmalloc_order 5.09MiB 6335 mm/memcontrol.c:2814 module:memcontrol func:alloc_slab_obj_exts 4.54MiB 78 mm/page_alloc.c:5777 module:page_alloc func:alloc_pages_exact 1.32MiB 338 include/asm-generic/pgalloc.h:63 module:pgtable func:__pte_alloc_one 1.16MiB 603 fs/xfs/xfs_log_priv.h:700 module:xfs func:xlog_kvmalloc 1.00MiB 256 mm/swap_cgroup.c:48 module:swap_cgroup func:swap_cgroup_prepare 734KiB 5380 fs/xfs/kmem.c:20 module:xfs func:kmem_alloc 640KiB 160 kernel/rcu/tree.c:3184 module:tree func:fill_page_cache_func 640KiB 160 drivers/char/virtio_console.c:452 module:virtio_console func:alloc_buf Support for more detailed allocation context including pid, tgid, task name, allocation size, timestamp and call stack is not posted in this patchset to keep it small. Implementation utilizes a more generic concept of code tagging, introduced as part of this patchset. Code tag is a structure identifying a specific location in the source code which is generated at compile time and can be embedded in an application-specific structure. A number of applications for code tagging have been presented in the original RFC [3]. Code tagging uses the old trick of "define a special elf section for objects of a given type so that we can iterate over them at runtime" and creates a proper library for it. To profile memory allocations, we instrument page, slab and percpu allocators to record total memory allocated in the associated code tag at every allocation in the codebase. Every time an allocation is performed by an instrumented allocator, the code tag at that location increments its counter by allocation size. Every time the memory is freed the counter is decremented. To decrement the counter upon freeing, allocated object needs a reference to its code tag. Page allocators use page_ext to record this reference while slab allocators use memcg_data (renamed into more generic slabobj_ext) of the slab page. Module allocations are accounted for the same way as other kernel allocations. Module loading and unloading is supported. If a module is unloaded while one or more of its allocations is still not freed (rather rare condition), its data section will be kept in memory to allow later code tag referencing when the allocation is freed later on. As part of this series we introduce several kernel configs: CONFIG_CODE_TAGGING - to enable code tagging framework. CONFIG_MEM_ALLOC_PROFILING - enables memory allocation profiling. CONFIG_MEM_ALLOC_PROFILING_ENABLED_BY_DEFAULT - enables memory allocation profiling by default. CONFIG_MEM_ALLOC_PROFILING_DEBUG - enables memory allocation profiling validation. Note: CONFIG_MEM_ALLOC_PROFILING enables CONFIG_PAGE_EXTENSION to store code tag reference in the page_ext object. /proc/sys/vm/mem_profiling sysctl is provided to enable/disable the functionality and avoid the performance overhead. Overhead To measure the overhead we are comparing the following configurations: (1) Baseline (2) Disabled by default (CONFIG_MEM_ALLOC_PROFILING & !CONFIG_MEM_ALLOC_PROFILING_BY_DEFAULT) (3) Enabled by default (CONFIG_MEM_ALLOC_PROFILING & CONFIG_MEM_ALLOC_PROFILING_BY_DEFAULT) (4) Enabled at runtime (CONFIG_MEM_ALLOC_PROFILING & !CONFIG_MEM_ALLOC_PROFILING_BY_DEFAULT & /proc/sys/vm/mem_profiling=1) (5) Memcg (CONFIG_MEMCG_KMEM) (6) Enabled by default with memcg (CONFIG_MEM_ALLOC_PROFILING & CONFIG_MEM_ALLOC_PROFILING_BY_DEFAULT & CONFIG_MEMCG_KMEM) Performance overhead: To evaluate performance we implemented an in-kernel test executing multiple get_free_page/free_page and kmalloc/kfree calls with allocation sizes growing from 8 to 240 bytes with CPU frequency set to max and CPU affinity set to a specific CPU to minimize the noise. Below is performance comparison between the baseline kernel, profiling when enabled, profiling when disabled and (for comparison purposes) baseline with CONFIG_MEMCG_KMEM enabled and allocations using __GFP_ACCOUNT: kmalloc pgalloc (1 baseline) 12.041s 49.190s (2 default disabled) 14.970s (+24.33%) 49.684s (+1.00%) (3 default enabled) 16.859s (+40.01%) 56.287s (+14.43%) (4 runtime enabled) 16.983s (+41.04%) 55.760s (+13.36%) (5 memcg) 33.831s (+180.96%) 51.433s (+4.56%) (6 enabled & memcg) 39.145s (+225.10%) 56.874s (+15.62%) Memory overhead: Kernel size: text data bss dec hex (1) 32638461 18286426 18325508 69250395 420ad5b (2) 32710110 18646586 18071556 69428252 423641c (3) 32706918 18646586 18071556 69425060 42357a4 (4) 32709664 18646586 18071556 69427806 423625e (5) 32715893 18345334 18239492 69300719 42171ef (6) 32786068 18701958 17993732 69481758 424351e Memory consumption on a 56 core Intel CPU with 125GB of memory running Fedora: Code tags: 192 kB PageExts: 262144 kB (256MB) SlabExts: 9876 kB (9.6MB) PcpuExts: 512 kB (0.5MB) Total overhead is 0.2% of total memory. [1] https://lore.kernel.org/all/20230501165450.15352-1-surenb@google.com/ [2] https://docs.google.com/presentation/d/1zQnuMbEfcq9lHUXgJRUZsd1McRAkr3Xq6Wk693YA0To/edit?usp=sharing [3] https://lore.kernel.org/all/20220830214919.53220-1-surenb@google.com/ [example 1]: typedef struct codetag { const char* file; int line; int counter; } codetag; void my_trampoline(func_ptr func, ...) { static codetag callsite_data __section("alloc_tags") = { __callsite_FILE, __callsite_LINE, 0 }; callsite_data.counter++; func(...); } __callsite_wrapper(my_trampoline) __attribute__ ((always_inline)) static inline void foo1(void) { printf("foo1 function\n"); } __callsite_wrapper(my_trampoline) __attribute__ ((always_inline)) static inline void foo2(void) { printf("foo2 function\n"); } void bar(void) { foo1(); } int main(int argc, char** argv) { foo1(); foo2(); bar(); return 0; } Kent Overstreet (16): lib/string_helpers: Add flags param to string_get_size() scripts/kallysms: Always include __start and __stop symbols fs: Convert alloc_inode_sb() to a macro nodemask: Split out include/linux/nodemask_types.h prandom: Remove unused include change alloc_pages name in ivpu_bo_ops to avoid conflicts mm/slub: Mark slab_free_freelist_hook() __always_inline mempool: Hook up to memory allocation profiling xfs: Memory allocation profiling fixups timekeeping: Fix a circular include dependency mm: percpu: Introduce pcpuobj_ext mm: percpu: Add codetag reference into pcpuobj_ext arm64: Fix circular header dependency mm: vmalloc: Enable memory allocation profiling rhashtable: Plumb through alloc tag MAINTAINERS: Add entries for code tagging and memory allocation profiling Suren Baghdasaryan (23): mm: enumerate all gfp flags mm: introduce slabobj_ext to support slab object extensions mm: introduce __GFP_NO_OBJ_EXT flag to selectively prevent slabobj_ext creation mm/slab: introduce SLAB_NO_OBJ_EXT to avoid obj_ext creation mm: prevent slabobj_ext allocations for slabobj_ext and kmem_cache objects slab: objext: introduce objext_flags as extension to page_memcg_data_flags lib: code tagging framework lib: code tagging module support lib: prevent module unloading if memory is not freed lib: add allocation tagging support for memory allocation profiling lib: introduce support for page allocation tagging change alloc_pages name in dma_map_ops to avoid name conflicts mm: enable page allocation tagging mm: create new codetag references during page splitting mm/page_ext: enable early_page_ext when CONFIG_MEM_ALLOC_PROFILING_DEBUG=y lib: add codetag reference into slabobj_ext mm/slab: add allocation accounting into slab allocation and free paths mm/slab: enable slab allocation tagging for kmalloc and friends mm: percpu: enable per-cpu allocation tagging lib: add memory allocations report in show_mem() codetag: debug: skip objext checking when it's for objext itself codetag: debug: mark codetags for reserved pages as empty codetag: debug: introduce OBJEXTS_ALLOC_FAIL to mark failed slab_ext allocations Documentation/admin-guide/sysctl/vm.rst | 16 ++ Documentation/filesystems/proc.rst | 28 ++ MAINTAINERS | 16 ++ arch/arm64/include/asm/spectre.h | 4 +- arch/powerpc/mm/book3s64/radix_pgtable.c | 2 +- arch/x86/kernel/amd_gart_64.c | 2 +- drivers/accel/ivpu/ivpu_gem.c | 8 +- drivers/accel/ivpu/ivpu_gem.h | 2 +- drivers/block/virtio_blk.c | 4 +- drivers/gpu/drm/gud/gud_drv.c | 2 +- drivers/iommu/dma-iommu.c | 2 +- drivers/mmc/core/block.c | 4 +- drivers/mtd/spi-nor/debugfs.c | 6 +- .../ethernet/chelsio/cxgb4/cxgb4_debugfs.c | 4 +- drivers/scsi/sd.c | 8 +- drivers/staging/media/atomisp/pci/hmm/hmm.c | 2 +- drivers/xen/grant-dma-ops.c | 2 +- drivers/xen/swiotlb-xen.c | 2 +- fs/xfs/kmem.c | 4 +- fs/xfs/kmem.h | 10 +- include/asm-generic/codetag.lds.h | 14 + include/asm-generic/vmlinux.lds.h | 3 + include/linux/alloc_tag.h | 188 +++++++++++++ include/linux/codetag.h | 83 ++++++ include/linux/dma-map-ops.h | 2 +- include/linux/fortify-string.h | 5 +- include/linux/fs.h | 6 +- include/linux/gfp.h | 111 +++++--- include/linux/gfp_types.h | 101 +++++-- include/linux/hrtimer.h | 2 +- include/linux/memcontrol.h | 56 +++- include/linux/mempool.h | 73 +++-- include/linux/mm.h | 8 + include/linux/mm_types.h | 4 +- include/linux/nodemask.h | 2 +- include/linux/nodemask_types.h | 9 + include/linux/page_ext.h | 1 - include/linux/pagemap.h | 9 +- include/linux/percpu.h | 23 +- include/linux/pgalloc_tag.h | 105 +++++++ include/linux/prandom.h | 1 - include/linux/rhashtable-types.h | 11 +- include/linux/sched.h | 26 +- include/linux/slab.h | 180 ++++++------ include/linux/slab_def.h | 2 +- include/linux/slub_def.h | 4 +- include/linux/string.h | 4 +- include/linux/string_helpers.h | 13 +- include/linux/time_namespace.h | 2 + include/linux/vmalloc.h | 60 +++- init/Kconfig | 4 + kernel/dma/mapping.c | 4 +- kernel/kallsyms_selftest.c | 2 +- kernel/module/main.c | 25 +- lib/Kconfig.debug | 31 +++ lib/Makefile | 3 + lib/alloc_tag.c | 212 ++++++++++++++ lib/codetag.c | 258 ++++++++++++++++++ lib/rhashtable.c | 52 +++- lib/string_helpers.c | 24 +- lib/test-string_helpers.c | 4 +- mm/compaction.c | 7 +- mm/filemap.c | 6 +- mm/huge_memory.c | 2 + mm/hugetlb.c | 8 +- mm/kfence/core.c | 14 +- mm/kfence/kfence.h | 4 +- mm/memcontrol.c | 56 +--- mm/mempolicy.c | 42 +-- mm/mempool.c | 34 +-- mm/mm_init.c | 1 + mm/page_alloc.c | 66 +++-- mm/page_ext.c | 13 + mm/page_owner.c | 2 +- mm/percpu-internal.h | 26 +- mm/percpu.c | 120 ++++---- mm/show_mem.c | 15 + mm/slab.c | 24 +- mm/slab.h | 246 +++++++++++++---- mm/slab_common.c | 96 +++++-- mm/slub.c | 26 +- mm/util.c | 44 +-- mm/vmalloc.c | 88 +++--- scripts/kallsyms.c | 13 + scripts/module.lds.S | 7 + 85 files changed, 2117 insertions(+), 698 deletions(-) create mode 100644 include/asm-generic/codetag.lds.h create mode 100644 include/linux/alloc_tag.h create mode 100644 include/linux/codetag.h create mode 100644 include/linux/nodemask_types.h create mode 100644 include/linux/pgalloc_tag.h create mode 100644 lib/alloc_tag.c create mode 100644 lib/codetag.c