diff options
author | Chris Mason <chris.mason@oracle.com> | 2010-06-11 11:17:59 -0700 |
---|---|---|
committer | Andy Grover <andy.grover@oracle.com> | 2010-09-08 18:15:28 -0700 |
commit | 6fa70da6081bbcf948801fd5ee0be4d222298a43 (patch) | |
tree | 519fef6c63abe15d80173ad25c7cd5aae673df55 /net/rds/xlist.h | |
parent | 0f4b1c7e89e699f588807a914ec6e6396c851a72 (diff) | |
download | op-kernel-dev-6fa70da6081bbcf948801fd5ee0be4d222298a43.zip op-kernel-dev-6fa70da6081bbcf948801fd5ee0be4d222298a43.tar.gz |
rds: recycle FMRs through lockless lists
FRM allocation and recycling is performance critical and fairly lock
intensive. The current code has a per connection lock that all
processes bang on and it becomes a major bottleneck on large systems.
This changes things to use a number of cmpxchg based lists instead,
allowing us to go through the whole FMR lifecycle without locking inside
RDS.
Zach Brown pointed out that our usage of cmpxchg for xlist removal is
racey if someone manages to remove and add back an FMR struct into the list
while another CPU can see the FMR's address at the head of the list.
The second CPU might assume the list hasn't changed when in fact any
number of operations might have happened in between the deletion and
reinsertion.
This commit maintains a per cpu count of CPUs that are currently
in xlist removal, and establishes a grace period to make sure that
nobody can see an entry we have just removed from the list.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
Diffstat (limited to 'net/rds/xlist.h')
-rw-r--r-- | net/rds/xlist.h | 110 |
1 files changed, 110 insertions, 0 deletions
diff --git a/net/rds/xlist.h b/net/rds/xlist.h new file mode 100644 index 0000000..8c21aca --- /dev/null +++ b/net/rds/xlist.h @@ -0,0 +1,110 @@ +#ifndef _LINUX_XLIST_H +#define _LINUX_XLIST_H + +#include <linux/stddef.h> +#include <linux/poison.h> +#include <linux/prefetch.h> +#include <asm/system.h> + +struct xlist_head { + struct xlist_head *next; +}; + +/* + * XLIST_PTR_TAIL can be used to prevent double insertion. See + * xlist_protect() + */ +#define XLIST_PTR_TAIL ((struct xlist_head *)0x1) + +static inline void xlist_add(struct xlist_head *new, struct xlist_head *tail, struct xlist_head *head) +{ + struct xlist_head *cur; + struct xlist_head *check; + + while (1) { + cur = head->next; + tail->next = cur; + check = cmpxchg(&head->next, cur, new); + if (check == cur) + break; + } +} + +/* + * To avoid duplicate insertion by two CPUs of the same xlist item + * you can call xlist_protect. It will stuff XLIST_PTR_TAIL + * into the entry->next pointer with xchg, and only return 1 + * if there was a NULL there before. + * + * if xlist_protect returns zero, someone else is busy working + * on this entry. Getting a NULL into the entry in a race + * free manner is the caller's job. + */ +static inline int xlist_protect(struct xlist_head *entry) +{ + struct xlist_head *val; + + val = xchg(&entry->next, XLIST_PTR_TAIL); + if (val == NULL) + return 1; + return 0; +} + +static inline struct xlist_head *xlist_del_head(struct xlist_head *head) +{ + struct xlist_head *cur; + struct xlist_head *check; + struct xlist_head *next; + + while (1) { + cur = head->next; + if (!cur) + goto out; + + if (cur == XLIST_PTR_TAIL) { + cur = NULL; + goto out; + } + + next = cur->next; + check = cmpxchg(&head->next, cur, next); + if (check == cur) + goto out; + } +out: + return cur; +} + +static inline struct xlist_head *xlist_del_head_fast(struct xlist_head *head) +{ + struct xlist_head *cur; + + cur = head->next; + if (!cur || cur == XLIST_PTR_TAIL) + return NULL; + + head->next = cur->next; + return cur; +} + +static inline void xlist_splice(struct xlist_head *list, + struct xlist_head *head) +{ + struct xlist_head *cur; + + WARN_ON(head->next); + cur = xchg(&list->next, NULL); + head->next = cur; +} + +static inline void INIT_XLIST_HEAD(struct xlist_head *list) +{ + list->next = NULL; +} + +static inline int xlist_empty(struct xlist_head *head) +{ + return head->next == NULL || head->next == XLIST_PTR_TAIL; +} + +#endif |