| Message ID | 20240618204750.79512-1-sidhartha.kumar@oracle.com (mailing list archive) |
|---|---|
| Headers | show |
| Series | Introduce a store type enum for the Maple tree | expand |
Thanks! Reviewed-by: Liam R. Howlett <Liam.Howlett@oracle.com> * Sidhartha Kumar <sidhartha.kumar@oracle.com> [240618 16:48]: > This series is rebased on top of mm-unstable + the patch: > maple_tree: modified return type of mas_wr_store_entry()[1]. Andrew could > you add that patch to mm-unstable before merging this series. > > > v2[2] -> v3: > - fix new line issues throughout the series > > - remove use of helper function in patch 13 > > > v1[3] -> v2: > - previous results were noisy, replaced with stress-ng-mmap config of > mmtests > > - document function parameter 'entry'in 'mas_prealloc_calc per Intel > Test Robot > > - explain why mas_reset() was dropeed in patch 4 > > - add locking and mas_destroy() to testcase in patch 4 > > - squash "set write store type" vs "set store type" into the > mas_wr_store_entry() modification > > - only set ret to xa_err() in the err case in mas_store_gfp() > > - move MT_BUG_ON() inside lock in patch 7 > > - noinline mas_wr_spanning store to reduce stack usage in > mas_wr_store_type() in patch 11 per Intel Test Robot > > - document function parameter gfp in mas_insert() per Intel > test robot. > > - further simplify the local variables in patch 11 > > ================================ OVERVIEW ================================ > > This series implements two work items[4]: "aligning mas_store_gfp() with > mas_preallocate()" and "enum for store type". > > mas_store_gfp() is modified to preallocate nodes. This simplies many of > the write helper functions by allowing them to use mas_store_gfp() rather > than open coding node allocation and error handling. > > The enum defines the following store types: > > enum store_type { > wr_invalid, > wr_new_root, > wr_store_root, > wr_exact_fit, > wr_spanning_store, > wr_split_store, > wr_rebalance, > wr_append, > wr_node_store, > wr_slot_store, > wr_bnode > }; > > In the current maple tree code, a walk down the tree is done in > mas_preallocate() to determine the number of nodes needed for this write. > After node allocation, mas_wr_store_entry() will perform another walk to > determine which write helper function to use to complete the write. > > Rather than performing the second walk, we can store the type of write > in the maple write state during node allocation and read this field to > complete the write. > > ================================ RESULTS ================================= > > v6.10_mmap v6.10_mmap_store_type > Duration User 9.80 8.69 > Duration System 2295.94 2243.85 > Duration Elapsed 1010.50 1009.44 > > ================================ TESTING ================================= > > Testing was done with the maple tree test suite. A new test case is also > added to validate the order in which we test for and assign the store type. > > [1]: https://lore.kernel.org/linux-mm/20240614092428.29491-1-rgbi3307@gmail.com/T/ > [2]: https://lore.kernel.org/linux-mm/20240607185257.963768-1-sidhartha.kumar@oracle.com/T/#ma3795b93f9d5c9c9286291e12f089bff45b87fed > [3]: https://lore.kernel.org/linux-mm/20240604174145.563900-1-sidhartha.kumar@oracle.com/ > [4]: https://lists.infradead.org/pipermail/maple-tree/2023-December/003098.html > > > Sidhartha Kumar (16): > maple_tree: introduce store_type enum > maple_tree: introduce mas_wr_prealloc_setup() > maple_tree: move up mas_wr_store_setup() and mas_wr_prealloc_setup() > maple_tree: introduce mas_wr_store_type() > maple_tree: remove mas_destroy() from mas_nomem() > maple_tree: use mas_store_gfp() in mas_erase() > maple_tree: use mas_store_gfp() in mtree_store_range() > maple_tree: print store type in mas_dump() > maple_tree: use store type in mas_wr_store_entry() > maple_tree: convert mas_insert() to preallocate nodes > maple_tree: simplify mas_commit_b_node() > maple_tree: remove mas_wr_modify() > maple_tree: have mas_store() allocate nodes if needed > maple_tree: remove node allocations from various write helper > functions > maple_tree: remove repeated sanity checks from mas_wr_append() > maple_tree: remove unneeded mas_wr_walk() in mas_store_prealloc() > > include/linux/maple_tree.h | 15 + > lib/maple_tree.c | 580 ++++++++++++++++++------------- > tools/testing/radix-tree/maple.c | 46 ++- > 3 files changed, 396 insertions(+), 245 deletions(-) > > -- > 2.45.2 >
On Tue, 18 Jun 2024, Sidhartha Kumar wrote: > This series is rebased on top of mm-unstable + the patch: > maple_tree: modified return type of mas_wr_store_entry()[1]. Andrew could > you add that patch to mm-unstable before merging this series. > > v2[2] -> v3: > - fix new line issues throughout the series > - remove use of helper function in patch 13 Please give tmpfs a try on the latest mm-unstable, with CONFIG_DEBUG_ATOMIC_SLEEP=y (maybe some of the messages below come from other config options like PROVE_LOCKING, but ATOMIC_SLEEP the main one). To the un-maple-trained eye, this series simply replaces a working maple tree preallocation scheme by a broken one, doing GFP_KERNEL allocations while holding spinlock. But I doubt that was the intention: maybe a line of code has gone missing or something, and you can quickly unbreak it. BUG: sleeping function called from invalid context at include/linux/sched/mm.h:337 in_atomic(): 1, irqs_disabled(): 0, non_block: 0, pid: 63, name: kdevtmpfs preempt_count: 1, expected: 0 RCU nest depth: 0, expected: 0 3 locks held by kdevtmpfs/63: #0: ffff8880008473f0 (sb_writers){.+.+}-{0:0}, at: mnt_want_write+0x19/0x40 #1: ffff8880008a0888 (&type->i_mutex_dir_key/1){+.+.}-{3:3}, at: filename_create+0x8a/0x120 #2: ffff8880008a0650 (&simple_offset_lock_class){+.+.}-{2:2}, at: mtree_alloc_cyclic+0x72/0xb0 Preemption disabled at: [<ffffffff81138b94>] preempt_count_add+0x54/0x60 CPU: 4 UID: 0 PID: 63 Comm: kdevtmpfs Not tainted 6.10.0-rc5-m25 #2 Hardware name: LENOVO 20XXS3LA00/20XXS3LA00, BIOS N32ET91W (1.67 ) 02/02/2024 Call Trace: <TASK> dump_stack_lvl+0x5d/0x80 ? preempt_count_add+0x54/0x60 dump_stack+0x10/0x20 __might_resched+0x23b/0x260 ? mas_alloc_nodes+0x71/0x160 __might_sleep+0x56/0x60 might_alloc+0x2a/0x40 kmem_cache_alloc_noprof+0x28/0x190 mas_alloc_nodes+0x71/0x160 ? lock_is_held+0xc/0x10 mas_node_count_gfp+0x2e/0x30 mas_wr_preallocate+0x43/0x60 mas_insert.isra.0+0x49/0xa0 mas_alloc_cyclic+0x9c/0x100 mtree_alloc_cyclic+0x92/0xb0 simple_offset_add+0x3c/0x60 shmem_mknod+0x55/0xb0 vfs_mknod+0x9c/0xc0 devtmpfs_work_loop+0x1c4/0x2a0 ? trace_hardirqs_on+0x37/0x40 ? _raw_spin_unlock_irqrestore+0x39/0x50 ? complete_with_flags+0x40/0x50 ? dmar_validate_one_drhd+0xa0/0xa0 devtmpfsd+0x25/0x30 kthread+0x100/0x110 ? list_del_init+0x30/0x30 ret_from_fork+0x22/0x40 ? list_del_init+0x30/0x30 ret_from_fork_asm+0x11/0x20 and lots more like that. Thanks, Hugh
This series is rebased on top of mm-unstable + the patch: maple_tree: modified return type of mas_wr_store_entry()[1]. Andrew could you add that patch to mm-unstable before merging this series. v2[2] -> v3: - fix new line issues throughout the series - remove use of helper function in patch 13 v1[3] -> v2: - previous results were noisy, replaced with stress-ng-mmap config of mmtests - document function parameter 'entry'in 'mas_prealloc_calc per Intel Test Robot - explain why mas_reset() was dropeed in patch 4 - add locking and mas_destroy() to testcase in patch 4 - squash "set write store type" vs "set store type" into the mas_wr_store_entry() modification - only set ret to xa_err() in the err case in mas_store_gfp() - move MT_BUG_ON() inside lock in patch 7 - noinline mas_wr_spanning store to reduce stack usage in mas_wr_store_type() in patch 11 per Intel Test Robot - document function parameter gfp in mas_insert() per Intel test robot. - further simplify the local variables in patch 11 ================================ OVERVIEW ================================ This series implements two work items[4]: "aligning mas_store_gfp() with mas_preallocate()" and "enum for store type". mas_store_gfp() is modified to preallocate nodes. This simplies many of the write helper functions by allowing them to use mas_store_gfp() rather than open coding node allocation and error handling. The enum defines the following store types: enum store_type { wr_invalid, wr_new_root, wr_store_root, wr_exact_fit, wr_spanning_store, wr_split_store, wr_rebalance, wr_append, wr_node_store, wr_slot_store, wr_bnode }; In the current maple tree code, a walk down the tree is done in mas_preallocate() to determine the number of nodes needed for this write. After node allocation, mas_wr_store_entry() will perform another walk to determine which write helper function to use to complete the write. Rather than performing the second walk, we can store the type of write in the maple write state during node allocation and read this field to complete the write. ================================ RESULTS ================================= v6.10_mmap v6.10_mmap_store_type Duration User 9.80 8.69 Duration System 2295.94 2243.85 Duration Elapsed 1010.50 1009.44 ================================ TESTING ================================= Testing was done with the maple tree test suite. A new test case is also added to validate the order in which we test for and assign the store type. [1]: https://lore.kernel.org/linux-mm/20240614092428.29491-1-rgbi3307@gmail.com/T/ [2]: https://lore.kernel.org/linux-mm/20240607185257.963768-1-sidhartha.kumar@oracle.com/T/#ma3795b93f9d5c9c9286291e12f089bff45b87fed [3]: https://lore.kernel.org/linux-mm/20240604174145.563900-1-sidhartha.kumar@oracle.com/ [4]: https://lists.infradead.org/pipermail/maple-tree/2023-December/003098.html Sidhartha Kumar (16): maple_tree: introduce store_type enum maple_tree: introduce mas_wr_prealloc_setup() maple_tree: move up mas_wr_store_setup() and mas_wr_prealloc_setup() maple_tree: introduce mas_wr_store_type() maple_tree: remove mas_destroy() from mas_nomem() maple_tree: use mas_store_gfp() in mas_erase() maple_tree: use mas_store_gfp() in mtree_store_range() maple_tree: print store type in mas_dump() maple_tree: use store type in mas_wr_store_entry() maple_tree: convert mas_insert() to preallocate nodes maple_tree: simplify mas_commit_b_node() maple_tree: remove mas_wr_modify() maple_tree: have mas_store() allocate nodes if needed maple_tree: remove node allocations from various write helper functions maple_tree: remove repeated sanity checks from mas_wr_append() maple_tree: remove unneeded mas_wr_walk() in mas_store_prealloc() include/linux/maple_tree.h | 15 + lib/maple_tree.c | 580 ++++++++++++++++++------------- tools/testing/radix-tree/maple.c | 46 ++- 3 files changed, 396 insertions(+), 245 deletions(-)