]> err.no Git - linux-2.6/commitdiff
drivers/edac: fix reset edac_mc pollmsec
authorDoug Thompson <dougthompson@xmission.com>
Thu, 26 Jul 2007 17:41:14 +0000 (10:41 -0700)
committerLinus Torvalds <torvalds@woody.linux-foundation.org>
Thu, 26 Jul 2007 18:35:18 +0000 (11:35 -0700)
This fixes a deadlock that could occur on a 'setup' and 'teardown' sequence of
the workq for a edac_mc control structure instance.  A similiar fix was
previously implemented for the edac_device code.

In addition, the edac_mc device code there was missing code to allow the workq
period valu to be altered via sysfs control.

This patch adds that fix on the code, and allows for the changing of the
period value as well.

Cc: Alan Cox <alan@lxorguk.ukuu.org.uk>
Signed-off-by: Doug Thompson <dougthompson@xmission.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
drivers/edac/edac_mc.c
drivers/edac/edac_mc_sysfs.c
drivers/edac/edac_module.h

index 4471be3625995e59a8264ca91a3c6737e163e604..063a1bffe38b026ba7005ee0b105a4e956baf6e9 100644 (file)
@@ -214,6 +214,13 @@ void edac_mc_free(struct mem_ctl_info *mci)
 }
 EXPORT_SYMBOL_GPL(edac_mc_free);
 
+
+/*
+ * find_mci_by_dev
+ *
+ *     scan list of controllers looking for the one that manages
+ *     the 'dev' device
+ */
 static struct mem_ctl_info *find_mci_by_dev(struct device *dev)
 {
        struct mem_ctl_info *mci;
@@ -268,12 +275,6 @@ static void edac_mc_workq_function(struct work_struct *work_req)
        if (edac_mc_assert_error_check_and_clear() && (mci->edac_check != NULL))
                mci->edac_check(mci);
 
-       /*
-        * FIXME: temp place holder for PCI checks,
-        * goes away when we break out PCI
-        */
-       edac_pci_do_parity_check();
-
        mutex_unlock(&mem_ctls_mutex);
 
        /* Reschedule */
@@ -314,36 +315,55 @@ static void edac_mc_workq_teardown(struct mem_ctl_info *mci)
 {
        int status;
 
-       /* if not running POLL, leave now */
-       if (mci->op_state == OP_RUNNING_POLL) {
-               status = cancel_delayed_work(&mci->work);
-               if (status == 0) {
-                       debugf0("%s() not canceled, flush the queue\n",
-                               __func__);
+       status = cancel_delayed_work(&mci->work);
+       if (status == 0) {
+               debugf0("%s() not canceled, flush the queue\n",
+                       __func__);
 
-                       /* workq instance might be running, wait for it */
-                       flush_workqueue(edac_workqueue);
-               }
+               /* workq instance might be running, wait for it */
+               flush_workqueue(edac_workqueue);
        }
 }
 
 /*
- * edac_reset_delay_period
+ * edac_mc_reset_delay_period(unsigned long value)
+ *
+ *     user space has updated our poll period value, need to
+ *     reset our workq delays
  */
-static void edac_reset_delay_period(struct mem_ctl_info *mci, unsigned long value)
+void edac_mc_reset_delay_period(int value)
 {
-       /* cancel the current workq request */
-       edac_mc_workq_teardown(mci);
+       struct mem_ctl_info *mci;
+       struct list_head *item;
 
-       /* lock the list of devices for the new setup */
        mutex_lock(&mem_ctls_mutex);
 
-       /* restart the workq request, with new delay value */
-       edac_mc_workq_setup(mci, value);
+       /* scan the list and turn off all workq timers, doing so under lock
+        */
+       list_for_each(item, &mc_devices) {
+               mci = list_entry(item, struct mem_ctl_info, link);
+
+               if (mci->op_state == OP_RUNNING_POLL)
+                       cancel_delayed_work(&mci->work);
+       }
+
+       mutex_unlock(&mem_ctls_mutex);
+
+
+       /* re-walk the list, and reset the poll delay */
+       mutex_lock(&mem_ctls_mutex);
+
+       list_for_each(item, &mc_devices) {
+               mci = list_entry(item, struct mem_ctl_info, link);
+
+               edac_mc_workq_setup(mci, (unsigned long) value);
+       }
 
        mutex_unlock(&mem_ctls_mutex);
 }
 
+
+
 /* Return 0 on success, 1 on failure.
  * Before calling this function, caller must
  * assign a unique value to mci->mc_idx.
index cd090b0677a78b2606f09aa2e694357003f42a2c..4a0576bd06fc93bab9b2028b695ff7907a13f9c4 100644 (file)
@@ -122,6 +122,23 @@ static ssize_t memctrl_int_store(void *ptr, const char *buffer, size_t count)
        return count;
 }
 
+/*
+ * mc poll_msec time value
+ */
+static ssize_t poll_msec_int_store(void *ptr, const char *buffer, size_t count)
+{
+       int *value = (int *)ptr;
+
+       if (isdigit(*buffer)) {
+               *value = simple_strtoul(buffer, NULL, 0);
+
+               /* notify edac_mc engine to reset the poll period */
+               edac_mc_reset_delay_period(*value);
+       }
+
+       return count;
+}
+
 
 /* EDAC sysfs CSROW data structures and methods
  */
@@ -704,7 +721,7 @@ MEMCTRL_ATTR(edac_mc_log_ce,
        S_IRUGO | S_IWUSR, memctrl_int_show, memctrl_int_store);
 
 MEMCTRL_ATTR(edac_mc_poll_msec,
-       S_IRUGO | S_IWUSR, memctrl_int_show, memctrl_int_store);
+       S_IRUGO | S_IWUSR, memctrl_int_show, poll_msec_int_store);
 
 /* Base Attributes of the memory ECC object */
 static struct memctrl_dev_attribute *memctrl_attr[] = {
index a2134dfc3cc691783122cddec1a9906d982b00ef..3664ae9ccd63853ad45349017c291a05134b8b92 100644 (file)
@@ -52,6 +52,8 @@ extern void edac_device_workq_setup(struct edac_device_ctl_info *edac_dev,
 extern void edac_device_workq_teardown(struct edac_device_ctl_info *edac_dev);
 extern void edac_device_reset_delay_period(struct edac_device_ctl_info
                                           *edac_dev, unsigned long value);
+extern void edac_mc_reset_delay_period(int value);
+
 extern void *edac_align_ptr(void *ptr, unsigned size);
 
 /*