This is an automated email from the ASF dual-hosted git repository. xiaoxiang pushed a commit to branch master in repository https://gitbox.apache.org/repos/asf/nuttx.git
commit 90a8e2e4220856b337d67fa2e5c36d46f928aa24 Author: Zhe Weng <[email protected]> AuthorDate: Tue Aug 29 22:12:23 2023 +0800 usrsock_rpmsg_server: Keep msg order in recursive call In `drivers/rptun/rptun.c`, we have a `rptun_is_recursive` function, which lets rptun thread run recursively. Then the `usrsock_rpmsg_ept_cb` may be called inside `usrsock_rpmsg_ept_cb`, which can cause some unexpected behavior, so we add a queue to keep the order of incoming messages to reduce the complexity. Signed-off-by: Zhe Weng <[email protected]> --- drivers/usrsock/usrsock_rpmsg_server.c | 98 +++++++++++++++++++++++++++++++--- 1 file changed, 90 insertions(+), 8 deletions(-) diff --git a/drivers/usrsock/usrsock_rpmsg_server.c b/drivers/usrsock/usrsock_rpmsg_server.c index 218a80aaf9..9020ec8201 100644 --- a/drivers/usrsock/usrsock_rpmsg_server.c +++ b/drivers/usrsock/usrsock_rpmsg_server.c @@ -35,6 +35,7 @@ #include <nuttx/mutex.h> #include <nuttx/net/dns.h> #include <nuttx/net/net.h> +#include <nuttx/queue.h> #include <nuttx/rptun/openamp.h> #include <nuttx/usrsock/usrsock_rpmsg.h> #ifdef CONFIG_NETDEV_WIRELESS_IOCTL @@ -53,11 +54,31 @@ struct usrsock_rpmsg_s struct pollfd pfds[CONFIG_NET_USRSOCK_RPMSG_SERVER_NSOCKS]; }; +/* Saving rpmsg requests to keep message order. */ + +struct usrsock_rpmsg_req_s +{ + sq_entry_t flink; + FAR void *data; + size_t len; + uint32_t src; +}; + struct usrsock_rpmsg_ept_s { struct rpmsg_endpoint ept; + + /* For sendto/recvfrom */ + struct iovec iov[CONFIG_NET_USRSOCK_RPMSG_SERVER_NIOVEC]; ssize_t remain; + + /* For keeping msg order, normally nIOVec is the max request we can get */ + + bool inuse; + sq_queue_t req_free; + sq_queue_t req_pending; + struct usrsock_rpmsg_req_s reqs[CONFIG_NET_USRSOCK_RPMSG_SERVER_NIOVEC]; }; /**************************************************************************** @@ -989,6 +1010,7 @@ static void usrsock_rpmsg_ns_bind(FAR struct rpmsg_device *rdev, FAR struct usrsock_rpmsg_s *priv = priv_; FAR struct usrsock_rpmsg_ept_s *uept; int ret; + int i; uept = kmm_zalloc(sizeof(*uept)); if (!uept) @@ -997,6 +1019,10 @@ static void usrsock_rpmsg_ns_bind(FAR struct rpmsg_device *rdev, } uept->ept.priv = priv; + for (i = 0; i < CONFIG_NET_USRSOCK_RPMSG_SERVER_NIOVEC; i++) + { + sq_addlast(&uept->reqs[i].flink, &uept->req_free); + } ret = rpmsg_create_ept(&uept->ept, rdev, USRSOCK_RPMSG_EPT_NAME, RPMSG_ADDR_ANY, dest, @@ -1042,28 +1068,84 @@ static void usrsock_rpmsg_ns_unbind(FAR struct rpmsg_endpoint *ept) kmm_free(ept); } -static int usrsock_rpmsg_ept_cb(FAR struct rpmsg_endpoint *ept, - FAR void *data, size_t len, uint32_t src, - FAR void *priv_) +static int usrsock_rpmsg_ept_do_cb(FAR struct usrsock_rpmsg_ept_s *uept, + FAR void *data, size_t len, uint32_t src, + FAR struct usrsock_rpmsg_s *priv) { FAR struct usrsock_request_common_s *common = data; - FAR struct usrsock_rpmsg_ept_s *uept = - (FAR struct usrsock_rpmsg_ept_s *)ept; - FAR struct usrsock_rpmsg_s *priv = priv_; if (uept->remain > 0) { - return usrsock_rpmsg_sendto_handler(ept, data, len, src, priv); + return usrsock_rpmsg_sendto_handler(&uept->ept, data, len, src, priv); } else if (common->reqid >= 0 && common->reqid <= USRSOCK_REQUEST__MAX) { - return g_usrsock_rpmsg_handler[common->reqid](ept, data, len, + return g_usrsock_rpmsg_handler[common->reqid](&uept->ept, data, len, src, priv); } return -EINVAL; } +static int usrsock_rpmsg_ept_cb(FAR struct rpmsg_endpoint *ept, + FAR void *data, size_t len, uint32_t src, + FAR void *priv_) +{ + FAR struct usrsock_rpmsg_s *priv = priv_; + FAR struct usrsock_rpmsg_ept_s *uept = + (FAR struct usrsock_rpmsg_ept_s *)ept; + FAR struct usrsock_rpmsg_req_s *req; + int ret; + + /* This callback is called from only one thread per ept, so don't need any + * lock for `inuse` state. + */ + + if (uept->inuse) + { + /* Avoid recursive call. */ + + req = (FAR struct usrsock_rpmsg_req_s *)sq_remfirst(&uept->req_free); + if (req == NULL) + { + return -ENOMEM; + } + + req->data = data; + req->len = len; + req->src = src; + sq_addlast(&req->flink, &uept->req_pending); + + rpmsg_hold_rx_buffer(ept, data); + return OK; + } + + uept->inuse = true; + ret = usrsock_rpmsg_ept_do_cb(uept, data, len, src, priv); + + /* Pop pending requests to proceed. */ + + while ((req = (FAR struct usrsock_rpmsg_req_s *) + sq_remfirst(&uept->req_pending)) != NULL) + { + data = req->data; + len = req->len; + src = req->src; + sq_addlast(&req->flink, &uept->req_free); + + ret = usrsock_rpmsg_ept_do_cb(uept, data, len, src, priv); + if (ret < 0) + { + nerr("ERROR: usrsock got error %d!", ret); + } + + rpmsg_release_rx_buffer(ept, data); + } + + uept->inuse = false; + return ret; +} + static void usrsock_rpmsg_poll_setup(FAR struct pollfd *pfds, pollevent_t events) {
