rapidio: update asynchronous discovery initialization
authorAlexandre Bounine <alexandre.bounine@idt.com>
Wed, 10 Oct 2012 22:53:59 +0000 (15:53 -0700)
committerLinus Torvalds <torvalds@linux-foundation.org>
Wed, 10 Oct 2012 23:50:15 +0000 (08:50 +0900)
Update discovery process initialization based on Andrew Morton's comments:
https://lkml.org/lkml/2012/10/3/552.

This update processes all enumerating mports first and schedules discovery
work after that. If the initialization routine fails to allocate resources
needed to execute discovery, it abandons discovery for all ports.

Signed-off-by: Alexandre Bounine <alexandre.bounine@idt.com>
Cc: Matt Porter <mporter@kernel.crashing.org>
Cc: Li Yang <leoli@freescale.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
drivers/rapidio/rio.c

index d4bd69013c501010fc5ba76ce58c949d46708a9e..c17ae22567e0494c381bf023c13e48dfc729aa8b 100644 (file)
@@ -1275,49 +1275,68 @@ static void __devinit disc_work_handler(struct work_struct *_work)
        pr_debug("RIO: discovery work for mport %d %s\n",
                 work->mport->id, work->mport->name);
        rio_disc_mport(work->mport);
-
-       kfree(work);
 }
 
 int __devinit rio_init_mports(void)
 {
        struct rio_mport *port;
        struct rio_disc_work *work;
-       int no_disc = 0;
+       int n = 0;
+
+       if (!next_portid)
+               return -ENODEV;
 
+       /*
+        * First, run enumerations and check if we need to perform discovery
+        * on any of the registered mports.
+        */
        list_for_each_entry(port, &rio_mports, node) {
                if (port->host_deviceid >= 0)
                        rio_enum_mport(port);
-               else if (!no_disc) {
-                       if (!rio_wq) {
-                               rio_wq = alloc_workqueue("riodisc", 0, 0);
-                               if (!rio_wq) {
-                                       pr_err("RIO: unable allocate rio_wq\n");
-                                       no_disc = 1;
-                                       continue;
-                               }
-                       }
-
-                       work = kzalloc(sizeof *work, GFP_KERNEL);
-                       if (!work) {
-                               pr_err("RIO: no memory for work struct\n");
-                               no_disc = 1;
-                               continue;
-                       }
-
-                       work->mport = port;
-                       INIT_WORK(&work->work, disc_work_handler);
-                       queue_work(rio_wq, &work->work);
-               }
+               else
+                       n++;
+       }
+
+       if (!n)
+               goto no_disc;
+
+       /*
+        * If we have mports that require discovery schedule a discovery work
+        * for each of them. If the code below fails to allocate needed
+        * resources, exit without error to keep results of enumeration
+        * process (if any).
+        * TODO: Implement restart of dicovery process for all or
+        * individual discovering mports.
+        */
+       rio_wq = alloc_workqueue("riodisc", 0, 0);
+       if (!rio_wq) {
+               pr_err("RIO: unable allocate rio_wq\n");
+               goto no_disc;
        }
 
-       if (rio_wq) {
-               pr_debug("RIO: flush discovery workqueue\n");
-               flush_workqueue(rio_wq);
-               pr_debug("RIO: flush discovery workqueue finished\n");
+       work = kcalloc(n, sizeof *work, GFP_KERNEL);
+       if (!work) {
+               pr_err("RIO: no memory for work struct\n");
                destroy_workqueue(rio_wq);
+               goto no_disc;
        }
 
+       n = 0;
+       list_for_each_entry(port, &rio_mports, node) {
+               if (port->host_deviceid < 0) {
+                       work[n].mport = port;
+                       INIT_WORK(&work[n].work, disc_work_handler);
+                       queue_work(rio_wq, &work[n].work);
+                       n++;
+               }
+       }
+
+       flush_workqueue(rio_wq);
+       pr_debug("RIO: destroy discovery workqueue\n");
+       destroy_workqueue(rio_wq);
+       kfree(work);
+
+no_disc:
        rio_init();
 
        return 0;