github/ocf
4
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Switch from global metadata locks to hash-bucket locks in engines

Browse Source 
Signed-off-by: Adam Rutkowski <adam.j.rutkowski@intel.com>
This commit is contained in:
Adam Rutkowski
2019-07-31 16:31:28 -04:00
parent b39bcf86d4
commit 3a70d68d38
12 changed files with 124 additions and 129 deletions
Download Patch File Download Diff File Expand all files Collapse all files

52
src/engine/engine_wt.c
View File

@@ -97,19 +97,17 @@ static inline void _ocf_write_wt_submit(struct ocf_request *req)
static void _ocf_write_wt_update_bits(struct ocf_request *req)
{
struct ocf_cache *cache = req->cache;
if (ocf_engine_is_miss(req)) {
OCF_METADATA_LOCK_RD();
ocf_req_hash_lock_rd(req);
/* Update valid status bits */
ocf_set_valid_map_info(req);
OCF_METADATA_UNLOCK_RD();
ocf_req_hash_unlock_rd(req);
}
if (req->info.dirty_any) {
OCF_METADATA_LOCK_WR();
ocf_req_hash_lock_wr(req);
/* Writes goes to SDD and HDD, need to update status bits from
* dirty to clean
@@ -117,20 +115,20 @@ static void _ocf_write_wt_update_bits(struct ocf_request *req)
ocf_set_clean_map_info(req);
OCF_METADATA_UNLOCK_WR();
ocf_req_hash_unlock_wr(req);
}
if (req->info.re_part) {
OCF_DEBUG_RQ(req, "Re-Part");
OCF_METADATA_LOCK_WR();
ocf_req_hash_lock_wr(req);
/* Probably some cache lines are assigned into wrong
* partition. Need to move it to new one
*/
ocf_part_move(req);
OCF_METADATA_UNLOCK_WR();
ocf_req_hash_unlock_wr(req);
}
}
@@ -164,8 +162,8 @@ int ocf_write_wt(struct ocf_request *req)
{
bool mapped;
int lock = OCF_LOCK_NOT_ACQUIRED;
struct ocf_cache *cache = req->cache;
bool promote = true;
struct ocf_metadata_lock *metadata_lock = &req->cache->metadata.lock;
ocf_io_start(&req->ioi.io);
@@ -175,7 +173,8 @@ int ocf_write_wt(struct ocf_request *req)
/* Set resume io_if */
req->io_if = &_io_if_wt_resume;
OCF_METADATA_LOCK_RD(); /*- Metadata READ access, No eviction --------*/
ocf_req_hash(req);
ocf_req_hash_lock_rd(req); /*- Metadata READ access, No eviction --------*/
/* Travers to check if request is mapped fully */
ocf_engine_traverse(req);
@@ -184,31 +183,32 @@ int ocf_write_wt(struct ocf_request *req)
if (mapped) {
/* All cache line are mapped, lock request for WRITE access */
lock = ocf_req_async_lock_wr(req, ocf_engine_on_resume);
}
if (!mapped) {
} else {
promote = ocf_promotion_req_should_promote(
cache->promotion_policy, req);
req->cache->promotion_policy, req);
}
OCF_METADATA_UNLOCK_RD(); /*- END Metadata READ access----------------*/
if (mapped || !promote) {
ocf_req_hash_unlock_rd(req);
} else {
/*- Metadata RD access ---------------------------------------*/
ocf_req_hash_lock_upgrade(req);
ocf_engine_map(req);
ocf_req_hash_unlock_wr(req);
if (!mapped && promote) {
OCF_METADATA_LOCK_WR(); /*- Metadata WR access, eviction -----*/
/* Now there is exclusive access for metadata. May traverse once
* again. If there are misses need to call eviction. This
* process is called 'mapping'.
*/
if (ocf_engine_evict(req) == LOOKUP_MAPPED)
ocf_engine_map(req);
if (req->info.mapping_error) {
/* Still not mapped - evict cachelines under global
* metadata write lock */
ocf_metadata_start_exclusive_access(metadata_lock);
if (ocf_engine_evict(req) == LOOKUP_MAPPED)
ocf_engine_map(req);
ocf_metadata_end_exclusive_access(metadata_lock);
}
if (!req->info.mapping_error) {
/* Lock request for WRITE access */
lock = ocf_req_async_lock_wr(req, ocf_engine_on_resume);
}
OCF_METADATA_UNLOCK_WR(); /*- END Metadata WR access ---------*/
}
if (promote && !req->info.mapping_error) {