On a large SMP systems with hundreds of CPUs, the current thresholds for allocating and freeing debug objects (256 and 1024 respectively) may not work well. This can cause a lot of needless calls to kmem_aloc() and kmem_free() on those systems. To alleviate this thrashing problem, the object freeing threshold is now increased to "1024 + # of CPUs * 32". Whereas the object allocation threshold is increased to "256 + # of CPUs * 4". That should make the debug objects subsystem scale better with the number of CPUs available in the system.
Showing the changes to lib/debugobjects.c (1 of 1 file). Show all
Unified · Split
| @@ -52,7 +52,10 @@ static int debug_objects_fixups __read_mostly; · 51 unmodified lines | |||
| 52 | static int debug_objects_warnings __read_mostly; | 52 | static int debug_objects_warnings __read_mostly; |
| 53 | static int debug_objects_enabled __read_mostly | 53 | static int debug_objects_enabled __read_mostly |
| 54 | = CONFIG_DEBUG_OBJECTS_ENABLE_DEFAULT; | 54 | = CONFIG_DEBUG_OBJECTS_ENABLE_DEFAULT; |
| 55 | 55 | static int debug_objects_pool_size __read_mostly | |
| 56 | = ODEBUG_POOL_SIZE; | ||
| 57 | static int debug_objects_pool_min_level __read_mostly | ||
| 58 | = ODEBUG_POOL_MIN_LEVEL; | ||
| 56 | static struct debug_obj_descr *descr_test __read_mostly; | 59 | static struct debug_obj_descr *descr_test __read_mostly; |
| 57 | 60 | ||
| 58 | /* | 61 | /* |
| @@ -94,13 +97,13 @@ static void fill_pool(void) · 35 unmodified lines | |||
| 94 | struct debug_obj *new; | 97 | struct debug_obj *new; |
| 95 | unsigned long flags; | 98 | unsigned long flags; |
| 96 | 99 | ||
| 97 | if (likely(obj_pool_free >= ODEBUG_POOL_MIN_LEVEL)) | 100 | if (likely(obj_pool_free >= debug_objects_pool_min_level)) |
| 98 | return; | 101 | return; |
| 99 | 102 | ||
| 100 | if (unlikely(!obj_cache)) | 103 | if (unlikely(!obj_cache)) |
| 101 | return; | 104 | return; |
| 102 | 105 | ||
| 103 | while (obj_pool_free < ODEBUG_POOL_MIN_LEVEL) { | 106 | while (obj_pool_free < debug_objects_pool_min_level) { |
| 104 | 107 | ||
| 105 | new = kmem_cache_zalloc(obj_cache, gfp); | 108 | new = kmem_cache_zalloc(obj_cache, gfp); |
| 106 | if (!new) | 109 | if (!new) |
| @@ -176,7 +179,7 @@ static void free_obj_work(struct work_struct *work) · 69 unmodified lines | |||
| 176 | unsigned long flags; | 179 | unsigned long flags; |
| 177 | 180 | ||
| 178 | raw_spin_lock_irqsave(&pool_lock, flags); | 181 | raw_spin_lock_irqsave(&pool_lock, flags); |
| 179 | while (obj_pool_free > ODEBUG_POOL_SIZE) { | 182 | while (obj_pool_free > debug_objects_pool_size) { |
| 180 | obj = hlist_entry(obj_pool.first, typeof(*obj), node); | 183 | obj = hlist_entry(obj_pool.first, typeof(*obj), node); |
| 181 | hlist_del(&obj->node); | 184 | hlist_del(&obj->node); |
| 182 | obj_pool_free--; | 185 | obj_pool_free--; |
| @@ -206,7 +209,7 @@ static void free_object(struct debug_obj *obj) · 23 unmodified lines | |||
| 206 | * schedule work when the pool is filled and the cache is | 209 | * schedule work when the pool is filled and the cache is |
| 207 | * initialized: | 210 | * initialized: |
| 208 | */ | 211 | */ |
| 209 | if (obj_pool_free > ODEBUG_POOL_SIZE && obj_cache) | 212 | if (obj_pool_free > debug_objects_pool_size && obj_cache) |
| 210 | sched = 1; | 213 | sched = 1; |
| 211 | hlist_add_head(&obj->node, &obj_pool); | 214 | hlist_add_head(&obj->node, &obj_pool); |
| 212 | obj_pool_free++; | 215 | obj_pool_free++; |
| @@ -1126,4 +1129,11 @@ void __init debug_objects_mem_init(void) · 913 unmodified lines | |||
| 1126 | pr_warn("out of memory.\n"); | 1129 | pr_warn("out of memory.\n"); |
| 1127 | } else | 1130 | } else |
| 1128 | debug_objects_selftest(); | 1131 | debug_objects_selftest(); |
| 1132 | |||
| 1133 | /* | ||
| 1134 | * Increase the thresholds for allocating and freeing objects | ||
| 1135 | * according to the number of possible CPUs available in the system. | ||
| 1136 | */ | ||
| 1137 | debug_objects_pool_size += num_possible_cpus() * 32; | ||
| 1138 | debug_objects_pool_min_level += num_possible_cpus() * 4; | ||
| 1129 | } | 1139 | } |