[ovs-dev] [PATCH] poll-loop: windows poll_block implementation

Linda Sun lsun at vmware.com
Wed Dec 18 22:52:17 UTC 2013


Use WaitForMultipleObjects for polling on windows.  This works on all kinds
 of objects, e.g. sockets, files, especially ioctl calls to the kernel.
 poll_fd_wait_event() is used if events need to be passed to pollfds.
latch is signaled with event, to be waited/polled by WaitForMultipleObjects()
 as well.
Changed array of fds to hmap to check for duplicate fds.

Signed-off-by: Linda Sun <lsun at vmware.com>
---
 include/linux/types.h |    2 +
 lib/latch.c           |    2 +-
 lib/latch.h           |    4 ++
 lib/poll-loop.c       |  119 +++++++++++++++++++++++++++++++++++++++----------
 lib/poll-loop.h       |    5 ++-
 lib/timeval.c         |   20 +++++++--
 lib/timeval.h         |    4 +-
 7 files changed, 124 insertions(+), 32 deletions(-)

diff --git a/include/linux/types.h b/include/linux/types.h
index b88fb1c..4114b0f 100644
--- a/include/linux/types.h
+++ b/include/linux/types.h
@@ -55,4 +55,6 @@ typedef uint32_t __bitwise__ __be32;
 typedef uint64_t __bitwise__ __be64;
 #endif	/* no <linux/types.h> */
 
+typedef __u32 HANDLE;
+
 #endif /* <linux/types.h> */
diff --git a/lib/latch.c b/lib/latch.c
index bf518b9..20a6575 100644
--- a/lib/latch.c
+++ b/lib/latch.c
@@ -83,5 +83,5 @@ latch_is_set(const struct latch *latch)
 void
 latch_wait_at(const struct latch *latch, const char *where)
 {
-    poll_fd_wait_at(latch->fds[0], POLLIN, where);
+    poll_fd_wait_at(latch->fds[0], 0, POLLIN, where);
 }
diff --git a/lib/latch.h b/lib/latch.h
index 0b6e8a3..250e48e 100644
--- a/lib/latch.h
+++ b/lib/latch.h
@@ -27,6 +27,10 @@
 
 struct latch {
     int fds[2];
+#ifdef WIN32
+    HANDLE wevent;
+    bool is_set;
+#endif
 };
 
 void latch_init(struct latch *);
diff --git a/lib/poll-loop.c b/lib/poll-loop.c
index 5e3618b..06911dc 100644
--- a/lib/poll-loop.c
+++ b/lib/poll-loop.c
@@ -30,18 +30,24 @@
 #include "socket-util.h"
 #include "timeval.h"
 #include "vlog.h"
+#include "hmap.h"
+#include "hash.h"
 
 VLOG_DEFINE_THIS_MODULE(poll_loop);
 
 COVERAGE_DEFINE(poll_fd_wait);
 COVERAGE_DEFINE(poll_zero_timeout);
 
+struct poll_node {
+    struct hmap_node    hmap_node;
+    struct pollfd       poll_fd;    /* Events to pass to time_poll() */
+    HANDLE              wevent;     /* events for waitformultipleobjects */
+    const char          *where;     /* where each pollfd was created */
+};
+
 struct poll_loop {
     /* All active poll waiters. */
-    struct pollfd *pollfds;     /* Events to pass to poll(). */
-    const char **where;         /* Where each pollfd was created. */
-    size_t n_waiters;           /* Number of elems in 'where' and 'pollfds'. */
-    size_t allocated_waiters;   /* Allocated elems in 'where' and 'pollfds'. */
+    struct hmap poll_nodes;
 
     /* Time at which to wake up the next call to poll_block(), LLONG_MIN to
      * wake up immediately, or LLONG_MAX to wait forever. */
@@ -51,6 +57,20 @@ struct poll_loop {
 
 static struct poll_loop *poll_loop(void);
 
+/* Look up the node with same fd and wevent */
+static struct poll_node *
+poll_fd_node_find(struct poll_loop *loop, int fd, uint32_t wevent)
+{
+    struct poll_node *node;
+
+    HMAP_FOR_EACH_WITH_HASH(node, hmap_node, hash_2words(fd, wevent), &loop->poll_nodes) {
+        if (node->poll_fd.fd == fd && node->wevent == wevent) {
+            return node;
+        }
+    }
+    return NULL;
+}
+
 /* Registers 'fd' as waiting for the specified 'events' (which should be POLLIN
  * or POLLOUT or POLLIN | POLLOUT).  The following call to poll_block() will
  * wake up when 'fd' becomes ready for one or more of the requested events.
@@ -63,23 +83,37 @@ static struct poll_loop *poll_loop(void);
  * automatically provide the caller's source file and line number for
  * 'where'.) */
 void
-poll_fd_wait_at(int fd, short int events, const char *where)
+poll_fd_wait_at(int fd, HANDLE wevent, short int events, const char *where)
 {
     struct poll_loop *loop = poll_loop();
+    struct poll_node *node;
 
     COVERAGE_INC(poll_fd_wait);
-    if (loop->n_waiters >= loop->allocated_waiters) {
-        loop->where = x2nrealloc(loop->where, &loop->allocated_waiters,
-                                 sizeof *loop->where);
-        loop->pollfds = xrealloc(loop->pollfds,
-                                 (loop->allocated_waiters
-                                  * sizeof *loop->pollfds));
+
+#ifdef WIN32
+    /* null event cannot be polled */
+    if (wevent == 0) {
+        VLOG_ERR("No event to wait fd %d\n", fd);
+        return;
     }
+#endif
 
-    loop->where[loop->n_waiters] = where;
-    loop->pollfds[loop->n_waiters].fd = fd;
-    loop->pollfds[loop->n_waiters].events = events;
-    loop->n_waiters++;
+    /* check for duplicate.  If found, "or" the event */
+    node = poll_fd_node_find(loop, fd, wevent);
+    if (node) {
+        node->poll_fd.events |= events;
+    } else {
+        node = xzalloc(sizeof *node);
+        if (node == NULL) {
+            return;
+        }
+        node->where = where;
+        node->poll_fd.fd = fd;
+        node->wevent = wevent;
+        node->poll_fd.events = events;
+        hmap_insert(&loop->poll_nodes, &node->hmap_node, 
+            hash_2words(fd, wevent));
+    }
 }
 
 /* Causes the following call to poll_block() to block for no more than 'msec'
@@ -215,8 +249,12 @@ void
 poll_block(void)
 {
     struct poll_loop *loop = poll_loop();
+    struct poll_node *node, *next;
+    struct pollfd *pollfds;
+    HANDLE *wevents = NULL;
     int elapsed;
     int retval;
+    int i = 0;
 
     /* Register fatal signal events before actually doing any real work for
      * poll_block. */
@@ -227,7 +265,29 @@ poll_block(void)
     }
 
     timewarp_wait();
-    retval = time_poll(loop->pollfds, loop->n_waiters,
+    pollfds = xzalloc(hmap_count(&loop->poll_nodes) * sizeof *pollfds);
+    if (pollfds == NULL) {
+        return;
+    }
+
+#ifdef WIN32
+    wevents = xzalloc(hmap_count(&loop->poll_nodes) * sizeof *wevents);
+    if (wevents == NULL) {
+        free(pollfds);
+        return;
+    }
+#endif
+
+    /* populate with all the fds and events */
+    HMAP_FOR_EACH(node, hmap_node, &loop->poll_nodes) {
+        memcpy(&pollfds[i], &node->poll_fd, sizeof node->poll_fd);
+#ifdef WIN32
+        wevents[i] = node->wevent;
+#endif
+        i++;
+    }
+
+    retval = time_poll(pollfds, hmap_count(&loop->poll_nodes), wevents,
                        loop->timeout_when, &elapsed);
     if (retval < 0) {
         static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
@@ -235,18 +295,23 @@ poll_block(void)
     } else if (!retval) {
         log_wakeup(loop->timeout_where, NULL, elapsed);
     } else if (get_cpu_usage() > 50 || VLOG_IS_DBG_ENABLED()) {
-        size_t i;
-
-        for (i = 0; i < loop->n_waiters; i++) {
-            if (loop->pollfds[i].revents) {
-                log_wakeup(loop->where[i], &loop->pollfds[i], 0);
+        HMAP_FOR_EACH(node, hmap_node, &loop->poll_nodes) {
+            if (node->poll_fd.revents) {
+                log_wakeup(node->where, &node->poll_fd, 0);
             }
         }
     }
 
+    HMAP_FOR_EACH_SAFE(node, next, hmap_node, &loop->poll_nodes) {
+        hmap_remove(&loop->poll_nodes, &node->hmap_node);
+        free(node);
+    }
+
     loop->timeout_when = LLONG_MAX;
     loop->timeout_where = NULL;
-    loop->n_waiters = 0;
+    free(pollfds);
+    if (wevents) 
+        free(wevents);
 
     /* Handle any pending signals before doing anything else. */
     fatal_signal_run();
@@ -258,9 +323,14 @@ static void
 free_poll_loop(void *loop_)
 {
     struct poll_loop *loop = loop_;
+    struct poll_node *node, *next;
 
-    free(loop->pollfds);
-    free(loop->where);
+    HMAP_FOR_EACH_SAFE(node, next, hmap_node, &loop->poll_nodes) {
+        hmap_remove(&loop->poll_nodes, &node->hmap_node);
+        free(node);
+    }
+    
+    hmap_destroy(&loop->poll_nodes);
     free(loop);
 }
 
@@ -279,6 +349,7 @@ poll_loop(void)
     loop = pthread_getspecific(key);
     if (!loop) {
         loop = xzalloc(sizeof *loop);
+        hmap_init(&loop->poll_nodes);
         xpthread_setspecific(key, loop);
     }
     return loop;
diff --git a/lib/poll-loop.h b/lib/poll-loop.h
index 0397853..2db3fff 100644
--- a/lib/poll-loop.h
+++ b/lib/poll-loop.h
@@ -50,8 +50,9 @@ extern "C" {
  * caller to supply a location explicitly, which is useful if the caller's own
  * caller would be more useful in log output.  See timer_wait_at() for an
  * example. */
-void poll_fd_wait_at(int fd, short int events, const char *where);
-#define poll_fd_wait(fd, events) poll_fd_wait_at(fd, events, SOURCE_LOCATOR)
+void poll_fd_wait_at(int fd, uint32_t wevent, short int events, const char *where);
+#define poll_fd_wait(fd, events) poll_fd_wait_at(fd, 0, events, SOURCE_LOCATOR)
+#define poll_fd_wait_event(fd, wevent, events) poll_fd_wait_at(fd, wevent, events, SOURCE_LOCATOR)
 
 void poll_timer_wait_at(long long int msec, const char *where);
 #define poll_timer_wait(msec) poll_timer_wait_at(msec, SOURCE_LOCATOR)
diff --git a/lib/timeval.c b/lib/timeval.c
index 2ce45fc..8548448 100644
--- a/lib/timeval.c
+++ b/lib/timeval.c
@@ -232,12 +232,12 @@ time_alarm(unsigned int secs)
  *
  * Stores the number of milliseconds elapsed during poll in '*elapsed'. */
 int
-time_poll(struct pollfd *pollfds, int n_pollfds, long long int timeout_when,
-          int *elapsed)
+time_poll(struct pollfd *pollfds, int n_pollfds, HANDLE *handles,
+          long long int timeout_when, int *elapsed)
 {
     long long int *last_wakeup = last_wakeup_get();
     long long int start;
-    int retval;
+    int retval = 0;
 
     time_init();
     coverage_clear();
@@ -261,10 +261,24 @@ time_poll(struct pollfd *pollfds, int n_pollfds, long long int timeout_when,
             time_left = timeout_when - now;
         }
 
+#ifndef WIN32
+        if (handles) {
+            /* non-windows platform shouldn't really create these handles */
+        }
         retval = poll(pollfds, n_pollfds, time_left);
         if (retval < 0) {
             retval = -errno;
         }
+#else
+        if (n_pollfds != 0) {
+            retval = WaitForMultipleObjects(n_pollfds, handles, FALSE, time_left);
+        }
+        if (retval < 0) {
+            /* This will be replace by a win error to errno conversion function */
+            retval = -WSAGetLastError();
+            retval = -EINVAL;
+        }
+#endif
 
         if (deadline <= time_msec()) {
             fatal_signal_handler(SIGALRM);
diff --git a/lib/timeval.h b/lib/timeval.h
index 1bbfd5c..4162f63 100644
--- a/lib/timeval.h
+++ b/lib/timeval.h
@@ -52,8 +52,8 @@ long long int time_wall_msec(void);
 void time_timespec(struct timespec *);
 void time_wall_timespec(struct timespec *);
 void time_alarm(unsigned int secs);
-int time_poll(struct pollfd *, int n_pollfds, long long int timeout_when,
-              int *elapsed);
+int time_poll(struct pollfd *, int n_pollfds, HANDLE *handles, 
+              long long int timeout_when, int *elapsed);
 
 long long int timespec_to_msec(const struct timespec *);
 long long int timeval_to_msec(const struct timeval *);
-- 
1.7.9.5




More information about the dev mailing list