[PATCH] blktrace: reduce context switches a lot, event count fix, etc

Transmit dropped events over the net as well, so the other end know if
we lost events.  Use POLLMSG from relayfs to indicate subbuf switch.

diff --git a/blktrace.c b/blktrace.c
index 89f0ce6b56ff70d97225e5c3a7eea4e27af4a7cf..29e55b84e6e87d6a95462657b4827f0943d77c16 100644 (file)
--- a/blktrace.c
+++ b/blktrace.c
@@ -186,7 +186,6 @@ struct thread_information {
        off_t ofile_offset;
        int ofile_stdout;
        int ofile_mmap;
-       volatile int sendfile_pending;
 
        int (*get_subbuf)(struct thread_information *, unsigned int);
        int (*flush_subbuf)(struct thread_information *, struct tip_subbuf *);
@@ -395,13 +394,16 @@ static void stop_all_traces(void)
        }
 }
 
-static void wait_for_data(struct thread_information *tip)
+static void wait_for_data(struct thread_information *tip, int events)
 {
-       struct pollfd pfd = { .fd = tip->fd, .events = POLLIN };
+       struct pollfd pfd = { .fd = tip->fd, .events = events };
 
        do {
-               poll(&pfd, 1, 100);
-               if (pfd.revents & POLLIN)
+               if (poll(&pfd, 1, 100) < 0) {
+                       perror("poll");
+                       break;
+               }
+               if (pfd.revents & events)
                        break;
                if (tip->ofile_stdout)
                        break;
@@ -414,7 +416,7 @@ static int read_data_file(struct thread_information *tip, void *buf,
        int ret = 0;
 
        do {
-               wait_for_data(tip);
+               wait_for_data(tip, POLLIN);
 
                ret = read(tip->fd, buf, len);
                if (!ret)
@@ -463,12 +465,6 @@ static int read_data_net(struct thread_information *tip, void *buf,
        return len - bytes_left;
 }
 
-static int read_data(struct thread_information *tip, void *buf,
-                    unsigned int len)
-{
-       return tip->read_data(tip, buf, len);
-}
-
 static inline struct tip_subbuf *
 subbuf_fifo_dequeue(struct thread_information *tip)
 {
@@ -539,7 +535,7 @@ static int mmap_subbuf(struct thread_information *tip, unsigned int maxlen)
                mlock(tip->fs_buf, tip->fs_buf_len);
        }
 
-       ret = read_data(tip, tip->fs_buf + tip->fs_off, maxlen);
+       ret = tip->read_data(tip, tip->fs_buf + tip->fs_off, maxlen);
        if (ret >= 0) {
                tip->data_read += ret;
                tip->fs_size += ret;
@@ -561,7 +557,7 @@ static int get_subbuf(struct thread_information *tip, unsigned int maxlen)
        ts->buf = malloc(buf_size);
        ts->max_len = maxlen;
 
-       ret = read_data(tip, ts->buf, ts->max_len);
+       ret = tip->read_data(tip, ts->buf, ts->max_len);
        if (ret > 0) {
                ts->len = ret;
                tip->data_read += ret;
@@ -579,7 +575,7 @@ static int get_subbuf_sendfile(struct thread_information *tip,
        struct stat sb;
        unsigned int ready;
 
-       wait_for_data(tip);
+       wait_for_data(tip, POLLMSG);
 
        /*
         * hack to get last data out, we can't use sendfile for that
@@ -587,18 +583,15 @@ static int get_subbuf_sendfile(struct thread_information *tip,
        if (is_done())
                return get_subbuf(tip, maxlen);
 
-       if (tip->sendfile_pending) {
-               usleep(100);
-               return 0;
-       }
-
        if (fstat(tip->fd, &sb) < 0) {
                perror("trace stat");
                return -1;
        }
        ready = sb.st_size - tip->data_queued;
-       if (!ready)
+       if (!ready) {
+               usleep(1000);
                return 0;
+       }
 
        ts = malloc(sizeof(*ts));
        ts->buf = NULL;
@@ -609,7 +602,6 @@ static int get_subbuf_sendfile(struct thread_information *tip,
        if (subbuf_fifo_queue(tip, ts))
                return -1;
 
-       tip->sendfile_pending++;
        return ready;
 }
 
@@ -722,13 +714,20 @@ static int net_send_header(struct thread_information *tip, unsigned int len)
  */
 static void net_client_send_close(void)
 {
+       struct device_information *dip;
        struct blktrace_net_hdr hdr;
+       int i;
 
        hdr.magic = BLK_IO_TRACE_MAGIC;
        hdr.cpu = 0;
        hdr.max_cpus = ncpus;
        hdr.len = 0;
 
+       for_each_dip(dip, i) {
+               strcpy(hdr.buts_name, dip->buts_name);
+               hdr.cpu += get_dropped_count(dip->buts_name);
+       }
+
        write_data_net(net_out_fd, &hdr, sizeof(hdr));
 }
 
@@ -736,13 +735,13 @@ static int flush_subbuf_net(struct thread_information *tip,
                            struct tip_subbuf *ts)
 {
        if (net_send_header(tip, ts->len))
-               return 1;
+               return -1;
        if (write_data_net(net_out_fd, ts->buf, ts->len))
-               return 1;
+               return -1;
 
        free(ts->buf);
        free(ts);
-       return 0;
+       return 1;
 }
 
 static int net_sendfile(struct thread_information *tip, struct tip_subbuf *ts)
@@ -763,7 +762,7 @@ static int net_sendfile(struct thread_information *tip, struct tip_subbuf *ts)
 static int flush_subbuf_sendfile(struct thread_information *tip,
                                 struct tip_subbuf *ts)
 {
-       int ret = 1;
+       int ret = -1;
 
        /*
         * currently we cannot use sendfile() on the last bytes read, as they
@@ -780,9 +779,8 @@ static int flush_subbuf_sendfile(struct thread_information *tip,
 
        tip->data_read += ts->len;
        tip->ofile_offset += buf_size;
-       ret = 0;
+       ret = 1;
 err:
-       tip->sendfile_pending--;
        free(ts);
        return ret;
 }
@@ -908,7 +906,7 @@ static void get_and_write_events(void)
                }
 
                if (!events)
-                       usleep(10);
+                       usleep(100000);
        }
 
        /*
@@ -945,7 +943,7 @@ static void wait_for_threads(void)
 
                do {
                        tips_running = 0;
-                       usleep(1000);
+                       usleep(100000);
 
                        for_each_dip(dip, i)
                                for_each_tip(dip, tip, j)
@@ -1318,10 +1316,10 @@ static int net_get_header(struct blktrace_net_hdr *bnh)
                                perror("recv header");
                                return 1;
                        }
-                       usleep(100);
+                       usleep(1000);
                        continue;
                } else if (!ret) {
-                       usleep(100);
+                       usleep(1000);
                        continue;
                } else {
                        p += ret;
@@ -1360,6 +1358,14 @@ static int net_server_loop(struct in_addr *cl_in_addr)
         * len == 0 means that the other end signalled end-of-run
         */
        if (!bnh.len) {
+               /*
+                * overload cpu count with dropped events
+                */
+               struct device_information *dip;
+
+               dip = net_get_dip(bnh.buts_name, cl_in_addr);
+               dip->drop_count = bnh.cpu;
+
                fprintf(stderr, "server: end of run\n");
                return 1;
        }