zbud是一个专门为存储压缩page而设计的内存分配器.用于将两个压缩的page存到一个单独的page中.zbud page 被分成chunks,每一个chunk的size是64byte,zbud目前只有被zswap使用.
我们来实际看看zbud的源码.
源码路径如下:source/mm/zbud.c
每一个被zbud管理的memory用zbud_pool来表示,每个zbud_pool有一个zbud_header。
126 #ifdef CONFIG_ZPOOL
216 MODULE_ALIAS("zpool-zbud");
217 #endif /* CONFIG_ZPOOL */
被CONFIG_ZPOOL 包起来的code是提供给zpool使用的接口,基本都是调用zbud的函数.zbud的使用一般是先调用zbud_zpool_create来新建一个pool,然后zbud_zpool_malloc 从这个pool中分配memory,如果想得到物理地址就调用zbud_zpool_map
228 /* Converts an allocation size in bytes to size in zbud chunks */
229 static int size_to_chunks(size_t size)
230 {
231 return (size + CHUNK_SIZE - 1) >> CHUNK_SHIFT;
232 }
由于zbud是以chunk 为单位管理memory的,因此将申请到size转成chunk index.
237 /* Initializes the zbud header of a newly allocated zbud page */
238 static struct zbud_header *init_zbud_page(struct page *page)
239 {
240 struct zbud_header *zhdr = page_address(page);
241 zhdr->first_chunks = 0;
242 zhdr->last_chunks = 0;
243 INIT_LIST_HEAD(&zhdr->buddy);
244 INIT_LIST_HEAD(&zhdr->lru);
245 zhdr->under_reclaim = 0;
246 return zhdr;
247 }
每个zbud都有一个head,init_zbud_page用于初始化zbud header
305 struct zbud_pool *zbud_create_pool(gfp_t gfp, const struct zbud_ops *ops)
306 {
307 struct zbud_pool *pool;
308 int i;
309
310 pool = kzalloc(sizeof(struct zbud_pool), gfp);
311 if (!pool)
312 return NULL;
313 spin_lock_init(&pool->lock);
314 for_each_unbuddied_list(i, 0)
315 INIT_LIST_HEAD(&pool->unbuddied[i]);
316 INIT_LIST_HEAD(&pool->buddied);
317 INIT_LIST_HEAD(&pool->lru);
318 pool->pages_nr = 0;
319 pool->ops = ops;
320 return pool;
321 }
新建一个zbud_pool通过kzalloc 申请内容为0的memory,然后做zbud_pool初始化,注意这个时候还没有申请要管理的memory.
int zbud_alloc(struct zbud_pool *pool, size_t size, gfp_t gfp,
354 unsigned long *handle)
355 {
356 int chunks, i, freechunks;
357 struct zbud_header *zhdr = NULL;
358 enum buddy bud;
359 struct page *page;
360
361 if (!size || (gfp & __GFP_HIGHMEM))
362 return -EINVAL;
363 if (size > PAGE_SIZE - ZHDR_SIZE_ALIGNED - CHUNK_SIZE)
364 return -ENOSPC;
365 chunks = size_to_chunks(size);
366 spin_lock(&pool->lock);
367
368 /* First, try to find an unbuddied zbud page. */
369 zhdr = NULL;
370 for_each_unbuddied_list(i, chunks) {
371 if (!list_empty(&pool->unbuddied[i])) {
372 zhdr = list_first_entry(&pool->unbuddied[i],
373 struct zbud_header, buddy);
374 list_del(&zhdr->buddy);
375 if (zhdr->first_chunks == 0)
376 bud = FIRST;
377 else
378 bud = LAST;
379 goto found;
380 }
381 }
382
383 /* Couldn't find unbuddied zbud page, create new one */
384 spin_unlock(&pool->lock);
385 page = alloc_page(gfp);
386 if (!page)
387 return -ENOMEM;
388 spin_lock(&pool->lock);
389 pool->pages_nr++;
390 zhdr = init_zbud_page(page);
391 bud = FIRST;
392
393 found:
394 if (bud == FIRST)
395 zhdr->first_chunks = chunks;
396 else
397 zhdr->last_chunks = chunks;
398
399 if (zhdr->first_chunks == 0 || zhdr->last_chunks == 0) {
400 /* Add to unbuddied list */
401 freechunks = num_free_chunks(zhdr);
402 list_add(&zhdr->buddy, &pool->unbuddied[freechunks]);
403 } else {
404 /* Add to buddied list */
405 list_add(&zhdr->buddy, &pool->buddied);
406 }
407
408 /* Add/move zbud page to beginning of LRU */
409 if (!list_empty(&zhdr->lru))
410 list_del(&zhdr->lru);
411 list_add(&zhdr->lru, &pool->lru);
412
413 *handle = encode_handle(zhdr, bud);
414 spin_unlock(&pool->lock);
415
416 return 0;
417 }
zswap 会调用这个函数从zbud要memory.
361不能申请highmemory,
363申请memory的size不能超过一个page.
365行将申请的size 转成chunk index.
370行判断pool->unbuddied 是否有memory可以用,由于是第一次申请,这个条件不成立。
385行通过alloc_page 向buddy system申请一个page的memory。
389行pages_nr++ 记录用于zbud管理的page 个数,第一次等于1
390 init_zbud_page 初始化呢header.
399行的zhdr->last_chunks == 0 成立,将free的chunk加到pool->unbuddied 列表中.
409条件不成立
411行将新申请的pool加到lru 中
413行调用encode_handle 返回在page中的offset.
259 static unsigned long encode_handle(struct zbud_header *zhdr, enum buddy bud)
260 {
261 unsigned long handle;
262
263 /*
264 * For now, the encoded handle is actually just the pointer to the data
265 * but this might not always be the case. A little information hiding.
266 * Add CHUNK_SIZE to the handle if it is the first allocation to jump
267 * over the zbud header in the first chunk.
268 */
269 handle = (unsigned long)zhdr;
270 if (bud == FIRST)
271 /* skip over zbud header */
272 handle += ZHDR_SIZE_ALIGNED;
273 else /* bud == LAST */
274 handle += PAGE_SIZE - (zhdr->last_chunks << CHUNK_SHIFT);
275 return handle;
276 }
关于zbud我有几点不明白FIRST和last 到底表示啥含义。为什么free chunks会是NCHUNKS - zhdr->first_chunks - zhdr->last_chunks
知道的大神指点下啊
我们来实际看看zbud的源码.
源码路径如下:source/mm/zbud.c
每一个被zbud管理的memory用zbud_pool来表示,每个zbud_pool有一个zbud_header。
126 #ifdef CONFIG_ZPOOL
216 MODULE_ALIAS("zpool-zbud");
217 #endif /* CONFIG_ZPOOL */
被CONFIG_ZPOOL 包起来的code是提供给zpool使用的接口,基本都是调用zbud的函数.zbud的使用一般是先调用zbud_zpool_create来新建一个pool,然后zbud_zpool_malloc 从这个pool中分配memory,如果想得到物理地址就调用zbud_zpool_map
228 /* Converts an allocation size in bytes to size in zbud chunks */
229 static int size_to_chunks(size_t size)
230 {
231 return (size + CHUNK_SIZE - 1) >> CHUNK_SHIFT;
232 }
由于zbud是以chunk 为单位管理memory的,因此将申请到size转成chunk index.
237 /* Initializes the zbud header of a newly allocated zbud page */
238 static struct zbud_header *init_zbud_page(struct page *page)
239 {
240 struct zbud_header *zhdr = page_address(page);
241 zhdr->first_chunks = 0;
242 zhdr->last_chunks = 0;
243 INIT_LIST_HEAD(&zhdr->buddy);
244 INIT_LIST_HEAD(&zhdr->lru);
245 zhdr->under_reclaim = 0;
246 return zhdr;
247 }
每个zbud都有一个head,init_zbud_page用于初始化zbud header
305 struct zbud_pool *zbud_create_pool(gfp_t gfp, const struct zbud_ops *ops)
306 {
307 struct zbud_pool *pool;
308 int i;
309
310 pool = kzalloc(sizeof(struct zbud_pool), gfp);
311 if (!pool)
312 return NULL;
313 spin_lock_init(&pool->lock);
314 for_each_unbuddied_list(i, 0)
315 INIT_LIST_HEAD(&pool->unbuddied[i]);
316 INIT_LIST_HEAD(&pool->buddied);
317 INIT_LIST_HEAD(&pool->lru);
318 pool->pages_nr = 0;
319 pool->ops = ops;
320 return pool;
321 }
新建一个zbud_pool通过kzalloc 申请内容为0的memory,然后做zbud_pool初始化,注意这个时候还没有申请要管理的memory.
int zbud_alloc(struct zbud_pool *pool, size_t size, gfp_t gfp,
354 unsigned long *handle)
355 {
356 int chunks, i, freechunks;
357 struct zbud_header *zhdr = NULL;
358 enum buddy bud;
359 struct page *page;
360
361 if (!size || (gfp & __GFP_HIGHMEM))
362 return -EINVAL;
363 if (size > PAGE_SIZE - ZHDR_SIZE_ALIGNED - CHUNK_SIZE)
364 return -ENOSPC;
365 chunks = size_to_chunks(size);
366 spin_lock(&pool->lock);
367
368 /* First, try to find an unbuddied zbud page. */
369 zhdr = NULL;
370 for_each_unbuddied_list(i, chunks) {
371 if (!list_empty(&pool->unbuddied[i])) {
372 zhdr = list_first_entry(&pool->unbuddied[i],
373 struct zbud_header, buddy);
374 list_del(&zhdr->buddy);
375 if (zhdr->first_chunks == 0)
376 bud = FIRST;
377 else
378 bud = LAST;
379 goto found;
380 }
381 }
382
383 /* Couldn't find unbuddied zbud page, create new one */
384 spin_unlock(&pool->lock);
385 page = alloc_page(gfp);
386 if (!page)
387 return -ENOMEM;
388 spin_lock(&pool->lock);
389 pool->pages_nr++;
390 zhdr = init_zbud_page(page);
391 bud = FIRST;
392
393 found:
394 if (bud == FIRST)
395 zhdr->first_chunks = chunks;
396 else
397 zhdr->last_chunks = chunks;
398
399 if (zhdr->first_chunks == 0 || zhdr->last_chunks == 0) {
400 /* Add to unbuddied list */
401 freechunks = num_free_chunks(zhdr);
402 list_add(&zhdr->buddy, &pool->unbuddied[freechunks]);
403 } else {
404 /* Add to buddied list */
405 list_add(&zhdr->buddy, &pool->buddied);
406 }
407
408 /* Add/move zbud page to beginning of LRU */
409 if (!list_empty(&zhdr->lru))
410 list_del(&zhdr->lru);
411 list_add(&zhdr->lru, &pool->lru);
412
413 *handle = encode_handle(zhdr, bud);
414 spin_unlock(&pool->lock);
415
416 return 0;
417 }
zswap 会调用这个函数从zbud要memory.
361不能申请highmemory,
363申请memory的size不能超过一个page.
365行将申请的size 转成chunk index.
370行判断pool->unbuddied 是否有memory可以用,由于是第一次申请,这个条件不成立。
385行通过alloc_page 向buddy system申请一个page的memory。
389行pages_nr++ 记录用于zbud管理的page 个数,第一次等于1
390 init_zbud_page 初始化呢header.
399行的zhdr->last_chunks == 0 成立,将free的chunk加到pool->unbuddied 列表中.
409条件不成立
411行将新申请的pool加到lru 中
413行调用encode_handle 返回在page中的offset.
259 static unsigned long encode_handle(struct zbud_header *zhdr, enum buddy bud)
260 {
261 unsigned long handle;
262
263 /*
264 * For now, the encoded handle is actually just the pointer to the data
265 * but this might not always be the case. A little information hiding.
266 * Add CHUNK_SIZE to the handle if it is the first allocation to jump
267 * over the zbud header in the first chunk.
268 */
269 handle = (unsigned long)zhdr;
270 if (bud == FIRST)
271 /* skip over zbud header */
272 handle += ZHDR_SIZE_ALIGNED;
273 else /* bud == LAST */
274 handle += PAGE_SIZE - (zhdr->last_chunks << CHUNK_SHIFT);
275 return handle;
276 }
关于zbud我有几点不明白FIRST和last 到底表示啥含义。为什么free chunks会是NCHUNKS - zhdr->first_chunks - zhdr->last_chunks
知道的大神指点下啊