aboutsummaryrefslogtreecommitdiffstats
path: root/include/linux/module.h
diff options
context:
space:
mode:
authorEric Dumazet <dada1@cosmosbay.com>2009-02-02 22:01:36 -0500
committerLinus Torvalds <torvalds@linux-foundation.org>2009-02-02 22:17:55 -0500
commit720eba31f47aeade8ec130ca7f4353223c49170f (patch)
tree4aef103b59a0fb7b9406fd3e4bb09bdf7e3ad2f7 /include/linux/module.h
parent27421e211a39784694b597dbf35848b88363c248 (diff)
modules: Use a better scheme for refcounting
Current refcounting for modules (done if CONFIG_MODULE_UNLOAD=y) is using a lot of memory. Each 'struct module' contains an [NR_CPUS] array of full cache lines. This patch uses existing infrastructure (percpu_modalloc() & percpu_modfree()) to allocate percpu space for the refcount storage. Instead of wasting NR_CPUS*128 bytes (on i386), we now use nr_cpu_ids*sizeof(local_t) bytes. On a typical distro, where NR_CPUS=8, shiping 2000 modules, we reduce size of module files by about 2 Mbytes. (1Kb per module) Instead of having all refcounters in the same memory node - with TLB misses because of vmalloc() - this new implementation permits to have better NUMA properties, since each CPU will use storage on its preferred node, thanks to percpu storage. Signed-off-by: Eric Dumazet <dada1@cosmosbay.com> Signed-off-by: Rusty Russell <rusty@rustcorp.com.au> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Diffstat (limited to 'include/linux/module.h')
-rw-r--r--include/linux/module.h25
1 files changed, 16 insertions, 9 deletions
diff --git a/include/linux/module.h b/include/linux/module.h
index 4f7ea12463d3..f3b8329eb5b8 100644
--- a/include/linux/module.h
+++ b/include/linux/module.h
@@ -219,11 +219,6 @@ void *__symbol_get_gpl(const char *symbol);
219 219
220#endif 220#endif
221 221
222struct module_ref
223{
224 local_t count;
225} ____cacheline_aligned;
226
227enum module_state 222enum module_state
228{ 223{
229 MODULE_STATE_LIVE, 224 MODULE_STATE_LIVE,
@@ -344,8 +339,11 @@ struct module
344 /* Destruction function. */ 339 /* Destruction function. */
345 void (*exit)(void); 340 void (*exit)(void);
346 341
347 /* Reference counts */ 342#ifdef CONFIG_SMP
348 struct module_ref ref[NR_CPUS]; 343 char *refptr;
344#else
345 local_t ref;
346#endif
349#endif 347#endif
350}; 348};
351#ifndef MODULE_ARCH_INIT 349#ifndef MODULE_ARCH_INIT
@@ -395,13 +393,22 @@ void __symbol_put(const char *symbol);
395#define symbol_put(x) __symbol_put(MODULE_SYMBOL_PREFIX #x) 393#define symbol_put(x) __symbol_put(MODULE_SYMBOL_PREFIX #x)
396void symbol_put_addr(void *addr); 394void symbol_put_addr(void *addr);
397 395
396static inline local_t *__module_ref_addr(struct module *mod, int cpu)
397{
398#ifdef CONFIG_SMP
399 return (local_t *) (mod->refptr + per_cpu_offset(cpu));
400#else
401 return &mod->ref;
402#endif
403}
404
398/* Sometimes we know we already have a refcount, and it's easier not 405/* Sometimes we know we already have a refcount, and it's easier not
399 to handle the error case (which only happens with rmmod --wait). */ 406 to handle the error case (which only happens with rmmod --wait). */
400static inline void __module_get(struct module *module) 407static inline void __module_get(struct module *module)
401{ 408{
402 if (module) { 409 if (module) {
403 BUG_ON(module_refcount(module) == 0); 410 BUG_ON(module_refcount(module) == 0);
404 local_inc(&module->ref[get_cpu()].count); 411 local_inc(__module_ref_addr(module, get_cpu()));
405 put_cpu(); 412 put_cpu();
406 } 413 }
407} 414}
@@ -413,7 +420,7 @@ static inline int try_module_get(struct module *module)
413 if (module) { 420 if (module) {
414 unsigned int cpu = get_cpu(); 421 unsigned int cpu = get_cpu();
415 if (likely(module_is_live(module))) 422 if (likely(module_is_live(module)))
416 local_inc(&module->ref[cpu].count); 423 local_inc(__module_ref_addr(module, cpu));
417 else 424 else
418 ret = 0; 425 ret = 0;
419 put_cpu(); 426 put_cpu();