On Tue, Aug 5, 2014 at 7:55 PM, Eric Blake <ebl...@redhat.com> wrote: > On 08/04/2014 09:33 PM, Ming Lei wrote: >> This patch introduces object allocation pool for speeding up >> object allocation in fast path. >> >> Signed-off-by: Ming Lei <ming....@canonical.com> >> --- >> include/qemu/obj_pool.h | 64 >> +++++++++++++++++++++++++++++++++++++++++++++++ >> 1 file changed, 64 insertions(+) >> create mode 100644 include/qemu/obj_pool.h >> >> diff --git a/include/qemu/obj_pool.h b/include/qemu/obj_pool.h >> new file mode 100644 >> index 0000000..94b5f49 >> --- /dev/null >> +++ b/include/qemu/obj_pool.h >> @@ -0,0 +1,64 @@ >> +#ifndef QEMU_OBJ_POOL_HEAD >> +#define QEMU_OBJ_POOL_HEAD > > Missing copyright boilerplate. According to LICENSE, that makes this > file GPLv2+, but I'd much rather you make it explicit. > >> + >> +typedef struct { >> + unsigned int size; >> + unsigned int cnt; > > size_t feels better for sizes. int may be okay in this case, but > definitely consider if size_t is appropriate.
Sounds good. > >> + >> + void **free_obj; >> + int free_idx; >> + >> + char *objs; >> +} ObjPool; >> + >> +static inline void obj_pool_init(ObjPool *op, void *objs_buf, void >> **free_objs, >> + unsigned int obj_size, unsigned cnt) >> +{ >> + int i; >> + >> + op->objs = (char *)objs_buf; > > Why the cast? This is C, not C++. Right, the cast isn't needed. > >> + op->free_obj = free_objs; >> + op->size = obj_size; >> + op->cnt = cnt; >> + >> + for (i = 0; i < op->cnt; i++) { >> + op->free_obj[i] = (void *)&op->objs[i * op->size]; > > Again, why the cast? Right too. > > >> +static inline bool obj_pool_has_obj(ObjPool *op, void *obj) >> +{ >> + return op && (unsigned long)obj >= (unsigned long)&op->objs[0] && >> + (unsigned long)obj <= >> + (unsigned long)&op->objs[(op->cnt - 1) * op->size]; > > uintptr_t, not unsigned long. You are asking for problems on 64-bit > mingw, where unsigned long is 32 bits but uintptr_t is 64 bits. Good point, it is the 1st time for me to know the mingw long magic. Thanks,