create rrpc slabs during rrpc module init,
thus eliminate the lock contention and slab
status check in rrpc_core_init. And destroy
them when rrpc exit.
Signed-off-by: Wenwei Tao <[email protected]>
---
drivers/lightnvm/rrpc.c | 54 ++++++++++++++++++++++++++++++-------------------
1 file changed, 33 insertions(+), 21 deletions(-)
diff --git a/drivers/lightnvm/rrpc.c b/drivers/lightnvm/rrpc.c
index 75e59c3..a36f39a 100644
--- a/drivers/lightnvm/rrpc.c
+++ b/drivers/lightnvm/rrpc.c
@@ -17,7 +17,6 @@
#include "rrpc.h"
static struct kmem_cache *rrpc_gcb_cache, *rrpc_rq_cache;
-static DECLARE_RWSEM(rrpc_lock);
static int rrpc_submit_io(struct rrpc *rrpc, struct bio *bio,
struct nvm_rq *rqd, unsigned long flags);
@@ -1019,26 +1018,6 @@ static int rrpc_map_init(struct rrpc *rrpc)
static int rrpc_core_init(struct rrpc *rrpc)
{
- down_write(&rrpc_lock);
- if (!rrpc_gcb_cache) {
- rrpc_gcb_cache = kmem_cache_create("rrpc_gcb",
- sizeof(struct rrpc_block_gc), 0, 0, NULL);
- if (!rrpc_gcb_cache) {
- up_write(&rrpc_lock);
- return -ENOMEM;
- }
-
- rrpc_rq_cache = kmem_cache_create("rrpc_rq",
- sizeof(struct nvm_rq) + sizeof(struct rrpc_rq),
- 0, 0, NULL);
- if (!rrpc_rq_cache) {
- kmem_cache_destroy(rrpc_gcb_cache);
- up_write(&rrpc_lock);
- return -ENOMEM;
- }
- }
- up_write(&rrpc_lock);
-
rrpc->page_pool = mempool_create_page_pool(PAGE_POOL_SIZE, 0);
if (!rrpc->page_pool)
return -ENOMEM;
@@ -1338,14 +1317,47 @@ static struct nvm_tgt_type tt_rrpc = {
.exit = rrpc_exit,
};
+static int __init rrpc_slab_init(void)
+{
+ rrpc_gcb_cache = kmem_cache_create("rrpc_gcb",
+ sizeof(struct rrpc_block_gc), 0, 0, NULL);
+ if (!rrpc_gcb_cache)
+ goto out;
+
+ rrpc_rq_cache = kmem_cache_create("rrpc_rq",
+ sizeof(struct nvm_rq) + sizeof(struct rrpc_rq),
+ 0, 0, NULL);
+ if (!rrpc_rq_cache)
+ goto out_free;
+
+ return 0;
+
+out_free:
+ kmem_cache_destroy(rrpc_gcb_cache);
+out:
+ return -ENOMEM;
+}
+
+static inline void rrpc_slab_free(void)
+{
+ kmem_cache_destroy(rrpc_gcb_cache);
+ kmem_cache_destroy(rrpc_rq_cache);
+}
+
static int __init rrpc_module_init(void)
{
+ int ret;
+
+ ret = rrpc_slab_init();
+ if (ret)
+ return ret;
return nvm_register_target(&tt_rrpc);
}
static void rrpc_module_exit(void)
{
nvm_unregister_target(&tt_rrpc);
+ rrpc_slab_free();
}
module_init(rrpc_module_init);
--
1.8.3.1
Hi Matias
In my understanding kmem_cache_create only allocate and setup some
basic structures, the actual slab memory is allocated from buddy
system when we use these slabs.
Do you think the memory consumed by these structures is a issue
compared to the lock contention and slab status check in
rrpc_core_init，or you have any other concerns ?
2015-12-08 3:45 GMT+08:00 Matias Bjørling <[email protected]>:
> On Mon, Dec 7, 2015 at 1:16 PM, Wenwei Tao <[email protected]> wrote:
>>
>> create rrpc slabs during rrpc module init,
>> thus eliminate the lock contention and slab
>> status check in rrpc_core_init. And destroy
>> them when rrpc exit.
>>
>> Signed-off-by: Wenwei Tao <[email protected]>
>> ---
>> drivers/lightnvm/rrpc.c | 54
>> ++++++++++++++++++++++++++++++-------------------
>> 1 file changed, 33 insertions(+), 21 deletions(-)
>>
>> diff --git a/drivers/lightnvm/rrpc.c b/drivers/lightnvm/rrpc.c
>> index 75e59c3..a36f39a 100644
>> --- a/drivers/lightnvm/rrpc.c
>> +++ b/drivers/lightnvm/rrpc.c
>> @@ -17,7 +17,6 @@
>> #include "rrpc.h"
>>
>> static struct kmem_cache *rrpc_gcb_cache, *rrpc_rq_cache;
>> -static DECLARE_RWSEM(rrpc_lock);
>>
>> static int rrpc_submit_io(struct rrpc *rrpc, struct bio *bio,
>> struct nvm_rq *rqd, unsigned long flags);
>> @@ -1019,26 +1018,6 @@ static int rrpc_map_init(struct rrpc *rrpc)
>>
>> static int rrpc_core_init(struct rrpc *rrpc)
>> {
>> - down_write(&rrpc_lock);
>> - if (!rrpc_gcb_cache) {
>> - rrpc_gcb_cache = kmem_cache_create("rrpc_gcb",
>> - sizeof(struct rrpc_block_gc), 0, 0, NULL);
>> - if (!rrpc_gcb_cache) {
>> - up_write(&rrpc_lock);
>> - return -ENOMEM;
>> - }
>> -
>> - rrpc_rq_cache = kmem_cache_create("rrpc_rq",
>> - sizeof(struct nvm_rq) + sizeof(struct
>> rrpc_rq),
>> - 0, 0, NULL);
>> - if (!rrpc_rq_cache) {
>> - kmem_cache_destroy(rrpc_gcb_cache);
>> - up_write(&rrpc_lock);
>> - return -ENOMEM;
>> - }
>> - }
>> - up_write(&rrpc_lock);
>> -
>> rrpc->page_pool = mempool_create_page_pool(PAGE_POOL_SIZE, 0);
>> if (!rrpc->page_pool)
>> return -ENOMEM;
>> @@ -1338,14 +1317,47 @@ static struct nvm_tgt_type tt_rrpc = {
>> .exit = rrpc_exit,
>> };
>>
>> +static int __init rrpc_slab_init(void)
>> +{
>> + rrpc_gcb_cache = kmem_cache_create("rrpc_gcb",
>> + sizeof(struct rrpc_block_gc), 0, 0, NULL);
>> + if (!rrpc_gcb_cache)
>> + goto out;
>> +
>> + rrpc_rq_cache = kmem_cache_create("rrpc_rq",
>> + sizeof(struct nvm_rq) + sizeof(struct rrpc_rq),
>> + 0, 0, NULL);
>> + if (!rrpc_rq_cache)
>> + goto out_free;
>> +
>> + return 0;
>> +
>> +out_free:
>> + kmem_cache_destroy(rrpc_gcb_cache);
>> +out:
>> + return -ENOMEM;
>> +}
>> +
>> +static inline void rrpc_slab_free(void)
>> +{
>> + kmem_cache_destroy(rrpc_gcb_cache);
>> + kmem_cache_destroy(rrpc_rq_cache);
>> +}
>> +
>> static int __init rrpc_module_init(void)
>> {
>> + int ret;
>> +
>> + ret = rrpc_slab_init();
>> + if (ret)
>> + return ret;
>> return nvm_register_target(&tt_rrpc);
>> }
>>
>> static void rrpc_module_exit(void)
>> {
>> nvm_unregister_target(&tt_rrpc);
>> + rrpc_slab_free();
>> }
>>
>> module_init(rrpc_module_init);
>
>
> Thanks Tao. I think the previous behavior is better. That way we don't
> consume any memory until the module is in use.
On 12/08/2015 12:36 PM, Wenwei Tao wrote:
> Hi Matias
> In my understanding kmem_cache_create only allocate and setup some
> basic structures, the actual slab memory is allocated from buddy
> system when we use these slabs.
> Do you think the memory consumed by these structures is a issue
> compared to the lock contention and slab status check in
> rrpc_core_init，or you have any other concerns ?
The rrpc_core_init is only run at registration time and isn't in a
critical path. So I rather not initialize something until we really it.
ps. it is custom not to top post :)