| Message ID | 20220718192844.1805158-15-yury.norov@gmail.com (mailing list archive) |
|---|---|
| State | New |
| Headers | show |
| Series | Introduce DEBUG_BITMAP config option and bitmap_check_params() | expand |
Hello, On Mon, Jul 18, 2022 at 12:28:42PM -0700, Yury Norov wrote: > Don't call bitmap_clear() to clear 0 bits. > > bitmap_clear() can handle 0-length requests properly, but it's not covered > with static optimizations, and falls to __bitmap_set(). So we are paying a > function call + prologue work cost just for nothing. > > Caught with CONFIG_DEBUG_BITMAP: > [ 45.571799] <TASK> > [ 45.571801] pcpu_alloc_area+0x194/0x340 > [ 45.571806] pcpu_alloc+0x2fb/0x8b0 > [ 45.571811] ? kmem_cache_alloc_trace+0x177/0x2a0 > [ 45.571815] __percpu_counter_init+0x22/0xa0 > [ 45.571819] fprop_local_init_percpu+0x14/0x30 > [ 45.571823] wb_get_create+0x15d/0x5f0 > [ 45.571828] cleanup_offline_cgwb+0x73/0x210 > [ 45.571831] cleanup_offline_cgwbs_workfn+0xcf/0x200 > [ 45.571835] process_one_work+0x1e5/0x3b0 > [ 45.571839] worker_thread+0x50/0x3a0 > [ 45.571843] ? rescuer_thread+0x390/0x390 > [ 45.571846] kthread+0xe8/0x110 > [ 45.571849] ? kthread_complete_and_exit+0x20/0x20 > [ 45.571853] ret_from_fork+0x22/0x30 > [ 45.571858] </TASK> > [ 45.571859] ---[ end trace 0000000000000000 ]--- > [ 45.571860] b1: ffffa8d5002e1000 > [ 45.571861] b2: 0 > [ 45.571861] b3: 0 > [ 45.571862] nbits: 44638 > [ 45.571863] start: 44638 > [ 45.571864] off: 0 > [ 45.571864] percpu: Bitmap: parameters check failed > [ 45.571865] percpu: include/linux/bitmap.h [538]: bitmap_clear > > Signed-off-by: Yury Norov <yury.norov@gmail.com> > --- > mm/percpu.c | 3 ++- > 1 file changed, 2 insertions(+), 1 deletion(-) > > diff --git a/mm/percpu.c b/mm/percpu.c > index 3633eeefaa0d..f720f7c36b91 100644 > --- a/mm/percpu.c > +++ b/mm/percpu.c > @@ -1239,7 +1239,8 @@ static int pcpu_alloc_area(struct pcpu_chunk *chunk, int alloc_bits, > > /* update boundary map */ > set_bit(bit_off, chunk->bound_map); > - bitmap_clear(chunk->bound_map, bit_off + 1, alloc_bits - 1); > + if (alloc_bits > 1) > + bitmap_clear(chunk->bound_map, bit_off + 1, alloc_bits - 1); > set_bit(bit_off + alloc_bits, chunk->bound_map); > > chunk->free_bytes -= alloc_bits * PCPU_MIN_ALLOC_SIZE; > -- > 2.34.1 > Acked-by: Dennis Zhou <dennis@kernel.org> Thanks, Dennis
diff --git a/mm/percpu.c b/mm/percpu.c index 3633eeefaa0d..f720f7c36b91 100644 --- a/mm/percpu.c +++ b/mm/percpu.c @@ -1239,7 +1239,8 @@ static int pcpu_alloc_area(struct pcpu_chunk *chunk, int alloc_bits, /* update boundary map */ set_bit(bit_off, chunk->bound_map); - bitmap_clear(chunk->bound_map, bit_off + 1, alloc_bits - 1); + if (alloc_bits > 1) + bitmap_clear(chunk->bound_map, bit_off + 1, alloc_bits - 1); set_bit(bit_off + alloc_bits, chunk->bound_map); chunk->free_bytes -= alloc_bits * PCPU_MIN_ALLOC_SIZE;
Don't call bitmap_clear() to clear 0 bits. bitmap_clear() can handle 0-length requests properly, but it's not covered with static optimizations, and falls to __bitmap_set(). So we are paying a function call + prologue work cost just for nothing. Caught with CONFIG_DEBUG_BITMAP: [ 45.571799] <TASK> [ 45.571801] pcpu_alloc_area+0x194/0x340 [ 45.571806] pcpu_alloc+0x2fb/0x8b0 [ 45.571811] ? kmem_cache_alloc_trace+0x177/0x2a0 [ 45.571815] __percpu_counter_init+0x22/0xa0 [ 45.571819] fprop_local_init_percpu+0x14/0x30 [ 45.571823] wb_get_create+0x15d/0x5f0 [ 45.571828] cleanup_offline_cgwb+0x73/0x210 [ 45.571831] cleanup_offline_cgwbs_workfn+0xcf/0x200 [ 45.571835] process_one_work+0x1e5/0x3b0 [ 45.571839] worker_thread+0x50/0x3a0 [ 45.571843] ? rescuer_thread+0x390/0x390 [ 45.571846] kthread+0xe8/0x110 [ 45.571849] ? kthread_complete_and_exit+0x20/0x20 [ 45.571853] ret_from_fork+0x22/0x30 [ 45.571858] </TASK> [ 45.571859] ---[ end trace 0000000000000000 ]--- [ 45.571860] b1: ffffa8d5002e1000 [ 45.571861] b2: 0 [ 45.571861] b3: 0 [ 45.571862] nbits: 44638 [ 45.571863] start: 44638 [ 45.571864] off: 0 [ 45.571864] percpu: Bitmap: parameters check failed [ 45.571865] percpu: include/linux/bitmap.h [538]: bitmap_clear Signed-off-by: Yury Norov <yury.norov@gmail.com> --- mm/percpu.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-)