Previously, we were putting an exclusive lock to prevent secondary
processes spinning up while we are sending our messages. However,
using exclusive locks had an effect of disallowing multiple
simultaenous unrelated messages/requests being sent, which was
not the intention behind locking.
Fix it to put a shared lock on the directory. That way, we still
prevent secondary process initializations while sending data over
IPC, but allow multiple unrelated transmissions to proceed.
Fixes: 89f1fe7e6d95 ("eal: lock IPC directory on init and send")
Cc: sta...@dpdk.org
Signed-off-by: Anatoly Burakov <anatoly.bura...@intel.com>
---
Notes:
This patch is needed for multiprocess hotplug support [1], to avoid cases
where multiple requests deadlock while in progress.
[1] http://patches.dpdk.org/project/dpdk/list/?series=252
lib/librte_eal/common/eal_common_proc.c | 6 +++---
1 file changed, 3 insertions(+), 3 deletions(-)
diff --git a/lib/librte_eal/common/eal_common_proc.c
b/lib/librte_eal/common/eal_common_proc.c
index 707d8ab30..f010ef59e 100644
--- a/lib/librte_eal/common/eal_common_proc.c
+++ b/lib/librte_eal/common/eal_common_proc.c
@@ -786,7 +786,7 @@ mp_send(struct rte_mp_msg *msg, const char *peer, int type)
dir_fd = dirfd(mp_dir);
/* lock the directory to prevent processes spinning up while we send */
- if (flock(dir_fd, LOCK_EX)) {
+ if (flock(dir_fd, LOCK_SH)) {
RTE_LOG(ERR, EAL, "Unable to lock directory %s\n",
mp_dir_path);
rte_errno = errno;
@@ -1020,7 +1020,7 @@ rte_mp_request_sync(struct rte_mp_msg *req, struct
rte_mp_reply *reply,
dir_fd = dirfd(mp_dir);
/* lock the directory to prevent processes spinning up while we send */
- if (flock(dir_fd, LOCK_EX)) {
+ if (flock(dir_fd, LOCK_SH)) {
RTE_LOG(ERR, EAL, "Unable to lock directory %s\n",
mp_dir_path);
closedir(mp_dir);
@@ -1146,7 +1146,7 @@ rte_mp_request_async(struct rte_mp_msg *req, const struct
timespec *ts,
dir_fd = dirfd(mp_dir);
/* lock the directory to prevent processes spinning up while we send */
- if (flock(dir_fd, LOCK_EX)) {
+ if (flock(dir_fd, LOCK_SH)) {
RTE_LOG(ERR, EAL, "Unable to lock directory %s\n",
mp_dir_path);
rte_errno = errno;
--
2.17.1
