Skip to contentSkip to conversation

SLOW_WORK: Fix the CONFIG_MODULES=n case

PATCH In review David Howells 2 participants 3 messages Dec 1, 2009 VFS

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 SLOW_WORK: Fix the CONFIG_MODULES=n case

From: David Howells <[email protected]>

Commits 3d7a641 ("SLOW_WORK: Wait for outstanding work items belonging to a module to clear") introduced some code to make sure that all of a module's slow-work items were complete before that module was removed, and commit 3bde31a ("SLOW_WORK: Allow a requeueable work item to sleep till the thread is needed") further extended that, breaking it in the process if CONFIG_MODULES=n:

CC kernel/slow-work.o
kernel/slow-work.c: In function 'slow_work_execute':
kernel/slow-work.c:313: error: 'slow_work_thread_processing' undeclared (first use in this function) kernel/slow-work.c:313: error: (Each undeclared identifier is reported only once kernel/slow-work.c:313: error: for each function it appears in.)
kernel/slow-work.c: In function 'slow_work_wait_for_items':
kernel/slow-work.c:950: error: 'slow_work_unreg_sync_lock' undeclared (first use in this function) kernel/slow-work.c:951: error: 'slow_work_unreg_wq' undeclared (first use in this function) kernel/slow-work.c:961: error: 'slow_work_unreg_work_item' undeclared (first use in this function) kernel/slow-work.c:974: error: 'slow_work_unreg_module' undeclared (first use in this function) kernel/slow-work.c:977: error: 'slow_work_thread_processing' undeclared (first use in this function) make[1]: *** [kernel/slow-work.o] Error 1

Fix this by:

(1) Extracting the bits of slow_work_execute() that are contingent on CONFIG_MODULES, and the bits that should be, into inline functions and placing them into the #ifdef'd section that defines the relevant variables and adding stubs for moduleless kernels. This allows the removal of some #ifdefs.

(2) #ifdef'ing out the contents of slow_work_wait_for_items() in moduleless kernels.

The four functions related to handling module unloading synchronisation (and their associated variables) could be offloaded into a separate .c file, but each function is only used once and three of them are tiny, so doing so would prevent them from being inlined.

---

 kernel/slow-work.c |   46 +++++++++++++++++++++++++++++-----------------
 1 files changed, 29 insertions(+), 17 deletions(-)
kernel/slow-work.c +29 -17
index da94f3c..b5c17f1 100644
@@ -109,6 +109,30 @@ static struct module *slow_work_unreg_module;
 static struct slow_work *slow_work_unreg_work_item;
 static DECLARE_WAIT_QUEUE_HEAD(slow_work_unreg_wq);
 static DEFINE_MUTEX(slow_work_unreg_sync_lock);
+
+static void slow_work_set_thread_processing(int id, struct slow_work *work)
+{
+	if (work)
+		slow_work_thread_processing[id] = work->owner;
+}
+static void slow_work_done_thread_processing(int id, struct slow_work *work)
+{
+	struct module *module = slow_work_thread_processing[id];
+
+	slow_work_thread_processing[id] = NULL;
+	smp_mb();
+	if (slow_work_unreg_work_item == work ||
+	    slow_work_unreg_module == module)
+		wake_up_all(&slow_work_unreg_wq);
+}
+static void slow_work_clear_thread_processing(int id)
+{
+	slow_work_thread_processing[id] = NULL;
+}
+#else
+static void slow_work_set_thread_processing(int id, struct slow_work *work) {}
+static void slow_work_done_thread_processing(int id, struct slow_work *work) {}
+static void slow_work_clear_thread_processing(int id) {}
 #endif
 
 /*
@@ -197,9 +221,6 @@ static unsigned slow_work_calc_vsmax(void)
  */
 static noinline bool slow_work_execute(int id)
 {
-#ifdef CONFIG_MODULES
-	struct module *module;
-#endif
 	struct slow_work *work = NULL;
 	unsigned vsmax;
 	bool very_slow;
@@ -236,10 +257,7 @@ static noinline bool slow_work_execute(int id)
 		very_slow = false; /* avoid the compiler warning */
 	}
 
-#ifdef CONFIG_MODULES
-	if (work)
-		slow_work_thread_processing[id] = work->owner;
-#endif
+	slow_work_set_thread_processing(id, work);
 	if (work) {
 		slow_work_mark_time(work);
 		slow_work_begin_exec(id, work);
@@ -287,15 +305,7 @@ static noinline bool slow_work_execute(int id)
 
 	/* sort out the race between module unloading and put_ref() */
 	slow_work_put_ref(work);
-
-#ifdef CONFIG_MODULES
-	module = slow_work_thread_processing[id];
-	slow_work_thread_processing[id] = NULL;
-	smp_mb();
-	if (slow_work_unreg_work_item == work ||
-	    slow_work_unreg_module == module)
-		wake_up_all(&slow_work_unreg_wq);
-#endif
+	slow_work_done_thread_processing(id, work);
 
 	return true;
 
@@ -310,7 +320,7 @@ auto_requeue:
 	else
 		list_add_tail(&work->link, &slow_work_queue);
 	spin_unlock_irq(&slow_work_queue_lock);
-	slow_work_thread_processing[id] = NULL;
+	slow_work_clear_thread_processing(id);
 	return true;
 }
 
@@ -943,6 +953,7 @@ EXPORT_SYMBOL(slow_work_register_user);
  */
 static void slow_work_wait_for_items(struct module *module)
 {
+#ifdef CONFIG_MODULES
 	DECLARE_WAITQUEUE(myself, current);
 	struct slow_work *work;
 	int loop;
@@ -989,6 +1000,7 @@ static void slow_work_wait_for_items(struct module *module)
 
 	remove_wait_queue(&slow_work_unreg_wq, &myself);
 	mutex_unlock(&slow_work_unreg_sync_lock);
+#endif /* CONFIG_MODULES */
 }
 
 /**
Acked-by on v1 1/1
#1 lore raw
Quoted 9 lines · David Howells
kernel/slow-work.c +1 -0
@@ -943,6 +953,7 @@ EXPORT_SYMBOL(slow_work_register_user);
  */
 static void slow_work_wait_for_items(struct module *module)
 {
+#ifdef CONFIG_MODULES
 	DECLARE_WAITQUEUE(myself, current);
 	struct slow_work *work;
 	int loop;

kernel/slow-work.c static void slow_work_wait_for_items(struct module *module) · line 1004

Quoted 5 lines
kernel/slow-work.c +1 -0
@@ -989,6 +1000,7 @@ static void slow_work_wait_for_items(struct module *module)
 
 	remove_wait_queue(&slow_work_unreg_wq, &myself);
 	mutex_unlock(&slow_work_unreg_sync_lock);
+#endif /* CONFIG_MODULES */
 }

this slow_work_wait_for_items() function should move into the #ifdef block too.

With that fixed it looks good to me for .33 (but i havent tested it):

In terms of .32 i guess it's OK too and the fix is needed - but i'd really not have done even the preceding changes - why again did we need /proc/slow_work_rq via 8fba10a and why did it have to happen right before the final kernel?

If then it should have been done in debugfs - we dont need yet another /proc ABI.

Also, a very small aesthetic detail: i think the title should use the 'slow-work: ' prefix, not 'SLOW_WORK: '.

Thanks,

Ingo

#2 lore raw
Quoted 2 lines · Ingo Molnar

this slow_work_wait_for_items() function should move into the #ifdef block too.

I disagree: I want to keep the variable declaration blocks small; I'd rather not even put the inline functions in there that I did. I only did that because you wanted the #ifdef count reduced.

Quoted 3 lines

In terms of .32 i guess it's OK too and the fix is needed - but i'd really not have done even the preceding changes - why again did we need /proc/slow_work_rq via 8fba10a

The slow_work_rq debugging interface is not strictly necessary, but it proved a useful debugging tool. I emailed Linus before I went on holiday and asked if he was willing to take these not-strictly-necessary patches on which other patches were built, or whether he'd prefer me to drop those patches and adjust the rest.

Quoted 1 line

and why did it have to happen right before the final kernel?

Because it did. That's when I finished my set of patches and published them before going on holiday for a week - and that in turn was related to when I came up with a better test case. Sometimes coincidences do happen.

Quoted 2 lines

If then it should have been done in debugfs - we dont need yet another /proc ABI.

Possibly. That just means we have a debugfs ABI instead of a proc ABI - it needs maintaining either way. On the other hand, it can be moved there easily and the docs changed, and doing so makes a reasonable amount of sense - except that debugfs isn't normally mounted by at least Fedora for some reason.

David